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 "messages.h"
13#include "misc.h"
14#include "ctree.h"
15#include "disk-io.h"
16#include "transaction.h"
17#include "bio.h"
18#include "print-tree.h"
19#include "compression.h"
20#include "fs.h"
21#include "accessors.h"
22#include "file-item.h"
23#include "super.h"
24
25#define __MAX_CSUM_ITEMS(r, size) ((unsigned long)(((BTRFS_LEAF_DATA_SIZE(r) - \
26 sizeof(struct btrfs_item) * 2) / \
27 size) - 1))
28
29#define MAX_CSUM_ITEMS(r, size) (min_t(u32, __MAX_CSUM_ITEMS(r, size), \
30 PAGE_SIZE))
31
32/*
33 * Set inode's size according to filesystem options.
34 *
35 * @inode: inode we want to update the disk_i_size for
36 * @new_i_size: i_size we want to set to, 0 if we use i_size
37 *
38 * With NO_HOLES set this simply sets the disk_is_size to whatever i_size_read()
39 * returns as it is perfectly fine with a file that has holes without hole file
40 * extent items.
41 *
42 * However without NO_HOLES we need to only return the area that is contiguous
43 * from the 0 offset of the file. Otherwise we could end up adjust i_size up
44 * to an extent that has a gap in between.
45 *
46 * Finally new_i_size should only be set in the case of truncate where we're not
47 * ready to use i_size_read() as the limiter yet.
48 */
49void btrfs_inode_safe_disk_i_size_write(struct btrfs_inode *inode, u64 new_i_size)
50{
51 struct btrfs_fs_info *fs_info = inode->root->fs_info;
52 u64 start, end, i_size;
53 int ret;
54
55 spin_lock(&inode->lock);
56 i_size = new_i_size ?: i_size_read(&inode->vfs_inode);
57 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
58 inode->disk_i_size = i_size;
59 goto out_unlock;
60 }
61
62 ret = find_contiguous_extent_bit(inode->file_extent_tree, 0, &start,
63 &end, EXTENT_DIRTY);
64 if (!ret && start == 0)
65 i_size = min(i_size, end + 1);
66 else
67 i_size = 0;
68 inode->disk_i_size = i_size;
69out_unlock:
70 spin_unlock(&inode->lock);
71}
72
73/*
74 * Mark range within a file as having a new extent inserted.
75 *
76 * @inode: inode being modified
77 * @start: start file offset of the file extent we've inserted
78 * @len: logical length of the file extent item
79 *
80 * Call when we are inserting a new file extent where there was none before.
81 * Does not need to call this in the case where we're replacing an existing file
82 * extent, however if not sure it's fine to call this multiple times.
83 *
84 * The start and len must match the file extent item, so thus must be sectorsize
85 * aligned.
86 */
87int btrfs_inode_set_file_extent_range(struct btrfs_inode *inode, u64 start,
88 u64 len)
89{
90 if (len == 0)
91 return 0;
92
93 ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize));
94
95 if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES))
96 return 0;
97 return set_extent_bit(inode->file_extent_tree, start, start + len - 1,
98 EXTENT_DIRTY, NULL);
99}
100
101/*
102 * Mark an inode range as not having a backing extent.
103 *
104 * @inode: inode being modified
105 * @start: start file offset of the file extent we've inserted
106 * @len: logical length of the file extent item
107 *
108 * Called when we drop a file extent, for example when we truncate. Doesn't
109 * need to be called for cases where we're replacing a file extent, like when
110 * we've COWed a file extent.
111 *
112 * The start and len must match the file extent item, so thus must be sectorsize
113 * aligned.
114 */
115int btrfs_inode_clear_file_extent_range(struct btrfs_inode *inode, u64 start,
116 u64 len)
117{
118 if (len == 0)
119 return 0;
120
121 ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize) ||
122 len == (u64)-1);
123
124 if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES))
125 return 0;
126 return clear_extent_bit(inode->file_extent_tree, start,
127 start + len - 1, EXTENT_DIRTY, NULL);
128}
129
130static size_t bytes_to_csum_size(const struct btrfs_fs_info *fs_info, u32 bytes)
131{
132 ASSERT(IS_ALIGNED(bytes, fs_info->sectorsize));
133
134 return (bytes >> fs_info->sectorsize_bits) * fs_info->csum_size;
135}
136
137static size_t csum_size_to_bytes(const struct btrfs_fs_info *fs_info, u32 csum_size)
138{
139 ASSERT(IS_ALIGNED(csum_size, fs_info->csum_size));
140
141 return (csum_size / fs_info->csum_size) << fs_info->sectorsize_bits;
142}
143
144static inline u32 max_ordered_sum_bytes(const struct btrfs_fs_info *fs_info)
145{
146 u32 max_csum_size = round_down(PAGE_SIZE - sizeof(struct btrfs_ordered_sum),
147 fs_info->csum_size);
148
149 return csum_size_to_bytes(fs_info, max_csum_size);
150}
151
152/*
153 * Calculate the total size needed to allocate for an ordered sum structure
154 * spanning @bytes in the file.
