Line data Source code
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 : */
49 17998501 : void btrfs_inode_safe_disk_i_size_write(struct btrfs_inode *inode, u64 new_i_size)
50 : {
51 17998501 : struct btrfs_fs_info *fs_info = inode->root->fs_info;
52 17998501 : u64 start, end, i_size;
53 17998501 : int ret;
54 :
55 17998501 : spin_lock(&inode->lock);
56 17998567 : i_size = new_i_size ?: i_size_read(&inode->vfs_inode);
57 17998567 : if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
58 17998167 : inode->disk_i_size = i_size;
59 17998167 : goto out_unlock;
60 : }
61 :
62 400 : ret = find_contiguous_extent_bit(&inode->file_extent_tree, 0, &start,
63 : &end, EXTENT_DIRTY);
64 400 : if (!ret && start == 0)
65 377 : i_size = min(i_size, end + 1);
66 : else
67 : i_size = 0;
68 400 : inode->disk_i_size = i_size;
69 17998567 : out_unlock:
70 17998567 : spin_unlock(&inode->lock);
71 17998569 : }
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 : */
87 21498884 : int btrfs_inode_set_file_extent_range(struct btrfs_inode *inode, u64 start,
88 : u64 len)
89 : {
90 21498884 : if (len == 0)
91 : return 0;
92 :
93 21459236 : ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize));
94 :
95 21459236 : if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES))
96 : return 0;
97 5765 : 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 : */
115 5117564 : int btrfs_inode_clear_file_extent_range(struct btrfs_inode *inode, u64 start,
116 : u64 len)
117 : {
118 5117564 : if (len == 0)
119 : return 0;
120 :
121 5100877 : ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize) ||
122 : len == (u64)-1);
123 :
124 5100877 : if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES))
125 : return 0;
126 15366 : return clear_extent_bit(&inode->file_extent_tree, start,
127 : start + len - 1, EXTENT_DIRTY, NULL);
128 : }
129 :
130 : static size_t bytes_to_csum_size(const struct btrfs_fs_info *fs_info, u32 bytes)
131 : {
132 8266564 : ASSERT(IS_ALIGNED(bytes, fs_info->sectorsize));
133 :
134 8266564 : return (bytes >> fs_info->sectorsize_bits) * fs_info->csum_size;
135 : }
136 :
137 : static size_t csum_size_to_bytes(const struct btrfs_fs_info *fs_info, u32 csum_size)
138 : {
139 3213282 : ASSERT(IS_ALIGNED(csum_size, fs_info->csum_size));
140 :
141 3213282 : return (csum_size / fs_info->csum_size) << fs_info->sectorsize_bits;
142 : }
143 :
144 : static inline u32 max_ordered_sum_bytes(const struct btrfs_fs_info *fs_info)
145 : {
146 1574036 : u32 max_csum_size = round_down(PAGE_SIZE - sizeof(struct btrfs_ordered_sum),
147 : fs_info->csum_size);
148 :
149 1574036 : 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 : */
156 : static int btrfs_ordered_sum_size(struct btrfs_fs_info *fs_info, unsigned long bytes)
157 : {
158 5107171 : return sizeof(struct btrfs_ordered_sum) + bytes_to_csum_size(fs_info, bytes);
159 : }
160 :
161 2608413 : int btrfs_insert_hole_extent(struct btrfs_trans_handle *trans,
162 : struct btrfs_root *root,
163 : u64 objectid, u64 pos, u64 num_bytes)
164 : {
165 2608413 : int ret = 0;
166 2608413 : struct btrfs_file_extent_item *item;
167 2608413 : struct btrfs_key file_key;
168 2608413 : struct btrfs_path *path;
169 2608413 : struct extent_buffer *leaf;
170 :
171 2608413 : path = btrfs_alloc_path();
172 2608412 : if (!path)
173 : return -ENOMEM;
174 2608412 : file_key.objectid = objectid;
175 2608412 : file_key.offset = pos;
176 2608412 : file_key.type = BTRFS_EXTENT_DATA_KEY;
177 :
178 2608412 : ret = btrfs_insert_empty_item(trans, root, path, &file_key,
179 : sizeof(*item));
180 2608413 : if (ret < 0)
181 0 : goto out;
182 2608413 : BUG_ON(ret); /* Can't happen */
183 2608413 : leaf = path->nodes[0];
184 2608413 : item = btrfs_item_ptr(leaf, path->slots[0],
185 : struct btrfs_file_extent_item);
186 2608413 : btrfs_set_file_extent_disk_bytenr(leaf, item, 0);
187 2608413 : btrfs_set_file_extent_disk_num_bytes(leaf, item, 0);
188 2608413 : btrfs_set_file_extent_offset(leaf, item, 0);
189 2608413 : btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
190 2608413 : btrfs_set_file_extent_ram_bytes(leaf, item, num_bytes);
191 2608413 : btrfs_set_file_extent_generation(leaf, item, trans->transid);
192 2608413 : btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
193 2608413 : btrfs_set_file_extent_compression(leaf, item, 0);
194 2608413 : btrfs_set_file_extent_encryption(leaf, item, 0);
195 2608413 : btrfs_set_file_extent_other_encoding(leaf, item, 0);
196 :
197 2608413 : btrfs_mark_buffer_dirty(leaf);
198 2608413 : out:
199 2608413 : btrfs_free_path(path);
200 2608413 : return ret;
201 : }
202 :
203 : static struct btrfs_csum_item *
204 7691555 : btrfs_lookup_csum(struct btrfs_trans_handle *trans,
205 : struct btrfs_root *root,
206 : struct btrfs_path *path,
207 : u64 bytenr, int cow)
208 : {
209 7691555 : struct btrfs_fs_info *fs_info = root->fs_info;
210 7691555 : int ret;
211 7691555 : struct btrfs_key file_key;
212 7691555 : struct btrfs_key found_key;
213 7691555 : struct btrfs_csum_item *item;
214 7691555 : struct extent_buffer *leaf;
215 7691555 : u64 csum_offset = 0;
216 7691555 : const u32 csum_size = fs_info->csum_size;
217 7691555 : int csums_in_item;
218 :
219 7691555 : file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
220 7691555 : file_key.offset = bytenr;
221 7691555 : file_key.type = BTRFS_EXTENT_CSUM_KEY;
222 7691555 : ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
223 7692555 : if (ret < 0)
224 0 : goto fail;
225 7692555 : leaf = path->nodes[0];
226 7692555 : if (ret > 0) {
227 7125436 : ret = 1;
228 7125436 : if (path->slots[0] == 0)
229 11943 : goto fail;
230 7113493 : path->slots[0]--;
231 7113493 : btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
232 7112802 : if (found_key.type != BTRFS_EXTENT_CSUM_KEY)
