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 btrfs_inode *inode, u64 new_i_size)
  42 {
  43         struct btrfs_fs_info *fs_info = inode->root->fs_info;
  44         u64 start, end, i_size;
  45         int ret;
  46
  47         i_size = new_i_size ?: i_size_read(&inode->vfs_inode);
  48         if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
  49                 inode->disk_i_size = i_size;
  50                 return;
  51         }
  52
  53         spin_lock(&inode->lock);
  54         ret = find_contiguous_extent_bit(&inode->file_extent_tree, 0, &start,
  55                                          &end, EXTENT_DIRTY);
  56         if (!ret && start == 0)
  57                 i_size = min(i_size, end + 1);
  58         else
  59                 i_size = 0;
  60         inode->disk_i_size = i_size;
  61         spin_unlock(&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         ret = btrfs_insert_empty_item(trans, root, path, &file_key,
 146                                       sizeof(*item));
 147         if (ret < 0)
 148                 goto out;
 149         BUG_ON(ret); /* Can't happen */
 150         leaf = path->nodes[0];
 151         item = btrfs_item_ptr(leaf, path->slots[0],
 152                               struct btrfs_file_extent_item);
 153         btrfs_set_file_extent_disk_bytenr(leaf, item, disk_offset);
 154         btrfs_set_file_extent_disk_num_bytes(leaf, item, disk_num_bytes);
 155         btrfs_set_file_extent_offset(leaf, item, offset);
 156         btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
 157         btrfs_set_file_extent_ram_bytes(leaf, item, ram_bytes);
 158         btrfs_set_file_extent_generation(leaf, item, trans->transid);
 159         btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
 160         btrfs_set_file_extent_compression(leaf, item, compression);
 161         btrfs_set_file_extent_encryption(leaf, item, encryption);
 162         btrfs_set_file_extent_other_encoding(leaf, item, other_encoding);
 163
 164         btrfs_mark_buffer_dirty(leaf);
 165 out:
 166         btrfs_free_path(path);
 167         return ret;
 168 }
 169
 170 static struct btrfs_csum_item *
 171 btrfs_lookup_csum(struct btrfs_trans_handle *trans,
 172                   struct btrfs_root *root,
 173                   struct btrfs_path *path,
 174                   u64 bytenr, int cow)
 175 {
 176         struct btrfs_fs_info *fs_info = root->fs_info;
 177         int ret;
 178         struct btrfs_key file_key;
 179         struct btrfs_key found_key;
 180         struct btrfs_csum_item *item;
 181         struct extent_buffer *leaf;
 182         u64 csum_offset = 0;
 183         const u32 csum_size = fs_info->csum_size;
 184         int csums_in_item;
 185
 186         file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
 187         file_key.offset = bytenr;
 188         file_key.type = BTRFS_EXTENT_CSUM_KEY;
 189         ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
 190         if (ret < 0)
 191                 goto fail;
 192         leaf = path->nodes[0];
 193         if (ret > 0) {
 194                 ret = 1;
 195                 if (path->slots[0] == 0)
 196                         goto fail;
 197                 path->slots[0]--;
 198                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 199                 if (found_key.type != BTRFS_EXTENT_CSUM_KEY)
 200                         goto fail;
 201
 202                 csum_offset = (bytenr - found_key.offset) >>
 203                                 fs_info->sectorsize_bits;
 204                 csums_in_item = btrfs_item_size_nr(leaf, path->slots[0]);
 205                 csums_in_item /= csum_size;
 206
 207                 if (csum_offset == csums_in_item) {
 208                         ret = -EFBIG;
 209                         goto fail;
 210                 } else if (csum_offset > csums_in_item) {
 211                         goto fail;
 212                 }
 213         }
 214         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
 215         item = (struct btrfs_csum_item *)((unsigned char *)item +
 216                                           csum_offset * csum_size);
 217         return item;
 218 fail:
 219         if (ret > 0)
 220                 ret = -ENOENT;
 221         return ERR_PTR(ret);
 222 }
 223
 224 int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
 225                              struct btrfs_root *root,
 226                              struct btrfs_path *path, u64 objectid,
 227                              u64 offset, int mod)
 228 {
 229         int ret;
 230         struct btrfs_key file_key;
 231         int ins_len = mod < 0 ? -1 : 0;
 232         int cow = mod != 0;
 233
 234         file_key.objectid = objectid;
 235         file_key.offset = offset;
 236         file_key.type = BTRFS_EXTENT_DATA_KEY;
 237         ret = btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
 238         return ret;
 239 }
 240
 241 /**
 242  * btrfs_lookup_bio_sums - Look up checksums for a read bio.
