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 "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}