Loading...
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_CSUM_ITEMS(r, size) (min(__MAX_CSUM_ITEMS(r, size), PAGE_CACHE_SIZE))
33
34#define MAX_ORDERED_SUM_BYTES(r) ((PAGE_SIZE - \
35 sizeof(struct btrfs_ordered_sum)) / \
36 sizeof(struct btrfs_sector_sum) * \
37 (r)->sectorsize - (r)->sectorsize)
38
39int btrfs_insert_file_extent(struct btrfs_trans_handle *trans,
40 struct btrfs_root *root,
41 u64 objectid, u64 pos,
42 u64 disk_offset, u64 disk_num_bytes,
43 u64 num_bytes, u64 offset, u64 ram_bytes,
44 u8 compression, u8 encryption, u16 other_encoding)
45{
46 int ret = 0;
47 struct btrfs_file_extent_item *item;
48 struct btrfs_key file_key;
49 struct btrfs_path *path;
50 struct extent_buffer *leaf;
51
52 path = btrfs_alloc_path();
53 if (!path)
54 return -ENOMEM;
55 file_key.objectid = objectid;
56 file_key.offset = pos;
57 btrfs_set_key_type(&file_key, BTRFS_EXTENT_DATA_KEY);
58
59 path->leave_spinning = 1;
60 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
61 sizeof(*item));
62 if (ret < 0)
63 goto out;
64 BUG_ON(ret); /* Can't happen */
65 leaf = path->nodes[0];
66 item = btrfs_item_ptr(leaf, path->slots[0],
67 struct btrfs_file_extent_item);
68 btrfs_set_file_extent_disk_bytenr(leaf, item, disk_offset);
69 btrfs_set_file_extent_disk_num_bytes(leaf, item, disk_num_bytes);
70 btrfs_set_file_extent_offset(leaf, item, offset);
71 btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
72 btrfs_set_file_extent_ram_bytes(leaf, item, ram_bytes);
73 btrfs_set_file_extent_generation(leaf, item, trans->transid);
74 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
75 btrfs_set_file_extent_compression(leaf, item, compression);
76 btrfs_set_file_extent_encryption(leaf, item, encryption);
77 btrfs_set_file_extent_other_encoding(leaf, item, other_encoding);
78
79 btrfs_mark_buffer_dirty(leaf);
80out:
81 btrfs_free_path(path);
82 return ret;
83}
84
85struct btrfs_csum_item *btrfs_lookup_csum(struct btrfs_trans_handle *trans,
86 struct btrfs_root *root,
87 struct btrfs_path *path,
88 u64 bytenr, int cow)
89{
90 int ret;
91 struct btrfs_key file_key;
92 struct btrfs_key found_key;
93 struct btrfs_csum_item *item;
94 struct extent_buffer *leaf;
95 u64 csum_offset = 0;
96 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
97 int csums_in_item;
98
99 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
100 file_key.offset = bytenr;
101 btrfs_set_key_type(&file_key, BTRFS_EXTENT_CSUM_KEY);
102 ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
103 if (ret < 0)
104 goto fail;
105 leaf = path->nodes[0];
106 if (ret > 0) {
107 ret = 1;
108 if (path->slots[0] == 0)
109 goto fail;
110 path->slots[0]--;
111 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
112 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_CSUM_KEY)
113 goto fail;
114
115 csum_offset = (bytenr - found_key.offset) >>
116 root->fs_info->sb->s_blocksize_bits;
117 csums_in_item = btrfs_item_size_nr(leaf, path->slots[0]);
118 csums_in_item /= csum_size;
119
120 if (csum_offset >= csums_in_item) {
121 ret = -EFBIG;
122 goto fail;
123 }
124 }
125 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
126 item = (struct btrfs_csum_item *)((unsigned char *)item +
127 csum_offset * csum_size);
128 return item;
129fail:
130 if (ret > 0)
131 ret = -ENOENT;
132 return ERR_PTR(ret);
133}
134
135
136int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
137 struct btrfs_root *root,
138 struct btrfs_path *path, u64 objectid,
139 u64 offset, int mod)
140{
141 int ret;
142 struct btrfs_key file_key;
143 int ins_len = mod < 0 ? -1 : 0;
144 int cow = mod != 0;
145
146 file_key.objectid = objectid;
147 file_key.offset = offset;
148 btrfs_set_key_type(&file_key, BTRFS_EXTENT_DATA_KEY);
149 ret = btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
150 return ret;
151}
152
153
154static int __btrfs_lookup_bio_sums(struct btrfs_root *root,
155 struct inode *inode, struct bio *bio,
156 u64 logical_offset, u32 *dst, int dio)
157{
158 u32 sum;
159 struct bio_vec *bvec = bio->bi_io_vec;
160 int bio_index = 0;
161 u64 offset = 0;
162 u64 item_start_offset = 0;
163 u64 item_last_offset = 0;
164 u64 disk_bytenr;
165 u32 diff;
166 u16 csum_size = 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 LIST_HEAD(tmplist);
290 unsigned long offset;
291 int ret;
292 size_t size;
293 u64 csum_end;
294 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
295
296 path = btrfs_alloc_path();
297 if (!path)
298 return -ENOMEM;
299
300 if (search_commit) {
301 path->skip_locking = 1;
302 path->reada = 2;
303 path->search_commit_root = 1;
304 }
305
306 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
307 key.offset = start;
308 key.type = BTRFS_EXTENT_CSUM_KEY;
309
310 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
311 if (ret < 0)
312 goto fail;
313 if (ret > 0 && path->slots[0] > 0) {
314 leaf = path->nodes[0];
315 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
316 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
317 key.type == BTRFS_EXTENT_CSUM_KEY) {
318 offset = (start - key.offset) >>
319 root->fs_info->sb->s_blocksize_bits;
320 if (offset * csum_size <
321 btrfs_item_size_nr(leaf, path->slots[0] - 1))
322 path->slots[0]--;
323 }
324 }
325
326 while (start <= end) {
327 leaf = path->nodes[0];
328 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
329 ret = btrfs_next_leaf(root, path);
330 if (ret < 0)
331 goto fail;
332 if (ret > 0)
333 break;
334 leaf = path->nodes[0];
335 }
336
337 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
338 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
339 key.type != BTRFS_EXTENT_CSUM_KEY)
340 break;
341
342 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
343 if (key.offset > end)
