Commit | Line | Data |
---|---|---|
a2de733c | 1 | /* |
b6bfebc1 | 2 | * Copyright (C) 2011, 2012 STRATO. All rights reserved. |
a2de733c AJ |
3 | * |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
a2de733c | 19 | #include <linux/blkdev.h> |
558540c1 | 20 | #include <linux/ratelimit.h> |
a2de733c AJ |
21 | #include "ctree.h" |
22 | #include "volumes.h" | |
23 | #include "disk-io.h" | |
24 | #include "ordered-data.h" | |
0ef8e451 | 25 | #include "transaction.h" |
558540c1 | 26 | #include "backref.h" |
5da6fcbc | 27 | #include "extent_io.h" |
ff023aac | 28 | #include "dev-replace.h" |
21adbd5c | 29 | #include "check-integrity.h" |
606686ee | 30 | #include "rcu-string.h" |
53b381b3 | 31 | #include "raid56.h" |
a2de733c AJ |
32 | |
33 | /* | |
34 | * This is only the first step towards a full-features scrub. It reads all | |
35 | * extent and super block and verifies the checksums. In case a bad checksum | |
36 | * is found or the extent cannot be read, good data will be written back if | |
37 | * any can be found. | |
38 | * | |
39 | * Future enhancements: | |
a2de733c AJ |
40 | * - In case an unrepairable extent is encountered, track which files are |
41 | * affected and report them | |
a2de733c | 42 | * - track and record media errors, throw out bad devices |
a2de733c | 43 | * - add a mode to also read unallocated space |
a2de733c AJ |
44 | */ |
45 | ||
b5d67f64 | 46 | struct scrub_block; |
d9d181c1 | 47 | struct scrub_ctx; |
a2de733c | 48 | |
ff023aac SB |
49 | /* |
50 | * the following three values only influence the performance. | |
51 | * The last one configures the number of parallel and outstanding I/O | |
52 | * operations. The first two values configure an upper limit for the number | |
53 | * of (dynamically allocated) pages that are added to a bio. | |
54 | */ | |
55 | #define SCRUB_PAGES_PER_RD_BIO 32 /* 128k per bio */ | |
56 | #define SCRUB_PAGES_PER_WR_BIO 32 /* 128k per bio */ | |
57 | #define SCRUB_BIOS_PER_SCTX 64 /* 8MB per device in flight */ | |
7a9e9987 SB |
58 | |
59 | /* | |
60 | * the following value times PAGE_SIZE needs to be large enough to match the | |
61 | * largest node/leaf/sector size that shall be supported. | |
62 | * Values larger than BTRFS_STRIPE_LEN are not supported. | |
63 | */ | |
b5d67f64 | 64 | #define SCRUB_MAX_PAGES_PER_BLOCK 16 /* 64k per node/leaf/sector */ |
a2de733c | 65 | |
af8e2d1d MX |
66 | struct scrub_recover { |
67 | atomic_t refs; | |
68 | struct btrfs_bio *bbio; | |
af8e2d1d MX |
69 | u64 map_length; |
70 | }; | |
71 | ||
a2de733c | 72 | struct scrub_page { |
b5d67f64 SB |
73 | struct scrub_block *sblock; |
74 | struct page *page; | |
442a4f63 | 75 | struct btrfs_device *dev; |
5a6ac9ea | 76 | struct list_head list; |
a2de733c AJ |
77 | u64 flags; /* extent flags */ |
78 | u64 generation; | |
b5d67f64 SB |
79 | u64 logical; |
80 | u64 physical; | |
ff023aac | 81 | u64 physical_for_dev_replace; |
57019345 | 82 | atomic_t refs; |
b5d67f64 SB |
83 | struct { |
84 | unsigned int mirror_num:8; | |
85 | unsigned int have_csum:1; | |
86 | unsigned int io_error:1; | |
87 | }; | |
a2de733c | 88 | u8 csum[BTRFS_CSUM_SIZE]; |
af8e2d1d MX |
89 | |
90 | struct scrub_recover *recover; | |
a2de733c AJ |
91 | }; |
92 | ||
93 | struct scrub_bio { | |
94 | int index; | |
d9d181c1 | 95 | struct scrub_ctx *sctx; |
a36cf8b8 | 96 | struct btrfs_device *dev; |
a2de733c AJ |
97 | struct bio *bio; |
98 | int err; | |
99 | u64 logical; | |
100 | u64 physical; | |
ff023aac SB |
101 | #if SCRUB_PAGES_PER_WR_BIO >= SCRUB_PAGES_PER_RD_BIO |
102 | struct scrub_page *pagev[SCRUB_PAGES_PER_WR_BIO]; | |
103 | #else | |
104 | struct scrub_page *pagev[SCRUB_PAGES_PER_RD_BIO]; | |
105 | #endif | |
b5d67f64 | 106 | int page_count; |
a2de733c AJ |
107 | int next_free; |
108 | struct btrfs_work work; | |
109 | }; | |
110 | ||
b5d67f64 | 111 | struct scrub_block { |
7a9e9987 | 112 | struct scrub_page *pagev[SCRUB_MAX_PAGES_PER_BLOCK]; |
b5d67f64 SB |
113 | int page_count; |
114 | atomic_t outstanding_pages; | |
57019345 | 115 | atomic_t refs; /* free mem on transition to zero */ |
d9d181c1 | 116 | struct scrub_ctx *sctx; |
5a6ac9ea | 117 | struct scrub_parity *sparity; |
b5d67f64 SB |
118 | struct { |
119 | unsigned int header_error:1; | |
120 | unsigned int checksum_error:1; | |
121 | unsigned int no_io_error_seen:1; | |
442a4f63 | 122 | unsigned int generation_error:1; /* also sets header_error */ |
5a6ac9ea MX |
123 | |
124 | /* The following is for the data used to check parity */ | |
125 | /* It is for the data with checksum */ | |
126 | unsigned int data_corrected:1; | |
b5d67f64 | 127 | }; |
73ff61db | 128 | struct btrfs_work work; |
b5d67f64 SB |
129 | }; |
130 | ||
5a6ac9ea MX |
131 | /* Used for the chunks with parity stripe such RAID5/6 */ |
132 | struct scrub_parity { | |
133 | struct scrub_ctx *sctx; | |
134 | ||
135 | struct btrfs_device *scrub_dev; | |
136 | ||
137 | u64 logic_start; | |
138 | ||
139 | u64 logic_end; | |
140 | ||
141 | int nsectors; | |
142 | ||
143 | int stripe_len; | |
144 | ||
57019345 | 145 | atomic_t refs; |
5a6ac9ea MX |
146 | |
147 | struct list_head spages; | |
148 | ||
149 | /* Work of parity check and repair */ | |
150 | struct btrfs_work work; | |
151 | ||
152 | /* Mark the parity blocks which have data */ | |
153 | unsigned long *dbitmap; | |
154 | ||
155 | /* | |
156 | * Mark the parity blocks which have data, but errors happen when | |
157 | * read data or check data | |
158 | */ | |
159 | unsigned long *ebitmap; | |
160 | ||
161 | unsigned long bitmap[0]; | |
162 | }; | |
163 | ||
ff023aac SB |
164 | struct scrub_wr_ctx { |
165 | struct scrub_bio *wr_curr_bio; | |
166 | struct btrfs_device *tgtdev; | |
167 | int pages_per_wr_bio; /* <= SCRUB_PAGES_PER_WR_BIO */ | |
168 | atomic_t flush_all_writes; | |
169 | struct mutex wr_lock; | |
170 | }; | |
171 | ||
d9d181c1 | 172 | struct scrub_ctx { |
ff023aac | 173 | struct scrub_bio *bios[SCRUB_BIOS_PER_SCTX]; |
a36cf8b8 | 174 | struct btrfs_root *dev_root; |
a2de733c AJ |
175 | int first_free; |
176 | int curr; | |
b6bfebc1 SB |
177 | atomic_t bios_in_flight; |
178 | atomic_t workers_pending; | |
a2de733c AJ |
179 | spinlock_t list_lock; |
180 | wait_queue_head_t list_wait; | |
181 | u16 csum_size; | |
182 | struct list_head csum_list; | |
183 | atomic_t cancel_req; | |
8628764e | 184 | int readonly; |
ff023aac | 185 | int pages_per_rd_bio; |
b5d67f64 SB |
186 | u32 sectorsize; |
187 | u32 nodesize; | |
63a212ab SB |
188 | |
189 | int is_dev_replace; | |
ff023aac | 190 | struct scrub_wr_ctx wr_ctx; |
63a212ab | 191 | |
a2de733c AJ |
192 | /* |
193 | * statistics | |
194 | */ | |
195 | struct btrfs_scrub_progress stat; | |
196 | spinlock_t stat_lock; | |
f55985f4 FM |
197 | |
198 | /* | |
199 | * Use a ref counter to avoid use-after-free issues. Scrub workers | |
200 | * decrement bios_in_flight and workers_pending and then do a wakeup | |
201 | * on the list_wait wait queue. We must ensure the main scrub task | |
202 | * doesn't free the scrub context before or while the workers are | |
203 | * doing the wakeup() call. | |
204 | */ | |
205 | atomic_t refs; | |
a2de733c AJ |
206 | }; |
207 | ||
0ef8e451 | 208 | struct scrub_fixup_nodatasum { |
d9d181c1 | 209 | struct scrub_ctx *sctx; |
a36cf8b8 | 210 | struct btrfs_device *dev; |
0ef8e451 JS |
211 | u64 logical; |
212 | struct btrfs_root *root; | |
213 | struct btrfs_work work; | |
214 | int mirror_num; | |
215 | }; | |
216 | ||
652f25a2 JB |
217 | struct scrub_nocow_inode { |
218 | u64 inum; | |
219 | u64 offset; | |
220 | u64 root; | |
221 | struct list_head list; | |
222 | }; | |
223 | ||
ff023aac SB |
224 | struct scrub_copy_nocow_ctx { |
225 | struct scrub_ctx *sctx; | |
226 | u64 logical; | |
227 | u64 len; | |
228 | int mirror_num; | |
229 | u64 physical_for_dev_replace; | |
652f25a2 | 230 | struct list_head inodes; |
ff023aac SB |
231 | struct btrfs_work work; |
232 | }; | |
233 | ||
558540c1 JS |
234 | struct scrub_warning { |
235 | struct btrfs_path *path; | |
236 | u64 extent_item_size; | |
558540c1 JS |
237 | const char *errstr; |
238 | sector_t sector; | |
239 | u64 logical; | |
240 | struct btrfs_device *dev; | |
558540c1 JS |
241 | }; |
242 | ||
b6bfebc1 SB |
243 | static void scrub_pending_bio_inc(struct scrub_ctx *sctx); |
244 | static void scrub_pending_bio_dec(struct scrub_ctx *sctx); | |
245 | static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx); | |
246 | static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx); | |
b5d67f64 | 247 | static int scrub_handle_errored_block(struct scrub_block *sblock_to_check); |
be50a8dd | 248 | static int scrub_setup_recheck_block(struct scrub_block *original_sblock, |
ff023aac | 249 | struct scrub_block *sblocks_for_recheck); |
34f5c8e9 SB |
250 | static void scrub_recheck_block(struct btrfs_fs_info *fs_info, |
251 | struct scrub_block *sblock, int is_metadata, | |
252 | int have_csum, u8 *csum, u64 generation, | |
af8e2d1d | 253 | u16 csum_size, int retry_failed_mirror); |
ba7cf988 | 254 | static void scrub_recheck_block_checksum(struct scrub_block *sblock); |
b5d67f64 | 255 | static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad, |
114ab50d | 256 | struct scrub_block *sblock_good); |
b5d67f64 SB |
257 | static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad, |
258 | struct scrub_block *sblock_good, | |
259 | int page_num, int force_write); | |
ff023aac SB |
260 | static void scrub_write_block_to_dev_replace(struct scrub_block *sblock); |
261 | static int scrub_write_page_to_dev_replace(struct scrub_block *sblock, | |
262 | int page_num); | |
b5d67f64 SB |
263 | static int scrub_checksum_data(struct scrub_block *sblock); |
264 | static int scrub_checksum_tree_block(struct scrub_block *sblock); | |
265 | static int scrub_checksum_super(struct scrub_block *sblock); | |
266 | static void scrub_block_get(struct scrub_block *sblock); | |
267 | static void scrub_block_put(struct scrub_block *sblock); | |
7a9e9987 SB |
268 | static void scrub_page_get(struct scrub_page *spage); |
269 | static void scrub_page_put(struct scrub_page *spage); | |
5a6ac9ea MX |
270 | static void scrub_parity_get(struct scrub_parity *sparity); |
271 | static void scrub_parity_put(struct scrub_parity *sparity); | |
ff023aac SB |
272 | static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx, |
273 | struct scrub_page *spage); | |
d9d181c1 | 274 | static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len, |
a36cf8b8 | 275 | u64 physical, struct btrfs_device *dev, u64 flags, |
ff023aac SB |
276 | u64 gen, int mirror_num, u8 *csum, int force, |
277 | u64 physical_for_dev_replace); | |
4246a0b6 | 278 | static void scrub_bio_end_io(struct bio *bio); |
b5d67f64 SB |
279 | static void scrub_bio_end_io_worker(struct btrfs_work *work); |
280 | static void scrub_block_complete(struct scrub_block *sblock); | |
ff023aac SB |
281 | static void scrub_remap_extent(struct btrfs_fs_info *fs_info, |
282 | u64 extent_logical, u64 extent_len, | |
283 | u64 *extent_physical, | |
284 | struct btrfs_device **extent_dev, | |
285 | int *extent_mirror_num); | |
286 | static int scrub_setup_wr_ctx(struct scrub_ctx *sctx, | |
287 | struct scrub_wr_ctx *wr_ctx, | |
288 | struct btrfs_fs_info *fs_info, | |
289 | struct btrfs_device *dev, | |
290 | int is_dev_replace); | |
291 | static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx); | |
292 | static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx, | |
293 | struct scrub_page *spage); | |
294 | static void scrub_wr_submit(struct scrub_ctx *sctx); | |
4246a0b6 | 295 | static void scrub_wr_bio_end_io(struct bio *bio); |
ff023aac SB |
296 | static void scrub_wr_bio_end_io_worker(struct btrfs_work *work); |
297 | static int write_page_nocow(struct scrub_ctx *sctx, | |
298 | u64 physical_for_dev_replace, struct page *page); | |
299 | static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root, | |
652f25a2 | 300 | struct scrub_copy_nocow_ctx *ctx); |
ff023aac SB |
301 | static int copy_nocow_pages(struct scrub_ctx *sctx, u64 logical, u64 len, |
302 | int mirror_num, u64 physical_for_dev_replace); | |
303 | static void copy_nocow_pages_worker(struct btrfs_work *work); | |
cb7ab021 | 304 | static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info); |
3cb0929a | 305 | static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info); |
f55985f4 | 306 | static void scrub_put_ctx(struct scrub_ctx *sctx); |
1623edeb SB |
307 | |
308 | ||
b6bfebc1 SB |
309 | static void scrub_pending_bio_inc(struct scrub_ctx *sctx) |
310 | { | |
f55985f4 | 311 | atomic_inc(&sctx->refs); |
b6bfebc1 SB |
312 | atomic_inc(&sctx->bios_in_flight); |
313 | } | |
314 | ||
315 | static void scrub_pending_bio_dec(struct scrub_ctx *sctx) | |
316 | { | |
317 | atomic_dec(&sctx->bios_in_flight); | |
318 | wake_up(&sctx->list_wait); | |
f55985f4 | 319 | scrub_put_ctx(sctx); |
b6bfebc1 SB |
320 | } |
321 | ||
cb7ab021 | 322 | static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info) |
3cb0929a WS |
323 | { |
324 | while (atomic_read(&fs_info->scrub_pause_req)) { | |
325 | mutex_unlock(&fs_info->scrub_lock); | |
326 | wait_event(fs_info->scrub_pause_wait, | |
327 | atomic_read(&fs_info->scrub_pause_req) == 0); | |
328 | mutex_lock(&fs_info->scrub_lock); | |
329 | } | |
330 | } | |
331 | ||
0e22be89 | 332 | static void scrub_pause_on(struct btrfs_fs_info *fs_info) |
cb7ab021 WS |
333 | { |
334 | atomic_inc(&fs_info->scrubs_paused); | |
335 | wake_up(&fs_info->scrub_pause_wait); | |
0e22be89 | 336 | } |
cb7ab021 | 337 | |
0e22be89 Z |
338 | static void scrub_pause_off(struct btrfs_fs_info *fs_info) |
339 | { | |
cb7ab021 WS |
340 | mutex_lock(&fs_info->scrub_lock); |
341 | __scrub_blocked_if_needed(fs_info); | |
342 | atomic_dec(&fs_info->scrubs_paused); | |
343 | mutex_unlock(&fs_info->scrub_lock); | |
344 | ||
345 | wake_up(&fs_info->scrub_pause_wait); | |
346 | } | |
347 | ||
0e22be89 Z |
348 | static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info) |
349 | { | |
350 | scrub_pause_on(fs_info); | |
351 | scrub_pause_off(fs_info); | |
352 | } | |
353 | ||
b6bfebc1 SB |
354 | /* |
355 | * used for workers that require transaction commits (i.e., for the | |
356 | * NOCOW case) | |
357 | */ | |
358 | static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx) | |
359 | { | |
360 | struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info; | |
361 | ||
f55985f4 | 362 | atomic_inc(&sctx->refs); |
b6bfebc1 SB |
363 | /* |
364 | * increment scrubs_running to prevent cancel requests from | |
365 | * completing as long as a worker is running. we must also | |
366 | * increment scrubs_paused to prevent deadlocking on pause | |
367 | * requests used for transactions commits (as the worker uses a | |
368 | * transaction context). it is safe to regard the worker | |
369 | * as paused for all matters practical. effectively, we only | |
370 | * avoid cancellation requests from completing. | |
371 | */ | |
372 | mutex_lock(&fs_info->scrub_lock); | |
373 | atomic_inc(&fs_info->scrubs_running); | |
374 | atomic_inc(&fs_info->scrubs_paused); | |
375 | mutex_unlock(&fs_info->scrub_lock); | |
32a44789 WS |
376 | |
377 | /* | |
378 | * check if @scrubs_running=@scrubs_paused condition | |
379 | * inside wait_event() is not an atomic operation. | |
380 | * which means we may inc/dec @scrub_running/paused | |
381 | * at any time. Let's wake up @scrub_pause_wait as | |
382 | * much as we can to let commit transaction blocked less. | |
383 | */ | |
384 | wake_up(&fs_info->scrub_pause_wait); | |
385 | ||
b6bfebc1 SB |
386 | atomic_inc(&sctx->workers_pending); |
387 | } | |
388 | ||
389 | /* used for workers that require transaction commits */ | |
390 | static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx) | |
391 | { | |
392 | struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info; | |
393 | ||
394 | /* | |
395 | * see scrub_pending_trans_workers_inc() why we're pretending | |
396 | * to be paused in the scrub counters | |
397 | */ | |
398 | mutex_lock(&fs_info->scrub_lock); | |
399 | atomic_dec(&fs_info->scrubs_running); | |
400 | atomic_dec(&fs_info->scrubs_paused); | |
401 | mutex_unlock(&fs_info->scrub_lock); | |
402 | atomic_dec(&sctx->workers_pending); | |
403 | wake_up(&fs_info->scrub_pause_wait); | |
404 | wake_up(&sctx->list_wait); | |
f55985f4 | 405 | scrub_put_ctx(sctx); |
b6bfebc1 SB |
406 | } |
407 | ||
d9d181c1 | 408 | static void scrub_free_csums(struct scrub_ctx *sctx) |
a2de733c | 409 | { |
d9d181c1 | 410 | while (!list_empty(&sctx->csum_list)) { |
a2de733c | 411 | struct btrfs_ordered_sum *sum; |
d9d181c1 | 412 | sum = list_first_entry(&sctx->csum_list, |
a2de733c AJ |
413 | struct btrfs_ordered_sum, list); |
414 | list_del(&sum->list); | |
415 | kfree(sum); | |
416 | } | |
417 | } | |
418 | ||
d9d181c1 | 419 | static noinline_for_stack void scrub_free_ctx(struct scrub_ctx *sctx) |
a2de733c AJ |
420 | { |
421 | int i; | |
a2de733c | 422 | |
d9d181c1 | 423 | if (!sctx) |
a2de733c AJ |
424 | return; |
425 | ||
ff023aac SB |
426 | scrub_free_wr_ctx(&sctx->wr_ctx); |
427 | ||
b5d67f64 | 428 | /* this can happen when scrub is cancelled */ |
d9d181c1 SB |
429 | if (sctx->curr != -1) { |
430 | struct scrub_bio *sbio = sctx->bios[sctx->curr]; | |
b5d67f64 SB |
431 | |
432 | for (i = 0; i < sbio->page_count; i++) { | |
ff023aac | 433 | WARN_ON(!sbio->pagev[i]->page); |
b5d67f64 SB |
434 | scrub_block_put(sbio->pagev[i]->sblock); |
435 | } | |
436 | bio_put(sbio->bio); | |
437 | } | |
438 | ||
ff023aac | 439 | for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) { |
d9d181c1 | 440 | struct scrub_bio *sbio = sctx->bios[i]; |
a2de733c AJ |
441 | |
442 | if (!sbio) | |
443 | break; | |
a2de733c AJ |
444 | kfree(sbio); |
445 | } | |
446 | ||
d9d181c1 SB |
447 | scrub_free_csums(sctx); |
448 | kfree(sctx); | |
a2de733c AJ |
449 | } |
450 | ||
f55985f4 FM |
451 | static void scrub_put_ctx(struct scrub_ctx *sctx) |
452 | { | |
453 | if (atomic_dec_and_test(&sctx->refs)) | |
454 | scrub_free_ctx(sctx); | |
455 | } | |
456 | ||
a2de733c | 457 | static noinline_for_stack |
63a212ab | 458 | struct scrub_ctx *scrub_setup_ctx(struct btrfs_device *dev, int is_dev_replace) |
a2de733c | 459 | { |
d9d181c1 | 460 | struct scrub_ctx *sctx; |
a2de733c | 461 | int i; |
a2de733c | 462 | struct btrfs_fs_info *fs_info = dev->dev_root->fs_info; |
ff023aac | 463 | int ret; |
a2de733c | 464 | |
d9d181c1 SB |
465 | sctx = kzalloc(sizeof(*sctx), GFP_NOFS); |
466 | if (!sctx) | |
a2de733c | 467 | goto nomem; |
f55985f4 | 468 | atomic_set(&sctx->refs, 1); |
63a212ab | 469 | sctx->is_dev_replace = is_dev_replace; |
b54ffb73 | 470 | sctx->pages_per_rd_bio = SCRUB_PAGES_PER_RD_BIO; |
d9d181c1 | 471 | sctx->curr = -1; |
a36cf8b8 | 472 | sctx->dev_root = dev->dev_root; |
ff023aac | 473 | for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) { |
a2de733c AJ |
474 | struct scrub_bio *sbio; |
475 | ||
476 | sbio = kzalloc(sizeof(*sbio), GFP_NOFS); | |
477 | if (!sbio) | |
478 | goto nomem; | |
d9d181c1 | 479 | sctx->bios[i] = sbio; |
a2de733c | 480 | |
a2de733c | 481 | sbio->index = i; |
d9d181c1 | 482 | sbio->sctx = sctx; |
b5d67f64 | 483 | sbio->page_count = 0; |
9e0af237 LB |
484 | btrfs_init_work(&sbio->work, btrfs_scrub_helper, |
485 | scrub_bio_end_io_worker, NULL, NULL); | |
a2de733c | 486 | |
ff023aac | 487 | if (i != SCRUB_BIOS_PER_SCTX - 1) |
d9d181c1 | 488 | sctx->bios[i]->next_free = i + 1; |
0ef8e451 | 489 | else |
d9d181c1 SB |
490 | sctx->bios[i]->next_free = -1; |
491 | } | |
492 | sctx->first_free = 0; | |
493 | sctx->nodesize = dev->dev_root->nodesize; | |
d9d181c1 | 494 | sctx->sectorsize = dev->dev_root->sectorsize; |
b6bfebc1 SB |
495 | atomic_set(&sctx->bios_in_flight, 0); |
496 | atomic_set(&sctx->workers_pending, 0); | |
d9d181c1 SB |
497 | atomic_set(&sctx->cancel_req, 0); |
498 | sctx->csum_size = btrfs_super_csum_size(fs_info->super_copy); | |
499 | INIT_LIST_HEAD(&sctx->csum_list); | |
500 | ||
501 | spin_lock_init(&sctx->list_lock); | |
502 | spin_lock_init(&sctx->stat_lock); | |
503 | init_waitqueue_head(&sctx->list_wait); | |
ff023aac SB |
504 | |
505 | ret = scrub_setup_wr_ctx(sctx, &sctx->wr_ctx, fs_info, | |
506 | fs_info->dev_replace.tgtdev, is_dev_replace); | |
507 | if (ret) { | |
508 | scrub_free_ctx(sctx); | |
509 | return ERR_PTR(ret); | |
510 | } | |
d9d181c1 | 511 | return sctx; |
a2de733c AJ |
512 | |
513 | nomem: | |
d9d181c1 | 514 | scrub_free_ctx(sctx); |
a2de733c AJ |
515 | return ERR_PTR(-ENOMEM); |
516 | } | |
517 | ||
ff023aac SB |
518 | static int scrub_print_warning_inode(u64 inum, u64 offset, u64 root, |
519 | void *warn_ctx) | |
558540c1 JS |
520 | { |
521 | u64 isize; | |
522 | u32 nlink; | |
523 | int ret; | |
524 | int i; | |
525 | struct extent_buffer *eb; | |
526 | struct btrfs_inode_item *inode_item; | |
ff023aac | 527 | struct scrub_warning *swarn = warn_ctx; |
558540c1 JS |
528 | struct btrfs_fs_info *fs_info = swarn->dev->dev_root->fs_info; |
529 | struct inode_fs_paths *ipath = NULL; | |
530 | struct btrfs_root *local_root; | |
531 | struct btrfs_key root_key; | |
