Commit | Line | Data |
---|---|---|
a2de733c | 1 | /* |
b6bfebc1 | 2 | * Copyright (C) 2011, 2012 STRATO. All rights reserved. |
a2de733c AJ |
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 | ||
a2de733c | 19 | #include <linux/blkdev.h> |
558540c1 | 20 | #include <linux/ratelimit.h> |
a2de733c AJ |
21 | #include "ctree.h" |
22 | #include "volumes.h" | |
23 | #include "disk-io.h" | |
24 | #include "ordered-data.h" | |
0ef8e451 | 25 | #include "transaction.h" |
558540c1 | 26 | #include "backref.h" |
5da6fcbc | 27 | #include "extent_io.h" |
21adbd5c | 28 | #include "check-integrity.h" |
606686ee | 29 | #include "rcu-string.h" |
a2de733c AJ |
30 | |
31 | /* | |
32 | * This is only the first step towards a full-features scrub. It reads all | |
33 | * extent and super block and verifies the checksums. In case a bad checksum | |
34 | * is found or the extent cannot be read, good data will be written back if | |
35 | * any can be found. | |
36 | * | |
37 | * Future enhancements: | |
a2de733c AJ |
38 | * - In case an unrepairable extent is encountered, track which files are |
39 | * affected and report them | |
a2de733c | 40 | * - track and record media errors, throw out bad devices |
a2de733c | 41 | * - add a mode to also read unallocated space |
a2de733c AJ |
42 | */ |
43 | ||
b5d67f64 | 44 | struct scrub_block; |
d9d181c1 | 45 | struct scrub_ctx; |
a2de733c AJ |
46 | |
47 | #define SCRUB_PAGES_PER_BIO 16 /* 64k per bio */ | |
d9d181c1 | 48 | #define SCRUB_BIOS_PER_CTX 16 /* 1 MB per device in flight */ |
7a9e9987 SB |
49 | |
50 | /* | |
51 | * the following value times PAGE_SIZE needs to be large enough to match the | |
52 | * largest node/leaf/sector size that shall be supported. | |
53 | * Values larger than BTRFS_STRIPE_LEN are not supported. | |
54 | */ | |
b5d67f64 | 55 | #define SCRUB_MAX_PAGES_PER_BLOCK 16 /* 64k per node/leaf/sector */ |
a2de733c AJ |
56 | |
57 | struct scrub_page { | |
b5d67f64 SB |
58 | struct scrub_block *sblock; |
59 | struct page *page; | |
442a4f63 | 60 | struct btrfs_device *dev; |
a2de733c AJ |
61 | u64 flags; /* extent flags */ |
62 | u64 generation; | |
b5d67f64 SB |
63 | u64 logical; |
64 | u64 physical; | |
7a9e9987 | 65 | atomic_t ref_count; |
b5d67f64 SB |
66 | struct { |
67 | unsigned int mirror_num:8; | |
68 | unsigned int have_csum:1; | |
69 | unsigned int io_error:1; | |
70 | }; | |
a2de733c AJ |
71 | u8 csum[BTRFS_CSUM_SIZE]; |
72 | }; | |
73 | ||
74 | struct scrub_bio { | |
75 | int index; | |
d9d181c1 | 76 | struct scrub_ctx *sctx; |
a36cf8b8 | 77 | struct btrfs_device *dev; |
a2de733c AJ |
78 | struct bio *bio; |
79 | int err; | |
80 | u64 logical; | |
81 | u64 physical; | |
b5d67f64 SB |
82 | struct scrub_page *pagev[SCRUB_PAGES_PER_BIO]; |
83 | int page_count; | |
a2de733c AJ |
84 | int next_free; |
85 | struct btrfs_work work; | |
86 | }; | |
87 | ||
b5d67f64 | 88 | struct scrub_block { |
7a9e9987 | 89 | struct scrub_page *pagev[SCRUB_MAX_PAGES_PER_BLOCK]; |
b5d67f64 SB |
90 | int page_count; |
91 | atomic_t outstanding_pages; | |
92 | atomic_t ref_count; /* free mem on transition to zero */ | |
d9d181c1 | 93 | struct scrub_ctx *sctx; |
b5d67f64 SB |
94 | struct { |
95 | unsigned int header_error:1; | |
96 | unsigned int checksum_error:1; | |
97 | unsigned int no_io_error_seen:1; | |
442a4f63 | 98 | unsigned int generation_error:1; /* also sets header_error */ |
b5d67f64 SB |
99 | }; |
100 | }; | |
101 | ||
d9d181c1 SB |
102 | struct scrub_ctx { |
103 | struct scrub_bio *bios[SCRUB_BIOS_PER_CTX]; | |
a36cf8b8 | 104 | struct btrfs_root *dev_root; |
a2de733c AJ |
105 | int first_free; |
106 | int curr; | |
b6bfebc1 SB |
107 | atomic_t bios_in_flight; |
108 | atomic_t workers_pending; | |
a2de733c AJ |
109 | spinlock_t list_lock; |
110 | wait_queue_head_t list_wait; | |
111 | u16 csum_size; | |
112 | struct list_head csum_list; | |
113 | atomic_t cancel_req; | |
8628764e | 114 | int readonly; |
b5d67f64 SB |
115 | int pages_per_bio; /* <= SCRUB_PAGES_PER_BIO */ |
116 | u32 sectorsize; | |
117 | u32 nodesize; | |
118 | u32 leafsize; | |
a2de733c AJ |
119 | /* |
120 | * statistics | |
121 | */ | |
122 | struct btrfs_scrub_progress stat; | |
123 | spinlock_t stat_lock; | |
124 | }; | |
125 | ||
0ef8e451 | 126 | struct scrub_fixup_nodatasum { |
d9d181c1 | 127 | struct scrub_ctx *sctx; |
a36cf8b8 | 128 | struct btrfs_device *dev; |
0ef8e451 JS |
129 | u64 logical; |
130 | struct btrfs_root *root; | |
131 | struct btrfs_work work; | |
132 | int mirror_num; | |
133 | }; | |
134 | ||
558540c1 JS |
135 | struct scrub_warning { |
136 | struct btrfs_path *path; | |
137 | u64 extent_item_size; | |
138 | char *scratch_buf; | |
139 | char *msg_buf; | |
140 | const char *errstr; | |
141 | sector_t sector; | |
142 | u64 logical; | |
143 | struct btrfs_device *dev; | |
144 | int msg_bufsize; | |
145 | int scratch_bufsize; | |
146 | }; | |
147 | ||
b5d67f64 | 148 | |
b6bfebc1 SB |
149 | static void scrub_pending_bio_inc(struct scrub_ctx *sctx); |
150 | static void scrub_pending_bio_dec(struct scrub_ctx *sctx); | |
151 | static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx); | |
152 | static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx); | |
b5d67f64 | 153 | static int scrub_handle_errored_block(struct scrub_block *sblock_to_check); |
d9d181c1 | 154 | static int scrub_setup_recheck_block(struct scrub_ctx *sctx, |
3ec706c8 | 155 | struct btrfs_fs_info *fs_info, |
b5d67f64 SB |
156 | u64 length, u64 logical, |
157 | struct scrub_block *sblock); | |
34f5c8e9 SB |
158 | static void scrub_recheck_block(struct btrfs_fs_info *fs_info, |
159 | struct scrub_block *sblock, int is_metadata, | |
160 | int have_csum, u8 *csum, u64 generation, | |
161 | u16 csum_size); | |
b5d67f64 SB |
162 | static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info, |
163 | struct scrub_block *sblock, | |
164 | int is_metadata, int have_csum, | |
165 | const u8 *csum, u64 generation, | |
166 | u16 csum_size); | |
167 | static void scrub_complete_bio_end_io(struct bio *bio, int err); | |
168 | static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad, | |
169 | struct scrub_block *sblock_good, | |
170 | int force_write); | |
171 | static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad, | |
172 | struct scrub_block *sblock_good, | |
173 | int page_num, int force_write); | |
174 | static int scrub_checksum_data(struct scrub_block *sblock); | |
175 | static int scrub_checksum_tree_block(struct scrub_block *sblock); | |
176 | static int scrub_checksum_super(struct scrub_block *sblock); | |
177 | static void scrub_block_get(struct scrub_block *sblock); | |
178 | static void scrub_block_put(struct scrub_block *sblock); | |
7a9e9987 SB |
179 | static void scrub_page_get(struct scrub_page *spage); |
180 | static void scrub_page_put(struct scrub_page *spage); | |
d9d181c1 | 181 | static int scrub_add_page_to_bio(struct scrub_ctx *sctx, |
b5d67f64 | 182 | struct scrub_page *spage); |
d9d181c1 | 183 | static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len, |
a36cf8b8 SB |
184 | u64 physical, struct btrfs_device *dev, u64 flags, |
185 | u64 gen, int mirror_num, u8 *csum, int force); | |
1623edeb | 186 | static void scrub_bio_end_io(struct bio *bio, int err); |
b5d67f64 SB |
187 | static void scrub_bio_end_io_worker(struct btrfs_work *work); |
188 | static void scrub_block_complete(struct scrub_block *sblock); | |
1623edeb SB |
189 | |
190 | ||
b6bfebc1 SB |
191 | static void scrub_pending_bio_inc(struct scrub_ctx *sctx) |
192 | { | |
193 | atomic_inc(&sctx->bios_in_flight); | |
194 | } | |
195 | ||
196 | static void scrub_pending_bio_dec(struct scrub_ctx *sctx) | |
197 | { | |
198 | atomic_dec(&sctx->bios_in_flight); | |
199 | wake_up(&sctx->list_wait); | |
200 | } | |
201 | ||
202 | /* | |
203 | * used for workers that require transaction commits (i.e., for the | |
204 | * NOCOW case) | |
205 | */ | |
206 | static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx) | |
207 | { | |
208 | struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info; | |
209 | ||
210 | /* | |
211 | * increment scrubs_running to prevent cancel requests from | |
212 | * completing as long as a worker is running. we must also | |
213 | * increment scrubs_paused to prevent deadlocking on pause | |
214 | * requests used for transactions commits (as the worker uses a | |
215 | * transaction context). it is safe to regard the worker | |
216 | * as paused for all matters practical. effectively, we only | |
217 | * avoid cancellation requests from completing. | |
218 | */ | |
219 | mutex_lock(&fs_info->scrub_lock); | |
220 | atomic_inc(&fs_info->scrubs_running); | |
221 | atomic_inc(&fs_info->scrubs_paused); | |
222 | mutex_unlock(&fs_info->scrub_lock); | |
223 | atomic_inc(&sctx->workers_pending); | |
224 | } | |
225 | ||
226 | /* used for workers that require transaction commits */ | |
227 | static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx) | |
228 | { | |
229 | struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info; | |
230 | ||
231 | /* | |
232 | * see scrub_pending_trans_workers_inc() why we're pretending | |
233 | * to be paused in the scrub counters | |
234 | */ | |
235 | mutex_lock(&fs_info->scrub_lock); | |
236 | atomic_dec(&fs_info->scrubs_running); | |
237 | atomic_dec(&fs_info->scrubs_paused); | |
238 | mutex_unlock(&fs_info->scrub_lock); | |
239 | atomic_dec(&sctx->workers_pending); | |
240 | wake_up(&fs_info->scrub_pause_wait); | |
241 | wake_up(&sctx->list_wait); | |
242 | } | |
243 | ||
d9d181c1 | 244 | static void scrub_free_csums(struct scrub_ctx *sctx) |
a2de733c | 245 | { |
d9d181c1 | 246 | while (!list_empty(&sctx->csum_list)) { |
a2de733c | 247 | struct btrfs_ordered_sum *sum; |
d9d181c1 | 248 | sum = list_first_entry(&sctx->csum_list, |
a2de733c AJ |
249 | struct btrfs_ordered_sum, list); |
250 | list_del(&sum->list); | |
251 | kfree(sum); | |
252 | } | |
253 | } | |
254 | ||
d9d181c1 | 255 | static noinline_for_stack void scrub_free_ctx(struct scrub_ctx *sctx) |
a2de733c AJ |
256 | { |
257 | int i; | |
a2de733c | 258 | |
d9d181c1 | 259 | if (!sctx) |
a2de733c AJ |
260 | return; |
261 | ||
b5d67f64 | 262 | /* this can happen when scrub is cancelled */ |
d9d181c1 SB |
263 | if (sctx->curr != -1) { |
264 | struct scrub_bio *sbio = sctx->bios[sctx->curr]; | |
b5d67f64 SB |
265 | |
266 | for (i = 0; i < sbio->page_count; i++) { | |
267 | BUG_ON(!sbio->pagev[i]); | |
268 | BUG_ON(!sbio->pagev[i]->page); | |
269 | scrub_block_put(sbio->pagev[i]->sblock); | |
270 | } | |
271 | bio_put(sbio->bio); | |
272 | } | |
273 | ||
d9d181c1 SB |
274 | for (i = 0; i < SCRUB_BIOS_PER_CTX; ++i) { |
275 | struct scrub_bio *sbio = sctx->bios[i]; | |
a2de733c AJ |
276 | |
277 | if (!sbio) | |
278 | break; | |
a2de733c AJ |
279 | kfree(sbio); |
280 | } | |
281 | ||
d9d181c1 SB |
282 | scrub_free_csums(sctx); |
283 | kfree(sctx); | |
a2de733c AJ |
284 | } |
285 | ||
286 | static noinline_for_stack | |
d9d181c1 | 287 | struct scrub_ctx *scrub_setup_ctx(struct btrfs_device *dev) |
a2de733c | 288 | { |
d9d181c1 | 289 | struct scrub_ctx *sctx; |
a2de733c | 290 | int i; |
a2de733c | 291 | struct btrfs_fs_info *fs_info = dev->dev_root->fs_info; |
b5d67f64 | 292 | int pages_per_bio; |
a2de733c | 293 | |
b5d67f64 SB |
294 | pages_per_bio = min_t(int, SCRUB_PAGES_PER_BIO, |
295 | bio_get_nr_vecs(dev->bdev)); | |
d9d181c1 SB |
296 | sctx = kzalloc(sizeof(*sctx), GFP_NOFS); |
297 | if (!sctx) | |
a2de733c | 298 | goto nomem; |
d9d181c1 SB |
299 | sctx->pages_per_bio = pages_per_bio; |
300 | sctx->curr = -1; | |
a36cf8b8 | 301 | sctx->dev_root = dev->dev_root; |
d9d181c1 | 302 | for (i = 0; i < SCRUB_BIOS_PER_CTX; ++i) { |
a2de733c AJ |
303 | struct scrub_bio *sbio; |
304 | ||
305 | sbio = kzalloc(sizeof(*sbio), GFP_NOFS); | |
306 | if (!sbio) | |
307 | goto nomem; | |
d9d181c1 | 308 | sctx->bios[i] = sbio; |
a2de733c | 309 | |
a2de733c | 310 | sbio->index = i; |
d9d181c1 | 311 | sbio->sctx = sctx; |
b5d67f64 SB |
312 | sbio->page_count = 0; |
313 | sbio->work.func = scrub_bio_end_io_worker; | |
a2de733c | 314 | |
d9d181c1 SB |
315 | if (i != SCRUB_BIOS_PER_CTX - 1) |
316 | sctx->bios[i]->next_free = i + 1; | |
0ef8e451 | 317 | else |
d9d181c1 SB |
318 | sctx->bios[i]->next_free = -1; |
319 | } | |
320 | sctx->first_free = 0; | |
321 | sctx->nodesize = dev->dev_root->nodesize; | |
322 | sctx->leafsize = dev->dev_root->leafsize; | |
