fs/btrfs/file-item.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/bio.h>
  20 #include <linux/slab.h>
  21 #include <linux/pagemap.h>
  22 #include <linux/highmem.h>
  23 #include "ctree.h"
  24 #include "disk-io.h"
  25 #include "transaction.h"
  26 #include "volumes.h"
  27 #include "print-tree.h"
  28 #include "compression.h"
  29
  30 #define __MAX_CSUM_ITEMS(r, size) ((unsigned long)(((BTRFS_LEAF_DATA_SIZE(r) - \
  31                                    sizeof(struct btrfs_item) * 2) / \
  32                                   size) - 1))
  33
  34 #define MAX_CSUM_ITEMS(r, size) (min_t(u32, __MAX_CSUM_ITEMS(r, size), \
  35                                        PAGE_SIZE))
  36
  37 #define MAX_ORDERED_SUM_BYTES(fs_info) ((PAGE_SIZE - \
  38                                    sizeof(struct btrfs_ordered_sum)) / \
  39                                    sizeof(u32) * (fs_info)->sectorsize)
  40
  41 int btrfs_insert_file_extent(struct btrfs_trans_handle *trans,
  42                              struct btrfs_root *root,
  43                              u64 objectid, u64 pos,
  44                              u64 disk_offset, u64 disk_num_bytes,
  45                              u64 num_bytes, u64 offset, u64 ram_bytes,
  46                              u8 compression, u8 encryption, u16 other_encoding)
  47 {
  48         int ret = 0;
  49         struct btrfs_file_extent_item *item;
  50         struct btrfs_key file_key;
  51         struct btrfs_path *path;
  52         struct extent_buffer *leaf;
  53
  54         path = btrfs_alloc_path();
  55         if (!path)
  56                 return -ENOMEM;
  57         file_key.objectid = objectid;
  58         file_key.offset = pos;
  59         file_key.type = BTRFS_EXTENT_DATA_KEY;
  60
  61         path->leave_spinning = 1;
  62         ret = btrfs_insert_empty_item(trans, root, path, &file_key,
  63                                       sizeof(*item));
  64         if (ret < 0)
  65                 goto out;
  66         BUG_ON(ret); /* Can't happen */
  67         leaf = path->nodes[0];
  68         item = btrfs_item_ptr(leaf, path->slots[0],
  69                               struct btrfs_file_extent_item);
  70         btrfs_set_file_extent_disk_bytenr(leaf, item, disk_offset);
  71         btrfs_set_file_extent_disk_num_bytes(leaf, item, disk_num_bytes);
  72         btrfs_set_file_extent_offset(leaf, item, offset);
  73         btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
  74         btrfs_set_file_extent_ram_bytes(leaf, item, ram_bytes);
  75         btrfs_set_file_extent_generation(leaf, item, trans->transid);
  76         btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
  77         btrfs_set_file_extent_compression(leaf, item, compression);
  78         btrfs_set_file_extent_encryption(leaf, item, encryption);
  79         btrfs_set_file_extent_other_encoding(leaf, item, other_encoding);
  80
  81         btrfs_mark_buffer_dirty(leaf);
  82 out:
  83         btrfs_free_path(path);
  84         return ret;
  85 }
  86
  87 static struct btrfs_csum_item *
  88 btrfs_lookup_csum(struct btrfs_trans_handle *trans,
  89                   struct btrfs_root *root,
  90                   struct btrfs_path *path,
  91                   u64 bytenr, int cow)
  92 {
  93         struct btrfs_fs_info *fs_info = root->fs_info;
  94         int ret;
  95         struct btrfs_key file_key;
  96         struct btrfs_key found_key;
  97         struct btrfs_csum_item *item;
  98         struct extent_buffer *leaf;
  99         u64 csum_offset = 0;
 100         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 101         int csums_in_item;
 102
 103         file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
 104         file_key.offset = bytenr;
 105         file_key.type = BTRFS_EXTENT_CSUM_KEY;
 106         ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
 107         if (ret < 0)
 108                 goto fail;
 109         leaf = path->nodes[0];
 110         if (ret > 0) {
 111                 ret = 1;
 112                 if (path->slots[0] == 0)
 113                         goto fail;
 114                 path->slots[0]--;