1d4c08e0 | 532 | struct btrfs_key key; |
558540c1 JS |
533 | |
534 | root_key.objectid = root; | |
535 | root_key.type = BTRFS_ROOT_ITEM_KEY; | |
536 | root_key.offset = (u64)-1; | |
537 | local_root = btrfs_read_fs_root_no_name(fs_info, &root_key); | |
538 | if (IS_ERR(local_root)) { | |
539 | ret = PTR_ERR(local_root); | |
540 | goto err; | |
541 | } | |
542 | ||
14692cc1 DS |
543 | /* |
544 | * this makes the path point to (inum INODE_ITEM ioff) | |
545 | */ | |
1d4c08e0 DS |
546 | key.objectid = inum; |
547 | key.type = BTRFS_INODE_ITEM_KEY; | |
548 | key.offset = 0; | |
549 | ||
550 | ret = btrfs_search_slot(NULL, local_root, &key, swarn->path, 0, 0); | |
558540c1 JS |
551 | if (ret) { |
552 | btrfs_release_path(swarn->path); | |
553 | goto err; | |
554 | } | |
555 | ||
556 | eb = swarn->path->nodes[0]; | |
557 | inode_item = btrfs_item_ptr(eb, swarn->path->slots[0], | |
558 | struct btrfs_inode_item); | |
559 | isize = btrfs_inode_size(eb, inode_item); | |
560 | nlink = btrfs_inode_nlink(eb, inode_item); | |
561 | btrfs_release_path(swarn->path); | |
562 | ||
563 | ipath = init_ipath(4096, local_root, swarn->path); | |
26bdef54 DC |
564 | if (IS_ERR(ipath)) { |
565 | ret = PTR_ERR(ipath); | |
566 | ipath = NULL; | |
567 | goto err; | |
568 | } | |
558540c1 JS |
569 | ret = paths_from_inode(inum, ipath); |
570 | ||
571 | if (ret < 0) | |
572 | goto err; | |
573 | ||
574 | /* | |
575 | * we deliberately ignore the bit ipath might have been too small to | |
576 | * hold all of the paths here | |
577 | */ | |
578 | for (i = 0; i < ipath->fspath->elem_cnt; ++i) | |
ecaeb14b | 579 | btrfs_warn_in_rcu(fs_info, "%s at logical %llu on dev " |
558540c1 | 580 | "%s, sector %llu, root %llu, inode %llu, offset %llu, " |
ecaeb14b | 581 | "length %llu, links %u (path: %s)", swarn->errstr, |
606686ee | 582 | swarn->logical, rcu_str_deref(swarn->dev->name), |
558540c1 JS |
583 | (unsigned long long)swarn->sector, root, inum, offset, |
584 | min(isize - offset, (u64)PAGE_SIZE), nlink, | |
745c4d8e | 585 | (char *)(unsigned long)ipath->fspath->val[i]); |
558540c1 JS |
586 | |
587 | free_ipath(ipath); | |
588 | return 0; | |
589 | ||
590 | err: | |
ecaeb14b | 591 | btrfs_warn_in_rcu(fs_info, "%s at logical %llu on dev " |
558540c1 | 592 | "%s, sector %llu, root %llu, inode %llu, offset %llu: path " |
ecaeb14b | 593 | "resolving failed with ret=%d", swarn->errstr, |
606686ee | 594 | swarn->logical, rcu_str_deref(swarn->dev->name), |
558540c1 JS |
595 | (unsigned long long)swarn->sector, root, inum, offset, ret); |
596 | ||
597 | free_ipath(ipath); | |
598 | return 0; | |
599 | } | |
600 | ||
b5d67f64 | 601 | static void scrub_print_warning(const char *errstr, struct scrub_block *sblock) |
558540c1 | 602 | { |
a36cf8b8 SB |
603 | struct btrfs_device *dev; |
604 | struct btrfs_fs_info *fs_info; | |
558540c1 JS |
605 | struct btrfs_path *path; |
606 | struct btrfs_key found_key; | |
607 | struct extent_buffer *eb; | |
608 | struct btrfs_extent_item *ei; | |
609 | struct scrub_warning swarn; | |
69917e43 LB |
610 | unsigned long ptr = 0; |
611 | u64 extent_item_pos; | |
612 | u64 flags = 0; | |
558540c1 | 613 | u64 ref_root; |
69917e43 | 614 | u32 item_size; |
558540c1 | 615 | u8 ref_level; |
69917e43 | 616 | int ret; |
558540c1 | 617 | |
a36cf8b8 | 618 | WARN_ON(sblock->page_count < 1); |
7a9e9987 | 619 | dev = sblock->pagev[0]->dev; |
a36cf8b8 SB |
620 | fs_info = sblock->sctx->dev_root->fs_info; |
621 | ||
558540c1 | 622 | path = btrfs_alloc_path(); |
8b9456da DS |
623 | if (!path) |
624 | return; | |
558540c1 | 625 | |
7a9e9987 SB |
626 | swarn.sector = (sblock->pagev[0]->physical) >> 9; |
627 | swarn.logical = sblock->pagev[0]->logical; | |
558540c1 | 628 | swarn.errstr = errstr; |
a36cf8b8 | 629 | swarn.dev = NULL; |
558540c1 | 630 | |
69917e43 LB |
631 | ret = extent_from_logical(fs_info, swarn.logical, path, &found_key, |
632 | &flags); | |
558540c1 JS |
633 | if (ret < 0) |
634 | goto out; | |
635 | ||
4692cf58 | 636 | extent_item_pos = swarn.logical - found_key.objectid; |
558540c1 JS |
637 | swarn.extent_item_size = found_key.offset; |
638 | ||
639 | eb = path->nodes[0]; | |
640 | ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item); | |
641 | item_size = btrfs_item_size_nr(eb, path->slots[0]); | |
642 | ||
69917e43 | 643 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { |
558540c1 | 644 | do { |
6eda71d0 LB |
645 | ret = tree_backref_for_extent(&ptr, eb, &found_key, ei, |
646 | item_size, &ref_root, | |
647 | &ref_level); | |
ecaeb14b DS |
648 | btrfs_warn_in_rcu(fs_info, |
649 | "%s at logical %llu on dev %s, " | |
558540c1 | 650 | "sector %llu: metadata %s (level %d) in tree " |
ecaeb14b | 651 | "%llu", errstr, swarn.logical, |
606686ee | 652 | rcu_str_deref(dev->name), |
558540c1 JS |
653 | (unsigned long long)swarn.sector, |
654 | ref_level ? "node" : "leaf", | |
655 | ret < 0 ? -1 : ref_level, | |
656 | ret < 0 ? -1 : ref_root); | |
657 | } while (ret != 1); | |
d8fe29e9 | 658 | btrfs_release_path(path); |
558540c1 | 659 | } else { |
d8fe29e9 | 660 | btrfs_release_path(path); |
558540c1 | 661 | swarn.path = path; |
a36cf8b8 | 662 | swarn.dev = dev; |
7a3ae2f8 JS |
663 | iterate_extent_inodes(fs_info, found_key.objectid, |
664 | extent_item_pos, 1, | |
558540c1 JS |
665 | scrub_print_warning_inode, &swarn); |
666 | } | |
667 | ||
668 | out: | |
669 | btrfs_free_path(path); | |
558540c1 JS |
670 | } |
671 | ||
ff023aac | 672 | static int scrub_fixup_readpage(u64 inum, u64 offset, u64 root, void *fixup_ctx) |
0ef8e451 | 673 | { |
5da6fcbc | 674 | struct page *page = NULL; |
0ef8e451 | 675 | unsigned long index; |
ff023aac | 676 | struct scrub_fixup_nodatasum *fixup = fixup_ctx; |
0ef8e451 | 677 | int ret; |
5da6fcbc | 678 | int corrected = 0; |
0ef8e451 | 679 | struct btrfs_key key; |
5da6fcbc | 680 | struct inode *inode = NULL; |
6f1c3605 | 681 | struct btrfs_fs_info *fs_info; |
0ef8e451 JS |
682 | u64 end = offset + PAGE_SIZE - 1; |
683 | struct btrfs_root *local_root; | |
6f1c3605 | 684 | int srcu_index; |
0ef8e451 JS |
685 | |
686 | key.objectid = root; | |
687 | key.type = BTRFS_ROOT_ITEM_KEY; | |
688 | key.offset = (u64)-1; | |
6f1c3605 LB |
689 | |
690 | fs_info = fixup->root->fs_info; | |
691 | srcu_index = srcu_read_lock(&fs_info->subvol_srcu); | |
692 | ||
693 | local_root = btrfs_read_fs_root_no_name(fs_info, &key); | |
694 | if (IS_ERR(local_root)) { | |
695 | srcu_read_unlock(&fs_info->subvol_srcu, srcu_index); | |
0ef8e451 | 696 | return PTR_ERR(local_root); |
6f1c3605 | 697 | } |
0ef8e451 JS |
698 | |
699 | key.type = BTRFS_INODE_ITEM_KEY; | |
700 | key.objectid = inum; | |
701 | key.offset = 0; | |
6f1c3605 LB |
702 | inode = btrfs_iget(fs_info->sb, &key, local_root, NULL); |
703 | srcu_read_unlock(&fs_info->subvol_srcu, srcu_index); | |
0ef8e451 JS |
704 | if (IS_ERR(inode)) |
705 | return PTR_ERR(inode); | |
706 | ||
0ef8e451 JS |
707 | index = offset >> PAGE_CACHE_SHIFT; |
708 | ||
709 | page = find_or_create_page(inode->i_mapping, index, GFP_NOFS); | |
5da6fcbc JS |
710 | if (!page) { |
711 | ret = -ENOMEM; | |
712 | goto out; | |
713 | } | |
714 | ||
715 | if (PageUptodate(page)) { | |
5da6fcbc JS |
716 | if (PageDirty(page)) { |
717 | /* | |
718 | * we need to write the data to the defect sector. the | |
719 | * data that was in that sector is not in memory, | |
720 | * because the page was modified. we must not write the | |
721 | * modified page to that sector. | |
722 | * | |
723 | * TODO: what could be done here: wait for the delalloc | |
724 | * runner to write out that page (might involve | |
725 | * COW) and see whether the sector is still | |
726 | * referenced afterwards. | |
727 | * | |
728 | * For the meantime, we'll treat this error | |
729 | * incorrectable, although there is a chance that a | |
730 | * later scrub will find the bad sector again and that | |
731 | * there's no dirty page in memory, then. | |
732 | */ | |
733 | ret = -EIO; | |
734 | goto out; | |
735 | } | |
1203b681 | 736 | ret = repair_io_failure(inode, offset, PAGE_SIZE, |
5da6fcbc | 737 | fixup->logical, page, |
ffdd2018 | 738 | offset - page_offset(page), |
5da6fcbc JS |
739 | fixup->mirror_num); |
740 | unlock_page(page); | |
741 | corrected = !ret; | |
742 | } else { | |
743 | /* | |
744 | * we need to get good data first. the general readpage path | |
745 | * will call repair_io_failure for us, we just have to make | |
746 | * sure we read the bad mirror. | |
747 | */ | |
748 | ret = set_extent_bits(&BTRFS_I(inode)->io_tree, offset, end, | |
749 | EXTENT_DAMAGED, GFP_NOFS); | |
750 | if (ret) { | |
751 | /* set_extent_bits should give proper error */ | |
752 | WARN_ON(ret > 0); | |
753 | if (ret > 0) | |
754 | ret = -EFAULT; | |
755 | goto out; | |
756 | } | |
757 | ||
758 | ret = extent_read_full_page(&BTRFS_I(inode)->io_tree, page, | |
759 | btrfs_get_extent, | |
760 | fixup->mirror_num); | |
761 | wait_on_page_locked(page); | |
762 | ||
763 | corrected = !test_range_bit(&BTRFS_I(inode)->io_tree, offset, | |
764 | end, EXTENT_DAMAGED, 0, NULL); | |
765 | if (!corrected) | |
766 | clear_extent_bits(&BTRFS_I(inode)->io_tree, offset, end, | |
767 | EXTENT_DAMAGED, GFP_NOFS); | |
768 | } | |
769 | ||
770 | out: | |
771 | if (page) | |
772 | put_page(page); | |
7fb18a06 TK |
773 | |
774 | iput(inode); | |
0ef8e451 JS |
775 | |
776 | if (ret < 0) | |
777 | return ret; | |
778 | ||
779 | if (ret == 0 && corrected) { | |
780 | /* | |
781 | * we only need to call readpage for one of the inodes belonging | |
782 | * to this extent. so make iterate_extent_inodes stop | |
783 | */ | |
784 | return 1; | |
785 | } | |
786 | ||
787 | return -EIO; | |
788 | } | |
789 | ||
790 | static void scrub_fixup_nodatasum(struct btrfs_work *work) | |
791 | { | |
792 | int ret; | |
793 | struct scrub_fixup_nodatasum *fixup; | |
d9d181c1 | 794 | struct scrub_ctx *sctx; |
0ef8e451 | 795 | struct btrfs_trans_handle *trans = NULL; |
0ef8e451 JS |
796 | struct btrfs_path *path; |
797 | int uncorrectable = 0; | |
798 | ||
799 | fixup = container_of(work, struct scrub_fixup_nodatasum, work); | |
d9d181c1 | 800 | sctx = fixup->sctx; |
0ef8e451 JS |
801 | |
802 | path = btrfs_alloc_path(); | |
803 | if (!path) { | |
d9d181c1 SB |
804 | spin_lock(&sctx->stat_lock); |
805 | ++sctx->stat.malloc_errors; | |
806 | spin_unlock(&sctx->stat_lock); | |
0ef8e451 JS |
807 | uncorrectable = 1; |
808 | goto out; | |
809 | } | |
810 | ||
811 | trans = btrfs_join_transaction(fixup->root); | |
812 | if (IS_ERR(trans)) { | |
813 | uncorrectable = 1; | |
814 | goto out; | |
815 | } | |
816 | ||
817 | /* | |
818 | * the idea is to trigger a regular read through the standard path. we | |
819 | * read a page from the (failed) logical address by specifying the | |
820 | * corresponding copynum of the failed sector. thus, that readpage is | |
821 | * expected to fail. | |
822 | * that is the point where on-the-fly error correction will kick in | |
823 | * (once it's finished) and rewrite the failed sector if a good copy | |
824 | * can be found. | |
825 | */ | |
826 | ret = iterate_inodes_from_logical(fixup->logical, fixup->root->fs_info, | |
827 | path, scrub_fixup_readpage, | |
828 | fixup); | |
829 | if (ret < 0) { | |
830 | uncorrectable = 1; | |
831 | goto out; | |
832 | } | |
833 | WARN_ON(ret != 1); | |
834 | ||
d9d181c1 SB |
835 | spin_lock(&sctx->stat_lock); |
836 | ++sctx->stat.corrected_errors; | |
837 | spin_unlock(&sctx->stat_lock); | |
0ef8e451 JS |
838 | |
839 | out: | |
840 | if (trans && !IS_ERR(trans)) | |
841 | btrfs_end_transaction(trans, fixup->root); | |
842 | if (uncorrectable) { | |
d9d181c1 SB |
843 | spin_lock(&sctx->stat_lock); |
844 | ++sctx->stat.uncorrectable_errors; | |
845 | spin_unlock(&sctx->stat_lock); | |
ff023aac SB |
846 | btrfs_dev_replace_stats_inc( |
847 | &sctx->dev_root->fs_info->dev_replace. | |
848 | num_uncorrectable_read_errors); | |
b14af3b4 DS |
849 | btrfs_err_rl_in_rcu(sctx->dev_root->fs_info, |
850 | "unable to fixup (nodatasum) error at logical %llu on dev %s", | |
c1c9ff7c | 851 | fixup->logical, rcu_str_deref(fixup->dev->name)); |
0ef8e451 JS |
852 | } |
853 | ||
854 | btrfs_free_path(path); | |
855 | kfree(fixup); | |
856 | ||
b6bfebc1 | 857 | scrub_pending_trans_workers_dec(sctx); |
0ef8e451 JS |
858 | } |
859 | ||
af8e2d1d MX |
860 | static inline void scrub_get_recover(struct scrub_recover *recover) |
861 | { | |
862 | atomic_inc(&recover->refs); | |
863 | } | |
864 | ||
865 | static inline void scrub_put_recover(struct scrub_recover *recover) | |
866 | { | |
867 | if (atomic_dec_and_test(&recover->refs)) { | |
6e9606d2 | 868 | btrfs_put_bbio(recover->bbio); |
af8e2d1d MX |
869 | kfree(recover); |
870 | } | |
871 | } | |
872 | ||
a2de733c | 873 | /* |
b5d67f64 SB |
874 | * scrub_handle_errored_block gets called when either verification of the |
875 | * pages failed or the bio failed to read, e.g. with EIO. In the latter | |
876 | * case, this function handles all pages in the bio, even though only one | |
877 | * may be bad. | |
878 | * The goal of this function is to repair the errored block by using the | |
879 | * contents of one of the mirrors. | |
a2de733c | 880 | */ |
b5d67f64 | 881 | static int scrub_handle_errored_block(struct scrub_block *sblock_to_check) |
a2de733c | 882 | { |
d9d181c1 | 883 | struct scrub_ctx *sctx = sblock_to_check->sctx; |
a36cf8b8 | 884 | struct btrfs_device *dev; |
b5d67f64 SB |
885 | struct btrfs_fs_info *fs_info; |
886 | u64 length; | |
887 | u64 logical; | |
888 | u64 generation; | |
889 | unsigned int failed_mirror_index; | |
890 | unsigned int is_metadata; | |
891 | unsigned int have_csum; | |
892 | u8 *csum; | |
893 | struct scrub_block *sblocks_for_recheck; /* holds one for each mirror */ | |
894 | struct scrub_block *sblock_bad; | |
895 | int ret; | |
896 | int mirror_index; | |
897 | int page_num; | |
898 | int success; | |
558540c1 | 899 | static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL, |
b5d67f64 SB |
900 | DEFAULT_RATELIMIT_BURST); |
901 | ||
902 | BUG_ON(sblock_to_check->page_count < 1); | |
a36cf8b8 | 903 | fs_info = sctx->dev_root->fs_info; |
4ded4f63 SB |
904 | if (sblock_to_check->pagev[0]->flags & BTRFS_EXTENT_FLAG_SUPER) { |
905 | /* | |
906 | * if we find an error in a super block, we just report it. | |
907 | * They will get written with the next transaction commit | |
908 | * anyway | |
909 | */ | |
910 | spin_lock(&sctx->stat_lock); | |
911 | ++sctx->stat.super_errors; | |
912 | spin_unlock(&sctx->stat_lock); | |
913 | return 0; | |
914 | } | |
b5d67f64 | 915 | length = sblock_to_check->page_count * PAGE_SIZE; |
7a9e9987 SB |
916 | logical = sblock_to_check->pagev[0]->logical; |
917 | generation = sblock_to_check->pagev[0]->generation; | |
918 | BUG_ON(sblock_to_check->pagev[0]->mirror_num < 1); | |
919 | failed_mirror_index = sblock_to_check->pagev[0]->mirror_num - 1; | |
920 | is_metadata = !(sblock_to_check->pagev[0]->flags & | |
b5d67f64 | 921 | BTRFS_EXTENT_FLAG_DATA); |
7a9e9987 SB |
922 | have_csum = sblock_to_check->pagev[0]->have_csum; |
923 | csum = sblock_to_check->pagev[0]->csum; | |
924 | dev = sblock_to_check->pagev[0]->dev; | |
13db62b7 | 925 | |
ff023aac SB |
926 | if (sctx->is_dev_replace && !is_metadata && !have_csum) { |
927 | sblocks_for_recheck = NULL; | |
928 | goto nodatasum_case; | |
929 | } | |
930 | ||
b5d67f64 SB |
931 | /* |
932 | * read all mirrors one after the other. This includes to | |
933 | * re-read the extent or metadata block that failed (that was | |
934 | * the cause that this fixup code is called) another time, | |
935 | * page by page this time in order to know which pages | |
936 | * caused I/O errors and which ones are good (for all mirrors). | |
937 | * It is the goal to handle the situation when more than one | |
938 | * mirror contains I/O errors, but the errors do not | |
939 | * overlap, i.e. the data can be repaired by selecting the | |
940 | * pages from those mirrors without I/O error on the | |
941 | * particular pages. One example (with blocks >= 2 * PAGE_SIZE) | |
942 | * would be that mirror #1 has an I/O error on the first page, | |
943 | * the second page is good, and mirror #2 has an I/O error on | |
944 | * the second page, but the first page is good. | |
945 | * Then the first page of the first mirror can be repaired by | |
946 | * taking the first page of the second mirror, and the | |
947 | * second page of the second mirror can be repaired by | |
948 | * copying the contents of the 2nd page of the 1st mirror. | |
949 | * One more note: if the pages of one mirror contain I/O | |
950 | * errors, the checksum cannot be verified. In order to get | |
951 | * the best data for repairing, the first attempt is to find | |
952 | * a mirror without I/O errors and with a validated checksum. | |
953 | * Only if this is not possible, the pages are picked from | |
954 | * mirrors with I/O errors without considering the checksum. | |
955 | * If the latter is the case, at the end, the checksum of the | |
956 | * repaired area is verified in order to correctly maintain | |
957 | * the statistics. | |
958 | */ | |
959 | ||
31e818fe DS |
960 | sblocks_for_recheck = kcalloc(BTRFS_MAX_MIRRORS, |
961 | sizeof(*sblocks_for_recheck), GFP_NOFS); | |
b5d67f64 | 962 | if (!sblocks_for_recheck) { |
d9d181c1 SB |
963 | spin_lock(&sctx->stat_lock); |
964 | sctx->stat.malloc_errors++; | |
965 | sctx->stat.read_errors++; | |
966 | sctx->stat.uncorrectable_errors++; | |
967 | spin_unlock(&sctx->stat_lock); | |
a36cf8b8 | 968 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS); |
b5d67f64 | 969 | goto out; |
a2de733c AJ |
970 | } |
971 | ||
b5d67f64 | 972 | /* setup the context, map the logical blocks and alloc the pages */ |
be50a8dd | 973 | ret = scrub_setup_recheck_block(sblock_to_check, sblocks_for_recheck); |
b5d67f64 | 974 | if (ret) { |
d9d181c1 SB |
975 | spin_lock(&sctx->stat_lock); |
976 | sctx->stat.read_errors++; | |
977 | sctx->stat.uncorrectable_errors++; | |
978 | spin_unlock(&sctx->stat_lock); | |
a36cf8b8 | 979 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS); |
b5d67f64 SB |
980 | goto out; |
981 | } | |
982 | BUG_ON(failed_mirror_index >= BTRFS_MAX_MIRRORS); | |
983 | sblock_bad = sblocks_for_recheck + failed_mirror_index; | |
13db62b7 | 984 | |
b5d67f64 | 985 | /* build and submit the bios for the failed mirror, check checksums */ |
34f5c8e9 | 986 | scrub_recheck_block(fs_info, sblock_bad, is_metadata, have_csum, |
af8e2d1d | 987 | csum, generation, sctx->csum_size, 1); |
a2de733c | 988 | |
b5d67f64 SB |
989 | if (!sblock_bad->header_error && !sblock_bad->checksum_error && |
990 | sblock_bad->no_io_error_seen) { | |
991 | /* | |
992 | * the error disappeared after reading page by page, or | |
993 | * the area was part of a huge bio and other parts of the | |
994 | * bio caused I/O errors, or the block layer merged several | |
995 | * read requests into one and the error is caused by a | |
996 | * different bio (usually one of the two latter cases is | |
997 | * the cause) | |
998 | */ | |
d9d181c1 SB |
999 | spin_lock(&sctx->stat_lock); |
1000 | sctx->stat.unverified_errors++; | |
5a6ac9ea | 1001 | sblock_to_check->data_corrected = 1; |
d9d181c1 | 1002 | spin_unlock(&sctx->stat_lock); |
a2de733c | 1003 | |
ff023aac SB |
1004 | if (sctx->is_dev_replace) |
1005 | scrub_write_block_to_dev_replace(sblock_bad); | |
b5d67f64 | 1006 | goto out; |
a2de733c | 1007 | } |
a2de733c | 1008 | |
b5d67f64 | 1009 | if (!sblock_bad->no_io_error_seen) { |
d9d181c1 SB |
1010 | spin_lock(&sctx->stat_lock); |
1011 | sctx->stat.read_errors++; | |
1012 | spin_unlock(&sctx->stat_lock); | |
b5d67f64 SB |
1013 | if (__ratelimit(&_rs)) |
1014 | scrub_print_warning("i/o error", sblock_to_check); | |
a36cf8b8 | 1015 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS); |
b5d67f64 | 1016 | } else if (sblock_bad->checksum_error) { |
d9d181c1 SB |
1017 | spin_lock(&sctx->stat_lock); |
1018 | sctx->stat.csum_errors++; | |
1019 | spin_unlock(&sctx->stat_lock); | |
b5d67f64 SB |
1020 | if (__ratelimit(&_rs)) |
1021 | scrub_print_warning("checksum error", sblock_to_check); | |
a36cf8b8 | 1022 | btrfs_dev_stat_inc_and_print(dev, |
442a4f63 | 1023 | BTRFS_DEV_STAT_CORRUPTION_ERRS); |
b5d67f64 | 1024 | } else if (sblock_bad->header_error) { |
d9d181c1 SB |
1025 | spin_lock(&sctx->stat_lock); |
1026 | sctx->stat.verify_errors++; | |
1027 | spin_unlock(&sctx->stat_lock); | |
b5d67f64 SB |
1028 | if (__ratelimit(&_rs)) |
1029 | scrub_print_warning("checksum/header error", | |
1030 | sblock_to_check); | |
442a4f63 | 1031 | if (sblock_bad->generation_error) |
a36cf8b8 | 1032 | btrfs_dev_stat_inc_and_print(dev, |
442a4f63 SB |
1033 | BTRFS_DEV_STAT_GENERATION_ERRS); |
1034 | else | |
a36cf8b8 | 1035 | btrfs_dev_stat_inc_and_print(dev, |
442a4f63 | 1036 | BTRFS_DEV_STAT_CORRUPTION_ERRS); |
b5d67f64 | 1037 | } |
a2de733c | 1038 | |
33ef30ad ID |
1039 | if (sctx->readonly) { |
1040 | ASSERT(!sctx->is_dev_replace); | |
1041 | goto out; | |
1042 | } | |
a2de733c | 1043 | |
b5d67f64 SB |
1044 | if (!is_metadata && !have_csum) { |
