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         path->leave_spinning = 1;
 146         ret = btrfs_insert_empty_item(trans, root, path, &file_key,
 147                                       sizeof(*item));
 148         if (ret < 0)
 149                 goto out;
 150         BUG_ON(ret); /* Can't happen */
 151         leaf = path->nodes[0];
 152         item = btrfs_item_ptr(leaf, path->slots[0],
 153                               struct btrfs_file_extent_item);
 154         btrfs_set_file_extent_disk_bytenr(leaf, item, disk_offset);
 155         btrfs_set_file_extent_disk_num_bytes(leaf, item, disk_num_bytes);
 156         btrfs_set_file_extent_offset(leaf, item, offset);
 157         btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
 158         btrfs_set_file_extent_ram_bytes(leaf, item, ram_bytes);
 159         btrfs_set_file_extent_generation(leaf, item, trans->transid);
 160         btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
 161         btrfs_set_file_extent_compression(leaf, item, compression);
 162         btrfs_set_file_extent_encryption(leaf, item, encryption);
 163         btrfs_set_file_extent_other_encoding(leaf, item, other_encoding);
 164
 165         btrfs_mark_buffer_dirty(leaf);
 166 out:
 167         btrfs_free_path(path);
 168         return ret;
 169 }
 170
 171 static struct btrfs_csum_item *
 172 btrfs_lookup_csum(struct btrfs_trans_handle *trans,
 173                   struct btrfs_root *root,
 174                   struct btrfs_path *path,
 175                   u64 bytenr, int cow)
 176 {
 177         struct btrfs_fs_info *fs_info = root->fs_info;
 178         int ret;
 179         struct btrfs_key file_key;
 180         struct btrfs_key found_key;
 181         struct btrfs_csum_item *item;
 182         struct extent_buffer *leaf;
 183         u64 csum_offset = 0;
 184         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 185         int csums_in_item;
 186
 187         file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
 188         file_key.offset = bytenr;
 189         file_key.type = BTRFS_EXTENT_CSUM_KEY;
 190         ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
 191         if (ret < 0)
 192                 goto fail;
 193         leaf = path->nodes[0];
 194         if (ret > 0) {
 195                 ret = 1;
 196                 if (path->slots[0] == 0)
 197                         goto fail;
 198                 path->slots[0]--;
 199                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 200                 if (found_key.type != BTRFS_EXTENT_CSUM_KEY)
 201                         goto fail;
 202
 203                 csum_offset = (bytenr - found_key.offset) >>
 204                                 fs_info->sb->s_blocksize_bits;
 205                 csums_in_item = btrfs_item_size_nr(leaf, path->slots[0]);
 206                 csums_in_item /= csum_size;
 207
 208                 if (csum_offset == csums_in_item) {
 209                         ret = -EFBIG;
 210                         goto fail;
 211                 } else if (csum_offset > csums_in_item) {
 212                         goto fail;
 213                 }
 214         }
 215         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
 216         item = (struct btrfs_csum_item *)((unsigned char *)item +
 217                                           csum_offset * csum_size);
 218         return item;
 219 fail:
 220         if (ret > 0)
 221                 ret = -ENOENT;
 222         return ERR_PTR(ret);
 223 }
 224
 225 int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
 226                              struct btrfs_root *root,
 227                              struct btrfs_path *path, u64 objectid,
 228                              u64 offset, int mod)
 229 {
 230         int ret;
 231         struct btrfs_key file_key;
 232         int ins_len = mod < 0 ? -1 : 0;
 233         int cow = mod != 0;
 234
 235         file_key.objectid = objectid;
 236         file_key.offset = offset;
 237         file_key.type = BTRFS_EXTENT_DATA_KEY;
 238         ret = btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
 239         return ret;
 240 }
 241
 242 /**
 243  * btrfs_lookup_bio_sums - Look up checksums for a bio.
 244  * @inode: inode that the bio is for.
 245  * @bio: bio to look up.
 246  * @offset: Unless (u64)-1, look up checksums for this offset in the file.
 247  *          If (u64)-1, use the page offsets from the bio instead.
 248  * @dst: Buffer of size nblocks * btrfs_super_csum_size() used to return
 249  *       checksum (nblocks = bio->bi_iter.bi_size / fs_info->sectorsize). If
 250  *       NULL, the checksum buffer is allocated and returned in
 251  *       btrfs_io_bio(bio)->csum instead.