 115                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 116                 if (found_key.type != BTRFS_EXTENT_CSUM_KEY)
 117                         goto fail;
 118
 119                 csum_offset = (bytenr - found_key.offset) >>
 120                                 fs_info->sb->s_blocksize_bits;
 121                 csums_in_item = btrfs_item_size_nr(leaf, path->slots[0]);
 122                 csums_in_item /= csum_size;
 123
 124                 if (csum_offset == csums_in_item) {
 125                         ret = -EFBIG;
 126                         goto fail;
 127                 } else if (csum_offset > csums_in_item) {
 128                         goto fail;
 129                 }
 130         }
 131         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
 132         item = (struct btrfs_csum_item *)((unsigned char *)item +
 133                                           csum_offset * csum_size);
 134         return item;
 135 fail:
 136         if (ret > 0)
 137                 ret = -ENOENT;
 138         return ERR_PTR(ret);
 139 }
 140
 141 int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
 142                              struct btrfs_root *root,
 143                              struct btrfs_path *path, u64 objectid,
 144                              u64 offset, int mod)
 145 {
 146         int ret;
 147         struct btrfs_key file_key;
 148         int ins_len = mod < 0 ? -1 : 0;
 149         int cow = mod != 0;
 150
 151         file_key.objectid = objectid;
 152         file_key.offset = offset;
 153         file_key.type = BTRFS_EXTENT_DATA_KEY;
 154         ret = btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
 155         return ret;
 156 }
 157
 158 static void btrfs_io_bio_endio_readpage(struct btrfs_io_bio *bio, int err)
 159 {
 160         kfree(bio->csum_allocated);
 161 }
 162
 163 static blk_status_t __btrfs_lookup_bio_sums(struct inode *inode, struct bio *bio,
 164                                    u64 logical_offset, u32 *dst, int dio)
 165 {
 166         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 167         struct bio_vec *bvec;
 168         struct btrfs_io_bio *btrfs_bio = btrfs_io_bio(bio);
 169         struct btrfs_csum_item *item = NULL;
 170         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 171         struct btrfs_path *path;
 172         u8 *csum;
 173         u64 offset = 0;
 174         u64 item_start_offset = 0;
 175         u64 item_last_offset = 0;
 176         u64 disk_bytenr;
 177         u64 page_bytes_left;
 178         u32 diff;
 179         int nblocks;
 180         int count = 0, i;
 181         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 182
 183         path = btrfs_alloc_path();
 184         if (!path)
 185                 return BLK_STS_RESOURCE;
 186
 187         nblocks = bio->bi_iter.bi_size >> inode->i_sb->s_blocksize_bits;
 188         if (!dst) {
 189                 if (nblocks * csum_size > BTRFS_BIO_INLINE_CSUM_SIZE) {
 190                         btrfs_bio->csum_allocated = kmalloc_array(nblocks,
 191                                         csum_size, GFP_NOFS);
 192                         if (!btrfs_bio->csum_allocated) {
 193                                 btrfs_free_path(path);
 194                                 return BLK_STS_RESOURCE;
 195                         }
 196                         btrfs_bio->csum = btrfs_bio->csum_allocated;
 197                         btrfs_bio->end_io = btrfs_io_bio_endio_readpage;
 198                 } else {
 199                         btrfs_bio->csum = btrfs_bio->csum_inline;
 200                 }
 201                 csum = btrfs_bio->csum;
 202         } else {
 203                 csum = (u8 *)dst;
 204         }
 205
 206         if (bio->bi_iter.bi_size > PAGE_SIZE * 8)
 207                 path->reada = READA_FORWARD;
 208
 209         WARN_ON(bio->bi_vcnt <= 0);
 210
 211         /*
 212          * the free space stuff is only read when it hasn't been
 213          * updated in the current transaction.  So, we can safely
 214          * read from the commit root and sidestep a nasty deadlock
 215          * between reading the free space cache and updating the csum tree.
