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