233 2410 : goto fail;
234 :
235 7110392 : csum_offset = (bytenr - found_key.offset) >>
236 7110392 : fs_info->sectorsize_bits;
237 7110392 : csums_in_item = btrfs_item_size(leaf, path->slots[0]);
238 7110186 : csums_in_item /= csum_size;
239 :
240 7110186 : if (csum_offset == csums_in_item) {
241 3477894 : ret = -EFBIG;
242 3477894 : goto fail;
243 3632292 : } else if (csum_offset > csums_in_item) {
244 1988122 : goto fail;
245 : }
246 : }
247 2211289 : item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
248 2211181 : item = (struct btrfs_csum_item *)((unsigned char *)item +
249 2211181 : csum_offset * csum_size);
250 2211181 : return item;
251 : fail:
252 5480369 : if (ret > 0)
253 : ret = -ENOENT;
254 5480369 : return ERR_PTR(ret);
255 : }
256 :
257 32724060 : int 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 32724060 : struct btrfs_key file_key;
263 32724060 : int ins_len = mod < 0 ? -1 : 0;
264 32724060 : int cow = mod != 0;
265 :
266 32724060 : file_key.objectid = objectid;
267 32724060 : file_key.offset = offset;
268 32724060 : file_key.type = BTRFS_EXTENT_DATA_KEY;
269 :
270 32724060 : 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 : */
282 2331906 : static 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 2331906 : struct btrfs_root *csum_root;
287 2331906 : struct btrfs_csum_item *item = NULL;
288 2331906 : struct btrfs_key key;
289 2331906 : const u32 sectorsize = fs_info->sectorsize;
290 2331906 : const u32 csum_size = fs_info->csum_size;
291 2331906 : u32 itemsize;
292 2331906 : int ret;
293 2331906 : u64 csum_start;
294 2331906 : u64 csum_len;
295 :
296 2331906 : ASSERT(IS_ALIGNED(disk_bytenr, sectorsize) &&
297 : IS_ALIGNED(len, sectorsize));
298 :
299 : /* Check if the current csum item covers disk_bytenr */
300 2331906 : if (path->nodes[0]) {
301 152816 : item = btrfs_item_ptr(path->nodes[0], path->slots[0],
302 : struct btrfs_csum_item);
303 152815 : btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
304 152815 : itemsize = btrfs_item_size(path->nodes[0], path->slots[0]);
305 :
306 152816 : csum_start = key.offset;
307 152816 : csum_len = (itemsize / csum_size) * sectorsize;
308 :
309 152816 : if (in_range(disk_bytenr, csum_start, csum_len))
310 1 : goto found;
311 : }
312 :
313 : /* Current item doesn't contain the desired range, search again */
314 2331905 : btrfs_release_path(path);
315 2332926 : csum_root = btrfs_csum_root(fs_info, disk_bytenr);
316 2333294 : item = btrfs_lookup_csum(NULL, csum_root, path, disk_bytenr, 0);
317 2333450 : if (IS_ERR(item)) {
318 121428 : ret = PTR_ERR(item);
319 121428 : goto out;
320 : }
321 2212022 : btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
322 2211707 : itemsize = btrfs_item_size(path->nodes[0], path->slots[0]);
323 :
324 2212355 : csum_start = key.offset;
325 2212355 : csum_len = (itemsize / csum_size) * sectorsize;
326 2212355 : ASSERT(in_range(disk_bytenr, csum_start, csum_len));
327 :
328 2212356 : found:
329 2212356 : ret = (min(csum_start + csum_len, disk_bytenr + len) -
330 2212356 : disk_bytenr) >> fs_info->sectorsize_bits;
331 2212356 : read_extent_buffer(path->nodes[0], dst, (unsigned long)item,
332 2212356 : ret * csum_size);
333 2333781 : out:
334 2333781 : if (ret == -ENOENT || ret == -EFBIG)
335 121428 : ret = 0;
336 2333781 : 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 : */
344 2181788 : blk_status_t btrfs_lookup_bio_sums(struct btrfs_bio *bbio)
345 : {
346 2181788 : struct btrfs_inode *inode = bbio->inode;
347 2181788 : struct btrfs_fs_info *fs_info = inode->root->fs_info;
348 2181788 : struct bio *bio = &bbio->bio;
349 2181788 : struct btrfs_path *path;
350 2181788 : const u32 sectorsize = fs_info->sectorsize;
351 2181788 : const u32 csum_size = fs_info->csum_size;
352 2181788 : u32 orig_len = bio->bi_iter.bi_size;
353 2181788 : u64 orig_disk_bytenr = bio->bi_iter.bi_sector << SECTOR_SHIFT;
354 2181788 : const unsigned int nblocks = orig_len >> fs_info->sectorsize_bits;
355 2181788 : blk_status_t ret = BLK_STS_OK;
356 2181788 : u32 bio_offset = 0;
357 :
358 4363240 : if ((inode->flags & BTRFS_INODE_NODATASUM) ||
359 2181452 : 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 2181452 : ASSERT(bio_op(bio) == REQ_OP_READ);
375 2181452 : path = btrfs_alloc_path();
376 2180741 : if (!path)
377 : return BLK_STS_RESOURCE;
378 :
379 2180741 : if (nblocks * csum_size > BTRFS_BIO_INLINE_CSUM_SIZE) {
380 64958 : bbio->csum = kmalloc_array(nblocks, csum_size, GFP_NOFS);
381 64959 : if (!bbio->csum) {
382 0 : btrfs_free_path(path);
383 0 : return BLK_STS_RESOURCE;
384 : }
385 : } else {
386 2115783 : 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 2180742 : if (nblocks > fs_info->csums_per_leaf)
394 0 : 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 4361484 : if (btrfs_is_free_space_inode(inode)) {
403 0 : path->search_commit_root = 1;
404 0 : path->skip_locking = 1;
405 : }
406 :
407 4514570 : while (bio_offset < orig_len) {
408 2333461 : int count;
409 2333461 : u64 cur_disk_bytenr = orig_disk_bytenr + bio_offset;
410 2333461 : u8 *csum_dst = bbio->csum +
411 2333461 : (bio_offset >> fs_info->sectorsize_bits) * csum_size;
412 :
413 2333461 : count = search_csum_tree(fs_info, path, cur_disk_bytenr,
414 2333461 : orig_len - bio_offset, csum_dst);
415 2333828 : if (count < 0) {
416 0 : ret = errno_to_blk_status(count);
417 0 : if (bbio->csum != bbio->csum_inline)
418 0 : kfree(bbio->csum);
419 0 : bbio->csum = NULL;
420 0 : 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 2333828 : if (count == 0) {
434 121428 : memset(csum_dst, 0, csum_size);
435 121428 : count = 1;
436 :
437 121428 : if (inode->root->root_key.objectid ==
438 : BTRFS_DATA_RELOC_TREE_OBJECTID) {
439 121428 : u64 file_offset = bbio->file_offset + bio_offset;
