]>
Commit | Line | Data |
---|---|---|
a2de733c AJ |
1 | /* |
2 | * Copyright (C) 2011 STRATO. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
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; | |
107 | atomic_t in_flight; | |
0ef8e451 | 108 | atomic_t fixup_cnt; |
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 SB |
148 | |
149 | static int scrub_handle_errored_block(struct scrub_block *sblock_to_check); | |
d9d181c1 | 150 | static int scrub_setup_recheck_block(struct scrub_ctx *sctx, |
b5d67f64 SB |
151 | struct btrfs_mapping_tree *map_tree, |
152 | u64 length, u64 logical, | |
153 | struct scrub_block *sblock); | |
154 | static int scrub_recheck_block(struct btrfs_fs_info *fs_info, | |
155 | struct scrub_block *sblock, int is_metadata, | |
156 | int have_csum, u8 *csum, u64 generation, | |
157 | u16 csum_size); | |
158 | static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info, | |
159 | struct scrub_block *sblock, | |
160 | int is_metadata, int have_csum, | |
161 | const u8 *csum, u64 generation, | |
162 | u16 csum_size); | |
163 | static void scrub_complete_bio_end_io(struct bio *bio, int err); | |
164 | static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad, | |
165 | struct scrub_block *sblock_good, | |
166 | int force_write); | |
167 | static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad, | |
168 | struct scrub_block *sblock_good, | |
169 | int page_num, int force_write); | |
170 | static int scrub_checksum_data(struct scrub_block *sblock); | |
171 | static int scrub_checksum_tree_block(struct scrub_block *sblock); | |
172 | static int scrub_checksum_super(struct scrub_block *sblock); | |
173 | static void scrub_block_get(struct scrub_block *sblock); | |
174 | static void scrub_block_put(struct scrub_block *sblock); | |
7a9e9987 SB |
175 | static void scrub_page_get(struct scrub_page *spage); |
176 | static void scrub_page_put(struct scrub_page *spage); | |
d9d181c1 | 177 | static int scrub_add_page_to_bio(struct scrub_ctx *sctx, |
b5d67f64 | 178 | struct scrub_page *spage); |
d9d181c1 | 179 | static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len, |
a36cf8b8 SB |
180 | u64 physical, struct btrfs_device *dev, u64 flags, |
181 | u64 gen, int mirror_num, u8 *csum, int force); | |
1623edeb | 182 | static void scrub_bio_end_io(struct bio *bio, int err); |
b5d67f64 SB |
183 | static void scrub_bio_end_io_worker(struct btrfs_work *work); |
184 | static void scrub_block_complete(struct scrub_block *sblock); | |
1623edeb SB |
185 | |
186 | ||
d9d181c1 | 187 | static void scrub_free_csums(struct scrub_ctx *sctx) |
a2de733c | 188 | { |
d9d181c1 | 189 | while (!list_empty(&sctx->csum_list)) { |
a2de733c | 190 | struct btrfs_ordered_sum *sum; |
d9d181c1 | 191 | sum = list_first_entry(&sctx->csum_list, |
a2de733c AJ |
192 | struct btrfs_ordered_sum, list); |
193 | list_del(&sum->list); | |
194 | kfree(sum); | |
195 | } | |
196 | } | |
197 | ||
d9d181c1 | 198 | static noinline_for_stack void scrub_free_ctx(struct scrub_ctx *sctx) |
a2de733c AJ |
199 | { |
200 | int i; | |
a2de733c | 201 | |
d9d181c1 | 202 | if (!sctx) |
a2de733c AJ |
203 | return; |
204 | ||
b5d67f64 | 205 | /* this can happen when scrub is cancelled */ |
d9d181c1 SB |
206 | if (sctx->curr != -1) { |
207 | struct scrub_bio *sbio = sctx->bios[sctx->curr]; | |
b5d67f64 SB |
208 | |
209 | for (i = 0; i < sbio->page_count; i++) { | |
210 | BUG_ON(!sbio->pagev[i]); | |
211 | BUG_ON(!sbio->pagev[i]->page); | |
212 | scrub_block_put(sbio->pagev[i]->sblock); | |
213 | } | |
214 | bio_put(sbio->bio); | |
215 | } | |
216 | ||
d9d181c1 SB |
217 | for (i = 0; i < SCRUB_BIOS_PER_CTX; ++i) { |
218 | struct scrub_bio *sbio = sctx->bios[i]; | |
a2de733c AJ |
219 | |
220 | if (!sbio) | |
221 | break; | |
a2de733c AJ |
222 | kfree(sbio); |
223 | } | |
224 | ||
d9d181c1 SB |
225 | scrub_free_csums(sctx); |
226 | kfree(sctx); | |
a2de733c AJ |
227 | } |
228 | ||
229 | static noinline_for_stack | |
d9d181c1 | 230 | struct scrub_ctx *scrub_setup_ctx(struct btrfs_device *dev) |
a2de733c | 231 | { |
d9d181c1 | 232 | struct scrub_ctx *sctx; |
a2de733c | 233 | int i; |
a2de733c | 234 | struct btrfs_fs_info *fs_info = dev->dev_root->fs_info; |
b5d67f64 | 235 | int pages_per_bio; |
a2de733c | 236 | |
b5d67f64 SB |
237 | pages_per_bio = min_t(int, SCRUB_PAGES_PER_BIO, |
238 | bio_get_nr_vecs(dev->bdev)); | |
d9d181c1 SB |
239 | sctx = kzalloc(sizeof(*sctx), GFP_NOFS); |
240 | if (!sctx) | |
a2de733c | 241 | goto nomem; |
d9d181c1 SB |
242 | sctx->pages_per_bio = pages_per_bio; |
243 | sctx->curr = -1; | |
a36cf8b8 | 244 | sctx->dev_root = dev->dev_root; |
d9d181c1 | 245 | for (i = 0; i < SCRUB_BIOS_PER_CTX; ++i) { |
a2de733c AJ |
246 | struct scrub_bio *sbio; |
247 | ||
248 | sbio = kzalloc(sizeof(*sbio), GFP_NOFS); | |
249 | if (!sbio) | |
250 | goto nomem; | |
d9d181c1 | 251 | sctx->bios[i] = sbio; |
a2de733c | 252 | |
a2de733c | 253 | sbio->index = i; |
d9d181c1 | 254 | sbio->sctx = sctx; |
b5d67f64 SB |
255 | sbio->page_count = 0; |
256 | sbio->work.func = scrub_bio_end_io_worker; | |
a2de733c | 257 | |
d9d181c1 SB |
258 | if (i != SCRUB_BIOS_PER_CTX - 1) |
259 | sctx->bios[i]->next_free = i + 1; | |
0ef8e451 | 260 | else |
d9d181c1 SB |
261 | sctx->bios[i]->next_free = -1; |
262 | } | |
263 | sctx->first_free = 0; | |
264 | sctx->nodesize = dev->dev_root->nodesize; | |
265 | sctx->leafsize = dev->dev_root->leafsize; | |
266 | sctx->sectorsize = dev->dev_root->sectorsize; | |
267 | atomic_set(&sctx->in_flight, 0); | |
268 | atomic_set(&sctx->fixup_cnt, 0); | |
269 | atomic_set(&sctx->cancel_req, 0); | |
270 | sctx->csum_size = btrfs_super_csum_size(fs_info->super_copy); | |
271 | INIT_LIST_HEAD(&sctx->csum_list); | |
272 | ||
273 | spin_lock_init(&sctx->list_lock); | |
274 | spin_lock_init(&sctx->stat_lock); | |
275 | init_waitqueue_head(&sctx->list_wait); | |
276 | return sctx; | |
a2de733c AJ |
277 | |
278 | nomem: | |
d9d181c1 | 279 | scrub_free_ctx(sctx); |
a2de733c AJ |
280 | return ERR_PTR(-ENOMEM); |
281 | } | |
282 | ||
558540c1 JS |
283 | static int scrub_print_warning_inode(u64 inum, u64 offset, u64 root, void *ctx) |
284 | { | |
285 | u64 isize; | |
286 | u32 nlink; | |
287 | int ret; | |
288 | int i; | |
289 | struct extent_buffer *eb; | |
290 | struct btrfs_inode_item *inode_item; | |
291 | struct scrub_warning *swarn = ctx; | |
292 | struct btrfs_fs_info *fs_info = swarn->dev->dev_root->fs_info; | |
293 | struct inode_fs_paths *ipath = NULL; | |
294 | struct btrfs_root *local_root; | |
295 | struct btrfs_key root_key; | |
296 | ||
297 | root_key.objectid = root; | |
298 | root_key.type = BTRFS_ROOT_ITEM_KEY; | |
299 | root_key.offset = (u64)-1; | |
300 | local_root = btrfs_read_fs_root_no_name(fs_info, &root_key); | |
301 | if (IS_ERR(local_root)) { | |
302 | ret = PTR_ERR(local_root); | |
303 | goto err; | |
304 | } | |
305 | ||
306 | ret = inode_item_info(inum, 0, local_root, swarn->path); | |
307 | if (ret) { | |
308 | btrfs_release_path(swarn->path); | |
309 | goto err; | |
310 | } | |
311 | ||
312 | eb = swarn->path->nodes[0]; | |
313 | inode_item = btrfs_item_ptr(eb, swarn->path->slots[0], | |
314 | struct btrfs_inode_item); | |
315 | isize = btrfs_inode_size(eb, inode_item); | |
316 | nlink = btrfs_inode_nlink(eb, inode_item); | |
317 | btrfs_release_path(swarn->path); | |
318 | ||
319 | ipath = init_ipath(4096, local_root, swarn->path); | |
26bdef54 DC |
320 | if (IS_ERR(ipath)) { |
321 | ret = PTR_ERR(ipath); | |
322 | ipath = NULL; | |
323 | goto err; | |
324 | } | |
558540c1 JS |
325 | ret = paths_from_inode(inum, ipath); |
326 | ||
327 | if (ret < 0) | |
328 | goto err; | |
329 | ||
330 | /* | |
331 | * we deliberately ignore the bit ipath might have been too small to | |
332 | * hold all of the paths here | |
333 | */ | |
334 | for (i = 0; i < ipath->fspath->elem_cnt; ++i) | |
606686ee | 335 | printk_in_rcu(KERN_WARNING "btrfs: %s at logical %llu on dev " |
558540c1 JS |
336 | "%s, sector %llu, root %llu, inode %llu, offset %llu, " |
337 | "length %llu, links %u (path: %s)\n", swarn->errstr, | |
606686ee | 338 | swarn->logical, rcu_str_deref(swarn->dev->name), |
558540c1 JS |
339 | (unsigned long long)swarn->sector, root, inum, offset, |
340 | min(isize - offset, (u64)PAGE_SIZE), nlink, | |
745c4d8e | 341 | (char *)(unsigned long)ipath->fspath->val[i]); |
558540c1 JS |
342 | |
343 | free_ipath(ipath); | |
344 | return 0; | |
345 | ||
346 | err: | |
606686ee | 347 | printk_in_rcu(KERN_WARNING "btrfs: %s at logical %llu on dev " |
558540c1 JS |
348 | "%s, sector %llu, root %llu, inode %llu, offset %llu: path " |
349 | "resolving failed with ret=%d\n", swarn->errstr, | |
606686ee | 350 | swarn->logical, rcu_str_deref(swarn->dev->name), |
558540c1 JS |
351 | (unsigned long long)swarn->sector, root, inum, offset, ret); |
352 | ||
353 | free_ipath(ipath); | |
354 | return 0; | |
355 | } | |
356 | ||
b5d67f64 | 357 | static void scrub_print_warning(const char *errstr, struct scrub_block *sblock) |
558540c1 | 358 | { |
a36cf8b8 SB |
359 | struct btrfs_device *dev; |
360 | struct btrfs_fs_info *fs_info; | |
558540c1 JS |
361 | struct btrfs_path *path; |
362 | struct btrfs_key found_key; | |
363 | struct extent_buffer *eb; | |
364 | struct btrfs_extent_item *ei; | |
365 | struct scrub_warning swarn; | |
69917e43 LB |
366 | unsigned long ptr = 0; |
367 | u64 extent_item_pos; | |
368 | u64 flags = 0; | |
558540c1 | 369 | u64 ref_root; |
69917e43 | 370 | u32 item_size; |
558540c1 | 371 | u8 ref_level; |
558540c1 | 372 | const int bufsize = 4096; |
69917e43 | 373 | int ret; |
558540c1 | 374 | |
a36cf8b8 | 375 | WARN_ON(sblock->page_count < 1); |
7a9e9987 | 376 | dev = sblock->pagev[0]->dev; |
a36cf8b8 SB |
377 | fs_info = sblock->sctx->dev_root->fs_info; |
378 | ||
558540c1 JS |
379 | path = btrfs_alloc_path(); |
380 | ||
381 | swarn.scratch_buf = kmalloc(bufsize, GFP_NOFS); | |
382 | swarn.msg_buf = kmalloc(bufsize, GFP_NOFS); | |
7a9e9987 SB |
383 | swarn.sector = (sblock->pagev[0]->physical) >> 9; |
384 | swarn.logical = sblock->pagev[0]->logical; | |
558540c1 | 385 | swarn.errstr = errstr; |
a36cf8b8 | 386 | swarn.dev = NULL; |
558540c1 JS |
387 | swarn.msg_bufsize = bufsize; |
388 | swarn.scratch_bufsize = bufsize; | |
389 | ||
390 | if (!path || !swarn.scratch_buf || !swarn.msg_buf) | |
391 | goto out; | |
392 | ||
69917e43 LB |
393 | ret = extent_from_logical(fs_info, swarn.logical, path, &found_key, |
394 | &flags); | |
558540c1 JS |
395 | if (ret < 0) |
396 | goto out; | |
397 | ||
4692cf58 | 398 | extent_item_pos = swarn.logical - found_key.objectid; |
558540c1 JS |
399 | swarn.extent_item_size = found_key.offset; |
400 | ||
401 | eb = path->nodes[0]; | |
402 | ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item); | |
403 | item_size = btrfs_item_size_nr(eb, path->slots[0]); | |
4692cf58 | 404 | btrfs_release_path(path); |
558540c1 | 405 | |
69917e43 | 406 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { |
558540c1 JS |
407 | do { |
408 | ret = tree_backref_for_extent(&ptr, eb, ei, item_size, | |
409 | &ref_root, &ref_level); | |
606686ee | 410 | printk_in_rcu(KERN_WARNING |
1623edeb | 411 | "btrfs: %s at logical %llu on dev %s, " |
558540c1 | 412 | "sector %llu: metadata %s (level %d) in tree " |
606686ee JB |
413 | "%llu\n", errstr, swarn.logical, |
414 | rcu_str_deref(dev->name), | |
558540c1 JS |
415 | (unsigned long long)swarn.sector, |
416 | ref_level ? "node" : "leaf", | |
417 | ret < 0 ? -1 : ref_level, | |
418 | ret < 0 ? -1 : ref_root); | |
419 | } while (ret != 1); | |
420 | } else { | |
421 | swarn.path = path; | |
a36cf8b8 | 422 | swarn.dev = dev; |
7a3ae2f8 JS |
423 | iterate_extent_inodes(fs_info, found_key.objectid, |
424 | extent_item_pos, 1, | |
