Loading...
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
26static inline u32 max_ordered_sum_bytes(struct btrfs_fs_info *fs_info,
27 u16 csum_size)
28{
29 u32 ncsums = (PAGE_SIZE - sizeof(struct btrfs_ordered_sum)) / csum_size;
30
31 return ncsums * fs_info->sectorsize;
32}
33
34int btrfs_insert_file_extent(struct btrfs_trans_handle *trans,
35 struct btrfs_root *root,
36 u64 objectid, u64 pos,
37 u64 disk_offset, u64 disk_num_bytes,
38 u64 num_bytes, u64 offset, u64 ram_bytes,
39 u8 compression, u8 encryption, u16 other_encoding)
40{
41 int ret = 0;
42 struct btrfs_file_extent_item *item;
43 struct btrfs_key file_key;
44 struct btrfs_path *path;
45 struct extent_buffer *leaf;
46
47 path = btrfs_alloc_path();
48 if (!path)
49 return -ENOMEM;
50 file_key.objectid = objectid;
51 file_key.offset = pos;
52 file_key.type = BTRFS_EXTENT_DATA_KEY;
53
54 path->leave_spinning = 1;
55 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
56 sizeof(*item));
57 if (ret < 0)
58 goto out;
59 BUG_ON(ret); /* Can't happen */
60 leaf = path->nodes[0];
61 item = btrfs_item_ptr(leaf, path->slots[0],
62 struct btrfs_file_extent_item);
63 btrfs_set_file_extent_disk_bytenr(leaf, item, disk_offset);
64 btrfs_set_file_extent_disk_num_bytes(leaf, item, disk_num_bytes);
65 btrfs_set_file_extent_offset(leaf, item, offset);
66 btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
67 btrfs_set_file_extent_ram_bytes(leaf, item, ram_bytes);
68 btrfs_set_file_extent_generation(leaf, item, trans->transid);
69 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
70 btrfs_set_file_extent_compression(leaf, item, compression);
71 btrfs_set_file_extent_encryption(leaf, item, encryption);
72 btrfs_set_file_extent_other_encoding(leaf, item, other_encoding);
73
74 btrfs_mark_buffer_dirty(leaf);
75out:
76 btrfs_free_path(path);
77 return ret;
78}
79
80static struct btrfs_csum_item *
81btrfs_lookup_csum(struct btrfs_trans_handle *trans,
82 struct btrfs_root *root,
83 struct btrfs_path *path,
84 u64 bytenr, int cow)
85{
86 struct btrfs_fs_info *fs_info = root->fs_info;
87 int ret;
88 struct btrfs_key file_key;
89 struct btrfs_key found_key;
90 struct btrfs_csum_item *item;
91 struct extent_buffer *leaf;
92 u64 csum_offset = 0;
93 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
94 int csums_in_item;
95
96 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
97 file_key.offset = bytenr;
98 file_key.type = BTRFS_EXTENT_CSUM_KEY;
99 ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
100 if (ret < 0)
101 goto fail;
102 leaf = path->nodes[0];
103 if (ret > 0) {
104 ret = 1;
105 if (path->slots[0] == 0)
106 goto fail;
107 path->slots[0]--;
108 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
109 if (found_key.type != BTRFS_EXTENT_CSUM_KEY)
110 goto fail;
111
112 csum_offset = (bytenr - found_key.offset) >>
113 fs_info->sb->s_blocksize_bits;
114 csums_in_item = btrfs_item_size_nr(leaf, path->slots[0]);
115 csums_in_item /= csum_size;
116
117 if (csum_offset == csums_in_item) {
118 ret = -EFBIG;
119 goto fail;
120 } else if (csum_offset > csums_in_item) {
121 goto fail;
122 }
123 }
124 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
125 item = (struct btrfs_csum_item *)((unsigned char *)item +
126 csum_offset * csum_size);
127 return item;
128fail:
129 if (ret > 0)
130 ret = -ENOENT;
131 return ERR_PTR(ret);
132}
133
134int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
135 struct btrfs_root *root,
136 struct btrfs_path *path, u64 objectid,
137 u64 offset, int mod)
138{
139 int ret;
140 struct btrfs_key file_key;
141 int ins_len = mod < 0 ? -1 : 0;
142 int cow = mod != 0;
143
144 file_key.objectid = objectid;
145 file_key.offset = offset;
146 file_key.type = BTRFS_EXTENT_DATA_KEY;
147 ret = btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
148 return ret;
149}
150
151static blk_status_t __btrfs_lookup_bio_sums(struct inode *inode, struct bio *bio,
152 u64 logical_offset, u8 *dst, int dio)
153{
154 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
155 struct bio_vec bvec;
156 struct bvec_iter iter;
157 struct btrfs_io_bio *btrfs_bio = btrfs_io_bio(bio);
158 struct btrfs_csum_item *item = NULL;
159 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
160 struct btrfs_path *path;
161 u8 *csum;
162 u64 offset = 0;
163 u64 item_start_offset = 0;
164 u64 item_last_offset = 0;
165 u64 disk_bytenr;
166 u64 page_bytes_left;
167 u32 diff;
168 int nblocks;
169 int count = 0;
170 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
171
172 path = btrfs_alloc_path();
173 if (!path)
174 return BLK_STS_RESOURCE;
175
176 nblocks = bio->bi_iter.bi_size >> inode->i_sb->s_blocksize_bits;
177 if (!dst) {
178 if (nblocks * csum_size > BTRFS_BIO_INLINE_CSUM_SIZE) {
179 btrfs_bio->csum = kmalloc_array(nblocks, csum_size,
180 GFP_NOFS);
181 if (!btrfs_bio->csum) {
182 btrfs_free_path(path);
183 return BLK_STS_RESOURCE;
184 }
185 } else {
186 btrfs_bio->csum = btrfs_bio->csum_inline;
187 }
188 csum = btrfs_bio->csum;
189 } else {
190 csum = dst;
191 }
192
193 if (bio->bi_iter.bi_size > PAGE_SIZE * 8)
194 path->reada = READA_FORWARD;
195
196 /*
197 * the free space stuff is only read when it hasn't been
198 * updated in the current transaction. So, we can safely
199 * read from the commit root and sidestep a nasty deadlock
200 * between reading the free space cache and updating the csum tree.
201 */
202 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
203 path->search_commit_root = 1;
204 path->skip_locking = 1;
205 }
206
207 disk_bytenr = (u64)bio->bi_iter.bi_sector << 9;
208 if (dio)
209 offset = logical_offset;
210
211 bio_for_each_segment(bvec, bio, iter) {
212 page_bytes_left = bvec.bv_len;
213 if (count)
214 goto next;
215
216 if (!dio)
217 offset = page_offset(bvec.bv_page) + bvec.bv_offset;
218 count = btrfs_find_ordered_sum(inode, offset, disk_bytenr,
219 csum, nblocks);
220 if (count)
221 goto found;
222
223 if (!item || disk_bytenr < item_start_offset ||
224 disk_bytenr >= item_last_offset) {
225 struct btrfs_key found_key;
226 u32 item_size;
227
228 if (item)
229 btrfs_release_path(path);
230 item = btrfs_lookup_csum(NULL, fs_info->csum_root,
231 path, disk_bytenr, 0);
232 if (IS_ERR(item)) {
233 count = 1;
234 memset(csum, 0, csum_size);
235 if (BTRFS_I(inode)->root->root_key.objectid ==
236 BTRFS_DATA_RELOC_TREE_OBJECTID) {
237 set_extent_bits(io_tree, offset,
238 offset + fs_info->sectorsize - 1,
239 EXTENT_NODATASUM);
240 } else {
241 btrfs_info_rl(fs_info,
242 "no csum found for inode %llu start %llu",
243 btrfs_ino(BTRFS_I(inode)), offset);
244 }
245 item = NULL;
246 btrfs_release_path(path);
247 goto found;
248 }
249 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
250 path->slots[0]);
251
252 item_start_offset = found_key.offset;
253 item_size = btrfs_item_size_nr(path->nodes[0],
254 path->slots[0]);
255 item_last_offset = item_start_offset +
256 (item_size / csum_size) *
257 fs_info->sectorsize;
258 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
259 struct btrfs_csum_item);
260 }
261 /*
262 * this byte range must be able to fit inside
263 * a single leaf so it will also fit inside a u32
264 */
265 diff = disk_bytenr - item_start_offset;
266 diff = diff / fs_info->sectorsize;
267 diff = diff * csum_size;
268 count = min_t(int, nblocks, (item_last_offset - disk_bytenr) >>
269 inode->i_sb->s_blocksize_bits);
270 read_extent_buffer(path->nodes[0], csum,
271 ((unsigned long)item) + diff,
272 csum_size * count);
273found:
274 csum += count * csum_size;
275 nblocks -= count;
276next:
277 while (count--) {
278 disk_bytenr += fs_info->sectorsize;
279 offset += fs_info->sectorsize;
280 page_bytes_left -= fs_info->sectorsize;
281 if (!page_bytes_left)
282 break; /* move to next bio */
283 }
284 }
285
286 WARN_ON_ONCE(count);
287 btrfs_free_path(path);
288 return 0;
289}
290
291blk_status_t btrfs_lookup_bio_sums(struct inode *inode, struct bio *bio,
292 u8 *dst)
293{
294 return __btrfs_lookup_bio_sums(inode, bio, 0, dst, 0);
295}
296
297blk_status_t btrfs_lookup_bio_sums_dio(struct inode *inode, struct bio *bio, u64 offset)
298{
299 return __btrfs_lookup_bio_sums(inode, bio, offset, NULL, 1);
300}
301
302int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
303 struct list_head *list, int search_commit)