440 :
441 121428 : set_extent_bit(&inode->io_tree, file_offset,
442 121428 : file_offset + sectorsize - 1,
443 : EXTENT_NODATASUM, NULL);
444 : } else {
445 0 : 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 2333828 : bio_offset += count * sectorsize;
451 : }
452 :
453 2181109 : btrfs_free_path(path);
454 2181109 : return ret;
455 : }
456 :
457 1874033 : int 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 1874033 : struct btrfs_fs_info *fs_info = root->fs_info;
462 1874033 : struct btrfs_key key;
463 1874033 : struct btrfs_path *path;
464 1874033 : struct extent_buffer *leaf;
465 1874033 : struct btrfs_ordered_sum *sums;
466 1874033 : struct btrfs_csum_item *item;
467 1874033 : LIST_HEAD(tmplist);
468 1874033 : int ret;
469 :
470 1874033 : ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
471 : IS_ALIGNED(end + 1, fs_info->sectorsize));
472 :
473 1874033 : path = btrfs_alloc_path();
474 1874034 : if (!path)
475 : return -ENOMEM;
476 :
477 1874034 : path->nowait = nowait;
478 1874034 : if (search_commit) {
479 0 : path->skip_locking = 1;
480 0 : path->reada = READA_FORWARD;
481 0 : path->search_commit_root = 1;
482 : }
483 :
484 1874034 : key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
485 1874034 : key.offset = start;
486 1874034 : key.type = BTRFS_EXTENT_CSUM_KEY;
487 :
488 1874034 : ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
489 1874036 : if (ret < 0)
490 0 : goto fail;
491 1874036 : if (ret > 0 && path->slots[0] > 0) {
492 1454599 : leaf = path->nodes[0];
493 1454599 : 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 1454598 : if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
511 1454059 : key.type == BTRFS_EXTENT_CSUM_KEY) {
512 1454058 : if (bytes_to_csum_size(fs_info, start - key.offset) <
513 1454059 : btrfs_item_size(leaf, path->slots[0] - 1))
514 1152163 : path->slots[0]--;
515 : }
516 : }
517 :
518 3447496 : while (start <= end) {
519 1877644 : u64 csum_end;
520 :
521 1877644 : leaf = path->nodes[0];
522 1877644 : if (path->slots[0] >= btrfs_header_nritems(leaf)) {
523 42273 : ret = btrfs_next_leaf(root, path);
524 42273 : if (ret < 0)
525 0 : goto fail;
526 42273 : if (ret > 0)
527 : break;
528 17907 : leaf = path->nodes[0];
529 : }
530 :
531 1853278 : btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
532 1853275 : if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
533 1853275 : key.type != BTRFS_EXTENT_CSUM_KEY ||
534 1853275 : key.offset > end)
535 : break;
536 :
537 1573465 : if (key.offset > start)
538 : start = key.offset;
539 :
540 1573465 : csum_end = key.offset + csum_size_to_bytes(fs_info,
541 : btrfs_item_size(leaf, path->slots[0]));
542 1573464 : if (csum_end <= start) {
543 18 : path->slots[0]++;
544 18 : continue;
545 : }
546 :
547 1573446 : csum_end = min(csum_end, end + 1);
548 1573446 : item = btrfs_item_ptr(path->nodes[0], path->slots[0],
549 : struct btrfs_csum_item);
550 3147480 : while (start < csum_end) {
551 1574036 : unsigned long offset;
552 1574036 : size_t size;
553 :
554 1574036 : size = min_t(size_t, csum_end - start,
555 : max_ordered_sum_bytes(fs_info));
556 1574036 : sums = kzalloc(btrfs_ordered_sum_size(fs_info, size),
557 : GFP_NOFS);
558 1574032 : if (!sums) {
559 0 : ret = -ENOMEM;
560 0 : goto fail;
561 : }
562 :
563 1574032 : sums->logical = start;
564 1574032 : sums->len = size;
565 :
566 1574032 : offset = bytes_to_csum_size(fs_info, start - key.offset);
567 :
568 1574032 : read_extent_buffer(path->nodes[0],
569 1574032 : sums->sums,
570 : ((unsigned long)item) + offset,
571 : bytes_to_csum_size(fs_info, size));
572 :
573 1574033 : start += size;
574 1574033 : list_add_tail(&sums->list, &tmplist);
575 : }
576 1573444 : path->slots[0]++;
577 : }
578 : ret = 0;
579 1874028 : fail:
580 1874027 : while (ret < 0 && !list_empty(&tmplist)) {
581 0 : sums = list_entry(tmplist.next, struct btrfs_ordered_sum, list);
582 0 : list_del(&sums->list);
583 0 : kfree(sums);
584 : }
585 1874027 : list_splice_tail(&tmplist, list);
586 :
587 1874027 : btrfs_free_path(path);
588 1874027 : 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 : */
600 65540 : int btrfs_lookup_csums_bitmap(struct btrfs_root *root, u64 start, u64 end,
601 : u8 *csum_buf, unsigned long *csum_bitmap,
602 : bool search_commit)
603 : {
604 65540 : struct btrfs_fs_info *fs_info = root->fs_info;
605 65540 : struct btrfs_key key;
606 65540 : struct btrfs_path *path;
607 65540 : struct extent_buffer *leaf;
608 65540 : struct btrfs_csum_item *item;
609 65540 : const u64 orig_start = start;
610 65540 : int ret;
611 :
612 65540 : ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
613 : IS_ALIGNED(end + 1, fs_info->sectorsize));
614 :
615 65540 : path = btrfs_alloc_path();
616 65540 : if (!path)
617 : return -ENOMEM;
618 :
619 65540 : if (search_commit) {
620 65540 : path->skip_locking = 1;
621 65540 : path->reada = READA_FORWARD;
622 65540 : path->search_commit_root = 1;
623 : }
624 :
625 65540 : key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
626 65540 : key.type = BTRFS_EXTENT_CSUM_KEY;
627 65540 : key.offset = start;
628 :
629 65540 : ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
630 65540 : if (ret < 0)
631 0 : goto fail;
632 65540 : if (ret > 0 && path->slots[0] > 0) {
633 65520 : leaf = path->nodes[0];
634 65520 : btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
635 :
636 : /*
637 : * There are two cases we can hit here for the previous csum
638 : * item:
639 : *
640 : * |<- search range ->|
641 : * |<- csum item ->|
642 : *
643 : * Or
644 : * |<- search range ->|
645 : * |<- csum item ->|
646 : *
647 : * Check if the previous csum item covers the leading part of
648 : * the search range. If so we have to start from previous csum
649 : * item.
