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         kfree(bio->csum_allocated);
 148 }
 149
 150 static blk_status_t __btrfs_lookup_bio_sums(struct inode *inode, struct bio *bio,
 151                                    u64 logical_offset, u32 *dst, int dio)
 152 {
 153         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 154         struct bio_vec bvec;
 155         struct bvec_iter iter;
 156         struct btrfs_io_bio *btrfs_bio = btrfs_io_bio(bio);
 157         struct btrfs_csum_item *item = NULL;
 158         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 159         struct btrfs_path *path;
 160         u8 *csum;
 161         u64 offset = 0;
 162         u64 item_start_offset = 0;
 163         u64 item_last_offset = 0;
 164         u64 disk_bytenr;
 165         u64 page_bytes_left;
 166         u32 diff;
 167         int nblocks;
 168         int count = 0;
 169         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 170
 171         path = btrfs_alloc_path();
 172         if (!path)
 173                 return BLK_STS_RESOURCE;
 174
 175         nblocks = bio->bi_iter.bi_size >> inode->i_sb->s_blocksize_bits;
 176         if (!dst) {
 177                 if (nblocks * csum_size > BTRFS_BIO_INLINE_CSUM_SIZE) {
 178                         btrfs_bio->csum_allocated = kmalloc_array(nblocks,
 179                                         csum_size, GFP_NOFS);
 180                         if (!btrfs_bio->csum_allocated) {
 181                                 btrfs_free_path(path);
 182                                 return BLK_STS_RESOURCE;
 183                         }
 184                         btrfs_bio->csum = btrfs_bio->csum_allocated;
 185                         btrfs_bio->end_io = btrfs_io_bio_endio_readpage;
 186                 } else {
 187                         btrfs_bio->csum = btrfs_bio->csum_inline;
 188                 }
 189                 csum = btrfs_bio->csum;
 190         } else {
 191                 csum = (u8 *)dst;
 192         }
 193
 194         if (bio->bi_iter.bi_size > PAGE_SIZE * 8)
 195                 path->reada = READA_FORWARD;
 196
 197         /*
 198          * the free space stuff is only read when it hasn't been
 199          * updated in the current transaction.  So, we can safely
 200          * read from the commit root and sidestep a nasty deadlock
 201          * between reading the free space cache and updating the csum tree.
 202          */
 203         if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
 204                 path->search_commit_root = 1;
 205                 path->skip_locking = 1;
 206         }
 207
 208         disk_bytenr = (u64)bio->bi_iter.bi_sector << 9;
 209         if (dio)
 210                 offset = logical_offset;
 211
 212         bio_for_each_segment(bvec, bio, iter) {
 213                 page_bytes_left = bvec.bv_len;
 214                 if (count)
 215                         goto next;
 216
 217                 if (!dio)
 218                         offset = page_offset(bvec.bv_page) + bvec.bv_offset;
 219                 count = btrfs_find_ordered_sum(inode, offset, disk_bytenr,
 220                                                (u32 *)csum, nblocks);
 221                 if (count)
 222                         goto found;
 223
 224                 if (!item || disk_bytenr < item_start_offset ||
 225                     disk_bytenr >= item_last_offset) {
 226                         struct btrfs_key found_key;
 227                         u32 item_size;
 228
 229                         if (item)
 230                                 btrfs_release_path(path);
 231                         item = btrfs_lookup_csum(NULL, fs_info->csum_root,
 232                                                  path, disk_bytenr, 0);
 233                         if (IS_ERR(item)) {
 234                                 count = 1;
 235                                 memset(csum, 0, csum_size);
 236                                 if (BTRFS_I(inode)->root->root_key.objectid ==
 237                                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 238                                         set_extent_bits(io_tree, offset,
 239                                                 offset + fs_info->sectorsize - 1,
 240                                                 EXTENT_NODATASUM);
 241                                 } else {
 242                                         btrfs_info_rl(fs_info,
 243                                                    "no csum found for inode %llu start %llu",
 244                                                btrfs_ino(BTRFS_I(inode)), offset);
 245                                 }
