fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/compat.h>
  34 #include <linux/bit_spinlock.h>
  35 #include <linux/xattr.h>
  36 #include <linux/posix_acl.h>
  37 #include <linux/falloc.h>
  38 #include <linux/slab.h>
  39 #include <linux/ratelimit.h>
  40 #include <linux/mount.h>
  41 #include <linux/btrfs.h>
  42 #include <linux/blkdev.h>
  43 #include <linux/posix_acl_xattr.h>
  44 #include <linux/uio.h>
  45 #include <linux/magic.h>
  46 #include "ctree.h"
  47 #include "disk-io.h"
  48 #include "transaction.h"
  49 #include "btrfs_inode.h"
  50 #include "print-tree.h"
  51 #include "ordered-data.h"
  52 #include "xattr.h"
  53 #include "tree-log.h"
  54 #include "volumes.h"
  55 #include "compression.h"
  56 #include "locking.h"
  57 #include "free-space-cache.h"
  58 #include "inode-map.h"
  59 #include "backref.h"
  60 #include "hash.h"
  61 #include "props.h"
  62 #include "qgroup.h"
  63 #include "dedupe.h"
  64
  65 struct btrfs_iget_args {
  66         struct btrfs_key *location;
  67         struct btrfs_root *root;
  68 };
  69
  70 struct btrfs_dio_data {
  71         u64 reserve;
  72         u64 unsubmitted_oe_range_start;
  73         u64 unsubmitted_oe_range_end;
  74         int overwrite;
  75 };
  76
  77 static const struct inode_operations btrfs_dir_inode_operations;
  78 static const struct inode_operations btrfs_symlink_inode_operations;
  79 static const struct inode_operations btrfs_dir_ro_inode_operations;
  80 static const struct inode_operations btrfs_special_inode_operations;
  81 static const struct inode_operations btrfs_file_inode_operations;
  82 static const struct address_space_operations btrfs_aops;
  83 static const struct address_space_operations btrfs_symlink_aops;
  84 static const struct file_operations btrfs_dir_file_operations;
  85 static const struct extent_io_ops btrfs_extent_io_ops;
  86
  87 static struct kmem_cache *btrfs_inode_cachep;
  88 struct kmem_cache *btrfs_trans_handle_cachep;
  89 struct kmem_cache *btrfs_path_cachep;
  90 struct kmem_cache *btrfs_free_space_cachep;
  91
  92 #define S_SHIFT 12
  93 static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  94         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  95         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  96         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  97         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  98         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  99         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
 100         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
 101 };
 102
 103 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
 104 static int btrfs_truncate(struct inode *inode);
 105 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
 106 static noinline int cow_file_range(struct inode *inode,
 107                                    struct page *locked_page,
 108                                    u64 start, u64 end, u64 delalloc_end,
 109                                    int *page_started, unsigned long *nr_written,
 110                                    int unlock, struct btrfs_dedupe_hash *hash);
 111 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
 112                                        u64 orig_start, u64 block_start,
 113                                        u64 block_len, u64 orig_block_len,
 114                                        u64 ram_bytes, int compress_type,
 115                                        int type);
 116
 117 static void __endio_write_update_ordered(struct inode *inode,
 118                                          const u64 offset, const u64 bytes,
 119                                          const bool uptodate);
 120
 121 /*
 122  * Cleanup all submitted ordered extents in specified range to handle errors
 123  * from the fill_dellaloc() callback.
 124  *
 125  * NOTE: caller must ensure that when an error happens, it can not call
 126  * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
 127  * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
 128  * to be released, which we want to happen only when finishing the ordered
 129  * extent (btrfs_finish_ordered_io()). Also note that the caller of the
 130  * fill_delalloc() callback already does proper cleanup for the first page of
 131  * the range, that is, it invokes the callback writepage_end_io_hook() for the
 132  * range of the first page.
 133  */
 134 static inline void btrfs_cleanup_ordered_extents(struct inode *inode,
 135                                                  const u64 offset,
 136                                                  const u64 bytes)
 137 {
 138         unsigned long index = offset >> PAGE_SHIFT;
 139         unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
 140         struct page *page;
 141
 142         while (index <= end_index) {
 143                 page = find_get_page(inode->i_mapping, index);
 144                 index++;
 145                 if (!page)
 146                         continue;
 147                 ClearPagePrivate2(page);
 148                 put_page(page);
 149         }
 150         return __endio_write_update_ordered(inode, offset + PAGE_SIZE,
 151                                             bytes - PAGE_SIZE, false);
 152 }
 153
 154 static int btrfs_dirty_inode(struct inode *inode);
 155
 156 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
 157 void btrfs_test_inode_set_ops(struct inode *inode)
 158 {
 159         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
 160 }
 161 #endif
 162
 163 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 164                                      struct inode *inode,  struct inode *dir,
 165                                      const struct qstr *qstr)
 166 {
 167         int err;
 168
 169         err = btrfs_init_acl(trans, inode, dir);
 170         if (!err)
 171                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 172         return err;
 173 }
 174
 175 /*
 176  * this does all the hard work for inserting an inline extent into
 177  * the btree.  The caller should have done a btrfs_drop_extents so that
 178  * no overlapping inline items exist in the btree
 179  */
 180 static int insert_inline_extent(struct btrfs_trans_handle *trans,
 181                                 struct btrfs_path *path, int extent_inserted,
 182                                 struct btrfs_root *root, struct inode *inode,
 183                                 u64 start, size_t size, size_t compressed_size,
 184                                 int compress_type,
 185                                 struct page **compressed_pages)
 186 {
 187         struct extent_buffer *leaf;
 188         struct page *page = NULL;
 189         char *kaddr;
 190         unsigned long ptr;
 191         struct btrfs_file_extent_item *ei;
 192         int ret;
 193         size_t cur_size = size;
 194         unsigned long offset;
 195
 196         if (compressed_size && compressed_pages)
 197                 cur_size = compressed_size;
 198
 199         inode_add_bytes(inode, size);
 200
 201         if (!extent_inserted) {
 202                 struct btrfs_key key;
 203                 size_t datasize;
 204
 205                 key.objectid = btrfs_ino(BTRFS_I(inode));
 206                 key.offset = start;
 207                 key.type = BTRFS_EXTENT_DATA_KEY;
 208
 209                 datasize = btrfs_file_extent_calc_inline_size(cur_size);
 210                 path->leave_spinning = 1;
 211                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 212                                               datasize);
 213                 if (ret)
 214                         goto fail;
 215         }
 216         leaf = path->nodes[0];
 217         ei = btrfs_item_ptr(leaf, path->slots[0],
 218                             struct btrfs_file_extent_item);
 219         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 220         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 221         btrfs_set_file_extent_encryption(leaf, ei, 0);
 222         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 223         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 224         ptr = btrfs_file_extent_inline_start(ei);
 225
 226         if (compress_type != BTRFS_COMPRESS_NONE) {
 227                 struct page *cpage;
 228                 int i = 0;
 229                 while (compressed_size > 0) {
 230                         cpage = compressed_pages[i];
 231                         cur_size = min_t(unsigned long, compressed_size,
 232                                        PAGE_SIZE);
 233
 234                         kaddr = kmap_atomic(cpage);
 235                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 236                         kunmap_atomic(kaddr);
 237
 238                         i++;
 239                         ptr += cur_size;
 240                         compressed_size -= cur_size;
 241                 }
 242                 btrfs_set_file_extent_compression(leaf, ei,
 243                                                   compress_type);
 244         } else {
 245                 page = find_get_page(inode->i_mapping,
 246                                      start >> PAGE_SHIFT);
 247                 btrfs_set_file_extent_compression(leaf, ei, 0);
 248                 kaddr = kmap_atomic(page);
 249                 offset = start & (PAGE_SIZE - 1);
 250                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 251                 kunmap_atomic(kaddr);
 252                 put_page(page);
 253         }
 254         btrfs_mark_buffer_dirty(leaf);
 255         btrfs_release_path(path);
 256
 257         /*
 258          * we're an inline extent, so nobody can
 259          * extend the file past i_size without locking
 260          * a page we already have locked.
 261          *
 262          * We must do any isize and inode updates
 263          * before we unlock the pages.  Otherwise we
 264          * could end up racing with unlink.
 265          */
 266         BTRFS_I(inode)->disk_i_size = inode->i_size;
 267         ret = btrfs_update_inode(trans, root, inode);
 268
 269 fail:
 270         return ret;
 271 }
 272
 273
 274 /*
 275  * conditionally insert an inline extent into the file.  This
 276  * does the checks required to make sure the data is small enough
 277  * to fit as an inline extent.
 278  */
 279 static noinline int cow_file_range_inline(struct btrfs_root *root,
 280                                           struct inode *inode, u64 start,
 281                                           u64 end, size_t compressed_size,
 282                                           int compress_type,
 283                                           struct page **compressed_pages)
 284 {
 285         struct btrfs_fs_info *fs_info = root->fs_info;
 286         struct btrfs_trans_handle *trans;
 287         u64 isize = i_size_read(inode);
 288         u64 actual_end = min(end + 1, isize);
 289         u64 inline_len = actual_end - start;
 290         u64 aligned_end = ALIGN(end, fs_info->sectorsize);
 291         u64 data_len = inline_len;
 292         int ret;
 293         struct btrfs_path *path;
 294         int extent_inserted = 0;
 295         u32 extent_item_size;
 296
 297         if (compressed_size)
 298                 data_len = compressed_size;
 299
 300         if (start > 0 ||
 301             actual_end > fs_info->sectorsize ||
 302             data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
 303             (!compressed_size &&
 304             (actual_end & (fs_info->sectorsize - 1)) == 0) ||
 305             end + 1 < isize ||
 306             data_len > fs_info->max_inline) {
 307                 return 1;
 308         }
 309
 310         path = btrfs_alloc_path();
 311         if (!path)
 312                 return -ENOMEM;
 313
 314         trans = btrfs_join_transaction(root);
 315         if (IS_ERR(trans)) {
 316                 btrfs_free_path(path);
 317                 return PTR_ERR(trans);
 318         }
 319         trans->block_rsv = &BTRFS_I(inode)->block_rsv;
 320
 321         if (compressed_size && compressed_pages)
 322                 extent_item_size = btrfs_file_extent_calc_inline_size(
 323                    compressed_size);
 324         else
 325                 extent_item_size = btrfs_file_extent_calc_inline_size(
 326                     inline_len);
 327
 328         ret = __btrfs_drop_extents(trans, root, inode, path,
 329                                    start, aligned_end, NULL,
 330                                    1, 1, extent_item_size, &extent_inserted);
 331         if (ret) {
 332                 btrfs_abort_transaction(trans, ret);
 333                 goto out;
 334         }
 335
 336         if (isize > actual_end)
 337                 inline_len = min_t(u64, isize, actual_end);
 338         ret = insert_inline_extent(trans, path, extent_inserted,
 339                                    root, inode, start,
 340                                    inline_len, compressed_size,
 341                                    compress_type, compressed_pages);
 342         if (ret && ret != -ENOSPC) {
 343                 btrfs_abort_transaction(trans, ret);
 344                 goto out;
 345         } else if (ret == -ENOSPC) {
 346                 ret = 1;
 347                 goto out;
 348         }
 349
 350         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
 351         btrfs_drop_extent_cache(BTRFS_I(inode), start, aligned_end - 1, 0);
 352 out:
 353         /*
 354          * Don't forget to free the reserved space, as for inlined extent
 355          * it won't count as data extent, free them directly here.
 356          * And at reserve time, it's always aligned to page size, so
 357          * just free one page here.
 358          */
 359         btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
 360         btrfs_free_path(path);
 361         btrfs_end_transaction(trans);
 362         return ret;
 363 }
 364
 365 struct async_extent {
 366         u64 start;
 367         u64 ram_size;
 368         u64 compressed_size;
 369         struct page **pages;
 370         unsigned long nr_pages;
 371         int compress_type;
 372         struct list_head list;
 373 };
 374
 375 struct async_cow {
 376         struct inode *inode;
 377         struct btrfs_root *root;
 378         struct page *locked_page;
 379         u64 start;
 380         u64 end;
 381         struct list_head extents;
 382         struct btrfs_work work;
 383 };
 384
 385 static noinline int add_async_extent(struct async_cow *cow,
 386                                      u64 start, u64 ram_size,
 387                                      u64 compressed_size,
 388                                      struct page **pages,
 389                                      unsigned long nr_pages,
 390                                      int compress_type)
 391 {
 392         struct async_extent *async_extent;
 393
 394         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 395         BUG_ON(!async_extent); /* -ENOMEM */
 396         async_extent->start = start;
 397         async_extent->ram_size = ram_size;
 398         async_extent->compressed_size = compressed_size;
 399         async_extent->pages = pages;
 400         async_extent->nr_pages = nr_pages;
 401         async_extent->compress_type = compress_type;
 402         list_add_tail(&async_extent->list, &cow->extents);
 403         return 0;
 404 }
 405
 406 static inline int inode_need_compress(struct inode *inode, u64 start, u64 end)
 407 {
 408         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 409
 410         /* force compress */
 411         if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
 412                 return 1;
 413         /* defrag ioctl */
 414         if (BTRFS_I(inode)->defrag_compress)
 415                 return 1;
 416         /* bad compression ratios */
 417         if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 418                 return 0;
 419         if (btrfs_test_opt(fs_info, COMPRESS) ||
 420             BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
 421             BTRFS_I(inode)->prop_compress)
 422                 return btrfs_compress_heuristic(inode, start, end);
 423         return 0;
 424 }
 425
 426 static inline void inode_should_defrag(struct btrfs_inode *inode,
 427                 u64 start, u64 end, u64 num_bytes, u64 small_write)
 428 {
 429         /* If this is a small write inside eof, kick off a defrag */
 430         if (num_bytes < small_write &&
 431             (start > 0 || end + 1 < inode->disk_i_size))
 432                 btrfs_add_inode_defrag(NULL, inode);
 433 }
 434
 435 /*
 436  * we create compressed extents in two phases.  The first
 437  * phase compresses a range of pages that have already been
 438  * locked (both pages and state bits are locked).
 439  *
 440  * This is done inside an ordered work queue, and the compression
 441  * is spread across many cpus.  The actual IO submission is step
 442  * two, and the ordered work queue takes care of making sure that
 443  * happens in the same order things were put onto the queue by
 444  * writepages and friends.
 445  *
 446  * If this code finds it can't get good compression, it puts an
 447  * entry onto the work queue to write the uncompressed bytes.  This
 448  * makes sure that both compressed inodes and uncompressed inodes
 449  * are written in the same order that the flusher thread sent them
 450  * down.
 451  */
 452 static noinline void compress_file_range(struct inode *inode,
 453                                         struct page *locked_page,
 454                                         u64 start, u64 end,
 455                                         struct async_cow *async_cow,
 456                                         int *num_added)
 457 {
 458         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 459         struct btrfs_root *root = BTRFS_I(inode)->root;
 460         u64 blocksize = fs_info->sectorsize;
 461         u64 actual_end;
 462         u64 isize = i_size_read(inode);
 463         int ret = 0;
 464         struct page **pages = NULL;
 465         unsigned long nr_pages;
 466         unsigned long total_compressed = 0;
 467         unsigned long total_in = 0;
 468         int i;
 469         int will_compress;
 470         int compress_type = fs_info->compress_type;
 471         int redirty = 0;
 472
 473         inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
 474                         SZ_16K);
 475
 476         actual_end = min_t(u64, isize, end + 1);
 477 again:
 478         will_compress = 0;
 479         nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
 480         BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
 481         nr_pages = min_t(unsigned long, nr_pages,
 482                         BTRFS_MAX_COMPRESSED / PAGE_SIZE);
 483
 484         /*
 485          * we don't want to send crud past the end of i_size through
 486          * compression, that's just a waste of CPU time.  So, if the
 487          * end of the file is before the start of our current
 488          * requested range of bytes, we bail out to the uncompressed
 489          * cleanup code that can deal with all of this.
 490          *
 491          * It isn't really the fastest way to fix things, but this is a
 492          * very uncommon corner.
 493          */
 494         if (actual_end <= start)
 495                 goto cleanup_and_bail_uncompressed;
 496
 497         total_compressed = actual_end - start;
 498
 499         /*
 500          * skip compression for a small file range(<=blocksize) that
 501          * isn't an inline extent, since it doesn't save disk space at all.
 502          */
 503         if (total_compressed <= blocksize &&
 504            (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 505                 goto cleanup_and_bail_uncompressed;
 506
 507         total_compressed = min_t(unsigned long, total_compressed,
 508                         BTRFS_MAX_UNCOMPRESSED);
 509         total_in = 0;
 510         ret = 0;
 511
 512         /*
 513          * we do compression for mount -o compress and when the
 514          * inode has not been flagged as nocompress.  This flag can
 515          * change at any time if we discover bad compression ratios.
 516          */
 517         if (inode_need_compress(inode, start, end)) {
 518                 WARN_ON(pages);
 519                 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
 520                 if (!pages) {
 521                         /* just bail out to the uncompressed code */
 522                         goto cont;
 523                 }
 524
 525                 if (BTRFS_I(inode)->defrag_compress)
 526                         compress_type = BTRFS_I(inode)->defrag_compress;
 527                 else if (BTRFS_I(inode)->prop_compress)
 528                         compress_type = BTRFS_I(inode)->prop_compress;
 529
 530                 /*
 531                  * we need to call clear_page_dirty_for_io on each
 532                  * page in the range.  Otherwise applications with the file
 533                  * mmap'd can wander in and change the page contents while
 534                  * we are compressing them.
 535                  *
 536                  * If the compression fails for any reason, we set the pages
 537                  * dirty again later on.
 538                  */
 539                 extent_range_clear_dirty_for_io(inode, start, end);
 540                 redirty = 1;
 541
 542                 /* Compression level is applied here and only here */
 543                 ret = btrfs_compress_pages(
 544                         compress_type | (fs_info->compress_level << 4),
 545                                            inode->i_mapping, start,
 546                                            pages,
 547                                            &nr_pages,
 548                                            &total_in,
 549                                            &total_compressed);
 550
 551                 if (!ret) {
 552                         unsigned long offset = total_compressed &
 553                                 (PAGE_SIZE - 1);
 554                         struct page *page = pages[nr_pages - 1];
 555                         char *kaddr;
 556
 557                         /* zero the tail end of the last page, we might be
 558                          * sending it down to disk
 559                          */
 560                         if (offset) {
 561                                 kaddr = kmap_atomic(page);
 562                                 memset(kaddr + offset, 0,
 563                                        PAGE_SIZE - offset);
 564                                 kunmap_atomic(kaddr);
 565                         }
 566                         will_compress = 1;
 567                 }
 568         }
 569 cont:
 570         if (start == 0) {
 571                 /* lets try to make an inline extent */
 572                 if (ret || total_in < actual_end) {
 573                         /* we didn't compress the entire range, try
 574                          * to make an uncompressed inline extent.
 575                          */
 576                         ret = cow_file_range_inline(root, inode, start, end,
 577                                             0, BTRFS_COMPRESS_NONE, NULL);
 578                 } else {
 579                         /* try making a compressed inline extent */
 580                         ret = cow_file_range_inline(root, inode, start, end,
 581                                                     total_compressed,
 582                                                     compress_type, pages);
 583                 }
 584                 if (ret <= 0) {
 585                         unsigned long clear_flags = EXTENT_DELALLOC |
 586                                 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
 587                                 EXTENT_DO_ACCOUNTING;
 588                         unsigned long page_error_op;
 589
 590                         page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
 591
 592                         /*
 593                          * inline extent creation worked or returned error,
 594                          * we don't need to create any more async work items.
 595                          * Unlock and free up our temp pages.
 596                          *
 597                          * We use DO_ACCOUNTING here because we need the
 598                          * delalloc_release_metadata to be done _after_ we drop
 599                          * our outstanding extent for clearing delalloc for this
 600                          * range.
 601                          */
 602                         extent_clear_unlock_delalloc(inode, start, end, end,
 603                                                      NULL, clear_flags,
 604                                                      PAGE_UNLOCK |
 605                                                      PAGE_CLEAR_DIRTY |
 606                                                      PAGE_SET_WRITEBACK |
 607                                                      page_error_op |
 608                                                      PAGE_END_WRITEBACK);
 609                         goto free_pages_out;
 610                 }
 611         }
 612
 613         if (will_compress) {
 614                 /*
 615                  * we aren't doing an inline extent round the compressed size
 616                  * up to a block size boundary so the allocator does sane
 617                  * things
 618                  */
 619                 total_compressed = ALIGN(total_compressed, blocksize);
 620
 621                 /*
 622                  * one last check to make sure the compression is really a
 623                  * win, compare the page count read with the blocks on disk,
 624                  * compression must free at least one sector size
 625                  */
 626                 total_in = ALIGN(total_in, PAGE_SIZE);
 627                 if (total_compressed + blocksize <= total_in) {
 628                         *num_added += 1;
 629
 630                         /*
 631                          * The async work queues will take care of doing actual
 632                          * allocation on disk for these compressed pages, and
 633                          * will submit them to the elevator.
 634                          */
 635                         add_async_extent(async_cow, start, total_in,
 636                                         total_compressed, pages, nr_pages,
 637                                         compress_type);
 638
 639                         if (start + total_in < end) {
 640                                 start += total_in;
 641                                 pages = NULL;
 642                                 cond_resched();
 643                                 goto again;
 644                         }
 645                         return;
 646                 }
 647         }
 648         if (pages) {
 649                 /*
 650                  * the compression code ran but failed to make things smaller,
 651                  * free any pages it allocated and our page pointer array
 652                  */
 653                 for (i = 0; i < nr_pages; i++) {
 654                         WARN_ON(pages[i]->mapping);
 655                         put_page(pages[i]);
 656                 }
 657                 kfree(pages);
 658                 pages = NULL;
 659                 total_compressed = 0;
 660                 nr_pages = 0;
 661
 662                 /* flag the file so we don't compress in the future */
 663                 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
 664                     !(BTRFS_I(inode)->prop_compress)) {
 665                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 666                 }
 667         }
 668 cleanup_and_bail_uncompressed:
 669         /*
 670          * No compression, but we still need to write the pages in the file
 671          * we've been given so far.  redirty the locked page if it corresponds
 672          * to our extent and set things up for the async work queue to run
 673          * cow_file_range to do the normal delalloc dance.
 674          */
 675         if (page_offset(locked_page) >= start &&
 676             page_offset(locked_page) <= end)
 677                 __set_page_dirty_nobuffers(locked_page);
 678                 /* unlocked later on in the async handlers */
 679
 680         if (redirty)
 681                 extent_range_redirty_for_io(inode, start, end);
 682         add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0,
 683                          BTRFS_COMPRESS_NONE);
 684         *num_added += 1;
 685
 686         return;
 687
 688 free_pages_out:
 689         for (i = 0; i < nr_pages; i++) {
 690                 WARN_ON(pages[i]->mapping);
 691                 put_page(pages[i]);
 692         }
 693         kfree(pages);
 694 }
 695
 696 static void free_async_extent_pages(struct async_extent *async_extent)
 697 {
 698         int i;
 699
 700         if (!async_extent->pages)
 701                 return;
 702
 703         for (i = 0; i < async_extent->nr_pages; i++) {
 704                 WARN_ON(async_extent->pages[i]->mapping);
 705                 put_page(async_extent->pages[i]);
 706         }
 707         kfree(async_extent->pages);
 708         async_extent->nr_pages = 0;
 709         async_extent->pages = NULL;
 710 }
 711
 712 /*
 713  * phase two of compressed writeback.  This is the ordered portion
 714  * of the code, which only gets called in the order the work was
 715  * queued.  We walk all the async extents created by compress_file_range
 716  * and send them down to the disk.
 717  */
 718 static noinline void submit_compressed_extents(struct inode *inode,
 719                                               struct async_cow *async_cow)
 720 {
 721         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 722         struct async_extent *async_extent;
 723         u64 alloc_hint = 0;
 724         struct btrfs_key ins;
 725         struct extent_map *em;
 726         struct btrfs_root *root = BTRFS_I(inode)->root;
 727         struct extent_io_tree *io_tree;
 728         int ret = 0;
 729
 730 again:
 731         while (!list_empty(&async_cow->extents)) {
 732                 async_extent = list_entry(async_cow->extents.next,
 733                                           struct async_extent, list);
 734                 list_del(&async_extent->list);
 735
 736                 io_tree = &BTRFS_I(inode)->io_tree;
 737
 738 retry:
 739                 /* did the compression code fall back to uncompressed IO? */
 740                 if (!async_extent->pages) {
 741                         int page_started = 0;
 742                         unsigned long nr_written = 0;
 743
 744                         lock_extent(io_tree, async_extent->start,
 745                                          async_extent->start +
 746                                          async_extent->ram_size - 1);
 747
 748                         /* allocate blocks */
 749                         ret = cow_file_range(inode, async_cow->locked_page,
 750                                              async_extent->start,
 751                                              async_extent->start +
 752                                              async_extent->ram_size - 1,
 753                                              async_extent->start +
 754                                              async_extent->ram_size - 1,
 755                                              &page_started, &nr_written, 0,
 756                                              NULL);
 757
 758                         /* JDM XXX */
 759
 760                         /*
 761                          * if page_started, cow_file_range inserted an
 762                          * inline extent and took care of all the unlocking
 763                          * and IO for us.  Otherwise, we need to submit
 764                          * all those pages down to the drive.
 765                          */
 766                         if (!page_started && !ret)
 767                                 extent_write_locked_range(io_tree,
 768                                                   inode, async_extent->start,
 769                                                   async_extent->start +
 770                                                   async_extent->ram_size - 1,
 771                                                   btrfs_get_extent,
 772                                                   WB_SYNC_ALL);
 773                         else if (ret)
 774                                 unlock_page(async_cow->locked_page);
 775                         kfree(async_extent);
 776                         cond_resched();
 777                         continue;
 778                 }
 779
 780                 lock_extent(io_tree, async_extent->start,
 781                             async_extent->start + async_extent->ram_size - 1);
 782
 783                 ret = btrfs_reserve_extent(root, async_extent->ram_size,
 784                                            async_extent->compressed_size,
 785                                            async_extent->compressed_size,
 786                                            0, alloc_hint, &ins, 1, 1);
 787                 if (ret) {
 788                         free_async_extent_pages(async_extent);
 789
 790                         if (ret == -ENOSPC) {
 791                                 unlock_extent(io_tree, async_extent->start,
 792                                               async_extent->start +
 793                                               async_extent->ram_size - 1);
 794
 795                                 /*
 796                                  * we need to redirty the pages if we decide to
 797                                  * fallback to uncompressed IO, otherwise we
 798                                  * will not submit these pages down to lower
 799                                  * layers.
 800                                  */
 801                                 extent_range_redirty_for_io(inode,
 802                                                 async_extent->start,
 803                                                 async_extent->start +
 804                                                 async_extent->ram_size - 1);
 805
 806                                 goto retry;
 807                         }
 808                         goto out_free;
 809                 }
 810                 /*
 811                  * here we're doing allocation and writeback of the
 812                  * compressed pages
 813                  */
 814                 em = create_io_em(inode, async_extent->start,
 815                                   async_extent->ram_size, /* len */
 816                                   async_extent->start, /* orig_start */
 817                                   ins.objectid, /* block_start */
 818                                   ins.offset, /* block_len */
 819                                   ins.offset, /* orig_block_len */
 820                                   async_extent->ram_size, /* ram_bytes */
 821                                   async_extent->compress_type,
 822                                   BTRFS_ORDERED_COMPRESSED);
 823                 if (IS_ERR(em))
 824                         /* ret value is not necessary due to void function */
 825                         goto out_free_reserve;
 826                 free_extent_map(em);
 827
 828                 ret = btrfs_add_ordered_extent_compress(inode,
 829                                                 async_extent->start,
 830                                                 ins.objectid,
 831                                                 async_extent->ram_size,
 832                                                 ins.offset,
 833                                                 BTRFS_ORDERED_COMPRESSED,
 834                                                 async_extent->compress_type);
 835                 if (ret) {
 836                         btrfs_drop_extent_cache(BTRFS_I(inode),
 837                                                 async_extent->start,
 838                                                 async_extent->start +
 839                                                 async_extent->ram_size - 1, 0);
 840                         goto out_free_reserve;
 841                 }
 842                 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
 843
 844                 /*
 845                  * clear dirty, set writeback and unlock the pages.
 846                  */
 847                 extent_clear_unlock_delalloc(inode, async_extent->start,
 848                                 async_extent->start +
 849                                 async_extent->ram_size - 1,
 850                                 async_extent->start +
 851                                 async_extent->ram_size - 1,
 852                                 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
 853                                 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 854                                 PAGE_SET_WRITEBACK);
 855                 if (btrfs_submit_compressed_write(inode,
 856                                     async_extent->start,
 857                                     async_extent->ram_size,
 858                                     ins.objectid,
 859                                     ins.offset, async_extent->pages,
 860                                     async_extent->nr_pages)) {
 861                         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
 862                         struct page *p = async_extent->pages[0];
 863                         const u64 start = async_extent->start;
 864                         const u64 end = start + async_extent->ram_size - 1;
 865
 866                         p->mapping = inode->i_mapping;
 867                         tree->ops->writepage_end_io_hook(p, start, end,
 868                                                          NULL, 0);
 869                         p->mapping = NULL;
 870                         extent_clear_unlock_delalloc(inode, start, end, end,
 871                                                      NULL, 0,
 872                                                      PAGE_END_WRITEBACK |
 873                                                      PAGE_SET_ERROR);
 874                         free_async_extent_pages(async_extent);
 875                 }
 876                 alloc_hint = ins.objectid + ins.offset;
 877                 kfree(async_extent);
 878                 cond_resched();
 879         }
 880         return;
 881 out_free_reserve:
 882         btrfs_dec_block_group_reservations(fs_info, ins.objectid);
 883         btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
 884 out_free:
 885         extent_clear_unlock_delalloc(inode, async_extent->start,
 886                                      async_extent->start +
 887                                      async_extent->ram_size - 1,
 888                                      async_extent->start +
 889                                      async_extent->ram_size - 1,
 890                                      NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
 891                                      EXTENT_DELALLOC_NEW |
 892                                      EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
 893                                      PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 894                                      PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
 895                                      PAGE_SET_ERROR);
 896         free_async_extent_pages(async_extent);
 897         kfree(async_extent);
 898         goto again;
 899 }
 900
 901 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 902                                       u64 num_bytes)
 903 {
 904         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 905         struct extent_map *em;
 906         u64 alloc_hint = 0;
 907
 908         read_lock(&em_tree->lock);
 909         em = search_extent_mapping(em_tree, start, num_bytes);
 910         if (em) {
 911                 /*
 912                  * if block start isn't an actual block number then find the
 913                  * first block in this inode and use that as a hint.  If that
 914                  * block is also bogus then just don't worry about it.
 915                  */
 916                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 917                         free_extent_map(em);
 918                         em = search_extent_mapping(em_tree, 0, 0);
 919                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 920                                 alloc_hint = em->block_start;
 921                         if (em)
 922                                 free_extent_map(em);
 923                 } else {
 924                         alloc_hint = em->block_start;
 925                         free_extent_map(em);
 926                 }
 927         }
 928         read_unlock(&em_tree->lock);
 929
 930         return alloc_hint;
 931 }
 932
 933 /*
 934  * when extent_io.c finds a delayed allocation range in the file,
 935  * the call backs end up in this code.  The basic idea is to
 936  * allocate extents on disk for the range, and create ordered data structs
 937  * in ram to track those extents.
 938  *
 939  * locked_page is the page that writepage had locked already.  We use
 940  * it to make sure we don't do extra locks or unlocks.
 941  *
 942  * *page_started is set to one if we unlock locked_page and do everything
 943  * required to start IO on it.  It may be clean and already done with
 944  * IO when we return.
 945  */
 946 static noinline int cow_file_range(struct inode *inode,
 947                                    struct page *locked_page,
 948                                    u64 start, u64 end, u64 delalloc_end,
 949                                    int *page_started, unsigned long *nr_written,
 950                                    int unlock, struct btrfs_dedupe_hash *hash)
 951 {
 952         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 953         struct btrfs_root *root = BTRFS_I(inode)->root;
 954         u64 alloc_hint = 0;
 955         u64 num_bytes;
 956         unsigned long ram_size;
 957         u64 disk_num_bytes;
 958         u64 cur_alloc_size = 0;
 959         u64 blocksize = fs_info->sectorsize;
 960         struct btrfs_key ins;
 961         struct extent_map *em;
 962         unsigned clear_bits;
 963         unsigned long page_ops;
 964         bool extent_reserved = false;
 965         int ret = 0;
 966
 967         if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
 968                 WARN_ON_ONCE(1);
 969                 ret = -EINVAL;
 970                 goto out_unlock;
 971         }
 972
 973         num_bytes = ALIGN(end - start + 1, blocksize);
 974         num_bytes = max(blocksize,  num_bytes);
 975         disk_num_bytes = num_bytes;
 976
 977         inode_should_defrag(BTRFS_I(inode), start, end, num_bytes, SZ_64K);
 978
 979         if (start == 0) {
 980                 /* lets try to make an inline extent */
 981                 ret = cow_file_range_inline(root, inode, start, end, 0,
 982                                         BTRFS_COMPRESS_NONE, NULL);
 983                 if (ret == 0) {
 984                         /*
 985                          * We use DO_ACCOUNTING here because we need the
 986                          * delalloc_release_metadata to be run _after_ we drop
 987                          * our outstanding extent for clearing delalloc for this
 988                          * range.
 989                          */
 990                         extent_clear_unlock_delalloc(inode, start, end,
 991                                      delalloc_end, NULL,
 992                                      EXTENT_LOCKED | EXTENT_DELALLOC |
 993                                      EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
 994                                      EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
 995                                      PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
 996                                      PAGE_END_WRITEBACK);
 997                         *nr_written = *nr_written +
 998                              (end - start + PAGE_SIZE) / PAGE_SIZE;
 999                         *page_started = 1;
1000                         goto out;
1001                 } else if (ret < 0) {
1002                         goto out_unlock;
1003                 }
1004         }
1005
1006         BUG_ON(disk_num_bytes >
1007                btrfs_super_total_bytes(fs_info->super_copy));
1008
1009         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1010         btrfs_drop_extent_cache(BTRFS_I(inode), start,
1011                         start + num_bytes - 1, 0);
1012
1013         while (disk_num_bytes > 0) {
1014                 cur_alloc_size = disk_num_bytes;
1015                 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1016                                            fs_info->sectorsize, 0, alloc_hint,
1017                                            &ins, 1, 1);
1018                 if (ret < 0)
1019                         goto out_unlock;
1020                 cur_alloc_size = ins.offset;
1021                 extent_reserved = true;
1022
1023                 ram_size = ins.offset;
1024                 em = create_io_em(inode, start, ins.offset, /* len */
1025                                   start, /* orig_start */
1026                                   ins.objectid, /* block_start */
1027                                   ins.offset, /* block_len */
1028                                   ins.offset, /* orig_block_len */
1029                                   ram_size, /* ram_bytes */
1030                                   BTRFS_COMPRESS_NONE, /* compress_type */
1031                                   BTRFS_ORDERED_REGULAR /* type */);
1032                 if (IS_ERR(em))
1033                         goto out_reserve;
1034                 free_extent_map(em);
1035
1036                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1037                                                ram_size, cur_alloc_size, 0);
1038                 if (ret)
1039                         goto out_drop_extent_cache;
1040
1041                 if (root->root_key.objectid ==
1042                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1043                         ret = btrfs_reloc_clone_csums(inode, start,
1044                                                       cur_alloc_size);
1045                         /*
1046                          * Only drop cache here, and process as normal.
1047                          *
1048                          * We must not allow extent_clear_unlock_delalloc()
1049                          * at out_unlock label to free meta of this ordered
1050                          * extent, as its meta should be freed by
1051                          * btrfs_finish_ordered_io().
1052                          *
1053                          * So we must continue until @start is increased to
1054                          * skip current ordered extent.
1055                          */
1056                         if (ret)
1057                                 btrfs_drop_extent_cache(BTRFS_I(inode), start,
1058                                                 start + ram_size - 1, 0);
1059                 }
1060
1061                 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1062
1063                 /* we're not doing compressed IO, don't unlock the first
1064                  * page (which the caller expects to stay locked), don't
1065                  * clear any dirty bits and don't set any writeback bits
1066                  *
1067                  * Do set the Private2 bit so we know this page was properly
1068                  * setup for writepage
1069                  */
1070                 page_ops = unlock ? PAGE_UNLOCK : 0;
1071                 page_ops |= PAGE_SET_PRIVATE2;
1072
1073                 extent_clear_unlock_delalloc(inode, start,
1074                                              start + ram_size - 1,
1075                                              delalloc_end, locked_page,
1076                                              EXTENT_LOCKED | EXTENT_DELALLOC,
1077                                              page_ops);
1078                 if (disk_num_bytes < cur_alloc_size)
1079                         disk_num_bytes = 0;
1080                 else
1081                         disk_num_bytes -= cur_alloc_size;
1082                 num_bytes -= cur_alloc_size;
1083                 alloc_hint = ins.objectid + ins.offset;
1084                 start += cur_alloc_size;
1085                 extent_reserved = false;
1086
1087                 /*
1088                  * btrfs_reloc_clone_csums() error, since start is increased
1089                  * extent_clear_unlock_delalloc() at out_unlock label won't
1090                  * free metadata of current ordered extent, we're OK to exit.
1091                  */
1092                 if (ret)
1093                         goto out_unlock;
1094         }
1095 out:
1096         return ret;
1097
1098 out_drop_extent_cache:
1099         btrfs_drop_extent_cache(BTRFS_I(inode), start, start + ram_size - 1, 0);
1100 out_reserve:
1101         btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1102         btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1103 out_unlock:
1104         clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1105                 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1106         page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1107                 PAGE_END_WRITEBACK;
1108         /*
1109          * If we reserved an extent for our delalloc range (or a subrange) and
1110          * failed to create the respective ordered extent, then it means that
1111          * when we reserved the extent we decremented the extent's size from
1112          * the data space_info's bytes_may_use counter and incremented the
1113          * space_info's bytes_reserved counter by the same amount. We must make
1114          * sure extent_clear_unlock_delalloc() does not try to decrement again
1115          * the data space_info's bytes_may_use counter, therefore we do not pass
1116          * it the flag EXTENT_CLEAR_DATA_RESV.
1117          */
1118         if (extent_reserved) {
1119                 extent_clear_unlock_delalloc(inode, start,
1120                                              start + cur_alloc_size,
1121                                              start + cur_alloc_size,
1122                                              locked_page,
1123                                              clear_bits,
1124                                              page_ops);
1125                 start += cur_alloc_size;
1126                 if (start >= end)
1127                         goto out;
1128         }
1129         extent_clear_unlock_delalloc(inode, start, end, delalloc_end,
1130                                      locked_page,
1131                                      clear_bits | EXTENT_CLEAR_DATA_RESV,
1132                                      page_ops);
1133         goto out;
1134 }
1135
1136 /*
1137  * work queue call back to started compression on a file and pages
1138  */
1139 static noinline void async_cow_start(struct btrfs_work *work)
1140 {
1141         struct async_cow *async_cow;
1142         int num_added = 0;
1143         async_cow = container_of(work, struct async_cow, work);
1144
1145         compress_file_range(async_cow->inode, async_cow->locked_page,
1146                             async_cow->start, async_cow->end, async_cow,
1147                             &num_added);
1148         if (num_added == 0) {
1149                 btrfs_add_delayed_iput(async_cow->inode);
1150                 async_cow->inode = NULL;
1151         }
1152 }
1153
1154 /*
1155  * work queue call back to submit previously compressed pages
1156  */
1157 static noinline void async_cow_submit(struct btrfs_work *work)
1158 {
1159         struct btrfs_fs_info *fs_info;
1160         struct async_cow *async_cow;
1161         struct btrfs_root *root;
1162         unsigned long nr_pages;
1163
1164         async_cow = container_of(work, struct async_cow, work);
1165
1166         root = async_cow->root;
1167         fs_info = root->fs_info;
1168         nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1169                 PAGE_SHIFT;
1170
1171         /*
1172          * atomic_sub_return implies a barrier for waitqueue_active
1173          */
1174         if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1175             5 * SZ_1M &&
1176             waitqueue_active(&fs_info->async_submit_wait))
1177                 wake_up(&fs_info->async_submit_wait);
1178
1179         if (async_cow->inode)
1180                 submit_compressed_extents(async_cow->inode, async_cow);
1181 }
1182
1183 static noinline void async_cow_free(struct btrfs_work *work)
1184 {
1185         struct async_cow *async_cow;
1186         async_cow = container_of(work, struct async_cow, work);
1187         if (async_cow->inode)
1188                 btrfs_add_delayed_iput(async_cow->inode);
1189         kfree(async_cow);
1190 }
1191
1192 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1193                                 u64 start, u64 end, int *page_started,
1194                                 unsigned long *nr_written)
1195 {
1196         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1197         struct async_cow *async_cow;
1198         struct btrfs_root *root = BTRFS_I(inode)->root;
1199         unsigned long nr_pages;
1200         u64 cur_end;
1201
1202         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1203                          1, 0, NULL, GFP_NOFS);
1204         while (start < end) {
1205                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1206                 BUG_ON(!async_cow); /* -ENOMEM */
1207                 async_cow->inode = igrab(inode);
1208                 async_cow->root = root;
1209                 async_cow->locked_page = locked_page;
1210                 async_cow->start = start;
1211
1212                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1213                     !btrfs_test_opt(fs_info, FORCE_COMPRESS))
1214                         cur_end = end;
1215                 else
1216                         cur_end = min(end, start + SZ_512K - 1);
1217
1218                 async_cow->end = cur_end;
1219                 INIT_LIST_HEAD(&async_cow->extents);
1220
1221                 btrfs_init_work(&async_cow->work,
1222                                 btrfs_delalloc_helper,
1223                                 async_cow_start, async_cow_submit,
1224                                 async_cow_free);
1225
1226                 nr_pages = (cur_end - start + PAGE_SIZE) >>
1227                         PAGE_SHIFT;
1228                 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1229
1230                 btrfs_queue_work(fs_info->delalloc_workers, &async_cow->work);
1231
1232                 *nr_written += nr_pages;
1233                 start = cur_end + 1;
1234         }
1235         *page_started = 1;
1236         return 0;
1237 }
1238
1239 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1240                                         u64 bytenr, u64 num_bytes)
1241 {
1242         int ret;
1243         struct btrfs_ordered_sum *sums;
1244         LIST_HEAD(list);
1245
1246         ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1247                                        bytenr + num_bytes - 1, &list, 0);
1248         if (ret == 0 && list_empty(&list))
1249                 return 0;
1250
1251         while (!list_empty(&list)) {
1252                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1253                 list_del(&sums->list);
1254                 kfree(sums);
1255         }
1256         return 1;
1257 }
1258
1259 /*
1260  * when nowcow writeback call back.  This checks for snapshots or COW copies
1261  * of the extents that exist in the file, and COWs the file as required.
