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/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19#include <linux/bio.h>
20#include <linux/slab.h>
21#include <linux/pagemap.h>
22#include <linux/highmem.h>
23#include "ctree.h"
24#include "disk-io.h"
25#include "transaction.h"
26#include "print-tree.h"
27
28#define MAX_CSUM_ITEMS(r, size) ((((BTRFS_LEAF_DATA_SIZE(r) - \
29 sizeof(struct btrfs_item) * 2) / \
30 size) - 1))
31
32#define MAX_ORDERED_SUM_BYTES(r) ((PAGE_SIZE - \
33 sizeof(struct btrfs_ordered_sum)) / \
34 sizeof(struct btrfs_sector_sum) * \
35 (r)->sectorsize - (r)->sectorsize)
36
37int btrfs_insert_file_extent(struct btrfs_trans_handle *trans,
38 struct btrfs_root *root,
39 u64 objectid, u64 pos,
40 u64 disk_offset, u64 disk_num_bytes,
41 u64 num_bytes, u64 offset, u64 ram_bytes,
42 u8 compression, u8 encryption, u16 other_encoding)
43{
44 int ret = 0;
45 struct btrfs_file_extent_item *item;
46 struct btrfs_key file_key;
47 struct btrfs_path *path;
48 struct extent_buffer *leaf;
49
50 path = btrfs_alloc_path();
51 if (!path)
52 return -ENOMEM;
53 file_key.objectid = objectid;
54 file_key.offset = pos;
55 btrfs_set_key_type(&file_key, BTRFS_EXTENT_DATA_KEY);
56
57 path->leave_spinning = 1;
58 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
59 sizeof(*item));
60 if (ret < 0)
61 goto out;
62 BUG_ON(ret);
63 leaf = path->nodes[0];
64 item = btrfs_item_ptr(leaf, path->slots[0],
65 struct btrfs_file_extent_item);
66 btrfs_set_file_extent_disk_bytenr(leaf, item, disk_offset);
67 btrfs_set_file_extent_disk_num_bytes(leaf, item, disk_num_bytes);
68 btrfs_set_file_extent_offset(leaf, item, offset);
69 btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
70 btrfs_set_file_extent_ram_bytes(leaf, item, ram_bytes);
71 btrfs_set_file_extent_generation(leaf, item, trans->transid);
72 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
73 btrfs_set_file_extent_compression(leaf, item, compression);
74 btrfs_set_file_extent_encryption(leaf, item, encryption);
75 btrfs_set_file_extent_other_encoding(leaf, item, other_encoding);
76
77 btrfs_mark_buffer_dirty(leaf);
78out:
79 btrfs_free_path(path);
80 return ret;
81}
82
83struct btrfs_csum_item *btrfs_lookup_csum(struct btrfs_trans_handle *trans,
84 struct btrfs_root *root,
85 struct btrfs_path *path,
86 u64 bytenr, int cow)
87{
88 int ret;
89 struct btrfs_key file_key;
90 struct btrfs_key found_key;
91 struct btrfs_csum_item *item;
92 struct extent_buffer *leaf;
93 u64 csum_offset = 0;
94 u16 csum_size =
95 btrfs_super_csum_size(&root->fs_info->super_copy);
96 int csums_in_item;
97
98 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
99 file_key.offset = bytenr;
100 btrfs_set_key_type(&file_key, BTRFS_EXTENT_CSUM_KEY);
101 ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
102 if (ret < 0)
103 goto fail;
104 leaf = path->nodes[0];
105 if (ret > 0) {
106 ret = 1;
107 if (path->slots[0] == 0)
108 goto fail;
109 path->slots[0]--;
110 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
111 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_CSUM_KEY)
112 goto fail;
113
114 csum_offset = (bytenr - found_key.offset) >>
115 root->fs_info->sb->s_blocksize_bits;
116 csums_in_item = btrfs_item_size_nr(leaf, path->slots[0]);
117 csums_in_item /= csum_size;
118