 243  *
 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         const u32 csum_size = fs_info->csum_size;
 274
 275         if (!fs_info->csum_root || (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
 276                 return BLK_STS_OK;
 277
 278         /*
 279          * This function is only called for read bio.
 280          *
 281          * This means two things:
 282          * - All our csums should only be in csum tree
 283          *   No ordered extents csums, as ordered extents are only for write
 284          *   path.
 285          */
 286         ASSERT(bio_op(bio) == REQ_OP_READ);
 287         path = btrfs_alloc_path();
 288         if (!path)
 289                 return BLK_STS_RESOURCE;
 290
 291         nblocks = bio->bi_iter.bi_size >> fs_info->sectorsize_bits;
 292         if (!dst) {
 293                 struct btrfs_io_bio *btrfs_bio = btrfs_io_bio(bio);
 294
 295                 if (nblocks * csum_size > BTRFS_BIO_INLINE_CSUM_SIZE) {
 296                         btrfs_bio->csum = kmalloc_array(nblocks, csum_size,
 297                                                         GFP_NOFS);
 298                         if (!btrfs_bio->csum) {
 299                                 btrfs_free_path(path);
 300                                 return BLK_STS_RESOURCE;
 301                         }
 302                 } else {
 303                         btrfs_bio->csum = btrfs_bio->csum_inline;
 304                 }
 305                 csum = btrfs_bio->csum;
 306         } else {
 307                 csum = dst;
 308         }
 309
 310         /*
 311          * If requested number of sectors is larger than one leaf can contain,
 312          * kick the readahead for csum tree.
 313          */
 314         if (nblocks > fs_info->csums_per_leaf)
 315                 path->reada = READA_FORWARD;
 316
 317         /*
 318          * the free space stuff is only read when it hasn't been
 319          * updated in the current transaction.  So, we can safely
 320          * read from the commit root and sidestep a nasty deadlock
 321          * between reading the free space cache and updating the csum tree.
 322          */
 323         if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
 324                 path->search_commit_root = 1;
 325                 path->skip_locking = 1;
 326         }
 327
 328         disk_bytenr = bio->bi_iter.bi_sector << 9;
 329
 330         bio_for_each_segment(bvec, bio, iter) {
 331                 page_bytes_left = bvec.bv_len;
 332                 if (count)
 333                         goto next;
 334
 335                 if (page_offsets)
 336                         offset = page_offset(bvec.bv_page) + bvec.bv_offset;
 337
 338                 if (!item || disk_bytenr < item_start_offset ||
 339                     disk_bytenr >= item_last_offset) {
 340                         struct btrfs_key found_key;
 341                         u32 item_size;
 342
 343                         if (item)
 344                                 btrfs_release_path(path);
 345                         item = btrfs_lookup_csum(NULL, fs_info->csum_root,
 346                                                  path, disk_bytenr, 0);
 347                         if (IS_ERR(item)) {
 348                                 count = 1;
 349                                 memset(csum, 0, csum_size);
 350                                 if (BTRFS_I(inode)->root->root_key.objectid ==
 351                                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 352                                         set_extent_bits(io_tree, offset,
 353                                                 offset + fs_info->sectorsize - 1,
 354                                                 EXTENT_NODATASUM);
 355                                 } else {
 356                                         btrfs_info_rl(fs_info,
 357                                                    "no csum found for inode %llu start %llu",
 358                                                btrfs_ino(BTRFS_I(inode)), offset);
 359                                 }
 360                                 item = NULL;
 361                                 btrfs_release_path(path);
 362                                 goto found;
 363                         }
 364                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 365                                               path->slots[0]);
 366
 367                         item_start_offset = found_key.offset;
 368                         item_size = btrfs_item_size_nr(path->nodes[0],
 369                                                        path->slots[0]);
 370                         item_last_offset = item_start_offset +
 371                                 (item_size / csum_size) *
 372                                 fs_info->sectorsize;
 373                         item = btrfs_item_ptr(path->nodes[0], path->slots[0],
 374                                               struct btrfs_csum_item);
 375                 }
 376                 /*
 377                  * this byte range must be able to fit inside
 378                  * a single leaf so it will also fit inside a u32
 379                  */
 380                 diff = disk_bytenr - item_start_offset;
 381                 diff = diff >> fs_info->sectorsize_bits;
 382                 diff = diff * csum_size;
 383                 count = min_t(int, nblocks, (item_last_offset - disk_bytenr) >>
 384                                             fs_info->sectorsize_bits);
 385                 read_extent_buffer(path->nodes[0], csum,
 386                                    ((unsigned long)item) + diff,
 387                                    csum_size * count);
 388 found:
 389                 csum += count * csum_size;
 390                 nblocks -= count;
 391 next:
 392                 while (count > 0) {
 393                         count--;
 394                         disk_bytenr += fs_info->sectorsize;