323 | sctx->sectorsize = dev->dev_root->sectorsize; | |
b6bfebc1 SB |
324 | atomic_set(&sctx->bios_in_flight, 0); |
325 | atomic_set(&sctx->workers_pending, 0); | |
d9d181c1 SB |
326 | atomic_set(&sctx->cancel_req, 0); |
327 | sctx->csum_size = btrfs_super_csum_size(fs_info->super_copy); | |
328 | INIT_LIST_HEAD(&sctx->csum_list); | |
329 | ||
330 | spin_lock_init(&sctx->list_lock); | |
331 | spin_lock_init(&sctx->stat_lock); | |
332 | init_waitqueue_head(&sctx->list_wait); | |
333 | return sctx; | |
a2de733c AJ |
334 | |
335 | nomem: | |
d9d181c1 | 336 | scrub_free_ctx(sctx); |
a2de733c AJ |
337 | return ERR_PTR(-ENOMEM); |
338 | } | |
339 | ||
558540c1 JS |
340 | static int scrub_print_warning_inode(u64 inum, u64 offset, u64 root, void *ctx) |
341 | { | |
342 | u64 isize; | |
343 | u32 nlink; | |
344 | int ret; | |
345 | int i; | |
346 | struct extent_buffer *eb; | |
347 | struct btrfs_inode_item *inode_item; | |
348 | struct scrub_warning *swarn = ctx; | |
349 | struct btrfs_fs_info *fs_info = swarn->dev->dev_root->fs_info; | |
350 | struct inode_fs_paths *ipath = NULL; | |
351 | struct btrfs_root *local_root; | |
352 | struct btrfs_key root_key; | |
353 | ||
354 | root_key.objectid = root; | |
355 | root_key.type = BTRFS_ROOT_ITEM_KEY; | |
356 | root_key.offset = (u64)-1; | |
357 | local_root = btrfs_read_fs_root_no_name(fs_info, &root_key); | |
358 | if (IS_ERR(local_root)) { | |
359 | ret = PTR_ERR(local_root); | |
360 | goto err; | |
361 | } | |
362 | ||
363 | ret = inode_item_info(inum, 0, local_root, swarn->path); | |
364 | if (ret) { | |
365 | btrfs_release_path(swarn->path); | |
366 | goto err; | |
367 | } | |
368 | ||
369 | eb = swarn->path->nodes[0]; | |
370 | inode_item = btrfs_item_ptr(eb, swarn->path->slots[0], | |
371 | struct btrfs_inode_item); | |
372 | isize = btrfs_inode_size(eb, inode_item); | |
373 | nlink = btrfs_inode_nlink(eb, inode_item); | |
374 | btrfs_release_path(swarn->path); | |
375 | ||
376 | ipath = init_ipath(4096, local_root, swarn->path); | |
26bdef54 DC |
377 | if (IS_ERR(ipath)) { |
378 | ret = PTR_ERR(ipath); | |
379 | ipath = NULL; | |
380 | goto err; | |
381 | } | |
558540c1 JS |
382 | ret = paths_from_inode(inum, ipath); |
383 | ||
384 | if (ret < 0) | |
385 | goto err; | |
386 | ||
387 | /* | |
388 | * we deliberately ignore the bit ipath might have been too small to | |
389 | * hold all of the paths here | |
390 | */ | |
391 | for (i = 0; i < ipath->fspath->elem_cnt; ++i) | |
606686ee | 392 | printk_in_rcu(KERN_WARNING "btrfs: %s at logical %llu on dev " |
558540c1 JS |
393 | "%s, sector %llu, root %llu, inode %llu, offset %llu, " |
394 | "length %llu, links %u (path: %s)\n", swarn->errstr, | |
606686ee | 395 | swarn->logical, rcu_str_deref(swarn->dev->name), |
558540c1 JS |
396 | (unsigned long long)swarn->sector, root, inum, offset, |
397 | min(isize - offset, (u64)PAGE_SIZE), nlink, | |
745c4d8e | 398 | (char *)(unsigned long)ipath->fspath->val[i]); |
558540c1 JS |
399 | |
400 | free_ipath(ipath); | |
401 | return 0; | |
402 | ||
403 | err: | |
606686ee | 404 | printk_in_rcu(KERN_WARNING "btrfs: %s at logical %llu on dev " |
558540c1 JS |
405 | "%s, sector %llu, root %llu, inode %llu, offset %llu: path " |
406 | "resolving failed with ret=%d\n", swarn->errstr, | |
606686ee | 407 | swarn->logical, rcu_str_deref(swarn->dev->name), |
558540c1 JS |
408 | (unsigned long long)swarn->sector, root, inum, offset, ret); |
409 | ||
410 | free_ipath(ipath); | |
411 | return 0; | |
412 | } | |
413 | ||
b5d67f64 | 414 | static void scrub_print_warning(const char *errstr, struct scrub_block *sblock) |
558540c1 | 415 | { |
a36cf8b8 SB |
416 | struct btrfs_device *dev; |
417 | struct btrfs_fs_info *fs_info; | |
558540c1 JS |
418 | struct btrfs_path *path; |
419 | struct btrfs_key found_key; | |
420 | struct extent_buffer *eb; | |
421 | struct btrfs_extent_item *ei; | |
422 | struct scrub_warning swarn; | |
69917e43 LB |
423 | unsigned long ptr = 0; |
424 | u64 extent_item_pos; | |
425 | u64 flags = 0; | |
558540c1 | 426 | u64 ref_root; |
69917e43 | 427 | u32 item_size; |
558540c1 | 428 | u8 ref_level; |
558540c1 | 429 | const int bufsize = 4096; |
69917e43 | 430 | int ret; |
558540c1 | 431 | |
a36cf8b8 | 432 | WARN_ON(sblock->page_count < 1); |
7a9e9987 | 433 | dev = sblock->pagev[0]->dev; |
a36cf8b8 SB |
434 | fs_info = sblock->sctx->dev_root->fs_info; |
435 | ||
558540c1 JS |
436 | path = btrfs_alloc_path(); |
437 | ||
438 | swarn.scratch_buf = kmalloc(bufsize, GFP_NOFS); | |
439 | swarn.msg_buf = kmalloc(bufsize, GFP_NOFS); | |
7a9e9987 SB |
440 | swarn.sector = (sblock->pagev[0]->physical) >> 9; |
441 | swarn.logical = sblock->pagev[0]->logical; | |
558540c1 | 442 | swarn.errstr = errstr; |
a36cf8b8 | 443 | swarn.dev = NULL; |
558540c1 JS |
444 | swarn.msg_bufsize = bufsize; |
445 | swarn.scratch_bufsize = bufsize; | |
446 | ||
447 | if (!path || !swarn.scratch_buf || !swarn.msg_buf) | |
448 | goto out; | |
449 | ||
69917e43 LB |
450 | ret = extent_from_logical(fs_info, swarn.logical, path, &found_key, |
451 | &flags); | |
558540c1 JS |
452 | if (ret < 0) |
453 | goto out; | |
454 | ||
4692cf58 | 455 | extent_item_pos = swarn.logical - found_key.objectid; |
558540c1 JS |
456 | swarn.extent_item_size = found_key.offset; |
457 | ||
458 | eb = path->nodes[0]; | |
459 | ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item); | |
460 | item_size = btrfs_item_size_nr(eb, path->slots[0]); | |
4692cf58 | 461 | btrfs_release_path(path); |
558540c1 | 462 | |
69917e43 | 463 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { |
558540c1 JS |
464 | do { |
465 | ret = tree_backref_for_extent(&ptr, eb, ei, item_size, | |
466 | &ref_root, &ref_level); | |
606686ee | 467 | printk_in_rcu(KERN_WARNING |
1623edeb | 468 | "btrfs: %s at logical %llu on dev %s, " |
558540c1 | 469 | "sector %llu: metadata %s (level %d) in tree " |
606686ee JB |
470 | "%llu\n", errstr, swarn.logical, |
471 | rcu_str_deref(dev->name), | |
558540c1 JS |
472 | (unsigned long long)swarn.sector, |
473 | ref_level ? "node" : "leaf", | |
474 | ret < 0 ? -1 : ref_level, | |
475 | ret < 0 ? -1 : ref_root); | |
476 | } while (ret != 1); | |
477 | } else { | |
478 | swarn.path = path; | |
a36cf8b8 | 479 | swarn.dev = dev; |
7a3ae2f8 JS |
480 | iterate_extent_inodes(fs_info, found_key.objectid, |
481 | extent_item_pos, 1, | |
558540c1 JS |
482 | scrub_print_warning_inode, &swarn); |
483 | } | |
484 | ||
485 | out: | |
486 | btrfs_free_path(path); | |
487 | kfree(swarn.scratch_buf); | |
488 | kfree(swarn.msg_buf); | |
489 | } | |
490 | ||
0ef8e451 JS |
491 | static int scrub_fixup_readpage(u64 inum, u64 offset, u64 root, void *ctx) |
492 | { | |
5da6fcbc | 493 | struct page *page = NULL; |
0ef8e451 JS |
494 | unsigned long index; |
495 | struct scrub_fixup_nodatasum *fixup = ctx; | |
496 | int ret; | |
5da6fcbc | 497 | int corrected = 0; |
0ef8e451 | 498 | struct btrfs_key key; |
5da6fcbc | 499 | struct inode *inode = NULL; |
0ef8e451 JS |
500 | u64 end = offset + PAGE_SIZE - 1; |
501 | struct btrfs_root *local_root; | |
502 | ||
503 | key.objectid = root; | |
504 | key.type = BTRFS_ROOT_ITEM_KEY; | |
505 | key.offset = (u64)-1; | |
506 | local_root = btrfs_read_fs_root_no_name(fixup->root->fs_info, &key); | |
507 | if (IS_ERR(local_root)) | |
508 | return PTR_ERR(local_root); | |
509 | ||
510 | key.type = BTRFS_INODE_ITEM_KEY; | |
511 | key.objectid = inum; | |
512 | key.offset = 0; | |
513 | inode = btrfs_iget(fixup->root->fs_info->sb, &key, local_root, NULL); | |
514 | if (IS_ERR(inode)) | |
515 | return PTR_ERR(inode); | |
516 | ||
0ef8e451 JS |
517 | index = offset >> PAGE_CACHE_SHIFT; |
518 | ||
519 | page = find_or_create_page(inode->i_mapping, index, GFP_NOFS); | |
5da6fcbc JS |
520 | if (!page) { |
521 | ret = -ENOMEM; | |
522 | goto out; | |
523 | } | |
524 | ||
525 | if (PageUptodate(page)) { | |
3ec706c8 | 526 | struct btrfs_fs_info *fs_info; |
5da6fcbc JS |
527 | if (PageDirty(page)) { |
528 | /* | |
529 | * we need to write the data to the defect sector. the | |
530 | * data that was in that sector is not in memory, | |
531 | * because the page was modified. we must not write the | |
532 | * modified page to that sector. | |
533 | * | |
534 | * TODO: what could be done here: wait for the delalloc | |
535 | * runner to write out that page (might involve | |
536 | * COW) and see whether the sector is still | |
537 | * referenced afterwards. | |
538 | * | |
539 | * For the meantime, we'll treat this error | |
540 | * incorrectable, although there is a chance that a | |
541 | * later scrub will find the bad sector again and that | |
542 | * there's no dirty page in memory, then. | |
543 | */ | |
544 | ret = -EIO; | |
545 | goto out; | |
546 | } | |
3ec706c8 SB |
547 | fs_info = BTRFS_I(inode)->root->fs_info; |
548 | ret = repair_io_failure(fs_info, offset, PAGE_SIZE, | |
5da6fcbc JS |
549 | fixup->logical, page, |
550 | fixup->mirror_num); | |
551 | unlock_page(page); | |
552 | corrected = !ret; | |
553 | } else { | |
554 | /* | |
555 | * we need to get good data first. the general readpage path | |
556 | * will call repair_io_failure for us, we just have to make | |
557 | * sure we read the bad mirror. | |
558 | */ | |
559 | ret = set_extent_bits(&BTRFS_I(inode)->io_tree, offset, end, | |
560 | EXTENT_DAMAGED, GFP_NOFS); | |
561 | if (ret) { | |
562 | /* set_extent_bits should give proper error */ | |
563 | WARN_ON(ret > 0); | |
564 | if (ret > 0) | |
565 | ret = -EFAULT; | |
566 | goto out; | |
567 | } | |
568 | ||
569 | ret = extent_read_full_page(&BTRFS_I(inode)->io_tree, page, | |
570 | btrfs_get_extent, | |
571 | fixup->mirror_num); | |
572 | wait_on_page_locked(page); | |
573 | ||
574 | corrected = !test_range_bit(&BTRFS_I(inode)->io_tree, offset, | |
575 | end, EXTENT_DAMAGED, 0, NULL); | |
576 | if (!corrected) | |
577 | clear_extent_bits(&BTRFS_I(inode)->io_tree, offset, end, | |
578 | EXTENT_DAMAGED, GFP_NOFS); | |
579 | } | |
580 | ||
581 | out: | |
582 | if (page) | |
583 | put_page(page); | |
584 | if (inode) | |
585 | iput(inode); | |
0ef8e451 JS |
586 | |
587 | if (ret < 0) | |
588 | return ret; | |
589 | ||
590 | if (ret == 0 && corrected) { | |
591 | /* | |
592 | * we only need to call readpage for one of the inodes belonging | |
593 | * to this extent. so make iterate_extent_inodes stop | |
594 | */ | |
595 | return 1; | |
596 | } | |
597 | ||
598 | return -EIO; | |
599 | } | |
600 | ||
601 | static void scrub_fixup_nodatasum(struct btrfs_work *work) | |
602 | { | |
603 | int ret; | |
604 | struct scrub_fixup_nodatasum *fixup; | |
d9d181c1 | 605 | struct scrub_ctx *sctx; |
0ef8e451 JS |
606 | struct btrfs_trans_handle *trans = NULL; |
607 | struct btrfs_fs_info *fs_info; | |
608 | struct btrfs_path *path; | |
609 | int uncorrectable = 0; | |
610 | ||
611 | fixup = container_of(work, struct scrub_fixup_nodatasum, work); | |
d9d181c1 | 612 | sctx = fixup->sctx; |
0ef8e451 JS |
613 | fs_info = fixup->root->fs_info; |
614 | ||
615 | path = btrfs_alloc_path(); | |
616 | if (!path) { | |
d9d181c1 SB |
617 | spin_lock(&sctx->stat_lock); |
618 | ++sctx->stat.malloc_errors; | |
619 | spin_unlock(&sctx->stat_lock); | |
0ef8e451 JS |
620 | uncorrectable = 1; |
621 | goto out; | |
622 | } | |
623 | ||
624 | trans = btrfs_join_transaction(fixup->root); | |
625 | if (IS_ERR(trans)) { | |
626 | uncorrectable = 1; | |
627 | goto out; | |
628 | } | |
629 | ||
630 | /* | |
631 | * the idea is to trigger a regular read through the standard path. we | |
632 | * read a page from the (failed) logical address by specifying the | |
633 | * corresponding copynum of the failed sector. thus, that readpage is | |
634 | * expected to fail. | |
635 | * that is the point where on-the-fly error correction will kick in | |
636 | * (once it's finished) and rewrite the failed sector if a good copy | |
637 | * can be found. | |
638 | */ | |
639 | ret = iterate_inodes_from_logical(fixup->logical, fixup->root->fs_info, | |
640 | path, scrub_fixup_readpage, | |
641 | fixup); | |
642 | if (ret < 0) { | |
643 | uncorrectable = 1; | |
644 | goto out; | |
645 | } | |
646 | WARN_ON(ret != 1); | |
647 | ||
d9d181c1 SB |
648 | spin_lock(&sctx->stat_lock); |
649 | ++sctx->stat.corrected_errors; | |
650 | spin_unlock(&sctx->stat_lock); | |
0ef8e451 JS |
651 | |
652 | out: | |
653 | if (trans && !IS_ERR(trans)) | |
654 | btrfs_end_transaction(trans, fixup->root); | |