119 if (csum_offset >= csums_in_item) {
120 ret = -EFBIG;
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
134
135int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
136 struct btrfs_root *root,
137 struct btrfs_path *path, u64 objectid,
138 u64 offset, int mod)
139{
140 int ret;
141 struct btrfs_key file_key;
142 int ins_len = mod < 0 ? -1 : 0;
143 int cow = mod != 0;
144
145 file_key.objectid = objectid;
146 file_key.offset = offset;
147 btrfs_set_key_type(&file_key, BTRFS_EXTENT_DATA_KEY);
148 ret = btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
149 return ret;
150}
151
152
153static int __btrfs_lookup_bio_sums(struct btrfs_root *root,
154 struct inode *inode, struct bio *bio,
155 u64 logical_offset, u32 *dst, int dio)
156{
157 u32 sum;
158 struct bio_vec *bvec = bio->bi_io_vec;
159 int bio_index = 0;
160 u64 offset = 0;
161 u64 item_start_offset = 0;
162 u64 item_last_offset = 0;
163 u64 disk_bytenr;
164 u32 diff;
165 u16 csum_size =
166 btrfs_super_csum_size(&root->fs_info->super_copy);
167 int ret;
168 struct btrfs_path *path;
169 struct btrfs_csum_item *item = NULL;
170 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
171
172 path = btrfs_alloc_path();
173 if (!path)
174 return -ENOMEM;
175 if (bio->bi_size > PAGE_CACHE_SIZE * 8)
176 path->reada = 2;
177
178 WARN_ON(bio->bi_vcnt <= 0);
179
180 /*
181 * the free space stuff is only read when it hasn't been
182 * updated in the current transaction. So, we can safely
183 * read from the commit root and sidestep a nasty deadlock
184 * between reading the free space cache and updating the csum tree.
185 */
186 if (btrfs_is_free_space_inode(root, inode)) {
187 path->search_commit_root = 1;
188 path->skip_locking = 1;
189 }
190
191 disk_bytenr = (u64)bio->bi_sector << 9;
192 if (dio)
193 offset = logical_offset;
194 while (bio_index < bio->bi_vcnt) {
195 if (!dio)
196 offset = page_offset(bvec->bv_page) + bvec->bv_offset;
197 ret = btrfs_find_ordered_sum(inode, offset, disk_bytenr, &sum);
198 if (ret == 0)
199 goto found;
200
201 if (!item || disk_bytenr < item_start_offset ||
202 disk_bytenr >= item_last_offset) {
203 struct btrfs_key found_key;
204 u32 item_size;
205
206 if (item)
207 btrfs_release_path(path);
208 item = btrfs_lookup_csum(NULL, root->fs_info->csum_root,
209 path, disk_bytenr, 0);
210 if (IS_ERR(item)) {
211 ret = PTR_ERR(item);
212 if (ret == -ENOENT || ret == -EFBIG)
213 ret = 0;
214 sum = 0;
215 if (BTRFS_I(inode)->root->root_key.objectid ==
216 BTRFS_DATA_RELOC_TREE_OBJECTID) {
217 set_extent_bits(io_tree, offset,
218 offset + bvec->bv_len - 1,
219 EXTENT_NODATASUM, GFP_NOFS);
220 } else {
221 printk(KERN_INFO "btrfs no csum found "
222 "for inode %llu start %llu\n",
223 (unsigned long long)
224 btrfs_ino(inode),
225 (unsigned long long)offset);
226 }
227 item = NULL;
228 btrfs_release_path(path);
229 goto found;
230 }
231 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
232 path->slots[0]);
233
234 item_start_offset = found_key.offset;
235 item_size = btrfs_item_size_nr(path->nodes[0],
236 path->slots[0]);
237 item_last_offset = item_start_offset +
238 (item_size / csum_size) *
239 root->sectorsize;
240 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
241 struct btrfs_csum_item);
242 }
243 /*
244 * this byte range must be able to fit inside