304{
305 struct btrfs_fs_info *fs_info = root->fs_info;
306 struct btrfs_key key;
307 struct btrfs_path *path;
308 struct extent_buffer *leaf;
309 struct btrfs_ordered_sum *sums;
310 struct btrfs_csum_item *item;
311 LIST_HEAD(tmplist);
312 unsigned long offset;
313 int ret;
314 size_t size;
315 u64 csum_end;
316 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
317
318 ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
319 IS_ALIGNED(end + 1, fs_info->sectorsize));
320
321 path = btrfs_alloc_path();
322 if (!path)
323 return -ENOMEM;
324
325 if (search_commit) {
326 path->skip_locking = 1;
327 path->reada = READA_FORWARD;
328 path->search_commit_root = 1;
329 }
330
331 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
332 key.offset = start;
333 key.type = BTRFS_EXTENT_CSUM_KEY;
334
335 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
336 if (ret < 0)
337 goto fail;
338 if (ret > 0 && path->slots[0] > 0) {
339 leaf = path->nodes[0];
340 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
341 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
342 key.type == BTRFS_EXTENT_CSUM_KEY) {
343 offset = (start - key.offset) >>
344 fs_info->sb->s_blocksize_bits;
345 if (offset * csum_size <
346 btrfs_item_size_nr(leaf, path->slots[0] - 1))
347 path->slots[0]--;
348 }
349 }
350
351 while (start <= end) {
352 leaf = path->nodes[0];
353 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
354 ret = btrfs_next_leaf(root, path);
355 if (ret < 0)
356 goto fail;
357 if (ret > 0)
358 break;
359 leaf = path->nodes[0];
360 }
361
362 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
363 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
364 key.type != BTRFS_EXTENT_CSUM_KEY ||
365 key.offset > end)
366 break;
367
368 if (key.offset > start)
369 start = key.offset;
370
371 size = btrfs_item_size_nr(leaf, path->slots[0]);
372 csum_end = key.offset + (size / csum_size) * fs_info->sectorsize;
373 if (csum_end <= start) {
374 path->slots[0]++;
375 continue;
376 }
377
378 csum_end = min(csum_end, end + 1);
379 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
380 struct btrfs_csum_item);
381 while (start < csum_end) {
382 size = min_t(size_t, csum_end - start,
383 max_ordered_sum_bytes(fs_info, csum_size));
384 sums = kzalloc(btrfs_ordered_sum_size(fs_info, size),
385 GFP_NOFS);
386 if (!sums) {
387 ret = -ENOMEM;
388 goto fail;
389 }
390
391 sums->bytenr = start;
392 sums->len = (int)size;
393
394 offset = (start - key.offset) >>
395 fs_info->sb->s_blocksize_bits;
396 offset *= csum_size;
397 size >>= fs_info->sb->s_blocksize_bits;
398
399 read_extent_buffer(path->nodes[0],
400 sums->sums,
401 ((unsigned long)item) + offset,
402 csum_size * size);
403
404 start += fs_info->sectorsize * size;
405 list_add_tail(&sums->list, &tmplist);
406 }
407 path->slots[0]++;
408 }
409 ret = 0;
410fail:
411 while (ret < 0 && !list_empty(&tmplist)) {
412 sums = list_entry(tmplist.next, struct btrfs_ordered_sum, list);
413 list_del(&sums->list);
414 kfree(sums);
415 }
416 list_splice_tail(&tmplist, list);
417
418 btrfs_free_path(path);
419 return ret;
420}
421
422/*
423 * btrfs_csum_one_bio - Calculates checksums of the data contained inside a bio
424 * @inode: Owner of the data inside the bio
425 * @bio: Contains the data to be checksummed
426 * @file_start: offset in file this bio begins to describe
427 * @contig: Boolean. If true/1 means all bio vecs in this bio are
428 * contiguous and they begin at @file_start in the file. False/0
429 * means this bio can contains potentially discontigous bio vecs
430 * so the logical offset of each should be calculated separately.
431 */
432blk_status_t btrfs_csum_one_bio(struct inode *inode, struct bio *bio,
433 u64 file_start, int contig)
434{
435 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
436 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
437 struct btrfs_ordered_sum *sums;
438 struct btrfs_ordered_extent *ordered = NULL;
439 char *data;
440 struct bvec_iter iter;
441 struct bio_vec bvec;
442 int index;
443 int nr_sectors;
444 unsigned long total_bytes = 0;
445 unsigned long this_sum_bytes = 0;
446 int i;
447 u64 offset;
448 unsigned nofs_flag;
449 const u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
450
451 nofs_flag = memalloc_nofs_save();
452 sums = kvzalloc(btrfs_ordered_sum_size(fs_info, bio->bi_iter.bi_size),
453 GFP_KERNEL);
454 memalloc_nofs_restore(nofs_flag);
455
456 if (!sums)
457 return BLK_STS_RESOURCE;
458
459 sums->len = bio->bi_iter.bi_size;
460 INIT_LIST_HEAD(&sums->list);
461
462 if (contig)
463 offset = file_start;
464 else
465 offset = 0; /* shut up gcc */
466
467 sums->bytenr = (u64)bio->bi_iter.bi_sector << 9;
468 index = 0;
469
470 shash->tfm = fs_info->csum_shash;
471
472 bio_for_each_segment(bvec, bio, iter) {
473 if (!contig)
474 offset = page_offset(bvec.bv_page) + bvec.bv_offset;
475
476 if (!ordered) {
477 ordered = btrfs_lookup_ordered_extent(inode, offset);
478 BUG_ON(!ordered); /* Logic error */
479 }
480
481 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info,
482 bvec.bv_len + fs_info->sectorsize
483 - 1);
484
485 for (i = 0; i < nr_sectors; i++) {
486 if (offset >= ordered->file_offset + ordered->len ||
487 offset < ordered->file_offset) {
488 unsigned long bytes_left;
489
490 sums->len = this_sum_bytes;
491 this_sum_bytes = 0;
492 btrfs_add_ordered_sum(ordered, sums);
493 btrfs_put_ordered_extent(ordered);
494
495 bytes_left = bio->bi_iter.bi_size - total_bytes;
496
497 nofs_flag = memalloc_nofs_save();
498 sums = kvzalloc(btrfs_ordered_sum_size(fs_info,
499 bytes_left), GFP_KERNEL);
500 memalloc_nofs_restore(nofs_flag);
501 BUG_ON(!sums); /* -ENOMEM */
502 sums->len = bytes_left;
503 ordered = btrfs_lookup_ordered_extent(inode,
504 offset);
505 ASSERT(ordered); /* Logic error */
506 sums->bytenr = ((u64)bio->bi_iter.bi_sector << 9)
507 + total_bytes;
508 index = 0;
509 }
510
511 crypto_shash_init(shash);
512 data = kmap_atomic(bvec.bv_page);
513 crypto_shash_update(shash, data + bvec.bv_offset
514 + (i * fs_info->sectorsize),
515 fs_info->sectorsize);
516 kunmap_atomic(data);
517 crypto_shash_final(shash, (char *)(sums->sums + index));
518 index += csum_size;
519 offset += fs_info->sectorsize;
520 this_sum_bytes += fs_info->sectorsize;
521 total_bytes += fs_info->sectorsize;
522 }
523
524 }
525 this_sum_bytes = 0;
526 btrfs_add_ordered_sum(ordered, sums);
527 btrfs_put_ordered_extent(ordered);
528 return 0;
529}
530
531/*
532 * helper function for csum removal, this expects the
533 * key to describe the csum pointed to by the path, and it expects
534 * the csum to overlap the range [bytenr, len]
535 *
536 * The csum should not be entirely contained in the range and the
537 * range should not be entirely contained in the csum.
538 *
539 * This calls btrfs_truncate_item with the correct args based on the
540 * overlap, and fixes up the key as required.
541 */
542static noinline void truncate_one_csum(struct btrfs_fs_info *fs_info,
543 struct btrfs_path *path,
544 struct btrfs_key *key,
545 u64 bytenr, u64 len)
546{
547 struct extent_buffer *leaf;
548 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
549 u64 csum_end;
550 u64 end_byte = bytenr + len;
551 u32 blocksize_bits = fs_info->sb->s_blocksize_bits;
552
553 leaf = path->nodes[0];
554 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
555 csum_end <<= fs_info->sb->s_blocksize_bits;
556 csum_end += key->offset;
557
558 if (key->offset < bytenr && csum_end <= end_byte) {
559 /*
560 * [ bytenr - len ]
561 * [ ]
562 * [csum ]
563 * A simple truncate off the end of the item
564 */
565 u32 new_size = (bytenr - key->offset) >> blocksize_bits;
566 new_size *= csum_size;
567 btrfs_truncate_item(path, new_size, 1);
568 } else if (key->offset >= bytenr && csum_end > end_byte &&
569 end_byte > key->offset) {
570 /*
571 * [ bytenr - len ]
572 * [ ]
573 * [csum ]
574 * we need to truncate from the beginning of the csum
575 */
576 u32 new_size = (csum_end - end_byte) >> blocksize_bits;
577 new_size *= csum_size;
578
579 btrfs_truncate_item(path, new_size, 0);
580
581 key->offset = end_byte;
582 btrfs_set_item_key_safe(fs_info, path, key);
583 } else {
584 BUG();
585 }
586}
587
588/*
589 * deletes the csum items from the csum tree for a given
590 * range of bytes.
