]>
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; |
7a9e9987 | 82 | atomic_t ref_count; |
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; | |
115 | atomic_t ref_count; /* 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 SB |
127 | }; |
128 | }; | |
129 | ||
5a6ac9ea MX |
130 | /* Used for the chunks with parity stripe such RAID5/6 */ |
131 | struct scrub_parity { | |
132 | struct scrub_ctx *sctx; | |
133 | ||
134 | struct btrfs_device *scrub_dev; | |
135 | ||
136 | u64 logic_start; | |
137 | ||
138 | u64 logic_end; | |
139 | ||
140 | int nsectors; | |
141 | ||
142 | int stripe_len; | |
143 | ||
144 | atomic_t ref_count; | |
145 | ||
146 | struct list_head spages; | |
147 | ||
148 | /* Work of parity check and repair */ | |
149 | struct btrfs_work work; | |
150 | ||
151 | /* Mark the parity blocks which have data */ | |
152 | unsigned long *dbitmap; | |
153 | ||
154 | /* | |
155 | * Mark the parity blocks which have data, but errors happen when | |
156 | * read data or check data | |
157 | */ | |
158 | unsigned long *ebitmap; | |
159 | ||
160 | unsigned long bitmap[0]; | |
161 | }; | |
162 | ||
ff023aac SB |
163 | struct scrub_wr_ctx { |
164 | struct scrub_bio *wr_curr_bio; | |
165 | struct btrfs_device *tgtdev; | |
166 | int pages_per_wr_bio; /* <= SCRUB_PAGES_PER_WR_BIO */ | |
167 | atomic_t flush_all_writes; | |
168 | struct mutex wr_lock; | |
169 | }; | |
170 | ||
d9d181c1 | 171 | struct scrub_ctx { |
ff023aac | 172 | struct scrub_bio *bios[SCRUB_BIOS_PER_SCTX]; |
a36cf8b8 | 173 | struct btrfs_root *dev_root; |
a2de733c AJ |
174 | int first_free; |
175 | int curr; | |
b6bfebc1 SB |
176 | atomic_t bios_in_flight; |
177 | atomic_t workers_pending; | |
a2de733c AJ |
178 | spinlock_t list_lock; |
179 | wait_queue_head_t list_wait; | |
180 | u16 csum_size; | |
181 | struct list_head csum_list; | |
182 | atomic_t cancel_req; | |
8628764e | 183 | int readonly; |
ff023aac | 184 | int pages_per_rd_bio; |
b5d67f64 SB |
185 | u32 sectorsize; |
186 | u32 nodesize; | |
63a212ab SB |
187 | |
188 | int is_dev_replace; | |
ff023aac | 189 | struct scrub_wr_ctx wr_ctx; |
63a212ab | 190 | |
a2de733c AJ |
191 | /* |
192 | * statistics | |
193 | */ | |
194 | struct btrfs_scrub_progress stat; | |
195 | spinlock_t stat_lock; | |
196 | }; | |
197 | ||
0ef8e451 | 198 | struct scrub_fixup_nodatasum { |
d9d181c1 | 199 | struct scrub_ctx *sctx; |
a36cf8b8 | 200 | struct btrfs_device *dev; |
0ef8e451 JS |
201 | u64 logical; |
202 | struct btrfs_root *root; | |
203 | struct btrfs_work work; | |
204 | int mirror_num; | |
205 | }; | |
206 | ||
652f25a2 JB |
207 | struct scrub_nocow_inode { |
208 | u64 inum; | |
209 | u64 offset; | |
210 | u64 root; | |
211 | struct list_head list; | |
212 | }; | |
213 | ||
ff023aac SB |
214 | struct scrub_copy_nocow_ctx { |
215 | struct scrub_ctx *sctx; | |
216 | u64 logical; | |
217 | u64 len; | |
218 | int mirror_num; | |
219 | u64 physical_for_dev_replace; | |
652f25a2 | 220 | struct list_head inodes; |
ff023aac SB |
221 | struct btrfs_work work; |
222 | }; | |
223 | ||
558540c1 JS |
224 | struct scrub_warning { |
225 | struct btrfs_path *path; | |
226 | u64 extent_item_size; | |
558540c1 JS |
227 | const char *errstr; |
228 | sector_t sector; | |
229 | u64 logical; | |
230 | struct btrfs_device *dev; | |
558540c1 JS |
231 | }; |
232 | ||
b6bfebc1 SB |
233 | static void scrub_pending_bio_inc(struct scrub_ctx *sctx); |
234 | static void scrub_pending_bio_dec(struct scrub_ctx *sctx); | |
235 | static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx); | |
236 | static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx); | |
b5d67f64 | 237 | static int scrub_handle_errored_block(struct scrub_block *sblock_to_check); |
d9d181c1 | 238 | static int scrub_setup_recheck_block(struct scrub_ctx *sctx, |
3ec706c8 | 239 | struct btrfs_fs_info *fs_info, |
ff023aac | 240 | struct scrub_block *original_sblock, |
b5d67f64 | 241 | u64 length, u64 logical, |
ff023aac | 242 | struct scrub_block *sblocks_for_recheck); |
34f5c8e9 SB |
243 | static void scrub_recheck_block(struct btrfs_fs_info *fs_info, |
244 | struct scrub_block *sblock, int is_metadata, | |
245 | int have_csum, u8 *csum, u64 generation, | |
af8e2d1d | 246 | u16 csum_size, int retry_failed_mirror); |
b5d67f64 SB |
247 | static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info, |
248 | struct scrub_block *sblock, | |
249 | int is_metadata, int have_csum, | |
250 | const u8 *csum, u64 generation, | |
251 | u16 csum_size); | |
b5d67f64 | 252 | static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad, |
114ab50d | 253 | struct scrub_block *sblock_good); |
b5d67f64 SB |
254 | static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad, |
255 | struct scrub_block *sblock_good, | |
256 | int page_num, int force_write); | |
ff023aac SB |
257 | static void scrub_write_block_to_dev_replace(struct scrub_block *sblock); |
258 | static int scrub_write_page_to_dev_replace(struct scrub_block *sblock, | |
259 | int page_num); | |
b5d67f64 SB |
260 | static int scrub_checksum_data(struct scrub_block *sblock); |
261 | static int scrub_checksum_tree_block(struct scrub_block *sblock); | |
262 | static int scrub_checksum_super(struct scrub_block *sblock); | |
263 | static void scrub_block_get(struct scrub_block *sblock); | |
264 | static void scrub_block_put(struct scrub_block *sblock); | |
7a9e9987 SB |
265 | static void scrub_page_get(struct scrub_page *spage); |
266 | static void scrub_page_put(struct scrub_page *spage); | |
5a6ac9ea MX |
267 | static void scrub_parity_get(struct scrub_parity *sparity); |
268 | static void scrub_parity_put(struct scrub_parity *sparity); | |
ff023aac SB |
269 | static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx, |
270 | struct scrub_page *spage); | |
d9d181c1 | 271 | static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len, |
a36cf8b8 | 272 | u64 physical, struct btrfs_device *dev, u64 flags, |
ff023aac SB |
273 | u64 gen, int mirror_num, u8 *csum, int force, |
274 | u64 physical_for_dev_replace); | |
1623edeb | 275 | static void scrub_bio_end_io(struct bio *bio, int err); |
b5d67f64 SB |
276 | static void scrub_bio_end_io_worker(struct btrfs_work *work); |
277 | static void scrub_block_complete(struct scrub_block *sblock); | |
ff023aac SB |
278 | static void scrub_remap_extent(struct btrfs_fs_info *fs_info, |
279 | u64 extent_logical, u64 extent_len, | |
280 | u64 *extent_physical, | |
281 | struct btrfs_device **extent_dev, | |
282 | int *extent_mirror_num); | |
283 | static int scrub_setup_wr_ctx(struct scrub_ctx *sctx, | |
284 | struct scrub_wr_ctx *wr_ctx, | |
285 | struct btrfs_fs_info *fs_info, | |
286 | struct btrfs_device *dev, | |
287 | int is_dev_replace); | |
288 | static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx); | |
289 | static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx, | |
290 | struct scrub_page *spage); | |
291 | static void scrub_wr_submit(struct scrub_ctx *sctx); | |
292 | static void scrub_wr_bio_end_io(struct bio *bio, int err); | |
293 | static void scrub_wr_bio_end_io_worker(struct btrfs_work *work); | |
294 | static int write_page_nocow(struct scrub_ctx *sctx, | |
295 | u64 physical_for_dev_replace, struct page *page); | |
296 | static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root, | |
652f25a2 | 297 | struct scrub_copy_nocow_ctx *ctx); |
ff023aac SB |
298 | static int copy_nocow_pages(struct scrub_ctx *sctx, u64 logical, u64 len, |
299 | int mirror_num, u64 physical_for_dev_replace); | |
300 | static void copy_nocow_pages_worker(struct btrfs_work *work); | |
cb7ab021 | 301 | static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info); |
3cb0929a | 302 | static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info); |
1623edeb SB |
303 | |
304 | ||
b6bfebc1 SB |
305 | static void scrub_pending_bio_inc(struct scrub_ctx *sctx) |
306 | { | |
307 | atomic_inc(&sctx->bios_in_flight); | |
308 | } | |
309 | ||
310 | static void scrub_pending_bio_dec(struct scrub_ctx *sctx) | |
311 | { | |
312 | atomic_dec(&sctx->bios_in_flight); | |
313 | wake_up(&sctx->list_wait); | |
314 | } | |
315 | ||
cb7ab021 | 316 | static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info) |
3cb0929a WS |
317 | { |
318 | while (atomic_read(&fs_info->scrub_pause_req)) { | |
319 | mutex_unlock(&fs_info->scrub_lock); | |
320 | wait_event(fs_info->scrub_pause_wait, | |
321 | atomic_read(&fs_info->scrub_pause_req) == 0); | |
322 | mutex_lock(&fs_info->scrub_lock); | |
323 | } | |
324 | } | |
325 | ||
cb7ab021 WS |
326 | static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info) |
327 | { | |
328 | atomic_inc(&fs_info->scrubs_paused); | |
329 | wake_up(&fs_info->scrub_pause_wait); | |
330 | ||
331 | mutex_lock(&fs_info->scrub_lock); | |
332 | __scrub_blocked_if_needed(fs_info); | |
333 | atomic_dec(&fs_info->scrubs_paused); | |
334 | mutex_unlock(&fs_info->scrub_lock); | |
335 | ||
336 | wake_up(&fs_info->scrub_pause_wait); | |
337 | } | |
338 | ||
b6bfebc1 SB |
339 | /* |
340 | * used for workers that require transaction commits (i.e., for the | |
341 | * NOCOW case) | |
342 | */ | |
343 | static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx) | |
344 | { | |
345 | struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info; | |
346 | ||
347 | /* | |
348 | * increment scrubs_running to prevent cancel requests from | |
349 | * completing as long as a worker is running. we must also | |
350 | * increment scrubs_paused to prevent deadlocking on pause | |
351 | * requests used for transactions commits (as the worker uses a | |
352 | * transaction context). it is safe to regard the worker | |
353 | * as paused for all matters practical. effectively, we only | |
354 | * avoid cancellation requests from completing. | |
355 | */ | |
356 | mutex_lock(&fs_info->scrub_lock); | |
357 | atomic_inc(&fs_info->scrubs_running); | |
358 | atomic_inc(&fs_info->scrubs_paused); | |
359 | mutex_unlock(&fs_info->scrub_lock); | |
32a44789 WS |
360 | |
361 | /* | |
362 | * check if @scrubs_running=@scrubs_paused condition | |
363 | * inside wait_event() is not an atomic operation. | |
364 | * which means we may inc/dec @scrub_running/paused | |
365 | * at any time. Let's wake up @scrub_pause_wait as | |
366 | * much as we can to let commit transaction blocked less. | |
367 | */ | |
368 | wake_up(&fs_info->scrub_pause_wait); | |
369 | ||
b6bfebc1 SB |
370 | atomic_inc(&sctx->workers_pending); |
371 | } | |
372 | ||
373 | /* used for workers that require transaction commits */ | |
374 | static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx) | |
375 | { | |
376 | struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info; | |
377 | ||
378 | /* | |
379 | * see scrub_pending_trans_workers_inc() why we're pretending | |
380 | * to be paused in the scrub counters | |
381 | */ | |
382 | mutex_lock(&fs_info->scrub_lock); | |
383 | atomic_dec(&fs_info->scrubs_running); | |
384 | atomic_dec(&fs_info->scrubs_paused); | |
385 | mutex_unlock(&fs_info->scrub_lock); | |
386 | atomic_dec(&sctx->workers_pending); | |
387 | wake_up(&fs_info->scrub_pause_wait); | |
388 | wake_up(&sctx->list_wait); | |
389 | } | |
390 | ||
d9d181c1 | 391 | static void scrub_free_csums(struct scrub_ctx *sctx) |
a2de733c | 392 | { |
d9d181c1 | 393 | while (!list_empty(&sctx->csum_list)) { |
a2de733c | 394 | struct btrfs_ordered_sum *sum; |
d9d181c1 | 395 | sum = list_first_entry(&sctx->csum_list, |
a2de733c AJ |
396 | struct btrfs_ordered_sum, list); |
397 | list_del(&sum->list); | |
398 | kfree(sum); | |
399 | } | |
400 | } | |
401 | ||
d9d181c1 | 402 | static noinline_for_stack void scrub_free_ctx(struct scrub_ctx *sctx) |
a2de733c AJ |
403 | { |
404 | int i; | |
a2de733c | 405 | |
d9d181c1 | 406 | if (!sctx) |
a2de733c AJ |
407 | return; |
408 | ||
ff023aac SB |
409 | scrub_free_wr_ctx(&sctx->wr_ctx); |
410 | ||
b5d67f64 | 411 | /* this can happen when scrub is cancelled */ |
d9d181c1 SB |
412 | if (sctx->curr != -1) { |
413 | struct scrub_bio *sbio = sctx->bios[sctx->curr]; | |
b5d67f64 SB |
414 | |
415 | for (i = 0; i < sbio->page_count; i++) { | |
ff023aac | 416 | WARN_ON(!sbio->pagev[i]->page); |
b5d67f64 SB |
417 | scrub_block_put(sbio->pagev[i]->sblock); |
418 | } | |
419 | bio_put(sbio->bio); | |
420 | } | |
421 | ||
ff023aac | 422 | for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) { |
d9d181c1 | 423 | struct scrub_bio *sbio = sctx->bios[i]; |
a2de733c AJ |
424 | |
425 | if (!sbio) | |
426 | break; | |
a2de733c AJ |
427 | kfree(sbio); |
428 | } | |
429 | ||
d9d181c1 SB |
430 | scrub_free_csums(sctx); |
431 | kfree(sctx); | |
a2de733c AJ |
432 | } |
433 | ||
434 | static noinline_for_stack | |
63a212ab | 435 | struct scrub_ctx *scrub_setup_ctx(struct btrfs_device *dev, int is_dev_replace) |
a2de733c | 436 | { |
d9d181c1 | 437 | struct scrub_ctx *sctx; |
a2de733c | 438 | int i; |
a2de733c | 439 | struct btrfs_fs_info *fs_info = dev->dev_root->fs_info; |
ff023aac SB |
440 | int pages_per_rd_bio; |
441 | int ret; | |
a2de733c | 442 | |
ff023aac SB |
443 | /* |
444 | * the setting of pages_per_rd_bio is correct for scrub but might | |
445 | * be wrong for the dev_replace code where we might read from | |
446 | * different devices in the initial huge bios. However, that | |
447 | * code is able to correctly handle the case when adding a page | |
448 | * to a bio fails. | |
449 | */ | |
450 | if (dev->bdev) | |
451 | pages_per_rd_bio = min_t(int, SCRUB_PAGES_PER_RD_BIO, | |
452 | bio_get_nr_vecs(dev->bdev)); | |
453 | else | |
454 | pages_per_rd_bio = SCRUB_PAGES_PER_RD_BIO; | |
d9d181c1 SB |
455 | sctx = kzalloc(sizeof(*sctx), GFP_NOFS); |
456 | if (!sctx) | |
a2de733c | 457 | goto nomem; |
63a212ab | 458 | sctx->is_dev_replace = is_dev_replace; |
ff023aac | 459 | sctx->pages_per_rd_bio = pages_per_rd_bio; |
d9d181c1 | 460 | sctx->curr = -1; |
a36cf8b8 | 461 | sctx->dev_root = dev->dev_root; |
ff023aac | 462 | for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) { |
a2de733c AJ |
463 | struct scrub_bio *sbio; |
464 | ||
465 | sbio = kzalloc(sizeof(*sbio), GFP_NOFS); | |
466 | if (!sbio) | |
467 | goto nomem; | |
d9d181c1 | 468 | sctx->bios[i] = sbio; |
a2de733c | 469 | |
a2de733c | 470 | sbio->index = i; |
d9d181c1 | 471 | sbio->sctx = sctx; |
b5d67f64 | 472 | sbio->page_count = 0; |
9e0af237 LB |
473 | btrfs_init_work(&sbio->work, btrfs_scrub_helper, |
474 | scrub_bio_end_io_worker, NULL, NULL); | |
a2de733c | 475 | |
ff023aac | 476 | if (i != SCRUB_BIOS_PER_SCTX - 1) |
d9d181c1 | 477 | sctx->bios[i]->next_free = i + 1; |
0ef8e451 | 478 | else |
d9d181c1 SB |
479 | sctx->bios[i]->next_free = -1; |
480 | } | |
481 | sctx->first_free = 0; | |
482 | sctx->nodesize = dev->dev_root->nodesize; | |
d9d181c1 | 483 | sctx->sectorsize = dev->dev_root->sectorsize; |
b6bfebc1 SB |
484 | atomic_set(&sctx->bios_in_flight, 0); |
485 | atomic_set(&sctx->workers_pending, 0); | |
d9d181c1 SB |
486 | atomic_set(&sctx->cancel_req, 0); |
487 | sctx->csum_size = btrfs_super_csum_size(fs_info->super_copy); | |
488 | INIT_LIST_HEAD(&sctx->csum_list); | |
489 | ||
490 | spin_lock_init(&sctx->list_lock); | |
491 | spin_lock_init(&sctx->stat_lock); | |
492 | init_waitqueue_head(&sctx->list_wait); | |
ff023aac SB |
493 | |
494 | ret = scrub_setup_wr_ctx(sctx, &sctx->wr_ctx, fs_info, | |
495 | fs_info->dev_replace.tgtdev, is_dev_replace); | |
496 | if (ret) { | |
497 | scrub_free_ctx(sctx); | |
498 | return ERR_PTR(ret); | |
499 | } | |
d9d181c1 | 500 | return sctx; |
a2de733c AJ |
501 | |
502 | nomem: | |
d9d181c1 | 503 | scrub_free_ctx(sctx); |
a2de733c AJ |
504 | return ERR_PTR(-ENOMEM); |
505 | } | |
506 | ||
ff023aac SB |
507 | static int scrub_print_warning_inode(u64 inum, u64 offset, u64 root, |
508 | void *warn_ctx) | |
558540c1 JS |
509 | { |
510 | u64 isize; | |
511 | u32 nlink; | |
512 | int ret; | |
513 | int i; | |
514 | struct extent_buffer *eb; | |
515 | struct btrfs_inode_item *inode_item; | |
ff023aac | 516 | struct scrub_warning *swarn = warn_ctx; |
558540c1 JS |
517 | struct btrfs_fs_info *fs_info = swarn->dev->dev_root->fs_info; |
518 | struct inode_fs_paths *ipath = NULL; | |
519 | struct btrfs_root *local_root; | |
520 | struct btrfs_key root_key; | |
1d4c08e0 | 521 | struct btrfs_key key; |
558540c1 JS |
522 | |
523 | root_key.objectid = root; | |
524 | root_key.type = BTRFS_ROOT_ITEM_KEY; | |
525 | root_key.offset = (u64)-1; | |
526 | local_root = btrfs_read_fs_root_no_name(fs_info, &root_key); | |
527 | if (IS_ERR(local_root)) { | |
528 | ret = PTR_ERR(local_root); | |
529 | goto err; | |
530 | } | |
531 | ||
14692cc1 DS |
532 | /* |
533 | * this makes the path point to (inum INODE_ITEM ioff) | |
534 | */ | |
1d4c08e0 DS |
535 | key.objectid = inum; |
536 | key.type = BTRFS_INODE_ITEM_KEY; | |
537 | key.offset = 0; | |
538 | ||
539 | ret = btrfs_search_slot(NULL, local_root, &key, swarn->path, 0, 0); | |
558540c1 JS |
540 | if (ret) { |
541 | btrfs_release_path(swarn->path); | |
542 | goto err; | |
543 | } | |
544 | ||
545 | eb = swarn->path->nodes[0]; | |
546 | inode_item = btrfs_item_ptr(eb, swarn->path->slots[0], | |
547 | struct btrfs_inode_item); | |
548 | isize = btrfs_inode_size(eb, inode_item); | |
549 | nlink = btrfs_inode_nlink(eb, inode_item); | |
550 | btrfs_release_path(swarn->path); | |
551 | ||
552 | ipath = init_ipath(4096, local_root, swarn->path); | |
26bdef54 DC |
553 | if (IS_ERR(ipath)) { |
554 | ret = PTR_ERR(ipath); | |
555 | ipath = NULL; | |
556 | goto err; | |
557 | } | |
558540c1 JS |
558 | ret = paths_from_inode(inum, ipath); |
559 | ||
560 | if (ret < 0) | |
561 | goto err; | |
562 | ||
563 | /* | |
564 | * we deliberately ignore the bit ipath might have been too small to | |
565 | * hold all of the paths here | |
566 | */ | |
567 | for (i = 0; i < ipath->fspath->elem_cnt; ++i) | |
efe120a0 | 568 | printk_in_rcu(KERN_WARNING "BTRFS: %s at logical %llu on dev " |
558540c1 JS |
569 | "%s, sector %llu, root %llu, inode %llu, offset %llu, " |
570 | "length %llu, links %u (path: %s)\n", swarn->errstr, | |
606686ee | 571 | swarn->logical, rcu_str_deref(swarn->dev->name), |
558540c1 JS |
572 | (unsigned long long)swarn->sector, root, inum, offset, |
573 | min(isize - offset, (u64)PAGE_SIZE), nlink, | |
745c4d8e | 574 | (char *)(unsigned long)ipath->fspath->val[i]); |
558540c1 JS |
575 | |
576 | free_ipath(ipath); | |
577 | return 0; | |
578 | ||
579 | err: | |
efe120a0 | 580 | printk_in_rcu(KERN_WARNING "BTRFS: %s at logical %llu on dev " |
558540c1 JS |
581 | "%s, sector %llu, root %llu, inode %llu, offset %llu: path " |
582 | "resolving failed with ret=%d\n", swarn->errstr, | |
606686ee | 583 | swarn->logical, rcu_str_deref(swarn->dev->name), |
558540c1 JS |
584 | (unsigned long long)swarn->sector, root, inum, offset, ret); |
585 | ||
586 | free_ipath(ipath); | |
587 | return 0; | |
588 | } | |
589 | ||
b5d67f64 | 590 | static void scrub_print_warning(const char *errstr, struct scrub_block *sblock) |
558540c1 | 591 | { |
a36cf8b8 SB |
592 | struct btrfs_device *dev; |
593 | struct btrfs_fs_info *fs_info; | |
558540c1 JS |
594 | struct btrfs_path *path; |
595 | struct btrfs_key found_key; | |
596 | struct extent_buffer *eb; | |
597 | struct btrfs_extent_item *ei; | |
598 | struct scrub_warning swarn; | |
69917e43 LB |
599 | unsigned long ptr = 0; |
600 | u64 extent_item_pos; | |
601 | u64 flags = 0; | |
558540c1 | 602 | u64 ref_root; |
69917e43 | 603 | u32 item_size; |
558540c1 | 604 | u8 ref_level; |
69917e43 | 605 | int ret; |
558540c1 | 606 | |
a36cf8b8 | 607 | WARN_ON(sblock->page_count < 1); |
7a9e9987 | 608 | dev = sblock->pagev[0]->dev; |
a36cf8b8 SB |
609 | fs_info = sblock->sctx->dev_root->fs_info; |
610 | ||
558540c1 | 611 | path = btrfs_alloc_path(); |
8b9456da DS |
612 | if (!path) |
613 | return; | |
