fs/btrfs/file-item.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2007 Oracle.  All rights reserved.
   4  */
   5
   6 #include <linux/bio.h>
   7 #include <linux/slab.h>
   8 #include <linux/pagemap.h>
   9 #include <linux/highmem.h>
  10 #include <linux/sched/mm.h>
  11 #include <crypto/hash.h>
  12 #include "ctree.h"
  13 #include "disk-io.h"
  14 #include "transaction.h"
  15 #include "volumes.h"
  16 #include "print-tree.h"
  17 #include "compression.h"
  18
  19 #define __MAX_CSUM_ITEMS(r, size) ((unsigned long)(((BTRFS_LEAF_DATA_SIZE(r) - \
  20                                    sizeof(struct btrfs_item) * 2) / \
  21                                   size) - 1))
  22
  23 #define MAX_CSUM_ITEMS(r, size) (min_t(u32, __MAX_CSUM_ITEMS(r, size), \
  24                                        PAGE_SIZE))
  25
  26 /**
  27  * @inode - the inode we want to update the disk_i_size for
  28  * @new_i_size - the i_size we want to set to, 0 if we use i_size
  29  *
  30  * With NO_HOLES set this simply sets the disk_is_size to whatever i_size_read()
  31  * returns as it is perfectly fine with a file that has holes without hole file
  32  * extent items.
  33  *
  34  * However without NO_HOLES we need to only return the area that is contiguous
  35  * from the 0 offset of the file.  Otherwise we could end up adjust i_size up
  36  * to an extent that has a gap in between.
  37  *
  38  * Finally new_i_size should only be set in the case of truncate where we're not
  39  * ready to use i_size_read() as the limiter yet.
  40  */
  41 void btrfs_inode_safe_disk_i_size_write(struct inode *inode, u64 new_i_size)
  42 {
  43         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
  44         u64 start, end, i_size;
  45         int ret;
  46
  47         i_size = new_i_size ?: i_size_read(inode);
  48         if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
  49                 BTRFS_I(inode)->disk_i_size = i_size;
  50                 return;
  51         }
  52
  53         spin_lock(&BTRFS_I(inode)->lock);
  54         ret = find_contiguous_extent_bit(&BTRFS_I(inode)->file_extent_tree, 0,
  55                                          &start, &end, EXTENT_DIRTY);
  56         if (!ret && start == 0)
  57                 i_size = min(i_size, end + 1);
  58         else
  59                 i_size = 0;
  60         BTRFS_I(inode)->disk_i_size = i_size;
  61         spin_unlock(&BTRFS_I(inode)->lock);
  62 }
  63
  64 /**
  65  * @inode - the inode we're modifying
  66  * @start - the start file offset of the file extent we've inserted
  67  * @len - the logical length of the file extent item
  68  *
  69  * Call when we are inserting a new file extent where there was none before.
  70  * Does not need to call this in the case where we're replacing an existing file
  71  * extent, however if not sure it's fine to call this multiple times.
  72  *
  73  * The start and len must match the file extent item, so thus must be sectorsize
  74  * aligned.
  75  */
  76 int btrfs_inode_set_file_extent_range(struct btrfs_inode *inode, u64 start,
  77                                       u64 len)
  78 {
  79         if (len == 0)
  80                 return 0;
  81
  82         ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize));
  83
  84         if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES))
  85                 return 0;
  86         return set_extent_bits(&inode->file_extent_tree, start, start + len - 1,
  87                                EXTENT_DIRTY);
  88 }
  89
  90 /**
  91  * @inode - the inode we're modifying
  92  * @start - the start file offset of the file extent we've inserted
  93  * @len - the logical length of the file extent item
  94  *
  95  * Called when we drop a file extent, for example when we truncate.  Doesn't
  96  * need to be called for cases where we're replacing a file extent, like when
  97  * we've COWed a file extent.
  98  *
  99  * The start and len must match the file extent item, so thus must be sectorsize
 100  * aligned.
