Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
a2de733c | 2 | /* |
b6bfebc1 | 3 | * Copyright (C) 2011, 2012 STRATO. All rights reserved. |
a2de733c AJ |
4 | */ |
5 | ||
a2de733c | 6 | #include <linux/blkdev.h> |
558540c1 | 7 | #include <linux/ratelimit.h> |
de2491fd | 8 | #include <linux/sched/mm.h> |
d5178578 | 9 | #include <crypto/hash.h> |
a2de733c | 10 | #include "ctree.h" |
6e80d4f8 | 11 | #include "discard.h" |
a2de733c AJ |
12 | #include "volumes.h" |
13 | #include "disk-io.h" | |
14 | #include "ordered-data.h" | |
0ef8e451 | 15 | #include "transaction.h" |
558540c1 | 16 | #include "backref.h" |
5da6fcbc | 17 | #include "extent_io.h" |
ff023aac | 18 | #include "dev-replace.h" |
21adbd5c | 19 | #include "check-integrity.h" |
606686ee | 20 | #include "rcu-string.h" |
53b381b3 | 21 | #include "raid56.h" |
aac0023c | 22 | #include "block-group.h" |
12659251 | 23 | #include "zoned.h" |
a2de733c AJ |
24 | |
25 | /* | |
26 | * This is only the first step towards a full-features scrub. It reads all | |
27 | * extent and super block and verifies the checksums. In case a bad checksum | |
28 | * is found or the extent cannot be read, good data will be written back if | |
29 | * any can be found. | |
30 | * | |
31 | * Future enhancements: | |
a2de733c AJ |
32 | * - In case an unrepairable extent is encountered, track which files are |
33 | * affected and report them | |
a2de733c | 34 | * - track and record media errors, throw out bad devices |
a2de733c | 35 | * - add a mode to also read unallocated space |
a2de733c AJ |
36 | */ |
37 | ||
b5d67f64 | 38 | struct scrub_block; |
d9d181c1 | 39 | struct scrub_ctx; |
a2de733c | 40 | |
ff023aac SB |
41 | /* |
42 | * the following three values only influence the performance. | |
43 | * The last one configures the number of parallel and outstanding I/O | |
44 | * operations. The first two values configure an upper limit for the number | |
45 | * of (dynamically allocated) pages that are added to a bio. | |
46 | */ | |
47 | #define SCRUB_PAGES_PER_RD_BIO 32 /* 128k per bio */ | |
48 | #define SCRUB_PAGES_PER_WR_BIO 32 /* 128k per bio */ | |
49 | #define SCRUB_BIOS_PER_SCTX 64 /* 8MB per device in flight */ | |
7a9e9987 SB |
50 | |
51 | /* | |
52 | * the following value times PAGE_SIZE needs to be large enough to match the | |
53 | * largest node/leaf/sector size that shall be supported. | |
54 | * Values larger than BTRFS_STRIPE_LEN are not supported. | |
55 | */ | |
b5d67f64 | 56 | #define SCRUB_MAX_PAGES_PER_BLOCK 16 /* 64k per node/leaf/sector */ |
a2de733c | 57 | |
af8e2d1d | 58 | struct scrub_recover { |
6f615018 | 59 | refcount_t refs; |
af8e2d1d | 60 | struct btrfs_bio *bbio; |
af8e2d1d MX |
61 | u64 map_length; |
62 | }; | |
63 | ||
a2de733c | 64 | struct scrub_page { |
b5d67f64 SB |
65 | struct scrub_block *sblock; |
66 | struct page *page; | |
442a4f63 | 67 | struct btrfs_device *dev; |
5a6ac9ea | 68 | struct list_head list; |
a2de733c AJ |
69 | u64 flags; /* extent flags */ |
70 | u64 generation; | |
b5d67f64 SB |
71 | u64 logical; |
72 | u64 physical; | |
ff023aac | 73 | u64 physical_for_dev_replace; |
57019345 | 74 | atomic_t refs; |
2c363954 QW |
75 | u8 mirror_num; |
76 | int have_csum:1; | |
77 | int io_error:1; | |
a2de733c | 78 | u8 csum[BTRFS_CSUM_SIZE]; |
af8e2d1d MX |
79 | |
80 | struct scrub_recover *recover; | |
a2de733c AJ |
81 | }; |
82 | ||
83 | struct scrub_bio { | |
84 | int index; | |
d9d181c1 | 85 | struct scrub_ctx *sctx; |
a36cf8b8 | 86 | struct btrfs_device *dev; |
a2de733c | 87 | struct bio *bio; |
4e4cbee9 | 88 | blk_status_t status; |
a2de733c AJ |
89 | u64 logical; |
90 | u64 physical; | |
ff023aac SB |
91 | #if SCRUB_PAGES_PER_WR_BIO >= SCRUB_PAGES_PER_RD_BIO |
92 | struct scrub_page *pagev[SCRUB_PAGES_PER_WR_BIO]; | |
93 | #else | |
94 | struct scrub_page *pagev[SCRUB_PAGES_PER_RD_BIO]; | |
95 | #endif | |
b5d67f64 | 96 | int page_count; |
a2de733c AJ |
97 | int next_free; |
98 | struct btrfs_work work; | |
99 | }; | |
100 | ||
b5d67f64 | 101 | struct scrub_block { |
7a9e9987 | 102 | struct scrub_page *pagev[SCRUB_MAX_PAGES_PER_BLOCK]; |
b5d67f64 SB |
103 | int page_count; |
104 | atomic_t outstanding_pages; | |
186debd6 | 105 | refcount_t refs; /* free mem on transition to zero */ |
d9d181c1 | 106 | struct scrub_ctx *sctx; |
5a6ac9ea | 107 | struct scrub_parity *sparity; |
b5d67f64 SB |
108 | struct { |
109 | unsigned int header_error:1; | |
110 | unsigned int checksum_error:1; | |
111 | unsigned int no_io_error_seen:1; | |
442a4f63 | 112 | unsigned int generation_error:1; /* also sets header_error */ |
5a6ac9ea MX |
113 | |
114 | /* The following is for the data used to check parity */ | |
115 | /* It is for the data with checksum */ | |
116 | unsigned int data_corrected:1; | |
b5d67f64 | 117 | }; |
73ff61db | 118 | struct btrfs_work work; |
b5d67f64 SB |
119 | }; |
120 | ||
5a6ac9ea MX |
121 | /* Used for the chunks with parity stripe such RAID5/6 */ |
122 | struct scrub_parity { | |
123 | struct scrub_ctx *sctx; | |
124 | ||
125 | struct btrfs_device *scrub_dev; | |
126 | ||
127 | u64 logic_start; | |
128 | ||
129 | u64 logic_end; | |
130 | ||
131 | int nsectors; | |
132 | ||
fa485d21 | 133 | u32 stripe_len; |
5a6ac9ea | 134 | |
78a76450 | 135 | refcount_t refs; |
5a6ac9ea MX |
136 | |
137 | struct list_head spages; | |
138 | ||
139 | /* Work of parity check and repair */ | |
140 | struct btrfs_work work; | |
141 | ||
142 | /* Mark the parity blocks which have data */ | |
143 | unsigned long *dbitmap; | |
144 | ||
145 | /* | |
146 | * Mark the parity blocks which have data, but errors happen when | |
147 | * read data or check data | |
148 | */ | |
149 | unsigned long *ebitmap; | |
150 | ||
a8753ee3 | 151 | unsigned long bitmap[]; |
5a6ac9ea MX |
152 | }; |
153 | ||
d9d181c1 | 154 | struct scrub_ctx { |
ff023aac | 155 | struct scrub_bio *bios[SCRUB_BIOS_PER_SCTX]; |
fb456252 | 156 | struct btrfs_fs_info *fs_info; |
a2de733c AJ |
157 | int first_free; |
158 | int curr; | |
b6bfebc1 SB |
159 | atomic_t bios_in_flight; |
160 | atomic_t workers_pending; | |
a2de733c AJ |
161 | spinlock_t list_lock; |
162 | wait_queue_head_t list_wait; | |
a2de733c AJ |
163 | struct list_head csum_list; |
164 | atomic_t cancel_req; | |
8628764e | 165 | int readonly; |
ff023aac | 166 | int pages_per_rd_bio; |
63a212ab SB |
167 | |
168 | int is_dev_replace; | |
de17addc | 169 | u64 write_pointer; |
3fb99303 DS |
170 | |
171 | struct scrub_bio *wr_curr_bio; | |
172 | struct mutex wr_lock; | |
173 | int pages_per_wr_bio; /* <= SCRUB_PAGES_PER_WR_BIO */ | |
3fb99303 | 174 | struct btrfs_device *wr_tgtdev; |
2073c4c2 | 175 | bool flush_all_writes; |
63a212ab | 176 | |
a2de733c AJ |
177 | /* |
178 | * statistics | |
179 | */ | |
180 | struct btrfs_scrub_progress stat; | |
181 | spinlock_t stat_lock; | |
f55985f4 FM |
182 | |
183 | /* | |
184 | * Use a ref counter to avoid use-after-free issues. Scrub workers | |
185 | * decrement bios_in_flight and workers_pending and then do a wakeup | |
186 | * on the list_wait wait queue. We must ensure the main scrub task | |
187 | * doesn't free the scrub context before or while the workers are | |
188 | * doing the wakeup() call. | |
189 | */ | |
99f4cdb1 | 190 | refcount_t refs; |
a2de733c AJ |
191 | }; |
192 | ||
558540c1 JS |
193 | struct scrub_warning { |
194 | struct btrfs_path *path; | |
195 | u64 extent_item_size; | |
558540c1 | 196 | const char *errstr; |
6aa21263 | 197 | u64 physical; |
558540c1 JS |
198 | u64 logical; |
199 | struct btrfs_device *dev; | |
558540c1 JS |
200 | }; |
201 | ||
0966a7b1 QW |
202 | struct full_stripe_lock { |
203 | struct rb_node node; | |
204 | u64 logical; | |
205 | u64 refs; | |
206 | struct mutex mutex; | |
207 | }; | |
208 | ||
b6bfebc1 SB |
209 | static void scrub_pending_bio_inc(struct scrub_ctx *sctx); |
210 | static void scrub_pending_bio_dec(struct scrub_ctx *sctx); | |
b5d67f64 | 211 | static int scrub_handle_errored_block(struct scrub_block *sblock_to_check); |
be50a8dd | 212 | static int scrub_setup_recheck_block(struct scrub_block *original_sblock, |
ff023aac | 213 | struct scrub_block *sblocks_for_recheck); |
34f5c8e9 | 214 | static void scrub_recheck_block(struct btrfs_fs_info *fs_info, |
affe4a5a ZL |
215 | struct scrub_block *sblock, |
216 | int retry_failed_mirror); | |
ba7cf988 | 217 | static void scrub_recheck_block_checksum(struct scrub_block *sblock); |
b5d67f64 | 218 | static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad, |
114ab50d | 219 | struct scrub_block *sblock_good); |
b5d67f64 SB |
220 | static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad, |
221 | struct scrub_block *sblock_good, | |
222 | int page_num, int force_write); | |
ff023aac SB |
223 | static void scrub_write_block_to_dev_replace(struct scrub_block *sblock); |
224 | static int scrub_write_page_to_dev_replace(struct scrub_block *sblock, | |
225 | int page_num); | |
b5d67f64 SB |
226 | static int scrub_checksum_data(struct scrub_block *sblock); |
227 | static int scrub_checksum_tree_block(struct scrub_block *sblock); | |
228 | static int scrub_checksum_super(struct scrub_block *sblock); | |
229 | static void scrub_block_get(struct scrub_block *sblock); | |
230 | static void scrub_block_put(struct scrub_block *sblock); | |
7a9e9987 SB |
231 | static void scrub_page_get(struct scrub_page *spage); |
232 | static void scrub_page_put(struct scrub_page *spage); | |
5a6ac9ea MX |
233 | static void scrub_parity_get(struct scrub_parity *sparity); |
234 | static void scrub_parity_put(struct scrub_parity *sparity); | |
ff023aac SB |
235 | static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx, |
236 | struct scrub_page *spage); | |
fa485d21 | 237 | static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u32 len, |
a36cf8b8 | 238 | u64 physical, struct btrfs_device *dev, u64 flags, |
96e63a45 | 239 | u64 gen, int mirror_num, u8 *csum, |
ff023aac | 240 | u64 physical_for_dev_replace); |
4246a0b6 | 241 | static void scrub_bio_end_io(struct bio *bio); |
b5d67f64 SB |
242 | static void scrub_bio_end_io_worker(struct btrfs_work *work); |
243 | static void scrub_block_complete(struct scrub_block *sblock); | |
ff023aac | 244 | static void scrub_remap_extent(struct btrfs_fs_info *fs_info, |
fa485d21 | 245 | u64 extent_logical, u32 extent_len, |
ff023aac SB |
246 | u64 *extent_physical, |
247 | struct btrfs_device **extent_dev, | |
248 | int *extent_mirror_num); | |
ff023aac SB |
249 | static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx, |
250 | struct scrub_page *spage); | |
251 | static void scrub_wr_submit(struct scrub_ctx *sctx); | |
4246a0b6 | 252 | static void scrub_wr_bio_end_io(struct bio *bio); |
ff023aac | 253 | static void scrub_wr_bio_end_io_worker(struct btrfs_work *work); |
cb7ab021 | 254 | static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info); |
3cb0929a | 255 | static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info); |
f55985f4 | 256 | static void scrub_put_ctx(struct scrub_ctx *sctx); |
1623edeb | 257 | |
261d2dcb | 258 | static inline int scrub_is_page_on_raid56(struct scrub_page *spage) |
762221f0 | 259 | { |
261d2dcb QW |
260 | return spage->recover && |
261 | (spage->recover->bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK); | |
762221f0 | 262 | } |
1623edeb | 263 | |
b6bfebc1 SB |
264 | static void scrub_pending_bio_inc(struct scrub_ctx *sctx) |
265 | { | |
99f4cdb1 | 266 | refcount_inc(&sctx->refs); |
b6bfebc1 SB |
267 | atomic_inc(&sctx->bios_in_flight); |
268 | } | |
269 | ||
270 | static void scrub_pending_bio_dec(struct scrub_ctx *sctx) | |
271 | { | |
272 | atomic_dec(&sctx->bios_in_flight); | |
273 | wake_up(&sctx->list_wait); | |
f55985f4 | 274 | scrub_put_ctx(sctx); |
b6bfebc1 SB |
275 | } |
276 | ||
cb7ab021 | 277 | static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info) |
3cb0929a WS |
278 | { |
279 | while (atomic_read(&fs_info->scrub_pause_req)) { | |
280 | mutex_unlock(&fs_info->scrub_lock); | |
281 | wait_event(fs_info->scrub_pause_wait, | |
282 | atomic_read(&fs_info->scrub_pause_req) == 0); | |
283 | mutex_lock(&fs_info->scrub_lock); | |
284 | } | |
285 | } | |
286 | ||
0e22be89 | 287 | static void scrub_pause_on(struct btrfs_fs_info *fs_info) |
cb7ab021 WS |
288 | { |
289 | atomic_inc(&fs_info->scrubs_paused); | |
290 | wake_up(&fs_info->scrub_pause_wait); | |
0e22be89 | 291 | } |
cb7ab021 | 292 | |
0e22be89 Z |
293 | static void scrub_pause_off(struct btrfs_fs_info *fs_info) |
294 | { | |
cb7ab021 WS |
295 | mutex_lock(&fs_info->scrub_lock); |
296 | __scrub_blocked_if_needed(fs_info); | |
297 | atomic_dec(&fs_info->scrubs_paused); | |
298 | mutex_unlock(&fs_info->scrub_lock); | |
299 | ||
300 | wake_up(&fs_info->scrub_pause_wait); | |
301 | } | |
302 | ||
0e22be89 Z |
303 | static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info) |
304 | { | |
305 | scrub_pause_on(fs_info); | |
306 | scrub_pause_off(fs_info); | |
307 | } | |
308 | ||
0966a7b1 QW |
309 | /* |
310 | * Insert new full stripe lock into full stripe locks tree | |
311 | * | |
312 | * Return pointer to existing or newly inserted full_stripe_lock structure if | |
313 | * everything works well. | |
314 | * Return ERR_PTR(-ENOMEM) if we failed to allocate memory | |
315 | * | |
316 | * NOTE: caller must hold full_stripe_locks_root->lock before calling this | |
317 | * function | |
318 | */ | |
319 | static struct full_stripe_lock *insert_full_stripe_lock( | |
320 | struct btrfs_full_stripe_locks_tree *locks_root, | |
321 | u64 fstripe_logical) | |
322 | { | |
323 | struct rb_node **p; | |
324 | struct rb_node *parent = NULL; | |
325 | struct full_stripe_lock *entry; | |
326 | struct full_stripe_lock *ret; | |
327 | ||
a32bf9a3 | 328 | lockdep_assert_held(&locks_root->lock); |
0966a7b1 QW |
329 | |
330 | p = &locks_root->root.rb_node; | |
331 | while (*p) { | |
332 | parent = *p; | |
333 | entry = rb_entry(parent, struct full_stripe_lock, node); | |
334 | if (fstripe_logical < entry->logical) { | |
335 | p = &(*p)->rb_left; | |
336 | } else if (fstripe_logical > entry->logical) { | |
337 | p = &(*p)->rb_right; | |
338 | } else { | |
339 | entry->refs++; | |
340 | return entry; | |
341 | } | |
342 | } | |
343 | ||
a5fb1142 FM |
344 | /* |
345 | * Insert new lock. | |
a5fb1142 | 346 | */ |
0966a7b1 QW |
347 | ret = kmalloc(sizeof(*ret), GFP_KERNEL); |
348 | if (!ret) | |
349 | return ERR_PTR(-ENOMEM); | |
350 | ret->logical = fstripe_logical; | |
351 | ret->refs = 1; | |
352 | mutex_init(&ret->mutex); | |
353 | ||
354 | rb_link_node(&ret->node, parent, p); | |
355 | rb_insert_color(&ret->node, &locks_root->root); | |
356 | return ret; | |
357 | } | |
358 | ||
359 | /* | |
360 | * Search for a full stripe lock of a block group | |
361 | * | |
362 | * Return pointer to existing full stripe lock if found | |
363 | * Return NULL if not found | |
364 | */ | |
365 | static struct full_stripe_lock *search_full_stripe_lock( | |
366 | struct btrfs_full_stripe_locks_tree *locks_root, | |
367 | u64 fstripe_logical) | |
368 | { | |
369 | struct rb_node *node; | |
370 | struct full_stripe_lock *entry; | |
371 | ||
a32bf9a3 | 372 | lockdep_assert_held(&locks_root->lock); |
0966a7b1 QW |
373 | |
374 | node = locks_root->root.rb_node; | |
375 | while (node) { | |
376 | entry = rb_entry(node, struct full_stripe_lock, node); | |
377 | if (fstripe_logical < entry->logical) | |
378 | node = node->rb_left; | |
379 | else if (fstripe_logical > entry->logical) | |
380 | node = node->rb_right; | |
381 | else | |
382 | return entry; | |
383 | } | |
384 | return NULL; | |
385 | } | |
386 | ||
387 | /* | |
388 | * Helper to get full stripe logical from a normal bytenr. | |
389 | * | |
390 | * Caller must ensure @cache is a RAID56 block group. | |
391 | */ | |
32da5386 | 392 | static u64 get_full_stripe_logical(struct btrfs_block_group *cache, u64 bytenr) |
0966a7b1 QW |
393 | { |
394 | u64 ret; | |
395 | ||
396 | /* | |
397 | * Due to chunk item size limit, full stripe length should not be | |
398 | * larger than U32_MAX. Just a sanity check here. | |
399 | */ | |
400 | WARN_ON_ONCE(cache->full_stripe_len >= U32_MAX); | |
401 | ||
402 | /* | |
403 | * round_down() can only handle power of 2, while RAID56 full | |
404 | * stripe length can be 64KiB * n, so we need to manually round down. | |
405 | */ | |
b3470b5d DS |
406 | ret = div64_u64(bytenr - cache->start, cache->full_stripe_len) * |
407 | cache->full_stripe_len + cache->start; | |
0966a7b1 QW |
408 | return ret; |
409 | } | |
410 | ||
411 | /* | |
412 | * Lock a full stripe to avoid concurrency of recovery and read | |
413 | * | |
414 | * It's only used for profiles with parities (RAID5/6), for other profiles it | |
415 | * does nothing. | |
416 | * | |
417 | * Return 0 if we locked full stripe covering @bytenr, with a mutex held. | |
418 | * So caller must call unlock_full_stripe() at the same context. | |
419 | * | |
420 | * Return <0 if encounters error. | |
421 | */ | |
422 | static int lock_full_stripe(struct btrfs_fs_info *fs_info, u64 bytenr, | |
423 | bool *locked_ret) | |
424 | { | |
32da5386 | 425 | struct btrfs_block_group *bg_cache; |
0966a7b1 QW |
426 | struct btrfs_full_stripe_locks_tree *locks_root; |
427 | struct full_stripe_lock *existing; | |
428 | u64 fstripe_start; | |
429 | int ret = 0; | |
430 | ||
431 | *locked_ret = false; | |
432 | bg_cache = btrfs_lookup_block_group(fs_info, bytenr); | |
433 | if (!bg_cache) { | |
434 | ASSERT(0); | |
435 | return -ENOENT; | |
436 | } | |
437 | ||
438 | /* Profiles not based on parity don't need full stripe lock */ | |
439 | if (!(bg_cache->flags & BTRFS_BLOCK_GROUP_RAID56_MASK)) | |
440 | goto out; | |
441 | locks_root = &bg_cache->full_stripe_locks_root; | |
442 | ||
443 | fstripe_start = get_full_stripe_logical(bg_cache, bytenr); | |
444 | ||
445 | /* Now insert the full stripe lock */ | |
446 | mutex_lock(&locks_root->lock); | |
447 | existing = insert_full_stripe_lock(locks_root, fstripe_start); | |
448 | mutex_unlock(&locks_root->lock); | |
449 | if (IS_ERR(existing)) { | |
450 | ret = PTR_ERR(existing); | |
451 | goto out; | |
452 | } | |
453 | mutex_lock(&existing->mutex); | |
454 | *locked_ret = true; | |
455 | out: | |
456 | btrfs_put_block_group(bg_cache); | |
457 | return ret; | |
458 | } | |
459 | ||
460 | /* | |
461 | * Unlock a full stripe. | |
462 | * | |
463 | * NOTE: Caller must ensure it's the same context calling corresponding | |
464 | * lock_full_stripe(). | |
465 | * | |
466 | * Return 0 if we unlock full stripe without problem. | |
467 | * Return <0 for error | |
468 | */ | |
469 | static int unlock_full_stripe(struct btrfs_fs_info *fs_info, u64 bytenr, | |
470 | bool locked) | |
471 | { | |
32da5386 | 472 | struct btrfs_block_group *bg_cache; |
0966a7b1 QW |
473 | struct btrfs_full_stripe_locks_tree *locks_root; |
474 | struct full_stripe_lock *fstripe_lock; | |
475 | u64 fstripe_start; | |
476 | bool freeit = false; | |
477 | int ret = 0; | |
478 | ||
479 | /* If we didn't acquire full stripe lock, no need to continue */ | |
480 | if (!locked) | |
481 | return 0; | |
482 | ||
483 | bg_cache = btrfs_lookup_block_group(fs_info, bytenr); | |
484 | if (!bg_cache) { | |
485 | ASSERT(0); | |
486 | return -ENOENT; | |
487 | } | |
488 | if (!(bg_cache->flags & BTRFS_BLOCK_GROUP_RAID56_MASK)) | |
489 | goto out; | |
490 | ||
491 | locks_root = &bg_cache->full_stripe_locks_root; | |
492 | fstripe_start = get_full_stripe_logical(bg_cache, bytenr); | |
493 | ||
494 | mutex_lock(&locks_root->lock); | |
495 | fstripe_lock = search_full_stripe_lock(locks_root, fstripe_start); | |
496 | /* Unpaired unlock_full_stripe() detected */ | |
497 | if (!fstripe_lock) { | |
498 | WARN_ON(1); | |
499 | ret = -ENOENT; | |
500 | mutex_unlock(&locks_root->lock); | |
501 | goto out; | |
502 | } | |
503 | ||
504 | if (fstripe_lock->refs == 0) { | |
505 | WARN_ON(1); | |
506 | btrfs_warn(fs_info, "full stripe lock at %llu refcount underflow", | |
507 | fstripe_lock->logical); | |
508 | } else { | |
509 | fstripe_lock->refs--; | |
510 | } | |
511 | ||
512 | if (fstripe_lock->refs == 0) { | |
513 | rb_erase(&fstripe_lock->node, &locks_root->root); | |
514 | freeit = true; | |
515 | } | |
516 | mutex_unlock(&locks_root->lock); | |
517 | ||
518 | mutex_unlock(&fstripe_lock->mutex); | |
519 | if (freeit) | |
520 | kfree(fstripe_lock); | |
521 | out: | |
522 | btrfs_put_block_group(bg_cache); | |
523 | return ret; | |
524 | } | |
525 | ||
d9d181c1 | 526 | static void scrub_free_csums(struct scrub_ctx *sctx) |
a2de733c | 527 | { |
d9d181c1 | 528 | while (!list_empty(&sctx->csum_list)) { |
a2de733c | 529 | struct btrfs_ordered_sum *sum; |
d9d181c1 | 530 | sum = list_first_entry(&sctx->csum_list, |
a2de733c AJ |
531 | struct btrfs_ordered_sum, list); |
532 | list_del(&sum->list); | |
533 | kfree(sum); | |
534 | } | |
535 | } | |
536 | ||
d9d181c1 | 537 | static noinline_for_stack void scrub_free_ctx(struct scrub_ctx *sctx) |
a2de733c AJ |
538 | { |
539 | int i; | |
a2de733c | 540 | |
d9d181c1 | 541 | if (!sctx) |
a2de733c AJ |
542 | return; |
543 | ||
b5d67f64 | 544 | /* this can happen when scrub is cancelled */ |
d9d181c1 SB |
545 | if (sctx->curr != -1) { |
546 | struct scrub_bio *sbio = sctx->bios[sctx->curr]; | |
b5d67f64 SB |
547 | |
548 | for (i = 0; i < sbio->page_count; i++) { | |