155 */
156static int btrfs_ordered_sum_size(struct btrfs_fs_info *fs_info, unsigned long bytes)
157{
158 return sizeof(struct btrfs_ordered_sum) + bytes_to_csum_size(fs_info, bytes);
159}
160
161int btrfs_insert_hole_extent(struct btrfs_trans_handle *trans,
162 struct btrfs_root *root,
163 u64 objectid, u64 pos, u64 num_bytes)
164{
165 int ret = 0;
166 struct btrfs_file_extent_item *item;
167 struct btrfs_key file_key;
168 struct btrfs_path *path;
169 struct extent_buffer *leaf;
170
171 path = btrfs_alloc_path();
172 if (!path)
173 return -ENOMEM;
174 file_key.objectid = objectid;
175 file_key.offset = pos;
176 file_key.type = BTRFS_EXTENT_DATA_KEY;
177
178 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
179 sizeof(*item));
180 if (ret < 0)
181 goto out;
182 BUG_ON(ret); /* Can't happen */
183 leaf = path->nodes[0];
184 item = btrfs_item_ptr(leaf, path->slots[0],
185 struct btrfs_file_extent_item);
186 btrfs_set_file_extent_disk_bytenr(leaf, item, 0);
187 btrfs_set_file_extent_disk_num_bytes(leaf, item, 0);
188 btrfs_set_file_extent_offset(leaf, item, 0);
189 btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
190 btrfs_set_file_extent_ram_bytes(leaf, item, num_bytes);
191 btrfs_set_file_extent_generation(leaf, item, trans->transid);
192 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
193 btrfs_set_file_extent_compression(leaf, item, 0);
194 btrfs_set_file_extent_encryption(leaf, item, 0);
195 btrfs_set_file_extent_other_encoding(leaf, item, 0);
196
197 btrfs_mark_buffer_dirty(trans, leaf);
198out:
199 btrfs_free_path(path);
200 return ret;
201}
202
203static struct btrfs_csum_item *
204btrfs_lookup_csum(struct btrfs_trans_handle *trans,
205 struct btrfs_root *root,
206 struct btrfs_path *path,
207 u64 bytenr, int cow)
208{
209 struct btrfs_fs_info *fs_info = root->fs_info;
210 int ret;
211 struct btrfs_key file_key;
212 struct btrfs_key found_key;
213 struct btrfs_csum_item *item;
214 struct extent_buffer *leaf;
215 u64 csum_offset = 0;
216 const u32 csum_size = fs_info->csum_size;
217 int csums_in_item;
218
219 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
220 file_key.offset = bytenr;
221 file_key.type = BTRFS_EXTENT_CSUM_KEY;
222 ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
223 if (ret < 0)
224 goto fail;
225 leaf = path->nodes[0];
226 if (ret > 0) {
227 ret = 1;
228 if (path->slots[0] == 0)
229 goto fail;
230 path->slots[0]--;
231 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
232 if (found_key.type != BTRFS_EXTENT_CSUM_KEY)
233 goto fail;
234
235 csum_offset = (bytenr - found_key.offset) >>
236 fs_info->sectorsize_bits;
237 csums_in_item = btrfs_item_size(leaf, path->slots[0]);
238 csums_in_item /= csum_size;
239
240 if (csum_offset == csums_in_item) {
241 ret = -EFBIG;
242 goto fail;
243 } else if (csum_offset > csums_in_item) {
244 goto fail;
245 }
246 }
247 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
248 item = (struct btrfs_csum_item *)((unsigned char *)item +
249 csum_offset * csum_size);
250 return item;
251fail:
252 if (ret > 0)
253 ret = -ENOENT;
254 return ERR_PTR(ret);
255}
256
257int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
258 struct btrfs_root *root,
259 struct btrfs_path *path, u64 objectid,
260 u64 offset, int mod)
261{
262 struct btrfs_key file_key;
263 int ins_len = mod < 0 ? -1 : 0;
264 int cow = mod != 0;
265
266 file_key.objectid = objectid;
267 file_key.offset = offset;
268 file_key.type = BTRFS_EXTENT_DATA_KEY;
269
270 return btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
271}
272
273/*
274 * Find checksums for logical bytenr range [disk_bytenr, disk_bytenr + len) and
275 * store the result to @dst.
276 *
277 * Return >0 for the number of sectors we found.
278 * Return 0 for the range [disk_bytenr, disk_bytenr + sectorsize) has no csum
279 * for it. Caller may want to try next sector until one range is hit.
280 * Return <0 for fatal error.
281 */
282static int search_csum_tree(struct btrfs_fs_info *fs_info,
283 struct btrfs_path *path, u64 disk_bytenr,
284 u64 len, u8 *dst)
285{
286 struct btrfs_root *csum_root;
287 struct btrfs_csum_item *item = NULL;
288 struct btrfs_key key;
289 const u32 sectorsize = fs_info->sectorsize;
290 const u32 csum_size = fs_info->csum_size;
291 u32 itemsize;
292 int ret;
293 u64 csum_start;
294 u64 csum_len;
295
296 ASSERT(IS_ALIGNED(disk_bytenr, sectorsize) &&
297 IS_ALIGNED(len, sectorsize));
298
299 /* Check if the current csum item covers disk_bytenr */
300 if (path->nodes[0]) {
301 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
302 struct btrfs_csum_item);
303 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
304 itemsize = btrfs_item_size(path->nodes[0], path->slots[0]);
305
306 csum_start = key.offset;
307 csum_len = (itemsize / csum_size) * sectorsize;
308
309 if (in_range(disk_bytenr, csum_start, csum_len))
310 goto found;
311 }
312
313 /* Current item doesn't contain the desired range, search again */
314 btrfs_release_path(path);
315 csum_root = btrfs_csum_root(fs_info, disk_bytenr);
316 item = btrfs_lookup_csum(NULL, csum_root, path, disk_bytenr, 0);
317 if (IS_ERR(item)) {
318 ret = PTR_ERR(item);
319 goto out;
320 }
321 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
322 itemsize = btrfs_item_size(path->nodes[0], path->slots[0]);
323
324 csum_start = key.offset;
325 csum_len = (itemsize / csum_size) * sectorsize;
326 ASSERT(in_range(disk_bytenr, csum_start, csum_len));
327
328found:
329 ret = (min(csum_start + csum_len, disk_bytenr + len) -
330 disk_bytenr) >> fs_info->sectorsize_bits;
331 read_extent_buffer(path->nodes[0], dst, (unsigned long)item,
332 ret * csum_size);
333out:
334 if (ret == -ENOENT || ret == -EFBIG)
335 ret = 0;
336 return ret;
337}
338
339/*
340 * Lookup the checksum for the read bio in csum tree.