245 * a single leaf so it will also fit inside a u32
246 */
247 diff = disk_bytenr - item_start_offset;
248 diff = diff / root->sectorsize;
249 diff = diff * csum_size;
250
251 read_extent_buffer(path->nodes[0], &sum,
252 ((unsigned long)item) + diff,
253 csum_size);
254found:
255 if (dst)
256 *dst++ = sum;
257 else
258 set_state_private(io_tree, offset, sum);
259 disk_bytenr += bvec->bv_len;
260 offset += bvec->bv_len;
261 bio_index++;
262 bvec++;
263 }
264 btrfs_free_path(path);
265 return 0;
266}
267
268int btrfs_lookup_bio_sums(struct btrfs_root *root, struct inode *inode,
269 struct bio *bio, u32 *dst)
270{
271 return __btrfs_lookup_bio_sums(root, inode, bio, 0, dst, 0);
272}
273
274int btrfs_lookup_bio_sums_dio(struct btrfs_root *root, struct inode *inode,
275 struct bio *bio, u64 offset, u32 *dst)
276{
277 return __btrfs_lookup_bio_sums(root, inode, bio, offset, dst, 1);
278}
279
280int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
281 struct list_head *list, int search_commit)
282{
283 struct btrfs_key key;
284 struct btrfs_path *path;
285 struct extent_buffer *leaf;
286 struct btrfs_ordered_sum *sums;
287 struct btrfs_sector_sum *sector_sum;
288 struct btrfs_csum_item *item;
289 unsigned long offset;
290 int ret;
291 size_t size;
292 u64 csum_end;
293 u16 csum_size = btrfs_super_csum_size(&root->fs_info->super_copy);
294
295 path = btrfs_alloc_path();
296 if (!path)
297 return -ENOMEM;
298
299 if (search_commit) {
300 path->skip_locking = 1;
301 path->reada = 2;
302 path->search_commit_root = 1;
303 }
304
305 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
306 key.offset = start;
307 key.type = BTRFS_EXTENT_CSUM_KEY;
308
309 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
310 if (ret < 0)
311 goto fail;
312 if (ret > 0 && path->slots[0] > 0) {
313 leaf = path->nodes[0];
314 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
315 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
316 key.type == BTRFS_EXTENT_CSUM_KEY) {
317 offset = (start - key.offset) >>
318 root->fs_info->sb->s_blocksize_bits;
319 if (offset * csum_size <
320 btrfs_item_size_nr(leaf, path->slots[0] - 1))
321 path->slots[0]--;
322 }
323 }
324
325 while (start <= end) {
326 leaf = path->nodes[0];
327 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
328 ret = btrfs_next_leaf(root, path);
329 if (ret < 0)
330 goto fail;
331 if (ret > 0)
332 break;
333 leaf = path->nodes[0];
334 }
335
336 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
337 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
338 key.type != BTRFS_EXTENT_CSUM_KEY)
339 break;
340
341 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
342 if (key.offset > end)
343 break;
344
345 if (key.offset > start)
346 start = key.offset;
347
348 size = btrfs_item_size_nr(leaf, path->slots[0]);
349 csum_end = key.offset + (size / csum_size) * root->sectorsize;
350 if (csum_end <= start) {
351 path->slots[0]++;
352 continue;
353 }
354
355 csum_end = min(csum_end, end + 1);
356 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
357 struct btrfs_csum_item);
358 while (start < csum_end) {
359 size = min_t(size_t, csum_end - start,
360 MAX_ORDERED_SUM_BYTES(root));
361 sums = kzalloc(btrfs_ordered_sum_size(root, size),
362 GFP_NOFS);