 252  *
 253  * Return: BLK_STS_RESOURCE if allocating memory fails, BLK_STS_OK otherwise.
 254  */
 255 blk_status_t btrfs_lookup_bio_sums(struct inode *inode, struct bio *bio,
 256                                    u64 offset, u8 *dst)
 257 {
 258         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 259         struct bio_vec bvec;
 260         struct bvec_iter iter;
 261         struct btrfs_csum_item *item = NULL;
 262         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 263         struct btrfs_path *path;
 264         const bool page_offsets = (offset == (u64)-1);
 265         u8 *csum;
 266         u64 item_start_offset = 0;
 267         u64 item_last_offset = 0;
 268         u64 disk_bytenr;
 269         u64 page_bytes_left;
 270         u32 diff;
 271         int nblocks;
 272         int count = 0;
 273         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 274
 275         path = btrfs_alloc_path();
 276         if (!path)
 277                 return BLK_STS_RESOURCE;
 278
 279         nblocks = bio->bi_iter.bi_size >> inode->i_sb->s_blocksize_bits;
 280         if (!dst) {
 281                 struct btrfs_io_bio *btrfs_bio = btrfs_io_bio(bio);
 282
 283                 if (nblocks * csum_size > BTRFS_BIO_INLINE_CSUM_SIZE) {
 284                         btrfs_bio->csum = kmalloc_array(nblocks, csum_size,
 285                                                         GFP_NOFS);
 286                         if (!btrfs_bio->csum) {
 287                                 btrfs_free_path(path);
 288                                 return BLK_STS_RESOURCE;
 289                         }
 290                 } else {
 291                         btrfs_bio->csum = btrfs_bio->csum_inline;
 292                 }
 293                 csum = btrfs_bio->csum;
 294         } else {
 295                 csum = dst;
 296         }
 297
 298         if (bio->bi_iter.bi_size > PAGE_SIZE * 8)
 299                 path->reada = READA_FORWARD;
 300
 301         /*
 302          * the free space stuff is only read when it hasn't been
 303          * updated in the current transaction.  So, we can safely
 304          * read from the commit root and sidestep a nasty deadlock
 305          * between reading the free space cache and updating the csum tree.
 306          */
 307         if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
 308                 path->search_commit_root = 1;
 309                 path->skip_locking = 1;
 310         }
 311
 312         disk_bytenr = (u64)bio->bi_iter.bi_sector << 9;
 313
 314         bio_for_each_segment(bvec, bio, iter) {
 315                 page_bytes_left = bvec.bv_len;
 316                 if (count)
 317                         goto next;
 318
 319                 if (page_offsets)
 320                         offset = page_offset(bvec.bv_page) + bvec.bv_offset;
 321                 count = btrfs_find_ordered_sum(inode, offset, disk_bytenr,
 322                                                csum, nblocks);
 323                 if (count)
 324                         goto found;
 325
 326                 if (!item || disk_bytenr < item_start_offset ||
 327                     disk_bytenr >= item_last_offset) {
 328                         struct btrfs_key found_key;
 329                         u32 item_size;
 330
 331                         if (item)
 332                                 btrfs_release_path(path);
 333                         item = btrfs_lookup_csum(NULL, fs_info->csum_root,
 334                                                  path, disk_bytenr, 0);
 335                         if (IS_ERR(item)) {
 336                                 count = 1;
 337                                 memset(csum, 0, csum_size);
 338                                 if (BTRFS_I(inode)->root->root_key.objectid ==
 339                                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 340                                         set_extent_bits(io_tree, offset,
 341                                                 offset + fs_info->sectorsize - 1,
 342                                                 EXTENT_NODATASUM);
 343                                 } else {
 344                                         btrfs_info_rl(fs_info,
 345                                                    "no csum found for inode %llu start %llu",
 346                                                btrfs_ino(BTRFS_I(inode)), offset);
 347                                 }
 348                                 item = NULL;
 349                                 btrfs_release_path(path);
 350                                 goto found;
 351                         }
 352                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 353                                               path->slots[0]);
 354
 355                         item_start_offset = found_key.offset;
 356                         item_size = btrfs_item_size_nr(path->nodes[0],
 357                                                        path->slots[0]);
 358                         item_last_offset = item_start_offset +