1262  *
1263  * If no cow copies or snapshots exist, we write directly to the existing
1264  * blocks on disk
1265  */
1266 static noinline int run_delalloc_nocow(struct inode *inode,
1267                                        struct page *locked_page,
1268                               u64 start, u64 end, int *page_started, int force,
1269                               unsigned long *nr_written)
1270 {
1271         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1272         struct btrfs_root *root = BTRFS_I(inode)->root;
1273         struct extent_buffer *leaf;
1274         struct btrfs_path *path;
1275         struct btrfs_file_extent_item *fi;
1276         struct btrfs_key found_key;
1277         struct extent_map *em;
1278         u64 cow_start;
1279         u64 cur_offset;
1280         u64 extent_end;
1281         u64 extent_offset;
1282         u64 disk_bytenr;
1283         u64 num_bytes;
1284         u64 disk_num_bytes;
1285         u64 ram_bytes;
1286         int extent_type;
1287         int ret, err;
1288         int type;
1289         int nocow;
1290         int check_prev = 1;
1291         bool nolock;
1292         u64 ino = btrfs_ino(BTRFS_I(inode));
1293
1294         path = btrfs_alloc_path();
1295         if (!path) {
1296                 extent_clear_unlock_delalloc(inode, start, end, end,
1297                                              locked_page,
1298                                              EXTENT_LOCKED | EXTENT_DELALLOC |
1299                                              EXTENT_DO_ACCOUNTING |
1300                                              EXTENT_DEFRAG, PAGE_UNLOCK |
1301                                              PAGE_CLEAR_DIRTY |
1302                                              PAGE_SET_WRITEBACK |
1303                                              PAGE_END_WRITEBACK);
1304                 return -ENOMEM;
1305         }
1306
1307         nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
1308
1309         cow_start = (u64)-1;
1310         cur_offset = start;
1311         while (1) {
1312                 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1313                                                cur_offset, 0);
1314                 if (ret < 0)
1315                         goto error;
1316                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1317                         leaf = path->nodes[0];
1318                         btrfs_item_key_to_cpu(leaf, &found_key,
1319                                               path->slots[0] - 1);
1320                         if (found_key.objectid == ino &&
1321                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1322                                 path->slots[0]--;
1323                 }
1324                 check_prev = 0;
1325 next_slot:
1326                 leaf = path->nodes[0];
1327                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1328                         ret = btrfs_next_leaf(root, path);
1329                         if (ret < 0)
1330                                 goto error;
1331                         if (ret > 0)
1332                                 break;
1333                         leaf = path->nodes[0];
1334                 }
1335
1336                 nocow = 0;
1337                 disk_bytenr = 0;
1338                 num_bytes = 0;
1339                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1340
1341                 if (found_key.objectid > ino)
1342                         break;
1343                 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1344                     found_key.type < BTRFS_EXTENT_DATA_KEY) {
1345                         path->slots[0]++;
1346                         goto next_slot;
1347                 }
1348                 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1349                     found_key.offset > end)
1350                         break;
1351
1352                 if (found_key.offset > cur_offset) {
1353                         extent_end = found_key.offset;
1354                         extent_type = 0;
1355                         goto out_check;
1356                 }
1357
1358                 fi = btrfs_item_ptr(leaf, path->slots[0],
1359                                     struct btrfs_file_extent_item);
1360                 extent_type = btrfs_file_extent_type(leaf, fi);
1361
1362                 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1363                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1364                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1365                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1366                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1367                         extent_end = found_key.offset +
1368                                 btrfs_file_extent_num_bytes(leaf, fi);
1369                         disk_num_bytes =
1370                                 btrfs_file_extent_disk_num_bytes(leaf, fi);
1371                         if (extent_end <= start) {
1372                                 path->slots[0]++;
1373                                 goto next_slot;
1374                         }
1375                         if (disk_bytenr == 0)
1376                                 goto out_check;
1377                         if (btrfs_file_extent_compression(leaf, fi) ||
1378                             btrfs_file_extent_encryption(leaf, fi) ||
1379                             btrfs_file_extent_other_encoding(leaf, fi))
1380                                 goto out_check;
1381                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1382                                 goto out_check;
1383                         if (btrfs_extent_readonly(fs_info, disk_bytenr))
1384                                 goto out_check;
1385                         if (btrfs_cross_ref_exist(root, ino,
1386                                                   found_key.offset -
1387                                                   extent_offset, disk_bytenr))
1388                                 goto out_check;
1389                         disk_bytenr += extent_offset;
1390                         disk_bytenr += cur_offset - found_key.offset;
1391                         num_bytes = min(end + 1, extent_end) - cur_offset;
1392                         /*
1393                          * if there are pending snapshots for this root,
1394                          * we fall into common COW way.
1395                          */
1396                         if (!nolock) {
1397                                 err = btrfs_start_write_no_snapshotting(root);
1398                                 if (!err)
1399                                         goto out_check;
1400                         }
1401                         /*
1402                          * force cow if csum exists in the range.
1403                          * this ensure that csum for a given extent are
1404                          * either valid or do not exist.
1405                          */
1406                         if (csum_exist_in_range(fs_info, disk_bytenr,
1407                                                 num_bytes)) {
1408                                 if (!nolock)
1409                                         btrfs_end_write_no_snapshotting(root);
1410                                 goto out_check;
1411                         }
1412                         if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr)) {
1413                                 if (!nolock)
1414                                         btrfs_end_write_no_snapshotting(root);
1415                                 goto out_check;
1416                         }
1417                         nocow = 1;
1418                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1419                         extent_end = found_key.offset +
1420                                 btrfs_file_extent_inline_len(leaf,
1421                                                      path->slots[0], fi);
1422                         extent_end = ALIGN(extent_end,
1423                                            fs_info->sectorsize);
1424                 } else {
1425                         BUG_ON(1);
1426                 }
1427 out_check:
1428                 if (extent_end <= start) {
1429                         path->slots[0]++;
1430                         if (!nolock && nocow)
1431                                 btrfs_end_write_no_snapshotting(root);
1432                         if (nocow)
1433                                 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1434                         goto next_slot;
1435                 }
1436                 if (!nocow) {
1437                         if (cow_start == (u64)-1)
1438                                 cow_start = cur_offset;
1439                         cur_offset = extent_end;
1440                         if (cur_offset > end)
1441                                 break;
1442                         path->slots[0]++;
1443                         goto next_slot;
1444                 }
1445
1446                 btrfs_release_path(path);
1447                 if (cow_start != (u64)-1) {
1448                         ret = cow_file_range(inode, locked_page,
1449                                              cow_start, found_key.offset - 1,
1450                                              end, page_started, nr_written, 1,
1451                                              NULL);
1452                         if (ret) {
1453                                 if (!nolock && nocow)
1454                                         btrfs_end_write_no_snapshotting(root);
1455                                 if (nocow)
1456                                         btrfs_dec_nocow_writers(fs_info,
1457                                                                 disk_bytenr);
1458                                 goto error;
1459                         }
1460                         cow_start = (u64)-1;
1461                 }
1462
1463                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1464                         u64 orig_start = found_key.offset - extent_offset;
1465
1466                         em = create_io_em(inode, cur_offset, num_bytes,
1467                                           orig_start,
1468                                           disk_bytenr, /* block_start */
1469                                           num_bytes, /* block_len */
1470                                           disk_num_bytes, /* orig_block_len */
1471                                           ram_bytes, BTRFS_COMPRESS_NONE,
1472                                           BTRFS_ORDERED_PREALLOC);
1473                         if (IS_ERR(em)) {
1474                                 if (!nolock && nocow)
1475                                         btrfs_end_write_no_snapshotting(root);
1476                                 if (nocow)
1477                                         btrfs_dec_nocow_writers(fs_info,
1478                                                                 disk_bytenr);
1479                                 ret = PTR_ERR(em);
1480                                 goto error;
1481                         }
1482                         free_extent_map(em);
1483                 }
1484
1485                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1486                         type = BTRFS_ORDERED_PREALLOC;
1487                 } else {
1488                         type = BTRFS_ORDERED_NOCOW;
1489                 }
1490
1491                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1492                                                num_bytes, num_bytes, type);
1493                 if (nocow)
1494                         btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1495                 BUG_ON(ret); /* -ENOMEM */
1496
1497                 if (root->root_key.objectid ==
1498                     BTRFS_DATA_RELOC_TREE_OBJECTID)
1499                         /*
1500                          * Error handled later, as we must prevent
1501                          * extent_clear_unlock_delalloc() in error handler
1502                          * from freeing metadata of created ordered extent.
1503                          */
1504                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1505                                                       num_bytes);
1506
1507                 extent_clear_unlock_delalloc(inode, cur_offset,
1508                                              cur_offset + num_bytes - 1, end,
1509                                              locked_page, EXTENT_LOCKED |
1510                                              EXTENT_DELALLOC |
1511                                              EXTENT_CLEAR_DATA_RESV,
1512                                              PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1513
1514                 if (!nolock && nocow)
1515                         btrfs_end_write_no_snapshotting(root);
1516                 cur_offset = extent_end;
1517
1518                 /*
1519                  * btrfs_reloc_clone_csums() error, now we're OK to call error
1520                  * handler, as metadata for created ordered extent will only
1521                  * be freed by btrfs_finish_ordered_io().
1522                  */
1523                 if (ret)
1524                         goto error;
1525                 if (cur_offset > end)
1526                         break;
1527         }
1528         btrfs_release_path(path);
1529
1530         if (cur_offset <= end && cow_start == (u64)-1) {
1531                 cow_start = cur_offset;
1532                 cur_offset = end;
1533         }
1534
1535         if (cow_start != (u64)-1) {
1536                 ret = cow_file_range(inode, locked_page, cow_start, end, end,
1537                                      page_started, nr_written, 1, NULL);
1538                 if (ret)
1539                         goto error;
1540         }
1541
1542 error:
1543         if (ret && cur_offset < end)
1544                 extent_clear_unlock_delalloc(inode, cur_offset, end, end,
1545                                              locked_page, EXTENT_LOCKED |
1546                                              EXTENT_DELALLOC | EXTENT_DEFRAG |
1547                                              EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1548                                              PAGE_CLEAR_DIRTY |
1549                                              PAGE_SET_WRITEBACK |
1550                                              PAGE_END_WRITEBACK);
1551         btrfs_free_path(path);
1552         return ret;
1553 }
1554
1555 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1556 {
1557
1558         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1559             !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1560                 return 0;
1561
1562         /*
1563          * @defrag_bytes is a hint value, no spinlock held here,
1564          * if is not zero, it means the file is defragging.
1565          * Force cow if given extent needs to be defragged.
1566          */
1567         if (BTRFS_I(inode)->defrag_bytes &&
1568             test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1569                            EXTENT_DEFRAG, 0, NULL))
1570                 return 1;
1571
1572         return 0;
1573 }
1574
1575 /*
1576  * extent_io.c call back to do delayed allocation processing
1577  */
1578 static int run_delalloc_range(void *private_data, struct page *locked_page,
1579                               u64 start, u64 end, int *page_started,
1580                               unsigned long *nr_written)
1581 {
1582         struct inode *inode = private_data;
1583         int ret;
1584         int force_cow = need_force_cow(inode, start, end);
1585
1586         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1587                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1588                                          page_started, 1, nr_written);
1589         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1590                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1591                                          page_started, 0, nr_written);
1592         } else if (!inode_need_compress(inode, start, end)) {
1593                 ret = cow_file_range(inode, locked_page, start, end, end,
1594                                       page_started, nr_written, 1, NULL);
1595         } else {
1596                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1597                         &BTRFS_I(inode)->runtime_flags);
1598                 ret = cow_file_range_async(inode, locked_page, start, end,
1599                                            page_started, nr_written);
1600         }
1601         if (ret)
1602                 btrfs_cleanup_ordered_extents(inode, start, end - start + 1);
1603         return ret;
1604 }
1605
1606 static void btrfs_split_extent_hook(void *private_data,
1607                                     struct extent_state *orig, u64 split)
1608 {
1609         struct inode *inode = private_data;
1610         u64 size;
1611
1612         /* not delalloc, ignore it */
1613         if (!(orig->state & EXTENT_DELALLOC))
1614                 return;
1615
1616         size = orig->end - orig->start + 1;
1617         if (size > BTRFS_MAX_EXTENT_SIZE) {
1618                 u32 num_extents;
1619                 u64 new_size;
1620
1621                 /*
1622                  * See the explanation in btrfs_merge_extent_hook, the same
1623                  * applies here, just in reverse.
1624                  */
1625                 new_size = orig->end - split + 1;
1626                 num_extents = count_max_extents(new_size);
1627                 new_size = split - orig->start;
1628                 num_extents += count_max_extents(new_size);
1629                 if (count_max_extents(size) >= num_extents)
1630                         return;
1631         }
1632
1633         spin_lock(&BTRFS_I(inode)->lock);
1634         btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1635         spin_unlock(&BTRFS_I(inode)->lock);
1636 }
1637
1638 /*
1639  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1640  * extents so we can keep track of new extents that are just merged onto old
1641  * extents, such as when we are doing sequential writes, so we can properly
1642  * account for the metadata space we'll need.
1643  */
1644 static void btrfs_merge_extent_hook(void *private_data,
1645                                     struct extent_state *new,
1646                                     struct extent_state *other)
1647 {
1648         struct inode *inode = private_data;
1649         u64 new_size, old_size;
1650         u32 num_extents;
1651
1652         /* not delalloc, ignore it */
1653         if (!(other->state & EXTENT_DELALLOC))
1654                 return;
1655
1656         if (new->start > other->start)
1657                 new_size = new->end - other->start + 1;
1658         else
1659                 new_size = other->end - new->start + 1;
1660
1661         /* we're not bigger than the max, unreserve the space and go */
1662         if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1663                 spin_lock(&BTRFS_I(inode)->lock);
1664                 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1665                 spin_unlock(&BTRFS_I(inode)->lock);
1666                 return;
1667         }
1668
1669         /*
1670          * We have to add up either side to figure out how many extents were
1671          * accounted for before we merged into one big extent.  If the number of
1672          * extents we accounted for is <= the amount we need for the new range
1673          * then we can return, otherwise drop.  Think of it like this
1674          *
1675          * [ 4k][MAX_SIZE]
1676          *
1677          * So we've grown the extent by a MAX_SIZE extent, this would mean we
1678          * need 2 outstanding extents, on one side we have 1 and the other side
1679          * we have 1 so they are == and we can return.  But in this case
1680          *
1681          * [MAX_SIZE+4k][MAX_SIZE+4k]
1682          *
1683          * Each range on their own accounts for 2 extents, but merged together
1684          * they are only 3 extents worth of accounting, so we need to drop in
1685          * this case.
1686          */
1687         old_size = other->end - other->start + 1;
1688         num_extents = count_max_extents(old_size);
1689         old_size = new->end - new->start + 1;
1690         num_extents += count_max_extents(old_size);
1691         if (count_max_extents(new_size) >= num_extents)
1692                 return;
1693
1694         spin_lock(&BTRFS_I(inode)->lock);
1695         btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1696         spin_unlock(&BTRFS_I(inode)->lock);
1697 }
1698
1699 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1700                                       struct inode *inode)
1701 {
1702         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1703
1704         spin_lock(&root->delalloc_lock);
1705         if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1706                 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1707                               &root->delalloc_inodes);
1708                 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1709                         &BTRFS_I(inode)->runtime_flags);
1710                 root->nr_delalloc_inodes++;
1711                 if (root->nr_delalloc_inodes == 1) {
1712                         spin_lock(&fs_info->delalloc_root_lock);
1713                         BUG_ON(!list_empty(&root->delalloc_root));
1714                         list_add_tail(&root->delalloc_root,
1715                                       &fs_info->delalloc_roots);
1716                         spin_unlock(&fs_info->delalloc_root_lock);
1717                 }
1718         }
1719         spin_unlock(&root->delalloc_lock);
1720 }
1721
1722 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1723                                      struct btrfs_inode *inode)
1724 {
1725         struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1726
1727         spin_lock(&root->delalloc_lock);
1728         if (!list_empty(&inode->delalloc_inodes)) {
1729                 list_del_init(&inode->delalloc_inodes);
1730                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1731                           &inode->runtime_flags);
1732                 root->nr_delalloc_inodes--;
1733                 if (!root->nr_delalloc_inodes) {
1734                         spin_lock(&fs_info->delalloc_root_lock);
1735                         BUG_ON(list_empty(&root->delalloc_root));
1736                         list_del_init(&root->delalloc_root);
1737                         spin_unlock(&fs_info->delalloc_root_lock);
1738                 }
1739         }
1740         spin_unlock(&root->delalloc_lock);
1741 }
1742
1743 /*
1744  * extent_io.c set_bit_hook, used to track delayed allocation
1745  * bytes in this file, and to maintain the list of inodes that
1746  * have pending delalloc work to be done.
1747  */
1748 static void btrfs_set_bit_hook(void *private_data,
1749                                struct extent_state *state, unsigned *bits)
1750 {
1751         struct inode *inode = private_data;
1752
1753         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1754
1755         if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1756                 WARN_ON(1);
1757         /*
1758          * set_bit and clear bit hooks normally require _irqsave/restore
1759          * but in this case, we are only testing for the DELALLOC
1760          * bit, which is only set or cleared with irqs on
1761          */
1762         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1763                 struct btrfs_root *root = BTRFS_I(inode)->root;
1764                 u64 len = state->end + 1 - state->start;
1765                 u32 num_extents = count_max_extents(len);
1766                 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
1767
1768                 spin_lock(&BTRFS_I(inode)->lock);
1769                 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
1770                 spin_unlock(&BTRFS_I(inode)->lock);
1771
1772                 /* For sanity tests */
1773                 if (btrfs_is_testing(fs_info))
1774                         return;
1775
1776                 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
1777                                          fs_info->delalloc_batch);
1778                 spin_lock(&BTRFS_I(inode)->lock);
1779                 BTRFS_I(inode)->delalloc_bytes += len;
1780                 if (*bits & EXTENT_DEFRAG)
1781                         BTRFS_I(inode)->defrag_bytes += len;
1782                 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1783                                          &BTRFS_I(inode)->runtime_flags))
1784                         btrfs_add_delalloc_inodes(root, inode);
1785                 spin_unlock(&BTRFS_I(inode)->lock);
1786         }
1787
1788         if (!(state->state & EXTENT_DELALLOC_NEW) &&
1789             (*bits & EXTENT_DELALLOC_NEW)) {
1790                 spin_lock(&BTRFS_I(inode)->lock);
1791                 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
1792                         state->start;
1793                 spin_unlock(&BTRFS_I(inode)->lock);
1794         }
1795 }
1796
1797 /*
1798  * extent_io.c clear_bit_hook, see set_bit_hook for why
1799  */
1800 static void btrfs_clear_bit_hook(void *private_data,
1801                                  struct extent_state *state,
1802                                  unsigned *bits)
1803 {
1804         struct btrfs_inode *inode = BTRFS_I((struct inode *)private_data);
1805         struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1806         u64 len = state->end + 1 - state->start;
1807         u32 num_extents = count_max_extents(len);
1808
1809         if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
1810                 spin_lock(&inode->lock);
1811                 inode->defrag_bytes -= len;
1812                 spin_unlock(&inode->lock);
1813         }
1814
1815         /*
1816          * set_bit and clear bit hooks normally require _irqsave/restore
1817          * but in this case, we are only testing for the DELALLOC
1818          * bit, which is only set or cleared with irqs on
1819          */
1820         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1821                 struct btrfs_root *root = inode->root;
1822                 bool do_list = !btrfs_is_free_space_inode(inode);
1823
1824                 spin_lock(&inode->lock);
1825                 btrfs_mod_outstanding_extents(inode, -num_extents);
1826                 spin_unlock(&inode->lock);
1827
1828                 /*
1829                  * We don't reserve metadata space for space cache inodes so we
1830                  * don't need to call dellalloc_release_metadata if there is an
1831                  * error.
1832                  */
1833                 if (*bits & EXTENT_CLEAR_META_RESV &&
1834                     root != fs_info->tree_root)
1835                         btrfs_delalloc_release_metadata(inode, len);
1836
1837                 /* For sanity tests. */
1838                 if (btrfs_is_testing(fs_info))
1839                         return;
1840
1841                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
1842                     do_list && !(state->state & EXTENT_NORESERVE) &&
1843                     (*bits & EXTENT_CLEAR_DATA_RESV))
1844                         btrfs_free_reserved_data_space_noquota(
1845                                         &inode->vfs_inode,
1846                                         state->start, len);
1847
1848                 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
1849                                          fs_info->delalloc_batch);
1850                 spin_lock(&inode->lock);
1851                 inode->delalloc_bytes -= len;
1852                 if (do_list && inode->delalloc_bytes == 0 &&
1853                     test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1854                                         &inode->runtime_flags))
1855                         btrfs_del_delalloc_inode(root, inode);
1856                 spin_unlock(&inode->lock);
1857         }
1858
1859         if ((state->state & EXTENT_DELALLOC_NEW) &&
1860             (*bits & EXTENT_DELALLOC_NEW)) {
1861                 spin_lock(&inode->lock);
1862                 ASSERT(inode->new_delalloc_bytes >= len);
1863                 inode->new_delalloc_bytes -= len;
1864                 spin_unlock(&inode->lock);
1865         }
1866 }
1867
1868 /*
1869  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1870  * we don't create bios that span stripes or chunks
1871  *
1872  * return 1 if page cannot be merged to bio
1873  * return 0 if page can be merged to bio
1874  * return error otherwise
1875  */
1876 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1877                          size_t size, struct bio *bio,
1878                          unsigned long bio_flags)
1879 {
1880         struct inode *inode = page->mapping->host;
1881         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1882         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1883         u64 length = 0;
1884         u64 map_length;
1885         int ret;
1886
1887         if (bio_flags & EXTENT_BIO_COMPRESSED)
1888                 return 0;
1889
1890         length = bio->bi_iter.bi_size;
1891         map_length = length;
1892         ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
1893                               NULL, 0);
1894         if (ret < 0)
1895                 return ret;
1896         if (map_length < length + size)
1897                 return 1;
1898         return 0;
1899 }
1900
1901 /*
1902  * in order to insert checksums into the metadata in large chunks,
1903  * we wait until bio submission time.   All the pages in the bio are
1904  * checksummed and sums are attached onto the ordered extent record.
1905  *
1906  * At IO completion time the cums attached on the ordered extent record
1907  * are inserted into the btree
1908  */
1909 static blk_status_t __btrfs_submit_bio_start(void *private_data, struct bio *bio,
1910                                     int mirror_num, unsigned long bio_flags,
1911                                     u64 bio_offset)
1912 {
1913         struct inode *inode = private_data;
1914         blk_status_t ret = 0;
1915
1916         ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1917         BUG_ON(ret); /* -ENOMEM */
1918         return 0;
1919 }
1920
1921 /*
1922  * in order to insert checksums into the metadata in large chunks,
1923  * we wait until bio submission time.   All the pages in the bio are
1924  * checksummed and sums are attached onto the ordered extent record.
1925  *
1926  * At IO completion time the cums attached on the ordered extent record
1927  * are inserted into the btree
1928  */
1929 static blk_status_t __btrfs_submit_bio_done(void *private_data, struct bio *bio,
1930                           int mirror_num, unsigned long bio_flags,
1931                           u64 bio_offset)
1932 {
1933         struct inode *inode = private_data;
1934         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1935         blk_status_t ret;
1936
1937         ret = btrfs_map_bio(fs_info, bio, mirror_num, 1);
1938         if (ret) {
1939                 bio->bi_status = ret;
1940                 bio_endio(bio);
1941         }
1942         return ret;
1943 }
1944
1945 /*
1946  * extent_io.c submission hook. This does the right thing for csum calculation
1947  * on write, or reading the csums from the tree before a read
1948  */
1949 static blk_status_t btrfs_submit_bio_hook(void *private_data, struct bio *bio,
1950                                  int mirror_num, unsigned long bio_flags,
1951                                  u64 bio_offset)
1952 {
1953         struct inode *inode = private_data;
1954         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1955         struct btrfs_root *root = BTRFS_I(inode)->root;
1956         enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1957         blk_status_t ret = 0;
1958         int skip_sum;
1959         int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1960
1961         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1962
1963         if (btrfs_is_free_space_inode(BTRFS_I(inode)))
1964                 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1965
1966         if (bio_op(bio) != REQ_OP_WRITE) {
1967                 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
1968                 if (ret)
1969                         goto out;
1970
1971                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1972                         ret = btrfs_submit_compressed_read(inode, bio,
1973                                                            mirror_num,
1974                                                            bio_flags);
1975                         goto out;
1976                 } else if (!skip_sum) {
1977                         ret = btrfs_lookup_bio_sums(inode, bio, NULL);
1978                         if (ret)
1979                                 goto out;
1980                 }
1981                 goto mapit;
1982         } else if (async && !skip_sum) {
1983                 /* csum items have already been cloned */
1984                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1985                         goto mapit;
1986                 /* we're doing a write, do the async checksumming */
1987                 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags,
1988                                           bio_offset, inode,
1989                                           __btrfs_submit_bio_start,
1990                                           __btrfs_submit_bio_done);
1991                 goto out;
1992         } else if (!skip_sum) {
1993                 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1994                 if (ret)
1995                         goto out;
1996         }
1997
1998 mapit:
1999         ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
2000
2001 out:
2002         if (ret) {
2003                 bio->bi_status = ret;
2004                 bio_endio(bio);
2005         }
2006         return ret;
2007 }
2008
2009 /*
2010  * given a list of ordered sums record them in the inode.  This happens
2011  * at IO completion time based on sums calculated at bio submission time.
2012  */
2013 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
2014                              struct inode *inode, struct list_head *list)
2015 {
2016         struct btrfs_ordered_sum *sum;
2017
2018         list_for_each_entry(sum, list, list) {
2019                 trans->adding_csums = 1;
2020                 btrfs_csum_file_blocks(trans,
2021                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
2022                 trans->adding_csums = 0;
2023         }
2024         return 0;
2025 }
2026
2027 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
2028                               struct extent_state **cached_state, int dedupe)
2029 {
2030         WARN_ON((end & (PAGE_SIZE - 1)) == 0);
2031         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2032                                    cached_state);
2033 }
2034
2035 /* see btrfs_writepage_start_hook for details on why this is required */
2036 struct btrfs_writepage_fixup {
2037         struct page *page;
2038         struct btrfs_work work;
2039 };
2040
2041 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2042 {
2043         struct btrfs_writepage_fixup *fixup;
2044         struct btrfs_ordered_extent *ordered;
2045         struct extent_state *cached_state = NULL;
2046         struct extent_changeset *data_reserved = NULL;
2047         struct page *page;
2048         struct inode *inode;
2049         u64 page_start;
2050         u64 page_end;
2051         int ret;
2052
2053         fixup = container_of(work, struct btrfs_writepage_fixup, work);
2054         page = fixup->page;
2055 again:
2056         lock_page(page);
2057         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2058                 ClearPageChecked(page);
2059                 goto out_page;
2060         }
2061
2062         inode = page->mapping->host;
2063         page_start = page_offset(page);
2064         page_end = page_offset(page) + PAGE_SIZE - 1;
2065
2066         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2067                          &cached_state);
2068
2069         /* already ordered? We're done */
2070         if (PagePrivate2(page))
2071                 goto out;
2072
2073         ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
2074                                         PAGE_SIZE);
2075         if (ordered) {
2076                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2077                                      page_end, &cached_state, GFP_NOFS);
2078                 unlock_page(page);
2079                 btrfs_start_ordered_extent(inode, ordered, 1);
2080                 btrfs_put_ordered_extent(ordered);
2081                 goto again;
2082         }
2083
2084         ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2085                                            PAGE_SIZE);
2086         if (ret) {
2087                 mapping_set_error(page->mapping, ret);
2088                 end_extent_writepage(page, ret, page_start, page_end);
2089                 ClearPageChecked(page);
2090                 goto out;
2091          }
2092
2093         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state,
2094                                   0);
2095         ClearPageChecked(page);
2096         set_page_dirty(page);
2097         btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
2098 out:
2099         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2100                              &cached_state, GFP_NOFS);
2101 out_page:
2102         unlock_page(page);
2103         put_page(page);
2104         kfree(fixup);
2105         extent_changeset_free(data_reserved);
2106 }
2107
2108 /*
2109  * There are a few paths in the higher layers of the kernel that directly
2110  * set the page dirty bit without asking the filesystem if it is a
2111  * good idea.  This causes problems because we want to make sure COW
2112  * properly happens and the data=ordered rules are followed.
2113  *
2114  * In our case any range that doesn't have the ORDERED bit set
2115  * hasn't been properly setup for IO.  We kick off an async process
2116  * to fix it up.  The async helper will wait for ordered extents, set
2117  * the delalloc bit and make it safe to write the page.
2118  */
2119 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2120 {
2121         struct inode *inode = page->mapping->host;
2122         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2123         struct btrfs_writepage_fixup *fixup;
2124
2125         /* this page is properly in the ordered list */
2126         if (TestClearPagePrivate2(page))
2127                 return 0;
2128
2129         if (PageChecked(page))
2130                 return -EAGAIN;
2131
2132         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2133         if (!fixup)
2134                 return -EAGAIN;
2135
2136         SetPageChecked(page);
2137         get_page(page);
2138         btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2139                         btrfs_writepage_fixup_worker, NULL, NULL);
2140         fixup->page = page;
2141         btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2142         return -EBUSY;
2143 }
2144
2145 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2146                                        struct inode *inode, u64 file_pos,
2147                                        u64 disk_bytenr, u64 disk_num_bytes,
2148                                        u64 num_bytes, u64 ram_bytes,
2149                                        u8 compression, u8 encryption,
2150                                        u16 other_encoding, int extent_type)
2151 {
2152         struct btrfs_root *root = BTRFS_I(inode)->root;
2153         struct btrfs_file_extent_item *fi;
2154         struct btrfs_path *path;
2155         struct extent_buffer *leaf;
2156         struct btrfs_key ins;
2157         u64 qg_released;
2158         int extent_inserted = 0;
2159         int ret;
2160
2161         path = btrfs_alloc_path();
2162         if (!path)
2163                 return -ENOMEM;
2164
2165         /*
2166          * we may be replacing one extent in the tree with another.
2167          * The new extent is pinned in the extent map, and we don't want
2168          * to drop it from the cache until it is completely in the btree.
2169          *
2170          * So, tell btrfs_drop_extents to leave this extent in the cache.
2171          * the caller is expected to unpin it and allow it to be merged
2172          * with the others.
2173          */
2174         ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2175                                    file_pos + num_bytes, NULL, 0,
2176                                    1, sizeof(*fi), &extent_inserted);
2177         if (ret)
2178                 goto out;
2179
2180         if (!extent_inserted) {
2181                 ins.objectid = btrfs_ino(BTRFS_I(inode));
2182                 ins.offset = file_pos;
2183                 ins.type = BTRFS_EXTENT_DATA_KEY;
2184
2185                 path->leave_spinning = 1;
2186                 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2187                                               sizeof(*fi));
2188                 if (ret)
2189                         goto out;
2190         }
2191         leaf = path->nodes[0];
2192         fi = btrfs_item_ptr(leaf, path->slots[0],
2193                             struct btrfs_file_extent_item);
2194         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2195         btrfs_set_file_extent_type(leaf, fi, extent_type);
2196         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2197         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2198         btrfs_set_file_extent_offset(leaf, fi, 0);
2199         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2200         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2201         btrfs_set_file_extent_compression(leaf, fi, compression);
2202         btrfs_set_file_extent_encryption(leaf, fi, encryption);
2203         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2204
2205         btrfs_mark_buffer_dirty(leaf);
2206         btrfs_release_path(path);
2207
2208         inode_add_bytes(inode, num_bytes);
2209
2210         ins.objectid = disk_bytenr;
2211         ins.offset = disk_num_bytes;
2212         ins.type = BTRFS_EXTENT_ITEM_KEY;
2213
2214         /*
2215          * Release the reserved range from inode dirty range map, as it is
2216          * already moved into delayed_ref_head
2217          */
2218         ret = btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2219         if (ret < 0)
2220                 goto out;
2221         qg_released = ret;
2222         ret = btrfs_alloc_reserved_file_extent(trans, root,
2223                                                btrfs_ino(BTRFS_I(inode)),
2224                                                file_pos, qg_released, &ins);
2225 out:
2226         btrfs_free_path(path);
2227
2228         return ret;
2229 }
2230
2231 /* snapshot-aware defrag */
2232 struct sa_defrag_extent_backref {
2233         struct rb_node node;
2234         struct old_sa_defrag_extent *old;
2235         u64 root_id;
2236         u64 inum;
2237         u64 file_pos;
2238         u64 extent_offset;
2239         u64 num_bytes;
2240         u64 generation;
2241 };
2242
2243 struct old_sa_defrag_extent {
2244         struct list_head list;
2245         struct new_sa_defrag_extent *new;
2246
2247         u64 extent_offset;
2248         u64 bytenr;
2249         u64 offset;
2250         u64 len;
2251         int count;
2252 };
2253
2254 struct new_sa_defrag_extent {
2255         struct rb_root root;
2256         struct list_head head;
2257         struct btrfs_path *path;
2258         struct inode *inode;
2259         u64 file_pos;
2260         u64 len;
2261         u64 bytenr;
2262         u64 disk_len;
2263         u8 compress_type;
2264 };
2265
2266 static int backref_comp(struct sa_defrag_extent_backref *b1,
2267                         struct sa_defrag_extent_backref *b2)
2268 {
2269         if (b1->root_id < b2->root_id)
2270                 return -1;
2271         else if (b1->root_id > b2->root_id)
2272                 return 1;
2273
2274         if (b1->inum < b2->inum)
2275                 return -1;
2276         else if (b1->inum > b2->inum)
2277                 return 1;
2278
2279         if (b1->file_pos < b2->file_pos)
2280                 return -1;
2281         else if (b1->file_pos > b2->file_pos)
2282                 return 1;
2283
2284         /*
2285          * [------------------------------] ===> (a range of space)
2286          *     |<--->|   |<---->| =============> (fs/file tree A)
2287          * |<---------------------------->| ===> (fs/file tree B)
2288          *
2289          * A range of space can refer to two file extents in one tree while
2290          * refer to only one file extent in another tree.
2291          *
2292          * So we may process a disk offset more than one time(two extents in A)
2293          * and locate at the same extent(one extent in B), then insert two same
2294          * backrefs(both refer to the extent in B).
2295          */
2296         return 0;
2297 }
2298
2299 static void backref_insert(struct rb_root *root,
2300                            struct sa_defrag_extent_backref *backref)
2301 {
2302         struct rb_node **p = &root->rb_node;
2303         struct rb_node *parent = NULL;
2304         struct sa_defrag_extent_backref *entry;
2305         int ret;
2306
2307         while (*p) {
2308                 parent = *p;
2309                 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2310
2311                 ret = backref_comp(backref, entry);
2312                 if (ret < 0)
2313                         p = &(*p)->rb_left;
2314                 else
2315                         p = &(*p)->rb_right;
2316         }
2317
2318         rb_link_node(&backref->node, parent, p);
2319         rb_insert_color(&backref->node, root);
2320 }
2321
2322 /*
2323  * Note the backref might has changed, and in this case we just return 0.
2324  */
2325 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2326                                        void *ctx)
2327 {
2328         struct btrfs_file_extent_item *extent;
2329         struct old_sa_defrag_extent *old = ctx;
2330         struct new_sa_defrag_extent *new = old->new;
2331         struct btrfs_path *path = new->path;
2332         struct btrfs_key key;
2333         struct btrfs_root *root;
2334         struct sa_defrag_extent_backref *backref;
2335         struct extent_buffer *leaf;
2336         struct inode *inode = new->inode;
2337         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2338         int slot;
2339         int ret;
2340         u64 extent_offset;
2341         u64 num_bytes;
2342
2343         if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2344             inum == btrfs_ino(BTRFS_I(inode)))
2345                 return 0;
2346
2347         key.objectid = root_id;
2348         key.type = BTRFS_ROOT_ITEM_KEY;
2349         key.offset = (u64)-1;
2350
2351         root = btrfs_read_fs_root_no_name(fs_info, &key);
2352         if (IS_ERR(root)) {
2353                 if (PTR_ERR(root) == -ENOENT)
2354                         return 0;
2355                 WARN_ON(1);
2356                 btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu",
2357                          inum, offset, root_id);
2358                 return PTR_ERR(root);
2359         }
2360
2361         key.objectid = inum;
2362         key.type = BTRFS_EXTENT_DATA_KEY;
2363         if (offset > (u64)-1 << 32)
2364                 key.offset = 0;
2365         else
2366                 key.offset = offset;
2367
2368         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2369         if (WARN_ON(ret < 0))
2370                 return ret;
2371         ret = 0;
2372
2373         while (1) {
2374                 cond_resched();
2375
2376                 leaf = path->nodes[0];
2377                 slot = path->slots[0];
2378
2379                 if (slot >= btrfs_header_nritems(leaf)) {
2380                         ret = btrfs_next_leaf(root, path);
2381                         if (ret < 0) {
2382                                 goto out;
2383                         } else if (ret > 0) {
2384                                 ret = 0;
2385                                 goto out;
2386                         }
2387                         continue;
2388                 }
2389
2390                 path->slots[0]++;
2391
2392                 btrfs_item_key_to_cpu(leaf, &key, slot);
2393
2394                 if (key.objectid > inum)
2395                         goto out;
2396
2397                 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2398                         continue;
2399
2400                 extent = btrfs_item_ptr(leaf, slot,
2401                                         struct btrfs_file_extent_item);
2402
2403                 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2404                         continue;
2405
2406                 /*
2407                  * 'offset' refers to the exact key.offset,
2408                  * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2409                  * (key.offset - extent_offset).
2410                  */
2411                 if (key.offset != offset)
2412                         continue;
2413
2414                 extent_offset = btrfs_file_extent_offset(leaf, extent);
2415                 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2416
2417                 if (extent_offset >= old->extent_offset + old->offset +
2418                     old->len || extent_offset + num_bytes <=
2419                     old->extent_offset + old->offset)
2420                         continue;
2421                 break;
2422         }
2423
2424         backref = kmalloc(sizeof(*backref), GFP_NOFS);
2425         if (!backref) {
2426                 ret = -ENOENT;
2427                 goto out;
2428         }
2429
2430         backref->root_id = root_id;
2431         backref->inum = inum;
2432         backref->file_pos = offset;
2433         backref->num_bytes = num_bytes;
2434         backref->extent_offset = extent_offset;
2435         backref->generation = btrfs_file_extent_generation(leaf, extent);
2436         backref->old = old;
2437         backref_insert(&new->root, backref);
2438         old->count++;
2439 out:
2440         btrfs_release_path(path);
2441         WARN_ON(ret);
2442         return ret;
2443 }
2444
2445 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2446                                    struct new_sa_defrag_extent *new)
2447 {
2448         struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2449         struct old_sa_defrag_extent *old, *tmp;
2450         int ret;
2451
2452         new->path = path;
2453
2454         list_for_each_entry_safe(old, tmp, &new->head, list) {
2455                 ret = iterate_inodes_from_logical(old->bytenr +
2456                                                   old->extent_offset, fs_info,
2457                                                   path, record_one_backref,
2458                                                   old, false);
2459                 if (ret < 0 && ret != -ENOENT)
2460                         return false;
2461
2462                 /* no backref to be processed for this extent */
2463                 if (!old->count) {
2464                         list_del(&old->list);
2465                         kfree(old);
2466                 }
2467         }
2468
2469         if (list_empty(&new->head))
2470                 return false;
2471
2472         return true;
2473 }
2474
2475 static int relink_is_mergable(struct extent_buffer *leaf,
2476                               struct btrfs_file_extent_item *fi,
2477                               struct new_sa_defrag_extent *new)
2478 {
2479         if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2480                 return 0;
2481
2482         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2483                 return 0;
2484
2485         if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2486                 return 0;
2487
2488         if (btrfs_file_extent_encryption(leaf, fi) ||
2489             btrfs_file_extent_other_encoding(leaf, fi))
2490                 return 0;
2491
2492         return 1;
2493 }
2494
2495 /*
2496  * Note the backref might has changed, and in this case we just return 0.