344 break;
345
346 if (key.offset > start)
347 start = key.offset;
348
349 size = btrfs_item_size_nr(leaf, path->slots[0]);
350 csum_end = key.offset + (size / csum_size) * root->sectorsize;
351 if (csum_end <= start) {
352 path->slots[0]++;
353 continue;
354 }
355
356 csum_end = min(csum_end, end + 1);
357 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
358 struct btrfs_csum_item);
359 while (start < csum_end) {
360 size = min_t(size_t, csum_end - start,
361 MAX_ORDERED_SUM_BYTES(root));
362 sums = kzalloc(btrfs_ordered_sum_size(root, size),
363 GFP_NOFS);
364 if (!sums) {
365 ret = -ENOMEM;
366 goto fail;
367 }
368
369 sector_sum = sums->sums;
370 sums->bytenr = start;
371 sums->len = size;
372
373 offset = (start - key.offset) >>
374 root->fs_info->sb->s_blocksize_bits;
375 offset *= csum_size;
376
377 while (size > 0) {
378 read_extent_buffer(path->nodes[0],
379 §or_sum->sum,
380 ((unsigned long)item) +
381 offset, csum_size);
382 sector_sum->bytenr = start;
383
384 size -= root->sectorsize;
385 start += root->sectorsize;
386 offset += csum_size;
387 sector_sum++;
388 }
389 list_add_tail(&sums->list, &tmplist);
390 }
391 path->slots[0]++;
392 }
393 ret = 0;
394fail:
395 while (ret < 0 && !list_empty(&tmplist)) {
396 sums = list_entry(&tmplist, struct btrfs_ordered_sum, list);
397 list_del(&sums->list);
398 kfree(sums);
399 }
400 list_splice_tail(&tmplist, list);
401
402 btrfs_free_path(path);
403 return ret;
404}
405
406int btrfs_csum_one_bio(struct btrfs_root *root, struct inode *inode,
407 struct bio *bio, u64 file_start, int contig)
408{
409 struct btrfs_ordered_sum *sums;
410 struct btrfs_sector_sum *sector_sum;
411 struct btrfs_ordered_extent *ordered;
412 char *data;
413 struct bio_vec *bvec = bio->bi_io_vec;
414 int bio_index = 0;
415 unsigned long total_bytes = 0;
416 unsigned long this_sum_bytes = 0;
417 u64 offset;
418 u64 disk_bytenr;
419
420 WARN_ON(bio->bi_vcnt <= 0);
421 sums = kzalloc(btrfs_ordered_sum_size(root, bio->bi_size), GFP_NOFS);
422 if (!sums)
423 return -ENOMEM;
424
425 sector_sum = sums->sums;
426 disk_bytenr = (u64)bio->bi_sector << 9;
427 sums->len = bio->bi_size;
428 INIT_LIST_HEAD(&sums->list);
429
430 if (contig)
431 offset = file_start;
432 else
433 offset = page_offset(bvec->bv_page) + bvec->bv_offset;
434
435 ordered = btrfs_lookup_ordered_extent(inode, offset);
436 BUG_ON(!ordered); /* Logic error */
437 sums->bytenr = ordered->start;
438
439 while (bio_index < bio->bi_vcnt) {
440 if (!contig)
441 offset = page_offset(bvec->bv_page) + bvec->bv_offset;
442
443 if (!contig && (offset >= ordered->file_offset + ordered->len ||
444 offset < ordered->file_offset)) {
445 unsigned long bytes_left;
446 sums->len = this_sum_bytes;
447 this_sum_bytes = 0;
448 btrfs_add_ordered_sum(inode, ordered, sums);
449 btrfs_put_ordered_extent(ordered);
450
451 bytes_left = bio->bi_size - total_bytes;
452
453 sums = kzalloc(btrfs_ordered_sum_size(root, bytes_left),
454 GFP_NOFS);
455 BUG_ON(!sums); /* -ENOMEM */
456 sector_sum = sums->sums;
457 sums->len = bytes_left;
458 ordered = btrfs_lookup_ordered_extent(inode, offset);
459 BUG_ON(!ordered); /* Logic error */
460 sums->bytenr = ordered->start;
461 }
462
463 data = kmap_atomic(bvec->bv_page);
464 sector_sum->sum = ~(u32)0;
465 sector_sum->sum = btrfs_csum_data(root,
466 data + bvec->bv_offset,
467 sector_sum->sum,
468 bvec->bv_len);
469 kunmap_atomic(data);
470 btrfs_csum_final(sector_sum->sum,
471 (char *)§or_sum->sum);
472 sector_sum->bytenr = disk_bytenr;
473
474 sector_sum++;
475 bio_index++;
476 total_bytes += bvec->bv_len;
477 this_sum_bytes += bvec->bv_len;
478 disk_bytenr += bvec->bv_len;
479 offset += bvec->bv_len;
480 bvec++;
481 }
482 this_sum_bytes = 0;
483 btrfs_add_ordered_sum(inode, ordered, sums);
484 btrfs_put_ordered_extent(ordered);
485 return 0;
486}
487
488/*
489 * helper function for csum removal, this expects the
490 * key to describe the csum pointed to by the path, and it expects
491 * the csum to overlap the range [bytenr, len]
492 *
493 * The csum should not be entirely contained in the range and the
494 * range should not be entirely contained in the csum.
495 *
496 * This calls btrfs_truncate_item with the correct args based on the
497 * overlap, and fixes up the key as required.
498 */
499static noinline void truncate_one_csum(struct btrfs_trans_handle *trans,
500 struct btrfs_root *root,
501 struct btrfs_path *path,
502 struct btrfs_key *key,
503 u64 bytenr, u64 len)
504{
505 struct extent_buffer *leaf;
506 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
507 u64 csum_end;
508 u64 end_byte = bytenr + len;
509 u32 blocksize_bits = root->fs_info->sb->s_blocksize_bits;
510
511 leaf = path->nodes[0];
512 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
513 csum_end <<= root->fs_info->sb->s_blocksize_bits;
514 csum_end += key->offset;
515
516 if (key->offset < bytenr && csum_end <= end_byte) {
517 /*
518 * [ bytenr - len ]
519 * [ ]
520 * [csum ]
521 * A simple truncate off the end of the item
522 */
523 u32 new_size = (bytenr - key->offset) >> blocksize_bits;
524 new_size *= csum_size;
525 btrfs_truncate_item(trans, root, path, new_size, 1);
526 } else if (key->offset >= bytenr && csum_end > end_byte &&
527 end_byte > key->offset) {
528 /*
529 * [ bytenr - len ]
530 * [ ]
531 * [csum ]
532 * we need to truncate from the beginning of the csum
533 */
534 u32 new_size = (csum_end - end_byte) >> blocksize_bits;
535 new_size *= csum_size;
536
537 btrfs_truncate_item(trans, root, path, new_size, 0);
538
539 key->offset = end_byte;
540 btrfs_set_item_key_safe(trans, root, path, key);
541 } else {
542 BUG();
543 }
544}
545
546/*
547 * deletes the csum items from the csum tree for a given
548 * range of bytes.