650 : */
651 65520 : if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
652 65520 : key.type == BTRFS_EXTENT_CSUM_KEY) {
653 65520 : if (bytes_to_csum_size(fs_info, start - key.offset) <
654 65520 : btrfs_item_size(leaf, path->slots[0] - 1))
655 65520 : path->slots[0]--;
656 : }
657 : }
658 :
659 131322 : while (start <= end) {
660 65782 : u64 csum_end;
661 :
662 65782 : leaf = path->nodes[0];
663 65782 : if (path->slots[0] >= btrfs_header_nritems(leaf)) {
664 242 : ret = btrfs_next_leaf(root, path);
665 242 : if (ret < 0)
666 0 : goto fail;
667 242 : if (ret > 0)
668 : break;
669 242 : leaf = path->nodes[0];
670 : }
671 :
672 65782 : btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
673 65782 : if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
674 65782 : key.type != BTRFS_EXTENT_CSUM_KEY ||
675 65782 : key.offset > end)
676 : break;
677 :
678 65782 : if (key.offset > start)
679 : start = key.offset;
680 :
681 65782 : csum_end = key.offset + csum_size_to_bytes(fs_info,
682 : btrfs_item_size(leaf, path->slots[0]));
683 65782 : if (csum_end <= start) {
684 0 : path->slots[0]++;
685 0 : continue;
686 : }
687 :
688 65782 : csum_end = min(csum_end, end + 1);
689 65782 : item = btrfs_item_ptr(path->nodes[0], path->slots[0],
690 : struct btrfs_csum_item);
691 131564 : while (start < csum_end) {
692 65782 : unsigned long offset;
693 65782 : size_t size;
694 65782 : u8 *csum_dest = csum_buf + bytes_to_csum_size(fs_info,
695 : start - orig_start);
696 :
697 65782 : size = min_t(size_t, csum_end - start, end + 1 - start);
698 :
699 65782 : offset = bytes_to_csum_size(fs_info, start - key.offset);
700 :
701 65782 : read_extent_buffer(path->nodes[0], csum_dest,
702 : ((unsigned long)item) + offset,
703 : bytes_to_csum_size(fs_info, size));
704 :
705 65782 : bitmap_set(csum_bitmap,
706 65782 : (start - orig_start) >> fs_info->sectorsize_bits,
707 65782 : size >> fs_info->sectorsize_bits);
708 :
709 65782 : start += size;
710 : }
711 65782 : path->slots[0]++;
712 : }
713 : ret = 0;
714 65540 : fail:
715 65540 : btrfs_free_path(path);
716 65540 : return ret;
717 : }
718 :
719 : /*
720 : * Calculate checksums of the data contained inside a bio.
721 : */
722 3533135 : blk_status_t btrfs_csum_one_bio(struct btrfs_bio *bbio)
723 : {
724 3533135 : struct btrfs_ordered_extent *ordered = bbio->ordered;
725 3533135 : struct btrfs_inode *inode = bbio->inode;
726 3533135 : struct btrfs_fs_info *fs_info = inode->root->fs_info;
727 3533135 : SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
728 3533135 : struct bio *bio = &bbio->bio;
729 3533135 : struct btrfs_ordered_sum *sums;
730 3533135 : char *data;
731 3533135 : struct bvec_iter iter;
732 3533135 : struct bio_vec bvec;
733 3533135 : int index;
734 3533135 : unsigned int blockcount;
735 3533135 : int i;
736 3533135 : unsigned nofs_flag;
737 :
738 3533135 : nofs_flag = memalloc_nofs_save();
739 3533135 : sums = kvzalloc(btrfs_ordered_sum_size(fs_info, bio->bi_iter.bi_size),
740 : GFP_KERNEL);
741 3533226 : memalloc_nofs_restore(nofs_flag);
742 :
743 3533226 : if (!sums)
744 : return BLK_STS_RESOURCE;
745 :
746 3533226 : sums->len = bio->bi_iter.bi_size;
747 3533226 : INIT_LIST_HEAD(&sums->list);
748 :
749 3533226 : sums->logical = bio->bi_iter.bi_sector << SECTOR_SHIFT;
750 3533226 : index = 0;
751 :
752 3533226 : shash->tfm = fs_info->csum_shash;
753 :
754 59769822 : bio_for_each_segment(bvec, bio, iter) {
755 56232929 : blockcount = BTRFS_BYTES_TO_BLKS(fs_info,
756 : bvec.bv_len + fs_info->sectorsize
757 : - 1);
758 :
759 112472010 : for (i = 0; i < blockcount; i++) {
760 56235414 : data = bvec_kmap_local(&bvec);
761 56235414 : crypto_shash_digest(shash,
762 56235414 : data + (i * fs_info->sectorsize),
763 : fs_info->sectorsize,
764 56235414 : sums->sums + index);
765 56239081 : kunmap_local(data);
766 56239081 : index += fs_info->csum_size;
767 : }
768 :
769 : }
770 :
771 3533211 : bbio->sums = sums;
772 3533211 : btrfs_add_ordered_sum(ordered, sums);
773 3533211 : return 0;
774 : }
775 :
776 : /*
777 : * Nodatasum I/O on zoned file systems still requires an btrfs_ordered_sum to
778 : * record the updated logical address on Zone Append completion.