 101  */
 102 int btrfs_inode_clear_file_extent_range(struct btrfs_inode *inode, u64 start,
 103                                         u64 len)
 104 {
 105         if (len == 0)
 106                 return 0;
 107
 108         ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize) ||
 109                len == (u64)-1);
 110
 111         if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES))
 112                 return 0;
 113         return clear_extent_bit(&inode->file_extent_tree, start,
 114                                 start + len - 1, EXTENT_DIRTY, 0, 0, NULL);
 115 }
 116
 117 static inline u32 max_ordered_sum_bytes(struct btrfs_fs_info *fs_info,
 118                                         u16 csum_size)
 119 {
 120         u32 ncsums = (PAGE_SIZE - sizeof(struct btrfs_ordered_sum)) / csum_size;
 121
 122         return ncsums * fs_info->sectorsize;
 123 }
 124
 125 int btrfs_insert_file_extent(struct btrfs_trans_handle *trans,
 126                              struct btrfs_root *root,
 127                              u64 objectid, u64 pos,
 128                              u64 disk_offset, u64 disk_num_bytes,
 129                              u64 num_bytes, u64 offset, u64 ram_bytes,
 130                              u8 compression, u8 encryption, u16 other_encoding)
 131 {
 132         int ret = 0;
 133         struct btrfs_file_extent_item *item;
 134         struct btrfs_key file_key;
 135         struct btrfs_path *path;
 136         struct extent_buffer *leaf;
 137
 138         path = btrfs_alloc_path();
 139         if (!path)
 140                 return -ENOMEM;
 141         file_key.objectid = objectid;
 142         file_key.offset = pos;
 143         file_key.type = BTRFS_EXTENT_DATA_KEY;
 144
 145         path->leave_spinning = 1;
 146         ret = btrfs_insert_empty_item(trans, root, path, &file_key,
 147                                       sizeof(*item));
 148         if (ret < 0)
 149                 goto out;
 150         BUG_ON(ret); /* Can't happen */
 151         leaf = path->nodes[0];
 152         item = btrfs_item_ptr(leaf, path->slots[0],
 153                               struct btrfs_file_extent_item);
 154         btrfs_set_file_extent_disk_bytenr(leaf, item, disk_offset);
 155         btrfs_set_file_extent_disk_num_bytes(leaf, item, disk_num_bytes);
 156         btrfs_set_file_extent_offset(leaf, item, offset);
 157         btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
 158         btrfs_set_file_extent_ram_bytes(leaf, item, ram_bytes);
 159         btrfs_set_file_extent_generation(leaf, item, trans->transid);
 160         btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
 161         btrfs_set_file_extent_compression(leaf, item, compression);
 162         btrfs_set_file_extent_encryption(leaf, item, encryption);
 163         btrfs_set_file_extent_other_encoding(leaf, item, other_encoding);
 164
 165         btrfs_mark_buffer_dirty(leaf);
 166 out:
 167         btrfs_free_path(path);
 168         return ret;
 169 }
 170
 171 static struct btrfs_csum_item *
 172 btrfs_lookup_csum(struct btrfs_trans_handle *trans,
 173                   struct btrfs_root *root,
 174                   struct btrfs_path *path,
 175                   u64 bytenr, int cow)
 176 {
 177         struct btrfs_fs_info *fs_info = root->fs_info;
 178         int ret;
 179         struct btrfs_key file_key;
 180         struct btrfs_key found_key;
 181         struct btrfs_csum_item *item;
 182         struct extent_buffer *leaf;
 183         u64 csum_offset = 0;
 184         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 185         int csums_in_item;
 186
 187         file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
 188         file_key.offset = bytenr;
 189         file_key.type = BTRFS_EXTENT_CSUM_KEY;
 190         ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
 191         if (ret < 0)
 192                 goto fail;
 193         leaf = path->nodes[0];
 194         if (ret > 0) {
 195                 ret = 1;
 196                 if (path->slots[0] == 0)
 197                         goto fail;
 198                 path->slots[0]--;
 199                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 200                 if (found_key.type != BTRFS_EXTENT_CSUM_KEY)
 201                         goto fail;
 202
 203                 csum_offset = (bytenr - found_key.offset) >>
 204                                 fs_info->sb->s_blocksize_bits;
 205                 csums_in_item = btrfs_item_size_nr(leaf, path->slots[0]);
 206                 csums_in_item /= csum_size;
 207
 208                 if (csum_offset == csums_in_item) {
 209                         ret = -EFBIG;
 210                         goto fail;
 211                 } else if (csum_offset > csums_in_item) {
 212                         goto fail;
 213                 }
 214         }
 215         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
 216         item = (struct btrfs_csum_item *)((unsigned char *)item +
 217                                           csum_offset * csum_size);
 218         return item;
 219 fail:
 220         if (ret > 0)
 221                 ret = -ENOENT;
 222         return ERR_PTR(ret);
 223 }
 224
 225 int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
 226                              struct btrfs_root *root,
 227                              struct btrfs_path *path, u64 objectid,
 228                              u64 offset, int mod)
 229 {
 230         int ret;
 231         struct btrfs_key file_key;
 232         int ins_len = mod < 0 ? -1 : 0;
 233         int cow = mod != 0;
 234
 235         file_key.objectid = objectid;
 236         file_key.offset = offset;
 237         file_key.type = BTRFS_EXTENT_DATA_KEY;
 238         ret = btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
 239         return ret;
 240 }
 241
 242 /**
 243  * btrfs_lookup_bio_sums - Look up checksums for a bio.
 244  * @inode: inode that the bio is for.
 245  * @bio: bio to look up.
 246  * @offset: Unless (u64)-1, look up checksums for this offset in the file.
 247  *          If (u64)-1, use the page offsets from the bio instead.
 248  * @dst: Buffer of size nblocks * btrfs_super_csum_size() used to return
 249  *       checksum (nblocks = bio->bi_iter.bi_size / fs_info->sectorsize). If
 250  *       NULL, the checksum buffer is allocated and returned in
 251  *       btrfs_io_bio(bio)->csum instead.
 252  *
 253  * Return: BLK_STS_RESOURCE if allocating memory fails, BLK_STS_OK otherwise.