 395                         offset += fs_info->sectorsize;
 396                         page_bytes_left -= fs_info->sectorsize;
 397                         if (!page_bytes_left)
 398                                 break; /* move to next bio */
 399                 }
 400         }
 401
 402         WARN_ON_ONCE(count);
 403         btrfs_free_path(path);
 404         return BLK_STS_OK;
 405 }
 406
 407 int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
 408                              struct list_head *list, int search_commit)
 409 {
 410         struct btrfs_fs_info *fs_info = root->fs_info;
 411         struct btrfs_key key;
 412         struct btrfs_path *path;
 413         struct extent_buffer *leaf;
 414         struct btrfs_ordered_sum *sums;
 415         struct btrfs_csum_item *item;
 416         LIST_HEAD(tmplist);
 417         unsigned long offset;
 418         int ret;
 419         size_t size;
 420         u64 csum_end;
 421         const u32 csum_size = fs_info->csum_size;
 422
 423         ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
 424                IS_ALIGNED(end + 1, fs_info->sectorsize));
 425
 426         path = btrfs_alloc_path();
 427         if (!path)
 428                 return -ENOMEM;
 429
 430         if (search_commit) {
 431                 path->skip_locking = 1;
 432                 path->reada = READA_FORWARD;
 433                 path->search_commit_root = 1;
 434         }
 435
 436         key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
 437         key.offset = start;
 438         key.type = BTRFS_EXTENT_CSUM_KEY;
 439
 440         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 441         if (ret < 0)
 442                 goto fail;
 443         if (ret > 0 && path->slots[0] > 0) {
 444                 leaf = path->nodes[0];
 445                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
 446                 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
 447                     key.type == BTRFS_EXTENT_CSUM_KEY) {
 448                         offset = (start - key.offset) >> fs_info->sectorsize_bits;
 449                         if (offset * csum_size <
 450                             btrfs_item_size_nr(leaf, path->slots[0] - 1))
 451                                 path->slots[0]--;
 452                 }
 453         }
 454
 455         while (start <= end) {
 456                 leaf = path->nodes[0];
 457                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 458                         ret = btrfs_next_leaf(root, path);
 459                         if (ret < 0)
 460                                 goto fail;
 461                         if (ret > 0)
 462                                 break;
 463                         leaf = path->nodes[0];
 464                 }
 465
 466                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 467                 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
 468                     key.type != BTRFS_EXTENT_CSUM_KEY ||
 469                     key.offset > end)
 470                         break;
 471
 472                 if (key.offset > start)
 473                         start = key.offset;
 474
 475                 size = btrfs_item_size_nr(leaf, path->slots[0]);
 476                 csum_end = key.offset + (size / csum_size) * fs_info->sectorsize;
 477                 if (csum_end <= start) {
 478                         path->slots[0]++;
 479                         continue;
 480                 }
 481
 482                 csum_end = min(csum_end, end + 1);
 483                 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
 484                                       struct btrfs_csum_item);
 485                 while (start < csum_end) {
 486                         size = min_t(size_t, csum_end - start,
 487                                      max_ordered_sum_bytes(fs_info, csum_size));
 488                         sums = kzalloc(btrfs_ordered_sum_size(fs_info, size),
 489                                        GFP_NOFS);
 490                         if (!sums) {
 491                                 ret = -ENOMEM;
 492                                 goto fail;
 493                         }
 494
 495                         sums->bytenr = start;
 496                         sums->len = (int)size;
 497
 498                         offset = (start - key.offset) >> fs_info->sectorsize_bits;
 499                         offset *= csum_size;
 500                         size >>= fs_info->sectorsize_bits;
 501
 502                         read_extent_buffer(path->nodes[0],
 503                                            sums->sums,
 504                                            ((unsigned long)item) + offset,
 505                                            csum_size * size);
 506
 507                         start += fs_info->sectorsize * size;
 508                         list_add_tail(&sums->list, &tmplist);
 509                 }
 510                 path->slots[0]++;
 511         }
 512         ret = 0;
 513 fail:
 514         while (ret < 0 && !list_empty(&tmplist)) {
 515                 sums = list_entry(tmplist.next, struct btrfs_ordered_sum, list);
 516                 list_del(&sums->list);
 517                 kfree(sums);
 518         }
 519         list_splice_tail(&tmplist, list);
 520
 521         btrfs_free_path(path);
 522         return ret;
 523 }
 524
 525 /*
 526  * btrfs_csum_one_bio - Calculates checksums of the data contained inside a bio
 527  * @inode:       Owner of the data inside the bio
 528  * @bio:         Contains the data to be checksummed
 529  * @file_start:  offset in file this bio begins to describe
 530  * @contig:      Boolean. If true/1 means all bio vecs in this bio are
 531  *               contiguous and they begin at @file_start in the file. False/0
 532  *               means this bio can contains potentially discontigous bio vecs
 533  *               so the logical offset of each should be calculated separately.