655 | if (uncorrectable) { | |
d9d181c1 SB |
656 | spin_lock(&sctx->stat_lock); |
657 | ++sctx->stat.uncorrectable_errors; | |
658 | spin_unlock(&sctx->stat_lock); | |
606686ee JB |
659 | |
660 | printk_ratelimited_in_rcu(KERN_ERR | |
b5d67f64 | 661 | "btrfs: unable to fixup (nodatasum) error at logical %llu on dev %s\n", |
606686ee | 662 | (unsigned long long)fixup->logical, |
a36cf8b8 | 663 | rcu_str_deref(fixup->dev->name)); |
0ef8e451 JS |
664 | } |
665 | ||
666 | btrfs_free_path(path); | |
667 | kfree(fixup); | |
668 | ||
b6bfebc1 | 669 | scrub_pending_trans_workers_dec(sctx); |
0ef8e451 JS |
670 | } |
671 | ||
a2de733c | 672 | /* |
b5d67f64 SB |
673 | * scrub_handle_errored_block gets called when either verification of the |
674 | * pages failed or the bio failed to read, e.g. with EIO. In the latter | |
675 | * case, this function handles all pages in the bio, even though only one | |
676 | * may be bad. | |
677 | * The goal of this function is to repair the errored block by using the | |
678 | * contents of one of the mirrors. | |
a2de733c | 679 | */ |
b5d67f64 | 680 | static int scrub_handle_errored_block(struct scrub_block *sblock_to_check) |
a2de733c | 681 | { |
d9d181c1 | 682 | struct scrub_ctx *sctx = sblock_to_check->sctx; |
a36cf8b8 | 683 | struct btrfs_device *dev; |
b5d67f64 SB |
684 | struct btrfs_fs_info *fs_info; |
685 | u64 length; | |
686 | u64 logical; | |
687 | u64 generation; | |
688 | unsigned int failed_mirror_index; | |
689 | unsigned int is_metadata; | |
690 | unsigned int have_csum; | |
691 | u8 *csum; | |
692 | struct scrub_block *sblocks_for_recheck; /* holds one for each mirror */ | |
693 | struct scrub_block *sblock_bad; | |
694 | int ret; | |
695 | int mirror_index; | |
696 | int page_num; | |
697 | int success; | |
558540c1 | 698 | static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL, |
b5d67f64 SB |
699 | DEFAULT_RATELIMIT_BURST); |
700 | ||
701 | BUG_ON(sblock_to_check->page_count < 1); | |
a36cf8b8 | 702 | fs_info = sctx->dev_root->fs_info; |
b5d67f64 | 703 | length = sblock_to_check->page_count * PAGE_SIZE; |
7a9e9987 SB |
704 | logical = sblock_to_check->pagev[0]->logical; |
705 | generation = sblock_to_check->pagev[0]->generation; | |
706 | BUG_ON(sblock_to_check->pagev[0]->mirror_num < 1); | |
707 | failed_mirror_index = sblock_to_check->pagev[0]->mirror_num - 1; | |
708 | is_metadata = !(sblock_to_check->pagev[0]->flags & | |
b5d67f64 | 709 | BTRFS_EXTENT_FLAG_DATA); |
7a9e9987 SB |
710 | have_csum = sblock_to_check->pagev[0]->have_csum; |
711 | csum = sblock_to_check->pagev[0]->csum; | |
712 | dev = sblock_to_check->pagev[0]->dev; | |
13db62b7 | 713 | |
b5d67f64 SB |
714 | /* |
715 | * read all mirrors one after the other. This includes to | |
716 | * re-read the extent or metadata block that failed (that was | |
717 | * the cause that this fixup code is called) another time, | |
718 | * page by page this time in order to know which pages | |
719 | * caused I/O errors and which ones are good (for all mirrors). | |
720 | * It is the goal to handle the situation when more than one | |
721 | * mirror contains I/O errors, but the errors do not | |
722 | * overlap, i.e. the data can be repaired by selecting the | |
723 | * pages from those mirrors without I/O error on the | |
724 | * particular pages. One example (with blocks >= 2 * PAGE_SIZE) | |
725 | * would be that mirror #1 has an I/O error on the first page, | |
726 | * the second page is good, and mirror #2 has an I/O error on | |
727 | * the second page, but the first page is good. | |
728 | * Then the first page of the first mirror can be repaired by | |
729 | * taking the first page of the second mirror, and the | |
730 | * second page of the second mirror can be repaired by | |
731 | * copying the contents of the 2nd page of the 1st mirror. | |
732 | * One more note: if the pages of one mirror contain I/O | |
733 | * errors, the checksum cannot be verified. In order to get | |
734 | * the best data for repairing, the first attempt is to find | |
735 | * a mirror without I/O errors and with a validated checksum. | |
736 | * Only if this is not possible, the pages are picked from | |
737 | * mirrors with I/O errors without considering the checksum. | |
738 | * If the latter is the case, at the end, the checksum of the | |
739 | * repaired area is verified in order to correctly maintain | |
740 | * the statistics. | |
741 | */ | |
742 | ||
743 | sblocks_for_recheck = kzalloc(BTRFS_MAX_MIRRORS * | |
744 | sizeof(*sblocks_for_recheck), | |
745 | GFP_NOFS); | |
746 | if (!sblocks_for_recheck) { | |
d9d181c1 SB |
747 | spin_lock(&sctx->stat_lock); |
748 | sctx->stat.malloc_errors++; | |
749 | sctx->stat.read_errors++; | |
750 | sctx->stat.uncorrectable_errors++; | |
751 | spin_unlock(&sctx->stat_lock); | |
a36cf8b8 | 752 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS); |
b5d67f64 | 753 | goto out; |
a2de733c AJ |
754 | } |
755 | ||
b5d67f64 | 756 | /* setup the context, map the logical blocks and alloc the pages */ |
3ec706c8 | 757 | ret = scrub_setup_recheck_block(sctx, fs_info, length, |
b5d67f64 SB |
758 | logical, sblocks_for_recheck); |
759 | if (ret) { | |
d9d181c1 SB |
760 | spin_lock(&sctx->stat_lock); |
761 | sctx->stat.read_errors++; | |
762 | sctx->stat.uncorrectable_errors++; | |
763 | spin_unlock(&sctx->stat_lock); | |
a36cf8b8 | 764 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS); |
b5d67f64 SB |
765 | goto out; |
766 | } | |
767 | BUG_ON(failed_mirror_index >= BTRFS_MAX_MIRRORS); | |
768 | sblock_bad = sblocks_for_recheck + failed_mirror_index; | |
13db62b7 | 769 | |
b5d67f64 | 770 | /* build and submit the bios for the failed mirror, check checksums */ |
34f5c8e9 SB |
771 | scrub_recheck_block(fs_info, sblock_bad, is_metadata, have_csum, |
772 | csum, generation, sctx->csum_size); | |
a2de733c | 773 | |
b5d67f64 SB |
774 | if (!sblock_bad->header_error && !sblock_bad->checksum_error && |
775 | sblock_bad->no_io_error_seen) { | |
776 | /* | |
777 | * the error disappeared after reading page by page, or | |
778 | * the area was part of a huge bio and other parts of the | |
779 | * bio caused I/O errors, or the block layer merged several | |
780 | * read requests into one and the error is caused by a | |
781 | * different bio (usually one of the two latter cases is | |
782 | * the cause) | |
783 | */ | |
d9d181c1 SB |
784 | spin_lock(&sctx->stat_lock); |
785 | sctx->stat.unverified_errors++; | |
786 | spin_unlock(&sctx->stat_lock); | |
a2de733c | 787 | |
b5d67f64 | 788 | goto out; |
a2de733c | 789 | } |
a2de733c | 790 | |
b5d67f64 | 791 | if (!sblock_bad->no_io_error_seen) { |
d9d181c1 SB |
792 | spin_lock(&sctx->stat_lock); |
793 | sctx->stat.read_errors++; | |
794 | spin_unlock(&sctx->stat_lock); | |
b5d67f64 SB |
795 | if (__ratelimit(&_rs)) |
796 | scrub_print_warning("i/o error", sblock_to_check); | |
a36cf8b8 | 797 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS); |
b5d67f64 | 798 | } else if (sblock_bad->checksum_error) { |
d9d181c1 SB |
799 | spin_lock(&sctx->stat_lock); |
800 | sctx->stat.csum_errors++; | |
801 | spin_unlock(&sctx->stat_lock); | |
b5d67f64 SB |
802 | if (__ratelimit(&_rs)) |
803 | scrub_print_warning("checksum error", sblock_to_check); | |
a36cf8b8 | 804 | btrfs_dev_stat_inc_and_print(dev, |
442a4f63 | 805 | BTRFS_DEV_STAT_CORRUPTION_ERRS); |
b5d67f64 | 806 | } else if (sblock_bad->header_error) { |
d9d181c1 SB |
807 | spin_lock(&sctx->stat_lock); |
808 | sctx->stat.verify_errors++; | |
809 | spin_unlock(&sctx->stat_lock); | |
b5d67f64 SB |
810 | if (__ratelimit(&_rs)) |
811 | scrub_print_warning("checksum/header error", | |
812 | sblock_to_check); | |
442a4f63 | 813 | if (sblock_bad->generation_error) |
a36cf8b8 | 814 | btrfs_dev_stat_inc_and_print(dev, |
442a4f63 SB |
815 | BTRFS_DEV_STAT_GENERATION_ERRS); |
816 | else | |
a36cf8b8 | 817 | btrfs_dev_stat_inc_and_print(dev, |
442a4f63 | 818 | BTRFS_DEV_STAT_CORRUPTION_ERRS); |
b5d67f64 | 819 | } |
a2de733c | 820 | |
d9d181c1 | 821 | if (sctx->readonly) |
b5d67f64 | 822 | goto did_not_correct_error; |
a2de733c | 823 | |
b5d67f64 SB |
824 | if (!is_metadata && !have_csum) { |
825 | struct scrub_fixup_nodatasum *fixup_nodatasum; | |
a2de733c | 826 | |
b5d67f64 SB |
827 | /* |
828 | * !is_metadata and !have_csum, this means that the data | |
829 | * might not be COW'ed, that it might be modified | |
830 | * concurrently. The general strategy to work on the | |
831 | * commit root does not help in the case when COW is not | |
832 | * used. | |
833 | */ | |
834 | fixup_nodatasum = kzalloc(sizeof(*fixup_nodatasum), GFP_NOFS); | |
835 | if (!fixup_nodatasum) | |
836 | goto did_not_correct_error; | |
d9d181c1 | 837 | fixup_nodatasum->sctx = sctx; |
a36cf8b8 | 838 | fixup_nodatasum->dev = dev; |
b5d67f64 SB |
839 | fixup_nodatasum->logical = logical; |
840 | fixup_nodatasum->root = fs_info->extent_root; | |
841 | fixup_nodatasum->mirror_num = failed_mirror_index + 1; | |
b6bfebc1 | 842 | scrub_pending_trans_workers_inc(sctx); |
b5d67f64 SB |
843 | fixup_nodatasum->work.func = scrub_fixup_nodatasum; |
844 | btrfs_queue_worker(&fs_info->scrub_workers, | |
845 | &fixup_nodatasum->work); | |
846 | goto out; | |
a2de733c AJ |
847 | } |
848 | ||
b5d67f64 SB |
849 | /* |
850 | * now build and submit the bios for the other mirrors, check | |
cb2ced73 SB |
851 | * checksums. |
852 | * First try to pick the mirror which is completely without I/O | |
b5d67f64 SB |
853 | * errors and also does not have a checksum error. |
854 | * If one is found, and if a checksum is present, the full block | |
855 | * that is known to contain an error is rewritten. Afterwards | |
856 | * the block is known to be corrected. | |
857 | * If a mirror is found which is completely correct, and no | |
858 | * checksum is present, only those pages are rewritten that had | |
859 | * an I/O error in the block to be repaired, since it cannot be | |
860 | * determined, which copy of the other pages is better (and it | |
861 | * could happen otherwise that a correct page would be | |
862 | * overwritten by a bad one). | |
863 | */ | |
864 | for (mirror_index = 0; | |
865 | mirror_index < BTRFS_MAX_MIRRORS && | |
866 | sblocks_for_recheck[mirror_index].page_count > 0; | |
867 | mirror_index++) { | |
cb2ced73 | 868 | struct scrub_block *sblock_other; |
b5d67f64 | 869 | |
cb2ced73 SB |
870 | if (mirror_index == failed_mirror_index) |
871 | continue; | |
872 | sblock_other = sblocks_for_recheck + mirror_index; | |
873 | ||
874 | /* build and submit the bios, check checksums */ | |
34f5c8e9 SB |
875 | scrub_recheck_block(fs_info, sblock_other, is_metadata, |
876 | have_csum, csum, generation, | |
877 | sctx->csum_size); | |
878 | ||
879 | if (!sblock_other->header_error && | |
b5d67f64 SB |
880 | !sblock_other->checksum_error && |
881 | sblock_other->no_io_error_seen) { | |
882 | int force_write = is_metadata || have_csum; | |
883 | ||
884 | ret = scrub_repair_block_from_good_copy(sblock_bad, | |
885 | sblock_other, | |
886 | force_write); | |
887 | if (0 == ret) | |
888 | goto corrected_error; | |
889 | } | |
890 | } | |
a2de733c AJ |
891 | |
892 | /* | |
b5d67f64 SB |
893 | * in case of I/O errors in the area that is supposed to be |
894 | * repaired, continue by picking good copies of those pages. | |
895 | * Select the good pages from mirrors to rewrite bad pages from | |
896 | * the area to fix. Afterwards verify the checksum of the block | |
897 | * that is supposed to be repaired. This verification step is | |
898 | * only done for the purpose of statistic counting and for the | |
899 | * final scrub report, whether errors remain. | |
900 | * A perfect algorithm could make use of the checksum and try | |
901 | * all possible combinations of pages from the different mirrors | |
902 | * until the checksum verification succeeds. For example, when | |
903 | * the 2nd page of mirror #1 faces I/O errors, and the 2nd page | |
904 | * of mirror #2 is readable but the final checksum test fails, | |
905 | * then the 2nd page of mirror #3 could be tried, whether now | |
906 | * the final checksum succeedes. But this would be a rare | |
907 | * exception and is therefore not implemented. At least it is | |
908 | * avoided that the good copy is overwritten. | |
909 | * A more useful improvement would be to pick the sectors | |
910 | * without I/O error based on sector sizes (512 bytes on legacy | |
