]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * raid1.c : Multiple Devices driver for Linux | |
3 | * | |
4 | * Copyright (C) 1999, 2000, 2001 Ingo Molnar, Red Hat | |
5 | * | |
6 | * Copyright (C) 1996, 1997, 1998 Ingo Molnar, Miguel de Icaza, Gadi Oxman | |
7 | * | |
8 | * RAID-1 management functions. | |
9 | * | |
10 | * Better read-balancing code written by Mika Kuoppala <miku@iki.fi>, 2000 | |
11 | * | |
96de0e25 | 12 | * Fixes to reconstruction by Jakob Østergaard" <jakob@ostenfeld.dk> |
1da177e4 LT |
13 | * Various fixes by Neil Brown <neilb@cse.unsw.edu.au> |
14 | * | |
191ea9b2 N |
15 | * Changes by Peter T. Breuer <ptb@it.uc3m.es> 31/1/2003 to support |
16 | * bitmapped intelligence in resync: | |
17 | * | |
18 | * - bitmap marked during normal i/o | |
19 | * - bitmap used to skip nondirty blocks during sync | |
20 | * | |
21 | * Additions to bitmap code, (C) 2003-2004 Paul Clements, SteelEye Technology: | |
22 | * - persistent bitmap code | |
23 | * | |
1da177e4 LT |
24 | * This program is free software; you can redistribute it and/or modify |
25 | * it under the terms of the GNU General Public License as published by | |
26 | * the Free Software Foundation; either version 2, or (at your option) | |
27 | * any later version. | |
28 | * | |
29 | * You should have received a copy of the GNU General Public License | |
30 | * (for example /usr/src/linux/COPYING); if not, write to the Free | |
31 | * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
32 | */ | |
33 | ||
5a0e3ad6 | 34 | #include <linux/slab.h> |
25570727 | 35 | #include <linux/delay.h> |
bff61975 | 36 | #include <linux/blkdev.h> |
056075c7 | 37 | #include <linux/module.h> |
bff61975 | 38 | #include <linux/seq_file.h> |
8bda470e | 39 | #include <linux/ratelimit.h> |
3f07c014 | 40 | |
109e3765 | 41 | #include <trace/events/block.h> |
3f07c014 | 42 | |
43b2e5d8 | 43 | #include "md.h" |
ef740c37 | 44 | #include "raid1.h" |
935fe098 | 45 | #include "md-bitmap.h" |
191ea9b2 | 46 | |
394ed8e4 SL |
47 | #define UNSUPPORTED_MDDEV_FLAGS \ |
48 | ((1L << MD_HAS_JOURNAL) | \ | |
ea0213e0 | 49 | (1L << MD_JOURNAL_CLEAN) | \ |
ddc08823 PB |
50 | (1L << MD_HAS_PPL) | \ |
51 | (1L << MD_HAS_MULTIPLE_PPLS)) | |
394ed8e4 | 52 | |
1da177e4 LT |
53 | /* |
54 | * Number of guaranteed r1bios in case of extreme VM load: | |
55 | */ | |
56 | #define NR_RAID1_BIOS 256 | |
57 | ||
473e87ce JB |
58 | /* when we get a read error on a read-only array, we redirect to another |
59 | * device without failing the first device, or trying to over-write to | |
60 | * correct the read error. To keep track of bad blocks on a per-bio | |
61 | * level, we store IO_BLOCKED in the appropriate 'bios' pointer | |
62 | */ | |
63 | #define IO_BLOCKED ((struct bio *)1) | |
64 | /* When we successfully write to a known bad-block, we need to remove the | |
65 | * bad-block marking which must be done from process context. So we record | |
66 | * the success by setting devs[n].bio to IO_MADE_GOOD | |
67 | */ | |
68 | #define IO_MADE_GOOD ((struct bio *)2) | |
69 | ||
70 | #define BIO_SPECIAL(bio) ((unsigned long)bio <= 2) | |
71 | ||
34db0cd6 N |
72 | /* When there are this many requests queue to be written by |
73 | * the raid1 thread, we become 'congested' to provide back-pressure | |
74 | * for writeback. | |
75 | */ | |
76 | static int max_queued_requests = 1024; | |
1da177e4 | 77 | |
fd76863e | 78 | static void allow_barrier(struct r1conf *conf, sector_t sector_nr); |
79 | static void lower_barrier(struct r1conf *conf, sector_t sector_nr); | |
1da177e4 | 80 | |
578b54ad N |
81 | #define raid1_log(md, fmt, args...) \ |