779 : * Allocate just the structure with an empty sums array here for that case.
780 : */
781 0 : blk_status_t btrfs_alloc_dummy_sum(struct btrfs_bio *bbio)
782 : {
783 0 : bbio->sums = kmalloc(sizeof(*bbio->sums), GFP_NOFS);
784 0 : if (!bbio->sums)
785 : return BLK_STS_RESOURCE;
786 0 : bbio->sums->len = bbio->bio.bi_iter.bi_size;
787 0 : bbio->sums->logical = bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT;
788 0 : btrfs_add_ordered_sum(bbio->ordered, bbio->sums);
789 0 : return 0;
790 : }
791 :
792 : /*
793 : * Remove one checksum overlapping a range.
794 : *
795 : * This expects the key to describe the csum pointed to by the path, and it
796 : * expects the csum to overlap the range [bytenr, len]
797 : *
798 : * The csum should not be entirely contained in the range and the range should
799 : * not be entirely contained in the csum.
800 : *
801 : * This calls btrfs_truncate_item with the correct args based on the overlap,
802 : * and fixes up the key as required.
803 : */
804 2248503 : static noinline void truncate_one_csum(struct btrfs_fs_info *fs_info,
805 : struct btrfs_path *path,
806 : struct btrfs_key *key,
807 : u64 bytenr, u64 len)
808 : {
809 2248503 : struct extent_buffer *leaf;
810 2248503 : const u32 csum_size = fs_info->csum_size;
811 2248503 : u64 csum_end;
812 2248503 : u64 end_byte = bytenr + len;
813 2248503 : u32 blocksize_bits = fs_info->sectorsize_bits;
814 :
815 2248503 : leaf = path->nodes[0];
816 2248503 : csum_end = btrfs_item_size(leaf, path->slots[0]) / csum_size;
817 2248503 : csum_end <<= blocksize_bits;
818 2248503 : csum_end += key->offset;
819 :
820 2495411 : if (key->offset < bytenr && csum_end <= end_byte) {
821 : /*
822 : * [ bytenr - len ]
823 : * [ ]
824 : * [csum ]
825 : * A simple truncate off the end of the item
826 : */
827 246908 : u32 new_size = (bytenr - key->offset) >> blocksize_bits;
828 246908 : new_size *= csum_size;
829 246908 : btrfs_truncate_item(path, new_size, 1);
830 2001595 : } else if (key->offset >= bytenr && csum_end > end_byte &&
831 2001596 : end_byte > key->offset) {
832 : /*
833 : * [ bytenr - len ]
834 : * [ ]
835 : * [csum ]
836 : * we need to truncate from the beginning of the csum
837 : */
838 2001595 : u32 new_size = (csum_end - end_byte) >> blocksize_bits;
839 2001595 : new_size *= csum_size;
840 :
841 2001595 : btrfs_truncate_item(path, new_size, 0);
842 :
843 2001596 : key->offset = end_byte;
844 2001596 : btrfs_set_item_key_safe(fs_info, path, key);
845 : } else {
846 0 : BUG();
847 : }
848 2248504 : }
849 :
850 : /*
851 : * Delete the csum items from the csum tree for a given range of bytes.
852 : */
853 3907169 : int btrfs_del_csums(struct btrfs_trans_handle *trans,
854 : struct btrfs_root *root, u64 bytenr, u64 len)
855 : {
856 3907169 : struct btrfs_fs_info *fs_info = trans->fs_info;
857 3907169 : struct btrfs_path *path;
858 3907169 : struct btrfs_key key;
859 3907169 : u64 end_byte = bytenr + len;
860 3907169 : u64 csum_end;
861 3907169 : struct extent_buffer *leaf;
862 3907169 : int ret = 0;
863 3907169 : const u32 csum_size = fs_info->csum_size;
864 3907169 : u32 blocksize_bits = fs_info->sectorsize_bits;
865 :
866 3907169 : ASSERT(root->root_key.objectid == BTRFS_CSUM_TREE_OBJECTID ||
867 : root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
868 :
869 3907169 : path = btrfs_alloc_path();
870 3907169 : if (!path)
871 : return -ENOMEM;
872 :
873 8471067 : while (1) {
874 6189118 : key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
875 6189118 : key.offset = end_byte - 1;
876 6189118 : key.type = BTRFS_EXTENT_CSUM_KEY;
877 :
878 6189118 : ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
879 6189113 : if (ret > 0) {
880 6189113 : ret = 0;
881 6189113 : if (path->slots[0] == 0)
882 : break;
883 6053107 : path->slots[0]--;
884 0 : } else if (ret < 0) {
885 : break;
886 : }
887 :
888 6053107 : leaf = path->nodes[0];
889 6053107 : btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
890 :
891 6053107 : if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
892 6052707 : key.type != BTRFS_EXTENT_CSUM_KEY) {
893 : break;
894 : }
895 :
896 6052707 : if (key.offset >= end_byte)
897 : break;
898 :
899 6052707 : csum_end = btrfs_item_size(leaf, path->slots[0]) / csum_size;
900 6052712 : csum_end <<= blocksize_bits;
901 6052712 : csum_end += key.offset;
902 :
903 : /* this csum ends before we start, we're done */
904 6052712 : if (csum_end <= bytenr)
905 : break;
906 :
907 : /* delete the entire item, it is inside our range */
908 3310345 : if (key.offset >= bytenr && csum_end <= end_byte) {
909 868063 : int del_nr = 1;
910 :
911 : /*
912 : * Check how many csum items preceding this one in this
913 : * leaf correspond to our range and then delete them all
914 : * at once.
