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 <linux/sched/mm.h>
  11 #include <crypto/hash.h>
  12 #include "ctree.h"
  13 #include "disk-io.h"
  14 #include "transaction.h"
  15 #include "volumes.h"
  16 #include "print-tree.h"
  17 #include "compression.h"
  18
  19 #define __MAX_CSUM_ITEMS(r, size) ((unsigned long)(((BTRFS_LEAF_DATA_SIZE(r) - \
  20                                    sizeof(struct btrfs_item) * 2) / \
  21                                   size) - 1))
  22
  23 #define MAX_CSUM_ITEMS(r, size) (min_t(u32, __MAX_CSUM_ITEMS(r, size), \
  24                                        PAGE_SIZE))
  25
  26 /**
  27  * @inode - the inode we want to update the disk_i_size for
  28  * @new_i_size - the i_size we want to set to, 0 if we use i_size
  29  *
  30  * With NO_HOLES set this simply sets the disk_is_size to whatever i_size_read()
  31  * returns as it is perfectly fine with a file that has holes without hole file
  32  * extent items.
  33  *
  34  * However without NO_HOLES we need to only return the area that is contiguous
  35  * from the 0 offset of the file.  Otherwise we could end up adjust i_size up
  36  * to an extent that has a gap in between.
  37  *
  38  * Finally new_i_size should only be set in the case of truncate where we're not
  39  * ready to use i_size_read() as the limiter yet.
  40  */
  41 void btrfs_inode_safe_disk_i_size_write(struct inode *inode, u64 new_i_size)
  42 {
  43         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
  44         u64 start, end, i_size;
  45         int ret;
  46
  47         i_size = new_i_size ?: i_size_read(inode);
  48         if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
  49                 BTRFS_I(inode)->disk_i_size = i_size;
  50                 return;
  51         }
  52
  53         spin_lock(&BTRFS_I(inode)->lock);
  54         ret = find_contiguous_extent_bit(&BTRFS_I(inode)->file_extent_tree, 0,
  55                                          &start, &end, EXTENT_DIRTY);
  56         if (!ret && start == 0)
  57                 i_size = min(i_size, end + 1);
  58         else
  59                 i_size = 0;
  60         BTRFS_I(inode)->disk_i_size = i_size;
  61         spin_unlock(&BTRFS_I(inode)->lock);
  62 }
  63
  64 /**
  65  * @inode - the inode we're modifying
  66  * @start - the start file offset of the file extent we've inserted
  67  * @len - the logical length of the file extent item
  68  *
  69  * Call when we are inserting a new file extent where there was none before.
  70  * Does not need to call this in the case where we're replacing an existing file
  71  * extent, however if not sure it's fine to call this multiple times.
  72  *
  73  * The start and len must match the file extent item, so thus must be sectorsize
  74  * aligned.
  75  */
  76 int btrfs_inode_set_file_extent_range(struct btrfs_inode *inode, u64 start,
  77                                       u64 len)
  78 {
  79         if (len == 0)
  80                 return 0;
  81
  82         ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize));
  83
  84         if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES))
  85                 return 0;
  86         return set_extent_bits(&inode->file_extent_tree, start, start + len - 1,
  87                                EXTENT_DIRTY);
  88 }
  89
  90 /**
  91  * @inode - the inode we're modifying
  92  * @start - the start file offset of the file extent we've inserted
  93  * @len - the logical length of the file extent item
  94  *
  95  * Called when we drop a file extent, for example when we truncate.  Doesn't
  96  * need to be called for cases where we're replacing a file extent, like when
  97  * we've COWed a file extent.
  98  *
  99  * The start and len must match the file extent item, so thus must be sectorsize
 100  * aligned.
 101  */
 102 int btrfs_inode_clear_file_extent_range(struct btrfs_inode *inode, u64 start,
 103                                         u64 len)
 104 {
 105         if (len == 0)
 106                 return 0;
 107
 108         ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize) ||
 109                len == (u64)-1);
 110
 111         if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES))
 112                 return 0;
 113         return clear_extent_bit(&inode->file_extent_tree, start,
 114                                 start + len - 1, EXTENT_DIRTY, 0, 0, NULL);
 115 }
 116
 117 static inline u32 max_ordered_sum_bytes(struct btrfs_fs_info *fs_info,
 118                                         u16 csum_size)
 119 {
 120         u32 ncsums = (PAGE_SIZE - sizeof(struct btrfs_ordered_sum)) / csum_size;
 121
 122         return ncsums * fs_info->sectorsize;
 123 }
 124
 125 int btrfs_insert_file_extent(struct btrfs_trans_handle *trans,
 126                              struct btrfs_root *root,
 127                              u64 objectid, u64 pos,
 128                              u64 disk_offset, u64 disk_num_bytes,
 129                              u64 num_bytes, u64 offset, u64 ram_bytes,
 130                              u8 compression, u8 encryption, u16 other_encoding)
 131 {