ff023aac | 549 | WARN_ON(!sbio->pagev[i]->page); |
b5d67f64 SB |
550 | scrub_block_put(sbio->pagev[i]->sblock); |
551 | } | |
552 | bio_put(sbio->bio); | |
553 | } | |
554 | ||
ff023aac | 555 | for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) { |
d9d181c1 | 556 | struct scrub_bio *sbio = sctx->bios[i]; |
a2de733c AJ |
557 | |
558 | if (!sbio) | |
559 | break; | |
a2de733c AJ |
560 | kfree(sbio); |
561 | } | |
562 | ||
3fb99303 | 563 | kfree(sctx->wr_curr_bio); |
d9d181c1 SB |
564 | scrub_free_csums(sctx); |
565 | kfree(sctx); | |
a2de733c AJ |
566 | } |
567 | ||
f55985f4 FM |
568 | static void scrub_put_ctx(struct scrub_ctx *sctx) |
569 | { | |
99f4cdb1 | 570 | if (refcount_dec_and_test(&sctx->refs)) |
f55985f4 FM |
571 | scrub_free_ctx(sctx); |
572 | } | |
573 | ||
92f7ba43 DS |
574 | static noinline_for_stack struct scrub_ctx *scrub_setup_ctx( |
575 | struct btrfs_fs_info *fs_info, int is_dev_replace) | |
a2de733c | 576 | { |
d9d181c1 | 577 | struct scrub_ctx *sctx; |
a2de733c | 578 | int i; |
a2de733c | 579 | |
58c4e173 | 580 | sctx = kzalloc(sizeof(*sctx), GFP_KERNEL); |
d9d181c1 | 581 | if (!sctx) |
a2de733c | 582 | goto nomem; |
99f4cdb1 | 583 | refcount_set(&sctx->refs, 1); |
63a212ab | 584 | sctx->is_dev_replace = is_dev_replace; |
b54ffb73 | 585 | sctx->pages_per_rd_bio = SCRUB_PAGES_PER_RD_BIO; |
d9d181c1 | 586 | sctx->curr = -1; |
92f7ba43 | 587 | sctx->fs_info = fs_info; |
e49be14b | 588 | INIT_LIST_HEAD(&sctx->csum_list); |
ff023aac | 589 | for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) { |
a2de733c AJ |
590 | struct scrub_bio *sbio; |
591 | ||
58c4e173 | 592 | sbio = kzalloc(sizeof(*sbio), GFP_KERNEL); |
a2de733c AJ |
593 | if (!sbio) |
594 | goto nomem; | |
d9d181c1 | 595 | sctx->bios[i] = sbio; |
a2de733c | 596 | |
a2de733c | 597 | sbio->index = i; |
d9d181c1 | 598 | sbio->sctx = sctx; |
b5d67f64 | 599 | sbio->page_count = 0; |
a0cac0ec OS |
600 | btrfs_init_work(&sbio->work, scrub_bio_end_io_worker, NULL, |
601 | NULL); | |
a2de733c | 602 | |
ff023aac | 603 | if (i != SCRUB_BIOS_PER_SCTX - 1) |
d9d181c1 | 604 | sctx->bios[i]->next_free = i + 1; |
0ef8e451 | 605 | else |
d9d181c1 SB |
606 | sctx->bios[i]->next_free = -1; |
607 | } | |
608 | sctx->first_free = 0; | |
b6bfebc1 SB |
609 | atomic_set(&sctx->bios_in_flight, 0); |
610 | atomic_set(&sctx->workers_pending, 0); | |
d9d181c1 | 611 | atomic_set(&sctx->cancel_req, 0); |
d9d181c1 SB |
612 | |
613 | spin_lock_init(&sctx->list_lock); | |
614 | spin_lock_init(&sctx->stat_lock); | |
615 | init_waitqueue_head(&sctx->list_wait); | |
ff023aac | 616 | |
3fb99303 DS |
617 | WARN_ON(sctx->wr_curr_bio != NULL); |
618 | mutex_init(&sctx->wr_lock); | |
619 | sctx->wr_curr_bio = NULL; | |
8fcdac3f | 620 | if (is_dev_replace) { |
ded56184 | 621 | WARN_ON(!fs_info->dev_replace.tgtdev); |
3fb99303 | 622 | sctx->pages_per_wr_bio = SCRUB_PAGES_PER_WR_BIO; |
ded56184 | 623 | sctx->wr_tgtdev = fs_info->dev_replace.tgtdev; |
2073c4c2 | 624 | sctx->flush_all_writes = false; |
ff023aac | 625 | } |
8fcdac3f | 626 | |
d9d181c1 | 627 | return sctx; |
a2de733c AJ |
628 | |
629 | nomem: | |
d9d181c1 | 630 | scrub_free_ctx(sctx); |
a2de733c AJ |
631 | return ERR_PTR(-ENOMEM); |
632 | } | |
633 | ||
ff023aac SB |
634 | static int scrub_print_warning_inode(u64 inum, u64 offset, u64 root, |
635 | void *warn_ctx) | |
558540c1 JS |
636 | { |
637 | u64 isize; | |
638 | u32 nlink; | |
639 | int ret; | |
640 | int i; | |
de2491fd | 641 | unsigned nofs_flag; |
558540c1 JS |
642 | struct extent_buffer *eb; |
643 | struct btrfs_inode_item *inode_item; | |
ff023aac | 644 | struct scrub_warning *swarn = warn_ctx; |
fb456252 | 645 | struct btrfs_fs_info *fs_info = swarn->dev->fs_info; |
558540c1 JS |
646 | struct inode_fs_paths *ipath = NULL; |
647 | struct btrfs_root *local_root; | |
1d4c08e0 | 648 | struct btrfs_key key; |
558540c1 | 649 | |
56e9357a | 650 | local_root = btrfs_get_fs_root(fs_info, root, true); |
558540c1 JS |
651 | if (IS_ERR(local_root)) { |
652 | ret = PTR_ERR(local_root); | |
653 | goto err; | |
654 | } | |
655 | ||
14692cc1 DS |
656 | /* |
657 | * this makes the path point to (inum INODE_ITEM ioff) | |
658 | */ | |
1d4c08e0 DS |
659 | key.objectid = inum; |
660 | key.type = BTRFS_INODE_ITEM_KEY; | |
661 | key.offset = 0; | |
662 | ||
663 | ret = btrfs_search_slot(NULL, local_root, &key, swarn->path, 0, 0); | |
558540c1 | 664 | if (ret) { |
00246528 | 665 | btrfs_put_root(local_root); |
558540c1 JS |
666 | btrfs_release_path(swarn->path); |
667 | goto err; | |
668 | } | |
669 | ||
670 | eb = swarn->path->nodes[0]; | |
671 | inode_item = btrfs_item_ptr(eb, swarn->path->slots[0], | |
672 | struct btrfs_inode_item); | |
673 | isize = btrfs_inode_size(eb, inode_item); | |
674 | nlink = btrfs_inode_nlink(eb, inode_item); | |
675 | btrfs_release_path(swarn->path); | |
676 | ||
de2491fd DS |
677 | /* |
678 | * init_path might indirectly call vmalloc, or use GFP_KERNEL. Scrub | |
679 | * uses GFP_NOFS in this context, so we keep it consistent but it does | |
680 | * not seem to be strictly necessary. | |
681 | */ | |
682 | nofs_flag = memalloc_nofs_save(); | |
558540c1 | 683 | ipath = init_ipath(4096, local_root, swarn->path); |
de2491fd | 684 | memalloc_nofs_restore(nofs_flag); |
26bdef54 | 685 | if (IS_ERR(ipath)) { |
00246528 | 686 | btrfs_put_root(local_root); |
26bdef54 DC |
687 | ret = PTR_ERR(ipath); |
688 | ipath = NULL; | |
689 | goto err; | |
690 | } | |
558540c1 JS |
691 | ret = paths_from_inode(inum, ipath); |
692 | ||
693 | if (ret < 0) | |
694 | goto err; | |
695 | ||
696 | /* | |
697 | * we deliberately ignore the bit ipath might have been too small to | |
698 | * hold all of the paths here | |
699 | */ | |
700 | for (i = 0; i < ipath->fspath->elem_cnt; ++i) | |
5d163e0e | 701 | btrfs_warn_in_rcu(fs_info, |
6aa21263 | 702 | "%s at logical %llu on dev %s, physical %llu, root %llu, inode %llu, offset %llu, length %llu, links %u (path: %s)", |
5d163e0e JM |
703 | swarn->errstr, swarn->logical, |
704 | rcu_str_deref(swarn->dev->name), | |
6aa21263 | 705 | swarn->physical, |
5d163e0e JM |
706 | root, inum, offset, |
707 | min(isize - offset, (u64)PAGE_SIZE), nlink, | |
708 | (char *)(unsigned long)ipath->fspath->val[i]); | |
558540c1 | 709 | |
00246528 | 710 | btrfs_put_root(local_root); |
558540c1 JS |
711 | free_ipath(ipath); |
712 | return 0; | |
713 | ||
714 | err: | |
5d163e0e | 715 | btrfs_warn_in_rcu(fs_info, |
6aa21263 | 716 | "%s at logical %llu on dev %s, physical %llu, root %llu, inode %llu, offset %llu: path resolving failed with ret=%d", |
5d163e0e JM |
717 | swarn->errstr, swarn->logical, |
718 | rcu_str_deref(swarn->dev->name), | |
6aa21263 | 719 | swarn->physical, |
5d163e0e | 720 | root, inum, offset, ret); |
558540c1 JS |
721 | |
722 | free_ipath(ipath); | |
723 | return 0; | |
724 | } | |
725 | ||
b5d67f64 | 726 | static void scrub_print_warning(const char *errstr, struct scrub_block *sblock) |
558540c1 | 727 | { |
a36cf8b8 SB |
728 | struct btrfs_device *dev; |
729 | struct btrfs_fs_info *fs_info; | |
558540c1 JS |
730 | struct btrfs_path *path; |
731 | struct btrfs_key found_key; | |
732 | struct extent_buffer *eb; | |
733 | struct btrfs_extent_item *ei; | |
734 | struct scrub_warning swarn; | |
69917e43 LB |
735 | unsigned long ptr = 0; |
736 | u64 extent_item_pos; | |
737 | u64 flags = 0; | |
558540c1 | 738 | u64 ref_root; |
69917e43 | 739 | u32 item_size; |
07c9a8e0 | 740 | u8 ref_level = 0; |
69917e43 | 741 | int ret; |
558540c1 | 742 | |
a36cf8b8 | 743 | WARN_ON(sblock->page_count < 1); |
7a9e9987 | 744 | dev = sblock->pagev[0]->dev; |
fb456252 | 745 | fs_info = sblock->sctx->fs_info; |
a36cf8b8 | 746 | |
558540c1 | 747 | path = btrfs_alloc_path(); |
8b9456da DS |
748 | if (!path) |
749 | return; | |
558540c1 | 750 | |
6aa21263 | 751 | swarn.physical = sblock->pagev[0]->physical; |
7a9e9987 | 752 | swarn.logical = sblock->pagev[0]->logical; |
558540c1 | 753 | swarn.errstr = errstr; |
a36cf8b8 | 754 | swarn.dev = NULL; |
558540c1 | 755 | |
69917e43 LB |
756 | ret = extent_from_logical(fs_info, swarn.logical, path, &found_key, |
757 | &flags); | |
558540c1 JS |
758 | if (ret < 0) |
759 | goto out; | |
760 | ||
4692cf58 | 761 | extent_item_pos = swarn.logical - found_key.objectid; |
558540c1 JS |
762 | swarn.extent_item_size = found_key.offset; |
763 | ||
764 | eb = path->nodes[0]; | |
765 | ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item); | |
766 | item_size = btrfs_item_size_nr(eb, path->slots[0]); | |
767 | ||
69917e43 | 768 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { |
558540c1 | 769 | do { |
6eda71d0 LB |
770 | ret = tree_backref_for_extent(&ptr, eb, &found_key, ei, |
771 | item_size, &ref_root, | |
772 | &ref_level); | |
ecaeb14b | 773 | btrfs_warn_in_rcu(fs_info, |
6aa21263 | 774 | "%s at logical %llu on dev %s, physical %llu: metadata %s (level %d) in tree %llu", |
5d163e0e | 775 | errstr, swarn.logical, |
606686ee | 776 | rcu_str_deref(dev->name), |
6aa21263 | 777 | swarn.physical, |
558540c1 JS |
778 | ref_level ? "node" : "leaf", |
779 | ret < 0 ? -1 : ref_level, | |
780 | ret < 0 ? -1 : ref_root); | |
781 | } while (ret != 1); | |
d8fe29e9 | 782 | btrfs_release_path(path); |
558540c1 | 783 | } else { |
d8fe29e9 | 784 | btrfs_release_path(path); |
558540c1 | 785 | swarn.path = path; |
a36cf8b8 | 786 | swarn.dev = dev; |
7a3ae2f8 JS |
787 | iterate_extent_inodes(fs_info, found_key.objectid, |
788 | extent_item_pos, 1, | |
c995ab3c | 789 | scrub_print_warning_inode, &swarn, false); |
558540c1 JS |
790 | } |
791 | ||
792 | out: | |
793 | btrfs_free_path(path); | |
558540c1 JS |
794 | } |
795 | ||
af8e2d1d MX |
796 | static inline void scrub_get_recover(struct scrub_recover *recover) |
797 | { | |
6f615018 | 798 | refcount_inc(&recover->refs); |
af8e2d1d MX |
799 | } |
800 | ||
e501bfe3 QW |
801 | static inline void scrub_put_recover(struct btrfs_fs_info *fs_info, |
802 | struct scrub_recover *recover) | |
af8e2d1d | 803 | { |
6f615018 | 804 | if (refcount_dec_and_test(&recover->refs)) { |
e501bfe3 | 805 | btrfs_bio_counter_dec(fs_info); |
6e9606d2 | 806 | btrfs_put_bbio(recover->bbio); |
af8e2d1d MX |
807 | kfree(recover); |
808 | } | |
809 | } | |
810 | ||
a2de733c | 811 | /* |
b5d67f64 SB |
812 | * scrub_handle_errored_block gets called when either verification of the |
813 | * pages failed or the bio failed to read, e.g. with EIO. In the latter | |
814 | * case, this function handles all pages in the bio, even though only one | |
815 | * may be bad. | |
816 | * The goal of this function is to repair the errored block by using the | |
817 | * contents of one of the mirrors. | |
a2de733c | 818 | */ |
b5d67f64 | 819 | static int scrub_handle_errored_block(struct scrub_block *sblock_to_check) |
a2de733c | 820 | { |
d9d181c1 | 821 | struct scrub_ctx *sctx = sblock_to_check->sctx; |
a36cf8b8 | 822 | struct btrfs_device *dev; |
b5d67f64 | 823 | struct btrfs_fs_info *fs_info; |
b5d67f64 | 824 | u64 logical; |
b5d67f64 SB |
825 | unsigned int failed_mirror_index; |
826 | unsigned int is_metadata; | |
827 | unsigned int have_csum; | |
b5d67f64 SB |
828 | struct scrub_block *sblocks_for_recheck; /* holds one for each mirror */ |
829 | struct scrub_block *sblock_bad; | |
830 | int ret; | |
831 | int mirror_index; | |
832 | int page_num; | |
833 | int success; | |
28d70e23 | 834 | bool full_stripe_locked; |
7c3c7cb9 | 835 | unsigned int nofs_flag; |
8bb1cf1b | 836 | static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL, |
b5d67f64 SB |
837 | DEFAULT_RATELIMIT_BURST); |
838 | ||
839 | BUG_ON(sblock_to_check->page_count < 1); | |
fb456252 | 840 | fs_info = sctx->fs_info; |
4ded4f63 SB |
841 | if (sblock_to_check->pagev[0]->flags & BTRFS_EXTENT_FLAG_SUPER) { |
842 | /* | |
843 | * if we find an error in a super block, we just report it. | |
844 | * They will get written with the next transaction commit | |
845 | * anyway | |
846 | */ | |
847 | spin_lock(&sctx->stat_lock); | |
848 | ++sctx->stat.super_errors; | |
849 | spin_unlock(&sctx->stat_lock); | |
850 | return 0; | |
851 | } | |
7a9e9987 | 852 | logical = sblock_to_check->pagev[0]->logical; |
7a9e9987 SB |
853 | BUG_ON(sblock_to_check->pagev[0]->mirror_num < 1); |
854 | failed_mirror_index = sblock_to_check->pagev[0]->mirror_num - 1; | |
855 | is_metadata = !(sblock_to_check->pagev[0]->flags & | |
b5d67f64 | 856 | BTRFS_EXTENT_FLAG_DATA); |
7a9e9987 | 857 | have_csum = sblock_to_check->pagev[0]->have_csum; |
7a9e9987 | 858 | dev = sblock_to_check->pagev[0]->dev; |
13db62b7 | 859 | |
f7ef5287 NA |
860 | if (btrfs_is_zoned(fs_info) && !sctx->is_dev_replace) |
861 | return btrfs_repair_one_zone(fs_info, logical); | |
862 | ||
7c3c7cb9 FM |
863 | /* |
864 | * We must use GFP_NOFS because the scrub task might be waiting for a | |
865 | * worker task executing this function and in turn a transaction commit | |
866 | * might be waiting the scrub task to pause (which needs to wait for all | |
867 | * the worker tasks to complete before pausing). | |
868 | * We do allocations in the workers through insert_full_stripe_lock() | |
869 | * and scrub_add_page_to_wr_bio(), which happens down the call chain of | |
870 | * this function. | |
871 | */ | |
872 | nofs_flag = memalloc_nofs_save(); | |
28d70e23 QW |
873 | /* |
874 | * For RAID5/6, race can happen for a different device scrub thread. | |
875 | * For data corruption, Parity and Data threads will both try | |
876 | * to recovery the data. | |
877 | * Race can lead to doubly added csum error, or even unrecoverable | |
878 | * error. | |
879 | */ | |
880 | ret = lock_full_stripe(fs_info, logical, &full_stripe_locked); | |
881 | if (ret < 0) { | |
7c3c7cb9 | 882 | memalloc_nofs_restore(nofs_flag); |
28d70e23 QW |
883 | spin_lock(&sctx->stat_lock); |
884 | if (ret == -ENOMEM) | |
885 | sctx->stat.malloc_errors++; | |
886 | sctx->stat.read_errors++; | |
887 | sctx->stat.uncorrectable_errors++; | |
888 | spin_unlock(&sctx->stat_lock); | |
889 | return ret; | |
890 | } | |
891 | ||
b5d67f64 SB |
892 | /* |
893 | * read all mirrors one after the other. This includes to | |
894 | * re-read the extent or metadata block that failed (that was | |
895 | * the cause that this fixup code is called) another time, | |
896 | * page by page this time in order to know which pages | |
897 | * caused I/O errors and which ones are good (for all mirrors). | |
898 | * It is the goal to handle the situation when more than one | |
899 | * mirror contains I/O errors, but the errors do not | |
900 | * overlap, i.e. the data can be repaired by selecting the | |
901 | * pages from those mirrors without I/O error on the | |
902 | * particular pages. One example (with blocks >= 2 * PAGE_SIZE) | |
903 | * would be that mirror #1 has an I/O error on the first page, | |
904 | * the second page is good, and mirror #2 has an I/O error on | |
905 | * the second page, but the first page is good. | |
906 | * Then the first page of the first mirror can be repaired by | |
907 | * taking the first page of the second mirror, and the | |
908 | * second page of the second mirror can be repaired by | |
909 | * copying the contents of the 2nd page of the 1st mirror. | |
910 | * One more note: if the pages of one mirror contain I/O | |
911 | * errors, the checksum cannot be verified. In order to get | |
912 | * the best data for repairing, the first attempt is to find | |
913 | * a mirror without I/O errors and with a validated checksum. | |
914 | * Only if this is not possible, the pages are picked from | |
915 | * mirrors with I/O errors without considering the checksum. | |
916 | * If the latter is the case, at the end, the checksum of the | |
917 | * repaired area is verified in order to correctly maintain | |
918 | * the statistics. | |
919 | */ | |
920 | ||
31e818fe | 921 | sblocks_for_recheck = kcalloc(BTRFS_MAX_MIRRORS, |
7c3c7cb9 | 922 | sizeof(*sblocks_for_recheck), GFP_KERNEL); |
b5d67f64 | 923 | if (!sblocks_for_recheck) { |
d9d181c1 SB |
924 | spin_lock(&sctx->stat_lock); |
925 | sctx->stat.malloc_errors++; | |
926 | sctx->stat.read_errors++; | |
927 | sctx->stat.uncorrectable_errors++; | |
928 | spin_unlock(&sctx->stat_lock); | |
a36cf8b8 | 929 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS); |
b5d67f64 | 930 | goto out; |
a2de733c AJ |
931 | } |
932 | ||
b5d67f64 | 933 | /* setup the context, map the logical blocks and alloc the pages */ |
be50a8dd | 934 | ret = scrub_setup_recheck_block(sblock_to_check, sblocks_for_recheck); |
b5d67f64 | 935 | if (ret) { |
d9d181c1 SB |
936 | spin_lock(&sctx->stat_lock); |
937 | sctx->stat.read_errors++; | |
938 | sctx->stat.uncorrectable_errors++; | |
939 | spin_unlock(&sctx->stat_lock); | |
a36cf8b8 | 940 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS); |
b5d67f64 SB |
941 | goto out; |
942 | } | |
943 | BUG_ON(failed_mirror_index >= BTRFS_MAX_MIRRORS); | |
944 | sblock_bad = sblocks_for_recheck + failed_mirror_index; | |
13db62b7 | 945 | |
b5d67f64 | 946 | /* build and submit the bios for the failed mirror, check checksums */ |
affe4a5a | 947 | scrub_recheck_block(fs_info, sblock_bad, 1); |
a2de733c | 948 | |
b5d67f64 SB |
949 | if (!sblock_bad->header_error && !sblock_bad->checksum_error && |
950 | sblock_bad->no_io_error_seen) { | |
951 | /* | |
952 | * the error disappeared after reading page by page, or | |
953 | * the area was part of a huge bio and other parts of the | |
954 | * bio caused I/O errors, or the block layer merged several | |
955 | * read requests into one and the error is caused by a | |
956 | * different bio (usually one of the two latter cases is | |
957 | * the cause) | |
958 | */ | |
d9d181c1 SB |
959 | spin_lock(&sctx->stat_lock); |
960 | sctx->stat.unverified_errors++; | |
5a6ac9ea | 961 | sblock_to_check->data_corrected = 1; |
d9d181c1 | 962 | spin_unlock(&sctx->stat_lock); |
a2de733c | 963 | |
ff023aac SB |
964 | if (sctx->is_dev_replace) |
965 | scrub_write_block_to_dev_replace(sblock_bad); | |
b5d67f64 | 966 | goto out; |
a2de733c | 967 | } |
a2de733c | 968 | |
b5d67f64 | 969 | if (!sblock_bad->no_io_error_seen) { |
d9d181c1 SB |
970 | spin_lock(&sctx->stat_lock); |
971 | sctx->stat.read_errors++; | |
972 | spin_unlock(&sctx->stat_lock); | |
8bb1cf1b | 973 | if (__ratelimit(&rs)) |
b5d67f64 | 974 | scrub_print_warning("i/o error", sblock_to_check); |
a36cf8b8 | 975 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS); |
b5d67f64 | 976 | } else if (sblock_bad->checksum_error) { |
d9d181c1 SB |
977 | spin_lock(&sctx->stat_lock); |
978 | sctx->stat.csum_errors++; | |
979 | spin_unlock(&sctx->stat_lock); | |
8bb1cf1b | 980 | if (__ratelimit(&rs)) |
b5d67f64 | 981 | scrub_print_warning("checksum error", sblock_to_check); |
a36cf8b8 | 982 | btrfs_dev_stat_inc_and_print(dev, |
442a4f63 | 983 | BTRFS_DEV_STAT_CORRUPTION_ERRS); |
b5d67f64 | 984 | } else if (sblock_bad->header_error) { |
d9d181c1 SB |
985 | spin_lock(&sctx->stat_lock); |
986 | sctx->stat.verify_errors++; | |
987 | spin_unlock(&sctx->stat_lock); | |
8bb1cf1b | 988 | if (__ratelimit(&rs)) |
b5d67f64 SB |
989 | scrub_print_warning("checksum/header error", |
990 | sblock_to_check); | |
442a4f63 | 991 | if (sblock_bad->generation_error) |
a36cf8b8 | 992 | btrfs_dev_stat_inc_and_print(dev, |
442a4f63 SB |
993 | BTRFS_DEV_STAT_GENERATION_ERRS); |
994 | else | |
a36cf8b8 | 995 | btrfs_dev_stat_inc_and_print(dev, |
442a4f63 | 996 | BTRFS_DEV_STAT_CORRUPTION_ERRS); |
b5d67f64 | 997 | } |
a2de733c | 998 | |
33ef30ad ID |
999 | if (sctx->readonly) { |
1000 | ASSERT(!sctx->is_dev_replace); | |
1001 | goto out; | |
1002 | } | |
a2de733c | 1003 | |
b5d67f64 SB |
1004 | /* |
1005 | * now build and submit the bios for the other mirrors, check | |
cb2ced73 SB |
1006 | * checksums. |
1007 | * First try to pick the mirror which is completely without I/O | |
b5d67f64 SB |
1008 | * errors and also does not have a checksum error. |
1009 | * If one is found, and if a checksum is present, the full block | |
1010 | * that is known to contain an error is rewritten. Afterwards | |
1011 | * the block is known to be corrected. | |
1012 | * If a mirror is found which is completely correct, and no | |
1013 | * checksum is present, only those pages are rewritten that had | |
1014 | * an I/O error in the block to be repaired, since it cannot be | |
1015 | * determined, which copy of the other pages is better (and it | |
1016 | * could happen otherwise that a correct page would be | |
1017 | * overwritten by a bad one). | |
1018 | */ | |
762221f0 | 1019 | for (mirror_index = 0; ;mirror_index++) { |
cb2ced73 | 1020 | struct scrub_block *sblock_other; |
b5d67f64 | 1021 | |
cb2ced73 SB |
1022 | if (mirror_index == failed_mirror_index) |
1023 | continue; | |
762221f0 LB |
1024 | |
1025 | /* raid56's mirror can be more than BTRFS_MAX_MIRRORS */ | |
1026 | if (!scrub_is_page_on_raid56(sblock_bad->pagev[0])) { | |
1027 | if (mirror_index >= BTRFS_MAX_MIRRORS) | |
1028 | break; | |
1029 | if (!sblocks_for_recheck[mirror_index].page_count) | |
1030 | break; | |
1031 | ||
1032 | sblock_other = sblocks_for_recheck + mirror_index; | |
1033 | } else { | |
1034 | struct scrub_recover *r = sblock_bad->pagev[0]->recover; | |
1035 | int max_allowed = r->bbio->num_stripes - | |
1036 | r->bbio->num_tgtdevs; | |
1037 | ||
1038 | if (mirror_index >= max_allowed) | |