558540c1 JS |
425 | scrub_print_warning_inode, &swarn); |
426 | } | |
427 | ||
428 | out: | |
429 | btrfs_free_path(path); | |
430 | kfree(swarn.scratch_buf); | |
431 | kfree(swarn.msg_buf); | |
432 | } | |
433 | ||
0ef8e451 JS |
434 | static int scrub_fixup_readpage(u64 inum, u64 offset, u64 root, void *ctx) |
435 | { | |
5da6fcbc | 436 | struct page *page = NULL; |
0ef8e451 JS |
437 | unsigned long index; |
438 | struct scrub_fixup_nodatasum *fixup = ctx; | |
439 | int ret; | |
5da6fcbc | 440 | int corrected = 0; |
0ef8e451 | 441 | struct btrfs_key key; |
5da6fcbc | 442 | struct inode *inode = NULL; |
0ef8e451 JS |
443 | u64 end = offset + PAGE_SIZE - 1; |
444 | struct btrfs_root *local_root; | |
445 | ||
446 | key.objectid = root; | |
447 | key.type = BTRFS_ROOT_ITEM_KEY; | |
448 | key.offset = (u64)-1; | |
449 | local_root = btrfs_read_fs_root_no_name(fixup->root->fs_info, &key); | |
450 | if (IS_ERR(local_root)) | |
451 | return PTR_ERR(local_root); | |
452 | ||
453 | key.type = BTRFS_INODE_ITEM_KEY; | |
454 | key.objectid = inum; | |
455 | key.offset = 0; | |
456 | inode = btrfs_iget(fixup->root->fs_info->sb, &key, local_root, NULL); | |
457 | if (IS_ERR(inode)) | |
458 | return PTR_ERR(inode); | |
459 | ||
0ef8e451 JS |
460 | index = offset >> PAGE_CACHE_SHIFT; |
461 | ||
462 | page = find_or_create_page(inode->i_mapping, index, GFP_NOFS); | |
5da6fcbc JS |
463 | if (!page) { |
464 | ret = -ENOMEM; | |
465 | goto out; | |
466 | } | |
467 | ||
468 | if (PageUptodate(page)) { | |
469 | struct btrfs_mapping_tree *map_tree; | |
470 | if (PageDirty(page)) { | |
471 | /* | |
472 | * we need to write the data to the defect sector. the | |
473 | * data that was in that sector is not in memory, | |
474 | * because the page was modified. we must not write the | |
475 | * modified page to that sector. | |
476 | * | |
477 | * TODO: what could be done here: wait for the delalloc | |
478 | * runner to write out that page (might involve | |
479 | * COW) and see whether the sector is still | |
480 | * referenced afterwards. | |
481 | * | |
482 | * For the meantime, we'll treat this error | |
483 | * incorrectable, although there is a chance that a | |
484 | * later scrub will find the bad sector again and that | |
485 | * there's no dirty page in memory, then. | |
486 | */ | |
487 | ret = -EIO; | |
488 | goto out; | |
489 | } | |
490 | map_tree = &BTRFS_I(inode)->root->fs_info->mapping_tree; | |
491 | ret = repair_io_failure(map_tree, offset, PAGE_SIZE, | |
492 | fixup->logical, page, | |
493 | fixup->mirror_num); | |
494 | unlock_page(page); | |
495 | corrected = !ret; | |
496 | } else { | |
497 | /* | |
498 | * we need to get good data first. the general readpage path | |
499 | * will call repair_io_failure for us, we just have to make | |
500 | * sure we read the bad mirror. | |
501 | */ | |
502 | ret = set_extent_bits(&BTRFS_I(inode)->io_tree, offset, end, | |
503 | EXTENT_DAMAGED, GFP_NOFS); | |
504 | if (ret) { | |
505 | /* set_extent_bits should give proper error */ | |
506 | WARN_ON(ret > 0); | |
507 | if (ret > 0) | |
508 | ret = -EFAULT; | |
509 | goto out; | |
510 | } | |
511 | ||
512 | ret = extent_read_full_page(&BTRFS_I(inode)->io_tree, page, | |
513 | btrfs_get_extent, | |
514 | fixup->mirror_num); | |
515 | wait_on_page_locked(page); | |
516 | ||
517 | corrected = !test_range_bit(&BTRFS_I(inode)->io_tree, offset, | |
518 | end, EXTENT_DAMAGED, 0, NULL); | |
519 | if (!corrected) | |
520 | clear_extent_bits(&BTRFS_I(inode)->io_tree, offset, end, | |
521 | EXTENT_DAMAGED, GFP_NOFS); | |
522 | } | |
523 | ||
524 | out: | |
525 | if (page) | |
526 | put_page(page); | |
527 | if (inode) | |
528 | iput(inode); | |
0ef8e451 JS |
529 | |
530 | if (ret < 0) | |
531 | return ret; | |
532 | ||
533 | if (ret == 0 && corrected) { | |
534 | /* | |
535 | * we only need to call readpage for one of the inodes belonging | |
536 | * to this extent. so make iterate_extent_inodes stop | |
537 | */ | |
538 | return 1; | |
539 | } | |
540 | ||
541 | return -EIO; | |
542 | } | |
543 | ||
544 | static void scrub_fixup_nodatasum(struct btrfs_work *work) | |
545 | { | |
546 | int ret; | |
547 | struct scrub_fixup_nodatasum *fixup; | |
d9d181c1 | 548 | struct scrub_ctx *sctx; |
0ef8e451 JS |
549 | struct btrfs_trans_handle *trans = NULL; |
550 | struct btrfs_fs_info *fs_info; | |
551 | struct btrfs_path *path; | |
552 | int uncorrectable = 0; | |
553 | ||
554 | fixup = container_of(work, struct scrub_fixup_nodatasum, work); | |
d9d181c1 | 555 | sctx = fixup->sctx; |
0ef8e451 JS |
556 | fs_info = fixup->root->fs_info; |
557 | ||
558 | path = btrfs_alloc_path(); | |
559 | if (!path) { | |
d9d181c1 SB |
560 | spin_lock(&sctx->stat_lock); |
561 | ++sctx->stat.malloc_errors; | |
562 | spin_unlock(&sctx->stat_lock); | |
0ef8e451 JS |
563 | uncorrectable = 1; |
564 | goto out; | |
565 | } | |
566 | ||
567 | trans = btrfs_join_transaction(fixup->root); | |
568 | if (IS_ERR(trans)) { | |
569 | uncorrectable = 1; | |
570 | goto out; | |
571 | } | |
572 | ||
573 | /* | |
574 | * the idea is to trigger a regular read through the standard path. we | |
575 | * read a page from the (failed) logical address by specifying the | |
576 | * corresponding copynum of the failed sector. thus, that readpage is | |
577 | * expected to fail. | |
578 | * that is the point where on-the-fly error correction will kick in | |
579 | * (once it's finished) and rewrite the failed sector if a good copy | |
580 | * can be found. | |
581 | */ | |
582 | ret = iterate_inodes_from_logical(fixup->logical, fixup->root->fs_info, | |
583 | path, scrub_fixup_readpage, | |
584 | fixup); | |
585 | if (ret < 0) { | |
586 | uncorrectable = 1; | |
587 | goto out; | |
588 | } | |
589 | WARN_ON(ret != 1); | |
590 | ||
d9d181c1 SB |
591 | spin_lock(&sctx->stat_lock); |
592 | ++sctx->stat.corrected_errors; | |
593 | spin_unlock(&sctx->stat_lock); | |
0ef8e451 JS |
594 | |
595 | out: | |
596 | if (trans && !IS_ERR(trans)) | |
597 | btrfs_end_transaction(trans, fixup->root); | |
598 | if (uncorrectable) { | |
d9d181c1 SB |
599 | spin_lock(&sctx->stat_lock); |
600 | ++sctx->stat.uncorrectable_errors; | |
601 | spin_unlock(&sctx->stat_lock); | |
606686ee JB |
602 | |
603 | printk_ratelimited_in_rcu(KERN_ERR | |
b5d67f64 | 604 | "btrfs: unable to fixup (nodatasum) error at logical %llu on dev %s\n", |
606686ee | 605 | (unsigned long long)fixup->logical, |
a36cf8b8 | 606 | rcu_str_deref(fixup->dev->name)); |
0ef8e451 JS |
607 | } |
608 | ||
609 | btrfs_free_path(path); | |
610 | kfree(fixup); | |
611 | ||
612 | /* see caller why we're pretending to be paused in the scrub counters */ | |
613 | mutex_lock(&fs_info->scrub_lock); | |
614 | atomic_dec(&fs_info->scrubs_running); | |
615 | atomic_dec(&fs_info->scrubs_paused); | |
616 | mutex_unlock(&fs_info->scrub_lock); | |
d9d181c1 | 617 | atomic_dec(&sctx->fixup_cnt); |
0ef8e451 | 618 | wake_up(&fs_info->scrub_pause_wait); |
d9d181c1 | 619 | wake_up(&sctx->list_wait); |
0ef8e451 JS |
620 | } |
621 | ||
a2de733c | 622 | /* |
b5d67f64 SB |
623 | * scrub_handle_errored_block gets called when either verification of the |
624 | * pages failed or the bio failed to read, e.g. with EIO. In the latter | |
625 | * case, this function handles all pages in the bio, even though only one | |
626 | * may be bad. | |
627 | * The goal of this function is to repair the errored block by using the | |
628 | * contents of one of the mirrors. | |
a2de733c | 629 | */ |
b5d67f64 | 630 | static int scrub_handle_errored_block(struct scrub_block *sblock_to_check) |
a2de733c | 631 | { |
d9d181c1 | 632 | struct scrub_ctx *sctx = sblock_to_check->sctx; |
a36cf8b8 | 633 | struct btrfs_device *dev; |
b5d67f64 SB |
634 | struct btrfs_fs_info *fs_info; |
635 | u64 length; | |
636 | u64 logical; | |
637 | u64 generation; | |
638 | unsigned int failed_mirror_index; | |
639 | unsigned int is_metadata; | |
640 | unsigned int have_csum; | |
641 | u8 *csum; | |
642 | struct scrub_block *sblocks_for_recheck; /* holds one for each mirror */ | |
643 | struct scrub_block *sblock_bad; | |
644 | int ret; | |
645 | int mirror_index; | |
646 | int page_num; | |
647 | int success; | |
558540c1 | 648 | static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL, |
b5d67f64 SB |
649 | DEFAULT_RATELIMIT_BURST); |
650 | ||
651 | BUG_ON(sblock_to_check->page_count < 1); | |
a36cf8b8 | 652 | fs_info = sctx->dev_root->fs_info; |
b5d67f64 | 653 | length = sblock_to_check->page_count * PAGE_SIZE; |
7a9e9987 SB |
654 | logical = sblock_to_check->pagev[0]->logical; |
655 | generation = sblock_to_check->pagev[0]->generation; | |
656 | BUG_ON(sblock_to_check->pagev[0]->mirror_num < 1); | |
657 | failed_mirror_index = sblock_to_check->pagev[0]->mirror_num - 1; | |
658 | is_metadata = !(sblock_to_check->pagev[0]->flags & | |
b5d67f64 | 659 | BTRFS_EXTENT_FLAG_DATA); |
7a9e9987 SB |
660 | have_csum = sblock_to_check->pagev[0]->have_csum; |
661 | csum = sblock_to_check->pagev[0]->csum; | |
662 | dev = sblock_to_check->pagev[0]->dev; | |
13db62b7 | 663 | |
b5d67f64 SB |
664 | /* |
665 | * read all mirrors one after the other. This includes to | |
666 | * re-read the extent or metadata block that failed (that was | |
667 | * the cause that this fixup code is called) another time, | |
668 | * page by page this time in order to know which pages | |
669 | * caused I/O errors and which ones are good (for all mirrors). | |
670 | * It is the goal to handle the situation when more than one | |
671 | * mirror contains I/O errors, but the errors do not | |
672 | * overlap, i.e. the data can be repaired by selecting the | |
673 | * pages from those mirrors without I/O error on the | |
674 | * particular pages. One example (with blocks >= 2 * PAGE_SIZE) | |
675 | * would be that mirror #1 has an I/O error on the first page, | |
676 | * the second page is good, and mirror #2 has an I/O error on | |
677 | * the second page, but the first page is good. | |
678 | * Then the first page of the first mirror can be repaired by | |
679 | * taking the first page of the second mirror, and the | |
680 | * second page of the second mirror can be repaired by | |
681 | * copying the contents of the 2nd page of the 1st mirror. | |
682 | * One more note: if the pages of one mirror contain I/O | |
683 | * errors, the checksum cannot be verified. In order to get | |
684 | * the best data for repairing, the first attempt is to find | |
685 | * a mirror without I/O errors and with a validated checksum. | |
686 | * Only if this is not possible, the pages are picked from | |
687 | * mirrors with I/O errors without considering the checksum. | |
688 | * If the latter is the case, at the end, the checksum of the | |
689 | * repaired area is verified in order to correctly maintain | |
690 | * the statistics. | |
691 | */ | |
692 | ||
693 | sblocks_for_recheck = kzalloc(BTRFS_MAX_MIRRORS * | |
694 | sizeof(*sblocks_for_recheck), | |
695 | GFP_NOFS); | |
696 | if (!sblocks_for_recheck) { | |
d9d181c1 SB |
697 | spin_lock(&sctx->stat_lock); |
698 | sctx->stat.malloc_errors++; | |
699 | sctx->stat.read_errors++; | |
700 | sctx->stat.uncorrectable_errors++; | |
701 | spin_unlock(&sctx->stat_lock); | |
a36cf8b8 | 702 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS); |
b5d67f64 | 703 | goto out; |
a2de733c AJ |
704 | } |
705 | ||
b5d67f64 | 706 | /* setup the context, map the logical blocks and alloc the pages */ |
d9d181c1 | 707 | ret = scrub_setup_recheck_block(sctx, &fs_info->mapping_tree, length, |
b5d67f64 SB |
708 | logical, sblocks_for_recheck); |
709 | if (ret) { | |
d9d181c1 SB |
710 | spin_lock(&sctx->stat_lock); |
711 | sctx->stat.read_errors++; | |
712 | sctx->stat.uncorrectable_errors++; | |