558540c1 | 614 | |
7a9e9987 SB |
615 | swarn.sector = (sblock->pagev[0]->physical) >> 9; |
616 | swarn.logical = sblock->pagev[0]->logical; | |
558540c1 | 617 | swarn.errstr = errstr; |
a36cf8b8 | 618 | swarn.dev = NULL; |
558540c1 | 619 | |
69917e43 LB |
620 | ret = extent_from_logical(fs_info, swarn.logical, path, &found_key, |
621 | &flags); | |
558540c1 JS |
622 | if (ret < 0) |
623 | goto out; | |
624 | ||
4692cf58 | 625 | extent_item_pos = swarn.logical - found_key.objectid; |
558540c1 JS |
626 | swarn.extent_item_size = found_key.offset; |
627 | ||
628 | eb = path->nodes[0]; | |
629 | ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item); | |
630 | item_size = btrfs_item_size_nr(eb, path->slots[0]); | |
631 | ||
69917e43 | 632 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { |
558540c1 | 633 | do { |
6eda71d0 LB |
634 | ret = tree_backref_for_extent(&ptr, eb, &found_key, ei, |
635 | item_size, &ref_root, | |
636 | &ref_level); | |
606686ee | 637 | printk_in_rcu(KERN_WARNING |
efe120a0 | 638 | "BTRFS: %s at logical %llu on dev %s, " |
558540c1 | 639 | "sector %llu: metadata %s (level %d) in tree " |
606686ee JB |
640 | "%llu\n", errstr, swarn.logical, |
641 | rcu_str_deref(dev->name), | |
558540c1 JS |
642 | (unsigned long long)swarn.sector, |
643 | ref_level ? "node" : "leaf", | |
644 | ret < 0 ? -1 : ref_level, | |
645 | ret < 0 ? -1 : ref_root); | |
646 | } while (ret != 1); | |
d8fe29e9 | 647 | btrfs_release_path(path); |
558540c1 | 648 | } else { |
d8fe29e9 | 649 | btrfs_release_path(path); |
558540c1 | 650 | swarn.path = path; |
a36cf8b8 | 651 | swarn.dev = dev; |
7a3ae2f8 JS |
652 | iterate_extent_inodes(fs_info, found_key.objectid, |
653 | extent_item_pos, 1, | |
558540c1 JS |
654 | scrub_print_warning_inode, &swarn); |
655 | } | |
656 | ||
657 | out: | |
658 | btrfs_free_path(path); | |
558540c1 JS |
659 | } |
660 | ||
ff023aac | 661 | static int scrub_fixup_readpage(u64 inum, u64 offset, u64 root, void *fixup_ctx) |
0ef8e451 | 662 | { |
5da6fcbc | 663 | struct page *page = NULL; |
0ef8e451 | 664 | unsigned long index; |
ff023aac | 665 | struct scrub_fixup_nodatasum *fixup = fixup_ctx; |
0ef8e451 | 666 | int ret; |
5da6fcbc | 667 | int corrected = 0; |
0ef8e451 | 668 | struct btrfs_key key; |
5da6fcbc | 669 | struct inode *inode = NULL; |
6f1c3605 | 670 | struct btrfs_fs_info *fs_info; |
0ef8e451 JS |
671 | u64 end = offset + PAGE_SIZE - 1; |
672 | struct btrfs_root *local_root; | |
6f1c3605 | 673 | int srcu_index; |
0ef8e451 JS |
674 | |
675 | key.objectid = root; | |
676 | key.type = BTRFS_ROOT_ITEM_KEY; | |
677 | key.offset = (u64)-1; | |
6f1c3605 LB |
678 | |
679 | fs_info = fixup->root->fs_info; | |
680 | srcu_index = srcu_read_lock(&fs_info->subvol_srcu); | |
681 | ||
682 | local_root = btrfs_read_fs_root_no_name(fs_info, &key); | |
683 | if (IS_ERR(local_root)) { | |
684 | srcu_read_unlock(&fs_info->subvol_srcu, srcu_index); | |
0ef8e451 | 685 | return PTR_ERR(local_root); |
6f1c3605 | 686 | } |
0ef8e451 JS |
687 | |
688 | key.type = BTRFS_INODE_ITEM_KEY; | |
689 | key.objectid = inum; | |
690 | key.offset = 0; | |
6f1c3605 LB |
691 | inode = btrfs_iget(fs_info->sb, &key, local_root, NULL); |
692 | srcu_read_unlock(&fs_info->subvol_srcu, srcu_index); | |
0ef8e451 JS |
693 | if (IS_ERR(inode)) |
694 | return PTR_ERR(inode); | |
695 | ||
0ef8e451 JS |
696 | index = offset >> PAGE_CACHE_SHIFT; |
697 | ||
698 | page = find_or_create_page(inode->i_mapping, index, GFP_NOFS); | |
5da6fcbc JS |
699 | if (!page) { |
700 | ret = -ENOMEM; | |
701 | goto out; | |
702 | } | |
703 | ||
704 | if (PageUptodate(page)) { | |
5da6fcbc JS |
705 | if (PageDirty(page)) { |
706 | /* | |
707 | * we need to write the data to the defect sector. the | |
708 | * data that was in that sector is not in memory, | |
709 | * because the page was modified. we must not write the | |
710 | * modified page to that sector. | |
711 | * | |
712 | * TODO: what could be done here: wait for the delalloc | |
713 | * runner to write out that page (might involve | |
714 | * COW) and see whether the sector is still | |
715 | * referenced afterwards. | |
716 | * | |
717 | * For the meantime, we'll treat this error | |
718 | * incorrectable, although there is a chance that a | |
719 | * later scrub will find the bad sector again and that | |
720 | * there's no dirty page in memory, then. | |
721 | */ | |
722 | ret = -EIO; | |
723 | goto out; | |
724 | } | |
1203b681 | 725 | ret = repair_io_failure(inode, offset, PAGE_SIZE, |
5da6fcbc | 726 | fixup->logical, page, |
ffdd2018 | 727 | offset - page_offset(page), |
5da6fcbc JS |
728 | fixup->mirror_num); |
729 | unlock_page(page); | |
730 | corrected = !ret; | |
731 | } else { | |
732 | /* | |
733 | * we need to get good data first. the general readpage path | |
734 | * will call repair_io_failure for us, we just have to make | |
735 | * sure we read the bad mirror. | |
736 | */ | |
737 | ret = set_extent_bits(&BTRFS_I(inode)->io_tree, offset, end, | |
738 | EXTENT_DAMAGED, GFP_NOFS); | |
739 | if (ret) { | |
740 | /* set_extent_bits should give proper error */ | |
741 | WARN_ON(ret > 0); | |
742 | if (ret > 0) | |
743 | ret = -EFAULT; | |
744 | goto out; | |
745 | } | |
746 | ||
747 | ret = extent_read_full_page(&BTRFS_I(inode)->io_tree, page, | |
748 | btrfs_get_extent, | |
749 | fixup->mirror_num); | |
750 | wait_on_page_locked(page); | |
751 | ||
752 | corrected = !test_range_bit(&BTRFS_I(inode)->io_tree, offset, | |
753 | end, EXTENT_DAMAGED, 0, NULL); | |
754 | if (!corrected) | |
755 | clear_extent_bits(&BTRFS_I(inode)->io_tree, offset, end, | |
756 | EXTENT_DAMAGED, GFP_NOFS); | |
757 | } | |
758 | ||
759 | out: | |
760 | if (page) | |
761 | put_page(page); | |
7fb18a06 TK |
762 | |
763 | iput(inode); | |
0ef8e451 JS |
764 | |
765 | if (ret < 0) | |
766 | return ret; | |
767 | ||
768 | if (ret == 0 && corrected) { | |
769 | /* | |
770 | * we only need to call readpage for one of the inodes belonging | |
771 | * to this extent. so make iterate_extent_inodes stop | |
772 | */ | |
773 | return 1; | |
774 | } | |
775 | ||
776 | return -EIO; | |
777 | } | |
778 | ||
779 | static void scrub_fixup_nodatasum(struct btrfs_work *work) | |
780 | { | |
781 | int ret; | |
782 | struct scrub_fixup_nodatasum *fixup; | |
d9d181c1 | 783 | struct scrub_ctx *sctx; |
0ef8e451 | 784 | struct btrfs_trans_handle *trans = NULL; |
0ef8e451 JS |
785 | struct btrfs_path *path; |
786 | int uncorrectable = 0; | |
787 | ||
788 | fixup = container_of(work, struct scrub_fixup_nodatasum, work); | |
d9d181c1 | 789 | sctx = fixup->sctx; |
0ef8e451 JS |
790 | |
791 | path = btrfs_alloc_path(); | |
792 | if (!path) { | |
d9d181c1 SB |
793 | spin_lock(&sctx->stat_lock); |
794 | ++sctx->stat.malloc_errors; | |
795 | spin_unlock(&sctx->stat_lock); | |
0ef8e451 JS |
796 | uncorrectable = 1; |
797 | goto out; | |
798 | } | |
799 | ||
800 | trans = btrfs_join_transaction(fixup->root); | |
801 | if (IS_ERR(trans)) { | |
802 | uncorrectable = 1; | |
803 | goto out; | |
804 | } | |
805 | ||
806 | /* | |
807 | * the idea is to trigger a regular read through the standard path. we | |
808 | * read a page from the (failed) logical address by specifying the | |
809 | * corresponding copynum of the failed sector. thus, that readpage is | |
810 | * expected to fail. | |
811 | * that is the point where on-the-fly error correction will kick in | |
812 | * (once it's finished) and rewrite the failed sector if a good copy | |
813 | * can be found. | |
814 | */ | |
815 | ret = iterate_inodes_from_logical(fixup->logical, fixup->root->fs_info, | |
816 | path, scrub_fixup_readpage, | |
817 | fixup); | |
818 | if (ret < 0) { | |
819 | uncorrectable = 1; | |
820 | goto out; | |
821 | } | |
822 | WARN_ON(ret != 1); | |
823 | ||
d9d181c1 SB |
824 | spin_lock(&sctx->stat_lock); |
825 | ++sctx->stat.corrected_errors; | |
826 | spin_unlock(&sctx->stat_lock); | |
0ef8e451 JS |
827 | |
828 | out: | |
829 | if (trans && !IS_ERR(trans)) | |
830 | btrfs_end_transaction(trans, fixup->root); | |
831 | if (uncorrectable) { | |
d9d181c1 SB |
832 | spin_lock(&sctx->stat_lock); |
833 | ++sctx->stat.uncorrectable_errors; | |
834 | spin_unlock(&sctx->stat_lock); | |
ff023aac SB |
835 | btrfs_dev_replace_stats_inc( |
836 | &sctx->dev_root->fs_info->dev_replace. | |
837 | num_uncorrectable_read_errors); | |
efe120a0 FH |
838 | printk_ratelimited_in_rcu(KERN_ERR "BTRFS: " |
839 | "unable to fixup (nodatasum) error at logical %llu on dev %s\n", | |
c1c9ff7c | 840 | fixup->logical, rcu_str_deref(fixup->dev->name)); |
0ef8e451 JS |
841 | } |
842 | ||
843 | btrfs_free_path(path); | |
844 | kfree(fixup); | |
845 | ||
b6bfebc1 | 846 | scrub_pending_trans_workers_dec(sctx); |
0ef8e451 JS |
847 | } |
848 | ||
af8e2d1d MX |
849 | static inline void scrub_get_recover(struct scrub_recover *recover) |
850 | { | |
851 | atomic_inc(&recover->refs); | |
852 | } | |
853 | ||
854 | static inline void scrub_put_recover(struct scrub_recover *recover) | |
855 | { | |
856 | if (atomic_dec_and_test(&recover->refs)) { | |
6e9606d2 | 857 | btrfs_put_bbio(recover->bbio); |
af8e2d1d MX |
858 | kfree(recover); |
859 | } | |
860 | } | |
861 | ||
a2de733c | 862 | /* |
b5d67f64 SB |
863 | * scrub_handle_errored_block gets called when either verification of the |
864 | * pages failed or the bio failed to read, e.g. with EIO. In the latter | |
865 | * case, this function handles all pages in the bio, even though only one | |
866 | * may be bad. | |
867 | * The goal of this function is to repair the errored block by using the | |
868 | * contents of one of the mirrors. | |
a2de733c | 869 | */ |
b5d67f64 | 870 | static int scrub_handle_errored_block(struct scrub_block *sblock_to_check) |
a2de733c | 871 | { |
d9d181c1 | 872 | struct scrub_ctx *sctx = sblock_to_check->sctx; |
a36cf8b8 | 873 | struct btrfs_device *dev; |
b5d67f64 SB |
874 | struct btrfs_fs_info *fs_info; |
875 | u64 length; | |
876 | u64 logical; | |
877 | u64 generation; | |
878 | unsigned int failed_mirror_index; | |
879 | unsigned int is_metadata; | |
880 | unsigned int have_csum; | |
881 | u8 *csum; | |
882 | struct scrub_block *sblocks_for_recheck; /* holds one for each mirror */ | |
883 | struct scrub_block *sblock_bad; | |
884 | int ret; | |
885 | int mirror_index; | |
886 | int page_num; | |
887 | int success; | |
558540c1 | 888 | static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL, |
b5d67f64 SB |
889 | DEFAULT_RATELIMIT_BURST); |
890 | ||
891 | BUG_ON(sblock_to_check->page_count < 1); | |
a36cf8b8 | 892 | fs_info = sctx->dev_root->fs_info; |
4ded4f63 SB |
893 | if (sblock_to_check->pagev[0]->flags & BTRFS_EXTENT_FLAG_SUPER) { |
894 | /* | |
895 | * if we find an error in a super block, we just report it. | |
896 | * They will get written with the next transaction commit | |
897 | * anyway | |
898 | */ | |
899 | spin_lock(&sctx->stat_lock); | |
900 | ++sctx->stat.super_errors; | |
901 | spin_unlock(&sctx->stat_lock); | |
902 | return 0; | |
903 | } | |
b5d67f64 | 904 | length = sblock_to_check->page_count * PAGE_SIZE; |
7a9e9987 SB |
905 | logical = sblock_to_check->pagev[0]->logical; |
906 | generation = sblock_to_check->pagev[0]->generation; | |
907 | BUG_ON(sblock_to_check->pagev[0]->mirror_num < 1); | |
908 | failed_mirror_index = sblock_to_check->pagev[0]->mirror_num - 1; | |
909 | is_metadata = !(sblock_to_check->pagev[0]->flags & | |
b5d67f64 | 910 | BTRFS_EXTENT_FLAG_DATA); |
7a9e9987 SB |
911 | have_csum = sblock_to_check->pagev[0]->have_csum; |
912 | csum = sblock_to_check->pagev[0]->csum; | |
913 | dev = sblock_to_check->pagev[0]->dev; | |
13db62b7 | 914 | |
ff023aac SB |
915 | if (sctx->is_dev_replace && !is_metadata && !have_csum) { |
916 | sblocks_for_recheck = NULL; | |
917 | goto nodatasum_case; | |
918 | } | |
919 | ||
b5d67f64 SB |
920 | /* |
921 | * read all mirrors one after the other. This includes to | |
922 | * re-read the extent or metadata block that failed (that was | |
923 | * the cause that this fixup code is called) another time, | |
924 | * page by page this time in order to know which pages | |
925 | * caused I/O errors and which ones are good (for all mirrors). | |
926 | * It is the goal to handle the situation when more than one | |
927 | * mirror contains I/O errors, but the errors do not | |
928 | * overlap, i.e. the data can be repaired by selecting the | |
929 | * pages from those mirrors without I/O error on the | |
930 | * particular pages. One example (with blocks >= 2 * PAGE_SIZE) | |
931 | * would be that mirror #1 has an I/O error on the first page, | |
932 | * the second page is good, and mirror #2 has an I/O error on | |
933 | * the second page, but the first page is good. | |
934 | * Then the first page of the first mirror can be repaired by | |
935 | * taking the first page of the second mirror, and the | |
936 | * second page of the second mirror can be repaired by | |
937 | * copying the contents of the 2nd page of the 1st mirror. | |
938 | * One more note: if the pages of one mirror contain I/O | |
939 | * errors, the checksum cannot be verified. In order to get | |
940 | * the best data for repairing, the first attempt is to find | |
941 | * a mirror without I/O errors and with a validated checksum. | |
942 | * Only if this is not possible, the pages are picked from | |
943 | * mirrors with I/O errors without considering the checksum. | |
944 | * If the latter is the case, at the end, the checksum of the | |
945 | * repaired area is verified in order to correctly maintain | |
946 | * the statistics. | |
947 | */ | |
948 | ||
949 | sblocks_for_recheck = kzalloc(BTRFS_MAX_MIRRORS * | |
950 | sizeof(*sblocks_for_recheck), | |
951 | GFP_NOFS); | |
952 | if (!sblocks_for_recheck) { | |
d9d181c1 SB |
953 | spin_lock(&sctx->stat_lock); |
954 | sctx->stat.malloc_errors++; | |
955 | sctx->stat.read_errors++; | |
956 | sctx->stat.uncorrectable_errors++; | |
957 | spin_unlock(&sctx->stat_lock); | |
a36cf8b8 | 958 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS); |
b5d67f64 | 959 | goto out; |
a2de733c AJ |
960 | } |
961 | ||
b5d67f64 | 962 | /* setup the context, map the logical blocks and alloc the pages */ |
ff023aac | 963 | ret = scrub_setup_recheck_block(sctx, fs_info, sblock_to_check, length, |
b5d67f64 SB |
964 | logical, sblocks_for_recheck); |
965 | if (ret) { | |
d9d181c1 SB |
966 | spin_lock(&sctx->stat_lock); |
967 | sctx->stat.read_errors++; | |
968 | sctx->stat.uncorrectable_errors++; | |
969 | spin_unlock(&sctx->stat_lock); | |
a36cf8b8 | 970 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS); |
b5d67f64 SB |
971 | goto out; |
972 | } | |
973 | BUG_ON(failed_mirror_index >= BTRFS_MAX_MIRRORS); | |
974 | sblock_bad = sblocks_for_recheck + failed_mirror_index; | |
13db62b7 | 975 | |
b5d67f64 | 976 | /* build and submit the bios for the failed mirror, check checksums */ |
34f5c8e9 | 977 | scrub_recheck_block(fs_info, sblock_bad, is_metadata, have_csum, |
af8e2d1d | 978 | csum, generation, sctx->csum_size, 1); |
a2de733c | 979 | |
b5d67f64 SB |
980 | if (!sblock_bad->header_error && !sblock_bad->checksum_error && |
981 | sblock_bad->no_io_error_seen) { | |
982 | /* | |
983 | * the error disappeared after reading page by page, or | |
984 | * the area was part of a huge bio and other parts of the | |
985 | * bio caused I/O errors, or the block layer merged several | |
986 | * read requests into one and the error is caused by a | |
987 | * different bio (usually one of the two latter cases is | |
988 | * the cause) | |
989 | */ | |
d9d181c1 SB |
990 | spin_lock(&sctx->stat_lock); |
991 | sctx->stat.unverified_errors++; | |
5a6ac9ea | 992 | sblock_to_check->data_corrected = 1; |
d9d181c1 | 993 | spin_unlock(&sctx->stat_lock); |
a2de733c | 994 | |
ff023aac SB |
995 | if (sctx->is_dev_replace) |
996 | scrub_write_block_to_dev_replace(sblock_bad); | |
b5d67f64 | 997 | goto out; |
a2de733c | 998 | } |
a2de733c | 999 | |
b5d67f64 | 1000 | if (!sblock_bad->no_io_error_seen) { |
d9d181c1 SB |
1001 | spin_lock(&sctx->stat_lock); |
1002 | sctx->stat.read_errors++; | |
1003 | spin_unlock(&sctx->stat_lock); | |
b5d67f64 SB |
1004 | if (__ratelimit(&_rs)) |
1005 | scrub_print_warning("i/o error", sblock_to_check); | |
a36cf8b8 | 1006 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS); |
b5d67f64 | 1007 | } else if (sblock_bad->checksum_error) { |
d9d181c1 SB |
1008 | spin_lock(&sctx->stat_lock); |
1009 | sctx->stat.csum_errors++; | |
1010 | spin_unlock(&sctx->stat_lock); | |
b5d67f64 SB |
1011 | if (__ratelimit(&_rs)) |
1012 | scrub_print_warning("checksum error", sblock_to_check); | |
a36cf8b8 | 1013 | btrfs_dev_stat_inc_and_print(dev, |
442a4f63 | 1014 | BTRFS_DEV_STAT_CORRUPTION_ERRS); |
b5d67f64 | 1015 | } else if (sblock_bad->header_error) { |
d9d181c1 SB |
1016 | spin_lock(&sctx->stat_lock); |
1017 | sctx->stat.verify_errors++; | |
1018 | spin_unlock(&sctx->stat_lock); | |
b5d67f64 SB |
1019 | if (__ratelimit(&_rs)) |
1020 | scrub_print_warning("checksum/header error", | |
1021 | sblock_to_check); | |
442a4f63 | 1022 | if (sblock_bad->generation_error) |
a36cf8b8 | 1023 | btrfs_dev_stat_inc_and_print(dev, |
442a4f63 SB |
1024 | BTRFS_DEV_STAT_GENERATION_ERRS); |
1025 | else | |
a36cf8b8 | 1026 | btrfs_dev_stat_inc_and_print(dev, |
442a4f63 | 1027 | BTRFS_DEV_STAT_CORRUPTION_ERRS); |
b5d67f64 | 1028 | } |
a2de733c | 1029 | |
33ef30ad ID |
1030 | if (sctx->readonly) { |
1031 | ASSERT(!sctx->is_dev_replace); | |
1032 | goto out; | |
1033 | } | |
a2de733c | 1034 | |
b5d67f64 SB |
1035 | if (!is_metadata && !have_csum) { |
1036 | struct scrub_fixup_nodatasum *fixup_nodatasum; | |
a2de733c | 1037 | |
ff023aac SB |
1038 | WARN_ON(sctx->is_dev_replace); |
1039 | ||
b25c94c5 ZL |
1040 | nodatasum_case: |
1041 | ||
b5d67f64 SB |
1042 | /* |
1043 | * !is_metadata and !have_csum, this means that the data | |
1044 | * might not be COW'ed, that it might be modified | |
1045 | * concurrently. The general strategy to work on the | |
1046 | * commit root does not help in the case when COW is not | |
1047 | * used. | |
1048 | */ | |
1049 | fixup_nodatasum = kzalloc(sizeof(*fixup_nodatasum), GFP_NOFS); | |
1050 | if (!fixup_nodatasum) | |
1051 | goto did_not_correct_error; | |
d9d181c1 | 1052 | fixup_nodatasum->sctx = sctx; |
a36cf8b8 | 1053 | fixup_nodatasum->dev = dev; |
b5d67f64 SB |
1054 | fixup_nodatasum->logical = logical; |
1055 | fixup_nodatasum->root = fs_info->extent_root; | |
1056 | fixup_nodatasum->mirror_num = failed_mirror_index + 1; | |
b6bfebc1 | 1057 | scrub_pending_trans_workers_inc(sctx); |
9e0af237 LB |
1058 | btrfs_init_work(&fixup_nodatasum->work, btrfs_scrub_helper, |
1059 | scrub_fixup_nodatasum, NULL, NULL); | |
0339ef2f QW |
1060 | btrfs_queue_work(fs_info->scrub_workers, |
1061 | &fixup_nodatasum->work); | |
b5d67f64 | 1062 | goto out; |
a2de733c AJ |
1063 | } |
1064 | ||
b5d67f64 SB |
1065 | /* |
1066 | * now build and submit the bios for the other mirrors, check | |
cb2ced73 SB |
1067 | * checksums. |
1068 | * First try to pick the mirror which is completely without I/O | |
b5d67f64 SB |
1069 | * errors and also does not have a checksum error. |
1070 | * If one is found, and if a checksum is present, the full block | |
1071 | * that is known to contain an error is rewritten. Afterwards | |
1072 | * the block is known to be corrected. | |
1073 | * If a mirror is found which is completely correct, and no | |
1074 | * checksum is present, only those pages are rewritten that had | |
1075 | * an I/O error in the block to be repaired, since it cannot be | |
1076 | * determined, which copy of the other pages is better (and it | |
1077 | * could happen otherwise that a correct page would be | |
1078 | * overwritten by a bad one). | |
1079 | */ | |
1080 | for (mirror_index = 0; | |