82 | do { if ((md)->queue) blk_add_trace_msg((md)->queue, "raid1 " fmt, ##args); } while (0) | |
83 | ||
fb0eb5df ML |
84 | #include "raid1-10.c" |
85 | ||
98d30c58 ML |
86 | /* |
87 | * for resync bio, r1bio pointer can be retrieved from the per-bio | |
88 | * 'struct resync_pages'. | |
89 | */ | |
90 | static inline struct r1bio *get_resync_r1bio(struct bio *bio) | |
91 | { | |
92 | return get_resync_pages(bio)->raid_bio; | |
93 | } | |
94 | ||
dd0fc66f | 95 | static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 LT |
96 | { |
97 | struct pool_info *pi = data; | |
9f2c9d12 | 98 | int size = offsetof(struct r1bio, bios[pi->raid_disks]); |
1da177e4 LT |
99 | |
100 | /* allocate a r1bio with room for raid_disks entries in the bios array */ | |
7eaceacc | 101 | return kzalloc(size, gfp_flags); |
1da177e4 LT |
102 | } |
103 | ||
104 | static void r1bio_pool_free(void *r1_bio, void *data) | |
105 | { | |
106 | kfree(r1_bio); | |
107 | } | |
108 | ||
8e005f7c | 109 | #define RESYNC_DEPTH 32 |
1da177e4 | 110 | #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9) |
8e005f7c | 111 | #define RESYNC_WINDOW (RESYNC_BLOCK_SIZE * RESYNC_DEPTH) |
112 | #define RESYNC_WINDOW_SECTORS (RESYNC_WINDOW >> 9) | |
c40f341f GR |
113 | #define CLUSTER_RESYNC_WINDOW (16 * RESYNC_WINDOW) |
114 | #define CLUSTER_RESYNC_WINDOW_SECTORS (CLUSTER_RESYNC_WINDOW >> 9) | |
1da177e4 | 115 | |
dd0fc66f | 116 | static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 LT |
117 | { |
118 | struct pool_info *pi = data; | |
9f2c9d12 | 119 | struct r1bio *r1_bio; |
1da177e4 | 120 | struct bio *bio; |
da1aab3d | 121 | int need_pages; |
98d30c58 ML |
122 | int j; |
123 | struct resync_pages *rps; | |
1da177e4 LT |
124 | |
125 | r1_bio = r1bio_pool_alloc(gfp_flags, pi); | |
7eaceacc | 126 | if (!r1_bio) |
1da177e4 | 127 | return NULL; |
1da177e4 | 128 | |
6da2ec56 KC |
129 | rps = kmalloc_array(pi->raid_disks, sizeof(struct resync_pages), |
130 | gfp_flags); | |
98d30c58 ML |
131 | if (!rps) |
132 | goto out_free_r1bio; | |
133 | ||
1da177e4 LT |
134 | /* |
135 | * Allocate bios : 1 for reading, n-1 for writing | |
136 | */ | |
137 | for (j = pi->raid_disks ; j-- ; ) { | |
6746557f | 138 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); |
1da177e4 LT |
139 | if (!bio) |
140 | goto out_free_bio; | |
141 | r1_bio->bios[j] = bio; | |
142 | } | |
143 | /* | |
144 | * Allocate RESYNC_PAGES data pages and attach them to | |
d11c171e N |
145 | * the first bio. |
146 | * If this is a user-requested check/repair, allocate | |
147 | * RESYNC_PAGES for each bio. | |
1da177e4 | 148 | */ |
d11c171e | 149 | if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) |
da1aab3d | 150 | need_pages = pi->raid_disks; |
d11c171e | 151 | else |
da1aab3d | 152 | need_pages = 1; |
98d30c58 ML |
153 | for (j = 0; j < pi->raid_disks; j++) { |
154 | struct resync_pages *rp = &rps[j]; | |
155 | ||
d11c171e | 156 | bio = r1_bio->bios[j]; |
d11c171e | 157 | |
98d30c58 ML |
158 | if (j < need_pages) { |
159 | if (resync_alloc_pages(rp, gfp_flags)) | |
160 | goto out_free_pages; | |
161 | } else { | |
162 | memcpy(rp, &rps[0], sizeof(*rp)); | |
163 | resync_get_all_pages(rp); | |
164 | } | |
165 | ||
98d30c58 ML |
166 | rp->raid_bio = r1_bio; |
167 | bio->bi_private = rp; | |
1da177e4 LT |
168 | } |
169 | ||
170 | r1_bio->master_bio = NULL; | |
171 | ||
172 | return r1_bio; | |
173 | ||
da1aab3d | 174 | out_free_pages: |
491221f8 | 175 | while (--j >= 0) |
98d30c58 | 176 | resync_free_pages(&rps[j]); |
da1aab3d | 177 | |
1da177e4 | 178 | out_free_bio: |
8f19ccb2 | 179 | while (++j < pi->raid_disks) |
1da177e4 | 180 | bio_put(r1_bio->bios[j]); |