363 BUG_ON(!sums);
364
365 sector_sum = sums->sums;
366 sums->bytenr = start;
367 sums->len = size;
368
369 offset = (start - key.offset) >>
370 root->fs_info->sb->s_blocksize_bits;
371 offset *= csum_size;
372
373 while (size > 0) {
374 read_extent_buffer(path->nodes[0],
375 §or_sum->sum,
376 ((unsigned long)item) +
377 offset, csum_size);
378 sector_sum->bytenr = start;
379
380 size -= root->sectorsize;
381 start += root->sectorsize;
382 offset += csum_size;
383 sector_sum++;
384 }
385 list_add_tail(&sums->list, list);
386 }
387 path->slots[0]++;
388 }
389 ret = 0;
390fail:
391 btrfs_free_path(path);
392 return ret;
393}
394
395int btrfs_csum_one_bio(struct btrfs_root *root, struct inode *inode,
396 struct bio *bio, u64 file_start, int contig)
397{
398 struct btrfs_ordered_sum *sums;
399 struct btrfs_sector_sum *sector_sum;
400 struct btrfs_ordered_extent *ordered;
401 char *data;
402 struct bio_vec *bvec = bio->bi_io_vec;
403 int bio_index = 0;
404 unsigned long total_bytes = 0;
405 unsigned long this_sum_bytes = 0;
406 u64 offset;
407 u64 disk_bytenr;
408
409 WARN_ON(bio->bi_vcnt <= 0);
410 sums = kzalloc(btrfs_ordered_sum_size(root, bio->bi_size), GFP_NOFS);
411 if (!sums)
412 return -ENOMEM;
413
414 sector_sum = sums->sums;
415 disk_bytenr = (u64)bio->bi_sector << 9;
416 sums->len = bio->bi_size;
417 INIT_LIST_HEAD(&sums->list);
418
419 if (contig)
420 offset = file_start;
421 else
422 offset = page_offset(bvec->bv_page) + bvec->bv_offset;
423
424 ordered = btrfs_lookup_ordered_extent(inode, offset);
425 BUG_ON(!ordered);
426 sums->bytenr = ordered->start;
427
428 while (bio_index < bio->bi_vcnt) {
429 if (!contig)
430 offset = page_offset(bvec->bv_page) + bvec->bv_offset;
431
432 if (!contig && (offset >= ordered->file_offset + ordered->len ||
433 offset < ordered->file_offset)) {
434 unsigned long bytes_left;
435 sums->len = this_sum_bytes;
436 this_sum_bytes = 0;
437 btrfs_add_ordered_sum(inode, ordered, sums);
438 btrfs_put_ordered_extent(ordered);
439
440 bytes_left = bio->bi_size - total_bytes;
441
442 sums = kzalloc(btrfs_ordered_sum_size(root, bytes_left),
443 GFP_NOFS);
444 BUG_ON(!sums);
445 sector_sum = sums->sums;
446 sums->len = bytes_left;
447 ordered = btrfs_lookup_ordered_extent(inode, offset);
448 BUG_ON(!ordered);
449 sums->bytenr = ordered->start;
450 }
451
452 data = kmap_atomic(bvec->bv_page, KM_USER0);
453 sector_sum->sum = ~(u32)0;
454 sector_sum->sum = btrfs_csum_data(root,
455 data + bvec->bv_offset,
456 sector_sum->sum,
457 bvec->bv_len);
458 kunmap_atomic(data, KM_USER0);
459 btrfs_csum_final(sector_sum->sum,
460 (char *)§or_sum->sum);
461 sector_sum->bytenr = disk_bytenr;
462
463 sector_sum++;
464 bio_index++;
465 total_bytes += bvec->bv_len;
466 this_sum_bytes += bvec->bv_len;
467 disk_bytenr += bvec->bv_len;
468 offset += bvec->bv_len;
469 bvec++;
470 }
471 this_sum_bytes = 0;
472 btrfs_add_ordered_sum(inode, ordered, sums);
473 btrfs_put_ordered_extent(ordered);
474 return 0;
475}
476
477/*
478 * helper function for csum removal, this expects the
479 * key to describe the csum pointed to by the path, and it expects
480 * the csum to overlap the range [bytenr, len]
481 *
482 * The csum should not be entirely contained in the range and the
483 * range should not be entirely contained in the csum.