591 */
592int btrfs_del_csums(struct btrfs_trans_handle *trans,
593 struct btrfs_fs_info *fs_info, u64 bytenr, u64 len)
594{
595 struct btrfs_root *root = fs_info->csum_root;
596 struct btrfs_path *path;
597 struct btrfs_key key;
598 u64 end_byte = bytenr + len;
599 u64 csum_end;
600 struct extent_buffer *leaf;
601 int ret;
602 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
603 int blocksize_bits = fs_info->sb->s_blocksize_bits;
604
605 path = btrfs_alloc_path();
606 if (!path)
607 return -ENOMEM;
608
609 while (1) {
610 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
611 key.offset = end_byte - 1;
612 key.type = BTRFS_EXTENT_CSUM_KEY;
613
614 path->leave_spinning = 1;
615 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
616 if (ret > 0) {
617 if (path->slots[0] == 0)
618 break;
619 path->slots[0]--;
620 } else if (ret < 0) {
621 break;
622 }
623
624 leaf = path->nodes[0];
625 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
626
627 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
628 key.type != BTRFS_EXTENT_CSUM_KEY) {
629 break;
630 }
631
632 if (key.offset >= end_byte)
633 break;
634
635 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
636 csum_end <<= blocksize_bits;
637 csum_end += key.offset;
638
639 /* this csum ends before we start, we're done */
640 if (csum_end <= bytenr)
641 break;
642
643 /* delete the entire item, it is inside our range */
644 if (key.offset >= bytenr && csum_end <= end_byte) {
645 int del_nr = 1;
646
647 /*
648 * Check how many csum items preceding this one in this
649 * leaf correspond to our range and then delete them all
650 * at once.
651 */
652 if (key.offset > bytenr && path->slots[0] > 0) {
653 int slot = path->slots[0] - 1;
654
655 while (slot >= 0) {
656 struct btrfs_key pk;
657
658 btrfs_item_key_to_cpu(leaf, &pk, slot);
659 if (pk.offset < bytenr ||
660 pk.type != BTRFS_EXTENT_CSUM_KEY ||
661 pk.objectid !=
662 BTRFS_EXTENT_CSUM_OBJECTID)
663 break;
664 path->slots[0] = slot;
665 del_nr++;
666 key.offset = pk.offset;
667 slot--;
668 }
669 }
670 ret = btrfs_del_items(trans, root, path,
671 path->slots[0], del_nr);
672 if (ret)
673 goto out;
674 if (key.offset == bytenr)
675 break;
676 } else if (key.offset < bytenr && csum_end > end_byte) {
677 unsigned long offset;
678 unsigned long shift_len;
679 unsigned long item_offset;
680 /*
681 * [ bytenr - len ]
682 * [csum ]
683 *
684 * Our bytes are in the middle of the csum,
685 * we need to split this item and insert a new one.
686 *
687 * But we can't drop the path because the
688 * csum could change, get removed, extended etc.
689 *
690 * The trick here is the max size of a csum item leaves
691 * enough room in the tree block for a single
692 * item header. So, we split the item in place,
693 * adding a new header pointing to the existing
694 * bytes. Then we loop around again and we have
695 * a nicely formed csum item that we can neatly
696 * truncate.
697 */
698 offset = (bytenr - key.offset) >> blocksize_bits;
699 offset *= csum_size;
700
701 shift_len = (len >> blocksize_bits) * csum_size;
702
703 item_offset = btrfs_item_ptr_offset(leaf,
704 path->slots[0]);
705
706 memzero_extent_buffer(leaf, item_offset + offset,
707 shift_len);
708 key.offset = bytenr;
709
710 /*
711 * btrfs_split_item returns -EAGAIN when the
712 * item changed size or key
713 */
714 ret = btrfs_split_item(trans, root, path, &key, offset);
715 if (ret && ret != -EAGAIN) {
716 btrfs_abort_transaction(trans, ret);
717 goto out;
718 }
719
720 key.offset = end_byte - 1;
721 } else {
722 truncate_one_csum(fs_info, path, &key, bytenr, len);
723 if (key.offset < bytenr)
724 break;
725 }
726 btrfs_release_path(path);
727 }
728 ret = 0;
729out:
730 btrfs_free_path(path);
731 return ret;
732}
733
734int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
735 struct btrfs_root *root,
736 struct btrfs_ordered_sum *sums)
737{
738 struct btrfs_fs_info *fs_info = root->fs_info;
739 struct btrfs_key file_key;
740 struct btrfs_key found_key;
741 struct btrfs_path *path;
742 struct btrfs_csum_item *item;
743 struct btrfs_csum_item *item_end;
744 struct extent_buffer *leaf = NULL;
745 u64 next_offset;
746 u64 total_bytes = 0;
747 u64 csum_offset;
748 u64 bytenr;
749 u32 nritems;
750 u32 ins_size;
751 int index = 0;
752 int found_next;
753 int ret;
754 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
755
756 path = btrfs_alloc_path();
757 if (!path)
758 return -ENOMEM;
759again:
760 next_offset = (u64)-1;
761 found_next = 0;
762 bytenr = sums->bytenr + total_bytes;
763 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
764 file_key.offset = bytenr;
765 file_key.type = BTRFS_EXTENT_CSUM_KEY;
766
767 item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
768 if (!IS_ERR(item)) {
769 ret = 0;
770 leaf = path->nodes[0];
771 item_end = btrfs_item_ptr(leaf, path->slots[0],
772 struct btrfs_csum_item);
773 item_end = (struct btrfs_csum_item *)((char *)item_end +
774 btrfs_item_size_nr(leaf, path->slots[0]));
775 goto found;
776 }
777 ret = PTR_ERR(item);
778 if (ret != -EFBIG && ret != -ENOENT)
779 goto fail_unlock;
780
781 if (ret == -EFBIG) {
782 u32 item_size;
783 /* we found one, but it isn't big enough yet */
784 leaf = path->nodes[0];
785 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
786 if ((item_size / csum_size) >=
787 MAX_CSUM_ITEMS(fs_info, csum_size)) {
788 /* already at max size, make a new one */
789 goto insert;
790 }
791 } else {
792 int slot = path->slots[0] + 1;
793 /* we didn't find a csum item, insert one */
794 nritems = btrfs_header_nritems(path->nodes[0]);
795 if (!nritems || (path->slots[0] >= nritems - 1)) {
796 ret = btrfs_next_leaf(root, path);
797 if (ret == 1)
798 found_next = 1;
799 if (ret != 0)
800 goto insert;
801 slot = path->slots[0];
802 }
803 btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
804 if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
805 found_key.type != BTRFS_EXTENT_CSUM_KEY) {
806 found_next = 1;
807 goto insert;
808 }
809 next_offset = found_key.offset;
810 found_next = 1;
811 goto insert;
812 }
813
814 /*
815 * at this point, we know the tree has an item, but it isn't big
816 * enough yet to put our csum in. Grow it
817 */
818 btrfs_release_path(path);
819 ret = btrfs_search_slot(trans, root, &file_key, path,
820 csum_size, 1);
821 if (ret < 0)
822 goto fail_unlock;
823
824 if (ret > 0) {
825 if (path->slots[0] == 0)
826 goto insert;
827 path->slots[0]--;
828 }
829
830 leaf = path->nodes[0];
831 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
832 csum_offset = (bytenr - found_key.offset) >>
833 fs_info->sb->s_blocksize_bits;
834
835 if (found_key.type != BTRFS_EXTENT_CSUM_KEY ||
836 found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
837 csum_offset >= MAX_CSUM_ITEMS(fs_info, csum_size)) {
838 goto insert;
839 }
840
841 if (csum_offset == btrfs_item_size_nr(leaf, path->slots[0]) /
842 csum_size) {
843 int extend_nr;
844 u64 tmp;
845 u32 diff;
846 u32 free_space;
847
848 if (btrfs_leaf_free_space(leaf) <
849 sizeof(struct btrfs_item) + csum_size * 2)
850 goto insert;
851
852 free_space = btrfs_leaf_free_space(leaf) -
853 sizeof(struct btrfs_item) - csum_size;
854 tmp = sums->len - total_bytes;
855 tmp >>= fs_info->sb->s_blocksize_bits;
856 WARN_ON(tmp < 1);
857
858 extend_nr = max_t(int, 1, (int)tmp);
859 diff = (csum_offset + extend_nr) * csum_size;
860 diff = min(diff,
861 MAX_CSUM_ITEMS(fs_info, csum_size) * csum_size);
862
863 diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
864 diff = min(free_space, diff);
865 diff /= csum_size;
866 diff *= csum_size;
867
868 btrfs_extend_item(path, diff);