1045 | struct scrub_fixup_nodatasum *fixup_nodatasum; | |
a2de733c | 1046 | |
ff023aac SB |
1047 | WARN_ON(sctx->is_dev_replace); |
1048 | ||
b25c94c5 ZL |
1049 | nodatasum_case: |
1050 | ||
b5d67f64 SB |
1051 | /* |
1052 | * !is_metadata and !have_csum, this means that the data | |
1053 | * might not be COW'ed, that it might be modified | |
1054 | * concurrently. The general strategy to work on the | |
1055 | * commit root does not help in the case when COW is not | |
1056 | * used. | |
1057 | */ | |
1058 | fixup_nodatasum = kzalloc(sizeof(*fixup_nodatasum), GFP_NOFS); | |
1059 | if (!fixup_nodatasum) | |
1060 | goto did_not_correct_error; | |
d9d181c1 | 1061 | fixup_nodatasum->sctx = sctx; |
a36cf8b8 | 1062 | fixup_nodatasum->dev = dev; |
b5d67f64 SB |
1063 | fixup_nodatasum->logical = logical; |
1064 | fixup_nodatasum->root = fs_info->extent_root; | |
1065 | fixup_nodatasum->mirror_num = failed_mirror_index + 1; | |
b6bfebc1 | 1066 | scrub_pending_trans_workers_inc(sctx); |
9e0af237 LB |
1067 | btrfs_init_work(&fixup_nodatasum->work, btrfs_scrub_helper, |
1068 | scrub_fixup_nodatasum, NULL, NULL); | |
0339ef2f QW |
1069 | btrfs_queue_work(fs_info->scrub_workers, |
1070 | &fixup_nodatasum->work); | |
b5d67f64 | 1071 | goto out; |
a2de733c AJ |
1072 | } |
1073 | ||
b5d67f64 SB |
1074 | /* |
1075 | * now build and submit the bios for the other mirrors, check | |
cb2ced73 SB |
1076 | * checksums. |
1077 | * First try to pick the mirror which is completely without I/O | |
b5d67f64 SB |
1078 | * errors and also does not have a checksum error. |
1079 | * If one is found, and if a checksum is present, the full block | |
1080 | * that is known to contain an error is rewritten. Afterwards | |
1081 | * the block is known to be corrected. | |
1082 | * If a mirror is found which is completely correct, and no | |
1083 | * checksum is present, only those pages are rewritten that had | |
1084 | * an I/O error in the block to be repaired, since it cannot be | |
1085 | * determined, which copy of the other pages is better (and it | |
1086 | * could happen otherwise that a correct page would be | |
1087 | * overwritten by a bad one). | |
1088 | */ | |
1089 | for (mirror_index = 0; | |
1090 | mirror_index < BTRFS_MAX_MIRRORS && | |
1091 | sblocks_for_recheck[mirror_index].page_count > 0; | |
1092 | mirror_index++) { | |
cb2ced73 | 1093 | struct scrub_block *sblock_other; |
b5d67f64 | 1094 | |
cb2ced73 SB |
1095 | if (mirror_index == failed_mirror_index) |
1096 | continue; | |
1097 | sblock_other = sblocks_for_recheck + mirror_index; | |
1098 | ||
1099 | /* build and submit the bios, check checksums */ | |
34f5c8e9 SB |
1100 | scrub_recheck_block(fs_info, sblock_other, is_metadata, |
1101 | have_csum, csum, generation, | |
af8e2d1d | 1102 | sctx->csum_size, 0); |
34f5c8e9 SB |
1103 | |
1104 | if (!sblock_other->header_error && | |
b5d67f64 SB |
1105 | !sblock_other->checksum_error && |
1106 | sblock_other->no_io_error_seen) { | |
ff023aac SB |
1107 | if (sctx->is_dev_replace) { |
1108 | scrub_write_block_to_dev_replace(sblock_other); | |
114ab50d | 1109 | goto corrected_error; |
ff023aac | 1110 | } else { |
ff023aac | 1111 | ret = scrub_repair_block_from_good_copy( |
114ab50d ZL |
1112 | sblock_bad, sblock_other); |
1113 | if (!ret) | |
1114 | goto corrected_error; | |
ff023aac | 1115 | } |
b5d67f64 SB |
1116 | } |
1117 | } | |
a2de733c | 1118 | |
b968fed1 ZL |
1119 | if (sblock_bad->no_io_error_seen && !sctx->is_dev_replace) |
1120 | goto did_not_correct_error; | |
ff023aac SB |
1121 | |
1122 | /* | |
ff023aac | 1123 | * In case of I/O errors in the area that is supposed to be |
b5d67f64 SB |
1124 | * repaired, continue by picking good copies of those pages. |
1125 | * Select the good pages from mirrors to rewrite bad pages from | |
1126 | * the area to fix. Afterwards verify the checksum of the block | |
1127 | * that is supposed to be repaired. This verification step is | |
1128 | * only done for the purpose of statistic counting and for the | |
1129 | * final scrub report, whether errors remain. | |
1130 | * A perfect algorithm could make use of the checksum and try | |
1131 | * all possible combinations of pages from the different mirrors | |
1132 | * until the checksum verification succeeds. For example, when | |
1133 | * the 2nd page of mirror #1 faces I/O errors, and the 2nd page | |
1134 | * of mirror #2 is readable but the final checksum test fails, | |
1135 | * then the 2nd page of mirror #3 could be tried, whether now | |
1136 | * the final checksum succeedes. But this would be a rare | |
1137 | * exception and is therefore not implemented. At least it is | |
1138 | * avoided that the good copy is overwritten. | |
1139 | * A more useful improvement would be to pick the sectors | |
1140 | * without I/O error based on sector sizes (512 bytes on legacy | |
1141 | * disks) instead of on PAGE_SIZE. Then maybe 512 byte of one | |
1142 | * mirror could be repaired by taking 512 byte of a different | |
1143 | * mirror, even if other 512 byte sectors in the same PAGE_SIZE | |
1144 | * area are unreadable. | |
a2de733c | 1145 | */ |
b5d67f64 | 1146 | success = 1; |
b968fed1 ZL |
1147 | for (page_num = 0; page_num < sblock_bad->page_count; |
1148 | page_num++) { | |
7a9e9987 | 1149 | struct scrub_page *page_bad = sblock_bad->pagev[page_num]; |
b968fed1 | 1150 | struct scrub_block *sblock_other = NULL; |
b5d67f64 | 1151 | |
b968fed1 ZL |
1152 | /* skip no-io-error page in scrub */ |
1153 | if (!page_bad->io_error && !sctx->is_dev_replace) | |
a2de733c | 1154 | continue; |
b5d67f64 | 1155 | |
b968fed1 ZL |
1156 | /* try to find no-io-error page in mirrors */ |
1157 | if (page_bad->io_error) { | |
1158 | for (mirror_index = 0; | |
1159 | mirror_index < BTRFS_MAX_MIRRORS && | |
1160 | sblocks_for_recheck[mirror_index].page_count > 0; | |
1161 | mirror_index++) { | |
1162 | if (!sblocks_for_recheck[mirror_index]. | |
1163 | pagev[page_num]->io_error) { | |
1164 | sblock_other = sblocks_for_recheck + | |
1165 | mirror_index; | |
1166 | break; | |
b5d67f64 SB |
1167 | } |
1168 | } | |
b968fed1 ZL |
1169 | if (!sblock_other) |
1170 | success = 0; | |
96e36920 | 1171 | } |
a2de733c | 1172 | |
b968fed1 ZL |
1173 | if (sctx->is_dev_replace) { |
1174 | /* | |
1175 | * did not find a mirror to fetch the page | |
1176 | * from. scrub_write_page_to_dev_replace() | |
1177 | * handles this case (page->io_error), by | |
1178 | * filling the block with zeros before | |
1179 | * submitting the write request | |
1180 | */ | |
1181 | if (!sblock_other) | |
1182 | sblock_other = sblock_bad; | |
1183 | ||
1184 | if (scrub_write_page_to_dev_replace(sblock_other, | |
1185 | page_num) != 0) { | |
1186 | btrfs_dev_replace_stats_inc( | |
1187 | &sctx->dev_root-> | |
1188 | fs_info->dev_replace. | |
1189 | num_write_errors); | |
1190 | success = 0; | |
1191 | } | |
1192 | } else if (sblock_other) { | |
1193 | ret = scrub_repair_page_from_good_copy(sblock_bad, | |
1194 | sblock_other, | |
1195 | page_num, 0); | |
1196 | if (0 == ret) | |
1197 | page_bad->io_error = 0; | |
1198 | else | |
1199 | success = 0; | |
b5d67f64 | 1200 | } |
a2de733c | 1201 | } |
a2de733c | 1202 | |
b968fed1 | 1203 | if (success && !sctx->is_dev_replace) { |
b5d67f64 SB |
1204 | if (is_metadata || have_csum) { |
1205 | /* | |
1206 | * need to verify the checksum now that all | |
1207 | * sectors on disk are repaired (the write | |
1208 | * request for data to be repaired is on its way). | |
1209 | * Just be lazy and use scrub_recheck_block() | |
1210 | * which re-reads the data before the checksum | |
1211 | * is verified, but most likely the data comes out | |
1212 | * of the page cache. | |
1213 | */ | |
34f5c8e9 SB |
1214 | scrub_recheck_block(fs_info, sblock_bad, |
1215 | is_metadata, have_csum, csum, | |
af8e2d1d | 1216 | generation, sctx->csum_size, 1); |
34f5c8e9 | 1217 | if (!sblock_bad->header_error && |
b5d67f64 SB |
1218 | !sblock_bad->checksum_error && |
1219 | sblock_bad->no_io_error_seen) | |
1220 | goto corrected_error; | |
1221 | else | |
1222 | goto did_not_correct_error; | |
1223 | } else { | |
1224 | corrected_error: | |
d9d181c1 SB |
1225 | spin_lock(&sctx->stat_lock); |
1226 | sctx->stat.corrected_errors++; | |
5a6ac9ea | 1227 | sblock_to_check->data_corrected = 1; |
d9d181c1 | 1228 | spin_unlock(&sctx->stat_lock); |
b14af3b4 DS |
1229 | btrfs_err_rl_in_rcu(fs_info, |
1230 | "fixed up error at logical %llu on dev %s", | |
c1c9ff7c | 1231 | logical, rcu_str_deref(dev->name)); |
8628764e | 1232 | } |
b5d67f64 SB |
1233 | } else { |
1234 | did_not_correct_error: | |
d9d181c1 SB |
1235 | spin_lock(&sctx->stat_lock); |
1236 | sctx->stat.uncorrectable_errors++; | |
1237 | spin_unlock(&sctx->stat_lock); | |
b14af3b4 DS |
1238 | btrfs_err_rl_in_rcu(fs_info, |
1239 | "unable to fixup (regular) error at logical %llu on dev %s", | |
c1c9ff7c | 1240 | logical, rcu_str_deref(dev->name)); |
96e36920 | 1241 | } |
a2de733c | 1242 | |
b5d67f64 SB |
1243 | out: |
1244 | if (sblocks_for_recheck) { | |
1245 | for (mirror_index = 0; mirror_index < BTRFS_MAX_MIRRORS; | |
1246 | mirror_index++) { | |
1247 | struct scrub_block *sblock = sblocks_for_recheck + | |
1248 | mirror_index; | |
af8e2d1d | 1249 | struct scrub_recover *recover; |
b5d67f64 SB |
1250 | int page_index; |
1251 | ||
7a9e9987 SB |
1252 | for (page_index = 0; page_index < sblock->page_count; |
1253 | page_index++) { | |
1254 | sblock->pagev[page_index]->sblock = NULL; | |
af8e2d1d MX |
1255 | recover = sblock->pagev[page_index]->recover; |
1256 | if (recover) { | |
1257 | scrub_put_recover(recover); | |
1258 | sblock->pagev[page_index]->recover = | |
1259 | NULL; | |
1260 | } | |
7a9e9987 SB |
1261 | scrub_page_put(sblock->pagev[page_index]); |
1262 | } | |
b5d67f64 SB |
1263 | } |
1264 | kfree(sblocks_for_recheck); | |
1265 | } | |
a2de733c | 1266 | |
b5d67f64 SB |
1267 | return 0; |
1268 | } | |
a2de733c | 1269 | |
8e5cfb55 | 1270 | static inline int scrub_nr_raid_mirrors(struct btrfs_bio *bbio) |
af8e2d1d | 1271 | { |
10f11900 ZL |
1272 | if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID5) |
1273 | return 2; | |
1274 | else if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID6) | |
1275 | return 3; | |
1276 | else | |
af8e2d1d | 1277 | return (int)bbio->num_stripes; |
af8e2d1d MX |
1278 | } |
1279 | ||
10f11900 ZL |
1280 | static inline void scrub_stripe_index_and_offset(u64 logical, u64 map_type, |
1281 | u64 *raid_map, | |
af8e2d1d MX |
1282 | u64 mapped_length, |
1283 | int nstripes, int mirror, | |
1284 | int *stripe_index, | |
1285 | u64 *stripe_offset) | |
1286 | { | |
1287 | int i; | |
1288 | ||
ffe2d203 | 1289 | if (map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
af8e2d1d MX |
1290 | /* RAID5/6 */ |
1291 | for (i = 0; i < nstripes; i++) { | |
1292 | if (raid_map[i] == RAID6_Q_STRIPE || | |
1293 | raid_map[i] == RAID5_P_STRIPE) | |
1294 | continue; | |
1295 | ||
1296 | if (logical >= raid_map[i] && | |
1297 | logical < raid_map[i] + mapped_length) | |
1298 | break; | |
1299 | } | |
1300 | ||
1301 | *stripe_index = i; | |
1302 | *stripe_offset = logical - raid_map[i]; | |
1303 | } else { | |
1304 | /* The other RAID type */ | |
1305 | *stripe_index = mirror; | |
1306 | *stripe_offset = 0; | |
1307 | } | |
1308 | } | |
1309 | ||
be50a8dd | 1310 | static int scrub_setup_recheck_block(struct scrub_block *original_sblock, |
b5d67f64 SB |
1311 | struct scrub_block *sblocks_for_recheck) |
1312 | { | |
be50a8dd ZL |
1313 | struct scrub_ctx *sctx = original_sblock->sctx; |
1314 | struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info; | |
1315 | u64 length = original_sblock->page_count * PAGE_SIZE; | |
1316 | u64 logical = original_sblock->pagev[0]->logical; | |
4734b7ed ZL |
1317 | u64 generation = original_sblock->pagev[0]->generation; |
1318 | u64 flags = original_sblock->pagev[0]->flags; | |
1319 | u64 have_csum = original_sblock->pagev[0]->have_csum; | |
af8e2d1d MX |
1320 | struct scrub_recover *recover; |
1321 | struct btrfs_bio *bbio; | |
af8e2d1d MX |
1322 | u64 sublen; |
1323 | u64 mapped_length; | |
1324 | u64 stripe_offset; | |
1325 | int stripe_index; | |
be50a8dd | 1326 | int page_index = 0; |
b5d67f64 | 1327 | int mirror_index; |
af8e2d1d | 1328 | int nmirrors; |
b5d67f64 SB |
1329 | int ret; |
1330 | ||
1331 | /* | |
57019345 | 1332 | * note: the two members refs and outstanding_pages |
b5d67f64 SB |
1333 | * are not used (and not set) in the blocks that are used for |
1334 | * the recheck procedure | |
1335 | */ | |
1336 | ||
b5d67f64 | 1337 | while (length > 0) { |
af8e2d1d MX |
1338 | sublen = min_t(u64, length, PAGE_SIZE); |
1339 | mapped_length = sublen; | |
1340 | bbio = NULL; | |
a2de733c | 1341 | |
b5d67f64 SB |
1342 | /* |
1343 | * with a length of PAGE_SIZE, each returned stripe | |
1344 | * represents one mirror | |
1345 | */ | |
af8e2d1d | 1346 | ret = btrfs_map_sblock(fs_info, REQ_GET_READ_MIRRORS, logical, |
8e5cfb55 | 1347 | &mapped_length, &bbio, 0, 1); |
b5d67f64 | 1348 | if (ret || !bbio || mapped_length < sublen) { |
6e9606d2 | 1349 | btrfs_put_bbio(bbio); |
b5d67f64 SB |
1350 | return -EIO; |
1351 | } | |
a2de733c | 1352 | |
af8e2d1d MX |
1353 | recover = kzalloc(sizeof(struct scrub_recover), GFP_NOFS); |
1354 | if (!recover) { | |
6e9606d2 | 1355 | btrfs_put_bbio(bbio); |
af8e2d1d MX |
1356 | return -ENOMEM; |
1357 | } | |
1358 | ||
1359 | atomic_set(&recover->refs, 1); | |
1360 | recover->bbio = bbio; | |
af8e2d1d MX |
1361 | recover->map_length = mapped_length; |
1362 | ||
ff023aac | 1363 | BUG_ON(page_index >= SCRUB_PAGES_PER_RD_BIO); |
af8e2d1d | 1364 | |
be50a8dd | 1365 | nmirrors = min(scrub_nr_raid_mirrors(bbio), BTRFS_MAX_MIRRORS); |
10f11900 | 1366 | |
af8e2d1d | 1367 | for (mirror_index = 0; mirror_index < nmirrors; |
b5d67f64 SB |
1368 | mirror_index++) { |
1369 | struct scrub_block *sblock; | |
1370 | struct scrub_page *page; | |
1371 | ||
b5d67f64 | 1372 | sblock = sblocks_for_recheck + mirror_index; |
7a9e9987 | 1373 | sblock->sctx = sctx; |
4734b7ed | 1374 | |
7a9e9987 SB |
1375 | page = kzalloc(sizeof(*page), GFP_NOFS); |
1376 | if (!page) { | |
1377 | leave_nomem: | |
d9d181c1 SB |
1378 | spin_lock(&sctx->stat_lock); |
1379 | sctx->stat.malloc_errors++; | |
1380 | spin_unlock(&sctx->stat_lock); | |
af8e2d1d | 1381 | scrub_put_recover(recover); |
b5d67f64 SB |
1382 | return -ENOMEM; |
1383 | } | |
7a9e9987 SB |
1384 | scrub_page_get(page); |
1385 | sblock->pagev[page_index] = page; | |
4734b7ed ZL |
1386 | page->sblock = sblock; |
1387 | page->flags = flags; | |
1388 | page->generation = generation; | |
7a9e9987 | 1389 | page->logical = logical; |
4734b7ed ZL |
1390 | page->have_csum = have_csum; |
1391 | if (have_csum) | |
1392 | memcpy(page->csum, | |
1393 | original_sblock->pagev[0]->csum, | |
1394 | sctx->csum_size); | |
af8e2d1d | 1395 | |
10f11900 ZL |
1396 | scrub_stripe_index_and_offset(logical, |
1397 | bbio->map_type, | |
1398 | bbio->raid_map, | |
af8e2d1d | 1399 | mapped_length, |
e34c330d ZL |
1400 | bbio->num_stripes - |
1401 | bbio->num_tgtdevs, | |
af8e2d1d MX |
1402 | mirror_index, |
1403 | &stripe_index, | |
1404 | &stripe_offset); | |
1405 | page->physical = bbio->stripes[stripe_index].physical + | |
1406 | stripe_offset; | |
1407 | page->dev = bbio->stripes[stripe_index].dev; | |
1408 | ||
ff023aac SB |
1409 | BUG_ON(page_index >= original_sblock->page_count); |
1410 | page->physical_for_dev_replace = | |
1411 | original_sblock->pagev[page_index]-> | |
1412 | physical_for_dev_replace; | |
7a9e9987 | 1413 | /* for missing devices, dev->bdev is NULL */ |
7a9e9987 | 1414 | page->mirror_num = mirror_index + 1; |
b5d67f64 | 1415 | sblock->page_count++; |
7a9e9987 SB |
1416 | page->page = alloc_page(GFP_NOFS); |
1417 | if (!page->page) | |
1418 | goto leave_nomem; | |
af8e2d1d MX |
1419 | |
1420 | scrub_get_recover(recover); | |
1421 | page->recover = recover; | |
b5d67f64 | 1422 | } |
af8e2d1d | 1423 | scrub_put_recover(recover); |
b5d67f64 SB |
1424 | length -= sublen; |
1425 | logical += sublen; | |
1426 | page_index++; | |
1427 | } | |
1428 | ||
1429 | return 0; | |
96e36920 ID |
1430 | } |
1431 | ||
af8e2d1d MX |
1432 | struct scrub_bio_ret { |
1433 | struct completion event; | |
1434 | int error; | |
1435 | }; | |
1436 | ||
4246a0b6 | 1437 | static void scrub_bio_wait_endio(struct bio *bio) |
af8e2d1d MX |
1438 | { |
1439 | struct scrub_bio_ret *ret = bio->bi_private; | |
1440 | ||
4246a0b6 | 1441 | ret->error = bio->bi_error; |
af8e2d1d MX |
1442 | complete(&ret->event); |
1443 | } | |
1444 | ||
1445 | static inline int scrub_is_page_on_raid56(struct scrub_page *page) | |
1446 | { | |
10f11900 | 1447 | return page->recover && |
ffe2d203 | 1448 | (page->recover->bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK); |
af8e2d1d MX |
1449 | } |
1450 | ||
1451 | static int scrub_submit_raid56_bio_wait(struct btrfs_fs_info *fs_info, | |
1452 | struct bio *bio, | |
1453 | struct scrub_page *page) | |
1454 | { | |
1455 | struct scrub_bio_ret done; | |
1456 | int ret; | |
1457 | ||
1458 | init_completion(&done.event); | |
1459 | done.error = 0; | |
1460 | bio->bi_iter.bi_sector = page->logical >> 9; | |
1461 | bio->bi_private = &done; | |
1462 | bio->bi_end_io = scrub_bio_wait_endio; | |
1463 | ||
1464 | ret = raid56_parity_recover(fs_info->fs_root, bio, page->recover->bbio, | |
af8e2d1d | 1465 | page->recover->map_length, |
4245215d | 1466 | page->mirror_num, 0); |
af8e2d1d MX |
1467 | if (ret) |
1468 | return ret; | |
1469 | ||
1470 | wait_for_completion(&done.event); | |
1471 | if (done.error) | |
1472 | return -EIO; | |
1473 | ||
1474 | return 0; | |
1475 | } | |
1476 | ||
b5d67f64 SB |
1477 | /* |
1478 | * this function will check the on disk data for checksum errors, header | |
1479 | * errors and read I/O errors. If any I/O errors happen, the exact pages | |
1480 | * which are errored are marked as being bad. The goal is to enable scrub | |
1481 | * to take those pages that are not errored from all the mirrors so that | |
1482 | * the pages that are errored in the just handled mirror can be repaired. | |
1483 | */ | |
34f5c8e9 SB |
1484 | static void scrub_recheck_block(struct btrfs_fs_info *fs_info, |
1485 | struct scrub_block *sblock, int is_metadata, | |
1486 | int have_csum, u8 *csum, u64 generation, | |
af8e2d1d | 1487 | u16 csum_size, int retry_failed_mirror) |
96e36920 | 1488 | { |
b5d67f64 | 1489 | int page_num; |
96e36920 | 1490 | |
b5d67f64 | 1491 | sblock->no_io_error_seen = 1; |
96e36920 | 1492 | |
b5d67f64 SB |
1493 | for (page_num = 0; page_num < sblock->page_count; page_num++) { |
1494 | struct bio *bio; | |
7a9e9987 | 1495 | struct scrub_page *page = sblock->pagev[page_num]; |
b5d67f64 | 1496 | |
442a4f63 | 1497 | if (page->dev->bdev == NULL) { |
ea9947b4 SB |
1498 | page->io_error = 1; |
1499 | sblock->no_io_error_seen = 0; | |
1500 | continue; | |
1501 | } | |
1502 | ||
7a9e9987 | 1503 | WARN_ON(!page->page); |
9be3395b | 1504 | bio = btrfs_io_bio_alloc(GFP_NOFS, 1); |
34f5c8e9 SB |
1505 | if (!bio) { |
1506 | page->io_error = 1; | |
1507 | sblock->no_io_error_seen = 0; | |
1508 | continue; | |
1509 | } | |
442a4f63 | 1510 | bio->bi_bdev = page->dev->bdev; |
b5d67f64 | 1511 | |
34f5c8e9 | 1512 | bio_add_page(bio, page->page, PAGE_SIZE, 0); |
af8e2d1d MX |
1513 | if (!retry_failed_mirror && scrub_is_page_on_raid56(page)) { |
1514 | if (scrub_submit_raid56_bio_wait(fs_info, bio, page)) | |
1515 | sblock->no_io_error_seen = 0; | |
1516 | } else { | |
1517 | bio->bi_iter.bi_sector = page->physical >> 9; | |
1518 | ||
1519 | if (btrfsic_submit_bio_wait(READ, bio)) | |
1520 | sblock->no_io_error_seen = 0; | |
1521 | } | |
33879d45 | 1522 | |
b5d67f64 SB |
1523 | bio_put(bio); |
1524 | } | |
96e36920 | 1525 | |
b5d67f64 | 1526 | if (sblock->no_io_error_seen) |
ba7cf988 | 1527 | scrub_recheck_block_checksum(sblock); |
b5d67f64 | 1528 | |
34f5c8e9 | 1529 | return; |
a2de733c AJ |
1530 | } |
1531 | ||
17a9be2f MX |
1532 | static inline int scrub_check_fsid(u8 fsid[], |
1533 | struct scrub_page *spage) | |
1534 | { | |
1535 | struct btrfs_fs_devices *fs_devices = spage->dev->fs_devices; | |
1536 | int ret; | |
1537 | ||
1538 | ret = memcmp(fsid, fs_devices->fsid, BTRFS_UUID_SIZE); | |
1539 | return !ret; | |
1540 | } | |
1541 | ||
ba7cf988 | 1542 | static void scrub_recheck_block_checksum(struct scrub_block *sblock) |
a2de733c | 1543 | { |
ba7cf988 ZL |
1544 | sblock->header_error = 0; |
1545 | sblock->checksum_error = 0; | |
1546 | sblock->generation_error = 0; | |
b5d67f64 | 1547 | |
ba7cf988 ZL |
1548 | if (sblock->pagev[0]->flags & BTRFS_EXTENT_FLAG_DATA) |
1549 | scrub_checksum_data(sblock); | |
1550 | else | |
1551 | scrub_checksum_tree_block(sblock); | |
a2de733c AJ |
1552 | } |
1553 | ||
b5d67f64 | 1554 | static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad, |
114ab50d | 1555 | struct scrub_block *sblock_good) |
b5d67f64 SB |
1556 | { |
1557 | int page_num; | |
1558 | int ret = 0; | |
96e36920 | 1559 | |
b5d67f64 SB |
1560 | for (page_num = 0; page_num < sblock_bad->page_count; page_num++) { |
1561 | int ret_sub; | |
96e36920 | 1562 | |
b5d67f64 SB |