 534  */
 535 blk_status_t btrfs_csum_one_bio(struct btrfs_inode *inode, struct bio *bio,
 536                        u64 file_start, int contig)
 537 {
 538         struct btrfs_fs_info *fs_info = inode->root->fs_info;
 539         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
 540         struct btrfs_ordered_sum *sums;
 541         struct btrfs_ordered_extent *ordered = NULL;
 542         char *data;
 543         struct bvec_iter iter;
 544         struct bio_vec bvec;
 545         int index;
 546         int nr_sectors;
 547         unsigned long total_bytes = 0;
 548         unsigned long this_sum_bytes = 0;
 549         int i;
 550         u64 offset;
 551         unsigned nofs_flag;
 552
 553         nofs_flag = memalloc_nofs_save();
 554         sums = kvzalloc(btrfs_ordered_sum_size(fs_info, bio->bi_iter.bi_size),
 555                        GFP_KERNEL);
 556         memalloc_nofs_restore(nofs_flag);
 557
 558         if (!sums)
 559                 return BLK_STS_RESOURCE;
 560
 561         sums->len = bio->bi_iter.bi_size;
 562         INIT_LIST_HEAD(&sums->list);
 563
 564         if (contig)
 565                 offset = file_start;
 566         else
 567                 offset = 0; /* shut up gcc */
 568
 569         sums->bytenr = bio->bi_iter.bi_sector << 9;
 570         index = 0;
 571
 572         shash->tfm = fs_info->csum_shash;
 573
 574         bio_for_each_segment(bvec, bio, iter) {
 575                 if (!contig)
 576                         offset = page_offset(bvec.bv_page) + bvec.bv_offset;
 577
 578                 if (!ordered) {
 579                         ordered = btrfs_lookup_ordered_extent(inode, offset);
 580                         BUG_ON(!ordered); /* Logic error */
 581                 }
 582
 583                 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info,
 584                                                  bvec.bv_len + fs_info->sectorsize
 585                                                  - 1);
 586
 587                 for (i = 0; i < nr_sectors; i++) {
 588                         if (offset >= ordered->file_offset + ordered->num_bytes ||
 589                             offset < ordered->file_offset) {
 590                                 unsigned long bytes_left;
 591
 592                                 sums->len = this_sum_bytes;
 593                                 this_sum_bytes = 0;
 594                                 btrfs_add_ordered_sum(ordered, sums);
 595                                 btrfs_put_ordered_extent(ordered);
 596
 597                                 bytes_left = bio->bi_iter.bi_size - total_bytes;
 598
 599                                 nofs_flag = memalloc_nofs_save();
 600                                 sums = kvzalloc(btrfs_ordered_sum_size(fs_info,
 601                                                       bytes_left), GFP_KERNEL);
 602                                 memalloc_nofs_restore(nofs_flag);
 603                                 BUG_ON(!sums); /* -ENOMEM */
 604                                 sums->len = bytes_left;
 605                                 ordered = btrfs_lookup_ordered_extent(inode,
 606                                                                 offset);
 607                                 ASSERT(ordered); /* Logic error */
 608                                 sums->bytenr = (bio->bi_iter.bi_sector << 9)
 609                                         + total_bytes;
 610                                 index = 0;
 611                         }
 612
 613                         data = kmap_atomic(bvec.bv_page);
 614                         crypto_shash_digest(shash, data + bvec.bv_offset
 615                                             + (i * fs_info->sectorsize),
 616                                             fs_info->sectorsize,
 617                                             sums->sums + index);
 618                         kunmap_atomic(data);
 619                         index += fs_info->csum_size;
 620                         offset += fs_info->sectorsize;
 621                         this_sum_bytes += fs_info->sectorsize;
 622                         total_bytes += fs_info->sectorsize;
 623                 }
 624
 625         }
 626         this_sum_bytes = 0;
 627         btrfs_add_ordered_sum(ordered, sums);
 628         btrfs_put_ordered_extent(ordered);
 629         return 0;
 630 }
 631
 632 /*
 633  * helper function for csum removal, this expects the
 634  * key to describe the csum pointed to by the path, and it expects
 635  * the csum to overlap the range [bytenr, len]
 636  *
 637  * The csum should not be entirely contained in the range and the
 638  * range should not be entirely contained in the csum.
