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--) {
 279                         disk_bytenr += fs_info->sectorsize;
 280                         offset += fs_info->sectorsize;
 281                         page_bytes_left -= fs_info->sectorsize;
 282                         if (!page_bytes_left)
 283                                 break; /* move to next bio */
 284                 }
 285         }
 286
 287         WARN_ON_ONCE(count);
 288         btrfs_free_path(path);
 289         return 0;
 290 }
 291
 292 blk_status_t btrfs_lookup_bio_sums(struct inode *inode, struct bio *bio, u32 *dst)
 293 {
 294         return __btrfs_lookup_bio_sums(inode, bio, 0, dst, 0);
 295 }
 296
 297 blk_status_t btrfs_lookup_bio_sums_dio(struct inode *inode, struct bio *bio, u64 offset)
 298 {
 299         return __btrfs_lookup_bio_sums(inode, bio, offset, NULL, 1);
 300 }
 301
 302 int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
 303                              struct list_head *list, int search_commit)
 304 {
 305         struct btrfs_fs_info *fs_info = root->fs_info;
 306         struct btrfs_key key;
 307         struct btrfs_path *path;
 308         struct extent_buffer *leaf;
 309         struct btrfs_ordered_sum *sums;
 310         struct btrfs_csum_item *item;
 311         LIST_HEAD(tmplist);
 312         unsigned long offset;
 313         int ret;
 314         size_t size;
 315         u64 csum_end;
 316         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 317
 318         ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
 319                IS_ALIGNED(end + 1, fs_info->sectorsize));
 320
 321         path = btrfs_alloc_path();
 322         if (!path)
 323                 return -ENOMEM;
 324
 325         if (search_commit) {
 326                 path->skip_locking = 1;
 327                 path->reada = READA_FORWARD;
 328                 path->search_commit_root = 1;
 329         }
 330
 331         key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
 332         key.offset = start;
 333         key.type = BTRFS_EXTENT_CSUM_KEY;
 334
 335         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 336         if (ret < 0)
 337                 goto fail;
 338         if (ret > 0 && path->slots[0] > 0) {
 339                 leaf = path->nodes[0];
 340                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
 341                 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
 342                     key.type == BTRFS_EXTENT_CSUM_KEY) {
 343                         offset = (start - key.offset) >>
 344                                  fs_info->sb->s_blocksize_bits;
 345                         if (offset * csum_size <
 346                             btrfs_item_size_nr(leaf, path->slots[0] - 1))
 347                                 path->slots[0]--;
 348                 }
 349         }
 350
 351         while (start <= end) {
 352                 leaf = path->nodes[0];
 353                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 354                         ret = btrfs_next_leaf(root, path);
 355                         if (ret < 0)
 356                                 goto fail;
 357                         if (ret > 0)
 358                                 break;
 359                         leaf = path->nodes[0];
 360                 }
 361
 362                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 363                 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
 364                     key.type != BTRFS_EXTENT_CSUM_KEY ||
 365                     key.offset > end)
 366                         break;
 367
 368                 if (key.offset > start)
 369                         start = key.offset;
 370
 371                 size = btrfs_item_size_nr(leaf, path->slots[0]);
 372                 csum_end = key.offset + (size / csum_size) * fs_info->sectorsize;
 373                 if (csum_end <= start) {
 374                         path->slots[0]++;
 375                         continue;
 376                 }
 377
 378                 csum_end = min(csum_end, end + 1);
 379                 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
 380                                       struct btrfs_csum_item);
 381                 while (start < csum_end) {
 382                         size = min_t(size_t, csum_end - start,
 383                                      MAX_ORDERED_SUM_BYTES(fs_info));
 384                         sums = kzalloc(btrfs_ordered_sum_size(fs_info, size),
 385                                        GFP_NOFS);
 386                         if (!sums) {
 387                                 ret = -ENOMEM;
 388                                 goto fail;
 389                         }
 390
 391                         sums->bytenr = start;
 392                         sums->len = (int)size;
 393
 394                         offset = (start - key.offset) >>
 395                                 fs_info->sb->s_blocksize_bits;
 396                         offset *= csum_size;
 397                         size >>= fs_info->sb->s_blocksize_bits;
 398
 399                         read_extent_buffer(path->nodes[0],
 400                                            sums->sums,
 401                                            ((unsigned long)item) + offset,