1081 | mirror_index < BTRFS_MAX_MIRRORS && | |
1082 | sblocks_for_recheck[mirror_index].page_count > 0; | |
1083 | mirror_index++) { | |
cb2ced73 | 1084 | struct scrub_block *sblock_other; |
b5d67f64 | 1085 | |
cb2ced73 SB |
1086 | if (mirror_index == failed_mirror_index) |
1087 | continue; | |
1088 | sblock_other = sblocks_for_recheck + mirror_index; | |
1089 | ||
1090 | /* build and submit the bios, check checksums */ | |
34f5c8e9 SB |
1091 | scrub_recheck_block(fs_info, sblock_other, is_metadata, |
1092 | have_csum, csum, generation, | |
af8e2d1d | 1093 | sctx->csum_size, 0); |
34f5c8e9 SB |
1094 | |
1095 | if (!sblock_other->header_error && | |
b5d67f64 SB |
1096 | !sblock_other->checksum_error && |
1097 | sblock_other->no_io_error_seen) { | |
ff023aac SB |
1098 | if (sctx->is_dev_replace) { |
1099 | scrub_write_block_to_dev_replace(sblock_other); | |
114ab50d | 1100 | goto corrected_error; |
ff023aac | 1101 | } else { |
ff023aac | 1102 | ret = scrub_repair_block_from_good_copy( |
114ab50d ZL |
1103 | sblock_bad, sblock_other); |
1104 | if (!ret) | |
1105 | goto corrected_error; | |
ff023aac | 1106 | } |
b5d67f64 SB |
1107 | } |
1108 | } | |
a2de733c AJ |
1109 | |
1110 | /* | |
ff023aac SB |
1111 | * for dev_replace, pick good pages and write to the target device. |
1112 | */ | |
1113 | if (sctx->is_dev_replace) { | |
1114 | success = 1; | |
1115 | for (page_num = 0; page_num < sblock_bad->page_count; | |
1116 | page_num++) { | |
8d6738c1 | 1117 | struct scrub_block *sblock_other = NULL; |
ff023aac | 1118 | |
ff023aac SB |
1119 | for (mirror_index = 0; |
1120 | mirror_index < BTRFS_MAX_MIRRORS && | |
1121 | sblocks_for_recheck[mirror_index].page_count > 0; | |
1122 | mirror_index++) { | |
8d6738c1 ZL |
1123 | if (!sblocks_for_recheck[mirror_index]. |
1124 | pagev[page_num]->io_error) { | |
1125 | sblock_other = sblocks_for_recheck + | |
1126 | mirror_index; | |
1127 | break; | |
ff023aac SB |
1128 | } |
1129 | } | |
1130 | ||
8d6738c1 | 1131 | if (!sblock_other) { |
ff023aac SB |
1132 | /* |
1133 | * did not find a mirror to fetch the page | |
1134 | * from. scrub_write_page_to_dev_replace() | |
1135 | * handles this case (page->io_error), by | |
1136 | * filling the block with zeros before | |
1137 | * submitting the write request | |
1138 | */ | |
8d6738c1 ZL |
1139 | sblock_other = sblock_bad; |
1140 | success = 0; | |
1141 | } | |
1142 | ||
1143 | if (scrub_write_page_to_dev_replace(sblock_other, | |
1144 | page_num) != 0) { | |
1145 | btrfs_dev_replace_stats_inc( | |
1146 | &sctx->dev_root-> | |
1147 | fs_info->dev_replace. | |
1148 | num_write_errors); | |
ff023aac | 1149 | success = 0; |
ff023aac SB |
1150 | } |
1151 | } | |
1152 | ||
1153 | goto out; | |
1154 | } | |
1155 | ||
1156 | /* | |
1157 | * for regular scrub, repair those pages that are errored. | |
1158 | * In case of I/O errors in the area that is supposed to be | |
b5d67f64 SB |
1159 | * repaired, continue by picking good copies of those pages. |
1160 | * Select the good pages from mirrors to rewrite bad pages from | |
1161 | * the area to fix. Afterwards verify the checksum of the block | |
1162 | * that is supposed to be repaired. This verification step is | |
1163 | * only done for the purpose of statistic counting and for the | |
1164 | * final scrub report, whether errors remain. | |
1165 | * A perfect algorithm could make use of the checksum and try | |
1166 | * all possible combinations of pages from the different mirrors | |
1167 | * until the checksum verification succeeds. For example, when | |
1168 | * the 2nd page of mirror #1 faces I/O errors, and the 2nd page | |
1169 | * of mirror #2 is readable but the final checksum test fails, | |
1170 | * then the 2nd page of mirror #3 could be tried, whether now | |
1171 | * the final checksum succeedes. But this would be a rare | |
1172 | * exception and is therefore not implemented. At least it is | |
1173 | * avoided that the good copy is overwritten. | |
1174 | * A more useful improvement would be to pick the sectors | |
1175 | * without I/O error based on sector sizes (512 bytes on legacy | |
1176 | * disks) instead of on PAGE_SIZE. Then maybe 512 byte of one | |
1177 | * mirror could be repaired by taking 512 byte of a different | |
1178 | * mirror, even if other 512 byte sectors in the same PAGE_SIZE | |
1179 | * area are unreadable. | |
a2de733c | 1180 | */ |
a2de733c | 1181 | |
b5d67f64 SB |
1182 | /* can only fix I/O errors from here on */ |
1183 | if (sblock_bad->no_io_error_seen) | |
1184 | goto did_not_correct_error; | |
1185 | ||
1186 | success = 1; | |
1187 | for (page_num = 0; page_num < sblock_bad->page_count; page_num++) { | |
7a9e9987 | 1188 | struct scrub_page *page_bad = sblock_bad->pagev[page_num]; |
b5d67f64 SB |
1189 | |
1190 | if (!page_bad->io_error) | |
a2de733c | 1191 | continue; |
b5d67f64 SB |
1192 | |
1193 | for (mirror_index = 0; | |
1194 | mirror_index < BTRFS_MAX_MIRRORS && | |
1195 | sblocks_for_recheck[mirror_index].page_count > 0; | |
1196 | mirror_index++) { | |
1197 | struct scrub_block *sblock_other = sblocks_for_recheck + | |
1198 | mirror_index; | |
7a9e9987 SB |
1199 | struct scrub_page *page_other = sblock_other->pagev[ |
1200 | page_num]; | |
b5d67f64 SB |
1201 | |
1202 | if (!page_other->io_error) { | |
1203 | ret = scrub_repair_page_from_good_copy( | |
1204 | sblock_bad, sblock_other, page_num, 0); | |
1205 | if (0 == ret) { | |
1206 | page_bad->io_error = 0; | |
1207 | break; /* succeeded for this page */ | |
1208 | } | |
1209 | } | |
96e36920 | 1210 | } |
a2de733c | 1211 | |
b5d67f64 SB |
1212 | if (page_bad->io_error) { |
1213 | /* did not find a mirror to copy the page from */ | |
1214 | success = 0; | |
1215 | } | |
a2de733c | 1216 | } |
a2de733c | 1217 | |
b5d67f64 SB |
1218 | if (success) { |
1219 | if (is_metadata || have_csum) { | |
1220 | /* | |
1221 | * need to verify the checksum now that all | |
1222 | * sectors on disk are repaired (the write | |
1223 | * request for data to be repaired is on its way). | |
1224 | * Just be lazy and use scrub_recheck_block() | |
1225 | * which re-reads the data before the checksum | |
1226 | * is verified, but most likely the data comes out | |
1227 | * of the page cache. | |
1228 | */ | |
34f5c8e9 SB |
1229 | scrub_recheck_block(fs_info, sblock_bad, |
1230 | is_metadata, have_csum, csum, | |
af8e2d1d | 1231 | generation, sctx->csum_size, 1); |
34f5c8e9 | 1232 | if (!sblock_bad->header_error && |
b5d67f64 SB |
1233 | !sblock_bad->checksum_error && |
1234 | sblock_bad->no_io_error_seen) | |
1235 | goto corrected_error; | |
1236 | else | |
1237 | goto did_not_correct_error; | |
1238 | } else { | |
1239 | corrected_error: | |
d9d181c1 SB |
1240 | spin_lock(&sctx->stat_lock); |
1241 | sctx->stat.corrected_errors++; | |
5a6ac9ea | 1242 | sblock_to_check->data_corrected = 1; |
d9d181c1 | 1243 | spin_unlock(&sctx->stat_lock); |
606686ee | 1244 | printk_ratelimited_in_rcu(KERN_ERR |
efe120a0 | 1245 | "BTRFS: fixed up error at logical %llu on dev %s\n", |
c1c9ff7c | 1246 | logical, rcu_str_deref(dev->name)); |
8628764e | 1247 | } |
b5d67f64 SB |
1248 | } else { |
1249 | did_not_correct_error: | |
d9d181c1 SB |
1250 | spin_lock(&sctx->stat_lock); |
1251 | sctx->stat.uncorrectable_errors++; | |
1252 | spin_unlock(&sctx->stat_lock); | |
606686ee | 1253 | printk_ratelimited_in_rcu(KERN_ERR |
efe120a0 | 1254 | "BTRFS: unable to fixup (regular) error at logical %llu on dev %s\n", |
c1c9ff7c | 1255 | logical, rcu_str_deref(dev->name)); |
96e36920 | 1256 | } |
a2de733c | 1257 | |
b5d67f64 SB |
1258 | out: |
1259 | if (sblocks_for_recheck) { | |
1260 | for (mirror_index = 0; mirror_index < BTRFS_MAX_MIRRORS; | |
1261 | mirror_index++) { | |
1262 | struct scrub_block *sblock = sblocks_for_recheck + | |
1263 | mirror_index; | |
af8e2d1d | 1264 | struct scrub_recover *recover; |
b5d67f64 SB |
1265 | int page_index; |
1266 | ||
7a9e9987 SB |
1267 | for (page_index = 0; page_index < sblock->page_count; |
1268 | page_index++) { | |
1269 | sblock->pagev[page_index]->sblock = NULL; | |
af8e2d1d MX |
1270 | recover = sblock->pagev[page_index]->recover; |
1271 | if (recover) { | |
1272 | scrub_put_recover(recover); | |
1273 | sblock->pagev[page_index]->recover = | |
1274 | NULL; | |
1275 | } | |
7a9e9987 SB |
1276 | scrub_page_put(sblock->pagev[page_index]); |
1277 | } | |
b5d67f64 SB |
1278 | } |
1279 | kfree(sblocks_for_recheck); | |
1280 | } | |
a2de733c | 1281 | |
b5d67f64 SB |
1282 | return 0; |
1283 | } | |
a2de733c | 1284 | |
8e5cfb55 | 1285 | static inline int scrub_nr_raid_mirrors(struct btrfs_bio *bbio) |
af8e2d1d | 1286 | { |
8e5cfb55 | 1287 | if (bbio->raid_map) { |
e34c330d ZL |
1288 | int real_stripes = bbio->num_stripes - bbio->num_tgtdevs; |
1289 | ||
8e5cfb55 | 1290 | if (bbio->raid_map[real_stripes - 1] == RAID6_Q_STRIPE) |
af8e2d1d MX |
1291 | return 3; |
1292 | else | |
1293 | return 2; | |
1294 | } else { | |
1295 | return (int)bbio->num_stripes; | |
1296 | } | |
1297 | } | |
1298 | ||
1299 | static inline void scrub_stripe_index_and_offset(u64 logical, u64 *raid_map, | |
1300 | u64 mapped_length, | |
1301 | int nstripes, int mirror, | |
1302 | int *stripe_index, | |
1303 | u64 *stripe_offset) | |
1304 | { | |
1305 | int i; | |
1306 | ||
1307 | if (raid_map) { | |
1308 | /* RAID5/6 */ | |
1309 | for (i = 0; i < nstripes; i++) { | |
1310 | if (raid_map[i] == RAID6_Q_STRIPE || | |
1311 | raid_map[i] == RAID5_P_STRIPE) | |
1312 | continue; | |
1313 | ||
1314 | if (logical >= raid_map[i] && | |
1315 | logical < raid_map[i] + mapped_length) | |
1316 | break; | |
1317 | } | |
1318 | ||
1319 | *stripe_index = i; | |
1320 | *stripe_offset = logical - raid_map[i]; | |
1321 | } else { | |
1322 | /* The other RAID type */ | |
1323 | *stripe_index = mirror; | |
1324 | *stripe_offset = 0; | |
1325 | } | |
1326 | } | |
1327 | ||
d9d181c1 | 1328 | static int scrub_setup_recheck_block(struct scrub_ctx *sctx, |
3ec706c8 | 1329 | struct btrfs_fs_info *fs_info, |
ff023aac | 1330 | struct scrub_block *original_sblock, |
b5d67f64 SB |
1331 | u64 length, u64 logical, |
1332 | struct scrub_block *sblocks_for_recheck) | |
1333 | { | |
af8e2d1d MX |
1334 | struct scrub_recover *recover; |
1335 | struct btrfs_bio *bbio; | |
af8e2d1d MX |
1336 | u64 sublen; |
1337 | u64 mapped_length; | |
1338 | u64 stripe_offset; | |
1339 | int stripe_index; | |
b5d67f64 SB |
1340 | int page_index; |
1341 | int mirror_index; | |
af8e2d1d | 1342 | int nmirrors; |
b5d67f64 SB |
1343 | int ret; |
1344 | ||
1345 | /* | |
7a9e9987 | 1346 | * note: the two members ref_count and outstanding_pages |
b5d67f64 SB |
1347 | * are not used (and not set) in the blocks that are used for |
1348 | * the recheck procedure | |
1349 | */ | |
1350 | ||
1351 | page_index = 0; | |
1352 | while (length > 0) { | |
af8e2d1d MX |
1353 | sublen = min_t(u64, length, PAGE_SIZE); |
1354 | mapped_length = sublen; | |
1355 | bbio = NULL; | |
a2de733c | 1356 | |
b5d67f64 SB |
1357 | /* |
1358 | * with a length of PAGE_SIZE, each returned stripe | |
1359 | * represents one mirror | |
1360 | */ | |
af8e2d1d | 1361 | ret = btrfs_map_sblock(fs_info, REQ_GET_READ_MIRRORS, logical, |
8e5cfb55 | 1362 | &mapped_length, &bbio, 0, 1); |
b5d67f64 | 1363 | if (ret || !bbio || mapped_length < sublen) { |
6e9606d2 | 1364 | btrfs_put_bbio(bbio); |
b5d67f64 SB |
1365 | return -EIO; |
1366 | } | |
a2de733c | 1367 | |
af8e2d1d MX |
1368 | recover = kzalloc(sizeof(struct scrub_recover), GFP_NOFS); |
1369 | if (!recover) { | |
6e9606d2 | 1370 | btrfs_put_bbio(bbio); |
af8e2d1d MX |
1371 | return -ENOMEM; |
1372 | } | |
1373 | ||
1374 | atomic_set(&recover->refs, 1); | |
1375 | recover->bbio = bbio; | |
af8e2d1d MX |
1376 | recover->map_length = mapped_length; |
1377 | ||
ff023aac | 1378 | BUG_ON(page_index >= SCRUB_PAGES_PER_RD_BIO); |
af8e2d1d | 1379 | |
8e5cfb55 | 1380 | nmirrors = scrub_nr_raid_mirrors(bbio); |
af8e2d1d | 1381 | for (mirror_index = 0; mirror_index < nmirrors; |
b5d67f64 SB |
1382 | mirror_index++) { |
1383 | struct scrub_block *sblock; | |
1384 | struct scrub_page *page; | |
1385 | ||
1386 | if (mirror_index >= BTRFS_MAX_MIRRORS) | |
dc5f7a3b | 1387 | break; |
b5d67f64 SB |
1388 | |
1389 | sblock = sblocks_for_recheck + mirror_index; | |
7a9e9987 SB |
1390 | sblock->sctx = sctx; |
1391 | page = kzalloc(sizeof(*page), GFP_NOFS); | |
1392 | if (!page) { | |
1393 | leave_nomem: | |
d9d181c1 SB |
1394 | spin_lock(&sctx->stat_lock); |
1395 | sctx->stat.malloc_errors++; | |
1396 | spin_unlock(&sctx->stat_lock); | |
af8e2d1d | 1397 | scrub_put_recover(recover); |
b5d67f64 SB |
1398 | return -ENOMEM; |
1399 | } | |
7a9e9987 SB |
1400 | scrub_page_get(page); |
1401 | sblock->pagev[page_index] = page; | |
1402 | page->logical = logical; | |
af8e2d1d | 1403 | |
8e5cfb55 | 1404 | scrub_stripe_index_and_offset(logical, bbio->raid_map, |
af8e2d1d | 1405 | mapped_length, |
e34c330d ZL |
1406 | bbio->num_stripes - |
1407 | bbio->num_tgtdevs, | |
af8e2d1d MX |
1408 | mirror_index, |
1409 | &stripe_index, | |
1410 | &stripe_offset); | |
1411 | page->physical = bbio->stripes[stripe_index].physical + | |
1412 | stripe_offset; | |
1413 | page->dev = bbio->stripes[stripe_index].dev; | |
1414 | ||
ff023aac SB |
1415 | BUG_ON(page_index >= original_sblock->page_count); |
1416 | page->physical_for_dev_replace = | |
1417 | original_sblock->pagev[page_index]-> | |
1418 | physical_for_dev_replace; | |
7a9e9987 | 1419 | /* for missing devices, dev->bdev is NULL */ |
7a9e9987 | 1420 | page->mirror_num = mirror_index + 1; |
b5d67f64 | 1421 | sblock->page_count++; |
7a9e9987 SB |
1422 | page->page = alloc_page(GFP_NOFS); |
1423 | if (!page->page) | |
1424 | goto leave_nomem; | |
af8e2d1d MX |
1425 | |
1426 | scrub_get_recover(recover); | |
1427 | page->recover = recover; | |
b5d67f64 | 1428 | } |
af8e2d1d | 1429 | scrub_put_recover(recover); |
b5d67f64 SB |
1430 | length -= sublen; |
1431 | logical += sublen; | |
1432 | page_index++; | |
1433 | } | |
1434 | ||
1435 | return 0; | |
96e36920 ID |
1436 | } |
1437 | ||
af8e2d1d MX |
1438 | struct scrub_bio_ret { |
1439 | struct completion event; | |
1440 | int error; | |
1441 | }; | |
1442 | ||
1443 | static void scrub_bio_wait_endio(struct bio *bio, int error) | |
1444 | { | |
1445 | struct scrub_bio_ret *ret = bio->bi_private; | |
1446 | ||
1447 | ret->error = error; | |
1448 | complete(&ret->event); | |
1449 | } | |
1450 | ||
1451 | static inline int scrub_is_page_on_raid56(struct scrub_page *page) | |
1452 | { | |
8e5cfb55 | 1453 | return page->recover && page->recover->bbio->raid_map; |
af8e2d1d MX |
1454 | } |
1455 | ||
1456 | static int scrub_submit_raid56_bio_wait(struct btrfs_fs_info *fs_info, | |
1457 | struct bio *bio, | |
1458 | struct scrub_page *page) | |
1459 | { | |
1460 | struct scrub_bio_ret done; | |
1461 | int ret; | |
1462 | ||
1463 | init_completion(&done.event); | |
1464 | done.error = 0; | |
1465 | bio->bi_iter.bi_sector = page->logical >> 9; | |
1466 | bio->bi_private = &done; | |
1467 | bio->bi_end_io = scrub_bio_wait_endio; | |
1468 | ||
1469 | ret = raid56_parity_recover(fs_info->fs_root, bio, page->recover->bbio, | |
af8e2d1d | 1470 | page->recover->map_length, |
4245215d | 1471 | page->mirror_num, 0); |
af8e2d1d MX |
1472 | if (ret) |
1473 | return ret; | |
1474 | ||
1475 | wait_for_completion(&done.event); | |
1476 | if (done.error) | |
1477 | return -EIO; | |
1478 | ||
1479 | return 0; | |
1480 | } | |
1481 | ||
b5d67f64 SB |
1482 | /* |
1483 | * this function will check the on disk data for checksum errors, header | |
1484 | * errors and read I/O errors. If any I/O errors happen, the exact pages | |
1485 | * which are errored are marked as being bad. The goal is to enable scrub | |
1486 | * to take those pages that are not errored from all the mirrors so that | |
1487 | * the pages that are errored in the just handled mirror can be repaired. | |
1488 | */ | |
34f5c8e9 SB |
1489 | static void scrub_recheck_block(struct btrfs_fs_info *fs_info, |
1490 | struct scrub_block *sblock, int is_metadata, | |
1491 | int have_csum, u8 *csum, u64 generation, | |
af8e2d1d | 1492 | u16 csum_size, int retry_failed_mirror) |
96e36920 | 1493 | { |
b5d67f64 | 1494 | int page_num; |
96e36920 | 1495 | |
b5d67f64 SB |
1496 | sblock->no_io_error_seen = 1; |
1497 | sblock->header_error = 0; | |
1498 | sblock->checksum_error = 0; | |
96e36920 | 1499 | |
b5d67f64 SB |
1500 | for (page_num = 0; page_num < sblock->page_count; page_num++) { |
1501 | struct bio *bio; | |
7a9e9987 | 1502 | struct scrub_page *page = sblock->pagev[page_num]; |
b5d67f64 | 1503 | |
442a4f63 | 1504 | if (page->dev->bdev == NULL) { |
ea9947b4 SB |
1505 | page->io_error = 1; |
1506 | sblock->no_io_error_seen = 0; | |
1507 | continue; | |
1508 | } | |
1509 | ||
7a9e9987 | 1510 | WARN_ON(!page->page); |
9be3395b | 1511 | bio = btrfs_io_bio_alloc(GFP_NOFS, 1); |
34f5c8e9 SB |
1512 | if (!bio) { |
1513 | page->io_error = 1; | |
1514 | sblock->no_io_error_seen = 0; | |
1515 | continue; | |
1516 | } | |
442a4f63 | 1517 | bio->bi_bdev = page->dev->bdev; |
b5d67f64 | 1518 | |
34f5c8e9 | 1519 | bio_add_page(bio, page->page, PAGE_SIZE, 0); |
af8e2d1d MX |
1520 | if (!retry_failed_mirror && scrub_is_page_on_raid56(page)) { |
1521 | if (scrub_submit_raid56_bio_wait(fs_info, bio, page)) | |
1522 | sblock->no_io_error_seen = 0; | |
1523 | } else { | |
1524 | bio->bi_iter.bi_sector = page->physical >> 9; | |
1525 | ||
1526 | if (btrfsic_submit_bio_wait(READ, bio)) | |
1527 | sblock->no_io_error_seen = 0; | |
1528 | } | |
33879d45 | 1529 | |
b5d67f64 SB |
1530 | bio_put(bio); |
1531 | } | |
96e36920 | 1532 | |
b5d67f64 SB |
1533 | if (sblock->no_io_error_seen) |
1534 | scrub_recheck_block_checksum(fs_info, sblock, is_metadata, | |
1535 | have_csum, csum, generation, | |
1536 | csum_size); | |
1537 | ||
34f5c8e9 | 1538 | return; |
a2de733c AJ |
1539 | } |
1540 | ||
17a9be2f MX |
1541 | static inline int scrub_check_fsid(u8 fsid[], |
1542 | struct scrub_page *spage) | |
1543 | { | |
1544 | struct btrfs_fs_devices *fs_devices = spage->dev->fs_devices; | |
1545 | int ret; | |
1546 | ||
1547 | ret = memcmp(fsid, fs_devices->fsid, BTRFS_UUID_SIZE); | |
1548 | return !ret; | |
1549 | } | |
1550 | ||
b5d67f64 SB |
1551 | static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info, |
1552 | struct scrub_block *sblock, | |
1553 | int is_metadata, int have_csum, | |
1554 | const u8 *csum, u64 generation, | |
1555 | u16 csum_size) | |
a2de733c | 1556 | { |
b5d67f64 SB |
1557 | int page_num; |
1558 | u8 calculated_csum[BTRFS_CSUM_SIZE]; | |
1559 | u32 crc = ~(u32)0; | |
b5d67f64 SB |
1560 | void *mapped_buffer; |
1561 | ||
7a9e9987 | 1562 | WARN_ON(!sblock->pagev[0]->page); |
b5d67f64 SB |
1563 | if (is_metadata) { |
1564 | struct btrfs_header *h; | |
1565 | ||
7a9e9987 | 1566 | mapped_buffer = kmap_atomic(sblock->pagev[0]->page); |
b5d67f64 SB |
1567 | h = (struct btrfs_header *)mapped_buffer; |
1568 | ||
3cae210f | 1569 | if (sblock->pagev[0]->logical != btrfs_stack_header_bytenr(h) || |
17a9be2f | 1570 | !scrub_check_fsid(h->fsid, sblock->pagev[0]) || |
b5d67f64 | 1571 | memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid, |
442a4f63 | 1572 | BTRFS_UUID_SIZE)) { |
b5d67f64 | 1573 | sblock->header_error = 1; |
3cae210f | 1574 | } else if (generation != btrfs_stack_header_generation(h)) { |
442a4f63 SB |
1575 | sblock->header_error = 1; |
1576 | sblock->generation_error = 1; | |
1577 | } | |
b5d67f64 SB |
1578 | csum = h->csum; |
1579 | } else { | |
1580 | if (!have_csum) | |
1581 | return; | |
a2de733c | 1582 | |
7a9e9987 | 1583 | mapped_buffer = kmap_atomic(sblock->pagev[0]->page); |
b5d67f64 | 1584 | } |
a2de733c | 1585 | |
b5d67f64 SB |
1586 | for (page_num = 0;;) { |
1587 | if (page_num == 0 && is_metadata) | |
b0496686 | 1588 | crc = btrfs_csum_data( |
b5d67f64 SB |