98d30c58 ML |
181 | kfree(rps); |
182 | ||
183 | out_free_r1bio: | |
1da177e4 LT |
184 | r1bio_pool_free(r1_bio, data); |
185 | return NULL; | |
186 | } | |
187 | ||
188 | static void r1buf_pool_free(void *__r1_bio, void *data) | |
189 | { | |
190 | struct pool_info *pi = data; | |
98d30c58 | 191 | int i; |
9f2c9d12 | 192 | struct r1bio *r1bio = __r1_bio; |
98d30c58 | 193 | struct resync_pages *rp = NULL; |
1da177e4 | 194 | |
98d30c58 ML |
195 | for (i = pi->raid_disks; i--; ) { |
196 | rp = get_resync_pages(r1bio->bios[i]); | |
197 | resync_free_pages(rp); | |
1da177e4 | 198 | bio_put(r1bio->bios[i]); |
98d30c58 ML |
199 | } |
200 | ||
201 | /* resync pages array stored in the 1st bio's .bi_private */ | |
202 | kfree(rp); | |
1da177e4 LT |
203 | |
204 | r1bio_pool_free(r1bio, data); | |
205 | } | |
206 | ||
e8096360 | 207 | static void put_all_bios(struct r1conf *conf, struct r1bio *r1_bio) |
1da177e4 LT |
208 | { |
209 | int i; | |
210 | ||
8f19ccb2 | 211 | for (i = 0; i < conf->raid_disks * 2; i++) { |
1da177e4 | 212 | struct bio **bio = r1_bio->bios + i; |
4367af55 | 213 | if (!BIO_SPECIAL(*bio)) |
1da177e4 LT |
214 | bio_put(*bio); |
215 | *bio = NULL; | |
216 | } | |
217 | } | |
218 | ||
9f2c9d12 | 219 | static void free_r1bio(struct r1bio *r1_bio) |
1da177e4 | 220 | { |
e8096360 | 221 | struct r1conf *conf = r1_bio->mddev->private; |
1da177e4 | 222 | |
1da177e4 | 223 | put_all_bios(conf, r1_bio); |
afeee514 | 224 | mempool_free(r1_bio, &conf->r1bio_pool); |
1da177e4 LT |
225 | } |
226 | ||
9f2c9d12 | 227 | static void put_buf(struct r1bio *r1_bio) |
1da177e4 | 228 | { |
e8096360 | 229 | struct r1conf *conf = r1_bio->mddev->private; |
af5f42a7 | 230 | sector_t sect = r1_bio->sector; |
3e198f78 N |
231 | int i; |
232 | ||
8f19ccb2 | 233 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3e198f78 N |
234 | struct bio *bio = r1_bio->bios[i]; |
235 | if (bio->bi_end_io) | |
236 | rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev); | |
237 | } | |
1da177e4 | 238 | |
afeee514 | 239 | mempool_free(r1_bio, &conf->r1buf_pool); |
1da177e4 | 240 | |
af5f42a7 | 241 | lower_barrier(conf, sect); |
1da177e4 LT |
242 | } |
243 | ||
9f2c9d12 | 244 | static void reschedule_retry(struct r1bio *r1_bio) |
1da177e4 LT |
245 | { |
246 | unsigned long flags; | |
fd01b88c | 247 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 248 | struct r1conf *conf = mddev->private; |
fd76863e | 249 | int idx; |
1da177e4 | 250 | |
fd76863e | 251 | idx = sector_to_idx(r1_bio->sector); |
1da177e4 LT |
252 | spin_lock_irqsave(&conf->device_lock, flags); |
253 | list_add(&r1_bio->retry_list, &conf->retry_list); | |
824e47da | 254 | atomic_inc(&conf->nr_queued[idx]); |
1da177e4 LT |
255 | spin_unlock_irqrestore(&conf->device_lock, flags); |
256 | ||
17999be4 | 257 | wake_up(&conf->wait_barrier); |
1da177e4 LT |
258 | md_wakeup_thread(mddev->thread); |
259 | } | |
260 | ||
261 | /* | |
262 | * raid_end_bio_io() is called when we have finished servicing a mirrored | |
263 | * operation and are ready to return a success/failure code to the buffer | |
264 | * cache layer. | |
265 | */ | |
9f2c9d12 | 266 | static void call_bio_endio(struct r1bio *r1_bio) |
d2eb35ac N |
267 | { |
268 | struct bio *bio = r1_bio->master_bio; | |
e8096360 | 269 | struct r1conf *conf = r1_bio->mddev->private; |
d2eb35ac N |
270 | |
271 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) | |
4e4cbee9 | 272 | bio->bi_status = BLK_STS_IOERR; |
4246a0b6 | 273 | |
37011e3a N |
274 | bio_endio(bio); |
275 | /* | |
276 | * Wake up any possible resync thread that waits for the device | |
277 | * to go idle. | |