 402                                            csum_size * size);
 403
 404                         start += fs_info->sectorsize * size;
 405                         list_add_tail(&sums->list, &tmplist);
 406                 }
 407                 path->slots[0]++;
 408         }
 409         ret = 0;
 410 fail:
 411         while (ret < 0 && !list_empty(&tmplist)) {
 412                 sums = list_entry(tmplist.next, struct btrfs_ordered_sum, list);
 413                 list_del(&sums->list);
 414                 kfree(sums);
 415         }
 416         list_splice_tail(&tmplist, list);
 417
 418         btrfs_free_path(path);
 419         return ret;
 420 }
 421
 422 blk_status_t btrfs_csum_one_bio(struct inode *inode, struct bio *bio,
 423                        u64 file_start, int contig)
 424 {
 425         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 426         struct btrfs_ordered_sum *sums;
 427         struct btrfs_ordered_extent *ordered = NULL;
 428         char *data;
 429         struct bvec_iter iter;
 430         struct bio_vec bvec;
 431         int index;
 432         int nr_sectors;
 433         unsigned long total_bytes = 0;
 434         unsigned long this_sum_bytes = 0;
 435         int i;
 436         u64 offset;
 437
 438         sums = kzalloc(btrfs_ordered_sum_size(fs_info, bio->bi_iter.bi_size),
 439                        GFP_NOFS);
 440         if (!sums)
 441                 return BLK_STS_RESOURCE;
 442
 443         sums->len = bio->bi_iter.bi_size;
 444         INIT_LIST_HEAD(&sums->list);
 445
 446         if (contig)
 447                 offset = file_start;
 448         else
 449                 offset = 0; /* shut up gcc */
 450
 451         sums->bytenr = (u64)bio->bi_iter.bi_sector << 9;
 452         index = 0;
 453
 454         bio_for_each_segment(bvec, bio, iter) {
 455                 if (!contig)
 456                         offset = page_offset(bvec.bv_page) + bvec.bv_offset;
 457
 458                 if (!ordered) {
 459                         ordered = btrfs_lookup_ordered_extent(inode, offset);
 460                         BUG_ON(!ordered); /* Logic error */
 461                 }
 462
 463                 data = kmap_atomic(bvec.bv_page);
 464
 465                 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info,
 466                                                  bvec.bv_len + fs_info->sectorsize
 467                                                  - 1);
 468
 469                 for (i = 0; i < nr_sectors; i++) {
 470                         if (offset >= ordered->file_offset + ordered->len ||
 471                                 offset < ordered->file_offset) {
 472                                 unsigned long bytes_left;
 473
 474                                 kunmap_atomic(data);
 475                                 sums->len = this_sum_bytes;
 476                                 this_sum_bytes = 0;
 477                                 btrfs_add_ordered_sum(inode, ordered, sums);
 478                                 btrfs_put_ordered_extent(ordered);
 479
 480                                 bytes_left = bio->bi_iter.bi_size - total_bytes;
 481
 482                                 sums = kzalloc(btrfs_ordered_sum_size(fs_info, bytes_left),
 483                                                GFP_NOFS);
 484                                 BUG_ON(!sums); /* -ENOMEM */
 485                                 sums->len = bytes_left;
 486                                 ordered = btrfs_lookup_ordered_extent(inode,
 487                                                                 offset);
 488                                 ASSERT(ordered); /* Logic error */
 489                                 sums->bytenr = ((u64)bio->bi_iter.bi_sector << 9)
 490                                         + total_bytes;
 491                                 index = 0;
 492
 493                                 data = kmap_atomic(bvec.bv_page);
 494                         }
 495
 496                         sums->sums[index] = ~(u32)0;
 497                         sums->sums[index]
 498                                 = btrfs_csum_data(data + bvec.bv_offset
 499                                                 + (i * fs_info->sectorsize),
 500                                                 sums->sums[index],
 501                                                 fs_info->sectorsize);
 502                         btrfs_csum_final(sums->sums[index],
 503                                         (char *)(sums->sums + index));
 504                         index++;
 505                         offset += fs_info->sectorsize;
 506                         this_sum_bytes += fs_info->sectorsize;
 507                         total_bytes += fs_info->sectorsize;
 508                 }
 509
 510                 kunmap_atomic(data);
 511         }
 512         this_sum_bytes = 0;
 513         btrfs_add_ordered_sum(inode, ordered, sums);
 514         btrfs_put_ordered_extent(ordered);
 515         return 0;
 516 }
 517
 518 /*
 519  * helper function for csum removal, this expects the
 520  * key to describe the csum pointed to by the path, and it expects
 521  * the csum to overlap the range [bytenr, len]
 522  *
 523  * The csum should not be entirely contained in the range and the
 524  * range should not be entirely contained in the csum.