1589 | ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE, |
1590 | crc, PAGE_SIZE - BTRFS_CSUM_SIZE); | |
1591 | else | |
b0496686 | 1592 | crc = btrfs_csum_data(mapped_buffer, crc, PAGE_SIZE); |
b5d67f64 | 1593 | |
9613bebb | 1594 | kunmap_atomic(mapped_buffer); |
b5d67f64 SB |
1595 | page_num++; |
1596 | if (page_num >= sblock->page_count) | |
1597 | break; | |
7a9e9987 | 1598 | WARN_ON(!sblock->pagev[page_num]->page); |
b5d67f64 | 1599 | |
7a9e9987 | 1600 | mapped_buffer = kmap_atomic(sblock->pagev[page_num]->page); |
b5d67f64 SB |
1601 | } |
1602 | ||
1603 | btrfs_csum_final(crc, calculated_csum); | |
1604 | if (memcmp(calculated_csum, csum, csum_size)) | |
1605 | sblock->checksum_error = 1; | |
a2de733c AJ |
1606 | } |
1607 | ||
b5d67f64 | 1608 | static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad, |
114ab50d | 1609 | struct scrub_block *sblock_good) |
b5d67f64 SB |
1610 | { |
1611 | int page_num; | |
1612 | int ret = 0; | |
96e36920 | 1613 | |
b5d67f64 SB |
1614 | for (page_num = 0; page_num < sblock_bad->page_count; page_num++) { |
1615 | int ret_sub; | |
96e36920 | 1616 | |
b5d67f64 SB |
1617 | ret_sub = scrub_repair_page_from_good_copy(sblock_bad, |
1618 | sblock_good, | |
114ab50d | 1619 | page_num, 1); |
b5d67f64 SB |
1620 | if (ret_sub) |
1621 | ret = ret_sub; | |
a2de733c | 1622 | } |
b5d67f64 SB |
1623 | |
1624 | return ret; | |
1625 | } | |
1626 | ||
1627 | static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad, | |
1628 | struct scrub_block *sblock_good, | |
1629 | int page_num, int force_write) | |
1630 | { | |
7a9e9987 SB |
1631 | struct scrub_page *page_bad = sblock_bad->pagev[page_num]; |
1632 | struct scrub_page *page_good = sblock_good->pagev[page_num]; | |
b5d67f64 | 1633 | |
7a9e9987 SB |
1634 | BUG_ON(page_bad->page == NULL); |
1635 | BUG_ON(page_good->page == NULL); | |
b5d67f64 SB |
1636 | if (force_write || sblock_bad->header_error || |
1637 | sblock_bad->checksum_error || page_bad->io_error) { | |
1638 | struct bio *bio; | |
1639 | int ret; | |
b5d67f64 | 1640 | |
ff023aac | 1641 | if (!page_bad->dev->bdev) { |
efe120a0 FH |
1642 | printk_ratelimited(KERN_WARNING "BTRFS: " |
1643 | "scrub_repair_page_from_good_copy(bdev == NULL) " | |
1644 | "is unexpected!\n"); | |
ff023aac SB |
1645 | return -EIO; |
1646 | } | |
1647 | ||
9be3395b | 1648 | bio = btrfs_io_bio_alloc(GFP_NOFS, 1); |
e627ee7b TI |
1649 | if (!bio) |
1650 | return -EIO; | |
442a4f63 | 1651 | bio->bi_bdev = page_bad->dev->bdev; |
4f024f37 | 1652 | bio->bi_iter.bi_sector = page_bad->physical >> 9; |
b5d67f64 SB |
1653 | |
1654 | ret = bio_add_page(bio, page_good->page, PAGE_SIZE, 0); | |
1655 | if (PAGE_SIZE != ret) { | |
1656 | bio_put(bio); | |
1657 | return -EIO; | |
13db62b7 | 1658 | } |
b5d67f64 | 1659 | |
33879d45 | 1660 | if (btrfsic_submit_bio_wait(WRITE, bio)) { |
442a4f63 SB |
1661 | btrfs_dev_stat_inc_and_print(page_bad->dev, |
1662 | BTRFS_DEV_STAT_WRITE_ERRS); | |
ff023aac SB |
1663 | btrfs_dev_replace_stats_inc( |
1664 | &sblock_bad->sctx->dev_root->fs_info-> | |
1665 | dev_replace.num_write_errors); | |
442a4f63 SB |
1666 | bio_put(bio); |
1667 | return -EIO; | |
1668 | } | |
b5d67f64 | 1669 | bio_put(bio); |
a2de733c AJ |
1670 | } |
1671 | ||
b5d67f64 SB |
1672 | return 0; |
1673 | } | |
1674 | ||
ff023aac SB |
1675 | static void scrub_write_block_to_dev_replace(struct scrub_block *sblock) |
1676 | { | |
1677 | int page_num; | |
1678 | ||
5a6ac9ea MX |
1679 | /* |
1680 | * This block is used for the check of the parity on the source device, | |
1681 | * so the data needn't be written into the destination device. | |
1682 | */ | |
1683 | if (sblock->sparity) | |
1684 | return; | |
1685 | ||
ff023aac SB |
1686 | for (page_num = 0; page_num < sblock->page_count; page_num++) { |
1687 | int ret; | |
1688 | ||
1689 | ret = scrub_write_page_to_dev_replace(sblock, page_num); | |
1690 | if (ret) | |
1691 | btrfs_dev_replace_stats_inc( | |
1692 | &sblock->sctx->dev_root->fs_info->dev_replace. | |
1693 | num_write_errors); | |
1694 | } | |
1695 | } | |
1696 | ||
1697 | static int scrub_write_page_to_dev_replace(struct scrub_block *sblock, | |
1698 | int page_num) | |
1699 | { | |
1700 | struct scrub_page *spage = sblock->pagev[page_num]; | |
1701 | ||
1702 | BUG_ON(spage->page == NULL); | |
1703 | if (spage->io_error) { | |
1704 | void *mapped_buffer = kmap_atomic(spage->page); | |
1705 | ||
1706 | memset(mapped_buffer, 0, PAGE_CACHE_SIZE); | |
1707 | flush_dcache_page(spage->page); | |
1708 | kunmap_atomic(mapped_buffer); | |
1709 | } | |
1710 | return scrub_add_page_to_wr_bio(sblock->sctx, spage); | |
1711 | } | |
1712 | ||
1713 | static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx, | |
1714 | struct scrub_page *spage) | |
1715 | { | |
1716 | struct scrub_wr_ctx *wr_ctx = &sctx->wr_ctx; | |
1717 | struct scrub_bio *sbio; | |
1718 | int ret; | |
1719 | ||
1720 | mutex_lock(&wr_ctx->wr_lock); | |
1721 | again: | |
1722 | if (!wr_ctx->wr_curr_bio) { | |
1723 | wr_ctx->wr_curr_bio = kzalloc(sizeof(*wr_ctx->wr_curr_bio), | |
1724 | GFP_NOFS); | |
1725 | if (!wr_ctx->wr_curr_bio) { | |
1726 | mutex_unlock(&wr_ctx->wr_lock); | |
1727 | return -ENOMEM; | |
1728 | } | |
1729 | wr_ctx->wr_curr_bio->sctx = sctx; | |
1730 | wr_ctx->wr_curr_bio->page_count = 0; | |
1731 | } | |
1732 | sbio = wr_ctx->wr_curr_bio; | |
1733 | if (sbio->page_count == 0) { | |
1734 | struct bio *bio; | |
1735 | ||
1736 | sbio->physical = spage->physical_for_dev_replace; | |
1737 | sbio->logical = spage->logical; | |
1738 | sbio->dev = wr_ctx->tgtdev; | |
1739 | bio = sbio->bio; | |
1740 | if (!bio) { | |
9be3395b | 1741 | bio = btrfs_io_bio_alloc(GFP_NOFS, wr_ctx->pages_per_wr_bio); |
ff023aac SB |
1742 | if (!bio) { |
1743 | mutex_unlock(&wr_ctx->wr_lock); | |
1744 | return -ENOMEM; | |
1745 | } | |
1746 | sbio->bio = bio; | |
1747 | } | |
1748 | ||
1749 | bio->bi_private = sbio; | |
1750 | bio->bi_end_io = scrub_wr_bio_end_io; | |
1751 | bio->bi_bdev = sbio->dev->bdev; | |
4f024f37 | 1752 | bio->bi_iter.bi_sector = sbio->physical >> 9; |
ff023aac SB |
1753 | sbio->err = 0; |
1754 | } else if (sbio->physical + sbio->page_count * PAGE_SIZE != | |
1755 | spage->physical_for_dev_replace || | |
1756 | sbio->logical + sbio->page_count * PAGE_SIZE != | |
1757 | spage->logical) { | |
1758 | scrub_wr_submit(sctx); | |
1759 | goto again; | |
1760 | } | |
1761 | ||
1762 | ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0); | |
1763 | if (ret != PAGE_SIZE) { | |
1764 | if (sbio->page_count < 1) { | |
1765 | bio_put(sbio->bio); | |
1766 | sbio->bio = NULL; | |
1767 | mutex_unlock(&wr_ctx->wr_lock); | |
1768 | return -EIO; | |
1769 | } | |
1770 | scrub_wr_submit(sctx); | |
1771 | goto again; | |
1772 | } | |
1773 | ||
1774 | sbio->pagev[sbio->page_count] = spage; | |
1775 | scrub_page_get(spage); | |
1776 | sbio->page_count++; | |
1777 | if (sbio->page_count == wr_ctx->pages_per_wr_bio) | |
1778 | scrub_wr_submit(sctx); | |
1779 | mutex_unlock(&wr_ctx->wr_lock); | |
1780 | ||
1781 | return 0; | |
1782 | } | |
1783 | ||
1784 | static void scrub_wr_submit(struct scrub_ctx *sctx) | |
1785 | { | |
1786 | struct scrub_wr_ctx *wr_ctx = &sctx->wr_ctx; | |
1787 | struct scrub_bio *sbio; | |
1788 | ||
1789 | if (!wr_ctx->wr_curr_bio) | |
1790 | return; | |
1791 | ||
1792 | sbio = wr_ctx->wr_curr_bio; | |
1793 | wr_ctx->wr_curr_bio = NULL; | |
1794 | WARN_ON(!sbio->bio->bi_bdev); | |
1795 | scrub_pending_bio_inc(sctx); | |
1796 | /* process all writes in a single worker thread. Then the block layer | |
1797 | * orders the requests before sending them to the driver which | |
1798 | * doubled the write performance on spinning disks when measured | |
1799 | * with Linux 3.5 */ | |
1800 | btrfsic_submit_bio(WRITE, sbio->bio); | |
1801 | } | |
1802 | ||
1803 | static void scrub_wr_bio_end_io(struct bio *bio, int err) | |
1804 | { | |
1805 | struct scrub_bio *sbio = bio->bi_private; | |
1806 | struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info; | |
1807 | ||
1808 | sbio->err = err; | |
1809 | sbio->bio = bio; | |
1810 | ||
9e0af237 LB |
1811 | btrfs_init_work(&sbio->work, btrfs_scrubwrc_helper, |
1812 | scrub_wr_bio_end_io_worker, NULL, NULL); | |
0339ef2f | 1813 | btrfs_queue_work(fs_info->scrub_wr_completion_workers, &sbio->work); |
ff023aac SB |
1814 | } |
1815 | ||
1816 | static void scrub_wr_bio_end_io_worker(struct btrfs_work *work) | |
1817 | { | |
1818 | struct scrub_bio *sbio = container_of(work, struct scrub_bio, work); | |
1819 | struct scrub_ctx *sctx = sbio->sctx; | |
1820 | int i; | |
1821 | ||
1822 | WARN_ON(sbio->page_count > SCRUB_PAGES_PER_WR_BIO); | |
1823 | if (sbio->err) { | |
1824 | struct btrfs_dev_replace *dev_replace = | |
1825 | &sbio->sctx->dev_root->fs_info->dev_replace; | |
1826 | ||
1827 | for (i = 0; i < sbio->page_count; i++) { | |
1828 | struct scrub_page *spage = sbio->pagev[i]; | |
1829 | ||
1830 | spage->io_error = 1; | |
1831 | btrfs_dev_replace_stats_inc(&dev_replace-> | |
1832 | num_write_errors); | |
1833 | } | |
1834 | } | |
1835 | ||
1836 | for (i = 0; i < sbio->page_count; i++) | |
1837 | scrub_page_put(sbio->pagev[i]); | |
1838 | ||
1839 | bio_put(sbio->bio); | |
1840 | kfree(sbio); | |
1841 | scrub_pending_bio_dec(sctx); | |
1842 | } | |
1843 | ||
1844 | static int scrub_checksum(struct scrub_block *sblock) | |
b5d67f64 SB |
1845 | { |
1846 | u64 flags; | |
1847 | int ret; | |
1848 | ||
7a9e9987 SB |
1849 | WARN_ON(sblock->page_count < 1); |
1850 | flags = sblock->pagev[0]->flags; | |
b5d67f64 SB |
1851 | ret = 0; |
1852 | if (flags & BTRFS_EXTENT_FLAG_DATA) | |
1853 | ret = scrub_checksum_data(sblock); | |
1854 | else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) | |
1855 | ret = scrub_checksum_tree_block(sblock); | |
1856 | else if (flags & BTRFS_EXTENT_FLAG_SUPER) | |
1857 | (void)scrub_checksum_super(sblock); | |
1858 | else | |
1859 | WARN_ON(1); | |
1860 | if (ret) | |
1861 | scrub_handle_errored_block(sblock); | |
ff023aac SB |
1862 | |
1863 | return ret; | |
a2de733c AJ |
1864 | } |
1865 | ||
b5d67f64 | 1866 | static int scrub_checksum_data(struct scrub_block *sblock) |
a2de733c | 1867 | { |
d9d181c1 | 1868 | struct scrub_ctx *sctx = sblock->sctx; |
a2de733c | 1869 | u8 csum[BTRFS_CSUM_SIZE]; |
b5d67f64 SB |
1870 | u8 *on_disk_csum; |
1871 | struct page *page; | |
1872 | void *buffer; | |
a2de733c AJ |
1873 | u32 crc = ~(u32)0; |
1874 | int fail = 0; | |
b5d67f64 SB |
1875 | u64 len; |
1876 | int index; | |
a2de733c | 1877 | |
b5d67f64 | 1878 | BUG_ON(sblock->page_count < 1); |
7a9e9987 | 1879 | if (!sblock->pagev[0]->have_csum) |
a2de733c AJ |
1880 | return 0; |
1881 | ||
7a9e9987 SB |
1882 | on_disk_csum = sblock->pagev[0]->csum; |
1883 | page = sblock->pagev[0]->page; | |
9613bebb | 1884 | buffer = kmap_atomic(page); |
b5d67f64 | 1885 | |
d9d181c1 | 1886 | len = sctx->sectorsize; |
b5d67f64 SB |
1887 | index = 0; |
1888 | for (;;) { | |
1889 | u64 l = min_t(u64, len, PAGE_SIZE); | |
1890 | ||
b0496686 | 1891 | crc = btrfs_csum_data(buffer, crc, l); |
9613bebb | 1892 | kunmap_atomic(buffer); |
b5d67f64 SB |
1893 | len -= l; |
1894 | if (len == 0) | |
1895 | break; | |
1896 | index++; | |
1897 | BUG_ON(index >= sblock->page_count); | |
7a9e9987 SB |
1898 | BUG_ON(!sblock->pagev[index]->page); |
1899 | page = sblock->pagev[index]->page; | |
9613bebb | 1900 | buffer = kmap_atomic(page); |
b5d67f64 SB |
1901 | } |
1902 | ||
a2de733c | 1903 | btrfs_csum_final(crc, csum); |
d9d181c1 | 1904 | if (memcmp(csum, on_disk_csum, sctx->csum_size)) |
a2de733c AJ |
1905 | fail = 1; |
1906 | ||
a2de733c AJ |
1907 | return fail; |
1908 | } | |
1909 | ||
b5d67f64 | 1910 | static int scrub_checksum_tree_block(struct scrub_block *sblock) |
a2de733c | 1911 | { |
d9d181c1 | 1912 | struct scrub_ctx *sctx = sblock->sctx; |
a2de733c | 1913 | struct btrfs_header *h; |
a36cf8b8 | 1914 | struct btrfs_root *root = sctx->dev_root; |
a2de733c | 1915 | struct btrfs_fs_info *fs_info = root->fs_info; |
b5d67f64 SB |
1916 | u8 calculated_csum[BTRFS_CSUM_SIZE]; |
1917 | u8 on_disk_csum[BTRFS_CSUM_SIZE]; | |
1918 | struct page *page; | |
1919 | void *mapped_buffer; | |
1920 | u64 mapped_size; | |
1921 | void *p; | |
a2de733c AJ |
1922 | u32 crc = ~(u32)0; |
1923 | int fail = 0; | |
1924 | int crc_fail = 0; | |
b5d67f64 SB |
1925 | u64 len; |
1926 | int index; | |
1927 | ||
1928 | BUG_ON(sblock->page_count < 1); | |
7a9e9987 | 1929 | page = sblock->pagev[0]->page; |
9613bebb | 1930 | mapped_buffer = kmap_atomic(page); |
b5d67f64 | 1931 | h = (struct btrfs_header *)mapped_buffer; |
d9d181c1 | 1932 | memcpy(on_disk_csum, h->csum, sctx->csum_size); |
a2de733c AJ |
1933 | |
1934 | /* | |
1935 | * we don't use the getter functions here, as we | |
1936 | * a) don't have an extent buffer and | |
1937 | * b) the page is already kmapped | |
1938 | */ | |
a2de733c | 1939 | |
3cae210f | 1940 | if (sblock->pagev[0]->logical != btrfs_stack_header_bytenr(h)) |
a2de733c AJ |
1941 | ++fail; |
1942 | ||
3cae210f | 1943 | if (sblock->pagev[0]->generation != btrfs_stack_header_generation(h)) |
a2de733c AJ |
1944 | ++fail; |
1945 | ||
17a9be2f | 1946 | if (!scrub_check_fsid(h->fsid, sblock->pagev[0])) |
a2de733c AJ |
1947 | ++fail; |
1948 | ||
1949 | if (memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid, | |
1950 | BTRFS_UUID_SIZE)) | |
1951 | ++fail; | |
1952 | ||
d9d181c1 | 1953 | len = sctx->nodesize - BTRFS_CSUM_SIZE; |
b5d67f64 SB |
1954 | mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE; |
1955 | p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE; | |
1956 | index = 0; | |
1957 | for (;;) { | |
1958 | u64 l = min_t(u64, len, mapped_size); | |
1959 | ||
b0496686 | 1960 | crc = btrfs_csum_data(p, crc, l); |
9613bebb | 1961 | kunmap_atomic(mapped_buffer); |
b5d67f64 SB |
1962 | len -= l; |
1963 | if (len == 0) | |
1964 | break; | |
1965 | index++; | |
1966 | BUG_ON(index >= sblock->page_count); | |
7a9e9987 SB |
1967 | BUG_ON(!sblock->pagev[index]->page); |
1968 | page = sblock->pagev[index]->page; | |
9613bebb | 1969 | mapped_buffer = kmap_atomic(page); |
b5d67f64 SB |
1970 | mapped_size = PAGE_SIZE; |
1971 | p = mapped_buffer; | |
1972 | } | |
1973 | ||
1974 | btrfs_csum_final(crc, calculated_csum); | |
d9d181c1 | 1975 | if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size)) |
a2de733c AJ |
1976 | ++crc_fail; |
1977 | ||
a2de733c AJ |
1978 | return fail || crc_fail; |
1979 | } | |
1980 | ||
b5d67f64 | 1981 | static int scrub_checksum_super(struct scrub_block *sblock) |
a2de733c AJ |
1982 | { |
1983 | struct btrfs_super_block *s; | |
d9d181c1 | 1984 | struct scrub_ctx *sctx = sblock->sctx; |
b5d67f64 SB |
1985 | u8 calculated_csum[BTRFS_CSUM_SIZE]; |
1986 | u8 on_disk_csum[BTRFS_CSUM_SIZE]; | |
1987 | struct page *page; | |
1988 | void *mapped_buffer; | |
1989 | u64 mapped_size; | |
1990 | void *p; | |
a2de733c | 1991 | u32 crc = ~(u32)0; |
442a4f63 SB |
1992 | int fail_gen = 0; |
1993 | int fail_cor = 0; | |
b5d67f64 SB |
1994 | u64 len; |
1995 | int index; | |
a2de733c | 1996 | |
b5d67f64 | 1997 | BUG_ON(sblock->page_count < 1); |
7a9e9987 | 1998 | page = sblock->pagev[0]->page; |
9613bebb | 1999 | mapped_buffer = kmap_atomic(page); |
b5d67f64 | 2000 | s = (struct btrfs_super_block *)mapped_buffer; |
d9d181c1 | 2001 | memcpy(on_disk_csum, s->csum, sctx->csum_size); |
a2de733c | 2002 | |
3cae210f | 2003 | if (sblock->pagev[0]->logical != btrfs_super_bytenr(s)) |
442a4f63 | 2004 | ++fail_cor; |
a2de733c | 2005 | |
3cae210f | 2006 | if (sblock->pagev[0]->generation != btrfs_super_generation(s)) |
442a4f63 | 2007 | ++fail_gen; |
a2de733c | 2008 | |
17a9be2f | 2009 | if (!scrub_check_fsid(s->fsid, sblock->pagev[0])) |
442a4f63 | 2010 | ++fail_cor; |
a2de733c | 2011 | |
b5d67f64 SB |
2012 | len = BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE; |
2013 | mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE; | |
2014 | p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE; | |
2015 | index = 0; | |
2016 | for (;;) { | |
2017 | u64 l = min_t(u64, len, mapped_size); | |
2018 | ||
b0496686 | 2019 | crc = btrfs_csum_data(p, crc, l); |
9613bebb | 2020 | kunmap_atomic(mapped_buffer); |
b5d67f64 SB |
2021 | len -= l; |
2022 | if (len == 0) | |
2023 | break; | |
2024 | index++; | |
2025 | BUG_ON(index >= sblock->page_count); | |
7a9e9987 SB |
2026 | BUG_ON(!sblock->pagev[index]->page); |
2027 | page = sblock->pagev[index]->page; | |
9613bebb | 2028 | mapped_buffer = kmap_atomic(page); |
b5d67f64 SB |
2029 | mapped_size = PAGE_SIZE; |
2030 | p = mapped_buffer; | |
2031 | } | |
2032 | ||
2033 | btrfs_csum_final(crc, calculated_csum); | |
d9d181c1 | 2034 | if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size)) |
442a4f63 | 2035 | ++fail_cor; |
a2de733c | 2036 | |
442a4f63 | 2037 | if (fail_cor + fail_gen) { |
a2de733c AJ |
2038 | /* |
2039 | * if we find an error in a super block, we just report it. | |
2040 | * They will get written with the next transaction commit | |
2041 | * anyway | |
2042 | */ | |
d9d181c1 SB |
2043 | spin_lock(&sctx->stat_lock); |
2044 | ++sctx->stat.super_errors; | |
2045 | spin_unlock(&sctx->stat_lock); | |
442a4f63 | 2046 | if (fail_cor) |
7a9e9987 | 2047 | btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev, |
442a4f63 SB |
2048 | BTRFS_DEV_STAT_CORRUPTION_ERRS); |
2049 | else | |
7a9e9987 | 2050 | btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev, |
442a4f63 | 2051 | BTRFS_DEV_STAT_GENERATION_ERRS); |
a2de733c AJ |
2052 | } |
2053 | ||
442a4f63 | 2054 | return fail_cor + fail_gen; |
a2de733c AJ |
2055 | } |
2056 | ||
b5d67f64 SB |
2057 | static void scrub_block_get(struct scrub_block *sblock) |
2058 | { | |
2059 | atomic_inc(&sblock->ref_count); | |
2060 | } | |
2061 | ||
2062 | static void scrub_block_put(struct scrub_block *sblock) | |
2063 | { | |
2064 | if (atomic_dec_and_test(&sblock->ref_count)) { | |
2065 | int i; | |
2066 | ||
5a6ac9ea MX |
2067 | if (sblock->sparity) |
2068 | scrub_parity_put(sblock->sparity); | |
2069 | ||
b5d67f64 | 2070 | for (i = 0; i < sblock->page_count; i++) |
7a9e9987 | 2071 | scrub_page_put(sblock->pagev[i]); |
b5d67f64 SB |
2072 | kfree(sblock); |
2073 | } | |
2074 | } | |
2075 | ||
7a9e9987 SB |
2076 | static void scrub_page_get(struct scrub_page *spage) |
2077 | { | |
2078 | atomic_inc(&spage->ref_count); | |
2079 | } | |
2080 | ||
2081 | static void scrub_page_put(struct scrub_page *spage) | |
2082 | { | |
2083 | if (atomic_dec_and_test(&spage->ref_count)) { | |
2084 | if (spage->page) | |
2085 | __free_page(spage->page); | |
2086 | kfree(spage); | |
2087 | } | |
2088 | } | |
2089 | ||
d9d181c1 | 2090 | static void scrub_submit(struct scrub_ctx *sctx) |
a2de733c AJ |
2091 | { |
2092 | struct scrub_bio *sbio; | |
2093 | ||
d9d181c1 | 2094 | if (sctx->curr == -1) |
1623edeb | 2095 | return; |
a2de733c | 2096 | |
d9d181c1 SB |
2097 | sbio = sctx->bios[sctx->curr]; |
2098 | sctx->curr = -1; | |
b6bfebc1 | 2099 | scrub_pending_bio_inc(sctx); |
a2de733c | 2100 | |
ff023aac SB |
2101 | if (!sbio->bio->bi_bdev) { |
2102 | /* | |
2103 | * this case should not happen. If btrfs_map_block() is | |
2104 | * wrong, it could happen for dev-replace operations on | |
2105 | * missing devices when no mirrors are available, but in | |
2106 | * this case it should already fail the mount. | |
2107 | * This case is handled correctly (but _very_ slowly). | |
2108 | */ | |
2109 | printk_ratelimited(KERN_WARNING | |
efe120a0 | 2110 | "BTRFS: scrub_submit(bio bdev == NULL) is unexpected!\n"); |
ff023aac SB |
2111 | bio_endio(sbio->bio, -EIO); |
2112 | } else { | |
2113 | btrfsic_submit_bio(READ, sbio->bio); | |
2114 | } | |
a2de733c AJ |
2115 | } |
2116 | ||
ff023aac SB |
2117 | static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx, |
2118 | struct scrub_page *spage) | |