549 */
550int btrfs_del_csums(struct btrfs_trans_handle *trans,
551 struct btrfs_root *root, u64 bytenr, u64 len)
552{
553 struct btrfs_path *path;
554 struct btrfs_key key;
555 u64 end_byte = bytenr + len;
556 u64 csum_end;
557 struct extent_buffer *leaf;
558 int ret;
559 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
560 int blocksize_bits = root->fs_info->sb->s_blocksize_bits;
561
562 root = root->fs_info->csum_root;
563
564 path = btrfs_alloc_path();
565 if (!path)
566 return -ENOMEM;
567
568 while (1) {
569 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
570 key.offset = end_byte - 1;
571 key.type = BTRFS_EXTENT_CSUM_KEY;
572
573 path->leave_spinning = 1;
574 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
575 if (ret > 0) {
576 if (path->slots[0] == 0)
577 break;
578 path->slots[0]--;
579 } else if (ret < 0) {
580 break;
581 }
582
583 leaf = path->nodes[0];
584 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
585
586 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
587 key.type != BTRFS_EXTENT_CSUM_KEY) {
588 break;
589 }
590
591 if (key.offset >= end_byte)
592 break;
593
594 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
595 csum_end <<= blocksize_bits;
596 csum_end += key.offset;
597
598 /* this csum ends before we start, we're done */
599 if (csum_end <= bytenr)
600 break;
601
602 /* delete the entire item, it is inside our range */
603 if (key.offset >= bytenr && csum_end <= end_byte) {
604 ret = btrfs_del_item(trans, root, path);
605 if (ret)
606 goto out;
607 if (key.offset == bytenr)
608 break;
609 } else if (key.offset < bytenr && csum_end > end_byte) {
610 unsigned long offset;
611 unsigned long shift_len;
612 unsigned long item_offset;
613 /*
614 * [ bytenr - len ]
615 * [csum ]
616 *
617 * Our bytes are in the middle of the csum,
618 * we need to split this item and insert a new one.
619 *
620 * But we can't drop the path because the
621 * csum could change, get removed, extended etc.
622 *
623 * The trick here is the max size of a csum item leaves
624 * enough room in the tree block for a single
625 * item header. So, we split the item in place,
626 * adding a new header pointing to the existing
627 * bytes. Then we loop around again and we have
628 * a nicely formed csum item that we can neatly
629 * truncate.
630 */
631 offset = (bytenr - key.offset) >> blocksize_bits;
632 offset *= csum_size;
633
634 shift_len = (len >> blocksize_bits) * csum_size;
635
636 item_offset = btrfs_item_ptr_offset(leaf,
637 path->slots[0]);
638
639 memset_extent_buffer(leaf, 0, item_offset + offset,
640 shift_len);
641 key.offset = bytenr;
642
643 /*
644 * btrfs_split_item returns -EAGAIN when the
645 * item changed size or key
646 */
647 ret = btrfs_split_item(trans, root, path, &key, offset);
648 if (ret && ret != -EAGAIN) {
649 btrfs_abort_transaction(trans, root, ret);
650 goto out;
651 }
652
653 key.offset = end_byte - 1;
654 } else {
655 truncate_one_csum(trans, root, path, &key, bytenr, len);
656 if (key.offset < bytenr)
657 break;
658 }
659 btrfs_release_path(path);
660 }
661 ret = 0;
662out:
663 btrfs_free_path(path);
664 return ret;
665}
666
667int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
668 struct btrfs_root *root,
669 struct btrfs_ordered_sum *sums)
670{
671 u64 bytenr;
672 int ret;
673 struct btrfs_key file_key;
674 struct btrfs_key found_key;
675 u64 next_offset;
676 u64 total_bytes = 0;
677 int found_next;
678 struct btrfs_path *path;
679 struct btrfs_csum_item *item;
680 struct btrfs_csum_item *item_end;
681 struct extent_buffer *leaf = NULL;
682 u64 csum_offset;
683 struct btrfs_sector_sum *sector_sum;
684 u32 nritems;
685 u32 ins_size;
686 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
687
688 path = btrfs_alloc_path();
689 if (!path)
690 return -ENOMEM;
691
692 sector_sum = sums->sums;
693again:
694 next_offset = (u64)-1;
695 found_next = 0;
696 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
697 file_key.offset = sector_sum->bytenr;
698 bytenr = sector_sum->bytenr;
699 btrfs_set_key_type(&file_key, BTRFS_EXTENT_CSUM_KEY);
700
701 item = btrfs_lookup_csum(trans, root, path, sector_sum->bytenr, 1);
702 if (!IS_ERR(item)) {
703 leaf = path->nodes[0];
704 ret = 0;
705 goto found;
706 }
707 ret = PTR_ERR(item);
708 if (ret != -EFBIG && ret != -ENOENT)
709 goto fail_unlock;
710
711 if (ret == -EFBIG) {
712 u32 item_size;
713 /* we found one, but it isn't big enough yet */
714 leaf = path->nodes[0];
715 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
716 if ((item_size / csum_size) >=
717 MAX_CSUM_ITEMS(root, csum_size)) {
718 /* already at max size, make a new one */
719 goto insert;
720 }
721 } else {
722 int slot = path->slots[0] + 1;
723 /* we didn't find a csum item, insert one */
724 nritems = btrfs_header_nritems(path->nodes[0]);
725 if (path->slots[0] >= nritems - 1) {
726 ret = btrfs_next_leaf(root, path);
727 if (ret == 1)
728 found_next = 1;
729 if (ret != 0)
730 goto insert;
731 slot = 0;
732 }
733 btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
734 if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
735 found_key.type != BTRFS_EXTENT_CSUM_KEY) {
736 found_next = 1;
737 goto insert;
738 }
739 next_offset = found_key.offset;
740 found_next = 1;
741 goto insert;
742 }
743
744 /*
745 * at this point, we know the tree has an item, but it isn't big