1563 | ret_sub = scrub_repair_page_from_good_copy(sblock_bad, |
1564 | sblock_good, | |
114ab50d | 1565 | page_num, 1); |
b5d67f64 SB |
1566 | if (ret_sub) |
1567 | ret = ret_sub; | |
a2de733c | 1568 | } |
b5d67f64 SB |
1569 | |
1570 | return ret; | |
1571 | } | |
1572 | ||
1573 | static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad, | |
1574 | struct scrub_block *sblock_good, | |
1575 | int page_num, int force_write) | |
1576 | { | |
7a9e9987 SB |
1577 | struct scrub_page *page_bad = sblock_bad->pagev[page_num]; |
1578 | struct scrub_page *page_good = sblock_good->pagev[page_num]; | |
b5d67f64 | 1579 | |
7a9e9987 SB |
1580 | BUG_ON(page_bad->page == NULL); |
1581 | BUG_ON(page_good->page == NULL); | |
b5d67f64 SB |
1582 | if (force_write || sblock_bad->header_error || |
1583 | sblock_bad->checksum_error || page_bad->io_error) { | |
1584 | struct bio *bio; | |
1585 | int ret; | |
b5d67f64 | 1586 | |
ff023aac | 1587 | if (!page_bad->dev->bdev) { |
94647322 | 1588 | btrfs_warn_rl(sblock_bad->sctx->dev_root->fs_info, |
efe120a0 | 1589 | "scrub_repair_page_from_good_copy(bdev == NULL) " |
94647322 | 1590 | "is unexpected"); |
ff023aac SB |
1591 | return -EIO; |
1592 | } | |
1593 | ||
9be3395b | 1594 | bio = btrfs_io_bio_alloc(GFP_NOFS, 1); |
e627ee7b TI |
1595 | if (!bio) |
1596 | return -EIO; | |
442a4f63 | 1597 | bio->bi_bdev = page_bad->dev->bdev; |
4f024f37 | 1598 | bio->bi_iter.bi_sector = page_bad->physical >> 9; |
b5d67f64 SB |
1599 | |
1600 | ret = bio_add_page(bio, page_good->page, PAGE_SIZE, 0); | |
1601 | if (PAGE_SIZE != ret) { | |
1602 | bio_put(bio); | |
1603 | return -EIO; | |
13db62b7 | 1604 | } |
b5d67f64 | 1605 | |
33879d45 | 1606 | if (btrfsic_submit_bio_wait(WRITE, bio)) { |
442a4f63 SB |
1607 | btrfs_dev_stat_inc_and_print(page_bad->dev, |
1608 | BTRFS_DEV_STAT_WRITE_ERRS); | |
ff023aac SB |
1609 | btrfs_dev_replace_stats_inc( |
1610 | &sblock_bad->sctx->dev_root->fs_info-> | |
1611 | dev_replace.num_write_errors); | |
442a4f63 SB |
1612 | bio_put(bio); |
1613 | return -EIO; | |
1614 | } | |
b5d67f64 | 1615 | bio_put(bio); |
a2de733c AJ |
1616 | } |
1617 | ||
b5d67f64 SB |
1618 | return 0; |
1619 | } | |
1620 | ||
ff023aac SB |
1621 | static void scrub_write_block_to_dev_replace(struct scrub_block *sblock) |
1622 | { | |
1623 | int page_num; | |
1624 | ||
5a6ac9ea MX |
1625 | /* |
1626 | * This block is used for the check of the parity on the source device, | |
1627 | * so the data needn't be written into the destination device. | |
1628 | */ | |
1629 | if (sblock->sparity) | |
1630 | return; | |
1631 | ||
ff023aac SB |
1632 | for (page_num = 0; page_num < sblock->page_count; page_num++) { |
1633 | int ret; | |
1634 | ||
1635 | ret = scrub_write_page_to_dev_replace(sblock, page_num); | |
1636 | if (ret) | |
1637 | btrfs_dev_replace_stats_inc( | |
1638 | &sblock->sctx->dev_root->fs_info->dev_replace. | |
1639 | num_write_errors); | |
1640 | } | |
1641 | } | |
1642 | ||
1643 | static int scrub_write_page_to_dev_replace(struct scrub_block *sblock, | |
1644 | int page_num) | |
1645 | { | |
1646 | struct scrub_page *spage = sblock->pagev[page_num]; | |
1647 | ||
1648 | BUG_ON(spage->page == NULL); | |
1649 | if (spage->io_error) { | |
1650 | void *mapped_buffer = kmap_atomic(spage->page); | |
1651 | ||
1652 | memset(mapped_buffer, 0, PAGE_CACHE_SIZE); | |
1653 | flush_dcache_page(spage->page); | |
1654 | kunmap_atomic(mapped_buffer); | |
1655 | } | |
1656 | return scrub_add_page_to_wr_bio(sblock->sctx, spage); | |
1657 | } | |
1658 | ||
1659 | static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx, | |
1660 | struct scrub_page *spage) | |
1661 | { | |
1662 | struct scrub_wr_ctx *wr_ctx = &sctx->wr_ctx; | |
1663 | struct scrub_bio *sbio; | |
1664 | int ret; | |
1665 | ||
1666 | mutex_lock(&wr_ctx->wr_lock); | |
1667 | again: | |
1668 | if (!wr_ctx->wr_curr_bio) { | |
1669 | wr_ctx->wr_curr_bio = kzalloc(sizeof(*wr_ctx->wr_curr_bio), | |
1670 | GFP_NOFS); | |
1671 | if (!wr_ctx->wr_curr_bio) { | |
1672 | mutex_unlock(&wr_ctx->wr_lock); | |
1673 | return -ENOMEM; | |
1674 | } | |
1675 | wr_ctx->wr_curr_bio->sctx = sctx; | |
1676 | wr_ctx->wr_curr_bio->page_count = 0; | |
1677 | } | |
1678 | sbio = wr_ctx->wr_curr_bio; | |
1679 | if (sbio->page_count == 0) { | |
1680 | struct bio *bio; | |
1681 | ||
1682 | sbio->physical = spage->physical_for_dev_replace; | |
1683 | sbio->logical = spage->logical; | |
1684 | sbio->dev = wr_ctx->tgtdev; | |
1685 | bio = sbio->bio; | |
1686 | if (!bio) { | |
9be3395b | 1687 | bio = btrfs_io_bio_alloc(GFP_NOFS, wr_ctx->pages_per_wr_bio); |
ff023aac SB |
1688 | if (!bio) { |
1689 | mutex_unlock(&wr_ctx->wr_lock); | |
1690 | return -ENOMEM; | |
1691 | } | |
1692 | sbio->bio = bio; | |
1693 | } | |
1694 | ||
1695 | bio->bi_private = sbio; | |
1696 | bio->bi_end_io = scrub_wr_bio_end_io; | |
1697 | bio->bi_bdev = sbio->dev->bdev; | |
4f024f37 | 1698 | bio->bi_iter.bi_sector = sbio->physical >> 9; |
ff023aac SB |
1699 | sbio->err = 0; |
1700 | } else if (sbio->physical + sbio->page_count * PAGE_SIZE != | |
1701 | spage->physical_for_dev_replace || | |
1702 | sbio->logical + sbio->page_count * PAGE_SIZE != | |
1703 | spage->logical) { | |
1704 | scrub_wr_submit(sctx); | |
1705 | goto again; | |
1706 | } | |
1707 | ||
1708 | ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0); | |
1709 | if (ret != PAGE_SIZE) { | |
1710 | if (sbio->page_count < 1) { | |
1711 | bio_put(sbio->bio); | |
1712 | sbio->bio = NULL; | |
1713 | mutex_unlock(&wr_ctx->wr_lock); | |
1714 | return -EIO; | |
1715 | } | |
1716 | scrub_wr_submit(sctx); | |
1717 | goto again; | |
1718 | } | |
1719 | ||
1720 | sbio->pagev[sbio->page_count] = spage; | |
1721 | scrub_page_get(spage); | |
1722 | sbio->page_count++; | |
1723 | if (sbio->page_count == wr_ctx->pages_per_wr_bio) | |
1724 | scrub_wr_submit(sctx); | |
1725 | mutex_unlock(&wr_ctx->wr_lock); | |
1726 | ||
1727 | return 0; | |
1728 | } | |
1729 | ||
1730 | static void scrub_wr_submit(struct scrub_ctx *sctx) | |
1731 | { | |
1732 | struct scrub_wr_ctx *wr_ctx = &sctx->wr_ctx; | |
1733 | struct scrub_bio *sbio; | |
1734 | ||
1735 | if (!wr_ctx->wr_curr_bio) | |
1736 | return; | |
1737 | ||
1738 | sbio = wr_ctx->wr_curr_bio; | |
1739 | wr_ctx->wr_curr_bio = NULL; | |
1740 | WARN_ON(!sbio->bio->bi_bdev); | |
1741 | scrub_pending_bio_inc(sctx); | |
1742 | /* process all writes in a single worker thread. Then the block layer | |
1743 | * orders the requests before sending them to the driver which | |
1744 | * doubled the write performance on spinning disks when measured | |
1745 | * with Linux 3.5 */ | |
1746 | btrfsic_submit_bio(WRITE, sbio->bio); | |
1747 | } | |
1748 | ||
4246a0b6 | 1749 | static void scrub_wr_bio_end_io(struct bio *bio) |
ff023aac SB |
1750 | { |
1751 | struct scrub_bio *sbio = bio->bi_private; | |
1752 | struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info; | |
1753 | ||
4246a0b6 | 1754 | sbio->err = bio->bi_error; |
ff023aac SB |
1755 | sbio->bio = bio; |
1756 | ||
9e0af237 LB |
1757 | btrfs_init_work(&sbio->work, btrfs_scrubwrc_helper, |
1758 | scrub_wr_bio_end_io_worker, NULL, NULL); | |
0339ef2f | 1759 | btrfs_queue_work(fs_info->scrub_wr_completion_workers, &sbio->work); |
ff023aac SB |
1760 | } |
1761 | ||
1762 | static void scrub_wr_bio_end_io_worker(struct btrfs_work *work) | |
1763 | { | |
1764 | struct scrub_bio *sbio = container_of(work, struct scrub_bio, work); | |
1765 | struct scrub_ctx *sctx = sbio->sctx; | |
1766 | int i; | |
1767 | ||
1768 | WARN_ON(sbio->page_count > SCRUB_PAGES_PER_WR_BIO); | |
1769 | if (sbio->err) { | |
1770 | struct btrfs_dev_replace *dev_replace = | |
1771 | &sbio->sctx->dev_root->fs_info->dev_replace; | |
1772 | ||
1773 | for (i = 0; i < sbio->page_count; i++) { | |
1774 | struct scrub_page *spage = sbio->pagev[i]; | |
1775 | ||
1776 | spage->io_error = 1; | |
1777 | btrfs_dev_replace_stats_inc(&dev_replace-> | |
1778 | num_write_errors); | |
1779 | } | |
1780 | } | |
1781 | ||
1782 | for (i = 0; i < sbio->page_count; i++) | |
1783 | scrub_page_put(sbio->pagev[i]); | |
1784 | ||
1785 | bio_put(sbio->bio); | |
1786 | kfree(sbio); | |
1787 | scrub_pending_bio_dec(sctx); | |
1788 | } | |
1789 | ||
1790 | static int scrub_checksum(struct scrub_block *sblock) | |
b5d67f64 SB |
1791 | { |
1792 | u64 flags; | |
1793 | int ret; | |
1794 | ||
ba7cf988 ZL |
1795 | /* |
1796 | * No need to initialize these stats currently, | |
1797 | * because this function only use return value | |
1798 | * instead of these stats value. | |
1799 | * | |
1800 | * Todo: | |
1801 | * always use stats | |
1802 | */ | |
1803 | sblock->header_error = 0; | |
1804 | sblock->generation_error = 0; | |
1805 | sblock->checksum_error = 0; | |
1806 | ||
7a9e9987 SB |
1807 | WARN_ON(sblock->page_count < 1); |
1808 | flags = sblock->pagev[0]->flags; | |
b5d67f64 SB |
1809 | ret = 0; |
1810 | if (flags & BTRFS_EXTENT_FLAG_DATA) | |
1811 | ret = scrub_checksum_data(sblock); | |
1812 | else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) | |
1813 | ret = scrub_checksum_tree_block(sblock); | |
1814 | else if (flags & BTRFS_EXTENT_FLAG_SUPER) | |
1815 | (void)scrub_checksum_super(sblock); | |
1816 | else | |
1817 | WARN_ON(1); | |
1818 | if (ret) | |
1819 | scrub_handle_errored_block(sblock); | |
ff023aac SB |
1820 | |
1821 | return ret; | |
a2de733c AJ |
1822 | } |
1823 | ||
b5d67f64 | 1824 | static int scrub_checksum_data(struct scrub_block *sblock) |
a2de733c | 1825 | { |
d9d181c1 | 1826 | struct scrub_ctx *sctx = sblock->sctx; |
a2de733c | 1827 | u8 csum[BTRFS_CSUM_SIZE]; |
b5d67f64 SB |
1828 | u8 *on_disk_csum; |
1829 | struct page *page; | |
1830 | void *buffer; | |
a2de733c | 1831 | u32 crc = ~(u32)0; |
b5d67f64 SB |
1832 | u64 len; |
1833 | int index; | |
a2de733c | 1834 | |
b5d67f64 | 1835 | BUG_ON(sblock->page_count < 1); |
7a9e9987 | 1836 | if (!sblock->pagev[0]->have_csum) |
a2de733c AJ |
1837 | return 0; |
1838 | ||
7a9e9987 SB |
1839 | on_disk_csum = sblock->pagev[0]->csum; |
1840 | page = sblock->pagev[0]->page; | |
9613bebb | 1841 | buffer = kmap_atomic(page); |
b5d67f64 | 1842 | |
d9d181c1 | 1843 | len = sctx->sectorsize; |
b5d67f64 SB |
1844 | index = 0; |
1845 | for (;;) { | |
1846 | u64 l = min_t(u64, len, PAGE_SIZE); | |
1847 | ||
b0496686 | 1848 | crc = btrfs_csum_data(buffer, crc, l); |
9613bebb | 1849 | kunmap_atomic(buffer); |
b5d67f64 SB |
1850 | len -= l; |
1851 | if (len == 0) | |
1852 | break; | |
1853 | index++; | |
1854 | BUG_ON(index >= sblock->page_count); | |
7a9e9987 SB |
1855 | BUG_ON(!sblock->pagev[index]->page); |
1856 | page = sblock->pagev[index]->page; | |
9613bebb | 1857 | buffer = kmap_atomic(page); |
b5d67f64 SB |
1858 | } |
1859 | ||
a2de733c | 1860 | btrfs_csum_final(crc, csum); |
d9d181c1 | 1861 | if (memcmp(csum, on_disk_csum, sctx->csum_size)) |
ba7cf988 | 1862 | sblock->checksum_error = 1; |
a2de733c | 1863 | |
ba7cf988 | 1864 | return sblock->checksum_error; |
a2de733c AJ |
1865 | } |
1866 | ||
b5d67f64 | 1867 | static int scrub_checksum_tree_block(struct scrub_block *sblock) |
a2de733c | 1868 | { |
d9d181c1 | 1869 | struct scrub_ctx *sctx = sblock->sctx; |
a2de733c | 1870 | struct btrfs_header *h; |
a36cf8b8 | 1871 | struct btrfs_root *root = sctx->dev_root; |
a2de733c | 1872 | struct btrfs_fs_info *fs_info = root->fs_info; |
b5d67f64 SB |
1873 | u8 calculated_csum[BTRFS_CSUM_SIZE]; |
1874 | u8 on_disk_csum[BTRFS_CSUM_SIZE]; | |
1875 | struct page *page; | |
1876 | void *mapped_buffer; | |
1877 | u64 mapped_size; | |
1878 | void *p; | |
a2de733c | 1879 | u32 crc = ~(u32)0; |
b5d67f64 SB |
1880 | u64 len; |
1881 | int index; | |
1882 | ||
1883 | BUG_ON(sblock->page_count < 1); | |
7a9e9987 | 1884 | page = sblock->pagev[0]->page; |
9613bebb | 1885 | mapped_buffer = kmap_atomic(page); |
b5d67f64 | 1886 | h = (struct btrfs_header *)mapped_buffer; |
d9d181c1 | 1887 | memcpy(on_disk_csum, h->csum, sctx->csum_size); |
a2de733c AJ |
1888 | |
1889 | /* | |
1890 | * we don't use the getter functions here, as we | |
1891 | * a) don't have an extent buffer and | |
1892 | * b) the page is already kmapped | |
1893 | */ | |
3cae210f | 1894 | if (sblock->pagev[0]->logical != btrfs_stack_header_bytenr(h)) |
ba7cf988 | 1895 | sblock->header_error = 1; |
a2de733c | 1896 | |
ba7cf988 ZL |
1897 | if (sblock->pagev[0]->generation != btrfs_stack_header_generation(h)) { |
1898 | sblock->header_error = 1; | |
1899 | sblock->generation_error = 1; | |
1900 | } | |
a2de733c | 1901 | |
17a9be2f | 1902 | if (!scrub_check_fsid(h->fsid, sblock->pagev[0])) |
ba7cf988 | 1903 | sblock->header_error = 1; |
a2de733c AJ |
1904 | |
1905 | if (memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid, | |
1906 | BTRFS_UUID_SIZE)) | |
ba7cf988 | 1907 | sblock->header_error = 1; |
a2de733c | 1908 | |
d9d181c1 | 1909 | len = sctx->nodesize - BTRFS_CSUM_SIZE; |
b5d67f64 SB |
1910 | mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE; |
1911 | p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE; | |
1912 | index = 0; | |
1913 | for (;;) { | |
1914 | u64 l = min_t(u64, len, mapped_size); | |
1915 | ||
b0496686 | 1916 | crc = btrfs_csum_data(p, crc, l); |
9613bebb | 1917 | kunmap_atomic(mapped_buffer); |
b5d67f64 SB |
1918 | len -= l; |
1919 | if (len == 0) | |
1920 | break; | |
1921 | index++; | |
1922 | BUG_ON(index >= sblock->page_count); | |
7a9e9987 SB |
1923 | BUG_ON(!sblock->pagev[index]->page); |
1924 | page = sblock->pagev[index]->page; | |
9613bebb | 1925 | mapped_buffer = kmap_atomic(page); |
b5d67f64 SB |
1926 | mapped_size = PAGE_SIZE; |
1927 | p = mapped_buffer; | |
1928 | } | |
1929 | ||
1930 | btrfs_csum_final(crc, calculated_csum); | |
d9d181c1 | 1931 | if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size)) |
ba7cf988 | 1932 | sblock->checksum_error = 1; |
a2de733c | 1933 | |
ba7cf988 | 1934 | return sblock->header_error || sblock->checksum_error; |
a2de733c AJ |
1935 | } |
1936 | ||
b5d67f64 | 1937 | static int scrub_checksum_super(struct scrub_block *sblock) |
a2de733c AJ |
1938 | { |
1939 | struct btrfs_super_block *s; | |
d9d181c1 | 1940 | struct scrub_ctx *sctx = sblock->sctx; |
b5d67f64 SB |
1941 | u8 calculated_csum[BTRFS_CSUM_SIZE]; |
1942 | u8 on_disk_csum[BTRFS_CSUM_SIZE]; | |
1943 | struct page *page; | |
1944 | void *mapped_buffer; | |
1945 | u64 mapped_size; | |
1946 | void *p; | |
a2de733c | 1947 | u32 crc = ~(u32)0; |
442a4f63 SB |
1948 | int fail_gen = 0; |
1949 | int fail_cor = 0; | |
b5d67f64 SB |
1950 | u64 len; |
1951 | int index; | |
a2de733c | 1952 | |
b5d67f64 | 1953 | BUG_ON(sblock->page_count < 1); |
7a9e9987 | 1954 | page = sblock->pagev[0]->page; |
9613bebb | 1955 | mapped_buffer = kmap_atomic(page); |
b5d67f64 | 1956 | s = (struct btrfs_super_block *)mapped_buffer; |
d9d181c1 | 1957 | memcpy(on_disk_csum, s->csum, sctx->csum_size); |
a2de733c | 1958 | |
3cae210f | 1959 | if (sblock->pagev[0]->logical != btrfs_super_bytenr(s)) |
442a4f63 | 1960 | ++fail_cor; |
a2de733c | 1961 | |
3cae210f | 1962 | if (sblock->pagev[0]->generation != btrfs_super_generation(s)) |
442a4f63 | 1963 | ++fail_gen; |
a2de733c | 1964 | |
17a9be2f | 1965 | if (!scrub_check_fsid(s->fsid, sblock->pagev[0])) |
442a4f63 | 1966 | ++fail_cor; |
a2de733c | 1967 | |
b5d67f64 SB |
1968 | len = BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE; |
1969 | mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE; | |
1970 | p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE; | |
1971 | index = 0; | |
1972 | for (;;) { | |
1973 | u64 l = min_t(u64, len, mapped_size); | |
1974 | ||
b0496686 | 1975 | crc = btrfs_csum_data(p, crc, l); |
9613bebb | 1976 | kunmap_atomic(mapped_buffer); |
b5d67f64 SB |
1977 | len -= l; |
1978 | if (len == 0) | |
1979 | break; | |
1980 | index++; | |
1981 | BUG_ON(index >= sblock->page_count); | |
7a9e9987 SB |
1982 | BUG_ON(!sblock->pagev[index]->page); |
1983 | page = sblock->pagev[index]->page; | |
9613bebb | 1984 | mapped_buffer = kmap_atomic(page); |
b5d67f64 SB |
1985 | mapped_size = PAGE_SIZE; |
1986 | p = mapped_buffer; | |
1987 | } | |
1988 | ||
1989 | btrfs_csum_final(crc, calculated_csum); | |
d9d181c1 | 1990 | if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size)) |
442a4f63 | 1991 | ++fail_cor; |
a2de733c | 1992 | |
442a4f63 | 1993 | if (fail_cor + fail_gen) { |
a2de733c AJ |
1994 | /* |
1995 | * if we find an error in a super block, we just report it. | |
1996 | * They will get written with the next transaction commit | |
1997 | * anyway | |
1998 | */ | |
d9d181c1 SB |
1999 | spin_lock(&sctx->stat_lock); |
2000 | ++sctx->stat.super_errors; | |
2001 | spin_unlock(&sctx->stat_lock); | |
442a4f63 | 2002 | if (fail_cor) |
7a9e9987 | 2003 | btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev, |
442a4f63 SB |
2004 | BTRFS_DEV_STAT_CORRUPTION_ERRS); |
2005 | else | |
7a9e9987 | 2006 | btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev, |
442a4f63 | 2007 | BTRFS_DEV_STAT_GENERATION_ERRS); |
a2de733c AJ |
2008 | } |
2009 | ||
442a4f63 | 2010 | return fail_cor + fail_gen; |
a2de733c AJ |
2011 | } |
2012 | ||
b5d67f64 SB |
2013 | static void scrub_block_get(struct scrub_block *sblock) |
2014 | { | |
57019345 | 2015 | atomic_inc(&sblock->refs); |
b5d67f64 SB |
2016 | } |
2017 | ||
2018 | static void scrub_block_put(struct scrub_block *sblock) | |
2019 | { | |
57019345 | 2020 | if (atomic_dec_and_test(&sblock->refs)) { |
b5d67f64 SB |
2021 | int i; |
2022 | ||
5a6ac9ea MX |
2023 | if (sblock->sparity) |
2024 | scrub_parity_put(sblock->sparity); | |
2025 | ||
b5d67f64 | 2026 | for (i = 0; i < sblock->page_count; i++) |
7a9e9987 | 2027 | scrub_page_put(sblock->pagev[i]); |
b5d67f64 SB |
2028 | kfree(sblock); |
2029 | } | |
2030 | } | |
2031 | ||
7a9e9987 SB |
2032 | static void scrub_page_get(struct scrub_page *spage) |
2033 | { | |
57019345 | 2034 | atomic_inc(&spage->refs); |
7a9e9987 SB |
2035 | } |
2036 | ||
2037 | static void scrub_page_put(struct scrub_page *spage) | |
2038 | { | |
57019345 | 2039 | if (atomic_dec_and_test(&spage->refs)) { |
7a9e9987 SB |
2040 | if (spage->page) |
2041 | __free_page(spage->page); | |
2042 | kfree(spage); | |
2043 | } | |
2044 | } | |
2045 | ||
d9d181c1 | 2046 | static void scrub_submit(struct scrub_ctx *sctx) |
a2de733c AJ |
2047 | { |
2048 | struct scrub_bio *sbio; | |
2049 | ||
d9d181c1 | 2050 | if (sctx->curr == -1) |
1623edeb | 2051 | return; |
a2de733c | 2052 | |
d9d181c1 SB |
2053 | sbio = sctx->bios[sctx->curr]; |
2054 | sctx->curr = -1; | |
b6bfebc1 | 2055 | scrub_pending_bio_inc(sctx); |
03679ade | 2056 | btrfsic_submit_bio(READ, sbio->bio); |
a2de733c AJ |
2057 | } |
2058 | ||
ff023aac SB |
2059 | static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx, |
2060 | struct scrub_page *spage) | |
a2de733c | 2061 | { |
b5d67f64 | 2062 | struct scrub_block *sblock = spage->sblock; |
a2de733c | 2063 | struct scrub_bio *sbio; |
69f4cb52 | 2064 | int ret; |
a2de733c AJ |
2065 | |
2066 | again: | |
2067 | /* | |
2068 | * grab a fresh bio or wait for one to become available | |
2069 | */ | |
d9d181c1 SB |
2070 | while (sctx->curr == -1) { |
2071 | spin_lock(&sctx->list_lock); | |
2072 | sctx->curr = sctx->first_free; | |
2073 | if (sctx->curr != -1) { | |
2074 | sctx->first_free = sctx->bios[sctx->curr]->next_free; | |
2075 | sctx->bios[sctx->curr]->next_free = -1; | |
2076 | sctx->bios[sctx->curr]->page_count = 0; | |
2077 | spin_unlock(&sctx->list_lock); | |
a2de733c | 2078 | } else { |
d9d181c1 SB |
2079 | spin_unlock(&sctx->list_lock); |
2080 | wait_event(sctx->list_wait, sctx->first_free != -1); | |
a2de733c AJ |
2081 | } |
2082 | } | |
d9d181c1 | 2083 | sbio = sctx->bios[sctx->curr]; |
b5d67f64 | 2084 | if (sbio->page_count == 0) { |
69f4cb52 AJ |
2085 | struct bio *bio; |
2086 | ||
b5d67f64 SB |
2087 | sbio->physical = spage->physical; |
2088 | sbio->logical = spage->logical; | |
a36cf8b8 | 2089 | sbio->dev = spage->dev; |
b5d67f64 SB |
2090 | bio = sbio->bio; |
2091 | if (!bio) { | |
9be3395b | 2092 | bio = btrfs_io_bio_alloc(GFP_NOFS, sctx->pages_per_rd_bio); |
b5d67f64 SB |
2093 | if (!bio) |
2094 | return -ENOMEM; | |
2095 | sbio->bio = bio; | |
2096 | } | |
69f4cb52 AJ |
2097 | |
2098 | bio->bi_private = sbio; | |
2099 | bio->bi_end_io = scrub_bio_end_io; | |
a36cf8b8 | 2100 | bio->bi_bdev = sbio->dev->bdev; |
4f024f37 | 2101 | bio->bi_iter.bi_sector = sbio->physical >> 9; |
69f4cb52 | 2102 | sbio->err = 0; |
b5d67f64 SB |
2103 | } else if (sbio->physical + sbio->page_count * PAGE_SIZE != |
2104 | spage->physical || | |
2105 | sbio->logical + sbio->page_count * PAGE_SIZE != | |
a36cf8b8 SB |
2106 | spage->logical || |
2107 | sbio->dev != spage->dev) { | |
d9d181c1 | 2108 | scrub_submit(sctx); |
a2de733c AJ |
2109 | goto again; |
2110 | } | |
69f4cb52 | 2111 | |
b5d67f64 SB |
2112 | sbio->pagev[sbio->page_count] = spage; |
2113 | ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0); | |
2114 | if (ret != PAGE_SIZE) { | |
2115 | if (sbio->page_count < 1) { | |
2116 | bio_put(sbio->bio); | |
2117 | sbio->bio = NULL; | |
2118 | return -EIO; | |
2119 | } | |
d9d181c1 | 2120 | scrub_submit(sctx); |
69f4cb52 AJ |