a2de733c | 2119 | { |
b5d67f64 | 2120 | struct scrub_block *sblock = spage->sblock; |
a2de733c | 2121 | struct scrub_bio *sbio; |
69f4cb52 | 2122 | int ret; |
a2de733c AJ |
2123 | |
2124 | again: | |
2125 | /* | |
2126 | * grab a fresh bio or wait for one to become available | |
2127 | */ | |
d9d181c1 SB |
2128 | while (sctx->curr == -1) { |
2129 | spin_lock(&sctx->list_lock); | |
2130 | sctx->curr = sctx->first_free; | |
2131 | if (sctx->curr != -1) { | |
2132 | sctx->first_free = sctx->bios[sctx->curr]->next_free; | |
2133 | sctx->bios[sctx->curr]->next_free = -1; | |
2134 | sctx->bios[sctx->curr]->page_count = 0; | |
2135 | spin_unlock(&sctx->list_lock); | |
a2de733c | 2136 | } else { |
d9d181c1 SB |
2137 | spin_unlock(&sctx->list_lock); |
2138 | wait_event(sctx->list_wait, sctx->first_free != -1); | |
a2de733c AJ |
2139 | } |
2140 | } | |
d9d181c1 | 2141 | sbio = sctx->bios[sctx->curr]; |
b5d67f64 | 2142 | if (sbio->page_count == 0) { |
69f4cb52 AJ |
2143 | struct bio *bio; |
2144 | ||
b5d67f64 SB |
2145 | sbio->physical = spage->physical; |
2146 | sbio->logical = spage->logical; | |
a36cf8b8 | 2147 | sbio->dev = spage->dev; |
b5d67f64 SB |
2148 | bio = sbio->bio; |
2149 | if (!bio) { | |
9be3395b | 2150 | bio = btrfs_io_bio_alloc(GFP_NOFS, sctx->pages_per_rd_bio); |
b5d67f64 SB |
2151 | if (!bio) |
2152 | return -ENOMEM; | |
2153 | sbio->bio = bio; | |
2154 | } | |
69f4cb52 AJ |
2155 | |
2156 | bio->bi_private = sbio; | |
2157 | bio->bi_end_io = scrub_bio_end_io; | |
a36cf8b8 | 2158 | bio->bi_bdev = sbio->dev->bdev; |
4f024f37 | 2159 | bio->bi_iter.bi_sector = sbio->physical >> 9; |
69f4cb52 | 2160 | sbio->err = 0; |
b5d67f64 SB |
2161 | } else if (sbio->physical + sbio->page_count * PAGE_SIZE != |
2162 | spage->physical || | |
2163 | sbio->logical + sbio->page_count * PAGE_SIZE != | |
a36cf8b8 SB |
2164 | spage->logical || |
2165 | sbio->dev != spage->dev) { | |
d9d181c1 | 2166 | scrub_submit(sctx); |
a2de733c AJ |
2167 | goto again; |
2168 | } | |
69f4cb52 | 2169 | |
b5d67f64 SB |
2170 | sbio->pagev[sbio->page_count] = spage; |
2171 | ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0); | |
2172 | if (ret != PAGE_SIZE) { | |
2173 | if (sbio->page_count < 1) { | |
2174 | bio_put(sbio->bio); | |
2175 | sbio->bio = NULL; | |
2176 | return -EIO; | |
2177 | } | |
d9d181c1 | 2178 | scrub_submit(sctx); |
69f4cb52 AJ |
2179 | goto again; |
2180 | } | |
2181 | ||
ff023aac | 2182 | scrub_block_get(sblock); /* one for the page added to the bio */ |
b5d67f64 SB |
2183 | atomic_inc(&sblock->outstanding_pages); |
2184 | sbio->page_count++; | |
ff023aac | 2185 | if (sbio->page_count == sctx->pages_per_rd_bio) |
d9d181c1 | 2186 | scrub_submit(sctx); |
b5d67f64 SB |
2187 | |
2188 | return 0; | |
2189 | } | |
2190 | ||
d9d181c1 | 2191 | static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len, |
a36cf8b8 | 2192 | u64 physical, struct btrfs_device *dev, u64 flags, |
ff023aac SB |
2193 | u64 gen, int mirror_num, u8 *csum, int force, |
2194 | u64 physical_for_dev_replace) | |
b5d67f64 SB |
2195 | { |
2196 | struct scrub_block *sblock; | |
2197 | int index; | |
2198 | ||
2199 | sblock = kzalloc(sizeof(*sblock), GFP_NOFS); | |
2200 | if (!sblock) { | |
d9d181c1 SB |
2201 | spin_lock(&sctx->stat_lock); |
2202 | sctx->stat.malloc_errors++; | |
2203 | spin_unlock(&sctx->stat_lock); | |
b5d67f64 | 2204 | return -ENOMEM; |
a2de733c | 2205 | } |
b5d67f64 | 2206 | |
7a9e9987 SB |
2207 | /* one ref inside this function, plus one for each page added to |
2208 | * a bio later on */ | |
b5d67f64 | 2209 | atomic_set(&sblock->ref_count, 1); |
d9d181c1 | 2210 | sblock->sctx = sctx; |
b5d67f64 SB |
2211 | sblock->no_io_error_seen = 1; |
2212 | ||
2213 | for (index = 0; len > 0; index++) { | |
7a9e9987 | 2214 | struct scrub_page *spage; |
b5d67f64 SB |
2215 | u64 l = min_t(u64, len, PAGE_SIZE); |
2216 | ||
7a9e9987 SB |
2217 | spage = kzalloc(sizeof(*spage), GFP_NOFS); |
2218 | if (!spage) { | |
2219 | leave_nomem: | |
d9d181c1 SB |
2220 | spin_lock(&sctx->stat_lock); |
2221 | sctx->stat.malloc_errors++; | |
2222 | spin_unlock(&sctx->stat_lock); | |
7a9e9987 | 2223 | scrub_block_put(sblock); |
b5d67f64 SB |
2224 | return -ENOMEM; |
2225 | } | |
7a9e9987 SB |
2226 | BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK); |
2227 | scrub_page_get(spage); | |
2228 | sblock->pagev[index] = spage; | |
b5d67f64 | 2229 | spage->sblock = sblock; |
a36cf8b8 | 2230 | spage->dev = dev; |
b5d67f64 SB |
2231 | spage->flags = flags; |
2232 | spage->generation = gen; | |
2233 | spage->logical = logical; | |
2234 | spage->physical = physical; | |
ff023aac | 2235 | spage->physical_for_dev_replace = physical_for_dev_replace; |
b5d67f64 SB |
2236 | spage->mirror_num = mirror_num; |
2237 | if (csum) { | |
2238 | spage->have_csum = 1; | |
d9d181c1 | 2239 | memcpy(spage->csum, csum, sctx->csum_size); |
b5d67f64 SB |
2240 | } else { |
2241 | spage->have_csum = 0; | |
2242 | } | |
2243 | sblock->page_count++; | |
7a9e9987 SB |
2244 | spage->page = alloc_page(GFP_NOFS); |
2245 | if (!spage->page) | |
2246 | goto leave_nomem; | |
b5d67f64 SB |
2247 | len -= l; |
2248 | logical += l; | |
2249 | physical += l; | |
ff023aac | 2250 | physical_for_dev_replace += l; |
b5d67f64 SB |
2251 | } |
2252 | ||
7a9e9987 | 2253 | WARN_ON(sblock->page_count == 0); |
b5d67f64 | 2254 | for (index = 0; index < sblock->page_count; index++) { |
7a9e9987 | 2255 | struct scrub_page *spage = sblock->pagev[index]; |
1bc87793 AJ |
2256 | int ret; |
2257 | ||
ff023aac | 2258 | ret = scrub_add_page_to_rd_bio(sctx, spage); |
b5d67f64 SB |
2259 | if (ret) { |
2260 | scrub_block_put(sblock); | |
1bc87793 | 2261 | return ret; |
b5d67f64 | 2262 | } |
1bc87793 | 2263 | } |
a2de733c | 2264 | |
b5d67f64 | 2265 | if (force) |
d9d181c1 | 2266 | scrub_submit(sctx); |
a2de733c | 2267 | |
b5d67f64 SB |
2268 | /* last one frees, either here or in bio completion for last page */ |
2269 | scrub_block_put(sblock); | |
a2de733c AJ |
2270 | return 0; |
2271 | } | |
2272 | ||
b5d67f64 SB |
2273 | static void scrub_bio_end_io(struct bio *bio, int err) |
2274 | { | |
2275 | struct scrub_bio *sbio = bio->bi_private; | |
a36cf8b8 | 2276 | struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info; |
b5d67f64 SB |
2277 | |
2278 | sbio->err = err; | |
2279 | sbio->bio = bio; | |
2280 | ||
0339ef2f | 2281 | btrfs_queue_work(fs_info->scrub_workers, &sbio->work); |
b5d67f64 SB |
2282 | } |
2283 | ||
2284 | static void scrub_bio_end_io_worker(struct btrfs_work *work) | |
2285 | { | |
2286 | struct scrub_bio *sbio = container_of(work, struct scrub_bio, work); | |
d9d181c1 | 2287 | struct scrub_ctx *sctx = sbio->sctx; |
b5d67f64 SB |
2288 | int i; |
2289 | ||
ff023aac | 2290 | BUG_ON(sbio->page_count > SCRUB_PAGES_PER_RD_BIO); |
b5d67f64 SB |
2291 | if (sbio->err) { |
2292 | for (i = 0; i < sbio->page_count; i++) { | |
2293 | struct scrub_page *spage = sbio->pagev[i]; | |
2294 | ||
2295 | spage->io_error = 1; | |
2296 | spage->sblock->no_io_error_seen = 0; | |
2297 | } | |
2298 | } | |
2299 | ||
2300 | /* now complete the scrub_block items that have all pages completed */ | |
2301 | for (i = 0; i < sbio->page_count; i++) { | |
2302 | struct scrub_page *spage = sbio->pagev[i]; | |
2303 | struct scrub_block *sblock = spage->sblock; | |
2304 | ||
2305 | if (atomic_dec_and_test(&sblock->outstanding_pages)) | |
2306 | scrub_block_complete(sblock); | |
2307 | scrub_block_put(sblock); | |
2308 | } | |
2309 | ||
b5d67f64 SB |
2310 | bio_put(sbio->bio); |
2311 | sbio->bio = NULL; | |
d9d181c1 SB |
2312 | spin_lock(&sctx->list_lock); |
2313 | sbio->next_free = sctx->first_free; | |
2314 | sctx->first_free = sbio->index; | |
2315 | spin_unlock(&sctx->list_lock); | |
ff023aac SB |
2316 | |
2317 | if (sctx->is_dev_replace && | |
2318 | atomic_read(&sctx->wr_ctx.flush_all_writes)) { | |
2319 | mutex_lock(&sctx->wr_ctx.wr_lock); | |
2320 | scrub_wr_submit(sctx); | |
2321 | mutex_unlock(&sctx->wr_ctx.wr_lock); | |
2322 | } | |
2323 | ||
b6bfebc1 | 2324 | scrub_pending_bio_dec(sctx); |
b5d67f64 SB |
2325 | } |
2326 | ||
5a6ac9ea MX |
2327 | static inline void __scrub_mark_bitmap(struct scrub_parity *sparity, |
2328 | unsigned long *bitmap, | |
2329 | u64 start, u64 len) | |
2330 | { | |
2331 | int offset; | |
2332 | int nsectors; | |
2333 | int sectorsize = sparity->sctx->dev_root->sectorsize; | |
2334 | ||
2335 | if (len >= sparity->stripe_len) { | |
2336 | bitmap_set(bitmap, 0, sparity->nsectors); | |
2337 | return; | |
2338 | } | |
2339 | ||
2340 | start -= sparity->logic_start; | |
2341 | offset = (int)do_div(start, sparity->stripe_len); | |
2342 | offset /= sectorsize; | |
2343 | nsectors = (int)len / sectorsize; | |
2344 | ||
2345 | if (offset + nsectors <= sparity->nsectors) { | |
2346 | bitmap_set(bitmap, offset, nsectors); | |
2347 | return; | |
2348 | } | |
2349 | ||
2350 | bitmap_set(bitmap, offset, sparity->nsectors - offset); | |
2351 | bitmap_set(bitmap, 0, nsectors - (sparity->nsectors - offset)); | |
2352 | } | |
2353 | ||
2354 | static inline void scrub_parity_mark_sectors_error(struct scrub_parity *sparity, | |
2355 | u64 start, u64 len) | |
2356 | { | |
2357 | __scrub_mark_bitmap(sparity, sparity->ebitmap, start, len); | |
2358 | } | |
2359 | ||
2360 | static inline void scrub_parity_mark_sectors_data(struct scrub_parity *sparity, | |
2361 | u64 start, u64 len) | |
2362 | { | |
2363 | __scrub_mark_bitmap(sparity, sparity->dbitmap, start, len); | |
2364 | } | |
2365 | ||
b5d67f64 SB |
2366 | static void scrub_block_complete(struct scrub_block *sblock) |
2367 | { | |
5a6ac9ea MX |
2368 | int corrupted = 0; |
2369 | ||
ff023aac | 2370 | if (!sblock->no_io_error_seen) { |
5a6ac9ea | 2371 | corrupted = 1; |
b5d67f64 | 2372 | scrub_handle_errored_block(sblock); |
ff023aac SB |
2373 | } else { |
2374 | /* | |
2375 | * if has checksum error, write via repair mechanism in | |
2376 | * dev replace case, otherwise write here in dev replace | |
2377 | * case. | |
2378 | */ | |
5a6ac9ea MX |
2379 | corrupted = scrub_checksum(sblock); |
2380 | if (!corrupted && sblock->sctx->is_dev_replace) | |
ff023aac SB |
2381 | scrub_write_block_to_dev_replace(sblock); |
2382 | } | |
5a6ac9ea MX |
2383 | |
2384 | if (sblock->sparity && corrupted && !sblock->data_corrected) { | |
2385 | u64 start = sblock->pagev[0]->logical; | |
2386 | u64 end = sblock->pagev[sblock->page_count - 1]->logical + | |
2387 | PAGE_SIZE; | |
2388 | ||
2389 | scrub_parity_mark_sectors_error(sblock->sparity, | |
2390 | start, end - start); | |
2391 | } | |
b5d67f64 SB |
2392 | } |
2393 | ||
d9d181c1 | 2394 | static int scrub_find_csum(struct scrub_ctx *sctx, u64 logical, u64 len, |
a2de733c AJ |
2395 | u8 *csum) |
2396 | { | |
2397 | struct btrfs_ordered_sum *sum = NULL; | |
f51a4a18 | 2398 | unsigned long index; |
a2de733c | 2399 | unsigned long num_sectors; |
a2de733c | 2400 | |
d9d181c1 SB |
2401 | while (!list_empty(&sctx->csum_list)) { |
2402 | sum = list_first_entry(&sctx->csum_list, | |
a2de733c AJ |
2403 | struct btrfs_ordered_sum, list); |
2404 | if (sum->bytenr > logical) | |
2405 | return 0; | |
2406 | if (sum->bytenr + sum->len > logical) | |
2407 | break; | |
2408 | ||
d9d181c1 | 2409 | ++sctx->stat.csum_discards; |
a2de733c AJ |
2410 | list_del(&sum->list); |
2411 | kfree(sum); | |
2412 | sum = NULL; | |
2413 | } | |
2414 | if (!sum) | |
2415 | return 0; | |
2416 | ||
f51a4a18 | 2417 | index = ((u32)(logical - sum->bytenr)) / sctx->sectorsize; |
d9d181c1 | 2418 | num_sectors = sum->len / sctx->sectorsize; |
f51a4a18 MX |
2419 | memcpy(csum, sum->sums + index, sctx->csum_size); |
2420 | if (index == num_sectors - 1) { | |
a2de733c AJ |
2421 | list_del(&sum->list); |
2422 | kfree(sum); | |
2423 | } | |
f51a4a18 | 2424 | return 1; |
a2de733c AJ |
2425 | } |
2426 | ||
2427 | /* scrub extent tries to collect up to 64 kB for each bio */ | |
d9d181c1 | 2428 | static int scrub_extent(struct scrub_ctx *sctx, u64 logical, u64 len, |
a36cf8b8 | 2429 | u64 physical, struct btrfs_device *dev, u64 flags, |
ff023aac | 2430 | u64 gen, int mirror_num, u64 physical_for_dev_replace) |
a2de733c AJ |
2431 | { |
2432 | int ret; | |
2433 | u8 csum[BTRFS_CSUM_SIZE]; | |
b5d67f64 SB |
2434 | u32 blocksize; |
2435 | ||
2436 | if (flags & BTRFS_EXTENT_FLAG_DATA) { | |
d9d181c1 SB |
2437 | blocksize = sctx->sectorsize; |
2438 | spin_lock(&sctx->stat_lock); | |
2439 | sctx->stat.data_extents_scrubbed++; | |
2440 | sctx->stat.data_bytes_scrubbed += len; | |
2441 | spin_unlock(&sctx->stat_lock); | |
b5d67f64 | 2442 | } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { |
d9d181c1 SB |
2443 | blocksize = sctx->nodesize; |
2444 | spin_lock(&sctx->stat_lock); | |
2445 | sctx->stat.tree_extents_scrubbed++; | |
2446 | sctx->stat.tree_bytes_scrubbed += len; | |
2447 | spin_unlock(&sctx->stat_lock); | |
b5d67f64 | 2448 | } else { |
d9d181c1 | 2449 | blocksize = sctx->sectorsize; |
ff023aac | 2450 | WARN_ON(1); |
b5d67f64 | 2451 | } |
a2de733c AJ |
2452 | |
2453 | while (len) { | |
b5d67f64 | 2454 | u64 l = min_t(u64, len, blocksize); |
a2de733c AJ |
2455 | int have_csum = 0; |
2456 | ||
2457 | if (flags & BTRFS_EXTENT_FLAG_DATA) { | |
2458 | /* push csums to sbio */ | |
d9d181c1 | 2459 | have_csum = scrub_find_csum(sctx, logical, l, csum); |
a2de733c | 2460 | if (have_csum == 0) |
d9d181c1 | 2461 | ++sctx->stat.no_csum; |
ff023aac SB |
2462 | if (sctx->is_dev_replace && !have_csum) { |
2463 | ret = copy_nocow_pages(sctx, logical, l, | |
2464 | mirror_num, | |
2465 | physical_for_dev_replace); | |
2466 | goto behind_scrub_pages; | |
2467 | } | |
a2de733c | 2468 | } |
a36cf8b8 | 2469 | ret = scrub_pages(sctx, logical, l, physical, dev, flags, gen, |
ff023aac SB |
2470 | mirror_num, have_csum ? csum : NULL, 0, |
2471 | physical_for_dev_replace); | |
2472 | behind_scrub_pages: | |
a2de733c AJ |
2473 | if (ret) |
2474 | return ret; | |
2475 | len -= l; | |
2476 | logical += l; | |
2477 | physical += l; | |
ff023aac | 2478 | physical_for_dev_replace += l; |
a2de733c AJ |
2479 | } |
2480 | return 0; | |
2481 | } | |
2482 | ||
5a6ac9ea MX |
2483 | static int scrub_pages_for_parity(struct scrub_parity *sparity, |
2484 | u64 logical, u64 len, | |
2485 | u64 physical, struct btrfs_device *dev, | |
2486 | u64 flags, u64 gen, int mirror_num, u8 *csum) | |
2487 | { | |
2488 | struct scrub_ctx *sctx = sparity->sctx; | |
2489 | struct scrub_block *sblock; | |
2490 | int index; | |
2491 | ||
2492 | sblock = kzalloc(sizeof(*sblock), GFP_NOFS); | |
2493 | if (!sblock) { | |
2494 | spin_lock(&sctx->stat_lock); | |
2495 | sctx->stat.malloc_errors++; | |
2496 | spin_unlock(&sctx->stat_lock); | |
2497 | return -ENOMEM; | |
2498 | } | |
2499 | ||
2500 | /* one ref inside this function, plus one for each page added to | |
2501 | * a bio later on */ | |
2502 | atomic_set(&sblock->ref_count, 1); | |
2503 | sblock->sctx = sctx; | |
2504 | sblock->no_io_error_seen = 1; | |
2505 | sblock->sparity = sparity; | |
2506 | scrub_parity_get(sparity); | |
2507 | ||
2508 | for (index = 0; len > 0; index++) { | |
2509 | struct scrub_page *spage; | |
2510 | u64 l = min_t(u64, len, PAGE_SIZE); | |
2511 | ||
2512 | spage = kzalloc(sizeof(*spage), GFP_NOFS); | |
2513 | if (!spage) { | |
2514 | leave_nomem: | |
2515 | spin_lock(&sctx->stat_lock); | |
2516 | sctx->stat.malloc_errors++; | |
2517 | spin_unlock(&sctx->stat_lock); | |
2518 | scrub_block_put(sblock); | |
2519 | return -ENOMEM; | |
2520 | } | |
2521 | BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK); | |
2522 | /* For scrub block */ | |
2523 | scrub_page_get(spage); | |
2524 | sblock->pagev[index] = spage; | |
2525 | /* For scrub parity */ | |
2526 | scrub_page_get(spage); | |
2527 | list_add_tail(&spage->list, &sparity->spages); | |
2528 | spage->sblock = sblock; | |
2529 | spage->dev = dev; | |
2530 | spage->flags = flags; | |
2531 | spage->generation = gen; | |
2532 | spage->logical = logical; | |
2533 | spage->physical = physical; | |
2534 | spage->mirror_num = mirror_num; | |
2535 | if (csum) { | |
2536 | spage->have_csum = 1; | |
2537 | memcpy(spage->csum, csum, sctx->csum_size); | |
2538 | } else { | |
2539 | spage->have_csum = 0; | |
2540 | } | |
2541 | sblock->page_count++; | |
2542 | spage->page = alloc_page(GFP_NOFS); | |
2543 | if (!spage->page) | |
2544 | goto leave_nomem; | |
2545 | len -= l; | |
2546 | logical += l; | |
2547 | physical += l; | |
2548 | } | |
2549 | ||
2550 | WARN_ON(sblock->page_count == 0); | |
2551 | for (index = 0; index < sblock->page_count; index++) { | |
2552 | struct scrub_page *spage = sblock->pagev[index]; | |
2553 | int ret; | |
2554 | ||
2555 | ret = scrub_add_page_to_rd_bio(sctx, spage); | |
2556 | if (ret) { | |
2557 | scrub_block_put(sblock); | |
2558 | return ret; | |
2559 | } | |
2560 | } | |
2561 | ||
2562 | /* last one frees, either here or in bio completion for last page */ | |
2563 | scrub_block_put(sblock); | |
2564 | return 0; | |
2565 | } | |
2566 | ||
2567 | static int scrub_extent_for_parity(struct scrub_parity *sparity, | |
2568 | u64 logical, u64 len, | |
2569 | u64 physical, struct btrfs_device *dev, | |
2570 | u64 flags, u64 gen, int mirror_num) | |
2571 | { | |
2572 | struct scrub_ctx *sctx = sparity->sctx; | |
2573 | int ret; | |
2574 | u8 csum[BTRFS_CSUM_SIZE]; | |
2575 | u32 blocksize; | |
2576 | ||
2577 | if (flags & BTRFS_EXTENT_FLAG_DATA) { | |
2578 | blocksize = sctx->sectorsize; | |
2579 | } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { | |
2580 | blocksize = sctx->nodesize; | |
2581 | } else { | |
2582 | blocksize = sctx->sectorsize; | |
2583 | WARN_ON(1); | |
2584 | } | |
2585 | ||
2586 | while (len) { | |
2587 | u64 l = min_t(u64, len, blocksize); | |
2588 | int have_csum = 0; | |
2589 | ||
2590 | if (flags & BTRFS_EXTENT_FLAG_DATA) { | |
2591 | /* push csums to sbio */ | |
2592 | have_csum = scrub_find_csum(sctx, logical, l, csum); | |
2593 | if (have_csum == 0) | |
2594 | goto skip; | |
2595 | } | |
2596 | ret = scrub_pages_for_parity(sparity, logical, l, physical, dev, | |
2597 | flags, gen, mirror_num, | |
2598 | have_csum ? csum : NULL); | |
5a6ac9ea MX |
2599 | if (ret) |
2600 | return ret; | |
6b6d24b3 | 2601 | skip: |
5a6ac9ea MX |
2602 | len -= l; |
2603 | logical += l; | |
2604 | physical += l; | |
2605 | } | |
2606 | return 0; | |
2607 | } | |
2608 | ||
3b080b25 WS |
2609 | /* |
2610 | * Given a physical address, this will calculate it's | |
2611 | * logical offset. if this is a parity stripe, it will return | |
2612 | * the most left data stripe's logical offset. | |
2613 | * | |
2614 | * return 0 if it is a data stripe, 1 means parity stripe. | |
2615 | */ | |
2616 | static int get_raid56_logic_offset(u64 physical, int num, | |
5a6ac9ea MX |
2617 | struct map_lookup *map, u64 *offset, |
2618 | u64 *stripe_start) | |
3b080b25 WS |
2619 | { |
2620 | int i; | |
2621 | int j = 0; | |
2622 | u64 stripe_nr; | |
2623 | u64 last_offset; | |
2624 | int stripe_index; | |
2625 | int rot; | |
2626 | ||
2627 | last_offset = (physical - map->stripes[num].physical) * | |
2628 | nr_data_stripes(map); | |
5a6ac9ea MX |
2629 | if (stripe_start) |
2630 | *stripe_start = last_offset; | |
2631 | ||
3b080b25 WS |
2632 | *offset = last_offset; |
2633 | for (i = 0; i < nr_data_stripes(map); i++) { | |
2634 | *offset = last_offset + i * map->stripe_len; | |
2635 | ||
2636 | stripe_nr = *offset; | |
2637 | do_div(stripe_nr, map->stripe_len); | |
2638 | do_div(stripe_nr, nr_data_stripes(map)); | |
2639 | ||
2640 | /* Work out the disk rotation on this stripe-set */ | |
2641 | rot = do_div(stripe_nr, map->num_stripes); | |
2642 | /* calculate which stripe this data locates */ | |
2643 | rot += i; | |
e4fbaee2 | 2644 | stripe_index = rot % map->num_stripes; |
3b080b25 WS |
2645 | if (stripe_index == num) |
2646 | return 0; | |
2647 | if (stripe_index < num) | |