341 *
342 * Return: BLK_STS_RESOURCE if allocating memory fails, BLK_STS_OK otherwise.
343 */
344blk_status_t btrfs_lookup_bio_sums(struct btrfs_bio *bbio)
345{
346 struct btrfs_inode *inode = bbio->inode;
347 struct btrfs_fs_info *fs_info = inode->root->fs_info;
348 struct bio *bio = &bbio->bio;
349 struct btrfs_path *path;
350 const u32 sectorsize = fs_info->sectorsize;
351 const u32 csum_size = fs_info->csum_size;
352 u32 orig_len = bio->bi_iter.bi_size;
353 u64 orig_disk_bytenr = bio->bi_iter.bi_sector << SECTOR_SHIFT;
354 const unsigned int nblocks = orig_len >> fs_info->sectorsize_bits;
355 blk_status_t ret = BLK_STS_OK;
356 u32 bio_offset = 0;
357
358 if ((inode->flags & BTRFS_INODE_NODATASUM) ||
359 test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state))
360 return BLK_STS_OK;
361
362 /*
363 * This function is only called for read bio.
364 *
365 * This means two things:
366 * - All our csums should only be in csum tree
367 * No ordered extents csums, as ordered extents are only for write
368 * path.
369 * - No need to bother any other info from bvec
370 * Since we're looking up csums, the only important info is the
371 * disk_bytenr and the length, which can be extracted from bi_iter
372 * directly.
373 */
374 ASSERT(bio_op(bio) == REQ_OP_READ);
375 path = btrfs_alloc_path();
376 if (!path)
377 return BLK_STS_RESOURCE;
378
379 if (nblocks * csum_size > BTRFS_BIO_INLINE_CSUM_SIZE) {
380 bbio->csum = kmalloc_array(nblocks, csum_size, GFP_NOFS);
381 if (!bbio->csum) {
382 btrfs_free_path(path);
383 return BLK_STS_RESOURCE;
384 }
385 } else {
386 bbio->csum = bbio->csum_inline;
387 }
388
389 /*
390 * If requested number of sectors is larger than one leaf can contain,
391 * kick the readahead for csum tree.
392 */
393 if (nblocks > fs_info->csums_per_leaf)
394 path->reada = READA_FORWARD;
395
396 /*
397 * the free space stuff is only read when it hasn't been
398 * updated in the current transaction. So, we can safely
399 * read from the commit root and sidestep a nasty deadlock
400 * between reading the free space cache and updating the csum tree.
401 */
402 if (btrfs_is_free_space_inode(inode)) {
403 path->search_commit_root = 1;
404 path->skip_locking = 1;
405 }
406
407 while (bio_offset < orig_len) {
408 int count;
409 u64 cur_disk_bytenr = orig_disk_bytenr + bio_offset;
410 u8 *csum_dst = bbio->csum +
411 (bio_offset >> fs_info->sectorsize_bits) * csum_size;
412
413 count = search_csum_tree(fs_info, path, cur_disk_bytenr,
414 orig_len - bio_offset, csum_dst);
415 if (count < 0) {
416 ret = errno_to_blk_status(count);
417 if (bbio->csum != bbio->csum_inline)
418 kfree(bbio->csum);
419 bbio->csum = NULL;
420 break;
421 }
422
423 /*
424 * We didn't find a csum for this range. We need to make sure
425 * we complain loudly about this, because we are not NODATASUM.
426 *
427 * However for the DATA_RELOC inode we could potentially be
428 * relocating data extents for a NODATASUM inode, so the inode
429 * itself won't be marked with NODATASUM, but the extent we're
430 * copying is in fact NODATASUM. If we don't find a csum we
431 * assume this is the case.
432 */
433 if (count == 0) {
434 memset(csum_dst, 0, csum_size);
435 count = 1;
436
437 if (inode->root->root_key.objectid ==
438 BTRFS_DATA_RELOC_TREE_OBJECTID) {
439 u64 file_offset = bbio->file_offset + bio_offset;
440
441 set_extent_bit(&inode->io_tree, file_offset,
442 file_offset + sectorsize - 1,
443 EXTENT_NODATASUM, NULL);
444 } else {
445 btrfs_warn_rl(fs_info,
446 "csum hole found for disk bytenr range [%llu, %llu)",
447 cur_disk_bytenr, cur_disk_bytenr + sectorsize);
448 }
449 }
450 bio_offset += count * sectorsize;
451 }
452
453 btrfs_free_path(path);
454 return ret;
455}
456
457int btrfs_lookup_csums_list(struct btrfs_root *root, u64 start, u64 end,
458 struct list_head *list, int search_commit,
459 bool nowait)
460{
461 struct btrfs_fs_info *fs_info = root->fs_info;
462 struct btrfs_key key;
463 struct btrfs_path *path;
464 struct extent_buffer *leaf;
465 struct btrfs_ordered_sum *sums;
466 struct btrfs_csum_item *item;
467 LIST_HEAD(tmplist);
468 int ret;
469
470 ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
471 IS_ALIGNED(end + 1, fs_info->sectorsize));
472
473 path = btrfs_alloc_path();
474 if (!path)
475 return -ENOMEM;
476
477 path->nowait = nowait;
478 if (search_commit) {
479 path->skip_locking = 1;
480 path->reada = READA_FORWARD;
481 path->search_commit_root = 1;
482 }
483
484 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
485 key.offset = start;
486 key.type = BTRFS_EXTENT_CSUM_KEY;
487
488 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
489 if (ret < 0)
490 goto fail;
491 if (ret > 0 && path->slots[0] > 0) {
492 leaf = path->nodes[0];
493 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
494
495 /*
496 * There are two cases we can hit here for the previous csum
497 * item:
498 *
499 * |<- search range ->|
500 * |<- csum item ->|
501 *
502 * Or
503 * |<- search range ->|
504 * |<- csum item ->|
505 *
506 * Check if the previous csum item covers the leading part of
507 * the search range. If so we have to start from previous csum
508 * item.