484 *
485 * This calls btrfs_truncate_item with the correct args based on the
486 * overlap, and fixes up the key as required.
487 */
488static noinline int truncate_one_csum(struct btrfs_trans_handle *trans,
489 struct btrfs_root *root,
490 struct btrfs_path *path,
491 struct btrfs_key *key,
492 u64 bytenr, u64 len)
493{
494 struct extent_buffer *leaf;
495 u16 csum_size =
496 btrfs_super_csum_size(&root->fs_info->super_copy);
497 u64 csum_end;
498 u64 end_byte = bytenr + len;
499 u32 blocksize_bits = root->fs_info->sb->s_blocksize_bits;
500 int ret;
501
502 leaf = path->nodes[0];
503 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
504 csum_end <<= root->fs_info->sb->s_blocksize_bits;
505 csum_end += key->offset;
506
507 if (key->offset < bytenr && csum_end <= end_byte) {
508 /*
509 * [ bytenr - len ]
510 * [ ]
511 * [csum ]
512 * A simple truncate off the end of the item
513 */
514 u32 new_size = (bytenr - key->offset) >> blocksize_bits;
515 new_size *= csum_size;
516 ret = btrfs_truncate_item(trans, root, path, new_size, 1);
517 } else if (key->offset >= bytenr && csum_end > end_byte &&
518 end_byte > key->offset) {
519 /*
520 * [ bytenr - len ]
521 * [ ]
522 * [csum ]
523 * we need to truncate from the beginning of the csum
524 */
525 u32 new_size = (csum_end - end_byte) >> blocksize_bits;
526 new_size *= csum_size;
527
528 ret = btrfs_truncate_item(trans, root, path, new_size, 0);
529
530 key->offset = end_byte;
531 ret = btrfs_set_item_key_safe(trans, root, path, key);
532 BUG_ON(ret);
533 } else {
534 BUG();
535 }
536 return 0;
537}
538
539/*
540 * deletes the csum items from the csum tree for a given
541 * range of bytes.
542 */
543int btrfs_del_csums(struct btrfs_trans_handle *trans,
544 struct btrfs_root *root, u64 bytenr, u64 len)
545{
546 struct btrfs_path *path;
547 struct btrfs_key key;
548 u64 end_byte = bytenr + len;
549 u64 csum_end;
550 struct extent_buffer *leaf;
551 int ret;
552 u16 csum_size =
553 btrfs_super_csum_size(&root->fs_info->super_copy);
554 int blocksize_bits = root->fs_info->sb->s_blocksize_bits;
555
556 root = root->fs_info->csum_root;
557
558 path = btrfs_alloc_path();
559 if (!path)
560 return -ENOMEM;
561
562 while (1) {
563 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
564 key.offset = end_byte - 1;
565 key.type = BTRFS_EXTENT_CSUM_KEY;
566
567 path->leave_spinning = 1;
568 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
569 if (ret > 0) {
570 if (path->slots[0] == 0)
571 break;
572 path->slots[0]--;
573 } else if (ret < 0) {
574 break;
575 }
576
577 leaf = path->nodes[0];
578 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
579
580 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
581 key.type != BTRFS_EXTENT_CSUM_KEY) {
582 break;
583 }
584
585 if (key.offset >= end_byte)
586 break;
587
588 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
589 csum_end <<= blocksize_bits;
590 csum_end += key.offset;
591
592 /* this csum ends before we start, we're done */
593 if (csum_end <= bytenr)
594 break;
595
596 /* delete the entire item, it is inside our range */
597 if (key.offset >= bytenr && csum_end <= end_byte) {
598 ret = btrfs_del_item(trans, root, path);
599 if (ret)
600 goto out;
601 if (key.offset == bytenr)
602 break;
603 } else if (key.offset < bytenr && csum_end > end_byte) {
604 unsigned long offset;
605 unsigned long shift_len;
606 unsigned long item_offset;
607 /*
608 * [ bytenr - len ]
609 * [csum ]
610 *
611 * Our bytes are in the middle of the csum,
612 * we need to split this item and insert a new one.