869 ret = 0;
870 goto csum;
871 }
872
873insert:
874 btrfs_release_path(path);
875 csum_offset = 0;
876 if (found_next) {
877 u64 tmp;
878
879 tmp = sums->len - total_bytes;
880 tmp >>= fs_info->sb->s_blocksize_bits;
881 tmp = min(tmp, (next_offset - file_key.offset) >>
882 fs_info->sb->s_blocksize_bits);
883
884 tmp = max_t(u64, 1, tmp);
885 tmp = min_t(u64, tmp, MAX_CSUM_ITEMS(fs_info, csum_size));
886 ins_size = csum_size * tmp;
887 } else {
888 ins_size = csum_size;
889 }
890 path->leave_spinning = 1;
891 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
892 ins_size);
893 path->leave_spinning = 0;
894 if (ret < 0)
895 goto fail_unlock;
896 if (WARN_ON(ret != 0))
897 goto fail_unlock;
898 leaf = path->nodes[0];
899csum:
900 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
901 item_end = (struct btrfs_csum_item *)((unsigned char *)item +
902 btrfs_item_size_nr(leaf, path->slots[0]));
903 item = (struct btrfs_csum_item *)((unsigned char *)item +
904 csum_offset * csum_size);
905found:
906 ins_size = (u32)(sums->len - total_bytes) >>
907 fs_info->sb->s_blocksize_bits;
908 ins_size *= csum_size;
909 ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item,
910 ins_size);
911 write_extent_buffer(leaf, sums->sums + index, (unsigned long)item,
912 ins_size);
913
914 index += ins_size;
915 ins_size /= csum_size;
916 total_bytes += ins_size * fs_info->sectorsize;
917
918 btrfs_mark_buffer_dirty(path->nodes[0]);
919 if (total_bytes < sums->len) {
920 btrfs_release_path(path);
921 cond_resched();
922 goto again;
923 }
924out:
925 btrfs_free_path(path);
926 return ret;
927
928fail_unlock:
929 goto out;
930}
931
932void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode,
933 const struct btrfs_path *path,
934 struct btrfs_file_extent_item *fi,
935 const bool new_inline,
936 struct extent_map *em)
937{
938 struct btrfs_fs_info *fs_info = inode->root->fs_info;
939 struct btrfs_root *root = inode->root;
940 struct extent_buffer *leaf = path->nodes[0];
941 const int slot = path->slots[0];
942 struct btrfs_key key;
943 u64 extent_start, extent_end;
944 u64 bytenr;
945 u8 type = btrfs_file_extent_type(leaf, fi);
946 int compress_type = btrfs_file_extent_compression(leaf, fi);
947
948 em->bdev = fs_info->fs_devices->latest_bdev;
949 btrfs_item_key_to_cpu(leaf, &key, slot);
950 extent_start = key.offset;
951
952 if (type == BTRFS_FILE_EXTENT_REG ||
953 type == BTRFS_FILE_EXTENT_PREALLOC) {
954 extent_end = extent_start +
955 btrfs_file_extent_num_bytes(leaf, fi);
956 } else if (type == BTRFS_FILE_EXTENT_INLINE) {
957 size_t size;
958 size = btrfs_file_extent_ram_bytes(leaf, fi);
959 extent_end = ALIGN(extent_start + size,
960 fs_info->sectorsize);
961 }
962
963 em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
964 if (type == BTRFS_FILE_EXTENT_REG ||
965 type == BTRFS_FILE_EXTENT_PREALLOC) {
966 em->start = extent_start;
967 em->len = extent_end - extent_start;
968 em->orig_start = extent_start -
969 btrfs_file_extent_offset(leaf, fi);
970 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
971 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
972 if (bytenr == 0) {
973 em->block_start = EXTENT_MAP_HOLE;
974 return;
975 }
976 if (compress_type != BTRFS_COMPRESS_NONE) {
977 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
978 em->compress_type = compress_type;
979 em->block_start = bytenr;
980 em->block_len = em->orig_block_len;
981 } else {
982 bytenr += btrfs_file_extent_offset(leaf, fi);
983 em->block_start = bytenr;
984 em->block_len = em->len;
985 if (type == BTRFS_FILE_EXTENT_PREALLOC)
986 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
987 }
988 } else if (type == BTRFS_FILE_EXTENT_INLINE) {
989 em->block_start = EXTENT_MAP_INLINE;
990 em->start = extent_start;
991 em->len = extent_end - extent_start;
992 /*
993 * Initialize orig_start and block_len with the same values
994 * as in inode.c:btrfs_get_extent().
995 */
996 em->orig_start = EXTENT_MAP_HOLE;
997 em->block_len = (u64)-1;
998 if (!new_inline && compress_type != BTRFS_COMPRESS_NONE) {
999 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1000 em->compress_type = compress_type;
1001 }
1002 } else {
1003 btrfs_err(fs_info,
1004 "unknown file extent item type %d, inode %llu, offset %llu, "
1005 "root %llu", type, btrfs_ino(inode), extent_start,
1006 root->root_key.objectid);
1007 }
1008}
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 "misc.h"
13#include "ctree.h"
14#include "disk-io.h"
15#include "transaction.h"
16#include "volumes.h"
17#include "print-tree.h"
18#include "compression.h"
19
20#define __MAX_CSUM_ITEMS(r, size) ((unsigned long)(((BTRFS_LEAF_DATA_SIZE(r) - \
21 sizeof(struct btrfs_item) * 2) / \
22 size) - 1))
23
24#define MAX_CSUM_ITEMS(r, size) (min_t(u32, __MAX_CSUM_ITEMS(r, size), \
25 PAGE_SIZE))
26
27/**
28 * Set inode's size according to filesystem options
29 *
30 * @inode: inode we want to update the disk_i_size for
31 * @new_i_size: i_size we want to set to, 0 if we use i_size
32 *
33 * With NO_HOLES set this simply sets the disk_is_size to whatever i_size_read()
34 * returns as it is perfectly fine with a file that has holes without hole file
35 * extent items.
36 *
37 * However without NO_HOLES we need to only return the area that is contiguous
38 * from the 0 offset of the file. Otherwise we could end up adjust i_size up
39 * to an extent that has a gap in between.
40 *
41 * Finally new_i_size should only be set in the case of truncate where we're not
42 * ready to use i_size_read() as the limiter yet.
43 */
44void btrfs_inode_safe_disk_i_size_write(struct btrfs_inode *inode, u64 new_i_size)
45{
46 struct btrfs_fs_info *fs_info = inode->root->fs_info;
47 u64 start, end, i_size;
48 int ret;
49
50 i_size = new_i_size ?: i_size_read(&inode->vfs_inode);
51 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
52 inode->disk_i_size = i_size;
53 return;
54 }
55
56 spin_lock(&inode->lock);
57 ret = find_contiguous_extent_bit(&inode->file_extent_tree, 0, &start,
58 &end, EXTENT_DIRTY);
59 if (!ret && start == 0)
60 i_size = min(i_size, end + 1);
61 else
62 i_size = 0;
63 inode->disk_i_size = i_size;
64 spin_unlock(&inode->lock);
65}
66
67/**
68 * Mark range within a file as having a new extent inserted
69 *
70 * @inode: inode being modified
71 * @start: start file offset of the file extent we've inserted
72 * @len: logical length of the file extent item
73 *
74 * Call when we are inserting a new file extent where there was none before.
75 * Does not need to call this in the case where we're replacing an existing file
76 * extent, however if not sure it's fine to call this multiple times.
77 *
78 * The start and len must match the file extent item, so thus must be sectorsize
79 * aligned.
80 */
81int btrfs_inode_set_file_extent_range(struct btrfs_inode *inode, u64 start,
82 u64 len)
83{
84 if (len == 0)
85 return 0;
86
87 ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize));
88
89 if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES))
90 return 0;
91 return set_extent_bits(&inode->file_extent_tree, start, start + len - 1,
92 EXTENT_DIRTY);
93}
94
95/**
96 * Marks an inode range as not having a backing extent
97 *
98 * @inode: inode being modified
99 * @start: start file offset of the file extent we've inserted
100 * @len: logical length of the file extent item
101 *
102 * Called when we drop a file extent, for example when we truncate. Doesn't
103 * need to be called for cases where we're replacing a file extent, like when
104 * we've COWed a file extent.
105 *
106 * The start and len must match the file extent item, so thus must be sectorsize
107 * aligned.
