]>
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); |
be50a8dd | 238 | static int scrub_setup_recheck_block(struct scrub_block *original_sblock, |
ff023aac | 239 | struct scrub_block *sblocks_for_recheck); |
34f5c8e9 SB |
240 | static void scrub_recheck_block(struct btrfs_fs_info *fs_info, |
241 | struct scrub_block *sblock, int is_metadata, | |
242 | int have_csum, u8 *csum, u64 generation, | |
af8e2d1d | 243 | u16 csum_size, int retry_failed_mirror); |
b5d67f64 SB |
244 | static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info, |
245 | struct scrub_block *sblock, | |
246 | int is_metadata, int have_csum, | |
247 | const u8 *csum, u64 generation, | |
248 | u16 csum_size); | |
b5d67f64 | 249 | static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad, |
114ab50d | 250 | struct scrub_block *sblock_good); |
b5d67f64 SB |
251 | static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad, |
252 | struct scrub_block *sblock_good, | |
253 | int page_num, int force_write); | |
ff023aac SB |
254 | static void scrub_write_block_to_dev_replace(struct scrub_block *sblock); |
255 | static int scrub_write_page_to_dev_replace(struct scrub_block *sblock, | |
256 | int page_num); | |
b5d67f64 SB |
257 | static int scrub_checksum_data(struct scrub_block *sblock); |
258 | static int scrub_checksum_tree_block(struct scrub_block *sblock); | |
259 | static int scrub_checksum_super(struct scrub_block *sblock); | |
260 | static void scrub_block_get(struct scrub_block *sblock); | |
261 | static void scrub_block_put(struct scrub_block *sblock); | |
7a9e9987 SB |
262 | static void scrub_page_get(struct scrub_page *spage); |
263 | static void scrub_page_put(struct scrub_page *spage); | |
5a6ac9ea MX |
264 | static void scrub_parity_get(struct scrub_parity *sparity); |
265 | static void scrub_parity_put(struct scrub_parity *sparity); | |
ff023aac SB |
266 | static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx, |
267 | struct scrub_page *spage); | |
d9d181c1 | 268 | static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len, |
a36cf8b8 | 269 | u64 physical, struct btrfs_device *dev, u64 flags, |
ff023aac SB |
270 | u64 gen, int mirror_num, u8 *csum, int force, |
271 | u64 physical_for_dev_replace); | |
1623edeb | 272 | static void scrub_bio_end_io(struct bio *bio, int err); |
b5d67f64 SB |
273 | static void scrub_bio_end_io_worker(struct btrfs_work *work); |
274 | static void scrub_block_complete(struct scrub_block *sblock); | |
ff023aac SB |
275 | static void scrub_remap_extent(struct btrfs_fs_info *fs_info, |
276 | u64 extent_logical, u64 extent_len, | |
277 | u64 *extent_physical, | |
278 | struct btrfs_device **extent_dev, | |
279 | int *extent_mirror_num); | |
280 | static int scrub_setup_wr_ctx(struct scrub_ctx *sctx, | |
281 | struct scrub_wr_ctx *wr_ctx, | |
282 | struct btrfs_fs_info *fs_info, | |
283 | struct btrfs_device *dev, | |
284 | int is_dev_replace); | |
285 | static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx); | |
286 | static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx, | |
287 | struct scrub_page *spage); | |
288 | static void scrub_wr_submit(struct scrub_ctx *sctx); | |
289 | static void scrub_wr_bio_end_io(struct bio *bio, int err); | |
290 | static void scrub_wr_bio_end_io_worker(struct btrfs_work *work); | |
291 | static int write_page_nocow(struct scrub_ctx *sctx, | |
292 | u64 physical_for_dev_replace, struct page *page); | |
293 | static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root, | |
652f25a2 | 294 | struct scrub_copy_nocow_ctx *ctx); |
ff023aac SB |
295 | static int copy_nocow_pages(struct scrub_ctx *sctx, u64 logical, u64 len, |
296 | int mirror_num, u64 physical_for_dev_replace); | |
297 | static void copy_nocow_pages_worker(struct btrfs_work *work); | |
cb7ab021 | 298 | static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info); |
3cb0929a | 299 | static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info); |
1623edeb SB |
300 | |
301 | ||
b6bfebc1 SB |
302 | static void scrub_pending_bio_inc(struct scrub_ctx *sctx) |
303 | { | |
304 | atomic_inc(&sctx->bios_in_flight); | |
305 | } | |
306 | ||
307 | static void scrub_pending_bio_dec(struct scrub_ctx *sctx) | |
308 | { | |
309 | atomic_dec(&sctx->bios_in_flight); | |
310 | wake_up(&sctx->list_wait); | |
311 | } | |
312 | ||
cb7ab021 | 313 | static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info) |
3cb0929a WS |
314 | { |
315 | while (atomic_read(&fs_info->scrub_pause_req)) { | |
316 | mutex_unlock(&fs_info->scrub_lock); | |
317 | wait_event(fs_info->scrub_pause_wait, | |
318 | atomic_read(&fs_info->scrub_pause_req) == 0); | |
319 | mutex_lock(&fs_info->scrub_lock); | |
320 | } | |
321 | } | |
322 | ||
cb7ab021 WS |
323 | static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info) |
324 | { | |
325 | atomic_inc(&fs_info->scrubs_paused); | |
326 | wake_up(&fs_info->scrub_pause_wait); | |
327 | ||
328 | mutex_lock(&fs_info->scrub_lock); | |
329 | __scrub_blocked_if_needed(fs_info); | |
330 | atomic_dec(&fs_info->scrubs_paused); | |
331 | mutex_unlock(&fs_info->scrub_lock); | |
332 | ||
333 | wake_up(&fs_info->scrub_pause_wait); | |
334 | } | |
335 | ||
b6bfebc1 SB |
336 | /* |
337 | * used for workers that require transaction commits (i.e., for the | |
338 | * NOCOW case) | |
339 | */ | |
340 | static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx) | |
341 | { | |
342 | struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info; | |
343 | ||
344 | /* | |
345 | * increment scrubs_running to prevent cancel requests from | |
346 | * completing as long as a worker is running. we must also | |
347 | * increment scrubs_paused to prevent deadlocking on pause | |
348 | * requests used for transactions commits (as the worker uses a | |
349 | * transaction context). it is safe to regard the worker | |
350 | * as paused for all matters practical. effectively, we only | |
351 | * avoid cancellation requests from completing. | |
352 | */ | |
353 | mutex_lock(&fs_info->scrub_lock); | |
354 | atomic_inc(&fs_info->scrubs_running); | |
355 | atomic_inc(&fs_info->scrubs_paused); | |
356 | mutex_unlock(&fs_info->scrub_lock); | |
32a44789 WS |
357 | |
358 | /* | |
359 | * check if @scrubs_running=@scrubs_paused condition | |
360 | * inside wait_event() is not an atomic operation. | |
361 | * which means we may inc/dec @scrub_running/paused | |
362 | * at any time. Let's wake up @scrub_pause_wait as | |
363 | * much as we can to let commit transaction blocked less. | |
364 | */ | |
365 | wake_up(&fs_info->scrub_pause_wait); | |
366 | ||
b6bfebc1 SB |
367 | atomic_inc(&sctx->workers_pending); |
368 | } | |
369 | ||
370 | /* used for workers that require transaction commits */ | |
371 | static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx) | |
372 | { | |
373 | struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info; | |
374 | ||
375 | /* | |
376 | * see scrub_pending_trans_workers_inc() why we're pretending | |
377 | * to be paused in the scrub counters | |
378 | */ | |
379 | mutex_lock(&fs_info->scrub_lock); | |
380 | atomic_dec(&fs_info->scrubs_running); | |
381 | atomic_dec(&fs_info->scrubs_paused); | |
382 | mutex_unlock(&fs_info->scrub_lock); | |
383 | atomic_dec(&sctx->workers_pending); | |
384 | wake_up(&fs_info->scrub_pause_wait); | |
385 | wake_up(&sctx->list_wait); | |
386 | } | |
387 | ||
d9d181c1 | 388 | static void scrub_free_csums(struct scrub_ctx *sctx) |
a2de733c | 389 | { |
d9d181c1 | 390 | while (!list_empty(&sctx->csum_list)) { |
a2de733c | 391 | struct btrfs_ordered_sum *sum; |
d9d181c1 | 392 | sum = list_first_entry(&sctx->csum_list, |
a2de733c AJ |
393 | struct btrfs_ordered_sum, list); |
394 | list_del(&sum->list); | |
395 | kfree(sum); | |
396 | } | |
397 | } | |
398 | ||
d9d181c1 | 399 | static noinline_for_stack void scrub_free_ctx(struct scrub_ctx *sctx) |
a2de733c AJ |
400 | { |
401 | int i; | |
a2de733c | 402 | |
d9d181c1 | 403 | if (!sctx) |
a2de733c AJ |
404 | return; |
405 | ||
ff023aac SB |
406 | scrub_free_wr_ctx(&sctx->wr_ctx); |
407 | ||
b5d67f64 | 408 | /* this can happen when scrub is cancelled */ |
d9d181c1 SB |
409 | if (sctx->curr != -1) { |
410 | struct scrub_bio *sbio = sctx->bios[sctx->curr]; | |
b5d67f64 SB |
411 | |
412 | for (i = 0; i < sbio->page_count; i++) { | |
ff023aac | 413 | WARN_ON(!sbio->pagev[i]->page); |
b5d67f64 SB |
414 | scrub_block_put(sbio->pagev[i]->sblock); |
415 | } | |
416 | bio_put(sbio->bio); | |
417 | } | |
418 | ||
ff023aac | 419 | for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) { |
d9d181c1 | 420 | struct scrub_bio *sbio = sctx->bios[i]; |
a2de733c AJ |
421 | |
422 | if (!sbio) | |
423 | break; | |
a2de733c AJ |
424 | kfree(sbio); |
425 | } | |
426 | ||
d9d181c1 SB |
427 | scrub_free_csums(sctx); |
428 | kfree(sctx); | |
a2de733c AJ |
429 | } |
430 | ||
431 | static noinline_for_stack | |
63a212ab | 432 | struct scrub_ctx *scrub_setup_ctx(struct btrfs_device *dev, int is_dev_replace) |
a2de733c | 433 | { |
d9d181c1 | 434 | struct scrub_ctx *sctx; |
a2de733c | 435 | int i; |
a2de733c | 436 | struct btrfs_fs_info *fs_info = dev->dev_root->fs_info; |
ff023aac SB |
437 | int pages_per_rd_bio; |
438 | int ret; | |
a2de733c | 439 | |
ff023aac SB |
440 | /* |
441 | * the setting of pages_per_rd_bio is correct for scrub but might | |
442 | * be wrong for the dev_replace code where we might read from | |
443 | * different devices in the initial huge bios. However, that | |
444 | * code is able to correctly handle the case when adding a page | |
445 | * to a bio fails. | |
446 | */ | |
447 | if (dev->bdev) | |
448 | pages_per_rd_bio = min_t(int, SCRUB_PAGES_PER_RD_BIO, | |
449 | bio_get_nr_vecs(dev->bdev)); | |
450 | else | |
451 | pages_per_rd_bio = SCRUB_PAGES_PER_RD_BIO; | |
d9d181c1 SB |
452 | sctx = kzalloc(sizeof(*sctx), GFP_NOFS); |
453 | if (!sctx) | |
a2de733c | 454 | goto nomem; |
63a212ab | 455 | sctx->is_dev_replace = is_dev_replace; |
ff023aac | 456 | sctx->pages_per_rd_bio = pages_per_rd_bio; |
d9d181c1 | 457 | sctx->curr = -1; |
a36cf8b8 | 458 | sctx->dev_root = dev->dev_root; |
ff023aac | 459 | for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) { |
a2de733c AJ |
460 | struct scrub_bio *sbio; |
461 | ||
462 | sbio = kzalloc(sizeof(*sbio), GFP_NOFS); | |
463 | if (!sbio) | |
464 | goto nomem; | |
d9d181c1 | 465 | sctx->bios[i] = sbio; |
a2de733c | 466 | |
a2de733c | 467 | sbio->index = i; |
d9d181c1 | 468 | sbio->sctx = sctx; |
b5d67f64 | 469 | sbio->page_count = 0; |
9e0af237 LB |
470 | btrfs_init_work(&sbio->work, btrfs_scrub_helper, |
471 | scrub_bio_end_io_worker, NULL, NULL); | |
a2de733c | 472 | |
ff023aac | 473 | if (i != SCRUB_BIOS_PER_SCTX - 1) |
d9d181c1 | 474 | sctx->bios[i]->next_free = i + 1; |
0ef8e451 | 475 | else |
d9d181c1 SB |
476 | sctx->bios[i]->next_free = -1; |
477 | } | |
478 | sctx->first_free = 0; | |
479 | sctx->nodesize = dev->dev_root->nodesize; | |
d9d181c1 | 480 | sctx->sectorsize = dev->dev_root->sectorsize; |
b6bfebc1 SB |
481 | atomic_set(&sctx->bios_in_flight, 0); |
482 | atomic_set(&sctx->workers_pending, 0); | |
d9d181c1 SB |
483 | atomic_set(&sctx->cancel_req, 0); |
484 | sctx->csum_size = btrfs_super_csum_size(fs_info->super_copy); | |
485 | INIT_LIST_HEAD(&sctx->csum_list); | |
486 | ||
487 | spin_lock_init(&sctx->list_lock); | |
488 | spin_lock_init(&sctx->stat_lock); | |
489 | init_waitqueue_head(&sctx->list_wait); | |
ff023aac SB |
490 | |
491 | ret = scrub_setup_wr_ctx(sctx, &sctx->wr_ctx, fs_info, | |
492 | fs_info->dev_replace.tgtdev, is_dev_replace); | |
493 | if (ret) { | |
494 | scrub_free_ctx(sctx); | |
495 | return ERR_PTR(ret); | |
496 | } | |
d9d181c1 | 497 | return sctx; |
a2de733c AJ |
498 | |
499 | nomem: | |
d9d181c1 | 500 | scrub_free_ctx(sctx); |
a2de733c AJ |
501 | return ERR_PTR(-ENOMEM); |
502 | } | |
503 | ||
ff023aac SB |
504 | static int scrub_print_warning_inode(u64 inum, u64 offset, u64 root, |
505 | void *warn_ctx) | |
558540c1 JS |
506 | { |
507 | u64 isize; | |
508 | u32 nlink; | |
509 | int ret; | |
510 | int i; | |
511 | struct extent_buffer *eb; | |
512 | struct btrfs_inode_item *inode_item; | |
ff023aac | 513 | struct scrub_warning *swarn = warn_ctx; |
558540c1 JS |
514 | struct btrfs_fs_info *fs_info = swarn->dev->dev_root->fs_info; |
515 | struct inode_fs_paths *ipath = NULL; | |
516 | struct btrfs_root *local_root; | |
517 | struct btrfs_key root_key; | |
1d4c08e0 | 518 | struct btrfs_key key; |
558540c1 JS |
519 | |
520 | root_key.objectid = root; | |
521 | root_key.type = BTRFS_ROOT_ITEM_KEY; | |
522 | root_key.offset = (u64)-1; | |
523 | local_root = btrfs_read_fs_root_no_name(fs_info, &root_key); | |
524 | if (IS_ERR(local_root)) { | |
525 | ret = PTR_ERR(local_root); | |
526 | goto err; | |
527 | } | |
528 | ||
14692cc1 DS |
529 | /* |
530 | * this makes the path point to (inum INODE_ITEM ioff) | |
531 | */ | |
1d4c08e0 DS |
532 | key.objectid = inum; |
533 | key.type = BTRFS_INODE_ITEM_KEY; | |
534 | key.offset = 0; | |
535 | ||
536 | ret = btrfs_search_slot(NULL, local_root, &key, swarn->path, 0, 0); | |
558540c1 JS |
537 | if (ret) { |
538 | btrfs_release_path(swarn->path); | |
539 | goto err; | |
540 | } | |
541 | ||
542 | eb = swarn->path->nodes[0]; | |
543 | inode_item = btrfs_item_ptr(eb, swarn->path->slots[0], | |
544 | struct btrfs_inode_item); | |
545 | isize = btrfs_inode_size(eb, inode_item); | |
546 | nlink = btrfs_inode_nlink(eb, inode_item); | |
547 | btrfs_release_path(swarn->path); | |
548 | ||
549 | ipath = init_ipath(4096, local_root, swarn->path); | |
26bdef54 DC |
550 | if (IS_ERR(ipath)) { |
551 | ret = PTR_ERR(ipath); | |
552 | ipath = NULL; | |
553 | goto err; | |
554 | } | |
558540c1 JS |
555 | ret = paths_from_inode(inum, ipath); |
556 | ||
557 | if (ret < 0) | |
558 | goto err; | |
559 | ||
560 | /* | |
561 | * we deliberately ignore the bit ipath might have been too small to | |
562 | * hold all of the paths here | |
563 | */ | |
564 | for (i = 0; i < ipath->fspath->elem_cnt; ++i) | |
efe120a0 | 565 | printk_in_rcu(KERN_WARNING "BTRFS: %s at logical %llu on dev " |
558540c1 JS |
566 | "%s, sector %llu, root %llu, inode %llu, offset %llu, " |
567 | "length %llu, links %u (path: %s)\n", swarn->errstr, | |
606686ee | 568 | swarn->logical, rcu_str_deref(swarn->dev->name), |
558540c1 JS |
569 | (unsigned long long)swarn->sector, root, inum, offset, |
570 | min(isize - offset, (u64)PAGE_SIZE), nlink, | |
745c4d8e | 571 | (char *)(unsigned long)ipath->fspath->val[i]); |
558540c1 JS |
572 | |
573 | free_ipath(ipath); | |
574 | return 0; | |
575 | ||
576 | err: | |
efe120a0 | 577 | printk_in_rcu(KERN_WARNING "BTRFS: %s at logical %llu on dev " |
558540c1 JS |
578 | "%s, sector %llu, root %llu, inode %llu, offset %llu: path " |
579 | "resolving failed with ret=%d\n", swarn->errstr, | |
606686ee | 580 | swarn->logical, rcu_str_deref(swarn->dev->name), |
558540c1 JS |
581 | (unsigned long long)swarn->sector, root, inum, offset, ret); |
582 | ||
583 | free_ipath(ipath); | |
584 | return 0; | |
585 | } | |
586 | ||
b5d67f64 | 587 | static void scrub_print_warning(const char *errstr, struct scrub_block *sblock) |
558540c1 | 588 | { |
a36cf8b8 SB |
589 | struct btrfs_device *dev; |
590 | struct btrfs_fs_info *fs_info; | |
558540c1 JS |
591 | struct btrfs_path *path; |
592 | struct btrfs_key found_key; | |
593 | struct extent_buffer *eb; | |
594 | struct btrfs_extent_item *ei; | |
595 | struct scrub_warning swarn; | |
69917e43 LB |
596 | unsigned long ptr = 0; |
597 | u64 extent_item_pos; | |
598 | u64 flags = 0; | |
558540c1 | 599 | u64 ref_root; |
69917e43 | 600 | u32 item_size; |
558540c1 | 601 | u8 ref_level; |
69917e43 | 602 | int ret; |
558540c1 | 603 | |
a36cf8b8 | 604 | WARN_ON(sblock->page_count < 1); |
7a9e9987 | 605 | dev = sblock->pagev[0]->dev; |
a36cf8b8 SB |
606 | fs_info = sblock->sctx->dev_root->fs_info; |
607 | ||
558540c1 | 608 | path = btrfs_alloc_path(); |
8b9456da DS |
609 | if (!path) |
610 | return; | |
558540c1 | 611 | |
7a9e9987 SB |
612 | swarn.sector = (sblock->pagev[0]->physical) >> 9; |
613 | swarn.logical = sblock->pagev[0]->logical; | |
558540c1 | 614 | swarn.errstr = errstr; |
a36cf8b8 | 615 | swarn.dev = NULL; |
558540c1 | 616 | |
69917e43 LB |
617 | ret = extent_from_logical(fs_info, swarn.logical, path, &found_key, |
618 | &flags); | |
558540c1 JS |
619 | if (ret < 0) |
620 | goto out; | |
621 | ||
4692cf58 | 622 | extent_item_pos = swarn.logical - found_key.objectid; |
558540c1 JS |
623 | swarn.extent_item_size = found_key.offset; |
624 | ||
625 | eb = path->nodes[0]; | |
626 | ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item); | |
627 | item_size = btrfs_item_size_nr(eb, path->slots[0]); | |
628 | ||
69917e43 | 629 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { |
558540c1 | 630 | do { |
6eda71d0 LB |
631 | ret = tree_backref_for_extent(&ptr, eb, &found_key, ei, |
632 | item_size, &ref_root, | |
633 | &ref_level); | |
606686ee | 634 | printk_in_rcu(KERN_WARNING |
efe120a0 | 635 | "BTRFS: %s at logical %llu on dev %s, " |
558540c1 | 636 | "sector %llu: metadata %s (level %d) in tree " |
606686ee JB |
637 | "%llu\n", errstr, swarn.logical, |
638 | rcu_str_deref(dev->name), | |
558540c1 JS |
639 | (unsigned long long)swarn.sector, |
640 | ref_level ? "node" : "leaf", | |
641 | ret < 0 ? -1 : ref_level, | |
642 | ret < 0 ? -1 : ref_root); | |
643 | } while (ret != 1); | |
d8fe29e9 | 644 | btrfs_release_path(path); |
558540c1 | 645 | } else { |
d8fe29e9 | 646 | btrfs_release_path(path); |
558540c1 | 647 | swarn.path = path; |
a36cf8b8 | 648 | swarn.dev = dev; |
7a3ae2f8 JS |
649 | iterate_extent_inodes(fs_info, found_key.objectid, |
650 | extent_item_pos, 1, | |
558540c1 JS |
651 | scrub_print_warning_inode, &swarn); |
652 | } | |
653 | ||
654 | out: | |
655 | btrfs_free_path(path); | |
558540c1 JS |
656 | } |
657 | ||
ff023aac | 658 | static int scrub_fixup_readpage(u64 inum, u64 offset, u64 root, void *fixup_ctx) |
0ef8e451 | 659 | { |
5da6fcbc | 660 | struct page *page = NULL; |
0ef8e451 | 661 | unsigned long index; |
ff023aac | 662 | struct scrub_fixup_nodatasum *fixup = fixup_ctx; |
0ef8e451 | 663 | int ret; |
5da6fcbc | 664 | int corrected = 0; |
0ef8e451 | 665 | struct btrfs_key key; |
5da6fcbc | 666 | struct inode *inode = NULL; |
6f1c3605 | 667 | struct btrfs_fs_info *fs_info; |
0ef8e451 JS |
668 | u64 end = offset + PAGE_SIZE - 1; |
669 | struct btrfs_root *local_root; | |
6f1c3605 | 670 | int srcu_index; |
0ef8e451 JS |
671 | |
672 | key.objectid = root; | |
673 | key.type = BTRFS_ROOT_ITEM_KEY; | |
674 | key.offset = (u64)-1; | |
6f1c3605 LB |
675 | |
676 | fs_info = fixup->root->fs_info; | |
677 | srcu_index = srcu_read_lock(&fs_info->subvol_srcu); | |
678 | ||
679 | local_root = btrfs_read_fs_root_no_name(fs_info, &key); | |
680 | if (IS_ERR(local_root)) { | |
681 | srcu_read_unlock(&fs_info->subvol_srcu, srcu_index); | |
0ef8e451 | 682 | return PTR_ERR(local_root); |
6f1c3605 | 683 | } |
0ef8e451 JS |
684 | |
685 | key.type = BTRFS_INODE_ITEM_KEY; | |
686 | key.objectid = inum; | |
687 | key.offset = 0; | |
6f1c3605 LB |
688 | inode = btrfs_iget(fs_info->sb, &key, local_root, NULL); |
689 | srcu_read_unlock(&fs_info->subvol_srcu, srcu_index); | |