 359                                 (item_size / csum_size) *
 360                                 fs_info->sectorsize;
 361                         item = btrfs_item_ptr(path->nodes[0], path->slots[0],
 362                                               struct btrfs_csum_item);
 363                 }
 364                 /*
 365                  * this byte range must be able to fit inside
 366                  * a single leaf so it will also fit inside a u32
 367                  */
 368                 diff = disk_bytenr - item_start_offset;
 369                 diff = diff / fs_info->sectorsize;
 370                 diff = diff * csum_size;
 371                 count = min_t(int, nblocks, (item_last_offset - disk_bytenr) >>
 372                                             inode->i_sb->s_blocksize_bits);
 373                 read_extent_buffer(path->nodes[0], csum,
 374                                    ((unsigned long)item) + diff,
 375                                    csum_size * count);
 376 found:
 377                 csum += count * csum_size;
 378                 nblocks -= count;
 379 next:
 380                 while (count > 0) {
 381                         count--;
 382                         disk_bytenr += fs_info->sectorsize;
 383                         offset += fs_info->sectorsize;
 384                         page_bytes_left -= fs_info->sectorsize;
 385                         if (!page_bytes_left)
 386                                 break; /* move to next bio */
 387                 }
 388         }
 389
 390         WARN_ON_ONCE(count);
 391         btrfs_free_path(path);
 392         return BLK_STS_OK;
 393 }
 394
 395 int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
 396                              struct list_head *list, int search_commit)
 397 {
 398         struct btrfs_fs_info *fs_info = root->fs_info;
 399         struct btrfs_key key;
 400         struct btrfs_path *path;
 401         struct extent_buffer *leaf;
 402         struct btrfs_ordered_sum *sums;
 403         struct btrfs_csum_item *item;
 404         LIST_HEAD(tmplist);
 405         unsigned long offset;
 406         int ret;
 407         size_t size;
 408         u64 csum_end;
 409         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 410
 411         ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
 412                IS_ALIGNED(end + 1, fs_info->sectorsize));
 413
 414         path = btrfs_alloc_path();
 415         if (!path)
 416                 return -ENOMEM;
 417
 418         if (search_commit) {
 419                 path->skip_locking = 1;
 420                 path->reada = READA_FORWARD;
 421                 path->search_commit_root = 1;
 422         }
 423
 424         key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
 425         key.offset = start;
 426         key.type = BTRFS_EXTENT_CSUM_KEY;
 427
 428         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 429         if (ret < 0)
 430                 goto fail;
 431         if (ret > 0 && path->slots[0] > 0) {
 432                 leaf = path->nodes[0];
 433                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
 434                 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
 435                     key.type == BTRFS_EXTENT_CSUM_KEY) {
 436                         offset = (start - key.offset) >>
 437                                  fs_info->sb->s_blocksize_bits;
 438                         if (offset * csum_size <
 439                             btrfs_item_size_nr(leaf, path->slots[0] - 1))
 440                                 path->slots[0]--;
 441                 }
 442         }
 443
 444         while (start <= end) {
 445                 leaf = path->nodes[0];
 446                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 447                         ret = btrfs_next_leaf(root, path);
 448                         if (ret < 0)
 449                                 goto fail;
 450                         if (ret > 0)
 451                                 break;
 452                         leaf = path->nodes[0];
 453                 }
 454
 455                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 456                 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
 457                     key.type != BTRFS_EXTENT_CSUM_KEY ||
 458                     key.offset > end)
 459                         break;
 460
 461                 if (key.offset > start)
 462                         start = key.offset;
 463
 464                 size = btrfs_item_size_nr(leaf, path->slots[0]);
 465                 csum_end = key.offset + (size / csum_size) * fs_info->sectorsize;
 466                 if (csum_end <= start) {
 467                         path->slots[0]++;
 468                         continue;
 469                 }
 470
 471                 csum_end = min(csum_end, end + 1);
 472                 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
 473                                       struct btrfs_csum_item);
 474                 while (start < csum_end) {