 132         int ret = 0;
 133         struct btrfs_file_extent_item *item;
 134         struct btrfs_key file_key;
 135         struct btrfs_path *path;
 136         struct extent_buffer *leaf;
 137
 138         path = btrfs_alloc_path();
 139         if (!path)
 140                 return -ENOMEM;
 141         file_key.objectid = objectid;
 142         file_key.offset = pos;
 143         file_key.type = BTRFS_EXTENT_DATA_KEY;
 144
 145         ret = btrfs_insert_empty_item(trans, root, path, &file_key,
 146                                       sizeof(*item));
 147         if (ret < 0)
 148                 goto out;
 149         BUG_ON(ret); /* Can't happen */
 150         leaf = path->nodes[0];
 151         item = btrfs_item_ptr(leaf, path->slots[0],
 152                               struct btrfs_file_extent_item);
 153         btrfs_set_file_extent_disk_bytenr(leaf, item, disk_offset);
 154         btrfs_set_file_extent_disk_num_bytes(leaf, item, disk_num_bytes);
 155         btrfs_set_file_extent_offset(leaf, item, offset);
 156         btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
 157         btrfs_set_file_extent_ram_bytes(leaf, item, ram_bytes);
 158         btrfs_set_file_extent_generation(leaf, item, trans->transid);
 159         btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
 160         btrfs_set_file_extent_compression(leaf, item, compression);
 161         btrfs_set_file_extent_encryption(leaf, item, encryption);
 162         btrfs_set_file_extent_other_encoding(leaf, item, other_encoding);
 163
 164         btrfs_mark_buffer_dirty(leaf);
 165 out:
 166         btrfs_free_path(path);
 167         return ret;
 168 }
 169
 170 static struct btrfs_csum_item *
 171 btrfs_lookup_csum(struct btrfs_trans_handle *trans,
 172                   struct btrfs_root *root,
 173                   struct btrfs_path *path,
 174                   u64 bytenr, int cow)
 175 {
 176         struct btrfs_fs_info *fs_info = root->fs_info;
 177         int ret;
 178         struct btrfs_key file_key;
 179         struct btrfs_key found_key;
 180         struct btrfs_csum_item *item;
 181         struct extent_buffer *leaf;
 182         u64 csum_offset = 0;
 183         const u32 csum_size = fs_info->csum_size;
 184         int csums_in_item;
 185
 186         file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
 187         file_key.offset = bytenr;
 188         file_key.type = BTRFS_EXTENT_CSUM_KEY;
 189         ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
 190         if (ret < 0)
 191                 goto fail;
 192         leaf = path->nodes[0];
 193         if (ret > 0) {
 194                 ret = 1;
 195                 if (path->slots[0] == 0)
 196                         goto fail;
 197                 path->slots[0]--;
 198                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 199                 if (found_key.type != BTRFS_EXTENT_CSUM_KEY)
 200                         goto fail;
 201
 202                 csum_offset = (bytenr - found_key.offset) >>
 203                                 fs_info->sectorsize_bits;
 204                 csums_in_item = btrfs_item_size_nr(leaf, path->slots[0]);
 205                 csums_in_item /= csum_size;
 206
 207                 if (csum_offset == csums_in_item) {
 208                         ret = -EFBIG;
 209                         goto fail;
 210                 } else if (csum_offset > csums_in_item) {
 211                         goto fail;
 212                 }
 213         }
 214         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
 215         item = (struct btrfs_csum_item *)((unsigned char *)item +
 216                                           csum_offset * csum_size);
 217         return item;
 218 fail:
 219         if (ret > 0)
 220                 ret = -ENOENT;
 221         return ERR_PTR(ret);
 222 }
 223
 224 int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
 225                              struct btrfs_root *root,
 226                              struct btrfs_path *path, u64 objectid,
 227                              u64 offset, int mod)
 228 {
 229         int ret;
 230         struct btrfs_key file_key;
 231         int ins_len = mod < 0 ? -1 : 0;
 232         int cow = mod != 0;
 233
 234         file_key.objectid = objectid;
 235         file_key.offset = offset;
 236         file_key.type = BTRFS_EXTENT_DATA_KEY;
 237         ret = btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
 238         return ret;
 239 }
 240
 241 /**
 242  * btrfs_lookup_bio_sums - Look up checksums for a bio.
 243  * @inode: inode that the bio is for.
 244  * @bio: bio to look up.
 245  * @offset: Unless (u64)-1, look up checksums for this offset in the file.
 246  *          If (u64)-1, use the page offsets from the bio instead.
 247  * @dst: Buffer of size nblocks * btrfs_super_csum_size() used to return
 248  *       checksum (nblocks = bio->bi_iter.bi_size / fs_info->sectorsize). If
 249  *       NULL, the checksum buffer is allocated and returned in
 250  *       btrfs_io_bio(bio)->csum instead.
 251  *
 252  * Return: BLK_STS_RESOURCE if allocating memory fails, BLK_STS_OK otherwise.