509 */
510 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
511 key.type == BTRFS_EXTENT_CSUM_KEY) {
512 if (bytes_to_csum_size(fs_info, start - key.offset) <
513 btrfs_item_size(leaf, path->slots[0] - 1))
514 path->slots[0]--;
515 }
516 }
517
518 while (start <= end) {
519 u64 csum_end;
520
521 leaf = path->nodes[0];
522 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
523 ret = btrfs_next_leaf(root, path);
524 if (ret < 0)
525 goto fail;
526 if (ret > 0)
527 break;
528 leaf = path->nodes[0];
529 }
530
531 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
532 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
533 key.type != BTRFS_EXTENT_CSUM_KEY ||
534 key.offset > end)
535 break;
536
537 if (key.offset > start)
538 start = key.offset;
539
540 csum_end = key.offset + csum_size_to_bytes(fs_info,
541 btrfs_item_size(leaf, path->slots[0]));
542 if (csum_end <= start) {
543 path->slots[0]++;
544 continue;
545 }
546
547 csum_end = min(csum_end, end + 1);
548 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
549 struct btrfs_csum_item);
550 while (start < csum_end) {
551 unsigned long offset;
552 size_t size;
553
554 size = min_t(size_t, csum_end - start,
555 max_ordered_sum_bytes(fs_info));
556 sums = kzalloc(btrfs_ordered_sum_size(fs_info, size),
557 GFP_NOFS);
558 if (!sums) {
559 ret = -ENOMEM;
560 goto fail;
561 }
562
563 sums->logical = start;
564 sums->len = size;
565
566 offset = bytes_to_csum_size(fs_info, start - key.offset);
567
568 read_extent_buffer(path->nodes[0],
569 sums->sums,
570 ((unsigned long)item) + offset,
571 bytes_to_csum_size(fs_info, size));
572
573 start += size;
574 list_add_tail(&sums->list, &tmplist);
575 }
576 path->slots[0]++;
577 }
578 ret = 0;
579fail:
580 while (ret < 0 && !list_empty(&tmplist)) {
581 sums = list_entry(tmplist.next, struct btrfs_ordered_sum, list);
582 list_del(&sums->list);
583 kfree(sums);
584 }
585 list_splice_tail(&tmplist, list);
586
587 btrfs_free_path(path);
588 return ret;
589}
590
591/*
592 * Do the same work as btrfs_lookup_csums_list(), the difference is in how
593 * we return the result.
594 *
595 * This version will set the corresponding bits in @csum_bitmap to represent
596 * that there is a csum found.
597 * Each bit represents a sector. Thus caller should ensure @csum_buf passed
598 * in is large enough to contain all csums.
599 */
600int btrfs_lookup_csums_bitmap(struct btrfs_root *root, struct btrfs_path *path,
601 u64 start, u64 end, u8 *csum_buf,
602 unsigned long *csum_bitmap)
603{
604 struct btrfs_fs_info *fs_info = root->fs_info;
605 struct btrfs_key key;
606 struct extent_buffer *leaf;
607 struct btrfs_csum_item *item;
608 const u64 orig_start = start;
609 bool free_path = false;
610 int ret;
611
612 ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
613 IS_ALIGNED(end + 1, fs_info->sectorsize));
614
615 if (!path) {
616 path = btrfs_alloc_path();
617 if (!path)
618 return -ENOMEM;
619 free_path = true;
620 }
621
622 /* Check if we can reuse the previous path. */
623 if (path->nodes[0]) {
624 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
625
626 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
627 key.type == BTRFS_EXTENT_CSUM_KEY &&
628 key.offset <= start)
629 goto search_forward;
630 btrfs_release_path(path);
631 }
632
633 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
634 key.type = BTRFS_EXTENT_CSUM_KEY;
635 key.offset = start;
636
637 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
638 if (ret < 0)
639 goto fail;
640 if (ret > 0 && path->slots[0] > 0) {
641 leaf = path->nodes[0];
642 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
643
644 /*
645 * There are two cases we can hit here for the previous csum
646 * item:
647 *
648 * |<- search range ->|
649 * |<- csum item ->|
650 *
651 * Or
652 * |<- search range ->|
653 * |<- csum item ->|
654 *
655 * Check if the previous csum item covers the leading part of
656 * the search range. If so we have to start from previous csum
657 * item.
658 */
659 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
660 key.type == BTRFS_EXTENT_CSUM_KEY) {
661 if (bytes_to_csum_size(fs_info, start - key.offset) <
662 btrfs_item_size(leaf, path->slots[0] - 1))
663 path->slots[0]--;
664 }
665 }
666
667search_forward:
668 while (start <= end) {
669 u64 csum_end;
670
671 leaf = path->nodes[0];
672 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
673 ret = btrfs_next_leaf(root, path);
674 if (ret < 0)
675 goto fail;
676 if (ret > 0)
677 break;
678 leaf = path->nodes[0];
679 }
680
681 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
682 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
683 key.type != BTRFS_EXTENT_CSUM_KEY ||
684 key.offset > end)
685 break;
686
687 if (key.offset > start)
688 start = key.offset;
689
690 csum_end = key.offset + csum_size_to_bytes(fs_info,
691 btrfs_item_size(leaf, path->slots[0]));
692 if (csum_end <= start) {
693 path->slots[0]++;
694 continue;
695 }
696
697 csum_end = min(csum_end, end + 1);
698 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
699 struct btrfs_csum_item);
700 while (start < csum_end) {
701 unsigned long offset;
702 size_t size;
703 u8 *csum_dest = csum_buf + bytes_to_csum_size(fs_info,
704 start - orig_start);
705
706 size = min_t(size_t, csum_end - start, end + 1 - start);
707
708 offset = bytes_to_csum_size(fs_info, start - key.offset);
709
710 read_extent_buffer(path->nodes[0], csum_dest,
711 ((unsigned long)item) + offset,
712 bytes_to_csum_size(fs_info, size));
713
714 bitmap_set(csum_bitmap,
715 (start - orig_start) >> fs_info->sectorsize_bits,
716 size >> fs_info->sectorsize_bits);
717
718 start += size;
719 }
720 path->slots[0]++;
721 }
722 ret = 0;
723fail:
724 if (free_path)
725 btrfs_free_path(path);
726 return ret;
727}
728
729/*
730 * Calculate checksums of the data contained inside a bio.