2648 | j++; | |
2649 | } | |
2650 | *offset = last_offset + j * map->stripe_len; | |
2651 | return 1; | |
2652 | } | |
2653 | ||
5a6ac9ea MX |
2654 | static void scrub_free_parity(struct scrub_parity *sparity) |
2655 | { | |
2656 | struct scrub_ctx *sctx = sparity->sctx; | |
2657 | struct scrub_page *curr, *next; | |
2658 | int nbits; | |
2659 | ||
2660 | nbits = bitmap_weight(sparity->ebitmap, sparity->nsectors); | |
2661 | if (nbits) { | |
2662 | spin_lock(&sctx->stat_lock); | |
2663 | sctx->stat.read_errors += nbits; | |
2664 | sctx->stat.uncorrectable_errors += nbits; | |
2665 | spin_unlock(&sctx->stat_lock); | |
2666 | } | |
2667 | ||
2668 | list_for_each_entry_safe(curr, next, &sparity->spages, list) { | |
2669 | list_del_init(&curr->list); | |
2670 | scrub_page_put(curr); | |
2671 | } | |
2672 | ||
2673 | kfree(sparity); | |
2674 | } | |
2675 | ||
2676 | static void scrub_parity_bio_endio(struct bio *bio, int error) | |
2677 | { | |
2678 | struct scrub_parity *sparity = (struct scrub_parity *)bio->bi_private; | |
2679 | struct scrub_ctx *sctx = sparity->sctx; | |
2680 | ||
2681 | if (error) | |
2682 | bitmap_or(sparity->ebitmap, sparity->ebitmap, sparity->dbitmap, | |
2683 | sparity->nsectors); | |
2684 | ||
2685 | scrub_free_parity(sparity); | |
2686 | scrub_pending_bio_dec(sctx); | |
2687 | bio_put(bio); | |
2688 | } | |
2689 | ||
2690 | static void scrub_parity_check_and_repair(struct scrub_parity *sparity) | |
2691 | { | |
2692 | struct scrub_ctx *sctx = sparity->sctx; | |
2693 | struct bio *bio; | |
2694 | struct btrfs_raid_bio *rbio; | |
2695 | struct scrub_page *spage; | |
2696 | struct btrfs_bio *bbio = NULL; | |
5a6ac9ea MX |
2697 | u64 length; |
2698 | int ret; | |
2699 | ||
2700 | if (!bitmap_andnot(sparity->dbitmap, sparity->dbitmap, sparity->ebitmap, | |
2701 | sparity->nsectors)) | |
2702 | goto out; | |
2703 | ||
2704 | length = sparity->logic_end - sparity->logic_start + 1; | |
76035976 | 2705 | ret = btrfs_map_sblock(sctx->dev_root->fs_info, WRITE, |
5a6ac9ea | 2706 | sparity->logic_start, |
8e5cfb55 ZL |
2707 | &length, &bbio, 0, 1); |
2708 | if (ret || !bbio || !bbio->raid_map) | |
5a6ac9ea MX |
2709 | goto bbio_out; |
2710 | ||
2711 | bio = btrfs_io_bio_alloc(GFP_NOFS, 0); | |
2712 | if (!bio) | |
2713 | goto bbio_out; | |
2714 | ||
2715 | bio->bi_iter.bi_sector = sparity->logic_start >> 9; | |
2716 | bio->bi_private = sparity; | |
2717 | bio->bi_end_io = scrub_parity_bio_endio; | |
2718 | ||
2719 | rbio = raid56_parity_alloc_scrub_rbio(sctx->dev_root, bio, bbio, | |
8e5cfb55 | 2720 | length, sparity->scrub_dev, |
5a6ac9ea MX |
2721 | sparity->dbitmap, |
2722 | sparity->nsectors); | |
2723 | if (!rbio) | |
2724 | goto rbio_out; | |
2725 | ||
2726 | list_for_each_entry(spage, &sparity->spages, list) | |
2727 | raid56_parity_add_scrub_pages(rbio, spage->page, | |
2728 | spage->logical); | |
2729 | ||
2730 | scrub_pending_bio_inc(sctx); | |
2731 | raid56_parity_submit_scrub_rbio(rbio); | |
2732 | return; | |
2733 | ||
2734 | rbio_out: | |
2735 | bio_put(bio); | |
2736 | bbio_out: | |
6e9606d2 | 2737 | btrfs_put_bbio(bbio); |
5a6ac9ea MX |
2738 | bitmap_or(sparity->ebitmap, sparity->ebitmap, sparity->dbitmap, |
2739 | sparity->nsectors); | |
2740 | spin_lock(&sctx->stat_lock); | |
2741 | sctx->stat.malloc_errors++; | |
2742 | spin_unlock(&sctx->stat_lock); | |
2743 | out: | |
2744 | scrub_free_parity(sparity); | |
2745 | } | |
2746 | ||
2747 | static inline int scrub_calc_parity_bitmap_len(int nsectors) | |
2748 | { | |
2749 | return DIV_ROUND_UP(nsectors, BITS_PER_LONG) * (BITS_PER_LONG / 8); | |
2750 | } | |
2751 | ||
2752 | static void scrub_parity_get(struct scrub_parity *sparity) | |
2753 | { | |
2754 | atomic_inc(&sparity->ref_count); | |
2755 | } | |
2756 | ||
2757 | static void scrub_parity_put(struct scrub_parity *sparity) | |
2758 | { | |
2759 | if (!atomic_dec_and_test(&sparity->ref_count)) | |
2760 | return; | |
2761 | ||
2762 | scrub_parity_check_and_repair(sparity); | |
2763 | } | |
2764 | ||
2765 | static noinline_for_stack int scrub_raid56_parity(struct scrub_ctx *sctx, | |
2766 | struct map_lookup *map, | |
2767 | struct btrfs_device *sdev, | |
2768 | struct btrfs_path *path, | |
2769 | u64 logic_start, | |
2770 | u64 logic_end) | |
2771 | { | |
2772 | struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info; | |
2773 | struct btrfs_root *root = fs_info->extent_root; | |
2774 | struct btrfs_root *csum_root = fs_info->csum_root; | |
2775 | struct btrfs_extent_item *extent; | |
2776 | u64 flags; | |
2777 | int ret; | |
2778 | int slot; | |
2779 | struct extent_buffer *l; | |
2780 | struct btrfs_key key; | |
2781 | u64 generation; | |
2782 | u64 extent_logical; | |
2783 | u64 extent_physical; | |
2784 | u64 extent_len; | |
2785 | struct btrfs_device *extent_dev; | |
2786 | struct scrub_parity *sparity; | |
2787 | int nsectors; | |
2788 | int bitmap_len; | |
2789 | int extent_mirror_num; | |
2790 | int stop_loop = 0; | |
2791 | ||
2792 | nsectors = map->stripe_len / root->sectorsize; | |
2793 | bitmap_len = scrub_calc_parity_bitmap_len(nsectors); | |
2794 | sparity = kzalloc(sizeof(struct scrub_parity) + 2 * bitmap_len, | |
2795 | GFP_NOFS); | |
2796 | if (!sparity) { | |
2797 | spin_lock(&sctx->stat_lock); | |
2798 | sctx->stat.malloc_errors++; | |
2799 | spin_unlock(&sctx->stat_lock); | |
2800 | return -ENOMEM; | |
2801 | } | |
2802 | ||
2803 | sparity->stripe_len = map->stripe_len; | |
2804 | sparity->nsectors = nsectors; | |
2805 | sparity->sctx = sctx; | |
2806 | sparity->scrub_dev = sdev; | |
2807 | sparity->logic_start = logic_start; | |
2808 | sparity->logic_end = logic_end; | |
2809 | atomic_set(&sparity->ref_count, 1); | |
2810 | INIT_LIST_HEAD(&sparity->spages); | |
2811 | sparity->dbitmap = sparity->bitmap; | |
2812 | sparity->ebitmap = (void *)sparity->bitmap + bitmap_len; | |
2813 | ||
2814 | ret = 0; | |
2815 | while (logic_start < logic_end) { | |
2816 | if (btrfs_fs_incompat(fs_info, SKINNY_METADATA)) | |
2817 | key.type = BTRFS_METADATA_ITEM_KEY; | |
2818 | else | |
2819 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2820 | key.objectid = logic_start; | |
2821 | key.offset = (u64)-1; | |
2822 | ||
2823 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2824 | if (ret < 0) | |
2825 | goto out; | |
2826 | ||
2827 | if (ret > 0) { | |
2828 | ret = btrfs_previous_extent_item(root, path, 0); | |
2829 | if (ret < 0) | |
2830 | goto out; | |
2831 | if (ret > 0) { | |
2832 | btrfs_release_path(path); | |
2833 | ret = btrfs_search_slot(NULL, root, &key, | |
2834 | path, 0, 0); | |
2835 | if (ret < 0) | |
2836 | goto out; | |
2837 | } | |
2838 | } | |
2839 | ||
2840 | stop_loop = 0; | |
2841 | while (1) { | |
2842 | u64 bytes; | |
2843 | ||
2844 | l = path->nodes[0]; | |
2845 | slot = path->slots[0]; | |
2846 | if (slot >= btrfs_header_nritems(l)) { | |
2847 | ret = btrfs_next_leaf(root, path); | |
2848 | if (ret == 0) | |
2849 | continue; | |
2850 | if (ret < 0) | |
2851 | goto out; | |
2852 | ||
2853 | stop_loop = 1; | |
2854 | break; | |
2855 | } | |
2856 | btrfs_item_key_to_cpu(l, &key, slot); | |
2857 | ||
2858 | if (key.type == BTRFS_METADATA_ITEM_KEY) | |
2859 | bytes = root->nodesize; | |
2860 | else | |
2861 | bytes = key.offset; | |
2862 | ||
2863 | if (key.objectid + bytes <= logic_start) | |
2864 | goto next; | |
2865 | ||
2866 | if (key.type != BTRFS_EXTENT_ITEM_KEY && | |
2867 | key.type != BTRFS_METADATA_ITEM_KEY) | |
2868 | goto next; | |
2869 | ||
2870 | if (key.objectid > logic_end) { | |
2871 | stop_loop = 1; | |
2872 | break; | |
2873 | } | |
2874 | ||
2875 | while (key.objectid >= logic_start + map->stripe_len) | |
2876 | logic_start += map->stripe_len; | |
2877 | ||
2878 | extent = btrfs_item_ptr(l, slot, | |
2879 | struct btrfs_extent_item); | |
2880 | flags = btrfs_extent_flags(l, extent); | |
2881 | generation = btrfs_extent_generation(l, extent); | |
2882 | ||
2883 | if (key.objectid < logic_start && | |
2884 | (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) { | |
2885 | btrfs_err(fs_info, | |
2886 | "scrub: tree block %llu spanning stripes, ignored. logical=%llu", | |
2887 | key.objectid, logic_start); | |
2888 | goto next; | |
2889 | } | |
2890 | again: | |
2891 | extent_logical = key.objectid; | |
2892 | extent_len = bytes; | |
2893 | ||
2894 | if (extent_logical < logic_start) { | |
2895 | extent_len -= logic_start - extent_logical; | |
2896 | extent_logical = logic_start; | |
2897 | } | |
2898 | ||
2899 | if (extent_logical + extent_len > | |
2900 | logic_start + map->stripe_len) | |
2901 | extent_len = logic_start + map->stripe_len - | |
2902 | extent_logical; | |
2903 | ||
2904 | scrub_parity_mark_sectors_data(sparity, extent_logical, | |
2905 | extent_len); | |
2906 | ||
2907 | scrub_remap_extent(fs_info, extent_logical, | |
2908 | extent_len, &extent_physical, | |
2909 | &extent_dev, | |
2910 | &extent_mirror_num); | |
2911 | ||
2912 | ret = btrfs_lookup_csums_range(csum_root, | |
2913 | extent_logical, | |
2914 | extent_logical + extent_len - 1, | |
2915 | &sctx->csum_list, 1); | |
2916 | if (ret) | |
2917 | goto out; | |
2918 | ||
2919 | ret = scrub_extent_for_parity(sparity, extent_logical, | |
2920 | extent_len, | |
2921 | extent_physical, | |
2922 | extent_dev, flags, | |
2923 | generation, | |
2924 | extent_mirror_num); | |
2925 | if (ret) | |
2926 | goto out; | |
2927 | ||
2928 | scrub_free_csums(sctx); | |
2929 | if (extent_logical + extent_len < | |
2930 | key.objectid + bytes) { | |
2931 | logic_start += map->stripe_len; | |
2932 | ||
2933 | if (logic_start >= logic_end) { | |
2934 | stop_loop = 1; | |
2935 | break; | |
2936 | } | |
2937 | ||
2938 | if (logic_start < key.objectid + bytes) { | |
2939 | cond_resched(); | |
2940 | goto again; | |
2941 | } | |
2942 | } | |
2943 | next: | |
2944 | path->slots[0]++; | |
2945 | } | |
2946 | ||
2947 | btrfs_release_path(path); | |
2948 | ||
2949 | if (stop_loop) | |
2950 | break; | |
2951 | ||
2952 | logic_start += map->stripe_len; | |
2953 | } | |
2954 | out: | |
2955 | if (ret < 0) | |
2956 | scrub_parity_mark_sectors_error(sparity, logic_start, | |
2957 | logic_end - logic_start + 1); | |
2958 | scrub_parity_put(sparity); | |
2959 | scrub_submit(sctx); | |
2960 | mutex_lock(&sctx->wr_ctx.wr_lock); | |
2961 | scrub_wr_submit(sctx); | |
2962 | mutex_unlock(&sctx->wr_ctx.wr_lock); | |
2963 | ||
2964 | btrfs_release_path(path); | |
2965 | return ret < 0 ? ret : 0; | |
2966 | } | |
2967 | ||
d9d181c1 | 2968 | static noinline_for_stack int scrub_stripe(struct scrub_ctx *sctx, |
a36cf8b8 SB |
2969 | struct map_lookup *map, |
2970 | struct btrfs_device *scrub_dev, | |
ff023aac SB |
2971 | int num, u64 base, u64 length, |
2972 | int is_dev_replace) | |
a2de733c | 2973 | { |
5a6ac9ea | 2974 | struct btrfs_path *path, *ppath; |
a36cf8b8 | 2975 | struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info; |
a2de733c AJ |
2976 | struct btrfs_root *root = fs_info->extent_root; |
2977 | struct btrfs_root *csum_root = fs_info->csum_root; | |
2978 | struct btrfs_extent_item *extent; | |
e7786c3a | 2979 | struct blk_plug plug; |
a2de733c AJ |
2980 | u64 flags; |
2981 | int ret; | |
2982 | int slot; | |
a2de733c | 2983 | u64 nstripes; |
a2de733c AJ |
2984 | struct extent_buffer *l; |
2985 | struct btrfs_key key; | |
2986 | u64 physical; | |
2987 | u64 logical; | |
625f1c8d | 2988 | u64 logic_end; |
3b080b25 | 2989 | u64 physical_end; |
a2de733c | 2990 | u64 generation; |
e12fa9cd | 2991 | int mirror_num; |
7a26285e AJ |
2992 | struct reada_control *reada1; |
2993 | struct reada_control *reada2; | |
2994 | struct btrfs_key key_start; | |
2995 | struct btrfs_key key_end; | |
a2de733c AJ |
2996 | u64 increment = map->stripe_len; |
2997 | u64 offset; | |
ff023aac SB |
2998 | u64 extent_logical; |
2999 | u64 extent_physical; | |
3000 | u64 extent_len; | |
5a6ac9ea MX |
3001 | u64 stripe_logical; |
3002 | u64 stripe_end; | |
ff023aac SB |
3003 | struct btrfs_device *extent_dev; |
3004 | int extent_mirror_num; | |
3b080b25 | 3005 | int stop_loop = 0; |
53b381b3 | 3006 | |
a2de733c | 3007 | nstripes = length; |
3b080b25 | 3008 | physical = map->stripes[num].physical; |
a2de733c AJ |
3009 | offset = 0; |
3010 | do_div(nstripes, map->stripe_len); | |
3011 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { | |
3012 | offset = map->stripe_len * num; | |
3013 | increment = map->stripe_len * map->num_stripes; | |
193ea74b | 3014 | mirror_num = 1; |
a2de733c AJ |
3015 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
3016 | int factor = map->num_stripes / map->sub_stripes; | |
3017 | offset = map->stripe_len * (num / map->sub_stripes); | |
3018 | increment = map->stripe_len * factor; | |
193ea74b | 3019 | mirror_num = num % map->sub_stripes + 1; |
a2de733c AJ |
3020 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { |
3021 | increment = map->stripe_len; | |
193ea74b | 3022 | mirror_num = num % map->num_stripes + 1; |
a2de733c AJ |
3023 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
3024 | increment = map->stripe_len; | |
193ea74b | 3025 | mirror_num = num % map->num_stripes + 1; |
3b080b25 WS |
3026 | } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | |
3027 | BTRFS_BLOCK_GROUP_RAID6)) { | |
5a6ac9ea | 3028 | get_raid56_logic_offset(physical, num, map, &offset, NULL); |
3b080b25 WS |
3029 | increment = map->stripe_len * nr_data_stripes(map); |
3030 | mirror_num = 1; | |
a2de733c AJ |
3031 | } else { |
3032 | increment = map->stripe_len; | |
193ea74b | 3033 | mirror_num = 1; |
a2de733c AJ |
3034 | } |
3035 | ||
3036 | path = btrfs_alloc_path(); | |
3037 | if (!path) | |
3038 | return -ENOMEM; | |
3039 | ||
5a6ac9ea MX |
3040 | ppath = btrfs_alloc_path(); |
3041 | if (!ppath) { | |
3042 | btrfs_free_path(ppath); | |
3043 | return -ENOMEM; | |
3044 | } | |
3045 | ||
b5d67f64 SB |
3046 | /* |
3047 | * work on commit root. The related disk blocks are static as | |
3048 | * long as COW is applied. This means, it is save to rewrite | |
3049 | * them to repair disk errors without any race conditions | |
3050 | */ | |
a2de733c AJ |
3051 | path->search_commit_root = 1; |
3052 | path->skip_locking = 1; | |
3053 | ||
3054 | /* | |
7a26285e AJ |
3055 | * trigger the readahead for extent tree csum tree and wait for |
3056 | * completion. During readahead, the scrub is officially paused | |
3057 | * to not hold off transaction commits | |
a2de733c AJ |
3058 | */ |
3059 | logical = base + offset; | |
3b080b25 WS |
3060 | physical_end = physical + nstripes * map->stripe_len; |
3061 | if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | | |
3062 | BTRFS_BLOCK_GROUP_RAID6)) { | |
3063 | get_raid56_logic_offset(physical_end, num, | |
5a6ac9ea | 3064 | map, &logic_end, NULL); |
3b080b25 WS |
3065 | logic_end += base; |
3066 | } else { | |
3067 | logic_end = logical + increment * nstripes; | |
3068 | } | |
d9d181c1 | 3069 | wait_event(sctx->list_wait, |
b6bfebc1 | 3070 | atomic_read(&sctx->bios_in_flight) == 0); |
cb7ab021 | 3071 | scrub_blocked_if_needed(fs_info); |
7a26285e AJ |
3072 | |
3073 | /* FIXME it might be better to start readahead at commit root */ | |
3074 | key_start.objectid = logical; | |
3075 | key_start.type = BTRFS_EXTENT_ITEM_KEY; | |
3076 | key_start.offset = (u64)0; | |
3b080b25 | 3077 | key_end.objectid = logic_end; |
3173a18f JB |
3078 | key_end.type = BTRFS_METADATA_ITEM_KEY; |
3079 | key_end.offset = (u64)-1; | |
7a26285e AJ |
3080 | reada1 = btrfs_reada_add(root, &key_start, &key_end); |
3081 | ||
3082 | key_start.objectid = BTRFS_EXTENT_CSUM_OBJECTID; | |
3083 | key_start.type = BTRFS_EXTENT_CSUM_KEY; | |
3084 | key_start.offset = logical; | |
3085 | key_end.objectid = BTRFS_EXTENT_CSUM_OBJECTID; | |
3086 | key_end.type = BTRFS_EXTENT_CSUM_KEY; | |
3b080b25 | 3087 | key_end.offset = logic_end; |
7a26285e AJ |
3088 | reada2 = btrfs_reada_add(csum_root, &key_start, &key_end); |
3089 | ||
3090 | if (!IS_ERR(reada1)) | |
3091 | btrfs_reada_wait(reada1); | |
3092 | if (!IS_ERR(reada2)) | |
3093 | btrfs_reada_wait(reada2); | |
3094 | ||
a2de733c AJ |
3095 | |
3096 | /* | |
3097 | * collect all data csums for the stripe to avoid seeking during | |
3098 | * the scrub. This might currently (crc32) end up to be about 1MB | |
3099 | */ | |
e7786c3a | 3100 | blk_start_plug(&plug); |
a2de733c | 3101 | |
a2de733c AJ |
3102 | /* |
3103 | * now find all extents for each stripe and scrub them | |
3104 | */ | |
a2de733c | 3105 | ret = 0; |
3b080b25 WS |
3106 | while (physical < physical_end) { |
3107 | /* for raid56, we skip parity stripe */ | |
3108 | if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | | |
3109 | BTRFS_BLOCK_GROUP_RAID6)) { | |
3110 | ret = get_raid56_logic_offset(physical, num, | |
5a6ac9ea | 3111 | map, &logical, &stripe_logical); |
3b080b25 | 3112 | logical += base; |
5a6ac9ea MX |
3113 | if (ret) { |
3114 | stripe_logical += base; | |
3115 | stripe_end = stripe_logical + increment - 1; | |
3116 | ret = scrub_raid56_parity(sctx, map, scrub_dev, | |
3117 | ppath, stripe_logical, | |
3118 | stripe_end); | |
3119 | if (ret) | |
3120 | goto out; | |
3b080b25 | 3121 | goto skip; |
5a6ac9ea | 3122 | } |
3b080b25 | 3123 | } |
a2de733c AJ |
3124 | /* |
3125 | * canceled? | |
3126 | */ | |
3127 | if (atomic_read(&fs_info->scrub_cancel_req) || | |
d9d181c1 | 3128 | atomic_read(&sctx->cancel_req)) { |
a2de733c AJ |
3129 | ret = -ECANCELED; |
3130 | goto out; | |
3131 | } | |
3132 | /* | |
3133 | * check to see if we have to pause | |
3134 | */ | |
3135 | if (atomic_read(&fs_info->scrub_pause_req)) { | |
3136 | /* push queued extents */ | |
ff023aac | 3137 | atomic_set(&sctx->wr_ctx.flush_all_writes, 1); |
d9d181c1 | 3138 | scrub_submit(sctx); |
ff023aac SB |
3139 | mutex_lock(&sctx->wr_ctx.wr_lock); |
3140 | scrub_wr_submit(sctx); | |
3141 | mutex_unlock(&sctx->wr_ctx.wr_lock); | |
d9d181c1 | 3142 | wait_event(sctx->list_wait, |
b6bfebc1 | 3143 | atomic_read(&sctx->bios_in_flight) == 0); |
ff023aac | 3144 | atomic_set(&sctx->wr_ctx.flush_all_writes, 0); |
3cb0929a | 3145 | scrub_blocked_if_needed(fs_info); |
a2de733c AJ |
3146 | } |
3147 | ||
7c76edb7 WS |
3148 | if (btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
3149 | key.type = BTRFS_METADATA_ITEM_KEY; | |
3150 | else | |
3151 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
a2de733c | 3152 | key.objectid = logical; |
625f1c8d | 3153 | key.offset = (u64)-1; |
a2de733c AJ |
3154 | |
3155 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
3156 | if (ret < 0) | |
3157 | goto out; | |
3173a18f | 3158 | |
8c51032f | 3159 | if (ret > 0) { |
ade2e0b3 | 3160 | ret = btrfs_previous_extent_item(root, path, 0); |
a2de733c AJ |
3161 | if (ret < 0) |
3162 | goto out; | |
8c51032f AJ |
3163 | if (ret > 0) { |
3164 | /* there's no smaller item, so stick with the | |
3165 | * larger one */ | |
3166 | btrfs_release_path(path); | |
3167 | ret = btrfs_search_slot(NULL, root, &key, | |
3168 | path, 0, 0); | |
3169 | if (ret < 0) | |
3170 | goto out; | |
3171 | } | |
a2de733c AJ |
3172 | } |
3173 | ||
625f1c8d | 3174 | stop_loop = 0; |
a2de733c | 3175 | while (1) { |
3173a18f JB |
3176 | u64 bytes; |
3177 | ||
a2de733c AJ |
3178 | l = path->nodes[0]; |
3179 | slot = path->slots[0]; | |
3180 | if (slot >= btrfs_header_nritems(l)) { | |
3181 | ret = btrfs_next_leaf(root, path); | |
3182 | if (ret == 0) | |
3183 | continue; | |
3184 | if (ret < 0) | |
3185 | goto out; | |
3186 | ||
625f1c8d | 3187 | stop_loop = 1; |
a2de733c AJ |
3188 | break; |
3189 | } | |
3190 | btrfs_item_key_to_cpu(l, &key, slot); | |
3191 | ||
3173a18f | 3192 | if (key.type == BTRFS_METADATA_ITEM_KEY) |
707e8a07 | 3193 | bytes = root->nodesize; |
3173a18f JB |
3194 | else |
3195 | bytes = key.offset; | |
3196 | ||
3197 | if (key.objectid + bytes <= logical) | |
a2de733c AJ |
3198 | goto next; |