911 | * disks) instead of on PAGE_SIZE. Then maybe 512 byte of one | |
912 | * mirror could be repaired by taking 512 byte of a different | |
913 | * mirror, even if other 512 byte sectors in the same PAGE_SIZE | |
914 | * area are unreadable. | |
a2de733c | 915 | */ |
a2de733c | 916 | |
b5d67f64 SB |
917 | /* can only fix I/O errors from here on */ |
918 | if (sblock_bad->no_io_error_seen) | |
919 | goto did_not_correct_error; | |
920 | ||
921 | success = 1; | |
922 | for (page_num = 0; page_num < sblock_bad->page_count; page_num++) { | |
7a9e9987 | 923 | struct scrub_page *page_bad = sblock_bad->pagev[page_num]; |
b5d67f64 SB |
924 | |
925 | if (!page_bad->io_error) | |
a2de733c | 926 | continue; |
b5d67f64 SB |
927 | |
928 | for (mirror_index = 0; | |
929 | mirror_index < BTRFS_MAX_MIRRORS && | |
930 | sblocks_for_recheck[mirror_index].page_count > 0; | |
931 | mirror_index++) { | |
932 | struct scrub_block *sblock_other = sblocks_for_recheck + | |
933 | mirror_index; | |
7a9e9987 SB |
934 | struct scrub_page *page_other = sblock_other->pagev[ |
935 | page_num]; | |
b5d67f64 SB |
936 | |
937 | if (!page_other->io_error) { | |
938 | ret = scrub_repair_page_from_good_copy( | |
939 | sblock_bad, sblock_other, page_num, 0); | |
940 | if (0 == ret) { | |
941 | page_bad->io_error = 0; | |
942 | break; /* succeeded for this page */ | |
943 | } | |
944 | } | |
96e36920 | 945 | } |
a2de733c | 946 | |
b5d67f64 SB |
947 | if (page_bad->io_error) { |
948 | /* did not find a mirror to copy the page from */ | |
949 | success = 0; | |
950 | } | |
a2de733c | 951 | } |
a2de733c | 952 | |
b5d67f64 SB |
953 | if (success) { |
954 | if (is_metadata || have_csum) { | |
955 | /* | |
956 | * need to verify the checksum now that all | |
957 | * sectors on disk are repaired (the write | |
958 | * request for data to be repaired is on its way). | |
959 | * Just be lazy and use scrub_recheck_block() | |
960 | * which re-reads the data before the checksum | |
961 | * is verified, but most likely the data comes out | |
962 | * of the page cache. | |
963 | */ | |
34f5c8e9 SB |
964 | scrub_recheck_block(fs_info, sblock_bad, |
965 | is_metadata, have_csum, csum, | |
966 | generation, sctx->csum_size); | |
967 | if (!sblock_bad->header_error && | |
b5d67f64 SB |
968 | !sblock_bad->checksum_error && |
969 | sblock_bad->no_io_error_seen) | |
970 | goto corrected_error; | |
971 | else | |
972 | goto did_not_correct_error; | |
973 | } else { | |
974 | corrected_error: | |
d9d181c1 SB |
975 | spin_lock(&sctx->stat_lock); |
976 | sctx->stat.corrected_errors++; | |
977 | spin_unlock(&sctx->stat_lock); | |
606686ee | 978 | printk_ratelimited_in_rcu(KERN_ERR |
b5d67f64 | 979 | "btrfs: fixed up error at logical %llu on dev %s\n", |
606686ee | 980 | (unsigned long long)logical, |
a36cf8b8 | 981 | rcu_str_deref(dev->name)); |
8628764e | 982 | } |
b5d67f64 SB |
983 | } else { |
984 | did_not_correct_error: | |
d9d181c1 SB |
985 | spin_lock(&sctx->stat_lock); |
986 | sctx->stat.uncorrectable_errors++; | |
987 | spin_unlock(&sctx->stat_lock); | |
606686ee | 988 | printk_ratelimited_in_rcu(KERN_ERR |
b5d67f64 | 989 | "btrfs: unable to fixup (regular) error at logical %llu on dev %s\n", |
606686ee | 990 | (unsigned long long)logical, |
a36cf8b8 | 991 | rcu_str_deref(dev->name)); |
96e36920 | 992 | } |
a2de733c | 993 | |
b5d67f64 SB |
994 | out: |
995 | if (sblocks_for_recheck) { | |
996 | for (mirror_index = 0; mirror_index < BTRFS_MAX_MIRRORS; | |
997 | mirror_index++) { | |
998 | struct scrub_block *sblock = sblocks_for_recheck + | |
999 | mirror_index; | |
1000 | int page_index; | |
1001 | ||
7a9e9987 SB |
1002 | for (page_index = 0; page_index < sblock->page_count; |
1003 | page_index++) { | |
1004 | sblock->pagev[page_index]->sblock = NULL; | |
1005 | scrub_page_put(sblock->pagev[page_index]); | |
1006 | } | |
b5d67f64 SB |
1007 | } |
1008 | kfree(sblocks_for_recheck); | |
1009 | } | |
a2de733c | 1010 | |
b5d67f64 SB |
1011 | return 0; |
1012 | } | |
a2de733c | 1013 | |
d9d181c1 | 1014 | static int scrub_setup_recheck_block(struct scrub_ctx *sctx, |
3ec706c8 | 1015 | struct btrfs_fs_info *fs_info, |
b5d67f64 SB |
1016 | u64 length, u64 logical, |
1017 | struct scrub_block *sblocks_for_recheck) | |
1018 | { | |
1019 | int page_index; | |
1020 | int mirror_index; | |
1021 | int ret; | |
1022 | ||
1023 | /* | |
7a9e9987 | 1024 | * note: the two members ref_count and outstanding_pages |
b5d67f64 SB |
1025 | * are not used (and not set) in the blocks that are used for |
1026 | * the recheck procedure | |
1027 | */ | |
1028 | ||
1029 | page_index = 0; | |
1030 | while (length > 0) { | |
1031 | u64 sublen = min_t(u64, length, PAGE_SIZE); | |
1032 | u64 mapped_length = sublen; | |
1033 | struct btrfs_bio *bbio = NULL; | |
a2de733c | 1034 | |
b5d67f64 SB |
1035 | /* |
1036 | * with a length of PAGE_SIZE, each returned stripe | |
1037 | * represents one mirror | |
1038 | */ | |
3ec706c8 | 1039 | ret = btrfs_map_block(fs_info, WRITE, logical, &mapped_length, |
b5d67f64 SB |
1040 | &bbio, 0); |
1041 | if (ret || !bbio || mapped_length < sublen) { | |
1042 | kfree(bbio); | |
1043 | return -EIO; | |
1044 | } | |
a2de733c | 1045 | |
b5d67f64 SB |
1046 | BUG_ON(page_index >= SCRUB_PAGES_PER_BIO); |
1047 | for (mirror_index = 0; mirror_index < (int)bbio->num_stripes; | |
1048 | mirror_index++) { | |
1049 | struct scrub_block *sblock; | |
1050 | struct scrub_page *page; | |
1051 | ||
1052 | if (mirror_index >= BTRFS_MAX_MIRRORS) | |
1053 | continue; | |
1054 | ||
1055 | sblock = sblocks_for_recheck + mirror_index; | |
7a9e9987 SB |
1056 | sblock->sctx = sctx; |
1057 | page = kzalloc(sizeof(*page), GFP_NOFS); | |
1058 | if (!page) { | |
1059 | leave_nomem: | |
d9d181c1 SB |
1060 | spin_lock(&sctx->stat_lock); |
1061 | sctx->stat.malloc_errors++; | |
1062 | spin_unlock(&sctx->stat_lock); | |
cf93dcce | 1063 | kfree(bbio); |
b5d67f64 SB |
1064 | return -ENOMEM; |
1065 | } | |
7a9e9987 SB |
1066 | scrub_page_get(page); |
1067 | sblock->pagev[page_index] = page; | |
1068 | page->logical = logical; | |
1069 | page->physical = bbio->stripes[mirror_index].physical; | |
1070 | /* for missing devices, dev->bdev is NULL */ | |
1071 | page->dev = bbio->stripes[mirror_index].dev; | |
1072 | page->mirror_num = mirror_index + 1; | |
b5d67f64 | 1073 | sblock->page_count++; |
7a9e9987 SB |
1074 | page->page = alloc_page(GFP_NOFS); |
1075 | if (!page->page) | |
1076 | goto leave_nomem; | |
b5d67f64 SB |
1077 | } |
1078 | kfree(bbio); | |
1079 | length -= sublen; | |
1080 | logical += sublen; | |
1081 | page_index++; | |
1082 | } | |
1083 | ||
1084 | return 0; | |
96e36920 ID |
1085 | } |
1086 | ||
b5d67f64 SB |
1087 | /* |
1088 | * this function will check the on disk data for checksum errors, header | |
1089 | * errors and read I/O errors. If any I/O errors happen, the exact pages | |
1090 | * which are errored are marked as being bad. The goal is to enable scrub | |
1091 | * to take those pages that are not errored from all the mirrors so that | |
1092 | * the pages that are errored in the just handled mirror can be repaired. | |
1093 | */ | |
34f5c8e9 SB |
1094 | static void scrub_recheck_block(struct btrfs_fs_info *fs_info, |
1095 | struct scrub_block *sblock, int is_metadata, | |
1096 | int have_csum, u8 *csum, u64 generation, | |
1097 | u16 csum_size) | |
96e36920 | 1098 | { |
b5d67f64 | 1099 | int page_num; |
96e36920 | 1100 | |
b5d67f64 SB |
1101 | sblock->no_io_error_seen = 1; |
1102 | sblock->header_error = 0; | |
1103 | sblock->checksum_error = 0; | |
96e36920 | 1104 | |
b5d67f64 SB |
1105 | for (page_num = 0; page_num < sblock->page_count; page_num++) { |
1106 | struct bio *bio; | |
7a9e9987 | 1107 | struct scrub_page *page = sblock->pagev[page_num]; |
b5d67f64 SB |
1108 | DECLARE_COMPLETION_ONSTACK(complete); |
1109 | ||
442a4f63 | 1110 | if (page->dev->bdev == NULL) { |
ea9947b4 SB |
1111 | page->io_error = 1; |
1112 | sblock->no_io_error_seen = 0; | |
1113 | continue; | |
1114 | } | |
1115 | ||
7a9e9987 | 1116 | WARN_ON(!page->page); |
b5d67f64 | 1117 | bio = bio_alloc(GFP_NOFS, 1); |
34f5c8e9 SB |
1118 | if (!bio) { |
1119 | page->io_error = 1; | |
1120 | sblock->no_io_error_seen = 0; | |
1121 | continue; | |
1122 | } | |
442a4f63 | 1123 | bio->bi_bdev = page->dev->bdev; |
b5d67f64 SB |
1124 | bio->bi_sector = page->physical >> 9; |
1125 | bio->bi_end_io = scrub_complete_bio_end_io; | |
1126 | bio->bi_private = &complete; | |
1127 | ||
34f5c8e9 | 1128 | bio_add_page(bio, page->page, PAGE_SIZE, 0); |
b5d67f64 | 1129 | btrfsic_submit_bio(READ, bio); |
96e36920 | 1130 | |
b5d67f64 SB |
1131 | /* this will also unplug the queue */ |
1132 | wait_for_completion(&complete); | |
96e36920 | 1133 | |
b5d67f64 SB |
1134 | page->io_error = !test_bit(BIO_UPTODATE, &bio->bi_flags); |
1135 | if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
1136 | sblock->no_io_error_seen = 0; | |
1137 | bio_put(bio); | |
1138 | } | |
96e36920 | 1139 | |
b5d67f64 SB |
1140 | if (sblock->no_io_error_seen) |
1141 | scrub_recheck_block_checksum(fs_info, sblock, is_metadata, | |
1142 | have_csum, csum, generation, | |
1143 | csum_size); | |
1144 | ||
34f5c8e9 | 1145 | return; |
a2de733c AJ |
1146 | } |
1147 | ||
b5d67f64 SB |
1148 | static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info, |
1149 | struct scrub_block *sblock, | |
1150 | int is_metadata, int have_csum, | |
1151 | const u8 *csum, u64 generation, | |
1152 | u16 csum_size) | |
a2de733c | 1153 | { |
b5d67f64 SB |
1154 | int page_num; |
1155 | u8 calculated_csum[BTRFS_CSUM_SIZE]; | |
1156 | u32 crc = ~(u32)0; | |
1157 | struct btrfs_root *root = fs_info->extent_root; | |
1158 | void *mapped_buffer; | |
1159 | ||
7a9e9987 | 1160 | WARN_ON(!sblock->pagev[0]->page); |
b5d67f64 SB |
1161 | if (is_metadata) { |
1162 | struct btrfs_header *h; | |
1163 | ||
7a9e9987 | 1164 | mapped_buffer = kmap_atomic(sblock->pagev[0]->page); |
b5d67f64 SB |
1165 | h = (struct btrfs_header *)mapped_buffer; |
1166 | ||
7a9e9987 | 1167 | if (sblock->pagev[0]->logical != le64_to_cpu(h->bytenr) || |
b5d67f64 SB |
1168 | memcmp(h->fsid, fs_info->fsid, BTRFS_UUID_SIZE) || |
1169 | memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid, | |
442a4f63 | 1170 | BTRFS_UUID_SIZE)) { |
b5d67f64 | 1171 | sblock->header_error = 1; |
442a4f63 SB |
1172 | } else if (generation != le64_to_cpu(h->generation)) { |
1173 | sblock->header_error = 1; | |
1174 | sblock->generation_error = 1; | |
1175 | } | |
b5d67f64 SB |
1176 | csum = h->csum; |
1177 | } else { | |
1178 | if (!have_csum) | |
1179 | return; | |
a2de733c | 1180 | |
7a9e9987 | 1181 | mapped_buffer = kmap_atomic(sblock->pagev[0]->page); |
b5d67f64 | 1182 | } |
a2de733c | 1183 | |
b5d67f64 SB |
1184 | for (page_num = 0;;) { |
1185 | if (page_num == 0 && is_metadata) | |
1186 | crc = btrfs_csum_data(root, | |
1187 | ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE, | |
1188 | crc, PAGE_SIZE - BTRFS_CSUM_SIZE); | |
1189 | else | |
1190 | crc = btrfs_csum_data(root, mapped_buffer, crc, | |
1191 | PAGE_SIZE); | |
1192 | ||
9613bebb | 1193 | kunmap_atomic(mapped_buffer); |
b5d67f64 SB |
1194 | page_num++; |
1195 | if (page_num >= sblock->page_count) | |
1196 | break; | |
7a9e9987 | 1197 | WARN_ON(!sblock->pagev[page_num]->page); |
b5d67f64 | 1198 | |
7a9e9987 | 1199 | mapped_buffer = kmap_atomic(sblock->pagev[page_num]->page); |
b5d67f64 SB |
1200 | } |
1201 | ||
1202 | btrfs_csum_final(crc, calculated_csum); | |
1203 | if (memcmp(calculated_csum, csum, csum_size)) | |
1204 | sblock->checksum_error = 1; | |
a2de733c AJ |
1205 | } |
1206 | ||
b5d67f64 | 1207 | static void scrub_complete_bio_end_io(struct bio *bio, int err) |
a2de733c | 1208 | { |
b5d67f64 SB |
1209 | complete((struct completion *)bio->bi_private); |
1210 | } | |
a2de733c | 1211 | |
b5d67f64 SB |
1212 | static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad, |
1213 | struct scrub_block *sblock_good, | |
1214 | int force_write) | |
1215 | { | |
1216 | int page_num; | |
1217 | int ret = 0; | |
96e36920 | 1218 | |
b5d67f64 SB |
1219 | for (page_num = 0; page_num < sblock_bad->page_count; page_num++) { |
1220 | int ret_sub; | |
96e36920 | 1221 | |
b5d67f64 SB |
1222 | ret_sub = scrub_repair_page_from_good_copy(sblock_bad, |
1223 | sblock_good, | |
1224 | page_num, | |
1225 | force_write); | |
1226 | if (ret_sub) | |
1227 | ret = ret_sub; | |
a2de733c | 1228 | } |
b5d67f64 SB |