 475                         size = min_t(size_t, csum_end - start,
 476                                      max_ordered_sum_bytes(fs_info, csum_size));
 477                         sums = kzalloc(btrfs_ordered_sum_size(fs_info, size),
 478                                        GFP_NOFS);
 479                         if (!sums) {
 480                                 ret = -ENOMEM;
 481                                 goto fail;
 482                         }
 483
 484                         sums->bytenr = start;
 485                         sums->len = (int)size;
 486
 487                         offset = (start - key.offset) >>
 488                                 fs_info->sb->s_blocksize_bits;
 489                         offset *= csum_size;
 490                         size >>= fs_info->sb->s_blocksize_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         const u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 543
 544         nofs_flag = memalloc_nofs_save();
 545         sums = kvzalloc(btrfs_ordered_sum_size(fs_info, bio->bi_iter.bi_size),
 546                        GFP_KERNEL);
 547         memalloc_nofs_restore(nofs_flag);
 548
 549         if (!sums)
 550                 return BLK_STS_RESOURCE;
 551
 552         sums->len = bio->bi_iter.bi_size;
 553         INIT_LIST_HEAD(&sums->list);
 554
 555         if (contig)
 556                 offset = file_start;
 557         else
 558                 offset = 0; /* shut up gcc */
 559
 560         sums->bytenr = (u64)bio->bi_iter.bi_sector << 9;
 561         index = 0;
 562
 563         shash->tfm = fs_info->csum_shash;
 564
 565         bio_for_each_segment(bvec, bio, iter) {
 566                 if (!contig)
 567                         offset = page_offset(bvec.bv_page) + bvec.bv_offset;
 568
 569                 if (!ordered) {
 570                         ordered = btrfs_lookup_ordered_extent(inode, offset);
 571                         BUG_ON(!ordered); /* Logic error */
 572                 }
 573
 574                 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info,
 575                                                  bvec.bv_len + fs_info->sectorsize
 576                                                  - 1);
 577
 578                 for (i = 0; i < nr_sectors; i++) {
 579                         if (offset >= ordered->file_offset + ordered->num_bytes ||
 580                             offset < ordered->file_offset) {
 581                                 unsigned long bytes_left;
 582
 583                                 sums->len = this_sum_bytes;
 584                                 this_sum_bytes = 0;
 585                                 btrfs_add_ordered_sum(ordered, sums);
 586                                 btrfs_put_ordered_extent(ordered);
 587
 588                                 bytes_left = bio->bi_iter.bi_size - total_bytes;
 589
 590                                 nofs_flag = memalloc_nofs_save();
 591                                 sums = kvzalloc(btrfs_ordered_sum_size(fs_info,
 592                                                       bytes_left), GFP_KERNEL);
 593                                 memalloc_nofs_restore(nofs_flag);
 594                                 BUG_ON(!sums); /* -ENOMEM */
 595                                 sums->len = bytes_left;
 596                                 ordered = btrfs_lookup_ordered_extent(inode,
 597                                                                 offset);
 598                                 ASSERT(ordered); /* Logic error */
 599                                 sums->bytenr = ((u64)bio->bi_iter.bi_sector << 9)
 600                                         + total_bytes;
 601                                 index = 0;
 602                         }
 603
 604                         data = kmap_atomic(bvec.bv_page);
 605                         crypto_shash_digest(shash, data + bvec.bv_offset
 606                                             + (i * fs_info->sectorsize),
 607                                             fs_info->sectorsize,
 608                                             sums->sums + index);
 609                         kunmap_atomic(data);
 610                         index += csum_size;
 611                         offset += fs_info->sectorsize;
 612                         this_sum_bytes += fs_info->sectorsize;
 613                         total_bytes += fs_info->sectorsize;
 614                 }
 615
 616         }
 617         this_sum_bytes = 0;
 618         btrfs_add_ordered_sum(ordered, sums);
 619         btrfs_put_ordered_extent(ordered);
 620         return 0;
 621 }
 622
 623 /*
 624  * helper function for csum removal, this expects the
 625  * key to describe the csum pointed to by the path, and it expects
 626  * the csum to overlap the range [bytenr, len]
 627  *
 628  * The csum should not be entirely contained in the range and the
 629  * range should not be entirely contained in the csum.
 630  *
 631  * This calls btrfs_truncate_item with the correct args based on the
 632  * overlap, and fixes up the key as required.