1039 | break; | |
1040 | if (!sblocks_for_recheck[1].page_count) | |
1041 | break; | |
1042 | ||
1043 | ASSERT(failed_mirror_index == 0); | |
1044 | sblock_other = sblocks_for_recheck + 1; | |
1045 | sblock_other->pagev[0]->mirror_num = 1 + mirror_index; | |
1046 | } | |
cb2ced73 SB |
1047 | |
1048 | /* build and submit the bios, check checksums */ | |
affe4a5a | 1049 | scrub_recheck_block(fs_info, sblock_other, 0); |
34f5c8e9 SB |
1050 | |
1051 | if (!sblock_other->header_error && | |
b5d67f64 SB |
1052 | !sblock_other->checksum_error && |
1053 | sblock_other->no_io_error_seen) { | |
ff023aac SB |
1054 | if (sctx->is_dev_replace) { |
1055 | scrub_write_block_to_dev_replace(sblock_other); | |
114ab50d | 1056 | goto corrected_error; |
ff023aac | 1057 | } else { |
ff023aac | 1058 | ret = scrub_repair_block_from_good_copy( |
114ab50d ZL |
1059 | sblock_bad, sblock_other); |
1060 | if (!ret) | |
1061 | goto corrected_error; | |
ff023aac | 1062 | } |
b5d67f64 SB |
1063 | } |
1064 | } | |
a2de733c | 1065 | |
b968fed1 ZL |
1066 | if (sblock_bad->no_io_error_seen && !sctx->is_dev_replace) |
1067 | goto did_not_correct_error; | |
ff023aac SB |
1068 | |
1069 | /* | |
ff023aac | 1070 | * In case of I/O errors in the area that is supposed to be |
b5d67f64 SB |
1071 | * repaired, continue by picking good copies of those pages. |
1072 | * Select the good pages from mirrors to rewrite bad pages from | |
1073 | * the area to fix. Afterwards verify the checksum of the block | |
1074 | * that is supposed to be repaired. This verification step is | |
1075 | * only done for the purpose of statistic counting and for the | |
1076 | * final scrub report, whether errors remain. | |
1077 | * A perfect algorithm could make use of the checksum and try | |
1078 | * all possible combinations of pages from the different mirrors | |
1079 | * until the checksum verification succeeds. For example, when | |
1080 | * the 2nd page of mirror #1 faces I/O errors, and the 2nd page | |
1081 | * of mirror #2 is readable but the final checksum test fails, | |
1082 | * then the 2nd page of mirror #3 could be tried, whether now | |
01327610 | 1083 | * the final checksum succeeds. But this would be a rare |
b5d67f64 SB |
1084 | * exception and is therefore not implemented. At least it is |
1085 | * avoided that the good copy is overwritten. | |
1086 | * A more useful improvement would be to pick the sectors | |
1087 | * without I/O error based on sector sizes (512 bytes on legacy | |
1088 | * disks) instead of on PAGE_SIZE. Then maybe 512 byte of one | |
1089 | * mirror could be repaired by taking 512 byte of a different | |
1090 | * mirror, even if other 512 byte sectors in the same PAGE_SIZE | |
1091 | * area are unreadable. | |
a2de733c | 1092 | */ |
b5d67f64 | 1093 | success = 1; |
b968fed1 ZL |
1094 | for (page_num = 0; page_num < sblock_bad->page_count; |
1095 | page_num++) { | |
261d2dcb | 1096 | struct scrub_page *spage_bad = sblock_bad->pagev[page_num]; |
b968fed1 | 1097 | struct scrub_block *sblock_other = NULL; |
b5d67f64 | 1098 | |
b968fed1 | 1099 | /* skip no-io-error page in scrub */ |
261d2dcb | 1100 | if (!spage_bad->io_error && !sctx->is_dev_replace) |
a2de733c | 1101 | continue; |
b5d67f64 | 1102 | |
4759700a LB |
1103 | if (scrub_is_page_on_raid56(sblock_bad->pagev[0])) { |
1104 | /* | |
1105 | * In case of dev replace, if raid56 rebuild process | |
1106 | * didn't work out correct data, then copy the content | |
1107 | * in sblock_bad to make sure target device is identical | |
1108 | * to source device, instead of writing garbage data in | |
1109 | * sblock_for_recheck array to target device. | |
1110 | */ | |
1111 | sblock_other = NULL; | |
261d2dcb | 1112 | } else if (spage_bad->io_error) { |
4759700a | 1113 | /* try to find no-io-error page in mirrors */ |
b968fed1 ZL |
1114 | for (mirror_index = 0; |
1115 | mirror_index < BTRFS_MAX_MIRRORS && | |
1116 | sblocks_for_recheck[mirror_index].page_count > 0; | |
1117 | mirror_index++) { | |
1118 | if (!sblocks_for_recheck[mirror_index]. | |
1119 | pagev[page_num]->io_error) { | |
1120 | sblock_other = sblocks_for_recheck + | |
1121 | mirror_index; | |
1122 | break; | |
b5d67f64 SB |
1123 | } |
1124 | } | |
b968fed1 ZL |
1125 | if (!sblock_other) |
1126 | success = 0; | |
96e36920 | 1127 | } |
a2de733c | 1128 | |
b968fed1 ZL |
1129 | if (sctx->is_dev_replace) { |
1130 | /* | |
1131 | * did not find a mirror to fetch the page | |
1132 | * from. scrub_write_page_to_dev_replace() | |
1133 | * handles this case (page->io_error), by | |
1134 | * filling the block with zeros before | |
1135 | * submitting the write request | |
1136 | */ | |
1137 | if (!sblock_other) | |
1138 | sblock_other = sblock_bad; | |
1139 | ||
1140 | if (scrub_write_page_to_dev_replace(sblock_other, | |
1141 | page_num) != 0) { | |
e37abe97 | 1142 | atomic64_inc( |
0b246afa | 1143 | &fs_info->dev_replace.num_write_errors); |
b968fed1 ZL |
1144 | success = 0; |
1145 | } | |
1146 | } else if (sblock_other) { | |
1147 | ret = scrub_repair_page_from_good_copy(sblock_bad, | |
1148 | sblock_other, | |
1149 | page_num, 0); | |
1150 | if (0 == ret) | |
261d2dcb | 1151 | spage_bad->io_error = 0; |
b968fed1 ZL |
1152 | else |
1153 | success = 0; | |
b5d67f64 | 1154 | } |
a2de733c | 1155 | } |
a2de733c | 1156 | |
b968fed1 | 1157 | if (success && !sctx->is_dev_replace) { |
b5d67f64 SB |
1158 | if (is_metadata || have_csum) { |
1159 | /* | |
1160 | * need to verify the checksum now that all | |
1161 | * sectors on disk are repaired (the write | |
1162 | * request for data to be repaired is on its way). | |
1163 | * Just be lazy and use scrub_recheck_block() | |
1164 | * which re-reads the data before the checksum | |
1165 | * is verified, but most likely the data comes out | |
1166 | * of the page cache. | |
1167 | */ | |
affe4a5a | 1168 | scrub_recheck_block(fs_info, sblock_bad, 1); |
34f5c8e9 | 1169 | if (!sblock_bad->header_error && |
b5d67f64 SB |
1170 | !sblock_bad->checksum_error && |
1171 | sblock_bad->no_io_error_seen) | |
1172 | goto corrected_error; | |
1173 | else | |
1174 | goto did_not_correct_error; | |
1175 | } else { | |
1176 | corrected_error: | |
d9d181c1 SB |
1177 | spin_lock(&sctx->stat_lock); |
1178 | sctx->stat.corrected_errors++; | |
5a6ac9ea | 1179 | sblock_to_check->data_corrected = 1; |
d9d181c1 | 1180 | spin_unlock(&sctx->stat_lock); |
b14af3b4 DS |
1181 | btrfs_err_rl_in_rcu(fs_info, |
1182 | "fixed up error at logical %llu on dev %s", | |
c1c9ff7c | 1183 | logical, rcu_str_deref(dev->name)); |
8628764e | 1184 | } |
b5d67f64 SB |
1185 | } else { |
1186 | did_not_correct_error: | |
d9d181c1 SB |
1187 | spin_lock(&sctx->stat_lock); |
1188 | sctx->stat.uncorrectable_errors++; | |
1189 | spin_unlock(&sctx->stat_lock); | |
b14af3b4 DS |
1190 | btrfs_err_rl_in_rcu(fs_info, |
1191 | "unable to fixup (regular) error at logical %llu on dev %s", | |
c1c9ff7c | 1192 | logical, rcu_str_deref(dev->name)); |
96e36920 | 1193 | } |
a2de733c | 1194 | |
b5d67f64 SB |
1195 | out: |
1196 | if (sblocks_for_recheck) { | |
1197 | for (mirror_index = 0; mirror_index < BTRFS_MAX_MIRRORS; | |
1198 | mirror_index++) { | |
1199 | struct scrub_block *sblock = sblocks_for_recheck + | |
1200 | mirror_index; | |
af8e2d1d | 1201 | struct scrub_recover *recover; |
b5d67f64 SB |
1202 | int page_index; |
1203 | ||
7a9e9987 SB |
1204 | for (page_index = 0; page_index < sblock->page_count; |
1205 | page_index++) { | |
1206 | sblock->pagev[page_index]->sblock = NULL; | |
af8e2d1d MX |
1207 | recover = sblock->pagev[page_index]->recover; |
1208 | if (recover) { | |
e501bfe3 | 1209 | scrub_put_recover(fs_info, recover); |
af8e2d1d MX |
1210 | sblock->pagev[page_index]->recover = |
1211 | NULL; | |
1212 | } | |
7a9e9987 SB |
1213 | scrub_page_put(sblock->pagev[page_index]); |
1214 | } | |
b5d67f64 SB |
1215 | } |
1216 | kfree(sblocks_for_recheck); | |
1217 | } | |
a2de733c | 1218 | |
28d70e23 | 1219 | ret = unlock_full_stripe(fs_info, logical, full_stripe_locked); |
7c3c7cb9 | 1220 | memalloc_nofs_restore(nofs_flag); |
28d70e23 QW |
1221 | if (ret < 0) |
1222 | return ret; | |
b5d67f64 SB |
1223 | return 0; |
1224 | } | |
a2de733c | 1225 | |
8e5cfb55 | 1226 | static inline int scrub_nr_raid_mirrors(struct btrfs_bio *bbio) |
af8e2d1d | 1227 | { |
10f11900 ZL |
1228 | if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID5) |
1229 | return 2; | |
1230 | else if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID6) | |
1231 | return 3; | |
1232 | else | |
af8e2d1d | 1233 | return (int)bbio->num_stripes; |
af8e2d1d MX |
1234 | } |
1235 | ||
10f11900 ZL |
1236 | static inline void scrub_stripe_index_and_offset(u64 logical, u64 map_type, |
1237 | u64 *raid_map, | |
af8e2d1d MX |
1238 | u64 mapped_length, |
1239 | int nstripes, int mirror, | |
1240 | int *stripe_index, | |
1241 | u64 *stripe_offset) | |
1242 | { | |
1243 | int i; | |
1244 | ||
ffe2d203 | 1245 | if (map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
af8e2d1d MX |
1246 | /* RAID5/6 */ |
1247 | for (i = 0; i < nstripes; i++) { | |
1248 | if (raid_map[i] == RAID6_Q_STRIPE || | |
1249 | raid_map[i] == RAID5_P_STRIPE) | |
1250 | continue; | |
1251 | ||
1252 | if (logical >= raid_map[i] && | |
1253 | logical < raid_map[i] + mapped_length) | |
1254 | break; | |
1255 | } | |
1256 | ||
1257 | *stripe_index = i; | |
1258 | *stripe_offset = logical - raid_map[i]; | |
1259 | } else { | |
1260 | /* The other RAID type */ | |
1261 | *stripe_index = mirror; | |
1262 | *stripe_offset = 0; | |
1263 | } | |
1264 | } | |
1265 | ||
be50a8dd | 1266 | static int scrub_setup_recheck_block(struct scrub_block *original_sblock, |
b5d67f64 SB |
1267 | struct scrub_block *sblocks_for_recheck) |
1268 | { | |
be50a8dd | 1269 | struct scrub_ctx *sctx = original_sblock->sctx; |
fb456252 | 1270 | struct btrfs_fs_info *fs_info = sctx->fs_info; |
be50a8dd ZL |
1271 | u64 length = original_sblock->page_count * PAGE_SIZE; |
1272 | u64 logical = original_sblock->pagev[0]->logical; | |
4734b7ed ZL |
1273 | u64 generation = original_sblock->pagev[0]->generation; |
1274 | u64 flags = original_sblock->pagev[0]->flags; | |
1275 | u64 have_csum = original_sblock->pagev[0]->have_csum; | |
af8e2d1d MX |
1276 | struct scrub_recover *recover; |
1277 | struct btrfs_bio *bbio; | |
af8e2d1d MX |
1278 | u64 sublen; |
1279 | u64 mapped_length; | |
1280 | u64 stripe_offset; | |
1281 | int stripe_index; | |
be50a8dd | 1282 | int page_index = 0; |
b5d67f64 | 1283 | int mirror_index; |
af8e2d1d | 1284 | int nmirrors; |
b5d67f64 SB |
1285 | int ret; |
1286 | ||
1287 | /* | |
57019345 | 1288 | * note: the two members refs and outstanding_pages |
b5d67f64 SB |
1289 | * are not used (and not set) in the blocks that are used for |
1290 | * the recheck procedure | |
1291 | */ | |
1292 | ||
b5d67f64 | 1293 | while (length > 0) { |
af8e2d1d MX |
1294 | sublen = min_t(u64, length, PAGE_SIZE); |
1295 | mapped_length = sublen; | |
1296 | bbio = NULL; | |
a2de733c | 1297 | |
b5d67f64 SB |
1298 | /* |
1299 | * with a length of PAGE_SIZE, each returned stripe | |
1300 | * represents one mirror | |
1301 | */ | |
e501bfe3 | 1302 | btrfs_bio_counter_inc_blocked(fs_info); |
cf8cddd3 | 1303 | ret = btrfs_map_sblock(fs_info, BTRFS_MAP_GET_READ_MIRRORS, |
825ad4c9 | 1304 | logical, &mapped_length, &bbio); |
b5d67f64 | 1305 | if (ret || !bbio || mapped_length < sublen) { |
6e9606d2 | 1306 | btrfs_put_bbio(bbio); |
e501bfe3 | 1307 | btrfs_bio_counter_dec(fs_info); |
b5d67f64 SB |
1308 | return -EIO; |
1309 | } | |
a2de733c | 1310 | |
af8e2d1d MX |
1311 | recover = kzalloc(sizeof(struct scrub_recover), GFP_NOFS); |
1312 | if (!recover) { | |
6e9606d2 | 1313 | btrfs_put_bbio(bbio); |
e501bfe3 | 1314 | btrfs_bio_counter_dec(fs_info); |
af8e2d1d MX |
1315 | return -ENOMEM; |
1316 | } | |
1317 | ||
6f615018 | 1318 | refcount_set(&recover->refs, 1); |
af8e2d1d | 1319 | recover->bbio = bbio; |
af8e2d1d MX |
1320 | recover->map_length = mapped_length; |
1321 | ||
24731149 | 1322 | BUG_ON(page_index >= SCRUB_MAX_PAGES_PER_BLOCK); |
af8e2d1d | 1323 | |
be50a8dd | 1324 | nmirrors = min(scrub_nr_raid_mirrors(bbio), BTRFS_MAX_MIRRORS); |
10f11900 | 1325 | |
af8e2d1d | 1326 | for (mirror_index = 0; mirror_index < nmirrors; |
b5d67f64 SB |
1327 | mirror_index++) { |
1328 | struct scrub_block *sblock; | |
261d2dcb | 1329 | struct scrub_page *spage; |
b5d67f64 | 1330 | |
b5d67f64 | 1331 | sblock = sblocks_for_recheck + mirror_index; |
7a9e9987 | 1332 | sblock->sctx = sctx; |
4734b7ed | 1333 | |
261d2dcb QW |
1334 | spage = kzalloc(sizeof(*spage), GFP_NOFS); |
1335 | if (!spage) { | |
7a9e9987 | 1336 | leave_nomem: |
d9d181c1 SB |
1337 | spin_lock(&sctx->stat_lock); |
1338 | sctx->stat.malloc_errors++; | |
1339 | spin_unlock(&sctx->stat_lock); | |
e501bfe3 | 1340 | scrub_put_recover(fs_info, recover); |
b5d67f64 SB |
1341 | return -ENOMEM; |
1342 | } | |
261d2dcb QW |
1343 | scrub_page_get(spage); |
1344 | sblock->pagev[page_index] = spage; | |
1345 | spage->sblock = sblock; | |
1346 | spage->flags = flags; | |
1347 | spage->generation = generation; | |
1348 | spage->logical = logical; | |
1349 | spage->have_csum = have_csum; | |
4734b7ed | 1350 | if (have_csum) |
261d2dcb | 1351 | memcpy(spage->csum, |
4734b7ed | 1352 | original_sblock->pagev[0]->csum, |
2ae0c2d8 | 1353 | sctx->fs_info->csum_size); |
af8e2d1d | 1354 | |
10f11900 ZL |
1355 | scrub_stripe_index_and_offset(logical, |
1356 | bbio->map_type, | |
1357 | bbio->raid_map, | |
af8e2d1d | 1358 | mapped_length, |
e34c330d ZL |
1359 | bbio->num_stripes - |
1360 | bbio->num_tgtdevs, | |
af8e2d1d MX |
1361 | mirror_index, |
1362 | &stripe_index, | |
1363 | &stripe_offset); | |
261d2dcb | 1364 | spage->physical = bbio->stripes[stripe_index].physical + |
af8e2d1d | 1365 | stripe_offset; |
261d2dcb | 1366 | spage->dev = bbio->stripes[stripe_index].dev; |
af8e2d1d | 1367 | |
ff023aac | 1368 | BUG_ON(page_index >= original_sblock->page_count); |
261d2dcb | 1369 | spage->physical_for_dev_replace = |
ff023aac SB |
1370 | original_sblock->pagev[page_index]-> |
1371 | physical_for_dev_replace; | |
7a9e9987 | 1372 | /* for missing devices, dev->bdev is NULL */ |
261d2dcb | 1373 | spage->mirror_num = mirror_index + 1; |
b5d67f64 | 1374 | sblock->page_count++; |
261d2dcb QW |
1375 | spage->page = alloc_page(GFP_NOFS); |
1376 | if (!spage->page) | |
7a9e9987 | 1377 | goto leave_nomem; |
af8e2d1d MX |
1378 | |
1379 | scrub_get_recover(recover); | |
261d2dcb | 1380 | spage->recover = recover; |
b5d67f64 | 1381 | } |
e501bfe3 | 1382 | scrub_put_recover(fs_info, recover); |
b5d67f64 SB |
1383 | length -= sublen; |
1384 | logical += sublen; | |
1385 | page_index++; | |
1386 | } | |
1387 | ||
1388 | return 0; | |
96e36920 ID |
1389 | } |
1390 | ||
4246a0b6 | 1391 | static void scrub_bio_wait_endio(struct bio *bio) |
af8e2d1d | 1392 | { |
b4ff5ad7 | 1393 | complete(bio->bi_private); |
af8e2d1d MX |
1394 | } |
1395 | ||
af8e2d1d MX |
1396 | static int scrub_submit_raid56_bio_wait(struct btrfs_fs_info *fs_info, |
1397 | struct bio *bio, | |
261d2dcb | 1398 | struct scrub_page *spage) |
af8e2d1d | 1399 | { |
b4ff5ad7 | 1400 | DECLARE_COMPLETION_ONSTACK(done); |
af8e2d1d | 1401 | int ret; |
762221f0 | 1402 | int mirror_num; |
af8e2d1d | 1403 | |
261d2dcb | 1404 | bio->bi_iter.bi_sector = spage->logical >> 9; |
af8e2d1d MX |
1405 | bio->bi_private = &done; |
1406 | bio->bi_end_io = scrub_bio_wait_endio; | |
1407 | ||
261d2dcb QW |
1408 | mirror_num = spage->sblock->pagev[0]->mirror_num; |
1409 | ret = raid56_parity_recover(fs_info, bio, spage->recover->bbio, | |
1410 | spage->recover->map_length, | |
762221f0 | 1411 | mirror_num, 0); |
af8e2d1d MX |
1412 | if (ret) |
1413 | return ret; | |
1414 | ||
b4ff5ad7 LB |
1415 | wait_for_completion_io(&done); |
1416 | return blk_status_to_errno(bio->bi_status); | |
af8e2d1d MX |
1417 | } |
1418 | ||
6ca1765b LB |
1419 | static void scrub_recheck_block_on_raid56(struct btrfs_fs_info *fs_info, |
1420 | struct scrub_block *sblock) | |
1421 | { | |
1422 | struct scrub_page *first_page = sblock->pagev[0]; | |
1423 | struct bio *bio; | |
1424 | int page_num; | |
1425 | ||
1426 | /* All pages in sblock belong to the same stripe on the same device. */ | |
1427 | ASSERT(first_page->dev); | |
1428 | if (!first_page->dev->bdev) | |
1429 | goto out; | |
1430 | ||
1431 | bio = btrfs_io_bio_alloc(BIO_MAX_PAGES); | |
1432 | bio_set_dev(bio, first_page->dev->bdev); | |
1433 | ||
1434 | for (page_num = 0; page_num < sblock->page_count; page_num++) { | |
261d2dcb | 1435 | struct scrub_page *spage = sblock->pagev[page_num]; |
6ca1765b | 1436 | |
261d2dcb QW |
1437 | WARN_ON(!spage->page); |
1438 | bio_add_page(bio, spage->page, PAGE_SIZE, 0); | |
6ca1765b LB |
1439 | } |
1440 | ||
1441 | if (scrub_submit_raid56_bio_wait(fs_info, bio, first_page)) { | |
1442 | bio_put(bio); | |
1443 | goto out; | |
1444 | } | |
1445 | ||
1446 | bio_put(bio); | |
1447 | ||
1448 | scrub_recheck_block_checksum(sblock); | |
1449 | ||
1450 | return; | |
1451 | out: | |
1452 | for (page_num = 0; page_num < sblock->page_count; page_num++) | |
1453 | sblock->pagev[page_num]->io_error = 1; | |
1454 | ||
1455 | sblock->no_io_error_seen = 0; | |
1456 | } | |
1457 | ||
b5d67f64 SB |
1458 | /* |
1459 | * this function will check the on disk data for checksum errors, header | |
1460 | * errors and read I/O errors. If any I/O errors happen, the exact pages | |
1461 | * which are errored are marked as being bad. The goal is to enable scrub | |
1462 | * to take those pages that are not errored from all the mirrors so that | |
1463 | * the pages that are errored in the just handled mirror can be repaired. | |
1464 | */ | |
34f5c8e9 | 1465 | static void scrub_recheck_block(struct btrfs_fs_info *fs_info, |
affe4a5a ZL |
1466 | struct scrub_block *sblock, |
1467 | int retry_failed_mirror) | |
96e36920 | 1468 | { |
b5d67f64 | 1469 | int page_num; |
96e36920 | 1470 | |
b5d67f64 | 1471 | sblock->no_io_error_seen = 1; |
96e36920 | 1472 | |
6ca1765b LB |
1473 | /* short cut for raid56 */ |
1474 | if (!retry_failed_mirror && scrub_is_page_on_raid56(sblock->pagev[0])) | |
1475 | return scrub_recheck_block_on_raid56(fs_info, sblock); | |
1476 | ||
b5d67f64 SB |
1477 | for (page_num = 0; page_num < sblock->page_count; page_num++) { |
1478 | struct bio *bio; | |
261d2dcb | 1479 | struct scrub_page *spage = sblock->pagev[page_num]; |
b5d67f64 | 1480 | |
261d2dcb QW |
1481 | if (spage->dev->bdev == NULL) { |
1482 | spage->io_error = 1; | |
ea9947b4 SB |
1483 | sblock->no_io_error_seen = 0; |
1484 | continue; | |
1485 | } | |
1486 | ||
261d2dcb | 1487 | WARN_ON(!spage->page); |
c5e4c3d7 | 1488 | bio = btrfs_io_bio_alloc(1); |
261d2dcb | 1489 | bio_set_dev(bio, spage->dev->bdev); |
b5d67f64 | 1490 | |
261d2dcb QW |
1491 | bio_add_page(bio, spage->page, PAGE_SIZE, 0); |
1492 | bio->bi_iter.bi_sector = spage->physical >> 9; | |
6ca1765b | 1493 | bio->bi_opf = REQ_OP_READ; |
af8e2d1d | 1494 | |
6ca1765b | 1495 | if (btrfsic_submit_bio_wait(bio)) { |
261d2dcb | 1496 | spage->io_error = 1; |
6ca1765b | 1497 | sblock->no_io_error_seen = 0; |
af8e2d1d | 1498 | } |
33879d45 | 1499 | |
b5d67f64 SB |
1500 | bio_put(bio); |
1501 | } | |
96e36920 | 1502 | |
b5d67f64 | 1503 | if (sblock->no_io_error_seen) |
ba7cf988 | 1504 | scrub_recheck_block_checksum(sblock); |
a2de733c AJ |
1505 | } |
1506 | ||
17a9be2f MX |
1507 | static inline int scrub_check_fsid(u8 fsid[], |
1508 | struct scrub_page *spage) | |
1509 | { | |
1510 | struct btrfs_fs_devices *fs_devices = spage->dev->fs_devices; | |
1511 | int ret; | |
1512 | ||
44880fdc | 1513 | ret = memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
17a9be2f MX |
1514 | return !ret; |
1515 | } | |
1516 | ||
ba7cf988 | 1517 | static void scrub_recheck_block_checksum(struct scrub_block *sblock) |
a2de733c | 1518 | { |
ba7cf988 ZL |
1519 | sblock->header_error = 0; |
1520 | sblock->checksum_error = 0; | |
1521 | sblock->generation_error = 0; | |
b5d67f64 | 1522 | |
ba7cf988 ZL |
1523 | if (sblock->pagev[0]->flags & BTRFS_EXTENT_FLAG_DATA) |
1524 | scrub_checksum_data(sblock); | |
1525 | else | |
1526 | scrub_checksum_tree_block(sblock); | |
a2de733c AJ |
1527 | } |
1528 | ||
b5d67f64 | 1529 | static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad, |
114ab50d | 1530 | struct scrub_block *sblock_good) |
b5d67f64 SB |
1531 | { |
1532 | int page_num; | |
1533 | int ret = 0; | |
96e36920 | 1534 | |
b5d67f64 SB |
1535 | for (page_num = 0; page_num < sblock_bad->page_count; page_num++) { |
1536 | int ret_sub; | |
96e36920 | 1537 | |
b5d67f64 SB |
1538 | ret_sub = scrub_repair_page_from_good_copy(sblock_bad, |
1539 | sblock_good, | |
114ab50d | 1540 | page_num, 1); |
b5d67f64 SB |
1541 | if (ret_sub) |
1542 | ret = ret_sub; | |
a2de733c | 1543 | } |
b5d67f64 SB |
1544 | |
1545 | return ret; | |
1546 | } | |
1547 | ||
1548 | static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad, | |
1549 | struct scrub_block *sblock_good, | |
1550 | int page_num, int force_write) | |
1551 | { | |
261d2dcb QW |
1552 | struct scrub_page *spage_bad = sblock_bad->pagev[page_num]; |
1553 | struct scrub_page *spage_good = sblock_good->pagev[page_num]; | |
0b246afa | 1554 | struct btrfs_fs_info *fs_info = sblock_bad->sctx->fs_info; |
b5d67f64 | 1555 | |