 216          */
 217         if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
 218                 path->search_commit_root = 1;
 219                 path->skip_locking = 1;
 220         }
 221
 222         disk_bytenr = (u64)bio->bi_iter.bi_sector << 9;
 223         if (dio)
 224                 offset = logical_offset;
 225
 226         bio_for_each_segment_all(bvec, bio, i) {
 227                 page_bytes_left = bvec->bv_len;
 228                 if (count)
 229                         goto next;
 230
 231                 if (!dio)
 232                         offset = page_offset(bvec->bv_page) + bvec->bv_offset;
 233                 count = btrfs_find_ordered_sum(inode, offset, disk_bytenr,
 234                                                (u32 *)csum, nblocks);
 235                 if (count)
 236                         goto found;
 237
 238                 if (!item || disk_bytenr < item_start_offset ||
 239                     disk_bytenr >= item_last_offset) {
 240                         struct btrfs_key found_key;
 241                         u32 item_size;
 242
 243                         if (item)
 244                                 btrfs_release_path(path);
 245                         item = btrfs_lookup_csum(NULL, fs_info->csum_root,
 246                                                  path, disk_bytenr, 0);
 247                         if (IS_ERR(item)) {
 248                                 count = 1;
 249                                 memset(csum, 0, csum_size);
 250                                 if (BTRFS_I(inode)->root->root_key.objectid ==
 251                                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 252                                         set_extent_bits(io_tree, offset,
 253                                                 offset + fs_info->sectorsize - 1,
 254                                                 EXTENT_NODATASUM);
 255                                 } else {
 256                                         btrfs_info_rl(fs_info,
 257                                                    "no csum found for inode %llu start %llu",
 258                                                btrfs_ino(BTRFS_I(inode)), offset);
 259                                 }
 260                                 item = NULL;
 261                                 btrfs_release_path(path);
 262                                 goto found;
 263                         }
 264                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 265                                               path->slots[0]);
 266
 267                         item_start_offset = found_key.offset;
 268                         item_size = btrfs_item_size_nr(path->nodes[0],
 269                                                        path->slots[0]);
 270                         item_last_offset = item_start_offset +
 271                                 (item_size / csum_size) *
 272                                 fs_info->sectorsize;
 273                         item = btrfs_item_ptr(path->nodes[0], path->slots[0],
 274                                               struct btrfs_csum_item);
 275                 }
 276                 /*
 277                  * this byte range must be able to fit inside
 278                  * a single leaf so it will also fit inside a u32
 279                  */
 280                 diff = disk_bytenr - item_start_offset;
 281                 diff = diff / fs_info->sectorsize;
 282                 diff = diff * csum_size;
 283                 count = min_t(int, nblocks, (item_last_offset - disk_bytenr) >>
 284                                             inode->i_sb->s_blocksize_bits);
 285                 read_extent_buffer(path->nodes[0], csum,
 286                                    ((unsigned long)item) + diff,
 287                                    csum_size * count);
 288 found:
 289                 csum += count * csum_size;
 290                 nblocks -= count;
 291 next:
 292                 while (count--) {
 293                         disk_bytenr += fs_info->sectorsize;
 294                         offset += fs_info->sectorsize;
 295                         page_bytes_left -= fs_info->sectorsize;
 296                         if (!page_bytes_left)
 297                                 break; /* move to next bio */
 298                 }
 299         }
 300
 301         WARN_ON_ONCE(count);
 302         btrfs_free_path(path);
 303         return 0;
 304 }
 305
 306 blk_status_t btrfs_lookup_bio_sums(struct inode *inode, struct bio *bio, u32 *dst)
 307 {
 308         return __btrfs_lookup_bio_sums(inode, bio, 0, dst, 0);
 309 }
 310
 311 blk_status_t btrfs_lookup_bio_sums_dio(struct inode *inode, struct bio *bio, u64 offset)
 312 {
 313         return __btrfs_lookup_bio_sums(inode, bio, offset, NULL, 1);
 314 }
 315
 316 int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
 317                              struct list_head *list, int search_commit)
 318 {
 319         struct btrfs_fs_info *fs_info = root->fs_info;
 320         struct btrfs_key key;
 321         struct btrfs_path *path;
 322         struct extent_buffer *leaf;
 323         struct btrfs_ordered_sum *sums;
 324         struct btrfs_csum_item *item;
 325         LIST_HEAD(tmplist);
 326         unsigned long offset;
 327         int ret;
 328         size_t size;
 329         u64 csum_end;
 330         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 331
 332         ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
 333                IS_ALIGNED(end + 1, fs_info->sectorsize));
 334
 335         path = btrfs_alloc_path();
 336         if (!path)
 337                 return -ENOMEM;
 338
 339         if (search_commit) {
 340                 path->skip_locking = 1;
 341                 path->reada = READA_FORWARD;
 342                 path->search_commit_root = 1;
 343         }
 344
 345         key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
 346         key.offset = start;
 347         key.type = BTRFS_EXTENT_CSUM_KEY;
 348
 349         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 350         if (ret < 0)
 351                 goto fail;
 352         if (ret > 0 && path->slots[0] > 0) {
 353                 leaf = path->nodes[0];
 354                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
 355                 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
 356                     key.type == BTRFS_EXTENT_CSUM_KEY) {
 357                         offset = (start - key.offset) >>
 358                                  fs_info->sb->s_blocksize_bits;
 359                         if (offset * csum_size <
 360                             btrfs_item_size_nr(leaf, path->slots[0] - 1))
 361                                 path->slots[0]--;
 362                 }
 363         }
 364
 365         while (start <= end) {
 366                 leaf = path->nodes[0];
 367                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 368                         ret = btrfs_next_leaf(root, path);
 369                         if (ret < 0)
 370                                 goto fail;
 371                         if (ret > 0)
 372                                 break;
 373                         leaf = path->nodes[0];
 374                 }
 375