3199 | ||
625f1c8d LB |
3200 | if (key.type != BTRFS_EXTENT_ITEM_KEY && |
3201 | key.type != BTRFS_METADATA_ITEM_KEY) | |
3202 | goto next; | |
a2de733c | 3203 | |
625f1c8d LB |
3204 | if (key.objectid >= logical + map->stripe_len) { |
3205 | /* out of this device extent */ | |
3206 | if (key.objectid >= logic_end) | |
3207 | stop_loop = 1; | |
3208 | break; | |
3209 | } | |
a2de733c AJ |
3210 | |
3211 | extent = btrfs_item_ptr(l, slot, | |
3212 | struct btrfs_extent_item); | |
3213 | flags = btrfs_extent_flags(l, extent); | |
3214 | generation = btrfs_extent_generation(l, extent); | |
3215 | ||
3216 | if (key.objectid < logical && | |
3217 | (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) { | |
efe120a0 FH |
3218 | btrfs_err(fs_info, |
3219 | "scrub: tree block %llu spanning " | |
3220 | "stripes, ignored. logical=%llu", | |
c1c9ff7c | 3221 | key.objectid, logical); |
a2de733c AJ |
3222 | goto next; |
3223 | } | |
3224 | ||
625f1c8d LB |
3225 | again: |
3226 | extent_logical = key.objectid; | |
3227 | extent_len = bytes; | |
3228 | ||
a2de733c AJ |
3229 | /* |
3230 | * trim extent to this stripe | |
3231 | */ | |
625f1c8d LB |
3232 | if (extent_logical < logical) { |
3233 | extent_len -= logical - extent_logical; | |
3234 | extent_logical = logical; | |
a2de733c | 3235 | } |
625f1c8d | 3236 | if (extent_logical + extent_len > |
a2de733c | 3237 | logical + map->stripe_len) { |
625f1c8d LB |
3238 | extent_len = logical + map->stripe_len - |
3239 | extent_logical; | |
a2de733c AJ |
3240 | } |
3241 | ||
625f1c8d | 3242 | extent_physical = extent_logical - logical + physical; |
ff023aac SB |
3243 | extent_dev = scrub_dev; |
3244 | extent_mirror_num = mirror_num; | |
3245 | if (is_dev_replace) | |
3246 | scrub_remap_extent(fs_info, extent_logical, | |
3247 | extent_len, &extent_physical, | |
3248 | &extent_dev, | |
3249 | &extent_mirror_num); | |
625f1c8d LB |
3250 | |
3251 | ret = btrfs_lookup_csums_range(csum_root, logical, | |
3252 | logical + map->stripe_len - 1, | |
3253 | &sctx->csum_list, 1); | |
3254 | if (ret) | |
3255 | goto out; | |
3256 | ||
ff023aac SB |
3257 | ret = scrub_extent(sctx, extent_logical, extent_len, |
3258 | extent_physical, extent_dev, flags, | |
3259 | generation, extent_mirror_num, | |
115930cb | 3260 | extent_logical - logical + physical); |
a2de733c AJ |
3261 | if (ret) |
3262 | goto out; | |
3263 | ||
d88d46c6 | 3264 | scrub_free_csums(sctx); |
625f1c8d LB |
3265 | if (extent_logical + extent_len < |
3266 | key.objectid + bytes) { | |
3b080b25 WS |
3267 | if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | |
3268 | BTRFS_BLOCK_GROUP_RAID6)) { | |
3269 | /* | |
3270 | * loop until we find next data stripe | |
3271 | * or we have finished all stripes. | |
3272 | */ | |
5a6ac9ea MX |
3273 | loop: |
3274 | physical += map->stripe_len; | |
3275 | ret = get_raid56_logic_offset(physical, | |
3276 | num, map, &logical, | |
3277 | &stripe_logical); | |
3278 | logical += base; | |
3279 | ||
3280 | if (ret && physical < physical_end) { | |
3281 | stripe_logical += base; | |
3282 | stripe_end = stripe_logical + | |
3283 | increment - 1; | |
3284 | ret = scrub_raid56_parity(sctx, | |
3285 | map, scrub_dev, ppath, | |
3286 | stripe_logical, | |
3287 | stripe_end); | |
3288 | if (ret) | |
3289 | goto out; | |
3290 | goto loop; | |
3291 | } | |
3b080b25 WS |
3292 | } else { |
3293 | physical += map->stripe_len; | |
3294 | logical += increment; | |
3295 | } | |
625f1c8d LB |
3296 | if (logical < key.objectid + bytes) { |
3297 | cond_resched(); | |
3298 | goto again; | |
3299 | } | |
3300 | ||
3b080b25 | 3301 | if (physical >= physical_end) { |
625f1c8d LB |
3302 | stop_loop = 1; |
3303 | break; | |
3304 | } | |
3305 | } | |
a2de733c AJ |
3306 | next: |
3307 | path->slots[0]++; | |
3308 | } | |
71267333 | 3309 | btrfs_release_path(path); |
3b080b25 | 3310 | skip: |
a2de733c AJ |
3311 | logical += increment; |
3312 | physical += map->stripe_len; | |
d9d181c1 | 3313 | spin_lock(&sctx->stat_lock); |
625f1c8d LB |
3314 | if (stop_loop) |
3315 | sctx->stat.last_physical = map->stripes[num].physical + | |
3316 | length; | |
3317 | else | |
3318 | sctx->stat.last_physical = physical; | |
d9d181c1 | 3319 | spin_unlock(&sctx->stat_lock); |
625f1c8d LB |
3320 | if (stop_loop) |
3321 | break; | |
a2de733c | 3322 | } |
ff023aac | 3323 | out: |
a2de733c | 3324 | /* push queued extents */ |
d9d181c1 | 3325 | scrub_submit(sctx); |
ff023aac SB |
3326 | mutex_lock(&sctx->wr_ctx.wr_lock); |
3327 | scrub_wr_submit(sctx); | |
3328 | mutex_unlock(&sctx->wr_ctx.wr_lock); | |
a2de733c | 3329 | |
e7786c3a | 3330 | blk_finish_plug(&plug); |
a2de733c | 3331 | btrfs_free_path(path); |
5a6ac9ea | 3332 | btrfs_free_path(ppath); |
a2de733c AJ |
3333 | return ret < 0 ? ret : 0; |
3334 | } | |
3335 | ||
d9d181c1 | 3336 | static noinline_for_stack int scrub_chunk(struct scrub_ctx *sctx, |
a36cf8b8 SB |
3337 | struct btrfs_device *scrub_dev, |
3338 | u64 chunk_tree, u64 chunk_objectid, | |
3339 | u64 chunk_offset, u64 length, | |
ff023aac | 3340 | u64 dev_offset, int is_dev_replace) |
a2de733c AJ |
3341 | { |
3342 | struct btrfs_mapping_tree *map_tree = | |
a36cf8b8 | 3343 | &sctx->dev_root->fs_info->mapping_tree; |
a2de733c AJ |
3344 | struct map_lookup *map; |
3345 | struct extent_map *em; | |
3346 | int i; | |
ff023aac | 3347 | int ret = 0; |
a2de733c AJ |
3348 | |
3349 | read_lock(&map_tree->map_tree.lock); | |
3350 | em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1); | |
3351 | read_unlock(&map_tree->map_tree.lock); | |
3352 | ||
3353 | if (!em) | |
3354 | return -EINVAL; | |
3355 | ||
3356 | map = (struct map_lookup *)em->bdev; | |
3357 | if (em->start != chunk_offset) | |
3358 | goto out; | |
3359 | ||
3360 | if (em->len < length) | |
3361 | goto out; | |
3362 | ||
3363 | for (i = 0; i < map->num_stripes; ++i) { | |
a36cf8b8 | 3364 | if (map->stripes[i].dev->bdev == scrub_dev->bdev && |
859acaf1 | 3365 | map->stripes[i].physical == dev_offset) { |
a36cf8b8 | 3366 | ret = scrub_stripe(sctx, map, scrub_dev, i, |
ff023aac SB |
3367 | chunk_offset, length, |
3368 | is_dev_replace); | |
a2de733c AJ |
3369 | if (ret) |
3370 | goto out; | |
3371 | } | |
3372 | } | |
3373 | out: | |
3374 | free_extent_map(em); | |
3375 | ||
3376 | return ret; | |
3377 | } | |
3378 | ||
3379 | static noinline_for_stack | |
a36cf8b8 | 3380 | int scrub_enumerate_chunks(struct scrub_ctx *sctx, |
ff023aac SB |
3381 | struct btrfs_device *scrub_dev, u64 start, u64 end, |
3382 | int is_dev_replace) | |
a2de733c AJ |
3383 | { |
3384 | struct btrfs_dev_extent *dev_extent = NULL; | |
3385 | struct btrfs_path *path; | |
a36cf8b8 | 3386 | struct btrfs_root *root = sctx->dev_root; |
a2de733c AJ |
3387 | struct btrfs_fs_info *fs_info = root->fs_info; |
3388 | u64 length; | |
3389 | u64 chunk_tree; | |
3390 | u64 chunk_objectid; | |
3391 | u64 chunk_offset; | |
3392 | int ret; | |
3393 | int slot; | |
3394 | struct extent_buffer *l; | |
3395 | struct btrfs_key key; | |
3396 | struct btrfs_key found_key; | |
3397 | struct btrfs_block_group_cache *cache; | |
ff023aac | 3398 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
a2de733c AJ |
3399 | |
3400 | path = btrfs_alloc_path(); | |
3401 | if (!path) | |
3402 | return -ENOMEM; | |
3403 | ||
3404 | path->reada = 2; | |
3405 | path->search_commit_root = 1; | |
3406 | path->skip_locking = 1; | |
3407 | ||
a36cf8b8 | 3408 | key.objectid = scrub_dev->devid; |
a2de733c AJ |
3409 | key.offset = 0ull; |
3410 | key.type = BTRFS_DEV_EXTENT_KEY; | |
3411 | ||
a2de733c AJ |
3412 | while (1) { |
3413 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
3414 | if (ret < 0) | |
8c51032f AJ |
3415 | break; |
3416 | if (ret > 0) { | |
3417 | if (path->slots[0] >= | |
3418 | btrfs_header_nritems(path->nodes[0])) { | |
3419 | ret = btrfs_next_leaf(root, path); | |
3420 | if (ret) | |
3421 | break; | |
3422 | } | |
3423 | } | |
a2de733c AJ |
3424 | |
3425 | l = path->nodes[0]; | |
3426 | slot = path->slots[0]; | |
3427 | ||
3428 | btrfs_item_key_to_cpu(l, &found_key, slot); | |
3429 | ||
a36cf8b8 | 3430 | if (found_key.objectid != scrub_dev->devid) |
a2de733c AJ |
3431 | break; |
3432 | ||
962a298f | 3433 | if (found_key.type != BTRFS_DEV_EXTENT_KEY) |
a2de733c AJ |
3434 | break; |
3435 | ||
3436 | if (found_key.offset >= end) | |
3437 | break; | |
3438 | ||
3439 | if (found_key.offset < key.offset) | |
3440 | break; | |
3441 | ||
3442 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
3443 | length = btrfs_dev_extent_length(l, dev_extent); | |
3444 | ||
ced96edc QW |
3445 | if (found_key.offset + length <= start) |
3446 | goto skip; | |
a2de733c AJ |
3447 | |
3448 | chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent); | |
3449 | chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent); | |
3450 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); | |
3451 | ||
3452 | /* | |
3453 | * get a reference on the corresponding block group to prevent | |
3454 | * the chunk from going away while we scrub it | |
3455 | */ | |
3456 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
ced96edc QW |
3457 | |
3458 | /* some chunks are removed but not committed to disk yet, | |
3459 | * continue scrubbing */ | |
3460 | if (!cache) | |
3461 | goto skip; | |
3462 | ||
ff023aac SB |
3463 | dev_replace->cursor_right = found_key.offset + length; |
3464 | dev_replace->cursor_left = found_key.offset; | |
3465 | dev_replace->item_needs_writeback = 1; | |
a36cf8b8 | 3466 | ret = scrub_chunk(sctx, scrub_dev, chunk_tree, chunk_objectid, |
ff023aac SB |
3467 | chunk_offset, length, found_key.offset, |
3468 | is_dev_replace); | |
3469 | ||
3470 | /* | |
3471 | * flush, submit all pending read and write bios, afterwards | |
3472 | * wait for them. | |
3473 | * Note that in the dev replace case, a read request causes | |
3474 | * write requests that are submitted in the read completion | |
3475 | * worker. Therefore in the current situation, it is required | |
3476 | * that all write requests are flushed, so that all read and | |
3477 | * write requests are really completed when bios_in_flight | |
3478 | * changes to 0. | |
3479 | */ | |
3480 | atomic_set(&sctx->wr_ctx.flush_all_writes, 1); | |
3481 | scrub_submit(sctx); | |
3482 | mutex_lock(&sctx->wr_ctx.wr_lock); | |
3483 | scrub_wr_submit(sctx); | |
3484 | mutex_unlock(&sctx->wr_ctx.wr_lock); | |
3485 | ||
3486 | wait_event(sctx->list_wait, | |
3487 | atomic_read(&sctx->bios_in_flight) == 0); | |
12cf9372 WS |
3488 | atomic_inc(&fs_info->scrubs_paused); |
3489 | wake_up(&fs_info->scrub_pause_wait); | |
3490 | ||
3491 | /* | |
3492 | * must be called before we decrease @scrub_paused. | |
3493 | * make sure we don't block transaction commit while | |
3494 | * we are waiting pending workers finished. | |
3495 | */ | |
ff023aac SB |
3496 | wait_event(sctx->list_wait, |
3497 | atomic_read(&sctx->workers_pending) == 0); | |
12cf9372 WS |
3498 | atomic_set(&sctx->wr_ctx.flush_all_writes, 0); |
3499 | ||
3500 | mutex_lock(&fs_info->scrub_lock); | |
3501 | __scrub_blocked_if_needed(fs_info); | |
3502 | atomic_dec(&fs_info->scrubs_paused); | |
3503 | mutex_unlock(&fs_info->scrub_lock); | |
3504 | wake_up(&fs_info->scrub_pause_wait); | |
ff023aac | 3505 | |
a2de733c AJ |
3506 | btrfs_put_block_group(cache); |
3507 | if (ret) | |
3508 | break; | |
af1be4f8 SB |
3509 | if (is_dev_replace && |
3510 | atomic64_read(&dev_replace->num_write_errors) > 0) { | |
ff023aac SB |
3511 | ret = -EIO; |
3512 | break; | |
3513 | } | |
3514 | if (sctx->stat.malloc_errors > 0) { | |
3515 | ret = -ENOMEM; | |
3516 | break; | |
3517 | } | |
a2de733c | 3518 | |
539f358a ID |
3519 | dev_replace->cursor_left = dev_replace->cursor_right; |
3520 | dev_replace->item_needs_writeback = 1; | |
ced96edc | 3521 | skip: |
a2de733c | 3522 | key.offset = found_key.offset + length; |
71267333 | 3523 | btrfs_release_path(path); |
a2de733c AJ |
3524 | } |
3525 | ||
a2de733c | 3526 | btrfs_free_path(path); |
8c51032f AJ |
3527 | |
3528 | /* | |
3529 | * ret can still be 1 from search_slot or next_leaf, | |
3530 | * that's not an error | |
3531 | */ | |
3532 | return ret < 0 ? ret : 0; | |
a2de733c AJ |
3533 | } |
3534 | ||
a36cf8b8 SB |
3535 | static noinline_for_stack int scrub_supers(struct scrub_ctx *sctx, |
3536 | struct btrfs_device *scrub_dev) | |
a2de733c AJ |
3537 | { |
3538 | int i; | |
3539 | u64 bytenr; | |
3540 | u64 gen; | |
3541 | int ret; | |
a36cf8b8 | 3542 | struct btrfs_root *root = sctx->dev_root; |
a2de733c | 3543 | |
87533c47 | 3544 | if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
79787eaa JM |
3545 | return -EIO; |
3546 | ||
5f546063 MX |
3547 | /* Seed devices of a new filesystem has their own generation. */ |
3548 | if (scrub_dev->fs_devices != root->fs_info->fs_devices) | |
3549 | gen = scrub_dev->generation; | |
3550 | else | |
3551 | gen = root->fs_info->last_trans_committed; | |
a2de733c AJ |
3552 | |
3553 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { | |
3554 | bytenr = btrfs_sb_offset(i); | |
935e5cc9 MX |
3555 | if (bytenr + BTRFS_SUPER_INFO_SIZE > |
3556 | scrub_dev->commit_total_bytes) | |
a2de733c AJ |
3557 | break; |
3558 | ||
d9d181c1 | 3559 | ret = scrub_pages(sctx, bytenr, BTRFS_SUPER_INFO_SIZE, bytenr, |
a36cf8b8 | 3560 | scrub_dev, BTRFS_EXTENT_FLAG_SUPER, gen, i, |
ff023aac | 3561 | NULL, 1, bytenr); |
a2de733c AJ |
3562 | if (ret) |
3563 | return ret; | |
3564 | } | |
b6bfebc1 | 3565 | wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0); |
a2de733c AJ |
3566 | |
3567 | return 0; | |
3568 | } | |
3569 | ||
3570 | /* | |
3571 | * get a reference count on fs_info->scrub_workers. start worker if necessary | |
3572 | */ | |
ff023aac SB |
3573 | static noinline_for_stack int scrub_workers_get(struct btrfs_fs_info *fs_info, |
3574 | int is_dev_replace) | |
a2de733c | 3575 | { |
0dc3b84a | 3576 | int ret = 0; |
0339ef2f QW |
3577 | int flags = WQ_FREEZABLE | WQ_UNBOUND; |
3578 | int max_active = fs_info->thread_pool_size; | |
a2de733c | 3579 | |
632dd772 | 3580 | if (fs_info->scrub_workers_refcnt == 0) { |
ff023aac | 3581 | if (is_dev_replace) |
0339ef2f QW |
3582 | fs_info->scrub_workers = |
3583 | btrfs_alloc_workqueue("btrfs-scrub", flags, | |
3584 | 1, 4); | |
ff023aac | 3585 | else |
0339ef2f QW |
3586 | fs_info->scrub_workers = |
3587 | btrfs_alloc_workqueue("btrfs-scrub", flags, | |
3588 | max_active, 4); | |
3589 | if (!fs_info->scrub_workers) { | |
3590 | ret = -ENOMEM; | |
0dc3b84a | 3591 | goto out; |
0339ef2f QW |
3592 | } |
3593 | fs_info->scrub_wr_completion_workers = | |
3594 | btrfs_alloc_workqueue("btrfs-scrubwrc", flags, | |
3595 | max_active, 2); | |
3596 | if (!fs_info->scrub_wr_completion_workers) { | |
3597 | ret = -ENOMEM; | |
ff023aac | 3598 | goto out; |
0339ef2f QW |
3599 | } |
3600 | fs_info->scrub_nocow_workers = | |
3601 | btrfs_alloc_workqueue("btrfs-scrubnc", flags, 1, 0); | |
3602 | if (!fs_info->scrub_nocow_workers) { | |
3603 | ret = -ENOMEM; | |
ff023aac | 3604 | goto out; |
0339ef2f | 3605 | } |
632dd772 | 3606 | } |
a2de733c | 3607 | ++fs_info->scrub_workers_refcnt; |
0dc3b84a | 3608 | out: |
0dc3b84a | 3609 | return ret; |
a2de733c AJ |
3610 | } |
3611 | ||
aa1b8cd4 | 3612 | static noinline_for_stack void scrub_workers_put(struct btrfs_fs_info *fs_info) |
a2de733c | 3613 | { |
ff023aac | 3614 | if (--fs_info->scrub_workers_refcnt == 0) { |
0339ef2f QW |
3615 | btrfs_destroy_workqueue(fs_info->scrub_workers); |
3616 | btrfs_destroy_workqueue(fs_info->scrub_wr_completion_workers); | |
3617 | btrfs_destroy_workqueue(fs_info->scrub_nocow_workers); | |
ff023aac | 3618 | } |
a2de733c | 3619 | WARN_ON(fs_info->scrub_workers_refcnt < 0); |
a2de733c AJ |
3620 | } |
3621 | ||
aa1b8cd4 SB |
3622 | int btrfs_scrub_dev(struct btrfs_fs_info *fs_info, u64 devid, u64 start, |
3623 | u64 end, struct btrfs_scrub_progress *progress, | |
63a212ab | 3624 | int readonly, int is_dev_replace) |
a2de733c | 3625 | { |
d9d181c1 | 3626 | struct scrub_ctx *sctx; |
a2de733c AJ |
3627 | int ret; |
3628 | struct btrfs_device *dev; | |
5d68da3b | 3629 | struct rcu_string *name; |
a2de733c | 3630 | |
aa1b8cd4 | 3631 | if (btrfs_fs_closing(fs_info)) |
a2de733c AJ |
3632 | return -EINVAL; |
3633 | ||
aa1b8cd4 | 3634 | if (fs_info->chunk_root->nodesize > BTRFS_STRIPE_LEN) { |
b5d67f64 SB |
3635 | /* |
3636 | * in this case scrub is unable to calculate the checksum | |
3637 | * the way scrub is implemented. Do not handle this | |
3638 | * situation at all because it won't ever happen. | |
3639 | */ | |
efe120a0 FH |
3640 | btrfs_err(fs_info, |
3641 | "scrub: size assumption nodesize <= BTRFS_STRIPE_LEN (%d <= %d) fails", | |
aa1b8cd4 | 3642 | fs_info->chunk_root->nodesize, BTRFS_STRIPE_LEN); |
b5d67f64 SB |
3643 | return -EINVAL; |
3644 | } | |
3645 | ||
aa1b8cd4 | 3646 | if (fs_info->chunk_root->sectorsize != PAGE_SIZE) { |
b5d67f64 | 3647 | /* not supported for data w/o checksums */ |
efe120a0 FH |
3648 | btrfs_err(fs_info, |
3649 | "scrub: size assumption sectorsize != PAGE_SIZE " | |
3650 | "(%d != %lu) fails", | |
27f9f023 | 3651 | fs_info->chunk_root->sectorsize, PAGE_SIZE); |
a2de733c AJ |
3652 | return -EINVAL; |
3653 | } | |
3654 | ||
7a9e9987 SB |
3655 | if (fs_info->chunk_root->nodesize > |
3656 | PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK || | |
3657 | fs_info->chunk_root->sectorsize > | |
3658 | PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK) { | |
3659 | /* | |
3660 | * would exhaust the array bounds of pagev member in | |
3661 | * struct scrub_block | |
3662 | */ | |
efe120a0 FH |
3663 | btrfs_err(fs_info, "scrub: size assumption nodesize and sectorsize " |
3664 | "<= SCRUB_MAX_PAGES_PER_BLOCK (%d <= %d && %d <= %d) fails", | |
7a9e9987 SB |
3665 | fs_info->chunk_root->nodesize, |
3666 | SCRUB_MAX_PAGES_PER_BLOCK, | |
3667 | fs_info->chunk_root->sectorsize, | |
3668 | SCRUB_MAX_PAGES_PER_BLOCK); | |
3669 | return -EINVAL; | |
3670 | } | |
3671 | ||
a2de733c | 3672 | |
aa1b8cd4 SB |
3673 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
3674 | dev = btrfs_find_device(fs_info, devid, NULL, NULL); | |
63a212ab | 3675 | if (!dev || (dev->missing && !is_dev_replace)) { |
aa1b8cd4 | 3676 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
a2de733c AJ |
3677 | return -ENODEV; |
3678 | } | |
a2de733c | 3679 | |
5d68da3b MX |
3680 | if (!is_dev_replace && !readonly && !dev->writeable) { |
3681 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
3682 | rcu_read_lock(); | |
3683 | name = rcu_dereference(dev->name); | |
3684 | btrfs_err(fs_info, "scrub: device %s is not writable", | |
3685 | name->str); | |
3686 | rcu_read_unlock(); | |
3687 | return -EROFS; | |
3688 | } | |
3689 | ||
3b7a016f | 3690 | mutex_lock(&fs_info->scrub_lock); |
63a212ab | 3691 | if (!dev->in_fs_metadata || dev->is_tgtdev_for_dev_replace) { |
a2de733c | 3692 | mutex_unlock(&fs_info->scrub_lock); |
aa1b8cd4 | 3693 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
aa1b8cd4 | 3694 | return -EIO; |
a2de733c AJ |
3695 | } |
3696 | ||
8dabb742 SB |
3697 | btrfs_dev_replace_lock(&fs_info->dev_replace); |
3698 | if (dev->scrub_device || | |
3699 | (!is_dev_replace && | |
3700 | btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))) { | |
3701 | btrfs_dev_replace_unlock(&fs_info->dev_replace); | |
a2de733c | 3702 | mutex_unlock(&fs_info->scrub_lock); |
aa1b8cd4 | 3703 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
a2de733c AJ |
3704 | return -EINPROGRESS; |
3705 | } | |
8dabb742 | 3706 | btrfs_dev_replace_unlock(&fs_info->dev_replace); |
3b7a016f WS |
3707 | |
3708 | ret = scrub_workers_get(fs_info, is_dev_replace); | |
3709 | if (ret) { | |
3710 | mutex_unlock(&fs_info->scrub_lock); | |
3711 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
3712 | return ret; | |
3713 | } | |
3714 | ||