 639  *
 640  * This calls btrfs_truncate_item with the correct args based on the
 641  * overlap, and fixes up the key as required.
 642  */
 643 static noinline void truncate_one_csum(struct btrfs_fs_info *fs_info,
 644                                        struct btrfs_path *path,
 645                                        struct btrfs_key *key,
 646                                        u64 bytenr, u64 len)
 647 {
 648         struct extent_buffer *leaf;
 649         const u32 csum_size = fs_info->csum_size;
 650         u64 csum_end;
 651         u64 end_byte = bytenr + len;
 652         u32 blocksize_bits = fs_info->sectorsize_bits;
 653
 654         leaf = path->nodes[0];
 655         csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
 656         csum_end <<= blocksize_bits;
 657         csum_end += key->offset;
 658
 659         if (key->offset < bytenr && csum_end <= end_byte) {
 660                 /*
 661                  *         [ bytenr - len ]
 662                  *         [   ]
 663                  *   [csum     ]
 664                  *   A simple truncate off the end of the item
 665                  */
 666                 u32 new_size = (bytenr - key->offset) >> blocksize_bits;
 667                 new_size *= csum_size;
 668                 btrfs_truncate_item(path, new_size, 1);
 669         } else if (key->offset >= bytenr && csum_end > end_byte &&
 670                    end_byte > key->offset) {
 671                 /*
 672                  *         [ bytenr - len ]
 673                  *                 [ ]
 674                  *                 [csum     ]
 675                  * we need to truncate from the beginning of the csum
 676                  */
 677                 u32 new_size = (csum_end - end_byte) >> blocksize_bits;
 678                 new_size *= csum_size;
 679
 680                 btrfs_truncate_item(path, new_size, 0);
 681
 682                 key->offset = end_byte;
 683                 btrfs_set_item_key_safe(fs_info, path, key);
 684         } else {
 685                 BUG();
 686         }
 687 }
 688
 689 /*
 690  * deletes the csum items from the csum tree for a given
 691  * range of bytes.
 692  */
 693 int btrfs_del_csums(struct btrfs_trans_handle *trans,
 694                     struct btrfs_root *root, u64 bytenr, u64 len)
 695 {
 696         struct btrfs_fs_info *fs_info = trans->fs_info;
 697         struct btrfs_path *path;
 698         struct btrfs_key key;
 699         u64 end_byte = bytenr + len;
 700         u64 csum_end;
 701         struct extent_buffer *leaf;
 702         int ret;
 703         const u32 csum_size = fs_info->csum_size;
 704         u32 blocksize_bits = fs_info->sectorsize_bits;
 705
 706         ASSERT(root == fs_info->csum_root ||
 707                root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
 708
 709         path = btrfs_alloc_path();
 710         if (!path)
 711                 return -ENOMEM;
 712
 713         while (1) {
 714                 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
 715                 key.offset = end_byte - 1;
 716                 key.type = BTRFS_EXTENT_CSUM_KEY;
 717
 718                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 719                 if (ret > 0) {
 720                         if (path->slots[0] == 0)
 721                                 break;
 722                         path->slots[0]--;
 723                 } else if (ret < 0) {
 724                         break;
 725                 }
 726
 727                 leaf = path->nodes[0];
 728                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 729
 730                 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
 731                     key.type != BTRFS_EXTENT_CSUM_KEY) {
 732                         break;
 733                 }
 734
 735                 if (key.offset >= end_byte)
 736                         break;
 737
 738                 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
 739                 csum_end <<= blocksize_bits;
 740                 csum_end += key.offset;
 741
 742                 /* this csum ends before we start, we're done */
 743                 if (csum_end <= bytenr)
 744                         break;
 745
 746                 /* delete the entire item, it is inside our range */
 747                 if (key.offset >= bytenr && csum_end <= end_byte) {
 748                         int del_nr = 1;
 749
 750                         /*
 751                          * Check how many csum items preceding this one in this
 752                          * leaf correspond to our range and then delete them all
 753                          * at once.