 376                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 377                 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
 378                     key.type != BTRFS_EXTENT_CSUM_KEY ||
 379                     key.offset > end)
 380                         break;
 381
 382                 if (key.offset > start)
 383                         start = key.offset;
 384
 385                 size = btrfs_item_size_nr(leaf, path->slots[0]);
 386                 csum_end = key.offset + (size / csum_size) * fs_info->sectorsize;
 387                 if (csum_end <= start) {
 388                         path->slots[0]++;
 389                         continue;
 390                 }
 391
 392                 csum_end = min(csum_end, end + 1);
 393                 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
 394                                       struct btrfs_csum_item);
 395                 while (start < csum_end) {
 396                         size = min_t(size_t, csum_end - start,
 397                                      MAX_ORDERED_SUM_BYTES(fs_info));
 398                         sums = kzalloc(btrfs_ordered_sum_size(fs_info, size),
 399                                        GFP_NOFS);
 400                         if (!sums) {
 401                                 ret = -ENOMEM;
 402                                 goto fail;
 403                         }
 404
 405                         sums->bytenr = start;
 406                         sums->len = (int)size;
 407
 408                         offset = (start - key.offset) >>
 409                                 fs_info->sb->s_blocksize_bits;
 410                         offset *= csum_size;
 411                         size >>= fs_info->sb->s_blocksize_bits;
 412
 413                         read_extent_buffer(path->nodes[0],
 414                                            sums->sums,
 415                                            ((unsigned long)item) + offset,
 416                                            csum_size * size);
 417
 418                         start += fs_info->sectorsize * size;
 419                         list_add_tail(&sums->list, &tmplist);
 420                 }
 421                 path->slots[0]++;
 422         }
 423         ret = 0;
 424 fail:
 425         while (ret < 0 && !list_empty(&tmplist)) {
 426                 sums = list_entry(tmplist.next, struct btrfs_ordered_sum, list);
 427                 list_del(&sums->list);
 428                 kfree(sums);
 429         }
 430         list_splice_tail(&tmplist, list);
 431
 432         btrfs_free_path(path);
 433         return ret;
 434 }
 435
 436 blk_status_t btrfs_csum_one_bio(struct inode *inode, struct bio *bio,
 437                        u64 file_start, int contig)
 438 {
 439         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 440         struct btrfs_ordered_sum *sums;
 441         struct btrfs_ordered_extent *ordered = NULL;
 442         char *data;
 443         struct bio_vec *bvec;
 444         int index;
 445         int nr_sectors;
 446         int i, j;
 447         unsigned long total_bytes = 0;
 448         unsigned long this_sum_bytes = 0;
 449         u64 offset;
 450
 451         WARN_ON(bio->bi_vcnt <= 0);
 452         sums = kzalloc(btrfs_ordered_sum_size(fs_info, bio->bi_iter.bi_size),
 453                        GFP_NOFS);
 454         if (!sums)
 455                 return BLK_STS_RESOURCE;
 456
 457         sums->len = bio->bi_iter.bi_size;
 458         INIT_LIST_HEAD(&sums->list);
 459
 460         if (contig)
 461                 offset = file_start;
 462         else
 463                 offset = 0; /* shut up gcc */
 464
 465         sums->bytenr = (u64)bio->bi_iter.bi_sector << 9;
 466         index = 0;
 467
 468         bio_for_each_segment_all(bvec, bio, j) {
 469                 if (!contig)
 470                         offset = page_offset(bvec->bv_page) + bvec->bv_offset;
 471
 472                 if (!ordered) {
 473                         ordered = btrfs_lookup_ordered_extent(inode, offset);
 474                         BUG_ON(!ordered); /* Logic error */
 475                 }
 476
 477                 data = kmap_atomic(bvec->bv_page);
 478
 479                 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info,
 480                                                  bvec->bv_len + fs_info->sectorsize
 481                                                  - 1);
 482
 483                 for (i = 0; i < nr_sectors; i++) {
 484                         if (offset >= ordered->file_offset + ordered->len ||
 485                                 offset < ordered->file_offset) {
 486                                 unsigned long bytes_left;
 487
 488                                 kunmap_atomic(data);
 489                                 sums->len = this_sum_bytes;
 490                                 this_sum_bytes = 0;
 491                                 btrfs_add_ordered_sum(inode, ordered, sums);
 492                                 btrfs_put_ordered_extent(ordered);
 493
 494                                 bytes_left = bio->bi_iter.bi_size - total_bytes;
 495
 496                                 sums = kzalloc(btrfs_ordered_sum_size(fs_info, bytes_left),
 497                                                GFP_NOFS);
 498                                 BUG_ON(!sums); /* -ENOMEM */
 499                                 sums->len = bytes_left;
 500                                 ordered = btrfs_lookup_ordered_extent(inode,
 501                                                                 offset);
 502                                 ASSERT(ordered); /* Logic error */
 503                                 sums->bytenr = ((u64)bio->bi_iter.bi_sector << 9)
 504                                         + total_bytes;
 505                                 index = 0;
 506
 507                                 data = kmap_atomic(bvec->bv_page);
 508                         }
 509
 510                         sums->sums[index] = ~(u32)0;
 511                         sums->sums[index]
 512                                 = btrfs_csum_data(data + bvec->bv_offset
 513                                                 + (i * fs_info->sectorsize),
 514                                                 sums->sums[index],
 515                                                 fs_info->sectorsize);
 516                         btrfs_csum_final(sums->sums[index],
 517                                         (char *)(sums->sums + index));
 518                         index++;
 519                         offset += fs_info->sectorsize;
 520                         this_sum_bytes += fs_info->sectorsize;
 521                         total_bytes += fs_info->sectorsize;
 522                 }
 523
 524                 kunmap_atomic(data);
 525         }
 526         this_sum_bytes = 0;
 527         btrfs_add_ordered_sum(inode, ordered, sums);
 528         btrfs_put_ordered_extent(ordered);
 529         return 0;
 530 }
 531
 532 /*
 533  * helper function for csum removal, this expects the
 534  * key to describe the csum pointed to by the path, and it expects
 535  * the csum to overlap the range [bytenr, len]
 536  *
 537  * The csum should not be entirely contained in the range and the
 538  * range should not be entirely contained in the csum.