 253  */
 254 blk_status_t btrfs_lookup_bio_sums(struct inode *inode, struct bio *bio,
 255                                    u64 offset, u8 *dst)
 256 {
 257         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 258         struct bio_vec bvec;
 259         struct bvec_iter iter;
 260         struct btrfs_csum_item *item = NULL;
 261         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 262         struct btrfs_path *path;
 263         const bool page_offsets = (offset == (u64)-1);
 264         u8 *csum;
 265         u64 item_start_offset = 0;
 266         u64 item_last_offset = 0;
 267         u64 disk_bytenr;
 268         u64 page_bytes_left;
 269         u32 diff;
 270         int nblocks;
 271         int count = 0;
 272         const u32 csum_size = fs_info->csum_size;
 273
 274         if (!fs_info->csum_root || (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
 275                 return BLK_STS_OK;
 276
 277         path = btrfs_alloc_path();
 278         if (!path)
 279                 return BLK_STS_RESOURCE;
 280
 281         nblocks = bio->bi_iter.bi_size >> fs_info->sectorsize_bits;
 282         if (!dst) {
 283                 struct btrfs_io_bio *btrfs_bio = btrfs_io_bio(bio);
 284
 285                 if (nblocks * csum_size > BTRFS_BIO_INLINE_CSUM_SIZE) {
 286                         btrfs_bio->csum = kmalloc_array(nblocks, csum_size,
 287                                                         GFP_NOFS);
 288                         if (!btrfs_bio->csum) {
 289                                 btrfs_free_path(path);
 290                                 return BLK_STS_RESOURCE;
 291                         }
 292                 } else {
 293                         btrfs_bio->csum = btrfs_bio->csum_inline;
 294                 }
 295                 csum = btrfs_bio->csum;
 296         } else {
 297                 csum = dst;
 298         }
 299
 300         if (bio->bi_iter.bi_size > PAGE_SIZE * 8)
 301                 path->reada = READA_FORWARD;
 302
 303         /*
 304          * the free space stuff is only read when it hasn't been
 305          * updated in the current transaction.  So, we can safely
 306          * read from the commit root and sidestep a nasty deadlock
 307          * between reading the free space cache and updating the csum tree.
 308          */
 309         if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
 310                 path->search_commit_root = 1;
 311                 path->skip_locking = 1;
 312         }
 313
 314         disk_bytenr = (u64)bio->bi_iter.bi_sector << 9;
 315
 316         bio_for_each_segment(bvec, bio, iter) {
 317                 page_bytes_left = bvec.bv_len;
 318                 if (count)
 319                         goto next;
 320
 321                 if (page_offsets)
 322                         offset = page_offset(bvec.bv_page) + bvec.bv_offset;
 323                 count = btrfs_find_ordered_sum(BTRFS_I(inode), offset,
 324                                                disk_bytenr, csum, nblocks);
 325                 if (count)
 326                         goto found;
 327
 328                 if (!item || disk_bytenr < item_start_offset ||
 329                     disk_bytenr >= item_last_offset) {
 330                         struct btrfs_key found_key;
 331                         u32 item_size;
 332
 333                         if (item)
 334                                 btrfs_release_path(path);
 335                         item = btrfs_lookup_csum(NULL, fs_info->csum_root,
 336                                                  path, disk_bytenr, 0);
 337                         if (IS_ERR(item)) {
 338                                 count = 1;
 339                                 memset(csum, 0, csum_size);
 340                                 if (BTRFS_I(inode)->root->root_key.objectid ==
 341                                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 342                                         set_extent_bits(io_tree, offset,
 343                                                 offset + fs_info->sectorsize - 1,
 344                                                 EXTENT_NODATASUM);
 345                                 } else {
 346                                         btrfs_info_rl(fs_info,
 347                                                    "no csum found for inode %llu start %llu",
 348                                                btrfs_ino(BTRFS_I(inode)), offset);
 349                                 }
 350                                 item = NULL;
 351                                 btrfs_release_path(path);
 352                                 goto found;
 353                         }
 354                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 355                                               path->slots[0]);
 356
 357                         item_start_offset = found_key.offset;
 358                         item_size = btrfs_item_size_nr(path->nodes[0],
 359                                                        path->slots[0]);
 360                         item_last_offset = item_start_offset +
 361                                 (item_size / csum_size) *
 362                                 fs_info->sectorsize;
 363                         item = btrfs_item_ptr(path->nodes[0], path->slots[0],
 364                                               struct btrfs_csum_item);
 365                 }
 366                 /*
 367                  * this byte range must be able to fit inside
 368                  * a single leaf so it will also fit inside a u32
 369                  */
 370                 diff = disk_bytenr - item_start_offset;
 371                 diff = diff >> fs_info->sectorsize_bits;
 372                 diff = diff * csum_size;
 373                 count = min_t(int, nblocks, (item_last_offset - disk_bytenr) >>
 374                                             fs_info->sectorsize_bits);
 375                 read_extent_buffer(path->nodes[0], csum,
 376                                    ((unsigned long)item) + diff,
 377                                    csum_size * count);
 378 found:
 379                 csum += count * csum_size;