 254  */
 255 blk_status_t btrfs_lookup_bio_sums(struct inode *inode, struct bio *bio,
 256                                    u64 offset, u8 *dst)
 257 {
 258         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 259         struct bio_vec bvec;
 260         struct bvec_iter iter;
 261         struct btrfs_csum_item *item = NULL;
 262         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 263         struct btrfs_path *path;
 264         const bool page_offsets = (offset == (u64)-1);
 265         u8 *csum;
 266         u64 item_start_offset = 0;
 267         u64 item_last_offset = 0;
 268         u64 disk_bytenr;
 269         u64 page_bytes_left;
 270         u32 diff;
 271         int nblocks;
 272         int count = 0;
 273         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 274
 275         if (!fs_info->csum_root || (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
 276                 return BLK_STS_OK;
 277
 278         path = btrfs_alloc_path();
 279         if (!path)
 280                 return BLK_STS_RESOURCE;
 281
 282         nblocks = bio->bi_iter.bi_size >> inode->i_sb->s_blocksize_bits;
 283         if (!dst) {
 284                 struct btrfs_io_bio *btrfs_bio = btrfs_io_bio(bio);
 285
 286                 if (nblocks * csum_size > BTRFS_BIO_INLINE_CSUM_SIZE) {
 287                         btrfs_bio->csum = kmalloc_array(nblocks, csum_size,
 288                                                         GFP_NOFS);
 289                         if (!btrfs_bio->csum) {
 290                                 btrfs_free_path(path);
 291                                 return BLK_STS_RESOURCE;
 292                         }
 293                 } else {
 294                         btrfs_bio->csum = btrfs_bio->csum_inline;
 295                 }
 296                 csum = btrfs_bio->csum;
 297         } else {
 298                 csum = dst;
 299         }
 300
 301         if (bio->bi_iter.bi_size > PAGE_SIZE * 8)
 302                 path->reada = READA_FORWARD;
 303
 304         /*
 305          * the free space stuff is only read when it hasn't been
 306          * updated in the current transaction.  So, we can safely
 307          * read from the commit root and sidestep a nasty deadlock
 308          * between reading the free space cache and updating the csum tree.
 309          */
 310         if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
 311                 path->search_commit_root = 1;
 312                 path->skip_locking = 1;
 313         }
 314
 315         disk_bytenr = (u64)bio->bi_iter.bi_sector << 9;
 316
 317         bio_for_each_segment(bvec, bio, iter) {
 318                 page_bytes_left = bvec.bv_len;
 319                 if (count)
 320                         goto next;
 321
 322                 if (page_offsets)
 323                         offset = page_offset(bvec.bv_page) + bvec.bv_offset;
 324                 count = btrfs_find_ordered_sum(BTRFS_I(inode), offset,
 325                                                disk_bytenr, csum, nblocks);
 326                 if (count)
 327                         goto found;
 328
 329                 if (!item || disk_bytenr < item_start_offset ||
 330                     disk_bytenr >= item_last_offset) {
 331                         struct btrfs_key found_key;
 332                         u32 item_size;
 333
 334                         if (item)
 335                                 btrfs_release_path(path);
 336                         item = btrfs_lookup_csum(NULL, fs_info->csum_root,
 337                                                  path, disk_bytenr, 0);
 338                         if (IS_ERR(item)) {
 339                                 count = 1;
 340                                 memset(csum, 0, csum_size);
 341                                 if (BTRFS_I(inode)->root->root_key.objectid ==
 342                                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 343                                         set_extent_bits(io_tree, offset,
 344                                                 offset + fs_info->sectorsize - 1,
 345                                                 EXTENT_NODATASUM);
 346                                 } else {
 347                                         btrfs_info_rl(fs_info,
 348                                                    "no csum found for inode %llu start %llu",
 349                                                btrfs_ino(BTRFS_I(inode)), offset);
 350                                 }
 351                                 item = NULL;
 352                                 btrfs_release_path(path);
 353                                 goto found;
 354                         }
 355                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 356                                               path->slots[0]);
 357
 358                         item_start_offset = found_key.offset;
 359                         item_size = btrfs_item_size_nr(path->nodes[0],
 360                                                        path->slots[0]);
 361                         item_last_offset = item_start_offset +
 362                                 (item_size / csum_size) *
 363                                 fs_info->sectorsize;
 364                         item = btrfs_item_ptr(path->nodes[0], path->slots[0],
 365                                               struct btrfs_csum_item);
 366                 }
 367                 /*
 368                  * this byte range must be able to fit inside
 369                  * a single leaf so it will also fit inside a u32
 370                  */
 371                 diff = disk_bytenr - item_start_offset;
 372                 diff = diff / fs_info->sectorsize;
 373                 diff = diff * csum_size;
 374                 count = min_t(int, nblocks, (item_last_offset - disk_bytenr) >>
 375                                             inode->i_sb->s_blocksize_bits);
 376                 read_extent_buffer(path->nodes[0], csum,
 377                                    ((unsigned long)item) + diff,
 378                                    csum_size * count);
 379 found:
 380                 csum += count * csum_size;
 381                 nblocks -= count;