915 : */
916 868063 : if (key.offset > bytenr && path->slots[0] > 0) {
917 74588 : int slot = path->slots[0] - 1;
918 :
919 130659 : while (slot >= 0) {
920 114285 : struct btrfs_key pk;
921 :
922 114285 : btrfs_item_key_to_cpu(leaf, &pk, slot);
923 114285 : if (pk.offset < bytenr ||
924 56071 : pk.type != BTRFS_EXTENT_CSUM_KEY ||
925 56071 : pk.objectid !=
926 : BTRFS_EXTENT_CSUM_OBJECTID)
927 : break;
928 56071 : path->slots[0] = slot;
929 56071 : del_nr++;
930 56071 : key.offset = pk.offset;
931 56071 : slot--;
932 : }
933 : }
934 868063 : ret = btrfs_del_items(trans, root, path,
935 : path->slots[0], del_nr);
936 868063 : if (ret)
937 : break;
938 868063 : if (key.offset == bytenr)
939 : break;
940 2442282 : } else if (key.offset < bytenr && csum_end > end_byte) {
941 193779 : unsigned long offset;
942 193779 : unsigned long shift_len;
943 193779 : unsigned long item_offset;
944 : /*
945 : * [ bytenr - len ]
946 : * [csum ]
947 : *
948 : * Our bytes are in the middle of the csum,
949 : * we need to split this item and insert a new one.
950 : *
951 : * But we can't drop the path because the
952 : * csum could change, get removed, extended etc.
953 : *
954 : * The trick here is the max size of a csum item leaves
955 : * enough room in the tree block for a single
956 : * item header. So, we split the item in place,
957 : * adding a new header pointing to the existing
958 : * bytes. Then we loop around again and we have
959 : * a nicely formed csum item that we can neatly
960 : * truncate.
961 : */
962 193779 : offset = (bytenr - key.offset) >> blocksize_bits;
963 193779 : offset *= csum_size;
964 :
965 193779 : shift_len = (len >> blocksize_bits) * csum_size;
966 :
967 193779 : item_offset = btrfs_item_ptr_offset(leaf,
968 : path->slots[0]);
969 :
970 193779 : memzero_extent_buffer(leaf, item_offset + offset,
971 : shift_len);
972 193779 : key.offset = bytenr;
973 :
974 : /*
975 : * btrfs_split_item returns -EAGAIN when the
976 : * item changed size or key
977 : */
978 193779 : ret = btrfs_split_item(trans, root, path, &key, offset);
979 193779 : if (ret && ret != -EAGAIN) {
980 0 : btrfs_abort_transaction(trans, ret);
981 0 : break;
982 : }
983 193779 : ret = 0;
984 :
985 193779 : key.offset = end_byte - 1;
986 : } else {
987 2248503 : truncate_one_csum(fs_info, path, &key, bytenr, len);
988 2248504 : if (key.offset < bytenr)
989 : break;
990 : }
991 2281949 : btrfs_release_path(path);
992 : }
993 3907170 : btrfs_free_path(path);
994 3907170 : return ret;
995 : }
996 :
997 2671852 : static int find_next_csum_offset(struct btrfs_root *root,
998 : struct btrfs_path *path,
999 : u64 *next_offset)
1000 : {
1001 2671852 : const u32 nritems = btrfs_header_nritems(path->nodes[0]);
1002 2671852 : struct btrfs_key found_key;
1003 2671852 : int slot = path->slots[0] + 1;
1004 2671852 : int ret;
1005 :
1006 2671852 : if (nritems == 0 || slot >= nritems) {
1007 544879 : ret = btrfs_next_leaf(root, path);
1008 544840 : if (ret < 0) {
1009 : return ret;
1010 544840 : } else if (ret > 0) {
1011 495887 : *next_offset = (u64)-1;
1012 495887 : return 0;
1013 : }
1014 48953 : slot = path->slots[0];
1015 : }
1016 :
1017 2175926 : btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
1018 :
1019 2175924 : if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1020 2175922 : found_key.type != BTRFS_EXTENT_CSUM_KEY)
1021 0 : *next_offset = (u64)-1;
1022 : else
1023 2175924 : *next_offset = found_key.offset;
1024 :
1025 : return 0;
1026 : }
1027 :
1028 5312367 : int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
1029 : struct btrfs_root *root,
1030 : struct btrfs_ordered_sum *sums)
1031 : {
1032 5312367 : struct btrfs_fs_info *fs_info = root->fs_info;
1033 5312367 : struct btrfs_key file_key;
1034 5312367 : struct btrfs_key found_key;
1035 5312367 : struct btrfs_path *path;
1036 5312367 : struct btrfs_csum_item *item;
1037 5312367 : struct btrfs_csum_item *item_end;
1038 5312367 : struct extent_buffer *leaf = NULL;
1039 5312367 : u64 next_offset;
1040 5312367 : u64 total_bytes = 0;
1041 5312367 : u64 csum_offset;
1042 5312367 : u64 bytenr;
1043 5312367 : u32 ins_size;
1044 5312367 : int index = 0;
1045 5312367 : int found_next;
1046 5312367 : int ret;
1047 5312367 : const u32 csum_size = fs_info->csum_size;
1048 :
1049 5312367 : path = btrfs_alloc_path();
1050 5312128 : if (!path)
1051 : return -ENOMEM;
1052 5312128 : again:
1053 5358511 : next_offset = (u64)-1;
1054 5358511 : found_next = 0;
1055 5358511 : bytenr = sums->logical + total_bytes;
1056 5358511 : file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
1057 5358511 : file_key.offset = bytenr;
1058 5358511 : file_key.type = BTRFS_EXTENT_CSUM_KEY;
1059 :
1060 5358511 : item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
1061 5358941 : if (!IS_ERR(item)) {
1062 0 : ret = 0;
1063 0 : leaf = path->nodes[0];
1064 0 : item_end = btrfs_item_ptr(leaf, path->slots[0],
1065 : struct btrfs_csum_item);
1066 0 : item_end = (struct btrfs_csum_item *)((char *)item_end +
1067 0 : btrfs_item_size(leaf, path->slots[0]));
1068 0 : goto found;
1069 : }
1070 5358941 : ret = PTR_ERR(item);
1071 5358941 : if (ret != -EFBIG && ret != -ENOENT)
1072 0 : goto out;
1073 :
1074 5358941 : if (ret == -EFBIG) {
1075 3477439 : u32 item_size;
1076 : /* we found one, but it isn't big enough yet */
1077 3477439 : leaf = path->nodes[0];
1078 3477439 : item_size = btrfs_item_size(leaf, path->slots[0]);
1079 3477417 : if ((item_size / csum_size) >=
1080 3477417 : MAX_CSUM_ITEMS(fs_info, csum_size)) {
1081 : /* already at max size, make a new one */
1082 6409 : goto insert;
1083 : }
1084 : } else {
1085 : /* We didn't find a csum item, insert one. */
1086 1881502 : ret = find_next_csum_offset(root, path, &next_offset);
1087 1881430 : if (ret < 0)
1088 0 : goto out;
1089 1881430 : found_next = 1;
1090 1881430 : goto insert;
1091 : }
1092 :
1093 : /*
1094 : * At this point, we know the tree has a checksum item that ends at an
1095 : * offset matching the start of the checksum range we want to insert.