 525  *
 526  * This calls btrfs_truncate_item with the correct args based on the
 527  * overlap, and fixes up the key as required.
 528  */
 529 static noinline void truncate_one_csum(struct btrfs_fs_info *fs_info,
 530                                        struct btrfs_path *path,
 531                                        struct btrfs_key *key,
 532                                        u64 bytenr, u64 len)
 533 {
 534         struct extent_buffer *leaf;
 535         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 536         u64 csum_end;
 537         u64 end_byte = bytenr + len;
 538         u32 blocksize_bits = fs_info->sb->s_blocksize_bits;
 539
 540         leaf = path->nodes[0];
 541         csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
 542         csum_end <<= fs_info->sb->s_blocksize_bits;
 543         csum_end += key->offset;
 544
 545         if (key->offset < bytenr && csum_end <= end_byte) {
 546                 /*
 547                  *         [ bytenr - len ]
 548                  *         [   ]
 549                  *   [csum     ]
 550                  *   A simple truncate off the end of the item
 551                  */
 552                 u32 new_size = (bytenr - key->offset) >> blocksize_bits;
 553                 new_size *= csum_size;
 554                 btrfs_truncate_item(fs_info, path, new_size, 1);
 555         } else if (key->offset >= bytenr && csum_end > end_byte &&
 556                    end_byte > key->offset) {
 557                 /*
 558                  *         [ bytenr - len ]
 559                  *                 [ ]
 560                  *                 [csum     ]
 561                  * we need to truncate from the beginning of the csum
 562                  */
 563                 u32 new_size = (csum_end - end_byte) >> blocksize_bits;
 564                 new_size *= csum_size;
 565
 566                 btrfs_truncate_item(fs_info, path, new_size, 0);
 567
 568                 key->offset = end_byte;
 569                 btrfs_set_item_key_safe(fs_info, path, key);
 570         } else {
 571                 BUG();
 572         }
 573 }
 574
 575 /*
 576  * deletes the csum items from the csum tree for a given
 577  * range of bytes.
 578  */
 579 int btrfs_del_csums(struct btrfs_trans_handle *trans,
 580                     struct btrfs_fs_info *fs_info, u64 bytenr, u64 len)
 581 {
 582         struct btrfs_root *root = fs_info->csum_root;
 583         struct btrfs_path *path;
 584         struct btrfs_key key;
 585         u64 end_byte = bytenr + len;
 586         u64 csum_end;
 587         struct extent_buffer *leaf;
 588         int ret;
 589         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 590         int blocksize_bits = fs_info->sb->s_blocksize_bits;
 591
 592         path = btrfs_alloc_path();
 593         if (!path)
 594                 return -ENOMEM;
 595
 596         while (1) {
 597                 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
 598                 key.offset = end_byte - 1;
 599                 key.type = BTRFS_EXTENT_CSUM_KEY;
 600
 601                 path->leave_spinning = 1;
 602                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 603                 if (ret > 0) {
 604                         if (path->slots[0] == 0)
 605                                 break;
 606                         path->slots[0]--;
 607                 } else if (ret < 0) {
 608                         break;
 609                 }
 610
 611                 leaf = path->nodes[0];
 612                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 613
 614                 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
 615                     key.type != BTRFS_EXTENT_CSUM_KEY) {
 616                         break;
 617                 }
 618
 619                 if (key.offset >= end_byte)
 620                         break;
 621
 622                 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
 623                 csum_end <<= blocksize_bits;
 624                 csum_end += key.offset;
 625
 626                 /* this csum ends before we start, we're done */
 627                 if (csum_end <= bytenr)
 628                         break;
 629
 630                 /* delete the entire item, it is inside our range */
 631                 if (key.offset >= bytenr && csum_end <= end_byte) {
 632                         int del_nr = 1;
 633
 634                         /*
 635                          * Check how many csum items preceding this one in this
 636                          * leaf correspond to our range and then delete them all
 637                          * at once.