731 */
732blk_status_t btrfs_csum_one_bio(struct btrfs_bio *bbio)
733{
734 struct btrfs_ordered_extent *ordered = bbio->ordered;
735 struct btrfs_inode *inode = bbio->inode;
736 struct btrfs_fs_info *fs_info = inode->root->fs_info;
737 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
738 struct bio *bio = &bbio->bio;
739 struct btrfs_ordered_sum *sums;
740 char *data;
741 struct bvec_iter iter;
742 struct bio_vec bvec;
743 int index;
744 unsigned int blockcount;
745 int i;
746 unsigned nofs_flag;
747
748 nofs_flag = memalloc_nofs_save();
749 sums = kvzalloc(btrfs_ordered_sum_size(fs_info, bio->bi_iter.bi_size),
750 GFP_KERNEL);
751 memalloc_nofs_restore(nofs_flag);
752
753 if (!sums)
754 return BLK_STS_RESOURCE;
755
756 sums->len = bio->bi_iter.bi_size;
757 INIT_LIST_HEAD(&sums->list);
758
759 sums->logical = bio->bi_iter.bi_sector << SECTOR_SHIFT;
760 index = 0;
761
762 shash->tfm = fs_info->csum_shash;
763
764 bio_for_each_segment(bvec, bio, iter) {
765 blockcount = BTRFS_BYTES_TO_BLKS(fs_info,
766 bvec.bv_len + fs_info->sectorsize
767 - 1);
768
769 for (i = 0; i < blockcount; i++) {
770 data = bvec_kmap_local(&bvec);
771 crypto_shash_digest(shash,
772 data + (i * fs_info->sectorsize),
773 fs_info->sectorsize,
774 sums->sums + index);
775 kunmap_local(data);
776 index += fs_info->csum_size;
777 }
778
779 }
780
781 bbio->sums = sums;
782 btrfs_add_ordered_sum(ordered, sums);
783 return 0;
784}
785
786/*
787 * Nodatasum I/O on zoned file systems still requires an btrfs_ordered_sum to
788 * record the updated logical address on Zone Append completion.
789 * Allocate just the structure with an empty sums array here for that case.
790 */
791blk_status_t btrfs_alloc_dummy_sum(struct btrfs_bio *bbio)
792{
793 bbio->sums = kmalloc(sizeof(*bbio->sums), GFP_NOFS);
794 if (!bbio->sums)
795 return BLK_STS_RESOURCE;
796 bbio->sums->len = bbio->bio.bi_iter.bi_size;
797 bbio->sums->logical = bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT;
798 btrfs_add_ordered_sum(bbio->ordered, bbio->sums);
799 return 0;
800}
801
802/*
803 * Remove one checksum overlapping a range.
804 *
805 * This expects the key to describe the csum pointed to by the path, and it
806 * expects the csum to overlap the range [bytenr, len]
807 *
808 * The csum should not be entirely contained in the range and the range should
809 * not be entirely contained in the csum.
810 *
811 * This calls btrfs_truncate_item with the correct args based on the overlap,
812 * and fixes up the key as required.
813 */
814static noinline void truncate_one_csum(struct btrfs_trans_handle *trans,
815 struct btrfs_path *path,
816 struct btrfs_key *key,
817 u64 bytenr, u64 len)
818{
819 struct btrfs_fs_info *fs_info = trans->fs_info;
820 struct extent_buffer *leaf;
821 const u32 csum_size = fs_info->csum_size;
822 u64 csum_end;
823 u64 end_byte = bytenr + len;
824 u32 blocksize_bits = fs_info->sectorsize_bits;
825
826 leaf = path->nodes[0];
827 csum_end = btrfs_item_size(leaf, path->slots[0]) / csum_size;
828 csum_end <<= blocksize_bits;
829 csum_end += key->offset;
830
831 if (key->offset < bytenr && csum_end <= end_byte) {
832 /*
833 * [ bytenr - len ]
834 * [ ]
835 * [csum ]
836 * A simple truncate off the end of the item
837 */
838 u32 new_size = (bytenr - key->offset) >> blocksize_bits;
839 new_size *= csum_size;
840 btrfs_truncate_item(trans, path, new_size, 1);
841 } else if (key->offset >= bytenr && csum_end > end_byte &&
842 end_byte > key->offset) {
843 /*
844 * [ bytenr - len ]
845 * [ ]
846 * [csum ]
847 * we need to truncate from the beginning of the csum
848 */
849 u32 new_size = (csum_end - end_byte) >> blocksize_bits;
850 new_size *= csum_size;
851
852 btrfs_truncate_item(trans, path, new_size, 0);
853
854 key->offset = end_byte;
855 btrfs_set_item_key_safe(trans, path, key);
856 } else {
857 BUG();
858 }
859}
860
861/*
862 * Delete the csum items from the csum tree for a given range of bytes.