 539  *
 540  * This calls btrfs_truncate_item with the correct args based on the
 541  * overlap, and fixes up the key as required.
 542  */
 543 static noinline void truncate_one_csum(struct btrfs_fs_info *fs_info,
 544                                        struct btrfs_path *path,
 545                                        struct btrfs_key *key,
 546                                        u64 bytenr, u64 len)
 547 {
 548         struct extent_buffer *leaf;
 549         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 550         u64 csum_end;
 551         u64 end_byte = bytenr + len;
 552         u32 blocksize_bits = fs_info->sb->s_blocksize_bits;
 553
 554         leaf = path->nodes[0];
 555         csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
 556         csum_end <<= fs_info->sb->s_blocksize_bits;
 557         csum_end += key->offset;
 558
 559         if (key->offset < bytenr && csum_end <= end_byte) {
 560                 /*
 561                  *         [ bytenr - len ]
 562                  *         [   ]
 563                  *   [csum     ]
 564                  *   A simple truncate off the end of the item
 565                  */
 566                 u32 new_size = (bytenr - key->offset) >> blocksize_bits;
 567                 new_size *= csum_size;
 568                 btrfs_truncate_item(fs_info, path, new_size, 1);
 569         } else if (key->offset >= bytenr && csum_end > end_byte &&
 570                    end_byte > key->offset) {
 571                 /*
 572                  *         [ bytenr - len ]
 573                  *                 [ ]
 574                  *                 [csum     ]
 575                  * we need to truncate from the beginning of the csum
 576                  */
 577                 u32 new_size = (csum_end - end_byte) >> blocksize_bits;
 578                 new_size *= csum_size;
 579
 580                 btrfs_truncate_item(fs_info, path, new_size, 0);
 581
 582                 key->offset = end_byte;
 583                 btrfs_set_item_key_safe(fs_info, path, key);
 584         } else {
 585                 BUG();
 586         }
 587 }
 588
 589 /*
 590  * deletes the csum items from the csum tree for a given
 591  * range of bytes.
 592  */
 593 int btrfs_del_csums(struct btrfs_trans_handle *trans,
 594                     struct btrfs_fs_info *fs_info, u64 bytenr, u64 len)
 595 {
 596         struct btrfs_root *root = fs_info->csum_root;
 597         struct btrfs_path *path;
 598         struct btrfs_key key;
 599         u64 end_byte = bytenr + len;
 600         u64 csum_end;
 601         struct extent_buffer *leaf;
 602         int ret;
 603         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 604         int blocksize_bits = fs_info->sb->s_blocksize_bits;
 605
 606         path = btrfs_alloc_path();
 607         if (!path)
 608                 return -ENOMEM;
 609
 610         while (1) {
 611                 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
 612                 key.offset = end_byte - 1;
 613                 key.type = BTRFS_EXTENT_CSUM_KEY;
 614
 615                 path->leave_spinning = 1;
 616                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 617                 if (ret > 0) {
 618                         if (path->slots[0] == 0)
 619                                 break;
 620                         path->slots[0]--;
 621                 } else if (ret < 0) {
 622                         break;
 623                 }
 624
 625                 leaf = path->nodes[0];
 626                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 627
 628                 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
 629                     key.type != BTRFS_EXTENT_CSUM_KEY) {
 630                         break;
 631                 }
 632
 633                 if (key.offset >= end_byte)
 634                         break;
 635
 636                 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
 637                 csum_end <<= blocksize_bits;
 638                 csum_end += key.offset;
 639
 640                 /* this csum ends before we start, we're done */
 641                 if (csum_end <= bytenr)
 642                         break;
 643
 644                 /* delete the entire item, it is inside our range */
 645                 if (key.offset >= bytenr && csum_end <= end_byte) {
 646                         int del_nr = 1;
 647
 648                         /*
 649                          * Check how many csum items preceding this one in this
 650                          * leaf correspond to our range and then delete them all
 651                          * at once.