261d2dcb QW |
1556 | BUG_ON(spage_bad->page == NULL); |
1557 | BUG_ON(spage_good->page == NULL); | |
b5d67f64 | 1558 | if (force_write || sblock_bad->header_error || |
261d2dcb | 1559 | sblock_bad->checksum_error || spage_bad->io_error) { |
b5d67f64 SB |
1560 | struct bio *bio; |
1561 | int ret; | |
b5d67f64 | 1562 | |
261d2dcb | 1563 | if (!spage_bad->dev->bdev) { |
0b246afa | 1564 | btrfs_warn_rl(fs_info, |
5d163e0e | 1565 | "scrub_repair_page_from_good_copy(bdev == NULL) is unexpected"); |
ff023aac SB |
1566 | return -EIO; |
1567 | } | |
1568 | ||
c5e4c3d7 | 1569 | bio = btrfs_io_bio_alloc(1); |
261d2dcb QW |
1570 | bio_set_dev(bio, spage_bad->dev->bdev); |
1571 | bio->bi_iter.bi_sector = spage_bad->physical >> 9; | |
ebcc3263 | 1572 | bio->bi_opf = REQ_OP_WRITE; |
b5d67f64 | 1573 | |
261d2dcb | 1574 | ret = bio_add_page(bio, spage_good->page, PAGE_SIZE, 0); |
b5d67f64 SB |
1575 | if (PAGE_SIZE != ret) { |
1576 | bio_put(bio); | |
1577 | return -EIO; | |
13db62b7 | 1578 | } |
b5d67f64 | 1579 | |
4e49ea4a | 1580 | if (btrfsic_submit_bio_wait(bio)) { |
261d2dcb | 1581 | btrfs_dev_stat_inc_and_print(spage_bad->dev, |
442a4f63 | 1582 | BTRFS_DEV_STAT_WRITE_ERRS); |
e37abe97 | 1583 | atomic64_inc(&fs_info->dev_replace.num_write_errors); |
442a4f63 SB |
1584 | bio_put(bio); |
1585 | return -EIO; | |
1586 | } | |
b5d67f64 | 1587 | bio_put(bio); |
a2de733c AJ |
1588 | } |
1589 | ||
b5d67f64 SB |
1590 | return 0; |
1591 | } | |
1592 | ||
ff023aac SB |
1593 | static void scrub_write_block_to_dev_replace(struct scrub_block *sblock) |
1594 | { | |
0b246afa | 1595 | struct btrfs_fs_info *fs_info = sblock->sctx->fs_info; |
ff023aac SB |
1596 | int page_num; |
1597 | ||
5a6ac9ea MX |
1598 | /* |
1599 | * This block is used for the check of the parity on the source device, | |
1600 | * so the data needn't be written into the destination device. | |
1601 | */ | |
1602 | if (sblock->sparity) | |
1603 | return; | |
1604 | ||
ff023aac SB |
1605 | for (page_num = 0; page_num < sblock->page_count; page_num++) { |
1606 | int ret; | |
1607 | ||
1608 | ret = scrub_write_page_to_dev_replace(sblock, page_num); | |
1609 | if (ret) | |
e37abe97 | 1610 | atomic64_inc(&fs_info->dev_replace.num_write_errors); |
ff023aac SB |
1611 | } |
1612 | } | |
1613 | ||
1614 | static int scrub_write_page_to_dev_replace(struct scrub_block *sblock, | |
1615 | int page_num) | |
1616 | { | |
1617 | struct scrub_page *spage = sblock->pagev[page_num]; | |
1618 | ||
1619 | BUG_ON(spage->page == NULL); | |
a8b3a890 DS |
1620 | if (spage->io_error) |
1621 | clear_page(page_address(spage->page)); | |
ff023aac | 1622 | |
ff023aac SB |
1623 | return scrub_add_page_to_wr_bio(sblock->sctx, spage); |
1624 | } | |
1625 | ||
de17addc NA |
1626 | static int fill_writer_pointer_gap(struct scrub_ctx *sctx, u64 physical) |
1627 | { | |
1628 | int ret = 0; | |
1629 | u64 length; | |
1630 | ||
1631 | if (!btrfs_is_zoned(sctx->fs_info)) | |
1632 | return 0; | |
1633 | ||
7db1c5d1 NA |
1634 | if (!btrfs_dev_is_sequential(sctx->wr_tgtdev, physical)) |
1635 | return 0; | |
1636 | ||
de17addc NA |
1637 | if (sctx->write_pointer < physical) { |
1638 | length = physical - sctx->write_pointer; | |
1639 | ||
1640 | ret = btrfs_zoned_issue_zeroout(sctx->wr_tgtdev, | |
1641 | sctx->write_pointer, length); | |
1642 | if (!ret) | |
1643 | sctx->write_pointer = physical; | |
1644 | } | |
1645 | return ret; | |
1646 | } | |
1647 | ||
ff023aac SB |
1648 | static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx, |
1649 | struct scrub_page *spage) | |
1650 | { | |
ff023aac SB |
1651 | struct scrub_bio *sbio; |
1652 | int ret; | |
1653 | ||
3fb99303 | 1654 | mutex_lock(&sctx->wr_lock); |
ff023aac | 1655 | again: |
3fb99303 DS |
1656 | if (!sctx->wr_curr_bio) { |
1657 | sctx->wr_curr_bio = kzalloc(sizeof(*sctx->wr_curr_bio), | |
58c4e173 | 1658 | GFP_KERNEL); |
3fb99303 DS |
1659 | if (!sctx->wr_curr_bio) { |
1660 | mutex_unlock(&sctx->wr_lock); | |
ff023aac SB |
1661 | return -ENOMEM; |
1662 | } | |
3fb99303 DS |
1663 | sctx->wr_curr_bio->sctx = sctx; |
1664 | sctx->wr_curr_bio->page_count = 0; | |
ff023aac | 1665 | } |
3fb99303 | 1666 | sbio = sctx->wr_curr_bio; |
ff023aac SB |
1667 | if (sbio->page_count == 0) { |
1668 | struct bio *bio; | |
1669 | ||
de17addc NA |
1670 | ret = fill_writer_pointer_gap(sctx, |
1671 | spage->physical_for_dev_replace); | |
1672 | if (ret) { | |
1673 | mutex_unlock(&sctx->wr_lock); | |
1674 | return ret; | |
1675 | } | |
1676 | ||
ff023aac SB |
1677 | sbio->physical = spage->physical_for_dev_replace; |
1678 | sbio->logical = spage->logical; | |
3fb99303 | 1679 | sbio->dev = sctx->wr_tgtdev; |
ff023aac SB |
1680 | bio = sbio->bio; |
1681 | if (!bio) { | |
c5e4c3d7 | 1682 | bio = btrfs_io_bio_alloc(sctx->pages_per_wr_bio); |
ff023aac SB |
1683 | sbio->bio = bio; |
1684 | } | |
1685 | ||
1686 | bio->bi_private = sbio; | |
1687 | bio->bi_end_io = scrub_wr_bio_end_io; | |
74d46992 | 1688 | bio_set_dev(bio, sbio->dev->bdev); |
4f024f37 | 1689 | bio->bi_iter.bi_sector = sbio->physical >> 9; |
ebcc3263 | 1690 | bio->bi_opf = REQ_OP_WRITE; |
4e4cbee9 | 1691 | sbio->status = 0; |
ff023aac SB |
1692 | } else if (sbio->physical + sbio->page_count * PAGE_SIZE != |
1693 | spage->physical_for_dev_replace || | |
1694 | sbio->logical + sbio->page_count * PAGE_SIZE != | |
1695 | spage->logical) { | |
1696 | scrub_wr_submit(sctx); | |
1697 | goto again; | |
1698 | } | |
1699 | ||
1700 | ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0); | |
1701 | if (ret != PAGE_SIZE) { | |
1702 | if (sbio->page_count < 1) { | |
1703 | bio_put(sbio->bio); | |
1704 | sbio->bio = NULL; | |
3fb99303 | 1705 | mutex_unlock(&sctx->wr_lock); |
ff023aac SB |
1706 | return -EIO; |
1707 | } | |
1708 | scrub_wr_submit(sctx); | |
1709 | goto again; | |
1710 | } | |
1711 | ||
1712 | sbio->pagev[sbio->page_count] = spage; | |
1713 | scrub_page_get(spage); | |
1714 | sbio->page_count++; | |
3fb99303 | 1715 | if (sbio->page_count == sctx->pages_per_wr_bio) |
ff023aac | 1716 | scrub_wr_submit(sctx); |
3fb99303 | 1717 | mutex_unlock(&sctx->wr_lock); |
ff023aac SB |
1718 | |
1719 | return 0; | |
1720 | } | |
1721 | ||
1722 | static void scrub_wr_submit(struct scrub_ctx *sctx) | |
1723 | { | |
ff023aac SB |
1724 | struct scrub_bio *sbio; |
1725 | ||
3fb99303 | 1726 | if (!sctx->wr_curr_bio) |
ff023aac SB |
1727 | return; |
1728 | ||
3fb99303 DS |
1729 | sbio = sctx->wr_curr_bio; |
1730 | sctx->wr_curr_bio = NULL; | |
74d46992 | 1731 | WARN_ON(!sbio->bio->bi_disk); |
ff023aac SB |
1732 | scrub_pending_bio_inc(sctx); |
1733 | /* process all writes in a single worker thread. Then the block layer | |
1734 | * orders the requests before sending them to the driver which | |
1735 | * doubled the write performance on spinning disks when measured | |
1736 | * with Linux 3.5 */ | |
4e49ea4a | 1737 | btrfsic_submit_bio(sbio->bio); |
de17addc NA |
1738 | |
1739 | if (btrfs_is_zoned(sctx->fs_info)) | |
1740 | sctx->write_pointer = sbio->physical + sbio->page_count * PAGE_SIZE; | |
ff023aac SB |
1741 | } |
1742 | ||
4246a0b6 | 1743 | static void scrub_wr_bio_end_io(struct bio *bio) |
ff023aac SB |
1744 | { |
1745 | struct scrub_bio *sbio = bio->bi_private; | |
fb456252 | 1746 | struct btrfs_fs_info *fs_info = sbio->dev->fs_info; |
ff023aac | 1747 | |
4e4cbee9 | 1748 | sbio->status = bio->bi_status; |
ff023aac SB |
1749 | sbio->bio = bio; |
1750 | ||
a0cac0ec | 1751 | btrfs_init_work(&sbio->work, scrub_wr_bio_end_io_worker, NULL, NULL); |
0339ef2f | 1752 | btrfs_queue_work(fs_info->scrub_wr_completion_workers, &sbio->work); |
ff023aac SB |
1753 | } |
1754 | ||
1755 | static void scrub_wr_bio_end_io_worker(struct btrfs_work *work) | |
1756 | { | |
1757 | struct scrub_bio *sbio = container_of(work, struct scrub_bio, work); | |
1758 | struct scrub_ctx *sctx = sbio->sctx; | |
1759 | int i; | |
1760 | ||
1761 | WARN_ON(sbio->page_count > SCRUB_PAGES_PER_WR_BIO); | |
4e4cbee9 | 1762 | if (sbio->status) { |
ff023aac | 1763 | struct btrfs_dev_replace *dev_replace = |
fb456252 | 1764 | &sbio->sctx->fs_info->dev_replace; |
ff023aac SB |
1765 | |
1766 | for (i = 0; i < sbio->page_count; i++) { | |
1767 | struct scrub_page *spage = sbio->pagev[i]; | |
1768 | ||
1769 | spage->io_error = 1; | |
e37abe97 | 1770 | atomic64_inc(&dev_replace->num_write_errors); |
ff023aac SB |
1771 | } |
1772 | } | |
1773 | ||
1774 | for (i = 0; i < sbio->page_count; i++) | |
1775 | scrub_page_put(sbio->pagev[i]); | |
1776 | ||
1777 | bio_put(sbio->bio); | |
1778 | kfree(sbio); | |
1779 | scrub_pending_bio_dec(sctx); | |
1780 | } | |
1781 | ||
1782 | static int scrub_checksum(struct scrub_block *sblock) | |
b5d67f64 SB |
1783 | { |
1784 | u64 flags; | |
1785 | int ret; | |
1786 | ||
ba7cf988 ZL |
1787 | /* |
1788 | * No need to initialize these stats currently, | |
1789 | * because this function only use return value | |
1790 | * instead of these stats value. | |
1791 | * | |
1792 | * Todo: | |
1793 | * always use stats | |
1794 | */ | |
1795 | sblock->header_error = 0; | |
1796 | sblock->generation_error = 0; | |
1797 | sblock->checksum_error = 0; | |
1798 | ||
7a9e9987 SB |
1799 | WARN_ON(sblock->page_count < 1); |
1800 | flags = sblock->pagev[0]->flags; | |
b5d67f64 SB |
1801 | ret = 0; |
1802 | if (flags & BTRFS_EXTENT_FLAG_DATA) | |
1803 | ret = scrub_checksum_data(sblock); | |
1804 | else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) | |
1805 | ret = scrub_checksum_tree_block(sblock); | |
1806 | else if (flags & BTRFS_EXTENT_FLAG_SUPER) | |
1807 | (void)scrub_checksum_super(sblock); | |
1808 | else | |
1809 | WARN_ON(1); | |
1810 | if (ret) | |
1811 | scrub_handle_errored_block(sblock); | |
ff023aac SB |
1812 | |
1813 | return ret; | |
a2de733c AJ |
1814 | } |
1815 | ||
b5d67f64 | 1816 | static int scrub_checksum_data(struct scrub_block *sblock) |
a2de733c | 1817 | { |
d9d181c1 | 1818 | struct scrub_ctx *sctx = sblock->sctx; |
d5178578 JT |
1819 | struct btrfs_fs_info *fs_info = sctx->fs_info; |
1820 | SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); | |
a2de733c | 1821 | u8 csum[BTRFS_CSUM_SIZE]; |
d41ebef2 | 1822 | struct scrub_page *spage; |
b0485252 | 1823 | char *kaddr; |
a2de733c | 1824 | |
b5d67f64 | 1825 | BUG_ON(sblock->page_count < 1); |
d41ebef2 DS |
1826 | spage = sblock->pagev[0]; |
1827 | if (!spage->have_csum) | |
a2de733c AJ |
1828 | return 0; |
1829 | ||
d41ebef2 | 1830 | kaddr = page_address(spage->page); |
b5d67f64 | 1831 | |
771aba0d DS |
1832 | shash->tfm = fs_info->csum_shash; |
1833 | crypto_shash_init(shash); | |
b5d67f64 | 1834 | |
b29dca44 QW |
1835 | /* |
1836 | * In scrub_pages() and scrub_pages_for_parity() we ensure each spage | |
1837 | * only contains one sector of data. | |
1838 | */ | |
1839 | crypto_shash_digest(shash, kaddr, fs_info->sectorsize, csum); | |
a2de733c | 1840 | |
b29dca44 QW |
1841 | if (memcmp(csum, spage->csum, fs_info->csum_size)) |
1842 | sblock->checksum_error = 1; | |
ba7cf988 | 1843 | return sblock->checksum_error; |
a2de733c AJ |
1844 | } |
1845 | ||
b5d67f64 | 1846 | static int scrub_checksum_tree_block(struct scrub_block *sblock) |
a2de733c | 1847 | { |
d9d181c1 | 1848 | struct scrub_ctx *sctx = sblock->sctx; |
a2de733c | 1849 | struct btrfs_header *h; |
0b246afa | 1850 | struct btrfs_fs_info *fs_info = sctx->fs_info; |
d5178578 | 1851 | SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); |
b5d67f64 SB |
1852 | u8 calculated_csum[BTRFS_CSUM_SIZE]; |
1853 | u8 on_disk_csum[BTRFS_CSUM_SIZE]; | |
53f3251d QW |
1854 | /* |
1855 | * This is done in sectorsize steps even for metadata as there's a | |
1856 | * constraint for nodesize to be aligned to sectorsize. This will need | |
1857 | * to change so we don't misuse data and metadata units like that. | |
1858 | */ | |
1859 | const u32 sectorsize = sctx->fs_info->sectorsize; | |
1860 | const int num_sectors = fs_info->nodesize >> fs_info->sectorsize_bits; | |
521e1022 | 1861 | int i; |
100aa5d9 | 1862 | struct scrub_page *spage; |
b0485252 | 1863 | char *kaddr; |
d5178578 | 1864 | |
b5d67f64 | 1865 | BUG_ON(sblock->page_count < 1); |
53f3251d QW |
1866 | |
1867 | /* Each member in pagev is just one block, not a full page */ | |
1868 | ASSERT(sblock->page_count == num_sectors); | |
1869 | ||
100aa5d9 DS |
1870 | spage = sblock->pagev[0]; |
1871 | kaddr = page_address(spage->page); | |
b0485252 | 1872 | h = (struct btrfs_header *)kaddr; |
2ae0c2d8 | 1873 | memcpy(on_disk_csum, h->csum, sctx->fs_info->csum_size); |
a2de733c AJ |
1874 | |
1875 | /* | |
1876 | * we don't use the getter functions here, as we | |
1877 | * a) don't have an extent buffer and | |
1878 | * b) the page is already kmapped | |
1879 | */ | |
100aa5d9 | 1880 | if (spage->logical != btrfs_stack_header_bytenr(h)) |
ba7cf988 | 1881 | sblock->header_error = 1; |
a2de733c | 1882 | |
100aa5d9 | 1883 | if (spage->generation != btrfs_stack_header_generation(h)) { |
ba7cf988 ZL |
1884 | sblock->header_error = 1; |
1885 | sblock->generation_error = 1; | |
1886 | } | |
a2de733c | 1887 | |
100aa5d9 | 1888 | if (!scrub_check_fsid(h->fsid, spage)) |
ba7cf988 | 1889 | sblock->header_error = 1; |
a2de733c AJ |
1890 | |
1891 | if (memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid, | |
1892 | BTRFS_UUID_SIZE)) | |
ba7cf988 | 1893 | sblock->header_error = 1; |
a2de733c | 1894 | |
521e1022 DS |
1895 | shash->tfm = fs_info->csum_shash; |
1896 | crypto_shash_init(shash); | |
1897 | crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE, | |
53f3251d | 1898 | sectorsize - BTRFS_CSUM_SIZE); |
b5d67f64 | 1899 | |
53f3251d | 1900 | for (i = 1; i < num_sectors; i++) { |
521e1022 | 1901 | kaddr = page_address(sblock->pagev[i]->page); |
53f3251d | 1902 | crypto_shash_update(shash, kaddr, sectorsize); |
b5d67f64 SB |
1903 | } |
1904 | ||
d5178578 | 1905 | crypto_shash_final(shash, calculated_csum); |
2ae0c2d8 | 1906 | if (memcmp(calculated_csum, on_disk_csum, sctx->fs_info->csum_size)) |
ba7cf988 | 1907 | sblock->checksum_error = 1; |
a2de733c | 1908 | |
ba7cf988 | 1909 | return sblock->header_error || sblock->checksum_error; |
a2de733c AJ |
1910 | } |
1911 | ||
b5d67f64 | 1912 | static int scrub_checksum_super(struct scrub_block *sblock) |
a2de733c AJ |
1913 | { |
1914 | struct btrfs_super_block *s; | |
d9d181c1 | 1915 | struct scrub_ctx *sctx = sblock->sctx; |
d5178578 JT |
1916 | struct btrfs_fs_info *fs_info = sctx->fs_info; |
1917 | SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); | |
b5d67f64 | 1918 | u8 calculated_csum[BTRFS_CSUM_SIZE]; |
c7460541 | 1919 | struct scrub_page *spage; |
b0485252 | 1920 | char *kaddr; |
442a4f63 SB |
1921 | int fail_gen = 0; |
1922 | int fail_cor = 0; | |
d5178578 | 1923 | |
b5d67f64 | 1924 | BUG_ON(sblock->page_count < 1); |
c7460541 DS |
1925 | spage = sblock->pagev[0]; |
1926 | kaddr = page_address(spage->page); | |
b0485252 | 1927 | s = (struct btrfs_super_block *)kaddr; |
a2de733c | 1928 | |
c7460541 | 1929 | if (spage->logical != btrfs_super_bytenr(s)) |
442a4f63 | 1930 | ++fail_cor; |
a2de733c | 1931 | |
c7460541 | 1932 | if (spage->generation != btrfs_super_generation(s)) |
442a4f63 | 1933 | ++fail_gen; |
a2de733c | 1934 | |
c7460541 | 1935 | if (!scrub_check_fsid(s->fsid, spage)) |
442a4f63 | 1936 | ++fail_cor; |
a2de733c | 1937 | |
83cf6d5e DS |
1938 | shash->tfm = fs_info->csum_shash; |
1939 | crypto_shash_init(shash); | |
1940 | crypto_shash_digest(shash, kaddr + BTRFS_CSUM_SIZE, | |
1941 | BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, calculated_csum); | |
b5d67f64 | 1942 | |
2ae0c2d8 | 1943 | if (memcmp(calculated_csum, s->csum, sctx->fs_info->csum_size)) |
442a4f63 | 1944 | ++fail_cor; |
a2de733c | 1945 | |
442a4f63 | 1946 | if (fail_cor + fail_gen) { |
a2de733c AJ |
1947 | /* |
1948 | * if we find an error in a super block, we just report it. | |
1949 | * They will get written with the next transaction commit | |
1950 | * anyway | |
1951 | */ | |
d9d181c1 SB |
1952 | spin_lock(&sctx->stat_lock); |
1953 | ++sctx->stat.super_errors; | |
1954 | spin_unlock(&sctx->stat_lock); | |
442a4f63 | 1955 | if (fail_cor) |
c7460541 | 1956 | btrfs_dev_stat_inc_and_print(spage->dev, |
442a4f63 SB |
1957 | BTRFS_DEV_STAT_CORRUPTION_ERRS); |
1958 | else | |
c7460541 | 1959 | btrfs_dev_stat_inc_and_print(spage->dev, |
442a4f63 | 1960 | BTRFS_DEV_STAT_GENERATION_ERRS); |
a2de733c AJ |
1961 | } |
1962 | ||
442a4f63 | 1963 | return fail_cor + fail_gen; |
a2de733c AJ |
1964 | } |
1965 | ||
b5d67f64 SB |
1966 | static void scrub_block_get(struct scrub_block *sblock) |
1967 | { | |
186debd6 | 1968 | refcount_inc(&sblock->refs); |
b5d67f64 SB |
1969 | } |
1970 | ||
1971 | static void scrub_block_put(struct scrub_block *sblock) | |
1972 | { | |
186debd6 | 1973 | if (refcount_dec_and_test(&sblock->refs)) { |
b5d67f64 SB |
1974 | int i; |
1975 | ||
5a6ac9ea MX |
1976 | if (sblock->sparity) |
1977 | scrub_parity_put(sblock->sparity); | |
1978 | ||
b5d67f64 | 1979 | for (i = 0; i < sblock->page_count; i++) |
7a9e9987 | 1980 | scrub_page_put(sblock->pagev[i]); |
b5d67f64 SB |
1981 | kfree(sblock); |
1982 | } | |
1983 | } | |
1984 | ||
7a9e9987 SB |
1985 | static void scrub_page_get(struct scrub_page *spage) |
1986 | { | |
57019345 | 1987 | atomic_inc(&spage->refs); |
7a9e9987 SB |
1988 | } |
1989 | ||
1990 | static void scrub_page_put(struct scrub_page *spage) | |
1991 | { | |
57019345 | 1992 | if (atomic_dec_and_test(&spage->refs)) { |
7a9e9987 SB |
1993 | if (spage->page) |
1994 | __free_page(spage->page); | |
1995 | kfree(spage); | |
1996 | } | |
1997 | } | |
1998 | ||
d9d181c1 | 1999 | static void scrub_submit(struct scrub_ctx *sctx) |
a2de733c AJ |
2000 | { |
2001 | struct scrub_bio *sbio; | |
2002 | ||
d9d181c1 | 2003 | if (sctx->curr == -1) |
1623edeb | 2004 | return; |
a2de733c | 2005 | |
d9d181c1 SB |
2006 | sbio = sctx->bios[sctx->curr]; |
2007 | sctx->curr = -1; | |
b6bfebc1 | 2008 | scrub_pending_bio_inc(sctx); |
4e49ea4a | 2009 | btrfsic_submit_bio(sbio->bio); |
a2de733c AJ |
2010 | } |
2011 | ||
ff023aac SB |
2012 | static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx, |
2013 | struct scrub_page *spage) | |
a2de733c | 2014 | { |
b5d67f64 | 2015 | struct scrub_block *sblock = spage->sblock; |
a2de733c | 2016 | struct scrub_bio *sbio; |
69f4cb52 | 2017 | int ret; |
a2de733c AJ |
2018 | |
2019 | again: | |
2020 | /* | |
2021 | * grab a fresh bio or wait for one to become available | |
2022 | */ | |
d9d181c1 SB |
2023 | while (sctx->curr == -1) { |
2024 | spin_lock(&sctx->list_lock); | |
2025 | sctx->curr = sctx->first_free; | |
2026 | if (sctx->curr != -1) { | |
2027 | sctx->first_free = sctx->bios[sctx->curr]->next_free; | |
2028 | sctx->bios[sctx->curr]->next_free = -1; | |
2029 | sctx->bios[sctx->curr]->page_count = 0; | |
2030 | spin_unlock(&sctx->list_lock); | |
a2de733c | 2031 | } else { |
d9d181c1 SB |
2032 | spin_unlock(&sctx->list_lock); |
2033 | wait_event(sctx->list_wait, sctx->first_free != -1); | |
a2de733c AJ |
2034 | } |
2035 | } | |
d9d181c1 | 2036 | sbio = sctx->bios[sctx->curr]; |
b5d67f64 | 2037 | if (sbio->page_count == 0) { |
69f4cb52 AJ |
2038 | struct bio *bio; |
2039 | ||
b5d67f64 SB |
2040 | sbio->physical = spage->physical; |
2041 | sbio->logical = spage->logical; | |
a36cf8b8 | 2042 | sbio->dev = spage->dev; |
b5d67f64 SB |
2043 | bio = sbio->bio; |
2044 | if (!bio) { | |
c5e4c3d7 | 2045 | bio = btrfs_io_bio_alloc(sctx->pages_per_rd_bio); |
b5d67f64 SB |
2046 | sbio->bio = bio; |
2047 | } | |
69f4cb52 AJ |
2048 | |
2049 | bio->bi_private = sbio; | |
2050 | bio->bi_end_io = scrub_bio_end_io; | |
74d46992 | 2051 | bio_set_dev(bio, sbio->dev->bdev); |
4f024f37 | 2052 | bio->bi_iter.bi_sector = sbio->physical >> 9; |
ebcc3263 | 2053 | bio->bi_opf = REQ_OP_READ; |
4e4cbee9 | 2054 | sbio->status = 0; |
b5d67f64 SB |
2055 | } else if (sbio->physical + sbio->page_count * PAGE_SIZE != |
2056 | spage->physical || | |
2057 | sbio->logical + sbio->page_count * PAGE_SIZE != | |
a36cf8b8 SB |
2058 | spage->logical || |
2059 | sbio->dev != spage->dev) { | |
d9d181c1 | 2060 | scrub_submit(sctx); |
a2de733c AJ |
2061 | goto again; |
2062 | } | |
69f4cb52 | 2063 | |
b5d67f64 SB |
2064 | sbio->pagev[sbio->page_count] = spage; |
2065 | ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0); | |
2066 | if (ret != PAGE_SIZE) { | |
2067 | if (sbio->page_count < 1) { | |
2068 | bio_put(sbio->bio); | |
2069 | sbio->bio = NULL; | |
2070 | return -EIO; | |
2071 | } | |
d9d181c1 | 2072 | scrub_submit(sctx); |
69f4cb52 AJ |
2073 | goto again; |
2074 | } | |
2075 | ||
ff023aac | 2076 | scrub_block_get(sblock); /* one for the page added to the bio */ |
b5d67f64 SB |
2077 | atomic_inc(&sblock->outstanding_pages); |
2078 | sbio->page_count++; | |
ff023aac | 2079 | if (sbio->page_count == sctx->pages_per_rd_bio) |
d9d181c1 | 2080 | scrub_submit(sctx); |
b5d67f64 SB |
2081 | |
2082 | return 0; | |
2083 | } | |
2084 | ||
22365979 | 2085 | static void scrub_missing_raid56_end_io(struct bio *bio) |
73ff61db OS |
2086 | { |
2087 | struct scrub_block *sblock = bio->bi_private; | |
fb456252 | 2088 | struct btrfs_fs_info *fs_info = sblock->sctx->fs_info; |
73ff61db | 2089 | |
4e4cbee9 | 2090 | if (bio->bi_status) |
73ff61db OS |
2091 | sblock->no_io_error_seen = 0; |
2092 | ||
4673272f ST |
2093 | bio_put(bio); |
2094 | ||
73ff61db OS |