108 */
109int btrfs_inode_clear_file_extent_range(struct btrfs_inode *inode, u64 start,
110 u64 len)
111{
112 if (len == 0)
113 return 0;
114
115 ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize) ||
116 len == (u64)-1);
117
118 if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES))
119 return 0;
120 return clear_extent_bit(&inode->file_extent_tree, start,
121 start + len - 1, EXTENT_DIRTY, 0, 0, NULL);
122}
123
124static inline u32 max_ordered_sum_bytes(struct btrfs_fs_info *fs_info,
125 u16 csum_size)
126{
127 u32 ncsums = (PAGE_SIZE - sizeof(struct btrfs_ordered_sum)) / csum_size;
128
129 return ncsums * fs_info->sectorsize;
130}
131
132int btrfs_insert_file_extent(struct btrfs_trans_handle *trans,
133 struct btrfs_root *root,
134 u64 objectid, u64 pos,
135 u64 disk_offset, u64 disk_num_bytes,
136 u64 num_bytes, u64 offset, u64 ram_bytes,
137 u8 compression, u8 encryption, u16 other_encoding)
138{
139 int ret = 0;
140 struct btrfs_file_extent_item *item;
141 struct btrfs_key file_key;
142 struct btrfs_path *path;
143 struct extent_buffer *leaf;
144
145 path = btrfs_alloc_path();
146 if (!path)
147 return -ENOMEM;
148 file_key.objectid = objectid;
149 file_key.offset = pos;
150 file_key.type = BTRFS_EXTENT_DATA_KEY;
151
152 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
153 sizeof(*item));
154 if (ret < 0)
155 goto out;
156 BUG_ON(ret); /* Can't happen */
157 leaf = path->nodes[0];
158 item = btrfs_item_ptr(leaf, path->slots[0],
159 struct btrfs_file_extent_item);
160 btrfs_set_file_extent_disk_bytenr(leaf, item, disk_offset);
161 btrfs_set_file_extent_disk_num_bytes(leaf, item, disk_num_bytes);
162 btrfs_set_file_extent_offset(leaf, item, offset);
163 btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
164 btrfs_set_file_extent_ram_bytes(leaf, item, ram_bytes);
165 btrfs_set_file_extent_generation(leaf, item, trans->transid);
166 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
167 btrfs_set_file_extent_compression(leaf, item, compression);
168 btrfs_set_file_extent_encryption(leaf, item, encryption);
169 btrfs_set_file_extent_other_encoding(leaf, item, other_encoding);
170
171 btrfs_mark_buffer_dirty(leaf);
172out:
173 btrfs_free_path(path);
174 return ret;
175}
176
177static struct btrfs_csum_item *
178btrfs_lookup_csum(struct btrfs_trans_handle *trans,
179 struct btrfs_root *root,
180 struct btrfs_path *path,
181 u64 bytenr, int cow)
182{
183 struct btrfs_fs_info *fs_info = root->fs_info;
184 int ret;
185 struct btrfs_key file_key;
186 struct btrfs_key found_key;
187 struct btrfs_csum_item *item;
188 struct extent_buffer *leaf;
189 u64 csum_offset = 0;
190 const u32 csum_size = fs_info->csum_size;
191 int csums_in_item;
192
193 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
194 file_key.offset = bytenr;
195 file_key.type = BTRFS_EXTENT_CSUM_KEY;
196 ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
197 if (ret < 0)
198 goto fail;
199 leaf = path->nodes[0];
200 if (ret > 0) {
201 ret = 1;
202 if (path->slots[0] == 0)
203 goto fail;
204 path->slots[0]--;
205 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
206 if (found_key.type != BTRFS_EXTENT_CSUM_KEY)
207 goto fail;
208
209 csum_offset = (bytenr - found_key.offset) >>
210 fs_info->sectorsize_bits;
211 csums_in_item = btrfs_item_size_nr(leaf, path->slots[0]);
212 csums_in_item /= csum_size;
213
214 if (csum_offset == csums_in_item) {
215 ret = -EFBIG;
216 goto fail;
217 } else if (csum_offset > csums_in_item) {
218 goto fail;
219 }
220 }
221 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
222 item = (struct btrfs_csum_item *)((unsigned char *)item +
223 csum_offset * csum_size);
224 return item;
225fail:
226 if (ret > 0)
227 ret = -ENOENT;
228 return ERR_PTR(ret);
229}
230
231int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
232 struct btrfs_root *root,
233 struct btrfs_path *path, u64 objectid,
234 u64 offset, int mod)
235{
236 int ret;
237 struct btrfs_key file_key;
238 int ins_len = mod < 0 ? -1 : 0;
239 int cow = mod != 0;
240
241 file_key.objectid = objectid;
242 file_key.offset = offset;
243 file_key.type = BTRFS_EXTENT_DATA_KEY;
244 ret = btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
245 return ret;
246}
247
248/*
249 * Find checksums for logical bytenr range [disk_bytenr, disk_bytenr + len) and
250 * estore the result to @dst.
251 *
252 * Return >0 for the number of sectors we found.
253 * Return 0 for the range [disk_bytenr, disk_bytenr + sectorsize) has no csum
254 * for it. Caller may want to try next sector until one range is hit.
255 * Return <0 for fatal error.
256 */
257static int search_csum_tree(struct btrfs_fs_info *fs_info,
258 struct btrfs_path *path, u64 disk_bytenr,
259 u64 len, u8 *dst)
260{
261 struct btrfs_csum_item *item = NULL;
262 struct btrfs_key key;
263 const u32 sectorsize = fs_info->sectorsize;
264 const u32 csum_size = fs_info->csum_size;
265 u32 itemsize;
266 int ret;
267 u64 csum_start;
268 u64 csum_len;
269
270 ASSERT(IS_ALIGNED(disk_bytenr, sectorsize) &&
271 IS_ALIGNED(len, sectorsize));
272
273 /* Check if the current csum item covers disk_bytenr */
274 if (path->nodes[0]) {
275 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
276 struct btrfs_csum_item);
277 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
278 itemsize = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
279
280 csum_start = key.offset;
281 csum_len = (itemsize / csum_size) * sectorsize;
282
283 if (in_range(disk_bytenr, csum_start, csum_len))
284 goto found;
285 }
286
287 /* Current item doesn't contain the desired range, search again */
288 btrfs_release_path(path);
289 item = btrfs_lookup_csum(NULL, fs_info->csum_root, path, disk_bytenr, 0);
290 if (IS_ERR(item)) {
291 ret = PTR_ERR(item);
292 goto out;
293 }
294 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
295 itemsize = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
296
297 csum_start = key.offset;
298 csum_len = (itemsize / csum_size) * sectorsize;
299 ASSERT(in_range(disk_bytenr, csum_start, csum_len));
300
301found:
302 ret = (min(csum_start + csum_len, disk_bytenr + len) -
303 disk_bytenr) >> fs_info->sectorsize_bits;
304 read_extent_buffer(path->nodes[0], dst, (unsigned long)item,
305 ret * csum_size);
306out:
307 if (ret == -ENOENT)
308 ret = 0;
309 return ret;
310}
311
312/*
313 * Locate the file_offset of @cur_disk_bytenr of a @bio.
314 *
315 * Bio of btrfs represents read range of
316 * [bi_sector << 9, bi_sector << 9 + bi_size).
317 * Knowing this, we can iterate through each bvec to locate the page belong to
318 * @cur_disk_bytenr and get the file offset.
319 *
320 * @inode is used to determine if the bvec page really belongs to @inode.
321 *
322 * Return 0 if we can't find the file offset
323 * Return >0 if we find the file offset and restore it to @file_offset_ret
324 */
325static int search_file_offset_in_bio(struct bio *bio, struct inode *inode,
326 u64 disk_bytenr, u64 *file_offset_ret)
327{
328 struct bvec_iter iter;
329 struct bio_vec bvec;
330 u64 cur = bio->bi_iter.bi_sector << SECTOR_SHIFT;
331 int ret = 0;
332
333 bio_for_each_segment(bvec, bio, iter) {
334 struct page *page = bvec.bv_page;
335
336 if (cur > disk_bytenr)
337 break;
338 if (cur + bvec.bv_len <= disk_bytenr) {
339 cur += bvec.bv_len;
340 continue;
341 }
342 ASSERT(in_range(disk_bytenr, cur, bvec.bv_len));
343 if (page->mapping && page->mapping->host &&
344 page->mapping->host == inode) {
345 ret = 1;
346 *file_offset_ret = page_offset(page) + bvec.bv_offset +
347 disk_bytenr - cur;
348 break;
349 }
350 }
351 return ret;
352}
353
354/**
355 * Lookup the checksum for the read bio in csum tree.
356 *
357 * @inode: inode that the bio is for.
358 * @bio: bio to look up.
359 * @dst: Buffer of size nblocks * btrfs_super_csum_size() used to return
360 * checksum (nblocks = bio->bi_iter.bi_size / fs_info->sectorsize). If
361 * NULL, the checksum buffer is allocated and returned in
362 * btrfs_io_bio(bio)->csum instead.
363 *
364 * Return: BLK_STS_RESOURCE if allocating memory fails, BLK_STS_OK otherwise.
365 */
366blk_status_t btrfs_lookup_bio_sums(struct inode *inode, struct bio *bio, u8 *dst)
367{
368 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
369 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
370 struct btrfs_path *path;
371 const u32 sectorsize = fs_info->sectorsize;
372 const u32 csum_size = fs_info->csum_size;
373 u32 orig_len = bio->bi_iter.bi_size;
374 u64 orig_disk_bytenr = bio->bi_iter.bi_sector << SECTOR_SHIFT;
375 u64 cur_disk_bytenr;
376 u8 *csum;
377 const unsigned int nblocks = orig_len >> fs_info->sectorsize_bits;
378 int count = 0;
379
380 if (!fs_info->csum_root || (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
381 return BLK_STS_OK;
382
383 /*
384 * This function is only called for read bio.
385 *
386 * This means two things:
387 * - All our csums should only be in csum tree
388 * No ordered extents csums, as ordered extents are only for write
389 * path.
390 * - No need to bother any other info from bvec
391 * Since we're looking up csums, the only important info is the
392 * disk_bytenr and the length, which can be extracted from bi_iter
393 * directly.
394 */
395 ASSERT(bio_op(bio) == REQ_OP_READ);
396 path = btrfs_alloc_path();
397 if (!path)
398 return BLK_STS_RESOURCE;
399
400 if (!dst) {
401 struct btrfs_io_bio *btrfs_bio = btrfs_io_bio(bio);
402
403 if (nblocks * csum_size > BTRFS_BIO_INLINE_CSUM_SIZE) {
404 btrfs_bio->csum = kmalloc_array(nblocks, csum_size,
405 GFP_NOFS);
406 if (!btrfs_bio->csum) {
407 btrfs_free_path(path);
408 return BLK_STS_RESOURCE;
409 }
410 } else {
411 btrfs_bio->csum = btrfs_bio->csum_inline;
412 }
413 csum = btrfs_bio->csum;
414 } else {
415 csum = dst;
416 }
417
418 /*
419 * If requested number of sectors is larger than one leaf can contain,
420 * kick the readahead for csum tree.
421 */
422 if (nblocks > fs_info->csums_per_leaf)
423 path->reada = READA_FORWARD;
424
425 /*
426 * the free space stuff is only read when it hasn't been
427 * updated in the current transaction. So, we can safely
428 * read from the commit root and sidestep a nasty deadlock
429 * between reading the free space cache and updating the csum tree.