 382 next:
 383                 while (count > 0) {
 384                         count--;
 385                         disk_bytenr += fs_info->sectorsize;
 386                         offset += fs_info->sectorsize;
 387                         page_bytes_left -= fs_info->sectorsize;
 388                         if (!page_bytes_left)
 389                                 break; /* move to next bio */
 390                 }
 391         }
 392
 393         WARN_ON_ONCE(count);
 394         btrfs_free_path(path);
 395         return BLK_STS_OK;
 396 }
 397
 398 int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
 399                              struct list_head *list, int search_commit)
 400 {
 401         struct btrfs_fs_info *fs_info = root->fs_info;
 402         struct btrfs_key key;
 403         struct btrfs_path *path;
 404         struct extent_buffer *leaf;
 405         struct btrfs_ordered_sum *sums;
 406         struct btrfs_csum_item *item;
 407         LIST_HEAD(tmplist);
 408         unsigned long offset;
 409         int ret;
 410         size_t size;
 411         u64 csum_end;
 412         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 413
 414         ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
 415                IS_ALIGNED(end + 1, fs_info->sectorsize));
 416
 417         path = btrfs_alloc_path();
 418         if (!path)
 419                 return -ENOMEM;
 420
 421         if (search_commit) {
 422                 path->skip_locking = 1;
 423                 path->reada = READA_FORWARD;
 424                 path->search_commit_root = 1;
 425         }
 426
 427         key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
 428         key.offset = start;
 429         key.type = BTRFS_EXTENT_CSUM_KEY;
 430
 431         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 432         if (ret < 0)
 433                 goto fail;
 434         if (ret > 0 && path->slots[0] > 0) {
 435                 leaf = path->nodes[0];
 436                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
 437                 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
 438                     key.type == BTRFS_EXTENT_CSUM_KEY) {
 439                         offset = (start - key.offset) >>
 440                                  fs_info->sb->s_blocksize_bits;
 441                         if (offset * csum_size <
 442                             btrfs_item_size_nr(leaf, path->slots[0] - 1))
 443                                 path->slots[0]--;
 444                 }
 445         }
 446
 447         while (start <= end) {
 448                 leaf = path->nodes[0];
 449                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 450                         ret = btrfs_next_leaf(root, path);
 451                         if (ret < 0)
 452                                 goto fail;
 453                         if (ret > 0)
 454                                 break;
 455                         leaf = path->nodes[0];
 456                 }
 457
 458                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 459                 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
 460                     key.type != BTRFS_EXTENT_CSUM_KEY ||
 461                     key.offset > end)
 462                         break;
 463
 464                 if (key.offset > start)
 465                         start = key.offset;
 466
 467                 size = btrfs_item_size_nr(leaf, path->slots[0]);
 468                 csum_end = key.offset + (size / csum_size) * fs_info->sectorsize;
 469                 if (csum_end <= start) {
 470                         path->slots[0]++;
 471                         continue;
 472                 }
 473
 474                 csum_end = min(csum_end, end + 1);
 475                 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
 476                                       struct btrfs_csum_item);
 477                 while (start < csum_end) {
 478                         size = min_t(size_t, csum_end - start,
 479                                      max_ordered_sum_bytes(fs_info, csum_size));
 480                         sums = kzalloc(btrfs_ordered_sum_size(fs_info, size),
 481                                        GFP_NOFS);
 482                         if (!sums) {
 483                                 ret = -ENOMEM;
 484                                 goto fail;
 485                         }
 486
 487                         sums->bytenr = start;
 488                         sums->len = (int)size;
 489
 490                         offset = (start - key.offset) >>
 491                                 fs_info->sb->s_blocksize_bits;
 492                         offset *= csum_size;
 493                         size >>= fs_info->sb->s_blocksize_bits;
 494
 495                         read_extent_buffer(path->nodes[0],
 496                                            sums->sums,
 497                                            ((unsigned long)item) + offset,
 498                                            csum_size * size);
 499
 500                         start += fs_info->sectorsize * size;
 501                         list_add_tail(&sums->list, &tmplist);
 502                 }
 503                 path->slots[0]++;
 504         }
 505         ret = 0;
 506 fail:
 507         while (ret < 0 && !list_empty(&tmplist)) {
 508                 sums = list_entry(tmplist.next, struct btrfs_ordered_sum, list);
 509                 list_del(&sums->list);
 510                 kfree(sums);
 511         }
 512         list_splice_tail(&tmplist, list);
 513
 514         btrfs_free_path(path);
 515         return ret;
 516 }
 517
 518 /*
 519  * btrfs_csum_one_bio - Calculates checksums of the data contained inside a bio
 520  * @inode:       Owner of the data inside the bio
 521  * @bio:         Contains the data to be checksummed
 522  * @file_start:  offset in file this bio begins to describe
 523  * @contig:      Boolean. If true/1 means all bio vecs in this bio are
 524  *               contiguous and they begin at @file_start in the file. False/0
 525  *               means this bio can contains potentially discontigous bio vecs
 526  *               so the logical offset of each should be calculated separately.