613 *
614 * But we can't drop the path because the
615 * csum could change, get removed, extended etc.
616 *
617 * The trick here is the max size of a csum item leaves
618 * enough room in the tree block for a single
619 * item header. So, we split the item in place,
620 * adding a new header pointing to the existing
621 * bytes. Then we loop around again and we have
622 * a nicely formed csum item that we can neatly
623 * truncate.
624 */
625 offset = (bytenr - key.offset) >> blocksize_bits;
626 offset *= csum_size;
627
628 shift_len = (len >> blocksize_bits) * csum_size;
629
630 item_offset = btrfs_item_ptr_offset(leaf,
631 path->slots[0]);
632
633 memset_extent_buffer(leaf, 0, item_offset + offset,
634 shift_len);
635 key.offset = bytenr;
636
637 /*
638 * btrfs_split_item returns -EAGAIN when the
639 * item changed size or key
640 */
641 ret = btrfs_split_item(trans, root, path, &key, offset);
642 BUG_ON(ret && ret != -EAGAIN);
643
644 key.offset = end_byte - 1;
645 } else {
646 ret = truncate_one_csum(trans, root, path,
647 &key, bytenr, len);
648 BUG_ON(ret);
649 if (key.offset < bytenr)
650 break;
651 }
652 btrfs_release_path(path);
653 }
654 ret = 0;
655out:
656 btrfs_free_path(path);
657 return ret;
658}
659
660int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
661 struct btrfs_root *root,
662 struct btrfs_ordered_sum *sums)
663{
664 u64 bytenr;
665 int ret;
666 struct btrfs_key file_key;
667 struct btrfs_key found_key;
668 u64 next_offset;
669 u64 total_bytes = 0;
670 int found_next;
671 struct btrfs_path *path;
672 struct btrfs_csum_item *item;
673 struct btrfs_csum_item *item_end;
674 struct extent_buffer *leaf = NULL;
675 u64 csum_offset;
676 struct btrfs_sector_sum *sector_sum;
677 u32 nritems;
678 u32 ins_size;
679 u16 csum_size =
680 btrfs_super_csum_size(&root->fs_info->super_copy);
681
682 path = btrfs_alloc_path();
683 if (!path)
684 return -ENOMEM;
685
686 sector_sum = sums->sums;
687again:
688 next_offset = (u64)-1;
689 found_next = 0;
690 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
691 file_key.offset = sector_sum->bytenr;
692 bytenr = sector_sum->bytenr;
693 btrfs_set_key_type(&file_key, BTRFS_EXTENT_CSUM_KEY);
694
695 item = btrfs_lookup_csum(trans, root, path, sector_sum->bytenr, 1);
696 if (!IS_ERR(item)) {
697 leaf = path->nodes[0];
698 ret = 0;
699 goto found;
700 }
701 ret = PTR_ERR(item);
702 if (ret != -EFBIG && ret != -ENOENT)
703 goto fail_unlock;
704
705 if (ret == -EFBIG) {
706 u32 item_size;
707 /* we found one, but it isn't big enough yet */
708 leaf = path->nodes[0];
709 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
710 if ((item_size / csum_size) >=
711 MAX_CSUM_ITEMS(root, csum_size)) {
712 /* already at max size, make a new one */
713 goto insert;
714 }
715 } else {
716 int slot = path->slots[0] + 1;
717 /* we didn't find a csum item, insert one */
718 nritems = btrfs_header_nritems(path->nodes[0]);
719 if (path->slots[0] >= nritems - 1) {
720 ret = btrfs_next_leaf(root, path);
721 if (ret == 1)
722 found_next = 1;
723 if (ret != 0)
724 goto insert;
725 slot = 0;
726 }
727 btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
728 if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
729 found_key.type != BTRFS_EXTENT_CSUM_KEY) {
730 found_next = 1;
731 goto insert;
732 }