2497  */
2498 static noinline int relink_extent_backref(struct btrfs_path *path,
2499                                  struct sa_defrag_extent_backref *prev,
2500                                  struct sa_defrag_extent_backref *backref)
2501 {
2502         struct btrfs_file_extent_item *extent;
2503         struct btrfs_file_extent_item *item;
2504         struct btrfs_ordered_extent *ordered;
2505         struct btrfs_trans_handle *trans;
2506         struct btrfs_root *root;
2507         struct btrfs_key key;
2508         struct extent_buffer *leaf;
2509         struct old_sa_defrag_extent *old = backref->old;
2510         struct new_sa_defrag_extent *new = old->new;
2511         struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2512         struct inode *inode;
2513         struct extent_state *cached = NULL;
2514         int ret = 0;
2515         u64 start;
2516         u64 len;
2517         u64 lock_start;
2518         u64 lock_end;
2519         bool merge = false;
2520         int index;
2521
2522         if (prev && prev->root_id == backref->root_id &&
2523             prev->inum == backref->inum &&
2524             prev->file_pos + prev->num_bytes == backref->file_pos)
2525                 merge = true;
2526
2527         /* step 1: get root */
2528         key.objectid = backref->root_id;
2529         key.type = BTRFS_ROOT_ITEM_KEY;
2530         key.offset = (u64)-1;
2531
2532         index = srcu_read_lock(&fs_info->subvol_srcu);
2533
2534         root = btrfs_read_fs_root_no_name(fs_info, &key);
2535         if (IS_ERR(root)) {
2536                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2537                 if (PTR_ERR(root) == -ENOENT)
2538                         return 0;
2539                 return PTR_ERR(root);
2540         }
2541
2542         if (btrfs_root_readonly(root)) {
2543                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2544                 return 0;
2545         }
2546
2547         /* step 2: get inode */
2548         key.objectid = backref->inum;
2549         key.type = BTRFS_INODE_ITEM_KEY;
2550         key.offset = 0;
2551
2552         inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2553         if (IS_ERR(inode)) {
2554                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2555                 return 0;
2556         }
2557
2558         srcu_read_unlock(&fs_info->subvol_srcu, index);
2559
2560         /* step 3: relink backref */
2561         lock_start = backref->file_pos;
2562         lock_end = backref->file_pos + backref->num_bytes - 1;
2563         lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2564                          &cached);
2565
2566         ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2567         if (ordered) {
2568                 btrfs_put_ordered_extent(ordered);
2569                 goto out_unlock;
2570         }
2571
2572         trans = btrfs_join_transaction(root);
2573         if (IS_ERR(trans)) {
2574                 ret = PTR_ERR(trans);
2575                 goto out_unlock;
2576         }
2577
2578         key.objectid = backref->inum;
2579         key.type = BTRFS_EXTENT_DATA_KEY;
2580         key.offset = backref->file_pos;
2581
2582         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2583         if (ret < 0) {
2584                 goto out_free_path;
2585         } else if (ret > 0) {
2586                 ret = 0;
2587                 goto out_free_path;
2588         }
2589
2590         extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2591                                 struct btrfs_file_extent_item);
2592
2593         if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2594             backref->generation)
2595                 goto out_free_path;
2596
2597         btrfs_release_path(path);
2598
2599         start = backref->file_pos;
2600         if (backref->extent_offset < old->extent_offset + old->offset)
2601                 start += old->extent_offset + old->offset -
2602                          backref->extent_offset;
2603
2604         len = min(backref->extent_offset + backref->num_bytes,
2605                   old->extent_offset + old->offset + old->len);
2606         len -= max(backref->extent_offset, old->extent_offset + old->offset);
2607
2608         ret = btrfs_drop_extents(trans, root, inode, start,
2609                                  start + len, 1);
2610         if (ret)
2611                 goto out_free_path;
2612 again:
2613         key.objectid = btrfs_ino(BTRFS_I(inode));
2614         key.type = BTRFS_EXTENT_DATA_KEY;
2615         key.offset = start;
2616
2617         path->leave_spinning = 1;
2618         if (merge) {
2619                 struct btrfs_file_extent_item *fi;
2620                 u64 extent_len;
2621                 struct btrfs_key found_key;
2622
2623                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2624                 if (ret < 0)
2625                         goto out_free_path;
2626
2627                 path->slots[0]--;
2628                 leaf = path->nodes[0];
2629                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2630
2631                 fi = btrfs_item_ptr(leaf, path->slots[0],
2632                                     struct btrfs_file_extent_item);
2633                 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2634
2635                 if (extent_len + found_key.offset == start &&
2636                     relink_is_mergable(leaf, fi, new)) {
2637                         btrfs_set_file_extent_num_bytes(leaf, fi,
2638                                                         extent_len + len);
2639                         btrfs_mark_buffer_dirty(leaf);
2640                         inode_add_bytes(inode, len);
2641
2642                         ret = 1;
2643                         goto out_free_path;
2644                 } else {
2645                         merge = false;
2646                         btrfs_release_path(path);
2647                         goto again;
2648                 }
2649         }
2650
2651         ret = btrfs_insert_empty_item(trans, root, path, &key,
2652                                         sizeof(*extent));
2653         if (ret) {
2654                 btrfs_abort_transaction(trans, ret);
2655                 goto out_free_path;
2656         }
2657
2658         leaf = path->nodes[0];
2659         item = btrfs_item_ptr(leaf, path->slots[0],
2660                                 struct btrfs_file_extent_item);
2661         btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2662         btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2663         btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2664         btrfs_set_file_extent_num_bytes(leaf, item, len);
2665         btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2666         btrfs_set_file_extent_generation(leaf, item, trans->transid);
2667         btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2668         btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2669         btrfs_set_file_extent_encryption(leaf, item, 0);
2670         btrfs_set_file_extent_other_encoding(leaf, item, 0);
2671
2672         btrfs_mark_buffer_dirty(leaf);
2673         inode_add_bytes(inode, len);
2674         btrfs_release_path(path);
2675
2676         ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2677                         new->disk_len, 0,
2678                         backref->root_id, backref->inum,
2679                         new->file_pos); /* start - extent_offset */
2680         if (ret) {
2681                 btrfs_abort_transaction(trans, ret);
2682                 goto out_free_path;
2683         }
2684
2685         ret = 1;
2686 out_free_path:
2687         btrfs_release_path(path);
2688         path->leave_spinning = 0;
2689         btrfs_end_transaction(trans);
2690 out_unlock:
2691         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2692                              &cached, GFP_NOFS);
2693         iput(inode);
2694         return ret;
2695 }
2696
2697 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2698 {
2699         struct old_sa_defrag_extent *old, *tmp;
2700
2701         if (!new)
2702                 return;
2703
2704         list_for_each_entry_safe(old, tmp, &new->head, list) {
2705                 kfree(old);
2706         }
2707         kfree(new);
2708 }
2709
2710 static void relink_file_extents(struct new_sa_defrag_extent *new)
2711 {
2712         struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2713         struct btrfs_path *path;
2714         struct sa_defrag_extent_backref *backref;
2715         struct sa_defrag_extent_backref *prev = NULL;
2716         struct inode *inode;
2717         struct btrfs_root *root;
2718         struct rb_node *node;
2719         int ret;
2720
2721         inode = new->inode;
2722         root = BTRFS_I(inode)->root;
2723
2724         path = btrfs_alloc_path();
2725         if (!path)
2726                 return;
2727
2728         if (!record_extent_backrefs(path, new)) {
2729                 btrfs_free_path(path);
2730                 goto out;
2731         }
2732         btrfs_release_path(path);
2733
2734         while (1) {
2735                 node = rb_first(&new->root);
2736                 if (!node)
2737                         break;
2738                 rb_erase(node, &new->root);
2739
2740                 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2741
2742                 ret = relink_extent_backref(path, prev, backref);
2743                 WARN_ON(ret < 0);
2744
2745                 kfree(prev);
2746
2747                 if (ret == 1)
2748                         prev = backref;
2749                 else
2750                         prev = NULL;
2751                 cond_resched();
2752         }
2753         kfree(prev);
2754
2755         btrfs_free_path(path);
2756 out:
2757         free_sa_defrag_extent(new);
2758
2759         atomic_dec(&fs_info->defrag_running);
2760         wake_up(&fs_info->transaction_wait);
2761 }
2762
2763 static struct new_sa_defrag_extent *
2764 record_old_file_extents(struct inode *inode,
2765                         struct btrfs_ordered_extent *ordered)
2766 {
2767         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2768         struct btrfs_root *root = BTRFS_I(inode)->root;
2769         struct btrfs_path *path;
2770         struct btrfs_key key;
2771         struct old_sa_defrag_extent *old;
2772         struct new_sa_defrag_extent *new;
2773         int ret;
2774
2775         new = kmalloc(sizeof(*new), GFP_NOFS);
2776         if (!new)
2777                 return NULL;
2778
2779         new->inode = inode;
2780         new->file_pos = ordered->file_offset;
2781         new->len = ordered->len;
2782         new->bytenr = ordered->start;
2783         new->disk_len = ordered->disk_len;
2784         new->compress_type = ordered->compress_type;
2785         new->root = RB_ROOT;
2786         INIT_LIST_HEAD(&new->head);
2787
2788         path = btrfs_alloc_path();
2789         if (!path)
2790                 goto out_kfree;
2791
2792         key.objectid = btrfs_ino(BTRFS_I(inode));
2793         key.type = BTRFS_EXTENT_DATA_KEY;
2794         key.offset = new->file_pos;
2795
2796         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2797         if (ret < 0)
2798                 goto out_free_path;
2799         if (ret > 0 && path->slots[0] > 0)
2800                 path->slots[0]--;
2801
2802         /* find out all the old extents for the file range */
2803         while (1) {
2804                 struct btrfs_file_extent_item *extent;
2805                 struct extent_buffer *l;
2806                 int slot;
2807                 u64 num_bytes;
2808                 u64 offset;
2809                 u64 end;
2810                 u64 disk_bytenr;
2811                 u64 extent_offset;
2812
2813                 l = path->nodes[0];
2814                 slot = path->slots[0];
2815
2816                 if (slot >= btrfs_header_nritems(l)) {
2817                         ret = btrfs_next_leaf(root, path);
2818                         if (ret < 0)
2819                                 goto out_free_path;
2820                         else if (ret > 0)
2821                                 break;
2822                         continue;
2823                 }
2824
2825                 btrfs_item_key_to_cpu(l, &key, slot);
2826
2827                 if (key.objectid != btrfs_ino(BTRFS_I(inode)))
2828                         break;
2829                 if (key.type != BTRFS_EXTENT_DATA_KEY)
2830                         break;
2831                 if (key.offset >= new->file_pos + new->len)
2832                         break;
2833
2834                 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2835
2836                 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2837                 if (key.offset + num_bytes < new->file_pos)
2838                         goto next;
2839
2840                 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2841                 if (!disk_bytenr)
2842                         goto next;
2843
2844                 extent_offset = btrfs_file_extent_offset(l, extent);
2845
2846                 old = kmalloc(sizeof(*old), GFP_NOFS);
2847                 if (!old)
2848                         goto out_free_path;
2849
2850                 offset = max(new->file_pos, key.offset);
2851                 end = min(new->file_pos + new->len, key.offset + num_bytes);
2852
2853                 old->bytenr = disk_bytenr;
2854                 old->extent_offset = extent_offset;
2855                 old->offset = offset - key.offset;
2856                 old->len = end - offset;
2857                 old->new = new;
2858                 old->count = 0;
2859                 list_add_tail(&old->list, &new->head);
2860 next:
2861                 path->slots[0]++;
2862                 cond_resched();
2863         }
2864
2865         btrfs_free_path(path);
2866         atomic_inc(&fs_info->defrag_running);
2867
2868         return new;
2869
2870 out_free_path:
2871         btrfs_free_path(path);
2872 out_kfree:
2873         free_sa_defrag_extent(new);
2874         return NULL;
2875 }
2876
2877 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2878                                          u64 start, u64 len)
2879 {
2880         struct btrfs_block_group_cache *cache;
2881
2882         cache = btrfs_lookup_block_group(fs_info, start);
2883         ASSERT(cache);
2884
2885         spin_lock(&cache->lock);
2886         cache->delalloc_bytes -= len;
2887         spin_unlock(&cache->lock);
2888
2889         btrfs_put_block_group(cache);
2890 }
2891
2892 /* as ordered data IO finishes, this gets called so we can finish
2893  * an ordered extent if the range of bytes in the file it covers are
2894  * fully written.
2895  */
2896 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2897 {
2898         struct inode *inode = ordered_extent->inode;
2899         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2900         struct btrfs_root *root = BTRFS_I(inode)->root;
2901         struct btrfs_trans_handle *trans = NULL;
2902         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2903         struct extent_state *cached_state = NULL;
2904         struct new_sa_defrag_extent *new = NULL;
2905         int compress_type = 0;
2906         int ret = 0;
2907         u64 logical_len = ordered_extent->len;
2908         bool nolock;
2909         bool truncated = false;
2910         bool range_locked = false;
2911         bool clear_new_delalloc_bytes = false;
2912
2913         if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2914             !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2915             !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2916                 clear_new_delalloc_bytes = true;
2917
2918         nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
2919
2920         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2921                 ret = -EIO;
2922                 goto out;
2923         }
2924
2925         btrfs_free_io_failure_record(BTRFS_I(inode),
2926                         ordered_extent->file_offset,
2927                         ordered_extent->file_offset +
2928                         ordered_extent->len - 1);
2929
2930         if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2931                 truncated = true;
2932                 logical_len = ordered_extent->truncated_len;
2933                 /* Truncated the entire extent, don't bother adding */
2934                 if (!logical_len)
2935                         goto out;
2936         }
2937
2938         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2939                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2940
2941                 /*
2942                  * For mwrite(mmap + memset to write) case, we still reserve
2943                  * space for NOCOW range.
2944                  * As NOCOW won't cause a new delayed ref, just free the space
2945                  */
2946                 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
2947                                        ordered_extent->len);
2948                 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2949                 if (nolock)
2950                         trans = btrfs_join_transaction_nolock(root);
2951                 else
2952                         trans = btrfs_join_transaction(root);
2953                 if (IS_ERR(trans)) {
2954                         ret = PTR_ERR(trans);
2955                         trans = NULL;
2956                         goto out;
2957                 }
2958                 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
2959                 ret = btrfs_update_inode_fallback(trans, root, inode);
2960                 if (ret) /* -ENOMEM or corruption */
2961                         btrfs_abort_transaction(trans, ret);
2962                 goto out;
2963         }
2964
2965         range_locked = true;
2966         lock_extent_bits(io_tree, ordered_extent->file_offset,
2967                          ordered_extent->file_offset + ordered_extent->len - 1,
2968                          &cached_state);
2969
2970         ret = test_range_bit(io_tree, ordered_extent->file_offset,
2971                         ordered_extent->file_offset + ordered_extent->len - 1,
2972                         EXTENT_DEFRAG, 0, cached_state);
2973         if (ret) {
2974                 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2975                 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2976                         /* the inode is shared */
2977                         new = record_old_file_extents(inode, ordered_extent);
2978
2979                 clear_extent_bit(io_tree, ordered_extent->file_offset,
2980                         ordered_extent->file_offset + ordered_extent->len - 1,
2981                         EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2982         }
2983
2984         if (nolock)
2985                 trans = btrfs_join_transaction_nolock(root);
2986         else
2987                 trans = btrfs_join_transaction(root);
2988         if (IS_ERR(trans)) {
2989                 ret = PTR_ERR(trans);
2990                 trans = NULL;
2991                 goto out;
2992         }
2993
2994         trans->block_rsv = &BTRFS_I(inode)->block_rsv;
2995
2996         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2997                 compress_type = ordered_extent->compress_type;
2998         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2999                 BUG_ON(compress_type);
3000                 ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
3001                                                 ordered_extent->file_offset,
3002                                                 ordered_extent->file_offset +
3003                                                 logical_len);
3004         } else {
3005                 BUG_ON(root == fs_info->tree_root);
3006                 ret = insert_reserved_file_extent(trans, inode,
3007                                                 ordered_extent->file_offset,
3008                                                 ordered_extent->start,
3009                                                 ordered_extent->disk_len,
3010                                                 logical_len, logical_len,
3011                                                 compress_type, 0, 0,
3012                                                 BTRFS_FILE_EXTENT_REG);
3013                 if (!ret)
3014                         btrfs_release_delalloc_bytes(fs_info,
3015                                                      ordered_extent->start,
3016                                                      ordered_extent->disk_len);
3017         }
3018         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
3019                            ordered_extent->file_offset, ordered_extent->len,
3020                            trans->transid);
3021         if (ret < 0) {
3022                 btrfs_abort_transaction(trans, ret);
3023                 goto out;
3024         }
3025
3026         add_pending_csums(trans, inode, &ordered_extent->list);
3027
3028         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
3029         ret = btrfs_update_inode_fallback(trans, root, inode);
3030         if (ret) { /* -ENOMEM or corruption */
3031                 btrfs_abort_transaction(trans, ret);
3032                 goto out;
3033         }
3034         ret = 0;
3035 out:
3036         if (range_locked || clear_new_delalloc_bytes) {
3037                 unsigned int clear_bits = 0;
3038
3039                 if (range_locked)
3040                         clear_bits |= EXTENT_LOCKED;
3041                 if (clear_new_delalloc_bytes)
3042                         clear_bits |= EXTENT_DELALLOC_NEW;
3043                 clear_extent_bit(&BTRFS_I(inode)->io_tree,
3044                                  ordered_extent->file_offset,
3045                                  ordered_extent->file_offset +
3046                                  ordered_extent->len - 1,
3047                                  clear_bits,
3048                                  (clear_bits & EXTENT_LOCKED) ? 1 : 0,
3049                                  0, &cached_state, GFP_NOFS);
3050         }
3051
3052         if (trans)
3053                 btrfs_end_transaction(trans);
3054
3055         if (ret || truncated) {
3056                 u64 start, end;
3057
3058                 if (truncated)
3059                         start = ordered_extent->file_offset + logical_len;
3060                 else
3061                         start = ordered_extent->file_offset;
3062                 end = ordered_extent->file_offset + ordered_extent->len - 1;
3063                 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
3064
3065                 /* Drop the cache for the part of the extent we didn't write. */
3066                 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
3067
3068                 /*
3069                  * If the ordered extent had an IOERR or something else went
3070                  * wrong we need to return the space for this ordered extent
3071                  * back to the allocator.  We only free the extent in the
3072                  * truncated case if we didn't write out the extent at all.
3073                  */
3074                 if ((ret || !logical_len) &&
3075                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3076                     !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
3077                         btrfs_free_reserved_extent(fs_info,
3078                                                    ordered_extent->start,
3079                                                    ordered_extent->disk_len, 1);
3080         }
3081
3082
3083         /*
3084          * This needs to be done to make sure anybody waiting knows we are done
3085          * updating everything for this ordered extent.
3086          */
3087         btrfs_remove_ordered_extent(inode, ordered_extent);
3088
3089         /* for snapshot-aware defrag */
3090         if (new) {
3091                 if (ret) {
3092                         free_sa_defrag_extent(new);
3093                         atomic_dec(&fs_info->defrag_running);
3094                 } else {
3095                         relink_file_extents(new);
3096                 }
3097         }
3098
3099         /* once for us */
3100         btrfs_put_ordered_extent(ordered_extent);
3101         /* once for the tree */
3102         btrfs_put_ordered_extent(ordered_extent);
3103
3104         return ret;
3105 }
3106
3107 static void finish_ordered_fn(struct btrfs_work *work)
3108 {
3109         struct btrfs_ordered_extent *ordered_extent;
3110         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3111         btrfs_finish_ordered_io(ordered_extent);
3112 }
3113
3114 static void btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3115                                 struct extent_state *state, int uptodate)
3116 {
3117         struct inode *inode = page->mapping->host;
3118         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3119         struct btrfs_ordered_extent *ordered_extent = NULL;
3120         struct btrfs_workqueue *wq;
3121         btrfs_work_func_t func;
3122
3123         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3124
3125         ClearPagePrivate2(page);
3126         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3127                                             end - start + 1, uptodate))
3128                 return;
3129
3130         if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
3131                 wq = fs_info->endio_freespace_worker;
3132                 func = btrfs_freespace_write_helper;
3133         } else {
3134                 wq = fs_info->endio_write_workers;
3135                 func = btrfs_endio_write_helper;
3136         }
3137
3138         btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3139                         NULL);
3140         btrfs_queue_work(wq, &ordered_extent->work);
3141 }
3142
3143 static int __readpage_endio_check(struct inode *inode,
3144                                   struct btrfs_io_bio *io_bio,
3145                                   int icsum, struct page *page,
3146                                   int pgoff, u64 start, size_t len)
3147 {
3148         char *kaddr;
3149         u32 csum_expected;
3150         u32 csum = ~(u32)0;
3151
3152         csum_expected = *(((u32 *)io_bio->csum) + icsum);
3153
3154         kaddr = kmap_atomic(page);
3155         csum = btrfs_csum_data(kaddr + pgoff, csum,  len);
3156         btrfs_csum_final(csum, (u8 *)&csum);
3157         if (csum != csum_expected)
3158                 goto zeroit;
3159
3160         kunmap_atomic(kaddr);
3161         return 0;
3162 zeroit:
3163         btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3164                                     io_bio->mirror_num);
3165         memset(kaddr + pgoff, 1, len);
3166         flush_dcache_page(page);
3167         kunmap_atomic(kaddr);
3168         return -EIO;
3169 }
3170
3171 /*
3172  * when reads are done, we need to check csums to verify the data is correct
3173  * if there's a match, we allow the bio to finish.  If not, the code in
3174  * extent_io.c will try to find good copies for us.
3175  */
3176 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3177                                       u64 phy_offset, struct page *page,
3178                                       u64 start, u64 end, int mirror)
3179 {
3180         size_t offset = start - page_offset(page);
3181         struct inode *inode = page->mapping->host;
3182         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3183         struct btrfs_root *root = BTRFS_I(inode)->root;
3184
3185         if (PageChecked(page)) {
3186                 ClearPageChecked(page);
3187                 return 0;
3188         }
3189
3190         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3191                 return 0;
3192
3193         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3194             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3195                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3196                 return 0;
3197         }
3198
3199         phy_offset >>= inode->i_sb->s_blocksize_bits;
3200         return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3201                                       start, (size_t)(end - start + 1));
3202 }
3203
3204 void btrfs_add_delayed_iput(struct inode *inode)
3205 {
3206         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3207         struct btrfs_inode *binode = BTRFS_I(inode);
3208
3209         if (atomic_add_unless(&inode->i_count, -1, 1))
3210                 return;
3211
3212         spin_lock(&fs_info->delayed_iput_lock);
3213         if (binode->delayed_iput_count == 0) {
3214                 ASSERT(list_empty(&binode->delayed_iput));
3215                 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3216         } else {
3217                 binode->delayed_iput_count++;
3218         }
3219         spin_unlock(&fs_info->delayed_iput_lock);
3220 }
3221
3222 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3223 {
3224
3225         spin_lock(&fs_info->delayed_iput_lock);
3226         while (!list_empty(&fs_info->delayed_iputs)) {
3227                 struct btrfs_inode *inode;
3228
3229                 inode = list_first_entry(&fs_info->delayed_iputs,
3230                                 struct btrfs_inode, delayed_iput);
3231                 if (inode->delayed_iput_count) {
3232                         inode->delayed_iput_count--;
3233                         list_move_tail(&inode->delayed_iput,
3234                                         &fs_info->delayed_iputs);
3235                 } else {
3236                         list_del_init(&inode->delayed_iput);
3237                 }
3238                 spin_unlock(&fs_info->delayed_iput_lock);
3239                 iput(&inode->vfs_inode);
3240                 spin_lock(&fs_info->delayed_iput_lock);
3241         }
3242         spin_unlock(&fs_info->delayed_iput_lock);
3243 }
3244
3245 /*
3246  * This is called in transaction commit time. If there are no orphan
3247  * files in the subvolume, it removes orphan item and frees block_rsv
3248  * structure.
3249  */
3250 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3251                               struct btrfs_root *root)
3252 {
3253         struct btrfs_fs_info *fs_info = root->fs_info;
3254         struct btrfs_block_rsv *block_rsv;
3255         int ret;
3256
3257         if (atomic_read(&root->orphan_inodes) ||
3258             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3259                 return;
3260
3261         spin_lock(&root->orphan_lock);
3262         if (atomic_read(&root->orphan_inodes)) {
3263                 spin_unlock(&root->orphan_lock);
3264                 return;
3265         }
3266
3267         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3268                 spin_unlock(&root->orphan_lock);
3269                 return;
3270         }
3271
3272         block_rsv = root->orphan_block_rsv;
3273         root->orphan_block_rsv = NULL;
3274         spin_unlock(&root->orphan_lock);
3275
3276         if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3277             btrfs_root_refs(&root->root_item) > 0) {
3278                 ret = btrfs_del_orphan_item(trans, fs_info->tree_root,
3279                                             root->root_key.objectid);
3280                 if (ret)
3281                         btrfs_abort_transaction(trans, ret);
3282                 else
3283                         clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3284                                   &root->state);
3285         }
3286
3287         if (block_rsv) {
3288                 WARN_ON(block_rsv->size > 0);
3289                 btrfs_free_block_rsv(fs_info, block_rsv);
3290         }
3291 }
3292
3293 /*
3294  * This creates an orphan entry for the given inode in case something goes
3295  * wrong in the middle of an unlink/truncate.
3296  *
3297  * NOTE: caller of this function should reserve 5 units of metadata for
3298  *       this function.
3299  */
3300 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3301                 struct btrfs_inode *inode)
3302 {
3303         struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
3304         struct btrfs_root *root = inode->root;
3305         struct btrfs_block_rsv *block_rsv = NULL;
3306         int reserve = 0;
3307         int insert = 0;
3308         int ret;
3309
3310         if (!root->orphan_block_rsv) {
3311                 block_rsv = btrfs_alloc_block_rsv(fs_info,
3312                                                   BTRFS_BLOCK_RSV_TEMP);
3313                 if (!block_rsv)
3314                         return -ENOMEM;
3315         }
3316
3317         spin_lock(&root->orphan_lock);
3318         if (!root->orphan_block_rsv) {
3319                 root->orphan_block_rsv = block_rsv;
3320         } else if (block_rsv) {
3321                 btrfs_free_block_rsv(fs_info, block_rsv);
3322                 block_rsv = NULL;
3323         }
3324
3325         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3326                               &inode->runtime_flags)) {
3327 #if 0
3328                 /*
3329                  * For proper ENOSPC handling, we should do orphan
3330                  * cleanup when mounting. But this introduces backward
3331                  * compatibility issue.
3332                  */
3333                 if (!xchg(&root->orphan_item_inserted, 1))
3334                         insert = 2;
3335                 else
3336                         insert = 1;
3337 #endif
3338                 insert = 1;
3339                 atomic_inc(&root->orphan_inodes);
3340         }
3341
3342         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3343                               &inode->runtime_flags))
3344                 reserve = 1;
3345         spin_unlock(&root->orphan_lock);
3346
3347         /* grab metadata reservation from transaction handle */
3348         if (reserve) {
3349                 ret = btrfs_orphan_reserve_metadata(trans, inode);
3350                 ASSERT(!ret);
3351                 if (ret) {
3352                         atomic_dec(&root->orphan_inodes);
3353                         clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3354                                   &inode->runtime_flags);
3355                         if (insert)
3356                                 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3357                                           &inode->runtime_flags);
3358                         return ret;
3359                 }
3360         }
3361
3362         /* insert an orphan item to track this unlinked/truncated file */
3363         if (insert >= 1) {
3364                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3365                 if (ret) {
3366                         atomic_dec(&root->orphan_inodes);
3367                         if (reserve) {
3368                                 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3369                                           &inode->runtime_flags);
3370                                 btrfs_orphan_release_metadata(inode);
3371                         }
3372                         if (ret != -EEXIST) {
3373                                 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3374                                           &inode->runtime_flags);
3375                                 btrfs_abort_transaction(trans, ret);
3376                                 return ret;
3377                         }
3378                 }
3379                 ret = 0;
3380         }
3381
3382         /* insert an orphan item to track subvolume contains orphan files */
3383         if (insert >= 2) {
3384                 ret = btrfs_insert_orphan_item(trans, fs_info->tree_root,
3385                                                root->root_key.objectid);
3386                 if (ret && ret != -EEXIST) {
3387                         btrfs_abort_transaction(trans, ret);
3388                         return ret;
3389                 }
3390         }
3391         return 0;
3392 }
3393
3394 /*
3395  * We have done the truncate/delete so we can go ahead and remove the orphan
3396  * item for this particular inode.
3397  */
3398 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3399                             struct btrfs_inode *inode)
3400 {
3401         struct btrfs_root *root = inode->root;
3402         int delete_item = 0;
3403         int release_rsv = 0;
3404         int ret = 0;
3405
3406         spin_lock(&root->orphan_lock);
3407         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3408                                &inode->runtime_flags))
3409                 delete_item = 1;
3410
3411         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3412                                &inode->runtime_flags))
3413                 release_rsv = 1;
3414         spin_unlock(&root->orphan_lock);
3415
3416         if (delete_item) {
3417                 atomic_dec(&root->orphan_inodes);
3418                 if (trans)
3419                         ret = btrfs_del_orphan_item(trans, root,
3420                                                     btrfs_ino(inode));
3421         }
3422
3423         if (release_rsv)
3424                 btrfs_orphan_release_metadata(inode);
3425
3426         return ret;
3427 }
3428
3429 /*
3430  * this cleans up any orphans that may be left on the list from the last use
3431  * of this root.
3432  */
3433 int btrfs_orphan_cleanup(struct btrfs_root *root)
3434 {
3435         struct btrfs_fs_info *fs_info = root->fs_info;
3436         struct btrfs_path *path;
3437         struct extent_buffer *leaf;
3438         struct btrfs_key key, found_key;
3439         struct btrfs_trans_handle *trans;
3440         struct inode *inode;
3441         u64 last_objectid = 0;
3442         int ret = 0, nr_unlink = 0, nr_truncate = 0;
3443
3444         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3445                 return 0;
3446
3447         path = btrfs_alloc_path();
3448         if (!path) {
3449                 ret = -ENOMEM;
3450                 goto out;
3451         }
3452         path->reada = READA_BACK;
3453
3454         key.objectid = BTRFS_ORPHAN_OBJECTID;
3455         key.type = BTRFS_ORPHAN_ITEM_KEY;
3456         key.offset = (u64)-1;
3457
3458         while (1) {
3459                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3460                 if (ret < 0)
3461                         goto out;
3462
3463                 /*
3464                  * if ret == 0 means we found what we were searching for, which
3465                  * is weird, but possible, so only screw with path if we didn't
3466                  * find the key and see if we have stuff that matches
3467                  */
3468                 if (ret > 0) {
3469                         ret = 0;
3470                         if (path->slots[0] == 0)
3471                                 break;
3472                         path->slots[0]--;
3473                 }
3474
3475                 /* pull out the item */
3476                 leaf = path->nodes[0];
3477                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3478
3479                 /* make sure the item matches what we want */
3480                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3481                         break;
3482                 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3483                         break;
3484
3485                 /* release the path since we're done with it */
3486                 btrfs_release_path(path);
3487
3488                 /*
3489                  * this is where we are basically btrfs_lookup, without the
3490                  * crossing root thing.  we store the inode number in the
3491                  * offset of the orphan item.
3492                  */
3493
3494                 if (found_key.offset == last_objectid) {
3495                         btrfs_err(fs_info,
3496                                   "Error removing orphan entry, stopping orphan cleanup");
3497                         ret = -EINVAL;
3498                         goto out;
3499                 }
3500
3501                 last_objectid = found_key.offset;
3502
3503                 found_key.objectid = found_key.offset;
3504                 found_key.type = BTRFS_INODE_ITEM_KEY;
3505                 found_key.offset = 0;
3506                 inode = btrfs_iget(fs_info->sb, &found_key, root, NULL);
3507                 ret = PTR_ERR_OR_ZERO(inode);
3508                 if (ret && ret != -ENOENT)
3509                         goto out;
3510
3511                 if (ret == -ENOENT && root == fs_info->tree_root) {
3512                         struct btrfs_root *dead_root;
3513                         struct btrfs_fs_info *fs_info = root->fs_info;
3514                         int is_dead_root = 0;
3515
3516                         /*
3517                          * this is an orphan in the tree root. Currently these
3518                          * could come from 2 sources:
3519                          *  a) a snapshot deletion in progress
3520                          *  b) a free space cache inode
3521                          * We need to distinguish those two, as the snapshot
3522                          * orphan must not get deleted.
3523                          * find_dead_roots already ran before us, so if this
3524                          * is a snapshot deletion, we should find the root
3525                          * in the dead_roots list
3526                          */
3527                         spin_lock(&fs_info->trans_lock);
3528                         list_for_each_entry(dead_root, &fs_info->dead_roots,
3529                                             root_list) {
3530                                 if (dead_root->root_key.objectid ==
3531                                     found_key.objectid) {
3532                                         is_dead_root = 1;
3533                                         break;
3534                                 }
3535                         }
3536                         spin_unlock(&fs_info->trans_lock);
3537                         if (is_dead_root) {
3538                                 /* prevent this orphan from being found again */
3539                                 key.offset = found_key.objectid - 1;
3540                                 continue;
3541                         }
3542                 }
3543                 /*
3544                  * Inode is already gone but the orphan item is still there,
3545                  * kill the orphan item.
3546                  */
3547                 if (ret == -ENOENT) {
3548                         trans = btrfs_start_transaction(root, 1);
3549                         if (IS_ERR(trans)) {
3550                                 ret = PTR_ERR(trans);
3551                                 goto out;
3552                         }
3553                         btrfs_debug(fs_info, "auto deleting %Lu",
3554                                     found_key.objectid);
3555                         ret = btrfs_del_orphan_item(trans, root,
3556                                                     found_key.objectid);
3557                         btrfs_end_transaction(trans);
3558                         if (ret)
3559                                 goto out;
3560                         continue;
3561                 }
3562
3563                 /*
3564                  * add this inode to the orphan list so btrfs_orphan_del does
3565                  * the proper thing when we hit it
3566                  */
3567                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3568                         &BTRFS_I(inode)->runtime_flags);
3569                 atomic_inc(&root->orphan_inodes);
3570
3571                 /* if we have links, this was a truncate, lets do that */
3572                 if (inode->i_nlink) {
3573                         if (WARN_ON(!S_ISREG(inode->i_mode))) {
3574                                 iput(inode);
3575                                 continue;
3576                         }
3577                         nr_truncate++;
3578
3579                         /* 1 for the orphan item deletion. */
3580                         trans = btrfs_start_transaction(root, 1);
3581                         if (IS_ERR(trans)) {
3582                                 iput(inode);
3583                                 ret = PTR_ERR(trans);
3584                                 goto out;
3585                         }
3586                         ret = btrfs_orphan_add(trans, BTRFS_I(inode));
3587                         btrfs_end_transaction(trans);
3588                         if (ret) {
3589                                 iput(inode);
3590                                 goto out;
3591                         }
3592
3593                         ret = btrfs_truncate(inode);
3594                         if (ret)
3595                                 btrfs_orphan_del(NULL, BTRFS_I(inode));
3596                 } else {
3597                         nr_unlink++;
3598                 }
3599
3600                 /* this will do delete_inode and everything for us */
3601                 iput(inode);
3602                 if (ret)
3603                         goto out;
3604         }
3605         /* release the path since we're done with it */
3606         btrfs_release_path(path);
3607
3608         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3609
3610         if (root->orphan_block_rsv)
3611                 btrfs_block_rsv_release(fs_info, root->orphan_block_rsv,
3612                                         (u64)-1);
3613
3614         if (root->orphan_block_rsv ||
3615             test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3616                 trans = btrfs_join_transaction(root);
3617                 if (!IS_ERR(trans))
3618                         btrfs_end_transaction(trans);
3619         }
3620
3621         if (nr_unlink)
3622                 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3623         if (nr_truncate)
3624                 btrfs_debug(fs_info, "truncated %d orphans", nr_truncate);
3625
3626 out:
3627         if (ret)
3628                 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3629         btrfs_free_path(path);
3630         return ret;
3631 }
3632
3633 /*
3634  * very simple check to peek ahead in the leaf looking for xattrs.  If we
3635  * don't find any xattrs, we know there can't be any acls.
3636  *
3637  * slot is the slot the inode is in, objectid is the objectid of the inode
3638  */
3639 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3640                                           int slot, u64 objectid,
3641                                           int *first_xattr_slot)
3642 {
3643         u32 nritems = btrfs_header_nritems(leaf);
3644         struct btrfs_key found_key;
3645         static u64 xattr_access = 0;
3646         static u64 xattr_default = 0;
3647         int scanned = 0;
3648
3649         if (!xattr_access) {
3650                 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3651                                         strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3652                 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3653                                         strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3654         }
3655
3656         slot++;
3657         *first_xattr_slot = -1;
3658         while (slot < nritems) {
3659                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3660
3661                 /* we found a different objectid, there must not be acls */
3662                 if (found_key.objectid != objectid)
3663                         return 0;
3664
3665                 /* we found an xattr, assume we've got an acl */
3666                 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3667                         if (*first_xattr_slot == -1)
3668                                 *first_xattr_slot = slot;
3669                         if (found_key.offset == xattr_access ||
3670                             found_key.offset == xattr_default)
3671                                 return 1;
3672                 }
3673
3674                 /*
3675                  * we found a key greater than an xattr key, there can't
3676                  * be any acls later on
3677                  */
3678                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3679                         return 0;
3680
3681                 slot++;
3682                 scanned++;
3683
3684                 /*
3685                  * it goes inode, inode backrefs, xattrs, extents,
3686                  * so if there are a ton of hard links to an inode there can
3687                  * be a lot of backrefs.  Don't waste time searching too hard,
3688                  * this is just an optimization
3689                  */
3690                 if (scanned >= 8)
3691                         break;
3692         }
3693         /* we hit the end of the leaf before we found an xattr or
3694          * something larger than an xattr.  We have to assume the inode
3695          * has acls
3696          */
3697         if (*first_xattr_slot == -1)
3698                 *first_xattr_slot = slot;
3699         return 1;
3700 }
3701
3702 /*
3703  * read an inode from the btree into the in-memory inode
3704  */
3705 static int btrfs_read_locked_inode(struct inode *inode)
3706 {
3707         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3708         struct btrfs_path *path;
3709         struct extent_buffer *leaf;
3710         struct btrfs_inode_item *inode_item;
3711         struct btrfs_root *root = BTRFS_I(inode)->root;
3712         struct btrfs_key location;
3713         unsigned long ptr;
3714         int maybe_acls;
3715         u32 rdev;
3716         int ret;
3717         bool filled = false;
3718         int first_xattr_slot;
3719
3720         ret = btrfs_fill_inode(inode, &rdev);
3721         if (!ret)
3722                 filled = true;
3723
3724         path = btrfs_alloc_path();
3725         if (!path) {
3726                 ret = -ENOMEM;
3727                 goto make_bad;
3728         }
3729
3730         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3731
3732         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3733         if (ret) {
3734                 if (ret > 0)
3735                         ret = -ENOENT;
3736                 goto make_bad;
3737         }
3738
3739         leaf = path->nodes[0];
3740
3741         if (filled)
3742                 goto cache_index;
3743
3744         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3745                                     struct btrfs_inode_item);
3746         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3747         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3748         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3749         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3750         btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3751
3752         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3753         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3754
3755         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3756         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3757
3758         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3759         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3760
3761         BTRFS_I(inode)->i_otime.tv_sec =
3762                 btrfs_timespec_sec(leaf, &inode_item->otime);
3763         BTRFS_I(inode)->i_otime.tv_nsec =
3764                 btrfs_timespec_nsec(leaf, &inode_item->otime);
3765
3766         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3767         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3768         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3769
3770         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3771         inode->i_generation = BTRFS_I(inode)->generation;
3772         inode->i_rdev = 0;
3773         rdev = btrfs_inode_rdev(leaf, inode_item);
3774
3775         BTRFS_I(inode)->index_cnt = (u64)-1;
3776         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3777
3778 cache_index:
3779         /*
3780          * If we were modified in the current generation and evicted from memory
3781          * and then re-read we need to do a full sync since we don't have any
3782          * idea about which extents were modified before we were evicted from
3783          * cache.
3784          *
3785          * This is required for both inode re-read from disk and delayed inode
3786          * in delayed_nodes_tree.
3787          */
3788         if (BTRFS_I(inode)->last_trans == fs_info->generation)
3789                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3790                         &BTRFS_I(inode)->runtime_flags);
3791
3792         /*
3793          * We don't persist the id of the transaction where an unlink operation
3794          * against the inode was last made. So here we assume the inode might
3795          * have been evicted, and therefore the exact value of last_unlink_trans
3796          * lost, and set it to last_trans to avoid metadata inconsistencies
3797          * between the inode and its parent if the inode is fsync'ed and the log
3798          * replayed. For example, in the scenario:
3799          *
3800          * touch mydir/foo
3801          * ln mydir/foo mydir/bar
3802          * sync
3803          * unlink mydir/bar
3804          * echo 2 > /proc/sys/vm/drop_caches   # evicts inode
3805          * xfs_io -c fsync mydir/foo
3806          * <power failure>
3807          * mount fs, triggers fsync log replay
3808          *
3809          * We must make sure that when we fsync our inode foo we also log its
3810          * parent inode, otherwise after log replay the parent still has the
3811          * dentry with the "bar" name but our inode foo has a link count of 1
3812          * and doesn't have an inode ref with the name "bar" anymore.
3813          *
3814          * Setting last_unlink_trans to last_trans is a pessimistic approach,
3815          * but it guarantees correctness at the expense of occasional full
3816          * transaction commits on fsync if our inode is a directory, or if our
3817          * inode is not a directory, logging its parent unnecessarily.
3818          */
3819         BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3820
3821         path->slots[0]++;
3822         if (inode->i_nlink != 1 ||
3823             path->slots[0] >= btrfs_header_nritems(leaf))
3824                 goto cache_acl;
3825
3826         btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3827         if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3828                 goto cache_acl;
3829
3830         ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3831         if (location.type == BTRFS_INODE_REF_KEY) {
3832                 struct btrfs_inode_ref *ref;
3833
3834                 ref = (struct btrfs_inode_ref *)ptr;
3835                 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3836         } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3837                 struct btrfs_inode_extref *extref;
3838
3839                 extref = (struct btrfs_inode_extref *)ptr;
3840                 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3841                                                                      extref);
3842         }
3843 cache_acl:
3844         /*
3845          * try to precache a NULL acl entry for files that don't have
3846          * any xattrs or acls
3847          */
3848         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3849                         btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3850         if (first_xattr_slot != -1) {
3851                 path->slots[0] = first_xattr_slot;
3852                 ret = btrfs_load_inode_props(inode, path);
3853                 if (ret)
3854                         btrfs_err(fs_info,
3855                                   "error loading props for ino %llu (root %llu): %d",
3856                                   btrfs_ino(BTRFS_I(inode)),
3857                                   root->root_key.objectid, ret);
3858         }
3859         btrfs_free_path(path);
3860
3861         if (!maybe_acls)
3862                 cache_no_acl(inode);
3863
3864         switch (inode->i_mode & S_IFMT) {
3865         case S_IFREG:
3866                 inode->i_mapping->a_ops = &btrfs_aops;
3867                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3868                 inode->i_fop = &btrfs_file_operations;
3869                 inode->i_op = &btrfs_file_inode_operations;
3870                 break;
3871         case S_IFDIR:
3872                 inode->i_fop = &btrfs_dir_file_operations;
3873                 inode->i_op = &btrfs_dir_inode_operations;
3874                 break;
3875         case S_IFLNK:
3876                 inode->i_op = &btrfs_symlink_inode_operations;
3877                 inode_nohighmem(inode);
3878                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3879                 break;
3880         default:
3881                 inode->i_op = &btrfs_special_inode_operations;
3882                 init_special_inode(inode, inode->i_mode, rdev);
3883                 break;
3884         }
3885
3886         btrfs_update_iflags(inode);
3887         return 0;
3888
3889 make_bad:
3890         btrfs_free_path(path);
3891         make_bad_inode(inode);
3892         return ret;
3893 }
3894
3895 /*
3896  * given a leaf and an inode, copy the inode fields into the leaf
3897  */
3898 static void fill_inode_item(struct btrfs_trans_handle *trans,
3899                             struct extent_buffer *leaf,
3900                             struct btrfs_inode_item *item,
3901                             struct inode *inode)
3902 {
3903         struct btrfs_map_token token;
3904
3905         btrfs_init_map_token(&token);
3906
3907         btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3908         btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3909         btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3910                                    &token);
3911         btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3912         btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3913
3914         btrfs_set_token_timespec_sec(leaf, &item->atime,
3915                                      inode->i_atime.tv_sec, &token);
3916         btrfs_set_token_timespec_nsec(leaf, &item->atime,
3917                                       inode->i_atime.tv_nsec, &token);
3918
3919         btrfs_set_token_timespec_sec(leaf, &item->mtime,
3920                                      inode->i_mtime.tv_sec, &token);
3921         btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3922                                       inode->i_mtime.tv_nsec, &token);
3923
3924         btrfs_set_token_timespec_sec(leaf, &item->ctime,
3925                                      inode->i_ctime.tv_sec, &token);
3926         btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3927                                       inode->i_ctime.tv_nsec, &token);
3928
3929         btrfs_set_token_timespec_sec(leaf, &item->otime,
3930                                      BTRFS_I(inode)->i_otime.tv_sec, &token);
3931         btrfs_set_token_timespec_nsec(leaf, &item->otime,
3932                                       BTRFS_I(inode)->i_otime.tv_nsec, &token);
3933
3934         btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3935                                      &token);
3936         btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3937                                          &token);
3938         btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3939         btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3940         btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3941         btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3942         btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3943 }
3944
3945 /*
3946  * copy everything in the in-memory inode into the btree.
3947  */
3948 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3949                                 struct btrfs_root *root, struct inode *inode)
3950 {
3951         struct btrfs_inode_item *inode_item;
3952         struct btrfs_path *path;
3953         struct extent_buffer *leaf;
3954         int ret;
3955
3956         path = btrfs_alloc_path();
3957         if (!path)
3958                 return -ENOMEM;
3959
3960         path->leave_spinning = 1;
3961         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3962                                  1);
3963         if (ret) {
3964                 if (ret > 0)
3965                         ret = -ENOENT;
3966                 goto failed;
3967         }
3968
3969         leaf = path->nodes[0];
3970         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3971                                     struct btrfs_inode_item);
3972
3973         fill_inode_item(trans, leaf, inode_item, inode);
3974         btrfs_mark_buffer_dirty(leaf);
3975         btrfs_set_inode_last_trans(trans, inode);
3976         ret = 0;
3977 failed:
3978         btrfs_free_path(path);
3979         return ret;
3980 }
3981
3982 /*
3983  * copy everything in the in-memory inode into the btree.
3984  */
3985 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3986                                 struct btrfs_root *root, struct inode *inode)
3987 {
3988         struct btrfs_fs_info *fs_info = root->fs_info;
3989         int ret;
3990
3991         /*
3992          * If the inode is a free space inode, we can deadlock during commit
3993          * if we put it into the delayed code.
3994          *
3995          * The data relocation inode should also be directly updated
3996          * without delay
3997          */
3998         if (!btrfs_is_free_space_inode(BTRFS_I(inode))
3999             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
4000             && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4001                 btrfs_update_root_times(trans, root);
4002
4003                 ret = btrfs_delayed_update_inode(trans, root, inode);
4004                 if (!ret)
4005                         btrfs_set_inode_last_trans(trans, inode);
4006                 return ret;
4007         }
4008
4009         return btrfs_update_inode_item(trans, root, inode);
4010 }
4011
4012 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4013                                          struct btrfs_root *root,
4014                                          struct inode *inode)
4015 {
4016         int ret;
4017
4018         ret = btrfs_update_inode(trans, root, inode);
4019         if (ret == -ENOSPC)
4020                 return btrfs_update_inode_item(trans, root, inode);
4021         return ret;
4022 }
4023
4024 /*
4025  * unlink helper that gets used here in inode.c and in the tree logging
4026  * recovery code.  It remove a link in a directory with a given name, and
4027  * also drops the back refs in the inode to the directory
4028  */
4029 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4030                                 struct btrfs_root *root,
4031                                 struct btrfs_inode *dir,
4032                                 struct btrfs_inode *inode,
4033                                 const char *name, int name_len)
4034 {
4035         struct btrfs_fs_info *fs_info = root->fs_info;
4036         struct btrfs_path *path;
4037         int ret = 0;
4038         struct extent_buffer *leaf;
4039         struct btrfs_dir_item *di;
4040         struct btrfs_key key;
4041         u64 index;
4042         u64 ino = btrfs_ino(inode);
4043         u64 dir_ino = btrfs_ino(dir);
4044
4045         path = btrfs_alloc_path();
4046         if (!path) {
4047                 ret = -ENOMEM;
4048                 goto out;
4049         }
4050
4051         path->leave_spinning = 1;
4052         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4053                                     name, name_len, -1);
4054         if (IS_ERR(di)) {
4055                 ret = PTR_ERR(di);
4056                 goto err;
4057         }
4058         if (!di) {
4059                 ret = -ENOENT;
4060                 goto err;
4061         }
4062         leaf = path->nodes[0];
4063         btrfs_dir_item_key_to_cpu(leaf, di, &key);
4064         ret = btrfs_delete_one_dir_name(trans, root, path, di);
4065         if (ret)
4066                 goto err;
4067         btrfs_release_path(path);
4068
4069         /*
4070          * If we don't have dir index, we have to get it by looking up
4071          * the inode ref, since we get the inode ref, remove it directly,
4072          * it is unnecessary to do delayed deletion.