0ef8e451 JS |
690 | if (IS_ERR(inode)) |
691 | return PTR_ERR(inode); | |
692 | ||
0ef8e451 JS |
693 | index = offset >> PAGE_CACHE_SHIFT; |
694 | ||
695 | page = find_or_create_page(inode->i_mapping, index, GFP_NOFS); | |
5da6fcbc JS |
696 | if (!page) { |
697 | ret = -ENOMEM; | |
698 | goto out; | |
699 | } | |
700 | ||
701 | if (PageUptodate(page)) { | |
5da6fcbc JS |
702 | if (PageDirty(page)) { |
703 | /* | |
704 | * we need to write the data to the defect sector. the | |
705 | * data that was in that sector is not in memory, | |
706 | * because the page was modified. we must not write the | |
707 | * modified page to that sector. | |
708 | * | |
709 | * TODO: what could be done here: wait for the delalloc | |
710 | * runner to write out that page (might involve | |
711 | * COW) and see whether the sector is still | |
712 | * referenced afterwards. | |
713 | * | |
714 | * For the meantime, we'll treat this error | |
715 | * incorrectable, although there is a chance that a | |
716 | * later scrub will find the bad sector again and that | |
717 | * there's no dirty page in memory, then. | |
718 | */ | |
719 | ret = -EIO; | |
720 | goto out; | |
721 | } | |
1203b681 | 722 | ret = repair_io_failure(inode, offset, PAGE_SIZE, |
5da6fcbc | 723 | fixup->logical, page, |
ffdd2018 | 724 | offset - page_offset(page), |
5da6fcbc JS |
725 | fixup->mirror_num); |
726 | unlock_page(page); | |
727 | corrected = !ret; | |
728 | } else { | |
729 | /* | |
730 | * we need to get good data first. the general readpage path | |
731 | * will call repair_io_failure for us, we just have to make | |
732 | * sure we read the bad mirror. | |
733 | */ | |
734 | ret = set_extent_bits(&BTRFS_I(inode)->io_tree, offset, end, | |
735 | EXTENT_DAMAGED, GFP_NOFS); | |
736 | if (ret) { | |
737 | /* set_extent_bits should give proper error */ | |
738 | WARN_ON(ret > 0); | |
739 | if (ret > 0) | |
740 | ret = -EFAULT; | |
741 | goto out; | |
742 | } | |
743 | ||
744 | ret = extent_read_full_page(&BTRFS_I(inode)->io_tree, page, | |
745 | btrfs_get_extent, | |
746 | fixup->mirror_num); | |
747 | wait_on_page_locked(page); | |
748 | ||
749 | corrected = !test_range_bit(&BTRFS_I(inode)->io_tree, offset, | |
750 | end, EXTENT_DAMAGED, 0, NULL); | |
751 | if (!corrected) | |
752 | clear_extent_bits(&BTRFS_I(inode)->io_tree, offset, end, | |
753 | EXTENT_DAMAGED, GFP_NOFS); | |
754 | } | |
755 | ||
756 | out: | |
757 | if (page) | |
758 | put_page(page); | |
7fb18a06 TK |
759 | |
760 | iput(inode); | |
0ef8e451 JS |
761 | |
762 | if (ret < 0) | |
763 | return ret; | |
764 | ||
765 | if (ret == 0 && corrected) { | |
766 | /* | |
767 | * we only need to call readpage for one of the inodes belonging | |
768 | * to this extent. so make iterate_extent_inodes stop | |
769 | */ | |
770 | return 1; | |
771 | } | |
772 | ||
773 | return -EIO; | |
774 | } | |
775 | ||
776 | static void scrub_fixup_nodatasum(struct btrfs_work *work) | |
777 | { | |
778 | int ret; | |
779 | struct scrub_fixup_nodatasum *fixup; | |
d9d181c1 | 780 | struct scrub_ctx *sctx; |
0ef8e451 | 781 | struct btrfs_trans_handle *trans = NULL; |
0ef8e451 JS |
782 | struct btrfs_path *path; |
783 | int uncorrectable = 0; | |
784 | ||
785 | fixup = container_of(work, struct scrub_fixup_nodatasum, work); | |
d9d181c1 | 786 | sctx = fixup->sctx; |
0ef8e451 JS |
787 | |
788 | path = btrfs_alloc_path(); | |
789 | if (!path) { | |
d9d181c1 SB |
790 | spin_lock(&sctx->stat_lock); |
791 | ++sctx->stat.malloc_errors; | |
792 | spin_unlock(&sctx->stat_lock); | |
0ef8e451 JS |
793 | uncorrectable = 1; |
794 | goto out; | |
795 | } | |
796 | ||
797 | trans = btrfs_join_transaction(fixup->root); | |
798 | if (IS_ERR(trans)) { | |
799 | uncorrectable = 1; | |
800 | goto out; | |
801 | } | |
802 | ||
803 | /* | |
804 | * the idea is to trigger a regular read through the standard path. we | |
805 | * read a page from the (failed) logical address by specifying the | |
806 | * corresponding copynum of the failed sector. thus, that readpage is | |
807 | * expected to fail. | |
808 | * that is the point where on-the-fly error correction will kick in | |
809 | * (once it's finished) and rewrite the failed sector if a good copy | |
810 | * can be found. | |
811 | */ | |
812 | ret = iterate_inodes_from_logical(fixup->logical, fixup->root->fs_info, | |
813 | path, scrub_fixup_readpage, | |
814 | fixup); | |
815 | if (ret < 0) { | |
816 | uncorrectable = 1; | |
817 | goto out; | |
818 | } | |
819 | WARN_ON(ret != 1); | |
820 | ||
d9d181c1 SB |
821 | spin_lock(&sctx->stat_lock); |
822 | ++sctx->stat.corrected_errors; | |
823 | spin_unlock(&sctx->stat_lock); | |
0ef8e451 JS |
824 | |
825 | out: | |
826 | if (trans && !IS_ERR(trans)) | |
827 | btrfs_end_transaction(trans, fixup->root); | |
828 | if (uncorrectable) { | |
d9d181c1 SB |
829 | spin_lock(&sctx->stat_lock); |
830 | ++sctx->stat.uncorrectable_errors; | |
831 | spin_unlock(&sctx->stat_lock); | |
ff023aac SB |
832 | btrfs_dev_replace_stats_inc( |
833 | &sctx->dev_root->fs_info->dev_replace. | |
834 | num_uncorrectable_read_errors); | |
efe120a0 FH |
835 | printk_ratelimited_in_rcu(KERN_ERR "BTRFS: " |
836 | "unable to fixup (nodatasum) error at logical %llu on dev %s\n", | |
c1c9ff7c | 837 | fixup->logical, rcu_str_deref(fixup->dev->name)); |
0ef8e451 JS |
838 | } |
839 | ||
840 | btrfs_free_path(path); | |
841 | kfree(fixup); | |
842 | ||
b6bfebc1 | 843 | scrub_pending_trans_workers_dec(sctx); |
0ef8e451 JS |
844 | } |
845 | ||
af8e2d1d MX |
846 | static inline void scrub_get_recover(struct scrub_recover *recover) |
847 | { | |
848 | atomic_inc(&recover->refs); | |
849 | } | |
850 | ||
851 | static inline void scrub_put_recover(struct scrub_recover *recover) | |
852 | { | |
853 | if (atomic_dec_and_test(&recover->refs)) { | |
6e9606d2 | 854 | btrfs_put_bbio(recover->bbio); |
af8e2d1d MX |
855 | kfree(recover); |
856 | } | |
857 | } | |
858 | ||
a2de733c | 859 | /* |
b5d67f64 SB |
860 | * scrub_handle_errored_block gets called when either verification of the |
861 | * pages failed or the bio failed to read, e.g. with EIO. In the latter | |
862 | * case, this function handles all pages in the bio, even though only one | |
863 | * may be bad. | |
864 | * The goal of this function is to repair the errored block by using the | |
865 | * contents of one of the mirrors. | |
a2de733c | 866 | */ |
b5d67f64 | 867 | static int scrub_handle_errored_block(struct scrub_block *sblock_to_check) |
a2de733c | 868 | { |
d9d181c1 | 869 | struct scrub_ctx *sctx = sblock_to_check->sctx; |
a36cf8b8 | 870 | struct btrfs_device *dev; |
b5d67f64 SB |
871 | struct btrfs_fs_info *fs_info; |
872 | u64 length; | |
873 | u64 logical; | |
874 | u64 generation; | |
875 | unsigned int failed_mirror_index; | |
876 | unsigned int is_metadata; | |
877 | unsigned int have_csum; | |
878 | u8 *csum; | |
879 | struct scrub_block *sblocks_for_recheck; /* holds one for each mirror */ | |
880 | struct scrub_block *sblock_bad; | |
881 | int ret; | |
882 | int mirror_index; | |
883 | int page_num; | |
884 | int success; | |
558540c1 | 885 | static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL, |
b5d67f64 SB |
886 | DEFAULT_RATELIMIT_BURST); |
887 | ||
888 | BUG_ON(sblock_to_check->page_count < 1); | |
a36cf8b8 | 889 | fs_info = sctx->dev_root->fs_info; |
4ded4f63 SB |
890 | if (sblock_to_check->pagev[0]->flags & BTRFS_EXTENT_FLAG_SUPER) { |
891 | /* | |
892 | * if we find an error in a super block, we just report it. | |
893 | * They will get written with the next transaction commit | |
894 | * anyway | |
895 | */ | |
896 | spin_lock(&sctx->stat_lock); | |
897 | ++sctx->stat.super_errors; | |
898 | spin_unlock(&sctx->stat_lock); | |
899 | return 0; | |
900 | } | |
b5d67f64 | 901 | length = sblock_to_check->page_count * PAGE_SIZE; |
7a9e9987 SB |
902 | logical = sblock_to_check->pagev[0]->logical; |
903 | generation = sblock_to_check->pagev[0]->generation; | |
904 | BUG_ON(sblock_to_check->pagev[0]->mirror_num < 1); | |
905 | failed_mirror_index = sblock_to_check->pagev[0]->mirror_num - 1; | |
906 | is_metadata = !(sblock_to_check->pagev[0]->flags & | |
b5d67f64 | 907 | BTRFS_EXTENT_FLAG_DATA); |
7a9e9987 SB |
908 | have_csum = sblock_to_check->pagev[0]->have_csum; |
909 | csum = sblock_to_check->pagev[0]->csum; | |
910 | dev = sblock_to_check->pagev[0]->dev; | |
13db62b7 | 911 | |
ff023aac SB |
912 | if (sctx->is_dev_replace && !is_metadata && !have_csum) { |
913 | sblocks_for_recheck = NULL; | |
914 | goto nodatasum_case; | |
915 | } | |
916 | ||
b5d67f64 SB |
917 | /* |
918 | * read all mirrors one after the other. This includes to | |
919 | * re-read the extent or metadata block that failed (that was | |
920 | * the cause that this fixup code is called) another time, | |
921 | * page by page this time in order to know which pages | |
922 | * caused I/O errors and which ones are good (for all mirrors). | |
923 | * It is the goal to handle the situation when more than one | |
924 | * mirror contains I/O errors, but the errors do not | |
925 | * overlap, i.e. the data can be repaired by selecting the | |
926 | * pages from those mirrors without I/O error on the | |
927 | * particular pages. One example (with blocks >= 2 * PAGE_SIZE) | |
928 | * would be that mirror #1 has an I/O error on the first page, | |
929 | * the second page is good, and mirror #2 has an I/O error on | |
930 | * the second page, but the first page is good. | |
931 | * Then the first page of the first mirror can be repaired by | |
932 | * taking the first page of the second mirror, and the | |
933 | * second page of the second mirror can be repaired by | |
934 | * copying the contents of the 2nd page of the 1st mirror. | |
935 | * One more note: if the pages of one mirror contain I/O | |
936 | * errors, the checksum cannot be verified. In order to get | |
937 | * the best data for repairing, the first attempt is to find | |
938 | * a mirror without I/O errors and with a validated checksum. | |
939 | * Only if this is not possible, the pages are picked from | |
940 | * mirrors with I/O errors without considering the checksum. | |
941 | * If the latter is the case, at the end, the checksum of the | |
942 | * repaired area is verified in order to correctly maintain | |
943 | * the statistics. | |
944 | */ | |
945 | ||
946 | sblocks_for_recheck = kzalloc(BTRFS_MAX_MIRRORS * | |
947 | sizeof(*sblocks_for_recheck), | |
948 | GFP_NOFS); | |
949 | if (!sblocks_for_recheck) { | |
d9d181c1 SB |
950 | spin_lock(&sctx->stat_lock); |
951 | sctx->stat.malloc_errors++; | |
952 | sctx->stat.read_errors++; | |
953 | sctx->stat.uncorrectable_errors++; | |
954 | spin_unlock(&sctx->stat_lock); | |
a36cf8b8 | 955 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS); |
b5d67f64 | 956 | goto out; |
a2de733c AJ |
957 | } |
958 | ||
b5d67f64 | 959 | /* setup the context, map the logical blocks and alloc the pages */ |
be50a8dd | 960 | ret = scrub_setup_recheck_block(sblock_to_check, sblocks_for_recheck); |
b5d67f64 | 961 | if (ret) { |
d9d181c1 SB |
962 | spin_lock(&sctx->stat_lock); |
963 | sctx->stat.read_errors++; | |
964 | sctx->stat.uncorrectable_errors++; | |
965 | spin_unlock(&sctx->stat_lock); | |
a36cf8b8 | 966 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS); |
b5d67f64 SB |
967 | goto out; |
968 | } | |
969 | BUG_ON(failed_mirror_index >= BTRFS_MAX_MIRRORS); | |
970 | sblock_bad = sblocks_for_recheck + failed_mirror_index; | |
13db62b7 | 971 | |
b5d67f64 | 972 | /* build and submit the bios for the failed mirror, check checksums */ |
34f5c8e9 | 973 | scrub_recheck_block(fs_info, sblock_bad, is_metadata, have_csum, |
af8e2d1d | 974 | csum, generation, sctx->csum_size, 1); |
a2de733c | 975 | |
b5d67f64 SB |
976 | if (!sblock_bad->header_error && !sblock_bad->checksum_error && |
977 | sblock_bad->no_io_error_seen) { | |
978 | /* | |
979 | * the error disappeared after reading page by page, or | |
980 | * the area was part of a huge bio and other parts of the | |
981 | * bio caused I/O errors, or the block layer merged several | |
982 | * read requests into one and the error is caused by a | |
983 | * different bio (usually one of the two latter cases is | |
984 | * the cause) | |
985 | */ | |
d9d181c1 SB |
986 | spin_lock(&sctx->stat_lock); |
987 | sctx->stat.unverified_errors++; | |
5a6ac9ea | 988 | sblock_to_check->data_corrected = 1; |
d9d181c1 | 989 | spin_unlock(&sctx->stat_lock); |
a2de733c | 990 | |
ff023aac SB |
991 | if (sctx->is_dev_replace) |
992 | scrub_write_block_to_dev_replace(sblock_bad); | |
b5d67f64 | 993 | goto out; |
a2de733c | 994 | } |
a2de733c | 995 | |
b5d67f64 | 996 | if (!sblock_bad->no_io_error_seen) { |
d9d181c1 SB |
997 | spin_lock(&sctx->stat_lock); |
998 | sctx->stat.read_errors++; | |
999 | spin_unlock(&sctx->stat_lock); | |
b5d67f64 SB |
1000 | if (__ratelimit(&_rs)) |
1001 | scrub_print_warning("i/o error", sblock_to_check); | |
a36cf8b8 | 1002 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS); |
b5d67f64 | 1003 | } else if (sblock_bad->checksum_error) { |
d9d181c1 SB |
1004 | spin_lock(&sctx->stat_lock); |
1005 | sctx->stat.csum_errors++; | |
1006 | spin_unlock(&sctx->stat_lock); | |
b5d67f64 SB |
1007 | if (__ratelimit(&_rs)) |
1008 | scrub_print_warning("checksum error", sblock_to_check); | |
a36cf8b8 | 1009 | btrfs_dev_stat_inc_and_print(dev, |
442a4f63 | 1010 | BTRFS_DEV_STAT_CORRUPTION_ERRS); |
b5d67f64 | 1011 | } else if (sblock_bad->header_error) { |
d9d181c1 SB |
1012 | spin_lock(&sctx->stat_lock); |
1013 | sctx->stat.verify_errors++; | |
1014 | spin_unlock(&sctx->stat_lock); | |
b5d67f64 SB |
1015 | if (__ratelimit(&_rs)) |
1016 | scrub_print_warning("checksum/header error", | |
1017 | sblock_to_check); | |
442a4f63 | 1018 | if (sblock_bad->generation_error) |
a36cf8b8 | 1019 | btrfs_dev_stat_inc_and_print(dev, |
442a4f63 SB |
1020 | BTRFS_DEV_STAT_GENERATION_ERRS); |
1021 | else | |
a36cf8b8 | 1022 | btrfs_dev_stat_inc_and_print(dev, |
442a4f63 | 1023 | BTRFS_DEV_STAT_CORRUPTION_ERRS); |
b5d67f64 | 1024 | } |
a2de733c | 1025 | |
33ef30ad ID |
1026 | if (sctx->readonly) { |
1027 | ASSERT(!sctx->is_dev_replace); | |
1028 | goto out; | |
1029 | } | |
a2de733c | 1030 | |
b5d67f64 SB |
1031 | if (!is_metadata && !have_csum) { |
1032 | struct scrub_fixup_nodatasum *fixup_nodatasum; | |
a2de733c | 1033 | |
ff023aac SB |
1034 | WARN_ON(sctx->is_dev_replace); |
1035 | ||
b25c94c5 ZL |
1036 | nodatasum_case: |
1037 | ||
b5d67f64 SB |
1038 | /* |
1039 | * !is_metadata and !have_csum, this means that the data | |
1040 | * might not be COW'ed, that it might be modified | |
1041 | * concurrently. The general strategy to work on the | |
1042 | * commit root does not help in the case when COW is not | |
1043 | * used. | |
1044 | */ | |
1045 | fixup_nodatasum = kzalloc(sizeof(*fixup_nodatasum), GFP_NOFS); | |
1046 | if (!fixup_nodatasum) | |
1047 | goto did_not_correct_error; | |
d9d181c1 | 1048 | fixup_nodatasum->sctx = sctx; |
a36cf8b8 | 1049 | fixup_nodatasum->dev = dev; |
b5d67f64 SB |
1050 | fixup_nodatasum->logical = logical; |
1051 | fixup_nodatasum->root = fs_info->extent_root; | |
1052 | fixup_nodatasum->mirror_num = failed_mirror_index + 1; | |
b6bfebc1 | 1053 | scrub_pending_trans_workers_inc(sctx); |
9e0af237 LB |
1054 | btrfs_init_work(&fixup_nodatasum->work, btrfs_scrub_helper, |
1055 | scrub_fixup_nodatasum, NULL, NULL); | |
0339ef2f QW |
1056 | btrfs_queue_work(fs_info->scrub_workers, |
1057 | &fixup_nodatasum->work); | |
b5d67f64 | 1058 | goto out; |
a2de733c AJ |
1059 | } |
1060 | ||
b5d67f64 SB |
1061 | /* |
1062 | * now build and submit the bios for the other mirrors, check | |
cb2ced73 SB |
1063 | * checksums. |
1064 | * First try to pick the mirror which is completely without I/O | |
b5d67f64 SB |
1065 | * errors and also does not have a checksum error. |
1066 | * If one is found, and if a checksum is present, the full block | |
1067 | * that is known to contain an error is rewritten. Afterwards | |
1068 | * the block is known to be corrected. | |
1069 | * If a mirror is found which is completely correct, and no | |
1070 | * checksum is present, only those pages are rewritten that had | |
1071 | * an I/O error in the block to be repaired, since it cannot be | |
1072 | * determined, which copy of the other pages is better (and it | |
1073 | * could happen otherwise that a correct page would be | |
1074 | * overwritten by a bad one). | |
1075 | */ | |
1076 | for (mirror_index = 0; | |
1077 | mirror_index < BTRFS_MAX_MIRRORS && | |
1078 | sblocks_for_recheck[mirror_index].page_count > 0; | |
1079 | mirror_index++) { | |
cb2ced73 | 1080 | struct scrub_block *sblock_other; |
b5d67f64 | 1081 | |
cb2ced73 SB |
1082 | if (mirror_index == failed_mirror_index) |
1083 | continue; | |
1084 | sblock_other = sblocks_for_recheck + mirror_index; | |
1085 | ||
1086 | /* build and submit the bios, check checksums */ | |
34f5c8e9 SB |
1087 | scrub_recheck_block(fs_info, sblock_other, is_metadata, |
1088 | have_csum, csum, generation, | |
af8e2d1d | 1089 | sctx->csum_size, 0); |
34f5c8e9 SB |
1090 | |
1091 | if (!sblock_other->header_error && | |
b5d67f64 SB |
1092 | !sblock_other->checksum_error && |
1093 | sblock_other->no_io_error_seen) { | |
ff023aac SB |
1094 | if (sctx->is_dev_replace) { |
1095 | scrub_write_block_to_dev_replace(sblock_other); | |
114ab50d | 1096 | goto corrected_error; |
ff023aac | 1097 | } else { |
ff023aac | 1098 | ret = scrub_repair_block_from_good_copy( |
114ab50d ZL |
1099 | sblock_bad, sblock_other); |
1100 | if (!ret) | |
1101 | goto corrected_error; | |
ff023aac | 1102 | } |
b5d67f64 SB |
1103 | } |
1104 | } | |
a2de733c | 1105 | |
b968fed1 ZL |
1106 | if (sblock_bad->no_io_error_seen && !sctx->is_dev_replace) |
1107 | goto did_not_correct_error; | |
ff023aac SB |
1108 | |
1109 | /* | |
ff023aac | 1110 | * In case of I/O errors in the area that is supposed to be |
b5d67f64 SB |
1111 | * repaired, continue by picking good copies of those pages. |
1112 | * Select the good pages from mirrors to rewrite bad pages from | |
1113 | * the area to fix. Afterwards verify the checksum of the block | |
1114 | * that is supposed to be repaired. This verification step is | |
1115 | * only done for the purpose of statistic counting and for the | |
1116 | * final scrub report, whether errors remain. | |
1117 | * A perfect algorithm could make use of the checksum and try | |
1118 | * all possible combinations of pages from the different mirrors | |
1119 | * until the checksum verification succeeds. For example, when | |
1120 | * the 2nd page of mirror #1 faces I/O errors, and the 2nd page | |
1121 | * of mirror #2 is readable but the final checksum test fails, | |
1122 | * then the 2nd page of mirror #3 could be tried, whether now | |
1123 | * the final checksum succeedes. But this would be a rare | |
1124 | * exception and is therefore not implemented. At least it is | |
1125 | * avoided that the good copy is overwritten. | |
1126 | * A more useful improvement would be to pick the sectors | |
1127 | * without I/O error based on sector sizes (512 bytes on legacy | |
1128 | * disks) instead of on PAGE_SIZE. Then maybe 512 byte of one | |