 638                          */
 639                         if (key.offset > bytenr && path->slots[0] > 0) {
 640                                 int slot = path->slots[0] - 1;
 641
 642                                 while (slot >= 0) {
 643                                         struct btrfs_key pk;
 644
 645                                         btrfs_item_key_to_cpu(leaf, &pk, slot);
 646                                         if (pk.offset < bytenr ||
 647                                             pk.type != BTRFS_EXTENT_CSUM_KEY ||
 648                                             pk.objectid !=
 649                                             BTRFS_EXTENT_CSUM_OBJECTID)
 650                                                 break;
 651                                         path->slots[0] = slot;
 652                                         del_nr++;
 653                                         key.offset = pk.offset;
 654                                         slot--;
 655                                 }
 656                         }
 657                         ret = btrfs_del_items(trans, root, path,
 658                                               path->slots[0], del_nr);
 659                         if (ret)
 660                                 goto out;
 661                         if (key.offset == bytenr)
 662                                 break;
 663                 } else if (key.offset < bytenr && csum_end > end_byte) {
 664                         unsigned long offset;
 665                         unsigned long shift_len;
 666                         unsigned long item_offset;
 667                         /*
 668                          *        [ bytenr - len ]
 669                          *     [csum                ]
 670                          *
 671                          * Our bytes are in the middle of the csum,
 672                          * we need to split this item and insert a new one.
 673                          *
 674                          * But we can't drop the path because the
 675                          * csum could change, get removed, extended etc.
 676                          *
 677                          * The trick here is the max size of a csum item leaves
 678                          * enough room in the tree block for a single
 679                          * item header.  So, we split the item in place,
 680                          * adding a new header pointing to the existing
 681                          * bytes.  Then we loop around again and we have
 682                          * a nicely formed csum item that we can neatly
 683                          * truncate.
 684                          */
 685                         offset = (bytenr - key.offset) >> blocksize_bits;
 686                         offset *= csum_size;
 687
 688                         shift_len = (len >> blocksize_bits) * csum_size;
 689
 690                         item_offset = btrfs_item_ptr_offset(leaf,
 691                                                             path->slots[0]);
 692
 693                         memzero_extent_buffer(leaf, item_offset + offset,
 694                                              shift_len);
 695                         key.offset = bytenr;
 696
 697                         /*
 698                          * btrfs_split_item returns -EAGAIN when the
 699                          * item changed size or key
 700                          */
 701                         ret = btrfs_split_item(trans, root, path, &key, offset);
 702                         if (ret && ret != -EAGAIN) {
 703                                 btrfs_abort_transaction(trans, ret);
 704                                 goto out;
 705                         }
 706
 707                         key.offset = end_byte - 1;
 708                 } else {
 709                         truncate_one_csum(fs_info, path, &key, bytenr, len);
 710                         if (key.offset < bytenr)
 711                                 break;
 712                 }
 713                 btrfs_release_path(path);
 714         }
 715         ret = 0;
 716 out:
 717         btrfs_free_path(path);
 718         return ret;
 719 }
 720
 721 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
 722                            struct btrfs_root *root,
 723                            struct btrfs_ordered_sum *sums)
 724 {
 725         struct btrfs_fs_info *fs_info = root->fs_info;
 726         struct btrfs_key file_key;
 727         struct btrfs_key found_key;
 728         struct btrfs_path *path;
 729         struct btrfs_csum_item *item;
 730         struct btrfs_csum_item *item_end;
 731         struct extent_buffer *leaf = NULL;
 732         u64 next_offset;
 733         u64 total_bytes = 0;
 734         u64 csum_offset;
 735         u64 bytenr;
 736         u32 nritems;
 737         u32 ins_size;
 738         int index = 0;
 739         int found_next;
 740         int ret;
 741         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 742
 743         path = btrfs_alloc_path();
 744         if (!path)
 745                 return -ENOMEM;
 746 again:
 747         next_offset = (u64)-1;
 748         found_next = 0;
 749         bytenr = sums->bytenr + total_bytes;
 750         file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
 751         file_key.offset = bytenr;
 752         file_key.type = BTRFS_EXTENT_CSUM_KEY;
 753