63a212ab | 3715 | sctx = scrub_setup_ctx(dev, is_dev_replace); |
d9d181c1 | 3716 | if (IS_ERR(sctx)) { |
a2de733c | 3717 | mutex_unlock(&fs_info->scrub_lock); |
aa1b8cd4 SB |
3718 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
3719 | scrub_workers_put(fs_info); | |
d9d181c1 | 3720 | return PTR_ERR(sctx); |
a2de733c | 3721 | } |
d9d181c1 SB |
3722 | sctx->readonly = readonly; |
3723 | dev->scrub_device = sctx; | |
3cb0929a | 3724 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
a2de733c | 3725 | |
3cb0929a WS |
3726 | /* |
3727 | * checking @scrub_pause_req here, we can avoid | |
3728 | * race between committing transaction and scrubbing. | |
3729 | */ | |
cb7ab021 | 3730 | __scrub_blocked_if_needed(fs_info); |
a2de733c AJ |
3731 | atomic_inc(&fs_info->scrubs_running); |
3732 | mutex_unlock(&fs_info->scrub_lock); | |
a2de733c | 3733 | |
ff023aac | 3734 | if (!is_dev_replace) { |
9b011adf WS |
3735 | /* |
3736 | * by holding device list mutex, we can | |
3737 | * kick off writing super in log tree sync. | |
3738 | */ | |
3cb0929a | 3739 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
ff023aac | 3740 | ret = scrub_supers(sctx, dev); |
3cb0929a | 3741 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
ff023aac | 3742 | } |
a2de733c AJ |
3743 | |
3744 | if (!ret) | |
ff023aac SB |
3745 | ret = scrub_enumerate_chunks(sctx, dev, start, end, |
3746 | is_dev_replace); | |
a2de733c | 3747 | |
b6bfebc1 | 3748 | wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0); |
a2de733c AJ |
3749 | atomic_dec(&fs_info->scrubs_running); |
3750 | wake_up(&fs_info->scrub_pause_wait); | |
3751 | ||
b6bfebc1 | 3752 | wait_event(sctx->list_wait, atomic_read(&sctx->workers_pending) == 0); |
0ef8e451 | 3753 | |
a2de733c | 3754 | if (progress) |
d9d181c1 | 3755 | memcpy(progress, &sctx->stat, sizeof(*progress)); |
a2de733c AJ |
3756 | |
3757 | mutex_lock(&fs_info->scrub_lock); | |
3758 | dev->scrub_device = NULL; | |
3b7a016f | 3759 | scrub_workers_put(fs_info); |
a2de733c AJ |
3760 | mutex_unlock(&fs_info->scrub_lock); |
3761 | ||
d9d181c1 | 3762 | scrub_free_ctx(sctx); |
a2de733c AJ |
3763 | |
3764 | return ret; | |
3765 | } | |
3766 | ||
143bede5 | 3767 | void btrfs_scrub_pause(struct btrfs_root *root) |
a2de733c AJ |
3768 | { |
3769 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3770 | ||
3771 | mutex_lock(&fs_info->scrub_lock); | |
3772 | atomic_inc(&fs_info->scrub_pause_req); | |
3773 | while (atomic_read(&fs_info->scrubs_paused) != | |
3774 | atomic_read(&fs_info->scrubs_running)) { | |
3775 | mutex_unlock(&fs_info->scrub_lock); | |
3776 | wait_event(fs_info->scrub_pause_wait, | |
3777 | atomic_read(&fs_info->scrubs_paused) == | |
3778 | atomic_read(&fs_info->scrubs_running)); | |
3779 | mutex_lock(&fs_info->scrub_lock); | |
3780 | } | |
3781 | mutex_unlock(&fs_info->scrub_lock); | |
a2de733c AJ |
3782 | } |
3783 | ||
143bede5 | 3784 | void btrfs_scrub_continue(struct btrfs_root *root) |
a2de733c AJ |
3785 | { |
3786 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3787 | ||
3788 | atomic_dec(&fs_info->scrub_pause_req); | |
3789 | wake_up(&fs_info->scrub_pause_wait); | |
a2de733c AJ |
3790 | } |
3791 | ||
aa1b8cd4 | 3792 | int btrfs_scrub_cancel(struct btrfs_fs_info *fs_info) |
a2de733c | 3793 | { |
a2de733c AJ |
3794 | mutex_lock(&fs_info->scrub_lock); |
3795 | if (!atomic_read(&fs_info->scrubs_running)) { | |
3796 | mutex_unlock(&fs_info->scrub_lock); | |
3797 | return -ENOTCONN; | |
3798 | } | |
3799 | ||
3800 | atomic_inc(&fs_info->scrub_cancel_req); | |
3801 | while (atomic_read(&fs_info->scrubs_running)) { | |
3802 | mutex_unlock(&fs_info->scrub_lock); | |
3803 | wait_event(fs_info->scrub_pause_wait, | |
3804 | atomic_read(&fs_info->scrubs_running) == 0); | |
3805 | mutex_lock(&fs_info->scrub_lock); | |
3806 | } | |
3807 | atomic_dec(&fs_info->scrub_cancel_req); | |
3808 | mutex_unlock(&fs_info->scrub_lock); | |
3809 | ||
3810 | return 0; | |
3811 | } | |
3812 | ||
aa1b8cd4 SB |
3813 | int btrfs_scrub_cancel_dev(struct btrfs_fs_info *fs_info, |
3814 | struct btrfs_device *dev) | |
49b25e05 | 3815 | { |
d9d181c1 | 3816 | struct scrub_ctx *sctx; |
a2de733c AJ |
3817 | |
3818 | mutex_lock(&fs_info->scrub_lock); | |
d9d181c1 SB |
3819 | sctx = dev->scrub_device; |
3820 | if (!sctx) { | |
a2de733c AJ |
3821 | mutex_unlock(&fs_info->scrub_lock); |
3822 | return -ENOTCONN; | |
3823 | } | |
d9d181c1 | 3824 | atomic_inc(&sctx->cancel_req); |
a2de733c AJ |
3825 | while (dev->scrub_device) { |
3826 | mutex_unlock(&fs_info->scrub_lock); | |
3827 | wait_event(fs_info->scrub_pause_wait, | |
3828 | dev->scrub_device == NULL); | |
3829 | mutex_lock(&fs_info->scrub_lock); | |
3830 | } | |
3831 | mutex_unlock(&fs_info->scrub_lock); | |
3832 | ||
3833 | return 0; | |
3834 | } | |
1623edeb | 3835 | |
a2de733c AJ |
3836 | int btrfs_scrub_progress(struct btrfs_root *root, u64 devid, |
3837 | struct btrfs_scrub_progress *progress) | |
3838 | { | |
3839 | struct btrfs_device *dev; | |
d9d181c1 | 3840 | struct scrub_ctx *sctx = NULL; |
a2de733c AJ |
3841 | |
3842 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
aa1b8cd4 | 3843 | dev = btrfs_find_device(root->fs_info, devid, NULL, NULL); |
a2de733c | 3844 | if (dev) |
d9d181c1 SB |
3845 | sctx = dev->scrub_device; |
3846 | if (sctx) | |
3847 | memcpy(progress, &sctx->stat, sizeof(*progress)); | |
a2de733c AJ |
3848 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
3849 | ||
d9d181c1 | 3850 | return dev ? (sctx ? 0 : -ENOTCONN) : -ENODEV; |
a2de733c | 3851 | } |
ff023aac SB |
3852 | |
3853 | static void scrub_remap_extent(struct btrfs_fs_info *fs_info, | |
3854 | u64 extent_logical, u64 extent_len, | |
3855 | u64 *extent_physical, | |
3856 | struct btrfs_device **extent_dev, | |
3857 | int *extent_mirror_num) | |
3858 | { | |
3859 | u64 mapped_length; | |
3860 | struct btrfs_bio *bbio = NULL; | |
3861 | int ret; | |
3862 | ||
3863 | mapped_length = extent_len; | |
3864 | ret = btrfs_map_block(fs_info, READ, extent_logical, | |
3865 | &mapped_length, &bbio, 0); | |
3866 | if (ret || !bbio || mapped_length < extent_len || | |
3867 | !bbio->stripes[0].dev->bdev) { | |
6e9606d2 | 3868 | btrfs_put_bbio(bbio); |
ff023aac SB |
3869 | return; |
3870 | } | |
3871 | ||
3872 | *extent_physical = bbio->stripes[0].physical; | |
3873 | *extent_mirror_num = bbio->mirror_num; | |
3874 | *extent_dev = bbio->stripes[0].dev; | |
6e9606d2 | 3875 | btrfs_put_bbio(bbio); |
ff023aac SB |
3876 | } |
3877 | ||
3878 | static int scrub_setup_wr_ctx(struct scrub_ctx *sctx, | |
3879 | struct scrub_wr_ctx *wr_ctx, | |
3880 | struct btrfs_fs_info *fs_info, | |
3881 | struct btrfs_device *dev, | |
3882 | int is_dev_replace) | |
3883 | { | |
3884 | WARN_ON(wr_ctx->wr_curr_bio != NULL); | |
3885 | ||
3886 | mutex_init(&wr_ctx->wr_lock); | |
3887 | wr_ctx->wr_curr_bio = NULL; | |
3888 | if (!is_dev_replace) | |
3889 | return 0; | |
3890 | ||
3891 | WARN_ON(!dev->bdev); | |
3892 | wr_ctx->pages_per_wr_bio = min_t(int, SCRUB_PAGES_PER_WR_BIO, | |
3893 | bio_get_nr_vecs(dev->bdev)); | |
3894 | wr_ctx->tgtdev = dev; | |
3895 | atomic_set(&wr_ctx->flush_all_writes, 0); | |
3896 | return 0; | |
3897 | } | |
3898 | ||
3899 | static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx) | |
3900 | { | |
3901 | mutex_lock(&wr_ctx->wr_lock); | |
3902 | kfree(wr_ctx->wr_curr_bio); | |
3903 | wr_ctx->wr_curr_bio = NULL; | |
3904 | mutex_unlock(&wr_ctx->wr_lock); | |
3905 | } | |
3906 | ||
3907 | static int copy_nocow_pages(struct scrub_ctx *sctx, u64 logical, u64 len, | |
3908 | int mirror_num, u64 physical_for_dev_replace) | |
3909 | { | |
3910 | struct scrub_copy_nocow_ctx *nocow_ctx; | |
3911 | struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info; | |
3912 | ||
3913 | nocow_ctx = kzalloc(sizeof(*nocow_ctx), GFP_NOFS); | |
3914 | if (!nocow_ctx) { | |
3915 | spin_lock(&sctx->stat_lock); | |
3916 | sctx->stat.malloc_errors++; | |
3917 | spin_unlock(&sctx->stat_lock); | |
3918 | return -ENOMEM; | |
3919 | } | |
3920 | ||
3921 | scrub_pending_trans_workers_inc(sctx); | |
3922 | ||
3923 | nocow_ctx->sctx = sctx; | |
3924 | nocow_ctx->logical = logical; | |
3925 | nocow_ctx->len = len; | |
3926 | nocow_ctx->mirror_num = mirror_num; | |
3927 | nocow_ctx->physical_for_dev_replace = physical_for_dev_replace; | |
9e0af237 LB |
3928 | btrfs_init_work(&nocow_ctx->work, btrfs_scrubnc_helper, |
3929 | copy_nocow_pages_worker, NULL, NULL); | |
652f25a2 | 3930 | INIT_LIST_HEAD(&nocow_ctx->inodes); |
0339ef2f QW |
3931 | btrfs_queue_work(fs_info->scrub_nocow_workers, |
3932 | &nocow_ctx->work); | |
ff023aac SB |
3933 | |
3934 | return 0; | |
3935 | } | |
3936 | ||
652f25a2 JB |
3937 | static int record_inode_for_nocow(u64 inum, u64 offset, u64 root, void *ctx) |
3938 | { | |
3939 | struct scrub_copy_nocow_ctx *nocow_ctx = ctx; | |
3940 | struct scrub_nocow_inode *nocow_inode; | |
3941 | ||
3942 | nocow_inode = kzalloc(sizeof(*nocow_inode), GFP_NOFS); | |
3943 | if (!nocow_inode) | |
3944 | return -ENOMEM; | |
3945 | nocow_inode->inum = inum; | |
3946 | nocow_inode->offset = offset; | |
3947 | nocow_inode->root = root; | |
3948 | list_add_tail(&nocow_inode->list, &nocow_ctx->inodes); | |
3949 | return 0; | |
3950 | } | |
3951 | ||
3952 | #define COPY_COMPLETE 1 | |
3953 | ||
ff023aac SB |
3954 | static void copy_nocow_pages_worker(struct btrfs_work *work) |
3955 | { | |
3956 | struct scrub_copy_nocow_ctx *nocow_ctx = | |
3957 | container_of(work, struct scrub_copy_nocow_ctx, work); | |
3958 | struct scrub_ctx *sctx = nocow_ctx->sctx; | |
3959 | u64 logical = nocow_ctx->logical; | |
3960 | u64 len = nocow_ctx->len; | |
3961 | int mirror_num = nocow_ctx->mirror_num; | |
3962 | u64 physical_for_dev_replace = nocow_ctx->physical_for_dev_replace; | |
3963 | int ret; | |
3964 | struct btrfs_trans_handle *trans = NULL; | |
3965 | struct btrfs_fs_info *fs_info; | |
3966 | struct btrfs_path *path; | |
3967 | struct btrfs_root *root; | |
3968 | int not_written = 0; | |
3969 | ||
3970 | fs_info = sctx->dev_root->fs_info; | |
3971 | root = fs_info->extent_root; | |
3972 | ||
3973 | path = btrfs_alloc_path(); | |
3974 | if (!path) { | |
3975 | spin_lock(&sctx->stat_lock); | |
3976 | sctx->stat.malloc_errors++; | |
3977 | spin_unlock(&sctx->stat_lock); | |
3978 | not_written = 1; | |
3979 | goto out; | |
3980 | } | |
3981 | ||
3982 | trans = btrfs_join_transaction(root); | |
3983 | if (IS_ERR(trans)) { | |
3984 | not_written = 1; | |
3985 | goto out; | |
3986 | } | |
3987 | ||
3988 | ret = iterate_inodes_from_logical(logical, fs_info, path, | |
652f25a2 | 3989 | record_inode_for_nocow, nocow_ctx); |
ff023aac | 3990 | if (ret != 0 && ret != -ENOENT) { |
efe120a0 FH |
3991 | btrfs_warn(fs_info, "iterate_inodes_from_logical() failed: log %llu, " |
3992 | "phys %llu, len %llu, mir %u, ret %d", | |
118a0a25 GU |
3993 | logical, physical_for_dev_replace, len, mirror_num, |
3994 | ret); | |
ff023aac SB |
3995 | not_written = 1; |
3996 | goto out; | |
3997 | } | |
3998 | ||
652f25a2 JB |
3999 | btrfs_end_transaction(trans, root); |
4000 | trans = NULL; | |
4001 | while (!list_empty(&nocow_ctx->inodes)) { | |
4002 | struct scrub_nocow_inode *entry; | |
4003 | entry = list_first_entry(&nocow_ctx->inodes, | |
4004 | struct scrub_nocow_inode, | |
4005 | list); | |
4006 | list_del_init(&entry->list); | |
4007 | ret = copy_nocow_pages_for_inode(entry->inum, entry->offset, | |
4008 | entry->root, nocow_ctx); | |
4009 | kfree(entry); | |
4010 | if (ret == COPY_COMPLETE) { | |
4011 | ret = 0; | |
4012 | break; | |
4013 | } else if (ret) { | |
4014 | break; | |
4015 | } | |
4016 | } | |
ff023aac | 4017 | out: |
652f25a2 JB |
4018 | while (!list_empty(&nocow_ctx->inodes)) { |
4019 | struct scrub_nocow_inode *entry; | |
4020 | entry = list_first_entry(&nocow_ctx->inodes, | |
4021 | struct scrub_nocow_inode, | |
4022 | list); | |
4023 | list_del_init(&entry->list); | |
4024 | kfree(entry); | |
4025 | } | |
ff023aac SB |
4026 | if (trans && !IS_ERR(trans)) |
4027 | btrfs_end_transaction(trans, root); | |
4028 | if (not_written) | |
4029 | btrfs_dev_replace_stats_inc(&fs_info->dev_replace. | |
4030 | num_uncorrectable_read_errors); | |
4031 | ||
4032 | btrfs_free_path(path); | |
4033 | kfree(nocow_ctx); | |
4034 | ||
4035 | scrub_pending_trans_workers_dec(sctx); | |
4036 | } | |
4037 | ||
32159242 GH |
4038 | static int check_extent_to_block(struct inode *inode, u64 start, u64 len, |
4039 | u64 logical) | |
4040 | { | |
4041 | struct extent_state *cached_state = NULL; | |
4042 | struct btrfs_ordered_extent *ordered; | |
4043 | struct extent_io_tree *io_tree; | |
4044 | struct extent_map *em; | |
4045 | u64 lockstart = start, lockend = start + len - 1; | |
4046 | int ret = 0; | |
4047 | ||
4048 | io_tree = &BTRFS_I(inode)->io_tree; | |
4049 | ||
4050 | lock_extent_bits(io_tree, lockstart, lockend, 0, &cached_state); | |
4051 | ordered = btrfs_lookup_ordered_range(inode, lockstart, len); | |
4052 | if (ordered) { | |
4053 | btrfs_put_ordered_extent(ordered); | |
4054 | ret = 1; | |
4055 | goto out_unlock; | |
4056 | } | |
4057 | ||
4058 | em = btrfs_get_extent(inode, NULL, 0, start, len, 0); | |
4059 | if (IS_ERR(em)) { | |
4060 | ret = PTR_ERR(em); | |
4061 | goto out_unlock; | |
4062 | } | |
4063 | ||
4064 | /* | |
4065 | * This extent does not actually cover the logical extent anymore, | |
4066 | * move on to the next inode. | |
4067 | */ | |
4068 | if (em->block_start > logical || | |
4069 | em->block_start + em->block_len < logical + len) { | |
4070 | free_extent_map(em); | |
4071 | ret = 1; | |
4072 | goto out_unlock; | |
4073 | } | |
4074 | free_extent_map(em); | |
4075 | ||
4076 | out_unlock: | |
4077 | unlock_extent_cached(io_tree, lockstart, lockend, &cached_state, | |
4078 | GFP_NOFS); | |
4079 | return ret; | |
4080 | } | |
4081 | ||
652f25a2 JB |
4082 | static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root, |
4083 | struct scrub_copy_nocow_ctx *nocow_ctx) | |
ff023aac | 4084 | { |
826aa0a8 | 4085 | struct btrfs_fs_info *fs_info = nocow_ctx->sctx->dev_root->fs_info; |
ff023aac | 4086 | struct btrfs_key key; |
826aa0a8 MX |
4087 | struct inode *inode; |
4088 | struct page *page; | |
ff023aac | 4089 | struct btrfs_root *local_root; |
652f25a2 | 4090 | struct extent_io_tree *io_tree; |
ff023aac | 4091 | u64 physical_for_dev_replace; |
32159242 | 4092 | u64 nocow_ctx_logical; |
652f25a2 | 4093 | u64 len = nocow_ctx->len; |
826aa0a8 | 4094 | unsigned long index; |
6f1c3605 | 4095 | int srcu_index; |
652f25a2 JB |
4096 | int ret = 0; |
4097 | int err = 0; | |
ff023aac SB |
4098 | |
4099 | key.objectid = root; | |
4100 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4101 | key.offset = (u64)-1; | |
6f1c3605 LB |
4102 | |
4103 | srcu_index = srcu_read_lock(&fs_info->subvol_srcu); | |
4104 | ||
ff023aac | 4105 | local_root = btrfs_read_fs_root_no_name(fs_info, &key); |
6f1c3605 LB |
4106 | if (IS_ERR(local_root)) { |
4107 | srcu_read_unlock(&fs_info->subvol_srcu, srcu_index); | |
ff023aac | 4108 | return PTR_ERR(local_root); |
6f1c3605 | 4109 | } |
ff023aac SB |
4110 | |
4111 | key.type = BTRFS_INODE_ITEM_KEY; | |
4112 | key.objectid = inum; | |
4113 | key.offset = 0; | |
4114 | inode = btrfs_iget(fs_info->sb, &key, local_root, NULL); | |
6f1c3605 | 4115 | srcu_read_unlock(&fs_info->subvol_srcu, srcu_index); |
ff023aac SB |
4116 | if (IS_ERR(inode)) |
4117 | return PTR_ERR(inode); | |
4118 | ||
edd1400b MX |
4119 | /* Avoid truncate/dio/punch hole.. */ |
4120 | mutex_lock(&inode->i_mutex); | |
4121 | inode_dio_wait(inode); | |
4122 | ||
ff023aac | 4123 | physical_for_dev_replace = nocow_ctx->physical_for_dev_replace; |
652f25a2 | 4124 | io_tree = &BTRFS_I(inode)->io_tree; |
32159242 | 4125 | nocow_ctx_logical = nocow_ctx->logical; |
652f25a2 | 4126 | |
32159242 GH |
4127 | ret = check_extent_to_block(inode, offset, len, nocow_ctx_logical); |
4128 | if (ret) { | |
4129 | ret = ret > 0 ? 0 : ret; | |
4130 | goto out; | |
652f25a2 | 4131 | } |
652f25a2 | 4132 | |
ff023aac | 4133 | while (len >= PAGE_CACHE_SIZE) { |
ff023aac | 4134 | index = offset >> PAGE_CACHE_SHIFT; |
edd1400b | 4135 | again: |
ff023aac SB |
4136 | page = find_or_create_page(inode->i_mapping, index, GFP_NOFS); |
4137 | if (!page) { | |
efe120a0 | 4138 | btrfs_err(fs_info, "find_or_create_page() failed"); |
ff023aac | 4139 | ret = -ENOMEM; |
826aa0a8 | 4140 | goto out; |
ff023aac SB |
4141 | } |
4142 | ||
4143 | if (PageUptodate(page)) { | |
4144 | if (PageDirty(page)) | |
4145 | goto next_page; | |
4146 | } else { | |
4147 | ClearPageError(page); | |
32159242 | 4148 | err = extent_read_full_page(io_tree, page, |
652f25a2 JB |
4149 | btrfs_get_extent, |
4150 | nocow_ctx->mirror_num); | |
826aa0a8 MX |
4151 | if (err) { |
4152 | ret = err; | |
ff023aac SB |
4153 | goto next_page; |
4154 | } | |
edd1400b | 4155 | |
26b25891 | 4156 | lock_page(page); |
edd1400b MX |
4157 | /* |
4158 | * If the page has been remove from the page cache, | |
4159 | * the data on it is meaningless, because it may be | |
4160 | * old one, the new data may be written into the new | |
4161 | * page in the page cache. | |
4162 | */ | |
4163 | if (page->mapping != inode->i_mapping) { | |
652f25a2 | 4164 | unlock_page(page); |
edd1400b MX |
4165 | page_cache_release(page); |
4166 | goto again; | |
4167 | } | |
ff023aac SB |
4168 | if (!PageUptodate(page)) { |
4169 | ret = -EIO; | |
4170 | goto next_page; | |
4171 | } | |
4172 | } | |
32159242 GH |
4173 | |
4174 | ret = check_extent_to_block(inode, offset, len, | |
4175 | nocow_ctx_logical); | |
4176 | if (ret) { | |
4177 | ret = ret > 0 ? 0 : ret; | |
4178 | goto next_page; | |
4179 | } | |
4180 | ||
826aa0a8 MX |
4181 | err = write_page_nocow(nocow_ctx->sctx, |
4182 | physical_for_dev_replace, page); | |
4183 | if (err) | |
4184 | ret = err; | |
ff023aac | 4185 | next_page: |
826aa0a8 MX |
4186 | unlock_page(page); |
4187 | page_cache_release(page); | |
4188 | ||
4189 | if (ret) | |
4190 | break; | |
4191 | ||
ff023aac SB |
4192 | offset += PAGE_CACHE_SIZE; |
4193 | physical_for_dev_replace += PAGE_CACHE_SIZE; | |
32159242 | 4194 | nocow_ctx_logical += PAGE_CACHE_SIZE; |
ff023aac SB |
4195 | len -= PAGE_CACHE_SIZE; |
4196 | } | |
652f25a2 | 4197 | ret = COPY_COMPLETE; |
826aa0a8 | 4198 | out: |
edd1400b | 4199 | mutex_unlock(&inode->i_mutex); |
826aa0a8 | 4200 | iput(inode); |
ff023aac SB |
4201 | return ret; |
4202 | } | |
4203 | ||
4204 | static int write_page_nocow(struct scrub_ctx *sctx, | |
4205 | u64 physical_for_dev_replace, struct page *page) | |
4206 | { | |
4207 | struct bio *bio; | |
4208 | struct btrfs_device *dev; | |
4209 | int ret; | |
ff023aac SB |
4210 | |
4211 | dev = sctx->wr_ctx.tgtdev; | |
4212 | if (!dev) | |
4213 | return -EIO; | |
4214 | if (!dev->bdev) { | |
4215 | printk_ratelimited(KERN_WARNING | |
efe120a0 | 4216 | "BTRFS: scrub write_page_nocow(bdev == NULL) is unexpected!\n"); |
ff023aac SB |
4217 | return -EIO; |
4218 | } | |
9be3395b | 4219 | bio = btrfs_io_bio_alloc(GFP_NOFS, 1); |
ff023aac SB |
4220 | if (!bio) { |
4221 | spin_lock(&sctx->stat_lock); | |
4222 | sctx->stat.malloc_errors++; | |
4223 | spin_unlock(&sctx->stat_lock); | |
4224 | return -ENOMEM; | |
4225 | } | |
4f024f37 KO |
4226 | bio->bi_iter.bi_size = 0; |
4227 | bio->bi_iter.bi_sector = physical_for_dev_replace >> 9; | |
ff023aac SB |
4228 | bio->bi_bdev = dev->bdev; |
4229 | ret = bio_add_page(bio, page, PAGE_CACHE_SIZE, 0); | |
4230 | if (ret != PAGE_CACHE_SIZE) { | |
4231 | leave_with_eio: | |
4232 | bio_put(bio); | |
4233 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS); | |
4234 | return -EIO; | |
4235 | } | |
ff023aac | 4236 | |
33879d45 | 4237 | if (btrfsic_submit_bio_wait(WRITE_SYNC, bio)) |
ff023aac SB |
4238 | goto leave_with_eio; |
4239 | ||
4240 | bio_put(bio); | |
4241 | return 0; | |
4242 | } |