1129 | * mirror could be repaired by taking 512 byte of a different | |
1130 | * mirror, even if other 512 byte sectors in the same PAGE_SIZE | |
1131 | * area are unreadable. | |
a2de733c | 1132 | */ |
b5d67f64 | 1133 | success = 1; |
b968fed1 ZL |
1134 | for (page_num = 0; page_num < sblock_bad->page_count; |
1135 | page_num++) { | |
7a9e9987 | 1136 | struct scrub_page *page_bad = sblock_bad->pagev[page_num]; |
b968fed1 | 1137 | struct scrub_block *sblock_other = NULL; |
b5d67f64 | 1138 | |
b968fed1 ZL |
1139 | /* skip no-io-error page in scrub */ |
1140 | if (!page_bad->io_error && !sctx->is_dev_replace) | |
a2de733c | 1141 | continue; |
b5d67f64 | 1142 | |
b968fed1 ZL |
1143 | /* try to find no-io-error page in mirrors */ |
1144 | if (page_bad->io_error) { | |
1145 | for (mirror_index = 0; | |
1146 | mirror_index < BTRFS_MAX_MIRRORS && | |
1147 | sblocks_for_recheck[mirror_index].page_count > 0; | |
1148 | mirror_index++) { | |
1149 | if (!sblocks_for_recheck[mirror_index]. | |
1150 | pagev[page_num]->io_error) { | |
1151 | sblock_other = sblocks_for_recheck + | |
1152 | mirror_index; | |
1153 | break; | |
b5d67f64 SB |
1154 | } |
1155 | } | |
b968fed1 ZL |
1156 | if (!sblock_other) |
1157 | success = 0; | |
96e36920 | 1158 | } |
a2de733c | 1159 | |
b968fed1 ZL |
1160 | if (sctx->is_dev_replace) { |
1161 | /* | |
1162 | * did not find a mirror to fetch the page | |
1163 | * from. scrub_write_page_to_dev_replace() | |
1164 | * handles this case (page->io_error), by | |
1165 | * filling the block with zeros before | |
1166 | * submitting the write request | |
1167 | */ | |
1168 | if (!sblock_other) | |
1169 | sblock_other = sblock_bad; | |
1170 | ||
1171 | if (scrub_write_page_to_dev_replace(sblock_other, | |
1172 | page_num) != 0) { | |
1173 | btrfs_dev_replace_stats_inc( | |
1174 | &sctx->dev_root-> | |
1175 | fs_info->dev_replace. | |
1176 | num_write_errors); | |
1177 | success = 0; | |
1178 | } | |
1179 | } else if (sblock_other) { | |
1180 | ret = scrub_repair_page_from_good_copy(sblock_bad, | |
1181 | sblock_other, | |
1182 | page_num, 0); | |
1183 | if (0 == ret) | |
1184 | page_bad->io_error = 0; | |
1185 | else | |
1186 | success = 0; | |
b5d67f64 | 1187 | } |
a2de733c | 1188 | } |
a2de733c | 1189 | |
b968fed1 | 1190 | if (success && !sctx->is_dev_replace) { |
b5d67f64 SB |
1191 | if (is_metadata || have_csum) { |
1192 | /* | |
1193 | * need to verify the checksum now that all | |
1194 | * sectors on disk are repaired (the write | |
1195 | * request for data to be repaired is on its way). | |
1196 | * Just be lazy and use scrub_recheck_block() | |
1197 | * which re-reads the data before the checksum | |
1198 | * is verified, but most likely the data comes out | |
1199 | * of the page cache. | |
1200 | */ | |
34f5c8e9 SB |
1201 | scrub_recheck_block(fs_info, sblock_bad, |
1202 | is_metadata, have_csum, csum, | |
af8e2d1d | 1203 | generation, sctx->csum_size, 1); |
34f5c8e9 | 1204 | if (!sblock_bad->header_error && |
b5d67f64 SB |
1205 | !sblock_bad->checksum_error && |
1206 | sblock_bad->no_io_error_seen) | |
1207 | goto corrected_error; | |
1208 | else | |
1209 | goto did_not_correct_error; | |
1210 | } else { | |
1211 | corrected_error: | |
d9d181c1 SB |
1212 | spin_lock(&sctx->stat_lock); |
1213 | sctx->stat.corrected_errors++; | |
5a6ac9ea | 1214 | sblock_to_check->data_corrected = 1; |
d9d181c1 | 1215 | spin_unlock(&sctx->stat_lock); |
606686ee | 1216 | printk_ratelimited_in_rcu(KERN_ERR |
efe120a0 | 1217 | "BTRFS: fixed up error at logical %llu on dev %s\n", |
c1c9ff7c | 1218 | logical, rcu_str_deref(dev->name)); |
8628764e | 1219 | } |
b5d67f64 SB |
1220 | } else { |
1221 | did_not_correct_error: | |
d9d181c1 SB |
1222 | spin_lock(&sctx->stat_lock); |
1223 | sctx->stat.uncorrectable_errors++; | |
1224 | spin_unlock(&sctx->stat_lock); | |
606686ee | 1225 | printk_ratelimited_in_rcu(KERN_ERR |
efe120a0 | 1226 | "BTRFS: unable to fixup (regular) error at logical %llu on dev %s\n", |
c1c9ff7c | 1227 | logical, rcu_str_deref(dev->name)); |
96e36920 | 1228 | } |
a2de733c | 1229 | |
b5d67f64 SB |
1230 | out: |
1231 | if (sblocks_for_recheck) { | |
1232 | for (mirror_index = 0; mirror_index < BTRFS_MAX_MIRRORS; | |
1233 | mirror_index++) { | |
1234 | struct scrub_block *sblock = sblocks_for_recheck + | |
1235 | mirror_index; | |
af8e2d1d | 1236 | struct scrub_recover *recover; |
b5d67f64 SB |
1237 | int page_index; |
1238 | ||
7a9e9987 SB |
1239 | for (page_index = 0; page_index < sblock->page_count; |
1240 | page_index++) { | |
1241 | sblock->pagev[page_index]->sblock = NULL; | |
af8e2d1d MX |
1242 | recover = sblock->pagev[page_index]->recover; |
1243 | if (recover) { | |
1244 | scrub_put_recover(recover); | |
1245 | sblock->pagev[page_index]->recover = | |
1246 | NULL; | |
1247 | } | |
7a9e9987 SB |
1248 | scrub_page_put(sblock->pagev[page_index]); |
1249 | } | |
b5d67f64 SB |
1250 | } |
1251 | kfree(sblocks_for_recheck); | |
1252 | } | |
a2de733c | 1253 | |
b5d67f64 SB |
1254 | return 0; |
1255 | } | |
a2de733c | 1256 | |
8e5cfb55 | 1257 | static inline int scrub_nr_raid_mirrors(struct btrfs_bio *bbio) |
af8e2d1d | 1258 | { |
8e5cfb55 | 1259 | if (bbio->raid_map) { |
e34c330d ZL |
1260 | int real_stripes = bbio->num_stripes - bbio->num_tgtdevs; |
1261 | ||
8e5cfb55 | 1262 | if (bbio->raid_map[real_stripes - 1] == RAID6_Q_STRIPE) |
af8e2d1d MX |
1263 | return 3; |
1264 | else | |
1265 | return 2; | |
1266 | } else { | |
1267 | return (int)bbio->num_stripes; | |
1268 | } | |
1269 | } | |
1270 | ||
1271 | static inline void scrub_stripe_index_and_offset(u64 logical, u64 *raid_map, | |
1272 | u64 mapped_length, | |
1273 | int nstripes, int mirror, | |
1274 | int *stripe_index, | |
1275 | u64 *stripe_offset) | |
1276 | { | |
1277 | int i; | |
1278 | ||
1279 | if (raid_map) { | |
1280 | /* RAID5/6 */ | |
1281 | for (i = 0; i < nstripes; i++) { | |
1282 | if (raid_map[i] == RAID6_Q_STRIPE || | |
1283 | raid_map[i] == RAID5_P_STRIPE) | |
1284 | continue; | |
1285 | ||
1286 | if (logical >= raid_map[i] && | |
1287 | logical < raid_map[i] + mapped_length) | |
1288 | break; | |
1289 | } | |
1290 | ||
1291 | *stripe_index = i; | |
1292 | *stripe_offset = logical - raid_map[i]; | |
1293 | } else { | |
1294 | /* The other RAID type */ | |
1295 | *stripe_index = mirror; | |
1296 | *stripe_offset = 0; | |
1297 | } | |
1298 | } | |
1299 | ||
be50a8dd | 1300 | static int scrub_setup_recheck_block(struct scrub_block *original_sblock, |
b5d67f64 SB |
1301 | struct scrub_block *sblocks_for_recheck) |
1302 | { | |
be50a8dd ZL |
1303 | struct scrub_ctx *sctx = original_sblock->sctx; |
1304 | struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info; | |
1305 | u64 length = original_sblock->page_count * PAGE_SIZE; | |
1306 | u64 logical = original_sblock->pagev[0]->logical; | |
af8e2d1d MX |
1307 | struct scrub_recover *recover; |
1308 | struct btrfs_bio *bbio; | |
af8e2d1d MX |
1309 | u64 sublen; |
1310 | u64 mapped_length; | |
1311 | u64 stripe_offset; | |
1312 | int stripe_index; | |
be50a8dd | 1313 | int page_index = 0; |
b5d67f64 | 1314 | int mirror_index; |
af8e2d1d | 1315 | int nmirrors; |
b5d67f64 SB |
1316 | int ret; |
1317 | ||
1318 | /* | |
7a9e9987 | 1319 | * note: the two members ref_count and outstanding_pages |
b5d67f64 SB |
1320 | * are not used (and not set) in the blocks that are used for |
1321 | * the recheck procedure | |
1322 | */ | |
1323 | ||
b5d67f64 | 1324 | while (length > 0) { |
af8e2d1d MX |
1325 | sublen = min_t(u64, length, PAGE_SIZE); |
1326 | mapped_length = sublen; | |
1327 | bbio = NULL; | |
a2de733c | 1328 | |
b5d67f64 SB |
1329 | /* |
1330 | * with a length of PAGE_SIZE, each returned stripe | |
1331 | * represents one mirror | |
1332 | */ | |
af8e2d1d | 1333 | ret = btrfs_map_sblock(fs_info, REQ_GET_READ_MIRRORS, logical, |
8e5cfb55 | 1334 | &mapped_length, &bbio, 0, 1); |
b5d67f64 | 1335 | if (ret || !bbio || mapped_length < sublen) { |
6e9606d2 | 1336 | btrfs_put_bbio(bbio); |
b5d67f64 SB |
1337 | return -EIO; |
1338 | } | |
a2de733c | 1339 | |
af8e2d1d MX |
1340 | recover = kzalloc(sizeof(struct scrub_recover), GFP_NOFS); |
1341 | if (!recover) { | |
6e9606d2 | 1342 | btrfs_put_bbio(bbio); |
af8e2d1d MX |
1343 | return -ENOMEM; |
1344 | } | |
1345 | ||
1346 | atomic_set(&recover->refs, 1); | |
1347 | recover->bbio = bbio; | |
af8e2d1d MX |
1348 | recover->map_length = mapped_length; |
1349 | ||
ff023aac | 1350 | BUG_ON(page_index >= SCRUB_PAGES_PER_RD_BIO); |
af8e2d1d | 1351 | |
be50a8dd | 1352 | nmirrors = min(scrub_nr_raid_mirrors(bbio), BTRFS_MAX_MIRRORS); |
af8e2d1d | 1353 | for (mirror_index = 0; mirror_index < nmirrors; |
b5d67f64 SB |
1354 | mirror_index++) { |
1355 | struct scrub_block *sblock; | |
1356 | struct scrub_page *page; | |
1357 | ||
b5d67f64 | 1358 | sblock = sblocks_for_recheck + mirror_index; |
7a9e9987 SB |
1359 | sblock->sctx = sctx; |
1360 | page = kzalloc(sizeof(*page), GFP_NOFS); | |
1361 | if (!page) { | |
1362 | leave_nomem: | |
d9d181c1 SB |
1363 | spin_lock(&sctx->stat_lock); |
1364 | sctx->stat.malloc_errors++; | |
1365 | spin_unlock(&sctx->stat_lock); | |
af8e2d1d | 1366 | scrub_put_recover(recover); |
b5d67f64 SB |
1367 | return -ENOMEM; |
1368 | } | |
7a9e9987 SB |
1369 | scrub_page_get(page); |
1370 | sblock->pagev[page_index] = page; | |
1371 | page->logical = logical; | |
af8e2d1d | 1372 | |
8e5cfb55 | 1373 | scrub_stripe_index_and_offset(logical, bbio->raid_map, |
af8e2d1d | 1374 | mapped_length, |
e34c330d ZL |
1375 | bbio->num_stripes - |
1376 | bbio->num_tgtdevs, | |
af8e2d1d MX |
1377 | mirror_index, |
1378 | &stripe_index, | |
1379 | &stripe_offset); | |
1380 | page->physical = bbio->stripes[stripe_index].physical + | |
1381 | stripe_offset; | |
1382 | page->dev = bbio->stripes[stripe_index].dev; | |
1383 | ||
ff023aac SB |
1384 | BUG_ON(page_index >= original_sblock->page_count); |
1385 | page->physical_for_dev_replace = | |
1386 | original_sblock->pagev[page_index]-> | |
1387 | physical_for_dev_replace; | |
7a9e9987 | 1388 | /* for missing devices, dev->bdev is NULL */ |
7a9e9987 | 1389 | page->mirror_num = mirror_index + 1; |
b5d67f64 | 1390 | sblock->page_count++; |
7a9e9987 SB |
1391 | page->page = alloc_page(GFP_NOFS); |
1392 | if (!page->page) | |
1393 | goto leave_nomem; | |
af8e2d1d MX |
1394 | |
1395 | scrub_get_recover(recover); | |
1396 | page->recover = recover; | |
b5d67f64 | 1397 | } |
af8e2d1d | 1398 | scrub_put_recover(recover); |
b5d67f64 SB |
1399 | length -= sublen; |
1400 | logical += sublen; | |
1401 | page_index++; | |
1402 | } | |
1403 | ||
1404 | return 0; | |
96e36920 ID |
1405 | } |
1406 | ||
af8e2d1d MX |
1407 | struct scrub_bio_ret { |
1408 | struct completion event; | |
1409 | int error; | |
1410 | }; | |
1411 | ||
1412 | static void scrub_bio_wait_endio(struct bio *bio, int error) | |
1413 | { | |
1414 | struct scrub_bio_ret *ret = bio->bi_private; | |
1415 | ||
1416 | ret->error = error; | |
1417 | complete(&ret->event); | |
1418 | } | |
1419 | ||
1420 | static inline int scrub_is_page_on_raid56(struct scrub_page *page) | |
1421 | { | |
8e5cfb55 | 1422 | return page->recover && page->recover->bbio->raid_map; |
af8e2d1d MX |
1423 | } |
1424 | ||
1425 | static int scrub_submit_raid56_bio_wait(struct btrfs_fs_info *fs_info, | |
1426 | struct bio *bio, | |
1427 | struct scrub_page *page) | |
1428 | { | |
1429 | struct scrub_bio_ret done; | |
1430 | int ret; | |
1431 | ||
1432 | init_completion(&done.event); | |
1433 | done.error = 0; | |
1434 | bio->bi_iter.bi_sector = page->logical >> 9; | |
1435 | bio->bi_private = &done; | |
1436 | bio->bi_end_io = scrub_bio_wait_endio; | |
1437 | ||
1438 | ret = raid56_parity_recover(fs_info->fs_root, bio, page->recover->bbio, | |
af8e2d1d | 1439 | page->recover->map_length, |
4245215d | 1440 | page->mirror_num, 0); |
af8e2d1d MX |
1441 | if (ret) |
1442 | return ret; | |
1443 | ||
1444 | wait_for_completion(&done.event); | |
1445 | if (done.error) | |
1446 | return -EIO; | |
1447 | ||
1448 | return 0; | |
1449 | } | |
1450 | ||
b5d67f64 SB |
1451 | /* |
1452 | * this function will check the on disk data for checksum errors, header | |
1453 | * errors and read I/O errors. If any I/O errors happen, the exact pages | |
1454 | * which are errored are marked as being bad. The goal is to enable scrub | |
1455 | * to take those pages that are not errored from all the mirrors so that | |
1456 | * the pages that are errored in the just handled mirror can be repaired. | |
1457 | */ | |
34f5c8e9 SB |
1458 | static void scrub_recheck_block(struct btrfs_fs_info *fs_info, |
1459 | struct scrub_block *sblock, int is_metadata, | |
1460 | int have_csum, u8 *csum, u64 generation, | |
af8e2d1d | 1461 | u16 csum_size, int retry_failed_mirror) |
96e36920 | 1462 | { |
b5d67f64 | 1463 | int page_num; |
96e36920 | 1464 | |
b5d67f64 SB |
1465 | sblock->no_io_error_seen = 1; |
1466 | sblock->header_error = 0; | |
1467 | sblock->checksum_error = 0; | |
96e36920 | 1468 | |
b5d67f64 SB |
1469 | for (page_num = 0; page_num < sblock->page_count; page_num++) { |
1470 | struct bio *bio; | |
7a9e9987 | 1471 | struct scrub_page *page = sblock->pagev[page_num]; |
b5d67f64 | 1472 | |
442a4f63 | 1473 | if (page->dev->bdev == NULL) { |
ea9947b4 SB |
1474 | page->io_error = 1; |
1475 | sblock->no_io_error_seen = 0; | |
1476 | continue; | |
1477 | } | |
1478 | ||
7a9e9987 | 1479 | WARN_ON(!page->page); |
9be3395b | 1480 | bio = btrfs_io_bio_alloc(GFP_NOFS, 1); |
34f5c8e9 SB |
1481 | if (!bio) { |
1482 | page->io_error = 1; | |
1483 | sblock->no_io_error_seen = 0; | |
1484 | continue; | |
1485 | } | |
442a4f63 | 1486 | bio->bi_bdev = page->dev->bdev; |
b5d67f64 | 1487 | |
34f5c8e9 | 1488 | bio_add_page(bio, page->page, PAGE_SIZE, 0); |
af8e2d1d MX |
1489 | if (!retry_failed_mirror && scrub_is_page_on_raid56(page)) { |
1490 | if (scrub_submit_raid56_bio_wait(fs_info, bio, page)) | |
1491 | sblock->no_io_error_seen = 0; | |
1492 | } else { | |
1493 | bio->bi_iter.bi_sector = page->physical >> 9; | |
1494 | ||
1495 | if (btrfsic_submit_bio_wait(READ, bio)) | |
1496 | sblock->no_io_error_seen = 0; | |
1497 | } | |
33879d45 | 1498 | |
b5d67f64 SB |
1499 | bio_put(bio); |
1500 | } | |
96e36920 | 1501 | |
b5d67f64 SB |
1502 | if (sblock->no_io_error_seen) |
1503 | scrub_recheck_block_checksum(fs_info, sblock, is_metadata, | |
1504 | have_csum, csum, generation, | |
1505 | csum_size); | |
1506 | ||
34f5c8e9 | 1507 | return; |
a2de733c AJ |
1508 | } |
1509 | ||
17a9be2f MX |
1510 | static inline int scrub_check_fsid(u8 fsid[], |
1511 | struct scrub_page *spage) | |
1512 | { | |
1513 | struct btrfs_fs_devices *fs_devices = spage->dev->fs_devices; | |
1514 | int ret; | |
1515 | ||
1516 | ret = memcmp(fsid, fs_devices->fsid, BTRFS_UUID_SIZE); | |
1517 | return !ret; | |
1518 | } | |
1519 | ||
b5d67f64 SB |
1520 | static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info, |
1521 | struct scrub_block *sblock, | |
1522 | int is_metadata, int have_csum, | |
1523 | const u8 *csum, u64 generation, | |
1524 | u16 csum_size) | |
a2de733c | 1525 | { |
b5d67f64 SB |
1526 | int page_num; |
1527 | u8 calculated_csum[BTRFS_CSUM_SIZE]; | |
1528 | u32 crc = ~(u32)0; | |
b5d67f64 SB |
1529 | void *mapped_buffer; |
1530 | ||
7a9e9987 | 1531 | WARN_ON(!sblock->pagev[0]->page); |
b5d67f64 SB |
1532 | if (is_metadata) { |
1533 | struct btrfs_header *h; | |
1534 | ||
7a9e9987 | 1535 | mapped_buffer = kmap_atomic(sblock->pagev[0]->page); |
b5d67f64 SB |
1536 | h = (struct btrfs_header *)mapped_buffer; |
1537 | ||
3cae210f | 1538 | if (sblock->pagev[0]->logical != btrfs_stack_header_bytenr(h) || |
17a9be2f | 1539 | !scrub_check_fsid(h->fsid, sblock->pagev[0]) || |
b5d67f64 | 1540 | memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid, |
442a4f63 | 1541 | BTRFS_UUID_SIZE)) { |
b5d67f64 | 1542 | sblock->header_error = 1; |
3cae210f | 1543 | } else if (generation != btrfs_stack_header_generation(h)) { |
442a4f63 SB |
1544 | sblock->header_error = 1; |
1545 | sblock->generation_error = 1; | |
1546 | } | |
b5d67f64 SB |
1547 | csum = h->csum; |
1548 | } else { | |
1549 | if (!have_csum) | |
1550 | return; | |
a2de733c | 1551 | |
7a9e9987 | 1552 | mapped_buffer = kmap_atomic(sblock->pagev[0]->page); |
b5d67f64 | 1553 | } |
a2de733c | 1554 | |
b5d67f64 SB |
1555 | for (page_num = 0;;) { |
1556 | if (page_num == 0 && is_metadata) | |
b0496686 | 1557 | crc = btrfs_csum_data( |
b5d67f64 SB |
1558 | ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE, |
1559 | crc, PAGE_SIZE - BTRFS_CSUM_SIZE); | |
1560 | else | |
b0496686 | 1561 | crc = btrfs_csum_data(mapped_buffer, crc, PAGE_SIZE); |
b5d67f64 | 1562 | |
9613bebb | 1563 | kunmap_atomic(mapped_buffer); |
b5d67f64 SB |
1564 | page_num++; |
1565 | if (page_num >= sblock->page_count) | |
1566 | break; | |
7a9e9987 | 1567 | WARN_ON(!sblock->pagev[page_num]->page); |
b5d67f64 | 1568 | |
7a9e9987 | 1569 | mapped_buffer = kmap_atomic(sblock->pagev[page_num]->page); |
b5d67f64 SB |
1570 | } |
1571 | ||
1572 | btrfs_csum_final(crc, calculated_csum); | |
1573 | if (memcmp(calculated_csum, csum, csum_size)) | |
1574 | sblock->checksum_error = 1; | |
a2de733c AJ |
1575 | } |
1576 | ||
b5d67f64 | 1577 | static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad, |
114ab50d | 1578 | struct scrub_block *sblock_good) |
b5d67f64 SB |
1579 | { |
1580 | int page_num; | |
1581 | int ret = 0; | |
96e36920 | 1582 | |
b5d67f64 SB |
1583 | for (page_num = 0; page_num < sblock_bad->page_count; page_num++) { |
1584 | int ret_sub; | |
96e36920 | 1585 | |
b5d67f64 SB |
1586 | ret_sub = scrub_repair_page_from_good_copy(sblock_bad, |
1587 | sblock_good, | |
114ab50d | 1588 | page_num, 1); |
b5d67f64 SB |
1589 | if (ret_sub) |
1590 | ret = ret_sub; | |
a2de733c | 1591 | } |
b5d67f64 SB |
1592 | |
1593 | return ret; | |
1594 | } | |
1595 | ||
1596 | static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad, | |
1597 | struct scrub_block *sblock_good, | |
1598 | int page_num, int force_write) | |
1599 | { | |
7a9e9987 SB |
1600 | struct scrub_page *page_bad = sblock_bad->pagev[page_num]; |
1601 | struct scrub_page *page_good = sblock_good->pagev[page_num]; | |
b5d67f64 | 1602 | |
7a9e9987 SB |
1603 | BUG_ON(page_bad->page == NULL); |
1604 | BUG_ON(page_good->page == NULL); | |
b5d67f64 SB |
1605 | if (force_write || sblock_bad->header_error || |
1606 | sblock_bad->checksum_error || page_bad->io_error) { | |
1607 | struct bio *bio; | |
1608 | int ret; | |
b5d67f64 | 1609 | |
ff023aac | 1610 | if (!page_bad->dev->bdev) { |
efe120a0 FH |
1611 | printk_ratelimited(KERN_WARNING "BTRFS: " |
1612 | "scrub_repair_page_from_good_copy(bdev == NULL) " | |
1613 | "is unexpected!\n"); | |
ff023aac SB |
1614 | return -EIO; |
1615 | } | |
1616 | ||
9be3395b | 1617 | bio = btrfs_io_bio_alloc(GFP_NOFS, 1); |
e627ee7b TI |
1618 | if (!bio) |
1619 | return -EIO; | |
442a4f63 | 1620 | bio->bi_bdev = page_bad->dev->bdev; |
4f024f37 | 1621 | bio->bi_iter.bi_sector = page_bad->physical >> 9; |
b5d67f64 SB |
1622 | |
1623 | ret = bio_add_page(bio, page_good->page, PAGE_SIZE, 0); | |
1624 | if (PAGE_SIZE != ret) { | |
1625 | bio_put(bio); | |
1626 | return -EIO; | |
13db62b7 | 1627 | } |
b5d67f64 | 1628 | |
33879d45 | 1629 | if (btrfsic_submit_bio_wait(WRITE, bio)) { |