863 */
864int btrfs_del_csums(struct btrfs_trans_handle *trans,
865 struct btrfs_root *root, u64 bytenr, u64 len)
866{
867 struct btrfs_fs_info *fs_info = trans->fs_info;
868 struct btrfs_path *path;
869 struct btrfs_key key;
870 u64 end_byte = bytenr + len;
871 u64 csum_end;
872 struct extent_buffer *leaf;
873 int ret = 0;
874 const u32 csum_size = fs_info->csum_size;
875 u32 blocksize_bits = fs_info->sectorsize_bits;
876
877 ASSERT(root->root_key.objectid == BTRFS_CSUM_TREE_OBJECTID ||
878 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
879
880 path = btrfs_alloc_path();
881 if (!path)
882 return -ENOMEM;
883
884 while (1) {
885 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
886 key.offset = end_byte - 1;
887 key.type = BTRFS_EXTENT_CSUM_KEY;
888
889 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
890 if (ret > 0) {
891 ret = 0;
892 if (path->slots[0] == 0)
893 break;
894 path->slots[0]--;
895 } else if (ret < 0) {
896 break;
897 }
898
899 leaf = path->nodes[0];
900 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
901
902 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
903 key.type != BTRFS_EXTENT_CSUM_KEY) {
904 break;
905 }
906
907 if (key.offset >= end_byte)
908 break;
909
910 csum_end = btrfs_item_size(leaf, path->slots[0]) / csum_size;
911 csum_end <<= blocksize_bits;
912 csum_end += key.offset;
913
914 /* this csum ends before we start, we're done */
915 if (csum_end <= bytenr)
916 break;
917
918 /* delete the entire item, it is inside our range */
919 if (key.offset >= bytenr && csum_end <= end_byte) {
920 int del_nr = 1;
921
922 /*
923 * Check how many csum items preceding this one in this
924 * leaf correspond to our range and then delete them all
925 * at once.
926 */
927 if (key.offset > bytenr && path->slots[0] > 0) {
928 int slot = path->slots[0] - 1;
929
930 while (slot >= 0) {
931 struct btrfs_key pk;
932
933 btrfs_item_key_to_cpu(leaf, &pk, slot);
934 if (pk.offset < bytenr ||
935 pk.type != BTRFS_EXTENT_CSUM_KEY ||
936 pk.objectid !=
937 BTRFS_EXTENT_CSUM_OBJECTID)
938 break;
939 path->slots[0] = slot;
940 del_nr++;
941 key.offset = pk.offset;
942 slot--;
943 }
944 }
945 ret = btrfs_del_items(trans, root, path,
946 path->slots[0], del_nr);
947 if (ret)
948 break;
949 if (key.offset == bytenr)
950 break;
951 } else if (key.offset < bytenr && csum_end > end_byte) {
952 unsigned long offset;
953 unsigned long shift_len;
954 unsigned long item_offset;
955 /*
956 * [ bytenr - len ]
957 * [csum ]
958 *
959 * Our bytes are in the middle of the csum,
960 * we need to split this item and insert a new one.
961 *
962 * But we can't drop the path because the
963 * csum could change, get removed, extended etc.
964 *
965 * The trick here is the max size of a csum item leaves
966 * enough room in the tree block for a single
967 * item header. So, we split the item in place,
968 * adding a new header pointing to the existing
969 * bytes. Then we loop around again and we have
970 * a nicely formed csum item that we can neatly
971 * truncate.
972 */
973 offset = (bytenr - key.offset) >> blocksize_bits;
974 offset *= csum_size;
975
976 shift_len = (len >> blocksize_bits) * csum_size;
977
978 item_offset = btrfs_item_ptr_offset(leaf,
979 path->slots[0]);
980
981 memzero_extent_buffer(leaf, item_offset + offset,
982 shift_len);
983 key.offset = bytenr;
984
985 /*
986 * btrfs_split_item returns -EAGAIN when the
987 * item changed size or key
988 */
989 ret = btrfs_split_item(trans, root, path, &key, offset);
990 if (ret && ret != -EAGAIN) {
991 btrfs_abort_transaction(trans, ret);
992 break;
993 }
994 ret = 0;
995
996 key.offset = end_byte - 1;
997 } else {
998 truncate_one_csum(trans, path, &key, bytenr, len);
999 if (key.offset < bytenr)
1000 break;
1001 }
1002 btrfs_release_path(path);
1003 }
1004 btrfs_free_path(path);
1005 return ret;
1006}
1007
1008static int find_next_csum_offset(struct btrfs_root *root,
1009 struct btrfs_path *path,
1010 u64 *next_offset)
1011{
1012 const u32 nritems = btrfs_header_nritems(path->nodes[0]);
1013 struct btrfs_key found_key;
1014 int slot = path->slots[0] + 1;
1015 int ret;
1016
1017 if (nritems == 0 || slot >= nritems) {
1018 ret = btrfs_next_leaf(root, path);
1019 if (ret < 0) {
1020 return ret;
1021 } else if (ret > 0) {
1022 *next_offset = (u64)-1;
1023 return 0;
1024 }
1025 slot = path->slots[0];
1026 }
1027
1028 btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
1029
1030 if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1031 found_key.type != BTRFS_EXTENT_CSUM_KEY)
1032 *next_offset = (u64)-1;
1033 else
1034 *next_offset = found_key.offset;
1035
1036 return 0;
1037}
1038
1039int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