 633  */
 634 static noinline void truncate_one_csum(struct btrfs_fs_info *fs_info,
 635                                        struct btrfs_path *path,
 636                                        struct btrfs_key *key,
 637                                        u64 bytenr, u64 len)
 638 {
 639         struct extent_buffer *leaf;
 640         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 641         u64 csum_end;
 642         u64 end_byte = bytenr + len;
 643         u32 blocksize_bits = fs_info->sb->s_blocksize_bits;
 644
 645         leaf = path->nodes[0];
 646         csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
 647         csum_end <<= fs_info->sb->s_blocksize_bits;
 648         csum_end += key->offset;
 649
 650         if (key->offset < bytenr && csum_end <= end_byte) {
 651                 /*
 652                  *         [ bytenr - len ]
 653                  *         [   ]
 654                  *   [csum     ]
 655                  *   A simple truncate off the end of the item
 656                  */
 657                 u32 new_size = (bytenr - key->offset) >> blocksize_bits;
 658                 new_size *= csum_size;
 659                 btrfs_truncate_item(path, new_size, 1);
 660         } else if (key->offset >= bytenr && csum_end > end_byte &&
 661                    end_byte > key->offset) {
 662                 /*
 663                  *         [ bytenr - len ]
 664                  *                 [ ]
 665                  *                 [csum     ]
 666                  * we need to truncate from the beginning of the csum
 667                  */
 668                 u32 new_size = (csum_end - end_byte) >> blocksize_bits;
 669                 new_size *= csum_size;
 670
 671                 btrfs_truncate_item(path, new_size, 0);
 672
 673                 key->offset = end_byte;
 674                 btrfs_set_item_key_safe(fs_info, path, key);
 675         } else {
 676                 BUG();
 677         }
 678 }
 679
 680 /*
 681  * deletes the csum items from the csum tree for a given
 682  * range of bytes.
 683  */
 684 int btrfs_del_csums(struct btrfs_trans_handle *trans,
 685                     struct btrfs_root *root, u64 bytenr, u64 len)
 686 {
 687         struct btrfs_fs_info *fs_info = trans->fs_info;
 688         struct btrfs_path *path;
 689         struct btrfs_key key;
 690         u64 end_byte = bytenr + len;
 691         u64 csum_end;
 692         struct extent_buffer *leaf;
 693         int ret;
 694         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 695         int blocksize_bits = fs_info->sb->s_blocksize_bits;
 696
 697         ASSERT(root == fs_info->csum_root ||
 698                root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
 699
 700         path = btrfs_alloc_path();
 701         if (!path)
 702                 return -ENOMEM;
 703
 704         while (1) {
 705                 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
 706                 key.offset = end_byte - 1;
 707                 key.type = BTRFS_EXTENT_CSUM_KEY;
 708
 709                 path->leave_spinning = 1;
 710                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 711                 if (ret > 0) {
 712                         if (path->slots[0] == 0)
 713                                 break;
 714                         path->slots[0]--;
 715                 } else if (ret < 0) {
 716                         break;
 717                 }
 718
 719                 leaf = path->nodes[0];
 720                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 721
 722                 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
 723                     key.type != BTRFS_EXTENT_CSUM_KEY) {
 724                         break;
 725                 }
 726
 727                 if (key.offset >= end_byte)
 728                         break;
 729
 730                 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
 731                 csum_end <<= blocksize_bits;
 732                 csum_end += key.offset;
 733
 734                 /* this csum ends before we start, we're done */
 735                 if (csum_end <= bytenr)
 736                         break;
 737
 738                 /* delete the entire item, it is inside our range */
 739                 if (key.offset >= bytenr && csum_end <= end_byte) {
 740                         int del_nr = 1;
 741
 742                         /*
 743                          * Check how many csum items preceding this one in this
 744                          * leaf correspond to our range and then delete them all
 745                          * at once.
 746                          */
 747                         if (key.offset > bytenr && path->slots[0] > 0) {
 748                                 int slot = path->slots[0] - 1;
 749
 750                                 while (slot >= 0) {
 751                                         struct btrfs_key pk;
 752
 753                                         btrfs_item_key_to_cpu(leaf, &pk, slot);
 754                                         if (pk.offset < bytenr ||
 755                                             pk.type != BTRFS_EXTENT_CSUM_KEY ||
 756                                             pk.objectid !=
 757                                             BTRFS_EXTENT_CSUM_OBJECTID)
 758                                                 break;
 759                                         path->slots[0] = slot;
 760                                         del_nr++;
 761                                         key.offset = pk.offset;
 762                                         slot--;
 763                                 }
 764                         }
 765                         ret = btrfs_del_items(trans, root, path,
 766                                               path->slots[0], del_nr);
 767                         if (ret)
 768                                 goto out;
 769                         if (key.offset == bytenr)
 770                                 break;
 771                 } else if (key.offset < bytenr && csum_end > end_byte) {
 772                         unsigned long offset;
 773                         unsigned long shift_len;
 774                         unsigned long item_offset;
 775                         /*
 776                          *        [ bytenr - len ]
 777                          *     [csum                ]
 778                          *
 779                          * Our bytes are in the middle of the csum,
 780                          * we need to split this item and insert a new one.