430 */
431 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
432 path->search_commit_root = 1;
433 path->skip_locking = 1;
434 }
435
436 for (cur_disk_bytenr = orig_disk_bytenr;
437 cur_disk_bytenr < orig_disk_bytenr + orig_len;
438 cur_disk_bytenr += (count * sectorsize)) {
439 u64 search_len = orig_disk_bytenr + orig_len - cur_disk_bytenr;
440 unsigned int sector_offset;
441 u8 *csum_dst;
442
443 /*
444 * Although both cur_disk_bytenr and orig_disk_bytenr is u64,
445 * we're calculating the offset to the bio start.
446 *
447 * Bio size is limited to UINT_MAX, thus unsigned int is large
448 * enough to contain the raw result, not to mention the right
449 * shifted result.
450 */
451 ASSERT(cur_disk_bytenr - orig_disk_bytenr < UINT_MAX);
452 sector_offset = (cur_disk_bytenr - orig_disk_bytenr) >>
453 fs_info->sectorsize_bits;
454 csum_dst = csum + sector_offset * csum_size;
455
456 count = search_csum_tree(fs_info, path, cur_disk_bytenr,
457 search_len, csum_dst);
458 if (count <= 0) {
459 /*
460 * Either we hit a critical error or we didn't find
461 * the csum.
462 * Either way, we put zero into the csums dst, and skip
463 * to the next sector.
464 */
465 memset(csum_dst, 0, csum_size);
466 count = 1;
467
468 /*
469 * For data reloc inode, we need to mark the range
470 * NODATASUM so that balance won't report false csum
471 * error.
472 */
473 if (BTRFS_I(inode)->root->root_key.objectid ==
474 BTRFS_DATA_RELOC_TREE_OBJECTID) {
475 u64 file_offset;
476 int ret;
477
478 ret = search_file_offset_in_bio(bio, inode,
479 cur_disk_bytenr, &file_offset);
480 if (ret)
481 set_extent_bits(io_tree, file_offset,
482 file_offset + sectorsize - 1,
483 EXTENT_NODATASUM);
484 } else {
485 btrfs_warn_rl(fs_info,
486 "csum hole found for disk bytenr range [%llu, %llu)",
487 cur_disk_bytenr, cur_disk_bytenr + sectorsize);
488 }
489 }
490 }
491
492 btrfs_free_path(path);
493 return BLK_STS_OK;
494}
495
496int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
497 struct list_head *list, int search_commit)
498{
499 struct btrfs_fs_info *fs_info = root->fs_info;
500 struct btrfs_key key;
501 struct btrfs_path *path;
502 struct extent_buffer *leaf;
503 struct btrfs_ordered_sum *sums;
504 struct btrfs_csum_item *item;
505 LIST_HEAD(tmplist);
506 unsigned long offset;
507 int ret;
508 size_t size;
509 u64 csum_end;
510 const u32 csum_size = fs_info->csum_size;
511
512 ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
513 IS_ALIGNED(end + 1, fs_info->sectorsize));
514
515 path = btrfs_alloc_path();
516 if (!path)
517 return -ENOMEM;
518
519 if (search_commit) {
520 path->skip_locking = 1;
521 path->reada = READA_FORWARD;
522 path->search_commit_root = 1;
523 }
524
525 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
526 key.offset = start;
527 key.type = BTRFS_EXTENT_CSUM_KEY;
528
529 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
530 if (ret < 0)
531 goto fail;
532 if (ret > 0 && path->slots[0] > 0) {
533 leaf = path->nodes[0];
534 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
535 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
536 key.type == BTRFS_EXTENT_CSUM_KEY) {
537 offset = (start - key.offset) >> fs_info->sectorsize_bits;
538 if (offset * csum_size <
539 btrfs_item_size_nr(leaf, path->slots[0] - 1))
540 path->slots[0]--;
541 }
542 }
543
544 while (start <= end) {
545 leaf = path->nodes[0];
546 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
547 ret = btrfs_next_leaf(root, path);
548 if (ret < 0)
549 goto fail;
550 if (ret > 0)
551 break;
552 leaf = path->nodes[0];
553 }
554
555 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
556 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
557 key.type != BTRFS_EXTENT_CSUM_KEY ||
558 key.offset > end)
559 break;
560
561 if (key.offset > start)
562 start = key.offset;
563
564 size = btrfs_item_size_nr(leaf, path->slots[0]);
565 csum_end = key.offset + (size / csum_size) * fs_info->sectorsize;
566 if (csum_end <= start) {
567 path->slots[0]++;
568 continue;
569 }
570
571 csum_end = min(csum_end, end + 1);
572 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
573 struct btrfs_csum_item);
574 while (start < csum_end) {
575 size = min_t(size_t, csum_end - start,
576 max_ordered_sum_bytes(fs_info, csum_size));
577 sums = kzalloc(btrfs_ordered_sum_size(fs_info, size),
578 GFP_NOFS);
579 if (!sums) {
580 ret = -ENOMEM;
581 goto fail;
582 }
583
584 sums->bytenr = start;
585 sums->len = (int)size;
586
587 offset = (start - key.offset) >> fs_info->sectorsize_bits;
588 offset *= csum_size;
589 size >>= fs_info->sectorsize_bits;
590
591 read_extent_buffer(path->nodes[0],
592 sums->sums,
593 ((unsigned long)item) + offset,
594 csum_size * size);
595
596 start += fs_info->sectorsize * size;
597 list_add_tail(&sums->list, &tmplist);
598 }
599 path->slots[0]++;
600 }
601 ret = 0;
602fail:
603 while (ret < 0 && !list_empty(&tmplist)) {
604 sums = list_entry(tmplist.next, struct btrfs_ordered_sum, list);
605 list_del(&sums->list);
606 kfree(sums);
607 }
608 list_splice_tail(&tmplist, list);
609
610 btrfs_free_path(path);
611 return ret;
612}
613
614/*
615 * btrfs_csum_one_bio - Calculates checksums of the data contained inside a bio
616 * @inode: Owner of the data inside the bio
617 * @bio: Contains the data to be checksummed
618 * @file_start: offset in file this bio begins to describe
619 * @contig: Boolean. If true/1 means all bio vecs in this bio are
620 * contiguous and they begin at @file_start in the file. False/0
621 * means this bio can contain potentially discontiguous bio vecs
622 * so the logical offset of each should be calculated separately.
623 */
624blk_status_t btrfs_csum_one_bio(struct btrfs_inode *inode, struct bio *bio,
625 u64 file_start, int contig)
626{
627 struct btrfs_fs_info *fs_info = inode->root->fs_info;
628 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
629 struct btrfs_ordered_sum *sums;
630 struct btrfs_ordered_extent *ordered = NULL;
631 char *data;
632 struct bvec_iter iter;
633 struct bio_vec bvec;
634 int index;
635 int nr_sectors;
636 unsigned long total_bytes = 0;
637 unsigned long this_sum_bytes = 0;
638 int i;
639 u64 offset;
640 unsigned nofs_flag;
641
642 nofs_flag = memalloc_nofs_save();
643 sums = kvzalloc(btrfs_ordered_sum_size(fs_info, bio->bi_iter.bi_size),
644 GFP_KERNEL);
645 memalloc_nofs_restore(nofs_flag);
646
647 if (!sums)
648 return BLK_STS_RESOURCE;
649
650 sums->len = bio->bi_iter.bi_size;
651 INIT_LIST_HEAD(&sums->list);
652
653 if (contig)
654 offset = file_start;
655 else
656 offset = 0; /* shut up gcc */
657
658 sums->bytenr = bio->bi_iter.bi_sector << 9;
659 index = 0;
660
661 shash->tfm = fs_info->csum_shash;
662
663 bio_for_each_segment(bvec, bio, iter) {
664 if (!contig)
665 offset = page_offset(bvec.bv_page) + bvec.bv_offset;
666
667 if (!ordered) {
668 ordered = btrfs_lookup_ordered_extent(inode, offset);
669 /*
670 * The bio range is not covered by any ordered extent,
671 * must be a code logic error.
672 */
673 if (unlikely(!ordered)) {
674 WARN(1, KERN_WARNING
675 "no ordered extent for root %llu ino %llu offset %llu\n",
676 inode->root->root_key.objectid,
677 btrfs_ino(inode), offset);
678 kvfree(sums);
679 return BLK_STS_IOERR;
680 }
681 }
682
683 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info,
684 bvec.bv_len + fs_info->sectorsize
685 - 1);
686
687 for (i = 0; i < nr_sectors; i++) {
688 if (offset >= ordered->file_offset + ordered->num_bytes ||
689 offset < ordered->file_offset) {
690 unsigned long bytes_left;
691
692 sums->len = this_sum_bytes;
693 this_sum_bytes = 0;
694 btrfs_add_ordered_sum(ordered, sums);
695 btrfs_put_ordered_extent(ordered);
696
697 bytes_left = bio->bi_iter.bi_size - total_bytes;
698
699 nofs_flag = memalloc_nofs_save();
700 sums = kvzalloc(btrfs_ordered_sum_size(fs_info,
701 bytes_left), GFP_KERNEL);
702 memalloc_nofs_restore(nofs_flag);
703 BUG_ON(!sums); /* -ENOMEM */
704 sums->len = bytes_left;
705 ordered = btrfs_lookup_ordered_extent(inode,
706 offset);
707 ASSERT(ordered); /* Logic error */
708 sums->bytenr = (bio->bi_iter.bi_sector << 9)
709 + total_bytes;
710 index = 0;
711 }
712
713 data = kmap_atomic(bvec.bv_page);
714 crypto_shash_digest(shash, data + bvec.bv_offset
715 + (i * fs_info->sectorsize),
716 fs_info->sectorsize,
717 sums->sums + index);
718 kunmap_atomic(data);
719 index += fs_info->csum_size;
720 offset += fs_info->sectorsize;
721 this_sum_bytes += fs_info->sectorsize;
722 total_bytes += fs_info->sectorsize;
723 }
724
725 }
726 this_sum_bytes = 0;
727 btrfs_add_ordered_sum(ordered, sums);
728 btrfs_put_ordered_extent(ordered);
729 return 0;
730}
731
732/*
733 * helper function for csum removal, this expects the
734 * key to describe the csum pointed to by the path, and it expects
735 * the csum to overlap the range [bytenr, len]
736 *
737 * The csum should not be entirely contained in the range and the
738 * range should not be entirely contained in the csum.