 754                          */
 755                         if (key.offset > bytenr && path->slots[0] > 0) {
 756                                 int slot = path->slots[0] - 1;
 757
 758                                 while (slot >= 0) {
 759                                         struct btrfs_key pk;
 760
 761                                         btrfs_item_key_to_cpu(leaf, &pk, slot);
 762                                         if (pk.offset < bytenr ||
 763                                             pk.type != BTRFS_EXTENT_CSUM_KEY ||
 764                                             pk.objectid !=
 765                                             BTRFS_EXTENT_CSUM_OBJECTID)
 766                                                 break;
 767                                         path->slots[0] = slot;
 768                                         del_nr++;
 769                                         key.offset = pk.offset;
 770                                         slot--;
 771                                 }
 772                         }
 773                         ret = btrfs_del_items(trans, root, path,
 774                                               path->slots[0], del_nr);
 775                         if (ret)
 776                                 goto out;
 777                         if (key.offset == bytenr)
 778                                 break;
 779                 } else if (key.offset < bytenr && csum_end > end_byte) {
 780                         unsigned long offset;
 781                         unsigned long shift_len;
 782                         unsigned long item_offset;
 783                         /*
 784                          *        [ bytenr - len ]
 785                          *     [csum                ]
 786                          *
 787                          * Our bytes are in the middle of the csum,
 788                          * we need to split this item and insert a new one.
 789                          *
 790                          * But we can't drop the path because the
 791                          * csum could change, get removed, extended etc.
 792                          *
 793                          * The trick here is the max size of a csum item leaves
 794                          * enough room in the tree block for a single
 795                          * item header.  So, we split the item in place,
 796                          * adding a new header pointing to the existing
 797                          * bytes.  Then we loop around again and we have
 798                          * a nicely formed csum item that we can neatly
 799                          * truncate.
 800                          */
 801                         offset = (bytenr - key.offset) >> blocksize_bits;
 802                         offset *= csum_size;
 803
 804                         shift_len = (len >> blocksize_bits) * csum_size;
 805
 806                         item_offset = btrfs_item_ptr_offset(leaf,
 807                                                             path->slots[0]);
 808
 809                         memzero_extent_buffer(leaf, item_offset + offset,
 810                                              shift_len);
 811                         key.offset = bytenr;
 812
 813                         /*
 814                          * btrfs_split_item returns -EAGAIN when the
 815                          * item changed size or key
 816                          */
 817                         ret = btrfs_split_item(trans, root, path, &key, offset);
 818                         if (ret && ret != -EAGAIN) {
 819                                 btrfs_abort_transaction(trans, ret);
 820                                 goto out;
 821                         }
 822
 823                         key.offset = end_byte - 1;
 824                 } else {
 825                         truncate_one_csum(fs_info, path, &key, bytenr, len);
 826                         if (key.offset < bytenr)
 827                                 break;
 828                 }
 829                 btrfs_release_path(path);
 830         }
 831         ret = 0;
 832 out:
 833         btrfs_free_path(path);
 834         return ret;
 835 }
 836
 837 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
 838                            struct btrfs_root *root,
 839                            struct btrfs_ordered_sum *sums)
 840 {
 841         struct btrfs_fs_info *fs_info = root->fs_info;
 842         struct btrfs_key file_key;
 843         struct btrfs_key found_key;
 844         struct btrfs_path *path;
 845         struct btrfs_csum_item *item;
 846         struct btrfs_csum_item *item_end;
 847         struct extent_buffer *leaf = NULL;
 848         u64 next_offset;
 849         u64 total_bytes = 0;
 850         u64 csum_offset;
 851         u64 bytenr;
 852         u32 nritems;
 853         u32 ins_size;
 854         int index = 0;
 855         int found_next;
 856         int ret;
 857         const u32 csum_size = fs_info->csum_size;
 858
 859         path = btrfs_alloc_path();
 860         if (!path)
 861                 return -ENOMEM;
 862 again:
 863         next_offset = (u64)-1;
 864         found_next = 0;
 865         bytenr = sums->bytenr + total_bytes;
 866         file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
 867         file_key.offset = bytenr;
 868         file_key.type = BTRFS_EXTENT_CSUM_KEY;
 869
 870         item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
 871         if (!IS_ERR(item)) {
 872                 ret = 0;
 873                 leaf = path->nodes[0];
 874                 item_end = btrfs_item_ptr(leaf, path->slots[0],
 875                                           struct btrfs_csum_item);
 876                 item_end = (struct btrfs_csum_item *)((char *)item_end +
 877                            btrfs_item_size_nr(leaf, path->slots[0]));
 878                 goto found;
 879         }
 880         ret = PTR_ERR(item);
 881         if (ret != -EFBIG && ret != -ENOENT)
 882                 goto out;
 883
 884         if (ret == -EFBIG) {
 885                 u32 item_size;
 886                 /* we found one, but it isn't big enough yet */
 887                 leaf = path->nodes[0];
 888                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
 889                 if ((item_size / csum_size) >=
 890                     MAX_CSUM_ITEMS(fs_info, csum_size)) {
 891                         /* already at max size, make a new one */
 892                         goto insert;
 893                 }
 894         } else {
 895                 int slot = path->slots[0] + 1;
 896                 /* we didn't find a csum item, insert one */
 897                 nritems = btrfs_header_nritems(path->nodes[0]);
 898                 if (!nritems || (path->slots[0] >= nritems - 1)) {
 899                         ret = btrfs_next_leaf(root, path);
 900                         if (ret < 0) {
 901                                 goto out;
 902                         } else if (ret > 0) {
 903                                 found_next = 1;
 904                                 goto insert;
 905                         }
 906                         slot = path->slots[0];
 907                 }
 908                 btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
 909                 if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
 910                     found_key.type != BTRFS_EXTENT_CSUM_KEY) {
 911                         found_next = 1;
 912                         goto insert;
 913                 }
 914                 next_offset = found_key.offset;
 915                 found_next = 1;
 916                 goto insert;
 917         }
 918
 919         /*
 920          * At this point, we know the tree has a checksum item that ends at an
 921          * offset matching the start of the checksum range we want to insert.