4073          *
4074          * But if we have dir index, needn't search inode ref to get it.
4075          * Since the inode ref is close to the inode item, it is better
4076          * that we delay to delete it, and just do this deletion when
4077          * we update the inode item.
4078          */
4079         if (inode->dir_index) {
4080                 ret = btrfs_delayed_delete_inode_ref(inode);
4081                 if (!ret) {
4082                         index = inode->dir_index;
4083                         goto skip_backref;
4084                 }
4085         }
4086
4087         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4088                                   dir_ino, &index);
4089         if (ret) {
4090                 btrfs_info(fs_info,
4091                         "failed to delete reference to %.*s, inode %llu parent %llu",
4092                         name_len, name, ino, dir_ino);
4093                 btrfs_abort_transaction(trans, ret);
4094                 goto err;
4095         }
4096 skip_backref:
4097         ret = btrfs_delete_delayed_dir_index(trans, fs_info, dir, index);
4098         if (ret) {
4099                 btrfs_abort_transaction(trans, ret);
4100                 goto err;
4101         }
4102
4103         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
4104                         dir_ino);
4105         if (ret != 0 && ret != -ENOENT) {
4106                 btrfs_abort_transaction(trans, ret);
4107                 goto err;
4108         }
4109
4110         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
4111                         index);
4112         if (ret == -ENOENT)
4113                 ret = 0;
4114         else if (ret)
4115                 btrfs_abort_transaction(trans, ret);
4116 err:
4117         btrfs_free_path(path);
4118         if (ret)
4119                 goto out;
4120
4121         btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
4122         inode_inc_iversion(&inode->vfs_inode);
4123         inode_inc_iversion(&dir->vfs_inode);
4124         inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
4125                 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4126         ret = btrfs_update_inode(trans, root, &dir->vfs_inode);
4127 out:
4128         return ret;
4129 }
4130
4131 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4132                        struct btrfs_root *root,
4133                        struct btrfs_inode *dir, struct btrfs_inode *inode,
4134                        const char *name, int name_len)
4135 {
4136         int ret;
4137         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4138         if (!ret) {
4139                 drop_nlink(&inode->vfs_inode);
4140                 ret = btrfs_update_inode(trans, root, &inode->vfs_inode);
4141         }
4142         return ret;
4143 }
4144
4145 /*
4146  * helper to start transaction for unlink and rmdir.
4147  *
4148  * unlink and rmdir are special in btrfs, they do not always free space, so
4149  * if we cannot make our reservations the normal way try and see if there is
4150  * plenty of slack room in the global reserve to migrate, otherwise we cannot
4151  * allow the unlink to occur.
4152  */
4153 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4154 {
4155         struct btrfs_root *root = BTRFS_I(dir)->root;
4156
4157         /*
4158          * 1 for the possible orphan item
4159          * 1 for the dir item
4160          * 1 for the dir index
4161          * 1 for the inode ref
4162          * 1 for the inode
4163          */
4164         return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4165 }
4166
4167 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4168 {
4169         struct btrfs_root *root = BTRFS_I(dir)->root;
4170         struct btrfs_trans_handle *trans;
4171         struct inode *inode = d_inode(dentry);
4172         int ret;
4173
4174         trans = __unlink_start_trans(dir);
4175         if (IS_ERR(trans))
4176                 return PTR_ERR(trans);
4177
4178         btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4179                         0);
4180
4181         ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4182                         BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4183                         dentry->d_name.len);
4184         if (ret)
4185                 goto out;
4186
4187         if (inode->i_nlink == 0) {
4188                 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4189                 if (ret)
4190                         goto out;
4191         }
4192
4193 out:
4194         btrfs_end_transaction(trans);
4195         btrfs_btree_balance_dirty(root->fs_info);
4196         return ret;
4197 }
4198
4199 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4200                         struct btrfs_root *root,
4201                         struct inode *dir, u64 objectid,
4202                         const char *name, int name_len)
4203 {
4204         struct btrfs_fs_info *fs_info = root->fs_info;
4205         struct btrfs_path *path;
4206         struct extent_buffer *leaf;
4207         struct btrfs_dir_item *di;
4208         struct btrfs_key key;
4209         u64 index;
4210         int ret;
4211         u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4212
4213         path = btrfs_alloc_path();
4214         if (!path)
4215                 return -ENOMEM;
4216
4217         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4218                                    name, name_len, -1);
4219         if (IS_ERR_OR_NULL(di)) {
4220                 if (!di)
4221                         ret = -ENOENT;
4222                 else
4223                         ret = PTR_ERR(di);
4224                 goto out;
4225         }
4226
4227         leaf = path->nodes[0];
4228         btrfs_dir_item_key_to_cpu(leaf, di, &key);
4229         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4230         ret = btrfs_delete_one_dir_name(trans, root, path, di);
4231         if (ret) {
4232                 btrfs_abort_transaction(trans, ret);
4233                 goto out;
4234         }
4235         btrfs_release_path(path);
4236
4237         ret = btrfs_del_root_ref(trans, fs_info, objectid,
4238                                  root->root_key.objectid, dir_ino,
4239                                  &index, name, name_len);
4240         if (ret < 0) {
4241                 if (ret != -ENOENT) {
4242                         btrfs_abort_transaction(trans, ret);
4243                         goto out;
4244                 }
4245                 di = btrfs_search_dir_index_item(root, path, dir_ino,
4246                                                  name, name_len);
4247                 if (IS_ERR_OR_NULL(di)) {
4248                         if (!di)
4249                                 ret = -ENOENT;
4250                         else
4251                                 ret = PTR_ERR(di);
4252                         btrfs_abort_transaction(trans, ret);
4253                         goto out;
4254                 }
4255
4256                 leaf = path->nodes[0];
4257                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4258                 btrfs_release_path(path);
4259                 index = key.offset;
4260         }
4261         btrfs_release_path(path);
4262
4263         ret = btrfs_delete_delayed_dir_index(trans, fs_info, BTRFS_I(dir), index);
4264         if (ret) {
4265                 btrfs_abort_transaction(trans, ret);
4266                 goto out;
4267         }
4268
4269         btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4270         inode_inc_iversion(dir);
4271         dir->i_mtime = dir->i_ctime = current_time(dir);
4272         ret = btrfs_update_inode_fallback(trans, root, dir);
4273         if (ret)
4274                 btrfs_abort_transaction(trans, ret);
4275 out:
4276         btrfs_free_path(path);
4277         return ret;
4278 }
4279
4280 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4281 {
4282         struct inode *inode = d_inode(dentry);
4283         int err = 0;
4284         struct btrfs_root *root = BTRFS_I(dir)->root;
4285         struct btrfs_trans_handle *trans;
4286         u64 last_unlink_trans;
4287
4288         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4289                 return -ENOTEMPTY;
4290         if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4291                 return -EPERM;
4292
4293         trans = __unlink_start_trans(dir);
4294         if (IS_ERR(trans))
4295                 return PTR_ERR(trans);
4296
4297         if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4298                 err = btrfs_unlink_subvol(trans, root, dir,
4299                                           BTRFS_I(inode)->location.objectid,
4300                                           dentry->d_name.name,
4301                                           dentry->d_name.len);
4302                 goto out;
4303         }
4304
4305         err = btrfs_orphan_add(trans, BTRFS_I(inode));
4306         if (err)
4307                 goto out;
4308
4309         last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4310
4311         /* now the directory is empty */
4312         err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4313                         BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4314                         dentry->d_name.len);
4315         if (!err) {
4316                 btrfs_i_size_write(BTRFS_I(inode), 0);
4317                 /*
4318                  * Propagate the last_unlink_trans value of the deleted dir to
4319                  * its parent directory. This is to prevent an unrecoverable
4320                  * log tree in the case we do something like this:
4321                  * 1) create dir foo
4322                  * 2) create snapshot under dir foo
4323                  * 3) delete the snapshot
4324                  * 4) rmdir foo
4325                  * 5) mkdir foo
4326                  * 6) fsync foo or some file inside foo
4327                  */
4328                 if (last_unlink_trans >= trans->transid)
4329                         BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4330         }
4331 out:
4332         btrfs_end_transaction(trans);
4333         btrfs_btree_balance_dirty(root->fs_info);
4334
4335         return err;
4336 }
4337
4338 static int truncate_space_check(struct btrfs_trans_handle *trans,
4339                                 struct btrfs_root *root,
4340                                 u64 bytes_deleted)
4341 {
4342         struct btrfs_fs_info *fs_info = root->fs_info;
4343         int ret;
4344
4345         /*
4346          * This is only used to apply pressure to the enospc system, we don't
4347          * intend to use this reservation at all.
4348          */
4349         bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted);
4350         bytes_deleted *= fs_info->nodesize;
4351         ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
4352                                   bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4353         if (!ret) {
4354                 trace_btrfs_space_reservation(fs_info, "transaction",
4355                                               trans->transid,
4356                                               bytes_deleted, 1);
4357                 trans->bytes_reserved += bytes_deleted;
4358         }
4359         return ret;
4360
4361 }
4362
4363 /*
4364  * Return this if we need to call truncate_block for the last bit of the
4365  * truncate.
4366  */
4367 #define NEED_TRUNCATE_BLOCK 1
4368
4369 /*
4370  * this can truncate away extent items, csum items and directory items.
4371  * It starts at a high offset and removes keys until it can't find
4372  * any higher than new_size
4373  *
4374  * csum items that cross the new i_size are truncated to the new size
4375  * as well.
4376  *
4377  * min_type is the minimum key type to truncate down to.  If set to 0, this
4378  * will kill all the items on this inode, including the INODE_ITEM_KEY.
4379  */
4380 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4381                                struct btrfs_root *root,
4382                                struct inode *inode,
4383                                u64 new_size, u32 min_type)
4384 {
4385         struct btrfs_fs_info *fs_info = root->fs_info;
4386         struct btrfs_path *path;
4387         struct extent_buffer *leaf;
4388         struct btrfs_file_extent_item *fi;
4389         struct btrfs_key key;
4390         struct btrfs_key found_key;
4391         u64 extent_start = 0;
4392         u64 extent_num_bytes = 0;
4393         u64 extent_offset = 0;
4394         u64 item_end = 0;
4395         u64 last_size = new_size;
4396         u32 found_type = (u8)-1;
4397         int found_extent;
4398         int del_item;
4399         int pending_del_nr = 0;
4400         int pending_del_slot = 0;
4401         int extent_type = -1;
4402         int ret;
4403         int err = 0;
4404         u64 ino = btrfs_ino(BTRFS_I(inode));
4405         u64 bytes_deleted = 0;
4406         bool be_nice = false;
4407         bool should_throttle = false;
4408         bool should_end = false;
4409
4410         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4411
4412         /*
4413          * for non-free space inodes and ref cows, we want to back off from
4414          * time to time
4415          */
4416         if (!btrfs_is_free_space_inode(BTRFS_I(inode)) &&
4417             test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4418                 be_nice = true;
4419
4420         path = btrfs_alloc_path();
4421         if (!path)
4422                 return -ENOMEM;
4423         path->reada = READA_BACK;
4424
4425         /*
4426          * We want to drop from the next block forward in case this new size is
4427          * not block aligned since we will be keeping the last block of the
4428          * extent just the way it is.
4429          */
4430         if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4431             root == fs_info->tree_root)
4432                 btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size,
4433                                         fs_info->sectorsize),
4434                                         (u64)-1, 0);
4435
4436         /*
4437          * This function is also used to drop the items in the log tree before
4438          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4439          * it is used to drop the loged items. So we shouldn't kill the delayed
4440          * items.
4441          */
4442         if (min_type == 0 && root == BTRFS_I(inode)->root)
4443                 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
4444
4445         key.objectid = ino;
4446         key.offset = (u64)-1;
4447         key.type = (u8)-1;
4448
4449 search_again:
4450         /*
4451          * with a 16K leaf size and 128MB extents, you can actually queue
4452          * up a huge file in a single leaf.  Most of the time that
4453          * bytes_deleted is > 0, it will be huge by the time we get here
4454          */
4455         if (be_nice && bytes_deleted > SZ_32M) {
4456                 if (btrfs_should_end_transaction(trans)) {
4457                         err = -EAGAIN;
4458                         goto error;
4459                 }
4460         }
4461
4462
4463         path->leave_spinning = 1;
4464         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4465         if (ret < 0) {
4466                 err = ret;
4467                 goto out;
4468         }
4469
4470         if (ret > 0) {
4471                 /* there are no items in the tree for us to truncate, we're
4472                  * done
4473                  */
4474                 if (path->slots[0] == 0)
4475                         goto out;
4476                 path->slots[0]--;
4477         }
4478
4479         while (1) {
4480                 fi = NULL;
4481                 leaf = path->nodes[0];
4482                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4483                 found_type = found_key.type;
4484
4485                 if (found_key.objectid != ino)
4486                         break;
4487
4488                 if (found_type < min_type)
4489                         break;
4490
4491                 item_end = found_key.offset;
4492                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4493                         fi = btrfs_item_ptr(leaf, path->slots[0],
4494                                             struct btrfs_file_extent_item);
4495                         extent_type = btrfs_file_extent_type(leaf, fi);
4496                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4497                                 item_end +=
4498                                     btrfs_file_extent_num_bytes(leaf, fi);
4499
4500                                 trace_btrfs_truncate_show_fi_regular(
4501                                         BTRFS_I(inode), leaf, fi,
4502                                         found_key.offset);
4503                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4504                                 item_end += btrfs_file_extent_inline_len(leaf,
4505                                                          path->slots[0], fi);
4506
4507                                 trace_btrfs_truncate_show_fi_inline(
4508                                         BTRFS_I(inode), leaf, fi, path->slots[0],
4509                                         found_key.offset);
4510                         }
4511                         item_end--;
4512                 }
4513                 if (found_type > min_type) {
4514                         del_item = 1;
4515                 } else {
4516                         if (item_end < new_size)
4517                                 break;
4518                         if (found_key.offset >= new_size)
4519                                 del_item = 1;
4520                         else
4521                                 del_item = 0;
4522                 }
4523                 found_extent = 0;
4524                 /* FIXME, shrink the extent if the ref count is only 1 */
4525                 if (found_type != BTRFS_EXTENT_DATA_KEY)
4526                         goto delete;
4527
4528                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4529                         u64 num_dec;
4530                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4531                         if (!del_item) {
4532                                 u64 orig_num_bytes =
4533                                         btrfs_file_extent_num_bytes(leaf, fi);
4534                                 extent_num_bytes = ALIGN(new_size -
4535                                                 found_key.offset,
4536                                                 fs_info->sectorsize);
4537                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4538                                                          extent_num_bytes);
4539                                 num_dec = (orig_num_bytes -
4540                                            extent_num_bytes);
4541                                 if (test_bit(BTRFS_ROOT_REF_COWS,
4542                                              &root->state) &&
4543                                     extent_start != 0)
4544                                         inode_sub_bytes(inode, num_dec);
4545                                 btrfs_mark_buffer_dirty(leaf);
4546                         } else {
4547                                 extent_num_bytes =
4548                                         btrfs_file_extent_disk_num_bytes(leaf,
4549                                                                          fi);
4550                                 extent_offset = found_key.offset -
4551                                         btrfs_file_extent_offset(leaf, fi);
4552
4553                                 /* FIXME blocksize != 4096 */
4554                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4555                                 if (extent_start != 0) {
4556                                         found_extent = 1;
4557                                         if (test_bit(BTRFS_ROOT_REF_COWS,
4558                                                      &root->state))
4559                                                 inode_sub_bytes(inode, num_dec);
4560                                 }
4561                         }
4562                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4563                         /*
4564                          * we can't truncate inline items that have had
4565                          * special encodings
4566                          */
4567                         if (!del_item &&
4568                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
4569                             btrfs_file_extent_other_encoding(leaf, fi) == 0 &&
4570                             btrfs_file_extent_compression(leaf, fi) == 0) {
4571                                 u32 size = (u32)(new_size - found_key.offset);
4572
4573                                 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4574                                 size = btrfs_file_extent_calc_inline_size(size);
4575                                 btrfs_truncate_item(root->fs_info, path, size, 1);
4576                         } else if (!del_item) {
4577                                 /*
4578                                  * We have to bail so the last_size is set to
4579                                  * just before this extent.
4580                                  */
4581                                 err = NEED_TRUNCATE_BLOCK;
4582                                 break;
4583                         }
4584
4585                         if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4586                                 inode_sub_bytes(inode, item_end + 1 - new_size);
4587                 }
4588 delete:
4589                 if (del_item)
4590                         last_size = found_key.offset;
4591                 else
4592                         last_size = new_size;
4593                 if (del_item) {
4594                         if (!pending_del_nr) {
4595                                 /* no pending yet, add ourselves */
4596                                 pending_del_slot = path->slots[0];
4597                                 pending_del_nr = 1;
4598                         } else if (pending_del_nr &&
4599                                    path->slots[0] + 1 == pending_del_slot) {
4600                                 /* hop on the pending chunk */
4601                                 pending_del_nr++;
4602                                 pending_del_slot = path->slots[0];
4603                         } else {
4604                                 BUG();
4605                         }
4606                 } else {
4607                         break;
4608                 }
4609                 should_throttle = false;
4610
4611                 if (found_extent &&
4612                     (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4613                      root == fs_info->tree_root)) {
4614                         btrfs_set_path_blocking(path);
4615                         bytes_deleted += extent_num_bytes;
4616                         ret = btrfs_free_extent(trans, root, extent_start,
4617                                                 extent_num_bytes, 0,
4618                                                 btrfs_header_owner(leaf),
4619                                                 ino, extent_offset);
4620                         BUG_ON(ret);
4621                         if (btrfs_should_throttle_delayed_refs(trans, fs_info))
4622                                 btrfs_async_run_delayed_refs(fs_info,
4623                                         trans->delayed_ref_updates * 2,
4624                                         trans->transid, 0);
4625                         if (be_nice) {
4626                                 if (truncate_space_check(trans, root,
4627                                                          extent_num_bytes)) {
4628                                         should_end = true;
4629                                 }
4630                                 if (btrfs_should_throttle_delayed_refs(trans,
4631                                                                        fs_info))
4632                                         should_throttle = true;
4633                         }
4634                 }
4635
4636                 if (found_type == BTRFS_INODE_ITEM_KEY)
4637                         break;
4638
4639                 if (path->slots[0] == 0 ||
4640                     path->slots[0] != pending_del_slot ||
4641                     should_throttle || should_end) {
4642                         if (pending_del_nr) {
4643                                 ret = btrfs_del_items(trans, root, path,
4644                                                 pending_del_slot,
4645                                                 pending_del_nr);
4646                                 if (ret) {
4647                                         btrfs_abort_transaction(trans, ret);
4648                                         goto error;
4649                                 }
4650                                 pending_del_nr = 0;
4651                         }
4652                         btrfs_release_path(path);
4653                         if (should_throttle) {
4654                                 unsigned long updates = trans->delayed_ref_updates;
4655                                 if (updates) {
4656                                         trans->delayed_ref_updates = 0;
4657                                         ret = btrfs_run_delayed_refs(trans,
4658                                                                    fs_info,
4659                                                                    updates * 2);
4660                                         if (ret && !err)
4661                                                 err = ret;
4662                                 }
4663                         }
4664                         /*
4665                          * if we failed to refill our space rsv, bail out
4666                          * and let the transaction restart
4667                          */
4668                         if (should_end) {
4669                                 err = -EAGAIN;
4670                                 goto error;
4671                         }
4672                         goto search_again;
4673                 } else {
4674                         path->slots[0]--;
4675                 }
4676         }
4677 out:
4678         if (pending_del_nr) {
4679                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4680                                       pending_del_nr);
4681                 if (ret)
4682                         btrfs_abort_transaction(trans, ret);
4683         }
4684 error:
4685         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4686                 ASSERT(last_size >= new_size);
4687                 if (!err && last_size > new_size)
4688                         last_size = new_size;
4689                 btrfs_ordered_update_i_size(inode, last_size, NULL);
4690         }
4691
4692         btrfs_free_path(path);
4693
4694         if (be_nice && bytes_deleted > SZ_32M) {
4695                 unsigned long updates = trans->delayed_ref_updates;
4696                 if (updates) {
4697                         trans->delayed_ref_updates = 0;
4698                         ret = btrfs_run_delayed_refs(trans, fs_info,
4699                                                      updates * 2);
4700                         if (ret && !err)
4701                                 err = ret;
4702                 }
4703         }
4704         return err;
4705 }
4706
4707 /*
4708  * btrfs_truncate_block - read, zero a chunk and write a block
4709  * @inode - inode that we're zeroing
4710  * @from - the offset to start zeroing
4711  * @len - the length to zero, 0 to zero the entire range respective to the
4712  *      offset
4713  * @front - zero up to the offset instead of from the offset on
4714  *
4715  * This will find the block for the "from" offset and cow the block and zero the
4716  * part we want to zero.  This is used with truncate and hole punching.
4717  */
4718 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4719                         int front)
4720 {
4721         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4722         struct address_space *mapping = inode->i_mapping;
4723         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4724         struct btrfs_ordered_extent *ordered;
4725         struct extent_state *cached_state = NULL;
4726         struct extent_changeset *data_reserved = NULL;
4727         char *kaddr;
4728         u32 blocksize = fs_info->sectorsize;
4729         pgoff_t index = from >> PAGE_SHIFT;
4730         unsigned offset = from & (blocksize - 1);
4731         struct page *page;
4732         gfp_t mask = btrfs_alloc_write_mask(mapping);
4733         int ret = 0;
4734         u64 block_start;
4735         u64 block_end;
4736
4737         if ((offset & (blocksize - 1)) == 0 &&
4738             (!len || ((len & (blocksize - 1)) == 0)))
4739                 goto out;
4740
4741         block_start = round_down(from, blocksize);
4742         block_end = block_start + blocksize - 1;
4743
4744         ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
4745                                            block_start, blocksize);
4746         if (ret)
4747                 goto out;
4748
4749 again:
4750         page = find_or_create_page(mapping, index, mask);
4751         if (!page) {
4752                 btrfs_delalloc_release_space(inode, data_reserved,
4753                                              block_start, blocksize);
4754                 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize);
4755                 ret = -ENOMEM;
4756                 goto out;
4757         }
4758
4759         if (!PageUptodate(page)) {
4760                 ret = btrfs_readpage(NULL, page);
4761                 lock_page(page);
4762                 if (page->mapping != mapping) {
4763                         unlock_page(page);
4764                         put_page(page);
4765                         goto again;
4766                 }
4767                 if (!PageUptodate(page)) {
4768                         ret = -EIO;
4769                         goto out_unlock;
4770                 }
4771         }
4772         wait_on_page_writeback(page);
4773
4774         lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4775         set_page_extent_mapped(page);
4776
4777         ordered = btrfs_lookup_ordered_extent(inode, block_start);
4778         if (ordered) {
4779                 unlock_extent_cached(io_tree, block_start, block_end,
4780                                      &cached_state, GFP_NOFS);
4781                 unlock_page(page);
4782                 put_page(page);
4783                 btrfs_start_ordered_extent(inode, ordered, 1);
4784                 btrfs_put_ordered_extent(ordered);
4785                 goto again;
4786         }
4787
4788         clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4789                           EXTENT_DIRTY | EXTENT_DELALLOC |
4790                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4791                           0, 0, &cached_state, GFP_NOFS);
4792
4793         ret = btrfs_set_extent_delalloc(inode, block_start, block_end,
4794                                         &cached_state, 0);
4795         if (ret) {
4796                 unlock_extent_cached(io_tree, block_start, block_end,
4797                                      &cached_state, GFP_NOFS);
4798                 goto out_unlock;
4799         }
4800
4801         if (offset != blocksize) {
4802                 if (!len)
4803                         len = blocksize - offset;
4804                 kaddr = kmap(page);
4805                 if (front)
4806                         memset(kaddr + (block_start - page_offset(page)),
4807                                 0, offset);
4808                 else
4809                         memset(kaddr + (block_start - page_offset(page)) +  offset,
4810                                 0, len);
4811                 flush_dcache_page(page);
4812                 kunmap(page);
4813         }
4814         ClearPageChecked(page);
4815         set_page_dirty(page);
4816         unlock_extent_cached(io_tree, block_start, block_end, &cached_state,
4817                              GFP_NOFS);
4818
4819 out_unlock:
4820         if (ret)
4821                 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4822                                              blocksize);
4823         btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize);
4824         unlock_page(page);
4825         put_page(page);
4826 out:
4827         extent_changeset_free(data_reserved);
4828         return ret;
4829 }
4830
4831 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4832                              u64 offset, u64 len)
4833 {
4834         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4835         struct btrfs_trans_handle *trans;
4836         int ret;
4837
4838         /*
4839          * Still need to make sure the inode looks like it's been updated so
4840          * that any holes get logged if we fsync.
4841          */
4842         if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4843                 BTRFS_I(inode)->last_trans = fs_info->generation;
4844                 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4845                 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4846                 return 0;
4847         }
4848
4849         /*
4850          * 1 - for the one we're dropping
4851          * 1 - for the one we're adding
4852          * 1 - for updating the inode.
4853          */
4854         trans = btrfs_start_transaction(root, 3);
4855         if (IS_ERR(trans))
4856                 return PTR_ERR(trans);
4857
4858         ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4859         if (ret) {
4860                 btrfs_abort_transaction(trans, ret);
4861                 btrfs_end_transaction(trans);
4862                 return ret;
4863         }
4864
4865         ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)),
4866                         offset, 0, 0, len, 0, len, 0, 0, 0);
4867         if (ret)
4868                 btrfs_abort_transaction(trans, ret);
4869         else
4870                 btrfs_update_inode(trans, root, inode);
4871         btrfs_end_transaction(trans);
4872         return ret;
4873 }
4874
4875 /*
4876  * This function puts in dummy file extents for the area we're creating a hole
4877  * for.  So if we are truncating this file to a larger size we need to insert
4878  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4879  * the range between oldsize and size
4880  */
4881 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4882 {
4883         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4884         struct btrfs_root *root = BTRFS_I(inode)->root;
4885         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4886         struct extent_map *em = NULL;
4887         struct extent_state *cached_state = NULL;
4888         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4889         u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4890         u64 block_end = ALIGN(size, fs_info->sectorsize);
4891         u64 last_byte;
4892         u64 cur_offset;
4893         u64 hole_size;
4894         int err = 0;
4895
4896         /*
4897          * If our size started in the middle of a block we need to zero out the
4898          * rest of the block before we expand the i_size, otherwise we could
4899          * expose stale data.
4900          */
4901         err = btrfs_truncate_block(inode, oldsize, 0, 0);
4902         if (err)
4903                 return err;
4904
4905         if (size <= hole_start)
4906                 return 0;
4907
4908         while (1) {
4909                 struct btrfs_ordered_extent *ordered;
4910
4911                 lock_extent_bits(io_tree, hole_start, block_end - 1,
4912                                  &cached_state);
4913                 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), hole_start,
4914                                                      block_end - hole_start);
4915                 if (!ordered)
4916                         break;
4917                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4918                                      &cached_state, GFP_NOFS);
4919                 btrfs_start_ordered_extent(inode, ordered, 1);
4920                 btrfs_put_ordered_extent(ordered);
4921         }
4922
4923         cur_offset = hole_start;
4924         while (1) {
4925                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
4926                                 block_end - cur_offset, 0);
4927                 if (IS_ERR(em)) {
4928                         err = PTR_ERR(em);
4929                         em = NULL;
4930                         break;
4931                 }
4932                 last_byte = min(extent_map_end(em), block_end);
4933                 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4934                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4935                         struct extent_map *hole_em;
4936                         hole_size = last_byte - cur_offset;
4937
4938                         err = maybe_insert_hole(root, inode, cur_offset,
4939                                                 hole_size);
4940                         if (err)
4941                                 break;
4942                         btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
4943                                                 cur_offset + hole_size - 1, 0);
4944                         hole_em = alloc_extent_map();
4945                         if (!hole_em) {
4946                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4947                                         &BTRFS_I(inode)->runtime_flags);
4948                                 goto next;
4949                         }
4950                         hole_em->start = cur_offset;
4951                         hole_em->len = hole_size;
4952                         hole_em->orig_start = cur_offset;
4953
4954                         hole_em->block_start = EXTENT_MAP_HOLE;
4955                         hole_em->block_len = 0;
4956                         hole_em->orig_block_len = 0;
4957                         hole_em->ram_bytes = hole_size;
4958                         hole_em->bdev = fs_info->fs_devices->latest_bdev;
4959                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
4960                         hole_em->generation = fs_info->generation;
4961
4962                         while (1) {
4963                                 write_lock(&em_tree->lock);
4964                                 err = add_extent_mapping(em_tree, hole_em, 1);
4965                                 write_unlock(&em_tree->lock);
4966                                 if (err != -EEXIST)
4967                                         break;
4968                                 btrfs_drop_extent_cache(BTRFS_I(inode),
4969                                                         cur_offset,
4970                                                         cur_offset +
4971                                                         hole_size - 1, 0);
4972                         }
4973                         free_extent_map(hole_em);
4974                 }
4975 next:
4976                 free_extent_map(em);
4977                 em = NULL;
4978                 cur_offset = last_byte;
4979                 if (cur_offset >= block_end)
4980                         break;
4981         }
4982         free_extent_map(em);
4983         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4984                              GFP_NOFS);
4985         return err;
4986 }
4987
4988 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4989 {
4990         struct btrfs_root *root = BTRFS_I(inode)->root;
4991         struct btrfs_trans_handle *trans;
4992         loff_t oldsize = i_size_read(inode);
4993         loff_t newsize = attr->ia_size;
4994         int mask = attr->ia_valid;
4995         int ret;
4996
4997         /*
4998          * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4999          * special case where we need to update the times despite not having
5000          * these flags set.  For all other operations the VFS set these flags
5001          * explicitly if it wants a timestamp update.
5002          */
5003         if (newsize != oldsize) {
5004                 inode_inc_iversion(inode);
5005                 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5006                         inode->i_ctime = inode->i_mtime =
5007                                 current_time(inode);
5008         }
5009
5010         if (newsize > oldsize) {
5011                 /*
5012                  * Don't do an expanding truncate while snapshotting is ongoing.
5013                  * This is to ensure the snapshot captures a fully consistent
5014                  * state of this file - if the snapshot captures this expanding
5015                  * truncation, it must capture all writes that happened before
5016                  * this truncation.
5017                  */
5018                 btrfs_wait_for_snapshot_creation(root);
5019                 ret = btrfs_cont_expand(inode, oldsize, newsize);
5020                 if (ret) {
5021                         btrfs_end_write_no_snapshotting(root);
5022                         return ret;
5023                 }
5024
5025                 trans = btrfs_start_transaction(root, 1);
5026                 if (IS_ERR(trans)) {
5027                         btrfs_end_write_no_snapshotting(root);
5028                         return PTR_ERR(trans);
5029                 }
5030
5031                 i_size_write(inode, newsize);
5032                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
5033                 pagecache_isize_extended(inode, oldsize, newsize);
5034                 ret = btrfs_update_inode(trans, root, inode);
5035                 btrfs_end_write_no_snapshotting(root);
5036                 btrfs_end_transaction(trans);
5037         } else {
5038
5039                 /*
5040                  * We're truncating a file that used to have good data down to
5041                  * zero. Make sure it gets into the ordered flush list so that
5042                  * any new writes get down to disk quickly.
5043                  */
5044                 if (newsize == 0)
5045                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
5046                                 &BTRFS_I(inode)->runtime_flags);
5047
5048                 /*
5049                  * 1 for the orphan item we're going to add
5050                  * 1 for the orphan item deletion.
5051                  */
5052                 trans = btrfs_start_transaction(root, 2);
5053                 if (IS_ERR(trans))
5054                         return PTR_ERR(trans);
5055
5056                 /*
5057                  * We need to do this in case we fail at _any_ point during the
5058                  * actual truncate.  Once we do the truncate_setsize we could
5059                  * invalidate pages which forces any outstanding ordered io to
5060                  * be instantly completed which will give us extents that need
5061                  * to be truncated.  If we fail to get an orphan inode down we
5062                  * could have left over extents that were never meant to live,
5063                  * so we need to guarantee from this point on that everything
5064                  * will be consistent.
5065                  */
5066                 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
5067                 btrfs_end_transaction(trans);
5068                 if (ret)
5069                         return ret;
5070
5071                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
5072                 truncate_setsize(inode, newsize);
5073
5074                 /* Disable nonlocked read DIO to avoid the end less truncate */
5075                 btrfs_inode_block_unlocked_dio(BTRFS_I(inode));
5076                 inode_dio_wait(inode);
5077                 btrfs_inode_resume_unlocked_dio(BTRFS_I(inode));
5078
5079                 ret = btrfs_truncate(inode);
5080                 if (ret && inode->i_nlink) {
5081                         int err;
5082
5083                         /* To get a stable disk_i_size */
5084                         err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5085                         if (err) {
5086                                 btrfs_orphan_del(NULL, BTRFS_I(inode));
5087                                 return err;
5088                         }
5089
5090                         /*
5091                          * failed to truncate, disk_i_size is only adjusted down
5092                          * as we remove extents, so it should represent the true
5093                          * size of the inode, so reset the in memory size and
5094                          * delete our orphan entry.
5095                          */
5096                         trans = btrfs_join_transaction(root);
5097                         if (IS_ERR(trans)) {
5098                                 btrfs_orphan_del(NULL, BTRFS_I(inode));
5099                                 return ret;
5100                         }
5101                         i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5102                         err = btrfs_orphan_del(trans, BTRFS_I(inode));
5103                         if (err)
5104                                 btrfs_abort_transaction(trans, err);
5105                         btrfs_end_transaction(trans);
5106                 }
5107         }
5108
5109         return ret;
5110 }
5111
5112 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5113 {
5114         struct inode *inode = d_inode(dentry);
5115         struct btrfs_root *root = BTRFS_I(inode)->root;
5116         int err;
5117
5118         if (btrfs_root_readonly(root))
5119                 return -EROFS;
5120
5121         err = setattr_prepare(dentry, attr);
5122         if (err)
5123                 return err;
5124
5125         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5126                 err = btrfs_setsize(inode, attr);
5127                 if (err)
5128                         return err;
5129         }
5130
5131         if (attr->ia_valid) {
5132                 setattr_copy(inode, attr);
5133                 inode_inc_iversion(inode);
5134                 err = btrfs_dirty_inode(inode);
5135
5136                 if (!err && attr->ia_valid & ATTR_MODE)
5137                         err = posix_acl_chmod(inode, inode->i_mode);
5138         }
5139
5140         return err;
5141 }
5142
5143 /*
5144  * While truncating the inode pages during eviction, we get the VFS calling
5145  * btrfs_invalidatepage() against each page of the inode. This is slow because
5146  * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5147  * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5148  * extent_state structures over and over, wasting lots of time.
5149  *
5150  * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5151  * those expensive operations on a per page basis and do only the ordered io
5152  * finishing, while we release here the extent_map and extent_state structures,
5153  * without the excessive merging and splitting.
5154  */
5155 static void evict_inode_truncate_pages(struct inode *inode)
5156 {
5157         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5158         struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5159         struct rb_node *node;
5160
5161         ASSERT(inode->i_state & I_FREEING);
5162         truncate_inode_pages_final(&inode->i_data);
5163
5164         write_lock(&map_tree->lock);
5165         while (!RB_EMPTY_ROOT(&map_tree->map)) {
5166                 struct extent_map *em;
5167
5168                 node = rb_first(&map_tree->map);
5169                 em = rb_entry(node, struct extent_map, rb_node);
5170                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5171                 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5172                 remove_extent_mapping(map_tree, em);
5173                 free_extent_map(em);
5174                 if (need_resched()) {
5175                         write_unlock(&map_tree->lock);
5176                         cond_resched();
5177                         write_lock(&map_tree->lock);
5178                 }
5179         }
5180         write_unlock(&map_tree->lock);
5181
5182         /*
5183          * Keep looping until we have no more ranges in the io tree.
5184          * We can have ongoing bios started by readpages (called from readahead)
5185          * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5186          * still in progress (unlocked the pages in the bio but did not yet
5187          * unlocked the ranges in the io tree). Therefore this means some
5188          * ranges can still be locked and eviction started because before
5189          * submitting those bios, which are executed by a separate task (work
5190          * queue kthread), inode references (inode->i_count) were not taken
5191          * (which would be dropped in the end io callback of each bio).
5192          * Therefore here we effectively end up waiting for those bios and
5193          * anyone else holding locked ranges without having bumped the inode's
5194          * reference count - if we don't do it, when they access the inode's
5195          * io_tree to unlock a range it may be too late, leading to an
5196          * use-after-free issue.
5197          */
5198         spin_lock(&io_tree->lock);
5199         while (!RB_EMPTY_ROOT(&io_tree->state)) {
5200                 struct extent_state *state;
5201                 struct extent_state *cached_state = NULL;
5202                 u64 start;
5203                 u64 end;
5204
5205                 node = rb_first(&io_tree->state);
5206                 state = rb_entry(node, struct extent_state, rb_node);
5207                 start = state->start;
5208                 end = state->end;
5209                 spin_unlock(&io_tree->lock);
5210
5211                 lock_extent_bits(io_tree, start, end, &cached_state);
5212
5213                 /*
5214                  * If still has DELALLOC flag, the extent didn't reach disk,
5215                  * and its reserved space won't be freed by delayed_ref.
5216                  * So we need to free its reserved space here.
5217                  * (Refer to comment in btrfs_invalidatepage, case 2)
5218                  *
5219                  * Note, end is the bytenr of last byte, so we need + 1 here.
5220                  */
5221                 if (state->state & EXTENT_DELALLOC)
5222                         btrfs_qgroup_free_data(inode, NULL, start, end - start + 1);
5223
5224                 clear_extent_bit(io_tree, start, end,
5225                                  EXTENT_LOCKED | EXTENT_DIRTY |
5226                                  EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5227                                  EXTENT_DEFRAG, 1, 1,
5228                                  &cached_state, GFP_NOFS);
5229
5230                 cond_resched();
5231                 spin_lock(&io_tree->lock);
5232         }
5233         spin_unlock(&io_tree->lock);
5234 }
5235
5236 void btrfs_evict_inode(struct inode *inode)
5237 {
5238         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5239         struct btrfs_trans_handle *trans;
5240         struct btrfs_root *root = BTRFS_I(inode)->root;
5241         struct btrfs_block_rsv *rsv, *global_rsv;
5242         int steal_from_global = 0;
5243         u64 min_size;
5244         int ret;
5245
5246         trace_btrfs_inode_evict(inode);
5247
5248         if (!root) {
5249                 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
5250                 return;
5251         }
5252
5253         min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
5254
5255         evict_inode_truncate_pages(inode);
5256
5257         if (inode->i_nlink &&
5258             ((btrfs_root_refs(&root->root_item) != 0 &&
5259               root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5260              btrfs_is_free_space_inode(BTRFS_I(inode))))
5261                 goto no_delete;
5262
5263         if (is_bad_inode(inode)) {
5264                 btrfs_orphan_del(NULL, BTRFS_I(inode));
5265                 goto no_delete;
5266         }
5267         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5268         if (!special_file(inode->i_mode))
5269                 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5270
5271         btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5272
5273         if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
5274                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5275                                  &BTRFS_I(inode)->runtime_flags));
5276                 goto no_delete;
5277         }
5278
5279         if (inode->i_nlink > 0) {
5280                 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5281                        root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5282                 goto no_delete;
5283         }
5284
5285         ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5286         if (ret) {
5287                 btrfs_orphan_del(NULL, BTRFS_I(inode));
5288                 goto no_delete;
5289         }
5290
5291         rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5292         if (!rsv) {
5293                 btrfs_orphan_del(NULL, BTRFS_I(inode));
5294                 goto no_delete;
5295         }
5296         rsv->size = min_size;
5297         rsv->failfast = 1;
5298         global_rsv = &fs_info->global_block_rsv;
5299
5300         btrfs_i_size_write(BTRFS_I(inode), 0);
5301
5302         /*
5303          * This is a bit simpler than btrfs_truncate since we've already
5304          * reserved our space for our orphan item in the unlink, so we just
5305          * need to reserve some slack space in case we add bytes and update
5306          * inode item when doing the truncate.
5307          */
5308         while (1) {
5309                 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5310                                              BTRFS_RESERVE_FLUSH_LIMIT);
5311
5312                 /*
5313                  * Try and steal from the global reserve since we will
5314                  * likely not use this space anyway, we want to try as
5315                  * hard as possible to get this to work.
5316                  */
5317                 if (ret)
5318                         steal_from_global++;
5319                 else
5320                         steal_from_global = 0;
5321                 ret = 0;
5322
5323                 /*
5324                  * steal_from_global == 0: we reserved stuff, hooray!
5325                  * steal_from_global == 1: we didn't reserve stuff, boo!
5326                  * steal_from_global == 2: we've committed, still not a lot of
5327                  * room but maybe we'll have room in the global reserve this
5328                  * time.
5329                  * steal_from_global == 3: abandon all hope!
5330                  */
5331                 if (steal_from_global > 2) {
5332                         btrfs_warn(fs_info,
5333                                    "Could not get space for a delete, will truncate on mount %d",
5334                                    ret);
5335                         btrfs_orphan_del(NULL, BTRFS_I(inode));
5336                         btrfs_free_block_rsv(fs_info, rsv);
5337                         goto no_delete;
5338                 }
5339
5340                 trans = btrfs_join_transaction(root);
5341                 if (IS_ERR(trans)) {
5342                         btrfs_orphan_del(NULL, BTRFS_I(inode));
5343                         btrfs_free_block_rsv(fs_info, rsv);
5344                         goto no_delete;
5345                 }
5346
5347                 /*
5348                  * We can't just steal from the global reserve, we need to make
5349                  * sure there is room to do it, if not we need to commit and try
5350                  * again.
5351                  */
5352                 if (steal_from_global) {
5353                         if (!btrfs_check_space_for_delayed_refs(trans, fs_info))
5354                                 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5355                                                               min_size, 0);
5356                         else
5357                                 ret = -ENOSPC;
5358                 }
5359
5360                 /*
5361                  * Couldn't steal from the global reserve, we have too much
5362                  * pending stuff built up, commit the transaction and try it
5363                  * again.
5364                  */
5365                 if (ret) {
5366                         ret = btrfs_commit_transaction(trans);
5367                         if (ret) {
5368                                 btrfs_orphan_del(NULL, BTRFS_I(inode));
5369                                 btrfs_free_block_rsv(fs_info, rsv);
5370                                 goto no_delete;
5371                         }
5372                         continue;
5373                 } else {
5374                         steal_from_global = 0;
5375                 }
5376
5377                 trans->block_rsv = rsv;
5378
5379                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5380                 if (ret != -ENOSPC && ret != -EAGAIN)
5381                         break;
5382
5383                 trans->block_rsv = &fs_info->trans_block_rsv;
5384                 btrfs_end_transaction(trans);
5385                 trans = NULL;
5386                 btrfs_btree_balance_dirty(fs_info);
5387         }
5388
5389         btrfs_free_block_rsv(fs_info, rsv);
5390
5391         /*
5392          * Errors here aren't a big deal, it just means we leave orphan items
5393          * in the tree.  They will be cleaned up on the next mount.
5394          */
5395         if (ret == 0) {
5396                 trans->block_rsv = root->orphan_block_rsv;
5397                 btrfs_orphan_del(trans, BTRFS_I(inode));
5398         } else {
5399                 btrfs_orphan_del(NULL, BTRFS_I(inode));
5400         }
5401
5402         trans->block_rsv = &fs_info->trans_block_rsv;
5403         if (!(root == fs_info->tree_root ||
5404               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5405                 btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
5406
5407         btrfs_end_transaction(trans);
5408         btrfs_btree_balance_dirty(fs_info);
5409 no_delete:
5410         btrfs_remove_delayed_node(BTRFS_I(inode));
5411         clear_inode(inode);
5412 }
5413
5414 /*
5415  * this returns the key found in the dir entry in the location pointer.