1040 struct btrfs_root *root,
1041 struct btrfs_ordered_sum *sums)
1042{
1043 struct btrfs_fs_info *fs_info = root->fs_info;
1044 struct btrfs_key file_key;
1045 struct btrfs_key found_key;
1046 struct btrfs_path *path;
1047 struct btrfs_csum_item *item;
1048 struct btrfs_csum_item *item_end;
1049 struct extent_buffer *leaf = NULL;
1050 u64 next_offset;
1051 u64 total_bytes = 0;
1052 u64 csum_offset;
1053 u64 bytenr;
1054 u32 ins_size;
1055 int index = 0;
1056 int found_next;
1057 int ret;
1058 const u32 csum_size = fs_info->csum_size;
1059
1060 path = btrfs_alloc_path();
1061 if (!path)
1062 return -ENOMEM;
1063again:
1064 next_offset = (u64)-1;
1065 found_next = 0;
1066 bytenr = sums->logical + total_bytes;
1067 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
1068 file_key.offset = bytenr;
1069 file_key.type = BTRFS_EXTENT_CSUM_KEY;
1070
1071 item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
1072 if (!IS_ERR(item)) {
1073 ret = 0;
1074 leaf = path->nodes[0];
1075 item_end = btrfs_item_ptr(leaf, path->slots[0],
1076 struct btrfs_csum_item);
1077 item_end = (struct btrfs_csum_item *)((char *)item_end +
1078 btrfs_item_size(leaf, path->slots[0]));
1079 goto found;
1080 }
1081 ret = PTR_ERR(item);
1082 if (ret != -EFBIG && ret != -ENOENT)
1083 goto out;
1084
1085 if (ret == -EFBIG) {
1086 u32 item_size;
1087 /* we found one, but it isn't big enough yet */
1088 leaf = path->nodes[0];
1089 item_size = btrfs_item_size(leaf, path->slots[0]);
1090 if ((item_size / csum_size) >=
1091 MAX_CSUM_ITEMS(fs_info, csum_size)) {
1092 /* already at max size, make a new one */
1093 goto insert;
1094 }
1095 } else {
1096 /* We didn't find a csum item, insert one. */
1097 ret = find_next_csum_offset(root, path, &next_offset);
1098 if (ret < 0)
1099 goto out;
1100 found_next = 1;
1101 goto insert;
1102 }
1103
1104 /*
1105 * At this point, we know the tree has a checksum item that ends at an
1106 * offset matching the start of the checksum range we want to insert.
1107 * We try to extend that item as much as possible and then add as many
1108 * checksums to it as they fit.
1109 *
1110 * First check if the leaf has enough free space for at least one
1111 * checksum. If it has go directly to the item extension code, otherwise
1112 * release the path and do a search for insertion before the extension.
1113 */
1114 if (btrfs_leaf_free_space(leaf) >= csum_size) {
1115 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1116 csum_offset = (bytenr - found_key.offset) >>
1117 fs_info->sectorsize_bits;
1118 goto extend_csum;
1119 }
1120
1121 btrfs_release_path(path);
1122 path->search_for_extension = 1;
1123 ret = btrfs_search_slot(trans, root, &file_key, path,
1124 csum_size, 1);
1125 path->search_for_extension = 0;
1126 if (ret < 0)
1127 goto out;
1128
1129 if (ret > 0) {
1130 if (path->slots[0] == 0)
1131 goto insert;
1132 path->slots[0]--;
1133 }
1134
1135 leaf = path->nodes[0];
1136 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1137 csum_offset = (bytenr - found_key.offset) >> fs_info->sectorsize_bits;
1138
1139 if (found_key.type != BTRFS_EXTENT_CSUM_KEY ||
1140 found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1141 csum_offset >= MAX_CSUM_ITEMS(fs_info, csum_size)) {
1142 goto insert;
1143 }
1144
1145extend_csum:
1146 if (csum_offset == btrfs_item_size(leaf, path->slots[0]) /
1147 csum_size) {
1148 int extend_nr;
1149 u64 tmp;
1150 u32 diff;
1151
1152 tmp = sums->len - total_bytes;
1153 tmp >>= fs_info->sectorsize_bits;
1154 WARN_ON(tmp < 1);
1155 extend_nr = max_t(int, 1, tmp);
1156
1157 /*
1158 * A log tree can already have checksum items with a subset of
1159 * the checksums we are trying to log. This can happen after
1160 * doing a sequence of partial writes into prealloc extents and
1161 * fsyncs in between, with a full fsync logging a larger subrange
1162 * of an extent for which a previous fast fsync logged a smaller
1163 * subrange. And this happens in particular due to merging file
1164 * extent items when we complete an ordered extent for a range
1165 * covered by a prealloc extent - this is done at
1166 * btrfs_mark_extent_written().
1167 *
1168 * So if we try to extend the previous checksum item, which has
1169 * a range that ends at the start of the range we want to insert,
1170 * make sure we don't extend beyond the start offset of the next
1171 * checksum item. If we are at the last item in the leaf, then
1172 * forget the optimization of extending and add a new checksum
1173 * item - it is not worth the complexity of releasing the path,
1174 * getting the first key for the next leaf, repeat the btree
1175 * search, etc, because log trees are temporary anyway and it
1176 * would only save a few bytes of leaf space.