713 | spin_unlock(&sctx->stat_lock); | |
a36cf8b8 | 714 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS); |
b5d67f64 SB |
715 | goto out; |
716 | } | |
717 | BUG_ON(failed_mirror_index >= BTRFS_MAX_MIRRORS); | |
718 | sblock_bad = sblocks_for_recheck + failed_mirror_index; | |
13db62b7 | 719 | |
b5d67f64 SB |
720 | /* build and submit the bios for the failed mirror, check checksums */ |
721 | ret = scrub_recheck_block(fs_info, sblock_bad, is_metadata, have_csum, | |
d9d181c1 | 722 | csum, generation, sctx->csum_size); |
b5d67f64 | 723 | if (ret) { |
d9d181c1 SB |
724 | spin_lock(&sctx->stat_lock); |
725 | sctx->stat.read_errors++; | |
726 | sctx->stat.uncorrectable_errors++; | |
727 | spin_unlock(&sctx->stat_lock); | |
a36cf8b8 | 728 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS); |
b5d67f64 SB |
729 | goto out; |
730 | } | |
a2de733c | 731 | |
b5d67f64 SB |
732 | if (!sblock_bad->header_error && !sblock_bad->checksum_error && |
733 | sblock_bad->no_io_error_seen) { | |
734 | /* | |
735 | * the error disappeared after reading page by page, or | |
736 | * the area was part of a huge bio and other parts of the | |
737 | * bio caused I/O errors, or the block layer merged several | |
738 | * read requests into one and the error is caused by a | |
739 | * different bio (usually one of the two latter cases is | |
740 | * the cause) | |
741 | */ | |
d9d181c1 SB |
742 | spin_lock(&sctx->stat_lock); |
743 | sctx->stat.unverified_errors++; | |
744 | spin_unlock(&sctx->stat_lock); | |
a2de733c | 745 | |
b5d67f64 | 746 | goto out; |
a2de733c | 747 | } |
a2de733c | 748 | |
b5d67f64 | 749 | if (!sblock_bad->no_io_error_seen) { |
d9d181c1 SB |
750 | spin_lock(&sctx->stat_lock); |
751 | sctx->stat.read_errors++; | |
752 | spin_unlock(&sctx->stat_lock); | |
b5d67f64 SB |
753 | if (__ratelimit(&_rs)) |
754 | scrub_print_warning("i/o error", sblock_to_check); | |
a36cf8b8 | 755 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS); |
b5d67f64 | 756 | } else if (sblock_bad->checksum_error) { |
d9d181c1 SB |
757 | spin_lock(&sctx->stat_lock); |
758 | sctx->stat.csum_errors++; | |
759 | spin_unlock(&sctx->stat_lock); | |
b5d67f64 SB |
760 | if (__ratelimit(&_rs)) |
761 | scrub_print_warning("checksum error", sblock_to_check); | |
a36cf8b8 | 762 | btrfs_dev_stat_inc_and_print(dev, |
442a4f63 | 763 | BTRFS_DEV_STAT_CORRUPTION_ERRS); |
b5d67f64 | 764 | } else if (sblock_bad->header_error) { |
d9d181c1 SB |
765 | spin_lock(&sctx->stat_lock); |
766 | sctx->stat.verify_errors++; | |
767 | spin_unlock(&sctx->stat_lock); | |
b5d67f64 SB |
768 | if (__ratelimit(&_rs)) |
769 | scrub_print_warning("checksum/header error", | |
770 | sblock_to_check); | |
442a4f63 | 771 | if (sblock_bad->generation_error) |
a36cf8b8 | 772 | btrfs_dev_stat_inc_and_print(dev, |
442a4f63 SB |
773 | BTRFS_DEV_STAT_GENERATION_ERRS); |
774 | else | |
a36cf8b8 | 775 | btrfs_dev_stat_inc_and_print(dev, |
442a4f63 | 776 | BTRFS_DEV_STAT_CORRUPTION_ERRS); |
b5d67f64 | 777 | } |
a2de733c | 778 | |
d9d181c1 | 779 | if (sctx->readonly) |
b5d67f64 | 780 | goto did_not_correct_error; |
a2de733c | 781 | |
b5d67f64 SB |
782 | if (!is_metadata && !have_csum) { |
783 | struct scrub_fixup_nodatasum *fixup_nodatasum; | |
a2de733c | 784 | |
b5d67f64 SB |
785 | /* |
786 | * !is_metadata and !have_csum, this means that the data | |
787 | * might not be COW'ed, that it might be modified | |
788 | * concurrently. The general strategy to work on the | |
789 | * commit root does not help in the case when COW is not | |
790 | * used. | |
791 | */ | |
792 | fixup_nodatasum = kzalloc(sizeof(*fixup_nodatasum), GFP_NOFS); | |
793 | if (!fixup_nodatasum) | |
794 | goto did_not_correct_error; | |
d9d181c1 | 795 | fixup_nodatasum->sctx = sctx; |
a36cf8b8 | 796 | fixup_nodatasum->dev = dev; |
b5d67f64 SB |
797 | fixup_nodatasum->logical = logical; |
798 | fixup_nodatasum->root = fs_info->extent_root; | |
799 | fixup_nodatasum->mirror_num = failed_mirror_index + 1; | |
a2de733c | 800 | /* |
0ef8e451 JS |
801 | * increment scrubs_running to prevent cancel requests from |
802 | * completing as long as a fixup worker is running. we must also | |
803 | * increment scrubs_paused to prevent deadlocking on pause | |
804 | * requests used for transactions commits (as the worker uses a | |
805 | * transaction context). it is safe to regard the fixup worker | |
806 | * as paused for all matters practical. effectively, we only | |
807 | * avoid cancellation requests from completing. | |
a2de733c | 808 | */ |
0ef8e451 JS |
809 | mutex_lock(&fs_info->scrub_lock); |
810 | atomic_inc(&fs_info->scrubs_running); | |
811 | atomic_inc(&fs_info->scrubs_paused); | |
812 | mutex_unlock(&fs_info->scrub_lock); | |
d9d181c1 | 813 | atomic_inc(&sctx->fixup_cnt); |
b5d67f64 SB |
814 | fixup_nodatasum->work.func = scrub_fixup_nodatasum; |
815 | btrfs_queue_worker(&fs_info->scrub_workers, | |
816 | &fixup_nodatasum->work); | |
817 | goto out; | |
a2de733c AJ |
818 | } |
819 | ||
b5d67f64 SB |
820 | /* |
821 | * now build and submit the bios for the other mirrors, check | |
822 | * checksums | |
823 | */ | |
824 | for (mirror_index = 0; | |
825 | mirror_index < BTRFS_MAX_MIRRORS && | |
826 | sblocks_for_recheck[mirror_index].page_count > 0; | |
827 | mirror_index++) { | |
828 | if (mirror_index == failed_mirror_index) | |
829 | continue; | |
830 | ||
831 | /* build and submit the bios, check checksums */ | |
832 | ret = scrub_recheck_block(fs_info, | |
833 | sblocks_for_recheck + mirror_index, | |
834 | is_metadata, have_csum, csum, | |
d9d181c1 | 835 | generation, sctx->csum_size); |
b5d67f64 SB |
836 | if (ret) |
837 | goto did_not_correct_error; | |
a2de733c AJ |
838 | } |
839 | ||
b5d67f64 SB |
840 | /* |
841 | * first try to pick the mirror which is completely without I/O | |
842 | * errors and also does not have a checksum error. | |
843 | * If one is found, and if a checksum is present, the full block | |
844 | * that is known to contain an error is rewritten. Afterwards | |
845 | * the block is known to be corrected. | |
846 | * If a mirror is found which is completely correct, and no | |
847 | * checksum is present, only those pages are rewritten that had | |
848 | * an I/O error in the block to be repaired, since it cannot be | |
849 | * determined, which copy of the other pages is better (and it | |
850 | * could happen otherwise that a correct page would be | |
851 | * overwritten by a bad one). | |
852 | */ | |
853 | for (mirror_index = 0; | |
854 | mirror_index < BTRFS_MAX_MIRRORS && | |
855 | sblocks_for_recheck[mirror_index].page_count > 0; | |
856 | mirror_index++) { | |
857 | struct scrub_block *sblock_other = sblocks_for_recheck + | |
858 | mirror_index; | |
859 | ||
860 | if (!sblock_other->header_error && | |
861 | !sblock_other->checksum_error && | |
862 | sblock_other->no_io_error_seen) { | |
863 | int force_write = is_metadata || have_csum; | |
864 | ||
865 | ret = scrub_repair_block_from_good_copy(sblock_bad, | |
866 | sblock_other, | |
867 | force_write); | |
868 | if (0 == ret) | |
869 | goto corrected_error; | |
870 | } | |
871 | } | |
a2de733c AJ |
872 | |
873 | /* | |
b5d67f64 SB |
874 | * in case of I/O errors in the area that is supposed to be |
875 | * repaired, continue by picking good copies of those pages. | |
876 | * Select the good pages from mirrors to rewrite bad pages from | |
877 | * the area to fix. Afterwards verify the checksum of the block | |
878 | * that is supposed to be repaired. This verification step is | |
879 | * only done for the purpose of statistic counting and for the | |
880 | * final scrub report, whether errors remain. | |
881 | * A perfect algorithm could make use of the checksum and try | |
882 | * all possible combinations of pages from the different mirrors | |
883 | * until the checksum verification succeeds. For example, when | |
884 | * the 2nd page of mirror #1 faces I/O errors, and the 2nd page | |
885 | * of mirror #2 is readable but the final checksum test fails, | |
886 | * then the 2nd page of mirror #3 could be tried, whether now | |
887 | * the final checksum succeedes. But this would be a rare | |
888 | * exception and is therefore not implemented. At least it is | |
889 | * avoided that the good copy is overwritten. | |
890 | * A more useful improvement would be to pick the sectors | |
891 | * without I/O error based on sector sizes (512 bytes on legacy | |
892 | * disks) instead of on PAGE_SIZE. Then maybe 512 byte of one | |
893 | * mirror could be repaired by taking 512 byte of a different | |
894 | * mirror, even if other 512 byte sectors in the same PAGE_SIZE | |
895 | * area are unreadable. | |
a2de733c | 896 | */ |
a2de733c | 897 | |
b5d67f64 SB |
898 | /* can only fix I/O errors from here on */ |
899 | if (sblock_bad->no_io_error_seen) | |
900 | goto did_not_correct_error; | |
901 | ||
902 | success = 1; | |
903 | for (page_num = 0; page_num < sblock_bad->page_count; page_num++) { | |
7a9e9987 | 904 | struct scrub_page *page_bad = sblock_bad->pagev[page_num]; |
b5d67f64 SB |
905 | |
906 | if (!page_bad->io_error) | |
a2de733c | 907 | continue; |
b5d67f64 SB |
908 | |
909 | for (mirror_index = 0; | |
910 | mirror_index < BTRFS_MAX_MIRRORS && | |
911 | sblocks_for_recheck[mirror_index].page_count > 0; | |
912 | mirror_index++) { | |
913 | struct scrub_block *sblock_other = sblocks_for_recheck + | |
914 | mirror_index; | |
7a9e9987 SB |
915 | struct scrub_page *page_other = sblock_other->pagev[ |
916 | page_num]; | |
b5d67f64 SB |
917 | |
918 | if (!page_other->io_error) { | |
919 | ret = scrub_repair_page_from_good_copy( | |
920 | sblock_bad, sblock_other, page_num, 0); | |
921 | if (0 == ret) { | |
922 | page_bad->io_error = 0; | |
923 | break; /* succeeded for this page */ | |
924 | } | |
925 | } | |
96e36920 | 926 | } |
a2de733c | 927 | |
b5d67f64 SB |
928 | if (page_bad->io_error) { |
929 | /* did not find a mirror to copy the page from */ | |
930 | success = 0; | |
931 | } | |
a2de733c | 932 | } |
a2de733c | 933 | |
b5d67f64 SB |
934 | if (success) { |
935 | if (is_metadata || have_csum) { | |
936 | /* | |
937 | * need to verify the checksum now that all | |
938 | * sectors on disk are repaired (the write | |
939 | * request for data to be repaired is on its way). | |
940 | * Just be lazy and use scrub_recheck_block() | |
941 | * which re-reads the data before the checksum | |
942 | * is verified, but most likely the data comes out | |
943 | * of the page cache. | |
944 | */ | |
945 | ret = scrub_recheck_block(fs_info, sblock_bad, | |
946 | is_metadata, have_csum, csum, | |
d9d181c1 | 947 | generation, sctx->csum_size); |
b5d67f64 SB |
948 | if (!ret && !sblock_bad->header_error && |
949 | !sblock_bad->checksum_error && | |
950 | sblock_bad->no_io_error_seen) | |
951 | goto corrected_error; | |
952 | else | |
953 | goto did_not_correct_error; | |
954 | } else { | |
955 | corrected_error: | |
d9d181c1 SB |
956 | spin_lock(&sctx->stat_lock); |
957 | sctx->stat.corrected_errors++; | |
958 | spin_unlock(&sctx->stat_lock); | |
606686ee | 959 | printk_ratelimited_in_rcu(KERN_ERR |
b5d67f64 | 960 | "btrfs: fixed up error at logical %llu on dev %s\n", |
606686ee | 961 | (unsigned long long)logical, |
a36cf8b8 | 962 | rcu_str_deref(dev->name)); |
8628764e | 963 | } |
b5d67f64 SB |
964 | } else { |
965 | did_not_correct_error: | |
d9d181c1 SB |
966 | spin_lock(&sctx->stat_lock); |
967 | sctx->stat.uncorrectable_errors++; | |
968 | spin_unlock(&sctx->stat_lock); | |
606686ee | 969 | printk_ratelimited_in_rcu(KERN_ERR |
b5d67f64 | 970 | "btrfs: unable to fixup (regular) error at logical %llu on dev %s\n", |
606686ee | 971 | (unsigned long long)logical, |
a36cf8b8 | 972 | rcu_str_deref(dev->name)); |
96e36920 | 973 | } |