746 * enough yet to put our csum in. Grow it
747 */
748 btrfs_release_path(path);
749 ret = btrfs_search_slot(trans, root, &file_key, path,
750 csum_size, 1);
751 if (ret < 0)
752 goto fail_unlock;
753
754 if (ret > 0) {
755 if (path->slots[0] == 0)
756 goto insert;
757 path->slots[0]--;
758 }
759
760 leaf = path->nodes[0];
761 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
762 csum_offset = (bytenr - found_key.offset) >>
763 root->fs_info->sb->s_blocksize_bits;
764
765 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_CSUM_KEY ||
766 found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
767 csum_offset >= MAX_CSUM_ITEMS(root, csum_size)) {
768 goto insert;
769 }
770
771 if (csum_offset >= btrfs_item_size_nr(leaf, path->slots[0]) /
772 csum_size) {
773 u32 diff = (csum_offset + 1) * csum_size;
774
775 /*
776 * is the item big enough already? we dropped our lock
777 * before and need to recheck
778 */
779 if (diff < btrfs_item_size_nr(leaf, path->slots[0]))
780 goto csum;
781
782 diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
783 if (diff != csum_size)
784 goto insert;
785
786 btrfs_extend_item(trans, root, path, diff);
787 goto csum;
788 }
789
790insert:
791 btrfs_release_path(path);
792 csum_offset = 0;
793 if (found_next) {
794 u64 tmp = total_bytes + root->sectorsize;
795 u64 next_sector = sector_sum->bytenr;
796 struct btrfs_sector_sum *next = sector_sum + 1;
797
798 while (tmp < sums->len) {
799 if (next_sector + root->sectorsize != next->bytenr)
800 break;
801 tmp += root->sectorsize;
802 next_sector = next->bytenr;
803 next++;
804 }
805 tmp = min(tmp, next_offset - file_key.offset);
806 tmp >>= root->fs_info->sb->s_blocksize_bits;
807 tmp = max((u64)1, tmp);
808 tmp = min(tmp, (u64)MAX_CSUM_ITEMS(root, csum_size));
809 ins_size = csum_size * tmp;
810 } else {
811 ins_size = csum_size;
812 }
813 path->leave_spinning = 1;
814 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
815 ins_size);
816 path->leave_spinning = 0;
817 if (ret < 0)
818 goto fail_unlock;
819 if (ret != 0) {
820 WARN_ON(1);
821 goto fail_unlock;
822 }
823csum:
824 leaf = path->nodes[0];
825 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
826 ret = 0;
827 item = (struct btrfs_csum_item *)((unsigned char *)item +
828 csum_offset * csum_size);
829found:
830 item_end = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
831 item_end = (struct btrfs_csum_item *)((unsigned char *)item_end +
832 btrfs_item_size_nr(leaf, path->slots[0]));
833next_sector:
834
835 write_extent_buffer(leaf, §or_sum->sum, (unsigned long)item, csum_size);
836
837 total_bytes += root->sectorsize;
838 sector_sum++;
839 if (total_bytes < sums->len) {
840 item = (struct btrfs_csum_item *)((char *)item +
841 csum_size);
842 if (item < item_end && bytenr + PAGE_CACHE_SIZE ==
843 sector_sum->bytenr) {
844 bytenr = sector_sum->bytenr;
845 goto next_sector;
846 }
847 }
848
849 btrfs_mark_buffer_dirty(path->nodes[0]);
850 if (total_bytes < sums->len) {
851 btrfs_release_path(path);
852 cond_resched();
853 goto again;
854 }
855out:
856 btrfs_free_path(path);
857 return ret;
858
859fail_unlock:
860 goto out;
861}
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 "ctree.h"
11#include "disk-io.h"
12#include "transaction.h"
13#include "volumes.h"
14#include "print-tree.h"
15#include "compression.h"
16
17#define __MAX_CSUM_ITEMS(r, size) ((unsigned long)(((BTRFS_LEAF_DATA_SIZE(r) - \
18 sizeof(struct btrfs_item) * 2) / \
19 size) - 1))
20
21#define MAX_CSUM_ITEMS(r, size) (min_t(u32, __MAX_CSUM_ITEMS(r, size), \
22 PAGE_SIZE))
23
24#define MAX_ORDERED_SUM_BYTES(fs_info) ((PAGE_SIZE - \
25 sizeof(struct btrfs_ordered_sum)) / \
26 sizeof(u32) * (fs_info)->sectorsize)
27
28int btrfs_insert_file_extent(struct btrfs_trans_handle *trans,
29 struct btrfs_root *root,
30 u64 objectid, u64 pos,
31 u64 disk_offset, u64 disk_num_bytes,
32 u64 num_bytes, u64 offset, u64 ram_bytes,
33 u8 compression, u8 encryption, u16 other_encoding)
34{
35 int ret = 0;
36 struct btrfs_file_extent_item *item;
37 struct btrfs_key file_key;
38 struct btrfs_path *path;
39 struct extent_buffer *leaf;
40
41 path = btrfs_alloc_path();
42 if (!path)
43 return -ENOMEM;
44 file_key.objectid = objectid;
45 file_key.offset = pos;
46 file_key.type = BTRFS_EXTENT_DATA_KEY;
47
48 path->leave_spinning = 1;
49 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
50 sizeof(*item));
51 if (ret < 0)
52 goto out;
53 BUG_ON(ret); /* Can't happen */
54 leaf = path->nodes[0];
55 item = btrfs_item_ptr(leaf, path->slots[0],
56 struct btrfs_file_extent_item);
57 btrfs_set_file_extent_disk_bytenr(leaf, item, disk_offset);
58 btrfs_set_file_extent_disk_num_bytes(leaf, item, disk_num_bytes);
59 btrfs_set_file_extent_offset(leaf, item, offset);
60 btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
61 btrfs_set_file_extent_ram_bytes(leaf, item, ram_bytes);
62 btrfs_set_file_extent_generation(leaf, item, trans->transid);
63 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
64 btrfs_set_file_extent_compression(leaf, item, compression);
65 btrfs_set_file_extent_encryption(leaf, item, encryption);
66 btrfs_set_file_extent_other_encoding(leaf, item, other_encoding);
67
68 btrfs_mark_buffer_dirty(leaf);
69out:
70 btrfs_free_path(path);
71 return ret;
72}
73
74static struct btrfs_csum_item *
75btrfs_lookup_csum(struct btrfs_trans_handle *trans,
76 struct btrfs_root *root,