1177 */
1178 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
1179 if (path->slots[0] + 1 >=
1180 btrfs_header_nritems(path->nodes[0])) {
1181 ret = find_next_csum_offset(root, path, &next_offset);
1182 if (ret < 0)
1183 goto out;
1184 found_next = 1;
1185 goto insert;
1186 }
1187
1188 ret = find_next_csum_offset(root, path, &next_offset);
1189 if (ret < 0)
1190 goto out;
1191
1192 tmp = (next_offset - bytenr) >> fs_info->sectorsize_bits;
1193 if (tmp <= INT_MAX)
1194 extend_nr = min_t(int, extend_nr, tmp);
1195 }
1196
1197 diff = (csum_offset + extend_nr) * csum_size;
1198 diff = min(diff,
1199 MAX_CSUM_ITEMS(fs_info, csum_size) * csum_size);
1200
1201 diff = diff - btrfs_item_size(leaf, path->slots[0]);
1202 diff = min_t(u32, btrfs_leaf_free_space(leaf), diff);
1203 diff /= csum_size;
1204 diff *= csum_size;
1205
1206 btrfs_extend_item(trans, path, diff);
1207 ret = 0;
1208 goto csum;
1209 }
1210
1211insert:
1212 btrfs_release_path(path);
1213 csum_offset = 0;
1214 if (found_next) {
1215 u64 tmp;
1216
1217 tmp = sums->len - total_bytes;
1218 tmp >>= fs_info->sectorsize_bits;
1219 tmp = min(tmp, (next_offset - file_key.offset) >>
1220 fs_info->sectorsize_bits);
1221
1222 tmp = max_t(u64, 1, tmp);
1223 tmp = min_t(u64, tmp, MAX_CSUM_ITEMS(fs_info, csum_size));
1224 ins_size = csum_size * tmp;
1225 } else {
1226 ins_size = csum_size;
1227 }
1228 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
1229 ins_size);
1230 if (ret < 0)
1231 goto out;
1232 if (WARN_ON(ret != 0))
1233 goto out;
1234 leaf = path->nodes[0];
1235csum:
1236 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
1237 item_end = (struct btrfs_csum_item *)((unsigned char *)item +
1238 btrfs_item_size(leaf, path->slots[0]));
1239 item = (struct btrfs_csum_item *)((unsigned char *)item +
1240 csum_offset * csum_size);
1241found:
1242 ins_size = (u32)(sums->len - total_bytes) >> fs_info->sectorsize_bits;
1243 ins_size *= csum_size;
1244 ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item,
1245 ins_size);
1246 write_extent_buffer(leaf, sums->sums + index, (unsigned long)item,
1247 ins_size);
1248
1249 index += ins_size;
1250 ins_size /= csum_size;
1251 total_bytes += ins_size * fs_info->sectorsize;
1252
1253 btrfs_mark_buffer_dirty(trans, path->nodes[0]);
1254 if (total_bytes < sums->len) {
1255 btrfs_release_path(path);
1256 cond_resched();
1257 goto again;
1258 }
1259out:
1260 btrfs_free_path(path);
1261 return ret;
1262}
1263
1264void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode,
1265 const struct btrfs_path *path,
1266 struct btrfs_file_extent_item *fi,
1267 struct extent_map *em)
1268{
1269 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1270 struct btrfs_root *root = inode->root;
1271 struct extent_buffer *leaf = path->nodes[0];
1272 const int slot = path->slots[0];
1273 struct btrfs_key key;
1274 u64 extent_start, extent_end;
1275 u64 bytenr;
1276 u8 type = btrfs_file_extent_type(leaf, fi);
1277 int compress_type = btrfs_file_extent_compression(leaf, fi);
1278
1279 btrfs_item_key_to_cpu(leaf, &key, slot);
1280 extent_start = key.offset;
1281 extent_end = btrfs_file_extent_end(path);
1282 em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1283 em->generation = btrfs_file_extent_generation(leaf, fi);
1284 if (type == BTRFS_FILE_EXTENT_REG ||
1285 type == BTRFS_FILE_EXTENT_PREALLOC) {
1286 em->start = extent_start;
1287 em->len = extent_end - extent_start;
1288 em->orig_start = extent_start -
1289 btrfs_file_extent_offset(leaf, fi);
1290 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
1291 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1292 if (bytenr == 0) {
1293 em->block_start = EXTENT_MAP_HOLE;
1294 return;
1295 }
1296 if (compress_type != BTRFS_COMPRESS_NONE) {
1297 extent_map_set_compression(em, compress_type);
1298 em->block_start = bytenr;
1299 em->block_len = em->orig_block_len;
1300 } else {
1301 bytenr += btrfs_file_extent_offset(leaf, fi);
1302 em->block_start = bytenr;
1303 em->block_len = em->len;
1304 if (type == BTRFS_FILE_EXTENT_PREALLOC)
1305 em->flags |= EXTENT_FLAG_PREALLOC;
1306 }
1307 } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1308 em->block_start = EXTENT_MAP_INLINE;
1309 em->start = extent_start;
1310 em->len = extent_end - extent_start;
1311 /*
1312 * Initialize orig_start and block_len with the same values
1313 * as in inode.c:btrfs_get_extent().
1314 */
1315 em->orig_start = EXTENT_MAP_HOLE;
1316 em->block_len = (u64)-1;
1317 extent_map_set_compression(em, compress_type);
1318 } else {
1319 btrfs_err(fs_info,
1320 "unknown file extent item type %d, inode %llu, offset %llu, "
1321 "root %llu", type, btrfs_ino(inode), extent_start,
1322 root->root_key.objectid);
1323 }
1324}
1325
1326/*
1327 * Returns the end offset (non inclusive) of the file extent item the given path
1328 * points to. If it points to an inline extent, the returned offset is rounded
1329 * up to the sector size.
1330 */
1331u64 btrfs_file_extent_end(const struct btrfs_path *path)
1332{
1333 const struct extent_buffer *leaf = path->nodes[0];
1334 const int slot = path->slots[0];
1335 struct btrfs_file_extent_item *fi;
1336 struct btrfs_key key;
1337 u64 end;
1338
1339 btrfs_item_key_to_cpu(leaf, &key, slot);
1340 ASSERT(key.type == BTRFS_EXTENT_DATA_KEY);
1341 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1342
1343 if (btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_INLINE) {
1344 end = btrfs_file_extent_ram_bytes(leaf, fi);
1345 end = ALIGN(key.offset + end, leaf->fs_info->sectorsize);
1346 } else {
1347 end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1348 }
1349
1350 return end;
1351}
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}