 652                          */
 653                         if (key.offset > bytenr && path->slots[0] > 0) {
 654                                 int slot = path->slots[0] - 1;
 655
 656                                 while (slot >= 0) {
 657                                         struct btrfs_key pk;
 658
 659                                         btrfs_item_key_to_cpu(leaf, &pk, slot);
 660                                         if (pk.offset < bytenr ||
 661                                             pk.type != BTRFS_EXTENT_CSUM_KEY ||
 662                                             pk.objectid !=
 663                                             BTRFS_EXTENT_CSUM_OBJECTID)
 664                                                 break;
 665                                         path->slots[0] = slot;
 666                                         del_nr++;
 667                                         key.offset = pk.offset;
 668                                         slot--;
 669                                 }
 670                         }
 671                         ret = btrfs_del_items(trans, root, path,
 672                                               path->slots[0], del_nr);
 673                         if (ret)
 674                                 goto out;
 675                         if (key.offset == bytenr)
 676                                 break;
 677                 } else if (key.offset < bytenr && csum_end > end_byte) {
 678                         unsigned long offset;
 679                         unsigned long shift_len;
 680                         unsigned long item_offset;
 681                         /*
 682                          *        [ bytenr - len ]
 683                          *     [csum                ]
 684                          *
 685                          * Our bytes are in the middle of the csum,
 686                          * we need to split this item and insert a new one.
 687                          *
 688                          * But we can't drop the path because the
 689                          * csum could change, get removed, extended etc.
 690                          *
 691                          * The trick here is the max size of a csum item leaves
 692                          * enough room in the tree block for a single
 693                          * item header.  So, we split the item in place,
 694                          * adding a new header pointing to the existing
 695                          * bytes.  Then we loop around again and we have
 696                          * a nicely formed csum item that we can neatly
 697                          * truncate.
 698                          */
 699                         offset = (bytenr - key.offset) >> blocksize_bits;
 700                         offset *= csum_size;
 701
 702                         shift_len = (len >> blocksize_bits) * csum_size;
 703
 704                         item_offset = btrfs_item_ptr_offset(leaf,
 705                                                             path->slots[0]);
 706
 707                         memzero_extent_buffer(leaf, item_offset + offset,
 708                                              shift_len);
 709                         key.offset = bytenr;
 710
 711                         /*
 712                          * btrfs_split_item returns -EAGAIN when the
 713                          * item changed size or key
 714                          */
 715                         ret = btrfs_split_item(trans, root, path, &key, offset);
 716                         if (ret && ret != -EAGAIN) {
 717                                 btrfs_abort_transaction(trans, ret);
 718                                 goto out;
 719                         }
 720
 721                         key.offset = end_byte - 1;
 722                 } else {
 723                         truncate_one_csum(fs_info, path, &key, bytenr, len);
 724                         if (key.offset < bytenr)
 725                                 break;
 726                 }
 727                 btrfs_release_path(path);
 728         }
 729         ret = 0;
 730 out:
 731         btrfs_free_path(path);
 732         return ret;
 733 }
 734
 735 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
 736                            struct btrfs_root *root,
 737                            struct btrfs_ordered_sum *sums)
 738 {
 739         struct btrfs_fs_info *fs_info = root->fs_info;
 740         struct btrfs_key file_key;
 741         struct btrfs_key found_key;
 742         struct btrfs_path *path;
 743         struct btrfs_csum_item *item;
 744         struct btrfs_csum_item *item_end;
 745         struct extent_buffer *leaf = NULL;
 746         u64 next_offset;
 747         u64 total_bytes = 0;
 748         u64 csum_offset;
 749         u64 bytenr;
 750         u32 nritems;
 751         u32 ins_size;
 752         int index = 0;
 753         int found_next;
 754         int ret;
 755         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 756
 757         path = btrfs_alloc_path();
 758         if (!path)
 759                 return -ENOMEM;
 760 again:
 761         next_offset = (u64)-1;
 762         found_next = 0;
 763         bytenr = sums->bytenr + total_bytes;
 764         file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
 765         file_key.offset = bytenr;
 766         file_key.type = BTRFS_EXTENT_CSUM_KEY;
 767