a2de733c | 974 | |
b5d67f64 SB |
975 | out: |
976 | if (sblocks_for_recheck) { | |
977 | for (mirror_index = 0; mirror_index < BTRFS_MAX_MIRRORS; | |
978 | mirror_index++) { | |
979 | struct scrub_block *sblock = sblocks_for_recheck + | |
980 | mirror_index; | |
981 | int page_index; | |
982 | ||
7a9e9987 SB |
983 | for (page_index = 0; page_index < sblock->page_count; |
984 | page_index++) { | |
985 | sblock->pagev[page_index]->sblock = NULL; | |
986 | scrub_page_put(sblock->pagev[page_index]); | |
987 | } | |
b5d67f64 SB |
988 | } |
989 | kfree(sblocks_for_recheck); | |
990 | } | |
a2de733c | 991 | |
b5d67f64 SB |
992 | return 0; |
993 | } | |
a2de733c | 994 | |
d9d181c1 | 995 | static int scrub_setup_recheck_block(struct scrub_ctx *sctx, |
b5d67f64 SB |
996 | struct btrfs_mapping_tree *map_tree, |
997 | u64 length, u64 logical, | |
998 | struct scrub_block *sblocks_for_recheck) | |
999 | { | |
1000 | int page_index; | |
1001 | int mirror_index; | |
1002 | int ret; | |
1003 | ||
1004 | /* | |
7a9e9987 | 1005 | * note: the two members ref_count and outstanding_pages |
b5d67f64 SB |
1006 | * are not used (and not set) in the blocks that are used for |
1007 | * the recheck procedure | |
1008 | */ | |
1009 | ||
1010 | page_index = 0; | |
1011 | while (length > 0) { | |
1012 | u64 sublen = min_t(u64, length, PAGE_SIZE); | |
1013 | u64 mapped_length = sublen; | |
1014 | struct btrfs_bio *bbio = NULL; | |
a2de733c | 1015 | |
b5d67f64 SB |
1016 | /* |
1017 | * with a length of PAGE_SIZE, each returned stripe | |
1018 | * represents one mirror | |
1019 | */ | |
1020 | ret = btrfs_map_block(map_tree, WRITE, logical, &mapped_length, | |
1021 | &bbio, 0); | |
1022 | if (ret || !bbio || mapped_length < sublen) { | |
1023 | kfree(bbio); | |
1024 | return -EIO; | |
1025 | } | |
a2de733c | 1026 | |
b5d67f64 SB |
1027 | BUG_ON(page_index >= SCRUB_PAGES_PER_BIO); |
1028 | for (mirror_index = 0; mirror_index < (int)bbio->num_stripes; | |
1029 | mirror_index++) { | |
1030 | struct scrub_block *sblock; | |
1031 | struct scrub_page *page; | |
1032 | ||
1033 | if (mirror_index >= BTRFS_MAX_MIRRORS) | |
1034 | continue; | |
1035 | ||
1036 | sblock = sblocks_for_recheck + mirror_index; | |
7a9e9987 SB |
1037 | sblock->sctx = sctx; |
1038 | page = kzalloc(sizeof(*page), GFP_NOFS); | |
1039 | if (!page) { | |
1040 | leave_nomem: | |
d9d181c1 SB |
1041 | spin_lock(&sctx->stat_lock); |
1042 | sctx->stat.malloc_errors++; | |
1043 | spin_unlock(&sctx->stat_lock); | |
cf93dcce | 1044 | kfree(bbio); |
b5d67f64 SB |
1045 | return -ENOMEM; |
1046 | } | |
7a9e9987 SB |
1047 | scrub_page_get(page); |
1048 | sblock->pagev[page_index] = page; | |
1049 | page->logical = logical; | |
1050 | page->physical = bbio->stripes[mirror_index].physical; | |
1051 | /* for missing devices, dev->bdev is NULL */ | |
1052 | page->dev = bbio->stripes[mirror_index].dev; | |
1053 | page->mirror_num = mirror_index + 1; | |
b5d67f64 | 1054 | sblock->page_count++; |
7a9e9987 SB |
1055 | page->page = alloc_page(GFP_NOFS); |
1056 | if (!page->page) | |
1057 | goto leave_nomem; | |
b5d67f64 SB |
1058 | } |
1059 | kfree(bbio); | |
1060 | length -= sublen; | |
1061 | logical += sublen; | |
1062 | page_index++; | |
1063 | } | |
1064 | ||
1065 | return 0; | |
96e36920 ID |
1066 | } |
1067 | ||
b5d67f64 SB |
1068 | /* |
1069 | * this function will check the on disk data for checksum errors, header | |
1070 | * errors and read I/O errors. If any I/O errors happen, the exact pages | |
1071 | * which are errored are marked as being bad. The goal is to enable scrub | |
1072 | * to take those pages that are not errored from all the mirrors so that | |
1073 | * the pages that are errored in the just handled mirror can be repaired. | |
1074 | */ | |
1075 | static int scrub_recheck_block(struct btrfs_fs_info *fs_info, | |
1076 | struct scrub_block *sblock, int is_metadata, | |
1077 | int have_csum, u8 *csum, u64 generation, | |
1078 | u16 csum_size) | |
96e36920 | 1079 | { |
b5d67f64 | 1080 | int page_num; |
96e36920 | 1081 | |
b5d67f64 SB |
1082 | sblock->no_io_error_seen = 1; |
1083 | sblock->header_error = 0; | |
1084 | sblock->checksum_error = 0; | |
96e36920 | 1085 | |
b5d67f64 SB |
1086 | for (page_num = 0; page_num < sblock->page_count; page_num++) { |
1087 | struct bio *bio; | |
1088 | int ret; | |
7a9e9987 | 1089 | struct scrub_page *page = sblock->pagev[page_num]; |
b5d67f64 SB |
1090 | DECLARE_COMPLETION_ONSTACK(complete); |
1091 | ||
442a4f63 | 1092 | if (page->dev->bdev == NULL) { |
ea9947b4 SB |
1093 | page->io_error = 1; |
1094 | sblock->no_io_error_seen = 0; | |
1095 | continue; | |
1096 | } | |
1097 | ||
7a9e9987 | 1098 | WARN_ON(!page->page); |
b5d67f64 | 1099 | bio = bio_alloc(GFP_NOFS, 1); |
e627ee7b TI |
1100 | if (!bio) |
1101 | return -EIO; | |
442a4f63 | 1102 | bio->bi_bdev = page->dev->bdev; |
b5d67f64 SB |
1103 | bio->bi_sector = page->physical >> 9; |
1104 | bio->bi_end_io = scrub_complete_bio_end_io; | |
1105 | bio->bi_private = &complete; | |
1106 | ||
1107 | ret = bio_add_page(bio, page->page, PAGE_SIZE, 0); | |
1108 | if (PAGE_SIZE != ret) { | |
1109 | bio_put(bio); | |
1110 | return -EIO; | |
1111 | } | |
1112 | btrfsic_submit_bio(READ, bio); | |
96e36920 | 1113 | |
b5d67f64 SB |
1114 | /* this will also unplug the queue */ |
1115 | wait_for_completion(&complete); | |
96e36920 | 1116 | |
b5d67f64 SB |
1117 | page->io_error = !test_bit(BIO_UPTODATE, &bio->bi_flags); |
1118 | if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
1119 | sblock->no_io_error_seen = 0; | |
1120 | bio_put(bio); | |
1121 | } | |
96e36920 | 1122 | |
b5d67f64 SB |
1123 | if (sblock->no_io_error_seen) |
1124 | scrub_recheck_block_checksum(fs_info, sblock, is_metadata, | |
1125 | have_csum, csum, generation, | |
1126 | csum_size); | |
1127 | ||
1128 | return 0; | |
a2de733c AJ |
1129 | } |
1130 | ||
b5d67f64 SB |
1131 | static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info, |
1132 | struct scrub_block *sblock, | |
1133 | int is_metadata, int have_csum, | |
1134 | const u8 *csum, u64 generation, | |
1135 | u16 csum_size) | |
a2de733c | 1136 | { |
b5d67f64 SB |
1137 | int page_num; |
1138 | u8 calculated_csum[BTRFS_CSUM_SIZE]; | |
1139 | u32 crc = ~(u32)0; | |
1140 | struct btrfs_root *root = fs_info->extent_root; | |
1141 | void *mapped_buffer; | |
1142 | ||
7a9e9987 | 1143 | WARN_ON(!sblock->pagev[0]->page); |
b5d67f64 SB |
1144 | if (is_metadata) { |
1145 | struct btrfs_header *h; | |
1146 | ||
7a9e9987 | 1147 | mapped_buffer = kmap_atomic(sblock->pagev[0]->page); |
b5d67f64 SB |
1148 | h = (struct btrfs_header *)mapped_buffer; |
1149 | ||
7a9e9987 | 1150 | if (sblock->pagev[0]->logical != le64_to_cpu(h->bytenr) || |
b5d67f64 SB |
1151 | memcmp(h->fsid, fs_info->fsid, BTRFS_UUID_SIZE) || |
1152 | memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid, | |
442a4f63 | 1153 | BTRFS_UUID_SIZE)) { |
b5d67f64 | 1154 | sblock->header_error = 1; |
442a4f63 SB |
1155 | } else if (generation != le64_to_cpu(h->generation)) { |
1156 | sblock->header_error = 1; | |
1157 | sblock->generation_error = 1; | |
1158 | } | |
b5d67f64 SB |
1159 | csum = h->csum; |
1160 | } else { | |
1161 | if (!have_csum) | |
1162 | return; | |
a2de733c | 1163 | |
7a9e9987 | 1164 | mapped_buffer = kmap_atomic(sblock->pagev[0]->page); |
b5d67f64 | 1165 | } |
a2de733c | 1166 | |
b5d67f64 SB |
1167 | for (page_num = 0;;) { |
1168 | if (page_num == 0 && is_metadata) | |
1169 | crc = btrfs_csum_data(root, | |
1170 | ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE, | |
1171 | crc, PAGE_SIZE - BTRFS_CSUM_SIZE); | |
1172 | else | |
1173 | crc = btrfs_csum_data(root, mapped_buffer, crc, | |
1174 | PAGE_SIZE); | |
1175 | ||
9613bebb | 1176 | kunmap_atomic(mapped_buffer); |
b5d67f64 SB |
1177 | page_num++; |
1178 | if (page_num >= sblock->page_count) | |
1179 | break; | |
7a9e9987 | 1180 | WARN_ON(!sblock->pagev[page_num]->page); |
b5d67f64 | 1181 | |
7a9e9987 | 1182 | mapped_buffer = kmap_atomic(sblock->pagev[page_num]->page); |
b5d67f64 SB |
1183 | } |
1184 | ||
1185 | btrfs_csum_final(crc, calculated_csum); | |
1186 | if (memcmp(calculated_csum, csum, csum_size)) | |
1187 | sblock->checksum_error = 1; | |
a2de733c AJ |
1188 | } |
1189 | ||
b5d67f64 | 1190 | static void scrub_complete_bio_end_io(struct bio *bio, int err) |
a2de733c | 1191 | { |
b5d67f64 SB |
1192 | complete((struct completion *)bio->bi_private); |
1193 | } | |
a2de733c | 1194 | |
b5d67f64 SB |
1195 | static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad, |
1196 | struct scrub_block *sblock_good, | |
1197 | int force_write) | |
1198 | { | |
1199 | int page_num; | |
1200 | int ret = 0; | |
96e36920 | 1201 | |
b5d67f64 SB |
1202 | for (page_num = 0; page_num < sblock_bad->page_count; page_num++) { |
1203 | int ret_sub; | |
96e36920 | 1204 | |
b5d67f64 SB |
1205 | ret_sub = scrub_repair_page_from_good_copy(sblock_bad, |
1206 | sblock_good, | |
1207 | page_num, | |
1208 | force_write); | |
1209 | if (ret_sub) | |
1210 | ret = ret_sub; | |
a2de733c | 1211 | } |
b5d67f64 SB |
1212 | |
1213 | return ret; | |
1214 | } | |
1215 | ||
1216 | static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad, | |
1217 | struct scrub_block *sblock_good, | |
1218 | int page_num, int force_write) | |
1219 | { | |
7a9e9987 SB |
1220 | struct scrub_page *page_bad = sblock_bad->pagev[page_num]; |
1221 | struct scrub_page *page_good = sblock_good->pagev[page_num]; | |
b5d67f64 | 1222 | |
7a9e9987 SB |
1223 | BUG_ON(page_bad->page == NULL); |
1224 | BUG_ON(page_good->page == NULL); | |
b5d67f64 SB |
1225 | if (force_write || sblock_bad->header_error || |
1226 | sblock_bad->checksum_error || page_bad->io_error) { | |
1227 | struct bio *bio; | |
1228 | int ret; | |
1229 | DECLARE_COMPLETION_ONSTACK(complete); | |
1230 | ||
1231 | bio = bio_alloc(GFP_NOFS, 1); | |
e627ee7b TI |
1232 | if (!bio) |
1233 | return -EIO; | |
442a4f63 | 1234 | bio->bi_bdev = page_bad->dev->bdev; |
b5d67f64 SB |
1235 | bio->bi_sector = page_bad->physical >> 9; |
1236 | bio->bi_end_io = scrub_complete_bio_end_io; | |
1237 | bio->bi_private = &complete; | |
1238 | ||
1239 | ret = bio_add_page(bio, page_good->page, PAGE_SIZE, 0); | |
1240 | if (PAGE_SIZE != ret) { | |
1241 | bio_put(bio); | |
1242 | return -EIO; | |
13db62b7 | 1243 | } |
b5d67f64 SB |
1244 | btrfsic_submit_bio(WRITE, bio); |
1245 | ||
1246 | /* this will also unplug the queue */ | |
1247 | wait_for_completion(&complete); | |
442a4f63 SB |
1248 | if (!bio_flagged(bio, BIO_UPTODATE)) { |
1249 | btrfs_dev_stat_inc_and_print(page_bad->dev, | |
1250 | BTRFS_DEV_STAT_WRITE_ERRS); | |
1251 | bio_put(bio); | |
1252 | return -EIO; | |
1253 | } | |
b5d67f64 | 1254 | bio_put(bio); |
a2de733c AJ |
1255 | } |
1256 | ||
b5d67f64 SB |
1257 | return 0; |
1258 | } | |
1259 | ||
1260 | static void scrub_checksum(struct scrub_block *sblock) | |
1261 | { | |
1262 | u64 flags; | |
1263 | int ret; | |
1264 | ||
7a9e9987 SB |
1265 | WARN_ON(sblock->page_count < 1); |
1266 | flags = sblock->pagev[0]->flags; | |
b5d67f64 SB |
1267 | ret = 0; |
1268 | if (flags & BTRFS_EXTENT_FLAG_DATA) | |
1269 | ret = scrub_checksum_data(sblock); | |
1270 | else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) | |
1271 | ret = scrub_checksum_tree_block(sblock); | |
1272 | else if (flags & BTRFS_EXTENT_FLAG_SUPER) | |
1273 | (void)scrub_checksum_super(sblock); | |
1274 | else | |
1275 | WARN_ON(1); | |
1276 | if (ret) | |
1277 | scrub_handle_errored_block(sblock); | |
a2de733c AJ |
1278 | } |
1279 | ||
b5d67f64 | 1280 | static int scrub_checksum_data(struct scrub_block *sblock) |
a2de733c | 1281 | { |
d9d181c1 | 1282 | struct scrub_ctx *sctx = sblock->sctx; |
a2de733c | 1283 | u8 csum[BTRFS_CSUM_SIZE]; |
b5d67f64 SB |