 380                 nblocks -= count;
 381 next:
 382                 while (count > 0) {
 383                         count--;
 384                         disk_bytenr += fs_info->sectorsize;
 385                         offset += fs_info->sectorsize;
 386                         page_bytes_left -= fs_info->sectorsize;
 387                         if (!page_bytes_left)
 388                                 break; /* move to next bio */
 389                 }
 390         }
 391
 392         WARN_ON_ONCE(count);
 393         btrfs_free_path(path);
 394         return BLK_STS_OK;
 395 }
 396
 397 int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
 398                              struct list_head *list, int search_commit)
 399 {
 400         struct btrfs_fs_info *fs_info = root->fs_info;
 401         struct btrfs_key key;
 402         struct btrfs_path *path;
 403         struct extent_buffer *leaf;
 404         struct btrfs_ordered_sum *sums;
 405         struct btrfs_csum_item *item;
 406         LIST_HEAD(tmplist);
 407         unsigned long offset;
 408         int ret;
 409         size_t size;
 410         u64 csum_end;
 411         const u32 csum_size = fs_info->csum_size;
 412
 413         ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
 414                IS_ALIGNED(end + 1, fs_info->sectorsize));
 415
 416         path = btrfs_alloc_path();
 417         if (!path)
 418                 return -ENOMEM;
 419
 420         if (search_commit) {
 421                 path->skip_locking = 1;
 422                 path->reada = READA_FORWARD;
 423                 path->search_commit_root = 1;
 424         }
 425
 426         key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
 427         key.offset = start;
 428         key.type = BTRFS_EXTENT_CSUM_KEY;
 429
 430         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 431         if (ret < 0)
 432                 goto fail;
 433         if (ret > 0 && path->slots[0] > 0) {
 434                 leaf = path->nodes[0];
 435                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
 436                 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
 437                     key.type == BTRFS_EXTENT_CSUM_KEY) {
 438                         offset = (start - key.offset) >> fs_info->sectorsize_bits;
 439                         if (offset * csum_size <
 440                             btrfs_item_size_nr(leaf, path->slots[0] - 1))
 441                                 path->slots[0]--;
 442                 }
 443         }
 444
 445         while (start <= end) {
 446                 leaf = path->nodes[0];
 447                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 448                         ret = btrfs_next_leaf(root, path);
 449                         if (ret < 0)
 450                                 goto fail;
 451                         if (ret > 0)
 452                                 break;
 453                         leaf = path->nodes[0];
 454                 }
 455
 456                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 457                 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
 458                     key.type != BTRFS_EXTENT_CSUM_KEY ||
 459                     key.offset > end)
 460                         break;
 461
 462                 if (key.offset > start)
 463                         start = key.offset;
 464
 465                 size = btrfs_item_size_nr(leaf, path->slots[0]);
 466                 csum_end = key.offset + (size / csum_size) * fs_info->sectorsize;
 467                 if (csum_end <= start) {
 468                         path->slots[0]++;
 469                         continue;
 470                 }
 471
 472                 csum_end = min(csum_end, end + 1);
 473                 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
 474                                       struct btrfs_csum_item);
 475                 while (start < csum_end) {
 476                         size = min_t(size_t, csum_end - start,
 477                                      max_ordered_sum_bytes(fs_info, csum_size));
 478                         sums = kzalloc(btrfs_ordered_sum_size(fs_info, size),
 479                                        GFP_NOFS);
 480                         if (!sums) {
 481                                 ret = -ENOMEM;
 482                                 goto fail;
 483                         }
 484
 485                         sums->bytenr = start;
 486                         sums->len = (int)size;
 487
 488                         offset = (start - key.offset) >> fs_info->sectorsize_bits;
 489                         offset *= csum_size;
 490                         size >>= fs_info->sectorsize_bits;
 491
 492                         read_extent_buffer(path->nodes[0],
 493                                            sums->sums,
 494                                            ((unsigned long)item) + offset,
 495                                            csum_size * size);
 496
 497                         start += fs_info->sectorsize * size;
 498                         list_add_tail(&sums->list, &tmplist);
 499                 }
 500                 path->slots[0]++;
 501         }
 502         ret = 0;
 503 fail:
 504         while (ret < 0 && !list_empty(&tmplist)) {
 505                 sums = list_entry(tmplist.next, struct btrfs_ordered_sum, list);
 506                 list_del(&sums->list);
 507                 kfree(sums);
 508         }
 509         list_splice_tail(&tmplist, list);
 510
 511         btrfs_free_path(path);
 512         return ret;
 513 }
 514
 515 /*
 516  * btrfs_csum_one_bio - Calculates checksums of the data contained inside a bio
 517  * @inode:       Owner of the data inside the bio
 518  * @bio:         Contains the data to be checksummed
 519  * @file_start:  offset in file this bio begins to describe
 520  * @contig:      Boolean. If true/1 means all bio vecs in this bio are
 521  *               contiguous and they begin at @file_start in the file. False/0
 522  *               means this bio can contains potentially discontigous bio vecs
 523  *               so the logical offset of each should be calculated separately.