 781                          *
 782                          * But we can't drop the path because the
 783                          * csum could change, get removed, extended etc.
 784                          *
 785                          * The trick here is the max size of a csum item leaves
 786                          * enough room in the tree block for a single
 787                          * item header.  So, we split the item in place,
 788                          * adding a new header pointing to the existing
 789                          * bytes.  Then we loop around again and we have
 790                          * a nicely formed csum item that we can neatly
 791                          * truncate.
 792                          */
 793                         offset = (bytenr - key.offset) >> blocksize_bits;
 794                         offset *= csum_size;
 795
 796                         shift_len = (len >> blocksize_bits) * csum_size;
 797
 798                         item_offset = btrfs_item_ptr_offset(leaf,
 799                                                             path->slots[0]);
 800
 801                         memzero_extent_buffer(leaf, item_offset + offset,
 802                                              shift_len);
 803                         key.offset = bytenr;
 804
 805                         /*
 806                          * btrfs_split_item returns -EAGAIN when the
 807                          * item changed size or key
 808                          */
 809                         ret = btrfs_split_item(trans, root, path, &key, offset);
 810                         if (ret && ret != -EAGAIN) {
 811                                 btrfs_abort_transaction(trans, ret);
 812                                 goto out;
 813                         }
 814
 815                         key.offset = end_byte - 1;
 816                 } else {
 817                         truncate_one_csum(fs_info, path, &key, bytenr, len);
 818                         if (key.offset < bytenr)
 819                                 break;
 820                 }
 821                 btrfs_release_path(path);
 822         }
 823         ret = 0;
 824 out:
 825         btrfs_free_path(path);
 826         return ret;
 827 }
 828
 829 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
 830                            struct btrfs_root *root,
 831                            struct btrfs_ordered_sum *sums)
 832 {
 833         struct btrfs_fs_info *fs_info = root->fs_info;
 834         struct btrfs_key file_key;
 835         struct btrfs_key found_key;
 836         struct btrfs_path *path;
 837         struct btrfs_csum_item *item;
 838         struct btrfs_csum_item *item_end;
 839         struct extent_buffer *leaf = NULL;
 840         u64 next_offset;
 841         u64 total_bytes = 0;
 842         u64 csum_offset;
 843         u64 bytenr;
 844         u32 nritems;
 845         u32 ins_size;
 846         int index = 0;
 847         int found_next;
 848         int ret;
 849         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 850
 851         path = btrfs_alloc_path();
 852         if (!path)
 853                 return -ENOMEM;
 854 again:
 855         next_offset = (u64)-1;
 856         found_next = 0;
 857         bytenr = sums->bytenr + total_bytes;
 858         file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
 859         file_key.offset = bytenr;
 860         file_key.type = BTRFS_EXTENT_CSUM_KEY;
 861
 862         item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
 863         if (!IS_ERR(item)) {
 864                 ret = 0;
 865                 leaf = path->nodes[0];
 866                 item_end = btrfs_item_ptr(leaf, path->slots[0],
 867                                           struct btrfs_csum_item);
 868                 item_end = (struct btrfs_csum_item *)((char *)item_end +
 869                            btrfs_item_size_nr(leaf, path->slots[0]));
 870                 goto found;
 871         }
 872         ret = PTR_ERR(item);
 873         if (ret != -EFBIG && ret != -ENOENT)
 874                 goto out;
 875
 876         if (ret == -EFBIG) {
 877                 u32 item_size;
 878                 /* we found one, but it isn't big enough yet */
 879                 leaf = path->nodes[0];
 880                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
 881                 if ((item_size / csum_size) >=
 882                     MAX_CSUM_ITEMS(fs_info, csum_size)) {
 883                         /* already at max size, make a new one */
 884                         goto insert;
 885                 }
 886         } else {
 887                 int slot = path->slots[0] + 1;
 888                 /* we didn't find a csum item, insert one */
 889                 nritems = btrfs_header_nritems(path->nodes[0]);
 890                 if (!nritems || (path->slots[0] >= nritems - 1)) {
 891                         ret = btrfs_next_leaf(root, path);
 892                         if (ret < 0) {
 893                                 goto out;
 894                         } else if (ret > 0) {
 895                                 found_next = 1;
 896                                 goto insert;
 897                         }
 898                         slot = path->slots[0];
 899                 }
 900                 btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
 901                 if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
 902                     found_key.type != BTRFS_EXTENT_CSUM_KEY) {
 903                         found_next = 1;
 904                         goto insert;
 905                 }
 906                 next_offset = found_key.offset;
 907                 found_next = 1;
 908                 goto insert;
 909         }
 910
 911         /*
 912          * At this point, we know the tree has a checksum item that ends at an
 913          * offset matching the start of the checksum range we want to insert.