2095 | btrfs_queue_work(fs_info->scrub_workers, &sblock->work); |
2096 | } | |
2097 | ||
2098 | static void scrub_missing_raid56_worker(struct btrfs_work *work) | |
2099 | { | |
2100 | struct scrub_block *sblock = container_of(work, struct scrub_block, work); | |
2101 | struct scrub_ctx *sctx = sblock->sctx; | |
0b246afa | 2102 | struct btrfs_fs_info *fs_info = sctx->fs_info; |
73ff61db OS |
2103 | u64 logical; |
2104 | struct btrfs_device *dev; | |
2105 | ||
73ff61db OS |
2106 | logical = sblock->pagev[0]->logical; |
2107 | dev = sblock->pagev[0]->dev; | |
2108 | ||
affe4a5a | 2109 | if (sblock->no_io_error_seen) |
ba7cf988 | 2110 | scrub_recheck_block_checksum(sblock); |
73ff61db OS |
2111 | |
2112 | if (!sblock->no_io_error_seen) { | |
2113 | spin_lock(&sctx->stat_lock); | |
2114 | sctx->stat.read_errors++; | |
2115 | spin_unlock(&sctx->stat_lock); | |
0b246afa | 2116 | btrfs_err_rl_in_rcu(fs_info, |
b14af3b4 | 2117 | "IO error rebuilding logical %llu for dev %s", |
73ff61db OS |
2118 | logical, rcu_str_deref(dev->name)); |
2119 | } else if (sblock->header_error || sblock->checksum_error) { | |
2120 | spin_lock(&sctx->stat_lock); | |
2121 | sctx->stat.uncorrectable_errors++; | |
2122 | spin_unlock(&sctx->stat_lock); | |
0b246afa | 2123 | btrfs_err_rl_in_rcu(fs_info, |
b14af3b4 | 2124 | "failed to rebuild valid logical %llu for dev %s", |
73ff61db OS |
2125 | logical, rcu_str_deref(dev->name)); |
2126 | } else { | |
2127 | scrub_write_block_to_dev_replace(sblock); | |
2128 | } | |
2129 | ||
2073c4c2 | 2130 | if (sctx->is_dev_replace && sctx->flush_all_writes) { |
3fb99303 | 2131 | mutex_lock(&sctx->wr_lock); |
73ff61db | 2132 | scrub_wr_submit(sctx); |
3fb99303 | 2133 | mutex_unlock(&sctx->wr_lock); |
73ff61db OS |
2134 | } |
2135 | ||
57d4f0b8 | 2136 | scrub_block_put(sblock); |
73ff61db OS |
2137 | scrub_pending_bio_dec(sctx); |
2138 | } | |
2139 | ||
2140 | static void scrub_missing_raid56_pages(struct scrub_block *sblock) | |
2141 | { | |
2142 | struct scrub_ctx *sctx = sblock->sctx; | |
fb456252 | 2143 | struct btrfs_fs_info *fs_info = sctx->fs_info; |
73ff61db OS |
2144 | u64 length = sblock->page_count * PAGE_SIZE; |
2145 | u64 logical = sblock->pagev[0]->logical; | |
f1fee653 | 2146 | struct btrfs_bio *bbio = NULL; |
73ff61db OS |
2147 | struct bio *bio; |
2148 | struct btrfs_raid_bio *rbio; | |
2149 | int ret; | |
2150 | int i; | |
2151 | ||
ae6529c3 | 2152 | btrfs_bio_counter_inc_blocked(fs_info); |
cf8cddd3 | 2153 | ret = btrfs_map_sblock(fs_info, BTRFS_MAP_GET_READ_MIRRORS, logical, |
825ad4c9 | 2154 | &length, &bbio); |
73ff61db OS |
2155 | if (ret || !bbio || !bbio->raid_map) |
2156 | goto bbio_out; | |
2157 | ||
2158 | if (WARN_ON(!sctx->is_dev_replace || | |
2159 | !(bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK))) { | |
2160 | /* | |
2161 | * We shouldn't be scrubbing a missing device. Even for dev | |
2162 | * replace, we should only get here for RAID 5/6. We either | |
2163 | * managed to mount something with no mirrors remaining or | |
2164 | * there's a bug in scrub_remap_extent()/btrfs_map_block(). | |
2165 | */ | |
2166 | goto bbio_out; | |
2167 | } | |
2168 | ||
c5e4c3d7 | 2169 | bio = btrfs_io_bio_alloc(0); |
73ff61db OS |
2170 | bio->bi_iter.bi_sector = logical >> 9; |
2171 | bio->bi_private = sblock; | |
2172 | bio->bi_end_io = scrub_missing_raid56_end_io; | |
2173 | ||
2ff7e61e | 2174 | rbio = raid56_alloc_missing_rbio(fs_info, bio, bbio, length); |
73ff61db OS |
2175 | if (!rbio) |
2176 | goto rbio_out; | |
2177 | ||
2178 | for (i = 0; i < sblock->page_count; i++) { | |
2179 | struct scrub_page *spage = sblock->pagev[i]; | |
2180 | ||
2181 | raid56_add_scrub_pages(rbio, spage->page, spage->logical); | |
2182 | } | |
2183 | ||
a0cac0ec | 2184 | btrfs_init_work(&sblock->work, scrub_missing_raid56_worker, NULL, NULL); |
73ff61db OS |
2185 | scrub_block_get(sblock); |
2186 | scrub_pending_bio_inc(sctx); | |
2187 | raid56_submit_missing_rbio(rbio); | |
2188 | return; | |
2189 | ||
2190 | rbio_out: | |
2191 | bio_put(bio); | |
2192 | bbio_out: | |
ae6529c3 | 2193 | btrfs_bio_counter_dec(fs_info); |
73ff61db OS |
2194 | btrfs_put_bbio(bbio); |
2195 | spin_lock(&sctx->stat_lock); | |
2196 | sctx->stat.malloc_errors++; | |
2197 | spin_unlock(&sctx->stat_lock); | |
2198 | } | |
2199 | ||
fa485d21 | 2200 | static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u32 len, |
a36cf8b8 | 2201 | u64 physical, struct btrfs_device *dev, u64 flags, |
96e63a45 | 2202 | u64 gen, int mirror_num, u8 *csum, |
ff023aac | 2203 | u64 physical_for_dev_replace) |
b5d67f64 SB |
2204 | { |
2205 | struct scrub_block *sblock; | |
d0a7a9c0 | 2206 | const u32 sectorsize = sctx->fs_info->sectorsize; |
b5d67f64 SB |
2207 | int index; |
2208 | ||
58c4e173 | 2209 | sblock = kzalloc(sizeof(*sblock), GFP_KERNEL); |
b5d67f64 | 2210 | if (!sblock) { |
d9d181c1 SB |
2211 | spin_lock(&sctx->stat_lock); |
2212 | sctx->stat.malloc_errors++; | |
2213 | spin_unlock(&sctx->stat_lock); | |
b5d67f64 | 2214 | return -ENOMEM; |
a2de733c | 2215 | } |
b5d67f64 | 2216 | |
7a9e9987 SB |
2217 | /* one ref inside this function, plus one for each page added to |
2218 | * a bio later on */ | |
186debd6 | 2219 | refcount_set(&sblock->refs, 1); |
d9d181c1 | 2220 | sblock->sctx = sctx; |
b5d67f64 SB |
2221 | sblock->no_io_error_seen = 1; |
2222 | ||
2223 | for (index = 0; len > 0; index++) { | |
7a9e9987 | 2224 | struct scrub_page *spage; |
d0a7a9c0 QW |
2225 | /* |
2226 | * Here we will allocate one page for one sector to scrub. | |
2227 | * This is fine if PAGE_SIZE == sectorsize, but will cost | |
2228 | * more memory for PAGE_SIZE > sectorsize case. | |
2229 | */ | |
2230 | u32 l = min(sectorsize, len); | |
b5d67f64 | 2231 | |
58c4e173 | 2232 | spage = kzalloc(sizeof(*spage), GFP_KERNEL); |
7a9e9987 SB |
2233 | if (!spage) { |
2234 | leave_nomem: | |
d9d181c1 SB |
2235 | spin_lock(&sctx->stat_lock); |
2236 | sctx->stat.malloc_errors++; | |
2237 | spin_unlock(&sctx->stat_lock); | |
7a9e9987 | 2238 | scrub_block_put(sblock); |
b5d67f64 SB |
2239 | return -ENOMEM; |
2240 | } | |
7a9e9987 SB |
2241 | BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK); |
2242 | scrub_page_get(spage); | |
2243 | sblock->pagev[index] = spage; | |
b5d67f64 | 2244 | spage->sblock = sblock; |
a36cf8b8 | 2245 | spage->dev = dev; |
b5d67f64 SB |
2246 | spage->flags = flags; |
2247 | spage->generation = gen; | |
2248 | spage->logical = logical; | |
2249 | spage->physical = physical; | |
ff023aac | 2250 | spage->physical_for_dev_replace = physical_for_dev_replace; |
b5d67f64 SB |
2251 | spage->mirror_num = mirror_num; |
2252 | if (csum) { | |
2253 | spage->have_csum = 1; | |
2ae0c2d8 | 2254 | memcpy(spage->csum, csum, sctx->fs_info->csum_size); |
b5d67f64 SB |
2255 | } else { |
2256 | spage->have_csum = 0; | |
2257 | } | |
2258 | sblock->page_count++; | |
58c4e173 | 2259 | spage->page = alloc_page(GFP_KERNEL); |
7a9e9987 SB |
2260 | if (!spage->page) |
2261 | goto leave_nomem; | |
b5d67f64 SB |
2262 | len -= l; |
2263 | logical += l; | |
2264 | physical += l; | |
ff023aac | 2265 | physical_for_dev_replace += l; |
b5d67f64 SB |
2266 | } |
2267 | ||
7a9e9987 | 2268 | WARN_ON(sblock->page_count == 0); |
e6e674bd | 2269 | if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) { |
73ff61db OS |
2270 | /* |
2271 | * This case should only be hit for RAID 5/6 device replace. See | |
2272 | * the comment in scrub_missing_raid56_pages() for details. | |
2273 | */ | |
2274 | scrub_missing_raid56_pages(sblock); | |
2275 | } else { | |
2276 | for (index = 0; index < sblock->page_count; index++) { | |
2277 | struct scrub_page *spage = sblock->pagev[index]; | |
2278 | int ret; | |
1bc87793 | 2279 | |
73ff61db OS |
2280 | ret = scrub_add_page_to_rd_bio(sctx, spage); |
2281 | if (ret) { | |
2282 | scrub_block_put(sblock); | |
2283 | return ret; | |
2284 | } | |
b5d67f64 | 2285 | } |
a2de733c | 2286 | |
96e63a45 | 2287 | if (flags & BTRFS_EXTENT_FLAG_SUPER) |
73ff61db OS |
2288 | scrub_submit(sctx); |
2289 | } | |
a2de733c | 2290 | |
b5d67f64 SB |
2291 | /* last one frees, either here or in bio completion for last page */ |
2292 | scrub_block_put(sblock); | |
a2de733c AJ |
2293 | return 0; |
2294 | } | |
2295 | ||
4246a0b6 | 2296 | static void scrub_bio_end_io(struct bio *bio) |
b5d67f64 SB |
2297 | { |
2298 | struct scrub_bio *sbio = bio->bi_private; | |
fb456252 | 2299 | struct btrfs_fs_info *fs_info = sbio->dev->fs_info; |
b5d67f64 | 2300 | |
4e4cbee9 | 2301 | sbio->status = bio->bi_status; |
b5d67f64 SB |
2302 | sbio->bio = bio; |
2303 | ||
0339ef2f | 2304 | btrfs_queue_work(fs_info->scrub_workers, &sbio->work); |
b5d67f64 SB |
2305 | } |
2306 | ||
2307 | static void scrub_bio_end_io_worker(struct btrfs_work *work) | |
2308 | { | |
2309 | struct scrub_bio *sbio = container_of(work, struct scrub_bio, work); | |
d9d181c1 | 2310 | struct scrub_ctx *sctx = sbio->sctx; |
b5d67f64 SB |
2311 | int i; |
2312 | ||
ff023aac | 2313 | BUG_ON(sbio->page_count > SCRUB_PAGES_PER_RD_BIO); |
4e4cbee9 | 2314 | if (sbio->status) { |
b5d67f64 SB |
2315 | for (i = 0; i < sbio->page_count; i++) { |
2316 | struct scrub_page *spage = sbio->pagev[i]; | |
2317 | ||
2318 | spage->io_error = 1; | |
2319 | spage->sblock->no_io_error_seen = 0; | |
2320 | } | |
2321 | } | |
2322 | ||
2323 | /* now complete the scrub_block items that have all pages completed */ | |
2324 | for (i = 0; i < sbio->page_count; i++) { | |
2325 | struct scrub_page *spage = sbio->pagev[i]; | |
2326 | struct scrub_block *sblock = spage->sblock; | |
2327 | ||
2328 | if (atomic_dec_and_test(&sblock->outstanding_pages)) | |
2329 | scrub_block_complete(sblock); | |
2330 | scrub_block_put(sblock); | |
2331 | } | |
2332 | ||
b5d67f64 SB |
2333 | bio_put(sbio->bio); |
2334 | sbio->bio = NULL; | |
d9d181c1 SB |
2335 | spin_lock(&sctx->list_lock); |
2336 | sbio->next_free = sctx->first_free; | |
2337 | sctx->first_free = sbio->index; | |
2338 | spin_unlock(&sctx->list_lock); | |
ff023aac | 2339 | |
2073c4c2 | 2340 | if (sctx->is_dev_replace && sctx->flush_all_writes) { |
3fb99303 | 2341 | mutex_lock(&sctx->wr_lock); |
ff023aac | 2342 | scrub_wr_submit(sctx); |
3fb99303 | 2343 | mutex_unlock(&sctx->wr_lock); |
ff023aac SB |
2344 | } |
2345 | ||
b6bfebc1 | 2346 | scrub_pending_bio_dec(sctx); |
b5d67f64 SB |
2347 | } |
2348 | ||
5a6ac9ea MX |
2349 | static inline void __scrub_mark_bitmap(struct scrub_parity *sparity, |
2350 | unsigned long *bitmap, | |
fa485d21 | 2351 | u64 start, u32 len) |
5a6ac9ea | 2352 | { |
972d7219 | 2353 | u64 offset; |
7736b0a4 | 2354 | u32 nsectors; |
ab108d99 | 2355 | u32 sectorsize_bits = sparity->sctx->fs_info->sectorsize_bits; |
5a6ac9ea MX |
2356 | |
2357 | if (len >= sparity->stripe_len) { | |
2358 | bitmap_set(bitmap, 0, sparity->nsectors); | |
2359 | return; | |
2360 | } | |
2361 | ||
2362 | start -= sparity->logic_start; | |
972d7219 | 2363 | start = div64_u64_rem(start, sparity->stripe_len, &offset); |
ab108d99 | 2364 | offset = offset >> sectorsize_bits; |
fa485d21 | 2365 | nsectors = len >> sectorsize_bits; |
5a6ac9ea MX |
2366 | |
2367 | if (offset + nsectors <= sparity->nsectors) { | |
2368 | bitmap_set(bitmap, offset, nsectors); | |
2369 | return; | |
2370 | } | |
2371 | ||
2372 | bitmap_set(bitmap, offset, sparity->nsectors - offset); | |
2373 | bitmap_set(bitmap, 0, nsectors - (sparity->nsectors - offset)); | |
2374 | } | |
2375 | ||
2376 | static inline void scrub_parity_mark_sectors_error(struct scrub_parity *sparity, | |
fa485d21 | 2377 | u64 start, u32 len) |
5a6ac9ea MX |
2378 | { |
2379 | __scrub_mark_bitmap(sparity, sparity->ebitmap, start, len); | |
2380 | } | |
2381 | ||
2382 | static inline void scrub_parity_mark_sectors_data(struct scrub_parity *sparity, | |
fa485d21 | 2383 | u64 start, u32 len) |
5a6ac9ea MX |
2384 | { |
2385 | __scrub_mark_bitmap(sparity, sparity->dbitmap, start, len); | |
2386 | } | |
2387 | ||
b5d67f64 SB |
2388 | static void scrub_block_complete(struct scrub_block *sblock) |
2389 | { | |
5a6ac9ea MX |
2390 | int corrupted = 0; |
2391 | ||
ff023aac | 2392 | if (!sblock->no_io_error_seen) { |
5a6ac9ea | 2393 | corrupted = 1; |
b5d67f64 | 2394 | scrub_handle_errored_block(sblock); |
ff023aac SB |
2395 | } else { |
2396 | /* | |
2397 | * if has checksum error, write via repair mechanism in | |
2398 | * dev replace case, otherwise write here in dev replace | |
2399 | * case. | |
2400 | */ | |
5a6ac9ea MX |
2401 | corrupted = scrub_checksum(sblock); |
2402 | if (!corrupted && sblock->sctx->is_dev_replace) | |
ff023aac SB |
2403 | scrub_write_block_to_dev_replace(sblock); |
2404 | } | |
5a6ac9ea MX |
2405 | |
2406 | if (sblock->sparity && corrupted && !sblock->data_corrected) { | |
2407 | u64 start = sblock->pagev[0]->logical; | |
2408 | u64 end = sblock->pagev[sblock->page_count - 1]->logical + | |
2409 | PAGE_SIZE; | |
2410 | ||
fa485d21 | 2411 | ASSERT(end - start <= U32_MAX); |
5a6ac9ea MX |
2412 | scrub_parity_mark_sectors_error(sblock->sparity, |
2413 | start, end - start); | |
2414 | } | |
b5d67f64 SB |
2415 | } |
2416 | ||
480a8ec8 QW |
2417 | static void drop_csum_range(struct scrub_ctx *sctx, struct btrfs_ordered_sum *sum) |
2418 | { | |
2419 | sctx->stat.csum_discards += sum->len >> sctx->fs_info->sectorsize_bits; | |
2420 | list_del(&sum->list); | |
2421 | kfree(sum); | |
2422 | } | |
2423 | ||
2424 | /* | |
2425 | * Find the desired csum for range [logical, logical + sectorsize), and store | |
2426 | * the csum into @csum. | |
2427 | * | |
2428 | * The search source is sctx->csum_list, which is a pre-populated list | |
2429 | * storing bytenr ordered csum ranges. We're reponsible to cleanup any range | |
2430 | * that is before @logical. | |
2431 | * | |
2432 | * Return 0 if there is no csum for the range. | |
2433 | * Return 1 if there is csum for the range and copied to @csum. | |
2434 | */ | |
3b5753ec | 2435 | static int scrub_find_csum(struct scrub_ctx *sctx, u64 logical, u8 *csum) |
a2de733c | 2436 | { |
480a8ec8 | 2437 | bool found = false; |
a2de733c | 2438 | |
d9d181c1 | 2439 | while (!list_empty(&sctx->csum_list)) { |
480a8ec8 QW |
2440 | struct btrfs_ordered_sum *sum = NULL; |
2441 | unsigned long index; | |
2442 | unsigned long num_sectors; | |
2443 | ||
d9d181c1 | 2444 | sum = list_first_entry(&sctx->csum_list, |
a2de733c | 2445 | struct btrfs_ordered_sum, list); |
480a8ec8 | 2446 | /* The current csum range is beyond our range, no csum found */ |
a2de733c | 2447 | if (sum->bytenr > logical) |
a2de733c AJ |
2448 | break; |
2449 | ||
480a8ec8 QW |
2450 | /* |
2451 | * The current sum is before our bytenr, since scrub is always | |
2452 | * done in bytenr order, the csum will never be used anymore, | |
2453 | * clean it up so that later calls won't bother with the range, | |
2454 | * and continue search the next range. | |
2455 | */ | |
2456 | if (sum->bytenr + sum->len <= logical) { | |
2457 | drop_csum_range(sctx, sum); | |
2458 | continue; | |
2459 | } | |
a2de733c | 2460 | |
480a8ec8 QW |
2461 | /* Now the csum range covers our bytenr, copy the csum */ |
2462 | found = true; | |
2463 | index = (logical - sum->bytenr) >> sctx->fs_info->sectorsize_bits; | |
2464 | num_sectors = sum->len >> sctx->fs_info->sectorsize_bits; | |
1d1bf92d | 2465 | |
480a8ec8 QW |
2466 | memcpy(csum, sum->sums + index * sctx->fs_info->csum_size, |
2467 | sctx->fs_info->csum_size); | |
2468 | ||
2469 | /* Cleanup the range if we're at the end of the csum range */ | |
2470 | if (index == num_sectors - 1) | |
2471 | drop_csum_range(sctx, sum); | |
2472 | break; | |
a2de733c | 2473 | } |
480a8ec8 QW |
2474 | if (!found) |
2475 | return 0; | |
f51a4a18 | 2476 | return 1; |
a2de733c AJ |
2477 | } |
2478 | ||
2479 | /* scrub extent tries to collect up to 64 kB for each bio */ | |
6ca1765b | 2480 | static int scrub_extent(struct scrub_ctx *sctx, struct map_lookup *map, |
fa485d21 | 2481 | u64 logical, u32 len, |
a36cf8b8 | 2482 | u64 physical, struct btrfs_device *dev, u64 flags, |
ff023aac | 2483 | u64 gen, int mirror_num, u64 physical_for_dev_replace) |
a2de733c AJ |
2484 | { |
2485 | int ret; | |
2486 | u8 csum[BTRFS_CSUM_SIZE]; | |
b5d67f64 SB |
2487 | u32 blocksize; |
2488 | ||
2489 | if (flags & BTRFS_EXTENT_FLAG_DATA) { | |
6ca1765b LB |
2490 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) |
2491 | blocksize = map->stripe_len; | |
2492 | else | |
2493 | blocksize = sctx->fs_info->sectorsize; | |
d9d181c1 SB |
2494 | spin_lock(&sctx->stat_lock); |
2495 | sctx->stat.data_extents_scrubbed++; | |
2496 | sctx->stat.data_bytes_scrubbed += len; | |
2497 | spin_unlock(&sctx->stat_lock); | |
b5d67f64 | 2498 | } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { |
6ca1765b LB |
2499 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) |
2500 | blocksize = map->stripe_len; | |
2501 | else | |
2502 | blocksize = sctx->fs_info->nodesize; | |
d9d181c1 SB |
2503 | spin_lock(&sctx->stat_lock); |
2504 | sctx->stat.tree_extents_scrubbed++; | |
2505 | sctx->stat.tree_bytes_scrubbed += len; | |
2506 | spin_unlock(&sctx->stat_lock); | |
b5d67f64 | 2507 | } else { |
25cc1226 | 2508 | blocksize = sctx->fs_info->sectorsize; |
ff023aac | 2509 | WARN_ON(1); |
b5d67f64 | 2510 | } |
a2de733c AJ |
2511 | |
2512 | while (len) { | |
fa485d21 | 2513 | u32 l = min(len, blocksize); |
a2de733c AJ |
2514 | int have_csum = 0; |
2515 | ||
2516 | if (flags & BTRFS_EXTENT_FLAG_DATA) { | |
2517 | /* push csums to sbio */ | |
3b5753ec | 2518 | have_csum = scrub_find_csum(sctx, logical, csum); |
a2de733c | 2519 | if (have_csum == 0) |
d9d181c1 | 2520 | ++sctx->stat.no_csum; |
a2de733c | 2521 | } |
a36cf8b8 | 2522 | ret = scrub_pages(sctx, logical, l, physical, dev, flags, gen, |
96e63a45 | 2523 | mirror_num, have_csum ? csum : NULL, |
ff023aac | 2524 | physical_for_dev_replace); |
a2de733c AJ |
2525 | if (ret) |
2526 | return ret; | |
2527 | len -= l; | |
2528 | logical += l; | |
2529 | physical += l; | |
ff023aac | 2530 | physical_for_dev_replace += l; |
a2de733c AJ |
2531 | } |
2532 | return 0; | |
2533 | } | |
2534 | ||
5a6ac9ea | 2535 | static int scrub_pages_for_parity(struct scrub_parity *sparity, |
fa485d21 | 2536 | u64 logical, u32 len, |
5a6ac9ea MX |
2537 | u64 physical, struct btrfs_device *dev, |
2538 | u64 flags, u64 gen, int mirror_num, u8 *csum) | |
2539 | { | |
2540 | struct scrub_ctx *sctx = sparity->sctx; | |
2541 | struct scrub_block *sblock; | |
d0a7a9c0 | 2542 | const u32 sectorsize = sctx->fs_info->sectorsize; |
5a6ac9ea MX |
2543 | int index; |
2544 | ||
d0a7a9c0 QW |
2545 | ASSERT(IS_ALIGNED(len, sectorsize)); |
2546 | ||
58c4e173 | 2547 | sblock = kzalloc(sizeof(*sblock), GFP_KERNEL); |
5a6ac9ea MX |
2548 | if (!sblock) { |
2549 | spin_lock(&sctx->stat_lock); | |
2550 | sctx->stat.malloc_errors++; | |
2551 | spin_unlock(&sctx->stat_lock); | |
2552 | return -ENOMEM; | |
2553 | } | |
2554 | ||
2555 | /* one ref inside this function, plus one for each page added to | |
2556 | * a bio later on */ | |
186debd6 | 2557 | refcount_set(&sblock->refs, 1); |
5a6ac9ea MX |
2558 | sblock->sctx = sctx; |
2559 | sblock->no_io_error_seen = 1; | |
2560 | sblock->sparity = sparity; | |
2561 | scrub_parity_get(sparity); | |
2562 | ||
2563 | for (index = 0; len > 0; index++) { | |
2564 | struct scrub_page *spage; | |
5a6ac9ea | 2565 | |
58c4e173 | 2566 | spage = kzalloc(sizeof(*spage), GFP_KERNEL); |
5a6ac9ea MX |
2567 | if (!spage) { |
2568 | leave_nomem: | |
2569 | spin_lock(&sctx->stat_lock); | |
2570 | sctx->stat.malloc_errors++; | |
2571 | spin_unlock(&sctx->stat_lock); | |
2572 | scrub_block_put(sblock); | |
2573 | return -ENOMEM; | |
2574 | } | |
2575 | BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK); | |
2576 | /* For scrub block */ | |
2577 | scrub_page_get(spage); | |
2578 | sblock->pagev[index] = spage; | |
2579 | /* For scrub parity */ | |
2580 | scrub_page_get(spage); | |
2581 | list_add_tail(&spage->list, &sparity->spages); | |
2582 | spage->sblock = sblock; | |
2583 | spage->dev = dev; | |
2584 | spage->flags = flags; | |
2585 | spage->generation = gen; | |
2586 | spage->logical = logical; | |
2587 | spage->physical = physical; | |
2588 | spage->mirror_num = mirror_num; | |
2589 | if (csum) { | |
2590 | spage->have_csum = 1; | |
2ae0c2d8 | 2591 | memcpy(spage->csum, csum, sctx->fs_info->csum_size); |
5a6ac9ea MX |
2592 | } else { |
2593 | spage->have_csum = 0; | |
2594 | } | |
2595 | sblock->page_count++; | |
58c4e173 | 2596 | spage->page = alloc_page(GFP_KERNEL); |
5a6ac9ea MX |
2597 | if (!spage->page) |
2598 | goto leave_nomem; | |
d0a7a9c0 QW |
2599 | |
2600 | ||
2601 | /* Iterate over the stripe range in sectorsize steps */ | |
2602 | len -= sectorsize; | |
2603 | logical += sectorsize; | |
2604 | physical += sectorsize; | |
5a6ac9ea MX |
2605 | } |
2606 | ||
2607 | WARN_ON(sblock->page_count == 0); | |
2608 | for (index = 0; index < sblock->page_count; index++) { | |
2609 | struct scrub_page *spage = sblock->pagev[index]; | |
2610 | int ret; | |
2611 | ||
2612 | ret = scrub_add_page_to_rd_bio(sctx, spage); | |
2613 | if (ret) { | |
2614 | scrub_block_put(sblock); | |
2615 | return ret; | |
2616 | } | |
2617 | } | |
2618 | ||
2619 | /* last one frees, either here or in bio completion for last page */ | |
2620 | scrub_block_put(sblock); | |
2621 | return 0; | |
2622 | } | |
2623 | ||