 524  */
 525 blk_status_t btrfs_csum_one_bio(struct btrfs_inode *inode, struct bio *bio,
 526                        u64 file_start, int contig)
 527 {
 528         struct btrfs_fs_info *fs_info = inode->root->fs_info;
 529         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
 530         struct btrfs_ordered_sum *sums;
 531         struct btrfs_ordered_extent *ordered = NULL;
 532         char *data;
 533         struct bvec_iter iter;
 534         struct bio_vec bvec;
 535         int index;
 536         int nr_sectors;
 537         unsigned long total_bytes = 0;
 538         unsigned long this_sum_bytes = 0;
 539         int i;
 540         u64 offset;
 541         unsigned nofs_flag;
 542
 543         nofs_flag = memalloc_nofs_save();
 544         sums = kvzalloc(btrfs_ordered_sum_size(fs_info, bio->bi_iter.bi_size),
 545                        GFP_KERNEL);
 546         memalloc_nofs_restore(nofs_flag);
 547
 548         if (!sums)
 549                 return BLK_STS_RESOURCE;
 550
 551         sums->len = bio->bi_iter.bi_size;
 552         INIT_LIST_HEAD(&sums->list);
 553
 554         if (contig)
 555                 offset = file_start;
 556         else
 557                 offset = 0; /* shut up gcc */
 558
 559         sums->bytenr = (u64)bio->bi_iter.bi_sector << 9;
 560         index = 0;
 561
 562         shash->tfm = fs_info->csum_shash;
 563
 564         bio_for_each_segment(bvec, bio, iter) {
 565                 if (!contig)
 566                         offset = page_offset(bvec.bv_page) + bvec.bv_offset;
 567
 568                 if (!ordered) {
 569                         ordered = btrfs_lookup_ordered_extent(inode, offset);
 570                         BUG_ON(!ordered); /* Logic error */
 571                 }
 572
 573                 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info,
 574                                                  bvec.bv_len + fs_info->sectorsize
 575                                                  - 1);
 576
 577                 for (i = 0; i < nr_sectors; i++) {
 578                         if (offset >= ordered->file_offset + ordered->num_bytes ||
 579                             offset < ordered->file_offset) {
 580                                 unsigned long bytes_left;
 581
 582                                 sums->len = this_sum_bytes;
 583                                 this_sum_bytes = 0;
 584                                 btrfs_add_ordered_sum(ordered, sums);
 585                                 btrfs_put_ordered_extent(ordered);
 586
 587                                 bytes_left = bio->bi_iter.bi_size - total_bytes;
 588
 589                                 nofs_flag = memalloc_nofs_save();
 590                                 sums = kvzalloc(btrfs_ordered_sum_size(fs_info,
 591                                                       bytes_left), GFP_KERNEL);
 592                                 memalloc_nofs_restore(nofs_flag);
 593                                 BUG_ON(!sums); /* -ENOMEM */
 594                                 sums->len = bytes_left;
 595                                 ordered = btrfs_lookup_ordered_extent(inode,
 596                                                                 offset);
 597                                 ASSERT(ordered); /* Logic error */
 598                                 sums->bytenr = ((u64)bio->bi_iter.bi_sector << 9)
 599                                         + total_bytes;
 600                                 index = 0;
 601                         }
 602
 603                         data = kmap_atomic(bvec.bv_page);
 604                         crypto_shash_digest(shash, data + bvec.bv_offset
 605                                             + (i * fs_info->sectorsize),
 606                                             fs_info->sectorsize,
 607                                             sums->sums + index);
 608                         kunmap_atomic(data);
 609                         index += fs_info->csum_size;
 610                         offset += fs_info->sectorsize;
 611                         this_sum_bytes += fs_info->sectorsize;
 612                         total_bytes += fs_info->sectorsize;
 613                 }
 614
 615         }
 616         this_sum_bytes = 0;
 617         btrfs_add_ordered_sum(ordered, sums);
 618         btrfs_put_ordered_extent(ordered);
 619         return 0;
 620 }
 621
 622 /*
 623  * helper function for csum removal, this expects the
 624  * key to describe the csum pointed to by the path, and it expects
 625  * the csum to overlap the range [bytenr, len]
 626  *
 627  * The csum should not be entirely contained in the range and the
 628  * range should not be entirely contained in the csum.
 629  *
 630  * This calls btrfs_truncate_item with the correct args based on the
 631  * overlap, and fixes up the key as required.
 632  */
 633 static noinline void truncate_one_csum(struct btrfs_fs_info *fs_info,
 634                                        struct btrfs_path *path,
 635                                        struct btrfs_key *key,
 636                                        u64 bytenr, u64 len)
 637 {
 638         struct extent_buffer *leaf;
 639         const u32 csum_size = fs_info->csum_size;
 640         u64 csum_end;
 641         u64 end_byte = bytenr + len;
 642         u32 blocksize_bits = fs_info->sectorsize_bits;
 643
 644         leaf = path->nodes[0];
 645         csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
 646         csum_end <<= blocksize_bits;
 647         csum_end += key->offset;
 648
 649         if (key->offset < bytenr && csum_end <= end_byte) {
 650                 /*
 651                  *         [ bytenr - len ]
 652                  *         [   ]
 653                  *   [csum     ]
 654                  *   A simple truncate off the end of the item
 655                  */
 656                 u32 new_size = (bytenr - key->offset) >> blocksize_bits;
 657                 new_size *= csum_size;
 658                 btrfs_truncate_item(path, new_size, 1);
 659         } else if (key->offset >= bytenr && csum_end > end_byte &&
 660                    end_byte > key->offset) {
 661                 /*
 662                  *         [ bytenr - len ]
 663                  *                 [ ]
 664                  *                 [csum     ]
 665                  * we need to truncate from the beginning of the csum
 666                  */
 667                 u32 new_size = (csum_end - end_byte) >> blocksize_bits;
 668                 new_size *= csum_size;
 669
 670                 btrfs_truncate_item(path, new_size, 0);
 671
 672                 key->offset = end_byte;
 673                 btrfs_set_item_key_safe(fs_info, path, key);
 674         } else {
 675                 BUG();
 676         }
 677 }
 678
 679 /*
 680  * deletes the csum items from the csum tree for a given
 681  * range of bytes.