 246                                 item = NULL;
 247                                 btrfs_release_path(path);
 248                                 goto found;
 249                         }
 250                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 251                                               path->slots[0]);
 252
 253                         item_start_offset = found_key.offset;
 254                         item_size = btrfs_item_size_nr(path->nodes[0],
 255                                                        path->slots[0]);
 256                         item_last_offset = item_start_offset +
 257                                 (item_size / csum_size) *
 258                                 fs_info->sectorsize;
 259                         item = btrfs_item_ptr(path->nodes[0], path->slots[0],
 260                                               struct btrfs_csum_item);
 261                 }
 262                 /*
 263                  * this byte range must be able to fit inside
 264                  * a single leaf so it will also fit inside a u32
 265                  */
 266                 diff = disk_bytenr - item_start_offset;
 267                 diff = diff / fs_info->sectorsize;
 268                 diff = diff * csum_size;
 269                 count = min_t(int, nblocks, (item_last_offset - disk_bytenr) >>
 270                                             inode->i_sb->s_blocksize_bits);
 271                 read_extent_buffer(path->nodes[0], csum,
 272                                    ((unsigned long)item) + diff,
 273                                    csum_size * count);
 274 found:
 275                 csum += count * csum_size;
 276                 nblocks -= count;
 277 next:
 278                 while (count > 0) {
 279                         count--;
 280                         disk_bytenr += fs_info->sectorsize;
 281                         offset += fs_info->sectorsize;
 282                         page_bytes_left -= fs_info->sectorsize;
 283                         if (!page_bytes_left)
 284                                 break; /* move to next bio */
 285                 }
 286         }
 287
 288         WARN_ON_ONCE(count);
 289         btrfs_free_path(path);
 290         return 0;
 291 }
 292
 293 blk_status_t btrfs_lookup_bio_sums(struct inode *inode, struct bio *bio, u32 *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));
 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 blk_status_t btrfs_csum_one_bio(struct inode *inode, struct bio *bio,
 424                        u64 file_start, int contig)
 425 {
 426         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 427         struct btrfs_ordered_sum *sums;
 428         struct btrfs_ordered_extent *ordered = NULL;
 429         char *data;
 430         struct bvec_iter iter;
 431         struct bio_vec bvec;
 432         int index;
 433         int nr_sectors;
 434         unsigned long total_bytes = 0;
 435         unsigned long this_sum_bytes = 0;
 436         int i;
 437         u64 offset;
 438
 439         sums = kzalloc(btrfs_ordered_sum_size(fs_info, bio->bi_iter.bi_size),
 440                        GFP_NOFS);
 441         if (!sums)
 442                 return BLK_STS_RESOURCE;
 443
 444         sums->len = bio->bi_iter.bi_size;
 445         INIT_LIST_HEAD(&sums->list);
 446
 447         if (contig)
 448                 offset = file_start;
 449         else
 450                 offset = 0; /* shut up gcc */
 451
 452         sums->bytenr = (u64)bio->bi_iter.bi_sector << 9;
 453         index = 0;
 454
 455         bio_for_each_segment(bvec, bio, iter) {
 456                 if (!contig)
 457                         offset = page_offset(bvec.bv_page) + bvec.bv_offset;
 458
 459                 if (!ordered) {
 460                         ordered = btrfs_lookup_ordered_extent(inode, offset);
 461                         BUG_ON(!ordered); /* Logic error */
 462                 }
 463
 464                 data = kmap_atomic(bvec.bv_page);
 465
 466                 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info,
 467                                                  bvec.bv_len + fs_info->sectorsize
 468                                                  - 1);
 469
 470                 for (i = 0; i < nr_sectors; i++) {
 471                         if (offset >= ordered->file_offset + ordered->len ||
 472                                 offset < ordered->file_offset) {
 473                                 unsigned long bytes_left;
 474
 475                                 kunmap_atomic(data);
 476                                 sums->len = this_sum_bytes;
 477                                 this_sum_bytes = 0;