2121 | goto again; |
2122 | } | |
2123 | ||
ff023aac | 2124 | scrub_block_get(sblock); /* one for the page added to the bio */ |
b5d67f64 SB |
2125 | atomic_inc(&sblock->outstanding_pages); |
2126 | sbio->page_count++; | |
ff023aac | 2127 | if (sbio->page_count == sctx->pages_per_rd_bio) |
d9d181c1 | 2128 | scrub_submit(sctx); |
b5d67f64 SB |
2129 | |
2130 | return 0; | |
2131 | } | |
2132 | ||
22365979 | 2133 | static void scrub_missing_raid56_end_io(struct bio *bio) |
73ff61db OS |
2134 | { |
2135 | struct scrub_block *sblock = bio->bi_private; | |
2136 | struct btrfs_fs_info *fs_info = sblock->sctx->dev_root->fs_info; | |
2137 | ||
22365979 | 2138 | if (bio->bi_error) |
73ff61db OS |
2139 | sblock->no_io_error_seen = 0; |
2140 | ||
2141 | btrfs_queue_work(fs_info->scrub_workers, &sblock->work); | |
2142 | } | |
2143 | ||
2144 | static void scrub_missing_raid56_worker(struct btrfs_work *work) | |
2145 | { | |
2146 | struct scrub_block *sblock = container_of(work, struct scrub_block, work); | |
2147 | struct scrub_ctx *sctx = sblock->sctx; | |
73ff61db OS |
2148 | u64 logical; |
2149 | struct btrfs_device *dev; | |
2150 | ||
73ff61db OS |
2151 | logical = sblock->pagev[0]->logical; |
2152 | dev = sblock->pagev[0]->dev; | |
2153 | ||
2154 | if (sblock->no_io_error_seen) { | |
ba7cf988 | 2155 | scrub_recheck_block_checksum(sblock); |
73ff61db OS |
2156 | } |
2157 | ||
2158 | if (!sblock->no_io_error_seen) { | |
2159 | spin_lock(&sctx->stat_lock); | |
2160 | sctx->stat.read_errors++; | |
2161 | spin_unlock(&sctx->stat_lock); | |
ba7cf988 | 2162 | btrfs_err_rl_in_rcu(sctx->dev_root->fs_info, |
b14af3b4 | 2163 | "IO error rebuilding logical %llu for dev %s", |
73ff61db OS |
2164 | logical, rcu_str_deref(dev->name)); |
2165 | } else if (sblock->header_error || sblock->checksum_error) { | |
2166 | spin_lock(&sctx->stat_lock); | |
2167 | sctx->stat.uncorrectable_errors++; | |
2168 | spin_unlock(&sctx->stat_lock); | |
ba7cf988 | 2169 | btrfs_err_rl_in_rcu(sctx->dev_root->fs_info, |
b14af3b4 | 2170 | "failed to rebuild valid logical %llu for dev %s", |
73ff61db OS |
2171 | logical, rcu_str_deref(dev->name)); |
2172 | } else { | |
2173 | scrub_write_block_to_dev_replace(sblock); | |
2174 | } | |
2175 | ||
2176 | scrub_block_put(sblock); | |
2177 | ||
2178 | if (sctx->is_dev_replace && | |
2179 | atomic_read(&sctx->wr_ctx.flush_all_writes)) { | |
2180 | mutex_lock(&sctx->wr_ctx.wr_lock); | |
2181 | scrub_wr_submit(sctx); | |
2182 | mutex_unlock(&sctx->wr_ctx.wr_lock); | |
2183 | } | |
2184 | ||
2185 | scrub_pending_bio_dec(sctx); | |
2186 | } | |
2187 | ||
2188 | static void scrub_missing_raid56_pages(struct scrub_block *sblock) | |
2189 | { | |
2190 | struct scrub_ctx *sctx = sblock->sctx; | |
2191 | struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info; | |
2192 | u64 length = sblock->page_count * PAGE_SIZE; | |
2193 | u64 logical = sblock->pagev[0]->logical; | |
2194 | struct btrfs_bio *bbio; | |
2195 | struct bio *bio; | |
2196 | struct btrfs_raid_bio *rbio; | |
2197 | int ret; | |
2198 | int i; | |
2199 | ||
2200 | ret = btrfs_map_sblock(fs_info, REQ_GET_READ_MIRRORS, logical, &length, | |
2201 | &bbio, 0, 1); | |
2202 | if (ret || !bbio || !bbio->raid_map) | |
2203 | goto bbio_out; | |
2204 | ||
2205 | if (WARN_ON(!sctx->is_dev_replace || | |
2206 | !(bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK))) { | |
2207 | /* | |
2208 | * We shouldn't be scrubbing a missing device. Even for dev | |
2209 | * replace, we should only get here for RAID 5/6. We either | |
2210 | * managed to mount something with no mirrors remaining or | |
2211 | * there's a bug in scrub_remap_extent()/btrfs_map_block(). | |
2212 | */ | |
2213 | goto bbio_out; | |
2214 | } | |
2215 | ||
2216 | bio = btrfs_io_bio_alloc(GFP_NOFS, 0); | |
2217 | if (!bio) | |
2218 | goto bbio_out; | |
2219 | ||
2220 | bio->bi_iter.bi_sector = logical >> 9; | |
2221 | bio->bi_private = sblock; | |
2222 | bio->bi_end_io = scrub_missing_raid56_end_io; | |
2223 | ||
2224 | rbio = raid56_alloc_missing_rbio(sctx->dev_root, bio, bbio, length); | |
2225 | if (!rbio) | |
2226 | goto rbio_out; | |
2227 | ||
2228 | for (i = 0; i < sblock->page_count; i++) { | |
2229 | struct scrub_page *spage = sblock->pagev[i]; | |
2230 | ||
2231 | raid56_add_scrub_pages(rbio, spage->page, spage->logical); | |
2232 | } | |
2233 | ||
2234 | btrfs_init_work(&sblock->work, btrfs_scrub_helper, | |
2235 | scrub_missing_raid56_worker, NULL, NULL); | |
2236 | scrub_block_get(sblock); | |
2237 | scrub_pending_bio_inc(sctx); | |
2238 | raid56_submit_missing_rbio(rbio); | |
2239 | return; | |
2240 | ||
2241 | rbio_out: | |
2242 | bio_put(bio); | |
2243 | bbio_out: | |
2244 | btrfs_put_bbio(bbio); | |
2245 | spin_lock(&sctx->stat_lock); | |
2246 | sctx->stat.malloc_errors++; | |
2247 | spin_unlock(&sctx->stat_lock); | |
2248 | } | |
2249 | ||
d9d181c1 | 2250 | static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len, |
a36cf8b8 | 2251 | u64 physical, struct btrfs_device *dev, u64 flags, |
ff023aac SB |
2252 | u64 gen, int mirror_num, u8 *csum, int force, |
2253 | u64 physical_for_dev_replace) | |
b5d67f64 SB |
2254 | { |
2255 | struct scrub_block *sblock; | |
2256 | int index; | |
2257 | ||
2258 | sblock = kzalloc(sizeof(*sblock), GFP_NOFS); | |
2259 | if (!sblock) { | |
d9d181c1 SB |
2260 | spin_lock(&sctx->stat_lock); |
2261 | sctx->stat.malloc_errors++; | |
2262 | spin_unlock(&sctx->stat_lock); | |
b5d67f64 | 2263 | return -ENOMEM; |
a2de733c | 2264 | } |
b5d67f64 | 2265 | |
7a9e9987 SB |
2266 | /* one ref inside this function, plus one for each page added to |
2267 | * a bio later on */ | |
57019345 | 2268 | atomic_set(&sblock->refs, 1); |
d9d181c1 | 2269 | sblock->sctx = sctx; |
b5d67f64 SB |
2270 | sblock->no_io_error_seen = 1; |
2271 | ||
2272 | for (index = 0; len > 0; index++) { | |
7a9e9987 | 2273 | struct scrub_page *spage; |
b5d67f64 SB |
2274 | u64 l = min_t(u64, len, PAGE_SIZE); |
2275 | ||
7a9e9987 SB |
2276 | spage = kzalloc(sizeof(*spage), GFP_NOFS); |
2277 | if (!spage) { | |
2278 | leave_nomem: | |
d9d181c1 SB |
2279 | spin_lock(&sctx->stat_lock); |
2280 | sctx->stat.malloc_errors++; | |
2281 | spin_unlock(&sctx->stat_lock); | |
7a9e9987 | 2282 | scrub_block_put(sblock); |
b5d67f64 SB |
2283 | return -ENOMEM; |
2284 | } | |
7a9e9987 SB |
2285 | BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK); |
2286 | scrub_page_get(spage); | |
2287 | sblock->pagev[index] = spage; | |
b5d67f64 | 2288 | spage->sblock = sblock; |
a36cf8b8 | 2289 | spage->dev = dev; |
b5d67f64 SB |
2290 | spage->flags = flags; |
2291 | spage->generation = gen; | |
2292 | spage->logical = logical; | |
2293 | spage->physical = physical; | |
ff023aac | 2294 | spage->physical_for_dev_replace = physical_for_dev_replace; |
b5d67f64 SB |
2295 | spage->mirror_num = mirror_num; |
2296 | if (csum) { | |
2297 | spage->have_csum = 1; | |
d9d181c1 | 2298 | memcpy(spage->csum, csum, sctx->csum_size); |
b5d67f64 SB |
2299 | } else { |
2300 | spage->have_csum = 0; | |
2301 | } | |
2302 | sblock->page_count++; | |
7a9e9987 SB |
2303 | spage->page = alloc_page(GFP_NOFS); |
2304 | if (!spage->page) | |
2305 | goto leave_nomem; | |
b5d67f64 SB |
2306 | len -= l; |
2307 | logical += l; | |
2308 | physical += l; | |
ff023aac | 2309 | physical_for_dev_replace += l; |
b5d67f64 SB |
2310 | } |
2311 | ||
7a9e9987 | 2312 | WARN_ON(sblock->page_count == 0); |
73ff61db OS |
2313 | if (dev->missing) { |
2314 | /* | |
2315 | * This case should only be hit for RAID 5/6 device replace. See | |
2316 | * the comment in scrub_missing_raid56_pages() for details. | |
2317 | */ | |
2318 | scrub_missing_raid56_pages(sblock); | |
2319 | } else { | |
2320 | for (index = 0; index < sblock->page_count; index++) { | |
2321 | struct scrub_page *spage = sblock->pagev[index]; | |
2322 | int ret; | |
1bc87793 | 2323 | |
73ff61db OS |
2324 | ret = scrub_add_page_to_rd_bio(sctx, spage); |
2325 | if (ret) { | |
2326 | scrub_block_put(sblock); | |
2327 | return ret; | |
2328 | } | |
b5d67f64 | 2329 | } |
a2de733c | 2330 | |
73ff61db OS |
2331 | if (force) |
2332 | scrub_submit(sctx); | |
2333 | } | |
a2de733c | 2334 | |
b5d67f64 SB |
2335 | /* last one frees, either here or in bio completion for last page */ |
2336 | scrub_block_put(sblock); | |
a2de733c AJ |
2337 | return 0; |
2338 | } | |
2339 | ||
4246a0b6 | 2340 | static void scrub_bio_end_io(struct bio *bio) |
b5d67f64 SB |
2341 | { |
2342 | struct scrub_bio *sbio = bio->bi_private; | |
a36cf8b8 | 2343 | struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info; |
b5d67f64 | 2344 | |
4246a0b6 | 2345 | sbio->err = bio->bi_error; |
b5d67f64 SB |
2346 | sbio->bio = bio; |
2347 | ||
0339ef2f | 2348 | btrfs_queue_work(fs_info->scrub_workers, &sbio->work); |
b5d67f64 SB |
2349 | } |
2350 | ||
2351 | static void scrub_bio_end_io_worker(struct btrfs_work *work) | |
2352 | { | |
2353 | struct scrub_bio *sbio = container_of(work, struct scrub_bio, work); | |
d9d181c1 | 2354 | struct scrub_ctx *sctx = sbio->sctx; |
b5d67f64 SB |
2355 | int i; |
2356 | ||
ff023aac | 2357 | BUG_ON(sbio->page_count > SCRUB_PAGES_PER_RD_BIO); |
b5d67f64 SB |
2358 | if (sbio->err) { |
2359 | for (i = 0; i < sbio->page_count; i++) { | |
2360 | struct scrub_page *spage = sbio->pagev[i]; | |
2361 | ||
2362 | spage->io_error = 1; | |
2363 | spage->sblock->no_io_error_seen = 0; | |
2364 | } | |
2365 | } | |
2366 | ||
2367 | /* now complete the scrub_block items that have all pages completed */ | |
2368 | for (i = 0; i < sbio->page_count; i++) { | |
2369 | struct scrub_page *spage = sbio->pagev[i]; | |
2370 | struct scrub_block *sblock = spage->sblock; | |
2371 | ||
2372 | if (atomic_dec_and_test(&sblock->outstanding_pages)) | |
2373 | scrub_block_complete(sblock); | |
2374 | scrub_block_put(sblock); | |
2375 | } | |
2376 | ||
b5d67f64 SB |
2377 | bio_put(sbio->bio); |
2378 | sbio->bio = NULL; | |
d9d181c1 SB |
2379 | spin_lock(&sctx->list_lock); |
2380 | sbio->next_free = sctx->first_free; | |
2381 | sctx->first_free = sbio->index; | |
2382 | spin_unlock(&sctx->list_lock); | |
ff023aac SB |
2383 | |
2384 | if (sctx->is_dev_replace && | |
2385 | atomic_read(&sctx->wr_ctx.flush_all_writes)) { | |
2386 | mutex_lock(&sctx->wr_ctx.wr_lock); | |
2387 | scrub_wr_submit(sctx); | |
2388 | mutex_unlock(&sctx->wr_ctx.wr_lock); | |
2389 | } | |
2390 | ||
b6bfebc1 | 2391 | scrub_pending_bio_dec(sctx); |
b5d67f64 SB |
2392 | } |
2393 | ||
5a6ac9ea MX |
2394 | static inline void __scrub_mark_bitmap(struct scrub_parity *sparity, |
2395 | unsigned long *bitmap, | |
2396 | u64 start, u64 len) | |
2397 | { | |
9d644a62 | 2398 | u32 offset; |
5a6ac9ea MX |
2399 | int nsectors; |
2400 | int sectorsize = sparity->sctx->dev_root->sectorsize; | |
2401 | ||
2402 | if (len >= sparity->stripe_len) { | |
2403 | bitmap_set(bitmap, 0, sparity->nsectors); | |
2404 | return; | |
2405 | } | |
2406 | ||
2407 | start -= sparity->logic_start; | |
47c5713f | 2408 | start = div_u64_rem(start, sparity->stripe_len, &offset); |
5a6ac9ea MX |
2409 | offset /= sectorsize; |
2410 | nsectors = (int)len / sectorsize; | |
2411 | ||
2412 | if (offset + nsectors <= sparity->nsectors) { | |
2413 | bitmap_set(bitmap, offset, nsectors); | |
2414 | return; | |
2415 | } | |
2416 | ||
2417 | bitmap_set(bitmap, offset, sparity->nsectors - offset); | |
2418 | bitmap_set(bitmap, 0, nsectors - (sparity->nsectors - offset)); | |
2419 | } | |
2420 | ||
2421 | static inline void scrub_parity_mark_sectors_error(struct scrub_parity *sparity, | |
2422 | u64 start, u64 len) | |
2423 | { | |
2424 | __scrub_mark_bitmap(sparity, sparity->ebitmap, start, len); | |
2425 | } | |
2426 | ||
2427 | static inline void scrub_parity_mark_sectors_data(struct scrub_parity *sparity, | |
2428 | u64 start, u64 len) | |
2429 | { | |
2430 | __scrub_mark_bitmap(sparity, sparity->dbitmap, start, len); | |
2431 | } | |
2432 | ||
b5d67f64 SB |
2433 | static void scrub_block_complete(struct scrub_block *sblock) |
2434 | { | |
5a6ac9ea MX |
2435 | int corrupted = 0; |
2436 | ||
ff023aac | 2437 | if (!sblock->no_io_error_seen) { |
5a6ac9ea | 2438 | corrupted = 1; |
b5d67f64 | 2439 | scrub_handle_errored_block(sblock); |
ff023aac SB |
2440 | } else { |
2441 | /* | |
2442 | * if has checksum error, write via repair mechanism in | |
2443 | * dev replace case, otherwise write here in dev replace | |
2444 | * case. | |
2445 | */ | |
5a6ac9ea MX |
2446 | corrupted = scrub_checksum(sblock); |
2447 | if (!corrupted && sblock->sctx->is_dev_replace) | |
ff023aac SB |
2448 | scrub_write_block_to_dev_replace(sblock); |
2449 | } | |
5a6ac9ea MX |
2450 | |
2451 | if (sblock->sparity && corrupted && !sblock->data_corrected) { | |
2452 | u64 start = sblock->pagev[0]->logical; | |
2453 | u64 end = sblock->pagev[sblock->page_count - 1]->logical + | |
2454 | PAGE_SIZE; | |
2455 | ||
2456 | scrub_parity_mark_sectors_error(sblock->sparity, | |
2457 | start, end - start); | |
2458 | } | |
b5d67f64 SB |
2459 | } |
2460 | ||
d9d181c1 | 2461 | static int scrub_find_csum(struct scrub_ctx *sctx, u64 logical, u64 len, |
a2de733c AJ |
2462 | u8 *csum) |
2463 | { | |
2464 | struct btrfs_ordered_sum *sum = NULL; | |
f51a4a18 | 2465 | unsigned long index; |
a2de733c | 2466 | unsigned long num_sectors; |
a2de733c | 2467 | |
d9d181c1 SB |
2468 | while (!list_empty(&sctx->csum_list)) { |
2469 | sum = list_first_entry(&sctx->csum_list, | |
a2de733c AJ |
2470 | struct btrfs_ordered_sum, list); |
2471 | if (sum->bytenr > logical) | |
2472 | return 0; | |
2473 | if (sum->bytenr + sum->len > logical) | |
2474 | break; | |
2475 | ||
d9d181c1 | 2476 | ++sctx->stat.csum_discards; |
a2de733c AJ |
2477 | list_del(&sum->list); |
2478 | kfree(sum); | |
2479 | sum = NULL; | |
2480 | } | |
2481 | if (!sum) | |
2482 | return 0; | |
2483 | ||
f51a4a18 | 2484 | index = ((u32)(logical - sum->bytenr)) / sctx->sectorsize; |
d9d181c1 | 2485 | num_sectors = sum->len / sctx->sectorsize; |
f51a4a18 MX |
2486 | memcpy(csum, sum->sums + index, sctx->csum_size); |
2487 | if (index == num_sectors - 1) { | |
a2de733c AJ |
2488 | list_del(&sum->list); |
2489 | kfree(sum); | |
2490 | } | |
f51a4a18 | 2491 | return 1; |
a2de733c AJ |
2492 | } |
2493 | ||
2494 | /* scrub extent tries to collect up to 64 kB for each bio */ | |
d9d181c1 | 2495 | static int scrub_extent(struct scrub_ctx *sctx, u64 logical, u64 len, |
a36cf8b8 | 2496 | u64 physical, struct btrfs_device *dev, u64 flags, |
ff023aac | 2497 | u64 gen, int mirror_num, u64 physical_for_dev_replace) |
a2de733c AJ |
2498 | { |
2499 | int ret; | |
2500 | u8 csum[BTRFS_CSUM_SIZE]; | |
b5d67f64 SB |
2501 | u32 blocksize; |
2502 | ||
2503 | if (flags & BTRFS_EXTENT_FLAG_DATA) { | |
d9d181c1 SB |
2504 | blocksize = sctx->sectorsize; |
2505 | spin_lock(&sctx->stat_lock); | |
2506 | sctx->stat.data_extents_scrubbed++; | |
2507 | sctx->stat.data_bytes_scrubbed += len; | |
2508 | spin_unlock(&sctx->stat_lock); | |
b5d67f64 | 2509 | } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { |
d9d181c1 SB |
2510 | blocksize = sctx->nodesize; |
2511 | spin_lock(&sctx->stat_lock); | |
2512 | sctx->stat.tree_extents_scrubbed++; | |
2513 | sctx->stat.tree_bytes_scrubbed += len; | |
2514 | spin_unlock(&sctx->stat_lock); | |
b5d67f64 | 2515 | } else { |
d9d181c1 | 2516 | blocksize = sctx->sectorsize; |
ff023aac | 2517 | WARN_ON(1); |
b5d67f64 | 2518 | } |
a2de733c AJ |
2519 | |
2520 | while (len) { | |
b5d67f64 | 2521 | u64 l = min_t(u64, len, blocksize); |
a2de733c AJ |
2522 | int have_csum = 0; |
2523 | ||
2524 | if (flags & BTRFS_EXTENT_FLAG_DATA) { | |
2525 | /* push csums to sbio */ | |
d9d181c1 | 2526 | have_csum = scrub_find_csum(sctx, logical, l, csum); |
a2de733c | 2527 | if (have_csum == 0) |
d9d181c1 | 2528 | ++sctx->stat.no_csum; |
ff023aac SB |
2529 | if (sctx->is_dev_replace && !have_csum) { |
2530 | ret = copy_nocow_pages(sctx, logical, l, | |
2531 | mirror_num, | |
2532 | physical_for_dev_replace); | |
2533 | goto behind_scrub_pages; | |
2534 | } | |
a2de733c | 2535 | } |
a36cf8b8 | 2536 | ret = scrub_pages(sctx, logical, l, physical, dev, flags, gen, |
ff023aac SB |
2537 | mirror_num, have_csum ? csum : NULL, 0, |
2538 | physical_for_dev_replace); | |
2539 | behind_scrub_pages: | |
a2de733c AJ |
2540 | if (ret) |
2541 | return ret; | |
2542 | len -= l; | |
2543 | logical += l; | |
2544 | physical += l; | |
ff023aac | 2545 | physical_for_dev_replace += l; |
a2de733c AJ |
2546 | } |
2547 | return 0; | |
2548 | } | |
2549 | ||
5a6ac9ea MX |
2550 | static int scrub_pages_for_parity(struct scrub_parity *sparity, |
2551 | u64 logical, u64 len, | |
2552 | u64 physical, struct btrfs_device *dev, | |
2553 | u64 flags, u64 gen, int mirror_num, u8 *csum) | |
2554 | { | |
2555 | struct scrub_ctx *sctx = sparity->sctx; | |
2556 | struct scrub_block *sblock; | |
2557 | int index; | |
2558 | ||
2559 | sblock = kzalloc(sizeof(*sblock), GFP_NOFS); | |
2560 | if (!sblock) { | |
2561 | spin_lock(&sctx->stat_lock); | |
2562 | sctx->stat.malloc_errors++; | |
2563 | spin_unlock(&sctx->stat_lock); | |
2564 | return -ENOMEM; | |
2565 | } | |
2566 | ||
2567 | /* one ref inside this function, plus one for each page added to | |
2568 | * a bio later on */ | |
57019345 | 2569 | atomic_set(&sblock->refs, 1); |
5a6ac9ea MX |
2570 | sblock->sctx = sctx; |
2571 | sblock->no_io_error_seen = 1; | |
2572 | sblock->sparity = sparity; | |
2573 | scrub_parity_get(sparity); | |
2574 | ||
2575 | for (index = 0; len > 0; index++) { | |
2576 | struct scrub_page *spage; | |
2577 | u64 l = min_t(u64, len, PAGE_SIZE); | |
2578 | ||
2579 | spage = kzalloc(sizeof(*spage), GFP_NOFS); | |
2580 | if (!spage) { | |
2581 | leave_nomem: | |
2582 | spin_lock(&sctx->stat_lock); | |
2583 | sctx->stat.malloc_errors++; | |
2584 | spin_unlock(&sctx->stat_lock); | |
2585 | scrub_block_put(sblock); | |
2586 | return -ENOMEM; | |
2587 | } | |
2588 | BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK); | |
2589 | /* For scrub block */ | |
2590 | scrub_page_get(spage); | |
2591 | sblock->pagev[index] = spage; | |
2592 | /* For scrub parity */ | |
2593 | scrub_page_get(spage); | |
2594 | list_add_tail(&spage->list, &sparity->spages); | |
2595 | spage->sblock = sblock; | |
2596 | spage->dev = dev; | |
2597 | spage->flags = flags; | |
2598 | spage->generation = gen; | |
2599 | spage->logical = logical; | |
2600 | spage->physical = physical; | |
2601 | spage->mirror_num = mirror_num; | |
2602 | if (csum) { | |
2603 | spage->have_csum = 1; | |
2604 | memcpy(spage->csum, csum, sctx->csum_size); | |
2605 | } else { | |
2606 | spage->have_csum = 0; | |
2607 | } | |
2608 | sblock->page_count++; | |
2609 | spage->page = alloc_page(GFP_NOFS); | |
2610 | if (!spage->page) | |
2611 | goto leave_nomem; | |
2612 | len -= l; | |
2613 | logical += l; | |
2614 | physical += l; | |
2615 | } | |
2616 | ||
2617 | WARN_ON(sblock->page_count == 0); | |
2618 | for (index = 0; index < sblock->page_count; index++) { | |
2619 | struct scrub_page *spage = sblock->pagev[index]; | |
2620 | int ret; | |
2621 | ||
2622 | ret = scrub_add_page_to_rd_bio(sctx, spage); | |
2623 | if (ret) { | |
2624 | scrub_block_put(sblock); | |
2625 | return ret; | |
2626 | } | |
2627 | } | |
2628 | ||
2629 | /* last one frees, either here or in bio completion for last page */ | |
2630 | scrub_block_put(sblock); | |
2631 | return 0; | |
2632 | } | |
2633 | ||
2634 | static int scrub_extent_for_parity(struct scrub_parity *sparity, | |
2635 | u64 logical, u64 len, | |
2636 | u64 physical, struct btrfs_device *dev, | |
2637 | u64 flags, u64 gen, int mirror_num) | |
2638 | { | |
2639 | struct scrub_ctx *sctx = sparity->sctx; | |
2640 | int ret; | |
2641 | u8 csum[BTRFS_CSUM_SIZE]; | |
2642 | u32 blocksize; | |
2643 | ||
4a770891 OS |
2644 | if (dev->missing) { |
2645 | scrub_parity_mark_sectors_error(sparity, logical, len); | |
2646 | return 0; | |
2647 | } | |
2648 | ||
5a6ac9ea MX |
2649 | if (flags & BTRFS_EXTENT_FLAG_DATA) { |
2650 | blocksize = sctx->sectorsize; | |
2651 | } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { | |
2652 | blocksize = sctx->nodesize; | |
2653 | } else { | |
2654 | blocksize = sctx->sectorsize; | |
2655 | WARN_ON(1); | |
2656 | } | |
2657 | ||
2658 | while (len) { | |
2659 | u64 l = min_t(u64, len, blocksize); | |
2660 | int have_csum = 0; | |
2661 | ||
2662 | if (flags & BTRFS_EXTENT_FLAG_DATA) { | |
2663 | /* push csums to sbio */ | |
2664 | have_csum = scrub_find_csum(sctx, logical, l, csum); | |
2665 | if (have_csum == 0) | |
2666 | goto skip; | |
2667 | } | |
2668 | ret = scrub_pages_for_parity(sparity, logical, l, physical, dev, | |
2669 | flags, gen, mirror_num, | |
2670 | have_csum ? csum : NULL); | |
5a6ac9ea MX |
2671 | if (ret) |
2672 | return ret; | |
6b6d24b3 | 2673 | skip: |
5a6ac9ea MX |
2674 | len -= l; |
2675 | logical += l; | |
2676 | physical += l; | |
2677 | } | |
2678 | return 0; | |
2679 | } | |
2680 | ||
3b080b25 WS |
2681 | /* |
2682 | * Given a physical address, this will calculate it's | |
2683 | * logical offset. if this is a parity stripe, it will return | |
2684 | * the most left data stripe's logical offset. | |