1229 | |
1230 | return ret; | |
1231 | } | |
1232 | ||
1233 | static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad, | |
1234 | struct scrub_block *sblock_good, | |
1235 | int page_num, int force_write) | |
1236 | { | |
7a9e9987 SB |
1237 | struct scrub_page *page_bad = sblock_bad->pagev[page_num]; |
1238 | struct scrub_page *page_good = sblock_good->pagev[page_num]; | |
b5d67f64 | 1239 | |
7a9e9987 SB |
1240 | BUG_ON(page_bad->page == NULL); |
1241 | BUG_ON(page_good->page == NULL); | |
b5d67f64 SB |
1242 | if (force_write || sblock_bad->header_error || |
1243 | sblock_bad->checksum_error || page_bad->io_error) { | |
1244 | struct bio *bio; | |
1245 | int ret; | |
1246 | DECLARE_COMPLETION_ONSTACK(complete); | |
1247 | ||
1248 | bio = bio_alloc(GFP_NOFS, 1); | |
e627ee7b TI |
1249 | if (!bio) |
1250 | return -EIO; | |
442a4f63 | 1251 | bio->bi_bdev = page_bad->dev->bdev; |
b5d67f64 SB |
1252 | bio->bi_sector = page_bad->physical >> 9; |
1253 | bio->bi_end_io = scrub_complete_bio_end_io; | |
1254 | bio->bi_private = &complete; | |
1255 | ||
1256 | ret = bio_add_page(bio, page_good->page, PAGE_SIZE, 0); | |
1257 | if (PAGE_SIZE != ret) { | |
1258 | bio_put(bio); | |
1259 | return -EIO; | |
13db62b7 | 1260 | } |
b5d67f64 SB |
1261 | btrfsic_submit_bio(WRITE, bio); |
1262 | ||
1263 | /* this will also unplug the queue */ | |
1264 | wait_for_completion(&complete); | |
442a4f63 SB |
1265 | if (!bio_flagged(bio, BIO_UPTODATE)) { |
1266 | btrfs_dev_stat_inc_and_print(page_bad->dev, | |
1267 | BTRFS_DEV_STAT_WRITE_ERRS); | |
1268 | bio_put(bio); | |
1269 | return -EIO; | |
1270 | } | |
b5d67f64 | 1271 | bio_put(bio); |
a2de733c AJ |
1272 | } |
1273 | ||
b5d67f64 SB |
1274 | return 0; |
1275 | } | |
1276 | ||
1277 | static void scrub_checksum(struct scrub_block *sblock) | |
1278 | { | |
1279 | u64 flags; | |
1280 | int ret; | |
1281 | ||
7a9e9987 SB |
1282 | WARN_ON(sblock->page_count < 1); |
1283 | flags = sblock->pagev[0]->flags; | |
b5d67f64 SB |
1284 | ret = 0; |
1285 | if (flags & BTRFS_EXTENT_FLAG_DATA) | |
1286 | ret = scrub_checksum_data(sblock); | |
1287 | else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) | |
1288 | ret = scrub_checksum_tree_block(sblock); | |
1289 | else if (flags & BTRFS_EXTENT_FLAG_SUPER) | |
1290 | (void)scrub_checksum_super(sblock); | |
1291 | else | |
1292 | WARN_ON(1); | |
1293 | if (ret) | |
1294 | scrub_handle_errored_block(sblock); | |
a2de733c AJ |
1295 | } |
1296 | ||
b5d67f64 | 1297 | static int scrub_checksum_data(struct scrub_block *sblock) |
a2de733c | 1298 | { |
d9d181c1 | 1299 | struct scrub_ctx *sctx = sblock->sctx; |
a2de733c | 1300 | u8 csum[BTRFS_CSUM_SIZE]; |
b5d67f64 SB |
1301 | u8 *on_disk_csum; |
1302 | struct page *page; | |
1303 | void *buffer; | |
a2de733c AJ |
1304 | u32 crc = ~(u32)0; |
1305 | int fail = 0; | |
a36cf8b8 | 1306 | struct btrfs_root *root = sctx->dev_root; |
b5d67f64 SB |
1307 | u64 len; |
1308 | int index; | |
a2de733c | 1309 | |
b5d67f64 | 1310 | BUG_ON(sblock->page_count < 1); |
7a9e9987 | 1311 | if (!sblock->pagev[0]->have_csum) |
a2de733c AJ |
1312 | return 0; |
1313 | ||
7a9e9987 SB |
1314 | on_disk_csum = sblock->pagev[0]->csum; |
1315 | page = sblock->pagev[0]->page; | |
9613bebb | 1316 | buffer = kmap_atomic(page); |
b5d67f64 | 1317 | |
d9d181c1 | 1318 | len = sctx->sectorsize; |
b5d67f64 SB |
1319 | index = 0; |
1320 | for (;;) { | |
1321 | u64 l = min_t(u64, len, PAGE_SIZE); | |
1322 | ||
1323 | crc = btrfs_csum_data(root, buffer, crc, l); | |
9613bebb | 1324 | kunmap_atomic(buffer); |
b5d67f64 SB |
1325 | len -= l; |
1326 | if (len == 0) | |
1327 | break; | |
1328 | index++; | |
1329 | BUG_ON(index >= sblock->page_count); | |
7a9e9987 SB |
1330 | BUG_ON(!sblock->pagev[index]->page); |
1331 | page = sblock->pagev[index]->page; | |
9613bebb | 1332 | buffer = kmap_atomic(page); |
b5d67f64 SB |
1333 | } |
1334 | ||
a2de733c | 1335 | btrfs_csum_final(crc, csum); |
d9d181c1 | 1336 | if (memcmp(csum, on_disk_csum, sctx->csum_size)) |
a2de733c AJ |
1337 | fail = 1; |
1338 | ||
a2de733c AJ |
1339 | return fail; |
1340 | } | |
1341 | ||
b5d67f64 | 1342 | static int scrub_checksum_tree_block(struct scrub_block *sblock) |
a2de733c | 1343 | { |
d9d181c1 | 1344 | struct scrub_ctx *sctx = sblock->sctx; |
a2de733c | 1345 | struct btrfs_header *h; |
a36cf8b8 | 1346 | struct btrfs_root *root = sctx->dev_root; |
a2de733c | 1347 | struct btrfs_fs_info *fs_info = root->fs_info; |
b5d67f64 SB |
1348 | u8 calculated_csum[BTRFS_CSUM_SIZE]; |
1349 | u8 on_disk_csum[BTRFS_CSUM_SIZE]; | |
1350 | struct page *page; | |
1351 | void *mapped_buffer; | |
1352 | u64 mapped_size; | |
1353 | void *p; | |
a2de733c AJ |
1354 | u32 crc = ~(u32)0; |
1355 | int fail = 0; | |
1356 | int crc_fail = 0; | |
b5d67f64 SB |
1357 | u64 len; |
1358 | int index; | |
1359 | ||
1360 | BUG_ON(sblock->page_count < 1); | |
7a9e9987 | 1361 | page = sblock->pagev[0]->page; |
9613bebb | 1362 | mapped_buffer = kmap_atomic(page); |
b5d67f64 | 1363 | h = (struct btrfs_header *)mapped_buffer; |
d9d181c1 | 1364 | memcpy(on_disk_csum, h->csum, sctx->csum_size); |
a2de733c AJ |
1365 | |
1366 | /* | |
1367 | * we don't use the getter functions here, as we | |
1368 | * a) don't have an extent buffer and | |
1369 | * b) the page is already kmapped | |
1370 | */ | |
a2de733c | 1371 | |
7a9e9987 | 1372 | if (sblock->pagev[0]->logical != le64_to_cpu(h->bytenr)) |
a2de733c AJ |
1373 | ++fail; |
1374 | ||
7a9e9987 | 1375 | if (sblock->pagev[0]->generation != le64_to_cpu(h->generation)) |
a2de733c AJ |
1376 | ++fail; |
1377 | ||
1378 | if (memcmp(h->fsid, fs_info->fsid, BTRFS_UUID_SIZE)) | |
1379 | ++fail; | |
1380 | ||
1381 | if (memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid, | |
1382 | BTRFS_UUID_SIZE)) | |
1383 | ++fail; | |
1384 | ||
d9d181c1 SB |
1385 | BUG_ON(sctx->nodesize != sctx->leafsize); |
1386 | len = sctx->nodesize - BTRFS_CSUM_SIZE; | |
b5d67f64 SB |
1387 | mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE; |
1388 | p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE; | |
1389 | index = 0; | |
1390 | for (;;) { | |
1391 | u64 l = min_t(u64, len, mapped_size); | |
1392 | ||
1393 | crc = btrfs_csum_data(root, p, crc, l); | |
9613bebb | 1394 | kunmap_atomic(mapped_buffer); |
b5d67f64 SB |
1395 | len -= l; |
1396 | if (len == 0) | |
1397 | break; | |
1398 | index++; | |
1399 | BUG_ON(index >= sblock->page_count); | |
7a9e9987 SB |
1400 | BUG_ON(!sblock->pagev[index]->page); |
1401 | page = sblock->pagev[index]->page; | |
9613bebb | 1402 | mapped_buffer = kmap_atomic(page); |
b5d67f64 SB |
1403 | mapped_size = PAGE_SIZE; |
1404 | p = mapped_buffer; | |
1405 | } | |
1406 | ||
1407 | btrfs_csum_final(crc, calculated_csum); | |
d9d181c1 | 1408 | if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size)) |
a2de733c AJ |
1409 | ++crc_fail; |
1410 | ||
a2de733c AJ |
1411 | return fail || crc_fail; |
1412 | } | |
1413 | ||
b5d67f64 | 1414 | static int scrub_checksum_super(struct scrub_block *sblock) |
a2de733c AJ |
1415 | { |
1416 | struct btrfs_super_block *s; | |
d9d181c1 | 1417 | struct scrub_ctx *sctx = sblock->sctx; |
a36cf8b8 | 1418 | struct btrfs_root *root = sctx->dev_root; |
a2de733c | 1419 | struct btrfs_fs_info *fs_info = root->fs_info; |
b5d67f64 SB |
1420 | u8 calculated_csum[BTRFS_CSUM_SIZE]; |
1421 | u8 on_disk_csum[BTRFS_CSUM_SIZE]; | |
1422 | struct page *page; | |
1423 | void *mapped_buffer; | |
1424 | u64 mapped_size; | |
1425 | void *p; | |
a2de733c | 1426 | u32 crc = ~(u32)0; |
442a4f63 SB |
1427 | int fail_gen = 0; |
1428 | int fail_cor = 0; | |
b5d67f64 SB |
1429 | u64 len; |
1430 | int index; | |
a2de733c | 1431 | |
b5d67f64 | 1432 | BUG_ON(sblock->page_count < 1); |
7a9e9987 | 1433 | page = sblock->pagev[0]->page; |
9613bebb | 1434 | mapped_buffer = kmap_atomic(page); |
b5d67f64 | 1435 | s = (struct btrfs_super_block *)mapped_buffer; |
d9d181c1 | 1436 | memcpy(on_disk_csum, s->csum, sctx->csum_size); |
a2de733c | 1437 | |
7a9e9987 | 1438 | if (sblock->pagev[0]->logical != le64_to_cpu(s->bytenr)) |
442a4f63 | 1439 | ++fail_cor; |
a2de733c | 1440 | |
7a9e9987 | 1441 | if (sblock->pagev[0]->generation != le64_to_cpu(s->generation)) |
442a4f63 | 1442 | ++fail_gen; |
a2de733c AJ |
1443 | |
1444 | if (memcmp(s->fsid, fs_info->fsid, BTRFS_UUID_SIZE)) | |
442a4f63 | 1445 | ++fail_cor; |
a2de733c | 1446 | |
b5d67f64 SB |
1447 | len = BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE; |
1448 | mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE; | |
1449 | p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE; | |
1450 | index = 0; | |
1451 | for (;;) { | |
1452 | u64 l = min_t(u64, len, mapped_size); | |
1453 | ||
1454 | crc = btrfs_csum_data(root, p, crc, l); | |
9613bebb | 1455 | kunmap_atomic(mapped_buffer); |
b5d67f64 SB |
1456 | len -= l; |
1457 | if (len == 0) | |
1458 | break; | |
1459 | index++; | |
1460 | BUG_ON(index >= sblock->page_count); | |
7a9e9987 SB |
1461 | BUG_ON(!sblock->pagev[index]->page); |
1462 | page = sblock->pagev[index]->page; | |
9613bebb | 1463 | mapped_buffer = kmap_atomic(page); |
b5d67f64 SB |
1464 | mapped_size = PAGE_SIZE; |
1465 | p = mapped_buffer; | |
1466 | } | |
1467 | ||
1468 | btrfs_csum_final(crc, calculated_csum); | |
d9d181c1 | 1469 | if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size)) |
442a4f63 | 1470 | ++fail_cor; |
a2de733c | 1471 | |
442a4f63 | 1472 | if (fail_cor + fail_gen) { |
a2de733c AJ |
1473 | /* |
1474 | * if we find an error in a super block, we just report it. | |
1475 | * They will get written with the next transaction commit | |
1476 | * anyway | |
1477 | */ | |
d9d181c1 SB |
1478 | spin_lock(&sctx->stat_lock); |
1479 | ++sctx->stat.super_errors; | |
1480 | spin_unlock(&sctx->stat_lock); | |
442a4f63 | 1481 | if (fail_cor) |
7a9e9987 | 1482 | btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev, |
442a4f63 SB |
1483 | BTRFS_DEV_STAT_CORRUPTION_ERRS); |
1484 | else | |
7a9e9987 | 1485 | btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev, |
442a4f63 | 1486 | BTRFS_DEV_STAT_GENERATION_ERRS); |
a2de733c AJ |
1487 | } |
1488 | ||
442a4f63 | 1489 | return fail_cor + fail_gen; |
a2de733c AJ |
1490 | } |
1491 | ||
b5d67f64 SB |
1492 | static void scrub_block_get(struct scrub_block *sblock) |
1493 | { | |
1494 | atomic_inc(&sblock->ref_count); | |
1495 | } | |
1496 | ||
1497 | static void scrub_block_put(struct scrub_block *sblock) | |
1498 | { | |
1499 | if (atomic_dec_and_test(&sblock->ref_count)) { | |
1500 | int i; | |
1501 | ||
1502 | for (i = 0; i < sblock->page_count; i++) | |
7a9e9987 | 1503 | scrub_page_put(sblock->pagev[i]); |
b5d67f64 SB |
1504 | kfree(sblock); |
1505 | } | |
1506 | } | |
1507 | ||
7a9e9987 SB |
1508 | static void scrub_page_get(struct scrub_page *spage) |
1509 | { | |
1510 | atomic_inc(&spage->ref_count); | |
1511 | } | |
1512 | ||
1513 | static void scrub_page_put(struct scrub_page *spage) | |
1514 | { | |
1515 | if (atomic_dec_and_test(&spage->ref_count)) { | |
1516 | if (spage->page) | |
1517 | __free_page(spage->page); | |
1518 | kfree(spage); | |
1519 | } | |
1520 | } | |
1521 | ||
d9d181c1 | 1522 | static void scrub_submit(struct scrub_ctx *sctx) |
a2de733c AJ |
1523 | { |
1524 | struct scrub_bio *sbio; | |
1525 | ||
d9d181c1 | 1526 | if (sctx->curr == -1) |
1623edeb | 1527 | return; |
a2de733c | 1528 | |
d9d181c1 SB |
1529 | sbio = sctx->bios[sctx->curr]; |
1530 | sctx->curr = -1; | |
b6bfebc1 | 1531 | scrub_pending_bio_inc(sctx); |
a2de733c | 1532 | |
21adbd5c | 1533 | btrfsic_submit_bio(READ, sbio->bio); |
a2de733c AJ |
1534 | } |
1535 | ||
d9d181c1 | 1536 | static int scrub_add_page_to_bio(struct scrub_ctx *sctx, |
b5d67f64 | 1537 | struct scrub_page *spage) |
a2de733c | 1538 | { |
b5d67f64 | 1539 | struct scrub_block *sblock = spage->sblock; |
a2de733c | 1540 | struct scrub_bio *sbio; |
69f4cb52 | 1541 | int ret; |
a2de733c AJ |
1542 | |
1543 | again: | |
1544 | /* | |
1545 | * grab a fresh bio or wait for one to become available | |
1546 | */ | |
d9d181c1 SB |
1547 | while (sctx->curr == -1) { |
1548 | spin_lock(&sctx->list_lock); | |
1549 | sctx->curr = sctx->first_free; | |
1550 | if (sctx->curr != -1) { | |
1551 | sctx->first_free = sctx->bios[sctx->curr]->next_free; | |
1552 | sctx->bios[sctx->curr]->next_free = -1; | |
1553 | sctx->bios[sctx->curr]->page_count = 0; | |
1554 | spin_unlock(&sctx->list_lock); | |
a2de733c | 1555 | } else { |
d9d181c1 SB |