 754         item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
 755         if (!IS_ERR(item)) {
 756                 ret = 0;
 757                 leaf = path->nodes[0];
 758                 item_end = btrfs_item_ptr(leaf, path->slots[0],
 759                                           struct btrfs_csum_item);
 760                 item_end = (struct btrfs_csum_item *)((char *)item_end +
 761                            btrfs_item_size_nr(leaf, path->slots[0]));
 762                 goto found;
 763         }
 764         ret = PTR_ERR(item);
 765         if (ret != -EFBIG && ret != -ENOENT)
 766                 goto fail_unlock;
 767
 768         if (ret == -EFBIG) {
 769                 u32 item_size;
 770                 /* we found one, but it isn't big enough yet */
 771                 leaf = path->nodes[0];
 772                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
 773                 if ((item_size / csum_size) >=
 774                     MAX_CSUM_ITEMS(fs_info, csum_size)) {
 775                         /* already at max size, make a new one */
 776                         goto insert;
 777                 }
 778         } else {
 779                 int slot = path->slots[0] + 1;
 780                 /* we didn't find a csum item, insert one */
 781                 nritems = btrfs_header_nritems(path->nodes[0]);
 782                 if (!nritems || (path->slots[0] >= nritems - 1)) {
 783                         ret = btrfs_next_leaf(root, path);
 784                         if (ret == 1)
 785                                 found_next = 1;
 786                         if (ret != 0)
 787                                 goto insert;
 788                         slot = path->slots[0];
 789                 }
 790                 btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
 791                 if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
 792                     found_key.type != BTRFS_EXTENT_CSUM_KEY) {
 793                         found_next = 1;
 794                         goto insert;
 795                 }
 796                 next_offset = found_key.offset;
 797                 found_next = 1;
 798                 goto insert;
 799         }
 800
 801         /*
 802          * at this point, we know the tree has an item, but it isn't big
 803          * enough yet to put our csum in.  Grow it
 804          */
 805         btrfs_release_path(path);
 806         ret = btrfs_search_slot(trans, root, &file_key, path,
 807                                 csum_size, 1);
 808         if (ret < 0)
 809                 goto fail_unlock;
 810
 811         if (ret > 0) {
 812                 if (path->slots[0] == 0)
 813                         goto insert;
 814                 path->slots[0]--;
 815         }
 816
 817         leaf = path->nodes[0];
 818         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 819         csum_offset = (bytenr - found_key.offset) >>
 820                         fs_info->sb->s_blocksize_bits;
 821
 822         if (found_key.type != BTRFS_EXTENT_CSUM_KEY ||
 823             found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
 824             csum_offset >= MAX_CSUM_ITEMS(fs_info, csum_size)) {
 825                 goto insert;
 826         }
 827
 828         if (csum_offset == btrfs_item_size_nr(leaf, path->slots[0]) /
 829             csum_size) {
 830                 int extend_nr;
 831                 u64 tmp;
 832                 u32 diff;
 833                 u32 free_space;
 834
 835                 if (btrfs_leaf_free_space(fs_info, leaf) <
 836                                  sizeof(struct btrfs_item) + csum_size * 2)
 837                         goto insert;
 838
 839                 free_space = btrfs_leaf_free_space(fs_info, leaf) -
 840                                          sizeof(struct btrfs_item) - csum_size;
 841                 tmp = sums->len - total_bytes;
 842                 tmp >>= fs_info->sb->s_blocksize_bits;
 843                 WARN_ON(tmp < 1);
 844
 845                 extend_nr = max_t(int, 1, (int)tmp);
 846                 diff = (csum_offset + extend_nr) * csum_size;
 847                 diff = min(diff,
 848                            MAX_CSUM_ITEMS(fs_info, csum_size) * csum_size);
 849
 850                 diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
 851                 diff = min(free_space, diff);
 852                 diff /= csum_size;
 853                 diff *= csum_size;
 854
 855                 btrfs_extend_item(fs_info, path, diff);
 856                 ret = 0;
 857                 goto csum;
 858         }
 859
 860 insert:
 861         btrfs_release_path(path);
 862         csum_offset = 0;
 863         if (found_next) {
 864                 u64 tmp;
 865
 866                 tmp = sums->len - total_bytes;
 867                 tmp >>= fs_info->sb->s_blocksize_bits;
 868                 tmp = min(tmp, (next_offset - file_key.offset) >>
 869                                          fs_info->sb->s_blocksize_bits);
 870
 871                 tmp = max_t(u64, 1, tmp);
 872                 tmp = min_t(u64, tmp, MAX_CSUM_ITEMS(fs_info, csum_size));
 873                 ins_size = csum_size * tmp;
 874         } else {
 875                 ins_size = csum_size;
 876         }
 877         path->leave_spinning = 1;
 878         ret = btrfs_insert_empty_item(trans, root, path, &file_key,
 879                                       ins_size);