442a4f63 SB |
1630 | btrfs_dev_stat_inc_and_print(page_bad->dev, |
1631 | BTRFS_DEV_STAT_WRITE_ERRS); | |
ff023aac SB |
1632 | btrfs_dev_replace_stats_inc( |
1633 | &sblock_bad->sctx->dev_root->fs_info-> | |
1634 | dev_replace.num_write_errors); | |
442a4f63 SB |
1635 | bio_put(bio); |
1636 | return -EIO; | |
1637 | } | |
b5d67f64 | 1638 | bio_put(bio); |
a2de733c AJ |
1639 | } |
1640 | ||
b5d67f64 SB |
1641 | return 0; |
1642 | } | |
1643 | ||
ff023aac SB |
1644 | static void scrub_write_block_to_dev_replace(struct scrub_block *sblock) |
1645 | { | |
1646 | int page_num; | |
1647 | ||
5a6ac9ea MX |
1648 | /* |
1649 | * This block is used for the check of the parity on the source device, | |
1650 | * so the data needn't be written into the destination device. | |
1651 | */ | |
1652 | if (sblock->sparity) | |
1653 | return; | |
1654 | ||
ff023aac SB |
1655 | for (page_num = 0; page_num < sblock->page_count; page_num++) { |
1656 | int ret; | |
1657 | ||
1658 | ret = scrub_write_page_to_dev_replace(sblock, page_num); | |
1659 | if (ret) | |
1660 | btrfs_dev_replace_stats_inc( | |
1661 | &sblock->sctx->dev_root->fs_info->dev_replace. | |
1662 | num_write_errors); | |
1663 | } | |
1664 | } | |
1665 | ||
1666 | static int scrub_write_page_to_dev_replace(struct scrub_block *sblock, | |
1667 | int page_num) | |
1668 | { | |
1669 | struct scrub_page *spage = sblock->pagev[page_num]; | |
1670 | ||
1671 | BUG_ON(spage->page == NULL); | |
1672 | if (spage->io_error) { | |
1673 | void *mapped_buffer = kmap_atomic(spage->page); | |
1674 | ||
1675 | memset(mapped_buffer, 0, PAGE_CACHE_SIZE); | |
1676 | flush_dcache_page(spage->page); | |
1677 | kunmap_atomic(mapped_buffer); | |
1678 | } | |
1679 | return scrub_add_page_to_wr_bio(sblock->sctx, spage); | |
1680 | } | |
1681 | ||
1682 | static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx, | |
1683 | struct scrub_page *spage) | |
1684 | { | |
1685 | struct scrub_wr_ctx *wr_ctx = &sctx->wr_ctx; | |
1686 | struct scrub_bio *sbio; | |
1687 | int ret; | |
1688 | ||
1689 | mutex_lock(&wr_ctx->wr_lock); | |
1690 | again: | |
1691 | if (!wr_ctx->wr_curr_bio) { | |
1692 | wr_ctx->wr_curr_bio = kzalloc(sizeof(*wr_ctx->wr_curr_bio), | |
1693 | GFP_NOFS); | |
1694 | if (!wr_ctx->wr_curr_bio) { | |
1695 | mutex_unlock(&wr_ctx->wr_lock); | |
1696 | return -ENOMEM; | |
1697 | } | |
1698 | wr_ctx->wr_curr_bio->sctx = sctx; | |
1699 | wr_ctx->wr_curr_bio->page_count = 0; | |
1700 | } | |
1701 | sbio = wr_ctx->wr_curr_bio; | |
1702 | if (sbio->page_count == 0) { | |
1703 | struct bio *bio; | |
1704 | ||
1705 | sbio->physical = spage->physical_for_dev_replace; | |
1706 | sbio->logical = spage->logical; | |
1707 | sbio->dev = wr_ctx->tgtdev; | |
1708 | bio = sbio->bio; | |
1709 | if (!bio) { | |
9be3395b | 1710 | bio = btrfs_io_bio_alloc(GFP_NOFS, wr_ctx->pages_per_wr_bio); |
ff023aac SB |
1711 | if (!bio) { |
1712 | mutex_unlock(&wr_ctx->wr_lock); | |
1713 | return -ENOMEM; | |
1714 | } | |
1715 | sbio->bio = bio; | |
1716 | } | |
1717 | ||
1718 | bio->bi_private = sbio; | |
1719 | bio->bi_end_io = scrub_wr_bio_end_io; | |
1720 | bio->bi_bdev = sbio->dev->bdev; | |
4f024f37 | 1721 | bio->bi_iter.bi_sector = sbio->physical >> 9; |
ff023aac SB |
1722 | sbio->err = 0; |
1723 | } else if (sbio->physical + sbio->page_count * PAGE_SIZE != | |
1724 | spage->physical_for_dev_replace || | |
1725 | sbio->logical + sbio->page_count * PAGE_SIZE != | |
1726 | spage->logical) { | |
1727 | scrub_wr_submit(sctx); | |
1728 | goto again; | |
1729 | } | |
1730 | ||
1731 | ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0); | |
1732 | if (ret != PAGE_SIZE) { | |
1733 | if (sbio->page_count < 1) { | |
1734 | bio_put(sbio->bio); | |
1735 | sbio->bio = NULL; | |
1736 | mutex_unlock(&wr_ctx->wr_lock); | |
1737 | return -EIO; | |
1738 | } | |
1739 | scrub_wr_submit(sctx); | |
1740 | goto again; | |
1741 | } | |
1742 | ||
1743 | sbio->pagev[sbio->page_count] = spage; | |
1744 | scrub_page_get(spage); | |
1745 | sbio->page_count++; | |
1746 | if (sbio->page_count == wr_ctx->pages_per_wr_bio) | |
1747 | scrub_wr_submit(sctx); | |
1748 | mutex_unlock(&wr_ctx->wr_lock); | |
1749 | ||
1750 | return 0; | |
1751 | } | |
1752 | ||
1753 | static void scrub_wr_submit(struct scrub_ctx *sctx) | |
1754 | { | |
1755 | struct scrub_wr_ctx *wr_ctx = &sctx->wr_ctx; | |
1756 | struct scrub_bio *sbio; | |
1757 | ||
1758 | if (!wr_ctx->wr_curr_bio) | |
1759 | return; | |
1760 | ||
1761 | sbio = wr_ctx->wr_curr_bio; | |
1762 | wr_ctx->wr_curr_bio = NULL; | |
1763 | WARN_ON(!sbio->bio->bi_bdev); | |
1764 | scrub_pending_bio_inc(sctx); | |
1765 | /* process all writes in a single worker thread. Then the block layer | |
1766 | * orders the requests before sending them to the driver which | |
1767 | * doubled the write performance on spinning disks when measured | |
1768 | * with Linux 3.5 */ | |
1769 | btrfsic_submit_bio(WRITE, sbio->bio); | |
1770 | } | |
1771 | ||
1772 | static void scrub_wr_bio_end_io(struct bio *bio, int err) | |
1773 | { | |
1774 | struct scrub_bio *sbio = bio->bi_private; | |
1775 | struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info; | |
1776 | ||
1777 | sbio->err = err; | |
1778 | sbio->bio = bio; | |
1779 | ||
9e0af237 LB |
1780 | btrfs_init_work(&sbio->work, btrfs_scrubwrc_helper, |
1781 | scrub_wr_bio_end_io_worker, NULL, NULL); | |
0339ef2f | 1782 | btrfs_queue_work(fs_info->scrub_wr_completion_workers, &sbio->work); |
ff023aac SB |
1783 | } |
1784 | ||
1785 | static void scrub_wr_bio_end_io_worker(struct btrfs_work *work) | |
1786 | { | |
1787 | struct scrub_bio *sbio = container_of(work, struct scrub_bio, work); | |
1788 | struct scrub_ctx *sctx = sbio->sctx; | |
1789 | int i; | |
1790 | ||
1791 | WARN_ON(sbio->page_count > SCRUB_PAGES_PER_WR_BIO); | |
1792 | if (sbio->err) { | |
1793 | struct btrfs_dev_replace *dev_replace = | |
1794 | &sbio->sctx->dev_root->fs_info->dev_replace; | |
1795 | ||
1796 | for (i = 0; i < sbio->page_count; i++) { | |
1797 | struct scrub_page *spage = sbio->pagev[i]; | |
1798 | ||
1799 | spage->io_error = 1; | |
1800 | btrfs_dev_replace_stats_inc(&dev_replace-> | |
1801 | num_write_errors); | |
1802 | } | |
1803 | } | |
1804 | ||
1805 | for (i = 0; i < sbio->page_count; i++) | |
1806 | scrub_page_put(sbio->pagev[i]); | |
1807 | ||
1808 | bio_put(sbio->bio); | |
1809 | kfree(sbio); | |
1810 | scrub_pending_bio_dec(sctx); | |
1811 | } | |
1812 | ||
1813 | static int scrub_checksum(struct scrub_block *sblock) | |
b5d67f64 SB |
1814 | { |
1815 | u64 flags; | |
1816 | int ret; | |
1817 | ||
7a9e9987 SB |
1818 | WARN_ON(sblock->page_count < 1); |
1819 | flags = sblock->pagev[0]->flags; | |
b5d67f64 SB |
1820 | ret = 0; |
1821 | if (flags & BTRFS_EXTENT_FLAG_DATA) | |
1822 | ret = scrub_checksum_data(sblock); | |
1823 | else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) | |
1824 | ret = scrub_checksum_tree_block(sblock); | |
1825 | else if (flags & BTRFS_EXTENT_FLAG_SUPER) | |
1826 | (void)scrub_checksum_super(sblock); | |
1827 | else | |
1828 | WARN_ON(1); | |
1829 | if (ret) | |
1830 | scrub_handle_errored_block(sblock); | |
ff023aac SB |
1831 | |
1832 | return ret; | |
a2de733c AJ |
1833 | } |
1834 | ||
b5d67f64 | 1835 | static int scrub_checksum_data(struct scrub_block *sblock) |
a2de733c | 1836 | { |
d9d181c1 | 1837 | struct scrub_ctx *sctx = sblock->sctx; |
a2de733c | 1838 | u8 csum[BTRFS_CSUM_SIZE]; |
b5d67f64 SB |
1839 | u8 *on_disk_csum; |
1840 | struct page *page; | |
1841 | void *buffer; | |
a2de733c AJ |
1842 | u32 crc = ~(u32)0; |
1843 | int fail = 0; | |
b5d67f64 SB |
1844 | u64 len; |
1845 | int index; | |
a2de733c | 1846 | |
b5d67f64 | 1847 | BUG_ON(sblock->page_count < 1); |
7a9e9987 | 1848 | if (!sblock->pagev[0]->have_csum) |
a2de733c AJ |
1849 | return 0; |
1850 | ||
7a9e9987 SB |
1851 | on_disk_csum = sblock->pagev[0]->csum; |
1852 | page = sblock->pagev[0]->page; | |
9613bebb | 1853 | buffer = kmap_atomic(page); |
b5d67f64 | 1854 | |
d9d181c1 | 1855 | len = sctx->sectorsize; |
b5d67f64 SB |
1856 | index = 0; |
1857 | for (;;) { | |
1858 | u64 l = min_t(u64, len, PAGE_SIZE); | |
1859 | ||
b0496686 | 1860 | crc = btrfs_csum_data(buffer, crc, l); |
9613bebb | 1861 | kunmap_atomic(buffer); |
b5d67f64 SB |
1862 | len -= l; |
1863 | if (len == 0) | |
1864 | break; | |
1865 | index++; | |
1866 | BUG_ON(index >= sblock->page_count); | |
7a9e9987 SB |
1867 | BUG_ON(!sblock->pagev[index]->page); |
1868 | page = sblock->pagev[index]->page; | |
9613bebb | 1869 | buffer = kmap_atomic(page); |
b5d67f64 SB |
1870 | } |
1871 | ||
a2de733c | 1872 | btrfs_csum_final(crc, csum); |
d9d181c1 | 1873 | if (memcmp(csum, on_disk_csum, sctx->csum_size)) |
a2de733c AJ |
1874 | fail = 1; |
1875 | ||
a2de733c AJ |
1876 | return fail; |
1877 | } | |
1878 | ||
b5d67f64 | 1879 | static int scrub_checksum_tree_block(struct scrub_block *sblock) |
a2de733c | 1880 | { |
d9d181c1 | 1881 | struct scrub_ctx *sctx = sblock->sctx; |
a2de733c | 1882 | struct btrfs_header *h; |
a36cf8b8 | 1883 | struct btrfs_root *root = sctx->dev_root; |
a2de733c | 1884 | struct btrfs_fs_info *fs_info = root->fs_info; |
b5d67f64 SB |
1885 | u8 calculated_csum[BTRFS_CSUM_SIZE]; |
1886 | u8 on_disk_csum[BTRFS_CSUM_SIZE]; | |
1887 | struct page *page; | |
1888 | void *mapped_buffer; | |
1889 | u64 mapped_size; | |
1890 | void *p; | |
a2de733c AJ |
1891 | u32 crc = ~(u32)0; |
1892 | int fail = 0; | |
1893 | int crc_fail = 0; | |
b5d67f64 SB |
1894 | u64 len; |
1895 | int index; | |
1896 | ||
1897 | BUG_ON(sblock->page_count < 1); | |
7a9e9987 | 1898 | page = sblock->pagev[0]->page; |
9613bebb | 1899 | mapped_buffer = kmap_atomic(page); |
b5d67f64 | 1900 | h = (struct btrfs_header *)mapped_buffer; |
d9d181c1 | 1901 | memcpy(on_disk_csum, h->csum, sctx->csum_size); |
a2de733c AJ |
1902 | |
1903 | /* | |
1904 | * we don't use the getter functions here, as we | |
1905 | * a) don't have an extent buffer and | |
1906 | * b) the page is already kmapped | |
1907 | */ | |
a2de733c | 1908 | |
3cae210f | 1909 | if (sblock->pagev[0]->logical != btrfs_stack_header_bytenr(h)) |
a2de733c AJ |
1910 | ++fail; |
1911 | ||
3cae210f | 1912 | if (sblock->pagev[0]->generation != btrfs_stack_header_generation(h)) |
a2de733c AJ |
1913 | ++fail; |
1914 | ||
17a9be2f | 1915 | if (!scrub_check_fsid(h->fsid, sblock->pagev[0])) |
a2de733c AJ |
1916 | ++fail; |
1917 | ||
1918 | if (memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid, | |
1919 | BTRFS_UUID_SIZE)) | |
1920 | ++fail; | |
1921 | ||
d9d181c1 | 1922 | len = sctx->nodesize - BTRFS_CSUM_SIZE; |
b5d67f64 SB |
1923 | mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE; |
1924 | p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE; | |
1925 | index = 0; | |
1926 | for (;;) { | |
1927 | u64 l = min_t(u64, len, mapped_size); | |
1928 | ||
b0496686 | 1929 | crc = btrfs_csum_data(p, crc, l); |
9613bebb | 1930 | kunmap_atomic(mapped_buffer); |
b5d67f64 SB |
1931 | len -= l; |
1932 | if (len == 0) | |
1933 | break; | |
1934 | index++; | |
1935 | BUG_ON(index >= sblock->page_count); | |
7a9e9987 SB |
1936 | BUG_ON(!sblock->pagev[index]->page); |
1937 | page = sblock->pagev[index]->page; | |
9613bebb | 1938 | mapped_buffer = kmap_atomic(page); |
b5d67f64 SB |
1939 | mapped_size = PAGE_SIZE; |
1940 | p = mapped_buffer; | |
1941 | } | |
1942 | ||
1943 | btrfs_csum_final(crc, calculated_csum); | |
d9d181c1 | 1944 | if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size)) |
a2de733c AJ |
1945 | ++crc_fail; |
1946 | ||
a2de733c AJ |
1947 | return fail || crc_fail; |
1948 | } | |
1949 | ||
b5d67f64 | 1950 | static int scrub_checksum_super(struct scrub_block *sblock) |
a2de733c AJ |
1951 | { |
1952 | struct btrfs_super_block *s; | |
d9d181c1 | 1953 | struct scrub_ctx *sctx = sblock->sctx; |
b5d67f64 SB |
1954 | u8 calculated_csum[BTRFS_CSUM_SIZE]; |
1955 | u8 on_disk_csum[BTRFS_CSUM_SIZE]; | |
1956 | struct page *page; | |
1957 | void *mapped_buffer; | |
1958 | u64 mapped_size; | |
1959 | void *p; | |
a2de733c | 1960 | u32 crc = ~(u32)0; |
442a4f63 SB |
1961 | int fail_gen = 0; |
1962 | int fail_cor = 0; | |
b5d67f64 SB |
1963 | u64 len; |
1964 | int index; | |
a2de733c | 1965 | |
b5d67f64 | 1966 | BUG_ON(sblock->page_count < 1); |
7a9e9987 | 1967 | page = sblock->pagev[0]->page; |
9613bebb | 1968 | mapped_buffer = kmap_atomic(page); |
b5d67f64 | 1969 | s = (struct btrfs_super_block *)mapped_buffer; |
d9d181c1 | 1970 | memcpy(on_disk_csum, s->csum, sctx->csum_size); |
a2de733c | 1971 | |
3cae210f | 1972 | if (sblock->pagev[0]->logical != btrfs_super_bytenr(s)) |
442a4f63 | 1973 | ++fail_cor; |
a2de733c | 1974 | |
3cae210f | 1975 | if (sblock->pagev[0]->generation != btrfs_super_generation(s)) |
442a4f63 | 1976 | ++fail_gen; |
a2de733c | 1977 | |
17a9be2f | 1978 | if (!scrub_check_fsid(s->fsid, sblock->pagev[0])) |
442a4f63 | 1979 | ++fail_cor; |
a2de733c | 1980 | |
b5d67f64 SB |
1981 | len = BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE; |
1982 | mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE; | |
1983 | p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE; | |
1984 | index = 0; | |
1985 | for (;;) { | |
1986 | u64 l = min_t(u64, len, mapped_size); | |
1987 | ||
b0496686 | 1988 | crc = btrfs_csum_data(p, crc, l); |
9613bebb | 1989 | kunmap_atomic(mapped_buffer); |
b5d67f64 SB |
1990 | len -= l; |
1991 | if (len == 0) | |
1992 | break; | |
1993 | index++; | |
1994 | BUG_ON(index >= sblock->page_count); | |
7a9e9987 SB |
1995 | BUG_ON(!sblock->pagev[index]->page); |
1996 | page = sblock->pagev[index]->page; | |
9613bebb | 1997 | mapped_buffer = kmap_atomic(page); |
b5d67f64 SB |
1998 | mapped_size = PAGE_SIZE; |
1999 | p = mapped_buffer; | |
2000 | } | |
2001 | ||
2002 | btrfs_csum_final(crc, calculated_csum); | |
d9d181c1 | 2003 | if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size)) |
442a4f63 | 2004 | ++fail_cor; |
a2de733c | 2005 | |
442a4f63 | 2006 | if (fail_cor + fail_gen) { |
a2de733c AJ |
2007 | /* |
2008 | * if we find an error in a super block, we just report it. | |
2009 | * They will get written with the next transaction commit | |
2010 | * anyway | |
2011 | */ | |
d9d181c1 SB |
2012 | spin_lock(&sctx->stat_lock); |
2013 | ++sctx->stat.super_errors; | |
2014 | spin_unlock(&sctx->stat_lock); | |
442a4f63 | 2015 | if (fail_cor) |
7a9e9987 | 2016 | btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev, |
442a4f63 SB |
2017 | BTRFS_DEV_STAT_CORRUPTION_ERRS); |
2018 | else | |
7a9e9987 | 2019 | btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev, |
442a4f63 | 2020 | BTRFS_DEV_STAT_GENERATION_ERRS); |
a2de733c AJ |
2021 | } |
2022 | ||
442a4f63 | 2023 | return fail_cor + fail_gen; |
a2de733c AJ |
2024 | } |
2025 | ||
b5d67f64 SB |
2026 | static void scrub_block_get(struct scrub_block *sblock) |
2027 | { | |
2028 | atomic_inc(&sblock->ref_count); | |
2029 | } | |
2030 | ||
2031 | static void scrub_block_put(struct scrub_block *sblock) | |
2032 | { | |
2033 | if (atomic_dec_and_test(&sblock->ref_count)) { | |
2034 | int i; | |
2035 | ||
5a6ac9ea MX |
2036 | if (sblock->sparity) |
2037 | scrub_parity_put(sblock->sparity); | |
2038 | ||
b5d67f64 | 2039 | for (i = 0; i < sblock->page_count; i++) |
7a9e9987 | 2040 | scrub_page_put(sblock->pagev[i]); |
b5d67f64 SB |
2041 | kfree(sblock); |
2042 | } | |
2043 | } | |
2044 | ||
7a9e9987 SB |
2045 | static void scrub_page_get(struct scrub_page *spage) |
2046 | { | |
2047 | atomic_inc(&spage->ref_count); | |
2048 | } | |
2049 | ||
2050 | static void scrub_page_put(struct scrub_page *spage) | |
2051 | { | |
2052 | if (atomic_dec_and_test(&spage->ref_count)) { | |
2053 | if (spage->page) | |
2054 | __free_page(spage->page); | |
2055 | kfree(spage); | |
2056 | } | |
2057 | } | |
2058 | ||
d9d181c1 | 2059 | static void scrub_submit(struct scrub_ctx *sctx) |
a2de733c AJ |
2060 | { |
2061 | struct scrub_bio *sbio; | |
2062 | ||
d9d181c1 | 2063 | if (sctx->curr == -1) |
1623edeb | 2064 | return; |
a2de733c | 2065 | |
d9d181c1 SB |
2066 | sbio = sctx->bios[sctx->curr]; |
2067 | sctx->curr = -1; | |
b6bfebc1 | 2068 | scrub_pending_bio_inc(sctx); |
a2de733c | 2069 | |
ff023aac SB |
2070 | if (!sbio->bio->bi_bdev) { |
2071 | /* | |
2072 | * this case should not happen. If btrfs_map_block() is | |
2073 | * wrong, it could happen for dev-replace operations on | |
2074 | * missing devices when no mirrors are available, but in | |
2075 | * this case it should already fail the mount. | |
2076 | * This case is handled correctly (but _very_ slowly). | |
2077 | */ | |
2078 | printk_ratelimited(KERN_WARNING | |
efe120a0 | 2079 | "BTRFS: scrub_submit(bio bdev == NULL) is unexpected!\n"); |
ff023aac SB |
2080 | bio_endio(sbio->bio, -EIO); |
2081 | } else { | |
2082 | btrfsic_submit_bio(READ, sbio->bio); | |
2083 | } | |
a2de733c AJ |
2084 | } |
2085 | ||
ff023aac SB |
2086 | static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx, |
2087 | struct scrub_page *spage) | |
a2de733c | 2088 | { |
b5d67f64 | 2089 | struct scrub_block *sblock = spage->sblock; |
a2de733c | 2090 | struct scrub_bio *sbio; |
69f4cb52 | 2091 | int ret; |
a2de733c AJ |
2092 | |
2093 | again: | |
2094 | /* | |
2095 | * grab a fresh bio or wait for one to become available | |
2096 | */ | |
d9d181c1 SB |
2097 | while (sctx->curr == -1) { |
2098 | spin_lock(&sctx->list_lock); | |
2099 | sctx->curr = sctx->first_free; | |
2100 | if (sctx->curr != -1) { | |
2101 | sctx->first_free = sctx->bios[sctx->curr]->next_free; | |
2102 | sctx->bios[sctx->curr]->next_free = -1; | |
2103 | sctx->bios[sctx->curr]->page_count = 0; | |
2104 | spin_unlock(&sctx->list_lock); | |
a2de733c | 2105 | } else { |
d9d181c1 SB |
2106 | spin_unlock(&sctx->list_lock); |
2107 | wait_event(sctx->list_wait, sctx->first_free != -1); | |
a2de733c AJ |
2108 | } |
2109 | } | |
d9d181c1 | 2110 | sbio = sctx->bios[sctx->curr]; |
b5d67f64 | 2111 | if (sbio->page_count == 0) { |
69f4cb52 AJ |
2112 | struct bio *bio; |
2113 | ||
b5d67f64 SB |
2114 | sbio->physical = spage->physical; |
2115 | sbio->logical = spage->logical; | |
a36cf8b8 | 2116 | sbio->dev = spage->dev; |
b5d67f64 SB |
2117 | bio = sbio->bio; |
2118 | if (!bio) { | |
9be3395b | 2119 | bio = btrfs_io_bio_alloc(GFP_NOFS, sctx->pages_per_rd_bio); |
b5d67f64 SB |
2120 | if (!bio) |
2121 | return -ENOMEM; | |
2122 | sbio->bio = bio; | |
2123 | } | |
69f4cb52 AJ |
2124 | |
2125 | bio->bi_private = sbio; | |
2126 | bio->bi_end_io = scrub_bio_end_io; | |
a36cf8b8 | 2127 | bio->bi_bdev = sbio->dev->bdev; |
4f024f37 | 2128 | bio->bi_iter.bi_sector = sbio->physical >> 9; |
69f4cb52 | 2129 | sbio->err = 0; |
b5d67f64 SB |
2130 | } else if (sbio->physical + sbio->page_count * PAGE_SIZE != |
2131 | spage->physical || | |
2132 | sbio->logical + sbio->page_count * PAGE_SIZE != | |
a36cf8b8 SB |
2133 | spage->logical || |
2134 | sbio->dev != spage->dev) { | |
d9d181c1 | 2135 | scrub_submit(sctx); |
a2de733c AJ |
2136 | goto again; |
2137 | } | |
69f4cb52 | 2138 | |
b5d67f64 SB |
2139 | sbio->pagev[sbio->page_count] = spage; |
2140 | ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0); | |
2141 | if (ret != PAGE_SIZE) { | |
2142 | if (sbio->page_count < 1) { | |
2143 | bio_put(sbio->bio); | |
2144 | sbio->bio = NULL; | |