 478                                 btrfs_add_ordered_sum(inode, ordered, sums);
 479                                 btrfs_put_ordered_extent(ordered);
 480
 481                                 bytes_left = bio->bi_iter.bi_size - total_bytes;
 482
 483                                 sums = kzalloc(btrfs_ordered_sum_size(fs_info, bytes_left),
 484                                                GFP_NOFS);
 485                                 BUG_ON(!sums); /* -ENOMEM */
 486                                 sums->len = bytes_left;
 487                                 ordered = btrfs_lookup_ordered_extent(inode,
 488                                                                 offset);
 489                                 ASSERT(ordered); /* Logic error */
 490                                 sums->bytenr = ((u64)bio->bi_iter.bi_sector << 9)
 491                                         + total_bytes;
 492                                 index = 0;
 493
 494                                 data = kmap_atomic(bvec.bv_page);
 495                         }
 496
 497                         sums->sums[index] = ~(u32)0;
 498                         sums->sums[index]
 499                                 = btrfs_csum_data(data + bvec.bv_offset
 500                                                 + (i * fs_info->sectorsize),
 501                                                 sums->sums[index],
 502                                                 fs_info->sectorsize);
 503                         btrfs_csum_final(sums->sums[index],
 504                                         (char *)(sums->sums + index));
 505                         index++;
 506                         offset += fs_info->sectorsize;
 507                         this_sum_bytes += fs_info->sectorsize;
 508                         total_bytes += fs_info->sectorsize;
 509                 }
 510
 511                 kunmap_atomic(data);
 512         }
 513         this_sum_bytes = 0;
 514         btrfs_add_ordered_sum(inode, ordered, sums);
 515         btrfs_put_ordered_extent(ordered);
 516         return 0;
 517 }
 518
 519 /*
 520  * helper function for csum removal, this expects the
 521  * key to describe the csum pointed to by the path, and it expects
 522  * the csum to overlap the range [bytenr, len]
 523  *
 524  * The csum should not be entirely contained in the range and the
 525  * range should not be entirely contained in the csum.
 526  *
 527  * This calls btrfs_truncate_item with the correct args based on the
 528  * overlap, and fixes up the key as required.
 529  */
 530 static noinline void truncate_one_csum(struct btrfs_fs_info *fs_info,
 531                                        struct btrfs_path *path,
 532                                        struct btrfs_key *key,
 533                                        u64 bytenr, u64 len)
 534 {
 535         struct extent_buffer *leaf;
 536         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 537         u64 csum_end;
 538         u64 end_byte = bytenr + len;
 539         u32 blocksize_bits = fs_info->sb->s_blocksize_bits;
 540
 541         leaf = path->nodes[0];
 542         csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
 543         csum_end <<= fs_info->sb->s_blocksize_bits;
 544         csum_end += key->offset;
 545
 546         if (key->offset < bytenr && csum_end <= end_byte) {
 547                 /*
 548                  *         [ bytenr - len ]
 549                  *         [   ]
 550                  *   [csum     ]
 551                  *   A simple truncate off the end of the item
 552                  */
 553                 u32 new_size = (bytenr - key->offset) >> blocksize_bits;
 554                 new_size *= csum_size;
 555                 btrfs_truncate_item(fs_info, path, new_size, 1);
 556         } else if (key->offset >= bytenr && csum_end > end_byte &&
 557                    end_byte > key->offset) {
 558                 /*
 559                  *         [ bytenr - len ]
 560                  *                 [ ]
 561                  *                 [csum     ]
 562                  * we need to truncate from the beginning of the csum
 563                  */
 564                 u32 new_size = (csum_end - end_byte) >> blocksize_bits;
 565                 new_size *= csum_size;
 566
 567                 btrfs_truncate_item(fs_info, path, new_size, 0);
 568
 569                 key->offset = end_byte;
 570                 btrfs_set_item_key_safe(fs_info, path, key);
 571         } else {
 572                 BUG();
 573         }
 574 }
 575
 576 /*
 577  * deletes the csum items from the csum tree for a given
 578  * range of bytes.