2685 | * | |
2686 | * return 0 if it is a data stripe, 1 means parity stripe. | |
2687 | */ | |
2688 | static int get_raid56_logic_offset(u64 physical, int num, | |
5a6ac9ea MX |
2689 | struct map_lookup *map, u64 *offset, |
2690 | u64 *stripe_start) | |
3b080b25 WS |
2691 | { |
2692 | int i; | |
2693 | int j = 0; | |
2694 | u64 stripe_nr; | |
2695 | u64 last_offset; | |
9d644a62 DS |
2696 | u32 stripe_index; |
2697 | u32 rot; | |
3b080b25 WS |
2698 | |
2699 | last_offset = (physical - map->stripes[num].physical) * | |
2700 | nr_data_stripes(map); | |
5a6ac9ea MX |
2701 | if (stripe_start) |
2702 | *stripe_start = last_offset; | |
2703 | ||
3b080b25 WS |
2704 | *offset = last_offset; |
2705 | for (i = 0; i < nr_data_stripes(map); i++) { | |
2706 | *offset = last_offset + i * map->stripe_len; | |
2707 | ||
b8b93add DS |
2708 | stripe_nr = div_u64(*offset, map->stripe_len); |
2709 | stripe_nr = div_u64(stripe_nr, nr_data_stripes(map)); | |
3b080b25 WS |
2710 | |
2711 | /* Work out the disk rotation on this stripe-set */ | |
47c5713f | 2712 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, &rot); |
3b080b25 WS |
2713 | /* calculate which stripe this data locates */ |
2714 | rot += i; | |
e4fbaee2 | 2715 | stripe_index = rot % map->num_stripes; |
3b080b25 WS |
2716 | if (stripe_index == num) |
2717 | return 0; | |
2718 | if (stripe_index < num) | |
2719 | j++; | |
2720 | } | |
2721 | *offset = last_offset + j * map->stripe_len; | |
2722 | return 1; | |
2723 | } | |
2724 | ||
5a6ac9ea MX |
2725 | static void scrub_free_parity(struct scrub_parity *sparity) |
2726 | { | |
2727 | struct scrub_ctx *sctx = sparity->sctx; | |
2728 | struct scrub_page *curr, *next; | |
2729 | int nbits; | |
2730 | ||
2731 | nbits = bitmap_weight(sparity->ebitmap, sparity->nsectors); | |
2732 | if (nbits) { | |
2733 | spin_lock(&sctx->stat_lock); | |
2734 | sctx->stat.read_errors += nbits; | |
2735 | sctx->stat.uncorrectable_errors += nbits; | |
2736 | spin_unlock(&sctx->stat_lock); | |
2737 | } | |
2738 | ||
2739 | list_for_each_entry_safe(curr, next, &sparity->spages, list) { | |
2740 | list_del_init(&curr->list); | |
2741 | scrub_page_put(curr); | |
2742 | } | |
2743 | ||
2744 | kfree(sparity); | |
2745 | } | |
2746 | ||
20b2e302 ZL |
2747 | static void scrub_parity_bio_endio_worker(struct btrfs_work *work) |
2748 | { | |
2749 | struct scrub_parity *sparity = container_of(work, struct scrub_parity, | |
2750 | work); | |
2751 | struct scrub_ctx *sctx = sparity->sctx; | |
2752 | ||
2753 | scrub_free_parity(sparity); | |
2754 | scrub_pending_bio_dec(sctx); | |
2755 | } | |
2756 | ||
4246a0b6 | 2757 | static void scrub_parity_bio_endio(struct bio *bio) |
5a6ac9ea MX |
2758 | { |
2759 | struct scrub_parity *sparity = (struct scrub_parity *)bio->bi_private; | |
5a6ac9ea | 2760 | |
4246a0b6 | 2761 | if (bio->bi_error) |
5a6ac9ea MX |
2762 | bitmap_or(sparity->ebitmap, sparity->ebitmap, sparity->dbitmap, |
2763 | sparity->nsectors); | |
2764 | ||
5a6ac9ea | 2765 | bio_put(bio); |
20b2e302 ZL |
2766 | |
2767 | btrfs_init_work(&sparity->work, btrfs_scrubparity_helper, | |
2768 | scrub_parity_bio_endio_worker, NULL, NULL); | |
2769 | btrfs_queue_work(sparity->sctx->dev_root->fs_info->scrub_parity_workers, | |
2770 | &sparity->work); | |
5a6ac9ea MX |
2771 | } |
2772 | ||
2773 | static void scrub_parity_check_and_repair(struct scrub_parity *sparity) | |
2774 | { | |
2775 | struct scrub_ctx *sctx = sparity->sctx; | |
2776 | struct bio *bio; | |
2777 | struct btrfs_raid_bio *rbio; | |
2778 | struct scrub_page *spage; | |
2779 | struct btrfs_bio *bbio = NULL; | |
5a6ac9ea MX |
2780 | u64 length; |
2781 | int ret; | |
2782 | ||
2783 | if (!bitmap_andnot(sparity->dbitmap, sparity->dbitmap, sparity->ebitmap, | |
2784 | sparity->nsectors)) | |
2785 | goto out; | |
2786 | ||
a0dd59de | 2787 | length = sparity->logic_end - sparity->logic_start; |
76035976 | 2788 | ret = btrfs_map_sblock(sctx->dev_root->fs_info, WRITE, |
5a6ac9ea | 2789 | sparity->logic_start, |
8e5cfb55 ZL |
2790 | &length, &bbio, 0, 1); |
2791 | if (ret || !bbio || !bbio->raid_map) | |
5a6ac9ea MX |
2792 | goto bbio_out; |
2793 | ||
2794 | bio = btrfs_io_bio_alloc(GFP_NOFS, 0); | |
2795 | if (!bio) | |
2796 | goto bbio_out; | |
2797 | ||
2798 | bio->bi_iter.bi_sector = sparity->logic_start >> 9; | |
2799 | bio->bi_private = sparity; | |
2800 | bio->bi_end_io = scrub_parity_bio_endio; | |
2801 | ||
2802 | rbio = raid56_parity_alloc_scrub_rbio(sctx->dev_root, bio, bbio, | |
8e5cfb55 | 2803 | length, sparity->scrub_dev, |
5a6ac9ea MX |
2804 | sparity->dbitmap, |
2805 | sparity->nsectors); | |
2806 | if (!rbio) | |
2807 | goto rbio_out; | |
2808 | ||
2809 | list_for_each_entry(spage, &sparity->spages, list) | |
b4ee1782 | 2810 | raid56_add_scrub_pages(rbio, spage->page, spage->logical); |
5a6ac9ea MX |
2811 | |
2812 | scrub_pending_bio_inc(sctx); | |
2813 | raid56_parity_submit_scrub_rbio(rbio); | |
2814 | return; | |
2815 | ||
2816 | rbio_out: | |
2817 | bio_put(bio); | |
2818 | bbio_out: | |
6e9606d2 | 2819 | btrfs_put_bbio(bbio); |
5a6ac9ea MX |
2820 | bitmap_or(sparity->ebitmap, sparity->ebitmap, sparity->dbitmap, |
2821 | sparity->nsectors); | |
2822 | spin_lock(&sctx->stat_lock); | |
2823 | sctx->stat.malloc_errors++; | |
2824 | spin_unlock(&sctx->stat_lock); | |
2825 | out: | |
2826 | scrub_free_parity(sparity); | |
2827 | } | |
2828 | ||
2829 | static inline int scrub_calc_parity_bitmap_len(int nsectors) | |
2830 | { | |
2831 | return DIV_ROUND_UP(nsectors, BITS_PER_LONG) * (BITS_PER_LONG / 8); | |
2832 | } | |
2833 | ||
2834 | static void scrub_parity_get(struct scrub_parity *sparity) | |
2835 | { | |
57019345 | 2836 | atomic_inc(&sparity->refs); |
5a6ac9ea MX |
2837 | } |
2838 | ||
2839 | static void scrub_parity_put(struct scrub_parity *sparity) | |
2840 | { | |
57019345 | 2841 | if (!atomic_dec_and_test(&sparity->refs)) |
5a6ac9ea MX |
2842 | return; |
2843 | ||
2844 | scrub_parity_check_and_repair(sparity); | |
2845 | } | |
2846 | ||
2847 | static noinline_for_stack int scrub_raid56_parity(struct scrub_ctx *sctx, | |
2848 | struct map_lookup *map, | |
2849 | struct btrfs_device *sdev, | |
2850 | struct btrfs_path *path, | |
2851 | u64 logic_start, | |
2852 | u64 logic_end) | |
2853 | { | |
2854 | struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info; | |
2855 | struct btrfs_root *root = fs_info->extent_root; | |
2856 | struct btrfs_root *csum_root = fs_info->csum_root; | |
2857 | struct btrfs_extent_item *extent; | |
4a770891 | 2858 | struct btrfs_bio *bbio = NULL; |
5a6ac9ea MX |
2859 | u64 flags; |
2860 | int ret; | |
2861 | int slot; | |
2862 | struct extent_buffer *l; | |
2863 | struct btrfs_key key; | |
2864 | u64 generation; | |
2865 | u64 extent_logical; | |
2866 | u64 extent_physical; | |
2867 | u64 extent_len; | |
4a770891 | 2868 | u64 mapped_length; |
5a6ac9ea MX |
2869 | struct btrfs_device *extent_dev; |
2870 | struct scrub_parity *sparity; | |
2871 | int nsectors; | |
2872 | int bitmap_len; | |
2873 | int extent_mirror_num; | |
2874 | int stop_loop = 0; | |
2875 | ||
2876 | nsectors = map->stripe_len / root->sectorsize; | |
2877 | bitmap_len = scrub_calc_parity_bitmap_len(nsectors); | |
2878 | sparity = kzalloc(sizeof(struct scrub_parity) + 2 * bitmap_len, | |
2879 | GFP_NOFS); | |
2880 | if (!sparity) { | |
2881 | spin_lock(&sctx->stat_lock); | |
2882 | sctx->stat.malloc_errors++; | |
2883 | spin_unlock(&sctx->stat_lock); | |
2884 | return -ENOMEM; | |
2885 | } | |
2886 | ||
2887 | sparity->stripe_len = map->stripe_len; | |
2888 | sparity->nsectors = nsectors; | |
2889 | sparity->sctx = sctx; | |
2890 | sparity->scrub_dev = sdev; | |
2891 | sparity->logic_start = logic_start; | |
2892 | sparity->logic_end = logic_end; | |
57019345 | 2893 | atomic_set(&sparity->refs, 1); |
5a6ac9ea MX |
2894 | INIT_LIST_HEAD(&sparity->spages); |
2895 | sparity->dbitmap = sparity->bitmap; | |
2896 | sparity->ebitmap = (void *)sparity->bitmap + bitmap_len; | |
2897 | ||
2898 | ret = 0; | |
2899 | while (logic_start < logic_end) { | |
2900 | if (btrfs_fs_incompat(fs_info, SKINNY_METADATA)) | |
2901 | key.type = BTRFS_METADATA_ITEM_KEY; | |
2902 | else | |
2903 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2904 | key.objectid = logic_start; | |
2905 | key.offset = (u64)-1; | |
2906 | ||
2907 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2908 | if (ret < 0) | |
2909 | goto out; | |
2910 | ||
2911 | if (ret > 0) { | |
2912 | ret = btrfs_previous_extent_item(root, path, 0); | |
2913 | if (ret < 0) | |
2914 | goto out; | |
2915 | if (ret > 0) { | |
2916 | btrfs_release_path(path); | |
2917 | ret = btrfs_search_slot(NULL, root, &key, | |
2918 | path, 0, 0); | |
2919 | if (ret < 0) | |
2920 | goto out; | |
2921 | } | |
2922 | } | |
2923 | ||
2924 | stop_loop = 0; | |
2925 | while (1) { | |
2926 | u64 bytes; | |
2927 | ||
2928 | l = path->nodes[0]; | |
2929 | slot = path->slots[0]; | |
2930 | if (slot >= btrfs_header_nritems(l)) { | |
2931 | ret = btrfs_next_leaf(root, path); | |
2932 | if (ret == 0) | |
2933 | continue; | |
2934 | if (ret < 0) | |
2935 | goto out; | |
2936 | ||
2937 | stop_loop = 1; | |
2938 | break; | |
2939 | } | |
2940 | btrfs_item_key_to_cpu(l, &key, slot); | |
2941 | ||
d7cad238 ZL |
2942 | if (key.type != BTRFS_EXTENT_ITEM_KEY && |
2943 | key.type != BTRFS_METADATA_ITEM_KEY) | |
2944 | goto next; | |
2945 | ||
5a6ac9ea MX |
2946 | if (key.type == BTRFS_METADATA_ITEM_KEY) |
2947 | bytes = root->nodesize; | |
2948 | else | |
2949 | bytes = key.offset; | |
2950 | ||
2951 | if (key.objectid + bytes <= logic_start) | |
2952 | goto next; | |
2953 | ||
a0dd59de | 2954 | if (key.objectid >= logic_end) { |
5a6ac9ea MX |
2955 | stop_loop = 1; |
2956 | break; | |
2957 | } | |
2958 | ||
2959 | while (key.objectid >= logic_start + map->stripe_len) | |
2960 | logic_start += map->stripe_len; | |
2961 | ||
2962 | extent = btrfs_item_ptr(l, slot, | |
2963 | struct btrfs_extent_item); | |
2964 | flags = btrfs_extent_flags(l, extent); | |
2965 | generation = btrfs_extent_generation(l, extent); | |
2966 | ||
a323e813 ZL |
2967 | if ((flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) && |
2968 | (key.objectid < logic_start || | |
2969 | key.objectid + bytes > | |
2970 | logic_start + map->stripe_len)) { | |
2971 | btrfs_err(fs_info, "scrub: tree block %llu spanning stripes, ignored. logical=%llu", | |
2972 | key.objectid, logic_start); | |
9799d2c3 ZL |
2973 | spin_lock(&sctx->stat_lock); |
2974 | sctx->stat.uncorrectable_errors++; | |
2975 | spin_unlock(&sctx->stat_lock); | |
5a6ac9ea MX |
2976 | goto next; |
2977 | } | |
2978 | again: | |
2979 | extent_logical = key.objectid; | |
2980 | extent_len = bytes; | |
2981 | ||
2982 | if (extent_logical < logic_start) { | |
2983 | extent_len -= logic_start - extent_logical; | |
2984 | extent_logical = logic_start; | |
2985 | } | |
2986 | ||
2987 | if (extent_logical + extent_len > | |
2988 | logic_start + map->stripe_len) | |
2989 | extent_len = logic_start + map->stripe_len - | |
2990 | extent_logical; | |
2991 | ||
2992 | scrub_parity_mark_sectors_data(sparity, extent_logical, | |
2993 | extent_len); | |
2994 | ||
4a770891 OS |
2995 | mapped_length = extent_len; |
2996 | ret = btrfs_map_block(fs_info, READ, extent_logical, | |
2997 | &mapped_length, &bbio, 0); | |
2998 | if (!ret) { | |
2999 | if (!bbio || mapped_length < extent_len) | |
3000 | ret = -EIO; | |
3001 | } | |
3002 | if (ret) { | |
3003 | btrfs_put_bbio(bbio); | |
3004 | goto out; | |
3005 | } | |
3006 | extent_physical = bbio->stripes[0].physical; | |
3007 | extent_mirror_num = bbio->mirror_num; | |
3008 | extent_dev = bbio->stripes[0].dev; | |
3009 | btrfs_put_bbio(bbio); | |
5a6ac9ea MX |
3010 | |
3011 | ret = btrfs_lookup_csums_range(csum_root, | |
3012 | extent_logical, | |
3013 | extent_logical + extent_len - 1, | |
3014 | &sctx->csum_list, 1); | |
3015 | if (ret) | |
3016 | goto out; | |
3017 | ||
3018 | ret = scrub_extent_for_parity(sparity, extent_logical, | |
3019 | extent_len, | |
3020 | extent_physical, | |
3021 | extent_dev, flags, | |
3022 | generation, | |
3023 | extent_mirror_num); | |
6fa96d72 ZL |
3024 | |
3025 | scrub_free_csums(sctx); | |
3026 | ||
5a6ac9ea MX |
3027 | if (ret) |
3028 | goto out; | |
3029 | ||
5a6ac9ea MX |
3030 | if (extent_logical + extent_len < |
3031 | key.objectid + bytes) { | |
3032 | logic_start += map->stripe_len; | |
3033 | ||
3034 | if (logic_start >= logic_end) { | |
3035 | stop_loop = 1; | |
3036 | break; | |
3037 | } | |
3038 | ||
3039 | if (logic_start < key.objectid + bytes) { | |
3040 | cond_resched(); | |
3041 | goto again; | |
3042 | } | |
3043 | } | |
3044 | next: | |
3045 | path->slots[0]++; | |
3046 | } | |
3047 | ||
3048 | btrfs_release_path(path); | |
3049 | ||
3050 | if (stop_loop) | |
3051 | break; | |
3052 | ||
3053 | logic_start += map->stripe_len; | |
3054 | } | |
3055 | out: | |
3056 | if (ret < 0) | |
3057 | scrub_parity_mark_sectors_error(sparity, logic_start, | |
a0dd59de | 3058 | logic_end - logic_start); |
5a6ac9ea MX |
3059 | scrub_parity_put(sparity); |
3060 | scrub_submit(sctx); | |
3061 | mutex_lock(&sctx->wr_ctx.wr_lock); | |
3062 | scrub_wr_submit(sctx); | |
3063 | mutex_unlock(&sctx->wr_ctx.wr_lock); | |
3064 | ||
3065 | btrfs_release_path(path); | |
3066 | return ret < 0 ? ret : 0; | |
3067 | } | |
3068 | ||
d9d181c1 | 3069 | static noinline_for_stack int scrub_stripe(struct scrub_ctx *sctx, |
a36cf8b8 SB |
3070 | struct map_lookup *map, |
3071 | struct btrfs_device *scrub_dev, | |
ff023aac SB |
3072 | int num, u64 base, u64 length, |
3073 | int is_dev_replace) | |
a2de733c | 3074 | { |
5a6ac9ea | 3075 | struct btrfs_path *path, *ppath; |
a36cf8b8 | 3076 | struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info; |
a2de733c AJ |
3077 | struct btrfs_root *root = fs_info->extent_root; |
3078 | struct btrfs_root *csum_root = fs_info->csum_root; | |
3079 | struct btrfs_extent_item *extent; | |
e7786c3a | 3080 | struct blk_plug plug; |
a2de733c AJ |
3081 | u64 flags; |
3082 | int ret; | |
3083 | int slot; | |
a2de733c | 3084 | u64 nstripes; |
a2de733c AJ |
3085 | struct extent_buffer *l; |
3086 | struct btrfs_key key; | |
3087 | u64 physical; | |
3088 | u64 logical; | |
625f1c8d | 3089 | u64 logic_end; |
3b080b25 | 3090 | u64 physical_end; |
a2de733c | 3091 | u64 generation; |
e12fa9cd | 3092 | int mirror_num; |
7a26285e AJ |
3093 | struct reada_control *reada1; |
3094 | struct reada_control *reada2; | |
3095 | struct btrfs_key key_start; | |
3096 | struct btrfs_key key_end; | |
a2de733c AJ |
3097 | u64 increment = map->stripe_len; |
3098 | u64 offset; | |
ff023aac SB |
3099 | u64 extent_logical; |
3100 | u64 extent_physical; | |
3101 | u64 extent_len; | |
5a6ac9ea MX |
3102 | u64 stripe_logical; |
3103 | u64 stripe_end; | |
ff023aac SB |
3104 | struct btrfs_device *extent_dev; |
3105 | int extent_mirror_num; | |
3b080b25 | 3106 | int stop_loop = 0; |
53b381b3 | 3107 | |
3b080b25 | 3108 | physical = map->stripes[num].physical; |
a2de733c | 3109 | offset = 0; |
b8b93add | 3110 | nstripes = div_u64(length, map->stripe_len); |
a2de733c AJ |
3111 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
3112 | offset = map->stripe_len * num; | |
3113 | increment = map->stripe_len * map->num_stripes; | |
193ea74b | 3114 | mirror_num = 1; |
a2de733c AJ |
3115 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
3116 | int factor = map->num_stripes / map->sub_stripes; | |
3117 | offset = map->stripe_len * (num / map->sub_stripes); | |
3118 | increment = map->stripe_len * factor; | |
193ea74b | 3119 | mirror_num = num % map->sub_stripes + 1; |
a2de733c AJ |
3120 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { |
3121 | increment = map->stripe_len; | |
193ea74b | 3122 | mirror_num = num % map->num_stripes + 1; |
a2de733c AJ |
3123 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
3124 | increment = map->stripe_len; | |
193ea74b | 3125 | mirror_num = num % map->num_stripes + 1; |
ffe2d203 | 3126 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
5a6ac9ea | 3127 | get_raid56_logic_offset(physical, num, map, &offset, NULL); |
3b080b25 WS |
3128 | increment = map->stripe_len * nr_data_stripes(map); |
3129 | mirror_num = 1; | |
a2de733c AJ |
3130 | } else { |
3131 | increment = map->stripe_len; | |
193ea74b | 3132 | mirror_num = 1; |
a2de733c AJ |
3133 | } |
3134 | ||
3135 | path = btrfs_alloc_path(); | |
3136 | if (!path) | |
3137 | return -ENOMEM; | |
3138 | ||
5a6ac9ea MX |
3139 | ppath = btrfs_alloc_path(); |
3140 | if (!ppath) { | |
379d6854 | 3141 | btrfs_free_path(path); |
5a6ac9ea MX |
3142 | return -ENOMEM; |
3143 | } | |
3144 | ||
b5d67f64 SB |
3145 | /* |
3146 | * work on commit root. The related disk blocks are static as | |
3147 | * long as COW is applied. This means, it is save to rewrite | |
3148 | * them to repair disk errors without any race conditions | |
3149 | */ | |
a2de733c AJ |
3150 | path->search_commit_root = 1; |
3151 | path->skip_locking = 1; | |
3152 | ||
063c54dc GH |
3153 | ppath->search_commit_root = 1; |
3154 | ppath->skip_locking = 1; | |
a2de733c | 3155 | /* |
7a26285e AJ |
3156 | * trigger the readahead for extent tree csum tree and wait for |
3157 | * completion. During readahead, the scrub is officially paused | |
3158 | * to not hold off transaction commits | |
a2de733c AJ |
3159 | */ |
3160 | logical = base + offset; | |
3b080b25 | 3161 | physical_end = physical + nstripes * map->stripe_len; |
ffe2d203 | 3162 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
3b080b25 | 3163 | get_raid56_logic_offset(physical_end, num, |
5a6ac9ea | 3164 | map, &logic_end, NULL); |
3b080b25 WS |
3165 | logic_end += base; |
3166 | } else { | |
3167 | logic_end = logical + increment * nstripes; | |
3168 | } | |
d9d181c1 | 3169 | wait_event(sctx->list_wait, |
b6bfebc1 | 3170 | atomic_read(&sctx->bios_in_flight) == 0); |
cb7ab021 | 3171 | scrub_blocked_if_needed(fs_info); |
7a26285e AJ |
3172 | |
3173 | /* FIXME it might be better to start readahead at commit root */ | |
3174 | key_start.objectid = logical; | |
3175 | key_start.type = BTRFS_EXTENT_ITEM_KEY; | |
3176 | key_start.offset = (u64)0; | |
3b080b25 | 3177 | key_end.objectid = logic_end; |
3173a18f JB |
3178 | key_end.type = BTRFS_METADATA_ITEM_KEY; |
3179 | key_end.offset = (u64)-1; | |
7a26285e AJ |
3180 | reada1 = btrfs_reada_add(root, &key_start, &key_end); |
3181 | ||
3182 | key_start.objectid = BTRFS_EXTENT_CSUM_OBJECTID; | |
3183 | key_start.type = BTRFS_EXTENT_CSUM_KEY; | |
3184 | key_start.offset = logical; | |
3185 | key_end.objectid = BTRFS_EXTENT_CSUM_OBJECTID; | |
3186 | key_end.type = BTRFS_EXTENT_CSUM_KEY; | |
3b080b25 | 3187 | key_end.offset = logic_end; |
7a26285e AJ |
3188 | reada2 = btrfs_reada_add(csum_root, &key_start, &key_end); |
3189 | ||
3190 | if (!IS_ERR(reada1)) | |
3191 | btrfs_reada_wait(reada1); | |
3192 | if (!IS_ERR(reada2)) | |
3193 | btrfs_reada_wait(reada2); | |
3194 | ||
a2de733c AJ |
3195 | |
3196 | /* | |
3197 | * collect all data csums for the stripe to avoid seeking during | |
3198 | * the scrub. This might currently (crc32) end up to be about 1MB | |
3199 | */ | |
e7786c3a | 3200 | blk_start_plug(&plug); |
a2de733c | 3201 | |
a2de733c AJ |
3202 | /* |
3203 | * now find all extents for each stripe and scrub them | |
3204 | */ | |
a2de733c | 3205 | ret = 0; |
3b080b25 | 3206 | while (physical < physical_end) { |
a2de733c AJ |
3207 | /* |
3208 | * canceled? | |
3209 | */ | |
3210 | if (atomic_read(&fs_info->scrub_cancel_req) || | |
d9d181c1 | 3211 | atomic_read(&sctx->cancel_req)) { |
a2de733c AJ |
3212 | ret = -ECANCELED; |
3213 | goto out; | |
3214 | } | |
3215 | /* | |
3216 | * check to see if we have to pause | |
3217 | */ | |
3218 | if (atomic_read(&fs_info->scrub_pause_req)) { | |
3219 | /* push queued extents */ | |
ff023aac | 3220 | atomic_set(&sctx->wr_ctx.flush_all_writes, 1); |
d9d181c1 | 3221 | scrub_submit(sctx); |
ff023aac SB |
3222 | mutex_lock(&sctx->wr_ctx.wr_lock); |
3223 | scrub_wr_submit(sctx); | |
3224 | mutex_unlock(&sctx->wr_ctx.wr_lock); | |
d9d181c1 | 3225 | wait_event(sctx->list_wait, |
b6bfebc1 | 3226 | atomic_read(&sctx->bios_in_flight) == 0); |
ff023aac | 3227 | atomic_set(&sctx->wr_ctx.flush_all_writes, 0); |
3cb0929a | 3228 | scrub_blocked_if_needed(fs_info); |
a2de733c AJ |
3229 | } |
3230 | ||
f2f66a2f ZL |
3231 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
3232 | ret = get_raid56_logic_offset(physical, num, map, | |
3233 | &logical, | |
3234 | &stripe_logical); | |
3235 | logical += base; | |
3236 | if (ret) { | |
7955323b | 3237 | /* it is parity strip */ |
f2f66a2f | 3238 | stripe_logical += base; |
a0dd59de | 3239 | stripe_end = stripe_logical + increment; |
f2f66a2f ZL |
3240 | ret = scrub_raid56_parity(sctx, map, scrub_dev, |
3241 | ppath, stripe_logical, | |
3242 | stripe_end); | |
3243 | if (ret) | |
3244 | goto out; | |
3245 | goto skip; | |
3246 | } | |
3247 | } | |
3248 | ||
7c76edb7 WS |
3249 | if (btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
3250 | key.type = BTRFS_METADATA_ITEM_KEY; | |
3251 | else | |