2145 | return -EIO; | |
2146 | } | |
d9d181c1 | 2147 | scrub_submit(sctx); |
69f4cb52 AJ |
2148 | goto again; |
2149 | } | |
2150 | ||
ff023aac | 2151 | scrub_block_get(sblock); /* one for the page added to the bio */ |
b5d67f64 SB |
2152 | atomic_inc(&sblock->outstanding_pages); |
2153 | sbio->page_count++; | |
ff023aac | 2154 | if (sbio->page_count == sctx->pages_per_rd_bio) |
d9d181c1 | 2155 | scrub_submit(sctx); |
b5d67f64 SB |
2156 | |
2157 | return 0; | |
2158 | } | |
2159 | ||
d9d181c1 | 2160 | static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len, |
a36cf8b8 | 2161 | u64 physical, struct btrfs_device *dev, u64 flags, |
ff023aac SB |
2162 | u64 gen, int mirror_num, u8 *csum, int force, |
2163 | u64 physical_for_dev_replace) | |
b5d67f64 SB |
2164 | { |
2165 | struct scrub_block *sblock; | |
2166 | int index; | |
2167 | ||
2168 | sblock = kzalloc(sizeof(*sblock), GFP_NOFS); | |
2169 | if (!sblock) { | |
d9d181c1 SB |
2170 | spin_lock(&sctx->stat_lock); |
2171 | sctx->stat.malloc_errors++; | |
2172 | spin_unlock(&sctx->stat_lock); | |
b5d67f64 | 2173 | return -ENOMEM; |
a2de733c | 2174 | } |
b5d67f64 | 2175 | |
7a9e9987 SB |
2176 | /* one ref inside this function, plus one for each page added to |
2177 | * a bio later on */ | |
b5d67f64 | 2178 | atomic_set(&sblock->ref_count, 1); |
d9d181c1 | 2179 | sblock->sctx = sctx; |
b5d67f64 SB |
2180 | sblock->no_io_error_seen = 1; |
2181 | ||
2182 | for (index = 0; len > 0; index++) { | |
7a9e9987 | 2183 | struct scrub_page *spage; |
b5d67f64 SB |
2184 | u64 l = min_t(u64, len, PAGE_SIZE); |
2185 | ||
7a9e9987 SB |
2186 | spage = kzalloc(sizeof(*spage), GFP_NOFS); |
2187 | if (!spage) { | |
2188 | leave_nomem: | |
d9d181c1 SB |
2189 | spin_lock(&sctx->stat_lock); |
2190 | sctx->stat.malloc_errors++; | |
2191 | spin_unlock(&sctx->stat_lock); | |
7a9e9987 | 2192 | scrub_block_put(sblock); |
b5d67f64 SB |
2193 | return -ENOMEM; |
2194 | } | |
7a9e9987 SB |
2195 | BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK); |
2196 | scrub_page_get(spage); | |
2197 | sblock->pagev[index] = spage; | |
b5d67f64 | 2198 | spage->sblock = sblock; |
a36cf8b8 | 2199 | spage->dev = dev; |
b5d67f64 SB |
2200 | spage->flags = flags; |
2201 | spage->generation = gen; | |
2202 | spage->logical = logical; | |
2203 | spage->physical = physical; | |
ff023aac | 2204 | spage->physical_for_dev_replace = physical_for_dev_replace; |
b5d67f64 SB |
2205 | spage->mirror_num = mirror_num; |
2206 | if (csum) { | |
2207 | spage->have_csum = 1; | |
d9d181c1 | 2208 | memcpy(spage->csum, csum, sctx->csum_size); |
b5d67f64 SB |
2209 | } else { |
2210 | spage->have_csum = 0; | |
2211 | } | |
2212 | sblock->page_count++; | |
7a9e9987 SB |
2213 | spage->page = alloc_page(GFP_NOFS); |
2214 | if (!spage->page) | |
2215 | goto leave_nomem; | |
b5d67f64 SB |
2216 | len -= l; |
2217 | logical += l; | |
2218 | physical += l; | |
ff023aac | 2219 | physical_for_dev_replace += l; |
b5d67f64 SB |
2220 | } |
2221 | ||
7a9e9987 | 2222 | WARN_ON(sblock->page_count == 0); |
b5d67f64 | 2223 | for (index = 0; index < sblock->page_count; index++) { |
7a9e9987 | 2224 | struct scrub_page *spage = sblock->pagev[index]; |
1bc87793 AJ |
2225 | int ret; |
2226 | ||
ff023aac | 2227 | ret = scrub_add_page_to_rd_bio(sctx, spage); |
b5d67f64 SB |
2228 | if (ret) { |
2229 | scrub_block_put(sblock); | |
1bc87793 | 2230 | return ret; |
b5d67f64 | 2231 | } |
1bc87793 | 2232 | } |
a2de733c | 2233 | |
b5d67f64 | 2234 | if (force) |
d9d181c1 | 2235 | scrub_submit(sctx); |
a2de733c | 2236 | |
b5d67f64 SB |
2237 | /* last one frees, either here or in bio completion for last page */ |
2238 | scrub_block_put(sblock); | |
a2de733c AJ |
2239 | return 0; |
2240 | } | |
2241 | ||
b5d67f64 SB |
2242 | static void scrub_bio_end_io(struct bio *bio, int err) |
2243 | { | |
2244 | struct scrub_bio *sbio = bio->bi_private; | |
a36cf8b8 | 2245 | struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info; |
b5d67f64 SB |
2246 | |
2247 | sbio->err = err; | |
2248 | sbio->bio = bio; | |
2249 | ||
0339ef2f | 2250 | btrfs_queue_work(fs_info->scrub_workers, &sbio->work); |
b5d67f64 SB |
2251 | } |
2252 | ||
2253 | static void scrub_bio_end_io_worker(struct btrfs_work *work) | |
2254 | { | |
2255 | struct scrub_bio *sbio = container_of(work, struct scrub_bio, work); | |
d9d181c1 | 2256 | struct scrub_ctx *sctx = sbio->sctx; |
b5d67f64 SB |
2257 | int i; |
2258 | ||
ff023aac | 2259 | BUG_ON(sbio->page_count > SCRUB_PAGES_PER_RD_BIO); |
b5d67f64 SB |
2260 | if (sbio->err) { |
2261 | for (i = 0; i < sbio->page_count; i++) { | |
2262 | struct scrub_page *spage = sbio->pagev[i]; | |
2263 | ||
2264 | spage->io_error = 1; | |
2265 | spage->sblock->no_io_error_seen = 0; | |
2266 | } | |
2267 | } | |
2268 | ||
2269 | /* now complete the scrub_block items that have all pages completed */ | |
2270 | for (i = 0; i < sbio->page_count; i++) { | |
2271 | struct scrub_page *spage = sbio->pagev[i]; | |
2272 | struct scrub_block *sblock = spage->sblock; | |
2273 | ||
2274 | if (atomic_dec_and_test(&sblock->outstanding_pages)) | |
2275 | scrub_block_complete(sblock); | |
2276 | scrub_block_put(sblock); | |
2277 | } | |
2278 | ||
b5d67f64 SB |
2279 | bio_put(sbio->bio); |
2280 | sbio->bio = NULL; | |
d9d181c1 SB |
2281 | spin_lock(&sctx->list_lock); |
2282 | sbio->next_free = sctx->first_free; | |
2283 | sctx->first_free = sbio->index; | |
2284 | spin_unlock(&sctx->list_lock); | |
ff023aac SB |
2285 | |
2286 | if (sctx->is_dev_replace && | |
2287 | atomic_read(&sctx->wr_ctx.flush_all_writes)) { | |
2288 | mutex_lock(&sctx->wr_ctx.wr_lock); | |
2289 | scrub_wr_submit(sctx); | |
2290 | mutex_unlock(&sctx->wr_ctx.wr_lock); | |
2291 | } | |
2292 | ||
b6bfebc1 | 2293 | scrub_pending_bio_dec(sctx); |
b5d67f64 SB |
2294 | } |
2295 | ||
5a6ac9ea MX |
2296 | static inline void __scrub_mark_bitmap(struct scrub_parity *sparity, |
2297 | unsigned long *bitmap, | |
2298 | u64 start, u64 len) | |
2299 | { | |
2300 | int offset; | |
2301 | int nsectors; | |
2302 | int sectorsize = sparity->sctx->dev_root->sectorsize; | |
2303 | ||
2304 | if (len >= sparity->stripe_len) { | |
2305 | bitmap_set(bitmap, 0, sparity->nsectors); | |
2306 | return; | |
2307 | } | |
2308 | ||
2309 | start -= sparity->logic_start; | |
2310 | offset = (int)do_div(start, sparity->stripe_len); | |
2311 | offset /= sectorsize; | |
2312 | nsectors = (int)len / sectorsize; | |
2313 | ||
2314 | if (offset + nsectors <= sparity->nsectors) { | |
2315 | bitmap_set(bitmap, offset, nsectors); | |
2316 | return; | |
2317 | } | |
2318 | ||
2319 | bitmap_set(bitmap, offset, sparity->nsectors - offset); | |
2320 | bitmap_set(bitmap, 0, nsectors - (sparity->nsectors - offset)); | |
2321 | } | |
2322 | ||
2323 | static inline void scrub_parity_mark_sectors_error(struct scrub_parity *sparity, | |
2324 | u64 start, u64 len) | |
2325 | { | |
2326 | __scrub_mark_bitmap(sparity, sparity->ebitmap, start, len); | |
2327 | } | |
2328 | ||
2329 | static inline void scrub_parity_mark_sectors_data(struct scrub_parity *sparity, | |
2330 | u64 start, u64 len) | |
2331 | { | |
2332 | __scrub_mark_bitmap(sparity, sparity->dbitmap, start, len); | |
2333 | } | |
2334 | ||
b5d67f64 SB |
2335 | static void scrub_block_complete(struct scrub_block *sblock) |
2336 | { | |
5a6ac9ea MX |
2337 | int corrupted = 0; |
2338 | ||
ff023aac | 2339 | if (!sblock->no_io_error_seen) { |
5a6ac9ea | 2340 | corrupted = 1; |
b5d67f64 | 2341 | scrub_handle_errored_block(sblock); |
ff023aac SB |
2342 | } else { |
2343 | /* | |
2344 | * if has checksum error, write via repair mechanism in | |
2345 | * dev replace case, otherwise write here in dev replace | |
2346 | * case. | |
2347 | */ | |
5a6ac9ea MX |
2348 | corrupted = scrub_checksum(sblock); |
2349 | if (!corrupted && sblock->sctx->is_dev_replace) | |
ff023aac SB |
2350 | scrub_write_block_to_dev_replace(sblock); |
2351 | } | |
5a6ac9ea MX |
2352 | |
2353 | if (sblock->sparity && corrupted && !sblock->data_corrected) { | |
2354 | u64 start = sblock->pagev[0]->logical; | |
2355 | u64 end = sblock->pagev[sblock->page_count - 1]->logical + | |
2356 | PAGE_SIZE; | |
2357 | ||
2358 | scrub_parity_mark_sectors_error(sblock->sparity, | |
2359 | start, end - start); | |
2360 | } | |
b5d67f64 SB |
2361 | } |
2362 | ||
d9d181c1 | 2363 | static int scrub_find_csum(struct scrub_ctx *sctx, u64 logical, u64 len, |
a2de733c AJ |
2364 | u8 *csum) |
2365 | { | |
2366 | struct btrfs_ordered_sum *sum = NULL; | |
f51a4a18 | 2367 | unsigned long index; |
a2de733c | 2368 | unsigned long num_sectors; |
a2de733c | 2369 | |
d9d181c1 SB |
2370 | while (!list_empty(&sctx->csum_list)) { |
2371 | sum = list_first_entry(&sctx->csum_list, | |
a2de733c AJ |
2372 | struct btrfs_ordered_sum, list); |
2373 | if (sum->bytenr > logical) | |
2374 | return 0; | |
2375 | if (sum->bytenr + sum->len > logical) | |
2376 | break; | |
2377 | ||
d9d181c1 | 2378 | ++sctx->stat.csum_discards; |
a2de733c AJ |
2379 | list_del(&sum->list); |
2380 | kfree(sum); | |
2381 | sum = NULL; | |
2382 | } | |
2383 | if (!sum) | |
2384 | return 0; | |
2385 | ||
f51a4a18 | 2386 | index = ((u32)(logical - sum->bytenr)) / sctx->sectorsize; |
d9d181c1 | 2387 | num_sectors = sum->len / sctx->sectorsize; |
f51a4a18 MX |
2388 | memcpy(csum, sum->sums + index, sctx->csum_size); |
2389 | if (index == num_sectors - 1) { | |
a2de733c AJ |
2390 | list_del(&sum->list); |
2391 | kfree(sum); | |
2392 | } | |
f51a4a18 | 2393 | return 1; |
a2de733c AJ |
2394 | } |
2395 | ||
2396 | /* scrub extent tries to collect up to 64 kB for each bio */ | |
d9d181c1 | 2397 | static int scrub_extent(struct scrub_ctx *sctx, u64 logical, u64 len, |
a36cf8b8 | 2398 | u64 physical, struct btrfs_device *dev, u64 flags, |
ff023aac | 2399 | u64 gen, int mirror_num, u64 physical_for_dev_replace) |
a2de733c AJ |
2400 | { |
2401 | int ret; | |
2402 | u8 csum[BTRFS_CSUM_SIZE]; | |
b5d67f64 SB |
2403 | u32 blocksize; |
2404 | ||
2405 | if (flags & BTRFS_EXTENT_FLAG_DATA) { | |
d9d181c1 SB |
2406 | blocksize = sctx->sectorsize; |
2407 | spin_lock(&sctx->stat_lock); | |
2408 | sctx->stat.data_extents_scrubbed++; | |
2409 | sctx->stat.data_bytes_scrubbed += len; | |
2410 | spin_unlock(&sctx->stat_lock); | |
b5d67f64 | 2411 | } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { |
d9d181c1 SB |
2412 | blocksize = sctx->nodesize; |
2413 | spin_lock(&sctx->stat_lock); | |
2414 | sctx->stat.tree_extents_scrubbed++; | |
2415 | sctx->stat.tree_bytes_scrubbed += len; | |
2416 | spin_unlock(&sctx->stat_lock); | |
b5d67f64 | 2417 | } else { |
d9d181c1 | 2418 | blocksize = sctx->sectorsize; |
ff023aac | 2419 | WARN_ON(1); |
b5d67f64 | 2420 | } |
a2de733c AJ |
2421 | |
2422 | while (len) { | |
b5d67f64 | 2423 | u64 l = min_t(u64, len, blocksize); |
a2de733c AJ |
2424 | int have_csum = 0; |
2425 | ||
2426 | if (flags & BTRFS_EXTENT_FLAG_DATA) { | |
2427 | /* push csums to sbio */ | |
d9d181c1 | 2428 | have_csum = scrub_find_csum(sctx, logical, l, csum); |
a2de733c | 2429 | if (have_csum == 0) |
d9d181c1 | 2430 | ++sctx->stat.no_csum; |
ff023aac SB |
2431 | if (sctx->is_dev_replace && !have_csum) { |
2432 | ret = copy_nocow_pages(sctx, logical, l, | |
2433 | mirror_num, | |
2434 | physical_for_dev_replace); | |
2435 | goto behind_scrub_pages; | |
2436 | } | |
a2de733c | 2437 | } |
a36cf8b8 | 2438 | ret = scrub_pages(sctx, logical, l, physical, dev, flags, gen, |
ff023aac SB |
2439 | mirror_num, have_csum ? csum : NULL, 0, |
2440 | physical_for_dev_replace); | |
2441 | behind_scrub_pages: | |
a2de733c AJ |
2442 | if (ret) |
2443 | return ret; | |
2444 | len -= l; | |
2445 | logical += l; | |
2446 | physical += l; | |
ff023aac | 2447 | physical_for_dev_replace += l; |
a2de733c AJ |
2448 | } |
2449 | return 0; | |
2450 | } | |
2451 | ||
5a6ac9ea MX |
2452 | static int scrub_pages_for_parity(struct scrub_parity *sparity, |
2453 | u64 logical, u64 len, | |
2454 | u64 physical, struct btrfs_device *dev, | |
2455 | u64 flags, u64 gen, int mirror_num, u8 *csum) | |
2456 | { | |
2457 | struct scrub_ctx *sctx = sparity->sctx; | |
2458 | struct scrub_block *sblock; | |
2459 | int index; | |
2460 | ||
2461 | sblock = kzalloc(sizeof(*sblock), GFP_NOFS); | |
2462 | if (!sblock) { | |
2463 | spin_lock(&sctx->stat_lock); | |
2464 | sctx->stat.malloc_errors++; | |
2465 | spin_unlock(&sctx->stat_lock); | |
2466 | return -ENOMEM; | |
2467 | } | |
2468 | ||
2469 | /* one ref inside this function, plus one for each page added to | |
2470 | * a bio later on */ | |
2471 | atomic_set(&sblock->ref_count, 1); | |
2472 | sblock->sctx = sctx; | |
2473 | sblock->no_io_error_seen = 1; | |
2474 | sblock->sparity = sparity; | |
2475 | scrub_parity_get(sparity); | |
2476 | ||
2477 | for (index = 0; len > 0; index++) { | |
2478 | struct scrub_page *spage; | |
2479 | u64 l = min_t(u64, len, PAGE_SIZE); | |
2480 | ||
2481 | spage = kzalloc(sizeof(*spage), GFP_NOFS); | |
2482 | if (!spage) { | |
2483 | leave_nomem: | |
2484 | spin_lock(&sctx->stat_lock); | |
2485 | sctx->stat.malloc_errors++; | |
2486 | spin_unlock(&sctx->stat_lock); | |
2487 | scrub_block_put(sblock); | |
2488 | return -ENOMEM; | |
2489 | } | |
2490 | BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK); | |
2491 | /* For scrub block */ | |
2492 | scrub_page_get(spage); | |
2493 | sblock->pagev[index] = spage; | |
2494 | /* For scrub parity */ | |
2495 | scrub_page_get(spage); | |
2496 | list_add_tail(&spage->list, &sparity->spages); | |
2497 | spage->sblock = sblock; | |
2498 | spage->dev = dev; | |
2499 | spage->flags = flags; | |
2500 | spage->generation = gen; | |
2501 | spage->logical = logical; | |
2502 | spage->physical = physical; | |
2503 | spage->mirror_num = mirror_num; | |
2504 | if (csum) { | |
2505 | spage->have_csum = 1; | |
2506 | memcpy(spage->csum, csum, sctx->csum_size); | |
2507 | } else { | |
2508 | spage->have_csum = 0; | |
2509 | } | |
2510 | sblock->page_count++; | |
2511 | spage->page = alloc_page(GFP_NOFS); | |
2512 | if (!spage->page) | |
2513 | goto leave_nomem; | |
2514 | len -= l; | |
2515 | logical += l; | |
2516 | physical += l; | |
2517 | } | |
2518 | ||
2519 | WARN_ON(sblock->page_count == 0); | |
2520 | for (index = 0; index < sblock->page_count; index++) { | |
2521 | struct scrub_page *spage = sblock->pagev[index]; | |
2522 | int ret; | |
2523 | ||
2524 | ret = scrub_add_page_to_rd_bio(sctx, spage); | |
2525 | if (ret) { | |
2526 | scrub_block_put(sblock); | |
2527 | return ret; | |
2528 | } | |
2529 | } | |
2530 | ||
2531 | /* last one frees, either here or in bio completion for last page */ | |
2532 | scrub_block_put(sblock); | |
2533 | return 0; | |
2534 | } | |
2535 | ||
2536 | static int scrub_extent_for_parity(struct scrub_parity *sparity, | |
2537 | u64 logical, u64 len, | |
2538 | u64 physical, struct btrfs_device *dev, | |
2539 | u64 flags, u64 gen, int mirror_num) | |
2540 | { | |
2541 | struct scrub_ctx *sctx = sparity->sctx; | |
2542 | int ret; | |
2543 | u8 csum[BTRFS_CSUM_SIZE]; | |
2544 | u32 blocksize; | |
2545 | ||
2546 | if (flags & BTRFS_EXTENT_FLAG_DATA) { | |
2547 | blocksize = sctx->sectorsize; | |
2548 | } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { | |
2549 | blocksize = sctx->nodesize; | |
2550 | } else { | |
2551 | blocksize = sctx->sectorsize; | |
2552 | WARN_ON(1); | |
2553 | } | |
2554 | ||
2555 | while (len) { | |
2556 | u64 l = min_t(u64, len, blocksize); | |
2557 | int have_csum = 0; | |
2558 | ||
2559 | if (flags & BTRFS_EXTENT_FLAG_DATA) { | |
2560 | /* push csums to sbio */ | |
2561 | have_csum = scrub_find_csum(sctx, logical, l, csum); | |
2562 | if (have_csum == 0) | |
2563 | goto skip; | |
2564 | } | |
2565 | ret = scrub_pages_for_parity(sparity, logical, l, physical, dev, | |
2566 | flags, gen, mirror_num, | |
2567 | have_csum ? csum : NULL); | |
5a6ac9ea MX |
2568 | if (ret) |
2569 | return ret; | |
6b6d24b3 | 2570 | skip: |
5a6ac9ea MX |
2571 | len -= l; |
2572 | logical += l; | |
2573 | physical += l; | |
2574 | } | |
2575 | return 0; | |
2576 | } | |
2577 | ||
3b080b25 WS |
2578 | /* |
2579 | * Given a physical address, this will calculate it's | |
2580 | * logical offset. if this is a parity stripe, it will return | |
2581 | * the most left data stripe's logical offset. | |
2582 | * | |
2583 | * return 0 if it is a data stripe, 1 means parity stripe. | |
2584 | */ | |
2585 | static int get_raid56_logic_offset(u64 physical, int num, | |
5a6ac9ea MX |
2586 | struct map_lookup *map, u64 *offset, |
2587 | u64 *stripe_start) | |
3b080b25 WS |
2588 | { |
2589 | int i; | |
2590 | int j = 0; | |
2591 | u64 stripe_nr; | |
2592 | u64 last_offset; | |
2593 | int stripe_index; | |
2594 | int rot; | |
2595 | ||
2596 | last_offset = (physical - map->stripes[num].physical) * | |
2597 | nr_data_stripes(map); | |
5a6ac9ea MX |
2598 | if (stripe_start) |
2599 | *stripe_start = last_offset; | |
2600 | ||
3b080b25 WS |
2601 | *offset = last_offset; |
2602 | for (i = 0; i < nr_data_stripes(map); i++) { | |
2603 | *offset = last_offset + i * map->stripe_len; | |
2604 | ||
2605 | stripe_nr = *offset; | |
2606 | do_div(stripe_nr, map->stripe_len); | |
2607 | do_div(stripe_nr, nr_data_stripes(map)); | |
2608 | ||
2609 | /* Work out the disk rotation on this stripe-set */ | |
2610 | rot = do_div(stripe_nr, map->num_stripes); | |
2611 | /* calculate which stripe this data locates */ | |
2612 | rot += i; | |
e4fbaee2 | 2613 | stripe_index = rot % map->num_stripes; |
3b080b25 WS |
2614 | if (stripe_index == num) |
2615 | return 0; | |
2616 | if (stripe_index < num) | |
2617 | j++; | |
2618 | } | |
2619 | *offset = last_offset + j * map->stripe_len; | |
2620 | return 1; | |
2621 | } | |
2622 | ||
5a6ac9ea MX |
2623 | static void scrub_free_parity(struct scrub_parity *sparity) |
2624 | { | |
2625 | struct scrub_ctx *sctx = sparity->sctx; | |
2626 | struct scrub_page *curr, *next; | |
2627 | int nbits; | |
2628 | ||
2629 | nbits = bitmap_weight(sparity->ebitmap, sparity->nsectors); | |
2630 | if (nbits) { | |
2631 | spin_lock(&sctx->stat_lock); | |
2632 | sctx->stat.read_errors += nbits; | |
2633 | sctx->stat.uncorrectable_errors += nbits; | |
2634 | spin_unlock(&sctx->stat_lock); | |
2635 | } | |
2636 | ||
2637 | list_for_each_entry_safe(curr, next, &sparity->spages, list) { | |
2638 | list_del_init(&curr->list); | |
2639 | scrub_page_put(curr); | |
2640 | } | |
2641 | ||
2642 | kfree(sparity); | |
2643 | } | |
2644 | ||
2645 | static void scrub_parity_bio_endio(struct bio *bio, int error) | |
2646 | { | |
2647 | struct scrub_parity *sparity = (struct scrub_parity *)bio->bi_private; | |
2648 | struct scrub_ctx *sctx = sparity->sctx; | |
2649 | ||
2650 | if (error) | |
2651 | bitmap_or(sparity->ebitmap, sparity->ebitmap, sparity->dbitmap, | |
2652 | sparity->nsectors); | |
2653 | ||
2654 | scrub_free_parity(sparity); | |
2655 | scrub_pending_bio_dec(sctx); | |
2656 | bio_put(bio); | |
2657 | } | |
2658 | ||
2659 | static void scrub_parity_check_and_repair(struct scrub_parity *sparity) | |
2660 | { | |
2661 | struct scrub_ctx *sctx = sparity->sctx; | |
2662 | struct bio *bio; | |
2663 | struct btrfs_raid_bio *rbio; | |
2664 | struct scrub_page *spage; | |
2665 | struct btrfs_bio *bbio = NULL; | |
5a6ac9ea MX |
2666 | u64 length; |