 579  */
 580 int btrfs_del_csums(struct btrfs_trans_handle *trans,
 581                     struct btrfs_root *root, u64 bytenr, u64 len)
 582 {
 583         struct btrfs_fs_info *fs_info = trans->fs_info;
 584         struct btrfs_path *path;
 585         struct btrfs_key key;
 586         u64 end_byte = bytenr + len;
 587         u64 csum_end;
 588         struct extent_buffer *leaf;
 589         int ret = 0;
 590         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 591         int blocksize_bits = fs_info->sb->s_blocksize_bits;
 592
 593         ASSERT(root == fs_info->csum_root ||
 594                root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
 595
 596         path = btrfs_alloc_path();
 597         if (!path)
 598                 return -ENOMEM;
 599
 600         while (1) {
 601                 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
 602                 key.offset = end_byte - 1;
 603                 key.type = BTRFS_EXTENT_CSUM_KEY;
 604
 605                 path->leave_spinning = 1;
 606                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 607                 if (ret > 0) {
 608                         ret = 0;
 609                         if (path->slots[0] == 0)
 610                                 break;
 611                         path->slots[0]--;
 612                 } else if (ret < 0) {
 613                         break;
 614                 }
 615
 616                 leaf = path->nodes[0];
 617                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 618
 619                 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
 620                     key.type != BTRFS_EXTENT_CSUM_KEY) {
 621                         break;
 622                 }
 623
 624                 if (key.offset >= end_byte)
 625                         break;
 626
 627                 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
 628                 csum_end <<= blocksize_bits;
 629                 csum_end += key.offset;
 630
 631                 /* this csum ends before we start, we're done */
 632                 if (csum_end <= bytenr)
 633                         break;
 634
 635                 /* delete the entire item, it is inside our range */
 636                 if (key.offset >= bytenr && csum_end <= end_byte) {
 637                         int del_nr = 1;
 638
 639                         /*
 640                          * Check how many csum items preceding this one in this
 641                          * leaf correspond to our range and then delete them all
 642                          * at once.
 643                          */
 644                         if (key.offset > bytenr && path->slots[0] > 0) {
 645                                 int slot = path->slots[0] - 1;
 646
 647                                 while (slot >= 0) {
 648                                         struct btrfs_key pk;
 649
 650                                         btrfs_item_key_to_cpu(leaf, &pk, slot);
 651                                         if (pk.offset < bytenr ||
 652                                             pk.type != BTRFS_EXTENT_CSUM_KEY ||
 653                                             pk.objectid !=
 654                                             BTRFS_EXTENT_CSUM_OBJECTID)
 655                                                 break;
 656                                         path->slots[0] = slot;
 657                                         del_nr++;
 658                                         key.offset = pk.offset;
 659                                         slot--;
 660                                 }
 661                         }
 662                         ret = btrfs_del_items(trans, root, path,
 663                                               path->slots[0], del_nr);
 664                         if (ret)
 665                                 break;
 666                         if (key.offset == bytenr)
 667                                 break;
 668                 } else if (key.offset < bytenr && csum_end > end_byte) {
 669                         unsigned long offset;
 670                         unsigned long shift_len;
 671                         unsigned long item_offset;
 672                         /*
 673                          *        [ bytenr - len ]
 674                          *     [csum                ]
 675                          *
 676                          * Our bytes are in the middle of the csum,
 677                          * we need to split this item and insert a new one.
 678                          *
 679                          * But we can't drop the path because the
 680                          * csum could change, get removed, extended etc.
 681                          *
 682                          * The trick here is the max size of a csum item leaves
 683                          * enough room in the tree block for a single
 684                          * item header.  So, we split the item in place,
 685                          * adding a new header pointing to the existing
 686                          * bytes.  Then we loop around again and we have
 687                          * a nicely formed csum item that we can neatly
 688                          * truncate.