3252 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
a2de733c | 3253 | key.objectid = logical; |
625f1c8d | 3254 | key.offset = (u64)-1; |
a2de733c AJ |
3255 | |
3256 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
3257 | if (ret < 0) | |
3258 | goto out; | |
3173a18f | 3259 | |
8c51032f | 3260 | if (ret > 0) { |
ade2e0b3 | 3261 | ret = btrfs_previous_extent_item(root, path, 0); |
a2de733c AJ |
3262 | if (ret < 0) |
3263 | goto out; | |
8c51032f AJ |
3264 | if (ret > 0) { |
3265 | /* there's no smaller item, so stick with the | |
3266 | * larger one */ | |
3267 | btrfs_release_path(path); | |
3268 | ret = btrfs_search_slot(NULL, root, &key, | |
3269 | path, 0, 0); | |
3270 | if (ret < 0) | |
3271 | goto out; | |
3272 | } | |
a2de733c AJ |
3273 | } |
3274 | ||
625f1c8d | 3275 | stop_loop = 0; |
a2de733c | 3276 | while (1) { |
3173a18f JB |
3277 | u64 bytes; |
3278 | ||
a2de733c AJ |
3279 | l = path->nodes[0]; |
3280 | slot = path->slots[0]; | |
3281 | if (slot >= btrfs_header_nritems(l)) { | |
3282 | ret = btrfs_next_leaf(root, path); | |
3283 | if (ret == 0) | |
3284 | continue; | |
3285 | if (ret < 0) | |
3286 | goto out; | |
3287 | ||
625f1c8d | 3288 | stop_loop = 1; |
a2de733c AJ |
3289 | break; |
3290 | } | |
3291 | btrfs_item_key_to_cpu(l, &key, slot); | |
3292 | ||
d7cad238 ZL |
3293 | if (key.type != BTRFS_EXTENT_ITEM_KEY && |
3294 | key.type != BTRFS_METADATA_ITEM_KEY) | |
3295 | goto next; | |
3296 | ||
3173a18f | 3297 | if (key.type == BTRFS_METADATA_ITEM_KEY) |
707e8a07 | 3298 | bytes = root->nodesize; |
3173a18f JB |
3299 | else |
3300 | bytes = key.offset; | |
3301 | ||
3302 | if (key.objectid + bytes <= logical) | |
a2de733c AJ |
3303 | goto next; |
3304 | ||
625f1c8d LB |
3305 | if (key.objectid >= logical + map->stripe_len) { |
3306 | /* out of this device extent */ | |
3307 | if (key.objectid >= logic_end) | |
3308 | stop_loop = 1; | |
3309 | break; | |
3310 | } | |
a2de733c AJ |
3311 | |
3312 | extent = btrfs_item_ptr(l, slot, | |
3313 | struct btrfs_extent_item); | |
3314 | flags = btrfs_extent_flags(l, extent); | |
3315 | generation = btrfs_extent_generation(l, extent); | |
3316 | ||
a323e813 ZL |
3317 | if ((flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) && |
3318 | (key.objectid < logical || | |
3319 | key.objectid + bytes > | |
3320 | logical + map->stripe_len)) { | |
efe120a0 FH |
3321 | btrfs_err(fs_info, |
3322 | "scrub: tree block %llu spanning " | |
3323 | "stripes, ignored. logical=%llu", | |
c1c9ff7c | 3324 | key.objectid, logical); |
9799d2c3 ZL |
3325 | spin_lock(&sctx->stat_lock); |
3326 | sctx->stat.uncorrectable_errors++; | |
3327 | spin_unlock(&sctx->stat_lock); | |
a2de733c AJ |
3328 | goto next; |
3329 | } | |
3330 | ||
625f1c8d LB |
3331 | again: |
3332 | extent_logical = key.objectid; | |
3333 | extent_len = bytes; | |
3334 | ||
a2de733c AJ |
3335 | /* |
3336 | * trim extent to this stripe | |
3337 | */ | |
625f1c8d LB |
3338 | if (extent_logical < logical) { |
3339 | extent_len -= logical - extent_logical; | |
3340 | extent_logical = logical; | |
a2de733c | 3341 | } |
625f1c8d | 3342 | if (extent_logical + extent_len > |
a2de733c | 3343 | logical + map->stripe_len) { |
625f1c8d LB |
3344 | extent_len = logical + map->stripe_len - |
3345 | extent_logical; | |
a2de733c AJ |
3346 | } |
3347 | ||
625f1c8d | 3348 | extent_physical = extent_logical - logical + physical; |
ff023aac SB |
3349 | extent_dev = scrub_dev; |
3350 | extent_mirror_num = mirror_num; | |
3351 | if (is_dev_replace) | |
3352 | scrub_remap_extent(fs_info, extent_logical, | |
3353 | extent_len, &extent_physical, | |
3354 | &extent_dev, | |
3355 | &extent_mirror_num); | |
625f1c8d | 3356 | |
fe8cf654 ZL |
3357 | ret = btrfs_lookup_csums_range(csum_root, |
3358 | extent_logical, | |
3359 | extent_logical + | |
3360 | extent_len - 1, | |
3361 | &sctx->csum_list, 1); | |
625f1c8d LB |
3362 | if (ret) |
3363 | goto out; | |
3364 | ||
ff023aac SB |
3365 | ret = scrub_extent(sctx, extent_logical, extent_len, |
3366 | extent_physical, extent_dev, flags, | |
3367 | generation, extent_mirror_num, | |
115930cb | 3368 | extent_logical - logical + physical); |
6fa96d72 ZL |
3369 | |
3370 | scrub_free_csums(sctx); | |
3371 | ||
a2de733c AJ |
3372 | if (ret) |
3373 | goto out; | |
3374 | ||
625f1c8d LB |
3375 | if (extent_logical + extent_len < |
3376 | key.objectid + bytes) { | |
ffe2d203 | 3377 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
3b080b25 WS |
3378 | /* |
3379 | * loop until we find next data stripe | |
3380 | * or we have finished all stripes. | |
3381 | */ | |
5a6ac9ea MX |
3382 | loop: |
3383 | physical += map->stripe_len; | |
3384 | ret = get_raid56_logic_offset(physical, | |
3385 | num, map, &logical, | |
3386 | &stripe_logical); | |
3387 | logical += base; | |
3388 | ||
3389 | if (ret && physical < physical_end) { | |
3390 | stripe_logical += base; | |
3391 | stripe_end = stripe_logical + | |
a0dd59de | 3392 | increment; |
5a6ac9ea MX |
3393 | ret = scrub_raid56_parity(sctx, |
3394 | map, scrub_dev, ppath, | |
3395 | stripe_logical, | |
3396 | stripe_end); | |
3397 | if (ret) | |
3398 | goto out; | |
3399 | goto loop; | |
3400 | } | |
3b080b25 WS |
3401 | } else { |
3402 | physical += map->stripe_len; | |
3403 | logical += increment; | |
3404 | } | |
625f1c8d LB |
3405 | if (logical < key.objectid + bytes) { |
3406 | cond_resched(); | |
3407 | goto again; | |
3408 | } | |
3409 | ||
3b080b25 | 3410 | if (physical >= physical_end) { |
625f1c8d LB |
3411 | stop_loop = 1; |
3412 | break; | |
3413 | } | |
3414 | } | |
a2de733c AJ |
3415 | next: |
3416 | path->slots[0]++; | |
3417 | } | |
71267333 | 3418 | btrfs_release_path(path); |
3b080b25 | 3419 | skip: |
a2de733c AJ |
3420 | logical += increment; |
3421 | physical += map->stripe_len; | |
d9d181c1 | 3422 | spin_lock(&sctx->stat_lock); |
625f1c8d LB |
3423 | if (stop_loop) |
3424 | sctx->stat.last_physical = map->stripes[num].physical + | |
3425 | length; | |
3426 | else | |
3427 | sctx->stat.last_physical = physical; | |
d9d181c1 | 3428 | spin_unlock(&sctx->stat_lock); |
625f1c8d LB |
3429 | if (stop_loop) |
3430 | break; | |
a2de733c | 3431 | } |
ff023aac | 3432 | out: |
a2de733c | 3433 | /* push queued extents */ |
d9d181c1 | 3434 | scrub_submit(sctx); |
ff023aac SB |
3435 | mutex_lock(&sctx->wr_ctx.wr_lock); |
3436 | scrub_wr_submit(sctx); | |
3437 | mutex_unlock(&sctx->wr_ctx.wr_lock); | |
a2de733c | 3438 | |
e7786c3a | 3439 | blk_finish_plug(&plug); |
a2de733c | 3440 | btrfs_free_path(path); |
5a6ac9ea | 3441 | btrfs_free_path(ppath); |
a2de733c AJ |
3442 | return ret < 0 ? ret : 0; |
3443 | } | |
3444 | ||
d9d181c1 | 3445 | static noinline_for_stack int scrub_chunk(struct scrub_ctx *sctx, |
a36cf8b8 | 3446 | struct btrfs_device *scrub_dev, |
a36cf8b8 | 3447 | u64 chunk_offset, u64 length, |
ff023aac | 3448 | u64 dev_offset, int is_dev_replace) |
a2de733c AJ |
3449 | { |
3450 | struct btrfs_mapping_tree *map_tree = | |
a36cf8b8 | 3451 | &sctx->dev_root->fs_info->mapping_tree; |
a2de733c AJ |
3452 | struct map_lookup *map; |
3453 | struct extent_map *em; | |
3454 | int i; | |
ff023aac | 3455 | int ret = 0; |
a2de733c AJ |
3456 | |
3457 | read_lock(&map_tree->map_tree.lock); | |
3458 | em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1); | |
3459 | read_unlock(&map_tree->map_tree.lock); | |
3460 | ||
3461 | if (!em) | |
3462 | return -EINVAL; | |
3463 | ||
3464 | map = (struct map_lookup *)em->bdev; | |
3465 | if (em->start != chunk_offset) | |
3466 | goto out; | |
3467 | ||
3468 | if (em->len < length) | |
3469 | goto out; | |
3470 | ||
3471 | for (i = 0; i < map->num_stripes; ++i) { | |
a36cf8b8 | 3472 | if (map->stripes[i].dev->bdev == scrub_dev->bdev && |
859acaf1 | 3473 | map->stripes[i].physical == dev_offset) { |
a36cf8b8 | 3474 | ret = scrub_stripe(sctx, map, scrub_dev, i, |
ff023aac SB |
3475 | chunk_offset, length, |
3476 | is_dev_replace); | |
a2de733c AJ |
3477 | if (ret) |
3478 | goto out; | |
3479 | } | |
3480 | } | |
3481 | out: | |
3482 | free_extent_map(em); | |
3483 | ||
3484 | return ret; | |
3485 | } | |
3486 | ||
3487 | static noinline_for_stack | |
a36cf8b8 | 3488 | int scrub_enumerate_chunks(struct scrub_ctx *sctx, |
ff023aac SB |
3489 | struct btrfs_device *scrub_dev, u64 start, u64 end, |
3490 | int is_dev_replace) | |
a2de733c AJ |
3491 | { |
3492 | struct btrfs_dev_extent *dev_extent = NULL; | |
3493 | struct btrfs_path *path; | |
a36cf8b8 | 3494 | struct btrfs_root *root = sctx->dev_root; |
a2de733c AJ |
3495 | struct btrfs_fs_info *fs_info = root->fs_info; |
3496 | u64 length; | |
a2de733c | 3497 | u64 chunk_offset; |
55e3a601 | 3498 | int ret = 0; |
a2de733c AJ |
3499 | int slot; |
3500 | struct extent_buffer *l; | |
3501 | struct btrfs_key key; | |
3502 | struct btrfs_key found_key; | |
3503 | struct btrfs_block_group_cache *cache; | |
ff023aac | 3504 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
a2de733c AJ |
3505 | |
3506 | path = btrfs_alloc_path(); | |
3507 | if (!path) | |
3508 | return -ENOMEM; | |
3509 | ||
3510 | path->reada = 2; | |
3511 | path->search_commit_root = 1; | |
3512 | path->skip_locking = 1; | |
3513 | ||
a36cf8b8 | 3514 | key.objectid = scrub_dev->devid; |
a2de733c AJ |
3515 | key.offset = 0ull; |
3516 | key.type = BTRFS_DEV_EXTENT_KEY; | |
3517 | ||
a2de733c AJ |
3518 | while (1) { |
3519 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
3520 | if (ret < 0) | |
8c51032f AJ |
3521 | break; |
3522 | if (ret > 0) { | |
3523 | if (path->slots[0] >= | |
3524 | btrfs_header_nritems(path->nodes[0])) { | |
3525 | ret = btrfs_next_leaf(root, path); | |
55e3a601 Z |
3526 | if (ret < 0) |
3527 | break; | |
3528 | if (ret > 0) { | |
3529 | ret = 0; | |
8c51032f | 3530 | break; |
55e3a601 Z |
3531 | } |
3532 | } else { | |
3533 | ret = 0; | |
8c51032f AJ |
3534 | } |
3535 | } | |
a2de733c AJ |
3536 | |
3537 | l = path->nodes[0]; | |
3538 | slot = path->slots[0]; | |
3539 | ||
3540 | btrfs_item_key_to_cpu(l, &found_key, slot); | |
3541 | ||
a36cf8b8 | 3542 | if (found_key.objectid != scrub_dev->devid) |
a2de733c AJ |
3543 | break; |
3544 | ||
962a298f | 3545 | if (found_key.type != BTRFS_DEV_EXTENT_KEY) |
a2de733c AJ |
3546 | break; |
3547 | ||
3548 | if (found_key.offset >= end) | |
3549 | break; | |
3550 | ||
3551 | if (found_key.offset < key.offset) | |
3552 | break; | |
3553 | ||
3554 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
3555 | length = btrfs_dev_extent_length(l, dev_extent); | |
3556 | ||
ced96edc QW |
3557 | if (found_key.offset + length <= start) |
3558 | goto skip; | |
a2de733c | 3559 | |
a2de733c AJ |
3560 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); |
3561 | ||
3562 | /* | |
3563 | * get a reference on the corresponding block group to prevent | |
3564 | * the chunk from going away while we scrub it | |
3565 | */ | |
3566 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
ced96edc QW |
3567 | |
3568 | /* some chunks are removed but not committed to disk yet, | |
3569 | * continue scrubbing */ | |
3570 | if (!cache) | |
3571 | goto skip; | |
3572 | ||
55e3a601 Z |
3573 | /* |
3574 | * we need call btrfs_inc_block_group_ro() with scrubs_paused, | |
3575 | * to avoid deadlock caused by: | |
3576 | * btrfs_inc_block_group_ro() | |
3577 | * -> btrfs_wait_for_commit() | |
3578 | * -> btrfs_commit_transaction() | |
3579 | * -> btrfs_scrub_pause() | |
3580 | */ | |
3581 | scrub_pause_on(fs_info); | |
3582 | ret = btrfs_inc_block_group_ro(root, cache); | |
3583 | scrub_pause_off(fs_info); | |
3584 | if (ret) { | |
3585 | btrfs_put_block_group(cache); | |
3586 | break; | |
3587 | } | |
3588 | ||
ff023aac SB |
3589 | dev_replace->cursor_right = found_key.offset + length; |
3590 | dev_replace->cursor_left = found_key.offset; | |
3591 | dev_replace->item_needs_writeback = 1; | |
8c204c96 ZL |
3592 | ret = scrub_chunk(sctx, scrub_dev, chunk_offset, length, |
3593 | found_key.offset, is_dev_replace); | |
ff023aac SB |
3594 | |
3595 | /* | |
3596 | * flush, submit all pending read and write bios, afterwards | |
3597 | * wait for them. | |
3598 | * Note that in the dev replace case, a read request causes | |
3599 | * write requests that are submitted in the read completion | |
3600 | * worker. Therefore in the current situation, it is required | |
3601 | * that all write requests are flushed, so that all read and | |
3602 | * write requests are really completed when bios_in_flight | |
3603 | * changes to 0. | |
3604 | */ | |
3605 | atomic_set(&sctx->wr_ctx.flush_all_writes, 1); | |
3606 | scrub_submit(sctx); | |
3607 | mutex_lock(&sctx->wr_ctx.wr_lock); | |
3608 | scrub_wr_submit(sctx); | |
3609 | mutex_unlock(&sctx->wr_ctx.wr_lock); | |
3610 | ||
3611 | wait_event(sctx->list_wait, | |
3612 | atomic_read(&sctx->bios_in_flight) == 0); | |
b708ce96 Z |
3613 | |
3614 | scrub_pause_on(fs_info); | |
12cf9372 WS |
3615 | |
3616 | /* | |
3617 | * must be called before we decrease @scrub_paused. | |
3618 | * make sure we don't block transaction commit while | |
3619 | * we are waiting pending workers finished. | |
3620 | */ | |
ff023aac SB |
3621 | wait_event(sctx->list_wait, |
3622 | atomic_read(&sctx->workers_pending) == 0); | |
12cf9372 WS |
3623 | atomic_set(&sctx->wr_ctx.flush_all_writes, 0); |
3624 | ||
b708ce96 | 3625 | scrub_pause_off(fs_info); |
ff023aac | 3626 | |
55e3a601 | 3627 | btrfs_dec_block_group_ro(root, cache); |
ff023aac | 3628 | |
a2de733c AJ |
3629 | btrfs_put_block_group(cache); |
3630 | if (ret) | |
3631 | break; | |
af1be4f8 SB |
3632 | if (is_dev_replace && |
3633 | atomic64_read(&dev_replace->num_write_errors) > 0) { | |
ff023aac SB |
3634 | ret = -EIO; |
3635 | break; | |
3636 | } | |
3637 | if (sctx->stat.malloc_errors > 0) { | |
3638 | ret = -ENOMEM; | |
3639 | break; | |
3640 | } | |
a2de733c | 3641 | |
539f358a ID |
3642 | dev_replace->cursor_left = dev_replace->cursor_right; |
3643 | dev_replace->item_needs_writeback = 1; | |
ced96edc | 3644 | skip: |
a2de733c | 3645 | key.offset = found_key.offset + length; |
71267333 | 3646 | btrfs_release_path(path); |
a2de733c AJ |
3647 | } |
3648 | ||
a2de733c | 3649 | btrfs_free_path(path); |
8c51032f | 3650 | |
55e3a601 | 3651 | return ret; |
a2de733c AJ |
3652 | } |
3653 | ||
a36cf8b8 SB |
3654 | static noinline_for_stack int scrub_supers(struct scrub_ctx *sctx, |
3655 | struct btrfs_device *scrub_dev) | |
a2de733c AJ |
3656 | { |
3657 | int i; | |
3658 | u64 bytenr; | |
3659 | u64 gen; | |
3660 | int ret; | |
a36cf8b8 | 3661 | struct btrfs_root *root = sctx->dev_root; |
a2de733c | 3662 | |
87533c47 | 3663 | if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
79787eaa JM |
3664 | return -EIO; |
3665 | ||
5f546063 MX |
3666 | /* Seed devices of a new filesystem has their own generation. */ |
3667 | if (scrub_dev->fs_devices != root->fs_info->fs_devices) | |
3668 | gen = scrub_dev->generation; | |
3669 | else | |
3670 | gen = root->fs_info->last_trans_committed; | |
a2de733c AJ |
3671 | |
3672 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { | |
3673 | bytenr = btrfs_sb_offset(i); | |
935e5cc9 MX |
3674 | if (bytenr + BTRFS_SUPER_INFO_SIZE > |
3675 | scrub_dev->commit_total_bytes) | |
a2de733c AJ |
3676 | break; |
3677 | ||
d9d181c1 | 3678 | ret = scrub_pages(sctx, bytenr, BTRFS_SUPER_INFO_SIZE, bytenr, |
a36cf8b8 | 3679 | scrub_dev, BTRFS_EXTENT_FLAG_SUPER, gen, i, |
ff023aac | 3680 | NULL, 1, bytenr); |
a2de733c AJ |
3681 | if (ret) |
3682 | return ret; | |
3683 | } | |
b6bfebc1 | 3684 | wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0); |
a2de733c AJ |
3685 | |
3686 | return 0; | |
3687 | } | |
3688 | ||
3689 | /* | |
3690 | * get a reference count on fs_info->scrub_workers. start worker if necessary | |
3691 | */ | |
ff023aac SB |
3692 | static noinline_for_stack int scrub_workers_get(struct btrfs_fs_info *fs_info, |
3693 | int is_dev_replace) | |
a2de733c | 3694 | { |
6f011058 | 3695 | unsigned int flags = WQ_FREEZABLE | WQ_UNBOUND; |
0339ef2f | 3696 | int max_active = fs_info->thread_pool_size; |
a2de733c | 3697 | |
632dd772 | 3698 | if (fs_info->scrub_workers_refcnt == 0) { |
ff023aac | 3699 | if (is_dev_replace) |
0339ef2f QW |
3700 | fs_info->scrub_workers = |
3701 | btrfs_alloc_workqueue("btrfs-scrub", flags, | |
3702 | 1, 4); | |
ff023aac | 3703 | else |
0339ef2f QW |
3704 | fs_info->scrub_workers = |
3705 | btrfs_alloc_workqueue("btrfs-scrub", flags, | |
3706 | max_active, 4); | |
e82afc52 ZL |
3707 | if (!fs_info->scrub_workers) |
3708 | goto fail_scrub_workers; | |
3709 | ||
0339ef2f QW |
3710 | fs_info->scrub_wr_completion_workers = |
3711 | btrfs_alloc_workqueue("btrfs-scrubwrc", flags, | |
3712 | max_active, 2); | |
e82afc52 ZL |
3713 | if (!fs_info->scrub_wr_completion_workers) |
3714 | goto fail_scrub_wr_completion_workers; | |
3715 | ||
0339ef2f QW |
3716 | fs_info->scrub_nocow_workers = |
3717 | btrfs_alloc_workqueue("btrfs-scrubnc", flags, 1, 0); | |
e82afc52 ZL |
3718 | if (!fs_info->scrub_nocow_workers) |
3719 | goto fail_scrub_nocow_workers; | |
20b2e302 ZL |
3720 | fs_info->scrub_parity_workers = |
3721 | btrfs_alloc_workqueue("btrfs-scrubparity", flags, | |
3722 | max_active, 2); | |
e82afc52 ZL |
3723 | if (!fs_info->scrub_parity_workers) |
3724 | goto fail_scrub_parity_workers; | |
632dd772 | 3725 | } |
a2de733c | 3726 | ++fs_info->scrub_workers_refcnt; |
e82afc52 ZL |
3727 | return 0; |
3728 | ||
3729 | fail_scrub_parity_workers: | |
3730 | btrfs_destroy_workqueue(fs_info->scrub_nocow_workers); | |
3731 | fail_scrub_nocow_workers: | |
3732 | btrfs_destroy_workqueue(fs_info->scrub_wr_completion_workers); | |
3733 | fail_scrub_wr_completion_workers: | |
3734 | btrfs_destroy_workqueue(fs_info->scrub_workers); | |
3735 | fail_scrub_workers: | |
3736 | return -ENOMEM; | |
a2de733c AJ |
3737 | } |
3738 | ||
aa1b8cd4 | 3739 | static noinline_for_stack void scrub_workers_put(struct btrfs_fs_info *fs_info) |
a2de733c | 3740 | { |
ff023aac | 3741 | if (--fs_info->scrub_workers_refcnt == 0) { |
0339ef2f QW |
3742 | btrfs_destroy_workqueue(fs_info->scrub_workers); |
3743 | btrfs_destroy_workqueue(fs_info->scrub_wr_completion_workers); | |
3744 | btrfs_destroy_workqueue(fs_info->scrub_nocow_workers); | |
20b2e302 | 3745 | btrfs_destroy_workqueue(fs_info->scrub_parity_workers); |
ff023aac | 3746 | } |
a2de733c | 3747 | WARN_ON(fs_info->scrub_workers_refcnt < 0); |
a2de733c AJ |
3748 | } |
3749 | ||
aa1b8cd4 SB |
3750 | int btrfs_scrub_dev(struct btrfs_fs_info *fs_info, u64 devid, u64 start, |
3751 | u64 end, struct btrfs_scrub_progress *progress, | |
63a212ab | 3752 | int readonly, int is_dev_replace) |
a2de733c | 3753 | { |
d9d181c1 | 3754 | struct scrub_ctx *sctx; |
a2de733c AJ |
3755 | int ret; |
3756 | struct btrfs_device *dev; | |
5d68da3b | 3757 | struct rcu_string *name; |
a2de733c | 3758 | |
aa1b8cd4 | 3759 | if (btrfs_fs_closing(fs_info)) |
a2de733c AJ |
3760 | return -EINVAL; |
3761 | ||
aa1b8cd4 | 3762 | if (fs_info->chunk_root->nodesize > BTRFS_STRIPE_LEN) { |
b5d67f64 SB |
3763 | /* |
3764 | * in this case scrub is unable to calculate the checksum | |
3765 | * the way scrub is implemented. Do not handle this | |
3766 | * situation at all because it won't ever happen. | |
3767 | */ | |
efe120a0 FH |
3768 | btrfs_err(fs_info, |
3769 | "scrub: size assumption nodesize <= BTRFS_STRIPE_LEN (%d <= %d) fails", | |
aa1b8cd4 | 3770 | fs_info->chunk_root->nodesize, BTRFS_STRIPE_LEN); |
b5d67f64 SB |
3771 | return -EINVAL; |
3772 | } | |
3773 | ||
aa1b8cd4 | 3774 | if (fs_info->chunk_root->sectorsize != PAGE_SIZE) { |
b5d67f64 | 3775 | /* not supported for data w/o checksums */ |
efe120a0 FH |
3776 | btrfs_err(fs_info, |
3777 | "scrub: size assumption sectorsize != PAGE_SIZE " | |
3778 | "(%d != %lu) fails", | |
27f9f023 | 3779 | fs_info->chunk_root->sectorsize, PAGE_SIZE); |
a2de733c AJ |
3780 | return -EINVAL; |
3781 | } | |
3782 | ||
7a9e9987 SB |
3783 | if (fs_info->chunk_root->nodesize > |
3784 | PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK || | |
3785 | fs_info->chunk_root->sectorsize > | |
3786 | PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK) { | |
3787 | /* | |
3788 | * would exhaust the array bounds of pagev member in | |
3789 | * struct scrub_block | |
3790 | */ | |
efe120a0 FH |
3791 | btrfs_err(fs_info, "scrub: size assumption nodesize and sectorsize " |
3792 | "<= SCRUB_MAX_PAGES_PER_BLOCK (%d <= %d && %d <= %d) fails", | |
7a9e9987 SB |
3793 | fs_info->chunk_root->nodesize, |
3794 | SCRUB_MAX_PAGES_PER_BLOCK, | |
3795 | fs_info->chunk_root->sectorsize, | |
3796 | SCRUB_MAX_PAGES_PER_BLOCK); | |
3797 | return -EINVAL; | |
3798 | } | |
3799 | ||
a2de733c | 3800 | |
aa1b8cd4 SB |
3801 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
3802 | dev = btrfs_find_device(fs_info, devid, NULL, NULL); | |
63a212ab | 3803 | if (!dev || (dev->missing && !is_dev_replace)) { |
aa1b8cd4 | 3804 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
a2de733c AJ |
3805 | return -ENODEV; |
3806 | } | |
a2de733c | 3807 | |
5d68da3b MX |
3808 | if (!is_dev_replace && !readonly && !dev->writeable) { |
3809 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
3810 | rcu_read_lock(); | |