2624 | static int scrub_extent_for_parity(struct scrub_parity *sparity, | |
fa485d21 | 2625 | u64 logical, u32 len, |
5a6ac9ea MX |
2626 | u64 physical, struct btrfs_device *dev, |
2627 | u64 flags, u64 gen, int mirror_num) | |
2628 | { | |
2629 | struct scrub_ctx *sctx = sparity->sctx; | |
2630 | int ret; | |
2631 | u8 csum[BTRFS_CSUM_SIZE]; | |
2632 | u32 blocksize; | |
2633 | ||
e6e674bd | 2634 | if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) { |
4a770891 OS |
2635 | scrub_parity_mark_sectors_error(sparity, logical, len); |
2636 | return 0; | |
2637 | } | |
2638 | ||
5a6ac9ea | 2639 | if (flags & BTRFS_EXTENT_FLAG_DATA) { |
6ca1765b | 2640 | blocksize = sparity->stripe_len; |
5a6ac9ea | 2641 | } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { |
6ca1765b | 2642 | blocksize = sparity->stripe_len; |
5a6ac9ea | 2643 | } else { |
25cc1226 | 2644 | blocksize = sctx->fs_info->sectorsize; |
5a6ac9ea MX |
2645 | WARN_ON(1); |
2646 | } | |
2647 | ||
2648 | while (len) { | |
fa485d21 | 2649 | u32 l = min(len, blocksize); |
5a6ac9ea MX |
2650 | int have_csum = 0; |
2651 | ||
2652 | if (flags & BTRFS_EXTENT_FLAG_DATA) { | |
2653 | /* push csums to sbio */ | |
3b5753ec | 2654 | have_csum = scrub_find_csum(sctx, logical, csum); |
5a6ac9ea MX |
2655 | if (have_csum == 0) |
2656 | goto skip; | |
2657 | } | |
2658 | ret = scrub_pages_for_parity(sparity, logical, l, physical, dev, | |
2659 | flags, gen, mirror_num, | |
2660 | have_csum ? csum : NULL); | |
5a6ac9ea MX |
2661 | if (ret) |
2662 | return ret; | |
6b6d24b3 | 2663 | skip: |
5a6ac9ea MX |
2664 | len -= l; |
2665 | logical += l; | |
2666 | physical += l; | |
2667 | } | |
2668 | return 0; | |
2669 | } | |
2670 | ||
3b080b25 WS |
2671 | /* |
2672 | * Given a physical address, this will calculate it's | |
2673 | * logical offset. if this is a parity stripe, it will return | |
2674 | * the most left data stripe's logical offset. | |
2675 | * | |
2676 | * return 0 if it is a data stripe, 1 means parity stripe. | |
2677 | */ | |
2678 | static int get_raid56_logic_offset(u64 physical, int num, | |
5a6ac9ea MX |
2679 | struct map_lookup *map, u64 *offset, |
2680 | u64 *stripe_start) | |
3b080b25 WS |
2681 | { |
2682 | int i; | |
2683 | int j = 0; | |
2684 | u64 stripe_nr; | |
2685 | u64 last_offset; | |
9d644a62 DS |
2686 | u32 stripe_index; |
2687 | u32 rot; | |
cff82672 | 2688 | const int data_stripes = nr_data_stripes(map); |
3b080b25 | 2689 | |
cff82672 | 2690 | last_offset = (physical - map->stripes[num].physical) * data_stripes; |
5a6ac9ea MX |
2691 | if (stripe_start) |
2692 | *stripe_start = last_offset; | |
2693 | ||
3b080b25 | 2694 | *offset = last_offset; |
cff82672 | 2695 | for (i = 0; i < data_stripes; i++) { |
3b080b25 WS |
2696 | *offset = last_offset + i * map->stripe_len; |
2697 | ||
42c61ab6 | 2698 | stripe_nr = div64_u64(*offset, map->stripe_len); |
cff82672 | 2699 | stripe_nr = div_u64(stripe_nr, data_stripes); |
3b080b25 WS |
2700 | |
2701 | /* Work out the disk rotation on this stripe-set */ | |
47c5713f | 2702 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, &rot); |
3b080b25 WS |
2703 | /* calculate which stripe this data locates */ |
2704 | rot += i; | |
e4fbaee2 | 2705 | stripe_index = rot % map->num_stripes; |
3b080b25 WS |
2706 | if (stripe_index == num) |
2707 | return 0; | |
2708 | if (stripe_index < num) | |
2709 | j++; | |
2710 | } | |
2711 | *offset = last_offset + j * map->stripe_len; | |
2712 | return 1; | |
2713 | } | |
2714 | ||
5a6ac9ea MX |
2715 | static void scrub_free_parity(struct scrub_parity *sparity) |
2716 | { | |
2717 | struct scrub_ctx *sctx = sparity->sctx; | |
2718 | struct scrub_page *curr, *next; | |
2719 | int nbits; | |
2720 | ||
2721 | nbits = bitmap_weight(sparity->ebitmap, sparity->nsectors); | |
2722 | if (nbits) { | |
2723 | spin_lock(&sctx->stat_lock); | |
2724 | sctx->stat.read_errors += nbits; | |
2725 | sctx->stat.uncorrectable_errors += nbits; | |
2726 | spin_unlock(&sctx->stat_lock); | |
2727 | } | |
2728 | ||
2729 | list_for_each_entry_safe(curr, next, &sparity->spages, list) { | |
2730 | list_del_init(&curr->list); | |
2731 | scrub_page_put(curr); | |
2732 | } | |
2733 | ||
2734 | kfree(sparity); | |
2735 | } | |
2736 | ||
20b2e302 ZL |
2737 | static void scrub_parity_bio_endio_worker(struct btrfs_work *work) |
2738 | { | |
2739 | struct scrub_parity *sparity = container_of(work, struct scrub_parity, | |
2740 | work); | |
2741 | struct scrub_ctx *sctx = sparity->sctx; | |
2742 | ||
2743 | scrub_free_parity(sparity); | |
2744 | scrub_pending_bio_dec(sctx); | |
2745 | } | |
2746 | ||
4246a0b6 | 2747 | static void scrub_parity_bio_endio(struct bio *bio) |
5a6ac9ea MX |
2748 | { |
2749 | struct scrub_parity *sparity = (struct scrub_parity *)bio->bi_private; | |
0b246afa | 2750 | struct btrfs_fs_info *fs_info = sparity->sctx->fs_info; |
5a6ac9ea | 2751 | |
4e4cbee9 | 2752 | if (bio->bi_status) |
5a6ac9ea MX |
2753 | bitmap_or(sparity->ebitmap, sparity->ebitmap, sparity->dbitmap, |
2754 | sparity->nsectors); | |
2755 | ||
5a6ac9ea | 2756 | bio_put(bio); |
20b2e302 | 2757 | |
a0cac0ec OS |
2758 | btrfs_init_work(&sparity->work, scrub_parity_bio_endio_worker, NULL, |
2759 | NULL); | |
0b246afa | 2760 | btrfs_queue_work(fs_info->scrub_parity_workers, &sparity->work); |
5a6ac9ea MX |
2761 | } |
2762 | ||
2763 | static void scrub_parity_check_and_repair(struct scrub_parity *sparity) | |
2764 | { | |
2765 | struct scrub_ctx *sctx = sparity->sctx; | |
0b246afa | 2766 | struct btrfs_fs_info *fs_info = sctx->fs_info; |
5a6ac9ea MX |
2767 | struct bio *bio; |
2768 | struct btrfs_raid_bio *rbio; | |
5a6ac9ea | 2769 | struct btrfs_bio *bbio = NULL; |
5a6ac9ea MX |
2770 | u64 length; |
2771 | int ret; | |
2772 | ||
2773 | if (!bitmap_andnot(sparity->dbitmap, sparity->dbitmap, sparity->ebitmap, | |
2774 | sparity->nsectors)) | |
2775 | goto out; | |
2776 | ||
a0dd59de | 2777 | length = sparity->logic_end - sparity->logic_start; |
ae6529c3 QW |
2778 | |
2779 | btrfs_bio_counter_inc_blocked(fs_info); | |
0b246afa | 2780 | ret = btrfs_map_sblock(fs_info, BTRFS_MAP_WRITE, sparity->logic_start, |
825ad4c9 | 2781 | &length, &bbio); |
8e5cfb55 | 2782 | if (ret || !bbio || !bbio->raid_map) |
5a6ac9ea MX |
2783 | goto bbio_out; |
2784 | ||
c5e4c3d7 | 2785 | bio = btrfs_io_bio_alloc(0); |
5a6ac9ea MX |
2786 | bio->bi_iter.bi_sector = sparity->logic_start >> 9; |
2787 | bio->bi_private = sparity; | |
2788 | bio->bi_end_io = scrub_parity_bio_endio; | |
2789 | ||
2ff7e61e | 2790 | rbio = raid56_parity_alloc_scrub_rbio(fs_info, bio, bbio, |
8e5cfb55 | 2791 | length, sparity->scrub_dev, |
5a6ac9ea MX |
2792 | sparity->dbitmap, |
2793 | sparity->nsectors); | |
2794 | if (!rbio) | |
2795 | goto rbio_out; | |
2796 | ||
5a6ac9ea MX |
2797 | scrub_pending_bio_inc(sctx); |
2798 | raid56_parity_submit_scrub_rbio(rbio); | |
2799 | return; | |
2800 | ||
2801 | rbio_out: | |
2802 | bio_put(bio); | |
2803 | bbio_out: | |
ae6529c3 | 2804 | btrfs_bio_counter_dec(fs_info); |
6e9606d2 | 2805 | btrfs_put_bbio(bbio); |
5a6ac9ea MX |
2806 | bitmap_or(sparity->ebitmap, sparity->ebitmap, sparity->dbitmap, |
2807 | sparity->nsectors); | |
2808 | spin_lock(&sctx->stat_lock); | |
2809 | sctx->stat.malloc_errors++; | |
2810 | spin_unlock(&sctx->stat_lock); | |
2811 | out: | |
2812 | scrub_free_parity(sparity); | |
2813 | } | |
2814 | ||
2815 | static inline int scrub_calc_parity_bitmap_len(int nsectors) | |
2816 | { | |
bfca9a6d | 2817 | return DIV_ROUND_UP(nsectors, BITS_PER_LONG) * sizeof(long); |
5a6ac9ea MX |
2818 | } |
2819 | ||
2820 | static void scrub_parity_get(struct scrub_parity *sparity) | |
2821 | { | |
78a76450 | 2822 | refcount_inc(&sparity->refs); |
5a6ac9ea MX |
2823 | } |
2824 | ||
2825 | static void scrub_parity_put(struct scrub_parity *sparity) | |
2826 | { | |
78a76450 | 2827 | if (!refcount_dec_and_test(&sparity->refs)) |
5a6ac9ea MX |
2828 | return; |
2829 | ||
2830 | scrub_parity_check_and_repair(sparity); | |
2831 | } | |
2832 | ||
2833 | static noinline_for_stack int scrub_raid56_parity(struct scrub_ctx *sctx, | |
2834 | struct map_lookup *map, | |
2835 | struct btrfs_device *sdev, | |
2836 | struct btrfs_path *path, | |
2837 | u64 logic_start, | |
2838 | u64 logic_end) | |
2839 | { | |
fb456252 | 2840 | struct btrfs_fs_info *fs_info = sctx->fs_info; |
5a6ac9ea MX |
2841 | struct btrfs_root *root = fs_info->extent_root; |
2842 | struct btrfs_root *csum_root = fs_info->csum_root; | |
2843 | struct btrfs_extent_item *extent; | |
4a770891 | 2844 | struct btrfs_bio *bbio = NULL; |
5a6ac9ea MX |
2845 | u64 flags; |
2846 | int ret; | |
2847 | int slot; | |
2848 | struct extent_buffer *l; | |
2849 | struct btrfs_key key; | |
2850 | u64 generation; | |
2851 | u64 extent_logical; | |
2852 | u64 extent_physical; | |
fa485d21 QW |
2853 | /* Check the comment in scrub_stripe() for why u32 is enough here */ |
2854 | u32 extent_len; | |
4a770891 | 2855 | u64 mapped_length; |
5a6ac9ea MX |
2856 | struct btrfs_device *extent_dev; |
2857 | struct scrub_parity *sparity; | |
2858 | int nsectors; | |
2859 | int bitmap_len; | |
2860 | int extent_mirror_num; | |
2861 | int stop_loop = 0; | |
2862 | ||
fa485d21 | 2863 | ASSERT(map->stripe_len <= U32_MAX); |
ab108d99 | 2864 | nsectors = map->stripe_len >> fs_info->sectorsize_bits; |
5a6ac9ea MX |
2865 | bitmap_len = scrub_calc_parity_bitmap_len(nsectors); |
2866 | sparity = kzalloc(sizeof(struct scrub_parity) + 2 * bitmap_len, | |
2867 | GFP_NOFS); | |
2868 | if (!sparity) { | |
2869 | spin_lock(&sctx->stat_lock); | |
2870 | sctx->stat.malloc_errors++; | |
2871 | spin_unlock(&sctx->stat_lock); | |
2872 | return -ENOMEM; | |
2873 | } | |
2874 | ||
fa485d21 | 2875 | ASSERT(map->stripe_len <= U32_MAX); |
5a6ac9ea MX |
2876 | sparity->stripe_len = map->stripe_len; |
2877 | sparity->nsectors = nsectors; | |
2878 | sparity->sctx = sctx; | |
2879 | sparity->scrub_dev = sdev; | |
2880 | sparity->logic_start = logic_start; | |
2881 | sparity->logic_end = logic_end; | |
78a76450 | 2882 | refcount_set(&sparity->refs, 1); |
5a6ac9ea MX |
2883 | INIT_LIST_HEAD(&sparity->spages); |
2884 | sparity->dbitmap = sparity->bitmap; | |
2885 | sparity->ebitmap = (void *)sparity->bitmap + bitmap_len; | |
2886 | ||
2887 | ret = 0; | |
2888 | while (logic_start < logic_end) { | |
2889 | if (btrfs_fs_incompat(fs_info, SKINNY_METADATA)) | |
2890 | key.type = BTRFS_METADATA_ITEM_KEY; | |
2891 | else | |
2892 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2893 | key.objectid = logic_start; | |
2894 | key.offset = (u64)-1; | |
2895 | ||
2896 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2897 | if (ret < 0) | |
2898 | goto out; | |
2899 | ||
2900 | if (ret > 0) { | |
2901 | ret = btrfs_previous_extent_item(root, path, 0); | |
2902 | if (ret < 0) | |
2903 | goto out; | |
2904 | if (ret > 0) { | |
2905 | btrfs_release_path(path); | |
2906 | ret = btrfs_search_slot(NULL, root, &key, | |
2907 | path, 0, 0); | |
2908 | if (ret < 0) | |
2909 | goto out; | |
2910 | } | |
2911 | } | |
2912 | ||
2913 | stop_loop = 0; | |
2914 | while (1) { | |
2915 | u64 bytes; | |
2916 | ||
2917 | l = path->nodes[0]; | |
2918 | slot = path->slots[0]; | |
2919 | if (slot >= btrfs_header_nritems(l)) { | |
2920 | ret = btrfs_next_leaf(root, path); | |
2921 | if (ret == 0) | |
2922 | continue; | |
2923 | if (ret < 0) | |
2924 | goto out; | |
2925 | ||
2926 | stop_loop = 1; | |
2927 | break; | |
2928 | } | |
2929 | btrfs_item_key_to_cpu(l, &key, slot); | |
2930 | ||
d7cad238 ZL |
2931 | if (key.type != BTRFS_EXTENT_ITEM_KEY && |
2932 | key.type != BTRFS_METADATA_ITEM_KEY) | |
2933 | goto next; | |
2934 | ||
5a6ac9ea | 2935 | if (key.type == BTRFS_METADATA_ITEM_KEY) |
0b246afa | 2936 | bytes = fs_info->nodesize; |
5a6ac9ea MX |
2937 | else |
2938 | bytes = key.offset; | |
2939 | ||
2940 | if (key.objectid + bytes <= logic_start) | |
2941 | goto next; | |
2942 | ||
a0dd59de | 2943 | if (key.objectid >= logic_end) { |
5a6ac9ea MX |
2944 | stop_loop = 1; |
2945 | break; | |
2946 | } | |
2947 | ||
2948 | while (key.objectid >= logic_start + map->stripe_len) | |
2949 | logic_start += map->stripe_len; | |
2950 | ||
2951 | extent = btrfs_item_ptr(l, slot, | |
2952 | struct btrfs_extent_item); | |
2953 | flags = btrfs_extent_flags(l, extent); | |
2954 | generation = btrfs_extent_generation(l, extent); | |
2955 | ||
a323e813 ZL |
2956 | if ((flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) && |
2957 | (key.objectid < logic_start || | |
2958 | key.objectid + bytes > | |
2959 | logic_start + map->stripe_len)) { | |
5d163e0e JM |
2960 | btrfs_err(fs_info, |
2961 | "scrub: tree block %llu spanning stripes, ignored. logical=%llu", | |
a323e813 | 2962 | key.objectid, logic_start); |
9799d2c3 ZL |
2963 | spin_lock(&sctx->stat_lock); |
2964 | sctx->stat.uncorrectable_errors++; | |
2965 | spin_unlock(&sctx->stat_lock); | |
5a6ac9ea MX |
2966 | goto next; |
2967 | } | |
2968 | again: | |
2969 | extent_logical = key.objectid; | |
fa485d21 | 2970 | ASSERT(bytes <= U32_MAX); |
5a6ac9ea MX |
2971 | extent_len = bytes; |
2972 | ||
2973 | if (extent_logical < logic_start) { | |
2974 | extent_len -= logic_start - extent_logical; | |
2975 | extent_logical = logic_start; | |
2976 | } | |
2977 | ||
2978 | if (extent_logical + extent_len > | |
2979 | logic_start + map->stripe_len) | |
2980 | extent_len = logic_start + map->stripe_len - | |
2981 | extent_logical; | |
2982 | ||
2983 | scrub_parity_mark_sectors_data(sparity, extent_logical, | |
2984 | extent_len); | |
2985 | ||
4a770891 | 2986 | mapped_length = extent_len; |
f1fee653 | 2987 | bbio = NULL; |
cf8cddd3 CH |
2988 | ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, |
2989 | extent_logical, &mapped_length, &bbio, | |
2990 | 0); | |
4a770891 OS |
2991 | if (!ret) { |
2992 | if (!bbio || mapped_length < extent_len) | |
2993 | ret = -EIO; | |
2994 | } | |
2995 | if (ret) { | |
2996 | btrfs_put_bbio(bbio); | |
2997 | goto out; | |
2998 | } | |
2999 | extent_physical = bbio->stripes[0].physical; | |
3000 | extent_mirror_num = bbio->mirror_num; | |
3001 | extent_dev = bbio->stripes[0].dev; | |
3002 | btrfs_put_bbio(bbio); | |
5a6ac9ea MX |
3003 | |
3004 | ret = btrfs_lookup_csums_range(csum_root, | |
3005 | extent_logical, | |
3006 | extent_logical + extent_len - 1, | |
3007 | &sctx->csum_list, 1); | |
3008 | if (ret) | |
3009 | goto out; | |
3010 | ||
3011 | ret = scrub_extent_for_parity(sparity, extent_logical, | |
3012 | extent_len, | |
3013 | extent_physical, | |
3014 | extent_dev, flags, | |
3015 | generation, | |
3016 | extent_mirror_num); | |
6fa96d72 ZL |
3017 | |
3018 | scrub_free_csums(sctx); | |
3019 | ||
5a6ac9ea MX |
3020 | if (ret) |
3021 | goto out; | |
3022 | ||
5a6ac9ea MX |
3023 | if (extent_logical + extent_len < |
3024 | key.objectid + bytes) { | |
3025 | logic_start += map->stripe_len; | |
3026 | ||
3027 | if (logic_start >= logic_end) { | |
3028 | stop_loop = 1; | |
3029 | break; | |
3030 | } | |
3031 | ||
3032 | if (logic_start < key.objectid + bytes) { | |
3033 | cond_resched(); | |
3034 | goto again; | |
3035 | } | |
3036 | } | |
3037 | next: | |
3038 | path->slots[0]++; | |
3039 | } | |
3040 | ||
3041 | btrfs_release_path(path); | |
3042 | ||
3043 | if (stop_loop) | |
3044 | break; | |
3045 | ||
3046 | logic_start += map->stripe_len; | |
3047 | } | |
3048 | out: | |
fa485d21 QW |
3049 | if (ret < 0) { |
3050 | ASSERT(logic_end - logic_start <= U32_MAX); | |
5a6ac9ea | 3051 | scrub_parity_mark_sectors_error(sparity, logic_start, |
a0dd59de | 3052 | logic_end - logic_start); |
fa485d21 | 3053 | } |
5a6ac9ea MX |
3054 | scrub_parity_put(sparity); |
3055 | scrub_submit(sctx); | |
3fb99303 | 3056 | mutex_lock(&sctx->wr_lock); |
5a6ac9ea | 3057 | scrub_wr_submit(sctx); |
3fb99303 | 3058 | mutex_unlock(&sctx->wr_lock); |
5a6ac9ea MX |
3059 | |
3060 | btrfs_release_path(path); | |
3061 | return ret < 0 ? ret : 0; | |
3062 | } | |
3063 | ||
de17addc NA |
3064 | static void sync_replace_for_zoned(struct scrub_ctx *sctx) |
3065 | { | |
3066 | if (!btrfs_is_zoned(sctx->fs_info)) | |
3067 | return; | |
3068 | ||
3069 | sctx->flush_all_writes = true; | |
3070 | scrub_submit(sctx); | |
3071 | mutex_lock(&sctx->wr_lock); | |
3072 | scrub_wr_submit(sctx); | |
3073 | mutex_unlock(&sctx->wr_lock); | |
3074 | ||
3075 | wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0); | |
3076 | } | |
3077 | ||
7db1c5d1 NA |
3078 | static int sync_write_pointer_for_zoned(struct scrub_ctx *sctx, u64 logical, |
3079 | u64 physical, u64 physical_end) | |
3080 | { | |
3081 | struct btrfs_fs_info *fs_info = sctx->fs_info; | |
3082 | int ret = 0; | |
3083 | ||
3084 | if (!btrfs_is_zoned(fs_info)) | |
3085 | return 0; | |
3086 | ||
3087 | wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0); | |
3088 | ||
3089 | mutex_lock(&sctx->wr_lock); | |
3090 | if (sctx->write_pointer < physical_end) { | |
3091 | ret = btrfs_sync_zone_write_pointer(sctx->wr_tgtdev, logical, | |
3092 | physical, | |
3093 | sctx->write_pointer); | |
3094 | if (ret) | |
3095 | btrfs_err(fs_info, | |
3096 | "zoned: failed to recover write pointer"); | |
3097 | } | |
3098 | mutex_unlock(&sctx->wr_lock); | |
3099 | btrfs_dev_clear_zone_empty(sctx->wr_tgtdev, physical); | |
3100 | ||
3101 | return ret; | |
3102 | } | |
3103 | ||
d9d181c1 | 3104 | static noinline_for_stack int scrub_stripe(struct scrub_ctx *sctx, |
a36cf8b8 SB |
3105 | struct map_lookup *map, |
3106 | struct btrfs_device *scrub_dev, | |
2473d24f FM |
3107 | int num, u64 base, u64 length, |
3108 | struct btrfs_block_group *cache) | |
a2de733c | 3109 | { |
5a6ac9ea | 3110 | struct btrfs_path *path, *ppath; |
fb456252 | 3111 | struct btrfs_fs_info *fs_info = sctx->fs_info; |
a2de733c AJ |
3112 | struct btrfs_root *root = fs_info->extent_root; |
3113 | struct btrfs_root *csum_root = fs_info->csum_root; | |
3114 | struct btrfs_extent_item *extent; | |
e7786c3a | 3115 | struct blk_plug plug; |
a2de733c AJ |
3116 | u64 flags; |
3117 | int ret; | |
3118 | int slot; | |
a2de733c | 3119 | u64 nstripes; |
a2de733c | 3120 | struct extent_buffer *l; |
a2de733c AJ |
3121 | u64 physical; |
3122 | u64 logical; | |
625f1c8d | 3123 | u64 logic_end; |
3b080b25 | 3124 | u64 physical_end; |
a2de733c | 3125 | u64 generation; |
e12fa9cd | 3126 | int mirror_num; |
7a26285e AJ |
3127 | struct reada_control *reada1; |
3128 | struct reada_control *reada2; | |
e6c11f9a | 3129 | struct btrfs_key key; |
7a26285e | 3130 | struct btrfs_key key_end; |
a2de733c AJ |
3131 | u64 increment = map->stripe_len; |
3132 | u64 offset; | |
ff023aac SB |
3133 | u64 extent_logical; |
3134 | u64 extent_physical; | |
fa485d21 QW |
3135 | /* |
3136 | * Unlike chunk length, extent length should never go beyond | |
3137 | * BTRFS_MAX_EXTENT_SIZE, thus u32 is enough here. | |
3138 | */ | |
3139 | u32 extent_len; | |
5a6ac9ea MX |
3140 | u64 stripe_logical; |
3141 | u64 stripe_end; | |
ff023aac SB |
3142 | struct btrfs_device *extent_dev; |
3143 | int extent_mirror_num; | |
3b080b25 | 3144 | int stop_loop = 0; |
53b381b3 | 3145 | |
3b080b25 | 3146 | physical = map->stripes[num].physical; |
a2de733c | 3147 | offset = 0; |
42c61ab6 | 3148 | nstripes = div64_u64(length, map->stripe_len); |
a2de733c AJ |
3149 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
3150 | offset = map->stripe_len * num; | |
3151 | increment = map->stripe_len * map->num_stripes; | |
193ea74b | 3152 | mirror_num = 1; |
a2de733c AJ |
3153 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
3154 | int factor = map->num_stripes / map->sub_stripes; | |
3155 | offset = map->stripe_len * (num / map->sub_stripes); | |
3156 | increment = map->stripe_len * factor; | |
193ea74b | 3157 | mirror_num = num % map->sub_stripes + 1; |
c7369b3f | 3158 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) { |
a2de733c | 3159 | increment = map->stripe_len; |
193ea74b | 3160 | mirror_num = num % map->num_stripes + 1; |
a2de733c AJ |
3161 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
3162 | increment = map->stripe_len; | |
193ea74b | 3163 | mirror_num = num % map->num_stripes + 1; |
ffe2d203 | 3164 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
5a6ac9ea | 3165 | get_raid56_logic_offset(physical, num, map, &offset, NULL); |
3b080b25 WS |
3166 | increment = map->stripe_len * nr_data_stripes(map); |
3167 | mirror_num = 1; | |
a2de733c AJ |
3168 | } else { |
3169 | increment = map->stripe_len; | |
193ea74b | 3170 | mirror_num = 1; |
a2de733c AJ |
3171 | } |
3172 | ||
3173 | path = btrfs_alloc_path(); | |
3174 | if (!path) | |
3175 | return -ENOMEM; | |
3176 | ||
5a6ac9ea MX |
3177 | ppath = btrfs_alloc_path(); |
3178 | if (!ppath) { | |
379d6854 | 3179 | btrfs_free_path(path); |
5a6ac9ea MX |
3180 | return -ENOMEM; |
3181 | } | |
3182 | ||
b5d67f64 SB |
3183 | /* |
3184 | * work on commit root. The related disk blocks are static as | |
3185 | * long as COW is applied. This means, it is save to rewrite | |
3186 | * them to repair disk errors without any race conditions | |
3187 | */ | |
a2de733c AJ |
3188 | path->search_commit_root = 1; |
3189 | path->skip_locking = 1; | |
3190 | ||
063c54dc GH |
3191 | ppath->search_commit_root = 1; |
3192 | ppath->skip_locking = 1; | |
a2de733c | 3193 | /* |
7a26285e AJ |
3194 | * trigger the readahead for extent tree csum tree and wait for |
