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