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