 682  */
 683 int btrfs_del_csums(struct btrfs_trans_handle *trans,
 684                     struct btrfs_root *root, u64 bytenr, u64 len)
 685 {
 686         struct btrfs_fs_info *fs_info = trans->fs_info;
 687         struct btrfs_path *path;
 688         struct btrfs_key key;
 689         u64 end_byte = bytenr + len;
 690         u64 csum_end;
 691         struct extent_buffer *leaf;
 692         int ret;
 693         const u32 csum_size = fs_info->csum_size;
 694         u32 blocksize_bits = fs_info->sectorsize_bits;
 695
 696         ASSERT(root == fs_info->csum_root ||
 697                root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
 698
 699         path = btrfs_alloc_path();
 700         if (!path)
 701                 return -ENOMEM;
 702
 703         while (1) {
 704                 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
 705                 key.offset = end_byte - 1;
 706                 key.type = BTRFS_EXTENT_CSUM_KEY;
 707
 708                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 709                 if (ret > 0) {
 710                         if (path->slots[0] == 0)
 711                                 break;
 712                         path->slots[0]--;
 713                 } else if (ret < 0) {
 714                         break;
 715                 }
 716
 717                 leaf = path->nodes[0];
 718                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 719
 720                 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
 721                     key.type != BTRFS_EXTENT_CSUM_KEY) {
 722                         break;
 723                 }
 724
 725                 if (key.offset >= end_byte)
 726                         break;
 727
 728                 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
 729                 csum_end <<= blocksize_bits;
 730                 csum_end += key.offset;
 731
 732                 /* this csum ends before we start, we're done */
 733                 if (csum_end <= bytenr)
 734                         break;
 735
 736                 /* delete the entire item, it is inside our range */
 737                 if (key.offset >= bytenr && csum_end <= end_byte) {
 738                         int del_nr = 1;
 739
 740                         /*
 741                          * Check how many csum items preceding this one in this
 742                          * leaf correspond to our range and then delete them all
 743                          * at once.
 744                          */
 745                         if (key.offset > bytenr && path->slots[0] > 0) {
 746                                 int slot = path->slots[0] - 1;
 747
 748                                 while (slot >= 0) {
 749                                         struct btrfs_key pk;
 750
 751                                         btrfs_item_key_to_cpu(leaf, &pk, slot);
 752                                         if (pk.offset < bytenr ||
 753                                             pk.type != BTRFS_EXTENT_CSUM_KEY ||
 754                                             pk.objectid !=
 755                                             BTRFS_EXTENT_CSUM_OBJECTID)
 756                                                 break;
 757                                         path->slots[0] = slot;
 758                                         del_nr++;
 759                                         key.offset = pk.offset;
 760                                         slot--;
 761                                 }
 762                         }
 763                         ret = btrfs_del_items(trans, root, path,
 764                                               path->slots[0], del_nr);
 765                         if (ret)
 766                                 goto out;
 767                         if (key.offset == bytenr)
 768                                 break;
 769                 } else if (key.offset < bytenr && csum_end > end_byte) {
 770                         unsigned long offset;
 771                         unsigned long shift_len;
 772                         unsigned long item_offset;
 773                         /*
 774                          *        [ bytenr - len ]
 775                          *     [csum                ]
 776                          *
 777                          * Our bytes are in the middle of the csum,
 778                          * we need to split this item and insert a new one.
 779                          *
 780                          * But we can't drop the path because the
 781                          * csum could change, get removed, extended etc.
 782                          *
 783                          * The trick here is the max size of a csum item leaves
 784                          * enough room in the tree block for a single
 785                          * item header.  So, we split the item in place,
 786                          * adding a new header pointing to the existing
 787                          * bytes.  Then we loop around again and we have
 788                          * a nicely formed csum item that we can neatly
 789                          * truncate.
 790                          */
 791                         offset = (bytenr - key.offset) >> blocksize_bits;
 792                         offset *= csum_size;
 793
 794                         shift_len = (len >> blocksize_bits) * csum_size;
 795
 796                         item_offset = btrfs_item_ptr_offset(leaf,
 797                                                             path->slots[0]);
 798
 799                         memzero_extent_buffer(leaf, item_offset + offset,
 800                                              shift_len);
 801                         key.offset = bytenr;
 802
 803                         /*
 804                          * btrfs_split_item returns -EAGAIN when the
 805                          * item changed size or key
 806                          */
 807                         ret = btrfs_split_item(trans, root, path, &key, offset);
 808                         if (ret && ret != -EAGAIN) {
 809                                 btrfs_abort_transaction(trans, ret);
 810                                 goto out;
 811                         }
 812
 813                         key.offset = end_byte - 1;
 814                 } else {
 815                         truncate_one_csum(fs_info, path, &key, bytenr, len);
 816                         if (key.offset < bytenr)
 817                                 break;