2667 | int ret; | |
2668 | ||
2669 | if (!bitmap_andnot(sparity->dbitmap, sparity->dbitmap, sparity->ebitmap, | |
2670 | sparity->nsectors)) | |
2671 | goto out; | |
2672 | ||
2673 | length = sparity->logic_end - sparity->logic_start + 1; | |
76035976 | 2674 | ret = btrfs_map_sblock(sctx->dev_root->fs_info, WRITE, |
5a6ac9ea | 2675 | sparity->logic_start, |
8e5cfb55 ZL |
2676 | &length, &bbio, 0, 1); |
2677 | if (ret || !bbio || !bbio->raid_map) | |
5a6ac9ea MX |
2678 | goto bbio_out; |
2679 | ||
2680 | bio = btrfs_io_bio_alloc(GFP_NOFS, 0); | |
2681 | if (!bio) | |
2682 | goto bbio_out; | |
2683 | ||
2684 | bio->bi_iter.bi_sector = sparity->logic_start >> 9; | |
2685 | bio->bi_private = sparity; | |
2686 | bio->bi_end_io = scrub_parity_bio_endio; | |
2687 | ||
2688 | rbio = raid56_parity_alloc_scrub_rbio(sctx->dev_root, bio, bbio, | |
8e5cfb55 | 2689 | length, sparity->scrub_dev, |
5a6ac9ea MX |
2690 | sparity->dbitmap, |
2691 | sparity->nsectors); | |
2692 | if (!rbio) | |
2693 | goto rbio_out; | |
2694 | ||
2695 | list_for_each_entry(spage, &sparity->spages, list) | |
2696 | raid56_parity_add_scrub_pages(rbio, spage->page, | |
2697 | spage->logical); | |
2698 | ||
2699 | scrub_pending_bio_inc(sctx); | |
2700 | raid56_parity_submit_scrub_rbio(rbio); | |
2701 | return; | |
2702 | ||
2703 | rbio_out: | |
2704 | bio_put(bio); | |
2705 | bbio_out: | |
6e9606d2 | 2706 | btrfs_put_bbio(bbio); |
5a6ac9ea MX |
2707 | bitmap_or(sparity->ebitmap, sparity->ebitmap, sparity->dbitmap, |
2708 | sparity->nsectors); | |
2709 | spin_lock(&sctx->stat_lock); | |
2710 | sctx->stat.malloc_errors++; | |
2711 | spin_unlock(&sctx->stat_lock); | |
2712 | out: | |
2713 | scrub_free_parity(sparity); | |
2714 | } | |
2715 | ||
2716 | static inline int scrub_calc_parity_bitmap_len(int nsectors) | |
2717 | { | |
2718 | return DIV_ROUND_UP(nsectors, BITS_PER_LONG) * (BITS_PER_LONG / 8); | |
2719 | } | |
2720 | ||
2721 | static void scrub_parity_get(struct scrub_parity *sparity) | |
2722 | { | |
2723 | atomic_inc(&sparity->ref_count); | |
2724 | } | |
2725 | ||
2726 | static void scrub_parity_put(struct scrub_parity *sparity) | |
2727 | { | |
2728 | if (!atomic_dec_and_test(&sparity->ref_count)) | |
2729 | return; | |
2730 | ||
2731 | scrub_parity_check_and_repair(sparity); | |
2732 | } | |
2733 | ||
2734 | static noinline_for_stack int scrub_raid56_parity(struct scrub_ctx *sctx, | |
2735 | struct map_lookup *map, | |
2736 | struct btrfs_device *sdev, | |
2737 | struct btrfs_path *path, | |
2738 | u64 logic_start, | |
2739 | u64 logic_end) | |
2740 | { | |
2741 | struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info; | |
2742 | struct btrfs_root *root = fs_info->extent_root; | |
2743 | struct btrfs_root *csum_root = fs_info->csum_root; | |
2744 | struct btrfs_extent_item *extent; | |
2745 | u64 flags; | |
2746 | int ret; | |
2747 | int slot; | |
2748 | struct extent_buffer *l; | |
2749 | struct btrfs_key key; | |
2750 | u64 generation; | |
2751 | u64 extent_logical; | |
2752 | u64 extent_physical; | |
2753 | u64 extent_len; | |
2754 | struct btrfs_device *extent_dev; | |
2755 | struct scrub_parity *sparity; | |
2756 | int nsectors; | |
2757 | int bitmap_len; | |
2758 | int extent_mirror_num; | |
2759 | int stop_loop = 0; | |
2760 | ||
2761 | nsectors = map->stripe_len / root->sectorsize; | |
2762 | bitmap_len = scrub_calc_parity_bitmap_len(nsectors); | |
2763 | sparity = kzalloc(sizeof(struct scrub_parity) + 2 * bitmap_len, | |
2764 | GFP_NOFS); | |
2765 | if (!sparity) { | |
2766 | spin_lock(&sctx->stat_lock); | |
2767 | sctx->stat.malloc_errors++; | |
2768 | spin_unlock(&sctx->stat_lock); | |
2769 | return -ENOMEM; | |
2770 | } | |
2771 | ||
2772 | sparity->stripe_len = map->stripe_len; | |
2773 | sparity->nsectors = nsectors; | |
2774 | sparity->sctx = sctx; | |
2775 | sparity->scrub_dev = sdev; | |
2776 | sparity->logic_start = logic_start; | |
2777 | sparity->logic_end = logic_end; | |
2778 | atomic_set(&sparity->ref_count, 1); | |
2779 | INIT_LIST_HEAD(&sparity->spages); | |
2780 | sparity->dbitmap = sparity->bitmap; | |
2781 | sparity->ebitmap = (void *)sparity->bitmap + bitmap_len; | |
2782 | ||
2783 | ret = 0; | |
2784 | while (logic_start < logic_end) { | |
2785 | if (btrfs_fs_incompat(fs_info, SKINNY_METADATA)) | |
2786 | key.type = BTRFS_METADATA_ITEM_KEY; | |
2787 | else | |
2788 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2789 | key.objectid = logic_start; | |
2790 | key.offset = (u64)-1; | |
2791 | ||
2792 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2793 | if (ret < 0) | |
2794 | goto out; | |
2795 | ||
2796 | if (ret > 0) { | |
2797 | ret = btrfs_previous_extent_item(root, path, 0); | |
2798 | if (ret < 0) | |
2799 | goto out; | |
2800 | if (ret > 0) { | |
2801 | btrfs_release_path(path); | |
2802 | ret = btrfs_search_slot(NULL, root, &key, | |
2803 | path, 0, 0); | |
2804 | if (ret < 0) | |
2805 | goto out; | |
2806 | } | |
2807 | } | |
2808 | ||
2809 | stop_loop = 0; | |
2810 | while (1) { | |
2811 | u64 bytes; | |
2812 | ||
2813 | l = path->nodes[0]; | |
2814 | slot = path->slots[0]; | |
2815 | if (slot >= btrfs_header_nritems(l)) { | |
2816 | ret = btrfs_next_leaf(root, path); | |
2817 | if (ret == 0) | |
2818 | continue; | |
2819 | if (ret < 0) | |
2820 | goto out; | |
2821 | ||
2822 | stop_loop = 1; | |
2823 | break; | |
2824 | } | |
2825 | btrfs_item_key_to_cpu(l, &key, slot); | |
2826 | ||
2827 | if (key.type == BTRFS_METADATA_ITEM_KEY) | |
2828 | bytes = root->nodesize; | |
2829 | else | |
2830 | bytes = key.offset; | |
2831 | ||
2832 | if (key.objectid + bytes <= logic_start) | |
2833 | goto next; | |
2834 | ||
2835 | if (key.type != BTRFS_EXTENT_ITEM_KEY && | |
2836 | key.type != BTRFS_METADATA_ITEM_KEY) | |
2837 | goto next; | |
2838 | ||
2839 | if (key.objectid > logic_end) { | |
2840 | stop_loop = 1; | |
2841 | break; | |
2842 | } | |
2843 | ||
2844 | while (key.objectid >= logic_start + map->stripe_len) | |
2845 | logic_start += map->stripe_len; | |
2846 | ||
2847 | extent = btrfs_item_ptr(l, slot, | |
2848 | struct btrfs_extent_item); | |
2849 | flags = btrfs_extent_flags(l, extent); | |
2850 | generation = btrfs_extent_generation(l, extent); | |
2851 | ||
2852 | if (key.objectid < logic_start && | |
2853 | (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) { | |
2854 | btrfs_err(fs_info, | |
2855 | "scrub: tree block %llu spanning stripes, ignored. logical=%llu", | |
2856 | key.objectid, logic_start); | |
2857 | goto next; | |
2858 | } | |
2859 | again: | |
2860 | extent_logical = key.objectid; | |
2861 | extent_len = bytes; | |
2862 | ||
2863 | if (extent_logical < logic_start) { | |
2864 | extent_len -= logic_start - extent_logical; | |
2865 | extent_logical = logic_start; | |
2866 | } | |
2867 | ||
2868 | if (extent_logical + extent_len > | |
2869 | logic_start + map->stripe_len) | |
2870 | extent_len = logic_start + map->stripe_len - | |
2871 | extent_logical; | |
2872 | ||
2873 | scrub_parity_mark_sectors_data(sparity, extent_logical, | |
2874 | extent_len); | |
2875 | ||
2876 | scrub_remap_extent(fs_info, extent_logical, | |
2877 | extent_len, &extent_physical, | |
2878 | &extent_dev, | |
2879 | &extent_mirror_num); | |
2880 | ||
2881 | ret = btrfs_lookup_csums_range(csum_root, | |
2882 | extent_logical, | |
2883 | extent_logical + extent_len - 1, | |
2884 | &sctx->csum_list, 1); | |
2885 | if (ret) | |
2886 | goto out; | |
2887 | ||
2888 | ret = scrub_extent_for_parity(sparity, extent_logical, | |
2889 | extent_len, | |
2890 | extent_physical, | |
2891 | extent_dev, flags, | |
2892 | generation, | |
2893 | extent_mirror_num); | |
2894 | if (ret) | |
2895 | goto out; | |
2896 | ||
2897 | scrub_free_csums(sctx); | |
2898 | if (extent_logical + extent_len < | |
2899 | key.objectid + bytes) { | |
2900 | logic_start += map->stripe_len; | |
2901 | ||
2902 | if (logic_start >= logic_end) { | |
2903 | stop_loop = 1; | |
2904 | break; | |
2905 | } | |
2906 | ||
2907 | if (logic_start < key.objectid + bytes) { | |
2908 | cond_resched(); | |
2909 | goto again; | |
2910 | } | |
2911 | } | |
2912 | next: | |
2913 | path->slots[0]++; | |
2914 | } | |
2915 | ||
2916 | btrfs_release_path(path); | |
2917 | ||
2918 | if (stop_loop) | |
2919 | break; | |
2920 | ||
2921 | logic_start += map->stripe_len; | |
2922 | } | |
2923 | out: | |
2924 | if (ret < 0) | |
2925 | scrub_parity_mark_sectors_error(sparity, logic_start, | |
2926 | logic_end - logic_start + 1); | |
2927 | scrub_parity_put(sparity); | |
2928 | scrub_submit(sctx); | |
2929 | mutex_lock(&sctx->wr_ctx.wr_lock); | |
2930 | scrub_wr_submit(sctx); | |
2931 | mutex_unlock(&sctx->wr_ctx.wr_lock); | |
2932 | ||
2933 | btrfs_release_path(path); | |
2934 | return ret < 0 ? ret : 0; | |
2935 | } | |
2936 | ||
d9d181c1 | 2937 | static noinline_for_stack int scrub_stripe(struct scrub_ctx *sctx, |
a36cf8b8 SB |
2938 | struct map_lookup *map, |
2939 | struct btrfs_device *scrub_dev, | |
ff023aac SB |
2940 | int num, u64 base, u64 length, |
2941 | int is_dev_replace) | |
a2de733c | 2942 | { |
5a6ac9ea | 2943 | struct btrfs_path *path, *ppath; |
a36cf8b8 | 2944 | struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info; |
a2de733c AJ |
2945 | struct btrfs_root *root = fs_info->extent_root; |
2946 | struct btrfs_root *csum_root = fs_info->csum_root; | |
2947 | struct btrfs_extent_item *extent; | |
e7786c3a | 2948 | struct blk_plug plug; |
a2de733c AJ |
2949 | u64 flags; |
2950 | int ret; | |
2951 | int slot; | |
a2de733c | 2952 | u64 nstripes; |
a2de733c AJ |
2953 | struct extent_buffer *l; |
2954 | struct btrfs_key key; | |
2955 | u64 physical; | |
2956 | u64 logical; | |
625f1c8d | 2957 | u64 logic_end; |
3b080b25 | 2958 | u64 physical_end; |
a2de733c | 2959 | u64 generation; |
e12fa9cd | 2960 | int mirror_num; |
7a26285e AJ |
2961 | struct reada_control *reada1; |
2962 | struct reada_control *reada2; | |
2963 | struct btrfs_key key_start; | |
2964 | struct btrfs_key key_end; | |
a2de733c AJ |
2965 | u64 increment = map->stripe_len; |
2966 | u64 offset; | |
ff023aac SB |
2967 | u64 extent_logical; |
2968 | u64 extent_physical; | |
2969 | u64 extent_len; | |
5a6ac9ea MX |
2970 | u64 stripe_logical; |
2971 | u64 stripe_end; | |
ff023aac SB |
2972 | struct btrfs_device *extent_dev; |
2973 | int extent_mirror_num; | |
3b080b25 | 2974 | int stop_loop = 0; |
53b381b3 | 2975 | |
a2de733c | 2976 | nstripes = length; |
3b080b25 | 2977 | physical = map->stripes[num].physical; |
a2de733c AJ |
2978 | offset = 0; |
2979 | do_div(nstripes, map->stripe_len); | |
2980 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { | |
2981 | offset = map->stripe_len * num; | |
2982 | increment = map->stripe_len * map->num_stripes; | |
193ea74b | 2983 | mirror_num = 1; |
a2de733c AJ |
2984 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
2985 | int factor = map->num_stripes / map->sub_stripes; | |
2986 | offset = map->stripe_len * (num / map->sub_stripes); | |
2987 | increment = map->stripe_len * factor; | |
193ea74b | 2988 | mirror_num = num % map->sub_stripes + 1; |
a2de733c AJ |
2989 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { |
2990 | increment = map->stripe_len; | |
193ea74b | 2991 | mirror_num = num % map->num_stripes + 1; |
a2de733c AJ |
2992 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
2993 | increment = map->stripe_len; | |
193ea74b | 2994 | mirror_num = num % map->num_stripes + 1; |
3b080b25 WS |
2995 | } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | |
2996 | BTRFS_BLOCK_GROUP_RAID6)) { | |
5a6ac9ea | 2997 | get_raid56_logic_offset(physical, num, map, &offset, NULL); |
3b080b25 WS |
2998 | increment = map->stripe_len * nr_data_stripes(map); |
2999 | mirror_num = 1; | |
a2de733c AJ |
3000 | } else { |
3001 | increment = map->stripe_len; | |
193ea74b | 3002 | mirror_num = 1; |
a2de733c AJ |
3003 | } |
3004 | ||
3005 | path = btrfs_alloc_path(); | |
3006 | if (!path) | |
3007 | return -ENOMEM; | |
3008 | ||
5a6ac9ea MX |
3009 | ppath = btrfs_alloc_path(); |
3010 | if (!ppath) { | |
3011 | btrfs_free_path(ppath); | |
3012 | return -ENOMEM; | |
3013 | } | |
3014 | ||
b5d67f64 SB |
3015 | /* |
3016 | * work on commit root. The related disk blocks are static as | |
3017 | * long as COW is applied. This means, it is save to rewrite | |
3018 | * them to repair disk errors without any race conditions | |
3019 | */ | |
a2de733c AJ |
3020 | path->search_commit_root = 1; |
3021 | path->skip_locking = 1; | |
3022 | ||
3023 | /* | |
7a26285e AJ |
3024 | * trigger the readahead for extent tree csum tree and wait for |
3025 | * completion. During readahead, the scrub is officially paused | |
3026 | * to not hold off transaction commits | |
a2de733c AJ |
3027 | */ |
3028 | logical = base + offset; | |
3b080b25 WS |
3029 | physical_end = physical + nstripes * map->stripe_len; |
3030 | if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | | |
3031 | BTRFS_BLOCK_GROUP_RAID6)) { | |
3032 | get_raid56_logic_offset(physical_end, num, | |
5a6ac9ea | 3033 | map, &logic_end, NULL); |
3b080b25 WS |
3034 | logic_end += base; |
3035 | } else { | |
3036 | logic_end = logical + increment * nstripes; | |
3037 | } | |
d9d181c1 | 3038 | wait_event(sctx->list_wait, |
b6bfebc1 | 3039 | atomic_read(&sctx->bios_in_flight) == 0); |
cb7ab021 | 3040 | scrub_blocked_if_needed(fs_info); |
7a26285e AJ |
3041 | |
3042 | /* FIXME it might be better to start readahead at commit root */ | |
3043 | key_start.objectid = logical; | |
3044 | key_start.type = BTRFS_EXTENT_ITEM_KEY; | |
3045 | key_start.offset = (u64)0; | |
3b080b25 | 3046 | key_end.objectid = logic_end; |
3173a18f JB |
3047 | key_end.type = BTRFS_METADATA_ITEM_KEY; |
3048 | key_end.offset = (u64)-1; | |
7a26285e AJ |
3049 | reada1 = btrfs_reada_add(root, &key_start, &key_end); |
3050 | ||
3051 | key_start.objectid = BTRFS_EXTENT_CSUM_OBJECTID; | |
3052 | key_start.type = BTRFS_EXTENT_CSUM_KEY; | |
3053 | key_start.offset = logical; | |
3054 | key_end.objectid = BTRFS_EXTENT_CSUM_OBJECTID; | |
3055 | key_end.type = BTRFS_EXTENT_CSUM_KEY; | |
3b080b25 | 3056 | key_end.offset = logic_end; |
7a26285e AJ |
3057 | reada2 = btrfs_reada_add(csum_root, &key_start, &key_end); |
3058 | ||
3059 | if (!IS_ERR(reada1)) | |
3060 | btrfs_reada_wait(reada1); | |
3061 | if (!IS_ERR(reada2)) | |
3062 | btrfs_reada_wait(reada2); | |
3063 | ||
a2de733c AJ |
3064 | |
3065 | /* | |
3066 | * collect all data csums for the stripe to avoid seeking during | |
3067 | * the scrub. This might currently (crc32) end up to be about 1MB | |
3068 | */ | |
e7786c3a | 3069 | blk_start_plug(&plug); |
a2de733c | 3070 | |
a2de733c AJ |
3071 | /* |
3072 | * now find all extents for each stripe and scrub them | |
3073 | */ | |
a2de733c | 3074 | ret = 0; |
3b080b25 WS |
3075 | while (physical < physical_end) { |
3076 | /* for raid56, we skip parity stripe */ | |
3077 | if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | | |
3078 | BTRFS_BLOCK_GROUP_RAID6)) { | |
3079 | ret = get_raid56_logic_offset(physical, num, | |
5a6ac9ea | 3080 | map, &logical, &stripe_logical); |
3b080b25 | 3081 | logical += base; |
5a6ac9ea MX |
3082 | if (ret) { |
3083 | stripe_logical += base; | |
3084 | stripe_end = stripe_logical + increment - 1; | |
3085 | ret = scrub_raid56_parity(sctx, map, scrub_dev, | |
3086 | ppath, stripe_logical, | |
3087 | stripe_end); | |
3088 | if (ret) | |
3089 | goto out; | |
3b080b25 | 3090 | goto skip; |
5a6ac9ea | 3091 | } |
3b080b25 | 3092 | } |
a2de733c AJ |
3093 | /* |
3094 | * canceled? | |
3095 | */ | |
3096 | if (atomic_read(&fs_info->scrub_cancel_req) || | |
d9d181c1 | 3097 | atomic_read(&sctx->cancel_req)) { |
a2de733c AJ |
3098 | ret = -ECANCELED; |
3099 | goto out; | |
3100 | } | |
3101 | /* | |
3102 | * check to see if we have to pause | |
3103 | */ | |
3104 | if (atomic_read(&fs_info->scrub_pause_req)) { | |
3105 | /* push queued extents */ | |
ff023aac | 3106 | atomic_set(&sctx->wr_ctx.flush_all_writes, 1); |
d9d181c1 | 3107 | scrub_submit(sctx); |
ff023aac SB |
3108 | mutex_lock(&sctx->wr_ctx.wr_lock); |
3109 | scrub_wr_submit(sctx); | |
3110 | mutex_unlock(&sctx->wr_ctx.wr_lock); | |
d9d181c1 | 3111 | wait_event(sctx->list_wait, |
b6bfebc1 | 3112 | atomic_read(&sctx->bios_in_flight) == 0); |
ff023aac | 3113 | atomic_set(&sctx->wr_ctx.flush_all_writes, 0); |
3cb0929a | 3114 | scrub_blocked_if_needed(fs_info); |
a2de733c AJ |
3115 | } |
3116 | ||
7c76edb7 WS |
3117 | if (btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
3118 | key.type = BTRFS_METADATA_ITEM_KEY; | |
3119 | else | |
3120 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
a2de733c | 3121 | key.objectid = logical; |
625f1c8d | 3122 | key.offset = (u64)-1; |
a2de733c AJ |
3123 | |
3124 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
3125 | if (ret < 0) | |
3126 | goto out; | |
3173a18f | 3127 | |
8c51032f | 3128 | if (ret > 0) { |
ade2e0b3 | 3129 | ret = btrfs_previous_extent_item(root, path, 0); |
a2de733c AJ |
3130 | if (ret < 0) |
3131 | goto out; | |
8c51032f AJ |
3132 | if (ret > 0) { |
3133 | /* there's no smaller item, so stick with the | |
3134 | * larger one */ | |
3135 | btrfs_release_path(path); | |
3136 | ret = btrfs_search_slot(NULL, root, &key, | |
3137 | path, 0, 0); | |
3138 | if (ret < 0) | |
3139 | goto out; | |
3140 | } | |
a2de733c AJ |
3141 | } |
3142 | ||
625f1c8d | 3143 | stop_loop = 0; |
a2de733c | 3144 | while (1) { |
3173a18f JB |
3145 | u64 bytes; |
3146 | ||
a2de733c AJ |
3147 | l = path->nodes[0]; |
3148 | slot = path->slots[0]; | |
3149 | if (slot >= btrfs_header_nritems(l)) { | |
3150 | ret = btrfs_next_leaf(root, path); | |
3151 | if (ret == 0) | |
3152 | continue; | |
3153 | if (ret < 0) | |
3154 | goto out; | |
3155 | ||
625f1c8d | 3156 | stop_loop = 1; |
a2de733c AJ |
3157 | break; |
3158 | } | |
3159 | btrfs_item_key_to_cpu(l, &key, slot); | |
3160 | ||
3173a18f | 3161 | if (key.type == BTRFS_METADATA_ITEM_KEY) |
707e8a07 | 3162 | bytes = root->nodesize; |
3173a18f JB |
3163 | else |
3164 | bytes = key.offset; | |
3165 | ||
3166 | if (key.objectid + bytes <= logical) | |
a2de733c AJ |
3167 | goto next; |
3168 | ||
625f1c8d LB |
3169 | if (key.type != BTRFS_EXTENT_ITEM_KEY && |
3170 | key.type != BTRFS_METADATA_ITEM_KEY) | |
3171 | goto next; | |
a2de733c | 3172 | |
625f1c8d LB |
3173 | if (key.objectid >= logical + map->stripe_len) { |
3174 | /* out of this device extent */ | |
3175 | if (key.objectid >= logic_end) | |
3176 | stop_loop = 1; | |
3177 | break; | |
3178 | } | |
a2de733c AJ |
3179 | |
3180 | extent = btrfs_item_ptr(l, slot, | |
3181 | struct btrfs_extent_item); | |
3182 | flags = btrfs_extent_flags(l, extent); | |
3183 | generation = btrfs_extent_generation(l, extent); | |
3184 | ||
3185 | if (key.objectid < logical && | |
3186 | (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) { | |
efe120a0 FH |
3187 | btrfs_err(fs_info, |
3188 | "scrub: tree block %llu spanning " | |
3189 | "stripes, ignored. logical=%llu", | |
c1c9ff7c | 3190 | key.objectid, logical); |
a2de733c AJ |
3191 | goto next; |
3192 | } | |
3193 | ||
625f1c8d LB |
3194 | again: |
3195 | extent_logical = key.objectid; | |
3196 | extent_len = bytes; | |
3197 | ||
a2de733c AJ |
3198 | /* |
3199 | * trim extent to this stripe | |