77 struct btrfs_path *path,
78 u64 bytenr, int cow)
79{
80 struct btrfs_fs_info *fs_info = root->fs_info;
81 int ret;
82 struct btrfs_key file_key;
83 struct btrfs_key found_key;
84 struct btrfs_csum_item *item;
85 struct extent_buffer *leaf;
86 u64 csum_offset = 0;
87 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
88 int csums_in_item;
89
90 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
91 file_key.offset = bytenr;
92 file_key.type = BTRFS_EXTENT_CSUM_KEY;
93 ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
94 if (ret < 0)
95 goto fail;
96 leaf = path->nodes[0];
97 if (ret > 0) {
98 ret = 1;
99 if (path->slots[0] == 0)
100 goto fail;
101 path->slots[0]--;
102 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
103 if (found_key.type != BTRFS_EXTENT_CSUM_KEY)
104 goto fail;
105
106 csum_offset = (bytenr - found_key.offset) >>
107 fs_info->sb->s_blocksize_bits;
108 csums_in_item = btrfs_item_size_nr(leaf, path->slots[0]);
109 csums_in_item /= csum_size;
110
111 if (csum_offset == csums_in_item) {
112 ret = -EFBIG;
113 goto fail;
114 } else if (csum_offset > csums_in_item) {
115 goto fail;
116 }
117 }
118 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
119 item = (struct btrfs_csum_item *)((unsigned char *)item +
120 csum_offset * csum_size);
121 return item;
122fail:
123 if (ret > 0)
124 ret = -ENOENT;
125 return ERR_PTR(ret);
126}
127
128int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
129 struct btrfs_root *root,
130 struct btrfs_path *path, u64 objectid,
131 u64 offset, int mod)
132{
133 int ret;
134 struct btrfs_key file_key;
135 int ins_len = mod < 0 ? -1 : 0;
136 int cow = mod != 0;
137
138 file_key.objectid = objectid;
139 file_key.offset = offset;
140 file_key.type = BTRFS_EXTENT_DATA_KEY;
141 ret = btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
142 return ret;
143}
144
145static void btrfs_io_bio_endio_readpage(struct btrfs_io_bio *bio, int err)
146{
147 kfree(bio->csum_allocated);
148}
149
150static blk_status_t __btrfs_lookup_bio_sums(struct inode *inode, struct bio *bio,
151 u64 logical_offset, u32 *dst, int dio)
152{
153 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
154 struct bio_vec bvec;
155 struct bvec_iter iter;
156 struct btrfs_io_bio *btrfs_bio = btrfs_io_bio(bio);
157 struct btrfs_csum_item *item = NULL;
158 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
159 struct btrfs_path *path;
160 u8 *csum;
161 u64 offset = 0;
162 u64 item_start_offset = 0;
163 u64 item_last_offset = 0;
164 u64 disk_bytenr;
165 u64 page_bytes_left;
166 u32 diff;
167 int nblocks;
168 int count = 0;
169 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
170
171 path = btrfs_alloc_path();
172 if (!path)
173 return BLK_STS_RESOURCE;
174
175 nblocks = bio->bi_iter.bi_size >> inode->i_sb->s_blocksize_bits;
176 if (!dst) {
177 if (nblocks * csum_size > BTRFS_BIO_INLINE_CSUM_SIZE) {
178 btrfs_bio->csum_allocated = kmalloc_array(nblocks,
179 csum_size, GFP_NOFS);
180 if (!btrfs_bio->csum_allocated) {
181 btrfs_free_path(path);
182 return BLK_STS_RESOURCE;
183 }
184 btrfs_bio->csum = btrfs_bio->csum_allocated;
185 btrfs_bio->end_io = btrfs_io_bio_endio_readpage;
186 } else {
187 btrfs_bio->csum = btrfs_bio->csum_inline;
188 }
189 csum = btrfs_bio->csum;
190 } else {
191 csum = (u8 *)dst;
192 }
193
194 if (bio->bi_iter.bi_size > PAGE_SIZE * 8)
195 path->reada = READA_FORWARD;
196
197 /*
198 * the free space stuff is only read when it hasn't been
199 * updated in the current transaction. So, we can safely
200 * read from the commit root and sidestep a nasty deadlock
201 * between reading the free space cache and updating the csum tree.
202 */
203 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
204 path->search_commit_root = 1;
205 path->skip_locking = 1;
206 }
207
208 disk_bytenr = (u64)bio->bi_iter.bi_sector << 9;
209 if (dio)
210 offset = logical_offset;
211
212 bio_for_each_segment(bvec, bio, iter) {
213 page_bytes_left = bvec.bv_len;
214 if (count)
215 goto next;
216
217 if (!dio)
218 offset = page_offset(bvec.bv_page) + bvec.bv_offset;
219 count = btrfs_find_ordered_sum(inode, offset, disk_bytenr,
220 (u32 *)csum, nblocks);
221 if (count)
222 goto found;
223
224 if (!item || disk_bytenr < item_start_offset ||
225 disk_bytenr >= item_last_offset) {
226 struct btrfs_key found_key;
227 u32 item_size;
228
229 if (item)
230 btrfs_release_path(path);
231 item = btrfs_lookup_csum(NULL, fs_info->csum_root,
232 path, disk_bytenr, 0);
233 if (IS_ERR(item)) {
234 count = 1;
235 memset(csum, 0, csum_size);
236 if (BTRFS_I(inode)->root->root_key.objectid ==
237 BTRFS_DATA_RELOC_TREE_OBJECTID) {
238 set_extent_bits(io_tree, offset,
239 offset + fs_info->sectorsize - 1,
240 EXTENT_NODATASUM);
241 } else {
242 btrfs_info_rl(fs_info,
243 "no csum found for inode %llu start %llu",
244 btrfs_ino(BTRFS_I(inode)), offset);
245 }
246 item = NULL;
247 btrfs_release_path(path);
248 goto found;
249 }
250 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
251 path->slots[0]);
252
253 item_start_offset = found_key.offset;
254 item_size = btrfs_item_size_nr(path->nodes[0],
255 path->slots[0]);
256 item_last_offset = item_start_offset +
257 (item_size / csum_size) *
258 fs_info->sectorsize;
259 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
260 struct btrfs_csum_item);
261 }
262 /*
263 * this byte range must be able to fit inside