 768         item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
 769         if (!IS_ERR(item)) {
 770                 ret = 0;
 771                 leaf = path->nodes[0];
 772                 item_end = btrfs_item_ptr(leaf, path->slots[0],
 773                                           struct btrfs_csum_item);
 774                 item_end = (struct btrfs_csum_item *)((char *)item_end +
 775                            btrfs_item_size_nr(leaf, path->slots[0]));
 776                 goto found;
 777         }
 778         ret = PTR_ERR(item);
 779         if (ret != -EFBIG && ret != -ENOENT)
 780                 goto fail_unlock;
 781
 782         if (ret == -EFBIG) {
 783                 u32 item_size;
 784                 /* we found one, but it isn't big enough yet */
 785                 leaf = path->nodes[0];
 786                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
 787                 if ((item_size / csum_size) >=
 788                     MAX_CSUM_ITEMS(fs_info, csum_size)) {
 789                         /* already at max size, make a new one */
 790                         goto insert;
 791                 }
 792         } else {
 793                 int slot = path->slots[0] + 1;
 794                 /* we didn't find a csum item, insert one */
 795                 nritems = btrfs_header_nritems(path->nodes[0]);
 796                 if (!nritems || (path->slots[0] >= nritems - 1)) {
 797                         ret = btrfs_next_leaf(root, path);
 798                         if (ret == 1)
 799                                 found_next = 1;
 800                         if (ret != 0)
 801                                 goto insert;
 802                         slot = path->slots[0];
 803                 }
 804                 btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
 805                 if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
 806                     found_key.type != BTRFS_EXTENT_CSUM_KEY) {
 807                         found_next = 1;
 808                         goto insert;
 809                 }
 810                 next_offset = found_key.offset;
 811                 found_next = 1;
 812                 goto insert;
 813         }
 814
 815         /*
 816          * at this point, we know the tree has an item, but it isn't big
 817          * enough yet to put our csum in.  Grow it
 818          */
 819         btrfs_release_path(path);
 820         ret = btrfs_search_slot(trans, root, &file_key, path,
 821                                 csum_size, 1);
 822         if (ret < 0)
 823                 goto fail_unlock;
 824
 825         if (ret > 0) {
 826                 if (path->slots[0] == 0)
 827                         goto insert;
 828                 path->slots[0]--;
 829         }
 830
 831         leaf = path->nodes[0];
 832         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 833         csum_offset = (bytenr - found_key.offset) >>
 834                         fs_info->sb->s_blocksize_bits;
 835
 836         if (found_key.type != BTRFS_EXTENT_CSUM_KEY ||
 837             found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
 838             csum_offset >= MAX_CSUM_ITEMS(fs_info, csum_size)) {
 839                 goto insert;
 840         }
 841
 842         if (csum_offset == btrfs_item_size_nr(leaf, path->slots[0]) /
 843             csum_size) {
 844                 int extend_nr;
 845                 u64 tmp;
 846                 u32 diff;
 847                 u32 free_space;
 848
 849                 if (btrfs_leaf_free_space(fs_info, leaf) <
 850                                  sizeof(struct btrfs_item) + csum_size * 2)
 851                         goto insert;
 852
 853                 free_space = btrfs_leaf_free_space(fs_info, leaf) -
 854                                          sizeof(struct btrfs_item) - csum_size;
 855                 tmp = sums->len - total_bytes;
 856                 tmp >>= fs_info->sb->s_blocksize_bits;
 857                 WARN_ON(tmp < 1);
 858
 859                 extend_nr = max_t(int, 1, (int)tmp);
 860                 diff = (csum_offset + extend_nr) * csum_size;
 861                 diff = min(diff,
 862                            MAX_CSUM_ITEMS(fs_info, csum_size) * csum_size);
 863
 864                 diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
 865                 diff = min(free_space, diff);
 866                 diff /= csum_size;
 867                 diff *= csum_size;
 868
 869                 btrfs_extend_item(fs_info, path, diff);
 870                 ret = 0;
 871                 goto csum;
 872         }
 873
 874 insert:
 875         btrfs_release_path(path);
 876         csum_offset = 0;
 877         if (found_next) {
 878                 u64 tmp;
 879
 880                 tmp = sums->len - total_bytes;
 881                 tmp >>= fs_info->sb->s_blocksize_bits;
 882                 tmp = min(tmp, (next_offset - file_key.offset) >>
 883                                          fs_info->sb->s_blocksize_bits);
 884
 885                 tmp = max_t(u64, 1, tmp);
 886                 tmp = min_t(u64, tmp, MAX_CSUM_ITEMS(fs_info, csum_size));
 887                 ins_size = csum_size * tmp;
 888         } else {
 889                 ins_size = csum_size;
 890         }
 891         path->leave_spinning = 1;
 892         ret = btrfs_insert_empty_item(trans, root, path, &file_key,
 893                                       ins_size);