3200 | */ | |
625f1c8d LB |
3201 | if (extent_logical < logical) { |
3202 | extent_len -= logical - extent_logical; | |
3203 | extent_logical = logical; | |
a2de733c | 3204 | } |
625f1c8d | 3205 | if (extent_logical + extent_len > |
a2de733c | 3206 | logical + map->stripe_len) { |
625f1c8d LB |
3207 | extent_len = logical + map->stripe_len - |
3208 | extent_logical; | |
a2de733c AJ |
3209 | } |
3210 | ||
625f1c8d | 3211 | extent_physical = extent_logical - logical + physical; |
ff023aac SB |
3212 | extent_dev = scrub_dev; |
3213 | extent_mirror_num = mirror_num; | |
3214 | if (is_dev_replace) | |
3215 | scrub_remap_extent(fs_info, extent_logical, | |
3216 | extent_len, &extent_physical, | |
3217 | &extent_dev, | |
3218 | &extent_mirror_num); | |
625f1c8d LB |
3219 | |
3220 | ret = btrfs_lookup_csums_range(csum_root, logical, | |
3221 | logical + map->stripe_len - 1, | |
3222 | &sctx->csum_list, 1); | |
3223 | if (ret) | |
3224 | goto out; | |
3225 | ||
ff023aac SB |
3226 | ret = scrub_extent(sctx, extent_logical, extent_len, |
3227 | extent_physical, extent_dev, flags, | |
3228 | generation, extent_mirror_num, | |
115930cb | 3229 | extent_logical - logical + physical); |
a2de733c AJ |
3230 | if (ret) |
3231 | goto out; | |
3232 | ||
d88d46c6 | 3233 | scrub_free_csums(sctx); |
625f1c8d LB |
3234 | if (extent_logical + extent_len < |
3235 | key.objectid + bytes) { | |
3b080b25 WS |
3236 | if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | |
3237 | BTRFS_BLOCK_GROUP_RAID6)) { | |
3238 | /* | |
3239 | * loop until we find next data stripe | |
3240 | * or we have finished all stripes. | |
3241 | */ | |
5a6ac9ea MX |
3242 | loop: |
3243 | physical += map->stripe_len; | |
3244 | ret = get_raid56_logic_offset(physical, | |
3245 | num, map, &logical, | |
3246 | &stripe_logical); | |
3247 | logical += base; | |
3248 | ||
3249 | if (ret && physical < physical_end) { | |
3250 | stripe_logical += base; | |
3251 | stripe_end = stripe_logical + | |
3252 | increment - 1; | |
3253 | ret = scrub_raid56_parity(sctx, | |
3254 | map, scrub_dev, ppath, | |
3255 | stripe_logical, | |
3256 | stripe_end); | |
3257 | if (ret) | |
3258 | goto out; | |
3259 | goto loop; | |
3260 | } | |
3b080b25 WS |
3261 | } else { |
3262 | physical += map->stripe_len; | |
3263 | logical += increment; | |
3264 | } | |
625f1c8d LB |
3265 | if (logical < key.objectid + bytes) { |
3266 | cond_resched(); | |
3267 | goto again; | |
3268 | } | |
3269 | ||
3b080b25 | 3270 | if (physical >= physical_end) { |
625f1c8d LB |
3271 | stop_loop = 1; |
3272 | break; | |
3273 | } | |
3274 | } | |
a2de733c AJ |
3275 | next: |
3276 | path->slots[0]++; | |
3277 | } | |
71267333 | 3278 | btrfs_release_path(path); |
3b080b25 | 3279 | skip: |
a2de733c AJ |
3280 | logical += increment; |
3281 | physical += map->stripe_len; | |
d9d181c1 | 3282 | spin_lock(&sctx->stat_lock); |
625f1c8d LB |
3283 | if (stop_loop) |
3284 | sctx->stat.last_physical = map->stripes[num].physical + | |
3285 | length; | |
3286 | else | |
3287 | sctx->stat.last_physical = physical; | |
d9d181c1 | 3288 | spin_unlock(&sctx->stat_lock); |
625f1c8d LB |
3289 | if (stop_loop) |
3290 | break; | |
a2de733c | 3291 | } |
ff023aac | 3292 | out: |
a2de733c | 3293 | /* push queued extents */ |
d9d181c1 | 3294 | scrub_submit(sctx); |
ff023aac SB |
3295 | mutex_lock(&sctx->wr_ctx.wr_lock); |
3296 | scrub_wr_submit(sctx); | |
3297 | mutex_unlock(&sctx->wr_ctx.wr_lock); | |
a2de733c | 3298 | |
e7786c3a | 3299 | blk_finish_plug(&plug); |
a2de733c | 3300 | btrfs_free_path(path); |
5a6ac9ea | 3301 | btrfs_free_path(ppath); |
a2de733c AJ |
3302 | return ret < 0 ? ret : 0; |
3303 | } | |
3304 | ||
d9d181c1 | 3305 | static noinline_for_stack int scrub_chunk(struct scrub_ctx *sctx, |
a36cf8b8 SB |
3306 | struct btrfs_device *scrub_dev, |
3307 | u64 chunk_tree, u64 chunk_objectid, | |
3308 | u64 chunk_offset, u64 length, | |
ff023aac | 3309 | u64 dev_offset, int is_dev_replace) |
a2de733c AJ |
3310 | { |
3311 | struct btrfs_mapping_tree *map_tree = | |
a36cf8b8 | 3312 | &sctx->dev_root->fs_info->mapping_tree; |
a2de733c AJ |
3313 | struct map_lookup *map; |
3314 | struct extent_map *em; | |
3315 | int i; | |
ff023aac | 3316 | int ret = 0; |
a2de733c AJ |
3317 | |
3318 | read_lock(&map_tree->map_tree.lock); | |
3319 | em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1); | |
3320 | read_unlock(&map_tree->map_tree.lock); | |
3321 | ||
3322 | if (!em) | |
3323 | return -EINVAL; | |
3324 | ||
3325 | map = (struct map_lookup *)em->bdev; | |
3326 | if (em->start != chunk_offset) | |
3327 | goto out; | |
3328 | ||
3329 | if (em->len < length) | |
3330 | goto out; | |
3331 | ||
3332 | for (i = 0; i < map->num_stripes; ++i) { | |
a36cf8b8 | 3333 | if (map->stripes[i].dev->bdev == scrub_dev->bdev && |
859acaf1 | 3334 | map->stripes[i].physical == dev_offset) { |
a36cf8b8 | 3335 | ret = scrub_stripe(sctx, map, scrub_dev, i, |
ff023aac SB |
3336 | chunk_offset, length, |
3337 | is_dev_replace); | |
a2de733c AJ |
3338 | if (ret) |
3339 | goto out; | |
3340 | } | |
3341 | } | |
3342 | out: | |
3343 | free_extent_map(em); | |
3344 | ||
3345 | return ret; | |
3346 | } | |
3347 | ||
3348 | static noinline_for_stack | |
a36cf8b8 | 3349 | int scrub_enumerate_chunks(struct scrub_ctx *sctx, |
ff023aac SB |
3350 | struct btrfs_device *scrub_dev, u64 start, u64 end, |
3351 | int is_dev_replace) | |
a2de733c AJ |
3352 | { |
3353 | struct btrfs_dev_extent *dev_extent = NULL; | |
3354 | struct btrfs_path *path; | |
a36cf8b8 | 3355 | struct btrfs_root *root = sctx->dev_root; |
a2de733c AJ |
3356 | struct btrfs_fs_info *fs_info = root->fs_info; |
3357 | u64 length; | |
3358 | u64 chunk_tree; | |
3359 | u64 chunk_objectid; | |
3360 | u64 chunk_offset; | |
3361 | int ret; | |
3362 | int slot; | |
3363 | struct extent_buffer *l; | |
3364 | struct btrfs_key key; | |
3365 | struct btrfs_key found_key; | |
3366 | struct btrfs_block_group_cache *cache; | |
ff023aac | 3367 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
a2de733c AJ |
3368 | |
3369 | path = btrfs_alloc_path(); | |
3370 | if (!path) | |
3371 | return -ENOMEM; | |
3372 | ||
3373 | path->reada = 2; | |
3374 | path->search_commit_root = 1; | |
3375 | path->skip_locking = 1; | |
3376 | ||
a36cf8b8 | 3377 | key.objectid = scrub_dev->devid; |
a2de733c AJ |
3378 | key.offset = 0ull; |
3379 | key.type = BTRFS_DEV_EXTENT_KEY; | |
3380 | ||
a2de733c AJ |
3381 | while (1) { |
3382 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
3383 | if (ret < 0) | |
8c51032f AJ |
3384 | break; |
3385 | if (ret > 0) { | |
3386 | if (path->slots[0] >= | |
3387 | btrfs_header_nritems(path->nodes[0])) { | |
3388 | ret = btrfs_next_leaf(root, path); | |
3389 | if (ret) | |
3390 | break; | |
3391 | } | |
3392 | } | |
a2de733c AJ |
3393 | |
3394 | l = path->nodes[0]; | |
3395 | slot = path->slots[0]; | |
3396 | ||
3397 | btrfs_item_key_to_cpu(l, &found_key, slot); | |
3398 | ||
a36cf8b8 | 3399 | if (found_key.objectid != scrub_dev->devid) |
a2de733c AJ |
3400 | break; |
3401 | ||
962a298f | 3402 | if (found_key.type != BTRFS_DEV_EXTENT_KEY) |
a2de733c AJ |
3403 | break; |
3404 | ||
3405 | if (found_key.offset >= end) | |
3406 | break; | |
3407 | ||
3408 | if (found_key.offset < key.offset) | |
3409 | break; | |
3410 | ||
3411 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
3412 | length = btrfs_dev_extent_length(l, dev_extent); | |
3413 | ||
ced96edc QW |
3414 | if (found_key.offset + length <= start) |
3415 | goto skip; | |
a2de733c AJ |
3416 | |
3417 | chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent); | |
3418 | chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent); | |
3419 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); | |
3420 | ||
3421 | /* | |
3422 | * get a reference on the corresponding block group to prevent | |
3423 | * the chunk from going away while we scrub it | |
3424 | */ | |
3425 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
ced96edc QW |
3426 | |
3427 | /* some chunks are removed but not committed to disk yet, | |
3428 | * continue scrubbing */ | |
3429 | if (!cache) | |
3430 | goto skip; | |
3431 | ||
ff023aac SB |
3432 | dev_replace->cursor_right = found_key.offset + length; |
3433 | dev_replace->cursor_left = found_key.offset; | |
3434 | dev_replace->item_needs_writeback = 1; | |
a36cf8b8 | 3435 | ret = scrub_chunk(sctx, scrub_dev, chunk_tree, chunk_objectid, |
ff023aac SB |
3436 | chunk_offset, length, found_key.offset, |
3437 | is_dev_replace); | |
3438 | ||
3439 | /* | |
3440 | * flush, submit all pending read and write bios, afterwards | |
3441 | * wait for them. | |
3442 | * Note that in the dev replace case, a read request causes | |
3443 | * write requests that are submitted in the read completion | |
3444 | * worker. Therefore in the current situation, it is required | |
3445 | * that all write requests are flushed, so that all read and | |
3446 | * write requests are really completed when bios_in_flight | |
3447 | * changes to 0. | |
3448 | */ | |
3449 | atomic_set(&sctx->wr_ctx.flush_all_writes, 1); | |
3450 | scrub_submit(sctx); | |
3451 | mutex_lock(&sctx->wr_ctx.wr_lock); | |
3452 | scrub_wr_submit(sctx); | |
3453 | mutex_unlock(&sctx->wr_ctx.wr_lock); | |
3454 | ||
3455 | wait_event(sctx->list_wait, | |
3456 | atomic_read(&sctx->bios_in_flight) == 0); | |
12cf9372 WS |
3457 | atomic_inc(&fs_info->scrubs_paused); |
3458 | wake_up(&fs_info->scrub_pause_wait); | |
3459 | ||
3460 | /* | |
3461 | * must be called before we decrease @scrub_paused. | |
3462 | * make sure we don't block transaction commit while | |
3463 | * we are waiting pending workers finished. | |
3464 | */ | |
ff023aac SB |
3465 | wait_event(sctx->list_wait, |
3466 | atomic_read(&sctx->workers_pending) == 0); | |
12cf9372 WS |
3467 | atomic_set(&sctx->wr_ctx.flush_all_writes, 0); |
3468 | ||
3469 | mutex_lock(&fs_info->scrub_lock); | |
3470 | __scrub_blocked_if_needed(fs_info); | |
3471 | atomic_dec(&fs_info->scrubs_paused); | |
3472 | mutex_unlock(&fs_info->scrub_lock); | |
3473 | wake_up(&fs_info->scrub_pause_wait); | |
ff023aac | 3474 | |
a2de733c AJ |
3475 | btrfs_put_block_group(cache); |
3476 | if (ret) | |
3477 | break; | |
af1be4f8 SB |
3478 | if (is_dev_replace && |
3479 | atomic64_read(&dev_replace->num_write_errors) > 0) { | |
ff023aac SB |
3480 | ret = -EIO; |
3481 | break; | |
3482 | } | |
3483 | if (sctx->stat.malloc_errors > 0) { | |
3484 | ret = -ENOMEM; | |
3485 | break; | |
3486 | } | |
a2de733c | 3487 | |
539f358a ID |
3488 | dev_replace->cursor_left = dev_replace->cursor_right; |
3489 | dev_replace->item_needs_writeback = 1; | |
ced96edc | 3490 | skip: |
a2de733c | 3491 | key.offset = found_key.offset + length; |
71267333 | 3492 | btrfs_release_path(path); |
a2de733c AJ |
3493 | } |
3494 | ||
a2de733c | 3495 | btrfs_free_path(path); |
8c51032f AJ |
3496 | |
3497 | /* | |
3498 | * ret can still be 1 from search_slot or next_leaf, | |
3499 | * that's not an error | |
3500 | */ | |
3501 | return ret < 0 ? ret : 0; | |
a2de733c AJ |
3502 | } |
3503 | ||
a36cf8b8 SB |
3504 | static noinline_for_stack int scrub_supers(struct scrub_ctx *sctx, |
3505 | struct btrfs_device *scrub_dev) | |
a2de733c AJ |
3506 | { |
3507 | int i; | |
3508 | u64 bytenr; | |
3509 | u64 gen; | |
3510 | int ret; | |
a36cf8b8 | 3511 | struct btrfs_root *root = sctx->dev_root; |
a2de733c | 3512 | |
87533c47 | 3513 | if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
79787eaa JM |
3514 | return -EIO; |
3515 | ||
5f546063 MX |
3516 | /* Seed devices of a new filesystem has their own generation. */ |
3517 | if (scrub_dev->fs_devices != root->fs_info->fs_devices) | |
3518 | gen = scrub_dev->generation; | |
3519 | else | |
3520 | gen = root->fs_info->last_trans_committed; | |
a2de733c AJ |
3521 | |
3522 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { | |
3523 | bytenr = btrfs_sb_offset(i); | |
935e5cc9 MX |
3524 | if (bytenr + BTRFS_SUPER_INFO_SIZE > |
3525 | scrub_dev->commit_total_bytes) | |
a2de733c AJ |
3526 | break; |
3527 | ||
d9d181c1 | 3528 | ret = scrub_pages(sctx, bytenr, BTRFS_SUPER_INFO_SIZE, bytenr, |
a36cf8b8 | 3529 | scrub_dev, BTRFS_EXTENT_FLAG_SUPER, gen, i, |
ff023aac | 3530 | NULL, 1, bytenr); |
a2de733c AJ |
3531 | if (ret) |
3532 | return ret; | |
3533 | } | |
b6bfebc1 | 3534 | wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0); |
a2de733c AJ |
3535 | |
3536 | return 0; | |
3537 | } | |
3538 | ||
3539 | /* | |
3540 | * get a reference count on fs_info->scrub_workers. start worker if necessary | |
3541 | */ | |
ff023aac SB |
3542 | static noinline_for_stack int scrub_workers_get(struct btrfs_fs_info *fs_info, |
3543 | int is_dev_replace) | |
a2de733c | 3544 | { |
0dc3b84a | 3545 | int ret = 0; |
0339ef2f QW |
3546 | int flags = WQ_FREEZABLE | WQ_UNBOUND; |
3547 | int max_active = fs_info->thread_pool_size; | |
a2de733c | 3548 | |
632dd772 | 3549 | if (fs_info->scrub_workers_refcnt == 0) { |
ff023aac | 3550 | if (is_dev_replace) |
0339ef2f QW |
3551 | fs_info->scrub_workers = |
3552 | btrfs_alloc_workqueue("btrfs-scrub", flags, | |
3553 | 1, 4); | |
ff023aac | 3554 | else |
0339ef2f QW |
3555 | fs_info->scrub_workers = |
3556 | btrfs_alloc_workqueue("btrfs-scrub", flags, | |
3557 | max_active, 4); | |
3558 | if (!fs_info->scrub_workers) { | |
3559 | ret = -ENOMEM; | |
0dc3b84a | 3560 | goto out; |
0339ef2f QW |
3561 | } |
3562 | fs_info->scrub_wr_completion_workers = | |
3563 | btrfs_alloc_workqueue("btrfs-scrubwrc", flags, | |
3564 | max_active, 2); | |
3565 | if (!fs_info->scrub_wr_completion_workers) { | |
3566 | ret = -ENOMEM; | |
ff023aac | 3567 | goto out; |
0339ef2f QW |
3568 | } |
3569 | fs_info->scrub_nocow_workers = | |
3570 | btrfs_alloc_workqueue("btrfs-scrubnc", flags, 1, 0); | |
3571 | if (!fs_info->scrub_nocow_workers) { | |
3572 | ret = -ENOMEM; | |
ff023aac | 3573 | goto out; |
0339ef2f | 3574 | } |
632dd772 | 3575 | } |
a2de733c | 3576 | ++fs_info->scrub_workers_refcnt; |
0dc3b84a | 3577 | out: |
0dc3b84a | 3578 | return ret; |
a2de733c AJ |
3579 | } |
3580 | ||
aa1b8cd4 | 3581 | static noinline_for_stack void scrub_workers_put(struct btrfs_fs_info *fs_info) |
a2de733c | 3582 | { |
ff023aac | 3583 | if (--fs_info->scrub_workers_refcnt == 0) { |
0339ef2f QW |
3584 | btrfs_destroy_workqueue(fs_info->scrub_workers); |
3585 | btrfs_destroy_workqueue(fs_info->scrub_wr_completion_workers); | |
3586 | btrfs_destroy_workqueue(fs_info->scrub_nocow_workers); | |
ff023aac | 3587 | } |
a2de733c | 3588 | WARN_ON(fs_info->scrub_workers_refcnt < 0); |
a2de733c AJ |
3589 | } |
3590 | ||
aa1b8cd4 SB |
3591 | int btrfs_scrub_dev(struct btrfs_fs_info *fs_info, u64 devid, u64 start, |
3592 | u64 end, struct btrfs_scrub_progress *progress, | |
63a212ab | 3593 | int readonly, int is_dev_replace) |
a2de733c | 3594 | { |
d9d181c1 | 3595 | struct scrub_ctx *sctx; |
a2de733c AJ |
3596 | int ret; |
3597 | struct btrfs_device *dev; | |
5d68da3b | 3598 | struct rcu_string *name; |
a2de733c | 3599 | |
aa1b8cd4 | 3600 | if (btrfs_fs_closing(fs_info)) |
a2de733c AJ |
3601 | return -EINVAL; |
3602 | ||
aa1b8cd4 | 3603 | if (fs_info->chunk_root->nodesize > BTRFS_STRIPE_LEN) { |
b5d67f64 SB |
3604 | /* |
3605 | * in this case scrub is unable to calculate the checksum | |
3606 | * the way scrub is implemented. Do not handle this | |
3607 | * situation at all because it won't ever happen. | |
3608 | */ | |
efe120a0 FH |
3609 | btrfs_err(fs_info, |
3610 | "scrub: size assumption nodesize <= BTRFS_STRIPE_LEN (%d <= %d) fails", | |
aa1b8cd4 | 3611 | fs_info->chunk_root->nodesize, BTRFS_STRIPE_LEN); |
b5d67f64 SB |
3612 | return -EINVAL; |
3613 | } | |
3614 | ||
aa1b8cd4 | 3615 | if (fs_info->chunk_root->sectorsize != PAGE_SIZE) { |
b5d67f64 | 3616 | /* not supported for data w/o checksums */ |
efe120a0 FH |
3617 | btrfs_err(fs_info, |
3618 | "scrub: size assumption sectorsize != PAGE_SIZE " | |
3619 | "(%d != %lu) fails", | |
27f9f023 | 3620 | fs_info->chunk_root->sectorsize, PAGE_SIZE); |
a2de733c AJ |
3621 | return -EINVAL; |
3622 | } | |
3623 | ||
7a9e9987 SB |
3624 | if (fs_info->chunk_root->nodesize > |
3625 | PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK || | |
3626 | fs_info->chunk_root->sectorsize > | |
3627 | PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK) { | |
3628 | /* | |
3629 | * would exhaust the array bounds of pagev member in | |
3630 | * struct scrub_block | |
3631 | */ | |
efe120a0 FH |
3632 | btrfs_err(fs_info, "scrub: size assumption nodesize and sectorsize " |
3633 | "<= SCRUB_MAX_PAGES_PER_BLOCK (%d <= %d && %d <= %d) fails", | |
7a9e9987 SB |
3634 | fs_info->chunk_root->nodesize, |
3635 | SCRUB_MAX_PAGES_PER_BLOCK, | |
3636 | fs_info->chunk_root->sectorsize, | |
3637 | SCRUB_MAX_PAGES_PER_BLOCK); | |
3638 | return -EINVAL; | |
3639 | } | |
3640 | ||
a2de733c | 3641 | |
aa1b8cd4 SB |
3642 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
3643 | dev = btrfs_find_device(fs_info, devid, NULL, NULL); | |
63a212ab | 3644 | if (!dev || (dev->missing && !is_dev_replace)) { |
aa1b8cd4 | 3645 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
a2de733c AJ |
3646 | return -ENODEV; |
3647 | } | |
a2de733c | 3648 | |
5d68da3b MX |
3649 | if (!is_dev_replace && !readonly && !dev->writeable) { |
3650 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
3651 | rcu_read_lock(); | |
3652 | name = rcu_dereference(dev->name); | |
3653 | btrfs_err(fs_info, "scrub: device %s is not writable", | |
3654 | name->str); | |
3655 | rcu_read_unlock(); | |
3656 | return -EROFS; | |
3657 | } | |
3658 | ||
3b7a016f | 3659 | mutex_lock(&fs_info->scrub_lock); |
63a212ab | 3660 | if (!dev->in_fs_metadata || dev->is_tgtdev_for_dev_replace) { |
a2de733c | 3661 | mutex_unlock(&fs_info->scrub_lock); |
aa1b8cd4 | 3662 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
aa1b8cd4 | 3663 | return -EIO; |
a2de733c AJ |
3664 | } |
3665 | ||
8dabb742 SB |
3666 | btrfs_dev_replace_lock(&fs_info->dev_replace); |
3667 | if (dev->scrub_device || | |
3668 | (!is_dev_replace && | |
3669 | btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))) { | |
3670 | btrfs_dev_replace_unlock(&fs_info->dev_replace); | |
a2de733c | 3671 | mutex_unlock(&fs_info->scrub_lock); |
aa1b8cd4 | 3672 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
a2de733c AJ |
3673 | return -EINPROGRESS; |
3674 | } | |
8dabb742 | 3675 | btrfs_dev_replace_unlock(&fs_info->dev_replace); |
3b7a016f WS |
3676 | |
3677 | ret = scrub_workers_get(fs_info, is_dev_replace); | |
3678 | if (ret) { | |
3679 | mutex_unlock(&fs_info->scrub_lock); | |
3680 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
3681 | return ret; | |
3682 | } | |
3683 | ||
63a212ab | 3684 | sctx = scrub_setup_ctx(dev, is_dev_replace); |
d9d181c1 | 3685 | if (IS_ERR(sctx)) { |
a2de733c | 3686 | mutex_unlock(&fs_info->scrub_lock); |
aa1b8cd4 SB |
3687 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
3688 | scrub_workers_put(fs_info); | |
d9d181c1 | 3689 | return PTR_ERR(sctx); |
a2de733c | 3690 | } |
d9d181c1 SB |
3691 | sctx->readonly = readonly; |
3692 | dev->scrub_device = sctx; | |
3cb0929a | 3693 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
a2de733c | 3694 | |
3cb0929a WS |
3695 | /* |
3696 | * checking @scrub_pause_req here, we can avoid | |