264 * a single leaf so it will also fit inside a u32
265 */
266 diff = disk_bytenr - item_start_offset;
267 diff = diff / fs_info->sectorsize;
268 diff = diff * csum_size;
269 count = min_t(int, nblocks, (item_last_offset - disk_bytenr) >>
270 inode->i_sb->s_blocksize_bits);
271 read_extent_buffer(path->nodes[0], csum,
272 ((unsigned long)item) + diff,
273 csum_size * count);
274found:
275 csum += count * csum_size;
276 nblocks -= count;
277next:
278 while (count--) {
279 disk_bytenr += fs_info->sectorsize;
280 offset += fs_info->sectorsize;
281 page_bytes_left -= fs_info->sectorsize;
282 if (!page_bytes_left)
283 break; /* move to next bio */
284 }
285 }
286
287 WARN_ON_ONCE(count);
288 btrfs_free_path(path);
289 return 0;
290}
291
292blk_status_t btrfs_lookup_bio_sums(struct inode *inode, struct bio *bio, u32 *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));
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
422blk_status_t btrfs_csum_one_bio(struct inode *inode, struct bio *bio,
423 u64 file_start, int contig)
424{
425 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
426 struct btrfs_ordered_sum *sums;
427 struct btrfs_ordered_extent *ordered = NULL;
428 char *data;
429 struct bvec_iter iter;
430 struct bio_vec bvec;
431 int index;
432 int nr_sectors;
433 unsigned long total_bytes = 0;
434 unsigned long this_sum_bytes = 0;
435 int i;
436 u64 offset;
437
438 sums = kzalloc(btrfs_ordered_sum_size(fs_info, bio->bi_iter.bi_size),
439 GFP_NOFS);
440 if (!sums)
441 return BLK_STS_RESOURCE;
442
443 sums->len = bio->bi_iter.bi_size;
444 INIT_LIST_HEAD(&sums->list);
445
446 if (contig)
447 offset = file_start;
448 else
449 offset = 0; /* shut up gcc */
450
451 sums->bytenr = (u64)bio->bi_iter.bi_sector << 9;
452 index = 0;
453
454 bio_for_each_segment(bvec, bio, iter) {
455 if (!contig)
456 offset = page_offset(bvec.bv_page) + bvec.bv_offset;
457
458 if (!ordered) {
459 ordered = btrfs_lookup_ordered_extent(inode, offset);
460 BUG_ON(!ordered); /* Logic error */
461 }
462
463 data = kmap_atomic(bvec.bv_page);
464
465 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info,
466 bvec.bv_len + fs_info->sectorsize
467 - 1);
468
469 for (i = 0; i < nr_sectors; i++) {
470 if (offset >= ordered->file_offset + ordered->len ||
471 offset < ordered->file_offset) {
472 unsigned long bytes_left;
473
474 kunmap_atomic(data);
475 sums->len = this_sum_bytes;
476 this_sum_bytes = 0;
477 btrfs_add_ordered_sum(inode, ordered, sums);
478 btrfs_put_ordered_extent(ordered);
479
480 bytes_left = bio->bi_iter.bi_size - total_bytes;
481
482 sums = kzalloc(btrfs_ordered_sum_size(fs_info, bytes_left),
483 GFP_NOFS);
484 BUG_ON(!sums); /* -ENOMEM */
485 sums->len = bytes_left;
486 ordered = btrfs_lookup_ordered_extent(inode,
487 offset);
488 ASSERT(ordered); /* Logic error */
489 sums->bytenr = ((u64)bio->bi_iter.bi_sector << 9)
490 + total_bytes;
491 index = 0;
492
493 data = kmap_atomic(bvec.bv_page);
494 }
495
496 sums->sums[index] = ~(u32)0;
497 sums->sums[index]
498 = btrfs_csum_data(data + bvec.bv_offset
499 + (i * fs_info->sectorsize),
500 sums->sums[index],
501 fs_info->sectorsize);
502 btrfs_csum_final(sums->sums[index],
503 (char *)(sums->sums + index));
504 index++;
505 offset += fs_info->sectorsize;
506 this_sum_bytes += fs_info->sectorsize;
507 total_bytes += fs_info->sectorsize;
508 }
509
510 kunmap_atomic(data);
511 }
512 this_sum_bytes = 0;
513 btrfs_add_ordered_sum(inode, ordered, sums);
514 btrfs_put_ordered_extent(ordered);
515 return 0;
516}
517
518/*
519 * helper function for csum removal, this expects the
520 * key to describe the csum pointed to by the path, and it expects
521 * the csum to overlap the range [bytenr, len]
522 *
523 * The csum should not be entirely contained in the range and the
524 * range should not be entirely contained in the csum.
525 *
526 * This calls btrfs_truncate_item with the correct args based on the
527 * overlap, and fixes up the key as required.
528 */
529static noinline void truncate_one_csum(struct btrfs_fs_info *fs_info,
530 struct btrfs_path *path,
531 struct btrfs_key *key,
532 u64 bytenr, u64 len)
533{
534 struct extent_buffer *leaf;
535 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
536 u64 csum_end;
537 u64 end_byte = bytenr + len;
538 u32 blocksize_bits = fs_info->sb->s_blocksize_bits;
539
540 leaf = path->nodes[0];
541 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
542 csum_end <<= fs_info->sb->s_blocksize_bits;
543 csum_end += key->offset;
544
545 if (key->offset < bytenr && csum_end <= end_byte) {
546 /*
547 * [ bytenr - len ]
548 * [ ]
549 * [csum ]
550 * A simple truncate off the end of the item
551 */
552 u32 new_size = (bytenr - key->offset) >> blocksize_bits;
553 new_size *= csum_size;
554 btrfs_truncate_item(fs_info, path, new_size, 1);
555 } else if (key->offset >= bytenr && csum_end > end_byte &&
556 end_byte > key->offset) {
557 /*
558 * [ bytenr - len ]
559 * [ ]
560 * [csum ]
561 * we need to truncate from the beginning of the csum
562 */
563 u32 new_size = (csum_end - end_byte) >> blocksize_bits;
564 new_size *= csum_size;
565
566 btrfs_truncate_item(fs_info, path, new_size, 0);
567
568 key->offset = end_byte;
569 btrfs_set_item_key_safe(fs_info, path, key);
570 } else {
571 BUG();
572 }
573}
574
575/*
576 * deletes the csum items from the csum tree for a given
577 * range of bytes.