5416  * If no dir entries were found, location->objectid is 0.
5417  */
5418 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5419                                struct btrfs_key *location)
5420 {
5421         const char *name = dentry->d_name.name;
5422         int namelen = dentry->d_name.len;
5423         struct btrfs_dir_item *di;
5424         struct btrfs_path *path;
5425         struct btrfs_root *root = BTRFS_I(dir)->root;
5426         int ret = 0;
5427
5428         path = btrfs_alloc_path();
5429         if (!path)
5430                 return -ENOMEM;
5431
5432         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5433                         name, namelen, 0);
5434         if (IS_ERR(di))
5435                 ret = PTR_ERR(di);
5436
5437         if (IS_ERR_OR_NULL(di))
5438                 goto out_err;
5439
5440         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5441 out:
5442         btrfs_free_path(path);
5443         return ret;
5444 out_err:
5445         location->objectid = 0;
5446         goto out;
5447 }
5448
5449 /*
5450  * when we hit a tree root in a directory, the btrfs part of the inode
5451  * needs to be changed to reflect the root directory of the tree root.  This
5452  * is kind of like crossing a mount point.
5453  */
5454 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5455                                     struct inode *dir,
5456                                     struct dentry *dentry,
5457                                     struct btrfs_key *location,
5458                                     struct btrfs_root **sub_root)
5459 {
5460         struct btrfs_path *path;
5461         struct btrfs_root *new_root;
5462         struct btrfs_root_ref *ref;
5463         struct extent_buffer *leaf;
5464         struct btrfs_key key;
5465         int ret;
5466         int err = 0;
5467
5468         path = btrfs_alloc_path();
5469         if (!path) {
5470                 err = -ENOMEM;
5471                 goto out;
5472         }
5473
5474         err = -ENOENT;
5475         key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5476         key.type = BTRFS_ROOT_REF_KEY;
5477         key.offset = location->objectid;
5478
5479         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5480         if (ret) {
5481                 if (ret < 0)
5482                         err = ret;
5483                 goto out;
5484         }
5485
5486         leaf = path->nodes[0];
5487         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5488         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5489             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5490                 goto out;
5491
5492         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5493                                    (unsigned long)(ref + 1),
5494                                    dentry->d_name.len);
5495         if (ret)
5496                 goto out;
5497
5498         btrfs_release_path(path);
5499
5500         new_root = btrfs_read_fs_root_no_name(fs_info, location);
5501         if (IS_ERR(new_root)) {
5502                 err = PTR_ERR(new_root);
5503                 goto out;
5504         }
5505
5506         *sub_root = new_root;
5507         location->objectid = btrfs_root_dirid(&new_root->root_item);
5508         location->type = BTRFS_INODE_ITEM_KEY;
5509         location->offset = 0;
5510         err = 0;
5511 out:
5512         btrfs_free_path(path);
5513         return err;
5514 }
5515
5516 static void inode_tree_add(struct inode *inode)
5517 {
5518         struct btrfs_root *root = BTRFS_I(inode)->root;
5519         struct btrfs_inode *entry;
5520         struct rb_node **p;
5521         struct rb_node *parent;
5522         struct rb_node *new = &BTRFS_I(inode)->rb_node;
5523         u64 ino = btrfs_ino(BTRFS_I(inode));
5524
5525         if (inode_unhashed(inode))
5526                 return;
5527         parent = NULL;
5528         spin_lock(&root->inode_lock);
5529         p = &root->inode_tree.rb_node;
5530         while (*p) {
5531                 parent = *p;
5532                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5533
5534                 if (ino < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5535                         p = &parent->rb_left;
5536                 else if (ino > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5537                         p = &parent->rb_right;
5538                 else {
5539                         WARN_ON(!(entry->vfs_inode.i_state &
5540                                   (I_WILL_FREE | I_FREEING)));
5541                         rb_replace_node(parent, new, &root->inode_tree);
5542                         RB_CLEAR_NODE(parent);
5543                         spin_unlock(&root->inode_lock);
5544                         return;
5545                 }
5546         }
5547         rb_link_node(new, parent, p);
5548         rb_insert_color(new, &root->inode_tree);
5549         spin_unlock(&root->inode_lock);
5550 }
5551
5552 static void inode_tree_del(struct inode *inode)
5553 {
5554         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5555         struct btrfs_root *root = BTRFS_I(inode)->root;
5556         int empty = 0;
5557
5558         spin_lock(&root->inode_lock);
5559         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5560                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5561                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5562                 empty = RB_EMPTY_ROOT(&root->inode_tree);
5563         }
5564         spin_unlock(&root->inode_lock);
5565
5566         if (empty && btrfs_root_refs(&root->root_item) == 0) {
5567                 synchronize_srcu(&fs_info->subvol_srcu);
5568                 spin_lock(&root->inode_lock);
5569                 empty = RB_EMPTY_ROOT(&root->inode_tree);
5570                 spin_unlock(&root->inode_lock);
5571                 if (empty)
5572                         btrfs_add_dead_root(root);
5573         }
5574 }
5575
5576 void btrfs_invalidate_inodes(struct btrfs_root *root)
5577 {
5578         struct btrfs_fs_info *fs_info = root->fs_info;
5579         struct rb_node *node;
5580         struct rb_node *prev;
5581         struct btrfs_inode *entry;
5582         struct inode *inode;
5583         u64 objectid = 0;
5584
5585         if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
5586                 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5587
5588         spin_lock(&root->inode_lock);
5589 again:
5590         node = root->inode_tree.rb_node;
5591         prev = NULL;
5592         while (node) {
5593                 prev = node;
5594                 entry = rb_entry(node, struct btrfs_inode, rb_node);
5595
5596                 if (objectid < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5597                         node = node->rb_left;
5598                 else if (objectid > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5599                         node = node->rb_right;
5600                 else
5601                         break;
5602         }
5603         if (!node) {
5604                 while (prev) {
5605                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
5606                         if (objectid <= btrfs_ino(BTRFS_I(&entry->vfs_inode))) {
5607                                 node = prev;
5608                                 break;
5609                         }
5610                         prev = rb_next(prev);
5611                 }
5612         }
5613         while (node) {
5614                 entry = rb_entry(node, struct btrfs_inode, rb_node);
5615                 objectid = btrfs_ino(BTRFS_I(&entry->vfs_inode)) + 1;
5616                 inode = igrab(&entry->vfs_inode);
5617                 if (inode) {
5618                         spin_unlock(&root->inode_lock);
5619                         if (atomic_read(&inode->i_count) > 1)
5620                                 d_prune_aliases(inode);
5621                         /*
5622                          * btrfs_drop_inode will have it removed from
5623                          * the inode cache when its usage count
5624                          * hits zero.
5625                          */
5626                         iput(inode);
5627                         cond_resched();
5628                         spin_lock(&root->inode_lock);
5629                         goto again;
5630                 }
5631
5632                 if (cond_resched_lock(&root->inode_lock))
5633                         goto again;
5634
5635                 node = rb_next(node);
5636         }
5637         spin_unlock(&root->inode_lock);
5638 }
5639
5640 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5641 {
5642         struct btrfs_iget_args *args = p;
5643         inode->i_ino = args->location->objectid;
5644         memcpy(&BTRFS_I(inode)->location, args->location,
5645                sizeof(*args->location));
5646         BTRFS_I(inode)->root = args->root;
5647         return 0;
5648 }
5649
5650 static int btrfs_find_actor(struct inode *inode, void *opaque)
5651 {
5652         struct btrfs_iget_args *args = opaque;
5653         return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5654                 args->root == BTRFS_I(inode)->root;
5655 }
5656
5657 static struct inode *btrfs_iget_locked(struct super_block *s,
5658                                        struct btrfs_key *location,
5659                                        struct btrfs_root *root)
5660 {
5661         struct inode *inode;
5662         struct btrfs_iget_args args;
5663         unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5664
5665         args.location = location;
5666         args.root = root;
5667
5668         inode = iget5_locked(s, hashval, btrfs_find_actor,
5669                              btrfs_init_locked_inode,
5670                              (void *)&args);
5671         return inode;
5672 }
5673
5674 /* Get an inode object given its location and corresponding root.
5675  * Returns in *is_new if the inode was read from disk
5676  */
5677 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5678                          struct btrfs_root *root, int *new)
5679 {
5680         struct inode *inode;
5681
5682         inode = btrfs_iget_locked(s, location, root);
5683         if (!inode)
5684                 return ERR_PTR(-ENOMEM);
5685
5686         if (inode->i_state & I_NEW) {
5687                 int ret;
5688
5689                 ret = btrfs_read_locked_inode(inode);
5690                 if (!is_bad_inode(inode)) {
5691                         inode_tree_add(inode);
5692                         unlock_new_inode(inode);
5693                         if (new)
5694                                 *new = 1;
5695                 } else {
5696                         unlock_new_inode(inode);
5697                         iput(inode);
5698                         ASSERT(ret < 0);
5699                         inode = ERR_PTR(ret < 0 ? ret : -ESTALE);
5700                 }
5701         }
5702
5703         return inode;
5704 }
5705
5706 static struct inode *new_simple_dir(struct super_block *s,
5707                                     struct btrfs_key *key,
5708                                     struct btrfs_root *root)
5709 {
5710         struct inode *inode = new_inode(s);
5711
5712         if (!inode)
5713                 return ERR_PTR(-ENOMEM);
5714
5715         BTRFS_I(inode)->root = root;
5716         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5717         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5718
5719         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5720         inode->i_op = &btrfs_dir_ro_inode_operations;
5721         inode->i_opflags &= ~IOP_XATTR;
5722         inode->i_fop = &simple_dir_operations;
5723         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5724         inode->i_mtime = current_time(inode);
5725         inode->i_atime = inode->i_mtime;
5726         inode->i_ctime = inode->i_mtime;
5727         BTRFS_I(inode)->i_otime = inode->i_mtime;
5728
5729         return inode;
5730 }
5731
5732 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5733 {
5734         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5735         struct inode *inode;
5736         struct btrfs_root *root = BTRFS_I(dir)->root;
5737         struct btrfs_root *sub_root = root;
5738         struct btrfs_key location;
5739         int index;
5740         int ret = 0;
5741
5742         if (dentry->d_name.len > BTRFS_NAME_LEN)
5743                 return ERR_PTR(-ENAMETOOLONG);
5744
5745         ret = btrfs_inode_by_name(dir, dentry, &location);
5746         if (ret < 0)
5747                 return ERR_PTR(ret);
5748
5749         if (location.objectid == 0)
5750                 return ERR_PTR(-ENOENT);
5751
5752         if (location.type == BTRFS_INODE_ITEM_KEY) {
5753                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5754                 return inode;
5755         }
5756
5757         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5758
5759         index = srcu_read_lock(&fs_info->subvol_srcu);
5760         ret = fixup_tree_root_location(fs_info, dir, dentry,
5761                                        &location, &sub_root);
5762         if (ret < 0) {
5763                 if (ret != -ENOENT)
5764                         inode = ERR_PTR(ret);
5765                 else
5766                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
5767         } else {
5768                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5769         }
5770         srcu_read_unlock(&fs_info->subvol_srcu, index);
5771
5772         if (!IS_ERR(inode) && root != sub_root) {
5773                 down_read(&fs_info->cleanup_work_sem);
5774                 if (!sb_rdonly(inode->i_sb))
5775                         ret = btrfs_orphan_cleanup(sub_root);
5776                 up_read(&fs_info->cleanup_work_sem);
5777                 if (ret) {
5778                         iput(inode);
5779                         inode = ERR_PTR(ret);
5780                 }
5781         }
5782
5783         return inode;
5784 }
5785
5786 static int btrfs_dentry_delete(const struct dentry *dentry)
5787 {
5788         struct btrfs_root *root;
5789         struct inode *inode = d_inode(dentry);
5790
5791         if (!inode && !IS_ROOT(dentry))
5792                 inode = d_inode(dentry->d_parent);
5793
5794         if (inode) {
5795                 root = BTRFS_I(inode)->root;
5796                 if (btrfs_root_refs(&root->root_item) == 0)
5797                         return 1;
5798
5799                 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5800                         return 1;
5801         }
5802         return 0;
5803 }
5804
5805 static void btrfs_dentry_release(struct dentry *dentry)
5806 {
5807         kfree(dentry->d_fsdata);
5808 }
5809
5810 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5811                                    unsigned int flags)
5812 {
5813         struct inode *inode;
5814
5815         inode = btrfs_lookup_dentry(dir, dentry);
5816         if (IS_ERR(inode)) {
5817                 if (PTR_ERR(inode) == -ENOENT)
5818                         inode = NULL;
5819                 else
5820                         return ERR_CAST(inode);
5821         }
5822
5823         return d_splice_alias(inode, dentry);
5824 }
5825
5826 unsigned char btrfs_filetype_table[] = {
5827         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5828 };
5829
5830 /*
5831  * All this infrastructure exists because dir_emit can fault, and we are holding
5832  * the tree lock when doing readdir.  For now just allocate a buffer and copy
5833  * our information into that, and then dir_emit from the buffer.  This is
5834  * similar to what NFS does, only we don't keep the buffer around in pagecache
5835  * because I'm afraid I'll mess that up.  Long term we need to make filldir do
5836  * copy_to_user_inatomic so we don't have to worry about page faulting under the
5837  * tree lock.
5838  */
5839 static int btrfs_opendir(struct inode *inode, struct file *file)
5840 {
5841         struct btrfs_file_private *private;
5842
5843         private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5844         if (!private)
5845                 return -ENOMEM;
5846         private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5847         if (!private->filldir_buf) {
5848                 kfree(private);
5849                 return -ENOMEM;
5850         }
5851         file->private_data = private;
5852         return 0;
5853 }
5854
5855 struct dir_entry {
5856         u64 ino;
5857         u64 offset;
5858         unsigned type;
5859         int name_len;
5860 };
5861
5862 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5863 {
5864         while (entries--) {
5865                 struct dir_entry *entry = addr;
5866                 char *name = (char *)(entry + 1);
5867
5868                 ctx->pos = entry->offset;
5869                 if (!dir_emit(ctx, name, entry->name_len, entry->ino,
5870                               entry->type))
5871                         return 1;
5872                 addr += sizeof(struct dir_entry) + entry->name_len;
5873                 ctx->pos++;
5874         }
5875         return 0;
5876 }
5877
5878 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5879 {
5880         struct inode *inode = file_inode(file);
5881         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5882         struct btrfs_root *root = BTRFS_I(inode)->root;
5883         struct btrfs_file_private *private = file->private_data;
5884         struct btrfs_dir_item *di;
5885         struct btrfs_key key;
5886         struct btrfs_key found_key;
5887         struct btrfs_path *path;
5888         void *addr;
5889         struct list_head ins_list;
5890         struct list_head del_list;
5891         int ret;
5892         struct extent_buffer *leaf;
5893         int slot;
5894         char *name_ptr;
5895         int name_len;
5896         int entries = 0;
5897         int total_len = 0;
5898         bool put = false;
5899         struct btrfs_key location;
5900
5901         if (!dir_emit_dots(file, ctx))
5902                 return 0;
5903
5904         path = btrfs_alloc_path();
5905         if (!path)
5906                 return -ENOMEM;
5907
5908         addr = private->filldir_buf;
5909         path->reada = READA_FORWARD;
5910
5911         INIT_LIST_HEAD(&ins_list);
5912         INIT_LIST_HEAD(&del_list);
5913         put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5914
5915 again:
5916         key.type = BTRFS_DIR_INDEX_KEY;
5917         key.offset = ctx->pos;
5918         key.objectid = btrfs_ino(BTRFS_I(inode));
5919
5920         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5921         if (ret < 0)
5922                 goto err;
5923
5924         while (1) {
5925                 struct dir_entry *entry;
5926
5927                 leaf = path->nodes[0];
5928                 slot = path->slots[0];
5929                 if (slot >= btrfs_header_nritems(leaf)) {
5930                         ret = btrfs_next_leaf(root, path);
5931                         if (ret < 0)
5932                                 goto err;
5933                         else if (ret > 0)
5934                                 break;
5935                         continue;
5936                 }
5937
5938                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5939
5940                 if (found_key.objectid != key.objectid)
5941                         break;
5942                 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5943                         break;
5944                 if (found_key.offset < ctx->pos)
5945                         goto next;
5946                 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5947                         goto next;
5948                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5949                 if (verify_dir_item(fs_info, leaf, slot, di))
5950                         goto next;
5951
5952                 name_len = btrfs_dir_name_len(leaf, di);
5953                 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5954                     PAGE_SIZE) {
5955                         btrfs_release_path(path);
5956                         ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5957                         if (ret)
5958                                 goto nopos;
5959                         addr = private->filldir_buf;
5960                         entries = 0;
5961                         total_len = 0;
5962                         goto again;
5963                 }
5964
5965                 entry = addr;
5966                 entry->name_len = name_len;
5967                 name_ptr = (char *)(entry + 1);
5968                 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5969                                    name_len);
5970                 entry->type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5971                 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5972                 entry->ino = location.objectid;
5973                 entry->offset = found_key.offset;
5974                 entries++;
5975                 addr += sizeof(struct dir_entry) + name_len;
5976                 total_len += sizeof(struct dir_entry) + name_len;
5977 next:
5978                 path->slots[0]++;
5979         }
5980         btrfs_release_path(path);
5981
5982         ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5983         if (ret)
5984                 goto nopos;
5985
5986         ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5987         if (ret)
5988                 goto nopos;
5989
5990         /*
5991          * Stop new entries from being returned after we return the last
5992          * entry.
5993          *
5994          * New directory entries are assigned a strictly increasing
5995          * offset.  This means that new entries created during readdir
5996          * are *guaranteed* to be seen in the future by that readdir.
5997          * This has broken buggy programs which operate on names as
5998          * they're returned by readdir.  Until we re-use freed offsets
5999          * we have this hack to stop new entries from being returned
6000          * under the assumption that they'll never reach this huge
6001          * offset.
6002          *
6003          * This is being careful not to overflow 32bit loff_t unless the
6004          * last entry requires it because doing so has broken 32bit apps
6005          * in the past.
6006          */
6007         if (ctx->pos >= INT_MAX)
6008                 ctx->pos = LLONG_MAX;
6009         else
6010                 ctx->pos = INT_MAX;
6011 nopos:
6012         ret = 0;
6013 err:
6014         if (put)
6015                 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6016         btrfs_free_path(path);
6017         return ret;
6018 }
6019
6020 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
6021 {
6022         struct btrfs_root *root = BTRFS_I(inode)->root;
6023         struct btrfs_trans_handle *trans;
6024         int ret = 0;
6025         bool nolock = false;
6026
6027         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6028                 return 0;
6029
6030         if (btrfs_fs_closing(root->fs_info) &&
6031                         btrfs_is_free_space_inode(BTRFS_I(inode)))
6032                 nolock = true;
6033
6034         if (wbc->sync_mode == WB_SYNC_ALL) {
6035                 if (nolock)
6036                         trans = btrfs_join_transaction_nolock(root);
6037                 else
6038                         trans = btrfs_join_transaction(root);
6039                 if (IS_ERR(trans))
6040                         return PTR_ERR(trans);
6041                 ret = btrfs_commit_transaction(trans);
6042         }
6043         return ret;
6044 }
6045
6046 /*
6047  * This is somewhat expensive, updating the tree every time the
6048  * inode changes.  But, it is most likely to find the inode in cache.
6049  * FIXME, needs more benchmarking...there are no reasons other than performance
6050  * to keep or drop this code.
6051  */
6052 static int btrfs_dirty_inode(struct inode *inode)
6053 {
6054         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6055         struct btrfs_root *root = BTRFS_I(inode)->root;
6056         struct btrfs_trans_handle *trans;
6057         int ret;
6058
6059         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6060                 return 0;
6061
6062         trans = btrfs_join_transaction(root);
6063         if (IS_ERR(trans))
6064                 return PTR_ERR(trans);
6065
6066         ret = btrfs_update_inode(trans, root, inode);
6067         if (ret && ret == -ENOSPC) {
6068                 /* whoops, lets try again with the full transaction */
6069                 btrfs_end_transaction(trans);
6070                 trans = btrfs_start_transaction(root, 1);
6071                 if (IS_ERR(trans))
6072                         return PTR_ERR(trans);
6073
6074                 ret = btrfs_update_inode(trans, root, inode);
6075         }
6076         btrfs_end_transaction(trans);
6077         if (BTRFS_I(inode)->delayed_node)
6078                 btrfs_balance_delayed_items(fs_info);
6079
6080         return ret;
6081 }
6082
6083 /*
6084  * This is a copy of file_update_time.  We need this so we can return error on
6085  * ENOSPC for updating the inode in the case of file write and mmap writes.
6086  */
6087 static int btrfs_update_time(struct inode *inode, struct timespec *now,
6088                              int flags)
6089 {
6090         struct btrfs_root *root = BTRFS_I(inode)->root;
6091
6092         if (btrfs_root_readonly(root))
6093                 return -EROFS;
6094
6095         if (flags & S_VERSION)
6096                 inode_inc_iversion(inode);
6097         if (flags & S_CTIME)
6098                 inode->i_ctime = *now;
6099         if (flags & S_MTIME)
6100                 inode->i_mtime = *now;
6101         if (flags & S_ATIME)
6102                 inode->i_atime = *now;
6103         return btrfs_dirty_inode(inode);
6104 }
6105
6106 /*
6107  * find the highest existing sequence number in a directory
6108  * and then set the in-memory index_cnt variable to reflect
6109  * free sequence numbers
6110  */
6111 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6112 {
6113         struct btrfs_root *root = inode->root;
6114         struct btrfs_key key, found_key;
6115         struct btrfs_path *path;
6116         struct extent_buffer *leaf;
6117         int ret;
6118
6119         key.objectid = btrfs_ino(inode);
6120         key.type = BTRFS_DIR_INDEX_KEY;
6121         key.offset = (u64)-1;
6122
6123         path = btrfs_alloc_path();
6124         if (!path)
6125                 return -ENOMEM;
6126
6127         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6128         if (ret < 0)
6129                 goto out;
6130         /* FIXME: we should be able to handle this */
6131         if (ret == 0)
6132                 goto out;
6133         ret = 0;
6134
6135         /*
6136          * MAGIC NUMBER EXPLANATION:
6137          * since we search a directory based on f_pos we have to start at 2
6138          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6139          * else has to start at 2
6140          */
6141         if (path->slots[0] == 0) {
6142                 inode->index_cnt = 2;
6143                 goto out;
6144         }
6145
6146         path->slots[0]--;
6147
6148         leaf = path->nodes[0];
6149         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6150
6151         if (found_key.objectid != btrfs_ino(inode) ||
6152             found_key.type != BTRFS_DIR_INDEX_KEY) {
6153                 inode->index_cnt = 2;
6154                 goto out;
6155         }
6156
6157         inode->index_cnt = found_key.offset + 1;
6158 out:
6159         btrfs_free_path(path);
6160         return ret;
6161 }
6162
6163 /*
6164  * helper to find a free sequence number in a given directory.  This current
6165  * code is very simple, later versions will do smarter things in the btree
6166  */
6167 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6168 {
6169         int ret = 0;
6170
6171         if (dir->index_cnt == (u64)-1) {
6172                 ret = btrfs_inode_delayed_dir_index_count(dir);
6173                 if (ret) {
6174                         ret = btrfs_set_inode_index_count(dir);
6175                         if (ret)
6176                                 return ret;
6177                 }
6178         }
6179
6180         *index = dir->index_cnt;
6181         dir->index_cnt++;
6182
6183         return ret;
6184 }
6185
6186 static int btrfs_insert_inode_locked(struct inode *inode)
6187 {
6188         struct btrfs_iget_args args;
6189         args.location = &BTRFS_I(inode)->location;
6190         args.root = BTRFS_I(inode)->root;
6191
6192         return insert_inode_locked4(inode,
6193                    btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6194                    btrfs_find_actor, &args);
6195 }
6196
6197 /*
6198  * Inherit flags from the parent inode.
6199  *
6200  * Currently only the compression flags and the cow flags are inherited.
6201  */
6202 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6203 {
6204         unsigned int flags;
6205
6206         if (!dir)
6207                 return;
6208
6209         flags = BTRFS_I(dir)->flags;
6210
6211         if (flags & BTRFS_INODE_NOCOMPRESS) {
6212                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6213                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6214         } else if (flags & BTRFS_INODE_COMPRESS) {
6215                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6216                 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6217         }
6218
6219         if (flags & BTRFS_INODE_NODATACOW) {
6220                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6221                 if (S_ISREG(inode->i_mode))
6222                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6223         }
6224
6225         btrfs_update_iflags(inode);
6226 }
6227
6228 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6229                                      struct btrfs_root *root,
6230                                      struct inode *dir,
6231                                      const char *name, int name_len,
6232                                      u64 ref_objectid, u64 objectid,
6233                                      umode_t mode, u64 *index)
6234 {
6235         struct btrfs_fs_info *fs_info = root->fs_info;
6236         struct inode *inode;
6237         struct btrfs_inode_item *inode_item;
6238         struct btrfs_key *location;
6239         struct btrfs_path *path;
6240         struct btrfs_inode_ref *ref;
6241         struct btrfs_key key[2];
6242         u32 sizes[2];
6243         int nitems = name ? 2 : 1;
6244         unsigned long ptr;
6245         int ret;
6246
6247         path = btrfs_alloc_path();
6248         if (!path)
6249                 return ERR_PTR(-ENOMEM);
6250
6251         inode = new_inode(fs_info->sb);
6252         if (!inode) {
6253                 btrfs_free_path(path);
6254                 return ERR_PTR(-ENOMEM);
6255         }
6256
6257         /*
6258          * O_TMPFILE, set link count to 0, so that after this point,
6259          * we fill in an inode item with the correct link count.
6260          */
6261         if (!name)
6262                 set_nlink(inode, 0);
6263
6264         /*
6265          * we have to initialize this early, so we can reclaim the inode
6266          * number if we fail afterwards in this function.
6267          */
6268         inode->i_ino = objectid;
6269
6270         if (dir && name) {
6271                 trace_btrfs_inode_request(dir);
6272
6273                 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6274                 if (ret) {
6275                         btrfs_free_path(path);
6276                         iput(inode);
6277                         return ERR_PTR(ret);
6278                 }
6279         } else if (dir) {
6280                 *index = 0;
6281         }
6282         /*
6283          * index_cnt is ignored for everything but a dir,
6284          * btrfs_get_inode_index_count has an explanation for the magic
6285          * number
6286          */
6287         BTRFS_I(inode)->index_cnt = 2;
6288         BTRFS_I(inode)->dir_index = *index;
6289         BTRFS_I(inode)->root = root;
6290         BTRFS_I(inode)->generation = trans->transid;
6291         inode->i_generation = BTRFS_I(inode)->generation;
6292
6293         /*
6294          * We could have gotten an inode number from somebody who was fsynced
6295          * and then removed in this same transaction, so let's just set full
6296          * sync since it will be a full sync anyway and this will blow away the
6297          * old info in the log.
6298          */
6299         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6300
6301         key[0].objectid = objectid;
6302         key[0].type = BTRFS_INODE_ITEM_KEY;
6303         key[0].offset = 0;
6304
6305         sizes[0] = sizeof(struct btrfs_inode_item);
6306
6307         if (name) {
6308                 /*
6309                  * Start new inodes with an inode_ref. This is slightly more
6310                  * efficient for small numbers of hard links since they will
6311                  * be packed into one item. Extended refs will kick in if we
6312                  * add more hard links than can fit in the ref item.
6313                  */
6314                 key[1].objectid = objectid;
6315                 key[1].type = BTRFS_INODE_REF_KEY;
6316                 key[1].offset = ref_objectid;
6317
6318                 sizes[1] = name_len + sizeof(*ref);
6319         }
6320
6321         location = &BTRFS_I(inode)->location;
6322         location->objectid = objectid;
6323         location->offset = 0;
6324         location->type = BTRFS_INODE_ITEM_KEY;
6325
6326         ret = btrfs_insert_inode_locked(inode);
6327         if (ret < 0)
6328                 goto fail;
6329
6330         path->leave_spinning = 1;
6331         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6332         if (ret != 0)
6333                 goto fail_unlock;
6334
6335         inode_init_owner(inode, dir, mode);
6336         inode_set_bytes(inode, 0);
6337
6338         inode->i_mtime = current_time(inode);
6339         inode->i_atime = inode->i_mtime;
6340         inode->i_ctime = inode->i_mtime;
6341         BTRFS_I(inode)->i_otime = inode->i_mtime;
6342
6343         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6344                                   struct btrfs_inode_item);
6345         memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6346                              sizeof(*inode_item));
6347         fill_inode_item(trans, path->nodes[0], inode_item, inode);
6348
6349         if (name) {
6350                 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6351                                      struct btrfs_inode_ref);
6352                 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6353                 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6354                 ptr = (unsigned long)(ref + 1);
6355                 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6356         }
6357
6358         btrfs_mark_buffer_dirty(path->nodes[0]);
6359         btrfs_free_path(path);
6360
6361         btrfs_inherit_iflags(inode, dir);
6362
6363         if (S_ISREG(mode)) {
6364                 if (btrfs_test_opt(fs_info, NODATASUM))
6365                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6366                 if (btrfs_test_opt(fs_info, NODATACOW))
6367                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6368                                 BTRFS_INODE_NODATASUM;
6369         }
6370
6371         inode_tree_add(inode);
6372
6373         trace_btrfs_inode_new(inode);
6374         btrfs_set_inode_last_trans(trans, inode);
6375
6376         btrfs_update_root_times(trans, root);
6377
6378         ret = btrfs_inode_inherit_props(trans, inode, dir);
6379         if (ret)
6380                 btrfs_err(fs_info,
6381                           "error inheriting props for ino %llu (root %llu): %d",
6382                         btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6383
6384         return inode;
6385
6386 fail_unlock:
6387         unlock_new_inode(inode);
6388 fail:
6389         if (dir && name)
6390                 BTRFS_I(dir)->index_cnt--;
6391         btrfs_free_path(path);
6392         iput(inode);
6393         return ERR_PTR(ret);
6394 }
6395
6396 static inline u8 btrfs_inode_type(struct inode *inode)
6397 {
6398         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6399 }
6400
6401 /*
6402  * utility function to add 'inode' into 'parent_inode' with
6403  * a give name and a given sequence number.
6404  * if 'add_backref' is true, also insert a backref from the
6405  * inode to the parent directory.
6406  */
6407 int btrfs_add_link(struct btrfs_trans_handle *trans,
6408                    struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6409                    const char *name, int name_len, int add_backref, u64 index)
6410 {
6411         struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6412         int ret = 0;
6413         struct btrfs_key key;
6414         struct btrfs_root *root = parent_inode->root;
6415         u64 ino = btrfs_ino(inode);
6416         u64 parent_ino = btrfs_ino(parent_inode);
6417
6418         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6419                 memcpy(&key, &inode->root->root_key, sizeof(key));
6420         } else {
6421                 key.objectid = ino;
6422                 key.type = BTRFS_INODE_ITEM_KEY;
6423                 key.offset = 0;
6424         }
6425
6426         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6427                 ret = btrfs_add_root_ref(trans, fs_info, key.objectid,
6428                                          root->root_key.objectid, parent_ino,
6429                                          index, name, name_len);
6430         } else if (add_backref) {
6431                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6432                                              parent_ino, index);
6433         }
6434
6435         /* Nothing to clean up yet */
6436         if (ret)
6437                 return ret;
6438
6439         ret = btrfs_insert_dir_item(trans, root, name, name_len,
6440                                     parent_inode, &key,
6441                                     btrfs_inode_type(&inode->vfs_inode), index);
6442         if (ret == -EEXIST || ret == -EOVERFLOW)
6443                 goto fail_dir_item;
6444         else if (ret) {
6445                 btrfs_abort_transaction(trans, ret);
6446                 return ret;
6447         }
6448
6449         btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6450                            name_len * 2);
6451         inode_inc_iversion(&parent_inode->vfs_inode);
6452         parent_inode->vfs_inode.i_mtime = parent_inode->vfs_inode.i_ctime =
6453                 current_time(&parent_inode->vfs_inode);
6454         ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode);
6455         if (ret)
6456                 btrfs_abort_transaction(trans, ret);
6457         return ret;
6458
6459 fail_dir_item:
6460         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6461                 u64 local_index;
6462                 int err;
6463                 err = btrfs_del_root_ref(trans, fs_info, key.objectid,
6464                                          root->root_key.objectid, parent_ino,
6465                                          &local_index, name, name_len);
6466
6467         } else if (add_backref) {
6468                 u64 local_index;
6469                 int err;
6470
6471                 err = btrfs_del_inode_ref(trans, root, name, name_len,
6472                                           ino, parent_ino, &local_index);
6473         }
6474         return ret;
6475 }
6476
6477 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6478                             struct btrfs_inode *dir, struct dentry *dentry,
6479                             struct btrfs_inode *inode, int backref, u64 index)
6480 {
6481         int err = btrfs_add_link(trans, dir, inode,
6482                                  dentry->d_name.name, dentry->d_name.len,
6483                                  backref, index);
6484         if (err > 0)
6485                 err = -EEXIST;
6486         return err;
6487 }
6488
6489 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6490                         umode_t mode, dev_t rdev)
6491 {
6492         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6493         struct btrfs_trans_handle *trans;
6494         struct btrfs_root *root = BTRFS_I(dir)->root;
6495         struct inode *inode = NULL;
6496         int err;
6497         int drop_inode = 0;
6498         u64 objectid;
6499         u64 index = 0;
6500
6501         /*
6502          * 2 for inode item and ref
6503          * 2 for dir items
6504          * 1 for xattr if selinux is on
6505          */
6506         trans = btrfs_start_transaction(root, 5);
6507         if (IS_ERR(trans))
6508                 return PTR_ERR(trans);
6509
6510         err = btrfs_find_free_ino(root, &objectid);
6511         if (err)
6512                 goto out_unlock;
6513
6514         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6515                         dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6516                         mode, &index);
6517         if (IS_ERR(inode)) {
6518                 err = PTR_ERR(inode);
6519                 goto out_unlock;
6520         }
6521
6522         /*
6523         * If the active LSM wants to access the inode during
6524         * d_instantiate it needs these. Smack checks to see
6525         * if the filesystem supports xattrs by looking at the
6526         * ops vector.
6527         */
6528         inode->i_op = &btrfs_special_inode_operations;
6529         init_special_inode(inode, inode->i_mode, rdev);
6530
6531         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6532         if (err)
6533                 goto out_unlock_inode;
6534
6535         err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6536                         0, index);
6537         if (err) {
6538                 goto out_unlock_inode;
6539         } else {
6540                 btrfs_update_inode(trans, root, inode);
6541                 unlock_new_inode(inode);
6542                 d_instantiate(dentry, inode);
6543         }
6544
6545 out_unlock:
6546         btrfs_end_transaction(trans);
6547         btrfs_balance_delayed_items(fs_info);
6548         btrfs_btree_balance_dirty(fs_info);
6549         if (drop_inode) {
6550                 inode_dec_link_count(inode);
6551                 iput(inode);
6552         }
6553         return err;
6554
6555 out_unlock_inode:
6556         drop_inode = 1;
6557         unlock_new_inode(inode);
6558         goto out_unlock;
6559
6560 }
6561
6562 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6563                         umode_t mode, bool excl)
6564 {
6565         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6566         struct btrfs_trans_handle *trans;
6567         struct btrfs_root *root = BTRFS_I(dir)->root;
6568         struct inode *inode = NULL;
6569         int drop_inode_on_err = 0;
6570         int err;
6571         u64 objectid;
6572         u64 index = 0;
6573
6574         /*
6575          * 2 for inode item and ref
6576          * 2 for dir items
6577          * 1 for xattr if selinux is on
6578          */
6579         trans = btrfs_start_transaction(root, 5);
6580         if (IS_ERR(trans))
6581                 return PTR_ERR(trans);
6582
6583         err = btrfs_find_free_ino(root, &objectid);
6584         if (err)
6585                 goto out_unlock;
6586
6587         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6588                         dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6589                         mode, &index);
6590         if (IS_ERR(inode)) {
6591                 err = PTR_ERR(inode);
6592                 goto out_unlock;
6593         }
6594         drop_inode_on_err = 1;
6595         /*
6596         * If the active LSM wants to access the inode during
6597         * d_instantiate it needs these. Smack checks to see
6598         * if the filesystem supports xattrs by looking at the
6599         * ops vector.
6600         */
6601         inode->i_fop = &btrfs_file_operations;
6602         inode->i_op = &btrfs_file_inode_operations;
6603         inode->i_mapping->a_ops = &btrfs_aops;
6604
6605         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6606         if (err)
6607                 goto out_unlock_inode;
6608
6609         err = btrfs_update_inode(trans, root, inode);
6610         if (err)
6611                 goto out_unlock_inode;
6612
6613         err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6614                         0, index);
6615         if (err)
6616                 goto out_unlock_inode;
6617
6618         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6619         unlock_new_inode(inode);
6620         d_instantiate(dentry, inode);
6621
6622 out_unlock:
6623         btrfs_end_transaction(trans);
6624         if (err && drop_inode_on_err) {
6625                 inode_dec_link_count(inode);
6626                 iput(inode);
6627         }
6628         btrfs_balance_delayed_items(fs_info);
6629         btrfs_btree_balance_dirty(fs_info);
6630         return err;
6631
6632 out_unlock_inode:
6633         unlock_new_inode(inode);
6634         goto out_unlock;
6635
6636 }
6637
6638 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6639                       struct dentry *dentry)
6640 {
6641         struct btrfs_trans_handle *trans = NULL;
6642         struct btrfs_root *root = BTRFS_I(dir)->root;
6643         struct inode *inode = d_inode(old_dentry);
6644         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6645         u64 index;
6646         int err;
6647         int drop_inode = 0;
6648
6649         /* do not allow sys_link's with other subvols of the same device */
6650         if (root->objectid != BTRFS_I(inode)->root->objectid)
6651                 return -EXDEV;
6652
6653         if (inode->i_nlink >= BTRFS_LINK_MAX)
6654                 return -EMLINK;
6655
6656         err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6657         if (err)
6658                 goto fail;
6659
6660         /*
6661          * 2 items for inode and inode ref
6662          * 2 items for dir items
6663          * 1 item for parent inode
6664          */
6665         trans = btrfs_start_transaction(root, 5);
6666         if (IS_ERR(trans)) {
6667                 err = PTR_ERR(trans);
6668                 trans = NULL;
6669                 goto fail;
6670         }
6671
6672         /* There are several dir indexes for this inode, clear the cache. */
6673         BTRFS_I(inode)->dir_index = 0ULL;
6674         inc_nlink(inode);
6675         inode_inc_iversion(inode);
6676         inode->i_ctime = current_time(inode);
6677         ihold(inode);
6678         set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6679
6680         err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6681                         1, index);
6682
6683         if (err) {
6684                 drop_inode = 1;
6685         } else {
6686                 struct dentry *parent = dentry->d_parent;
6687                 err = btrfs_update_inode(trans, root, inode);
6688                 if (err)
6689                         goto fail;
6690                 if (inode->i_nlink == 1) {
6691                         /*
6692                          * If new hard link count is 1, it's a file created
6693                          * with open(2) O_TMPFILE flag.
6694                          */
6695                         err = btrfs_orphan_del(trans, BTRFS_I(inode));
6696                         if (err)
6697                                 goto fail;
6698                 }
6699                 d_instantiate(dentry, inode);
6700                 btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent);
6701         }
6702
6703         btrfs_balance_delayed_items(fs_info);
6704 fail:
6705         if (trans)
6706                 btrfs_end_transaction(trans);
6707         if (drop_inode) {
6708                 inode_dec_link_count(inode);
6709                 iput(inode);
6710         }
6711         btrfs_btree_balance_dirty(fs_info);
6712         return err;
6713 }
6714
6715 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6716 {
6717         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6718         struct inode *inode = NULL;
6719         struct btrfs_trans_handle *trans;
6720         struct btrfs_root *root = BTRFS_I(dir)->root;
6721         int err = 0;
6722         int drop_on_err = 0;
6723         u64 objectid = 0;
6724         u64 index = 0;
6725
6726         /*
6727          * 2 items for inode and ref
6728          * 2 items for dir items
6729          * 1 for xattr if selinux is on
6730          */
6731         trans = btrfs_start_transaction(root, 5);
6732         if (IS_ERR(trans))
6733                 return PTR_ERR(trans);
6734
6735         err = btrfs_find_free_ino(root, &objectid);
6736         if (err)
6737                 goto out_fail;
6738
6739         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6740                         dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6741                         S_IFDIR | mode, &index);
6742         if (IS_ERR(inode)) {
6743                 err = PTR_ERR(inode);
6744                 goto out_fail;
6745         }
6746
6747         drop_on_err = 1;
6748         /* these must be set before we unlock the inode */
6749         inode->i_op = &btrfs_dir_inode_operations;
6750         inode->i_fop = &btrfs_dir_file_operations;
6751
6752         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6753         if (err)
6754                 goto out_fail_inode;
6755
6756         btrfs_i_size_write(BTRFS_I(inode), 0);
6757         err = btrfs_update_inode(trans, root, inode);
6758         if (err)
6759                 goto out_fail_inode;
6760
6761         err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6762                         dentry->d_name.name,
6763                         dentry->d_name.len, 0, index);
6764         if (err)
6765                 goto out_fail_inode;
6766
6767         d_instantiate(dentry, inode);
6768         /*
6769          * mkdir is special.  We're unlocking after we call d_instantiate
6770          * to avoid a race with nfsd calling d_instantiate.
6771          */
6772         unlock_new_inode(inode);
6773         drop_on_err = 0;
6774
6775 out_fail:
6776         btrfs_end_transaction(trans);
6777         if (drop_on_err) {
6778                 inode_dec_link_count(inode);
6779                 iput(inode);
6780         }
6781         btrfs_balance_delayed_items(fs_info);
6782         btrfs_btree_balance_dirty(fs_info);
6783         return err;
6784
6785 out_fail_inode:
6786         unlock_new_inode(inode);
6787         goto out_fail;
6788 }
6789
6790 /* Find next extent map of a given extent map, caller needs to ensure locks */
6791 static struct extent_map *next_extent_map(struct extent_map *em)
6792 {
6793         struct rb_node *next;
6794
6795         next = rb_next(&em->rb_node);
6796         if (!next)
6797                 return NULL;
6798         return container_of(next, struct extent_map, rb_node);
6799 }
6800
6801 static struct extent_map *prev_extent_map(struct extent_map *em)
6802 {
6803         struct rb_node *prev;
6804
6805         prev = rb_prev(&em->rb_node);
6806         if (!prev)
6807                 return NULL;
6808         return container_of(prev, struct extent_map, rb_node);
6809 }
6810
6811 /* helper for btfs_get_extent.  Given an existing extent in the tree,
6812  * the existing extent is the nearest extent to map_start,
6813  * and an extent that you want to insert, deal with overlap and insert
6814  * the best fitted new extent into the tree.
6815  */
6816 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6817                                 struct extent_map *existing,
6818                                 struct extent_map *em,
6819                                 u64 map_start)
6820 {
6821         struct extent_map *prev;
6822         struct extent_map *next;
6823         u64 start;
6824         u64 end;
6825         u64 start_diff;
6826
6827         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6828
6829         if (existing->start > map_start) {
6830                 next = existing;
6831                 prev = prev_extent_map(next);
6832         } else {
6833                 prev = existing;
6834                 next = next_extent_map(prev);
6835         }
6836
6837         start = prev ? extent_map_end(prev) : em->start;
6838         start = max_t(u64, start, em->start);
6839         end = next ? next->start : extent_map_end(em);
6840         end = min_t(u64, end, extent_map_end(em));
6841         start_diff = start - em->start;
6842         em->start = start;
6843         em->len = end - start;
6844         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6845             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6846                 em->block_start += start_diff;
6847                 em->block_len -= start_diff;
6848         }
6849         return add_extent_mapping(em_tree, em, 0);
6850 }
6851
6852 static noinline int uncompress_inline(struct btrfs_path *path,
6853                                       struct page *page,
6854                                       size_t pg_offset, u64 extent_offset,
6855                                       struct btrfs_file_extent_item *item)
6856 {
6857         int ret;
6858         struct extent_buffer *leaf = path->nodes[0];
6859         char *tmp;
6860         size_t max_size;
6861         unsigned long inline_size;
6862         unsigned long ptr;
6863         int compress_type;
6864
6865         WARN_ON(pg_offset != 0);
6866         compress_type = btrfs_file_extent_compression(leaf, item);
6867         max_size = btrfs_file_extent_ram_bytes(leaf, item);
6868         inline_size = btrfs_file_extent_inline_item_len(leaf,
6869                                         btrfs_item_nr(path->slots[0]));
6870         tmp = kmalloc(inline_size, GFP_NOFS);
6871         if (!tmp)
6872                 return -ENOMEM;
6873         ptr = btrfs_file_extent_inline_start(item);
6874
6875         read_extent_buffer(leaf, tmp, ptr, inline_size);
6876
6877         max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6878         ret = btrfs_decompress(compress_type, tmp, page,
6879                                extent_offset, inline_size, max_size);
6880
6881         /*
6882          * decompression code contains a memset to fill in any space between the end
6883          * of the uncompressed data and the end of max_size in case the decompressed
6884          * data ends up shorter than ram_bytes.  That doesn't cover the hole between
6885          * the end of an inline extent and the beginning of the next block, so we
6886          * cover that region here.