 689                          */
 690                         offset = (bytenr - key.offset) >> blocksize_bits;
 691                         offset *= csum_size;
 692
 693                         shift_len = (len >> blocksize_bits) * csum_size;
 694
 695                         item_offset = btrfs_item_ptr_offset(leaf,
 696                                                             path->slots[0]);
 697
 698                         memzero_extent_buffer(leaf, item_offset + offset,
 699                                              shift_len);
 700                         key.offset = bytenr;
 701
 702                         /*
 703                          * btrfs_split_item returns -EAGAIN when the
 704                          * item changed size or key
 705                          */
 706                         ret = btrfs_split_item(trans, root, path, &key, offset);
 707                         if (ret && ret != -EAGAIN) {
 708                                 btrfs_abort_transaction(trans, ret);
 709                                 break;
 710                         }
 711                         ret = 0;
 712
 713                         key.offset = end_byte - 1;
 714                 } else {
 715                         truncate_one_csum(fs_info, path, &key, bytenr, len);
 716                         if (key.offset < bytenr)
 717                                 break;
 718                 }
 719                 btrfs_release_path(path);
 720         }
 721         btrfs_free_path(path);
 722         return ret;
 723 }
 724
 725 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
 726                            struct btrfs_root *root,
 727                            struct btrfs_ordered_sum *sums)
 728 {
 729         struct btrfs_fs_info *fs_info = root->fs_info;
 730         struct btrfs_key file_key;
 731         struct btrfs_key found_key;
 732         struct btrfs_path *path;
 733         struct btrfs_csum_item *item;
 734         struct btrfs_csum_item *item_end;
 735         struct extent_buffer *leaf = NULL;
 736         u64 next_offset;
 737         u64 total_bytes = 0;
 738         u64 csum_offset;
 739         u64 bytenr;
 740         u32 nritems;
 741         u32 ins_size;
 742         int index = 0;
 743         int found_next;
 744         int ret;
 745         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 746
 747         path = btrfs_alloc_path();
 748         if (!path)
 749                 return -ENOMEM;
 750 again:
 751         next_offset = (u64)-1;
 752         found_next = 0;
 753         bytenr = sums->bytenr + total_bytes;
 754         file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
 755         file_key.offset = bytenr;
 756         file_key.type = BTRFS_EXTENT_CSUM_KEY;
 757
 758         item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
 759         if (!IS_ERR(item)) {
 760                 ret = 0;
 761                 leaf = path->nodes[0];
 762                 item_end = btrfs_item_ptr(leaf, path->slots[0],
 763                                           struct btrfs_csum_item);
 764                 item_end = (struct btrfs_csum_item *)((char *)item_end +
 765                            btrfs_item_size_nr(leaf, path->slots[0]));
 766                 goto found;
 767         }
 768         ret = PTR_ERR(item);
 769         if (ret != -EFBIG && ret != -ENOENT)
 770                 goto fail_unlock;
 771
 772         if (ret == -EFBIG) {
 773                 u32 item_size;
 774                 /* we found one, but it isn't big enough yet */
 775                 leaf = path->nodes[0];
 776                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
 777                 if ((item_size / csum_size) >=
 778                     MAX_CSUM_ITEMS(fs_info, csum_size)) {
 779                         /* already at max size, make a new one */
 780                         goto insert;
 781                 }
 782         } else {
 783                 int slot = path->slots[0] + 1;
 784                 /* we didn't find a csum item, insert one */
 785                 nritems = btrfs_header_nritems(path->nodes[0]);
 786                 if (!nritems || (path->slots[0] >= nritems - 1)) {
 787                         ret = btrfs_next_leaf(root, path);
 788                         if (ret < 0) {
 789                                 goto out;
 790                         } else if (ret > 0) {
 791                                 found_next = 1;
 792                                 goto insert;
 793                         }
 794                         slot = path->slots[0];