578 */
579int btrfs_del_csums(struct btrfs_trans_handle *trans,
580 struct btrfs_fs_info *fs_info, u64 bytenr, u64 len)
581{
582 struct btrfs_root *root = fs_info->csum_root;
583 struct btrfs_path *path;
584 struct btrfs_key key;
585 u64 end_byte = bytenr + len;
586 u64 csum_end;
587 struct extent_buffer *leaf;
588 int ret;
589 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
590 int blocksize_bits = fs_info->sb->s_blocksize_bits;
591
592 path = btrfs_alloc_path();
593 if (!path)
594 return -ENOMEM;
595
596 while (1) {
597 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
598 key.offset = end_byte - 1;
599 key.type = BTRFS_EXTENT_CSUM_KEY;
600
601 path->leave_spinning = 1;
602 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
603 if (ret > 0) {
604 if (path->slots[0] == 0)
605 break;
606 path->slots[0]--;
607 } else if (ret < 0) {
608 break;
609 }
610
611 leaf = path->nodes[0];
612 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
613
614 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
615 key.type != BTRFS_EXTENT_CSUM_KEY) {
616 break;
617 }
618
619 if (key.offset >= end_byte)
620 break;
621
622 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
623 csum_end <<= blocksize_bits;
624 csum_end += key.offset;
625
626 /* this csum ends before we start, we're done */
627 if (csum_end <= bytenr)
628 break;
629
630 /* delete the entire item, it is inside our range */
631 if (key.offset >= bytenr && csum_end <= end_byte) {
632 int del_nr = 1;
633
634 /*
635 * Check how many csum items preceding this one in this
636 * leaf correspond to our range and then delete them all
637 * at once.
638 */
639 if (key.offset > bytenr && path->slots[0] > 0) {
640 int slot = path->slots[0] - 1;
641
642 while (slot >= 0) {
643 struct btrfs_key pk;
644
645 btrfs_item_key_to_cpu(leaf, &pk, slot);
646 if (pk.offset < bytenr ||
647 pk.type != BTRFS_EXTENT_CSUM_KEY ||
648 pk.objectid !=
649 BTRFS_EXTENT_CSUM_OBJECTID)
650 break;
651 path->slots[0] = slot;
652 del_nr++;
653 key.offset = pk.offset;
654 slot--;
655 }
656 }
657 ret = btrfs_del_items(trans, root, path,
658 path->slots[0], del_nr);
659 if (ret)
660 goto out;
661 if (key.offset == bytenr)
662 break;
663 } else if (key.offset < bytenr && csum_end > end_byte) {
664 unsigned long offset;
665 unsigned long shift_len;
666 unsigned long item_offset;
667 /*
668 * [ bytenr - len ]
669 * [csum ]
670 *
671 * Our bytes are in the middle of the csum,
672 * we need to split this item and insert a new one.
673 *
674 * But we can't drop the path because the
675 * csum could change, get removed, extended etc.
676 *
677 * The trick here is the max size of a csum item leaves
678 * enough room in the tree block for a single
679 * item header. So, we split the item in place,
680 * adding a new header pointing to the existing
681 * bytes. Then we loop around again and we have
682 * a nicely formed csum item that we can neatly
683 * truncate.
684 */
685 offset = (bytenr - key.offset) >> blocksize_bits;
686 offset *= csum_size;
687
688 shift_len = (len >> blocksize_bits) * csum_size;
689
690 item_offset = btrfs_item_ptr_offset(leaf,
691 path->slots[0]);
692
693 memzero_extent_buffer(leaf, item_offset + offset,
694 shift_len);
695 key.offset = bytenr;
696
697 /*
698 * btrfs_split_item returns -EAGAIN when the
699 * item changed size or key
700 */
701 ret = btrfs_split_item(trans, root, path, &key, offset);
702 if (ret && ret != -EAGAIN) {
703 btrfs_abort_transaction(trans, ret);
704 goto out;
705 }
706
707 key.offset = end_byte - 1;
708 } else {
709 truncate_one_csum(fs_info, path, &key, bytenr, len);
710 if (key.offset < bytenr)
711 break;
712 }
713 btrfs_release_path(path);
714 }
715 ret = 0;
716out:
717 btrfs_free_path(path);
718 return ret;
719}
720
721int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
722 struct btrfs_root *root,
723 struct btrfs_ordered_sum *sums)
724{
725 struct btrfs_fs_info *fs_info = root->fs_info;
726 struct btrfs_key file_key;
727 struct btrfs_key found_key;
728 struct btrfs_path *path;
729 struct btrfs_csum_item *item;
730 struct btrfs_csum_item *item_end;
731 struct extent_buffer *leaf = NULL;
732 u64 next_offset;
733 u64 total_bytes = 0;
734 u64 csum_offset;
735 u64 bytenr;
736 u32 nritems;
737 u32 ins_size;
738 int index = 0;
739 int found_next;
740 int ret;
741 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
742
743 path = btrfs_alloc_path();
744 if (!path)
745 return -ENOMEM;
746again:
747 next_offset = (u64)-1;
748 found_next = 0;
749 bytenr = sums->bytenr + total_bytes;
750 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
751 file_key.offset = bytenr;
752 file_key.type = BTRFS_EXTENT_CSUM_KEY;
753
754 item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
755 if (!IS_ERR(item)) {
756 ret = 0;
757 leaf = path->nodes[0];
758 item_end = btrfs_item_ptr(leaf, path->slots[0],
759 struct btrfs_csum_item);
760 item_end = (struct btrfs_csum_item *)((char *)item_end +
761 btrfs_item_size_nr(leaf, path->slots[0]));
762 goto found;
763 }
764 ret = PTR_ERR(item);
765 if (ret != -EFBIG && ret != -ENOENT)
766 goto fail_unlock;
767
768 if (ret == -EFBIG) {
769 u32 item_size;
770 /* we found one, but it isn't big enough yet */
771 leaf = path->nodes[0];
772 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
773 if ((item_size / csum_size) >=
774 MAX_CSUM_ITEMS(fs_info, csum_size)) {
775 /* already at max size, make a new one */
776 goto insert;
777 }
778 } else {
779 int slot = path->slots[0] + 1;
780 /* we didn't find a csum item, insert one */
781 nritems = btrfs_header_nritems(path->nodes[0]);
782 if (!nritems || (path->slots[0] >= nritems - 1)) {
783 ret = btrfs_next_leaf(root, path);
784 if (ret == 1)
785 found_next = 1;
786 if (ret != 0)
787 goto insert;
788 slot = path->slots[0];
789 }
790 btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
791 if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
792 found_key.type != BTRFS_EXTENT_CSUM_KEY) {
793 found_next = 1;
794 goto insert;
795 }