733 next_offset = found_key.offset;
734 found_next = 1;
735 goto insert;
736 }
737
738 /*
739 * at this point, we know the tree has an item, but it isn't big
740 * enough yet to put our csum in. Grow it
741 */
742 btrfs_release_path(path);
743 ret = btrfs_search_slot(trans, root, &file_key, path,
744 csum_size, 1);
745 if (ret < 0)
746 goto fail_unlock;
747
748 if (ret > 0) {
749 if (path->slots[0] == 0)
750 goto insert;
751 path->slots[0]--;
752 }
753
754 leaf = path->nodes[0];
755 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
756 csum_offset = (bytenr - found_key.offset) >>
757 root->fs_info->sb->s_blocksize_bits;
758
759 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_CSUM_KEY ||
760 found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
761 csum_offset >= MAX_CSUM_ITEMS(root, csum_size)) {
762 goto insert;
763 }
764
765 if (csum_offset >= btrfs_item_size_nr(leaf, path->slots[0]) /
766 csum_size) {
767 u32 diff = (csum_offset + 1) * csum_size;
768
769 /*
770 * is the item big enough already? we dropped our lock
771 * before and need to recheck
772 */
773 if (diff < btrfs_item_size_nr(leaf, path->slots[0]))
774 goto csum;
775
776 diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
777 if (diff != csum_size)
778 goto insert;
779
780 ret = btrfs_extend_item(trans, root, path, diff);
781 goto csum;
782 }
783
784insert:
785 btrfs_release_path(path);
786 csum_offset = 0;
787 if (found_next) {
788 u64 tmp = total_bytes + root->sectorsize;
789 u64 next_sector = sector_sum->bytenr;
790 struct btrfs_sector_sum *next = sector_sum + 1;
791
792 while (tmp < sums->len) {
793 if (next_sector + root->sectorsize != next->bytenr)
794 break;
795 tmp += root->sectorsize;
796 next_sector = next->bytenr;
797 next++;
798 }
799 tmp = min(tmp, next_offset - file_key.offset);
800 tmp >>= root->fs_info->sb->s_blocksize_bits;
801 tmp = max((u64)1, tmp);
802 tmp = min(tmp, (u64)MAX_CSUM_ITEMS(root, csum_size));
803 ins_size = csum_size * tmp;
804 } else {
805 ins_size = csum_size;
806 }
807 path->leave_spinning = 1;
808 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
809 ins_size);
810 path->leave_spinning = 0;
811 if (ret < 0)
812 goto fail_unlock;
813 if (ret != 0) {
814 WARN_ON(1);
815 goto fail_unlock;
816 }
817csum:
818 leaf = path->nodes[0];
819 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
820 ret = 0;
821 item = (struct btrfs_csum_item *)((unsigned char *)item +
822 csum_offset * csum_size);
823found:
824 item_end = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
825 item_end = (struct btrfs_csum_item *)((unsigned char *)item_end +
826 btrfs_item_size_nr(leaf, path->slots[0]));
827next_sector:
828
829 write_extent_buffer(leaf, §or_sum->sum, (unsigned long)item, csum_size);
830
831 total_bytes += root->sectorsize;
832 sector_sum++;
833 if (total_bytes < sums->len) {
834 item = (struct btrfs_csum_item *)((char *)item +
835 csum_size);
836 if (item < item_end && bytenr + PAGE_CACHE_SIZE ==
837 sector_sum->bytenr) {
838 bytenr = sector_sum->bytenr;
839 goto next_sector;
840 }
841 }
842
843 btrfs_mark_buffer_dirty(path->nodes[0]);
844 if (total_bytes < sums->len) {
845 btrfs_release_path(path);
846 cond_resched();
847 goto again;
848 }
849out:
850 btrfs_free_path(path);
851 return ret;
852
853fail_unlock:
854 goto out;
855}