 527  */
 528 blk_status_t btrfs_csum_one_bio(struct btrfs_inode *inode, struct bio *bio,
 529                        u64 file_start, int contig)
 530 {
 531         struct btrfs_fs_info *fs_info = inode->root->fs_info;
 532         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
 533         struct btrfs_ordered_sum *sums;
 534         struct btrfs_ordered_extent *ordered = NULL;
 535         char *data;
 536         struct bvec_iter iter;
 537         struct bio_vec bvec;
 538         int index;
 539         int nr_sectors;
 540         unsigned long total_bytes = 0;
 541         unsigned long this_sum_bytes = 0;
 542         int i;
 543         u64 offset;
 544         unsigned nofs_flag;
 545         const u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 546
 547         nofs_flag = memalloc_nofs_save();
 548         sums = kvzalloc(btrfs_ordered_sum_size(fs_info, bio->bi_iter.bi_size),
 549                        GFP_KERNEL);
 550         memalloc_nofs_restore(nofs_flag);
 551
 552         if (!sums)
 553                 return BLK_STS_RESOURCE;
 554
 555         sums->len = bio->bi_iter.bi_size;
 556         INIT_LIST_HEAD(&sums->list);
 557
 558         if (contig)
 559                 offset = file_start;
 560         else
 561                 offset = 0; /* shut up gcc */
 562
 563         sums->bytenr = (u64)bio->bi_iter.bi_sector << 9;
 564         index = 0;
 565
 566         shash->tfm = fs_info->csum_shash;
 567
 568         bio_for_each_segment(bvec, bio, iter) {
 569                 if (!contig)
 570                         offset = page_offset(bvec.bv_page) + bvec.bv_offset;
 571
 572                 if (!ordered) {
 573                         ordered = btrfs_lookup_ordered_extent(inode, offset);
 574                         BUG_ON(!ordered); /* Logic error */
 575                 }
 576
 577                 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info,
 578                                                  bvec.bv_len + fs_info->sectorsize
 579                                                  - 1);
 580
 581                 for (i = 0; i < nr_sectors; i++) {
 582                         if (offset >= ordered->file_offset + ordered->num_bytes ||
 583                             offset < ordered->file_offset) {
 584                                 unsigned long bytes_left;
 585
 586                                 sums->len = this_sum_bytes;
 587                                 this_sum_bytes = 0;
 588                                 btrfs_add_ordered_sum(ordered, sums);
 589                                 btrfs_put_ordered_extent(ordered);
 590
 591                                 bytes_left = bio->bi_iter.bi_size - total_bytes;
 592
 593                                 nofs_flag = memalloc_nofs_save();
 594                                 sums = kvzalloc(btrfs_ordered_sum_size(fs_info,
 595                                                       bytes_left), GFP_KERNEL);
 596                                 memalloc_nofs_restore(nofs_flag);
 597                                 BUG_ON(!sums); /* -ENOMEM */
 598                                 sums->len = bytes_left;
 599                                 ordered = btrfs_lookup_ordered_extent(inode,
 600                                                                 offset);
 601                                 ASSERT(ordered); /* Logic error */
 602                                 sums->bytenr = ((u64)bio->bi_iter.bi_sector << 9)
 603                                         + total_bytes;
 604                                 index = 0;
 605                         }
 606
 607                         data = kmap_atomic(bvec.bv_page);
 608                         crypto_shash_digest(shash, data + bvec.bv_offset
 609                                             + (i * fs_info->sectorsize),
 610                                             fs_info->sectorsize,
 611                                             sums->sums + index);
 612                         kunmap_atomic(data);
 613                         index += csum_size;
 614                         offset += fs_info->sectorsize;
 615                         this_sum_bytes += fs_info->sectorsize;
 616                         total_bytes += fs_info->sectorsize;
 617                 }
 618
 619         }
 620         this_sum_bytes = 0;
 621         btrfs_add_ordered_sum(ordered, sums);
 622         btrfs_put_ordered_extent(ordered);
 623         return 0;
 624 }
 625
 626 /*
 627  * helper function for csum removal, this expects the
 628  * key to describe the csum pointed to by the path, and it expects
 629  * the csum to overlap the range [bytenr, len]
 630  *
 631  * The csum should not be entirely contained in the range and the
 632  * range should not be entirely contained in the csum.
 633  *
 634  * This calls btrfs_truncate_item with the correct args based on the
 635  * overlap, and fixes up the key as required.
 636  */
 637 static noinline void truncate_one_csum(struct btrfs_fs_info *fs_info,
 638                                        struct btrfs_path *path,
 639                                        struct btrfs_key *key,
 640                                        u64 bytenr, u64 len)
 641 {
 642         struct extent_buffer *leaf;
 643         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 644         u64 csum_end;
 645         u64 end_byte = bytenr + len;
 646         u32 blocksize_bits = fs_info->sb->s_blocksize_bits;
 647
 648         leaf = path->nodes[0];
 649         csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
 650         csum_end <<= fs_info->sb->s_blocksize_bits;
 651         csum_end += key->offset;
 652
 653         if (key->offset < bytenr && csum_end <= end_byte) {
 654                 /*
 655                  *         [ bytenr - len ]
 656                  *         [   ]
 657                  *   [csum     ]
 658                  *   A simple truncate off the end of the item
 659                  */
 660                 u32 new_size = (bytenr - key->offset) >> blocksize_bits;
 661                 new_size *= csum_size;
 662                 btrfs_truncate_item(path, new_size, 1);
 663         } else if (key->offset >= bytenr && csum_end > end_byte &&
 664                    end_byte > key->offset) {
 665                 /*
 666                  *         [ bytenr - len ]
 667                  *                 [ ]
 668                  *                 [csum     ]
 669                  * we need to truncate from the beginning of the csum
 670                  */
 671                 u32 new_size = (csum_end - end_byte) >> blocksize_bits;
 672                 new_size *= csum_size;
 673
 674                 btrfs_truncate_item(path, new_size, 0);
 675
 676                 key->offset = end_byte;
 677                 btrfs_set_item_key_safe(fs_info, path, key);
 678         } else {
 679                 BUG();
 680         }
 681 }
 682
 683 /*
 684  * deletes the csum items from the csum tree for a given
 685  * range of bytes.