 818                 }
 819                 btrfs_release_path(path);
 820         }
 821         ret = 0;
 822 out:
 823         btrfs_free_path(path);
 824         return ret;
 825 }
 826
 827 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
 828                            struct btrfs_root *root,
 829                            struct btrfs_ordered_sum *sums)
 830 {
 831         struct btrfs_fs_info *fs_info = root->fs_info;
 832         struct btrfs_key file_key;
 833         struct btrfs_key found_key;
 834         struct btrfs_path *path;
 835         struct btrfs_csum_item *item;
 836         struct btrfs_csum_item *item_end;
 837         struct extent_buffer *leaf = NULL;
 838         u64 next_offset;
 839         u64 total_bytes = 0;
 840         u64 csum_offset;
 841         u64 bytenr;
 842         u32 nritems;
 843         u32 ins_size;
 844         int index = 0;
 845         int found_next;
 846         int ret;
 847         const u32 csum_size = fs_info->csum_size;
 848
 849         path = btrfs_alloc_path();
 850         if (!path)
 851                 return -ENOMEM;
 852 again:
 853         next_offset = (u64)-1;
 854         found_next = 0;
 855         bytenr = sums->bytenr + total_bytes;
 856         file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
 857         file_key.offset = bytenr;
 858         file_key.type = BTRFS_EXTENT_CSUM_KEY;
 859
 860         item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
 861         if (!IS_ERR(item)) {
 862                 ret = 0;
 863                 leaf = path->nodes[0];
 864                 item_end = btrfs_item_ptr(leaf, path->slots[0],
 865                                           struct btrfs_csum_item);
 866                 item_end = (struct btrfs_csum_item *)((char *)item_end +
 867                            btrfs_item_size_nr(leaf, path->slots[0]));
 868                 goto found;
 869         }
 870         ret = PTR_ERR(item);
 871         if (ret != -EFBIG && ret != -ENOENT)
 872                 goto out;
 873
 874         if (ret == -EFBIG) {
 875                 u32 item_size;
 876                 /* we found one, but it isn't big enough yet */
 877                 leaf = path->nodes[0];
 878                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
 879                 if ((item_size / csum_size) >=
 880                     MAX_CSUM_ITEMS(fs_info, csum_size)) {
 881                         /* already at max size, make a new one */
 882                         goto insert;
 883                 }
 884         } else {
 885                 int slot = path->slots[0] + 1;
 886                 /* we didn't find a csum item, insert one */
 887                 nritems = btrfs_header_nritems(path->nodes[0]);
 888                 if (!nritems || (path->slots[0] >= nritems - 1)) {
 889                         ret = btrfs_next_leaf(root, path);
 890                         if (ret < 0) {
 891                                 goto out;
 892                         } else if (ret > 0) {
 893                                 found_next = 1;
 894                                 goto insert;
 895                         }
 896                         slot = path->slots[0];
 897                 }
 898                 btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
 899                 if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
 900                     found_key.type != BTRFS_EXTENT_CSUM_KEY) {
 901                         found_next = 1;
 902                         goto insert;
 903                 }
 904                 next_offset = found_key.offset;
 905                 found_next = 1;
 906                 goto insert;
 907         }
 908
 909         /*
 910          * At this point, we know the tree has a checksum item that ends at an
 911          * offset matching the start of the checksum range we want to insert.
 912          * We try to extend that item as much as possible and then add as many
 913          * checksums to it as they fit.
 914          *
 915          * First check if the leaf has enough free space for at least one
 916          * checksum. If it has go directly to the item extension code, otherwise
 917          * release the path and do a search for insertion before the extension.
 918          */
 919         if (btrfs_leaf_free_space(leaf) >= csum_size) {
 920                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 921                 csum_offset = (bytenr - found_key.offset) >>
 922                         fs_info->sectorsize_bits;
 923                 goto extend_csum;
 924         }
 925
 926         btrfs_release_path(path);
 927         ret = btrfs_search_slot(trans, root, &file_key, path,
 928                                 csum_size, 1);
 929         if (ret < 0)
 930                 goto out;
 931
 932         if (ret > 0) {
 933                 if (path->slots[0] == 0)
 934                         goto insert;
 935                 path->slots[0]--;
 936         }
 937
 938         leaf = path->nodes[0];
 939         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 940         csum_offset = (bytenr - found_key.offset) >> fs_info->sectorsize_bits;
 941
 942         if (found_key.type != BTRFS_EXTENT_CSUM_KEY ||
 943             found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
 944             csum_offset >= MAX_CSUM_ITEMS(fs_info, csum_size)) {
 945                 goto insert;
 946         }
 947
 948 extend_csum:
 949         if (csum_offset == btrfs_item_size_nr(leaf, path->slots[0]) /
 950             csum_size) {
 951                 int extend_nr;
 952                 u64 tmp;
 953                 u32 diff;
 954
 955                 tmp = sums->len - total_bytes;
 956                 tmp >>= fs_info->sectorsize_bits;
 957                 WARN_ON(tmp < 1);
 958
 959                 extend_nr = max_t(int, 1, (int)tmp);
 960                 diff = (csum_offset + extend_nr) * csum_size;
 961                 diff = min(diff,
 962                            MAX_CSUM_ITEMS(fs_info, csum_size) * csum_size);
 963
 964                 diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
 965                 diff = min_t(u32, btrfs_leaf_free_space(leaf), diff);
 966                 diff /= csum_size;
 967                 diff *= csum_size;
 968
 969                 btrfs_extend_item(path, diff);
 970                 ret = 0;
 971                 goto csum;
 972         }
 973
 974 insert:
 975         btrfs_release_path(path);
 976         csum_offset = 0;