796 next_offset = found_key.offset;
797 found_next = 1;
798 goto insert;
799 }
800
801 /*
802 * at this point, we know the tree has an item, but it isn't big
803 * enough yet to put our csum in. Grow it
804 */
805 btrfs_release_path(path);
806 ret = btrfs_search_slot(trans, root, &file_key, path,
807 csum_size, 1);
808 if (ret < 0)
809 goto fail_unlock;
810
811 if (ret > 0) {
812 if (path->slots[0] == 0)
813 goto insert;
814 path->slots[0]--;
815 }
816
817 leaf = path->nodes[0];
818 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
819 csum_offset = (bytenr - found_key.offset) >>
820 fs_info->sb->s_blocksize_bits;
821
822 if (found_key.type != BTRFS_EXTENT_CSUM_KEY ||
823 found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
824 csum_offset >= MAX_CSUM_ITEMS(fs_info, csum_size)) {
825 goto insert;
826 }
827
828 if (csum_offset == btrfs_item_size_nr(leaf, path->slots[0]) /
829 csum_size) {
830 int extend_nr;
831 u64 tmp;
832 u32 diff;
833 u32 free_space;
834
835 if (btrfs_leaf_free_space(fs_info, leaf) <
836 sizeof(struct btrfs_item) + csum_size * 2)
837 goto insert;
838
839 free_space = btrfs_leaf_free_space(fs_info, leaf) -
840 sizeof(struct btrfs_item) - csum_size;
841 tmp = sums->len - total_bytes;
842 tmp >>= fs_info->sb->s_blocksize_bits;
843 WARN_ON(tmp < 1);
844
845 extend_nr = max_t(int, 1, (int)tmp);
846 diff = (csum_offset + extend_nr) * csum_size;
847 diff = min(diff,
848 MAX_CSUM_ITEMS(fs_info, csum_size) * csum_size);
849
850 diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
851 diff = min(free_space, diff);
852 diff /= csum_size;
853 diff *= csum_size;
854
855 btrfs_extend_item(fs_info, path, diff);
856 ret = 0;
857 goto csum;
858 }
859
860insert:
861 btrfs_release_path(path);
862 csum_offset = 0;
863 if (found_next) {
864 u64 tmp;
865
866 tmp = sums->len - total_bytes;
867 tmp >>= fs_info->sb->s_blocksize_bits;
868 tmp = min(tmp, (next_offset - file_key.offset) >>
869 fs_info->sb->s_blocksize_bits);
870
871 tmp = max_t(u64, 1, tmp);
872 tmp = min_t(u64, tmp, MAX_CSUM_ITEMS(fs_info, csum_size));
873 ins_size = csum_size * tmp;
874 } else {
875 ins_size = csum_size;
876 }
877 path->leave_spinning = 1;
878 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
879 ins_size);
880 path->leave_spinning = 0;
881 if (ret < 0)
882 goto fail_unlock;
883 if (WARN_ON(ret != 0))
884 goto fail_unlock;
885 leaf = path->nodes[0];
886csum:
887 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
888 item_end = (struct btrfs_csum_item *)((unsigned char *)item +
889 btrfs_item_size_nr(leaf, path->slots[0]));
890 item = (struct btrfs_csum_item *)((unsigned char *)item +
891 csum_offset * csum_size);
892found:
893 ins_size = (u32)(sums->len - total_bytes) >>
894 fs_info->sb->s_blocksize_bits;
895 ins_size *= csum_size;
896 ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item,
897 ins_size);
898 write_extent_buffer(leaf, sums->sums + index, (unsigned long)item,
899 ins_size);
900
901 ins_size /= csum_size;
902 total_bytes += ins_size * fs_info->sectorsize;
903 index += ins_size;
904
905 btrfs_mark_buffer_dirty(path->nodes[0]);
906 if (total_bytes < sums->len) {
907 btrfs_release_path(path);
908 cond_resched();
909 goto again;
910 }
911out:
912 btrfs_free_path(path);
913 return ret;
914
915fail_unlock:
916 goto out;
917}
918
919void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode,
920 const struct btrfs_path *path,
921 struct btrfs_file_extent_item *fi,
922 const bool new_inline,
923 struct extent_map *em)
924{
925 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
926 struct btrfs_root *root = inode->root;
927 struct extent_buffer *leaf = path->nodes[0];
928 const int slot = path->slots[0];
929 struct btrfs_key key;
930 u64 extent_start, extent_end;
931 u64 bytenr;
932 u8 type = btrfs_file_extent_type(leaf, fi);
933 int compress_type = btrfs_file_extent_compression(leaf, fi);
934
935 em->bdev = fs_info->fs_devices->latest_bdev;
936 btrfs_item_key_to_cpu(leaf, &key, slot);
937 extent_start = key.offset;
938
939 if (type == BTRFS_FILE_EXTENT_REG ||
940 type == BTRFS_FILE_EXTENT_PREALLOC) {
941 extent_end = extent_start +
942 btrfs_file_extent_num_bytes(leaf, fi);
943 } else if (type == BTRFS_FILE_EXTENT_INLINE) {
944 size_t size;
945 size = btrfs_file_extent_inline_len(leaf, slot, fi);
946 extent_end = ALIGN(extent_start + size,
947 fs_info->sectorsize);
948 }
949
950 em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
951 if (type == BTRFS_FILE_EXTENT_REG ||
952 type == BTRFS_FILE_EXTENT_PREALLOC) {
953 em->start = extent_start;
954 em->len = extent_end - extent_start;
955 em->orig_start = extent_start -
956 btrfs_file_extent_offset(leaf, fi);
957 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
958 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
959 if (bytenr == 0) {
960 em->block_start = EXTENT_MAP_HOLE;
961 return;
962 }
963 if (compress_type != BTRFS_COMPRESS_NONE) {
964 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
965 em->compress_type = compress_type;
966 em->block_start = bytenr;
967 em->block_len = em->orig_block_len;
968 } else {
969 bytenr += btrfs_file_extent_offset(leaf, fi);
970 em->block_start = bytenr;
971 em->block_len = em->len;
972 if (type == BTRFS_FILE_EXTENT_PREALLOC)
973 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
974 }
975 } else if (type == BTRFS_FILE_EXTENT_INLINE) {
976 em->block_start = EXTENT_MAP_INLINE;
977 em->start = extent_start;
978 em->len = extent_end - extent_start;
979 /*
980 * Initialize orig_start and block_len with the same values
981 * as in inode.c:btrfs_get_extent().
982 */
983 em->orig_start = EXTENT_MAP_HOLE;
984 em->block_len = (u64)-1;
985 if (!new_inline && compress_type != BTRFS_COMPRESS_NONE) {
986 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
987 em->compress_type = compress_type;
988 }
989 } else {
990 btrfs_err(fs_info,
991 "unknown file extent item type %d, inode %llu, offset %llu, "
992 "root %llu", type, btrfs_ino(inode), extent_start,
993 root->root_key.objectid);
994 }
995}