 922          * We try to extend that item as much as possible and then add as many
 923          * checksums to it as they fit.
 924          *
 925          * First check if the leaf has enough free space for at least one
 926          * checksum. If it has go directly to the item extension code, otherwise
 927          * release the path and do a search for insertion before the extension.
 928          */
 929         if (btrfs_leaf_free_space(leaf) >= csum_size) {
 930                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 931                 csum_offset = (bytenr - found_key.offset) >>
 932                         fs_info->sectorsize_bits;
 933                 goto extend_csum;
 934         }
 935
 936         btrfs_release_path(path);
 937         ret = btrfs_search_slot(trans, root, &file_key, path,
 938                                 csum_size, 1);
 939         if (ret < 0)
 940                 goto out;
 941
 942         if (ret > 0) {
 943                 if (path->slots[0] == 0)
 944                         goto insert;
 945                 path->slots[0]--;
 946         }
 947
 948         leaf = path->nodes[0];
 949         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 950         csum_offset = (bytenr - found_key.offset) >> fs_info->sectorsize_bits;
 951
 952         if (found_key.type != BTRFS_EXTENT_CSUM_KEY ||
 953             found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
 954             csum_offset >= MAX_CSUM_ITEMS(fs_info, csum_size)) {
 955                 goto insert;
 956         }
 957
 958 extend_csum:
 959         if (csum_offset == btrfs_item_size_nr(leaf, path->slots[0]) /
 960             csum_size) {
 961                 int extend_nr;
 962                 u64 tmp;
 963                 u32 diff;
 964
 965                 tmp = sums->len - total_bytes;
 966                 tmp >>= fs_info->sectorsize_bits;
 967                 WARN_ON(tmp < 1);
 968
 969                 extend_nr = max_t(int, 1, (int)tmp);
 970                 diff = (csum_offset + extend_nr) * csum_size;
 971                 diff = min(diff,
 972                            MAX_CSUM_ITEMS(fs_info, csum_size) * csum_size);
 973
 974                 diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
 975                 diff = min_t(u32, btrfs_leaf_free_space(leaf), diff);
 976                 diff /= csum_size;
 977                 diff *= csum_size;
 978
 979                 btrfs_extend_item(path, diff);
 980                 ret = 0;
 981                 goto csum;
 982         }
 983
 984 insert:
 985         btrfs_release_path(path);
 986         csum_offset = 0;
 987         if (found_next) {
 988                 u64 tmp;
 989
 990                 tmp = sums->len - total_bytes;
 991                 tmp >>= fs_info->sectorsize_bits;
 992                 tmp = min(tmp, (next_offset - file_key.offset) >>
 993                                          fs_info->sectorsize_bits);
 994
 995                 tmp = max_t(u64, 1, tmp);
 996                 tmp = min_t(u64, tmp, MAX_CSUM_ITEMS(fs_info, csum_size));
 997                 ins_size = csum_size * tmp;
 998         } else {
 999                 ins_size = csum_size;
1000         }
1001         ret = btrfs_insert_empty_item(trans, root, path, &file_key,
1002                                       ins_size);
1003         if (ret < 0)
1004                 goto out;
1005         if (WARN_ON(ret != 0))
1006                 goto out;
1007         leaf = path->nodes[0];
1008 csum:
1009         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
1010         item_end = (struct btrfs_csum_item *)((unsigned char *)item +
1011                                       btrfs_item_size_nr(leaf, path->slots[0]));
1012         item = (struct btrfs_csum_item *)((unsigned char *)item +
1013                                           csum_offset * csum_size);
1014 found:
1015         ins_size = (u32)(sums->len - total_bytes) >> fs_info->sectorsize_bits;
1016         ins_size *= csum_size;
1017         ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item,
1018                               ins_size);
1019         write_extent_buffer(leaf, sums->sums + index, (unsigned long)item,
1020                             ins_size);
1021
1022         index += ins_size;
1023         ins_size /= csum_size;
1024         total_bytes += ins_size * fs_info->sectorsize;
1025
1026         btrfs_mark_buffer_dirty(path->nodes[0]);
1027         if (total_bytes < sums->len) {
1028                 btrfs_release_path(path);
1029                 cond_resched();
1030                 goto again;