 686  */
 687 int btrfs_del_csums(struct btrfs_trans_handle *trans,
 688                     struct btrfs_root *root, u64 bytenr, u64 len)
 689 {
 690         struct btrfs_fs_info *fs_info = trans->fs_info;
 691         struct btrfs_path *path;
 692         struct btrfs_key key;
 693         u64 end_byte = bytenr + len;
 694         u64 csum_end;
 695         struct extent_buffer *leaf;
 696         int ret;
 697         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 698         int blocksize_bits = fs_info->sb->s_blocksize_bits;
 699
 700         ASSERT(root == fs_info->csum_root ||
 701                root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
 702
 703         path = btrfs_alloc_path();
 704         if (!path)
 705                 return -ENOMEM;
 706
 707         while (1) {
 708                 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
 709                 key.offset = end_byte - 1;
 710                 key.type = BTRFS_EXTENT_CSUM_KEY;
 711
 712                 path->leave_spinning = 1;
 713                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 714                 if (ret > 0) {
 715                         if (path->slots[0] == 0)
 716                                 break;
 717                         path->slots[0]--;
 718                 } else if (ret < 0) {
 719                         break;
 720                 }
 721
 722                 leaf = path->nodes[0];
 723                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 724
 725                 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
 726                     key.type != BTRFS_EXTENT_CSUM_KEY) {
 727                         break;
 728                 }
 729
 730                 if (key.offset >= end_byte)
 731                         break;
 732
 733                 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
 734                 csum_end <<= blocksize_bits;
 735                 csum_end += key.offset;
 736
 737                 /* this csum ends before we start, we're done */
 738                 if (csum_end <= bytenr)
 739                         break;
 740
 741                 /* delete the entire item, it is inside our range */
 742                 if (key.offset >= bytenr && csum_end <= end_byte) {
 743                         int del_nr = 1;
 744
 745                         /*
 746                          * Check how many csum items preceding this one in this
 747                          * leaf correspond to our range and then delete them all
 748                          * at once.
 749                          */
 750                         if (key.offset > bytenr && path->slots[0] > 0) {
 751                                 int slot = path->slots[0] - 1;
 752
 753                                 while (slot >= 0) {
 754                                         struct btrfs_key pk;
 755
 756                                         btrfs_item_key_to_cpu(leaf, &pk, slot);
 757                                         if (pk.offset < bytenr ||
 758                                             pk.type != BTRFS_EXTENT_CSUM_KEY ||
 759                                             pk.objectid !=
 760                                             BTRFS_EXTENT_CSUM_OBJECTID)
 761                                                 break;
 762                                         path->slots[0] = slot;
 763                                         del_nr++;
 764                                         key.offset = pk.offset;
 765                                         slot--;
 766                                 }
 767                         }
 768                         ret = btrfs_del_items(trans, root, path,
 769                                               path->slots[0], del_nr);
 770                         if (ret)
 771                                 goto out;
 772                         if (key.offset == bytenr)
 773                                 break;
 774                 } else if (key.offset < bytenr && csum_end > end_byte) {
 775                         unsigned long offset;
 776                         unsigned long shift_len;
 777                         unsigned long item_offset;
 778                         /*
 779                          *        [ bytenr - len ]
 780                          *     [csum                ]
 781                          *
 782                          * Our bytes are in the middle of the csum,
 783                          * we need to split this item and insert a new one.
 784                          *
 785                          * But we can't drop the path because the
 786                          * csum could change, get removed, extended etc.
 787                          *
 788                          * The trick here is the max size of a csum item leaves
 789                          * enough room in the tree block for a single
 790                          * item header.  So, we split the item in place,
 791                          * adding a new header pointing to the existing
 792                          * bytes.  Then we loop around again and we have
 793                          * a nicely formed csum item that we can neatly
 794                          * truncate.
 795                          */
 796                         offset = (bytenr - key.offset) >> blocksize_bits;
 797                         offset *= csum_size;
 798
 799                         shift_len = (len >> blocksize_bits) * csum_size;
 800
 801                         item_offset = btrfs_item_ptr_offset(leaf,
 802                                                             path->slots[0]);
 803
 804                         memzero_extent_buffer(leaf, item_offset + offset,
 805                                              shift_len);
 806                         key.offset = bytenr;
 807
 808                         /*
 809                          * btrfs_split_item returns -EAGAIN when the
 810                          * item changed size or key
 811                          */
 812                         ret = btrfs_split_item(trans, root, path, &key, offset);
 813                         if (ret && ret != -EAGAIN) {
 814                                 btrfs_abort_transaction(trans, ret);
 815                                 goto out;
 816                         }
 817
 818                         key.offset = end_byte - 1;
 819                 } else {
 820                         truncate_one_csum(fs_info, path, &key, bytenr, len);
 821                         if (key.offset < bytenr)
 822                                 break;