739 *
740 * This calls btrfs_truncate_item with the correct args based on the
741 * overlap, and fixes up the key as required.
742 */
743static noinline void truncate_one_csum(struct btrfs_fs_info *fs_info,
744 struct btrfs_path *path,
745 struct btrfs_key *key,
746 u64 bytenr, u64 len)
747{
748 struct extent_buffer *leaf;
749 const u32 csum_size = fs_info->csum_size;
750 u64 csum_end;
751 u64 end_byte = bytenr + len;
752 u32 blocksize_bits = fs_info->sectorsize_bits;
753
754 leaf = path->nodes[0];
755 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
756 csum_end <<= blocksize_bits;
757 csum_end += key->offset;
758
759 if (key->offset < bytenr && csum_end <= end_byte) {
760 /*
761 * [ bytenr - len ]
762 * [ ]
763 * [csum ]
764 * A simple truncate off the end of the item
765 */
766 u32 new_size = (bytenr - key->offset) >> blocksize_bits;
767 new_size *= csum_size;
768 btrfs_truncate_item(path, new_size, 1);
769 } else if (key->offset >= bytenr && csum_end > end_byte &&
770 end_byte > key->offset) {
771 /*
772 * [ bytenr - len ]
773 * [ ]
774 * [csum ]
775 * we need to truncate from the beginning of the csum
776 */
777 u32 new_size = (csum_end - end_byte) >> blocksize_bits;
778 new_size *= csum_size;
779
780 btrfs_truncate_item(path, new_size, 0);
781
782 key->offset = end_byte;
783 btrfs_set_item_key_safe(fs_info, path, key);
784 } else {
785 BUG();
786 }
787}
788
789/*
790 * deletes the csum items from the csum tree for a given
791 * range of bytes.
792 */
793int btrfs_del_csums(struct btrfs_trans_handle *trans,
794 struct btrfs_root *root, u64 bytenr, u64 len)
795{
796 struct btrfs_fs_info *fs_info = trans->fs_info;
797 struct btrfs_path *path;
798 struct btrfs_key key;
799 u64 end_byte = bytenr + len;
800 u64 csum_end;
801 struct extent_buffer *leaf;
802 int ret = 0;
803 const u32 csum_size = fs_info->csum_size;
804 u32 blocksize_bits = fs_info->sectorsize_bits;
805
806 ASSERT(root == fs_info->csum_root ||
807 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
808
809 path = btrfs_alloc_path();
810 if (!path)
811 return -ENOMEM;
812
813 while (1) {
814 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
815 key.offset = end_byte - 1;
816 key.type = BTRFS_EXTENT_CSUM_KEY;
817
818 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
819 if (ret > 0) {
820 ret = 0;
821 if (path->slots[0] == 0)
822 break;
823 path->slots[0]--;
824 } else if (ret < 0) {
825 break;
826 }
827
828 leaf = path->nodes[0];
829 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
830
831 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
832 key.type != BTRFS_EXTENT_CSUM_KEY) {
833 break;
834 }
835
836 if (key.offset >= end_byte)
837 break;
838
839 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
840 csum_end <<= blocksize_bits;
841 csum_end += key.offset;
842
843 /* this csum ends before we start, we're done */
844 if (csum_end <= bytenr)
845 break;
846
847 /* delete the entire item, it is inside our range */
848 if (key.offset >= bytenr && csum_end <= end_byte) {
849 int del_nr = 1;
850
851 /*
852 * Check how many csum items preceding this one in this
853 * leaf correspond to our range and then delete them all
854 * at once.
855 */
856 if (key.offset > bytenr && path->slots[0] > 0) {
857 int slot = path->slots[0] - 1;
858
859 while (slot >= 0) {
860 struct btrfs_key pk;
861
862 btrfs_item_key_to_cpu(leaf, &pk, slot);
863 if (pk.offset < bytenr ||
864 pk.type != BTRFS_EXTENT_CSUM_KEY ||
865 pk.objectid !=
866 BTRFS_EXTENT_CSUM_OBJECTID)
867 break;
868 path->slots[0] = slot;
869 del_nr++;
870 key.offset = pk.offset;
871 slot--;
872 }
873 }
874 ret = btrfs_del_items(trans, root, path,
875 path->slots[0], del_nr);
876 if (ret)
877 break;
878 if (key.offset == bytenr)
879 break;
880 } else if (key.offset < bytenr && csum_end > end_byte) {
881 unsigned long offset;
882 unsigned long shift_len;
883 unsigned long item_offset;
884 /*
885 * [ bytenr - len ]
886 * [csum ]
887 *
888 * Our bytes are in the middle of the csum,
889 * we need to split this item and insert a new one.
890 *
891 * But we can't drop the path because the
892 * csum could change, get removed, extended etc.
893 *
894 * The trick here is the max size of a csum item leaves
895 * enough room in the tree block for a single
896 * item header. So, we split the item in place,
897 * adding a new header pointing to the existing
898 * bytes. Then we loop around again and we have
899 * a nicely formed csum item that we can neatly
900 * truncate.
901 */
902 offset = (bytenr - key.offset) >> blocksize_bits;
903 offset *= csum_size;
904
905 shift_len = (len >> blocksize_bits) * csum_size;
906
907 item_offset = btrfs_item_ptr_offset(leaf,
908 path->slots[0]);
909
910 memzero_extent_buffer(leaf, item_offset + offset,
911 shift_len);
912 key.offset = bytenr;
913
914 /*
915 * btrfs_split_item returns -EAGAIN when the
916 * item changed size or key
917 */
918 ret = btrfs_split_item(trans, root, path, &key, offset);
919 if (ret && ret != -EAGAIN) {
920 btrfs_abort_transaction(trans, ret);
921 break;
922 }
923 ret = 0;
924
925 key.offset = end_byte - 1;
926 } else {
927 truncate_one_csum(fs_info, path, &key, bytenr, len);
928 if (key.offset < bytenr)
929 break;
930 }
931 btrfs_release_path(path);
932 }
933 btrfs_free_path(path);
934 return ret;
935}
936
937static int find_next_csum_offset(struct btrfs_root *root,
938 struct btrfs_path *path,
939 u64 *next_offset)
940{
941 const u32 nritems = btrfs_header_nritems(path->nodes[0]);
942 struct btrfs_key found_key;
943 int slot = path->slots[0] + 1;
944 int ret;
945
946 if (nritems == 0 || slot >= nritems) {
947 ret = btrfs_next_leaf(root, path);
948 if (ret < 0) {
949 return ret;
950 } else if (ret > 0) {
951 *next_offset = (u64)-1;
952 return 0;
953 }
954 slot = path->slots[0];
955 }
956
957 btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
958
959 if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
960 found_key.type != BTRFS_EXTENT_CSUM_KEY)
961 *next_offset = (u64)-1;
962 else
963 *next_offset = found_key.offset;
964
965 return 0;
966}
967
968int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
969 struct btrfs_root *root,
970 struct btrfs_ordered_sum *sums)
971{
972 struct btrfs_fs_info *fs_info = root->fs_info;
973 struct btrfs_key file_key;
974 struct btrfs_key found_key;
975 struct btrfs_path *path;
976 struct btrfs_csum_item *item;
977 struct btrfs_csum_item *item_end;
978 struct extent_buffer *leaf = NULL;
979 u64 next_offset;
980 u64 total_bytes = 0;
981 u64 csum_offset;
982 u64 bytenr;
983 u32 ins_size;
984 int index = 0;
985 int found_next;
986 int ret;
987 const u32 csum_size = fs_info->csum_size;
988
989 path = btrfs_alloc_path();
990 if (!path)
991 return -ENOMEM;
992again:
993 next_offset = (u64)-1;
994 found_next = 0;
995 bytenr = sums->bytenr + total_bytes;
996 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
997 file_key.offset = bytenr;
998 file_key.type = BTRFS_EXTENT_CSUM_KEY;
999
1000 item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
1001 if (!IS_ERR(item)) {
1002 ret = 0;
1003 leaf = path->nodes[0];
1004 item_end = btrfs_item_ptr(leaf, path->slots[0],
1005 struct btrfs_csum_item);
1006 item_end = (struct btrfs_csum_item *)((char *)item_end +
1007 btrfs_item_size_nr(leaf, path->slots[0]));
1008 goto found;
1009 }
1010 ret = PTR_ERR(item);
1011 if (ret != -EFBIG && ret != -ENOENT)
1012 goto out;
1013
1014 if (ret == -EFBIG) {
1015 u32 item_size;
1016 /* we found one, but it isn't big enough yet */
1017 leaf = path->nodes[0];
1018 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1019 if ((item_size / csum_size) >=
1020 MAX_CSUM_ITEMS(fs_info, csum_size)) {
1021 /* already at max size, make a new one */
1022 goto insert;
1023 }
1024 } else {
1025 /* We didn't find a csum item, insert one. */
1026 ret = find_next_csum_offset(root, path, &next_offset);
1027 if (ret < 0)
1028 goto out;
1029 found_next = 1;
1030 goto insert;
1031 }
1032
1033 /*
1034 * At this point, we know the tree has a checksum item that ends at an
1035 * offset matching the start of the checksum range we want to insert.