 977         if (found_next) {
 978                 u64 tmp;
 979
 980                 tmp = sums->len - total_bytes;
 981                 tmp >>= fs_info->sectorsize_bits;
 982                 tmp = min(tmp, (next_offset - file_key.offset) >>
 983                                          fs_info->sectorsize_bits);
 984
 985                 tmp = max_t(u64, 1, tmp);
 986                 tmp = min_t(u64, tmp, MAX_CSUM_ITEMS(fs_info, csum_size));
 987                 ins_size = csum_size * tmp;
 988         } else {
 989                 ins_size = csum_size;
 990         }
 991         ret = btrfs_insert_empty_item(trans, root, path, &file_key,
 992                                       ins_size);
 993         if (ret < 0)
 994                 goto out;
 995         if (WARN_ON(ret != 0))
 996                 goto out;
 997         leaf = path->nodes[0];
 998 csum:
 999         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
1000         item_end = (struct btrfs_csum_item *)((unsigned char *)item +
1001                                       btrfs_item_size_nr(leaf, path->slots[0]));
1002         item = (struct btrfs_csum_item *)((unsigned char *)item +
1003                                           csum_offset * csum_size);
1004 found:
1005         ins_size = (u32)(sums->len - total_bytes) >> fs_info->sectorsize_bits;
1006         ins_size *= csum_size;
1007         ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item,
1008                               ins_size);
1009         write_extent_buffer(leaf, sums->sums + index, (unsigned long)item,
1010                             ins_size);
1011
1012         index += ins_size;
1013         ins_size /= csum_size;
1014         total_bytes += ins_size * fs_info->sectorsize;
1015
1016         btrfs_mark_buffer_dirty(path->nodes[0]);
1017         if (total_bytes < sums->len) {
1018                 btrfs_release_path(path);
1019                 cond_resched();
1020                 goto again;
1021         }
1022 out:
1023         btrfs_free_path(path);
1024         return ret;
1025 }
1026
1027 void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode,
1028                                      const struct btrfs_path *path,
1029                                      struct btrfs_file_extent_item *fi,
1030                                      const bool new_inline,
1031                                      struct extent_map *em)
1032 {
1033         struct btrfs_fs_info *fs_info = inode->root->fs_info;
1034         struct btrfs_root *root = inode->root;
1035         struct extent_buffer *leaf = path->nodes[0];
1036         const int slot = path->slots[0];
1037         struct btrfs_key key;
1038         u64 extent_start, extent_end;
1039         u64 bytenr;
1040         u8 type = btrfs_file_extent_type(leaf, fi);
1041         int compress_type = btrfs_file_extent_compression(leaf, fi);
1042
1043         btrfs_item_key_to_cpu(leaf, &key, slot);
1044         extent_start = key.offset;
1045         extent_end = btrfs_file_extent_end(path);
1046         em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1047         if (type == BTRFS_FILE_EXTENT_REG ||
1048             type == BTRFS_FILE_EXTENT_PREALLOC) {
1049                 em->start = extent_start;
1050                 em->len = extent_end - extent_start;
1051                 em->orig_start = extent_start -
1052                         btrfs_file_extent_offset(leaf, fi);
1053                 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
1054                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1055                 if (bytenr == 0) {
1056                         em->block_start = EXTENT_MAP_HOLE;
1057                         return;
1058                 }
1059                 if (compress_type != BTRFS_COMPRESS_NONE) {
1060                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1061                         em->compress_type = compress_type;
1062                         em->block_start = bytenr;
1063                         em->block_len = em->orig_block_len;
1064                 } else {
1065                         bytenr += btrfs_file_extent_offset(leaf, fi);
1066                         em->block_start = bytenr;
1067                         em->block_len = em->len;
1068                         if (type == BTRFS_FILE_EXTENT_PREALLOC)
1069                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
1070                 }
1071         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1072                 em->block_start = EXTENT_MAP_INLINE;
1073                 em->start = extent_start;
1074                 em->len = extent_end - extent_start;
1075                 /*
1076                  * Initialize orig_start and block_len with the same values
1077                  * as in inode.c:btrfs_get_extent().
1078                  */
1079                 em->orig_start = EXTENT_MAP_HOLE;
1080                 em->block_len = (u64)-1;
1081                 if (!new_inline && compress_type != BTRFS_COMPRESS_NONE) {
1082                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1083                         em->compress_type = compress_type;
1084                 }
1085         } else {
1086                 btrfs_err(fs_info,
1087                           "unknown file extent item type %d, inode %llu, offset %llu, "
1088                           "root %llu", type, btrfs_ino(inode), extent_start,
1089                           root->root_key.objectid);
1090         }
1091 }
1092
1093 /*
1094  * Returns the end offset (non inclusive) of the file extent item the given path
1095  * points to. If it points to an inline extent, the returned offset is rounded
1096  * up to the sector size.
1097  */
1098 u64 btrfs_file_extent_end(const struct btrfs_path *path)
1099 {
1100         const struct extent_buffer *leaf = path->nodes[0];
1101         const int slot = path->slots[0];
1102         struct btrfs_file_extent_item *fi;
1103         struct btrfs_key key;
1104         u64 end;
1105
1106         btrfs_item_key_to_cpu(leaf, &key, slot);
1107         ASSERT(key.type == BTRFS_EXTENT_DATA_KEY);
1108         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1109
1110         if (btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_INLINE) {
1111                 end = btrfs_file_extent_ram_bytes(leaf, fi);
1112                 end = ALIGN(key.offset + end, leaf->fs_info->sectorsize);
1113         } else {
1114                 end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1115         }
1116
1117         return end;
1118 }