6887          */
6888
6889         if (max_size + pg_offset < PAGE_SIZE) {
6890                 char *map = kmap(page);
6891                 memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6892                 kunmap(page);
6893         }
6894         kfree(tmp);
6895         return ret;
6896 }
6897
6898 /*
6899  * a bit scary, this does extent mapping from logical file offset to the disk.
6900  * the ugly parts come from merging extents from the disk with the in-ram
6901  * representation.  This gets more complex because of the data=ordered code,
6902  * where the in-ram extents might be locked pending data=ordered completion.
6903  *
6904  * This also copies inline extents directly into the page.
6905  */
6906 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6907                 struct page *page,
6908             size_t pg_offset, u64 start, u64 len,
6909                 int create)
6910 {
6911         struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6912         int ret;
6913         int err = 0;
6914         u64 extent_start = 0;
6915         u64 extent_end = 0;
6916         u64 objectid = btrfs_ino(inode);
6917         u32 found_type;
6918         struct btrfs_path *path = NULL;
6919         struct btrfs_root *root = inode->root;
6920         struct btrfs_file_extent_item *item;
6921         struct extent_buffer *leaf;
6922         struct btrfs_key found_key;
6923         struct extent_map *em = NULL;
6924         struct extent_map_tree *em_tree = &inode->extent_tree;
6925         struct extent_io_tree *io_tree = &inode->io_tree;
6926         struct btrfs_trans_handle *trans = NULL;
6927         const bool new_inline = !page || create;
6928
6929 again:
6930         read_lock(&em_tree->lock);
6931         em = lookup_extent_mapping(em_tree, start, len);
6932         if (em)
6933                 em->bdev = fs_info->fs_devices->latest_bdev;
6934         read_unlock(&em_tree->lock);
6935
6936         if (em) {
6937                 if (em->start > start || em->start + em->len <= start)
6938                         free_extent_map(em);
6939                 else if (em->block_start == EXTENT_MAP_INLINE && page)
6940                         free_extent_map(em);
6941                 else
6942                         goto out;
6943         }
6944         em = alloc_extent_map();
6945         if (!em) {
6946                 err = -ENOMEM;
6947                 goto out;
6948         }
6949         em->bdev = fs_info->fs_devices->latest_bdev;
6950         em->start = EXTENT_MAP_HOLE;
6951         em->orig_start = EXTENT_MAP_HOLE;
6952         em->len = (u64)-1;
6953         em->block_len = (u64)-1;
6954
6955         if (!path) {
6956                 path = btrfs_alloc_path();
6957                 if (!path) {
6958                         err = -ENOMEM;
6959                         goto out;
6960                 }
6961                 /*
6962                  * Chances are we'll be called again, so go ahead and do
6963                  * readahead
6964                  */
6965                 path->reada = READA_FORWARD;
6966         }
6967
6968         ret = btrfs_lookup_file_extent(trans, root, path,
6969                                        objectid, start, trans != NULL);
6970         if (ret < 0) {
6971                 err = ret;
6972                 goto out;
6973         }
6974
6975         if (ret != 0) {
6976                 if (path->slots[0] == 0)
6977                         goto not_found;
6978                 path->slots[0]--;
6979         }
6980
6981         leaf = path->nodes[0];
6982         item = btrfs_item_ptr(leaf, path->slots[0],
6983                               struct btrfs_file_extent_item);
6984         /* are we inside the extent that was found? */
6985         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6986         found_type = found_key.type;
6987         if (found_key.objectid != objectid ||
6988             found_type != BTRFS_EXTENT_DATA_KEY) {
6989                 /*
6990                  * If we backup past the first extent we want to move forward
6991                  * and see if there is an extent in front of us, otherwise we'll
6992                  * say there is a hole for our whole search range which can
6993                  * cause problems.
6994                  */
6995                 extent_end = start;
6996                 goto next;
6997         }
6998
6999         found_type = btrfs_file_extent_type(leaf, item);
7000         extent_start = found_key.offset;
7001         if (found_type == BTRFS_FILE_EXTENT_REG ||
7002             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7003                 extent_end = extent_start +
7004                        btrfs_file_extent_num_bytes(leaf, item);
7005
7006                 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
7007                                                        extent_start);
7008         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
7009                 size_t size;
7010                 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
7011                 extent_end = ALIGN(extent_start + size,
7012                                    fs_info->sectorsize);
7013
7014                 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
7015                                                       path->slots[0],
7016                                                       extent_start);
7017         }
7018 next:
7019         if (start >= extent_end) {
7020                 path->slots[0]++;
7021                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
7022                         ret = btrfs_next_leaf(root, path);
7023                         if (ret < 0) {
7024                                 err = ret;
7025                                 goto out;
7026                         }
7027                         if (ret > 0)
7028                                 goto not_found;
7029                         leaf = path->nodes[0];
7030                 }
7031                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7032                 if (found_key.objectid != objectid ||
7033                     found_key.type != BTRFS_EXTENT_DATA_KEY)
7034                         goto not_found;
7035                 if (start + len <= found_key.offset)
7036                         goto not_found;
7037                 if (start > found_key.offset)
7038                         goto next;
7039                 em->start = start;
7040                 em->orig_start = start;
7041                 em->len = found_key.offset - start;
7042                 goto not_found_em;
7043         }
7044
7045         btrfs_extent_item_to_extent_map(inode, path, item,
7046                         new_inline, em);
7047
7048         if (found_type == BTRFS_FILE_EXTENT_REG ||
7049             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7050                 goto insert;
7051         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
7052                 unsigned long ptr;
7053                 char *map;
7054                 size_t size;
7055                 size_t extent_offset;
7056                 size_t copy_size;
7057
7058                 if (new_inline)
7059                         goto out;
7060
7061                 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
7062                 extent_offset = page_offset(page) + pg_offset - extent_start;
7063                 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
7064                                   size - extent_offset);
7065                 em->start = extent_start + extent_offset;
7066                 em->len = ALIGN(copy_size, fs_info->sectorsize);
7067                 em->orig_block_len = em->len;
7068                 em->orig_start = em->start;
7069                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
7070                 if (create == 0 && !PageUptodate(page)) {
7071                         if (btrfs_file_extent_compression(leaf, item) !=
7072                             BTRFS_COMPRESS_NONE) {
7073                                 ret = uncompress_inline(path, page, pg_offset,
7074                                                         extent_offset, item);
7075                                 if (ret) {
7076                                         err = ret;
7077                                         goto out;
7078                                 }
7079                         } else {
7080                                 map = kmap(page);
7081                                 read_extent_buffer(leaf, map + pg_offset, ptr,
7082                                                    copy_size);
7083                                 if (pg_offset + copy_size < PAGE_SIZE) {
7084                                         memset(map + pg_offset + copy_size, 0,
7085                                                PAGE_SIZE - pg_offset -
7086                                                copy_size);
7087                                 }
7088                                 kunmap(page);
7089                         }
7090                         flush_dcache_page(page);
7091                 } else if (create && PageUptodate(page)) {
7092                         BUG();
7093                         if (!trans) {
7094                                 kunmap(page);
7095                                 free_extent_map(em);
7096                                 em = NULL;
7097
7098                                 btrfs_release_path(path);
7099                                 trans = btrfs_join_transaction(root);
7100
7101                                 if (IS_ERR(trans))
7102                                         return ERR_CAST(trans);
7103                                 goto again;
7104                         }
7105                         map = kmap(page);
7106                         write_extent_buffer(leaf, map + pg_offset, ptr,
7107                                             copy_size);
7108                         kunmap(page);
7109                         btrfs_mark_buffer_dirty(leaf);
7110                 }
7111                 set_extent_uptodate(io_tree, em->start,
7112                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
7113                 goto insert;
7114         }
7115 not_found:
7116         em->start = start;
7117         em->orig_start = start;
7118         em->len = len;
7119 not_found_em:
7120         em->block_start = EXTENT_MAP_HOLE;
7121         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
7122 insert:
7123         btrfs_release_path(path);
7124         if (em->start > start || extent_map_end(em) <= start) {
7125                 btrfs_err(fs_info,
7126                           "bad extent! em: [%llu %llu] passed [%llu %llu]",
7127                           em->start, em->len, start, len);
7128                 err = -EIO;
7129                 goto out;
7130         }
7131
7132         err = 0;
7133         write_lock(&em_tree->lock);
7134         ret = add_extent_mapping(em_tree, em, 0);
7135         /* it is possible that someone inserted the extent into the tree
7136          * while we had the lock dropped.  It is also possible that
7137          * an overlapping map exists in the tree
7138          */
7139         if (ret == -EEXIST) {
7140                 struct extent_map *existing;
7141
7142                 ret = 0;
7143
7144                 existing = search_extent_mapping(em_tree, start, len);
7145                 /*
7146                  * existing will always be non-NULL, since there must be
7147                  * extent causing the -EEXIST.
7148                  */
7149                 if (existing->start == em->start &&
7150                     extent_map_end(existing) >= extent_map_end(em) &&
7151                     em->block_start == existing->block_start) {
7152                         /*
7153                          * The existing extent map already encompasses the
7154                          * entire extent map we tried to add.
7155                          */
7156                         free_extent_map(em);
7157                         em = existing;
7158                         err = 0;
7159
7160                 } else if (start >= extent_map_end(existing) ||
7161                     start <= existing->start) {
7162                         /*
7163                          * The existing extent map is the one nearest to
7164                          * the [start, start + len) range which overlaps
7165                          */
7166                         err = merge_extent_mapping(em_tree, existing,
7167                                                    em, start);
7168                         free_extent_map(existing);
7169                         if (err) {
7170                                 free_extent_map(em);
7171                                 em = NULL;
7172                         }
7173                 } else {
7174                         free_extent_map(em);
7175                         em = existing;
7176                         err = 0;
7177                 }
7178         }
7179         write_unlock(&em_tree->lock);
7180 out:
7181
7182         trace_btrfs_get_extent(root, inode, em);
7183
7184         btrfs_free_path(path);
7185         if (trans) {
7186                 ret = btrfs_end_transaction(trans);
7187                 if (!err)
7188                         err = ret;
7189         }
7190         if (err) {
7191                 free_extent_map(em);
7192                 return ERR_PTR(err);
7193         }
7194         BUG_ON(!em); /* Error is always set */
7195         return em;
7196 }
7197
7198 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
7199                 struct page *page,
7200                 size_t pg_offset, u64 start, u64 len,
7201                 int create)
7202 {
7203         struct extent_map *em;
7204         struct extent_map *hole_em = NULL;
7205         u64 range_start = start;
7206         u64 end;
7207         u64 found;
7208         u64 found_end;
7209         int err = 0;
7210
7211         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7212         if (IS_ERR(em))
7213                 return em;
7214         /*
7215          * If our em maps to:
7216          * - a hole or
7217          * - a pre-alloc extent,
7218          * there might actually be delalloc bytes behind it.
7219          */
7220         if (em->block_start != EXTENT_MAP_HOLE &&
7221             !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7222                 return em;
7223         else
7224                 hole_em = em;
7225
7226         /* check to see if we've wrapped (len == -1 or similar) */
7227         end = start + len;
7228         if (end < start)
7229                 end = (u64)-1;
7230         else
7231                 end -= 1;
7232
7233         em = NULL;
7234
7235         /* ok, we didn't find anything, lets look for delalloc */
7236         found = count_range_bits(&inode->io_tree, &range_start,
7237                                  end, len, EXTENT_DELALLOC, 1);
7238         found_end = range_start + found;
7239         if (found_end < range_start)
7240                 found_end = (u64)-1;
7241
7242         /*
7243          * we didn't find anything useful, return
7244          * the original results from get_extent()
7245          */
7246         if (range_start > end || found_end <= start) {
7247                 em = hole_em;
7248                 hole_em = NULL;
7249                 goto out;
7250         }
7251
7252         /* adjust the range_start to make sure it doesn't
7253          * go backwards from the start they passed in
7254          */
7255         range_start = max(start, range_start);
7256         found = found_end - range_start;
7257
7258         if (found > 0) {
7259                 u64 hole_start = start;
7260                 u64 hole_len = len;
7261
7262                 em = alloc_extent_map();
7263                 if (!em) {
7264                         err = -ENOMEM;
7265                         goto out;
7266                 }
7267                 /*
7268                  * when btrfs_get_extent can't find anything it
7269                  * returns one huge hole
7270                  *
7271                  * make sure what it found really fits our range, and
7272                  * adjust to make sure it is based on the start from
7273                  * the caller
7274                  */
7275                 if (hole_em) {
7276                         u64 calc_end = extent_map_end(hole_em);
7277
7278                         if (calc_end <= start || (hole_em->start > end)) {
7279                                 free_extent_map(hole_em);
7280                                 hole_em = NULL;
7281                         } else {
7282                                 hole_start = max(hole_em->start, start);
7283                                 hole_len = calc_end - hole_start;
7284                         }
7285                 }
7286                 em->bdev = NULL;
7287                 if (hole_em && range_start > hole_start) {
7288                         /* our hole starts before our delalloc, so we
7289                          * have to return just the parts of the hole
7290                          * that go until  the delalloc starts
7291                          */
7292                         em->len = min(hole_len,
7293                                       range_start - hole_start);
7294                         em->start = hole_start;
7295                         em->orig_start = hole_start;
7296                         /*
7297                          * don't adjust block start at all,
7298                          * it is fixed at EXTENT_MAP_HOLE
7299                          */
7300                         em->block_start = hole_em->block_start;
7301                         em->block_len = hole_len;
7302                         if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7303                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7304                 } else {
7305                         em->start = range_start;
7306                         em->len = found;
7307                         em->orig_start = range_start;
7308                         em->block_start = EXTENT_MAP_DELALLOC;
7309                         em->block_len = found;
7310                 }
7311         } else if (hole_em) {
7312                 return hole_em;
7313         }
7314 out:
7315
7316         free_extent_map(hole_em);
7317         if (err) {
7318                 free_extent_map(em);
7319                 return ERR_PTR(err);
7320         }
7321         return em;
7322 }
7323
7324 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7325                                                   const u64 start,
7326                                                   const u64 len,
7327                                                   const u64 orig_start,
7328                                                   const u64 block_start,
7329                                                   const u64 block_len,
7330                                                   const u64 orig_block_len,
7331                                                   const u64 ram_bytes,
7332                                                   const int type)
7333 {
7334         struct extent_map *em = NULL;
7335         int ret;
7336
7337         if (type != BTRFS_ORDERED_NOCOW) {
7338                 em = create_io_em(inode, start, len, orig_start,
7339                                   block_start, block_len, orig_block_len,
7340                                   ram_bytes,
7341                                   BTRFS_COMPRESS_NONE, /* compress_type */
7342                                   type);
7343                 if (IS_ERR(em))
7344                         goto out;
7345         }
7346         ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7347                                            len, block_len, type);
7348         if (ret) {
7349                 if (em) {
7350                         free_extent_map(em);
7351                         btrfs_drop_extent_cache(BTRFS_I(inode), start,
7352                                                 start + len - 1, 0);
7353                 }
7354                 em = ERR_PTR(ret);
7355         }
7356  out:
7357
7358         return em;
7359 }
7360
7361 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7362                                                   u64 start, u64 len)
7363 {
7364         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7365         struct btrfs_root *root = BTRFS_I(inode)->root;
7366         struct extent_map *em;
7367         struct btrfs_key ins;
7368         u64 alloc_hint;
7369         int ret;
7370
7371         alloc_hint = get_extent_allocation_hint(inode, start, len);
7372         ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7373                                    0, alloc_hint, &ins, 1, 1);
7374         if (ret)
7375                 return ERR_PTR(ret);
7376
7377         em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7378                                      ins.objectid, ins.offset, ins.offset,
7379                                      ins.offset, BTRFS_ORDERED_REGULAR);
7380         btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7381         if (IS_ERR(em))
7382                 btrfs_free_reserved_extent(fs_info, ins.objectid,
7383                                            ins.offset, 1);
7384
7385         return em;
7386 }
7387
7388 /*
7389  * returns 1 when the nocow is safe, < 1 on error, 0 if the
7390  * block must be cow'd
7391  */
7392 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7393                               u64 *orig_start, u64 *orig_block_len,
7394                               u64 *ram_bytes)
7395 {
7396         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7397         struct btrfs_path *path;
7398         int ret;
7399         struct extent_buffer *leaf;
7400         struct btrfs_root *root = BTRFS_I(inode)->root;
7401         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7402         struct btrfs_file_extent_item *fi;
7403         struct btrfs_key key;
7404         u64 disk_bytenr;
7405         u64 backref_offset;
7406         u64 extent_end;
7407         u64 num_bytes;
7408         int slot;
7409         int found_type;
7410         bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7411
7412         path = btrfs_alloc_path();
7413         if (!path)
7414                 return -ENOMEM;
7415
7416         ret = btrfs_lookup_file_extent(NULL, root, path,
7417                         btrfs_ino(BTRFS_I(inode)), offset, 0);
7418         if (ret < 0)
7419                 goto out;
7420
7421         slot = path->slots[0];
7422         if (ret == 1) {
7423                 if (slot == 0) {
7424                         /* can't find the item, must cow */
7425                         ret = 0;
7426                         goto out;
7427                 }
7428                 slot--;
7429         }
7430         ret = 0;
7431         leaf = path->nodes[0];
7432         btrfs_item_key_to_cpu(leaf, &key, slot);
7433         if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7434             key.type != BTRFS_EXTENT_DATA_KEY) {
7435                 /* not our file or wrong item type, must cow */
7436                 goto out;
7437         }
7438
7439         if (key.offset > offset) {
7440                 /* Wrong offset, must cow */
7441                 goto out;
7442         }
7443
7444         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7445         found_type = btrfs_file_extent_type(leaf, fi);
7446         if (found_type != BTRFS_FILE_EXTENT_REG &&
7447             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7448                 /* not a regular extent, must cow */
7449                 goto out;
7450         }
7451
7452         if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7453                 goto out;
7454
7455         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7456         if (extent_end <= offset)
7457                 goto out;
7458
7459         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7460         if (disk_bytenr == 0)
7461                 goto out;
7462
7463         if (btrfs_file_extent_compression(leaf, fi) ||
7464             btrfs_file_extent_encryption(leaf, fi) ||
7465             btrfs_file_extent_other_encoding(leaf, fi))
7466                 goto out;
7467
7468         backref_offset = btrfs_file_extent_offset(leaf, fi);
7469
7470         if (orig_start) {
7471                 *orig_start = key.offset - backref_offset;
7472                 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7473                 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7474         }
7475
7476         if (btrfs_extent_readonly(fs_info, disk_bytenr))
7477                 goto out;
7478
7479         num_bytes = min(offset + *len, extent_end) - offset;
7480         if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7481                 u64 range_end;
7482
7483                 range_end = round_up(offset + num_bytes,
7484                                      root->fs_info->sectorsize) - 1;
7485                 ret = test_range_bit(io_tree, offset, range_end,
7486                                      EXTENT_DELALLOC, 0, NULL);
7487                 if (ret) {
7488                         ret = -EAGAIN;
7489                         goto out;
7490                 }
7491         }
7492
7493         btrfs_release_path(path);
7494
7495         /*
7496          * look for other files referencing this extent, if we
7497          * find any we must cow
7498          */
7499
7500         ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7501                                     key.offset - backref_offset, disk_bytenr);
7502         if (ret) {
7503                 ret = 0;
7504                 goto out;
7505         }
7506
7507         /*
7508          * adjust disk_bytenr and num_bytes to cover just the bytes
7509          * in this extent we are about to write.  If there
7510          * are any csums in that range we have to cow in order
7511          * to keep the csums correct
7512          */
7513         disk_bytenr += backref_offset;
7514         disk_bytenr += offset - key.offset;
7515         if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7516                 goto out;
7517         /*
7518          * all of the above have passed, it is safe to overwrite this extent
7519          * without cow
7520          */
7521         *len = num_bytes;
7522         ret = 1;
7523 out:
7524         btrfs_free_path(path);
7525         return ret;
7526 }
7527
7528 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7529 {
7530         struct radix_tree_root *root = &inode->i_mapping->page_tree;
7531         bool found = false;
7532         void **pagep = NULL;
7533         struct page *page = NULL;
7534         unsigned long start_idx;
7535         unsigned long end_idx;
7536
7537         start_idx = start >> PAGE_SHIFT;
7538
7539         /*
7540          * end is the last byte in the last page.  end == start is legal
7541          */
7542         end_idx = end >> PAGE_SHIFT;
7543
7544         rcu_read_lock();
7545
7546         /* Most of the code in this while loop is lifted from
7547          * find_get_page.  It's been modified to begin searching from a
7548          * page and return just the first page found in that range.  If the
7549          * found idx is less than or equal to the end idx then we know that
7550          * a page exists.  If no pages are found or if those pages are
7551          * outside of the range then we're fine (yay!) */
7552         while (page == NULL &&
7553                radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7554                 page = radix_tree_deref_slot(pagep);
7555                 if (unlikely(!page))
7556                         break;
7557
7558                 if (radix_tree_exception(page)) {
7559                         if (radix_tree_deref_retry(page)) {
7560                                 page = NULL;
7561                                 continue;
7562                         }
7563                         /*
7564                          * Otherwise, shmem/tmpfs must be storing a swap entry
7565                          * here as an exceptional entry: so return it without
7566                          * attempting to raise page count.
7567                          */
7568                         page = NULL;
7569                         break; /* TODO: Is this relevant for this use case? */
7570                 }
7571
7572                 if (!page_cache_get_speculative(page)) {
7573                         page = NULL;
7574                         continue;
7575                 }
7576
7577                 /*
7578                  * Has the page moved?
7579                  * This is part of the lockless pagecache protocol. See
7580                  * include/linux/pagemap.h for details.
7581                  */
7582                 if (unlikely(page != *pagep)) {
7583                         put_page(page);
7584                         page = NULL;
7585                 }
7586         }
7587
7588         if (page) {
7589                 if (page->index <= end_idx)
7590                         found = true;
7591                 put_page(page);
7592         }
7593
7594         rcu_read_unlock();
7595         return found;
7596 }
7597
7598 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7599                               struct extent_state **cached_state, int writing)
7600 {
7601         struct btrfs_ordered_extent *ordered;
7602         int ret = 0;
7603
7604         while (1) {
7605                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7606                                  cached_state);
7607                 /*
7608                  * We're concerned with the entire range that we're going to be
7609                  * doing DIO to, so we need to make sure there's no ordered
7610                  * extents in this range.
7611                  */
7612                 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7613                                                      lockend - lockstart + 1);
7614
7615                 /*
7616                  * We need to make sure there are no buffered pages in this
7617                  * range either, we could have raced between the invalidate in
7618                  * generic_file_direct_write and locking the extent.  The
7619                  * invalidate needs to happen so that reads after a write do not
7620                  * get stale data.
7621                  */
7622                 if (!ordered &&
7623                     (!writing ||
7624                      !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7625                         break;
7626
7627                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7628                                      cached_state, GFP_NOFS);
7629
7630                 if (ordered) {
7631                         /*
7632                          * If we are doing a DIO read and the ordered extent we
7633                          * found is for a buffered write, we can not wait for it
7634                          * to complete and retry, because if we do so we can
7635                          * deadlock with concurrent buffered writes on page
7636                          * locks. This happens only if our DIO read covers more
7637                          * than one extent map, if at this point has already
7638                          * created an ordered extent for a previous extent map
7639                          * and locked its range in the inode's io tree, and a
7640                          * concurrent write against that previous extent map's
7641                          * range and this range started (we unlock the ranges
7642                          * in the io tree only when the bios complete and
7643                          * buffered writes always lock pages before attempting
7644                          * to lock range in the io tree).
7645                          */
7646                         if (writing ||
7647                             test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7648                                 btrfs_start_ordered_extent(inode, ordered, 1);
7649                         else
7650                                 ret = -ENOTBLK;
7651                         btrfs_put_ordered_extent(ordered);
7652                 } else {
7653                         /*
7654                          * We could trigger writeback for this range (and wait
7655                          * for it to complete) and then invalidate the pages for
7656                          * this range (through invalidate_inode_pages2_range()),
7657                          * but that can lead us to a deadlock with a concurrent
7658                          * call to readpages() (a buffered read or a defrag call
7659                          * triggered a readahead) on a page lock due to an
7660                          * ordered dio extent we created before but did not have
7661                          * yet a corresponding bio submitted (whence it can not
7662                          * complete), which makes readpages() wait for that
7663                          * ordered extent to complete while holding a lock on
7664                          * that page.
7665                          */
7666                         ret = -ENOTBLK;
7667                 }
7668
7669                 if (ret)
7670                         break;
7671
7672                 cond_resched();
7673         }
7674
7675         return ret;
7676 }
7677
7678 /* The callers of this must take lock_extent() */
7679 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
7680                                        u64 orig_start, u64 block_start,
7681                                        u64 block_len, u64 orig_block_len,
7682                                        u64 ram_bytes, int compress_type,
7683                                        int type)
7684 {
7685         struct extent_map_tree *em_tree;
7686         struct extent_map *em;
7687         struct btrfs_root *root = BTRFS_I(inode)->root;
7688         int ret;
7689
7690         ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7691                type == BTRFS_ORDERED_COMPRESSED ||
7692                type == BTRFS_ORDERED_NOCOW ||
7693                type == BTRFS_ORDERED_REGULAR);
7694
7695         em_tree = &BTRFS_I(inode)->extent_tree;
7696         em = alloc_extent_map();
7697         if (!em)
7698                 return ERR_PTR(-ENOMEM);
7699
7700         em->start = start;
7701         em->orig_start = orig_start;
7702         em->len = len;
7703         em->block_len = block_len;
7704         em->block_start = block_start;
7705         em->bdev = root->fs_info->fs_devices->latest_bdev;
7706         em->orig_block_len = orig_block_len;
7707         em->ram_bytes = ram_bytes;
7708         em->generation = -1;
7709         set_bit(EXTENT_FLAG_PINNED, &em->flags);
7710         if (type == BTRFS_ORDERED_PREALLOC) {
7711                 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7712         } else if (type == BTRFS_ORDERED_COMPRESSED) {
7713                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7714                 em->compress_type = compress_type;
7715         }
7716
7717         do {
7718                 btrfs_drop_extent_cache(BTRFS_I(inode), em->start,
7719                                 em->start + em->len - 1, 0);
7720                 write_lock(&em_tree->lock);
7721                 ret = add_extent_mapping(em_tree, em, 1);
7722                 write_unlock(&em_tree->lock);
7723                 /*
7724                  * The caller has taken lock_extent(), who could race with us
7725                  * to add em?
7726                  */
7727         } while (ret == -EEXIST);
7728
7729         if (ret) {
7730                 free_extent_map(em);
7731                 return ERR_PTR(ret);
7732         }
7733
7734         /* em got 2 refs now, callers needs to do free_extent_map once. */
7735         return em;
7736 }
7737
7738 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7739                                    struct buffer_head *bh_result, int create)
7740 {
7741         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7742         struct extent_map *em;
7743         struct extent_state *cached_state = NULL;
7744         struct btrfs_dio_data *dio_data = NULL;
7745         u64 start = iblock << inode->i_blkbits;
7746         u64 lockstart, lockend;
7747         u64 len = bh_result->b_size;
7748         int unlock_bits = EXTENT_LOCKED;
7749         int ret = 0;
7750
7751         if (create)
7752                 unlock_bits |= EXTENT_DIRTY;
7753         else
7754                 len = min_t(u64, len, fs_info->sectorsize);
7755
7756         lockstart = start;
7757         lockend = start + len - 1;
7758
7759         if (current->journal_info) {
7760                 /*
7761                  * Need to pull our outstanding extents and set journal_info to NULL so
7762                  * that anything that needs to check if there's a transaction doesn't get
7763                  * confused.
7764                  */
7765                 dio_data = current->journal_info;
7766                 current->journal_info = NULL;
7767         }
7768
7769         /*
7770          * If this errors out it's because we couldn't invalidate pagecache for
7771          * this range and we need to fallback to buffered.
7772          */
7773         if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7774                                create)) {
7775                 ret = -ENOTBLK;
7776                 goto err;
7777         }
7778
7779         em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
7780         if (IS_ERR(em)) {
7781                 ret = PTR_ERR(em);
7782                 goto unlock_err;
7783         }
7784
7785         /*
7786          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7787          * io.  INLINE is special, and we could probably kludge it in here, but
7788          * it's still buffered so for safety lets just fall back to the generic
7789          * buffered path.
7790          *
7791          * For COMPRESSED we _have_ to read the entire extent in so we can
7792          * decompress it, so there will be buffering required no matter what we
7793          * do, so go ahead and fallback to buffered.
7794          *
7795          * We return -ENOTBLK because that's what makes DIO go ahead and go back
7796          * to buffered IO.  Don't blame me, this is the price we pay for using
7797          * the generic code.
7798          */
7799         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7800             em->block_start == EXTENT_MAP_INLINE) {
7801                 free_extent_map(em);
7802                 ret = -ENOTBLK;
7803                 goto unlock_err;
7804         }
7805
7806         /* Just a good old fashioned hole, return */
7807         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7808                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7809                 free_extent_map(em);
7810                 goto unlock_err;
7811         }
7812
7813         /*
7814          * We don't allocate a new extent in the following cases
7815          *
7816          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
7817          * existing extent.
7818          * 2) The extent is marked as PREALLOC.  We're good to go here and can
7819          * just use the extent.
7820          *
7821          */
7822         if (!create) {
7823                 len = min(len, em->len - (start - em->start));
7824                 lockstart = start + len;
7825                 goto unlock;
7826         }
7827
7828         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7829             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7830              em->block_start != EXTENT_MAP_HOLE)) {
7831                 int type;
7832                 u64 block_start, orig_start, orig_block_len, ram_bytes;
7833
7834                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7835                         type = BTRFS_ORDERED_PREALLOC;
7836                 else
7837                         type = BTRFS_ORDERED_NOCOW;
7838                 len = min(len, em->len - (start - em->start));
7839                 block_start = em->block_start + (start - em->start);
7840
7841                 if (can_nocow_extent(inode, start, &len, &orig_start,
7842                                      &orig_block_len, &ram_bytes) == 1 &&
7843                     btrfs_inc_nocow_writers(fs_info, block_start)) {
7844                         struct extent_map *em2;
7845
7846                         em2 = btrfs_create_dio_extent(inode, start, len,
7847                                                       orig_start, block_start,
7848                                                       len, orig_block_len,
7849                                                       ram_bytes, type);
7850                         btrfs_dec_nocow_writers(fs_info, block_start);
7851                         if (type == BTRFS_ORDERED_PREALLOC) {
7852                                 free_extent_map(em);
7853                                 em = em2;
7854                         }
7855                         if (em2 && IS_ERR(em2)) {
7856                                 ret = PTR_ERR(em2);
7857                                 goto unlock_err;
7858                         }
7859                         /*
7860                          * For inode marked NODATACOW or extent marked PREALLOC,
7861                          * use the existing or preallocated extent, so does not
7862                          * need to adjust btrfs_space_info's bytes_may_use.
7863                          */
7864                         btrfs_free_reserved_data_space_noquota(inode,
7865                                         start, len);
7866                         goto unlock;
7867                 }
7868         }
7869
7870         /*
7871          * this will cow the extent, reset the len in case we changed
7872          * it above
7873          */
7874         len = bh_result->b_size;
7875         free_extent_map(em);
7876         em = btrfs_new_extent_direct(inode, start, len);
7877         if (IS_ERR(em)) {
7878                 ret = PTR_ERR(em);
7879                 goto unlock_err;
7880         }
7881         len = min(len, em->len - (start - em->start));
7882 unlock:
7883         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7884                 inode->i_blkbits;
7885         bh_result->b_size = len;
7886         bh_result->b_bdev = em->bdev;
7887         set_buffer_mapped(bh_result);
7888         if (create) {
7889                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7890                         set_buffer_new(bh_result);
7891
7892                 /*
7893                  * Need to update the i_size under the extent lock so buffered
7894                  * readers will get the updated i_size when we unlock.
7895                  */
7896                 if (!dio_data->overwrite && start + len > i_size_read(inode))
7897                         i_size_write(inode, start + len);
7898
7899                 WARN_ON(dio_data->reserve < len);
7900                 dio_data->reserve -= len;
7901                 dio_data->unsubmitted_oe_range_end = start + len;
7902                 current->journal_info = dio_data;
7903         }
7904
7905         /*
7906          * In the case of write we need to clear and unlock the entire range,
7907          * in the case of read we need to unlock only the end area that we
7908          * aren't using if there is any left over space.
7909          */
7910         if (lockstart < lockend) {
7911                 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7912                                  lockend, unlock_bits, 1, 0,
7913                                  &cached_state, GFP_NOFS);
7914         } else {
7915                 free_extent_state(cached_state);
7916         }
7917
7918         free_extent_map(em);
7919
7920         return 0;
7921
7922 unlock_err:
7923         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7924                          unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7925 err:
7926         if (dio_data)
7927                 current->journal_info = dio_data;
7928         return ret;
7929 }
7930
7931 static inline blk_status_t submit_dio_repair_bio(struct inode *inode,
7932                                                  struct bio *bio,
7933                                                  int mirror_num)
7934 {
7935         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7936         blk_status_t ret;
7937
7938         BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7939
7940         bio_get(bio);
7941
7942         ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR);
7943         if (ret)
7944                 goto err;
7945
7946         ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
7947 err:
7948         bio_put(bio);
7949         return ret;
7950 }
7951
7952 static int btrfs_check_dio_repairable(struct inode *inode,
7953                                       struct bio *failed_bio,
7954                                       struct io_failure_record *failrec,
7955                                       int failed_mirror)
7956 {
7957         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7958         int num_copies;
7959
7960         num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
7961         if (num_copies == 1) {
7962                 /*
7963                  * we only have a single copy of the data, so don't bother with
7964                  * all the retry and error correction code that follows. no
7965                  * matter what the error is, it is very likely to persist.
7966                  */
7967                 btrfs_debug(fs_info,
7968                         "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
7969                         num_copies, failrec->this_mirror, failed_mirror);
7970                 return 0;
7971         }
7972
7973         failrec->failed_mirror = failed_mirror;
7974         failrec->this_mirror++;
7975         if (failrec->this_mirror == failed_mirror)
7976                 failrec->this_mirror++;
7977
7978         if (failrec->this_mirror > num_copies) {
7979                 btrfs_debug(fs_info,
7980                         "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
7981                         num_copies, failrec->this_mirror, failed_mirror);
7982                 return 0;
7983         }
7984
7985         return 1;
7986 }
7987
7988 static blk_status_t dio_read_error(struct inode *inode, struct bio *failed_bio,
7989                                    struct page *page, unsigned int pgoff,
7990                                    u64 start, u64 end, int failed_mirror,
7991                                    bio_end_io_t *repair_endio, void *repair_arg)
7992 {
7993         struct io_failure_record *failrec;
7994         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7995         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7996         struct bio *bio;
7997         int isector;
7998         unsigned int read_mode = 0;
7999         int segs;
8000         int ret;
8001         blk_status_t status;
8002
8003         BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
8004
8005         ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
8006         if (ret)
8007                 return errno_to_blk_status(ret);
8008
8009         ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
8010                                          failed_mirror);
8011         if (!ret) {
8012                 free_io_failure(failure_tree, io_tree, failrec);
8013                 return BLK_STS_IOERR;
8014         }
8015
8016         segs = bio_segments(failed_bio);
8017         if (segs > 1 ||
8018             (failed_bio->bi_io_vec->bv_len > btrfs_inode_sectorsize(inode)))
8019                 read_mode |= REQ_FAILFAST_DEV;
8020
8021         isector = start - btrfs_io_bio(failed_bio)->logical;
8022         isector >>= inode->i_sb->s_blocksize_bits;
8023         bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
8024                                 pgoff, isector, repair_endio, repair_arg);
8025         bio_set_op_attrs(bio, REQ_OP_READ, read_mode);
8026
8027         btrfs_debug(BTRFS_I(inode)->root->fs_info,
8028                     "repair DIO read error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d",
8029                     read_mode, failrec->this_mirror, failrec->in_validation);
8030
8031         status = submit_dio_repair_bio(inode, bio, failrec->this_mirror);
8032         if (status) {
8033                 free_io_failure(failure_tree, io_tree, failrec);
8034                 bio_put(bio);
8035         }
8036
8037         return status;
8038 }
8039
8040 struct btrfs_retry_complete {
8041         struct completion done;
8042         struct inode *inode;
8043         u64 start;
8044         int uptodate;
8045 };
8046
8047 static void btrfs_retry_endio_nocsum(struct bio *bio)
8048 {
8049         struct btrfs_retry_complete *done = bio->bi_private;
8050         struct inode *inode = done->inode;
8051         struct bio_vec *bvec;
8052         struct extent_io_tree *io_tree, *failure_tree;
8053         int i;
8054
8055         if (bio->bi_status)
8056                 goto end;
8057
8058         ASSERT(bio->bi_vcnt == 1);
8059         io_tree = &BTRFS_I(inode)->io_tree;
8060         failure_tree = &BTRFS_I(inode)->io_failure_tree;
8061         ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(inode));
8062
8063         done->uptodate = 1;
8064         ASSERT(!bio_flagged(bio, BIO_CLONED));
8065         bio_for_each_segment_all(bvec, bio, i)
8066                 clean_io_failure(BTRFS_I(inode)->root->fs_info, failure_tree,
8067                                  io_tree, done->start, bvec->bv_page,
8068                                  btrfs_ino(BTRFS_I(inode)), 0);
8069 end:
8070         complete(&done->done);
8071         bio_put(bio);
8072 }
8073
8074 static blk_status_t __btrfs_correct_data_nocsum(struct inode *inode,
8075                                                 struct btrfs_io_bio *io_bio)
8076 {
8077         struct btrfs_fs_info *fs_info;
8078         struct bio_vec bvec;
8079         struct bvec_iter iter;
8080         struct btrfs_retry_complete done;
8081         u64 start;
8082         unsigned int pgoff;
8083         u32 sectorsize;
8084         int nr_sectors;
8085         blk_status_t ret;
8086         blk_status_t err = BLK_STS_OK;
8087
8088         fs_info = BTRFS_I(inode)->root->fs_info;
8089         sectorsize = fs_info->sectorsize;
8090
8091         start = io_bio->logical;
8092         done.inode = inode;
8093         io_bio->bio.bi_iter = io_bio->iter;
8094
8095         bio_for_each_segment(bvec, &io_bio->bio, iter) {
8096                 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
8097                 pgoff = bvec.bv_offset;
8098
8099 next_block_or_try_again:
8100                 done.uptodate = 0;
8101                 done.start = start;
8102                 init_completion(&done.done);
8103
8104                 ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
8105                                 pgoff, start, start + sectorsize - 1,
8106                                 io_bio->mirror_num,
8107                                 btrfs_retry_endio_nocsum, &done);
8108                 if (ret) {
8109                         err = ret;
8110                         goto next;
8111                 }
8112
8113                 wait_for_completion_io(&done.done);
8114
8115                 if (!done.uptodate) {
8116                         /* We might have another mirror, so try again */
8117                         goto next_block_or_try_again;
8118                 }
8119
8120 next:
8121                 start += sectorsize;
8122
8123                 nr_sectors--;
8124                 if (nr_sectors) {
8125                         pgoff += sectorsize;
8126                         ASSERT(pgoff < PAGE_SIZE);
8127                         goto next_block_or_try_again;
8128                 }
8129         }
8130
8131         return err;
8132 }
8133
8134 static void btrfs_retry_endio(struct bio *bio)
8135 {
8136         struct btrfs_retry_complete *done = bio->bi_private;
8137         struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8138         struct extent_io_tree *io_tree, *failure_tree;
8139         struct inode *inode = done->inode;
8140         struct bio_vec *bvec;
8141         int uptodate;
8142         int ret;
8143         int i;
8144
8145         if (bio->bi_status)
8146                 goto end;
8147
8148         uptodate = 1;
8149
8150         ASSERT(bio->bi_vcnt == 1);
8151         ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(done->inode));
8152
8153         io_tree = &BTRFS_I(inode)->io_tree;
8154         failure_tree = &BTRFS_I(inode)->io_failure_tree;
8155
8156         ASSERT(!bio_flagged(bio, BIO_CLONED));
8157         bio_for_each_segment_all(bvec, bio, i) {
8158                 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
8159                                              bvec->bv_offset, done->start,
8160                                              bvec->bv_len);
8161                 if (!ret)
8162                         clean_io_failure(BTRFS_I(inode)->root->fs_info,
8163                                          failure_tree, io_tree, done->start,
8164                                          bvec->bv_page,
8165                                          btrfs_ino(BTRFS_I(inode)),
8166                                          bvec->bv_offset);
8167                 else
8168                         uptodate = 0;
8169         }
8170
8171         done->uptodate = uptodate;
8172 end:
8173         complete(&done->done);
8174         bio_put(bio);
8175 }
8176
8177 static blk_status_t __btrfs_subio_endio_read(struct inode *inode,
8178                 struct btrfs_io_bio *io_bio, blk_status_t err)
8179 {
8180         struct btrfs_fs_info *fs_info;
8181         struct bio_vec bvec;
8182         struct bvec_iter iter;
8183         struct btrfs_retry_complete done;
8184         u64 start;
8185         u64 offset = 0;
8186         u32 sectorsize;
8187         int nr_sectors;
8188         unsigned int pgoff;
8189         int csum_pos;
8190         bool uptodate = (err == 0);
8191         int ret;
8192         blk_status_t status;
8193
8194         fs_info = BTRFS_I(inode)->root->fs_info;
8195         sectorsize = fs_info->sectorsize;
8196
8197         err = BLK_STS_OK;
8198         start = io_bio->logical;
8199         done.inode = inode;
8200         io_bio->bio.bi_iter = io_bio->iter;
8201
8202         bio_for_each_segment(bvec, &io_bio->bio, iter) {
8203                 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
8204
8205                 pgoff = bvec.bv_offset;
8206 next_block:
8207                 if (uptodate) {
8208                         csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
8209                         ret = __readpage_endio_check(inode, io_bio, csum_pos,
8210                                         bvec.bv_page, pgoff, start, sectorsize);
8211                         if (likely(!ret))
8212                                 goto next;
8213                 }
8214 try_again:
8215                 done.uptodate = 0;
8216                 done.start = start;
8217                 init_completion(&done.done);
8218
8219                 status = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
8220                                         pgoff, start, start + sectorsize - 1,
8221                                         io_bio->mirror_num, btrfs_retry_endio,
8222                                         &done);
8223                 if (status) {
8224                         err = status;
8225                         goto next;
8226                 }
8227
8228                 wait_for_completion_io(&done.done);
8229
8230                 if (!done.uptodate) {
8231                         /* We might have another mirror, so try again */
8232                         goto try_again;
8233                 }
8234 next:
8235                 offset += sectorsize;
8236                 start += sectorsize;
8237
8238                 ASSERT(nr_sectors);
8239
8240                 nr_sectors--;
8241                 if (nr_sectors) {
8242                         pgoff += sectorsize;
8243                         ASSERT(pgoff < PAGE_SIZE);
8244                         goto next_block;
8245                 }
8246         }
8247
8248         return err;
8249 }
8250
8251 static blk_status_t btrfs_subio_endio_read(struct inode *inode,
8252                 struct btrfs_io_bio *io_bio, blk_status_t err)
8253 {
8254         bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8255
8256         if (skip_csum) {
8257                 if (unlikely(err))
8258                         return __btrfs_correct_data_nocsum(inode, io_bio);
8259                 else
8260                         return BLK_STS_OK;
8261         } else {
8262                 return __btrfs_subio_endio_read(inode, io_bio, err);
8263         }
8264 }
8265
8266 static void btrfs_endio_direct_read(struct bio *bio)
8267 {
8268         struct btrfs_dio_private *dip = bio->bi_private;
8269         struct inode *inode = dip->inode;
8270         struct bio *dio_bio;
8271         struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8272         blk_status_t err = bio->bi_status;
8273
8274         if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8275                 err = btrfs_subio_endio_read(inode, io_bio, err);
8276
8277         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8278                       dip->logical_offset + dip->bytes - 1);
8279         dio_bio = dip->dio_bio;
8280
8281         kfree(dip);
8282
8283         dio_bio->bi_status = err;
8284         dio_end_io(dio_bio);
8285
8286         if (io_bio->end_io)
8287                 io_bio->end_io(io_bio, blk_status_to_errno(err));
8288         bio_put(bio);
8289 }
8290
8291 static void __endio_write_update_ordered(struct inode *inode,
8292                                          const u64 offset, const u64 bytes,
8293                                          const bool uptodate)
8294 {
8295         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8296         struct btrfs_ordered_extent *ordered = NULL;
8297         struct btrfs_workqueue *wq;
8298         btrfs_work_func_t func;
8299         u64 ordered_offset = offset;
8300         u64 ordered_bytes = bytes;
8301         u64 last_offset;
8302         int ret;
8303
8304         if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
8305                 wq = fs_info->endio_freespace_worker;
8306                 func = btrfs_freespace_write_helper;
8307         } else {
8308                 wq = fs_info->endio_write_workers;
8309                 func = btrfs_endio_write_helper;
8310         }
8311
8312 again:
8313         last_offset = ordered_offset;
8314         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8315                                                    &ordered_offset,
8316                                                    ordered_bytes,
8317                                                    uptodate);
8318         if (!ret)
8319                 goto out_test;
8320
8321         btrfs_init_work(&ordered->work, func, finish_ordered_fn, NULL, NULL);
8322         btrfs_queue_work(wq, &ordered->work);
8323 out_test:
8324         /*
8325          * If btrfs_dec_test_ordered_pending does not find any ordered extent
8326          * in the range, we can exit.