1556 | spin_unlock(&sctx->list_lock); |
1557 | wait_event(sctx->list_wait, sctx->first_free != -1); | |
a2de733c AJ |
1558 | } |
1559 | } | |
d9d181c1 | 1560 | sbio = sctx->bios[sctx->curr]; |
b5d67f64 | 1561 | if (sbio->page_count == 0) { |
69f4cb52 AJ |
1562 | struct bio *bio; |
1563 | ||
b5d67f64 SB |
1564 | sbio->physical = spage->physical; |
1565 | sbio->logical = spage->logical; | |
a36cf8b8 | 1566 | sbio->dev = spage->dev; |
b5d67f64 SB |
1567 | bio = sbio->bio; |
1568 | if (!bio) { | |
d9d181c1 | 1569 | bio = bio_alloc(GFP_NOFS, sctx->pages_per_bio); |
b5d67f64 SB |
1570 | if (!bio) |
1571 | return -ENOMEM; | |
1572 | sbio->bio = bio; | |
1573 | } | |
69f4cb52 AJ |
1574 | |
1575 | bio->bi_private = sbio; | |
1576 | bio->bi_end_io = scrub_bio_end_io; | |
a36cf8b8 SB |
1577 | bio->bi_bdev = sbio->dev->bdev; |
1578 | bio->bi_sector = sbio->physical >> 9; | |
69f4cb52 | 1579 | sbio->err = 0; |
b5d67f64 SB |
1580 | } else if (sbio->physical + sbio->page_count * PAGE_SIZE != |
1581 | spage->physical || | |
1582 | sbio->logical + sbio->page_count * PAGE_SIZE != | |
a36cf8b8 SB |
1583 | spage->logical || |
1584 | sbio->dev != spage->dev) { | |
d9d181c1 | 1585 | scrub_submit(sctx); |
a2de733c AJ |
1586 | goto again; |
1587 | } | |
69f4cb52 | 1588 | |
b5d67f64 SB |
1589 | sbio->pagev[sbio->page_count] = spage; |
1590 | ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0); | |
1591 | if (ret != PAGE_SIZE) { | |
1592 | if (sbio->page_count < 1) { | |
1593 | bio_put(sbio->bio); | |
1594 | sbio->bio = NULL; | |
1595 | return -EIO; | |
1596 | } | |
d9d181c1 | 1597 | scrub_submit(sctx); |
69f4cb52 AJ |
1598 | goto again; |
1599 | } | |
1600 | ||
b5d67f64 SB |
1601 | scrub_block_get(sblock); /* one for the added page */ |
1602 | atomic_inc(&sblock->outstanding_pages); | |
1603 | sbio->page_count++; | |
d9d181c1 SB |
1604 | if (sbio->page_count == sctx->pages_per_bio) |
1605 | scrub_submit(sctx); | |
b5d67f64 SB |
1606 | |
1607 | return 0; | |
1608 | } | |
1609 | ||
d9d181c1 | 1610 | static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len, |
a36cf8b8 SB |
1611 | u64 physical, struct btrfs_device *dev, u64 flags, |
1612 | u64 gen, int mirror_num, u8 *csum, int force) | |
b5d67f64 SB |
1613 | { |
1614 | struct scrub_block *sblock; | |
1615 | int index; | |
1616 | ||
1617 | sblock = kzalloc(sizeof(*sblock), GFP_NOFS); | |
1618 | if (!sblock) { | |
d9d181c1 SB |
1619 | spin_lock(&sctx->stat_lock); |
1620 | sctx->stat.malloc_errors++; | |
1621 | spin_unlock(&sctx->stat_lock); | |
b5d67f64 | 1622 | return -ENOMEM; |
a2de733c | 1623 | } |
b5d67f64 | 1624 | |
7a9e9987 SB |
1625 | /* one ref inside this function, plus one for each page added to |
1626 | * a bio later on */ | |
b5d67f64 | 1627 | atomic_set(&sblock->ref_count, 1); |
d9d181c1 | 1628 | sblock->sctx = sctx; |
b5d67f64 SB |
1629 | sblock->no_io_error_seen = 1; |
1630 | ||
1631 | for (index = 0; len > 0; index++) { | |
7a9e9987 | 1632 | struct scrub_page *spage; |
b5d67f64 SB |
1633 | u64 l = min_t(u64, len, PAGE_SIZE); |
1634 | ||
7a9e9987 SB |
1635 | spage = kzalloc(sizeof(*spage), GFP_NOFS); |
1636 | if (!spage) { | |
1637 | leave_nomem: | |
d9d181c1 SB |
1638 | spin_lock(&sctx->stat_lock); |
1639 | sctx->stat.malloc_errors++; | |
1640 | spin_unlock(&sctx->stat_lock); | |
7a9e9987 | 1641 | scrub_block_put(sblock); |
b5d67f64 SB |
1642 | return -ENOMEM; |
1643 | } | |
7a9e9987 SB |
1644 | BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK); |
1645 | scrub_page_get(spage); | |
1646 | sblock->pagev[index] = spage; | |
b5d67f64 | 1647 | spage->sblock = sblock; |
a36cf8b8 | 1648 | spage->dev = dev; |
b5d67f64 SB |
1649 | spage->flags = flags; |
1650 | spage->generation = gen; | |
1651 | spage->logical = logical; | |
1652 | spage->physical = physical; | |
1653 | spage->mirror_num = mirror_num; | |
1654 | if (csum) { | |
1655 | spage->have_csum = 1; | |
d9d181c1 | 1656 | memcpy(spage->csum, csum, sctx->csum_size); |
b5d67f64 SB |
1657 | } else { |
1658 | spage->have_csum = 0; | |
1659 | } | |
1660 | sblock->page_count++; | |
7a9e9987 SB |
1661 | spage->page = alloc_page(GFP_NOFS); |
1662 | if (!spage->page) | |
1663 | goto leave_nomem; | |
b5d67f64 SB |
1664 | len -= l; |
1665 | logical += l; | |
1666 | physical += l; | |
1667 | } | |
1668 | ||
7a9e9987 | 1669 | WARN_ON(sblock->page_count == 0); |
b5d67f64 | 1670 | for (index = 0; index < sblock->page_count; index++) { |
7a9e9987 | 1671 | struct scrub_page *spage = sblock->pagev[index]; |
1bc87793 AJ |
1672 | int ret; |
1673 | ||
d9d181c1 | 1674 | ret = scrub_add_page_to_bio(sctx, spage); |
b5d67f64 SB |
1675 | if (ret) { |
1676 | scrub_block_put(sblock); | |
1bc87793 | 1677 | return ret; |
b5d67f64 | 1678 | } |
1bc87793 | 1679 | } |
a2de733c | 1680 | |
b5d67f64 | 1681 | if (force) |
d9d181c1 | 1682 | scrub_submit(sctx); |
a2de733c | 1683 | |
b5d67f64 SB |
1684 | /* last one frees, either here or in bio completion for last page */ |
1685 | scrub_block_put(sblock); | |
a2de733c AJ |
1686 | return 0; |
1687 | } | |
1688 | ||
b5d67f64 SB |
1689 | static void scrub_bio_end_io(struct bio *bio, int err) |
1690 | { | |
1691 | struct scrub_bio *sbio = bio->bi_private; | |
a36cf8b8 | 1692 | struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info; |
b5d67f64 SB |
1693 | |
1694 | sbio->err = err; | |
1695 | sbio->bio = bio; | |
1696 | ||
1697 | btrfs_queue_worker(&fs_info->scrub_workers, &sbio->work); | |
1698 | } | |
1699 | ||
1700 | static void scrub_bio_end_io_worker(struct btrfs_work *work) | |
1701 | { | |
1702 | struct scrub_bio *sbio = container_of(work, struct scrub_bio, work); | |
d9d181c1 | 1703 | struct scrub_ctx *sctx = sbio->sctx; |
b5d67f64 SB |
1704 | int i; |
1705 | ||
1706 | BUG_ON(sbio->page_count > SCRUB_PAGES_PER_BIO); | |
1707 | if (sbio->err) { | |
1708 | for (i = 0; i < sbio->page_count; i++) { | |
1709 | struct scrub_page *spage = sbio->pagev[i]; | |
1710 | ||
1711 | spage->io_error = 1; | |
1712 | spage->sblock->no_io_error_seen = 0; | |
1713 | } | |
1714 | } | |
1715 | ||
1716 | /* now complete the scrub_block items that have all pages completed */ | |
1717 | for (i = 0; i < sbio->page_count; i++) { | |
1718 | struct scrub_page *spage = sbio->pagev[i]; | |
1719 | struct scrub_block *sblock = spage->sblock; | |
1720 | ||
1721 | if (atomic_dec_and_test(&sblock->outstanding_pages)) | |
1722 | scrub_block_complete(sblock); | |
1723 | scrub_block_put(sblock); | |
1724 | } | |
1725 | ||
b5d67f64 SB |
1726 | bio_put(sbio->bio); |
1727 | sbio->bio = NULL; | |
d9d181c1 SB |
1728 | spin_lock(&sctx->list_lock); |
1729 | sbio->next_free = sctx->first_free; | |
1730 | sctx->first_free = sbio->index; | |
1731 | spin_unlock(&sctx->list_lock); | |
b6bfebc1 | 1732 | scrub_pending_bio_dec(sctx); |
b5d67f64 SB |
1733 | } |
1734 | ||
1735 | static void scrub_block_complete(struct scrub_block *sblock) | |
1736 | { | |
1737 | if (!sblock->no_io_error_seen) | |
1738 | scrub_handle_errored_block(sblock); | |
1739 | else | |
1740 | scrub_checksum(sblock); | |
1741 | } | |
1742 | ||
d9d181c1 | 1743 | static int scrub_find_csum(struct scrub_ctx *sctx, u64 logical, u64 len, |
a2de733c AJ |
1744 | u8 *csum) |
1745 | { | |
1746 | struct btrfs_ordered_sum *sum = NULL; | |
1747 | int ret = 0; | |
1748 | unsigned long i; | |
1749 | unsigned long num_sectors; | |
a2de733c | 1750 | |
d9d181c1 SB |
1751 | while (!list_empty(&sctx->csum_list)) { |
1752 | sum = list_first_entry(&sctx->csum_list, | |
a2de733c AJ |
1753 | struct btrfs_ordered_sum, list); |
1754 | if (sum->bytenr > logical) | |
1755 | return 0; | |
1756 | if (sum->bytenr + sum->len > logical) | |
1757 | break; | |
1758 | ||
d9d181c1 | 1759 | ++sctx->stat.csum_discards; |
a2de733c AJ |
1760 | list_del(&sum->list); |
1761 | kfree(sum); | |
1762 | sum = NULL; | |
1763 | } | |
1764 | if (!sum) | |
1765 | return 0; | |
1766 | ||
d9d181c1 | 1767 | num_sectors = sum->len / sctx->sectorsize; |
a2de733c AJ |
1768 | for (i = 0; i < num_sectors; ++i) { |
1769 | if (sum->sums[i].bytenr == logical) { | |
d9d181c1 | 1770 | memcpy(csum, &sum->sums[i].sum, sctx->csum_size); |
a2de733c AJ |
1771 | ret = 1; |
1772 | break; | |
1773 | } | |
1774 | } | |
1775 | if (ret && i == num_sectors - 1) { | |
1776 | list_del(&sum->list); | |
1777 | kfree(sum); | |
1778 | } | |
1779 | return ret; | |
1780 | } | |
1781 | ||
1782 | /* scrub extent tries to collect up to 64 kB for each bio */ | |
d9d181c1 | 1783 | static int scrub_extent(struct scrub_ctx *sctx, u64 logical, u64 len, |
a36cf8b8 SB |
1784 | u64 physical, struct btrfs_device *dev, u64 flags, |
1785 | u64 gen, int mirror_num) | |
a2de733c AJ |
1786 | { |
1787 | int ret; | |
1788 | u8 csum[BTRFS_CSUM_SIZE]; | |
b5d67f64 SB |
1789 | u32 blocksize; |
1790 | ||
1791 | if (flags & BTRFS_EXTENT_FLAG_DATA) { | |
d9d181c1 SB |
1792 | blocksize = sctx->sectorsize; |
1793 | spin_lock(&sctx->stat_lock); | |
1794 | sctx->stat.data_extents_scrubbed++; | |
1795 | sctx->stat.data_bytes_scrubbed += len; | |
1796 | spin_unlock(&sctx->stat_lock); | |
b5d67f64 | 1797 | } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { |
d9d181c1 SB |
1798 | BUG_ON(sctx->nodesize != sctx->leafsize); |
1799 | blocksize = sctx->nodesize; | |
1800 | spin_lock(&sctx->stat_lock); | |
1801 | sctx->stat.tree_extents_scrubbed++; | |
1802 | sctx->stat.tree_bytes_scrubbed += len; | |
1803 | spin_unlock(&sctx->stat_lock); | |
b5d67f64 | 1804 | } else { |
d9d181c1 | 1805 | blocksize = sctx->sectorsize; |
b5d67f64 SB |
1806 | BUG_ON(1); |
1807 | } | |
a2de733c AJ |
1808 | |
1809 | while (len) { | |
b5d67f64 | 1810 | u64 l = min_t(u64, len, blocksize); |
a2de733c AJ |
1811 | int have_csum = 0; |
1812 | ||
1813 | if (flags & BTRFS_EXTENT_FLAG_DATA) { | |
1814 | /* push csums to sbio */ | |
d9d181c1 | 1815 | have_csum = scrub_find_csum(sctx, logical, l, csum); |
a2de733c | 1816 | if (have_csum == 0) |
d9d181c1 | 1817 | ++sctx->stat.no_csum; |
a2de733c | 1818 | } |
a36cf8b8 | 1819 | ret = scrub_pages(sctx, logical, l, physical, dev, flags, gen, |
b5d67f64 | 1820 | mirror_num, have_csum ? csum : NULL, 0); |
a2de733c AJ |
1821 | if (ret) |
1822 | return ret; | |
1823 | len -= l; | |
1824 | logical += l; | |
1825 | physical += l; | |
1826 | } | |
1827 | return 0; | |
1828 | } | |
1829 | ||
d9d181c1 | 1830 | static noinline_for_stack int scrub_stripe(struct scrub_ctx *sctx, |
a36cf8b8 SB |
1831 | struct map_lookup *map, |
1832 | struct btrfs_device *scrub_dev, | |
1833 | int num, u64 base, u64 length) | |
a2de733c AJ |
1834 | { |
1835 | struct btrfs_path *path; | |
a36cf8b8 | 1836 | struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info; |
a2de733c AJ |
1837 | struct btrfs_root *root = fs_info->extent_root; |
1838 | struct btrfs_root *csum_root = fs_info->csum_root; | |
1839 | struct btrfs_extent_item *extent; | |
e7786c3a | 1840 | struct blk_plug plug; |
a2de733c AJ |
1841 | u64 flags; |
1842 | int ret; | |
1843 | int slot; | |
1844 | int i; | |
1845 | u64 nstripes; | |
a2de733c AJ |
1846 | struct extent_buffer *l; |
1847 | struct btrfs_key key; | |
1848 | u64 physical; | |
1849 | u64 logical; | |
1850 | u64 generation; | |
e12fa9cd | 1851 | int mirror_num; |
7a26285e AJ |
1852 | struct reada_control *reada1; |
1853 | struct reada_control *reada2; | |
1854 | struct btrfs_key key_start; | |
1855 | struct btrfs_key key_end; | |
a2de733c AJ |
1856 | u64 increment = map->stripe_len; |
1857 | u64 offset; | |
1858 | ||
1859 | nstripes = length; | |
1860 | offset = 0; | |
1861 | do_div(nstripes, map->stripe_len); | |
1862 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { | |
1863 | offset = map->stripe_len * num; | |
1864 | increment = map->stripe_len * map->num_stripes; | |
193ea74b | 1865 | mirror_num = 1; |
a2de733c AJ |
1866 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
1867 | int factor = map->num_stripes / map->sub_stripes; | |
1868 | offset = map->stripe_len * (num / map->sub_stripes); | |
1869 | increment = map->stripe_len * factor; | |
193ea74b | 1870 | mirror_num = num % map->sub_stripes + 1; |
a2de733c AJ |
1871 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { |
1872 | increment = map->stripe_len; | |
193ea74b | 1873 | mirror_num = num % map->num_stripes + 1; |
a2de733c AJ |
1874 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
1875 | increment = map->stripe_len; | |
193ea74b | 1876 | mirror_num = num % map->num_stripes + 1; |
a2de733c AJ |
1877 | } else { |