1284 | u8 *on_disk_csum; |
1285 | struct page *page; | |
1286 | void *buffer; | |
a2de733c AJ |
1287 | u32 crc = ~(u32)0; |
1288 | int fail = 0; | |
a36cf8b8 | 1289 | struct btrfs_root *root = sctx->dev_root; |
b5d67f64 SB |
1290 | u64 len; |
1291 | int index; | |
a2de733c | 1292 | |
b5d67f64 | 1293 | BUG_ON(sblock->page_count < 1); |
7a9e9987 | 1294 | if (!sblock->pagev[0]->have_csum) |
a2de733c AJ |
1295 | return 0; |
1296 | ||
7a9e9987 SB |
1297 | on_disk_csum = sblock->pagev[0]->csum; |
1298 | page = sblock->pagev[0]->page; | |
9613bebb | 1299 | buffer = kmap_atomic(page); |
b5d67f64 | 1300 | |
d9d181c1 | 1301 | len = sctx->sectorsize; |
b5d67f64 SB |
1302 | index = 0; |
1303 | for (;;) { | |
1304 | u64 l = min_t(u64, len, PAGE_SIZE); | |
1305 | ||
1306 | crc = btrfs_csum_data(root, buffer, crc, l); | |
9613bebb | 1307 | kunmap_atomic(buffer); |
b5d67f64 SB |
1308 | len -= l; |
1309 | if (len == 0) | |
1310 | break; | |
1311 | index++; | |
1312 | BUG_ON(index >= sblock->page_count); | |
7a9e9987 SB |
1313 | BUG_ON(!sblock->pagev[index]->page); |
1314 | page = sblock->pagev[index]->page; | |
9613bebb | 1315 | buffer = kmap_atomic(page); |
b5d67f64 SB |
1316 | } |
1317 | ||
a2de733c | 1318 | btrfs_csum_final(crc, csum); |
d9d181c1 | 1319 | if (memcmp(csum, on_disk_csum, sctx->csum_size)) |
a2de733c AJ |
1320 | fail = 1; |
1321 | ||
a2de733c AJ |
1322 | return fail; |
1323 | } | |
1324 | ||
b5d67f64 | 1325 | static int scrub_checksum_tree_block(struct scrub_block *sblock) |
a2de733c | 1326 | { |
d9d181c1 | 1327 | struct scrub_ctx *sctx = sblock->sctx; |
a2de733c | 1328 | struct btrfs_header *h; |
a36cf8b8 | 1329 | struct btrfs_root *root = sctx->dev_root; |
a2de733c | 1330 | struct btrfs_fs_info *fs_info = root->fs_info; |
b5d67f64 SB |
1331 | u8 calculated_csum[BTRFS_CSUM_SIZE]; |
1332 | u8 on_disk_csum[BTRFS_CSUM_SIZE]; | |
1333 | struct page *page; | |
1334 | void *mapped_buffer; | |
1335 | u64 mapped_size; | |
1336 | void *p; | |
a2de733c AJ |
1337 | u32 crc = ~(u32)0; |
1338 | int fail = 0; | |
1339 | int crc_fail = 0; | |
b5d67f64 SB |
1340 | u64 len; |
1341 | int index; | |
1342 | ||
1343 | BUG_ON(sblock->page_count < 1); | |
7a9e9987 | 1344 | page = sblock->pagev[0]->page; |
9613bebb | 1345 | mapped_buffer = kmap_atomic(page); |
b5d67f64 | 1346 | h = (struct btrfs_header *)mapped_buffer; |
d9d181c1 | 1347 | memcpy(on_disk_csum, h->csum, sctx->csum_size); |
a2de733c AJ |
1348 | |
1349 | /* | |
1350 | * we don't use the getter functions here, as we | |
1351 | * a) don't have an extent buffer and | |
1352 | * b) the page is already kmapped | |
1353 | */ | |
a2de733c | 1354 | |
7a9e9987 | 1355 | if (sblock->pagev[0]->logical != le64_to_cpu(h->bytenr)) |
a2de733c AJ |
1356 | ++fail; |
1357 | ||
7a9e9987 | 1358 | if (sblock->pagev[0]->generation != le64_to_cpu(h->generation)) |
a2de733c AJ |
1359 | ++fail; |
1360 | ||
1361 | if (memcmp(h->fsid, fs_info->fsid, BTRFS_UUID_SIZE)) | |
1362 | ++fail; | |
1363 | ||
1364 | if (memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid, | |
1365 | BTRFS_UUID_SIZE)) | |
1366 | ++fail; | |
1367 | ||
d9d181c1 SB |
1368 | BUG_ON(sctx->nodesize != sctx->leafsize); |
1369 | len = sctx->nodesize - BTRFS_CSUM_SIZE; | |
b5d67f64 SB |
1370 | mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE; |
1371 | p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE; | |
1372 | index = 0; | |
1373 | for (;;) { | |
1374 | u64 l = min_t(u64, len, mapped_size); | |
1375 | ||
1376 | crc = btrfs_csum_data(root, p, crc, l); | |
9613bebb | 1377 | kunmap_atomic(mapped_buffer); |
b5d67f64 SB |
1378 | len -= l; |
1379 | if (len == 0) | |
1380 | break; | |
1381 | index++; | |
1382 | BUG_ON(index >= sblock->page_count); | |
7a9e9987 SB |
1383 | BUG_ON(!sblock->pagev[index]->page); |
1384 | page = sblock->pagev[index]->page; | |
9613bebb | 1385 | mapped_buffer = kmap_atomic(page); |
b5d67f64 SB |
1386 | mapped_size = PAGE_SIZE; |
1387 | p = mapped_buffer; | |
1388 | } | |
1389 | ||
1390 | btrfs_csum_final(crc, calculated_csum); | |
d9d181c1 | 1391 | if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size)) |
a2de733c AJ |
1392 | ++crc_fail; |
1393 | ||
a2de733c AJ |
1394 | return fail || crc_fail; |
1395 | } | |
1396 | ||
b5d67f64 | 1397 | static int scrub_checksum_super(struct scrub_block *sblock) |
a2de733c AJ |
1398 | { |
1399 | struct btrfs_super_block *s; | |
d9d181c1 | 1400 | struct scrub_ctx *sctx = sblock->sctx; |
a36cf8b8 | 1401 | struct btrfs_root *root = sctx->dev_root; |
a2de733c | 1402 | struct btrfs_fs_info *fs_info = root->fs_info; |
b5d67f64 SB |
1403 | u8 calculated_csum[BTRFS_CSUM_SIZE]; |
1404 | u8 on_disk_csum[BTRFS_CSUM_SIZE]; | |
1405 | struct page *page; | |
1406 | void *mapped_buffer; | |
1407 | u64 mapped_size; | |
1408 | void *p; | |
a2de733c | 1409 | u32 crc = ~(u32)0; |
442a4f63 SB |
1410 | int fail_gen = 0; |
1411 | int fail_cor = 0; | |
b5d67f64 SB |
1412 | u64 len; |
1413 | int index; | |
a2de733c | 1414 | |
b5d67f64 | 1415 | BUG_ON(sblock->page_count < 1); |
7a9e9987 | 1416 | page = sblock->pagev[0]->page; |
9613bebb | 1417 | mapped_buffer = kmap_atomic(page); |
b5d67f64 | 1418 | s = (struct btrfs_super_block *)mapped_buffer; |
d9d181c1 | 1419 | memcpy(on_disk_csum, s->csum, sctx->csum_size); |
a2de733c | 1420 | |
7a9e9987 | 1421 | if (sblock->pagev[0]->logical != le64_to_cpu(s->bytenr)) |
442a4f63 | 1422 | ++fail_cor; |
a2de733c | 1423 | |
7a9e9987 | 1424 | if (sblock->pagev[0]->generation != le64_to_cpu(s->generation)) |
442a4f63 | 1425 | ++fail_gen; |
a2de733c AJ |
1426 | |
1427 | if (memcmp(s->fsid, fs_info->fsid, BTRFS_UUID_SIZE)) | |
442a4f63 | 1428 | ++fail_cor; |
a2de733c | 1429 | |
b5d67f64 SB |
1430 | len = BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE; |
1431 | mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE; | |
1432 | p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE; | |
1433 | index = 0; | |
1434 | for (;;) { | |
1435 | u64 l = min_t(u64, len, mapped_size); | |
1436 | ||
1437 | crc = btrfs_csum_data(root, p, crc, l); | |
9613bebb | 1438 | kunmap_atomic(mapped_buffer); |
b5d67f64 SB |
1439 | len -= l; |
1440 | if (len == 0) | |
1441 | break; | |
1442 | index++; | |
1443 | BUG_ON(index >= sblock->page_count); | |
7a9e9987 SB |
1444 | BUG_ON(!sblock->pagev[index]->page); |
1445 | page = sblock->pagev[index]->page; | |
9613bebb | 1446 | mapped_buffer = kmap_atomic(page); |
b5d67f64 SB |
1447 | mapped_size = PAGE_SIZE; |
1448 | p = mapped_buffer; | |
1449 | } | |
1450 | ||
1451 | btrfs_csum_final(crc, calculated_csum); | |
d9d181c1 | 1452 | if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size)) |
442a4f63 | 1453 | ++fail_cor; |
a2de733c | 1454 | |
442a4f63 | 1455 | if (fail_cor + fail_gen) { |
a2de733c AJ |
1456 | /* |
1457 | * if we find an error in a super block, we just report it. | |
1458 | * They will get written with the next transaction commit | |
1459 | * anyway | |
1460 | */ | |
d9d181c1 SB |
1461 | spin_lock(&sctx->stat_lock); |
1462 | ++sctx->stat.super_errors; | |
1463 | spin_unlock(&sctx->stat_lock); | |
442a4f63 | 1464 | if (fail_cor) |
7a9e9987 | 1465 | btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev, |
442a4f63 SB |
1466 | BTRFS_DEV_STAT_CORRUPTION_ERRS); |
1467 | else | |
7a9e9987 | 1468 | btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev, |
442a4f63 | 1469 | BTRFS_DEV_STAT_GENERATION_ERRS); |
a2de733c AJ |
1470 | } |
1471 | ||
442a4f63 | 1472 | return fail_cor + fail_gen; |
a2de733c AJ |
1473 | } |
1474 | ||
b5d67f64 SB |
1475 | static void scrub_block_get(struct scrub_block *sblock) |
1476 | { | |
1477 | atomic_inc(&sblock->ref_count); | |
1478 | } | |
1479 | ||
1480 | static void scrub_block_put(struct scrub_block *sblock) | |
1481 | { | |
1482 | if (atomic_dec_and_test(&sblock->ref_count)) { | |
1483 | int i; | |
1484 | ||
1485 | for (i = 0; i < sblock->page_count; i++) | |
7a9e9987 | 1486 | scrub_page_put(sblock->pagev[i]); |
b5d67f64 SB |
1487 | kfree(sblock); |
1488 | } | |
1489 | } | |
1490 | ||
7a9e9987 SB |
1491 | static void scrub_page_get(struct scrub_page *spage) |
1492 | { | |
1493 | atomic_inc(&spage->ref_count); | |
1494 | } | |
1495 | ||
1496 | static void scrub_page_put(struct scrub_page *spage) | |
1497 | { | |
1498 | if (atomic_dec_and_test(&spage->ref_count)) { | |
1499 | if (spage->page) | |
1500 | __free_page(spage->page); | |
1501 | kfree(spage); | |
1502 | } | |
1503 | } | |
1504 | ||
d9d181c1 | 1505 | static void scrub_submit(struct scrub_ctx *sctx) |
a2de733c AJ |
1506 | { |
1507 | struct scrub_bio *sbio; | |
1508 | ||
d9d181c1 | 1509 | if (sctx->curr == -1) |
1623edeb | 1510 | return; |
a2de733c | 1511 | |
d9d181c1 SB |
1512 | sbio = sctx->bios[sctx->curr]; |
1513 | sctx->curr = -1; | |
1514 | atomic_inc(&sctx->in_flight); | |
a2de733c | 1515 | |
21adbd5c | 1516 | btrfsic_submit_bio(READ, sbio->bio); |
a2de733c AJ |
1517 | } |
1518 | ||
d9d181c1 | 1519 | static int scrub_add_page_to_bio(struct scrub_ctx *sctx, |
b5d67f64 | 1520 | struct scrub_page *spage) |
a2de733c | 1521 | { |
b5d67f64 | 1522 | struct scrub_block *sblock = spage->sblock; |
a2de733c | 1523 | struct scrub_bio *sbio; |
69f4cb52 | 1524 | int ret; |
a2de733c AJ |
1525 | |
1526 | again: | |
1527 | /* | |
1528 | * grab a fresh bio or wait for one to become available | |
1529 | */ | |
d9d181c1 SB |
1530 | while (sctx->curr == -1) { |
1531 | spin_lock(&sctx->list_lock); | |
1532 | sctx->curr = sctx->first_free; | |
1533 | if (sctx->curr != -1) { | |
1534 | sctx->first_free = sctx->bios[sctx->curr]->next_free; | |
1535 | sctx->bios[sctx->curr]->next_free = -1; | |
1536 | sctx->bios[sctx->curr]->page_count = 0; | |
1537 | spin_unlock(&sctx->list_lock); | |
a2de733c | 1538 | } else { |
d9d181c1 SB |
1539 | spin_unlock(&sctx->list_lock); |
1540 | wait_event(sctx->list_wait, sctx->first_free != -1); | |
a2de733c AJ |
1541 | } |
1542 | } | |
d9d181c1 | 1543 | sbio = sctx->bios[sctx->curr]; |
b5d67f64 | 1544 | if (sbio->page_count == 0) { |
69f4cb52 AJ |
1545 | struct bio *bio; |
1546 | ||
b5d67f64 SB |
1547 | sbio->physical = spage->physical; |
1548 | sbio->logical = spage->logical; | |
a36cf8b8 | 1549 | sbio->dev = spage->dev; |
b5d67f64 SB |
1550 | bio = sbio->bio; |
1551 | if (!bio) { | |
d9d181c1 | 1552 | bio = bio_alloc(GFP_NOFS, sctx->pages_per_bio); |
b5d67f64 SB |
1553 | if (!bio) |
1554 | return -ENOMEM; | |
1555 | sbio->bio = bio; | |
1556 | } | |
69f4cb52 AJ |
1557 | |
1558 | bio->bi_private = sbio; | |
1559 | bio->bi_end_io = scrub_bio_end_io; | |
a36cf8b8 SB |
1560 | bio->bi_bdev = sbio->dev->bdev; |
1561 | bio->bi_sector = sbio->physical >> 9; | |
69f4cb52 | 1562 | sbio->err = 0; |
b5d67f64 SB |
1563 | } else if (sbio->physical + sbio->page_count * PAGE_SIZE != |
1564 | spage->physical || | |
1565 | sbio->logical + sbio->page_count * PAGE_SIZE != | |
a36cf8b8 SB |
1566 | spage->logical || |
1567 | sbio->dev != spage->dev) { | |
d9d181c1 | 1568 | scrub_submit(sctx); |
a2de733c AJ |
1569 | goto again; |
1570 | } | |
69f4cb52 | 1571 | |
b5d67f64 SB |
1572 | sbio->pagev[sbio->page_count] = spage; |
1573 | ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0); | |
1574 | if (ret != PAGE_SIZE) { | |
1575 | if (sbio->page_count < 1) { | |
1576 | bio_put(sbio->bio); | |
1577 | sbio->bio = NULL; | |
1578 | return -EIO; | |
1579 | } | |
d9d181c1 | 1580 | scrub_submit(sctx); |
69f4cb52 AJ |
1581 | goto again; |
1582 | } | |
1583 | ||
b5d67f64 SB |
1584 | scrub_block_get(sblock); /* one for the added page */ |
1585 | atomic_inc(&sblock->outstanding_pages); | |
1586 | sbio->page_count++; | |
d9d181c1 SB |
1587 | if (sbio->page_count == sctx->pages_per_bio) |
1588 | scrub_submit(sctx); | |
b5d67f64 SB |
1589 | |
1590 | return 0; | |
1591 | } | |
1592 | ||