8327          */
8328         if (ordered_offset == last_offset)
8329                 return;
8330         /*
8331          * our bio might span multiple ordered extents.  If we haven't
8332          * completed the accounting for the whole dio, go back and try again
8333          */
8334         if (ordered_offset < offset + bytes) {
8335                 ordered_bytes = offset + bytes - ordered_offset;
8336                 ordered = NULL;
8337                 goto again;
8338         }
8339 }
8340
8341 static void btrfs_endio_direct_write(struct bio *bio)
8342 {
8343         struct btrfs_dio_private *dip = bio->bi_private;
8344         struct bio *dio_bio = dip->dio_bio;
8345
8346         __endio_write_update_ordered(dip->inode, dip->logical_offset,
8347                                      dip->bytes, !bio->bi_status);
8348
8349         kfree(dip);
8350
8351         dio_bio->bi_status = bio->bi_status;
8352         dio_end_io(dio_bio);
8353         bio_put(bio);
8354 }
8355
8356 static blk_status_t __btrfs_submit_bio_start_direct_io(void *private_data,
8357                                     struct bio *bio, int mirror_num,
8358                                     unsigned long bio_flags, u64 offset)
8359 {
8360         struct inode *inode = private_data;
8361         blk_status_t ret;
8362         ret = btrfs_csum_one_bio(inode, bio, offset, 1);
8363         BUG_ON(ret); /* -ENOMEM */
8364         return 0;
8365 }
8366
8367 static void btrfs_end_dio_bio(struct bio *bio)
8368 {
8369         struct btrfs_dio_private *dip = bio->bi_private;
8370         blk_status_t err = bio->bi_status;
8371
8372         if (err)
8373                 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8374                            "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8375                            btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8376                            bio->bi_opf,
8377                            (unsigned long long)bio->bi_iter.bi_sector,
8378                            bio->bi_iter.bi_size, err);
8379
8380         if (dip->subio_endio)
8381                 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8382
8383         if (err) {
8384                 dip->errors = 1;
8385
8386                 /*
8387                  * before atomic variable goto zero, we must make sure
8388                  * dip->errors is perceived to be set.
8389                  */
8390                 smp_mb__before_atomic();
8391         }
8392
8393         /* if there are more bios still pending for this dio, just exit */
8394         if (!atomic_dec_and_test(&dip->pending_bios))
8395                 goto out;
8396
8397         if (dip->errors) {
8398                 bio_io_error(dip->orig_bio);
8399         } else {
8400                 dip->dio_bio->bi_status = BLK_STS_OK;
8401                 bio_endio(dip->orig_bio);
8402         }
8403 out:
8404         bio_put(bio);
8405 }
8406
8407 static inline blk_status_t btrfs_lookup_and_bind_dio_csum(struct inode *inode,
8408                                                  struct btrfs_dio_private *dip,
8409                                                  struct bio *bio,
8410                                                  u64 file_offset)
8411 {
8412         struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8413         struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8414         blk_status_t ret;
8415
8416         /*
8417          * We load all the csum data we need when we submit
8418          * the first bio to reduce the csum tree search and
8419          * contention.
8420          */
8421         if (dip->logical_offset == file_offset) {
8422                 ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio,
8423                                                 file_offset);
8424                 if (ret)
8425                         return ret;
8426         }
8427
8428         if (bio == dip->orig_bio)
8429                 return 0;
8430
8431         file_offset -= dip->logical_offset;
8432         file_offset >>= inode->i_sb->s_blocksize_bits;
8433         io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8434
8435         return 0;
8436 }
8437
8438 static inline blk_status_t
8439 __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, u64 file_offset,
8440                        int async_submit)
8441 {
8442         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8443         struct btrfs_dio_private *dip = bio->bi_private;
8444         bool write = bio_op(bio) == REQ_OP_WRITE;
8445         blk_status_t ret;
8446
8447         if (async_submit)
8448                 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8449
8450         bio_get(bio);
8451
8452         if (!write) {
8453                 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8454                 if (ret)
8455                         goto err;
8456         }
8457
8458         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
8459                 goto map;
8460
8461         if (write && async_submit) {
8462                 ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0,
8463                                           file_offset, inode,
8464                                           __btrfs_submit_bio_start_direct_io,
8465                                           __btrfs_submit_bio_done);
8466                 goto err;
8467         } else if (write) {
8468                 /*
8469                  * If we aren't doing async submit, calculate the csum of the
8470                  * bio now.
8471                  */
8472                 ret = btrfs_csum_one_bio(inode, bio, file_offset, 1);
8473                 if (ret)
8474                         goto err;
8475         } else {
8476                 ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio,
8477                                                      file_offset);
8478                 if (ret)
8479                         goto err;
8480         }
8481 map:
8482         ret = btrfs_map_bio(fs_info, bio, 0, 0);
8483 err:
8484         bio_put(bio);
8485         return ret;
8486 }
8487
8488 static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip)
8489 {
8490         struct inode *inode = dip->inode;
8491         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8492         struct bio *bio;
8493         struct bio *orig_bio = dip->orig_bio;
8494         u64 start_sector = orig_bio->bi_iter.bi_sector;
8495         u64 file_offset = dip->logical_offset;
8496         u64 map_length;
8497         int async_submit = 0;
8498         u64 submit_len;
8499         int clone_offset = 0;
8500         int clone_len;
8501         int ret;
8502         blk_status_t status;
8503
8504         map_length = orig_bio->bi_iter.bi_size;
8505         submit_len = map_length;
8506         ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9,
8507                               &map_length, NULL, 0);
8508         if (ret)
8509                 return -EIO;
8510
8511         if (map_length >= submit_len) {
8512                 bio = orig_bio;
8513                 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8514                 goto submit;
8515         }
8516
8517         /* async crcs make it difficult to collect full stripe writes. */
8518         if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8519                 async_submit = 0;
8520         else
8521                 async_submit = 1;
8522
8523         /* bio split */
8524         ASSERT(map_length <= INT_MAX);
8525         atomic_inc(&dip->pending_bios);
8526         do {
8527                 clone_len = min_t(int, submit_len, map_length);
8528
8529                 /*
8530                  * This will never fail as it's passing GPF_NOFS and
8531                  * the allocation is backed by btrfs_bioset.
8532                  */
8533                 bio = btrfs_bio_clone_partial(orig_bio, clone_offset,
8534                                               clone_len);
8535                 bio->bi_private = dip;
8536                 bio->bi_end_io = btrfs_end_dio_bio;
8537                 btrfs_io_bio(bio)->logical = file_offset;
8538
8539                 ASSERT(submit_len >= clone_len);
8540                 submit_len -= clone_len;
8541                 if (submit_len == 0)
8542                         break;
8543
8544                 /*
8545                  * Increase the count before we submit the bio so we know
8546                  * the end IO handler won't happen before we increase the
8547                  * count. Otherwise, the dip might get freed before we're
8548                  * done setting it up.
8549                  */
8550                 atomic_inc(&dip->pending_bios);
8551
8552                 status = __btrfs_submit_dio_bio(bio, inode, file_offset,
8553                                                 async_submit);
8554                 if (status) {
8555                         bio_put(bio);
8556                         atomic_dec(&dip->pending_bios);
8557                         goto out_err;
8558                 }
8559
8560                 clone_offset += clone_len;
8561                 start_sector += clone_len >> 9;
8562                 file_offset += clone_len;
8563
8564                 map_length = submit_len;
8565                 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio),
8566                                       start_sector << 9, &map_length, NULL, 0);
8567                 if (ret)
8568                         goto out_err;
8569         } while (submit_len > 0);
8570
8571 submit:
8572         status = __btrfs_submit_dio_bio(bio, inode, file_offset, async_submit);
8573         if (!status)
8574                 return 0;
8575
8576         bio_put(bio);
8577 out_err:
8578         dip->errors = 1;
8579         /*
8580          * before atomic variable goto zero, we must
8581          * make sure dip->errors is perceived to be set.
8582          */
8583         smp_mb__before_atomic();
8584         if (atomic_dec_and_test(&dip->pending_bios))
8585                 bio_io_error(dip->orig_bio);
8586
8587         /* bio_end_io() will handle error, so we needn't return it */
8588         return 0;
8589 }
8590
8591 static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode,
8592                                 loff_t file_offset)
8593 {
8594         struct btrfs_dio_private *dip = NULL;
8595         struct bio *bio = NULL;
8596         struct btrfs_io_bio *io_bio;
8597         bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
8598         int ret = 0;
8599
8600         bio = btrfs_bio_clone(dio_bio);
8601
8602         dip = kzalloc(sizeof(*dip), GFP_NOFS);
8603         if (!dip) {
8604                 ret = -ENOMEM;
8605                 goto free_ordered;
8606         }
8607
8608         dip->private = dio_bio->bi_private;
8609         dip->inode = inode;
8610         dip->logical_offset = file_offset;
8611         dip->bytes = dio_bio->bi_iter.bi_size;
8612         dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8613         bio->bi_private = dip;
8614         dip->orig_bio = bio;
8615         dip->dio_bio = dio_bio;
8616         atomic_set(&dip->pending_bios, 0);
8617         io_bio = btrfs_io_bio(bio);
8618         io_bio->logical = file_offset;
8619
8620         if (write) {
8621                 bio->bi_end_io = btrfs_endio_direct_write;
8622         } else {
8623                 bio->bi_end_io = btrfs_endio_direct_read;
8624                 dip->subio_endio = btrfs_subio_endio_read;
8625         }
8626
8627         /*
8628          * Reset the range for unsubmitted ordered extents (to a 0 length range)
8629          * even if we fail to submit a bio, because in such case we do the
8630          * corresponding error handling below and it must not be done a second
8631          * time by btrfs_direct_IO().
8632          */
8633         if (write) {
8634                 struct btrfs_dio_data *dio_data = current->journal_info;
8635
8636                 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8637                         dip->bytes;
8638                 dio_data->unsubmitted_oe_range_start =
8639                         dio_data->unsubmitted_oe_range_end;
8640         }
8641
8642         ret = btrfs_submit_direct_hook(dip);
8643         if (!ret)
8644                 return;
8645
8646         if (io_bio->end_io)
8647                 io_bio->end_io(io_bio, ret);
8648
8649 free_ordered:
8650         /*
8651          * If we arrived here it means either we failed to submit the dip
8652          * or we either failed to clone the dio_bio or failed to allocate the
8653          * dip. If we cloned the dio_bio and allocated the dip, we can just
8654          * call bio_endio against our io_bio so that we get proper resource
8655          * cleanup if we fail to submit the dip, otherwise, we must do the
8656          * same as btrfs_endio_direct_[write|read] because we can't call these
8657          * callbacks - they require an allocated dip and a clone of dio_bio.
8658          */
8659         if (bio && dip) {
8660                 bio_io_error(bio);
8661                 /*
8662                  * The end io callbacks free our dip, do the final put on bio
8663                  * and all the cleanup and final put for dio_bio (through
8664                  * dio_end_io()).
8665                  */
8666                 dip = NULL;
8667                 bio = NULL;
8668         } else {
8669                 if (write)
8670                         __endio_write_update_ordered(inode,
8671                                                 file_offset,
8672                                                 dio_bio->bi_iter.bi_size,
8673                                                 false);
8674                 else
8675                         unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8676                               file_offset + dio_bio->bi_iter.bi_size - 1);
8677
8678                 dio_bio->bi_status = BLK_STS_IOERR;
8679                 /*
8680                  * Releases and cleans up our dio_bio, no need to bio_put()
8681                  * nor bio_endio()/bio_io_error() against dio_bio.
8682                  */
8683                 dio_end_io(dio_bio);
8684         }
8685         if (bio)
8686                 bio_put(bio);
8687         kfree(dip);
8688 }
8689
8690 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
8691                                const struct iov_iter *iter, loff_t offset)
8692 {
8693         int seg;
8694         int i;
8695         unsigned int blocksize_mask = fs_info->sectorsize - 1;
8696         ssize_t retval = -EINVAL;
8697
8698         if (offset & blocksize_mask)
8699                 goto out;
8700
8701         if (iov_iter_alignment(iter) & blocksize_mask)
8702                 goto out;
8703
8704         /* If this is a write we don't need to check anymore */
8705         if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter))
8706                 return 0;
8707         /*
8708          * Check to make sure we don't have duplicate iov_base's in this
8709          * iovec, if so return EINVAL, otherwise we'll get csum errors
8710          * when reading back.
8711          */
8712         for (seg = 0; seg < iter->nr_segs; seg++) {
8713                 for (i = seg + 1; i < iter->nr_segs; i++) {
8714                         if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8715                                 goto out;
8716                 }
8717         }
8718         retval = 0;
8719 out:
8720         return retval;
8721 }
8722
8723 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8724 {
8725         struct file *file = iocb->ki_filp;
8726         struct inode *inode = file->f_mapping->host;
8727         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8728         struct btrfs_dio_data dio_data = { 0 };
8729         struct extent_changeset *data_reserved = NULL;
8730         loff_t offset = iocb->ki_pos;
8731         size_t count = 0;
8732         int flags = 0;
8733         bool wakeup = true;
8734         bool relock = false;
8735         ssize_t ret;
8736
8737         if (check_direct_IO(fs_info, iter, offset))
8738                 return 0;
8739
8740         inode_dio_begin(inode);
8741
8742         /*
8743          * The generic stuff only does filemap_write_and_wait_range, which
8744          * isn't enough if we've written compressed pages to this area, so
8745          * we need to flush the dirty pages again to make absolutely sure
8746          * that any outstanding dirty pages are on disk.
8747          */
8748         count = iov_iter_count(iter);
8749         if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8750                      &BTRFS_I(inode)->runtime_flags))
8751                 filemap_fdatawrite_range(inode->i_mapping, offset,
8752                                          offset + count - 1);
8753
8754         if (iov_iter_rw(iter) == WRITE) {
8755                 /*
8756                  * If the write DIO is beyond the EOF, we need update
8757                  * the isize, but it is protected by i_mutex. So we can
8758                  * not unlock the i_mutex at this case.
8759                  */
8760                 if (offset + count <= inode->i_size) {
8761                         dio_data.overwrite = 1;
8762                         inode_unlock(inode);
8763                         relock = true;
8764                 } else if (iocb->ki_flags & IOCB_NOWAIT) {
8765                         ret = -EAGAIN;
8766                         goto out;
8767                 }
8768                 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
8769                                                    offset, count);
8770                 if (ret)
8771                         goto out;
8772
8773                 /*
8774                  * We need to know how many extents we reserved so that we can
8775                  * do the accounting properly if we go over the number we
8776                  * originally calculated.  Abuse current->journal_info for this.
8777                  */
8778                 dio_data.reserve = round_up(count,
8779                                             fs_info->sectorsize);
8780                 dio_data.unsubmitted_oe_range_start = (u64)offset;
8781                 dio_data.unsubmitted_oe_range_end = (u64)offset;
8782                 current->journal_info = &dio_data;
8783                 down_read(&BTRFS_I(inode)->dio_sem);
8784         } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8785                                      &BTRFS_I(inode)->runtime_flags)) {
8786                 inode_dio_end(inode);
8787                 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8788                 wakeup = false;
8789         }
8790
8791         ret = __blockdev_direct_IO(iocb, inode,
8792                                    fs_info->fs_devices->latest_bdev,
8793                                    iter, btrfs_get_blocks_direct, NULL,
8794                                    btrfs_submit_direct, flags);
8795         if (iov_iter_rw(iter) == WRITE) {
8796                 up_read(&BTRFS_I(inode)->dio_sem);
8797                 current->journal_info = NULL;
8798                 if (ret < 0 && ret != -EIOCBQUEUED) {
8799                         if (dio_data.reserve)
8800                                 btrfs_delalloc_release_space(inode, data_reserved,
8801                                         offset, dio_data.reserve);
8802                         /*
8803                          * On error we might have left some ordered extents
8804                          * without submitting corresponding bios for them, so
8805                          * cleanup them up to avoid other tasks getting them
8806                          * and waiting for them to complete forever.
8807                          */
8808                         if (dio_data.unsubmitted_oe_range_start <
8809                             dio_data.unsubmitted_oe_range_end)
8810                                 __endio_write_update_ordered(inode,
8811                                         dio_data.unsubmitted_oe_range_start,
8812                                         dio_data.unsubmitted_oe_range_end -
8813                                         dio_data.unsubmitted_oe_range_start,
8814                                         false);
8815                 } else if (ret >= 0 && (size_t)ret < count)
8816                         btrfs_delalloc_release_space(inode, data_reserved,
8817                                         offset, count - (size_t)ret);
8818                 btrfs_delalloc_release_extents(BTRFS_I(inode), count);
8819         }
8820 out:
8821         if (wakeup)
8822                 inode_dio_end(inode);
8823         if (relock)
8824                 inode_lock(inode);
8825
8826         extent_changeset_free(data_reserved);
8827         return ret;
8828 }
8829
8830 #define BTRFS_FIEMAP_FLAGS      (FIEMAP_FLAG_SYNC)
8831
8832 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8833                 __u64 start, __u64 len)
8834 {
8835         int     ret;
8836
8837         ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8838         if (ret)
8839                 return ret;
8840
8841         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8842 }
8843
8844 int btrfs_readpage(struct file *file, struct page *page)
8845 {
8846         struct extent_io_tree *tree;
8847         tree = &BTRFS_I(page->mapping->host)->io_tree;
8848         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8849 }
8850
8851 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8852 {
8853         struct extent_io_tree *tree;
8854         struct inode *inode = page->mapping->host;
8855         int ret;
8856
8857         if (current->flags & PF_MEMALLOC) {
8858                 redirty_page_for_writepage(wbc, page);
8859                 unlock_page(page);
8860                 return 0;
8861         }
8862
8863         /*
8864          * If we are under memory pressure we will call this directly from the
8865          * VM, we need to make sure we have the inode referenced for the ordered
8866          * extent.  If not just return like we didn't do anything.
8867          */
8868         if (!igrab(inode)) {
8869                 redirty_page_for_writepage(wbc, page);
8870                 return AOP_WRITEPAGE_ACTIVATE;
8871         }
8872         tree = &BTRFS_I(page->mapping->host)->io_tree;
8873         ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8874         btrfs_add_delayed_iput(inode);
8875         return ret;
8876 }
8877
8878 static int btrfs_writepages(struct address_space *mapping,
8879                             struct writeback_control *wbc)
8880 {
8881         struct extent_io_tree *tree;
8882
8883         tree = &BTRFS_I(mapping->host)->io_tree;
8884         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8885 }
8886
8887 static int
8888 btrfs_readpages(struct file *file, struct address_space *mapping,
8889                 struct list_head *pages, unsigned nr_pages)
8890 {
8891         struct extent_io_tree *tree;
8892         tree = &BTRFS_I(mapping->host)->io_tree;
8893         return extent_readpages(tree, mapping, pages, nr_pages,
8894                                 btrfs_get_extent);
8895 }
8896 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8897 {
8898         struct extent_io_tree *tree;
8899         struct extent_map_tree *map;
8900         int ret;
8901
8902         tree = &BTRFS_I(page->mapping->host)->io_tree;
8903         map = &BTRFS_I(page->mapping->host)->extent_tree;
8904         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8905         if (ret == 1) {
8906                 ClearPagePrivate(page);
8907                 set_page_private(page, 0);
8908                 put_page(page);
8909         }
8910         return ret;
8911 }
8912
8913 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8914 {
8915         if (PageWriteback(page) || PageDirty(page))
8916                 return 0;
8917         return __btrfs_releasepage(page, gfp_flags);
8918 }
8919
8920 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8921                                  unsigned int length)
8922 {
8923         struct inode *inode = page->mapping->host;
8924         struct extent_io_tree *tree;
8925         struct btrfs_ordered_extent *ordered;
8926         struct extent_state *cached_state = NULL;
8927         u64 page_start = page_offset(page);
8928         u64 page_end = page_start + PAGE_SIZE - 1;
8929         u64 start;
8930         u64 end;
8931         int inode_evicting = inode->i_state & I_FREEING;
8932
8933         /*
8934          * we have the page locked, so new writeback can't start,
8935          * and the dirty bit won't be cleared while we are here.
8936          *
8937          * Wait for IO on this page so that we can safely clear
8938          * the PagePrivate2 bit and do ordered accounting
8939          */
8940         wait_on_page_writeback(page);
8941
8942         tree = &BTRFS_I(inode)->io_tree;
8943         if (offset) {
8944                 btrfs_releasepage(page, GFP_NOFS);
8945                 return;
8946         }
8947
8948         if (!inode_evicting)
8949                 lock_extent_bits(tree, page_start, page_end, &cached_state);
8950 again:
8951         start = page_start;
8952         ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
8953                                         page_end - start + 1);
8954         if (ordered) {
8955                 end = min(page_end, ordered->file_offset + ordered->len - 1);
8956                 /*
8957                  * IO on this page will never be started, so we need
8958                  * to account for any ordered extents now
8959                  */
8960                 if (!inode_evicting)
8961                         clear_extent_bit(tree, start, end,
8962                                          EXTENT_DIRTY | EXTENT_DELALLOC |
8963                                          EXTENT_DELALLOC_NEW |
8964                                          EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8965                                          EXTENT_DEFRAG, 1, 0, &cached_state,
8966                                          GFP_NOFS);
8967                 /*
8968                  * whoever cleared the private bit is responsible
8969                  * for the finish_ordered_io
8970                  */
8971                 if (TestClearPagePrivate2(page)) {
8972                         struct btrfs_ordered_inode_tree *tree;
8973                         u64 new_len;
8974
8975                         tree = &BTRFS_I(inode)->ordered_tree;
8976
8977                         spin_lock_irq(&tree->lock);
8978                         set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8979                         new_len = start - ordered->file_offset;
8980                         if (new_len < ordered->truncated_len)
8981                                 ordered->truncated_len = new_len;
8982                         spin_unlock_irq(&tree->lock);
8983
8984                         if (btrfs_dec_test_ordered_pending(inode, &ordered,
8985                                                            start,
8986                                                            end - start + 1, 1))
8987                                 btrfs_finish_ordered_io(ordered);
8988                 }
8989                 btrfs_put_ordered_extent(ordered);
8990                 if (!inode_evicting) {
8991                         cached_state = NULL;
8992                         lock_extent_bits(tree, start, end,
8993                                          &cached_state);
8994                 }
8995
8996                 start = end + 1;
8997                 if (start < page_end)
8998                         goto again;
8999         }
9000
9001         /*
9002          * Qgroup reserved space handler
9003          * Page here will be either
9004          * 1) Already written to disk
9005          *    In this case, its reserved space is released from data rsv map
9006          *    and will be freed by delayed_ref handler finally.
9007          *    So even we call qgroup_free_data(), it won't decrease reserved
9008          *    space.
9009          * 2) Not written to disk
9010          *    This means the reserved space should be freed here. However,
9011          *    if a truncate invalidates the page (by clearing PageDirty)
9012          *    and the page is accounted for while allocating extent
9013          *    in btrfs_check_data_free_space() we let delayed_ref to
9014          *    free the entire extent.
9015          */
9016         if (PageDirty(page))
9017                 btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
9018         if (!inode_evicting) {
9019                 clear_extent_bit(tree, page_start, page_end,
9020                                  EXTENT_LOCKED | EXTENT_DIRTY |
9021                                  EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
9022                                  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
9023                                  &cached_state, GFP_NOFS);
9024
9025                 __btrfs_releasepage(page, GFP_NOFS);
9026         }
9027
9028         ClearPageChecked(page);
9029         if (PagePrivate(page)) {
9030                 ClearPagePrivate(page);
9031                 set_page_private(page, 0);
9032                 put_page(page);
9033         }
9034 }
9035
9036 /*
9037  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
9038  * called from a page fault handler when a page is first dirtied. Hence we must
9039  * be careful to check for EOF conditions here. We set the page up correctly
9040  * for a written page which means we get ENOSPC checking when writing into
9041  * holes and correct delalloc and unwritten extent mapping on filesystems that
9042  * support these features.
9043  *
9044  * We are not allowed to take the i_mutex here so we have to play games to
9045  * protect against truncate races as the page could now be beyond EOF.  Because
9046  * vmtruncate() writes the inode size before removing pages, once we have the
9047  * page lock we can determine safely if the page is beyond EOF. If it is not
9048  * beyond EOF, then the page is guaranteed safe against truncation until we
9049  * unlock the page.
9050  */
9051 int btrfs_page_mkwrite(struct vm_fault *vmf)
9052 {
9053         struct page *page = vmf->page;
9054         struct inode *inode = file_inode(vmf->vma->vm_file);
9055         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9056         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
9057         struct btrfs_ordered_extent *ordered;
9058         struct extent_state *cached_state = NULL;
9059         struct extent_changeset *data_reserved = NULL;
9060         char *kaddr;
9061         unsigned long zero_start;
9062         loff_t size;
9063         int ret;
9064         int reserved = 0;
9065         u64 reserved_space;
9066         u64 page_start;
9067         u64 page_end;
9068         u64 end;
9069
9070         reserved_space = PAGE_SIZE;
9071
9072         sb_start_pagefault(inode->i_sb);
9073         page_start = page_offset(page);
9074         page_end = page_start + PAGE_SIZE - 1;
9075         end = page_end;
9076
9077         /*
9078          * Reserving delalloc space after obtaining the page lock can lead to
9079          * deadlock. For example, if a dirty page is locked by this function
9080          * and the call to btrfs_delalloc_reserve_space() ends up triggering
9081          * dirty page write out, then the btrfs_writepage() function could
9082          * end up waiting indefinitely to get a lock on the page currently
9083          * being processed by btrfs_page_mkwrite() function.
9084          */
9085         ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
9086                                            reserved_space);
9087         if (!ret) {
9088                 ret = file_update_time(vmf->vma->vm_file);
9089                 reserved = 1;
9090         }
9091         if (ret) {
9092                 if (ret == -ENOMEM)
9093                         ret = VM_FAULT_OOM;
9094                 else /* -ENOSPC, -EIO, etc */
9095                         ret = VM_FAULT_SIGBUS;
9096                 if (reserved)
9097                         goto out;
9098                 goto out_noreserve;
9099         }
9100
9101         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
9102 again:
9103         lock_page(page);
9104         size = i_size_read(inode);
9105
9106         if ((page->mapping != inode->i_mapping) ||
9107             (page_start >= size)) {
9108                 /* page got truncated out from underneath us */
9109                 goto out_unlock;
9110         }
9111         wait_on_page_writeback(page);
9112
9113         lock_extent_bits(io_tree, page_start, page_end, &cached_state);
9114         set_page_extent_mapped(page);
9115
9116         /*
9117          * we can't set the delalloc bits if there are pending ordered
9118          * extents.  Drop our locks and wait for them to finish
9119          */
9120         ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
9121                         PAGE_SIZE);
9122         if (ordered) {
9123                 unlock_extent_cached(io_tree, page_start, page_end,
9124                                      &cached_state, GFP_NOFS);
9125                 unlock_page(page);
9126                 btrfs_start_ordered_extent(inode, ordered, 1);
9127                 btrfs_put_ordered_extent(ordered);
9128                 goto again;
9129         }
9130
9131         if (page->index == ((size - 1) >> PAGE_SHIFT)) {
9132                 reserved_space = round_up(size - page_start,
9133                                           fs_info->sectorsize);
9134                 if (reserved_space < PAGE_SIZE) {
9135                         end = page_start + reserved_space - 1;
9136                         btrfs_delalloc_release_space(inode, data_reserved,
9137                                         page_start, PAGE_SIZE - reserved_space);
9138                 }
9139         }
9140
9141         /*
9142          * page_mkwrite gets called when the page is firstly dirtied after it's
9143          * faulted in, but write(2) could also dirty a page and set delalloc
9144          * bits, thus in this case for space account reason, we still need to
9145          * clear any delalloc bits within this page range since we have to
9146          * reserve data&meta space before lock_page() (see above comments).
9147          */
9148         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
9149                           EXTENT_DIRTY | EXTENT_DELALLOC |
9150                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
9151                           0, 0, &cached_state, GFP_NOFS);
9152
9153         ret = btrfs_set_extent_delalloc(inode, page_start, end,
9154                                         &cached_state, 0);
9155         if (ret) {
9156                 unlock_extent_cached(io_tree, page_start, page_end,
9157                                      &cached_state, GFP_NOFS);
9158                 ret = VM_FAULT_SIGBUS;
9159                 goto out_unlock;
9160         }
9161         ret = 0;
9162
9163         /* page is wholly or partially inside EOF */
9164         if (page_start + PAGE_SIZE > size)
9165                 zero_start = size & ~PAGE_MASK;
9166         else
9167                 zero_start = PAGE_SIZE;
9168
9169         if (zero_start != PAGE_SIZE) {
9170                 kaddr = kmap(page);
9171                 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
9172                 flush_dcache_page(page);
9173                 kunmap(page);
9174         }
9175         ClearPageChecked(page);
9176         set_page_dirty(page);
9177         SetPageUptodate(page);
9178
9179         BTRFS_I(inode)->last_trans = fs_info->generation;
9180         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
9181         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
9182
9183         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9184
9185 out_unlock:
9186         if (!ret) {
9187                 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
9188                 sb_end_pagefault(inode->i_sb);
9189                 extent_changeset_free(data_reserved);
9190                 return VM_FAULT_LOCKED;
9191         }
9192         unlock_page(page);
9193 out:
9194         btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
9195         btrfs_delalloc_release_space(inode, data_reserved, page_start,
9196                                      reserved_space);
9197 out_noreserve:
9198         sb_end_pagefault(inode->i_sb);
9199         extent_changeset_free(data_reserved);
9200         return ret;
9201 }
9202
9203 static int btrfs_truncate(struct inode *inode)
9204 {
9205         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9206         struct btrfs_root *root = BTRFS_I(inode)->root;
9207         struct btrfs_block_rsv *rsv;
9208         int ret = 0;
9209         int err = 0;
9210         struct btrfs_trans_handle *trans;
9211         u64 mask = fs_info->sectorsize - 1;
9212         u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
9213
9214         ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
9215                                        (u64)-1);
9216         if (ret)
9217                 return ret;
9218
9219         /*
9220          * Yes ladies and gentlemen, this is indeed ugly.  The fact is we have
9221          * 3 things going on here
9222          *
9223          * 1) We need to reserve space for our orphan item and the space to
9224          * delete our orphan item.  Lord knows we don't want to have a dangling
9225          * orphan item because we didn't reserve space to remove it.
9226          *
9227          * 2) We need to reserve space to update our inode.
9228          *
9229          * 3) We need to have something to cache all the space that is going to
9230          * be free'd up by the truncate operation, but also have some slack
9231          * space reserved in case it uses space during the truncate (thank you
9232          * very much snapshotting).
9233          *
9234          * And we need these to all be separate.  The fact is we can use a lot of
9235          * space doing the truncate, and we have no earthly idea how much space
9236          * we will use, so we need the truncate reservation to be separate so it
9237          * doesn't end up using space reserved for updating the inode or
9238          * removing the orphan item.  We also need to be able to stop the
9239          * transaction and start a new one, which means we need to be able to
9240          * update the inode several times, and we have no idea of knowing how
9241          * many times that will be, so we can't just reserve 1 item for the
9242          * entirety of the operation, so that has to be done separately as well.
9243          * Then there is the orphan item, which does indeed need to be held on
9244          * to for the whole operation, and we need nobody to touch this reserved
9245          * space except the orphan code.
9246          *
9247          * So that leaves us with
9248          *
9249          * 1) root->orphan_block_rsv - for the orphan deletion.
9250          * 2) rsv - for the truncate reservation, which we will steal from the
9251          * transaction reservation.
9252          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9253          * updating the inode.
9254          */
9255         rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
9256         if (!rsv)
9257                 return -ENOMEM;
9258         rsv->size = min_size;
9259         rsv->failfast = 1;
9260
9261         /*
9262          * 1 for the truncate slack space
9263          * 1 for updating the inode.
9264          */
9265         trans = btrfs_start_transaction(root, 2);
9266         if (IS_ERR(trans)) {
9267                 err = PTR_ERR(trans);
9268                 goto out;
9269         }
9270
9271         /* Migrate the slack space for the truncate to our reserve */
9272         ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
9273                                       min_size, 0);
9274         BUG_ON(ret);
9275
9276         /*
9277          * So if we truncate and then write and fsync we normally would just
9278          * write the extents that changed, which is a problem if we need to
9279          * first truncate that entire inode.  So set this flag so we write out
9280          * all of the extents in the inode to the sync log so we're completely
9281          * safe.
9282          */
9283         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9284         trans->block_rsv = rsv;
9285
9286         while (1) {
9287                 ret = btrfs_truncate_inode_items(trans, root, inode,
9288                                                  inode->i_size,
9289                                                  BTRFS_EXTENT_DATA_KEY);
9290                 trans->block_rsv = &fs_info->trans_block_rsv;
9291                 if (ret != -ENOSPC && ret != -EAGAIN) {
9292                         err = ret;
9293                         break;
9294                 }
9295
9296                 ret = btrfs_update_inode(trans, root, inode);
9297                 if (ret) {
9298                         err = ret;
9299                         break;
9300                 }
9301
9302                 btrfs_end_transaction(trans);
9303                 btrfs_btree_balance_dirty(fs_info);
9304
9305                 trans = btrfs_start_transaction(root, 2);
9306                 if (IS_ERR(trans)) {
9307                         ret = err = PTR_ERR(trans);
9308                         trans = NULL;
9309                         break;
9310                 }
9311
9312                 btrfs_block_rsv_release(fs_info, rsv, -1);
9313                 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
9314                                               rsv, min_size, 0);
9315                 BUG_ON(ret);    /* shouldn't happen */
9316                 trans->block_rsv = rsv;
9317         }
9318
9319         /*
9320          * We can't call btrfs_truncate_block inside a trans handle as we could
9321          * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
9322          * we've truncated everything except the last little bit, and can do
9323          * btrfs_truncate_block and then update the disk_i_size.
9324          */
9325         if (ret == NEED_TRUNCATE_BLOCK) {
9326                 btrfs_end_transaction(trans);
9327                 btrfs_btree_balance_dirty(fs_info);
9328
9329                 ret = btrfs_truncate_block(inode, inode->i_size, 0, 0);
9330                 if (ret)
9331                         goto out;
9332                 trans = btrfs_start_transaction(root, 1);
9333                 if (IS_ERR(trans)) {
9334                         ret = PTR_ERR(trans);
9335                         goto out;
9336                 }
9337                 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
9338         }
9339
9340         if (ret == 0 && inode->i_nlink > 0) {
9341                 trans->block_rsv = root->orphan_block_rsv;
9342                 ret = btrfs_orphan_del(trans, BTRFS_I(inode));
9343                 if (ret)
9344                         err = ret;
9345         }
9346
9347         if (trans) {
9348                 trans->block_rsv = &fs_info->trans_block_rsv;
9349                 ret = btrfs_update_inode(trans, root, inode);
9350                 if (ret && !err)
9351                         err = ret;
9352
9353                 ret = btrfs_end_transaction(trans);
9354                 btrfs_btree_balance_dirty(fs_info);
9355         }
9356 out:
9357         btrfs_free_block_rsv(fs_info, rsv);
9358
9359         if (ret && !err)
9360                 err = ret;
9361
9362         return err;
9363 }
9364
9365 /*
9366  * create a new subvolume directory/inode (helper for the ioctl).
9367  */
9368 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9369                              struct btrfs_root *new_root,
9370                              struct btrfs_root *parent_root,
9371                              u64 new_dirid)
9372 {
9373         struct inode *inode;
9374         int err;
9375         u64 index = 0;
9376
9377         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9378                                 new_dirid, new_dirid,
9379                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
9380                                 &index);
9381         if (IS_ERR(inode))
9382                 return PTR_ERR(inode);
9383         inode->i_op = &btrfs_dir_inode_operations;
9384         inode->i_fop = &btrfs_dir_file_operations;
9385
9386         set_nlink(inode, 1);
9387         btrfs_i_size_write(BTRFS_I(inode), 0);
9388         unlock_new_inode(inode);
9389
9390         err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9391         if (err)
9392                 btrfs_err(new_root->fs_info,
9393                           "error inheriting subvolume %llu properties: %d",
9394                           new_root->root_key.objectid, err);
9395
9396         err = btrfs_update_inode(trans, new_root, inode);
9397
9398         iput(inode);
9399         return err;
9400 }
9401
9402 struct inode *btrfs_alloc_inode(struct super_block *sb)
9403 {
9404         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
9405         struct btrfs_inode *ei;
9406         struct inode *inode;
9407
9408         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9409         if (!ei)
9410                 return NULL;
9411
9412         ei->root = NULL;
9413         ei->generation = 0;
9414         ei->last_trans = 0;
9415         ei->last_sub_trans = 0;
9416         ei->logged_trans = 0;
9417         ei->delalloc_bytes = 0;
9418         ei->new_delalloc_bytes = 0;
9419         ei->defrag_bytes = 0;
9420         ei->disk_i_size = 0;
9421         ei->flags = 0;
9422         ei->csum_bytes = 0;
9423         ei->index_cnt = (u64)-1;
9424         ei->dir_index = 0;
9425         ei->last_unlink_trans = 0;
9426         ei->last_log_commit = 0;
9427         ei->delayed_iput_count = 0;
9428
9429         spin_lock_init(&ei->lock);
9430         ei->outstanding_extents = 0;
9431         if (sb->s_magic != BTRFS_TEST_MAGIC)
9432                 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
9433                                               BTRFS_BLOCK_RSV_DELALLOC);
9434         ei->runtime_flags = 0;
9435         ei->prop_compress = BTRFS_COMPRESS_NONE;
9436         ei->defrag_compress = BTRFS_COMPRESS_NONE;
9437
9438         ei->delayed_node = NULL;
9439
9440         ei->i_otime.tv_sec = 0;
9441         ei->i_otime.tv_nsec = 0;
9442
9443         inode = &ei->vfs_inode;
9444         extent_map_tree_init(&ei->extent_tree);
9445         extent_io_tree_init(&ei->io_tree, inode);
9446         extent_io_tree_init(&ei->io_failure_tree, inode);
9447         ei->io_tree.track_uptodate = 1;
9448         ei->io_failure_tree.track_uptodate = 1;
9449         atomic_set(&ei->sync_writers, 0);
9450         mutex_init(&ei->log_mutex);
9451         mutex_init(&ei->delalloc_mutex);
9452         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9453         INIT_LIST_HEAD(&ei->delalloc_inodes);
9454         INIT_LIST_HEAD(&ei->delayed_iput);
9455         RB_CLEAR_NODE(&ei->rb_node);
9456         init_rwsem(&ei->dio_sem);
9457
9458         return inode;
9459 }
9460
9461 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9462 void btrfs_test_destroy_inode(struct inode *inode)
9463 {
9464         btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9465         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9466 }
9467 #endif
9468
9469 static void btrfs_i_callback(struct rcu_head *head)
9470 {
9471         struct inode *inode = container_of(head, struct inode, i_rcu);
9472         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9473 }
9474
9475 void btrfs_destroy_inode(struct inode *inode)
9476 {
9477         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9478         struct btrfs_ordered_extent *ordered;
9479         struct btrfs_root *root = BTRFS_I(inode)->root;
9480
9481         WARN_ON(!hlist_empty(&inode->i_dentry));
9482         WARN_ON(inode->i_data.nrpages);
9483         WARN_ON(BTRFS_I(inode)->block_rsv.reserved);
9484         WARN_ON(BTRFS_I(inode)->block_rsv.size);
9485         WARN_ON(BTRFS_I(inode)->outstanding_extents);
9486         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9487         WARN_ON(BTRFS_I(inode)->new_delalloc_bytes);
9488         WARN_ON(BTRFS_I(inode)->csum_bytes);
9489         WARN_ON(BTRFS_I(inode)->defrag_bytes);
9490
9491         /*
9492          * This can happen where we create an inode, but somebody else also
9493          * created the same inode and we need to destroy the one we already
9494          * created.
9495          */
9496         if (!root)
9497                 goto free;
9498
9499         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9500                      &BTRFS_I(inode)->runtime_flags)) {
9501                 btrfs_info(fs_info, "inode %llu still on the orphan list",
9502                            btrfs_ino(BTRFS_I(inode)));
9503                 atomic_dec(&root->orphan_inodes);
9504         }
9505
9506         while (1) {
9507                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9508                 if (!ordered)
9509                         break;
9510                 else {
9511                         btrfs_err(fs_info,
9512                                   "found ordered extent %llu %llu on inode cleanup",
9513                                   ordered->file_offset, ordered->len);
9514                         btrfs_remove_ordered_extent(inode, ordered);
9515                         btrfs_put_ordered_extent(ordered);
9516                         btrfs_put_ordered_extent(ordered);
9517                 }
9518         }
9519         btrfs_qgroup_check_reserved_leak(inode);
9520         inode_tree_del(inode);
9521         btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9522 free:
9523         call_rcu(&inode->i_rcu, btrfs_i_callback);
9524 }
9525
9526 int btrfs_drop_inode(struct inode *inode)
9527 {
9528         struct btrfs_root *root = BTRFS_I(inode)->root;
9529
9530         if (root == NULL)
9531                 return 1;
9532
9533         /* the snap/subvol tree is on deleting */
9534         if (btrfs_root_refs(&root->root_item) == 0)
9535                 return 1;
9536         else
9537                 return generic_drop_inode(inode);
9538 }
9539
9540 static void init_once(void *foo)
9541 {
9542         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9543
9544         inode_init_once(&ei->vfs_inode);
9545 }
9546
9547 void btrfs_destroy_cachep(void)
9548 {
9549         /*
9550          * Make sure all delayed rcu free inodes are flushed before we
9551          * destroy cache.