 914          * We try to extend that item as much as possible and then add as many
 915          * checksums to it as they fit.
 916          *
 917          * First check if the leaf has enough free space for at least one
 918          * checksum. If it has go directly to the item extension code, otherwise
 919          * release the path and do a search for insertion before the extension.
 920          */
 921         if (btrfs_leaf_free_space(leaf) >= csum_size) {
 922                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 923                 csum_offset = (bytenr - found_key.offset) >>
 924                         fs_info->sb->s_blocksize_bits;
 925                 goto extend_csum;
 926         }
 927
 928         btrfs_release_path(path);
 929         ret = btrfs_search_slot(trans, root, &file_key, path,
 930                                 csum_size, 1);
 931         if (ret < 0)
 932                 goto out;
 933
 934         if (ret > 0) {
 935                 if (path->slots[0] == 0)
 936                         goto insert;
 937                 path->slots[0]--;
 938         }
 939
 940         leaf = path->nodes[0];
 941         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 942         csum_offset = (bytenr - found_key.offset) >>
 943                         fs_info->sb->s_blocksize_bits;
 944
 945         if (found_key.type != BTRFS_EXTENT_CSUM_KEY ||
 946             found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
 947             csum_offset >= MAX_CSUM_ITEMS(fs_info, csum_size)) {
 948                 goto insert;
 949         }
 950
 951 extend_csum:
 952         if (csum_offset == btrfs_item_size_nr(leaf, path->slots[0]) /
 953             csum_size) {
 954                 int extend_nr;
 955                 u64 tmp;
 956                 u32 diff;
 957
 958                 tmp = sums->len - total_bytes;
 959                 tmp >>= fs_info->sb->s_blocksize_bits;
 960                 WARN_ON(tmp < 1);
 961
 962                 extend_nr = max_t(int, 1, (int)tmp);
 963                 diff = (csum_offset + extend_nr) * csum_size;
 964                 diff = min(diff,
 965                            MAX_CSUM_ITEMS(fs_info, csum_size) * csum_size);
 966
 967                 diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
 968                 diff = min_t(u32, btrfs_leaf_free_space(leaf), diff);
 969                 diff /= csum_size;
 970                 diff *= csum_size;
 971
 972                 btrfs_extend_item(path, diff);
 973                 ret = 0;
 974                 goto csum;
 975         }
 976
 977 insert:
 978         btrfs_release_path(path);
 979         csum_offset = 0;
 980         if (found_next) {
 981                 u64 tmp;
 982
 983                 tmp = sums->len - total_bytes;
 984                 tmp >>= fs_info->sb->s_blocksize_bits;
 985                 tmp = min(tmp, (next_offset - file_key.offset) >>
 986                                          fs_info->sb->s_blocksize_bits);
 987
 988                 tmp = max_t(u64, 1, tmp);
 989                 tmp = min_t(u64, tmp, MAX_CSUM_ITEMS(fs_info, csum_size));
 990                 ins_size = csum_size * tmp;
 991         } else {
 992                 ins_size = csum_size;
 993         }
 994         path->leave_spinning = 1;
 995         ret = btrfs_insert_empty_item(trans, root, path, &file_key,
 996                                       ins_size);
 997         path->leave_spinning = 0;
 998         if (ret < 0)
 999                 goto out;
1000         if (WARN_ON(ret != 0))
1001                 goto out;
1002         leaf = path->nodes[0];
1003 csum:
1004         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
1005         item_end = (struct btrfs_csum_item *)((unsigned char *)item +
1006                                       btrfs_item_size_nr(leaf, path->slots[0]));
1007         item = (struct btrfs_csum_item *)((unsigned char *)item +
1008                                           csum_offset * csum_size);
1009 found:
1010         ins_size = (u32)(sums->len - total_bytes) >>
1011                    fs_info->sb->s_blocksize_bits;
1012         ins_size *= csum_size;
1013         ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item,
1014                               ins_size);
1015         write_extent_buffer(leaf, sums->sums + index, (unsigned long)item,
1016                             ins_size);
1017
1018         index += ins_size;
1019         ins_size /= csum_size;
1020         total_bytes += ins_size * fs_info->sectorsize;
1021
1022         btrfs_mark_buffer_dirty(path->nodes[0]);
1023         if (total_bytes < sums->len) {
1024                 btrfs_release_path(path);