3195 | * completion. During readahead, the scrub is officially paused | |
3196 | * to not hold off transaction commits | |
a2de733c AJ |
3197 | */ |
3198 | logical = base + offset; | |
3b080b25 | 3199 | physical_end = physical + nstripes * map->stripe_len; |
ffe2d203 | 3200 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
3b080b25 | 3201 | get_raid56_logic_offset(physical_end, num, |
5a6ac9ea | 3202 | map, &logic_end, NULL); |
3b080b25 WS |
3203 | logic_end += base; |
3204 | } else { | |
3205 | logic_end = logical + increment * nstripes; | |
3206 | } | |
d9d181c1 | 3207 | wait_event(sctx->list_wait, |
b6bfebc1 | 3208 | atomic_read(&sctx->bios_in_flight) == 0); |
cb7ab021 | 3209 | scrub_blocked_if_needed(fs_info); |
7a26285e AJ |
3210 | |
3211 | /* FIXME it might be better to start readahead at commit root */ | |
e6c11f9a DS |
3212 | key.objectid = logical; |
3213 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
3214 | key.offset = (u64)0; | |
3b080b25 | 3215 | key_end.objectid = logic_end; |
3173a18f JB |
3216 | key_end.type = BTRFS_METADATA_ITEM_KEY; |
3217 | key_end.offset = (u64)-1; | |
e6c11f9a | 3218 | reada1 = btrfs_reada_add(root, &key, &key_end); |
7a26285e | 3219 | |
a6889caf FM |
3220 | if (cache->flags & BTRFS_BLOCK_GROUP_DATA) { |
3221 | key.objectid = BTRFS_EXTENT_CSUM_OBJECTID; | |
3222 | key.type = BTRFS_EXTENT_CSUM_KEY; | |
3223 | key.offset = logical; | |
3224 | key_end.objectid = BTRFS_EXTENT_CSUM_OBJECTID; | |
3225 | key_end.type = BTRFS_EXTENT_CSUM_KEY; | |
3226 | key_end.offset = logic_end; | |
3227 | reada2 = btrfs_reada_add(csum_root, &key, &key_end); | |
3228 | } else { | |
3229 | reada2 = NULL; | |
3230 | } | |
7a26285e AJ |
3231 | |
3232 | if (!IS_ERR(reada1)) | |
3233 | btrfs_reada_wait(reada1); | |
a6889caf | 3234 | if (!IS_ERR_OR_NULL(reada2)) |
7a26285e AJ |
3235 | btrfs_reada_wait(reada2); |
3236 | ||
a2de733c AJ |
3237 | |
3238 | /* | |
3239 | * collect all data csums for the stripe to avoid seeking during | |
3240 | * the scrub. This might currently (crc32) end up to be about 1MB | |
3241 | */ | |
e7786c3a | 3242 | blk_start_plug(&plug); |
a2de733c | 3243 | |
de17addc NA |
3244 | if (sctx->is_dev_replace && |
3245 | btrfs_dev_is_sequential(sctx->wr_tgtdev, physical)) { | |
3246 | mutex_lock(&sctx->wr_lock); | |
3247 | sctx->write_pointer = physical; | |
3248 | mutex_unlock(&sctx->wr_lock); | |
3249 | sctx->flush_all_writes = true; | |
3250 | } | |
3251 | ||
a2de733c AJ |
3252 | /* |
3253 | * now find all extents for each stripe and scrub them | |
3254 | */ | |
a2de733c | 3255 | ret = 0; |
3b080b25 | 3256 | while (physical < physical_end) { |
a2de733c AJ |
3257 | /* |
3258 | * canceled? | |
3259 | */ | |
3260 | if (atomic_read(&fs_info->scrub_cancel_req) || | |
d9d181c1 | 3261 | atomic_read(&sctx->cancel_req)) { |
a2de733c AJ |
3262 | ret = -ECANCELED; |
3263 | goto out; | |
3264 | } | |
3265 | /* | |
3266 | * check to see if we have to pause | |
3267 | */ | |
3268 | if (atomic_read(&fs_info->scrub_pause_req)) { | |
3269 | /* push queued extents */ | |
2073c4c2 | 3270 | sctx->flush_all_writes = true; |
d9d181c1 | 3271 | scrub_submit(sctx); |
3fb99303 | 3272 | mutex_lock(&sctx->wr_lock); |
ff023aac | 3273 | scrub_wr_submit(sctx); |
3fb99303 | 3274 | mutex_unlock(&sctx->wr_lock); |
d9d181c1 | 3275 | wait_event(sctx->list_wait, |
b6bfebc1 | 3276 | atomic_read(&sctx->bios_in_flight) == 0); |
2073c4c2 | 3277 | sctx->flush_all_writes = false; |
3cb0929a | 3278 | scrub_blocked_if_needed(fs_info); |
a2de733c AJ |
3279 | } |
3280 | ||
f2f66a2f ZL |
3281 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
3282 | ret = get_raid56_logic_offset(physical, num, map, | |
3283 | &logical, | |
3284 | &stripe_logical); | |
3285 | logical += base; | |
3286 | if (ret) { | |
7955323b | 3287 | /* it is parity strip */ |
f2f66a2f | 3288 | stripe_logical += base; |
a0dd59de | 3289 | stripe_end = stripe_logical + increment; |
f2f66a2f ZL |
3290 | ret = scrub_raid56_parity(sctx, map, scrub_dev, |
3291 | ppath, stripe_logical, | |
3292 | stripe_end); | |
3293 | if (ret) | |
3294 | goto out; | |
3295 | goto skip; | |
3296 | } | |
3297 | } | |
3298 | ||
7c76edb7 WS |
3299 | if (btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
3300 | key.type = BTRFS_METADATA_ITEM_KEY; | |
3301 | else | |
3302 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
a2de733c | 3303 | key.objectid = logical; |
625f1c8d | 3304 | key.offset = (u64)-1; |
a2de733c AJ |
3305 | |
3306 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
3307 | if (ret < 0) | |
3308 | goto out; | |
3173a18f | 3309 | |
8c51032f | 3310 | if (ret > 0) { |
ade2e0b3 | 3311 | ret = btrfs_previous_extent_item(root, path, 0); |
a2de733c AJ |
3312 | if (ret < 0) |
3313 | goto out; | |
8c51032f AJ |
3314 | if (ret > 0) { |
3315 | /* there's no smaller item, so stick with the | |
3316 | * larger one */ | |
3317 | btrfs_release_path(path); | |
3318 | ret = btrfs_search_slot(NULL, root, &key, | |
3319 | path, 0, 0); | |
3320 | if (ret < 0) | |
3321 | goto out; | |
3322 | } | |
a2de733c AJ |
3323 | } |
3324 | ||
625f1c8d | 3325 | stop_loop = 0; |
a2de733c | 3326 | while (1) { |
3173a18f JB |
3327 | u64 bytes; |
3328 | ||
a2de733c AJ |
3329 | l = path->nodes[0]; |
3330 | slot = path->slots[0]; | |
3331 | if (slot >= btrfs_header_nritems(l)) { | |
3332 | ret = btrfs_next_leaf(root, path); | |
3333 | if (ret == 0) | |
3334 | continue; | |
3335 | if (ret < 0) | |
3336 | goto out; | |
3337 | ||
625f1c8d | 3338 | stop_loop = 1; |
a2de733c AJ |
3339 | break; |
3340 | } | |
3341 | btrfs_item_key_to_cpu(l, &key, slot); | |
3342 | ||
d7cad238 ZL |
3343 | if (key.type != BTRFS_EXTENT_ITEM_KEY && |
3344 | key.type != BTRFS_METADATA_ITEM_KEY) | |
3345 | goto next; | |
3346 | ||
3173a18f | 3347 | if (key.type == BTRFS_METADATA_ITEM_KEY) |
0b246afa | 3348 | bytes = fs_info->nodesize; |
3173a18f JB |
3349 | else |
3350 | bytes = key.offset; | |
3351 | ||
3352 | if (key.objectid + bytes <= logical) | |
a2de733c AJ |
3353 | goto next; |
3354 | ||
625f1c8d LB |
3355 | if (key.objectid >= logical + map->stripe_len) { |
3356 | /* out of this device extent */ | |
3357 | if (key.objectid >= logic_end) | |
3358 | stop_loop = 1; | |
3359 | break; | |
3360 | } | |
a2de733c | 3361 | |
2473d24f FM |
3362 | /* |
3363 | * If our block group was removed in the meanwhile, just | |
3364 | * stop scrubbing since there is no point in continuing. | |
3365 | * Continuing would prevent reusing its device extents | |
3366 | * for new block groups for a long time. | |
3367 | */ | |
3368 | spin_lock(&cache->lock); | |
3369 | if (cache->removed) { | |
3370 | spin_unlock(&cache->lock); | |
3371 | ret = 0; | |
3372 | goto out; | |
3373 | } | |
3374 | spin_unlock(&cache->lock); | |
3375 | ||
a2de733c AJ |
3376 | extent = btrfs_item_ptr(l, slot, |
3377 | struct btrfs_extent_item); | |
3378 | flags = btrfs_extent_flags(l, extent); | |
3379 | generation = btrfs_extent_generation(l, extent); | |
3380 | ||
a323e813 ZL |
3381 | if ((flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) && |
3382 | (key.objectid < logical || | |
3383 | key.objectid + bytes > | |
3384 | logical + map->stripe_len)) { | |
efe120a0 | 3385 | btrfs_err(fs_info, |
5d163e0e | 3386 | "scrub: tree block %llu spanning stripes, ignored. logical=%llu", |
c1c9ff7c | 3387 | key.objectid, logical); |
9799d2c3 ZL |
3388 | spin_lock(&sctx->stat_lock); |
3389 | sctx->stat.uncorrectable_errors++; | |
3390 | spin_unlock(&sctx->stat_lock); | |
a2de733c AJ |
3391 | goto next; |
3392 | } | |
3393 | ||
625f1c8d LB |
3394 | again: |
3395 | extent_logical = key.objectid; | |
fa485d21 | 3396 | ASSERT(bytes <= U32_MAX); |
625f1c8d LB |
3397 | extent_len = bytes; |
3398 | ||
a2de733c AJ |
3399 | /* |
3400 | * trim extent to this stripe | |
3401 | */ | |
625f1c8d LB |
3402 | if (extent_logical < logical) { |
3403 | extent_len -= logical - extent_logical; | |
3404 | extent_logical = logical; | |
a2de733c | 3405 | } |
625f1c8d | 3406 | if (extent_logical + extent_len > |
a2de733c | 3407 | logical + map->stripe_len) { |
625f1c8d LB |
3408 | extent_len = logical + map->stripe_len - |
3409 | extent_logical; | |
a2de733c AJ |
3410 | } |
3411 | ||
625f1c8d | 3412 | extent_physical = extent_logical - logical + physical; |
ff023aac SB |
3413 | extent_dev = scrub_dev; |
3414 | extent_mirror_num = mirror_num; | |
32934280 | 3415 | if (sctx->is_dev_replace) |
ff023aac SB |
3416 | scrub_remap_extent(fs_info, extent_logical, |
3417 | extent_len, &extent_physical, | |
3418 | &extent_dev, | |
3419 | &extent_mirror_num); | |
625f1c8d | 3420 | |
89490303 FM |
3421 | if (flags & BTRFS_EXTENT_FLAG_DATA) { |
3422 | ret = btrfs_lookup_csums_range(csum_root, | |
3423 | extent_logical, | |
3424 | extent_logical + extent_len - 1, | |
3425 | &sctx->csum_list, 1); | |
3426 | if (ret) | |
3427 | goto out; | |
3428 | } | |
625f1c8d | 3429 | |
6ca1765b | 3430 | ret = scrub_extent(sctx, map, extent_logical, extent_len, |
ff023aac SB |
3431 | extent_physical, extent_dev, flags, |
3432 | generation, extent_mirror_num, | |
115930cb | 3433 | extent_logical - logical + physical); |
6fa96d72 ZL |
3434 | |
3435 | scrub_free_csums(sctx); | |
3436 | ||
a2de733c AJ |
3437 | if (ret) |
3438 | goto out; | |
3439 | ||
de17addc NA |
3440 | if (sctx->is_dev_replace) |
3441 | sync_replace_for_zoned(sctx); | |
3442 | ||
625f1c8d LB |
3443 | if (extent_logical + extent_len < |
3444 | key.objectid + bytes) { | |
ffe2d203 | 3445 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
3b080b25 WS |
3446 | /* |
3447 | * loop until we find next data stripe | |
3448 | * or we have finished all stripes. | |
3449 | */ | |
5a6ac9ea MX |
3450 | loop: |
3451 | physical += map->stripe_len; | |
3452 | ret = get_raid56_logic_offset(physical, | |
3453 | num, map, &logical, | |
3454 | &stripe_logical); | |
3455 | logical += base; | |
3456 | ||
3457 | if (ret && physical < physical_end) { | |
3458 | stripe_logical += base; | |
3459 | stripe_end = stripe_logical + | |
a0dd59de | 3460 | increment; |
5a6ac9ea MX |
3461 | ret = scrub_raid56_parity(sctx, |
3462 | map, scrub_dev, ppath, | |
3463 | stripe_logical, | |
3464 | stripe_end); | |
3465 | if (ret) | |
3466 | goto out; | |
3467 | goto loop; | |
3468 | } | |
3b080b25 WS |
3469 | } else { |
3470 | physical += map->stripe_len; | |
3471 | logical += increment; | |
3472 | } | |
625f1c8d LB |
3473 | if (logical < key.objectid + bytes) { |
3474 | cond_resched(); | |
3475 | goto again; | |
3476 | } | |
3477 | ||
3b080b25 | 3478 | if (physical >= physical_end) { |
625f1c8d LB |
3479 | stop_loop = 1; |
3480 | break; | |
3481 | } | |
3482 | } | |
a2de733c AJ |
3483 | next: |
3484 | path->slots[0]++; | |
3485 | } | |
71267333 | 3486 | btrfs_release_path(path); |
3b080b25 | 3487 | skip: |
a2de733c AJ |
3488 | logical += increment; |
3489 | physical += map->stripe_len; | |
d9d181c1 | 3490 | spin_lock(&sctx->stat_lock); |
625f1c8d LB |
3491 | if (stop_loop) |
3492 | sctx->stat.last_physical = map->stripes[num].physical + | |
3493 | length; | |
3494 | else | |
3495 | sctx->stat.last_physical = physical; | |
d9d181c1 | 3496 | spin_unlock(&sctx->stat_lock); |
625f1c8d LB |
3497 | if (stop_loop) |
3498 | break; | |
a2de733c | 3499 | } |
ff023aac | 3500 | out: |
a2de733c | 3501 | /* push queued extents */ |
d9d181c1 | 3502 | scrub_submit(sctx); |
3fb99303 | 3503 | mutex_lock(&sctx->wr_lock); |
ff023aac | 3504 | scrub_wr_submit(sctx); |
3fb99303 | 3505 | mutex_unlock(&sctx->wr_lock); |
a2de733c | 3506 | |
e7786c3a | 3507 | blk_finish_plug(&plug); |
a2de733c | 3508 | btrfs_free_path(path); |
5a6ac9ea | 3509 | btrfs_free_path(ppath); |
7db1c5d1 NA |
3510 | |
3511 | if (sctx->is_dev_replace && ret >= 0) { | |
3512 | int ret2; | |
3513 | ||
3514 | ret2 = sync_write_pointer_for_zoned(sctx, base + offset, | |
3515 | map->stripes[num].physical, | |
3516 | physical_end); | |
3517 | if (ret2) | |
3518 | ret = ret2; | |
3519 | } | |
3520 | ||
a2de733c AJ |
3521 | return ret < 0 ? ret : 0; |
3522 | } | |
3523 | ||
d9d181c1 | 3524 | static noinline_for_stack int scrub_chunk(struct scrub_ctx *sctx, |
a36cf8b8 | 3525 | struct btrfs_device *scrub_dev, |
a36cf8b8 | 3526 | u64 chunk_offset, u64 length, |
020d5b73 | 3527 | u64 dev_offset, |
32da5386 | 3528 | struct btrfs_block_group *cache) |
a2de733c | 3529 | { |
fb456252 | 3530 | struct btrfs_fs_info *fs_info = sctx->fs_info; |
c8bf1b67 | 3531 | struct extent_map_tree *map_tree = &fs_info->mapping_tree; |
a2de733c AJ |
3532 | struct map_lookup *map; |
3533 | struct extent_map *em; | |
3534 | int i; | |
ff023aac | 3535 | int ret = 0; |
a2de733c | 3536 | |
c8bf1b67 DS |
3537 | read_lock(&map_tree->lock); |
3538 | em = lookup_extent_mapping(map_tree, chunk_offset, 1); | |
3539 | read_unlock(&map_tree->lock); | |
a2de733c | 3540 | |
020d5b73 FM |
3541 | if (!em) { |
3542 | /* | |
3543 | * Might have been an unused block group deleted by the cleaner | |
3544 | * kthread or relocation. | |
3545 | */ | |
3546 | spin_lock(&cache->lock); | |
3547 | if (!cache->removed) | |
3548 | ret = -EINVAL; | |
3549 | spin_unlock(&cache->lock); | |
3550 | ||
3551 | return ret; | |
3552 | } | |
a2de733c | 3553 | |
95617d69 | 3554 | map = em->map_lookup; |
a2de733c AJ |
3555 | if (em->start != chunk_offset) |
3556 | goto out; | |
3557 | ||
3558 | if (em->len < length) | |
3559 | goto out; | |
3560 | ||
3561 | for (i = 0; i < map->num_stripes; ++i) { | |
a36cf8b8 | 3562 | if (map->stripes[i].dev->bdev == scrub_dev->bdev && |
859acaf1 | 3563 | map->stripes[i].physical == dev_offset) { |
a36cf8b8 | 3564 | ret = scrub_stripe(sctx, map, scrub_dev, i, |
2473d24f | 3565 | chunk_offset, length, cache); |
a2de733c AJ |
3566 | if (ret) |
3567 | goto out; | |
3568 | } | |
3569 | } | |
3570 | out: | |
3571 | free_extent_map(em); | |
3572 | ||
3573 | return ret; | |
3574 | } | |
3575 | ||
de17addc NA |
3576 | static int finish_extent_writes_for_zoned(struct btrfs_root *root, |
3577 | struct btrfs_block_group *cache) | |
3578 | { | |
3579 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
3580 | struct btrfs_trans_handle *trans; | |
3581 | ||
3582 | if (!btrfs_is_zoned(fs_info)) | |
3583 | return 0; | |
3584 | ||
3585 | btrfs_wait_block_group_reservations(cache); | |
3586 | btrfs_wait_nocow_writers(cache); | |
3587 | btrfs_wait_ordered_roots(fs_info, U64_MAX, cache->start, cache->length); | |
3588 | ||
3589 | trans = btrfs_join_transaction(root); | |
3590 | if (IS_ERR(trans)) | |
3591 | return PTR_ERR(trans); | |
3592 | return btrfs_commit_transaction(trans); | |
3593 | } | |
3594 | ||
a2de733c | 3595 | static noinline_for_stack |
a36cf8b8 | 3596 | int scrub_enumerate_chunks(struct scrub_ctx *sctx, |
32934280 | 3597 | struct btrfs_device *scrub_dev, u64 start, u64 end) |
a2de733c AJ |
3598 | { |
3599 | struct btrfs_dev_extent *dev_extent = NULL; | |
3600 | struct btrfs_path *path; | |
0b246afa JM |
3601 | struct btrfs_fs_info *fs_info = sctx->fs_info; |
3602 | struct btrfs_root *root = fs_info->dev_root; | |
a2de733c | 3603 | u64 length; |
a2de733c | 3604 | u64 chunk_offset; |
55e3a601 | 3605 | int ret = 0; |
76a8efa1 | 3606 | int ro_set; |
a2de733c AJ |
3607 | int slot; |
3608 | struct extent_buffer *l; | |
3609 | struct btrfs_key key; | |
3610 | struct btrfs_key found_key; | |
32da5386 | 3611 | struct btrfs_block_group *cache; |
ff023aac | 3612 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
a2de733c AJ |
3613 | |
3614 | path = btrfs_alloc_path(); | |
3615 | if (!path) | |
3616 | return -ENOMEM; | |
3617 | ||
e4058b54 | 3618 | path->reada = READA_FORWARD; |
a2de733c AJ |
3619 | path->search_commit_root = 1; |
3620 | path->skip_locking = 1; | |
3621 | ||
a36cf8b8 | 3622 | key.objectid = scrub_dev->devid; |
a2de733c AJ |
3623 | key.offset = 0ull; |
3624 | key.type = BTRFS_DEV_EXTENT_KEY; | |
3625 | ||
a2de733c AJ |
3626 | while (1) { |
3627 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
3628 | if (ret < 0) | |
8c51032f AJ |
3629 | break; |
3630 | if (ret > 0) { | |
3631 | if (path->slots[0] >= | |
3632 | btrfs_header_nritems(path->nodes[0])) { | |
3633 | ret = btrfs_next_leaf(root, path); | |
55e3a601 Z |
3634 | if (ret < 0) |
3635 | break; | |
3636 | if (ret > 0) { | |
3637 | ret = 0; | |
8c51032f | 3638 | break; |
55e3a601 Z |
3639 | } |
3640 | } else { | |
3641 | ret = 0; | |
8c51032f AJ |
3642 | } |
3643 | } | |
a2de733c AJ |
3644 | |
3645 | l = path->nodes[0]; | |
3646 | slot = path->slots[0]; | |
3647 | ||
3648 | btrfs_item_key_to_cpu(l, &found_key, slot); | |
3649 | ||
a36cf8b8 | 3650 | if (found_key.objectid != scrub_dev->devid) |
a2de733c AJ |
3651 | break; |
3652 | ||
962a298f | 3653 | if (found_key.type != BTRFS_DEV_EXTENT_KEY) |
a2de733c AJ |
3654 | break; |
3655 | ||
3656 | if (found_key.offset >= end) | |
3657 | break; | |
3658 | ||
3659 | if (found_key.offset < key.offset) | |
3660 | break; | |
3661 | ||
3662 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
3663 | length = btrfs_dev_extent_length(l, dev_extent); | |
3664 | ||
ced96edc QW |
3665 | if (found_key.offset + length <= start) |
3666 | goto skip; | |
a2de733c | 3667 | |
a2de733c AJ |
3668 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); |
3669 | ||
3670 | /* | |
3671 | * get a reference on the corresponding block group to prevent | |
3672 | * the chunk from going away while we scrub it | |
3673 | */ | |
3674 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
ced96edc QW |
3675 | |
3676 | /* some chunks are removed but not committed to disk yet, | |
3677 | * continue scrubbing */ | |
3678 | if (!cache) | |
3679 | goto skip; | |
3680 | ||
78ce9fc2 NA |
3681 | if (sctx->is_dev_replace && btrfs_is_zoned(fs_info)) { |
3682 | spin_lock(&cache->lock); | |
3683 | if (!cache->to_copy) { | |
3684 | spin_unlock(&cache->lock); | |
3685 | ro_set = 0; | |
3686 | goto done; | |
3687 | } | |
3688 | spin_unlock(&cache->lock); | |
3689 | } | |
3690 | ||
2473d24f FM |
3691 | /* |
3692 | * Make sure that while we are scrubbing the corresponding block | |
3693 | * group doesn't get its logical address and its device extents | |
3694 | * reused for another block group, which can possibly be of a | |
3695 | * different type and different profile. We do this to prevent | |
3696 | * false error detections and crashes due to bogus attempts to | |
3697 | * repair extents. | |
3698 | */ | |
3699 | spin_lock(&cache->lock); | |
3700 | if (cache->removed) { | |
3701 | spin_unlock(&cache->lock); | |
3702 | btrfs_put_block_group(cache); | |
3703 | goto skip; | |
3704 | } | |
6b7304af | 3705 | btrfs_freeze_block_group(cache); |
2473d24f FM |
3706 | spin_unlock(&cache->lock); |
3707 | ||
55e3a601 Z |
3708 | /* |
3709 | * we need call btrfs_inc_block_group_ro() with scrubs_paused, | |
3710 | * to avoid deadlock caused by: | |
3711 | * btrfs_inc_block_group_ro() | |
3712 | * -> btrfs_wait_for_commit() | |
3713 | * -> btrfs_commit_transaction() | |
3714 | * -> btrfs_scrub_pause() | |
3715 | */ | |
3716 | scrub_pause_on(fs_info); | |
b12de528 QW |
3717 | |
3718 | /* | |
3719 | * Don't do chunk preallocation for scrub. | |
3720 | * | |
3721 | * This is especially important for SYSTEM bgs, or we can hit | |
3722 | * -EFBIG from btrfs_finish_chunk_alloc() like: | |
3723 | * 1. The only SYSTEM bg is marked RO. | |
3724 | * Since SYSTEM bg is small, that's pretty common. | |
3725 | * 2. New SYSTEM bg will be allocated | |
3726 | * Due to regular version will allocate new chunk. | |
3727 | * 3. New SYSTEM bg is empty and will get cleaned up | |
3728 | * Before cleanup really happens, it's marked RO again. | |
3729 | * 4. Empty SYSTEM bg get scrubbed | |
3730 | * We go back to 2. | |
3731 | * | |
3732 | * This can easily boost the amount of SYSTEM chunks if cleaner | |
3733 | * thread can't be triggered fast enough, and use up all space | |
3734 | * of btrfs_super_block::sys_chunk_array | |
1bbb97b8 QW |
3735 | * |
3736 | * While for dev replace, we need to try our best to mark block | |
3737 | * group RO, to prevent race between: | |
3738 | * - Write duplication | |
3739 | * Contains latest data | |
3740 | * - Scrub copy | |
3741 | * Contains data from commit tree | |
3742 | * | |
3743 | * If target block group is not marked RO, nocow writes can | |
3744 | * be overwritten by scrub copy, causing data corruption. | |
3745 | * So for dev-replace, it's not allowed to continue if a block | |
3746 | * group is not RO. | |
b12de528 | 3747 | */ |
1bbb97b8 | 3748 | ret = btrfs_inc_block_group_ro(cache, sctx->is_dev_replace); |
de17addc NA |
3749 | if (!ret && sctx->is_dev_replace) { |
3750 | ret = finish_extent_writes_for_zoned(root, cache); | |
3751 | if (ret) { | |
3752 | btrfs_dec_block_group_ro(cache); | |
3753 | scrub_pause_off(fs_info); | |
3754 | btrfs_put_block_group(cache); | |
3755 | break; | |
3756 | } | |
3757 | } | |
3758 | ||
76a8efa1 Z |
3759 | if (ret == 0) { |
3760 | ro_set = 1; | |
1bbb97b8 | 3761 | } else if (ret == -ENOSPC && !sctx->is_dev_replace) { |
76a8efa1 Z |
3762 | /* |
3763 | * btrfs_inc_block_group_ro return -ENOSPC when it | |
3764 | * failed in creating new chunk for metadata. | |
1bbb97b8 | 3765 | * It is not a problem for scrub, because |
76a8efa1 Z |
3766 | * metadata are always cowed, and our scrub paused |
3767 | * commit_transactions. | |
3768 | */ | |
3769 | ro_set = 0; | |
3770 | } else { | |
5d163e0e | 3771 | btrfs_warn(fs_info, |