 795                 }
 796                 btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
 797                 if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
 798                     found_key.type != BTRFS_EXTENT_CSUM_KEY) {
 799                         found_next = 1;
 800                         goto insert;
 801                 }
 802                 next_offset = found_key.offset;
 803                 found_next = 1;
 804                 goto insert;
 805         }
 806
 807         /*
 808          * at this point, we know the tree has an item, but it isn't big
 809          * enough yet to put our csum in.  Grow it
 810          */
 811         btrfs_release_path(path);
 812         ret = btrfs_search_slot(trans, root, &file_key, path,
 813                                 csum_size, 1);
 814         if (ret < 0)
 815                 goto fail_unlock;
 816
 817         if (ret > 0) {
 818                 if (path->slots[0] == 0)
 819                         goto insert;
 820                 path->slots[0]--;
 821         }
 822
 823         leaf = path->nodes[0];
 824         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 825         csum_offset = (bytenr - found_key.offset) >>
 826                         fs_info->sb->s_blocksize_bits;
 827
 828         if (found_key.type != BTRFS_EXTENT_CSUM_KEY ||
 829             found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
 830             csum_offset >= MAX_CSUM_ITEMS(fs_info, csum_size)) {
 831                 goto insert;
 832         }
 833
 834         if (csum_offset == btrfs_item_size_nr(leaf, path->slots[0]) /
 835             csum_size) {
 836                 int extend_nr;
 837                 u64 tmp;
 838                 u32 diff;
 839                 u32 free_space;
 840
 841                 if (btrfs_leaf_free_space(fs_info, leaf) <
 842                                  sizeof(struct btrfs_item) + csum_size * 2)
 843                         goto insert;
 844
 845                 free_space = btrfs_leaf_free_space(fs_info, leaf) -
 846                                          sizeof(struct btrfs_item) - csum_size;
 847                 tmp = sums->len - total_bytes;
 848                 tmp >>= fs_info->sb->s_blocksize_bits;
 849                 WARN_ON(tmp < 1);
 850
 851                 extend_nr = max_t(int, 1, (int)tmp);
 852                 diff = (csum_offset + extend_nr) * csum_size;
 853                 diff = min(diff,
 854                            MAX_CSUM_ITEMS(fs_info, csum_size) * csum_size);
 855
 856                 diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
 857                 diff = min(free_space, diff);
 858                 diff /= csum_size;
 859                 diff *= csum_size;
 860
 861                 btrfs_extend_item(fs_info, path, diff);
 862                 ret = 0;
 863                 goto csum;
 864         }
 865
 866 insert:
 867         btrfs_release_path(path);
 868         csum_offset = 0;
 869         if (found_next) {
 870                 u64 tmp;
 871
 872                 tmp = sums->len - total_bytes;
 873                 tmp >>= fs_info->sb->s_blocksize_bits;
 874                 tmp = min(tmp, (next_offset - file_key.offset) >>
 875                                          fs_info->sb->s_blocksize_bits);
 876
 877                 tmp = max_t(u64, 1, tmp);
 878                 tmp = min_t(u64, tmp, MAX_CSUM_ITEMS(fs_info, csum_size));
 879                 ins_size = csum_size * tmp;
 880         } else {
 881                 ins_size = csum_size;
 882         }
 883         path->leave_spinning = 1;
 884         ret = btrfs_insert_empty_item(trans, root, path, &file_key,
 885                                       ins_size);
 886         path->leave_spinning = 0;
 887         if (ret < 0)
 888                 goto fail_unlock;
 889         if (WARN_ON(ret != 0))
 890                 goto fail_unlock;
 891         leaf = path->nodes[0];
 892 csum:
 893         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
 894         item_end = (struct btrfs_csum_item *)((unsigned char *)item +
 895                                       btrfs_item_size_nr(leaf, path->slots[0]));
 896         item = (struct btrfs_csum_item *)((unsigned char *)item +
 897                                           csum_offset * csum_size);
 898 found:
 899         ins_size = (u32)(sums->len - total_bytes) >>
 900                    fs_info->sb->s_blocksize_bits;
 901         ins_size *= csum_size;
 902         ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item,
 903                               ins_size);