1025                 cond_resched();
1026                 goto again;
1027         }
1028 out:
1029         btrfs_free_path(path);
1030         return ret;
1031 }
1032
1033 void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode,
1034                                      const struct btrfs_path *path,
1035                                      struct btrfs_file_extent_item *fi,
1036                                      const bool new_inline,
1037                                      struct extent_map *em)
1038 {
1039         struct btrfs_fs_info *fs_info = inode->root->fs_info;
1040         struct btrfs_root *root = inode->root;
1041         struct extent_buffer *leaf = path->nodes[0];
1042         const int slot = path->slots[0];
1043         struct btrfs_key key;
1044         u64 extent_start, extent_end;
1045         u64 bytenr;
1046         u8 type = btrfs_file_extent_type(leaf, fi);
1047         int compress_type = btrfs_file_extent_compression(leaf, fi);
1048
1049         btrfs_item_key_to_cpu(leaf, &key, slot);
1050         extent_start = key.offset;
1051         extent_end = btrfs_file_extent_end(path);
1052         em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1053         if (type == BTRFS_FILE_EXTENT_REG ||
1054             type == BTRFS_FILE_EXTENT_PREALLOC) {
1055                 em->start = extent_start;
1056                 em->len = extent_end - extent_start;
1057                 em->orig_start = extent_start -
1058                         btrfs_file_extent_offset(leaf, fi);
1059                 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
1060                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1061                 if (bytenr == 0) {
1062                         em->block_start = EXTENT_MAP_HOLE;
1063                         return;
1064                 }
1065                 if (compress_type != BTRFS_COMPRESS_NONE) {
1066                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1067                         em->compress_type = compress_type;
1068                         em->block_start = bytenr;
1069                         em->block_len = em->orig_block_len;
1070                 } else {
1071                         bytenr += btrfs_file_extent_offset(leaf, fi);
1072                         em->block_start = bytenr;
1073                         em->block_len = em->len;
1074                         if (type == BTRFS_FILE_EXTENT_PREALLOC)
1075                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
1076                 }
1077         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1078                 em->block_start = EXTENT_MAP_INLINE;
1079                 em->start = extent_start;
1080                 em->len = extent_end - extent_start;
1081                 /*
1082                  * Initialize orig_start and block_len with the same values
1083                  * as in inode.c:btrfs_get_extent().
1084                  */
1085                 em->orig_start = EXTENT_MAP_HOLE;
1086                 em->block_len = (u64)-1;
1087                 if (!new_inline && compress_type != BTRFS_COMPRESS_NONE) {
1088                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1089                         em->compress_type = compress_type;
1090                 }
1091         } else {
1092                 btrfs_err(fs_info,
1093                           "unknown file extent item type %d, inode %llu, offset %llu, "
1094                           "root %llu", type, btrfs_ino(inode), extent_start,
1095                           root->root_key.objectid);
1096         }
1097 }
1098
1099 /*
1100  * Returns the end offset (non inclusive) of the file extent item the given path
1101  * points to. If it points to an inline extent, the returned offset is rounded
1102  * up to the sector size.
1103  */
1104 u64 btrfs_file_extent_end(const struct btrfs_path *path)
1105 {
1106         const struct extent_buffer *leaf = path->nodes[0];
1107         const int slot = path->slots[0];
1108         struct btrfs_file_extent_item *fi;
1109         struct btrfs_key key;
1110         u64 end;
1111
1112         btrfs_item_key_to_cpu(leaf, &key, slot);
1113         ASSERT(key.type == BTRFS_EXTENT_DATA_KEY);
1114         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1115
1116         if (btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_INLINE) {
1117                 end = btrfs_file_extent_ram_bytes(leaf, fi);
1118                 end = ALIGN(key.offset + end, leaf->fs_info->sectorsize);
1119         } else {
1120                 end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1121         }
1122
1123         return end;
1124 }