1878 | increment = map->stripe_len; | |
193ea74b | 1879 | mirror_num = 1; |
a2de733c AJ |
1880 | } |
1881 | ||
1882 | path = btrfs_alloc_path(); | |
1883 | if (!path) | |
1884 | return -ENOMEM; | |
1885 | ||
b5d67f64 SB |
1886 | /* |
1887 | * work on commit root. The related disk blocks are static as | |
1888 | * long as COW is applied. This means, it is save to rewrite | |
1889 | * them to repair disk errors without any race conditions | |
1890 | */ | |
a2de733c AJ |
1891 | path->search_commit_root = 1; |
1892 | path->skip_locking = 1; | |
1893 | ||
1894 | /* | |
7a26285e AJ |
1895 | * trigger the readahead for extent tree csum tree and wait for |
1896 | * completion. During readahead, the scrub is officially paused | |
1897 | * to not hold off transaction commits | |
a2de733c AJ |
1898 | */ |
1899 | logical = base + offset; | |
a2de733c | 1900 | |
d9d181c1 | 1901 | wait_event(sctx->list_wait, |
b6bfebc1 | 1902 | atomic_read(&sctx->bios_in_flight) == 0); |
7a26285e AJ |
1903 | atomic_inc(&fs_info->scrubs_paused); |
1904 | wake_up(&fs_info->scrub_pause_wait); | |
1905 | ||
1906 | /* FIXME it might be better to start readahead at commit root */ | |
1907 | key_start.objectid = logical; | |
1908 | key_start.type = BTRFS_EXTENT_ITEM_KEY; | |
1909 | key_start.offset = (u64)0; | |
1910 | key_end.objectid = base + offset + nstripes * increment; | |
1911 | key_end.type = BTRFS_EXTENT_ITEM_KEY; | |
1912 | key_end.offset = (u64)0; | |
1913 | reada1 = btrfs_reada_add(root, &key_start, &key_end); | |
1914 | ||
1915 | key_start.objectid = BTRFS_EXTENT_CSUM_OBJECTID; | |
1916 | key_start.type = BTRFS_EXTENT_CSUM_KEY; | |
1917 | key_start.offset = logical; | |
1918 | key_end.objectid = BTRFS_EXTENT_CSUM_OBJECTID; | |
1919 | key_end.type = BTRFS_EXTENT_CSUM_KEY; | |
1920 | key_end.offset = base + offset + nstripes * increment; | |
1921 | reada2 = btrfs_reada_add(csum_root, &key_start, &key_end); | |
1922 | ||
1923 | if (!IS_ERR(reada1)) | |
1924 | btrfs_reada_wait(reada1); | |
1925 | if (!IS_ERR(reada2)) | |
1926 | btrfs_reada_wait(reada2); | |
1927 | ||
1928 | mutex_lock(&fs_info->scrub_lock); | |
1929 | while (atomic_read(&fs_info->scrub_pause_req)) { | |
1930 | mutex_unlock(&fs_info->scrub_lock); | |
1931 | wait_event(fs_info->scrub_pause_wait, | |
1932 | atomic_read(&fs_info->scrub_pause_req) == 0); | |
1933 | mutex_lock(&fs_info->scrub_lock); | |
a2de733c | 1934 | } |
7a26285e AJ |
1935 | atomic_dec(&fs_info->scrubs_paused); |
1936 | mutex_unlock(&fs_info->scrub_lock); | |
1937 | wake_up(&fs_info->scrub_pause_wait); | |
a2de733c AJ |
1938 | |
1939 | /* | |
1940 | * collect all data csums for the stripe to avoid seeking during | |
1941 | * the scrub. This might currently (crc32) end up to be about 1MB | |
1942 | */ | |
e7786c3a | 1943 | blk_start_plug(&plug); |
a2de733c | 1944 | |
a2de733c AJ |
1945 | /* |
1946 | * now find all extents for each stripe and scrub them | |
1947 | */ | |
7a26285e AJ |
1948 | logical = base + offset; |
1949 | physical = map->stripes[num].physical; | |
a2de733c | 1950 | ret = 0; |
7a26285e | 1951 | for (i = 0; i < nstripes; ++i) { |
a2de733c AJ |
1952 | /* |
1953 | * canceled? | |
1954 | */ | |
1955 | if (atomic_read(&fs_info->scrub_cancel_req) || | |
d9d181c1 | 1956 | atomic_read(&sctx->cancel_req)) { |
a2de733c AJ |
1957 | ret = -ECANCELED; |
1958 | goto out; | |
1959 | } | |
1960 | /* | |
1961 | * check to see if we have to pause | |
1962 | */ | |
1963 | if (atomic_read(&fs_info->scrub_pause_req)) { | |
1964 | /* push queued extents */ | |
d9d181c1 SB |
1965 | scrub_submit(sctx); |
1966 | wait_event(sctx->list_wait, | |
b6bfebc1 | 1967 | atomic_read(&sctx->bios_in_flight) == 0); |
a2de733c AJ |
1968 | atomic_inc(&fs_info->scrubs_paused); |
1969 | wake_up(&fs_info->scrub_pause_wait); | |
1970 | mutex_lock(&fs_info->scrub_lock); | |
1971 | while (atomic_read(&fs_info->scrub_pause_req)) { | |
1972 | mutex_unlock(&fs_info->scrub_lock); | |
1973 | wait_event(fs_info->scrub_pause_wait, | |
1974 | atomic_read(&fs_info->scrub_pause_req) == 0); | |
1975 | mutex_lock(&fs_info->scrub_lock); | |
1976 | } | |
1977 | atomic_dec(&fs_info->scrubs_paused); | |
1978 | mutex_unlock(&fs_info->scrub_lock); | |
1979 | wake_up(&fs_info->scrub_pause_wait); | |
a2de733c AJ |
1980 | } |
1981 | ||
7a26285e AJ |
1982 | ret = btrfs_lookup_csums_range(csum_root, logical, |
1983 | logical + map->stripe_len - 1, | |
d9d181c1 | 1984 | &sctx->csum_list, 1); |
7a26285e AJ |
1985 | if (ret) |
1986 | goto out; | |
1987 | ||
a2de733c AJ |
1988 | key.objectid = logical; |
1989 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
1990 | key.offset = (u64)0; | |
1991 | ||
1992 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1993 | if (ret < 0) | |
1994 | goto out; | |
8c51032f | 1995 | if (ret > 0) { |
a2de733c AJ |
1996 | ret = btrfs_previous_item(root, path, 0, |
1997 | BTRFS_EXTENT_ITEM_KEY); | |
1998 | if (ret < 0) | |
1999 | goto out; | |
8c51032f AJ |
2000 | if (ret > 0) { |
2001 | /* there's no smaller item, so stick with the | |
2002 | * larger one */ | |
2003 | btrfs_release_path(path); | |
2004 | ret = btrfs_search_slot(NULL, root, &key, | |
2005 | path, 0, 0); | |
2006 | if (ret < 0) | |
2007 | goto out; | |
2008 | } | |
a2de733c AJ |
2009 | } |
2010 | ||
2011 | while (1) { | |
2012 | l = path->nodes[0]; | |
2013 | slot = path->slots[0]; | |
2014 | if (slot >= btrfs_header_nritems(l)) { | |
2015 | ret = btrfs_next_leaf(root, path); | |
2016 | if (ret == 0) | |
2017 | continue; | |
2018 | if (ret < 0) | |
2019 | goto out; | |
2020 | ||
2021 | break; | |
2022 | } | |
2023 | btrfs_item_key_to_cpu(l, &key, slot); | |
2024 | ||
2025 | if (key.objectid + key.offset <= logical) | |
2026 | goto next; | |
2027 | ||
2028 | if (key.objectid >= logical + map->stripe_len) | |
2029 | break; | |
2030 | ||
2031 | if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY) | |
2032 | goto next; | |
2033 | ||
2034 | extent = btrfs_item_ptr(l, slot, | |
2035 | struct btrfs_extent_item); | |
2036 | flags = btrfs_extent_flags(l, extent); | |
2037 | generation = btrfs_extent_generation(l, extent); | |
2038 | ||
2039 | if (key.objectid < logical && | |
2040 | (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) { | |
2041 | printk(KERN_ERR | |
2042 | "btrfs scrub: tree block %llu spanning " | |
2043 | "stripes, ignored. logical=%llu\n", | |
2044 | (unsigned long long)key.objectid, | |
2045 | (unsigned long long)logical); | |
2046 | goto next; | |
2047 | } | |
2048 | ||
2049 | /* | |
2050 | * trim extent to this stripe | |
2051 | */ | |
2052 | if (key.objectid < logical) { | |
2053 | key.offset -= logical - key.objectid; | |
2054 | key.objectid = logical; | |
2055 | } | |
2056 | if (key.objectid + key.offset > | |
2057 | logical + map->stripe_len) { | |
2058 | key.offset = logical + map->stripe_len - | |
2059 | key.objectid; | |
2060 | } | |
2061 | ||
d9d181c1 | 2062 | ret = scrub_extent(sctx, key.objectid, key.offset, |
a2de733c | 2063 | key.objectid - logical + physical, |
a36cf8b8 SB |
2064 | scrub_dev, flags, generation, |
2065 | mirror_num); | |
a2de733c AJ |
2066 | if (ret) |
2067 | goto out; | |
2068 | ||
2069 | next: | |
2070 | path->slots[0]++; | |
2071 | } | |
71267333 | 2072 | btrfs_release_path(path); |
a2de733c AJ |
2073 | logical += increment; |
2074 | physical += map->stripe_len; | |
d9d181c1 SB |
2075 | spin_lock(&sctx->stat_lock); |
2076 | sctx->stat.last_physical = physical; | |
2077 | spin_unlock(&sctx->stat_lock); | |
a2de733c AJ |
2078 | } |
2079 | /* push queued extents */ | |
d9d181c1 | 2080 | scrub_submit(sctx); |
a2de733c AJ |
2081 | |
2082 | out: | |
e7786c3a | 2083 | blk_finish_plug(&plug); |
a2de733c AJ |
2084 | btrfs_free_path(path); |
2085 | return ret < 0 ? ret : 0; | |
2086 | } | |
2087 | ||
d9d181c1 | 2088 | static noinline_for_stack int scrub_chunk(struct scrub_ctx *sctx, |
a36cf8b8 SB |
2089 | struct btrfs_device *scrub_dev, |
2090 | u64 chunk_tree, u64 chunk_objectid, | |
2091 | u64 chunk_offset, u64 length, | |
2092 | u64 dev_offset) | |
a2de733c AJ |
2093 | { |
2094 | struct btrfs_mapping_tree *map_tree = | |
a36cf8b8 | 2095 | &sctx->dev_root->fs_info->mapping_tree; |
a2de733c AJ |
2096 | struct map_lookup *map; |
2097 | struct extent_map *em; | |
2098 | int i; | |
2099 | int ret = -EINVAL; | |
2100 | ||
2101 | read_lock(&map_tree->map_tree.lock); | |
2102 | em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1); | |
2103 | read_unlock(&map_tree->map_tree.lock); | |
2104 | ||
2105 | if (!em) | |
2106 | return -EINVAL; | |
2107 | ||
2108 | map = (struct map_lookup *)em->bdev; | |
2109 | if (em->start != chunk_offset) | |
2110 | goto out; | |
2111 | ||
2112 | if (em->len < length) | |
2113 | goto out; | |
2114 | ||
2115 | for (i = 0; i < map->num_stripes; ++i) { | |
a36cf8b8 | 2116 | if (map->stripes[i].dev->bdev == scrub_dev->bdev && |
859acaf1 | 2117 | map->stripes[i].physical == dev_offset) { |
a36cf8b8 SB |
2118 | ret = scrub_stripe(sctx, map, scrub_dev, i, |
2119 | chunk_offset, length); | |
a2de733c AJ |
2120 | if (ret) |
2121 | goto out; | |
2122 | } | |
2123 | } | |
2124 | out: | |
2125 | free_extent_map(em); | |
2126 | ||
2127 | return ret; | |
2128 | } | |
2129 | ||
2130 | static noinline_for_stack | |
a36cf8b8 SB |
2131 | int scrub_enumerate_chunks(struct scrub_ctx *sctx, |
2132 | struct btrfs_device *scrub_dev, u64 start, u64 end) | |
a2de733c AJ |
2133 | { |
2134 | struct btrfs_dev_extent *dev_extent = NULL; | |
2135 | struct btrfs_path *path; | |
a36cf8b8 | 2136 | struct btrfs_root *root = sctx->dev_root; |
a2de733c AJ |
2137 | struct btrfs_fs_info *fs_info = root->fs_info; |
2138 | u64 length; | |
2139 | u64 chunk_tree; | |
2140 | u64 chunk_objectid; | |
2141 | u64 chunk_offset; | |
2142 | int ret; | |
2143 | int slot; | |
2144 | struct extent_buffer *l; | |
2145 | struct btrfs_key key; | |
2146 | struct btrfs_key found_key; | |
2147 | struct btrfs_block_group_cache *cache; | |
2148 | ||
2149 | path = btrfs_alloc_path(); | |
2150 | if (!path) | |
2151 | return -ENOMEM; | |
2152 | ||
2153 | path->reada = 2; | |
2154 | path->search_commit_root = 1; | |
2155 | path->skip_locking = 1; | |
2156 | ||
a36cf8b8 | 2157 | key.objectid = scrub_dev->devid; |
a2de733c AJ |
2158 | key.offset = 0ull; |
2159 | key.type = BTRFS_DEV_EXTENT_KEY; | |
2160 | ||
a2de733c AJ |
2161 | while (1) { |
2162 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2163 | if (ret < 0) | |
8c51032f AJ |
2164 | break; |
2165 | if (ret > 0) { | |
2166 | if (path->slots[0] >= | |
2167 | btrfs_header_nritems(path->nodes[0])) { | |
2168 | ret = btrfs_next_leaf(root, path); | |
2169 | if (ret) | |
2170 | break; | |
2171 | } | |
2172 | } | |
a2de733c AJ |
2173 | |
2174 | l = path->nodes[0]; | |
2175 | slot = path->slots[0]; | |
2176 | ||
2177 | btrfs_item_key_to_cpu(l, &found_key, slot); | |
2178 | ||
a36cf8b8 | 2179 | if (found_key.objectid != scrub_dev->devid) |
a2de733c AJ |
2180 | break; |
2181 | ||
8c51032f | 2182 | if (btrfs_key_type(&found_key) != BTRFS_DEV_EXTENT_KEY) |
a2de733c AJ |
2183 | break; |
2184 | ||
2185 | if (found_key.offset >= end) | |
2186 | break; | |
2187 | ||
2188 | if (found_key.offset < key.offset) | |
2189 | break; | |
2190 | ||
2191 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
2192 | length = btrfs_dev_extent_length(l, dev_extent); | |
2193 | ||
2194 | if (found_key.offset + length <= start) { | |
2195 | key.offset = found_key.offset + length; | |
71267333 | 2196 | btrfs_release_path(path); |
a2de733c AJ |
2197 | continue; |
2198 | } | |
2199 | ||
2200 | chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent); | |
2201 | chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent); | |
2202 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); | |
2203 | ||
2204 | /* | |
2205 | * get a reference on the corresponding block group to prevent | |
2206 | * the chunk from going away while we scrub it | |
2207 | */ | |
2208 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
2209 | if (!cache) { | |
2210 | ret = -ENOENT; | |
8c51032f | 2211 | break; |
a2de733c | 2212 | } |
a36cf8b8 | 2213 | ret = scrub_chunk(sctx, scrub_dev, chunk_tree, chunk_objectid, |
859acaf1 | 2214 | chunk_offset, length, found_key.offset); |
a2de733c AJ |
2215 | btrfs_put_block_group(cache); |
2216 | if (ret) | |
2217 | break; | |
2218 | ||
2219 | key.offset = found_key.offset + length; | |
71267333 | 2220 | btrfs_release_path(path); |
a2de733c AJ |
2221 | } |