d9d181c1 | 1593 | static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len, |
a36cf8b8 SB |
1594 | u64 physical, struct btrfs_device *dev, u64 flags, |
1595 | u64 gen, int mirror_num, u8 *csum, int force) | |
b5d67f64 SB |
1596 | { |
1597 | struct scrub_block *sblock; | |
1598 | int index; | |
1599 | ||
1600 | sblock = kzalloc(sizeof(*sblock), GFP_NOFS); | |
1601 | if (!sblock) { | |
d9d181c1 SB |
1602 | spin_lock(&sctx->stat_lock); |
1603 | sctx->stat.malloc_errors++; | |
1604 | spin_unlock(&sctx->stat_lock); | |
b5d67f64 | 1605 | return -ENOMEM; |
a2de733c | 1606 | } |
b5d67f64 | 1607 | |
7a9e9987 SB |
1608 | /* one ref inside this function, plus one for each page added to |
1609 | * a bio later on */ | |
b5d67f64 | 1610 | atomic_set(&sblock->ref_count, 1); |
d9d181c1 | 1611 | sblock->sctx = sctx; |
b5d67f64 SB |
1612 | sblock->no_io_error_seen = 1; |
1613 | ||
1614 | for (index = 0; len > 0; index++) { | |
7a9e9987 | 1615 | struct scrub_page *spage; |
b5d67f64 SB |
1616 | u64 l = min_t(u64, len, PAGE_SIZE); |
1617 | ||
7a9e9987 SB |
1618 | spage = kzalloc(sizeof(*spage), GFP_NOFS); |
1619 | if (!spage) { | |
1620 | leave_nomem: | |
d9d181c1 SB |
1621 | spin_lock(&sctx->stat_lock); |
1622 | sctx->stat.malloc_errors++; | |
1623 | spin_unlock(&sctx->stat_lock); | |
7a9e9987 | 1624 | scrub_block_put(sblock); |
b5d67f64 SB |
1625 | return -ENOMEM; |
1626 | } | |
7a9e9987 SB |
1627 | BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK); |
1628 | scrub_page_get(spage); | |
1629 | sblock->pagev[index] = spage; | |
b5d67f64 | 1630 | spage->sblock = sblock; |
a36cf8b8 | 1631 | spage->dev = dev; |
b5d67f64 SB |
1632 | spage->flags = flags; |
1633 | spage->generation = gen; | |
1634 | spage->logical = logical; | |
1635 | spage->physical = physical; | |
1636 | spage->mirror_num = mirror_num; | |
1637 | if (csum) { | |
1638 | spage->have_csum = 1; | |
d9d181c1 | 1639 | memcpy(spage->csum, csum, sctx->csum_size); |
b5d67f64 SB |
1640 | } else { |
1641 | spage->have_csum = 0; | |
1642 | } | |
1643 | sblock->page_count++; | |
7a9e9987 SB |
1644 | spage->page = alloc_page(GFP_NOFS); |
1645 | if (!spage->page) | |
1646 | goto leave_nomem; | |
b5d67f64 SB |
1647 | len -= l; |
1648 | logical += l; | |
1649 | physical += l; | |
1650 | } | |
1651 | ||
7a9e9987 | 1652 | WARN_ON(sblock->page_count == 0); |
b5d67f64 | 1653 | for (index = 0; index < sblock->page_count; index++) { |
7a9e9987 | 1654 | struct scrub_page *spage = sblock->pagev[index]; |
1bc87793 AJ |
1655 | int ret; |
1656 | ||
d9d181c1 | 1657 | ret = scrub_add_page_to_bio(sctx, spage); |
b5d67f64 SB |
1658 | if (ret) { |
1659 | scrub_block_put(sblock); | |
1bc87793 | 1660 | return ret; |
b5d67f64 | 1661 | } |
1bc87793 | 1662 | } |
a2de733c | 1663 | |
b5d67f64 | 1664 | if (force) |
d9d181c1 | 1665 | scrub_submit(sctx); |
a2de733c | 1666 | |
b5d67f64 SB |
1667 | /* last one frees, either here or in bio completion for last page */ |
1668 | scrub_block_put(sblock); | |
a2de733c AJ |
1669 | return 0; |
1670 | } | |
1671 | ||
b5d67f64 SB |
1672 | static void scrub_bio_end_io(struct bio *bio, int err) |
1673 | { | |
1674 | struct scrub_bio *sbio = bio->bi_private; | |
a36cf8b8 | 1675 | struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info; |
b5d67f64 SB |
1676 | |
1677 | sbio->err = err; | |
1678 | sbio->bio = bio; | |
1679 | ||
1680 | btrfs_queue_worker(&fs_info->scrub_workers, &sbio->work); | |
1681 | } | |
1682 | ||
1683 | static void scrub_bio_end_io_worker(struct btrfs_work *work) | |
1684 | { | |
1685 | struct scrub_bio *sbio = container_of(work, struct scrub_bio, work); | |
d9d181c1 | 1686 | struct scrub_ctx *sctx = sbio->sctx; |
b5d67f64 SB |
1687 | int i; |
1688 | ||
1689 | BUG_ON(sbio->page_count > SCRUB_PAGES_PER_BIO); | |
1690 | if (sbio->err) { | |
1691 | for (i = 0; i < sbio->page_count; i++) { | |
1692 | struct scrub_page *spage = sbio->pagev[i]; | |
1693 | ||
1694 | spage->io_error = 1; | |
1695 | spage->sblock->no_io_error_seen = 0; | |
1696 | } | |
1697 | } | |
1698 | ||
1699 | /* now complete the scrub_block items that have all pages completed */ | |
1700 | for (i = 0; i < sbio->page_count; i++) { | |
1701 | struct scrub_page *spage = sbio->pagev[i]; | |
1702 | struct scrub_block *sblock = spage->sblock; | |
1703 | ||
1704 | if (atomic_dec_and_test(&sblock->outstanding_pages)) | |
1705 | scrub_block_complete(sblock); | |
1706 | scrub_block_put(sblock); | |
1707 | } | |
1708 | ||
b5d67f64 SB |
1709 | bio_put(sbio->bio); |
1710 | sbio->bio = NULL; | |
d9d181c1 SB |
1711 | spin_lock(&sctx->list_lock); |
1712 | sbio->next_free = sctx->first_free; | |
1713 | sctx->first_free = sbio->index; | |
1714 | spin_unlock(&sctx->list_lock); | |
1715 | atomic_dec(&sctx->in_flight); | |
1716 | wake_up(&sctx->list_wait); | |
b5d67f64 SB |
1717 | } |
1718 | ||
1719 | static void scrub_block_complete(struct scrub_block *sblock) | |
1720 | { | |
1721 | if (!sblock->no_io_error_seen) | |
1722 | scrub_handle_errored_block(sblock); | |
1723 | else | |
1724 | scrub_checksum(sblock); | |
1725 | } | |
1726 | ||
d9d181c1 | 1727 | static int scrub_find_csum(struct scrub_ctx *sctx, u64 logical, u64 len, |
a2de733c AJ |
1728 | u8 *csum) |
1729 | { | |
1730 | struct btrfs_ordered_sum *sum = NULL; | |
1731 | int ret = 0; | |
1732 | unsigned long i; | |
1733 | unsigned long num_sectors; | |
a2de733c | 1734 | |
d9d181c1 SB |
1735 | while (!list_empty(&sctx->csum_list)) { |
1736 | sum = list_first_entry(&sctx->csum_list, | |
a2de733c AJ |
1737 | struct btrfs_ordered_sum, list); |
1738 | if (sum->bytenr > logical) | |
1739 | return 0; | |
1740 | if (sum->bytenr + sum->len > logical) | |
1741 | break; | |
1742 | ||
d9d181c1 | 1743 | ++sctx->stat.csum_discards; |
a2de733c AJ |
1744 | list_del(&sum->list); |
1745 | kfree(sum); | |
1746 | sum = NULL; | |
1747 | } | |
1748 | if (!sum) | |
1749 | return 0; | |
1750 | ||
d9d181c1 | 1751 | num_sectors = sum->len / sctx->sectorsize; |
a2de733c AJ |
1752 | for (i = 0; i < num_sectors; ++i) { |
1753 | if (sum->sums[i].bytenr == logical) { | |
d9d181c1 | 1754 | memcpy(csum, &sum->sums[i].sum, sctx->csum_size); |
a2de733c AJ |
1755 | ret = 1; |
1756 | break; | |
1757 | } | |
1758 | } | |
1759 | if (ret && i == num_sectors - 1) { | |
1760 | list_del(&sum->list); | |
1761 | kfree(sum); | |
1762 | } | |
1763 | return ret; | |
1764 | } | |
1765 | ||
1766 | /* scrub extent tries to collect up to 64 kB for each bio */ | |
d9d181c1 | 1767 | static int scrub_extent(struct scrub_ctx *sctx, u64 logical, u64 len, |
a36cf8b8 SB |
1768 | u64 physical, struct btrfs_device *dev, u64 flags, |
1769 | u64 gen, int mirror_num) | |
a2de733c AJ |
1770 | { |
1771 | int ret; | |
1772 | u8 csum[BTRFS_CSUM_SIZE]; | |
b5d67f64 SB |
1773 | u32 blocksize; |
1774 | ||
1775 | if (flags & BTRFS_EXTENT_FLAG_DATA) { | |
d9d181c1 SB |
1776 | blocksize = sctx->sectorsize; |
1777 | spin_lock(&sctx->stat_lock); | |
1778 | sctx->stat.data_extents_scrubbed++; | |
1779 | sctx->stat.data_bytes_scrubbed += len; | |
1780 | spin_unlock(&sctx->stat_lock); | |
b5d67f64 | 1781 | } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { |
d9d181c1 SB |
1782 | BUG_ON(sctx->nodesize != sctx->leafsize); |
1783 | blocksize = sctx->nodesize; | |
1784 | spin_lock(&sctx->stat_lock); | |
1785 | sctx->stat.tree_extents_scrubbed++; | |
1786 | sctx->stat.tree_bytes_scrubbed += len; | |
1787 | spin_unlock(&sctx->stat_lock); | |
b5d67f64 | 1788 | } else { |
d9d181c1 | 1789 | blocksize = sctx->sectorsize; |
b5d67f64 SB |
1790 | BUG_ON(1); |
1791 | } | |
a2de733c AJ |
1792 | |
1793 | while (len) { | |
b5d67f64 | 1794 | u64 l = min_t(u64, len, blocksize); |
a2de733c AJ |
1795 | int have_csum = 0; |
1796 | ||
1797 | if (flags & BTRFS_EXTENT_FLAG_DATA) { | |
1798 | /* push csums to sbio */ | |
d9d181c1 | 1799 | have_csum = scrub_find_csum(sctx, logical, l, csum); |
a2de733c | 1800 | if (have_csum == 0) |
d9d181c1 | 1801 | ++sctx->stat.no_csum; |
a2de733c | 1802 | } |
a36cf8b8 | 1803 | ret = scrub_pages(sctx, logical, l, physical, dev, flags, gen, |
b5d67f64 | 1804 | mirror_num, have_csum ? csum : NULL, 0); |
a2de733c AJ |
1805 | if (ret) |
1806 | return ret; | |
1807 | len -= l; | |
1808 | logical += l; | |
1809 | physical += l; | |
1810 | } | |
1811 | return 0; | |
1812 | } | |
1813 | ||
d9d181c1 | 1814 | static noinline_for_stack int scrub_stripe(struct scrub_ctx *sctx, |
a36cf8b8 SB |
1815 | struct map_lookup *map, |
1816 | struct btrfs_device *scrub_dev, | |
1817 | int num, u64 base, u64 length) | |
a2de733c AJ |
1818 | { |
1819 | struct btrfs_path *path; | |
a36cf8b8 | 1820 | struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info; |
a2de733c AJ |
1821 | struct btrfs_root *root = fs_info->extent_root; |
1822 | struct btrfs_root *csum_root = fs_info->csum_root; | |
1823 | struct btrfs_extent_item *extent; | |
e7786c3a | 1824 | struct blk_plug plug; |
a2de733c AJ |
1825 | u64 flags; |
1826 | int ret; | |
1827 | int slot; | |
1828 | int i; | |
1829 | u64 nstripes; | |
a2de733c AJ |
1830 | struct extent_buffer *l; |
1831 | struct btrfs_key key; | |
1832 | u64 physical; | |
1833 | u64 logical; | |
1834 | u64 generation; | |
e12fa9cd | 1835 | int mirror_num; |
7a26285e AJ |
1836 | struct reada_control *reada1; |
1837 | struct reada_control *reada2; | |
1838 | struct btrfs_key key_start; | |
1839 | struct btrfs_key key_end; | |
a2de733c AJ |
1840 | u64 increment = map->stripe_len; |
1841 | u64 offset; | |
1842 | ||
1843 | nstripes = length; | |
1844 | offset = 0; | |
1845 | do_div(nstripes, map->stripe_len); | |
1846 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { | |
1847 | offset = map->stripe_len * num; | |
1848 | increment = map->stripe_len * map->num_stripes; | |
193ea74b | 1849 | mirror_num = 1; |
a2de733c AJ |
1850 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
1851 | int factor = map->num_stripes / map->sub_stripes; | |
1852 | offset = map->stripe_len * (num / map->sub_stripes); | |
1853 | increment = map->stripe_len * factor; | |
193ea74b | 1854 | mirror_num = num % map->sub_stripes + 1; |
a2de733c AJ |
1855 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { |
1856 | increment = map->stripe_len; | |
193ea74b | 1857 | mirror_num = num % map->num_stripes + 1; |
a2de733c AJ |
1858 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
1859 | increment = map->stripe_len; | |
193ea74b | 1860 | mirror_num = num % map->num_stripes + 1; |
a2de733c AJ |
1861 | } else { |
1862 | increment = map->stripe_len; | |
193ea74b | 1863 | mirror_num = 1; |
a2de733c AJ |
1864 | } |
1865 | ||
1866 | path = btrfs_alloc_path(); | |
1867 | if (!path) | |
1868 | return -ENOMEM; | |
1869 | ||
b5d67f64 SB |
1870 | /* |
1871 | * work on commit root. The related disk blocks are static as | |
1872 | * long as COW is applied. This means, it is save to rewrite | |
1873 | * them to repair disk errors without any race conditions | |
1874 | */ | |
a2de733c AJ |
1875 | path->search_commit_root = 1; |
1876 | path->skip_locking = 1; | |
1877 | ||
1878 | /* | |
7a26285e AJ |
1879 | * trigger the readahead for extent tree csum tree and wait for |
1880 | * completion. During readahead, the scrub is officially paused | |
1881 | * to not hold off transaction commits | |
a2de733c AJ |
1882 | */ |
1883 | logical = base + offset; | |
a2de733c | 1884 | |
d9d181c1 SB |
1885 | wait_event(sctx->list_wait, |
1886 | atomic_read(&sctx->in_flight) == 0); | |
7a26285e AJ |
1887 | atomic_inc(&fs_info->scrubs_paused); |
1888 | wake_up(&fs_info->scrub_pause_wait); | |
1889 | ||
1890 | /* FIXME it might be better to start readahead at commit root */ | |
1891 | key_start.objectid = logical; | |
1892 | key_start.type = BTRFS_EXTENT_ITEM_KEY; | |
1893 | key_start.offset = (u64)0; | |
1894 | key_end.objectid = base + offset + nstripes * increment; | |