 823                 }
 824                 btrfs_release_path(path);
 825         }
 826         ret = 0;
 827 out:
 828         btrfs_free_path(path);
 829         return ret;
 830 }
 831
 832 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
 833                            struct btrfs_root *root,
 834                            struct btrfs_ordered_sum *sums)
 835 {
 836         struct btrfs_fs_info *fs_info = root->fs_info;
 837         struct btrfs_key file_key;
 838         struct btrfs_key found_key;
 839         struct btrfs_path *path;
 840         struct btrfs_csum_item *item;
 841         struct btrfs_csum_item *item_end;
 842         struct extent_buffer *leaf = NULL;
 843         u64 next_offset;
 844         u64 total_bytes = 0;
 845         u64 csum_offset;
 846         u64 bytenr;
 847         u32 nritems;
 848         u32 ins_size;
 849         int index = 0;
 850         int found_next;
 851         int ret;
 852         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 853
 854         path = btrfs_alloc_path();
 855         if (!path)
 856                 return -ENOMEM;
 857 again:
 858         next_offset = (u64)-1;
 859         found_next = 0;
 860         bytenr = sums->bytenr + total_bytes;
 861         file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
 862         file_key.offset = bytenr;
 863         file_key.type = BTRFS_EXTENT_CSUM_KEY;
 864
 865         item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
 866         if (!IS_ERR(item)) {
 867                 ret = 0;
 868                 leaf = path->nodes[0];
 869                 item_end = btrfs_item_ptr(leaf, path->slots[0],
 870                                           struct btrfs_csum_item);
 871                 item_end = (struct btrfs_csum_item *)((char *)item_end +
 872                            btrfs_item_size_nr(leaf, path->slots[0]));
 873                 goto found;
 874         }
 875         ret = PTR_ERR(item);
 876         if (ret != -EFBIG && ret != -ENOENT)
 877                 goto out;
 878
 879         if (ret == -EFBIG) {
 880                 u32 item_size;
 881                 /* we found one, but it isn't big enough yet */
 882                 leaf = path->nodes[0];
 883                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
 884                 if ((item_size / csum_size) >=
 885                     MAX_CSUM_ITEMS(fs_info, csum_size)) {
 886                         /* already at max size, make a new one */
 887                         goto insert;
 888                 }
 889         } else {
 890                 int slot = path->slots[0] + 1;
 891                 /* we didn't find a csum item, insert one */
 892                 nritems = btrfs_header_nritems(path->nodes[0]);
 893                 if (!nritems || (path->slots[0] >= nritems - 1)) {
 894                         ret = btrfs_next_leaf(root, path);
 895                         if (ret < 0) {
 896                                 goto out;
 897                         } else if (ret > 0) {
 898                                 found_next = 1;
 899                                 goto insert;
 900                         }
 901                         slot = path->slots[0];
 902                 }
 903                 btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
 904                 if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
 905                     found_key.type != BTRFS_EXTENT_CSUM_KEY) {
 906                         found_next = 1;
 907                         goto insert;
 908                 }
 909                 next_offset = found_key.offset;
 910                 found_next = 1;
 911                 goto insert;
 912         }
 913
 914         /*
 915          * At this point, we know the tree has a checksum item that ends at an
 916          * offset matching the start of the checksum range we want to insert.
 917          * We try to extend that item as much as possible and then add as many
 918          * checksums to it as they fit.
 919          *
 920          * First check if the leaf has enough free space for at least one
 921          * checksum. If it has go directly to the item extension code, otherwise
 922          * release the path and do a search for insertion before the extension.
 923          */
 924         if (btrfs_leaf_free_space(leaf) >= csum_size) {
 925                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 926                 csum_offset = (bytenr - found_key.offset) >>
 927                         fs_info->sb->s_blocksize_bits;
 928                 goto extend_csum;
 929         }
 930
 931         btrfs_release_path(path);
 932         ret = btrfs_search_slot(trans, root, &file_key, path,
 933                                 csum_size, 1);
 934         if (ret < 0)
 935                 goto out;
 936
 937         if (ret > 0) {
 938                 if (path->slots[0] == 0)
 939                         goto insert;
 940                 path->slots[0]--;
 941         }
 942
 943         leaf = path->nodes[0];
 944         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 945         csum_offset = (bytenr - found_key.offset) >>
 946                         fs_info->sb->s_blocksize_bits;
 947
 948         if (found_key.type != BTRFS_EXTENT_CSUM_KEY ||
 949             found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
 950             csum_offset >= MAX_CSUM_ITEMS(fs_info, csum_size)) {
 951                 goto insert;
 952         }
 953
 954 extend_csum:
 955         if (csum_offset == btrfs_item_size_nr(leaf, path->slots[0]) /
 956             csum_size) {
 957                 int extend_nr;
 958                 u64 tmp;
 959                 u32 diff;
 960
 961                 tmp = sums->len - total_bytes;
 962                 tmp >>= fs_info->sb->s_blocksize_bits;
 963                 WARN_ON(tmp < 1);
 964
 965                 extend_nr = max_t(int, 1, (int)tmp);
 966                 diff = (csum_offset + extend_nr) * csum_size;
 967                 diff = min(diff,
 968                            MAX_CSUM_ITEMS(fs_info, csum_size) * csum_size);
 969
 970                 diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
 971                 diff = min_t(u32, btrfs_leaf_free_space(leaf), diff);
 972                 diff /= csum_size;
 973                 diff *= csum_size;
 974
 975                 btrfs_extend_item(path, diff);
 976                 ret = 0;
 977                 goto csum;
 978         }
 979
 980 insert:
 981         btrfs_release_path(path);
 982         csum_offset = 0;
 983         if (found_next) {
 984                 u64 tmp;