 904         write_extent_buffer(leaf, sums->sums + index, (unsigned long)item,
 905                             ins_size);
 906
 907         ins_size /= csum_size;
 908         total_bytes += ins_size * fs_info->sectorsize;
 909         index += ins_size;
 910
 911         btrfs_mark_buffer_dirty(path->nodes[0]);
 912         if (total_bytes < sums->len) {
 913                 btrfs_release_path(path);
 914                 cond_resched();
 915                 goto again;
 916         }
 917 out:
 918         btrfs_free_path(path);
 919         return ret;
 920
 921 fail_unlock:
 922         goto out;
 923 }
 924
 925 void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode,
 926                                      const struct btrfs_path *path,
 927                                      struct btrfs_file_extent_item *fi,
 928                                      const bool new_inline,
 929                                      struct extent_map *em)
 930 {
 931         struct btrfs_fs_info *fs_info = inode->root->fs_info;
 932         struct btrfs_root *root = inode->root;
 933         struct extent_buffer *leaf = path->nodes[0];
 934         const int slot = path->slots[0];
 935         struct btrfs_key key;
 936         u64 extent_start, extent_end;
 937         u64 bytenr;
 938         u8 type = btrfs_file_extent_type(leaf, fi);
 939         int compress_type = btrfs_file_extent_compression(leaf, fi);
 940
 941         em->bdev = fs_info->fs_devices->latest_bdev;
 942         btrfs_item_key_to_cpu(leaf, &key, slot);
 943         extent_start = key.offset;
 944
 945         if (type == BTRFS_FILE_EXTENT_REG ||
 946             type == BTRFS_FILE_EXTENT_PREALLOC) {
 947                 extent_end = extent_start +
 948                         btrfs_file_extent_num_bytes(leaf, fi);
 949         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
 950                 size_t size;
 951                 size = btrfs_file_extent_ram_bytes(leaf, fi);
 952                 extent_end = ALIGN(extent_start + size,
 953                                    fs_info->sectorsize);
 954         }
 955
 956         em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
 957         if (type == BTRFS_FILE_EXTENT_REG ||
 958             type == BTRFS_FILE_EXTENT_PREALLOC) {
 959                 em->start = extent_start;
 960                 em->len = extent_end - extent_start;
 961                 em->orig_start = extent_start -
 962                         btrfs_file_extent_offset(leaf, fi);
 963                 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
 964                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
 965                 if (bytenr == 0) {
 966                         em->block_start = EXTENT_MAP_HOLE;
 967                         return;
 968                 }
 969                 if (compress_type != BTRFS_COMPRESS_NONE) {
 970                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 971                         em->compress_type = compress_type;
 972                         em->block_start = bytenr;
 973                         em->block_len = em->orig_block_len;
 974                 } else {
 975                         bytenr += btrfs_file_extent_offset(leaf, fi);
 976                         em->block_start = bytenr;
 977                         em->block_len = em->len;
 978                         if (type == BTRFS_FILE_EXTENT_PREALLOC)
 979                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
 980                 }
 981         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
 982                 em->block_start = EXTENT_MAP_INLINE;
 983                 em->start = extent_start;
 984                 em->len = extent_end - extent_start;
 985                 /*
 986                  * Initialize orig_start and block_len with the same values
 987                  * as in inode.c:btrfs_get_extent().
 988                  */
 989                 em->orig_start = EXTENT_MAP_HOLE;
 990                 em->block_len = (u64)-1;
 991                 if (!new_inline && compress_type != BTRFS_COMPRESS_NONE) {
 992                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 993                         em->compress_type = compress_type;
 994                 }
 995         } else {
 996                 btrfs_err(fs_info,
 997                           "unknown file extent item type %d, inode %llu, offset %llu, "
 998                           "root %llu", type, btrfs_ino(inode), extent_start,
 999                           root->root_key.objectid);
1000         }
1001 }