 894         path->leave_spinning = 0;
 895         if (ret < 0)
 896                 goto fail_unlock;
 897         if (WARN_ON(ret != 0))
 898                 goto fail_unlock;
 899         leaf = path->nodes[0];
 900 csum:
 901         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
 902         item_end = (struct btrfs_csum_item *)((unsigned char *)item +
 903                                       btrfs_item_size_nr(leaf, path->slots[0]));
 904         item = (struct btrfs_csum_item *)((unsigned char *)item +
 905                                           csum_offset * csum_size);
 906 found:
 907         ins_size = (u32)(sums->len - total_bytes) >>
 908                    fs_info->sb->s_blocksize_bits;
 909         ins_size *= csum_size;
 910         ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item,
 911                               ins_size);
 912         write_extent_buffer(leaf, sums->sums + index, (unsigned long)item,
 913                             ins_size);
 914
 915         ins_size /= csum_size;
 916         total_bytes += ins_size * fs_info->sectorsize;
 917         index += ins_size;
 918
 919         btrfs_mark_buffer_dirty(path->nodes[0]);
 920         if (total_bytes < sums->len) {
 921                 btrfs_release_path(path);
 922                 cond_resched();
 923                 goto again;
 924         }
 925 out:
 926         btrfs_free_path(path);
 927         return ret;
 928
 929 fail_unlock:
 930         goto out;
 931 }
 932
 933 void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode,
 934                                      const struct btrfs_path *path,
 935                                      struct btrfs_file_extent_item *fi,
 936                                      const bool new_inline,
 937                                      struct extent_map *em)
 938 {
 939         struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
 940         struct btrfs_root *root = inode->root;
 941         struct extent_buffer *leaf = path->nodes[0];
 942         const int slot = path->slots[0];
 943         struct btrfs_key key;
 944         u64 extent_start, extent_end;
 945         u64 bytenr;
 946         u8 type = btrfs_file_extent_type(leaf, fi);
 947         int compress_type = btrfs_file_extent_compression(leaf, fi);
 948
 949         em->bdev = fs_info->fs_devices->latest_bdev;
 950         btrfs_item_key_to_cpu(leaf, &key, slot);
 951         extent_start = key.offset;
 952
 953         if (type == BTRFS_FILE_EXTENT_REG ||
 954             type == BTRFS_FILE_EXTENT_PREALLOC) {
 955                 extent_end = extent_start +
 956                         btrfs_file_extent_num_bytes(leaf, fi);
 957         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
 958                 size_t size;
 959                 size = btrfs_file_extent_inline_len(leaf, slot, fi);
 960                 extent_end = ALIGN(extent_start + size,
 961                                    fs_info->sectorsize);
 962         }
 963
 964         em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
 965         if (type == BTRFS_FILE_EXTENT_REG ||
 966             type == BTRFS_FILE_EXTENT_PREALLOC) {
 967                 em->start = extent_start;
 968                 em->len = extent_end - extent_start;
 969                 em->orig_start = extent_start -
 970                         btrfs_file_extent_offset(leaf, fi);
 971                 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
 972                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
 973                 if (bytenr == 0) {
 974                         em->block_start = EXTENT_MAP_HOLE;
 975                         return;
 976                 }
 977                 if (compress_type != BTRFS_COMPRESS_NONE) {
 978                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 979                         em->compress_type = compress_type;
 980                         em->block_start = bytenr;
 981                         em->block_len = em->orig_block_len;
 982                 } else {
 983                         bytenr += btrfs_file_extent_offset(leaf, fi);
 984                         em->block_start = bytenr;
 985                         em->block_len = em->len;
 986                         if (type == BTRFS_FILE_EXTENT_PREALLOC)
 987                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
 988                 }
 989         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
 990                 em->block_start = EXTENT_MAP_INLINE;
 991                 em->start = extent_start;
 992                 em->len = extent_end - extent_start;
 993                 /*
 994                  * Initialize orig_start and block_len with the same values
 995                  * as in inode.c:btrfs_get_extent().
 996                  */
 997                 em->orig_start = EXTENT_MAP_HOLE;
 998                 em->block_len = (u64)-1;
 999                 if (!new_inline && compress_type != BTRFS_COMPRESS_NONE) {
1000                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1001                         em->compress_type = compress_type;
1002                 }
1003         } else {
1004                 btrfs_err(fs_info,
1005                           "unknown file extent item type %d, inode %llu, offset %llu, "
1006                           "root %llu", type, btrfs_ino(inode), extent_start,
1007                           root->root_key.objectid);
1008         }
1009 }