3811 | name = rcu_dereference(dev->name); | |
3812 | btrfs_err(fs_info, "scrub: device %s is not writable", | |
3813 | name->str); | |
3814 | rcu_read_unlock(); | |
3815 | return -EROFS; | |
3816 | } | |
3817 | ||
3b7a016f | 3818 | mutex_lock(&fs_info->scrub_lock); |
63a212ab | 3819 | if (!dev->in_fs_metadata || dev->is_tgtdev_for_dev_replace) { |
a2de733c | 3820 | mutex_unlock(&fs_info->scrub_lock); |
aa1b8cd4 | 3821 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
aa1b8cd4 | 3822 | return -EIO; |
a2de733c AJ |
3823 | } |
3824 | ||
8dabb742 SB |
3825 | btrfs_dev_replace_lock(&fs_info->dev_replace); |
3826 | if (dev->scrub_device || | |
3827 | (!is_dev_replace && | |
3828 | btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))) { | |
3829 | btrfs_dev_replace_unlock(&fs_info->dev_replace); | |
a2de733c | 3830 | mutex_unlock(&fs_info->scrub_lock); |
aa1b8cd4 | 3831 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
a2de733c AJ |
3832 | return -EINPROGRESS; |
3833 | } | |
8dabb742 | 3834 | btrfs_dev_replace_unlock(&fs_info->dev_replace); |
3b7a016f WS |
3835 | |
3836 | ret = scrub_workers_get(fs_info, is_dev_replace); | |
3837 | if (ret) { | |
3838 | mutex_unlock(&fs_info->scrub_lock); | |
3839 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
3840 | return ret; | |
3841 | } | |
3842 | ||
63a212ab | 3843 | sctx = scrub_setup_ctx(dev, is_dev_replace); |
d9d181c1 | 3844 | if (IS_ERR(sctx)) { |
a2de733c | 3845 | mutex_unlock(&fs_info->scrub_lock); |
aa1b8cd4 SB |
3846 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
3847 | scrub_workers_put(fs_info); | |
d9d181c1 | 3848 | return PTR_ERR(sctx); |
a2de733c | 3849 | } |
d9d181c1 SB |
3850 | sctx->readonly = readonly; |
3851 | dev->scrub_device = sctx; | |
3cb0929a | 3852 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
a2de733c | 3853 | |
3cb0929a WS |
3854 | /* |
3855 | * checking @scrub_pause_req here, we can avoid | |
3856 | * race between committing transaction and scrubbing. | |
3857 | */ | |
cb7ab021 | 3858 | __scrub_blocked_if_needed(fs_info); |
a2de733c AJ |
3859 | atomic_inc(&fs_info->scrubs_running); |
3860 | mutex_unlock(&fs_info->scrub_lock); | |
a2de733c | 3861 | |
ff023aac | 3862 | if (!is_dev_replace) { |
9b011adf WS |
3863 | /* |
3864 | * by holding device list mutex, we can | |
3865 | * kick off writing super in log tree sync. | |
3866 | */ | |
3cb0929a | 3867 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
ff023aac | 3868 | ret = scrub_supers(sctx, dev); |
3cb0929a | 3869 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
ff023aac | 3870 | } |
a2de733c AJ |
3871 | |
3872 | if (!ret) | |
ff023aac SB |
3873 | ret = scrub_enumerate_chunks(sctx, dev, start, end, |
3874 | is_dev_replace); | |
a2de733c | 3875 | |
b6bfebc1 | 3876 | wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0); |
a2de733c AJ |
3877 | atomic_dec(&fs_info->scrubs_running); |
3878 | wake_up(&fs_info->scrub_pause_wait); | |
3879 | ||
b6bfebc1 | 3880 | wait_event(sctx->list_wait, atomic_read(&sctx->workers_pending) == 0); |
0ef8e451 | 3881 | |
a2de733c | 3882 | if (progress) |
d9d181c1 | 3883 | memcpy(progress, &sctx->stat, sizeof(*progress)); |
a2de733c AJ |
3884 | |
3885 | mutex_lock(&fs_info->scrub_lock); | |
3886 | dev->scrub_device = NULL; | |
3b7a016f | 3887 | scrub_workers_put(fs_info); |
a2de733c AJ |
3888 | mutex_unlock(&fs_info->scrub_lock); |
3889 | ||
f55985f4 | 3890 | scrub_put_ctx(sctx); |
a2de733c AJ |
3891 | |
3892 | return ret; | |
3893 | } | |
3894 | ||
143bede5 | 3895 | void btrfs_scrub_pause(struct btrfs_root *root) |
a2de733c AJ |
3896 | { |
3897 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3898 | ||
3899 | mutex_lock(&fs_info->scrub_lock); | |
3900 | atomic_inc(&fs_info->scrub_pause_req); | |
3901 | while (atomic_read(&fs_info->scrubs_paused) != | |
3902 | atomic_read(&fs_info->scrubs_running)) { | |
3903 | mutex_unlock(&fs_info->scrub_lock); | |
3904 | wait_event(fs_info->scrub_pause_wait, | |
3905 | atomic_read(&fs_info->scrubs_paused) == | |
3906 | atomic_read(&fs_info->scrubs_running)); | |
3907 | mutex_lock(&fs_info->scrub_lock); | |
3908 | } | |
3909 | mutex_unlock(&fs_info->scrub_lock); | |
a2de733c AJ |
3910 | } |
3911 | ||
143bede5 | 3912 | void btrfs_scrub_continue(struct btrfs_root *root) |
a2de733c AJ |
3913 | { |
3914 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3915 | ||
3916 | atomic_dec(&fs_info->scrub_pause_req); | |
3917 | wake_up(&fs_info->scrub_pause_wait); | |
a2de733c AJ |
3918 | } |
3919 | ||
aa1b8cd4 | 3920 | int btrfs_scrub_cancel(struct btrfs_fs_info *fs_info) |
a2de733c | 3921 | { |
a2de733c AJ |
3922 | mutex_lock(&fs_info->scrub_lock); |
3923 | if (!atomic_read(&fs_info->scrubs_running)) { | |
3924 | mutex_unlock(&fs_info->scrub_lock); | |
3925 | return -ENOTCONN; | |
3926 | } | |
3927 | ||
3928 | atomic_inc(&fs_info->scrub_cancel_req); | |
3929 | while (atomic_read(&fs_info->scrubs_running)) { | |
3930 | mutex_unlock(&fs_info->scrub_lock); | |
3931 | wait_event(fs_info->scrub_pause_wait, | |
3932 | atomic_read(&fs_info->scrubs_running) == 0); | |
3933 | mutex_lock(&fs_info->scrub_lock); | |
3934 | } | |
3935 | atomic_dec(&fs_info->scrub_cancel_req); | |
3936 | mutex_unlock(&fs_info->scrub_lock); | |
3937 | ||
3938 | return 0; | |
3939 | } | |
3940 | ||
aa1b8cd4 SB |
3941 | int btrfs_scrub_cancel_dev(struct btrfs_fs_info *fs_info, |
3942 | struct btrfs_device *dev) | |
49b25e05 | 3943 | { |
d9d181c1 | 3944 | struct scrub_ctx *sctx; |
a2de733c AJ |
3945 | |
3946 | mutex_lock(&fs_info->scrub_lock); | |
d9d181c1 SB |
3947 | sctx = dev->scrub_device; |
3948 | if (!sctx) { | |
a2de733c AJ |
3949 | mutex_unlock(&fs_info->scrub_lock); |
3950 | return -ENOTCONN; | |
3951 | } | |
d9d181c1 | 3952 | atomic_inc(&sctx->cancel_req); |
a2de733c AJ |
3953 | while (dev->scrub_device) { |
3954 | mutex_unlock(&fs_info->scrub_lock); | |
3955 | wait_event(fs_info->scrub_pause_wait, | |
3956 | dev->scrub_device == NULL); | |
3957 | mutex_lock(&fs_info->scrub_lock); | |
3958 | } | |
3959 | mutex_unlock(&fs_info->scrub_lock); | |
3960 | ||
3961 | return 0; | |
3962 | } | |
1623edeb | 3963 | |
a2de733c AJ |
3964 | int btrfs_scrub_progress(struct btrfs_root *root, u64 devid, |
3965 | struct btrfs_scrub_progress *progress) | |
3966 | { | |
3967 | struct btrfs_device *dev; | |
d9d181c1 | 3968 | struct scrub_ctx *sctx = NULL; |
a2de733c AJ |
3969 | |
3970 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
aa1b8cd4 | 3971 | dev = btrfs_find_device(root->fs_info, devid, NULL, NULL); |
a2de733c | 3972 | if (dev) |
d9d181c1 SB |
3973 | sctx = dev->scrub_device; |
3974 | if (sctx) | |
3975 | memcpy(progress, &sctx->stat, sizeof(*progress)); | |
a2de733c AJ |
3976 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
3977 | ||
d9d181c1 | 3978 | return dev ? (sctx ? 0 : -ENOTCONN) : -ENODEV; |
a2de733c | 3979 | } |
ff023aac SB |
3980 | |
3981 | static void scrub_remap_extent(struct btrfs_fs_info *fs_info, | |
3982 | u64 extent_logical, u64 extent_len, | |
3983 | u64 *extent_physical, | |
3984 | struct btrfs_device **extent_dev, | |
3985 | int *extent_mirror_num) | |
3986 | { | |
3987 | u64 mapped_length; | |
3988 | struct btrfs_bio *bbio = NULL; | |
3989 | int ret; | |
3990 | ||
3991 | mapped_length = extent_len; | |
3992 | ret = btrfs_map_block(fs_info, READ, extent_logical, | |
3993 | &mapped_length, &bbio, 0); | |
3994 | if (ret || !bbio || mapped_length < extent_len || | |
3995 | !bbio->stripes[0].dev->bdev) { | |
6e9606d2 | 3996 | btrfs_put_bbio(bbio); |
ff023aac SB |
3997 | return; |
3998 | } | |
3999 | ||
4000 | *extent_physical = bbio->stripes[0].physical; | |
4001 | *extent_mirror_num = bbio->mirror_num; | |
4002 | *extent_dev = bbio->stripes[0].dev; | |
6e9606d2 | 4003 | btrfs_put_bbio(bbio); |
ff023aac SB |
4004 | } |
4005 | ||
4006 | static int scrub_setup_wr_ctx(struct scrub_ctx *sctx, | |
4007 | struct scrub_wr_ctx *wr_ctx, | |
4008 | struct btrfs_fs_info *fs_info, | |
4009 | struct btrfs_device *dev, | |
4010 | int is_dev_replace) | |
4011 | { | |
4012 | WARN_ON(wr_ctx->wr_curr_bio != NULL); | |
4013 | ||
4014 | mutex_init(&wr_ctx->wr_lock); | |
4015 | wr_ctx->wr_curr_bio = NULL; | |
4016 | if (!is_dev_replace) | |
4017 | return 0; | |
4018 | ||
4019 | WARN_ON(!dev->bdev); | |
b54ffb73 | 4020 | wr_ctx->pages_per_wr_bio = SCRUB_PAGES_PER_WR_BIO; |
ff023aac SB |
4021 | wr_ctx->tgtdev = dev; |
4022 | atomic_set(&wr_ctx->flush_all_writes, 0); | |
4023 | return 0; | |
4024 | } | |
4025 | ||
4026 | static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx) | |
4027 | { | |
4028 | mutex_lock(&wr_ctx->wr_lock); | |
4029 | kfree(wr_ctx->wr_curr_bio); | |
4030 | wr_ctx->wr_curr_bio = NULL; | |
4031 | mutex_unlock(&wr_ctx->wr_lock); | |
4032 | } | |
4033 | ||
4034 | static int copy_nocow_pages(struct scrub_ctx *sctx, u64 logical, u64 len, | |
4035 | int mirror_num, u64 physical_for_dev_replace) | |
4036 | { | |
4037 | struct scrub_copy_nocow_ctx *nocow_ctx; | |
4038 | struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info; | |
4039 | ||
4040 | nocow_ctx = kzalloc(sizeof(*nocow_ctx), GFP_NOFS); | |
4041 | if (!nocow_ctx) { | |
4042 | spin_lock(&sctx->stat_lock); | |
4043 | sctx->stat.malloc_errors++; | |
4044 | spin_unlock(&sctx->stat_lock); | |
4045 | return -ENOMEM; | |
4046 | } | |
4047 | ||
4048 | scrub_pending_trans_workers_inc(sctx); | |
4049 | ||
4050 | nocow_ctx->sctx = sctx; | |
4051 | nocow_ctx->logical = logical; | |
4052 | nocow_ctx->len = len; | |
4053 | nocow_ctx->mirror_num = mirror_num; | |
4054 | nocow_ctx->physical_for_dev_replace = physical_for_dev_replace; | |
9e0af237 LB |
4055 | btrfs_init_work(&nocow_ctx->work, btrfs_scrubnc_helper, |
4056 | copy_nocow_pages_worker, NULL, NULL); | |
652f25a2 | 4057 | INIT_LIST_HEAD(&nocow_ctx->inodes); |
0339ef2f QW |
4058 | btrfs_queue_work(fs_info->scrub_nocow_workers, |
4059 | &nocow_ctx->work); | |
ff023aac SB |
4060 | |
4061 | return 0; | |
4062 | } | |
4063 | ||
652f25a2 JB |
4064 | static int record_inode_for_nocow(u64 inum, u64 offset, u64 root, void *ctx) |
4065 | { | |
4066 | struct scrub_copy_nocow_ctx *nocow_ctx = ctx; | |
4067 | struct scrub_nocow_inode *nocow_inode; | |
4068 | ||
4069 | nocow_inode = kzalloc(sizeof(*nocow_inode), GFP_NOFS); | |
4070 | if (!nocow_inode) | |
4071 | return -ENOMEM; | |
4072 | nocow_inode->inum = inum; | |
4073 | nocow_inode->offset = offset; | |
4074 | nocow_inode->root = root; | |
4075 | list_add_tail(&nocow_inode->list, &nocow_ctx->inodes); | |
4076 | return 0; | |
4077 | } | |
4078 | ||
4079 | #define COPY_COMPLETE 1 | |
4080 | ||
ff023aac SB |
4081 | static void copy_nocow_pages_worker(struct btrfs_work *work) |
4082 | { | |
4083 | struct scrub_copy_nocow_ctx *nocow_ctx = | |
4084 | container_of(work, struct scrub_copy_nocow_ctx, work); | |
4085 | struct scrub_ctx *sctx = nocow_ctx->sctx; | |
4086 | u64 logical = nocow_ctx->logical; | |
4087 | u64 len = nocow_ctx->len; | |
4088 | int mirror_num = nocow_ctx->mirror_num; | |
4089 | u64 physical_for_dev_replace = nocow_ctx->physical_for_dev_replace; | |
4090 | int ret; | |
4091 | struct btrfs_trans_handle *trans = NULL; | |
4092 | struct btrfs_fs_info *fs_info; | |
4093 | struct btrfs_path *path; | |
4094 | struct btrfs_root *root; | |
4095 | int not_written = 0; | |
4096 | ||
4097 | fs_info = sctx->dev_root->fs_info; | |
4098 | root = fs_info->extent_root; | |
4099 | ||
4100 | path = btrfs_alloc_path(); | |
4101 | if (!path) { | |
4102 | spin_lock(&sctx->stat_lock); | |
4103 | sctx->stat.malloc_errors++; | |
4104 | spin_unlock(&sctx->stat_lock); | |
4105 | not_written = 1; | |
4106 | goto out; | |
4107 | } | |
4108 | ||
4109 | trans = btrfs_join_transaction(root); | |
4110 | if (IS_ERR(trans)) { | |
4111 | not_written = 1; | |
4112 | goto out; | |
4113 | } | |
4114 | ||
4115 | ret = iterate_inodes_from_logical(logical, fs_info, path, | |
652f25a2 | 4116 | record_inode_for_nocow, nocow_ctx); |
ff023aac | 4117 | if (ret != 0 && ret != -ENOENT) { |
efe120a0 FH |
4118 | btrfs_warn(fs_info, "iterate_inodes_from_logical() failed: log %llu, " |
4119 | "phys %llu, len %llu, mir %u, ret %d", | |
118a0a25 GU |
4120 | logical, physical_for_dev_replace, len, mirror_num, |
4121 | ret); | |
ff023aac SB |
4122 | not_written = 1; |
4123 | goto out; | |
4124 | } | |
4125 | ||
652f25a2 JB |
4126 | btrfs_end_transaction(trans, root); |
4127 | trans = NULL; | |
4128 | while (!list_empty(&nocow_ctx->inodes)) { | |
4129 | struct scrub_nocow_inode *entry; | |
4130 | entry = list_first_entry(&nocow_ctx->inodes, | |
4131 | struct scrub_nocow_inode, | |
4132 | list); | |
4133 | list_del_init(&entry->list); | |
4134 | ret = copy_nocow_pages_for_inode(entry->inum, entry->offset, | |
4135 | entry->root, nocow_ctx); | |
4136 | kfree(entry); | |
4137 | if (ret == COPY_COMPLETE) { | |
4138 | ret = 0; | |
4139 | break; | |
4140 | } else if (ret) { | |
4141 | break; | |
4142 | } | |
4143 | } | |
ff023aac | 4144 | out: |
652f25a2 JB |
4145 | while (!list_empty(&nocow_ctx->inodes)) { |
4146 | struct scrub_nocow_inode *entry; | |
4147 | entry = list_first_entry(&nocow_ctx->inodes, | |
4148 | struct scrub_nocow_inode, | |
4149 | list); | |
4150 | list_del_init(&entry->list); | |
4151 | kfree(entry); | |
4152 | } | |
ff023aac SB |
4153 | if (trans && !IS_ERR(trans)) |
4154 | btrfs_end_transaction(trans, root); | |
4155 | if (not_written) | |
4156 | btrfs_dev_replace_stats_inc(&fs_info->dev_replace. | |
4157 | num_uncorrectable_read_errors); | |
4158 | ||
4159 | btrfs_free_path(path); | |
4160 | kfree(nocow_ctx); | |
4161 | ||
4162 | scrub_pending_trans_workers_dec(sctx); | |
4163 | } | |
4164 | ||
32159242 GH |
4165 | static int check_extent_to_block(struct inode *inode, u64 start, u64 len, |
4166 | u64 logical) | |
4167 | { | |
4168 | struct extent_state *cached_state = NULL; | |
4169 | struct btrfs_ordered_extent *ordered; | |
4170 | struct extent_io_tree *io_tree; | |
4171 | struct extent_map *em; | |
4172 | u64 lockstart = start, lockend = start + len - 1; | |
4173 | int ret = 0; | |
4174 | ||
4175 | io_tree = &BTRFS_I(inode)->io_tree; | |
4176 | ||
4177 | lock_extent_bits(io_tree, lockstart, lockend, 0, &cached_state); | |
4178 | ordered = btrfs_lookup_ordered_range(inode, lockstart, len); | |
4179 | if (ordered) { | |
4180 | btrfs_put_ordered_extent(ordered); | |
4181 | ret = 1; | |
4182 | goto out_unlock; | |
4183 | } | |
4184 | ||
4185 | em = btrfs_get_extent(inode, NULL, 0, start, len, 0); | |
4186 | if (IS_ERR(em)) { | |
4187 | ret = PTR_ERR(em); | |
4188 | goto out_unlock; | |
4189 | } | |
4190 | ||
4191 | /* | |
4192 | * This extent does not actually cover the logical extent anymore, | |
4193 | * move on to the next inode. | |
4194 | */ | |
4195 | if (em->block_start > logical || | |
4196 | em->block_start + em->block_len < logical + len) { | |
4197 | free_extent_map(em); | |
4198 | ret = 1; | |
4199 | goto out_unlock; | |
4200 | } | |
4201 | free_extent_map(em); | |
4202 | ||
4203 | out_unlock: | |
4204 | unlock_extent_cached(io_tree, lockstart, lockend, &cached_state, | |
4205 | GFP_NOFS); | |
4206 | return ret; | |
4207 | } | |
4208 | ||
652f25a2 JB |
4209 | static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root, |
4210 | struct scrub_copy_nocow_ctx *nocow_ctx) | |
ff023aac | 4211 | { |
826aa0a8 | 4212 | struct btrfs_fs_info *fs_info = nocow_ctx->sctx->dev_root->fs_info; |
ff023aac | 4213 | struct btrfs_key key; |
826aa0a8 MX |
4214 | struct inode *inode; |
4215 | struct page *page; | |
ff023aac | 4216 | struct btrfs_root *local_root; |
652f25a2 | 4217 | struct extent_io_tree *io_tree; |
ff023aac | 4218 | u64 physical_for_dev_replace; |
32159242 | 4219 | u64 nocow_ctx_logical; |
652f25a2 | 4220 | u64 len = nocow_ctx->len; |
826aa0a8 | 4221 | unsigned long index; |
6f1c3605 | 4222 | int srcu_index; |
652f25a2 JB |
4223 | int ret = 0; |
4224 | int err = 0; | |
ff023aac SB |
4225 | |
4226 | key.objectid = root; | |
4227 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4228 | key.offset = (u64)-1; | |
6f1c3605 LB |
4229 | |
4230 | srcu_index = srcu_read_lock(&fs_info->subvol_srcu); | |
4231 | ||
ff023aac | 4232 | local_root = btrfs_read_fs_root_no_name(fs_info, &key); |
6f1c3605 LB |
4233 | if (IS_ERR(local_root)) { |
4234 | srcu_read_unlock(&fs_info->subvol_srcu, srcu_index); | |
ff023aac | 4235 | return PTR_ERR(local_root); |
6f1c3605 | 4236 | } |
ff023aac SB |
4237 | |
4238 | key.type = BTRFS_INODE_ITEM_KEY; | |
4239 | key.objectid = inum; | |
4240 | key.offset = 0; | |
4241 | inode = btrfs_iget(fs_info->sb, &key, local_root, NULL); | |
6f1c3605 | 4242 | srcu_read_unlock(&fs_info->subvol_srcu, srcu_index); |
ff023aac SB |
4243 | if (IS_ERR(inode)) |
4244 | return PTR_ERR(inode); | |
4245 | ||
edd1400b MX |
4246 | /* Avoid truncate/dio/punch hole.. */ |
4247 | mutex_lock(&inode->i_mutex); | |
4248 | inode_dio_wait(inode); | |
4249 | ||
ff023aac | 4250 | physical_for_dev_replace = nocow_ctx->physical_for_dev_replace; |
652f25a2 | 4251 | io_tree = &BTRFS_I(inode)->io_tree; |
32159242 | 4252 | nocow_ctx_logical = nocow_ctx->logical; |
652f25a2 | 4253 | |
32159242 GH |
4254 | ret = check_extent_to_block(inode, offset, len, nocow_ctx_logical); |
4255 | if (ret) { | |
4256 | ret = ret > 0 ? 0 : ret; | |
4257 | goto out; | |
652f25a2 | 4258 | } |
652f25a2 | 4259 | |
ff023aac | 4260 | while (len >= PAGE_CACHE_SIZE) { |
ff023aac | 4261 | index = offset >> PAGE_CACHE_SHIFT; |
edd1400b | 4262 | again: |
ff023aac SB |
4263 | page = find_or_create_page(inode->i_mapping, index, GFP_NOFS); |
4264 | if (!page) { | |
efe120a0 | 4265 | btrfs_err(fs_info, "find_or_create_page() failed"); |
ff023aac | 4266 | ret = -ENOMEM; |
826aa0a8 | 4267 | goto out; |
ff023aac SB |
4268 | } |
4269 | ||
4270 | if (PageUptodate(page)) { | |
4271 | if (PageDirty(page)) | |
4272 | goto next_page; | |
4273 | } else { | |
4274 | ClearPageError(page); | |
32159242 | 4275 | err = extent_read_full_page(io_tree, page, |
652f25a2 JB |
4276 | btrfs_get_extent, |
4277 | nocow_ctx->mirror_num); | |
826aa0a8 MX |
4278 | if (err) { |
4279 | ret = err; | |
ff023aac SB |
4280 | goto next_page; |
4281 | } | |
edd1400b | 4282 | |
26b25891 | 4283 | lock_page(page); |
edd1400b MX |
4284 | /* |
4285 | * If the page has been remove from the page cache, | |
4286 | * the data on it is meaningless, because it may be | |
4287 | * old one, the new data may be written into the new | |
4288 | * page in the page cache. | |
4289 | */ | |
4290 | if (page->mapping != inode->i_mapping) { | |
652f25a2 | 4291 | unlock_page(page); |
edd1400b MX |
4292 | page_cache_release(page); |
4293 | goto again; | |
4294 | } | |
ff023aac SB |
4295 | if (!PageUptodate(page)) { |
4296 | ret = -EIO; | |
4297 | goto next_page; | |
4298 | } | |
4299 | } | |
32159242 GH |
4300 | |
4301 | ret = check_extent_to_block(inode, offset, len, | |
4302 | nocow_ctx_logical); | |
4303 | if (ret) { | |
4304 | ret = ret > 0 ? 0 : ret; | |
4305 | goto next_page; | |
4306 | } | |
4307 | ||
826aa0a8 MX |
4308 | err = write_page_nocow(nocow_ctx->sctx, |
4309 | physical_for_dev_replace, page); | |
4310 | if (err) | |
4311 | ret = err; | |
ff023aac | 4312 | next_page: |
826aa0a8 MX |
4313 | unlock_page(page); |
4314 | page_cache_release(page); | |
4315 | ||
4316 | if (ret) | |
4317 | break; | |
4318 | ||
ff023aac SB |
4319 | offset += PAGE_CACHE_SIZE; |
4320 | physical_for_dev_replace += PAGE_CACHE_SIZE; | |
32159242 | 4321 | nocow_ctx_logical += PAGE_CACHE_SIZE; |
ff023aac SB |
4322 | len -= PAGE_CACHE_SIZE; |
4323 | } | |
652f25a2 | 4324 | ret = COPY_COMPLETE; |
826aa0a8 | 4325 | out: |
edd1400b | 4326 | mutex_unlock(&inode->i_mutex); |
826aa0a8 | 4327 | iput(inode); |
ff023aac SB |
4328 | return ret; |
4329 | } | |
4330 | ||
4331 | static int write_page_nocow(struct scrub_ctx *sctx, | |
4332 | u64 physical_for_dev_replace, struct page *page) | |
4333 | { | |
4334 | struct bio *bio; | |
4335 | struct btrfs_device *dev; | |
4336 | int ret; | |
ff023aac SB |
4337 | |
4338 | dev = sctx->wr_ctx.tgtdev; | |
4339 | if (!dev) | |
4340 | return -EIO; | |
4341 | if (!dev->bdev) { | |
94647322 DS |
4342 | btrfs_warn_rl(dev->dev_root->fs_info, |
4343 | "scrub write_page_nocow(bdev == NULL) is unexpected"); | |
ff023aac SB |
4344 | return -EIO; |
4345 | } | |
9be3395b | 4346 | bio = btrfs_io_bio_alloc(GFP_NOFS, 1); |
ff023aac SB |
4347 | if (!bio) { |
4348 | spin_lock(&sctx->stat_lock); | |
4349 | sctx->stat.malloc_errors++; | |
4350 | spin_unlock(&sctx->stat_lock); | |
4351 | return -ENOMEM; | |
4352 | } | |
4f024f37 KO |
4353 | bio->bi_iter.bi_size = 0; |
4354 | bio->bi_iter.bi_sector = physical_for_dev_replace >> 9; | |
ff023aac SB |
4355 | bio->bi_bdev = dev->bdev; |
4356 | ret = bio_add_page(bio, page, PAGE_CACHE_SIZE, 0); | |
4357 | if (ret != PAGE_CACHE_SIZE) { | |
4358 | leave_with_eio: | |
4359 | bio_put(bio); | |
4360 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS); | |
4361 | return -EIO; | |
4362 | } | |
ff023aac | 4363 | |
33879d45 | 4364 | if (btrfsic_submit_bio_wait(WRITE_SYNC, bio)) |
ff023aac SB |
4365 | goto leave_with_eio; |
4366 | ||
4367 | bio_put(bio); | |
4368 | return 0; | |
4369 | } |