 880         path->leave_spinning = 0;
 881         if (ret < 0)
 882                 goto fail_unlock;
 883         if (WARN_ON(ret != 0))
 884                 goto fail_unlock;
 885         leaf = path->nodes[0];
 886 csum:
 887         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
 888         item_end = (struct btrfs_csum_item *)((unsigned char *)item +
 889                                       btrfs_item_size_nr(leaf, path->slots[0]));
 890         item = (struct btrfs_csum_item *)((unsigned char *)item +
 891                                           csum_offset * csum_size);
 892 found:
 893         ins_size = (u32)(sums->len - total_bytes) >>
 894                    fs_info->sb->s_blocksize_bits;
 895         ins_size *= csum_size;
 896         ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item,
 897                               ins_size);
 898         write_extent_buffer(leaf, sums->sums + index, (unsigned long)item,
 899                             ins_size);
 900
 901         ins_size /= csum_size;
 902         total_bytes += ins_size * fs_info->sectorsize;
 903         index += ins_size;
 904
 905         btrfs_mark_buffer_dirty(path->nodes[0]);
 906         if (total_bytes < sums->len) {
 907                 btrfs_release_path(path);
 908                 cond_resched();
 909                 goto again;
 910         }
 911 out:
 912         btrfs_free_path(path);
 913         return ret;
 914
 915 fail_unlock:
 916         goto out;
 917 }
 918
 919 void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode,
 920                                      const struct btrfs_path *path,
 921                                      struct btrfs_file_extent_item *fi,
 922                                      const bool new_inline,
 923                                      struct extent_map *em)
 924 {
 925         struct btrfs_fs_info *fs_info = inode->root->fs_info;
 926         struct btrfs_root *root = inode->root;
 927         struct extent_buffer *leaf = path->nodes[0];
 928         const int slot = path->slots[0];
 929         struct btrfs_key key;
 930         u64 extent_start, extent_end;
 931         u64 bytenr;
 932         u8 type = btrfs_file_extent_type(leaf, fi);
 933         int compress_type = btrfs_file_extent_compression(leaf, fi);
 934
 935         em->bdev = fs_info->fs_devices->latest_bdev;
 936         btrfs_item_key_to_cpu(leaf, &key, slot);
 937         extent_start = key.offset;
 938
 939         if (type == BTRFS_FILE_EXTENT_REG ||
 940             type == BTRFS_FILE_EXTENT_PREALLOC) {
 941                 extent_end = extent_start +
 942                         btrfs_file_extent_num_bytes(leaf, fi);
 943         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
 944                 size_t size;
 945                 size = btrfs_file_extent_ram_bytes(leaf, fi);
 946                 extent_end = ALIGN(extent_start + size,
 947                                    fs_info->sectorsize);
 948         }
 949
 950         em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
 951         if (type == BTRFS_FILE_EXTENT_REG ||
 952             type == BTRFS_FILE_EXTENT_PREALLOC) {
 953                 em->start = extent_start;
 954                 em->len = extent_end - extent_start;
 955                 em->orig_start = extent_start -
 956                         btrfs_file_extent_offset(leaf, fi);
 957                 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
 958                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
 959                 if (bytenr == 0) {
 960                         em->block_start = EXTENT_MAP_HOLE;
 961                         return;
 962                 }
 963                 if (compress_type != BTRFS_COMPRESS_NONE) {
 964                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 965                         em->compress_type = compress_type;
 966                         em->block_start = bytenr;
 967                         em->block_len = em->orig_block_len;
 968                 } else {
 969                         bytenr += btrfs_file_extent_offset(leaf, fi);
 970                         em->block_start = bytenr;
 971                         em->block_len = em->len;
 972                         if (type == BTRFS_FILE_EXTENT_PREALLOC)
 973                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
 974                 }
 975         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
 976                 em->block_start = EXTENT_MAP_INLINE;
 977                 em->start = extent_start;
 978                 em->len = extent_end - extent_start;
 979                 /*
 980                  * Initialize orig_start and block_len with the same values
 981                  * as in inode.c:btrfs_get_extent().
 982                  */
 983                 em->orig_start = EXTENT_MAP_HOLE;
 984                 em->block_len = (u64)-1;
 985                 if (!new_inline && compress_type != BTRFS_COMPRESS_NONE) {
 986                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 987                         em->compress_type = compress_type;
 988                 }
 989         } else {
 990                 btrfs_err(fs_info,
 991                           "unknown file extent item type %d, inode %llu, offset %llu, "
 992                           "root %llu", type, btrfs_ino(inode), extent_start,
 993                           root->root_key.objectid);
 994         }
 995 }