3697 | * race between committing transaction and scrubbing. | |
3698 | */ | |
cb7ab021 | 3699 | __scrub_blocked_if_needed(fs_info); |
a2de733c AJ |
3700 | atomic_inc(&fs_info->scrubs_running); |
3701 | mutex_unlock(&fs_info->scrub_lock); | |
a2de733c | 3702 | |
ff023aac | 3703 | if (!is_dev_replace) { |
9b011adf WS |
3704 | /* |
3705 | * by holding device list mutex, we can | |
3706 | * kick off writing super in log tree sync. | |
3707 | */ | |
3cb0929a | 3708 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
ff023aac | 3709 | ret = scrub_supers(sctx, dev); |
3cb0929a | 3710 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
ff023aac | 3711 | } |
a2de733c AJ |
3712 | |
3713 | if (!ret) | |
ff023aac SB |
3714 | ret = scrub_enumerate_chunks(sctx, dev, start, end, |
3715 | is_dev_replace); | |
a2de733c | 3716 | |
b6bfebc1 | 3717 | wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0); |
a2de733c AJ |
3718 | atomic_dec(&fs_info->scrubs_running); |
3719 | wake_up(&fs_info->scrub_pause_wait); | |
3720 | ||
b6bfebc1 | 3721 | wait_event(sctx->list_wait, atomic_read(&sctx->workers_pending) == 0); |
0ef8e451 | 3722 | |
a2de733c | 3723 | if (progress) |
d9d181c1 | 3724 | memcpy(progress, &sctx->stat, sizeof(*progress)); |
a2de733c AJ |
3725 | |
3726 | mutex_lock(&fs_info->scrub_lock); | |
3727 | dev->scrub_device = NULL; | |
3b7a016f | 3728 | scrub_workers_put(fs_info); |
a2de733c AJ |
3729 | mutex_unlock(&fs_info->scrub_lock); |
3730 | ||
d9d181c1 | 3731 | scrub_free_ctx(sctx); |
a2de733c AJ |
3732 | |
3733 | return ret; | |
3734 | } | |
3735 | ||
143bede5 | 3736 | void btrfs_scrub_pause(struct btrfs_root *root) |
a2de733c AJ |
3737 | { |
3738 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3739 | ||
3740 | mutex_lock(&fs_info->scrub_lock); | |
3741 | atomic_inc(&fs_info->scrub_pause_req); | |
3742 | while (atomic_read(&fs_info->scrubs_paused) != | |
3743 | atomic_read(&fs_info->scrubs_running)) { | |
3744 | mutex_unlock(&fs_info->scrub_lock); | |
3745 | wait_event(fs_info->scrub_pause_wait, | |
3746 | atomic_read(&fs_info->scrubs_paused) == | |
3747 | atomic_read(&fs_info->scrubs_running)); | |
3748 | mutex_lock(&fs_info->scrub_lock); | |
3749 | } | |
3750 | mutex_unlock(&fs_info->scrub_lock); | |
a2de733c AJ |
3751 | } |
3752 | ||
143bede5 | 3753 | void btrfs_scrub_continue(struct btrfs_root *root) |
a2de733c AJ |
3754 | { |
3755 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3756 | ||
3757 | atomic_dec(&fs_info->scrub_pause_req); | |
3758 | wake_up(&fs_info->scrub_pause_wait); | |
a2de733c AJ |
3759 | } |
3760 | ||
aa1b8cd4 | 3761 | int btrfs_scrub_cancel(struct btrfs_fs_info *fs_info) |
a2de733c | 3762 | { |
a2de733c AJ |
3763 | mutex_lock(&fs_info->scrub_lock); |
3764 | if (!atomic_read(&fs_info->scrubs_running)) { | |
3765 | mutex_unlock(&fs_info->scrub_lock); | |
3766 | return -ENOTCONN; | |
3767 | } | |
3768 | ||
3769 | atomic_inc(&fs_info->scrub_cancel_req); | |
3770 | while (atomic_read(&fs_info->scrubs_running)) { | |
3771 | mutex_unlock(&fs_info->scrub_lock); | |
3772 | wait_event(fs_info->scrub_pause_wait, | |
3773 | atomic_read(&fs_info->scrubs_running) == 0); | |
3774 | mutex_lock(&fs_info->scrub_lock); | |
3775 | } | |
3776 | atomic_dec(&fs_info->scrub_cancel_req); | |
3777 | mutex_unlock(&fs_info->scrub_lock); | |
3778 | ||
3779 | return 0; | |
3780 | } | |
3781 | ||
aa1b8cd4 SB |
3782 | int btrfs_scrub_cancel_dev(struct btrfs_fs_info *fs_info, |
3783 | struct btrfs_device *dev) | |
49b25e05 | 3784 | { |
d9d181c1 | 3785 | struct scrub_ctx *sctx; |
a2de733c AJ |
3786 | |
3787 | mutex_lock(&fs_info->scrub_lock); | |
d9d181c1 SB |
3788 | sctx = dev->scrub_device; |
3789 | if (!sctx) { | |
a2de733c AJ |
3790 | mutex_unlock(&fs_info->scrub_lock); |
3791 | return -ENOTCONN; | |
3792 | } | |
d9d181c1 | 3793 | atomic_inc(&sctx->cancel_req); |
a2de733c AJ |
3794 | while (dev->scrub_device) { |
3795 | mutex_unlock(&fs_info->scrub_lock); | |
3796 | wait_event(fs_info->scrub_pause_wait, | |
3797 | dev->scrub_device == NULL); | |
3798 | mutex_lock(&fs_info->scrub_lock); | |
3799 | } | |
3800 | mutex_unlock(&fs_info->scrub_lock); | |
3801 | ||
3802 | return 0; | |
3803 | } | |
1623edeb | 3804 | |
a2de733c AJ |
3805 | int btrfs_scrub_progress(struct btrfs_root *root, u64 devid, |
3806 | struct btrfs_scrub_progress *progress) | |
3807 | { | |
3808 | struct btrfs_device *dev; | |
d9d181c1 | 3809 | struct scrub_ctx *sctx = NULL; |
a2de733c AJ |
3810 | |
3811 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
aa1b8cd4 | 3812 | dev = btrfs_find_device(root->fs_info, devid, NULL, NULL); |
a2de733c | 3813 | if (dev) |
d9d181c1 SB |
3814 | sctx = dev->scrub_device; |
3815 | if (sctx) | |
3816 | memcpy(progress, &sctx->stat, sizeof(*progress)); | |
a2de733c AJ |
3817 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
3818 | ||
d9d181c1 | 3819 | return dev ? (sctx ? 0 : -ENOTCONN) : -ENODEV; |
a2de733c | 3820 | } |
ff023aac SB |
3821 | |
3822 | static void scrub_remap_extent(struct btrfs_fs_info *fs_info, | |
3823 | u64 extent_logical, u64 extent_len, | |
3824 | u64 *extent_physical, | |
3825 | struct btrfs_device **extent_dev, | |
3826 | int *extent_mirror_num) | |
3827 | { | |
3828 | u64 mapped_length; | |
3829 | struct btrfs_bio *bbio = NULL; | |
3830 | int ret; | |
3831 | ||
3832 | mapped_length = extent_len; | |
3833 | ret = btrfs_map_block(fs_info, READ, extent_logical, | |
3834 | &mapped_length, &bbio, 0); | |
3835 | if (ret || !bbio || mapped_length < extent_len || | |
3836 | !bbio->stripes[0].dev->bdev) { | |
6e9606d2 | 3837 | btrfs_put_bbio(bbio); |
ff023aac SB |
3838 | return; |
3839 | } | |
3840 | ||
3841 | *extent_physical = bbio->stripes[0].physical; | |
3842 | *extent_mirror_num = bbio->mirror_num; | |
3843 | *extent_dev = bbio->stripes[0].dev; | |
6e9606d2 | 3844 | btrfs_put_bbio(bbio); |
ff023aac SB |
3845 | } |
3846 | ||
3847 | static int scrub_setup_wr_ctx(struct scrub_ctx *sctx, | |
3848 | struct scrub_wr_ctx *wr_ctx, | |
3849 | struct btrfs_fs_info *fs_info, | |
3850 | struct btrfs_device *dev, | |
3851 | int is_dev_replace) | |
3852 | { | |
3853 | WARN_ON(wr_ctx->wr_curr_bio != NULL); | |
3854 | ||
3855 | mutex_init(&wr_ctx->wr_lock); | |
3856 | wr_ctx->wr_curr_bio = NULL; | |
3857 | if (!is_dev_replace) | |
3858 | return 0; | |
3859 | ||
3860 | WARN_ON(!dev->bdev); | |
3861 | wr_ctx->pages_per_wr_bio = min_t(int, SCRUB_PAGES_PER_WR_BIO, | |
3862 | bio_get_nr_vecs(dev->bdev)); | |
3863 | wr_ctx->tgtdev = dev; | |
3864 | atomic_set(&wr_ctx->flush_all_writes, 0); | |
3865 | return 0; | |
3866 | } | |
3867 | ||
3868 | static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx) | |
3869 | { | |
3870 | mutex_lock(&wr_ctx->wr_lock); | |
3871 | kfree(wr_ctx->wr_curr_bio); | |
3872 | wr_ctx->wr_curr_bio = NULL; | |
3873 | mutex_unlock(&wr_ctx->wr_lock); | |
3874 | } | |
3875 | ||
3876 | static int copy_nocow_pages(struct scrub_ctx *sctx, u64 logical, u64 len, | |
3877 | int mirror_num, u64 physical_for_dev_replace) | |
3878 | { | |
3879 | struct scrub_copy_nocow_ctx *nocow_ctx; | |
3880 | struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info; | |
3881 | ||
3882 | nocow_ctx = kzalloc(sizeof(*nocow_ctx), GFP_NOFS); | |
3883 | if (!nocow_ctx) { | |
3884 | spin_lock(&sctx->stat_lock); | |
3885 | sctx->stat.malloc_errors++; | |
3886 | spin_unlock(&sctx->stat_lock); | |
3887 | return -ENOMEM; | |
3888 | } | |
3889 | ||
3890 | scrub_pending_trans_workers_inc(sctx); | |
3891 | ||
3892 | nocow_ctx->sctx = sctx; | |
3893 | nocow_ctx->logical = logical; | |
3894 | nocow_ctx->len = len; | |
3895 | nocow_ctx->mirror_num = mirror_num; | |
3896 | nocow_ctx->physical_for_dev_replace = physical_for_dev_replace; | |
9e0af237 LB |
3897 | btrfs_init_work(&nocow_ctx->work, btrfs_scrubnc_helper, |
3898 | copy_nocow_pages_worker, NULL, NULL); | |
652f25a2 | 3899 | INIT_LIST_HEAD(&nocow_ctx->inodes); |
0339ef2f QW |
3900 | btrfs_queue_work(fs_info->scrub_nocow_workers, |
3901 | &nocow_ctx->work); | |
ff023aac SB |
3902 | |
3903 | return 0; | |
3904 | } | |
3905 | ||
652f25a2 JB |
3906 | static int record_inode_for_nocow(u64 inum, u64 offset, u64 root, void *ctx) |
3907 | { | |
3908 | struct scrub_copy_nocow_ctx *nocow_ctx = ctx; | |
3909 | struct scrub_nocow_inode *nocow_inode; | |
3910 | ||
3911 | nocow_inode = kzalloc(sizeof(*nocow_inode), GFP_NOFS); | |
3912 | if (!nocow_inode) | |
3913 | return -ENOMEM; | |
3914 | nocow_inode->inum = inum; | |
3915 | nocow_inode->offset = offset; | |
3916 | nocow_inode->root = root; | |
3917 | list_add_tail(&nocow_inode->list, &nocow_ctx->inodes); | |
3918 | return 0; | |
3919 | } | |
3920 | ||
3921 | #define COPY_COMPLETE 1 | |
3922 | ||
ff023aac SB |
3923 | static void copy_nocow_pages_worker(struct btrfs_work *work) |
3924 | { | |
3925 | struct scrub_copy_nocow_ctx *nocow_ctx = | |
3926 | container_of(work, struct scrub_copy_nocow_ctx, work); | |
3927 | struct scrub_ctx *sctx = nocow_ctx->sctx; | |
3928 | u64 logical = nocow_ctx->logical; | |
3929 | u64 len = nocow_ctx->len; | |
3930 | int mirror_num = nocow_ctx->mirror_num; | |
3931 | u64 physical_for_dev_replace = nocow_ctx->physical_for_dev_replace; | |
3932 | int ret; | |
3933 | struct btrfs_trans_handle *trans = NULL; | |
3934 | struct btrfs_fs_info *fs_info; | |
3935 | struct btrfs_path *path; | |
3936 | struct btrfs_root *root; | |
3937 | int not_written = 0; | |
3938 | ||
3939 | fs_info = sctx->dev_root->fs_info; | |
3940 | root = fs_info->extent_root; | |
3941 | ||
3942 | path = btrfs_alloc_path(); | |
3943 | if (!path) { | |
3944 | spin_lock(&sctx->stat_lock); | |
3945 | sctx->stat.malloc_errors++; | |
3946 | spin_unlock(&sctx->stat_lock); | |
3947 | not_written = 1; | |
3948 | goto out; | |
3949 | } | |
3950 | ||
3951 | trans = btrfs_join_transaction(root); | |
3952 | if (IS_ERR(trans)) { | |
3953 | not_written = 1; | |
3954 | goto out; | |
3955 | } | |
3956 | ||
3957 | ret = iterate_inodes_from_logical(logical, fs_info, path, | |
652f25a2 | 3958 | record_inode_for_nocow, nocow_ctx); |
ff023aac | 3959 | if (ret != 0 && ret != -ENOENT) { |
efe120a0 FH |
3960 | btrfs_warn(fs_info, "iterate_inodes_from_logical() failed: log %llu, " |
3961 | "phys %llu, len %llu, mir %u, ret %d", | |
118a0a25 GU |
3962 | logical, physical_for_dev_replace, len, mirror_num, |
3963 | ret); | |
ff023aac SB |
3964 | not_written = 1; |
3965 | goto out; | |
3966 | } | |
3967 | ||
652f25a2 JB |
3968 | btrfs_end_transaction(trans, root); |
3969 | trans = NULL; | |
3970 | while (!list_empty(&nocow_ctx->inodes)) { | |
3971 | struct scrub_nocow_inode *entry; | |
3972 | entry = list_first_entry(&nocow_ctx->inodes, | |
3973 | struct scrub_nocow_inode, | |
3974 | list); | |
3975 | list_del_init(&entry->list); | |
3976 | ret = copy_nocow_pages_for_inode(entry->inum, entry->offset, | |
3977 | entry->root, nocow_ctx); | |
3978 | kfree(entry); | |
3979 | if (ret == COPY_COMPLETE) { | |
3980 | ret = 0; | |
3981 | break; | |
3982 | } else if (ret) { | |
3983 | break; | |
3984 | } | |
3985 | } | |
ff023aac | 3986 | out: |
652f25a2 JB |
3987 | while (!list_empty(&nocow_ctx->inodes)) { |
3988 | struct scrub_nocow_inode *entry; | |
3989 | entry = list_first_entry(&nocow_ctx->inodes, | |
3990 | struct scrub_nocow_inode, | |
3991 | list); | |
3992 | list_del_init(&entry->list); | |
3993 | kfree(entry); | |
3994 | } | |
ff023aac SB |
3995 | if (trans && !IS_ERR(trans)) |
3996 | btrfs_end_transaction(trans, root); | |
3997 | if (not_written) | |
3998 | btrfs_dev_replace_stats_inc(&fs_info->dev_replace. | |
3999 | num_uncorrectable_read_errors); | |
4000 | ||
4001 | btrfs_free_path(path); | |
4002 | kfree(nocow_ctx); | |
4003 | ||
4004 | scrub_pending_trans_workers_dec(sctx); | |
4005 | } | |
4006 | ||
32159242 GH |
4007 | static int check_extent_to_block(struct inode *inode, u64 start, u64 len, |
4008 | u64 logical) | |
4009 | { | |
4010 | struct extent_state *cached_state = NULL; | |
4011 | struct btrfs_ordered_extent *ordered; | |
4012 | struct extent_io_tree *io_tree; | |
4013 | struct extent_map *em; | |
4014 | u64 lockstart = start, lockend = start + len - 1; | |
4015 | int ret = 0; | |
4016 | ||
4017 | io_tree = &BTRFS_I(inode)->io_tree; | |
4018 | ||
4019 | lock_extent_bits(io_tree, lockstart, lockend, 0, &cached_state); | |
4020 | ordered = btrfs_lookup_ordered_range(inode, lockstart, len); | |
4021 | if (ordered) { | |
4022 | btrfs_put_ordered_extent(ordered); | |
4023 | ret = 1; | |
4024 | goto out_unlock; | |
4025 | } | |
4026 | ||
4027 | em = btrfs_get_extent(inode, NULL, 0, start, len, 0); | |
4028 | if (IS_ERR(em)) { | |
4029 | ret = PTR_ERR(em); | |
4030 | goto out_unlock; | |
4031 | } | |
4032 | ||
4033 | /* | |
4034 | * This extent does not actually cover the logical extent anymore, | |
4035 | * move on to the next inode. | |
4036 | */ | |
4037 | if (em->block_start > logical || | |
4038 | em->block_start + em->block_len < logical + len) { | |
4039 | free_extent_map(em); | |
4040 | ret = 1; | |
4041 | goto out_unlock; | |
4042 | } | |
4043 | free_extent_map(em); | |
4044 | ||
4045 | out_unlock: | |
4046 | unlock_extent_cached(io_tree, lockstart, lockend, &cached_state, | |
4047 | GFP_NOFS); | |
4048 | return ret; | |
4049 | } | |
4050 | ||
652f25a2 JB |
4051 | static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root, |
4052 | struct scrub_copy_nocow_ctx *nocow_ctx) | |
ff023aac | 4053 | { |
826aa0a8 | 4054 | struct btrfs_fs_info *fs_info = nocow_ctx->sctx->dev_root->fs_info; |
ff023aac | 4055 | struct btrfs_key key; |
826aa0a8 MX |
4056 | struct inode *inode; |
4057 | struct page *page; | |
ff023aac | 4058 | struct btrfs_root *local_root; |
652f25a2 | 4059 | struct extent_io_tree *io_tree; |
ff023aac | 4060 | u64 physical_for_dev_replace; |
32159242 | 4061 | u64 nocow_ctx_logical; |
652f25a2 | 4062 | u64 len = nocow_ctx->len; |
826aa0a8 | 4063 | unsigned long index; |
6f1c3605 | 4064 | int srcu_index; |
652f25a2 JB |
4065 | int ret = 0; |
4066 | int err = 0; | |
ff023aac SB |
4067 | |
4068 | key.objectid = root; | |
4069 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4070 | key.offset = (u64)-1; | |
6f1c3605 LB |
4071 | |
4072 | srcu_index = srcu_read_lock(&fs_info->subvol_srcu); | |
4073 | ||
ff023aac | 4074 | local_root = btrfs_read_fs_root_no_name(fs_info, &key); |
6f1c3605 LB |
4075 | if (IS_ERR(local_root)) { |
4076 | srcu_read_unlock(&fs_info->subvol_srcu, srcu_index); | |
ff023aac | 4077 | return PTR_ERR(local_root); |
6f1c3605 | 4078 | } |
ff023aac SB |
4079 | |
4080 | key.type = BTRFS_INODE_ITEM_KEY; | |
4081 | key.objectid = inum; | |
4082 | key.offset = 0; | |
4083 | inode = btrfs_iget(fs_info->sb, &key, local_root, NULL); | |
6f1c3605 | 4084 | srcu_read_unlock(&fs_info->subvol_srcu, srcu_index); |
ff023aac SB |
4085 | if (IS_ERR(inode)) |
4086 | return PTR_ERR(inode); | |
4087 | ||
edd1400b MX |
4088 | /* Avoid truncate/dio/punch hole.. */ |
4089 | mutex_lock(&inode->i_mutex); | |
4090 | inode_dio_wait(inode); | |
4091 | ||
ff023aac | 4092 | physical_for_dev_replace = nocow_ctx->physical_for_dev_replace; |
652f25a2 | 4093 | io_tree = &BTRFS_I(inode)->io_tree; |
32159242 | 4094 | nocow_ctx_logical = nocow_ctx->logical; |
652f25a2 | 4095 | |
32159242 GH |
4096 | ret = check_extent_to_block(inode, offset, len, nocow_ctx_logical); |
4097 | if (ret) { | |
4098 | ret = ret > 0 ? 0 : ret; | |
4099 | goto out; | |
652f25a2 | 4100 | } |
652f25a2 | 4101 | |
ff023aac | 4102 | while (len >= PAGE_CACHE_SIZE) { |
ff023aac | 4103 | index = offset >> PAGE_CACHE_SHIFT; |
edd1400b | 4104 | again: |
ff023aac SB |
4105 | page = find_or_create_page(inode->i_mapping, index, GFP_NOFS); |
4106 | if (!page) { | |
efe120a0 | 4107 | btrfs_err(fs_info, "find_or_create_page() failed"); |
ff023aac | 4108 | ret = -ENOMEM; |
826aa0a8 | 4109 | goto out; |
ff023aac SB |
4110 | } |
4111 | ||
4112 | if (PageUptodate(page)) { | |
4113 | if (PageDirty(page)) | |
4114 | goto next_page; | |
4115 | } else { | |
4116 | ClearPageError(page); | |
32159242 | 4117 | err = extent_read_full_page(io_tree, page, |
652f25a2 JB |
4118 | btrfs_get_extent, |
4119 | nocow_ctx->mirror_num); | |
826aa0a8 MX |
4120 | if (err) { |
4121 | ret = err; | |
ff023aac SB |
4122 | goto next_page; |
4123 | } | |
edd1400b | 4124 | |
26b25891 | 4125 | lock_page(page); |
edd1400b MX |
4126 | /* |
4127 | * If the page has been remove from the page cache, | |
4128 | * the data on it is meaningless, because it may be | |
4129 | * old one, the new data may be written into the new | |
4130 | * page in the page cache. | |
4131 | */ | |
4132 | if (page->mapping != inode->i_mapping) { | |
652f25a2 | 4133 | unlock_page(page); |
edd1400b MX |
4134 | page_cache_release(page); |
4135 | goto again; | |
4136 | } | |
ff023aac SB |
4137 | if (!PageUptodate(page)) { |
4138 | ret = -EIO; | |
4139 | goto next_page; | |
4140 | } | |
4141 | } | |
32159242 GH |
4142 | |
4143 | ret = check_extent_to_block(inode, offset, len, | |
4144 | nocow_ctx_logical); | |
4145 | if (ret) { | |
4146 | ret = ret > 0 ? 0 : ret; | |
4147 | goto next_page; | |
4148 | } | |
4149 | ||
826aa0a8 MX |
4150 | err = write_page_nocow(nocow_ctx->sctx, |
4151 | physical_for_dev_replace, page); | |
4152 | if (err) | |
4153 | ret = err; | |
ff023aac | 4154 | next_page: |
826aa0a8 MX |
4155 | unlock_page(page); |
4156 | page_cache_release(page); | |
4157 | ||
4158 | if (ret) | |
4159 | break; | |
4160 | ||
ff023aac SB |
4161 | offset += PAGE_CACHE_SIZE; |
4162 | physical_for_dev_replace += PAGE_CACHE_SIZE; | |
32159242 | 4163 | nocow_ctx_logical += PAGE_CACHE_SIZE; |
ff023aac SB |
4164 | len -= PAGE_CACHE_SIZE; |
4165 | } | |
652f25a2 | 4166 | ret = COPY_COMPLETE; |
826aa0a8 | 4167 | out: |
edd1400b | 4168 | mutex_unlock(&inode->i_mutex); |
826aa0a8 | 4169 | iput(inode); |
ff023aac SB |
4170 | return ret; |
4171 | } | |
4172 | ||
4173 | static int write_page_nocow(struct scrub_ctx *sctx, | |
4174 | u64 physical_for_dev_replace, struct page *page) | |
4175 | { | |
4176 | struct bio *bio; | |
4177 | struct btrfs_device *dev; | |
4178 | int ret; | |
ff023aac SB |
4179 | |
4180 | dev = sctx->wr_ctx.tgtdev; | |
4181 | if (!dev) | |
4182 | return -EIO; | |
4183 | if (!dev->bdev) { | |
4184 | printk_ratelimited(KERN_WARNING | |
efe120a0 | 4185 | "BTRFS: scrub write_page_nocow(bdev == NULL) is unexpected!\n"); |
ff023aac SB |
4186 | return -EIO; |
4187 | } | |
9be3395b | 4188 | bio = btrfs_io_bio_alloc(GFP_NOFS, 1); |
ff023aac SB |
4189 | if (!bio) { |
4190 | spin_lock(&sctx->stat_lock); | |
4191 | sctx->stat.malloc_errors++; | |
4192 | spin_unlock(&sctx->stat_lock); | |
4193 | return -ENOMEM; | |
4194 | } | |
4f024f37 KO |
4195 | bio->bi_iter.bi_size = 0; |
4196 | bio->bi_iter.bi_sector = physical_for_dev_replace >> 9; | |
ff023aac SB |
4197 | bio->bi_bdev = dev->bdev; |
4198 | ret = bio_add_page(bio, page, PAGE_CACHE_SIZE, 0); | |
4199 | if (ret != PAGE_CACHE_SIZE) { | |
4200 | leave_with_eio: | |
4201 | bio_put(bio); | |
4202 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS); | |
4203 | return -EIO; | |
4204 | } | |
ff023aac | 4205 | |
33879d45 | 4206 | if (btrfsic_submit_bio_wait(WRITE_SYNC, bio)) |
ff023aac SB |
4207 | goto leave_with_eio; |
4208 | ||
4209 | bio_put(bio); | |
4210 | return 0; | |
4211 | } |