9552          */
9553         rcu_barrier();
9554         kmem_cache_destroy(btrfs_inode_cachep);
9555         kmem_cache_destroy(btrfs_trans_handle_cachep);
9556         kmem_cache_destroy(btrfs_path_cachep);
9557         kmem_cache_destroy(btrfs_free_space_cachep);
9558 }
9559
9560 int btrfs_init_cachep(void)
9561 {
9562         btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9563                         sizeof(struct btrfs_inode), 0,
9564                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9565                         init_once);
9566         if (!btrfs_inode_cachep)
9567                 goto fail;
9568
9569         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9570                         sizeof(struct btrfs_trans_handle), 0,
9571                         SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9572         if (!btrfs_trans_handle_cachep)
9573                 goto fail;
9574
9575         btrfs_path_cachep = kmem_cache_create("btrfs_path",
9576                         sizeof(struct btrfs_path), 0,
9577                         SLAB_MEM_SPREAD, NULL);
9578         if (!btrfs_path_cachep)
9579                 goto fail;
9580
9581         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9582                         sizeof(struct btrfs_free_space), 0,
9583                         SLAB_MEM_SPREAD, NULL);
9584         if (!btrfs_free_space_cachep)
9585                 goto fail;
9586
9587         return 0;
9588 fail:
9589         btrfs_destroy_cachep();
9590         return -ENOMEM;
9591 }
9592
9593 static int btrfs_getattr(const struct path *path, struct kstat *stat,
9594                          u32 request_mask, unsigned int flags)
9595 {
9596         u64 delalloc_bytes;
9597         struct inode *inode = d_inode(path->dentry);
9598         u32 blocksize = inode->i_sb->s_blocksize;
9599         u32 bi_flags = BTRFS_I(inode)->flags;
9600
9601         stat->result_mask |= STATX_BTIME;
9602         stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9603         stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9604         if (bi_flags & BTRFS_INODE_APPEND)
9605                 stat->attributes |= STATX_ATTR_APPEND;
9606         if (bi_flags & BTRFS_INODE_COMPRESS)
9607                 stat->attributes |= STATX_ATTR_COMPRESSED;
9608         if (bi_flags & BTRFS_INODE_IMMUTABLE)
9609                 stat->attributes |= STATX_ATTR_IMMUTABLE;
9610         if (bi_flags & BTRFS_INODE_NODUMP)
9611                 stat->attributes |= STATX_ATTR_NODUMP;
9612
9613         stat->attributes_mask |= (STATX_ATTR_APPEND |
9614                                   STATX_ATTR_COMPRESSED |
9615                                   STATX_ATTR_IMMUTABLE |
9616                                   STATX_ATTR_NODUMP);
9617
9618         generic_fillattr(inode, stat);
9619         stat->dev = BTRFS_I(inode)->root->anon_dev;
9620
9621         spin_lock(&BTRFS_I(inode)->lock);
9622         delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9623         spin_unlock(&BTRFS_I(inode)->lock);
9624         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9625                         ALIGN(delalloc_bytes, blocksize)) >> 9;
9626         return 0;
9627 }
9628
9629 static int btrfs_rename_exchange(struct inode *old_dir,
9630                               struct dentry *old_dentry,
9631                               struct inode *new_dir,
9632                               struct dentry *new_dentry)
9633 {
9634         struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9635         struct btrfs_trans_handle *trans;
9636         struct btrfs_root *root = BTRFS_I(old_dir)->root;
9637         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9638         struct inode *new_inode = new_dentry->d_inode;
9639         struct inode *old_inode = old_dentry->d_inode;
9640         struct timespec ctime = current_time(old_inode);
9641         struct dentry *parent;
9642         u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9643         u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9644         u64 old_idx = 0;
9645         u64 new_idx = 0;
9646         u64 root_objectid;
9647         int ret;
9648         bool root_log_pinned = false;
9649         bool dest_log_pinned = false;
9650
9651         /* we only allow rename subvolume link between subvolumes */
9652         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9653                 return -EXDEV;
9654
9655         /* close the race window with snapshot create/destroy ioctl */
9656         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9657                 down_read(&fs_info->subvol_sem);
9658         if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9659                 down_read(&fs_info->subvol_sem);
9660
9661         /*
9662          * We want to reserve the absolute worst case amount of items.  So if
9663          * both inodes are subvols and we need to unlink them then that would
9664          * require 4 item modifications, but if they are both normal inodes it
9665          * would require 5 item modifications, so we'll assume their normal
9666          * inodes.  So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9667          * should cover the worst case number of items we'll modify.
9668          */
9669         trans = btrfs_start_transaction(root, 12);
9670         if (IS_ERR(trans)) {
9671                 ret = PTR_ERR(trans);
9672                 goto out_notrans;
9673         }
9674
9675         /*
9676          * We need to find a free sequence number both in the source and
9677          * in the destination directory for the exchange.
9678          */
9679         ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9680         if (ret)
9681                 goto out_fail;
9682         ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9683         if (ret)
9684                 goto out_fail;
9685
9686         BTRFS_I(old_inode)->dir_index = 0ULL;
9687         BTRFS_I(new_inode)->dir_index = 0ULL;
9688
9689         /* Reference for the source. */
9690         if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9691                 /* force full log commit if subvolume involved. */
9692                 btrfs_set_log_full_commit(fs_info, trans);
9693         } else {
9694                 btrfs_pin_log_trans(root);
9695                 root_log_pinned = true;
9696                 ret = btrfs_insert_inode_ref(trans, dest,
9697                                              new_dentry->d_name.name,
9698                                              new_dentry->d_name.len,
9699                                              old_ino,
9700                                              btrfs_ino(BTRFS_I(new_dir)),
9701                                              old_idx);
9702                 if (ret)
9703                         goto out_fail;
9704         }
9705
9706         /* And now for the dest. */
9707         if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9708                 /* force full log commit if subvolume involved. */
9709                 btrfs_set_log_full_commit(fs_info, trans);
9710         } else {
9711                 btrfs_pin_log_trans(dest);
9712                 dest_log_pinned = true;
9713                 ret = btrfs_insert_inode_ref(trans, root,
9714                                              old_dentry->d_name.name,
9715                                              old_dentry->d_name.len,
9716                                              new_ino,
9717                                              btrfs_ino(BTRFS_I(old_dir)),
9718                                              new_idx);
9719                 if (ret)
9720                         goto out_fail;
9721         }
9722
9723         /* Update inode version and ctime/mtime. */
9724         inode_inc_iversion(old_dir);
9725         inode_inc_iversion(new_dir);
9726         inode_inc_iversion(old_inode);
9727         inode_inc_iversion(new_inode);
9728         old_dir->i_ctime = old_dir->i_mtime = ctime;
9729         new_dir->i_ctime = new_dir->i_mtime = ctime;
9730         old_inode->i_ctime = ctime;
9731         new_inode->i_ctime = ctime;
9732
9733         if (old_dentry->d_parent != new_dentry->d_parent) {
9734                 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9735                                 BTRFS_I(old_inode), 1);
9736                 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9737                                 BTRFS_I(new_inode), 1);
9738         }
9739
9740         /* src is a subvolume */
9741         if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9742                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9743                 ret = btrfs_unlink_subvol(trans, root, old_dir,
9744                                           root_objectid,
9745                                           old_dentry->d_name.name,
9746                                           old_dentry->d_name.len);
9747         } else { /* src is an inode */
9748                 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9749                                            BTRFS_I(old_dentry->d_inode),
9750                                            old_dentry->d_name.name,
9751                                            old_dentry->d_name.len);
9752                 if (!ret)
9753                         ret = btrfs_update_inode(trans, root, old_inode);
9754         }
9755         if (ret) {
9756                 btrfs_abort_transaction(trans, ret);
9757                 goto out_fail;
9758         }
9759
9760         /* dest is a subvolume */
9761         if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9762                 root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9763                 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9764                                           root_objectid,
9765                                           new_dentry->d_name.name,
9766                                           new_dentry->d_name.len);
9767         } else { /* dest is an inode */
9768                 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9769                                            BTRFS_I(new_dentry->d_inode),
9770                                            new_dentry->d_name.name,
9771                                            new_dentry->d_name.len);
9772                 if (!ret)
9773                         ret = btrfs_update_inode(trans, dest, new_inode);
9774         }
9775         if (ret) {
9776                 btrfs_abort_transaction(trans, ret);
9777                 goto out_fail;
9778         }
9779
9780         ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9781                              new_dentry->d_name.name,
9782                              new_dentry->d_name.len, 0, old_idx);
9783         if (ret) {
9784                 btrfs_abort_transaction(trans, ret);
9785                 goto out_fail;
9786         }
9787
9788         ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9789                              old_dentry->d_name.name,
9790                              old_dentry->d_name.len, 0, new_idx);
9791         if (ret) {
9792                 btrfs_abort_transaction(trans, ret);
9793                 goto out_fail;
9794         }
9795
9796         if (old_inode->i_nlink == 1)
9797                 BTRFS_I(old_inode)->dir_index = old_idx;
9798         if (new_inode->i_nlink == 1)
9799                 BTRFS_I(new_inode)->dir_index = new_idx;
9800
9801         if (root_log_pinned) {
9802                 parent = new_dentry->d_parent;
9803                 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9804                                 parent);
9805                 btrfs_end_log_trans(root);
9806                 root_log_pinned = false;
9807         }
9808         if (dest_log_pinned) {
9809                 parent = old_dentry->d_parent;
9810                 btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir),
9811                                 parent);
9812                 btrfs_end_log_trans(dest);
9813                 dest_log_pinned = false;
9814         }
9815 out_fail:
9816         /*
9817          * If we have pinned a log and an error happened, we unpin tasks
9818          * trying to sync the log and force them to fallback to a transaction
9819          * commit if the log currently contains any of the inodes involved in
9820          * this rename operation (to ensure we do not persist a log with an
9821          * inconsistent state for any of these inodes or leading to any
9822          * inconsistencies when replayed). If the transaction was aborted, the
9823          * abortion reason is propagated to userspace when attempting to commit
9824          * the transaction. If the log does not contain any of these inodes, we
9825          * allow the tasks to sync it.
9826          */
9827         if (ret && (root_log_pinned || dest_log_pinned)) {
9828                 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9829                     btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9830                     btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9831                     (new_inode &&
9832                      btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9833                         btrfs_set_log_full_commit(fs_info, trans);
9834
9835                 if (root_log_pinned) {
9836                         btrfs_end_log_trans(root);
9837                         root_log_pinned = false;
9838                 }
9839                 if (dest_log_pinned) {
9840                         btrfs_end_log_trans(dest);
9841                         dest_log_pinned = false;
9842                 }
9843         }
9844         ret = btrfs_end_transaction(trans);
9845 out_notrans:
9846         if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9847                 up_read(&fs_info->subvol_sem);
9848         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9849                 up_read(&fs_info->subvol_sem);
9850
9851         return ret;
9852 }
9853
9854 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9855                                      struct btrfs_root *root,
9856                                      struct inode *dir,
9857                                      struct dentry *dentry)
9858 {
9859         int ret;
9860         struct inode *inode;
9861         u64 objectid;
9862         u64 index;
9863
9864         ret = btrfs_find_free_ino(root, &objectid);
9865         if (ret)
9866                 return ret;
9867
9868         inode = btrfs_new_inode(trans, root, dir,
9869                                 dentry->d_name.name,
9870                                 dentry->d_name.len,
9871                                 btrfs_ino(BTRFS_I(dir)),
9872                                 objectid,
9873                                 S_IFCHR | WHITEOUT_MODE,
9874                                 &index);
9875
9876         if (IS_ERR(inode)) {
9877                 ret = PTR_ERR(inode);
9878                 return ret;
9879         }
9880
9881         inode->i_op = &btrfs_special_inode_operations;
9882         init_special_inode(inode, inode->i_mode,
9883                 WHITEOUT_DEV);
9884
9885         ret = btrfs_init_inode_security(trans, inode, dir,
9886                                 &dentry->d_name);
9887         if (ret)
9888                 goto out;
9889
9890         ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9891                                 BTRFS_I(inode), 0, index);
9892         if (ret)
9893                 goto out;
9894
9895         ret = btrfs_update_inode(trans, root, inode);
9896 out:
9897         unlock_new_inode(inode);
9898         if (ret)
9899                 inode_dec_link_count(inode);
9900         iput(inode);
9901
9902         return ret;
9903 }
9904
9905 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9906                            struct inode *new_dir, struct dentry *new_dentry,
9907                            unsigned int flags)
9908 {
9909         struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9910         struct btrfs_trans_handle *trans;
9911         unsigned int trans_num_items;
9912         struct btrfs_root *root = BTRFS_I(old_dir)->root;
9913         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9914         struct inode *new_inode = d_inode(new_dentry);
9915         struct inode *old_inode = d_inode(old_dentry);
9916         u64 index = 0;
9917         u64 root_objectid;
9918         int ret;
9919         u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9920         bool log_pinned = false;
9921
9922         if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9923                 return -EPERM;
9924
9925         /* we only allow rename subvolume link between subvolumes */
9926         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9927                 return -EXDEV;
9928
9929         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9930             (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9931                 return -ENOTEMPTY;
9932
9933         if (S_ISDIR(old_inode->i_mode) && new_inode &&
9934             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9935                 return -ENOTEMPTY;
9936
9937
9938         /* check for collisions, even if the  name isn't there */
9939         ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9940                              new_dentry->d_name.name,
9941                              new_dentry->d_name.len);
9942
9943         if (ret) {
9944                 if (ret == -EEXIST) {
9945                         /* we shouldn't get
9946                          * eexist without a new_inode */
9947                         if (WARN_ON(!new_inode)) {
9948                                 return ret;
9949                         }
9950                 } else {
9951                         /* maybe -EOVERFLOW */
9952                         return ret;
9953                 }
9954         }
9955         ret = 0;
9956
9957         /*
9958          * we're using rename to replace one file with another.  Start IO on it
9959          * now so  we don't add too much work to the end of the transaction
9960          */
9961         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9962                 filemap_flush(old_inode->i_mapping);
9963
9964         /* close the racy window with snapshot create/destroy ioctl */
9965         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9966                 down_read(&fs_info->subvol_sem);
9967         /*
9968          * We want to reserve the absolute worst case amount of items.  So if
9969          * both inodes are subvols and we need to unlink them then that would
9970          * require 4 item modifications, but if they are both normal inodes it
9971          * would require 5 item modifications, so we'll assume they are normal
9972          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9973          * should cover the worst case number of items we'll modify.
9974          * If our rename has the whiteout flag, we need more 5 units for the
9975          * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9976          * when selinux is enabled).
9977          */
9978         trans_num_items = 11;
9979         if (flags & RENAME_WHITEOUT)
9980                 trans_num_items += 5;
9981         trans = btrfs_start_transaction(root, trans_num_items);
9982         if (IS_ERR(trans)) {
9983                 ret = PTR_ERR(trans);
9984                 goto out_notrans;
9985         }
9986
9987         if (dest != root)
9988                 btrfs_record_root_in_trans(trans, dest);
9989
9990         ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9991         if (ret)
9992                 goto out_fail;
9993
9994         BTRFS_I(old_inode)->dir_index = 0ULL;
9995         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9996                 /* force full log commit if subvolume involved. */
9997                 btrfs_set_log_full_commit(fs_info, trans);
9998         } else {
9999                 btrfs_pin_log_trans(root);
10000                 log_pinned = true;
10001                 ret = btrfs_insert_inode_ref(trans, dest,
10002                                              new_dentry->d_name.name,
10003                                              new_dentry->d_name.len,
10004                                              old_ino,
10005                                              btrfs_ino(BTRFS_I(new_dir)), index);
10006                 if (ret)
10007                         goto out_fail;
10008         }
10009
10010         inode_inc_iversion(old_dir);
10011         inode_inc_iversion(new_dir);
10012         inode_inc_iversion(old_inode);
10013         old_dir->i_ctime = old_dir->i_mtime =
10014         new_dir->i_ctime = new_dir->i_mtime =
10015         old_inode->i_ctime = current_time(old_dir);
10016
10017         if (old_dentry->d_parent != new_dentry->d_parent)
10018                 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
10019                                 BTRFS_I(old_inode), 1);
10020
10021         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
10022                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
10023                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
10024                                         old_dentry->d_name.name,
10025                                         old_dentry->d_name.len);
10026         } else {
10027                 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
10028                                         BTRFS_I(d_inode(old_dentry)),
10029                                         old_dentry->d_name.name,
10030                                         old_dentry->d_name.len);
10031                 if (!ret)
10032                         ret = btrfs_update_inode(trans, root, old_inode);
10033         }
10034         if (ret) {
10035                 btrfs_abort_transaction(trans, ret);
10036                 goto out_fail;
10037         }
10038
10039         if (new_inode) {
10040                 inode_inc_iversion(new_inode);
10041                 new_inode->i_ctime = current_time(new_inode);
10042                 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
10043                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
10044                         root_objectid = BTRFS_I(new_inode)->location.objectid;
10045                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
10046                                                 root_objectid,
10047                                                 new_dentry->d_name.name,
10048                                                 new_dentry->d_name.len);
10049                         BUG_ON(new_inode->i_nlink == 0);
10050                 } else {
10051                         ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
10052                                                  BTRFS_I(d_inode(new_dentry)),
10053                                                  new_dentry->d_name.name,
10054                                                  new_dentry->d_name.len);
10055                 }
10056                 if (!ret && new_inode->i_nlink == 0)
10057                         ret = btrfs_orphan_add(trans,
10058                                         BTRFS_I(d_inode(new_dentry)));
10059                 if (ret) {
10060                         btrfs_abort_transaction(trans, ret);
10061                         goto out_fail;
10062                 }
10063         }
10064
10065         ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
10066                              new_dentry->d_name.name,
10067                              new_dentry->d_name.len, 0, index);
10068         if (ret) {
10069                 btrfs_abort_transaction(trans, ret);
10070                 goto out_fail;
10071         }
10072
10073         if (old_inode->i_nlink == 1)
10074                 BTRFS_I(old_inode)->dir_index = index;
10075
10076         if (log_pinned) {
10077                 struct dentry *parent = new_dentry->d_parent;
10078
10079                 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
10080                                 parent);
10081                 btrfs_end_log_trans(root);
10082                 log_pinned = false;
10083         }
10084
10085         if (flags & RENAME_WHITEOUT) {
10086                 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
10087                                                 old_dentry);
10088
10089                 if (ret) {
10090                         btrfs_abort_transaction(trans, ret);
10091                         goto out_fail;
10092                 }
10093         }
10094 out_fail:
10095         /*
10096          * If we have pinned the log and an error happened, we unpin tasks
10097          * trying to sync the log and force them to fallback to a transaction
10098          * commit if the log currently contains any of the inodes involved in
10099          * this rename operation (to ensure we do not persist a log with an
10100          * inconsistent state for any of these inodes or leading to any
10101          * inconsistencies when replayed). If the transaction was aborted, the
10102          * abortion reason is propagated to userspace when attempting to commit
10103          * the transaction. If the log does not contain any of these inodes, we
10104          * allow the tasks to sync it.
10105          */
10106         if (ret && log_pinned) {
10107                 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
10108                     btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
10109                     btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
10110                     (new_inode &&
10111                      btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
10112                         btrfs_set_log_full_commit(fs_info, trans);
10113
10114                 btrfs_end_log_trans(root);
10115                 log_pinned = false;
10116         }
10117         btrfs_end_transaction(trans);
10118 out_notrans:
10119         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
10120                 up_read(&fs_info->subvol_sem);
10121
10122         return ret;
10123 }
10124
10125 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
10126                          struct inode *new_dir, struct dentry *new_dentry,
10127                          unsigned int flags)
10128 {
10129         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
10130                 return -EINVAL;
10131
10132         if (flags & RENAME_EXCHANGE)
10133                 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
10134                                           new_dentry);
10135
10136         return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
10137 }
10138
10139 static void btrfs_run_delalloc_work(struct btrfs_work *work)
10140 {
10141         struct btrfs_delalloc_work *delalloc_work;
10142         struct inode *inode;
10143
10144         delalloc_work = container_of(work, struct btrfs_delalloc_work,
10145                                      work);
10146         inode = delalloc_work->inode;
10147         filemap_flush(inode->i_mapping);
10148         if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
10149                                 &BTRFS_I(inode)->runtime_flags))
10150                 filemap_flush(inode->i_mapping);
10151
10152         if (delalloc_work->delay_iput)
10153                 btrfs_add_delayed_iput(inode);
10154         else
10155                 iput(inode);
10156         complete(&delalloc_work->completion);
10157 }
10158
10159 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
10160                                                     int delay_iput)
10161 {
10162         struct btrfs_delalloc_work *work;
10163
10164         work = kmalloc(sizeof(*work), GFP_NOFS);
10165         if (!work)
10166                 return NULL;
10167
10168         init_completion(&work->completion);
10169         INIT_LIST_HEAD(&work->list);
10170         work->inode = inode;
10171         work->delay_iput = delay_iput;
10172         WARN_ON_ONCE(!inode);
10173         btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
10174                         btrfs_run_delalloc_work, NULL, NULL);
10175
10176         return work;
10177 }
10178
10179 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
10180 {
10181         wait_for_completion(&work->completion);
10182         kfree(work);
10183 }
10184
10185 /*
10186  * some fairly slow code that needs optimization. This walks the list
10187  * of all the inodes with pending delalloc and forces them to disk.
10188  */
10189 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
10190                                    int nr)
10191 {
10192         struct btrfs_inode *binode;
10193         struct inode *inode;
10194         struct btrfs_delalloc_work *work, *next;
10195         struct list_head works;
10196         struct list_head splice;
10197         int ret = 0;
10198
10199         INIT_LIST_HEAD(&works);
10200         INIT_LIST_HEAD(&splice);
10201
10202         mutex_lock(&root->delalloc_mutex);
10203         spin_lock(&root->delalloc_lock);
10204         list_splice_init(&root->delalloc_inodes, &splice);
10205         while (!list_empty(&splice)) {
10206                 binode = list_entry(splice.next, struct btrfs_inode,
10207                                     delalloc_inodes);
10208
10209                 list_move_tail(&binode->delalloc_inodes,
10210                                &root->delalloc_inodes);
10211                 inode = igrab(&binode->vfs_inode);
10212                 if (!inode) {
10213                         cond_resched_lock(&root->delalloc_lock);
10214                         continue;
10215                 }
10216                 spin_unlock(&root->delalloc_lock);
10217
10218                 work = btrfs_alloc_delalloc_work(inode, delay_iput);
10219                 if (!work) {
10220                         if (delay_iput)
10221                                 btrfs_add_delayed_iput(inode);
10222                         else
10223                                 iput(inode);
10224                         ret = -ENOMEM;
10225                         goto out;
10226                 }
10227                 list_add_tail(&work->list, &works);
10228                 btrfs_queue_work(root->fs_info->flush_workers,
10229                                  &work->work);
10230                 ret++;
10231                 if (nr != -1 && ret >= nr)
10232                         goto out;
10233                 cond_resched();
10234                 spin_lock(&root->delalloc_lock);
10235         }
10236         spin_unlock(&root->delalloc_lock);
10237
10238 out:
10239         list_for_each_entry_safe(work, next, &works, list) {
10240                 list_del_init(&work->list);
10241                 btrfs_wait_and_free_delalloc_work(work);
10242         }
10243
10244         if (!list_empty_careful(&splice)) {
10245                 spin_lock(&root->delalloc_lock);
10246                 list_splice_tail(&splice, &root->delalloc_inodes);
10247                 spin_unlock(&root->delalloc_lock);
10248         }
10249         mutex_unlock(&root->delalloc_mutex);
10250         return ret;
10251 }
10252
10253 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
10254 {
10255         struct btrfs_fs_info *fs_info = root->fs_info;
10256         int ret;
10257
10258         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10259                 return -EROFS;
10260
10261         ret = __start_delalloc_inodes(root, delay_iput, -1);
10262         if (ret > 0)
10263                 ret = 0;
10264         return ret;
10265 }
10266
10267 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
10268                                int nr)
10269 {
10270         struct btrfs_root *root;
10271         struct list_head splice;
10272         int ret;
10273
10274         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10275                 return -EROFS;
10276
10277         INIT_LIST_HEAD(&splice);
10278
10279         mutex_lock(&fs_info->delalloc_root_mutex);
10280         spin_lock(&fs_info->delalloc_root_lock);
10281         list_splice_init(&fs_info->delalloc_roots, &splice);
10282         while (!list_empty(&splice) && nr) {
10283                 root = list_first_entry(&splice, struct btrfs_root,
10284                                         delalloc_root);
10285                 root = btrfs_grab_fs_root(root);
10286                 BUG_ON(!root);
10287                 list_move_tail(&root->delalloc_root,
10288                                &fs_info->delalloc_roots);
10289                 spin_unlock(&fs_info->delalloc_root_lock);
10290
10291                 ret = __start_delalloc_inodes(root, delay_iput, nr);
10292                 btrfs_put_fs_root(root);
10293                 if (ret < 0)
10294                         goto out;
10295
10296                 if (nr != -1) {
10297                         nr -= ret;
10298                         WARN_ON(nr < 0);
10299                 }
10300                 spin_lock(&fs_info->delalloc_root_lock);
10301         }
10302         spin_unlock(&fs_info->delalloc_root_lock);
10303
10304         ret = 0;
10305 out:
10306         if (!list_empty_careful(&splice)) {
10307                 spin_lock(&fs_info->delalloc_root_lock);
10308                 list_splice_tail(&splice, &fs_info->delalloc_roots);
10309                 spin_unlock(&fs_info->delalloc_root_lock);
10310         }
10311         mutex_unlock(&fs_info->delalloc_root_mutex);
10312         return ret;
10313 }
10314
10315 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10316                          const char *symname)
10317 {
10318         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10319         struct btrfs_trans_handle *trans;
10320         struct btrfs_root *root = BTRFS_I(dir)->root;
10321         struct btrfs_path *path;
10322         struct btrfs_key key;
10323         struct inode *inode = NULL;
10324         int err;
10325         int drop_inode = 0;
10326         u64 objectid;
10327         u64 index = 0;
10328         int name_len;
10329         int datasize;
10330         unsigned long ptr;
10331         struct btrfs_file_extent_item *ei;
10332         struct extent_buffer *leaf;
10333
10334         name_len = strlen(symname);
10335         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
10336                 return -ENAMETOOLONG;
10337
10338         /*
10339          * 2 items for inode item and ref
10340          * 2 items for dir items
10341          * 1 item for updating parent inode item
10342          * 1 item for the inline extent item
10343          * 1 item for xattr if selinux is on
10344          */
10345         trans = btrfs_start_transaction(root, 7);
10346         if (IS_ERR(trans))
10347                 return PTR_ERR(trans);
10348
10349         err = btrfs_find_free_ino(root, &objectid);
10350         if (err)
10351                 goto out_unlock;
10352
10353         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10354                                 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
10355                                 objectid, S_IFLNK|S_IRWXUGO, &index);
10356         if (IS_ERR(inode)) {
10357                 err = PTR_ERR(inode);
10358                 goto out_unlock;
10359         }
10360
10361         /*
10362         * If the active LSM wants to access the inode during
10363         * d_instantiate it needs these. Smack checks to see
10364         * if the filesystem supports xattrs by looking at the
10365         * ops vector.
10366         */
10367         inode->i_fop = &btrfs_file_operations;
10368         inode->i_op = &btrfs_file_inode_operations;
10369         inode->i_mapping->a_ops = &btrfs_aops;
10370         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10371
10372         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10373         if (err)
10374                 goto out_unlock_inode;
10375
10376         path = btrfs_alloc_path();
10377         if (!path) {
10378                 err = -ENOMEM;
10379                 goto out_unlock_inode;
10380         }
10381         key.objectid = btrfs_ino(BTRFS_I(inode));
10382         key.offset = 0;
10383         key.type = BTRFS_EXTENT_DATA_KEY;
10384         datasize = btrfs_file_extent_calc_inline_size(name_len);
10385         err = btrfs_insert_empty_item(trans, root, path, &key,
10386                                       datasize);
10387         if (err) {
10388                 btrfs_free_path(path);
10389                 goto out_unlock_inode;
10390         }
10391         leaf = path->nodes[0];
10392         ei = btrfs_item_ptr(leaf, path->slots[0],
10393                             struct btrfs_file_extent_item);
10394         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10395         btrfs_set_file_extent_type(leaf, ei,
10396                                    BTRFS_FILE_EXTENT_INLINE);
10397         btrfs_set_file_extent_encryption(leaf, ei, 0);
10398         btrfs_set_file_extent_compression(leaf, ei, 0);
10399         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10400         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10401
10402         ptr = btrfs_file_extent_inline_start(ei);
10403         write_extent_buffer(leaf, symname, ptr, name_len);
10404         btrfs_mark_buffer_dirty(leaf);
10405         btrfs_free_path(path);
10406
10407         inode->i_op = &btrfs_symlink_inode_operations;
10408         inode_nohighmem(inode);
10409         inode->i_mapping->a_ops = &btrfs_symlink_aops;
10410         inode_set_bytes(inode, name_len);
10411         btrfs_i_size_write(BTRFS_I(inode), name_len);
10412         err = btrfs_update_inode(trans, root, inode);
10413         /*
10414          * Last step, add directory indexes for our symlink inode. This is the
10415          * last step to avoid extra cleanup of these indexes if an error happens
10416          * elsewhere above.
10417          */
10418         if (!err)
10419                 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
10420                                 BTRFS_I(inode), 0, index);
10421         if (err) {
10422                 drop_inode = 1;
10423                 goto out_unlock_inode;
10424         }
10425
10426         unlock_new_inode(inode);
10427         d_instantiate(dentry, inode);
10428
10429 out_unlock:
10430         btrfs_end_transaction(trans);
10431         if (drop_inode) {
10432                 inode_dec_link_count(inode);
10433                 iput(inode);
10434         }
10435         btrfs_btree_balance_dirty(fs_info);
10436         return err;
10437
10438 out_unlock_inode:
10439         drop_inode = 1;
10440         unlock_new_inode(inode);
10441         goto out_unlock;
10442 }
10443
10444 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10445                                        u64 start, u64 num_bytes, u64 min_size,
10446                                        loff_t actual_len, u64 *alloc_hint,
10447                                        struct btrfs_trans_handle *trans)
10448 {
10449         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
10450         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10451         struct extent_map *em;
10452         struct btrfs_root *root = BTRFS_I(inode)->root;
10453         struct btrfs_key ins;
10454         u64 cur_offset = start;
10455         u64 i_size;
10456         u64 cur_bytes;
10457         u64 last_alloc = (u64)-1;
10458         int ret = 0;
10459         bool own_trans = true;
10460         u64 end = start + num_bytes - 1;
10461
10462         if (trans)
10463                 own_trans = false;
10464         while (num_bytes > 0) {
10465                 if (own_trans) {
10466                         trans = btrfs_start_transaction(root, 3);
10467                         if (IS_ERR(trans)) {
10468                                 ret = PTR_ERR(trans);
10469                                 break;
10470                         }
10471                 }
10472
10473                 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10474                 cur_bytes = max(cur_bytes, min_size);
10475                 /*
10476                  * If we are severely fragmented we could end up with really
10477                  * small allocations, so if the allocator is returning small
10478                  * chunks lets make its job easier by only searching for those
10479                  * sized chunks.
10480                  */
10481                 cur_bytes = min(cur_bytes, last_alloc);
10482                 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10483                                 min_size, 0, *alloc_hint, &ins, 1, 0);
10484                 if (ret) {
10485                         if (own_trans)
10486                                 btrfs_end_transaction(trans);
10487                         break;
10488                 }
10489                 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10490
10491                 last_alloc = ins.offset;
10492                 ret = insert_reserved_file_extent(trans, inode,
10493                                                   cur_offset, ins.objectid,
10494                                                   ins.offset, ins.offset,
10495                                                   ins.offset, 0, 0, 0,
10496                                                   BTRFS_FILE_EXTENT_PREALLOC);
10497                 if (ret) {
10498                         btrfs_free_reserved_extent(fs_info, ins.objectid,
10499                                                    ins.offset, 0);
10500                         btrfs_abort_transaction(trans, ret);
10501                         if (own_trans)
10502                                 btrfs_end_transaction(trans);
10503                         break;
10504                 }
10505
10506                 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10507                                         cur_offset + ins.offset -1, 0);
10508
10509                 em = alloc_extent_map();
10510                 if (!em) {
10511                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10512                                 &BTRFS_I(inode)->runtime_flags);
10513                         goto next;
10514                 }
10515
10516                 em->start = cur_offset;
10517                 em->orig_start = cur_offset;
10518                 em->len = ins.offset;
10519                 em->block_start = ins.objectid;
10520                 em->block_len = ins.offset;
10521                 em->orig_block_len = ins.offset;
10522                 em->ram_bytes = ins.offset;
10523                 em->bdev = fs_info->fs_devices->latest_bdev;
10524                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10525                 em->generation = trans->transid;
10526
10527                 while (1) {
10528                         write_lock(&em_tree->lock);
10529                         ret = add_extent_mapping(em_tree, em, 1);
10530                         write_unlock(&em_tree->lock);
10531                         if (ret != -EEXIST)
10532                                 break;
10533                         btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10534                                                 cur_offset + ins.offset - 1,
10535                                                 0);
10536                 }
10537                 free_extent_map(em);
10538 next:
10539                 num_bytes -= ins.offset;
10540                 cur_offset += ins.offset;
10541                 *alloc_hint = ins.objectid + ins.offset;
10542
10543                 inode_inc_iversion(inode);
10544                 inode->i_ctime = current_time(inode);
10545                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10546                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10547                     (actual_len > inode->i_size) &&
10548                     (cur_offset > inode->i_size)) {
10549                         if (cur_offset > actual_len)
10550                                 i_size = actual_len;
10551                         else
10552                                 i_size = cur_offset;
10553                         i_size_write(inode, i_size);
10554                         btrfs_ordered_update_i_size(inode, i_size, NULL);
10555                 }
10556
10557                 ret = btrfs_update_inode(trans, root, inode);
10558
10559                 if (ret) {
10560                         btrfs_abort_transaction(trans, ret);
10561                         if (own_trans)
10562                                 btrfs_end_transaction(trans);
10563                         break;
10564                 }
10565
10566                 if (own_trans)
10567                         btrfs_end_transaction(trans);
10568         }
10569         if (cur_offset < end)
10570                 btrfs_free_reserved_data_space(inode, NULL, cur_offset,
10571                         end - cur_offset + 1);
10572         return ret;
10573 }
10574
10575 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10576                               u64 start, u64 num_bytes, u64 min_size,
10577                               loff_t actual_len, u64 *alloc_hint)
10578 {
10579         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10580                                            min_size, actual_len, alloc_hint,
10581                                            NULL);
10582 }
10583
10584 int btrfs_prealloc_file_range_trans(struct inode *inode,
10585                                     struct btrfs_trans_handle *trans, int mode,
10586                                     u64 start, u64 num_bytes, u64 min_size,
10587                                     loff_t actual_len, u64 *alloc_hint)
10588 {
10589         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10590                                            min_size, actual_len, alloc_hint, trans);
10591 }
10592
10593 static int btrfs_set_page_dirty(struct page *page)
10594 {
10595         return __set_page_dirty_nobuffers(page);
10596 }
10597
10598 static int btrfs_permission(struct inode *inode, int mask)
10599 {
10600         struct btrfs_root *root = BTRFS_I(inode)->root;
10601         umode_t mode = inode->i_mode;
10602
10603         if (mask & MAY_WRITE &&
10604             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10605                 if (btrfs_root_readonly(root))
10606                         return -EROFS;
10607                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10608                         return -EACCES;
10609         }
10610         return generic_permission(inode, mask);
10611 }
10612
10613 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10614 {
10615         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10616         struct btrfs_trans_handle *trans;
10617         struct btrfs_root *root = BTRFS_I(dir)->root;
10618         struct inode *inode = NULL;
10619         u64 objectid;
10620         u64 index;
10621         int ret = 0;
10622
10623         /*
10624          * 5 units required for adding orphan entry
10625          */
10626         trans = btrfs_start_transaction(root, 5);
10627         if (IS_ERR(trans))
10628                 return PTR_ERR(trans);
10629
10630         ret = btrfs_find_free_ino(root, &objectid);
10631         if (ret)
10632                 goto out;
10633
10634         inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10635                         btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10636         if (IS_ERR(inode)) {
10637                 ret = PTR_ERR(inode);
10638                 inode = NULL;
10639                 goto out;
10640         }
10641
10642         inode->i_fop = &btrfs_file_operations;
10643         inode->i_op = &btrfs_file_inode_operations;
10644
10645         inode->i_mapping->a_ops = &btrfs_aops;
10646         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10647
10648         ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10649         if (ret)
10650                 goto out_inode;
10651
10652         ret = btrfs_update_inode(trans, root, inode);
10653         if (ret)
10654                 goto out_inode;
10655         ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10656         if (ret)
10657                 goto out_inode;
10658
10659         /*
10660          * We set number of links to 0 in btrfs_new_inode(), and here we set
10661          * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10662          * through:
10663          *
10664          *    d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10665          */
10666         set_nlink(inode, 1);
10667         unlock_new_inode(inode);
10668         d_tmpfile(dentry, inode);
10669         mark_inode_dirty(inode);
10670
10671 out:
10672         btrfs_end_transaction(trans);
10673         if (ret)
10674                 iput(inode);
10675         btrfs_balance_delayed_items(fs_info);
10676         btrfs_btree_balance_dirty(fs_info);
10677         return ret;
10678
10679 out_inode:
10680         unlock_new_inode(inode);
10681         goto out;
10682
10683 }
10684
10685 __attribute__((const))
10686 static int btrfs_readpage_io_failed_hook(struct page *page, int failed_mirror)
10687 {
10688         return -EAGAIN;
10689 }
10690
10691 static struct btrfs_fs_info *iotree_fs_info(void *private_data)
10692 {
10693         struct inode *inode = private_data;
10694         return btrfs_sb(inode->i_sb);
10695 }
10696
10697 static void btrfs_check_extent_io_range(void *private_data, const char *caller,
10698                                         u64 start, u64 end)
10699 {
10700         struct inode *inode = private_data;
10701         u64 isize;
10702
10703         isize = i_size_read(inode);
10704         if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
10705                 btrfs_debug_rl(BTRFS_I(inode)->root->fs_info,
10706                     "%s: ino %llu isize %llu odd range [%llu,%llu]",
10707                         caller, btrfs_ino(BTRFS_I(inode)), isize, start, end);
10708         }
10709 }
10710
10711 void btrfs_set_range_writeback(void *private_data, u64 start, u64 end)
10712 {
10713         struct inode *inode = private_data;
10714         unsigned long index = start >> PAGE_SHIFT;
10715         unsigned long end_index = end >> PAGE_SHIFT;
10716         struct page *page;
10717
10718         while (index <= end_index) {
10719                 page = find_get_page(inode->i_mapping, index);
10720                 ASSERT(page); /* Pages should be in the extent_io_tree */
10721                 set_page_writeback(page);
10722                 put_page(page);
10723                 index++;
10724         }
10725 }
10726
10727 static const struct inode_operations btrfs_dir_inode_operations = {
10728         .getattr        = btrfs_getattr,
10729         .lookup         = btrfs_lookup,
10730         .create         = btrfs_create,
10731         .unlink         = btrfs_unlink,
10732         .link           = btrfs_link,
10733         .mkdir          = btrfs_mkdir,
10734         .rmdir          = btrfs_rmdir,
10735         .rename         = btrfs_rename2,
10736         .symlink        = btrfs_symlink,
10737         .setattr        = btrfs_setattr,
10738         .mknod          = btrfs_mknod,
10739         .listxattr      = btrfs_listxattr,
10740         .permission     = btrfs_permission,
10741         .get_acl        = btrfs_get_acl,
10742         .set_acl        = btrfs_set_acl,
10743         .update_time    = btrfs_update_time,
10744         .tmpfile        = btrfs_tmpfile,
10745 };
10746 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10747         .lookup         = btrfs_lookup,
10748         .permission     = btrfs_permission,
10749         .update_time    = btrfs_update_time,
10750 };
10751
10752 static const struct file_operations btrfs_dir_file_operations = {
10753         .llseek         = generic_file_llseek,
10754         .read           = generic_read_dir,
10755         .iterate_shared = btrfs_real_readdir,
10756         .open           = btrfs_opendir,
10757         .unlocked_ioctl = btrfs_ioctl,
10758 #ifdef CONFIG_COMPAT
10759         .compat_ioctl   = btrfs_compat_ioctl,
10760 #endif
10761         .release        = btrfs_release_file,
10762         .fsync          = btrfs_sync_file,
10763 };
10764
10765 static const struct extent_io_ops btrfs_extent_io_ops = {
10766         /* mandatory callbacks */
10767         .submit_bio_hook = btrfs_submit_bio_hook,
10768         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10769         .merge_bio_hook = btrfs_merge_bio_hook,
10770         .readpage_io_failed_hook = btrfs_readpage_io_failed_hook,
10771         .tree_fs_info = iotree_fs_info,
10772         .set_range_writeback = btrfs_set_range_writeback,
10773
10774         /* optional callbacks */
10775         .fill_delalloc = run_delalloc_range,
10776         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10777         .writepage_start_hook = btrfs_writepage_start_hook,
10778         .set_bit_hook = btrfs_set_bit_hook,
10779         .clear_bit_hook = btrfs_clear_bit_hook,
10780         .merge_extent_hook = btrfs_merge_extent_hook,
10781         .split_extent_hook = btrfs_split_extent_hook,
10782         .check_extent_io_range = btrfs_check_extent_io_range,
10783 };
10784
10785 /*
10786  * btrfs doesn't support the bmap operation because swapfiles
10787  * use bmap to make a mapping of extents in the file.  They assume
10788  * these extents won't change over the life of the file and they
10789  * use the bmap result to do IO directly to the drive.
10790  *
10791  * the btrfs bmap call would return logical addresses that aren't
10792  * suitable for IO and they also will change frequently as COW
10793  * operations happen.  So, swapfile + btrfs == corruption.
10794  *
10795  * For now we're avoiding this by dropping bmap.
10796  */
10797 static const struct address_space_operations btrfs_aops = {
10798         .readpage       = btrfs_readpage,
10799         .writepage      = btrfs_writepage,
10800         .writepages     = btrfs_writepages,
10801         .readpages      = btrfs_readpages,
10802         .direct_IO      = btrfs_direct_IO,
10803         .invalidatepage = btrfs_invalidatepage,
10804         .releasepage    = btrfs_releasepage,
10805         .set_page_dirty = btrfs_set_page_dirty,
10806         .error_remove_page = generic_error_remove_page,
10807 };
10808
10809 static const struct address_space_operations btrfs_symlink_aops = {
10810         .readpage       = btrfs_readpage,
10811         .writepage      = btrfs_writepage,
10812         .invalidatepage = btrfs_invalidatepage,
10813         .releasepage    = btrfs_releasepage,
10814 };
10815
10816 static const struct inode_operations btrfs_file_inode_operations = {
10817         .getattr        = btrfs_getattr,
10818         .setattr        = btrfs_setattr,
10819         .listxattr      = btrfs_listxattr,
10820         .permission     = btrfs_permission,
10821         .fiemap         = btrfs_fiemap,
10822         .get_acl        = btrfs_get_acl,
10823         .set_acl        = btrfs_set_acl,
10824         .update_time    = btrfs_update_time,
10825 };
10826 static const struct inode_operations btrfs_special_inode_operations = {
10827         .getattr        = btrfs_getattr,
10828         .setattr        = btrfs_setattr,
10829         .permission     = btrfs_permission,
10830         .listxattr      = btrfs_listxattr,
10831         .get_acl        = btrfs_get_acl,
10832         .set_acl        = btrfs_set_acl,
10833         .update_time    = btrfs_update_time,
10834 };
10835 static const struct inode_operations btrfs_symlink_inode_operations = {
10836         .get_link       = page_get_link,
10837         .getattr        = btrfs_getattr,
10838         .setattr        = btrfs_setattr,
10839         .permission     = btrfs_permission,
10840         .listxattr      = btrfs_listxattr,
10841         .update_time    = btrfs_update_time,
10842 };
10843
10844 const struct dentry_operations btrfs_dentry_operations = {
10845         .d_delete       = btrfs_dentry_delete,
10846         .d_release      = btrfs_dentry_release,
10847 };