1096 : * We try to extend that item as much as possible and then add as many
1097 : * checksums to it as they fit.
1098 : *
1099 : * First check if the leaf has enough free space for at least one
1100 : * checksum. If it has go directly to the item extension code, otherwise
1101 : * release the path and do a search for insertion before the extension.
1102 : */
1103 3471008 : if (btrfs_leaf_free_space(leaf) >= csum_size) {
1104 3436067 : btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1105 3436071 : csum_offset = (bytenr - found_key.offset) >>
1106 3436071 : fs_info->sectorsize_bits;
1107 3436071 : goto extend_csum;
1108 : }
1109 :
1110 34956 : btrfs_release_path(path);
1111 34956 : path->search_for_extension = 1;
1112 34956 : ret = btrfs_search_slot(trans, root, &file_key, path,
1113 : csum_size, 1);
1114 34956 : path->search_for_extension = 0;
1115 34956 : if (ret < 0)
1116 0 : goto out;
1117 :
1118 34956 : if (ret > 0) {
1119 34956 : if (path->slots[0] == 0)
1120 10377 : goto insert;
1121 24579 : path->slots[0]--;
1122 : }
1123 :
1124 24579 : leaf = path->nodes[0];
1125 24579 : btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1126 24579 : csum_offset = (bytenr - found_key.offset) >> fs_info->sectorsize_bits;
1127 :
1128 24579 : if (found_key.type != BTRFS_EXTENT_CSUM_KEY ||
1129 24579 : found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1130 24579 : csum_offset >= MAX_CSUM_ITEMS(fs_info, csum_size)) {
1131 0 : goto insert;
1132 : }
1133 :
1134 24579 : extend_csum:
1135 3460650 : if (csum_offset == btrfs_item_size(leaf, path->slots[0]) /
1136 : csum_size) {
1137 3460643 : int extend_nr;
1138 3460643 : u64 tmp;
1139 3460643 : u32 diff;
1140 :
1141 3460643 : tmp = sums->len - total_bytes;
1142 3460643 : tmp >>= fs_info->sectorsize_bits;
1143 3460643 : WARN_ON(tmp < 1);
1144 3460643 : extend_nr = max_t(int, 1, tmp);
1145 :
1146 : /*
1147 : * A log tree can already have checksum items with a subset of
1148 : * the checksums we are trying to log. This can happen after
1149 : * doing a sequence of partial writes into prealloc extents and
1150 : * fsyncs in between, with a full fsync logging a larger subrange
1151 : * of an extent for which a previous fast fsync logged a smaller
1152 : * subrange. And this happens in particular due to merging file
1153 : * extent items when we complete an ordered extent for a range
1154 : * covered by a prealloc extent - this is done at
1155 : * btrfs_mark_extent_written().
1156 : *
1157 : * So if we try to extend the previous checksum item, which has
1158 : * a range that ends at the start of the range we want to insert,
1159 : * make sure we don't extend beyond the start offset of the next
1160 : * checksum item. If we are at the last item in the leaf, then
1161 : * forget the optimization of extending and add a new checksum
1162 : * item - it is not worth the complexity of releasing the path,
1163 : * getting the first key for the next leaf, repeat the btree
1164 : * search, etc, because log trees are temporary anyway and it
1165 : * would only save a few bytes of leaf space.
1166 : */
1167 3460643 : if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
1168 790372 : if (path->slots[0] + 1 >=
1169 790372 : btrfs_header_nritems(path->nodes[0])) {
1170 99200 : ret = find_next_csum_offset(root, path, &next_offset);
1171 99200 : if (ret < 0)
1172 0 : goto out;
1173 99200 : found_next = 1;
1174 99200 : goto insert;
1175 : }
1176 :
1177 691172 : ret = find_next_csum_offset(root, path, &next_offset);
1178 691173 : if (ret < 0)
1179 0 : goto out;
1180 :
1181 691173 : tmp = (next_offset - bytenr) >> fs_info->sectorsize_bits;
1182 691173 : if (tmp <= INT_MAX)
1183 691173 : extend_nr = min_t(int, extend_nr, tmp);
1184 : }
1185 :
1186 3361444 : diff = (csum_offset + extend_nr) * csum_size;
1187 3361444 : diff = min(diff,
1188 : MAX_CSUM_ITEMS(fs_info, csum_size) * csum_size);
1189 :
1190 3361444 : diff = diff - btrfs_item_size(leaf, path->slots[0]);
1191 3361424 : diff = min_t(u32, btrfs_leaf_free_space(leaf), diff);
1192 3361432 : diff /= csum_size;
1193 3361432 : diff *= csum_size;
1194 :
1195 3361432 : btrfs_extend_item(path, diff);
1196 3361479 : ret = 0;
1197 3361479 : goto csum;
1198 : }
1199 :
1200 0 : insert:
1201 1997416 : btrfs_release_path(path);
1202 1997508 : csum_offset = 0;
1203 1997508 : if (found_next) {
1204 1980719 : u64 tmp;
1205 :
1206 1980719 : tmp = sums->len - total_bytes;
1207 1980719 : tmp >>= fs_info->sectorsize_bits;
1208 1980719 : tmp = min(tmp, (next_offset - file_key.offset) >>
1209 : fs_info->sectorsize_bits);
1210 :
1211 1980719 : tmp = max_t(u64, 1, tmp);
1212 1980719 : tmp = min_t(u64, tmp, MAX_CSUM_ITEMS(fs_info, csum_size));
1213 1980719 : ins_size = csum_size * tmp;
1214 : } else {
1215 : ins_size = csum_size;
1216 : }
1217 1997508 : ret = btrfs_insert_empty_item(trans, root, path, &file_key,
1218 : ins_size);
1219 1997499 : if (ret < 0)
1220 0 : goto out;
1221 1997499 : if (WARN_ON(ret != 0))
1222 0 : goto out;
1223 1997499 : leaf = path->nodes[0];
1224 5358978 : csum:
1225 5358978 : item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
1226 10717859 : item_end = (struct btrfs_csum_item *)((unsigned char *)item +
1227 5358932 : btrfs_item_size(leaf, path->slots[0]));
1228 5358927 : item = (struct btrfs_csum_item *)((unsigned char *)item +
1229 5358927 : csum_offset * csum_size);
1230 5358927 : found:
1231 5358927 : ins_size = (u32)(sums->len - total_bytes) >> fs_info->sectorsize_bits;
1232 5358927 : ins_size *= csum_size;
1233 5358927 : ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item,
1234 : ins_size);
1235 5358927 : write_extent_buffer(leaf, sums->sums + index, (unsigned long)item,
1236 : ins_size);
1237 :
1238 5358935 : index += ins_size;
1239 5358935 : ins_size /= csum_size;
1240 5358935 : total_bytes += ins_size * fs_info->sectorsize;
1241 :
1242 5358935 : btrfs_mark_buffer_dirty(path->nodes[0]);
1243 5358994 : if (total_bytes < sums->len) {
1244 46383 : btrfs_release_path(path);
1245 46383 : cond_resched();
1246 46383 : goto again;
1247 : }
1248 5312611 : out:
1249 5312611 : btrfs_free_path(path);
1250 5312611 : return ret;
1251 : }
1252 :
1253 1176246 : void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode,
1254 : const struct btrfs_path *path,
1255 : struct btrfs_file_extent_item *fi,
1256 : struct extent_map *em)
1257 : {
1258 1176246 : struct btrfs_fs_info *fs_info = inode->root->fs_info;
1259 1176246 : struct btrfs_root *root = inode->root;
1260 1176246 : struct extent_buffer *leaf = path->nodes[0];
1261 1176246 : const int slot = path->slots[0];
1262 1176246 : struct btrfs_key key;
1263 1176246 : u64 extent_start, extent_end;
1264 1176246 : u64 bytenr;
1265 1176246 : u8 type = btrfs_file_extent_type(leaf, fi);
1266 1176225 : int compress_type = btrfs_file_extent_compression(leaf, fi);
1267 :
1268 1176244 : btrfs_item_key_to_cpu(leaf, &key, slot);
1269 1176255 : extent_start = key.offset;
1270 1176255 : extent_end = btrfs_file_extent_end(path);
1271 1176282 : em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1272 1176279 : em->generation = btrfs_file_extent_generation(leaf, fi);
1273 1176282 : if (type == BTRFS_FILE_EXTENT_REG ||
1274 : type == BTRFS_FILE_EXTENT_PREALLOC) {
1275 940747 : em->start = extent_start;
1276 940747 : em->len = extent_end - extent_start;
1277 940747 : em->orig_start = extent_start -
1278 : btrfs_file_extent_offset(leaf, fi);
1279 940742 : em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
1280 940740 : bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1281 940733 : if (bytenr == 0) {
1282 5051 : em->block_start = EXTENT_MAP_HOLE;
1283 5051 : return;
1284 : }
1285 935682 : if (compress_type != BTRFS_COMPRESS_NONE) {
1286 1177 : set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1287 1177 : em->compress_type = compress_type;
1288 1177 : em->block_start = bytenr;
1289 1177 : em->block_len = em->orig_block_len;
1290 : } else {
1291 934505 : bytenr += btrfs_file_extent_offset(leaf, fi);
1292 934514 : em->block_start = bytenr;
1293 934514 : em->block_len = em->len;
1294 934514 : if (type == BTRFS_FILE_EXTENT_PREALLOC)
1295 170144 : set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
1296 : }
1297 235535 : } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1298 235535 : em->block_start = EXTENT_MAP_INLINE;
1299 235535 : em->start = extent_start;
1300 235535 : em->len = extent_end - extent_start;
1301 : /*
1302 : * Initialize orig_start and block_len with the same values
1303 : * as in inode.c:btrfs_get_extent().
1304 : */
1305 235535 : em->orig_start = EXTENT_MAP_HOLE;
1306 235535 : em->block_len = (u64)-1;
1307 235535 : em->compress_type = compress_type;
1308 235535 : if (compress_type != BTRFS_COMPRESS_NONE)
1309 19 : set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1310 : } else {
1311 0 : btrfs_err(fs_info,
1312 : "unknown file extent item type %d, inode %llu, offset %llu, "
1313 : "root %llu", type, btrfs_ino(inode), extent_start,
1314 : root->root_key.objectid);
1315 : }
1316 : }
1317 :
1318 : /*
1319 : * Returns the end offset (non inclusive) of the file extent item the given path
1320 : * points to. If it points to an inline extent, the returned offset is rounded
1321 : * up to the sector size.
1322 : */
1323 15828719 : u64 btrfs_file_extent_end(const struct btrfs_path *path)
1324 : {
1325 15828719 : const struct extent_buffer *leaf = path->nodes[0];
1326 15828719 : const int slot = path->slots[0];
1327 15828719 : struct btrfs_file_extent_item *fi;
1328 15828719 : struct btrfs_key key;
1329 15828719 : u64 end;
1330 :
1331 15828719 : btrfs_item_key_to_cpu(leaf, &key, slot);
1332 15828667 : ASSERT(key.type == BTRFS_EXTENT_DATA_KEY);
1333 15828667 : fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1334 :
1335 15828668 : if (btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_INLINE) {
1336 474398 : end = btrfs_file_extent_ram_bytes(leaf, fi);
1337 474398 : end = ALIGN(key.offset + end, leaf->fs_info->sectorsize);
1338 : } else {
1339 15354237 : end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1340 : }
1341 :
1342 15828565 : return end;
1343 : }
|