1031         }
1032 out:
1033         btrfs_free_path(path);
1034         return ret;
1035 }
1036
1037 void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode,
1038                                      const struct btrfs_path *path,
1039                                      struct btrfs_file_extent_item *fi,
1040                                      const bool new_inline,
1041                                      struct extent_map *em)
1042 {
1043         struct btrfs_fs_info *fs_info = inode->root->fs_info;
1044         struct btrfs_root *root = inode->root;
1045         struct extent_buffer *leaf = path->nodes[0];
1046         const int slot = path->slots[0];
1047         struct btrfs_key key;
1048         u64 extent_start, extent_end;
1049         u64 bytenr;
1050         u8 type = btrfs_file_extent_type(leaf, fi);
1051         int compress_type = btrfs_file_extent_compression(leaf, fi);
1052
1053         btrfs_item_key_to_cpu(leaf, &key, slot);
1054         extent_start = key.offset;
1055         extent_end = btrfs_file_extent_end(path);
1056         em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1057         if (type == BTRFS_FILE_EXTENT_REG ||
1058             type == BTRFS_FILE_EXTENT_PREALLOC) {
1059                 em->start = extent_start;
1060                 em->len = extent_end - extent_start;
1061                 em->orig_start = extent_start -
1062                         btrfs_file_extent_offset(leaf, fi);
1063                 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
1064                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1065                 if (bytenr == 0) {
1066                         em->block_start = EXTENT_MAP_HOLE;
1067                         return;
1068                 }
1069                 if (compress_type != BTRFS_COMPRESS_NONE) {
1070                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1071                         em->compress_type = compress_type;
1072                         em->block_start = bytenr;
1073                         em->block_len = em->orig_block_len;
1074                 } else {
1075                         bytenr += btrfs_file_extent_offset(leaf, fi);
1076                         em->block_start = bytenr;
1077                         em->block_len = em->len;
1078                         if (type == BTRFS_FILE_EXTENT_PREALLOC)
1079                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
1080                 }
1081         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1082                 em->block_start = EXTENT_MAP_INLINE;
1083                 em->start = extent_start;
1084                 em->len = extent_end - extent_start;
1085                 /*
1086                  * Initialize orig_start and block_len with the same values
1087                  * as in inode.c:btrfs_get_extent().
1088                  */
1089                 em->orig_start = EXTENT_MAP_HOLE;
1090                 em->block_len = (u64)-1;
1091                 if (!new_inline && compress_type != BTRFS_COMPRESS_NONE) {
1092                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1093                         em->compress_type = compress_type;
1094                 }
1095         } else {
1096                 btrfs_err(fs_info,
1097                           "unknown file extent item type %d, inode %llu, offset %llu, "
1098                           "root %llu", type, btrfs_ino(inode), extent_start,
1099                           root->root_key.objectid);
1100         }
1101 }
1102
1103 /*
1104  * Returns the end offset (non inclusive) of the file extent item the given path
1105  * points to. If it points to an inline extent, the returned offset is rounded
1106  * up to the sector size.
1107  */
1108 u64 btrfs_file_extent_end(const struct btrfs_path *path)
1109 {
1110         const struct extent_buffer *leaf = path->nodes[0];
1111         const int slot = path->slots[0];
1112         struct btrfs_file_extent_item *fi;
1113         struct btrfs_key key;
1114         u64 end;
1115
1116         btrfs_item_key_to_cpu(leaf, &key, slot);
1117         ASSERT(key.type == BTRFS_EXTENT_DATA_KEY);
1118         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1119
1120         if (btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_INLINE) {
1121                 end = btrfs_file_extent_ram_bytes(leaf, fi);
1122                 end = ALIGN(key.offset + end, leaf->fs_info->sectorsize);
1123         } else {
1124                 end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1125         }
1126
1127         return end;
1128 }