913e1535 | 3772 | "failed setting block group ro: %d", ret); |
6b7304af | 3773 | btrfs_unfreeze_block_group(cache); |
55e3a601 | 3774 | btrfs_put_block_group(cache); |
1bbb97b8 | 3775 | scrub_pause_off(fs_info); |
55e3a601 Z |
3776 | break; |
3777 | } | |
3778 | ||
1bbb97b8 QW |
3779 | /* |
3780 | * Now the target block is marked RO, wait for nocow writes to | |
3781 | * finish before dev-replace. | |
3782 | * COW is fine, as COW never overwrites extents in commit tree. | |
3783 | */ | |
3784 | if (sctx->is_dev_replace) { | |
3785 | btrfs_wait_nocow_writers(cache); | |
3786 | btrfs_wait_ordered_roots(fs_info, U64_MAX, cache->start, | |
3787 | cache->length); | |
3788 | } | |
3789 | ||
3790 | scrub_pause_off(fs_info); | |
3ec17a67 | 3791 | down_write(&dev_replace->rwsem); |
ff023aac SB |
3792 | dev_replace->cursor_right = found_key.offset + length; |
3793 | dev_replace->cursor_left = found_key.offset; | |
3794 | dev_replace->item_needs_writeback = 1; | |
cb5583dd DS |
3795 | up_write(&dev_replace->rwsem); |
3796 | ||
8c204c96 | 3797 | ret = scrub_chunk(sctx, scrub_dev, chunk_offset, length, |
32934280 | 3798 | found_key.offset, cache); |
ff023aac SB |
3799 | |
3800 | /* | |
3801 | * flush, submit all pending read and write bios, afterwards | |
3802 | * wait for them. | |
3803 | * Note that in the dev replace case, a read request causes | |
3804 | * write requests that are submitted in the read completion | |
3805 | * worker. Therefore in the current situation, it is required | |
3806 | * that all write requests are flushed, so that all read and | |
3807 | * write requests are really completed when bios_in_flight | |
3808 | * changes to 0. | |
3809 | */ | |
2073c4c2 | 3810 | sctx->flush_all_writes = true; |
ff023aac | 3811 | scrub_submit(sctx); |
3fb99303 | 3812 | mutex_lock(&sctx->wr_lock); |
ff023aac | 3813 | scrub_wr_submit(sctx); |
3fb99303 | 3814 | mutex_unlock(&sctx->wr_lock); |
ff023aac SB |
3815 | |
3816 | wait_event(sctx->list_wait, | |
3817 | atomic_read(&sctx->bios_in_flight) == 0); | |
b708ce96 Z |
3818 | |
3819 | scrub_pause_on(fs_info); | |
12cf9372 WS |
3820 | |
3821 | /* | |
3822 | * must be called before we decrease @scrub_paused. | |
3823 | * make sure we don't block transaction commit while | |
3824 | * we are waiting pending workers finished. | |
3825 | */ | |
ff023aac SB |
3826 | wait_event(sctx->list_wait, |
3827 | atomic_read(&sctx->workers_pending) == 0); | |
2073c4c2 | 3828 | sctx->flush_all_writes = false; |
12cf9372 | 3829 | |
b708ce96 | 3830 | scrub_pause_off(fs_info); |
ff023aac | 3831 | |
78ce9fc2 NA |
3832 | if (sctx->is_dev_replace && |
3833 | !btrfs_finish_block_group_to_copy(dev_replace->srcdev, | |
3834 | cache, found_key.offset)) | |
3835 | ro_set = 0; | |
3836 | ||
3837 | done: | |
3ec17a67 | 3838 | down_write(&dev_replace->rwsem); |
1a1a8b73 FM |
3839 | dev_replace->cursor_left = dev_replace->cursor_right; |
3840 | dev_replace->item_needs_writeback = 1; | |
3ec17a67 | 3841 | up_write(&dev_replace->rwsem); |
1a1a8b73 | 3842 | |
76a8efa1 | 3843 | if (ro_set) |
2ff7e61e | 3844 | btrfs_dec_block_group_ro(cache); |
ff023aac | 3845 | |
758f2dfc FM |
3846 | /* |
3847 | * We might have prevented the cleaner kthread from deleting | |
3848 | * this block group if it was already unused because we raced | |
3849 | * and set it to RO mode first. So add it back to the unused | |
3850 | * list, otherwise it might not ever be deleted unless a manual | |
3851 | * balance is triggered or it becomes used and unused again. | |
3852 | */ | |
3853 | spin_lock(&cache->lock); | |
3854 | if (!cache->removed && !cache->ro && cache->reserved == 0 && | |
bf38be65 | 3855 | cache->used == 0) { |
758f2dfc | 3856 | spin_unlock(&cache->lock); |
6e80d4f8 DZ |
3857 | if (btrfs_test_opt(fs_info, DISCARD_ASYNC)) |
3858 | btrfs_discard_queue_work(&fs_info->discard_ctl, | |
3859 | cache); | |
3860 | else | |
3861 | btrfs_mark_bg_unused(cache); | |
758f2dfc FM |
3862 | } else { |
3863 | spin_unlock(&cache->lock); | |
3864 | } | |
3865 | ||
6b7304af | 3866 | btrfs_unfreeze_block_group(cache); |
a2de733c AJ |
3867 | btrfs_put_block_group(cache); |
3868 | if (ret) | |
3869 | break; | |
32934280 | 3870 | if (sctx->is_dev_replace && |
af1be4f8 | 3871 | atomic64_read(&dev_replace->num_write_errors) > 0) { |
ff023aac SB |
3872 | ret = -EIO; |
3873 | break; | |
3874 | } | |
3875 | if (sctx->stat.malloc_errors > 0) { | |
3876 | ret = -ENOMEM; | |
3877 | break; | |
3878 | } | |
ced96edc | 3879 | skip: |
a2de733c | 3880 | key.offset = found_key.offset + length; |
71267333 | 3881 | btrfs_release_path(path); |
a2de733c AJ |
3882 | } |
3883 | ||
a2de733c | 3884 | btrfs_free_path(path); |
8c51032f | 3885 | |
55e3a601 | 3886 | return ret; |
a2de733c AJ |
3887 | } |
3888 | ||
a36cf8b8 SB |
3889 | static noinline_for_stack int scrub_supers(struct scrub_ctx *sctx, |
3890 | struct btrfs_device *scrub_dev) | |
a2de733c AJ |
3891 | { |
3892 | int i; | |
3893 | u64 bytenr; | |
3894 | u64 gen; | |
3895 | int ret; | |
0b246afa | 3896 | struct btrfs_fs_info *fs_info = sctx->fs_info; |
a2de733c | 3897 | |
0b246afa | 3898 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
fbabd4a3 | 3899 | return -EROFS; |
79787eaa | 3900 | |
5f546063 | 3901 | /* Seed devices of a new filesystem has their own generation. */ |
0b246afa | 3902 | if (scrub_dev->fs_devices != fs_info->fs_devices) |
5f546063 MX |
3903 | gen = scrub_dev->generation; |
3904 | else | |
0b246afa | 3905 | gen = fs_info->last_trans_committed; |
a2de733c AJ |
3906 | |
3907 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { | |
3908 | bytenr = btrfs_sb_offset(i); | |
935e5cc9 MX |
3909 | if (bytenr + BTRFS_SUPER_INFO_SIZE > |
3910 | scrub_dev->commit_total_bytes) | |
a2de733c | 3911 | break; |
12659251 NA |
3912 | if (!btrfs_check_super_location(scrub_dev, bytenr)) |
3913 | continue; | |
a2de733c | 3914 | |
d9d181c1 | 3915 | ret = scrub_pages(sctx, bytenr, BTRFS_SUPER_INFO_SIZE, bytenr, |
a36cf8b8 | 3916 | scrub_dev, BTRFS_EXTENT_FLAG_SUPER, gen, i, |
96e63a45 | 3917 | NULL, bytenr); |
a2de733c AJ |
3918 | if (ret) |
3919 | return ret; | |
3920 | } | |
b6bfebc1 | 3921 | wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0); |
a2de733c AJ |
3922 | |
3923 | return 0; | |
3924 | } | |
3925 | ||
e89c4a9c JB |
3926 | static void scrub_workers_put(struct btrfs_fs_info *fs_info) |
3927 | { | |
3928 | if (refcount_dec_and_mutex_lock(&fs_info->scrub_workers_refcnt, | |
3929 | &fs_info->scrub_lock)) { | |
3930 | struct btrfs_workqueue *scrub_workers = NULL; | |
3931 | struct btrfs_workqueue *scrub_wr_comp = NULL; | |
3932 | struct btrfs_workqueue *scrub_parity = NULL; | |
3933 | ||
3934 | scrub_workers = fs_info->scrub_workers; | |
3935 | scrub_wr_comp = fs_info->scrub_wr_completion_workers; | |
3936 | scrub_parity = fs_info->scrub_parity_workers; | |
3937 | ||
3938 | fs_info->scrub_workers = NULL; | |
3939 | fs_info->scrub_wr_completion_workers = NULL; | |
3940 | fs_info->scrub_parity_workers = NULL; | |
3941 | mutex_unlock(&fs_info->scrub_lock); | |
3942 | ||
3943 | btrfs_destroy_workqueue(scrub_workers); | |
3944 | btrfs_destroy_workqueue(scrub_wr_comp); | |
3945 | btrfs_destroy_workqueue(scrub_parity); | |
3946 | } | |
3947 | } | |
3948 | ||
a2de733c AJ |
3949 | /* |
3950 | * get a reference count on fs_info->scrub_workers. start worker if necessary | |
3951 | */ | |
ff023aac SB |
3952 | static noinline_for_stack int scrub_workers_get(struct btrfs_fs_info *fs_info, |
3953 | int is_dev_replace) | |
a2de733c | 3954 | { |
e89c4a9c JB |
3955 | struct btrfs_workqueue *scrub_workers = NULL; |
3956 | struct btrfs_workqueue *scrub_wr_comp = NULL; | |
3957 | struct btrfs_workqueue *scrub_parity = NULL; | |
6f011058 | 3958 | unsigned int flags = WQ_FREEZABLE | WQ_UNBOUND; |
0339ef2f | 3959 | int max_active = fs_info->thread_pool_size; |
e89c4a9c | 3960 | int ret = -ENOMEM; |
a2de733c | 3961 | |
e89c4a9c JB |
3962 | if (refcount_inc_not_zero(&fs_info->scrub_workers_refcnt)) |
3963 | return 0; | |
eb4318e5 | 3964 | |
e89c4a9c JB |
3965 | scrub_workers = btrfs_alloc_workqueue(fs_info, "scrub", flags, |
3966 | is_dev_replace ? 1 : max_active, 4); | |
3967 | if (!scrub_workers) | |
3968 | goto fail_scrub_workers; | |
e82afc52 | 3969 | |
e89c4a9c | 3970 | scrub_wr_comp = btrfs_alloc_workqueue(fs_info, "scrubwrc", flags, |
20b2e302 | 3971 | max_active, 2); |
e89c4a9c JB |
3972 | if (!scrub_wr_comp) |
3973 | goto fail_scrub_wr_completion_workers; | |
ff09c4ca | 3974 | |
e89c4a9c JB |
3975 | scrub_parity = btrfs_alloc_workqueue(fs_info, "scrubparity", flags, |
3976 | max_active, 2); | |
3977 | if (!scrub_parity) | |
3978 | goto fail_scrub_parity_workers; | |
3979 | ||
3980 | mutex_lock(&fs_info->scrub_lock); | |
3981 | if (refcount_read(&fs_info->scrub_workers_refcnt) == 0) { | |
3982 | ASSERT(fs_info->scrub_workers == NULL && | |
3983 | fs_info->scrub_wr_completion_workers == NULL && | |
3984 | fs_info->scrub_parity_workers == NULL); | |
3985 | fs_info->scrub_workers = scrub_workers; | |
3986 | fs_info->scrub_wr_completion_workers = scrub_wr_comp; | |
3987 | fs_info->scrub_parity_workers = scrub_parity; | |
ff09c4ca | 3988 | refcount_set(&fs_info->scrub_workers_refcnt, 1); |
e89c4a9c JB |
3989 | mutex_unlock(&fs_info->scrub_lock); |
3990 | return 0; | |
632dd772 | 3991 | } |
e89c4a9c JB |
3992 | /* Other thread raced in and created the workers for us */ |
3993 | refcount_inc(&fs_info->scrub_workers_refcnt); | |
3994 | mutex_unlock(&fs_info->scrub_lock); | |
e82afc52 | 3995 | |
e89c4a9c JB |
3996 | ret = 0; |
3997 | btrfs_destroy_workqueue(scrub_parity); | |
e82afc52 | 3998 | fail_scrub_parity_workers: |
e89c4a9c | 3999 | btrfs_destroy_workqueue(scrub_wr_comp); |
e82afc52 | 4000 | fail_scrub_wr_completion_workers: |
e89c4a9c | 4001 | btrfs_destroy_workqueue(scrub_workers); |
e82afc52 | 4002 | fail_scrub_workers: |
e89c4a9c | 4003 | return ret; |
a2de733c AJ |
4004 | } |
4005 | ||
aa1b8cd4 SB |
4006 | int btrfs_scrub_dev(struct btrfs_fs_info *fs_info, u64 devid, u64 start, |
4007 | u64 end, struct btrfs_scrub_progress *progress, | |
63a212ab | 4008 | int readonly, int is_dev_replace) |
a2de733c | 4009 | { |
d9d181c1 | 4010 | struct scrub_ctx *sctx; |
a2de733c AJ |
4011 | int ret; |
4012 | struct btrfs_device *dev; | |
a5fb1142 | 4013 | unsigned int nofs_flag; |
a2de733c | 4014 | |
aa1b8cd4 | 4015 | if (btrfs_fs_closing(fs_info)) |
6c3abeda | 4016 | return -EAGAIN; |
a2de733c | 4017 | |
da17066c | 4018 | if (fs_info->nodesize > BTRFS_STRIPE_LEN) { |
b5d67f64 SB |
4019 | /* |
4020 | * in this case scrub is unable to calculate the checksum | |
4021 | * the way scrub is implemented. Do not handle this | |
4022 | * situation at all because it won't ever happen. | |
4023 | */ | |
efe120a0 FH |
4024 | btrfs_err(fs_info, |
4025 | "scrub: size assumption nodesize <= BTRFS_STRIPE_LEN (%d <= %d) fails", | |
da17066c JM |
4026 | fs_info->nodesize, |
4027 | BTRFS_STRIPE_LEN); | |
b5d67f64 SB |
4028 | return -EINVAL; |
4029 | } | |
4030 | ||
da17066c | 4031 | if (fs_info->nodesize > |
7a9e9987 | 4032 | PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK || |
da17066c | 4033 | fs_info->sectorsize > PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK) { |
7a9e9987 SB |
4034 | /* |
4035 | * would exhaust the array bounds of pagev member in | |
4036 | * struct scrub_block | |
4037 | */ | |
5d163e0e JM |
4038 | btrfs_err(fs_info, |
4039 | "scrub: size assumption nodesize and sectorsize <= SCRUB_MAX_PAGES_PER_BLOCK (%d <= %d && %d <= %d) fails", | |
da17066c | 4040 | fs_info->nodesize, |
7a9e9987 | 4041 | SCRUB_MAX_PAGES_PER_BLOCK, |
da17066c | 4042 | fs_info->sectorsize, |
7a9e9987 SB |
4043 | SCRUB_MAX_PAGES_PER_BLOCK); |
4044 | return -EINVAL; | |
4045 | } | |
4046 | ||
0e94c4f4 DS |
4047 | /* Allocate outside of device_list_mutex */ |
4048 | sctx = scrub_setup_ctx(fs_info, is_dev_replace); | |
4049 | if (IS_ERR(sctx)) | |
4050 | return PTR_ERR(sctx); | |
a2de733c | 4051 | |
e89c4a9c JB |
4052 | ret = scrub_workers_get(fs_info, is_dev_replace); |
4053 | if (ret) | |
4054 | goto out_free_ctx; | |
4055 | ||
aa1b8cd4 | 4056 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
b2598edf | 4057 | dev = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL); |
e6e674bd AJ |
4058 | if (!dev || (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) && |
4059 | !is_dev_replace)) { | |
aa1b8cd4 | 4060 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
0e94c4f4 | 4061 | ret = -ENODEV; |
e89c4a9c | 4062 | goto out; |
a2de733c | 4063 | } |
a2de733c | 4064 | |
ebbede42 AJ |
4065 | if (!is_dev_replace && !readonly && |
4066 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) { | |
5d68da3b | 4067 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
a4852cf2 DS |
4068 | btrfs_err_in_rcu(fs_info, |
4069 | "scrub on devid %llu: filesystem on %s is not writable", | |
4070 | devid, rcu_str_deref(dev->name)); | |
0e94c4f4 | 4071 | ret = -EROFS; |
e89c4a9c | 4072 | goto out; |
5d68da3b MX |
4073 | } |
4074 | ||
3b7a016f | 4075 | mutex_lock(&fs_info->scrub_lock); |
e12c9621 | 4076 | if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || |
401e29c1 | 4077 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &dev->dev_state)) { |
a2de733c | 4078 | mutex_unlock(&fs_info->scrub_lock); |
aa1b8cd4 | 4079 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
0e94c4f4 | 4080 | ret = -EIO; |
e89c4a9c | 4081 | goto out; |
a2de733c AJ |
4082 | } |
4083 | ||
cb5583dd | 4084 | down_read(&fs_info->dev_replace.rwsem); |
cadbc0a0 | 4085 | if (dev->scrub_ctx || |
8dabb742 SB |
4086 | (!is_dev_replace && |
4087 | btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))) { | |
cb5583dd | 4088 | up_read(&fs_info->dev_replace.rwsem); |
a2de733c | 4089 | mutex_unlock(&fs_info->scrub_lock); |
aa1b8cd4 | 4090 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
0e94c4f4 | 4091 | ret = -EINPROGRESS; |
e89c4a9c | 4092 | goto out; |
a2de733c | 4093 | } |
cb5583dd | 4094 | up_read(&fs_info->dev_replace.rwsem); |
3b7a016f | 4095 | |
d9d181c1 | 4096 | sctx->readonly = readonly; |
cadbc0a0 | 4097 | dev->scrub_ctx = sctx; |
3cb0929a | 4098 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
a2de733c | 4099 | |
3cb0929a WS |
4100 | /* |
4101 | * checking @scrub_pause_req here, we can avoid | |
4102 | * race between committing transaction and scrubbing. | |
4103 | */ | |
cb7ab021 | 4104 | __scrub_blocked_if_needed(fs_info); |
a2de733c AJ |
4105 | atomic_inc(&fs_info->scrubs_running); |
4106 | mutex_unlock(&fs_info->scrub_lock); | |
a2de733c | 4107 | |
a5fb1142 FM |
4108 | /* |
4109 | * In order to avoid deadlock with reclaim when there is a transaction | |
4110 | * trying to pause scrub, make sure we use GFP_NOFS for all the | |
4111 | * allocations done at btrfs_scrub_pages() and scrub_pages_for_parity() | |
4112 | * invoked by our callees. The pausing request is done when the | |
4113 | * transaction commit starts, and it blocks the transaction until scrub | |
4114 | * is paused (done at specific points at scrub_stripe() or right above | |
4115 | * before incrementing fs_info->scrubs_running). | |
4116 | */ | |
4117 | nofs_flag = memalloc_nofs_save(); | |
ff023aac | 4118 | if (!is_dev_replace) { |
d1e14420 | 4119 | btrfs_info(fs_info, "scrub: started on devid %llu", devid); |
9b011adf WS |
4120 | /* |
4121 | * by holding device list mutex, we can | |
4122 | * kick off writing super in log tree sync. | |
4123 | */ | |
3cb0929a | 4124 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
ff023aac | 4125 | ret = scrub_supers(sctx, dev); |
3cb0929a | 4126 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
ff023aac | 4127 | } |
a2de733c AJ |
4128 | |
4129 | if (!ret) | |
32934280 | 4130 | ret = scrub_enumerate_chunks(sctx, dev, start, end); |
a5fb1142 | 4131 | memalloc_nofs_restore(nofs_flag); |
a2de733c | 4132 | |
b6bfebc1 | 4133 | wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0); |
a2de733c AJ |
4134 | atomic_dec(&fs_info->scrubs_running); |
4135 | wake_up(&fs_info->scrub_pause_wait); | |
4136 | ||
b6bfebc1 | 4137 | wait_event(sctx->list_wait, atomic_read(&sctx->workers_pending) == 0); |
0ef8e451 | 4138 | |
a2de733c | 4139 | if (progress) |
d9d181c1 | 4140 | memcpy(progress, &sctx->stat, sizeof(*progress)); |
a2de733c | 4141 | |
d1e14420 AJ |
4142 | if (!is_dev_replace) |
4143 | btrfs_info(fs_info, "scrub: %s on devid %llu with status: %d", | |
4144 | ret ? "not finished" : "finished", devid, ret); | |
4145 | ||
a2de733c | 4146 | mutex_lock(&fs_info->scrub_lock); |
cadbc0a0 | 4147 | dev->scrub_ctx = NULL; |
a2de733c AJ |
4148 | mutex_unlock(&fs_info->scrub_lock); |
4149 | ||
e89c4a9c | 4150 | scrub_workers_put(fs_info); |
f55985f4 | 4151 | scrub_put_ctx(sctx); |
a2de733c | 4152 | |
0e94c4f4 | 4153 | return ret; |
e89c4a9c JB |
4154 | out: |
4155 | scrub_workers_put(fs_info); | |
0e94c4f4 DS |
4156 | out_free_ctx: |
4157 | scrub_free_ctx(sctx); | |
4158 | ||
a2de733c AJ |
4159 | return ret; |
4160 | } | |
4161 | ||
2ff7e61e | 4162 | void btrfs_scrub_pause(struct btrfs_fs_info *fs_info) |
a2de733c | 4163 | { |
a2de733c AJ |
4164 | mutex_lock(&fs_info->scrub_lock); |
4165 | atomic_inc(&fs_info->scrub_pause_req); | |
4166 | while (atomic_read(&fs_info->scrubs_paused) != | |
4167 | atomic_read(&fs_info->scrubs_running)) { | |
4168 | mutex_unlock(&fs_info->scrub_lock); | |
4169 | wait_event(fs_info->scrub_pause_wait, | |
4170 | atomic_read(&fs_info->scrubs_paused) == | |
4171 | atomic_read(&fs_info->scrubs_running)); | |
4172 | mutex_lock(&fs_info->scrub_lock); | |
4173 | } | |
4174 | mutex_unlock(&fs_info->scrub_lock); | |
a2de733c AJ |
4175 | } |
4176 | ||
2ff7e61e | 4177 | void btrfs_scrub_continue(struct btrfs_fs_info *fs_info) |
a2de733c | 4178 | { |
a2de733c AJ |
4179 | atomic_dec(&fs_info->scrub_pause_req); |
4180 | wake_up(&fs_info->scrub_pause_wait); | |
a2de733c AJ |
4181 | } |
4182 | ||
aa1b8cd4 | 4183 | int btrfs_scrub_cancel(struct btrfs_fs_info *fs_info) |
a2de733c | 4184 | { |
a2de733c AJ |
4185 | mutex_lock(&fs_info->scrub_lock); |
4186 | if (!atomic_read(&fs_info->scrubs_running)) { | |
4187 | mutex_unlock(&fs_info->scrub_lock); | |
4188 | return -ENOTCONN; | |
4189 | } | |
4190 | ||
4191 | atomic_inc(&fs_info->scrub_cancel_req); | |
4192 | while (atomic_read(&fs_info->scrubs_running)) { | |
4193 | mutex_unlock(&fs_info->scrub_lock); | |
4194 | wait_event(fs_info->scrub_pause_wait, | |
4195 | atomic_read(&fs_info->scrubs_running) == 0); | |
4196 | mutex_lock(&fs_info->scrub_lock); | |
4197 | } | |
4198 | atomic_dec(&fs_info->scrub_cancel_req); | |
4199 | mutex_unlock(&fs_info->scrub_lock); | |
4200 | ||
4201 | return 0; | |
4202 | } | |
4203 | ||
163e97ee | 4204 | int btrfs_scrub_cancel_dev(struct btrfs_device *dev) |
49b25e05 | 4205 | { |
163e97ee | 4206 | struct btrfs_fs_info *fs_info = dev->fs_info; |
d9d181c1 | 4207 | struct scrub_ctx *sctx; |
a2de733c AJ |
4208 | |
4209 | mutex_lock(&fs_info->scrub_lock); | |
cadbc0a0 | 4210 | sctx = dev->scrub_ctx; |
d9d181c1 | 4211 | if (!sctx) { |
a2de733c AJ |
4212 | mutex_unlock(&fs_info->scrub_lock); |
4213 | return -ENOTCONN; | |
4214 | } | |
d9d181c1 | 4215 | atomic_inc(&sctx->cancel_req); |
cadbc0a0 | 4216 | while (dev->scrub_ctx) { |
a2de733c AJ |
4217 | mutex_unlock(&fs_info->scrub_lock); |
4218 | wait_event(fs_info->scrub_pause_wait, | |
cadbc0a0 | 4219 | dev->scrub_ctx == NULL); |
a2de733c AJ |
4220 | mutex_lock(&fs_info->scrub_lock); |
4221 | } | |
4222 | mutex_unlock(&fs_info->scrub_lock); | |
4223 | ||
4224 | return 0; | |
4225 | } | |
1623edeb | 4226 | |
2ff7e61e | 4227 | int btrfs_scrub_progress(struct btrfs_fs_info *fs_info, u64 devid, |
a2de733c AJ |
4228 | struct btrfs_scrub_progress *progress) |
4229 | { | |
4230 | struct btrfs_device *dev; | |
d9d181c1 | 4231 | struct scrub_ctx *sctx = NULL; |
a2de733c | 4232 | |
0b246afa | 4233 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
b2598edf | 4234 | dev = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL); |
a2de733c | 4235 | if (dev) |
cadbc0a0 | 4236 | sctx = dev->scrub_ctx; |
d9d181c1 SB |
4237 | if (sctx) |
4238 | memcpy(progress, &sctx->stat, sizeof(*progress)); | |
0b246afa | 4239 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
a2de733c | 4240 | |
d9d181c1 | 4241 | return dev ? (sctx ? 0 : -ENOTCONN) : -ENODEV; |
a2de733c | 4242 | } |
ff023aac SB |
4243 | |
4244 | static void scrub_remap_extent(struct btrfs_fs_info *fs_info, | |
fa485d21 | 4245 | u64 extent_logical, u32 extent_len, |
ff023aac SB |
4246 | u64 *extent_physical, |
4247 | struct btrfs_device **extent_dev, | |
4248 | int *extent_mirror_num) | |
4249 | { | |
4250 | u64 mapped_length; | |
4251 | struct btrfs_bio *bbio = NULL; | |
4252 | int ret; | |
4253 | ||
4254 | mapped_length = extent_len; | |
cf8cddd3 | 4255 | ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, extent_logical, |
ff023aac SB |
4256 | &mapped_length, &bbio, 0); |
4257 | if (ret || !bbio || mapped_length < extent_len || | |
4258 | !bbio->stripes[0].dev->bdev) { | |
6e9606d2 | 4259 | btrfs_put_bbio(bbio); |
ff023aac SB |
4260 | return; |
4261 | } | |
4262 | ||
4263 | *extent_physical = bbio->stripes[0].physical; | |
4264 | *extent_mirror_num = bbio->mirror_num; | |
4265 | *extent_dev = bbio->stripes[0].dev; | |
6e9606d2 | 4266 | btrfs_put_bbio(bbio); |
ff023aac | 4267 | } |