1895 | key_end.type = BTRFS_EXTENT_ITEM_KEY; | |
1896 | key_end.offset = (u64)0; | |
1897 | reada1 = btrfs_reada_add(root, &key_start, &key_end); | |
1898 | ||
1899 | key_start.objectid = BTRFS_EXTENT_CSUM_OBJECTID; | |
1900 | key_start.type = BTRFS_EXTENT_CSUM_KEY; | |
1901 | key_start.offset = logical; | |
1902 | key_end.objectid = BTRFS_EXTENT_CSUM_OBJECTID; | |
1903 | key_end.type = BTRFS_EXTENT_CSUM_KEY; | |
1904 | key_end.offset = base + offset + nstripes * increment; | |
1905 | reada2 = btrfs_reada_add(csum_root, &key_start, &key_end); | |
1906 | ||
1907 | if (!IS_ERR(reada1)) | |
1908 | btrfs_reada_wait(reada1); | |
1909 | if (!IS_ERR(reada2)) | |
1910 | btrfs_reada_wait(reada2); | |
1911 | ||
1912 | mutex_lock(&fs_info->scrub_lock); | |
1913 | while (atomic_read(&fs_info->scrub_pause_req)) { | |
1914 | mutex_unlock(&fs_info->scrub_lock); | |
1915 | wait_event(fs_info->scrub_pause_wait, | |
1916 | atomic_read(&fs_info->scrub_pause_req) == 0); | |
1917 | mutex_lock(&fs_info->scrub_lock); | |
a2de733c | 1918 | } |
7a26285e AJ |
1919 | atomic_dec(&fs_info->scrubs_paused); |
1920 | mutex_unlock(&fs_info->scrub_lock); | |
1921 | wake_up(&fs_info->scrub_pause_wait); | |
a2de733c AJ |
1922 | |
1923 | /* | |
1924 | * collect all data csums for the stripe to avoid seeking during | |
1925 | * the scrub. This might currently (crc32) end up to be about 1MB | |
1926 | */ | |
e7786c3a | 1927 | blk_start_plug(&plug); |
a2de733c | 1928 | |
a2de733c AJ |
1929 | /* |
1930 | * now find all extents for each stripe and scrub them | |
1931 | */ | |
7a26285e AJ |
1932 | logical = base + offset; |
1933 | physical = map->stripes[num].physical; | |
a2de733c | 1934 | ret = 0; |
7a26285e | 1935 | for (i = 0; i < nstripes; ++i) { |
a2de733c AJ |
1936 | /* |
1937 | * canceled? | |
1938 | */ | |
1939 | if (atomic_read(&fs_info->scrub_cancel_req) || | |
d9d181c1 | 1940 | atomic_read(&sctx->cancel_req)) { |
a2de733c AJ |
1941 | ret = -ECANCELED; |
1942 | goto out; | |
1943 | } | |
1944 | /* | |
1945 | * check to see if we have to pause | |
1946 | */ | |
1947 | if (atomic_read(&fs_info->scrub_pause_req)) { | |
1948 | /* push queued extents */ | |
d9d181c1 SB |
1949 | scrub_submit(sctx); |
1950 | wait_event(sctx->list_wait, | |
1951 | atomic_read(&sctx->in_flight) == 0); | |
a2de733c AJ |
1952 | atomic_inc(&fs_info->scrubs_paused); |
1953 | wake_up(&fs_info->scrub_pause_wait); | |
1954 | mutex_lock(&fs_info->scrub_lock); | |
1955 | while (atomic_read(&fs_info->scrub_pause_req)) { | |
1956 | mutex_unlock(&fs_info->scrub_lock); | |
1957 | wait_event(fs_info->scrub_pause_wait, | |
1958 | atomic_read(&fs_info->scrub_pause_req) == 0); | |
1959 | mutex_lock(&fs_info->scrub_lock); | |
1960 | } | |
1961 | atomic_dec(&fs_info->scrubs_paused); | |
1962 | mutex_unlock(&fs_info->scrub_lock); | |
1963 | wake_up(&fs_info->scrub_pause_wait); | |
a2de733c AJ |
1964 | } |
1965 | ||
7a26285e AJ |
1966 | ret = btrfs_lookup_csums_range(csum_root, logical, |
1967 | logical + map->stripe_len - 1, | |
d9d181c1 | 1968 | &sctx->csum_list, 1); |
7a26285e AJ |
1969 | if (ret) |
1970 | goto out; | |
1971 | ||
a2de733c AJ |
1972 | key.objectid = logical; |
1973 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
1974 | key.offset = (u64)0; | |
1975 | ||
1976 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1977 | if (ret < 0) | |
1978 | goto out; | |
8c51032f | 1979 | if (ret > 0) { |
a2de733c AJ |
1980 | ret = btrfs_previous_item(root, path, 0, |
1981 | BTRFS_EXTENT_ITEM_KEY); | |
1982 | if (ret < 0) | |
1983 | goto out; | |
8c51032f AJ |
1984 | if (ret > 0) { |
1985 | /* there's no smaller item, so stick with the | |
1986 | * larger one */ | |
1987 | btrfs_release_path(path); | |
1988 | ret = btrfs_search_slot(NULL, root, &key, | |
1989 | path, 0, 0); | |
1990 | if (ret < 0) | |
1991 | goto out; | |
1992 | } | |
a2de733c AJ |
1993 | } |
1994 | ||
1995 | while (1) { | |
1996 | l = path->nodes[0]; | |
1997 | slot = path->slots[0]; | |
1998 | if (slot >= btrfs_header_nritems(l)) { | |
1999 | ret = btrfs_next_leaf(root, path); | |
2000 | if (ret == 0) | |
2001 | continue; | |
2002 | if (ret < 0) | |
2003 | goto out; | |
2004 | ||
2005 | break; | |
2006 | } | |
2007 | btrfs_item_key_to_cpu(l, &key, slot); | |
2008 | ||
2009 | if (key.objectid + key.offset <= logical) | |
2010 | goto next; | |
2011 | ||
2012 | if (key.objectid >= logical + map->stripe_len) | |
2013 | break; | |
2014 | ||
2015 | if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY) | |
2016 | goto next; | |
2017 | ||
2018 | extent = btrfs_item_ptr(l, slot, | |
2019 | struct btrfs_extent_item); | |
2020 | flags = btrfs_extent_flags(l, extent); | |
2021 | generation = btrfs_extent_generation(l, extent); | |
2022 | ||
2023 | if (key.objectid < logical && | |
2024 | (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) { | |
2025 | printk(KERN_ERR | |
2026 | "btrfs scrub: tree block %llu spanning " | |
2027 | "stripes, ignored. logical=%llu\n", | |
2028 | (unsigned long long)key.objectid, | |
2029 | (unsigned long long)logical); | |
2030 | goto next; | |
2031 | } | |
2032 | ||
2033 | /* | |
2034 | * trim extent to this stripe | |
2035 | */ | |
2036 | if (key.objectid < logical) { | |
2037 | key.offset -= logical - key.objectid; | |
2038 | key.objectid = logical; | |
2039 | } | |
2040 | if (key.objectid + key.offset > | |
2041 | logical + map->stripe_len) { | |
2042 | key.offset = logical + map->stripe_len - | |
2043 | key.objectid; | |
2044 | } | |
2045 | ||
d9d181c1 | 2046 | ret = scrub_extent(sctx, key.objectid, key.offset, |
a2de733c | 2047 | key.objectid - logical + physical, |
a36cf8b8 SB |
2048 | scrub_dev, flags, generation, |
2049 | mirror_num); | |
a2de733c AJ |
2050 | if (ret) |
2051 | goto out; | |
2052 | ||
2053 | next: | |
2054 | path->slots[0]++; | |
2055 | } | |
71267333 | 2056 | btrfs_release_path(path); |
a2de733c AJ |
2057 | logical += increment; |
2058 | physical += map->stripe_len; | |
d9d181c1 SB |
2059 | spin_lock(&sctx->stat_lock); |
2060 | sctx->stat.last_physical = physical; | |
2061 | spin_unlock(&sctx->stat_lock); | |
a2de733c AJ |
2062 | } |
2063 | /* push queued extents */ | |
d9d181c1 | 2064 | scrub_submit(sctx); |
a2de733c AJ |
2065 | |
2066 | out: | |
e7786c3a | 2067 | blk_finish_plug(&plug); |
a2de733c AJ |
2068 | btrfs_free_path(path); |
2069 | return ret < 0 ? ret : 0; | |
2070 | } | |
2071 | ||
d9d181c1 | 2072 | static noinline_for_stack int scrub_chunk(struct scrub_ctx *sctx, |
a36cf8b8 SB |
2073 | struct btrfs_device *scrub_dev, |
2074 | u64 chunk_tree, u64 chunk_objectid, | |
2075 | u64 chunk_offset, u64 length, | |
2076 | u64 dev_offset) | |
a2de733c AJ |
2077 | { |
2078 | struct btrfs_mapping_tree *map_tree = | |
a36cf8b8 | 2079 | &sctx->dev_root->fs_info->mapping_tree; |
a2de733c AJ |
2080 | struct map_lookup *map; |
2081 | struct extent_map *em; | |
2082 | int i; | |
2083 | int ret = -EINVAL; | |
2084 | ||
2085 | read_lock(&map_tree->map_tree.lock); | |
2086 | em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1); | |
2087 | read_unlock(&map_tree->map_tree.lock); | |
2088 | ||
2089 | if (!em) | |
2090 | return -EINVAL; | |
2091 | ||
2092 | map = (struct map_lookup *)em->bdev; | |
2093 | if (em->start != chunk_offset) | |
2094 | goto out; | |
2095 | ||
2096 | if (em->len < length) | |
2097 | goto out; | |
2098 | ||
2099 | for (i = 0; i < map->num_stripes; ++i) { | |
a36cf8b8 | 2100 | if (map->stripes[i].dev->bdev == scrub_dev->bdev && |
859acaf1 | 2101 | map->stripes[i].physical == dev_offset) { |
a36cf8b8 SB |
2102 | ret = scrub_stripe(sctx, map, scrub_dev, i, |
2103 | chunk_offset, length); | |
a2de733c AJ |
2104 | if (ret) |
2105 | goto out; | |
2106 | } | |
2107 | } | |
2108 | out: | |
2109 | free_extent_map(em); | |
2110 | ||
2111 | return ret; | |
2112 | } | |
2113 | ||
2114 | static noinline_for_stack | |
a36cf8b8 SB |
2115 | int scrub_enumerate_chunks(struct scrub_ctx *sctx, |
2116 | struct btrfs_device *scrub_dev, u64 start, u64 end) | |
a2de733c AJ |
2117 | { |
2118 | struct btrfs_dev_extent *dev_extent = NULL; | |
2119 | struct btrfs_path *path; | |
a36cf8b8 | 2120 | struct btrfs_root *root = sctx->dev_root; |
a2de733c AJ |
2121 | struct btrfs_fs_info *fs_info = root->fs_info; |
2122 | u64 length; | |
2123 | u64 chunk_tree; | |
2124 | u64 chunk_objectid; | |
2125 | u64 chunk_offset; | |
2126 | int ret; | |
2127 | int slot; | |
2128 | struct extent_buffer *l; | |
2129 | struct btrfs_key key; | |
2130 | struct btrfs_key found_key; | |
2131 | struct btrfs_block_group_cache *cache; | |
2132 | ||
2133 | path = btrfs_alloc_path(); | |
2134 | if (!path) | |
2135 | return -ENOMEM; | |
2136 | ||
2137 | path->reada = 2; | |
2138 | path->search_commit_root = 1; | |
2139 | path->skip_locking = 1; | |
2140 | ||
a36cf8b8 | 2141 | key.objectid = scrub_dev->devid; |
a2de733c AJ |
2142 | key.offset = 0ull; |
2143 | key.type = BTRFS_DEV_EXTENT_KEY; | |
2144 | ||
a2de733c AJ |
2145 | while (1) { |
2146 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2147 | if (ret < 0) | |
8c51032f AJ |
2148 | break; |
2149 | if (ret > 0) { | |
2150 | if (path->slots[0] >= | |
2151 | btrfs_header_nritems(path->nodes[0])) { | |
2152 | ret = btrfs_next_leaf(root, path); | |
2153 | if (ret) | |
2154 | break; | |
2155 | } | |
2156 | } | |
a2de733c AJ |
2157 | |
2158 | l = path->nodes[0]; | |
2159 | slot = path->slots[0]; | |
2160 | ||
2161 | btrfs_item_key_to_cpu(l, &found_key, slot); | |
2162 | ||
a36cf8b8 | 2163 | if (found_key.objectid != scrub_dev->devid) |
a2de733c AJ |
2164 | break; |
2165 | ||
8c51032f | 2166 | if (btrfs_key_type(&found_key) != BTRFS_DEV_EXTENT_KEY) |
a2de733c AJ |
2167 | break; |
2168 | ||
2169 | if (found_key.offset >= end) | |
2170 | break; | |
2171 | ||
2172 | if (found_key.offset < key.offset) | |
2173 | break; | |
2174 | ||
2175 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
2176 | length = btrfs_dev_extent_length(l, dev_extent); | |
2177 | ||
2178 | if (found_key.offset + length <= start) { | |
2179 | key.offset = found_key.offset + length; | |
71267333 | 2180 | btrfs_release_path(path); |
a2de733c AJ |
2181 | continue; |
2182 | } | |
2183 | ||
2184 | chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent); | |
2185 | chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent); | |
2186 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); | |
2187 | ||
2188 | /* | |
2189 | * get a reference on the corresponding block group to prevent | |
2190 | * the chunk from going away while we scrub it | |
2191 | */ | |
2192 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
2193 | if (!cache) { | |
2194 | ret = -ENOENT; | |
8c51032f | 2195 | break; |
a2de733c | 2196 | } |
a36cf8b8 | 2197 | ret = scrub_chunk(sctx, scrub_dev, chunk_tree, chunk_objectid, |
859acaf1 | 2198 | chunk_offset, length, found_key.offset); |
a2de733c AJ |
2199 | btrfs_put_block_group(cache); |
2200 | if (ret) | |
2201 | break; | |
2202 | ||
2203 | key.offset = found_key.offset + length; | |
71267333 | 2204 | btrfs_release_path(path); |
a2de733c AJ |
2205 | } |
2206 | ||
a2de733c | 2207 | btrfs_free_path(path); |
8c51032f AJ |
2208 | |
2209 | /* | |
2210 | * ret can still be 1 from search_slot or next_leaf, | |
2211 | * that's not an error | |
2212 | */ | |
2213 | return ret < 0 ? ret : 0; | |
a2de733c AJ |
2214 | } |
2215 | ||
a36cf8b8 SB |
2216 | static noinline_for_stack int scrub_supers(struct scrub_ctx *sctx, |
2217 | struct btrfs_device *scrub_dev) | |
a2de733c AJ |
2218 | { |
2219 | int i; | |
2220 | u64 bytenr; | |
2221 | u64 gen; | |
2222 | int ret; | |
a36cf8b8 | 2223 | struct btrfs_root *root = sctx->dev_root; |
a2de733c | 2224 | |
79787eaa JM |
2225 | if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) |