1036 * We try to extend that item as much as possible and then add as many
1037 * checksums to it as they fit.
1038 *
1039 * First check if the leaf has enough free space for at least one
1040 * checksum. If it has go directly to the item extension code, otherwise
1041 * release the path and do a search for insertion before the extension.
1042 */
1043 if (btrfs_leaf_free_space(leaf) >= csum_size) {
1044 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1045 csum_offset = (bytenr - found_key.offset) >>
1046 fs_info->sectorsize_bits;
1047 goto extend_csum;
1048 }
1049
1050 btrfs_release_path(path);
1051 path->search_for_extension = 1;
1052 ret = btrfs_search_slot(trans, root, &file_key, path,
1053 csum_size, 1);
1054 path->search_for_extension = 0;
1055 if (ret < 0)
1056 goto out;
1057
1058 if (ret > 0) {
1059 if (path->slots[0] == 0)
1060 goto insert;
1061 path->slots[0]--;
1062 }
1063
1064 leaf = path->nodes[0];
1065 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1066 csum_offset = (bytenr - found_key.offset) >> fs_info->sectorsize_bits;
1067
1068 if (found_key.type != BTRFS_EXTENT_CSUM_KEY ||
1069 found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1070 csum_offset >= MAX_CSUM_ITEMS(fs_info, csum_size)) {
1071 goto insert;
1072 }
1073
1074extend_csum:
1075 if (csum_offset == btrfs_item_size_nr(leaf, path->slots[0]) /
1076 csum_size) {
1077 int extend_nr;
1078 u64 tmp;
1079 u32 diff;
1080
1081 tmp = sums->len - total_bytes;
1082 tmp >>= fs_info->sectorsize_bits;
1083 WARN_ON(tmp < 1);
1084 extend_nr = max_t(int, 1, tmp);
1085
1086 /*
1087 * A log tree can already have checksum items with a subset of
1088 * the checksums we are trying to log. This can happen after
1089 * doing a sequence of partial writes into prealloc extents and
1090 * fsyncs in between, with a full fsync logging a larger subrange
1091 * of an extent for which a previous fast fsync logged a smaller
1092 * subrange. And this happens in particular due to merging file
1093 * extent items when we complete an ordered extent for a range
1094 * covered by a prealloc extent - this is done at
1095 * btrfs_mark_extent_written().
1096 *
1097 * So if we try to extend the previous checksum item, which has
1098 * a range that ends at the start of the range we want to insert,
1099 * make sure we don't extend beyond the start offset of the next
1100 * checksum item. If we are at the last item in the leaf, then
1101 * forget the optimization of extending and add a new checksum
1102 * item - it is not worth the complexity of releasing the path,
1103 * getting the first key for the next leaf, repeat the btree
1104 * search, etc, because log trees are temporary anyway and it
1105 * would only save a few bytes of leaf space.
1106 */
1107 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
1108 if (path->slots[0] + 1 >=
1109 btrfs_header_nritems(path->nodes[0])) {
1110 ret = find_next_csum_offset(root, path, &next_offset);
1111 if (ret < 0)
1112 goto out;
1113 found_next = 1;
1114 goto insert;
1115 }
1116
1117 ret = find_next_csum_offset(root, path, &next_offset);
1118 if (ret < 0)
1119 goto out;
1120
1121 tmp = (next_offset - bytenr) >> fs_info->sectorsize_bits;
1122 if (tmp <= INT_MAX)
1123 extend_nr = min_t(int, extend_nr, tmp);
1124 }
1125
1126 diff = (csum_offset + extend_nr) * csum_size;
1127 diff = min(diff,
1128 MAX_CSUM_ITEMS(fs_info, csum_size) * csum_size);
1129
1130 diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
1131 diff = min_t(u32, btrfs_leaf_free_space(leaf), diff);
1132 diff /= csum_size;
1133 diff *= csum_size;
1134
1135 btrfs_extend_item(path, diff);
1136 ret = 0;
1137 goto csum;
1138 }
1139
1140insert:
1141 btrfs_release_path(path);
1142 csum_offset = 0;
1143 if (found_next) {
1144 u64 tmp;
1145
1146 tmp = sums->len - total_bytes;
1147 tmp >>= fs_info->sectorsize_bits;
1148 tmp = min(tmp, (next_offset - file_key.offset) >>
1149 fs_info->sectorsize_bits);
1150
1151 tmp = max_t(u64, 1, tmp);
1152 tmp = min_t(u64, tmp, MAX_CSUM_ITEMS(fs_info, csum_size));
1153 ins_size = csum_size * tmp;
1154 } else {
1155 ins_size = csum_size;
1156 }
1157 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
1158 ins_size);
1159 if (ret < 0)
1160 goto out;
1161 if (WARN_ON(ret != 0))
1162 goto out;
1163 leaf = path->nodes[0];
1164csum:
1165 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
1166 item_end = (struct btrfs_csum_item *)((unsigned char *)item +
1167 btrfs_item_size_nr(leaf, path->slots[0]));
1168 item = (struct btrfs_csum_item *)((unsigned char *)item +
1169 csum_offset * csum_size);
1170found:
1171 ins_size = (u32)(sums->len - total_bytes) >> fs_info->sectorsize_bits;
1172 ins_size *= csum_size;
1173 ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item,
1174 ins_size);
1175 write_extent_buffer(leaf, sums->sums + index, (unsigned long)item,
1176 ins_size);
1177
1178 index += ins_size;
1179 ins_size /= csum_size;
1180 total_bytes += ins_size * fs_info->sectorsize;
1181
1182 btrfs_mark_buffer_dirty(path->nodes[0]);
1183 if (total_bytes < sums->len) {
1184 btrfs_release_path(path);
1185 cond_resched();
1186 goto again;
1187 }
1188out:
1189 btrfs_free_path(path);
1190 return ret;
1191}
1192
1193void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode,
1194 const struct btrfs_path *path,
1195 struct btrfs_file_extent_item *fi,
1196 const bool new_inline,
1197 struct extent_map *em)
1198{
1199 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1200 struct btrfs_root *root = inode->root;
1201 struct extent_buffer *leaf = path->nodes[0];
1202 const int slot = path->slots[0];
1203 struct btrfs_key key;
1204 u64 extent_start, extent_end;
1205 u64 bytenr;
1206 u8 type = btrfs_file_extent_type(leaf, fi);
1207 int compress_type = btrfs_file_extent_compression(leaf, fi);
1208
1209 btrfs_item_key_to_cpu(leaf, &key, slot);
1210 extent_start = key.offset;
1211 extent_end = btrfs_file_extent_end(path);
1212 em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1213 if (type == BTRFS_FILE_EXTENT_REG ||
1214 type == BTRFS_FILE_EXTENT_PREALLOC) {
1215 em->start = extent_start;
1216 em->len = extent_end - extent_start;
1217 em->orig_start = extent_start -
1218 btrfs_file_extent_offset(leaf, fi);
1219 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
1220 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1221 if (bytenr == 0) {
1222 em->block_start = EXTENT_MAP_HOLE;
1223 return;
1224 }
1225 if (compress_type != BTRFS_COMPRESS_NONE) {
1226 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1227 em->compress_type = compress_type;
1228 em->block_start = bytenr;
1229 em->block_len = em->orig_block_len;
1230 } else {
1231 bytenr += btrfs_file_extent_offset(leaf, fi);
1232 em->block_start = bytenr;
1233 em->block_len = em->len;
1234 if (type == BTRFS_FILE_EXTENT_PREALLOC)
1235 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
1236 }
1237 } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1238 em->block_start = EXTENT_MAP_INLINE;
1239 em->start = extent_start;
1240 em->len = extent_end - extent_start;
1241 /*
1242 * Initialize orig_start and block_len with the same values
1243 * as in inode.c:btrfs_get_extent().
1244 */
1245 em->orig_start = EXTENT_MAP_HOLE;
1246 em->block_len = (u64)-1;
1247 if (!new_inline && compress_type != BTRFS_COMPRESS_NONE) {
1248 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1249 em->compress_type = compress_type;
1250 }
1251 } else {
1252 btrfs_err(fs_info,
1253 "unknown file extent item type %d, inode %llu, offset %llu, "
1254 "root %llu", type, btrfs_ino(inode), extent_start,
1255 root->root_key.objectid);
1256 }
1257}
1258
1259/*
1260 * Returns the end offset (non inclusive) of the file extent item the given path
1261 * points to. If it points to an inline extent, the returned offset is rounded
1262 * up to the sector size.
1263 */
1264u64 btrfs_file_extent_end(const struct btrfs_path *path)
1265{
1266 const struct extent_buffer *leaf = path->nodes[0];
1267 const int slot = path->slots[0];
1268 struct btrfs_file_extent_item *fi;
1269 struct btrfs_key key;
1270 u64 end;
1271
1272 btrfs_item_key_to_cpu(leaf, &key, slot);
1273 ASSERT(key.type == BTRFS_EXTENT_DATA_KEY);
1274 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1275
1276 if (btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_INLINE) {
1277 end = btrfs_file_extent_ram_bytes(leaf, fi);
1278 end = ALIGN(key.offset + end, leaf->fs_info->sectorsize);
1279 } else {
1280 end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1281 }
1282
1283 return end;
1284}