 985
 986                 tmp = sums->len - total_bytes;
 987                 tmp >>= fs_info->sb->s_blocksize_bits;
 988                 tmp = min(tmp, (next_offset - file_key.offset) >>
 989                                          fs_info->sb->s_blocksize_bits);
 990
 991                 tmp = max_t(u64, 1, tmp);
 992                 tmp = min_t(u64, tmp, MAX_CSUM_ITEMS(fs_info, csum_size));
 993                 ins_size = csum_size * tmp;
 994         } else {
 995                 ins_size = csum_size;
 996         }
 997         path->leave_spinning = 1;
 998         ret = btrfs_insert_empty_item(trans, root, path, &file_key,
 999                                       ins_size);
1000         path->leave_spinning = 0;
1001         if (ret < 0)
1002                 goto out;
1003         if (WARN_ON(ret != 0))
1004                 goto out;
1005         leaf = path->nodes[0];
1006 csum:
1007         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
1008         item_end = (struct btrfs_csum_item *)((unsigned char *)item +
1009                                       btrfs_item_size_nr(leaf, path->slots[0]));
1010         item = (struct btrfs_csum_item *)((unsigned char *)item +
1011                                           csum_offset * csum_size);
1012 found:
1013         ins_size = (u32)(sums->len - total_bytes) >>
1014                    fs_info->sb->s_blocksize_bits;
1015         ins_size *= csum_size;
1016         ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item,
1017                               ins_size);
1018         write_extent_buffer(leaf, sums->sums + index, (unsigned long)item,
1019                             ins_size);
1020
1021         index += ins_size;
1022         ins_size /= csum_size;
1023         total_bytes += ins_size * fs_info->sectorsize;
1024
1025         btrfs_mark_buffer_dirty(path->nodes[0]);
1026         if (total_bytes < sums->len) {
1027                 btrfs_release_path(path);
1028                 cond_resched();
1029                 goto again;
1030         }
1031 out:
1032         btrfs_free_path(path);
1033         return ret;
1034 }
1035
1036 void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode,
1037                                      const struct btrfs_path *path,
1038                                      struct btrfs_file_extent_item *fi,
1039                                      const bool new_inline,
1040                                      struct extent_map *em)
1041 {
1042         struct btrfs_fs_info *fs_info = inode->root->fs_info;
1043         struct btrfs_root *root = inode->root;
1044         struct extent_buffer *leaf = path->nodes[0];
1045         const int slot = path->slots[0];
1046         struct btrfs_key key;
1047         u64 extent_start, extent_end;
1048         u64 bytenr;
1049         u8 type = btrfs_file_extent_type(leaf, fi);
1050         int compress_type = btrfs_file_extent_compression(leaf, fi);
1051
1052         btrfs_item_key_to_cpu(leaf, &key, slot);
1053         extent_start = key.offset;
1054         extent_end = btrfs_file_extent_end(path);
1055         em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1056         if (type == BTRFS_FILE_EXTENT_REG ||
1057             type == BTRFS_FILE_EXTENT_PREALLOC) {
1058                 em->start = extent_start;
1059                 em->len = extent_end - extent_start;
1060                 em->orig_start = extent_start -
1061                         btrfs_file_extent_offset(leaf, fi);
1062                 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
1063                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1064                 if (bytenr == 0) {
1065                         em->block_start = EXTENT_MAP_HOLE;
1066                         return;
1067                 }
1068                 if (compress_type != BTRFS_COMPRESS_NONE) {
1069                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1070                         em->compress_type = compress_type;
1071                         em->block_start = bytenr;
1072                         em->block_len = em->orig_block_len;
1073                 } else {
1074                         bytenr += btrfs_file_extent_offset(leaf, fi);
1075                         em->block_start = bytenr;
1076                         em->block_len = em->len;
1077                         if (type == BTRFS_FILE_EXTENT_PREALLOC)
1078                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
1079                 }
1080         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1081                 em->block_start = EXTENT_MAP_INLINE;
1082                 em->start = extent_start;
1083                 em->len = extent_end - extent_start;
1084                 /*
1085                  * Initialize orig_start and block_len with the same values
1086                  * as in inode.c:btrfs_get_extent().
1087                  */
1088                 em->orig_start = EXTENT_MAP_HOLE;
1089                 em->block_len = (u64)-1;
1090                 if (!new_inline && compress_type != BTRFS_COMPRESS_NONE) {
1091                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1092                         em->compress_type = compress_type;
1093                 }
1094         } else {
1095                 btrfs_err(fs_info,
1096                           "unknown file extent item type %d, inode %llu, offset %llu, "
1097                           "root %llu", type, btrfs_ino(inode), extent_start,
1098                           root->root_key.objectid);
1099         }
1100 }
1101
1102 /*
1103  * Returns the end offset (non inclusive) of the file extent item the given path
1104  * points to. If it points to an inline extent, the returned offset is rounded
1105  * up to the sector size.
1106  */
1107 u64 btrfs_file_extent_end(const struct btrfs_path *path)
1108 {
1109         const struct extent_buffer *leaf = path->nodes[0];
1110         const int slot = path->slots[0];
1111         struct btrfs_file_extent_item *fi;
1112         struct btrfs_key key;
1113         u64 end;
1114
1115         btrfs_item_key_to_cpu(leaf, &key, slot);
1116         ASSERT(key.type == BTRFS_EXTENT_DATA_KEY);
1117         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1118
1119         if (btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_INLINE) {
1120                 end = btrfs_file_extent_ram_bytes(leaf, fi);
1121                 end = ALIGN(key.offset + end, leaf->fs_info->sectorsize);
1122         } else {
1123                 end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1124         }
1125
1126         return end;
1127 }