2226 | return -EIO; | |
2227 | ||
a2de733c AJ |
2228 | gen = root->fs_info->last_trans_committed; |
2229 | ||
2230 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { | |
2231 | bytenr = btrfs_sb_offset(i); | |
a36cf8b8 | 2232 | if (bytenr + BTRFS_SUPER_INFO_SIZE > scrub_dev->total_bytes) |
a2de733c AJ |
2233 | break; |
2234 | ||
d9d181c1 | 2235 | ret = scrub_pages(sctx, bytenr, BTRFS_SUPER_INFO_SIZE, bytenr, |
a36cf8b8 SB |
2236 | scrub_dev, BTRFS_EXTENT_FLAG_SUPER, gen, i, |
2237 | NULL, 1); | |
a2de733c AJ |
2238 | if (ret) |
2239 | return ret; | |
2240 | } | |
d9d181c1 | 2241 | wait_event(sctx->list_wait, atomic_read(&sctx->in_flight) == 0); |
a2de733c AJ |
2242 | |
2243 | return 0; | |
2244 | } | |
2245 | ||
2246 | /* | |
2247 | * get a reference count on fs_info->scrub_workers. start worker if necessary | |
2248 | */ | |
2249 | static noinline_for_stack int scrub_workers_get(struct btrfs_root *root) | |
2250 | { | |
2251 | struct btrfs_fs_info *fs_info = root->fs_info; | |
0dc3b84a | 2252 | int ret = 0; |
a2de733c AJ |
2253 | |
2254 | mutex_lock(&fs_info->scrub_lock); | |
632dd772 AJ |
2255 | if (fs_info->scrub_workers_refcnt == 0) { |
2256 | btrfs_init_workers(&fs_info->scrub_workers, "scrub", | |
2257 | fs_info->thread_pool_size, &fs_info->generic_worker); | |
2258 | fs_info->scrub_workers.idle_thresh = 4; | |
0dc3b84a JB |
2259 | ret = btrfs_start_workers(&fs_info->scrub_workers); |
2260 | if (ret) | |
2261 | goto out; | |
632dd772 | 2262 | } |
a2de733c | 2263 | ++fs_info->scrub_workers_refcnt; |
0dc3b84a | 2264 | out: |
a2de733c AJ |
2265 | mutex_unlock(&fs_info->scrub_lock); |
2266 | ||
0dc3b84a | 2267 | return ret; |
a2de733c AJ |
2268 | } |
2269 | ||
2270 | static noinline_for_stack void scrub_workers_put(struct btrfs_root *root) | |
2271 | { | |
2272 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2273 | ||
2274 | mutex_lock(&fs_info->scrub_lock); | |
2275 | if (--fs_info->scrub_workers_refcnt == 0) | |
2276 | btrfs_stop_workers(&fs_info->scrub_workers); | |
2277 | WARN_ON(fs_info->scrub_workers_refcnt < 0); | |
2278 | mutex_unlock(&fs_info->scrub_lock); | |
2279 | } | |
2280 | ||
2281 | ||
2282 | int btrfs_scrub_dev(struct btrfs_root *root, u64 devid, u64 start, u64 end, | |
8628764e | 2283 | struct btrfs_scrub_progress *progress, int readonly) |
a2de733c | 2284 | { |
d9d181c1 | 2285 | struct scrub_ctx *sctx; |
a2de733c AJ |
2286 | struct btrfs_fs_info *fs_info = root->fs_info; |
2287 | int ret; | |
2288 | struct btrfs_device *dev; | |
2289 | ||
7841cb28 | 2290 | if (btrfs_fs_closing(root->fs_info)) |
a2de733c AJ |
2291 | return -EINVAL; |
2292 | ||
2293 | /* | |
2294 | * check some assumptions | |
2295 | */ | |
b5d67f64 SB |
2296 | if (root->nodesize != root->leafsize) { |
2297 | printk(KERN_ERR | |
2298 | "btrfs_scrub: size assumption nodesize == leafsize (%d == %d) fails\n", | |
2299 | root->nodesize, root->leafsize); | |
2300 | return -EINVAL; | |
2301 | } | |
2302 | ||
2303 | if (root->nodesize > BTRFS_STRIPE_LEN) { | |
2304 | /* | |
2305 | * in this case scrub is unable to calculate the checksum | |
2306 | * the way scrub is implemented. Do not handle this | |
2307 | * situation at all because it won't ever happen. | |
2308 | */ | |
2309 | printk(KERN_ERR | |
2310 | "btrfs_scrub: size assumption nodesize <= BTRFS_STRIPE_LEN (%d <= %d) fails\n", | |
2311 | root->nodesize, BTRFS_STRIPE_LEN); | |
2312 | return -EINVAL; | |
2313 | } | |
2314 | ||
2315 | if (root->sectorsize != PAGE_SIZE) { | |
2316 | /* not supported for data w/o checksums */ | |
2317 | printk(KERN_ERR | |
2318 | "btrfs_scrub: size assumption sectorsize != PAGE_SIZE (%d != %lld) fails\n", | |
2319 | root->sectorsize, (unsigned long long)PAGE_SIZE); | |
a2de733c AJ |
2320 | return -EINVAL; |
2321 | } | |
2322 | ||
7a9e9987 SB |
2323 | if (fs_info->chunk_root->nodesize > |
2324 | PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK || | |
2325 | fs_info->chunk_root->sectorsize > | |
2326 | PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK) { | |
2327 | /* | |
2328 | * would exhaust the array bounds of pagev member in | |
2329 | * struct scrub_block | |
2330 | */ | |
2331 | pr_err("btrfs_scrub: size assumption nodesize and sectorsize <= SCRUB_MAX_PAGES_PER_BLOCK (%d <= %d && %d <= %d) fails\n", | |
2332 | fs_info->chunk_root->nodesize, | |
2333 | SCRUB_MAX_PAGES_PER_BLOCK, | |
2334 | fs_info->chunk_root->sectorsize, | |
2335 | SCRUB_MAX_PAGES_PER_BLOCK); | |
2336 | return -EINVAL; | |
2337 | } | |
2338 | ||
a2de733c AJ |
2339 | ret = scrub_workers_get(root); |
2340 | if (ret) | |
2341 | return ret; | |
2342 | ||
2343 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
2344 | dev = btrfs_find_device(root, devid, NULL, NULL); | |
2345 | if (!dev || dev->missing) { | |
2346 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); | |
2347 | scrub_workers_put(root); | |
2348 | return -ENODEV; | |
2349 | } | |
2350 | mutex_lock(&fs_info->scrub_lock); | |
2351 | ||
2352 | if (!dev->in_fs_metadata) { | |
2353 | mutex_unlock(&fs_info->scrub_lock); | |
2354 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); | |
2355 | scrub_workers_put(root); | |
2356 | return -ENODEV; | |
2357 | } | |
2358 | ||
2359 | if (dev->scrub_device) { | |
2360 | mutex_unlock(&fs_info->scrub_lock); | |
2361 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); | |
2362 | scrub_workers_put(root); | |
2363 | return -EINPROGRESS; | |
2364 | } | |
d9d181c1 SB |
2365 | sctx = scrub_setup_ctx(dev); |
2366 | if (IS_ERR(sctx)) { | |
a2de733c AJ |
2367 | mutex_unlock(&fs_info->scrub_lock); |
2368 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); | |
2369 | scrub_workers_put(root); | |
d9d181c1 | 2370 | return PTR_ERR(sctx); |
a2de733c | 2371 | } |
d9d181c1 SB |
2372 | sctx->readonly = readonly; |
2373 | dev->scrub_device = sctx; | |
a2de733c AJ |
2374 | |
2375 | atomic_inc(&fs_info->scrubs_running); | |
2376 | mutex_unlock(&fs_info->scrub_lock); | |
2377 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); | |
2378 | ||
2379 | down_read(&fs_info->scrub_super_lock); | |
a36cf8b8 | 2380 | ret = scrub_supers(sctx, dev); |
a2de733c AJ |
2381 | up_read(&fs_info->scrub_super_lock); |
2382 | ||
2383 | if (!ret) | |
a36cf8b8 | 2384 | ret = scrub_enumerate_chunks(sctx, dev, start, end); |
a2de733c | 2385 | |
d9d181c1 | 2386 | wait_event(sctx->list_wait, atomic_read(&sctx->in_flight) == 0); |
a2de733c AJ |
2387 | atomic_dec(&fs_info->scrubs_running); |
2388 | wake_up(&fs_info->scrub_pause_wait); | |
2389 | ||
d9d181c1 | 2390 | wait_event(sctx->list_wait, atomic_read(&sctx->fixup_cnt) == 0); |
0ef8e451 | 2391 | |
a2de733c | 2392 | if (progress) |
d9d181c1 | 2393 | memcpy(progress, &sctx->stat, sizeof(*progress)); |
a2de733c AJ |
2394 | |
2395 | mutex_lock(&fs_info->scrub_lock); | |
2396 | dev->scrub_device = NULL; | |
2397 | mutex_unlock(&fs_info->scrub_lock); | |
2398 | ||
d9d181c1 | 2399 | scrub_free_ctx(sctx); |
a2de733c AJ |
2400 | scrub_workers_put(root); |
2401 | ||
2402 | return ret; | |
2403 | } | |
2404 | ||
143bede5 | 2405 | void btrfs_scrub_pause(struct btrfs_root *root) |
a2de733c AJ |
2406 | { |
2407 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2408 | ||
2409 | mutex_lock(&fs_info->scrub_lock); | |
2410 | atomic_inc(&fs_info->scrub_pause_req); | |
2411 | while (atomic_read(&fs_info->scrubs_paused) != | |
2412 | atomic_read(&fs_info->scrubs_running)) { | |
2413 | mutex_unlock(&fs_info->scrub_lock); | |
2414 | wait_event(fs_info->scrub_pause_wait, | |
2415 | atomic_read(&fs_info->scrubs_paused) == | |
2416 | atomic_read(&fs_info->scrubs_running)); | |
2417 | mutex_lock(&fs_info->scrub_lock); | |
2418 | } | |
2419 | mutex_unlock(&fs_info->scrub_lock); | |
a2de733c AJ |
2420 | } |
2421 | ||
143bede5 | 2422 | void btrfs_scrub_continue(struct btrfs_root *root) |
a2de733c AJ |
2423 | { |
2424 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2425 | ||
2426 | atomic_dec(&fs_info->scrub_pause_req); | |
2427 | wake_up(&fs_info->scrub_pause_wait); | |
a2de733c AJ |
2428 | } |
2429 | ||
143bede5 | 2430 | void btrfs_scrub_pause_super(struct btrfs_root *root) |
a2de733c AJ |
2431 | { |
2432 | down_write(&root->fs_info->scrub_super_lock); | |
a2de733c AJ |
2433 | } |
2434 | ||
143bede5 | 2435 | void btrfs_scrub_continue_super(struct btrfs_root *root) |
a2de733c AJ |
2436 | { |
2437 | up_write(&root->fs_info->scrub_super_lock); | |
a2de733c AJ |
2438 | } |
2439 | ||
49b25e05 | 2440 | int __btrfs_scrub_cancel(struct btrfs_fs_info *fs_info) |
a2de733c | 2441 | { |
a2de733c AJ |
2442 | |
2443 | mutex_lock(&fs_info->scrub_lock); | |
2444 | if (!atomic_read(&fs_info->scrubs_running)) { | |
2445 | mutex_unlock(&fs_info->scrub_lock); | |
2446 | return -ENOTCONN; | |
2447 | } | |
2448 | ||
2449 | atomic_inc(&fs_info->scrub_cancel_req); | |
2450 | while (atomic_read(&fs_info->scrubs_running)) { | |
2451 | mutex_unlock(&fs_info->scrub_lock); | |
2452 | wait_event(fs_info->scrub_pause_wait, | |
2453 | atomic_read(&fs_info->scrubs_running) == 0); | |
2454 | mutex_lock(&fs_info->scrub_lock); | |
2455 | } | |
2456 | atomic_dec(&fs_info->scrub_cancel_req); | |
2457 | mutex_unlock(&fs_info->scrub_lock); | |
2458 | ||
2459 | return 0; | |
2460 | } | |
2461 | ||
49b25e05 JM |
2462 | int btrfs_scrub_cancel(struct btrfs_root *root) |
2463 | { | |
2464 | return __btrfs_scrub_cancel(root->fs_info); | |
2465 | } | |
2466 | ||
a2de733c AJ |
2467 | int btrfs_scrub_cancel_dev(struct btrfs_root *root, struct btrfs_device *dev) |
2468 | { | |
2469 | struct btrfs_fs_info *fs_info = root->fs_info; | |
d9d181c1 | 2470 | struct scrub_ctx *sctx; |
a2de733c AJ |
2471 | |
2472 | mutex_lock(&fs_info->scrub_lock); | |
d9d181c1 SB |
2473 | sctx = dev->scrub_device; |
2474 | if (!sctx) { | |
a2de733c AJ |
2475 | mutex_unlock(&fs_info->scrub_lock); |
2476 | return -ENOTCONN; | |
2477 | } | |
d9d181c1 | 2478 | atomic_inc(&sctx->cancel_req); |
a2de733c AJ |
2479 | while (dev->scrub_device) { |
2480 | mutex_unlock(&fs_info->scrub_lock); | |
2481 | wait_event(fs_info->scrub_pause_wait, | |
2482 | dev->scrub_device == NULL); | |
2483 | mutex_lock(&fs_info->scrub_lock); | |
2484 | } | |
2485 | mutex_unlock(&fs_info->scrub_lock); | |
2486 | ||
2487 | return 0; | |
2488 | } | |
1623edeb | 2489 | |
a2de733c AJ |
2490 | int btrfs_scrub_cancel_devid(struct btrfs_root *root, u64 devid) |
2491 | { | |
2492 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2493 | struct btrfs_device *dev; | |
2494 | int ret; | |
2495 | ||
2496 | /* | |
2497 | * we have to hold the device_list_mutex here so the device | |
2498 | * does not go away in cancel_dev. FIXME: find a better solution | |
2499 | */ | |
2500 | mutex_lock(&fs_info->fs_devices->device_list_mutex); | |
2501 | dev = btrfs_find_device(root, devid, NULL, NULL); | |
2502 | if (!dev) { | |
2503 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
2504 | return -ENODEV; | |
2505 | } | |
2506 | ret = btrfs_scrub_cancel_dev(root, dev); | |
2507 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
2508 | ||
2509 | return ret; | |
2510 | } | |
2511 | ||
2512 | int btrfs_scrub_progress(struct btrfs_root *root, u64 devid, | |
2513 | struct btrfs_scrub_progress *progress) | |
2514 | { | |
2515 | struct btrfs_device *dev; | |
d9d181c1 | 2516 | struct scrub_ctx *sctx = NULL; |
a2de733c AJ |
2517 | |
2518 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
2519 | dev = btrfs_find_device(root, devid, NULL, NULL); | |
2520 | if (dev) | |
d9d181c1 SB |
2521 | sctx = dev->scrub_device; |
2522 | if (sctx) | |
2523 | memcpy(progress, &sctx->stat, sizeof(*progress)); | |
a2de733c AJ |
2524 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
2525 | ||
d9d181c1 | 2526 | return dev ? (sctx ? 0 : -ENOTCONN) : -ENODEV; |
a2de733c | 2527 | } |