2222 | ||
a2de733c | 2223 | btrfs_free_path(path); |
8c51032f AJ |
2224 | |
2225 | /* | |
2226 | * ret can still be 1 from search_slot or next_leaf, | |
2227 | * that's not an error | |
2228 | */ | |
2229 | return ret < 0 ? ret : 0; | |
a2de733c AJ |
2230 | } |
2231 | ||
a36cf8b8 SB |
2232 | static noinline_for_stack int scrub_supers(struct scrub_ctx *sctx, |
2233 | struct btrfs_device *scrub_dev) | |
a2de733c AJ |
2234 | { |
2235 | int i; | |
2236 | u64 bytenr; | |
2237 | u64 gen; | |
2238 | int ret; | |
a36cf8b8 | 2239 | struct btrfs_root *root = sctx->dev_root; |
a2de733c | 2240 | |
79787eaa JM |
2241 | if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) |
2242 | return -EIO; | |
2243 | ||
a2de733c AJ |
2244 | gen = root->fs_info->last_trans_committed; |
2245 | ||
2246 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { | |
2247 | bytenr = btrfs_sb_offset(i); | |
a36cf8b8 | 2248 | if (bytenr + BTRFS_SUPER_INFO_SIZE > scrub_dev->total_bytes) |
a2de733c AJ |
2249 | break; |
2250 | ||
d9d181c1 | 2251 | ret = scrub_pages(sctx, bytenr, BTRFS_SUPER_INFO_SIZE, bytenr, |
a36cf8b8 SB |
2252 | scrub_dev, BTRFS_EXTENT_FLAG_SUPER, gen, i, |
2253 | NULL, 1); | |
a2de733c AJ |
2254 | if (ret) |
2255 | return ret; | |
2256 | } | |
b6bfebc1 | 2257 | wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0); |
a2de733c AJ |
2258 | |
2259 | return 0; | |
2260 | } | |
2261 | ||
2262 | /* | |
2263 | * get a reference count on fs_info->scrub_workers. start worker if necessary | |
2264 | */ | |
2265 | static noinline_for_stack int scrub_workers_get(struct btrfs_root *root) | |
2266 | { | |
2267 | struct btrfs_fs_info *fs_info = root->fs_info; | |
0dc3b84a | 2268 | int ret = 0; |
a2de733c AJ |
2269 | |
2270 | mutex_lock(&fs_info->scrub_lock); | |
632dd772 AJ |
2271 | if (fs_info->scrub_workers_refcnt == 0) { |
2272 | btrfs_init_workers(&fs_info->scrub_workers, "scrub", | |
2273 | fs_info->thread_pool_size, &fs_info->generic_worker); | |
2274 | fs_info->scrub_workers.idle_thresh = 4; | |
0dc3b84a JB |
2275 | ret = btrfs_start_workers(&fs_info->scrub_workers); |
2276 | if (ret) | |
2277 | goto out; | |
632dd772 | 2278 | } |
a2de733c | 2279 | ++fs_info->scrub_workers_refcnt; |
0dc3b84a | 2280 | out: |
a2de733c AJ |
2281 | mutex_unlock(&fs_info->scrub_lock); |
2282 | ||
0dc3b84a | 2283 | return ret; |
a2de733c AJ |
2284 | } |
2285 | ||
2286 | static noinline_for_stack void scrub_workers_put(struct btrfs_root *root) | |
2287 | { | |
2288 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2289 | ||
2290 | mutex_lock(&fs_info->scrub_lock); | |
2291 | if (--fs_info->scrub_workers_refcnt == 0) | |
2292 | btrfs_stop_workers(&fs_info->scrub_workers); | |
2293 | WARN_ON(fs_info->scrub_workers_refcnt < 0); | |
2294 | mutex_unlock(&fs_info->scrub_lock); | |
2295 | } | |
2296 | ||
2297 | ||
2298 | int btrfs_scrub_dev(struct btrfs_root *root, u64 devid, u64 start, u64 end, | |
8628764e | 2299 | struct btrfs_scrub_progress *progress, int readonly) |
a2de733c | 2300 | { |
d9d181c1 | 2301 | struct scrub_ctx *sctx; |
a2de733c AJ |
2302 | struct btrfs_fs_info *fs_info = root->fs_info; |
2303 | int ret; | |
2304 | struct btrfs_device *dev; | |
2305 | ||
7841cb28 | 2306 | if (btrfs_fs_closing(root->fs_info)) |
a2de733c AJ |
2307 | return -EINVAL; |
2308 | ||
2309 | /* | |
2310 | * check some assumptions | |
2311 | */ | |
b5d67f64 SB |
2312 | if (root->nodesize != root->leafsize) { |
2313 | printk(KERN_ERR | |
2314 | "btrfs_scrub: size assumption nodesize == leafsize (%d == %d) fails\n", | |
2315 | root->nodesize, root->leafsize); | |
2316 | return -EINVAL; | |
2317 | } | |
2318 | ||
2319 | if (root->nodesize > BTRFS_STRIPE_LEN) { | |
2320 | /* | |
2321 | * in this case scrub is unable to calculate the checksum | |
2322 | * the way scrub is implemented. Do not handle this | |
2323 | * situation at all because it won't ever happen. | |
2324 | */ | |
2325 | printk(KERN_ERR | |
2326 | "btrfs_scrub: size assumption nodesize <= BTRFS_STRIPE_LEN (%d <= %d) fails\n", | |
2327 | root->nodesize, BTRFS_STRIPE_LEN); | |
2328 | return -EINVAL; | |
2329 | } | |
2330 | ||
2331 | if (root->sectorsize != PAGE_SIZE) { | |
2332 | /* not supported for data w/o checksums */ | |
2333 | printk(KERN_ERR | |
2334 | "btrfs_scrub: size assumption sectorsize != PAGE_SIZE (%d != %lld) fails\n", | |
2335 | root->sectorsize, (unsigned long long)PAGE_SIZE); | |
a2de733c AJ |
2336 | return -EINVAL; |
2337 | } | |
2338 | ||
7a9e9987 SB |
2339 | if (fs_info->chunk_root->nodesize > |
2340 | PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK || | |
2341 | fs_info->chunk_root->sectorsize > | |
2342 | PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK) { | |
2343 | /* | |
2344 | * would exhaust the array bounds of pagev member in | |
2345 | * struct scrub_block | |
2346 | */ | |
2347 | pr_err("btrfs_scrub: size assumption nodesize and sectorsize <= SCRUB_MAX_PAGES_PER_BLOCK (%d <= %d && %d <= %d) fails\n", | |
2348 | fs_info->chunk_root->nodesize, | |
2349 | SCRUB_MAX_PAGES_PER_BLOCK, | |
2350 | fs_info->chunk_root->sectorsize, | |
2351 | SCRUB_MAX_PAGES_PER_BLOCK); | |
2352 | return -EINVAL; | |
2353 | } | |
2354 | ||
a2de733c AJ |
2355 | ret = scrub_workers_get(root); |
2356 | if (ret) | |
2357 | return ret; | |
2358 | ||
2359 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
2360 | dev = btrfs_find_device(root, devid, NULL, NULL); | |
2361 | if (!dev || dev->missing) { | |
2362 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); | |
2363 | scrub_workers_put(root); | |
2364 | return -ENODEV; | |
2365 | } | |
2366 | mutex_lock(&fs_info->scrub_lock); | |
2367 | ||
2368 | if (!dev->in_fs_metadata) { | |
2369 | mutex_unlock(&fs_info->scrub_lock); | |
2370 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); | |
2371 | scrub_workers_put(root); | |
2372 | return -ENODEV; | |
2373 | } | |
2374 | ||
2375 | if (dev->scrub_device) { | |
2376 | mutex_unlock(&fs_info->scrub_lock); | |
2377 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); | |
2378 | scrub_workers_put(root); | |
2379 | return -EINPROGRESS; | |
2380 | } | |
d9d181c1 SB |
2381 | sctx = scrub_setup_ctx(dev); |
2382 | if (IS_ERR(sctx)) { | |
a2de733c AJ |
2383 | mutex_unlock(&fs_info->scrub_lock); |
2384 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); | |
2385 | scrub_workers_put(root); | |
d9d181c1 | 2386 | return PTR_ERR(sctx); |
a2de733c | 2387 | } |
d9d181c1 SB |
2388 | sctx->readonly = readonly; |
2389 | dev->scrub_device = sctx; | |
a2de733c AJ |
2390 | |
2391 | atomic_inc(&fs_info->scrubs_running); | |
2392 | mutex_unlock(&fs_info->scrub_lock); | |
2393 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); | |
2394 | ||
2395 | down_read(&fs_info->scrub_super_lock); | |
a36cf8b8 | 2396 | ret = scrub_supers(sctx, dev); |
a2de733c AJ |
2397 | up_read(&fs_info->scrub_super_lock); |
2398 | ||
2399 | if (!ret) | |
a36cf8b8 | 2400 | ret = scrub_enumerate_chunks(sctx, dev, start, end); |
a2de733c | 2401 | |
b6bfebc1 | 2402 | wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0); |
a2de733c AJ |
2403 | atomic_dec(&fs_info->scrubs_running); |
2404 | wake_up(&fs_info->scrub_pause_wait); | |
2405 | ||
b6bfebc1 | 2406 | wait_event(sctx->list_wait, atomic_read(&sctx->workers_pending) == 0); |
0ef8e451 | 2407 | |
a2de733c | 2408 | if (progress) |
d9d181c1 | 2409 | memcpy(progress, &sctx->stat, sizeof(*progress)); |
a2de733c AJ |
2410 | |
2411 | mutex_lock(&fs_info->scrub_lock); | |
2412 | dev->scrub_device = NULL; | |
2413 | mutex_unlock(&fs_info->scrub_lock); | |
2414 | ||
d9d181c1 | 2415 | scrub_free_ctx(sctx); |
a2de733c AJ |
2416 | scrub_workers_put(root); |
2417 | ||
2418 | return ret; | |
2419 | } | |
2420 | ||
143bede5 | 2421 | void btrfs_scrub_pause(struct btrfs_root *root) |
a2de733c AJ |
2422 | { |
2423 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2424 | ||
2425 | mutex_lock(&fs_info->scrub_lock); | |
2426 | atomic_inc(&fs_info->scrub_pause_req); | |
2427 | while (atomic_read(&fs_info->scrubs_paused) != | |
2428 | atomic_read(&fs_info->scrubs_running)) { | |
2429 | mutex_unlock(&fs_info->scrub_lock); | |
2430 | wait_event(fs_info->scrub_pause_wait, | |
2431 | atomic_read(&fs_info->scrubs_paused) == | |
2432 | atomic_read(&fs_info->scrubs_running)); | |
2433 | mutex_lock(&fs_info->scrub_lock); | |
2434 | } | |
2435 | mutex_unlock(&fs_info->scrub_lock); | |
a2de733c AJ |
2436 | } |
2437 | ||
143bede5 | 2438 | void btrfs_scrub_continue(struct btrfs_root *root) |
a2de733c AJ |
2439 | { |
2440 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2441 | ||
2442 | atomic_dec(&fs_info->scrub_pause_req); | |
2443 | wake_up(&fs_info->scrub_pause_wait); | |
a2de733c AJ |
2444 | } |
2445 | ||
143bede5 | 2446 | void btrfs_scrub_pause_super(struct btrfs_root *root) |
a2de733c AJ |
2447 | { |
2448 | down_write(&root->fs_info->scrub_super_lock); | |
a2de733c AJ |
2449 | } |
2450 | ||
143bede5 | 2451 | void btrfs_scrub_continue_super(struct btrfs_root *root) |
a2de733c AJ |
2452 | { |
2453 | up_write(&root->fs_info->scrub_super_lock); | |
a2de733c AJ |
2454 | } |
2455 | ||
49b25e05 | 2456 | int __btrfs_scrub_cancel(struct btrfs_fs_info *fs_info) |
a2de733c | 2457 | { |
a2de733c AJ |
2458 | |
2459 | mutex_lock(&fs_info->scrub_lock); | |
2460 | if (!atomic_read(&fs_info->scrubs_running)) { | |
2461 | mutex_unlock(&fs_info->scrub_lock); | |
2462 | return -ENOTCONN; | |
2463 | } | |
2464 | ||
2465 | atomic_inc(&fs_info->scrub_cancel_req); | |
2466 | while (atomic_read(&fs_info->scrubs_running)) { | |
2467 | mutex_unlock(&fs_info->scrub_lock); | |
2468 | wait_event(fs_info->scrub_pause_wait, | |
2469 | atomic_read(&fs_info->scrubs_running) == 0); | |
2470 | mutex_lock(&fs_info->scrub_lock); | |
2471 | } | |
2472 | atomic_dec(&fs_info->scrub_cancel_req); | |
2473 | mutex_unlock(&fs_info->scrub_lock); | |
2474 | ||
2475 | return 0; | |
2476 | } | |
2477 | ||
49b25e05 JM |
2478 | int btrfs_scrub_cancel(struct btrfs_root *root) |
2479 | { | |
2480 | return __btrfs_scrub_cancel(root->fs_info); | |
2481 | } | |
2482 | ||
a2de733c AJ |
2483 | int btrfs_scrub_cancel_dev(struct btrfs_root *root, struct btrfs_device *dev) |
2484 | { | |
2485 | struct btrfs_fs_info *fs_info = root->fs_info; | |
d9d181c1 | 2486 | struct scrub_ctx *sctx; |
a2de733c AJ |
2487 | |
2488 | mutex_lock(&fs_info->scrub_lock); | |
d9d181c1 SB |
2489 | sctx = dev->scrub_device; |
2490 | if (!sctx) { | |
a2de733c AJ |
2491 | mutex_unlock(&fs_info->scrub_lock); |
2492 | return -ENOTCONN; | |
2493 | } | |
d9d181c1 | 2494 | atomic_inc(&sctx->cancel_req); |
a2de733c AJ |
2495 | while (dev->scrub_device) { |
2496 | mutex_unlock(&fs_info->scrub_lock); | |
2497 | wait_event(fs_info->scrub_pause_wait, | |
2498 | dev->scrub_device == NULL); | |
2499 | mutex_lock(&fs_info->scrub_lock); | |
2500 | } | |
2501 | mutex_unlock(&fs_info->scrub_lock); | |
2502 | ||
2503 | return 0; | |
2504 | } | |
1623edeb | 2505 | |
a2de733c AJ |
2506 | int btrfs_scrub_cancel_devid(struct btrfs_root *root, u64 devid) |
2507 | { | |
2508 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2509 | struct btrfs_device *dev; | |
2510 | int ret; | |
2511 | ||
2512 | /* | |
2513 | * we have to hold the device_list_mutex here so the device | |
2514 | * does not go away in cancel_dev. FIXME: find a better solution | |
2515 | */ | |
2516 | mutex_lock(&fs_info->fs_devices->device_list_mutex); | |
2517 | dev = btrfs_find_device(root, devid, NULL, NULL); | |
2518 | if (!dev) { | |
2519 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
2520 | return -ENODEV; | |
2521 | } | |
2522 | ret = btrfs_scrub_cancel_dev(root, dev); | |
2523 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
2524 | ||
2525 | return ret; | |
2526 | } | |
2527 | ||
2528 | int btrfs_scrub_progress(struct btrfs_root *root, u64 devid, | |
2529 | struct btrfs_scrub_progress *progress) | |
2530 | { | |
2531 | struct btrfs_device *dev; | |
d9d181c1 | 2532 | struct scrub_ctx *sctx = NULL; |
a2de733c AJ |
2533 | |
2534 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
2535 | dev = btrfs_find_device(root, devid, NULL, NULL); | |
2536 | if (dev) | |
d9d181c1 SB |
2537 | sctx = dev->scrub_device; |
2538 | if (sctx) | |
2539 | memcpy(progress, &sctx->stat, sizeof(*progress)); | |
a2de733c AJ |
2540 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
2541 | ||
d9d181c1 | 2542 | return dev ? (sctx ? 0 : -ENOTCONN) : -ENODEV; |
a2de733c | 2543 | } |