278 | */ | |
279 | allow_barrier(conf, r1_bio->sector); | |
d2eb35ac N |
280 | } |
281 | ||
9f2c9d12 | 282 | static void raid_end_bio_io(struct r1bio *r1_bio) |
1da177e4 LT |
283 | { |
284 | struct bio *bio = r1_bio->master_bio; | |
285 | ||
4b6d287f N |
286 | /* if nobody has done the final endio yet, do it now */ |
287 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { | |
36a4e1fe N |
288 | pr_debug("raid1: sync end %s on sectors %llu-%llu\n", |
289 | (bio_data_dir(bio) == WRITE) ? "write" : "read", | |
4f024f37 KO |
290 | (unsigned long long) bio->bi_iter.bi_sector, |
291 | (unsigned long long) bio_end_sector(bio) - 1); | |
4b6d287f | 292 | |
d2eb35ac | 293 | call_bio_endio(r1_bio); |
4b6d287f | 294 | } |
1da177e4 LT |
295 | free_r1bio(r1_bio); |
296 | } | |
297 | ||
298 | /* | |
299 | * Update disk head position estimator based on IRQ completion info. | |
300 | */ | |
9f2c9d12 | 301 | static inline void update_head_pos(int disk, struct r1bio *r1_bio) |
1da177e4 | 302 | { |
e8096360 | 303 | struct r1conf *conf = r1_bio->mddev->private; |
1da177e4 LT |
304 | |
305 | conf->mirrors[disk].head_position = | |
306 | r1_bio->sector + (r1_bio->sectors); | |
307 | } | |
308 | ||
ba3ae3be NK |
309 | /* |
310 | * Find the disk number which triggered given bio | |
311 | */ | |
9f2c9d12 | 312 | static int find_bio_disk(struct r1bio *r1_bio, struct bio *bio) |
ba3ae3be NK |
313 | { |
314 | int mirror; | |
30194636 N |
315 | struct r1conf *conf = r1_bio->mddev->private; |
316 | int raid_disks = conf->raid_disks; | |
ba3ae3be | 317 | |
8f19ccb2 | 318 | for (mirror = 0; mirror < raid_disks * 2; mirror++) |
ba3ae3be NK |
319 | if (r1_bio->bios[mirror] == bio) |
320 | break; | |
321 | ||
8f19ccb2 | 322 | BUG_ON(mirror == raid_disks * 2); |
ba3ae3be NK |
323 | update_head_pos(mirror, r1_bio); |
324 | ||
325 | return mirror; | |
326 | } | |
327 | ||
4246a0b6 | 328 | static void raid1_end_read_request(struct bio *bio) |
1da177e4 | 329 | { |
4e4cbee9 | 330 | int uptodate = !bio->bi_status; |
9f2c9d12 | 331 | struct r1bio *r1_bio = bio->bi_private; |
e8096360 | 332 | struct r1conf *conf = r1_bio->mddev->private; |
e5872d58 | 333 | struct md_rdev *rdev = conf->mirrors[r1_bio->read_disk].rdev; |
1da177e4 | 334 | |
1da177e4 LT |
335 | /* |
336 | * this branch is our 'one mirror IO has finished' event handler: | |
337 | */ | |
e5872d58 | 338 | update_head_pos(r1_bio->read_disk, r1_bio); |
ddaf22ab | 339 | |
dd00a99e N |
340 | if (uptodate) |
341 | set_bit(R1BIO_Uptodate, &r1_bio->state); | |
2e52d449 N |
342 | else if (test_bit(FailFast, &rdev->flags) && |
343 | test_bit(R1BIO_FailFast, &r1_bio->state)) | |
344 | /* This was a fail-fast read so we definitely | |
345 | * want to retry */ | |
346 | ; | |
dd00a99e N |
347 | else { |
348 | /* If all other devices have failed, we want to return | |
349 | * the error upwards rather than fail the last device. | |
350 | * Here we redefine "uptodate" to mean "Don't want to retry" | |
1da177e4 | 351 | */ |
dd00a99e N |
352 | unsigned long flags; |
353 | spin_lock_irqsave(&conf->device_lock, flags); | |
354 | if (r1_bio->mddev->degraded == conf->raid_disks || | |
355 | (r1_bio->mddev->degraded == conf->raid_disks-1 && | |
e5872d58 | 356 | test_bit(In_sync, &rdev->flags))) |
dd00a99e N |
357 | uptodate = 1; |
358 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
359 | } | |
1da177e4 | 360 | |
7ad4d4a6 | 361 | if (uptodate) { |
1da177e4 | 362 | raid_end_bio_io(r1_bio); |
e5872d58 | 363 | rdev_dec_pending(rdev, conf->mddev); |
7ad4d4a6 | 364 | } else { |
1da177e4 LT |
365 | /* |
366 | * oops, read error: | |
367 | */ | |
368 | char b[BDEVNAME_SIZE]; | |
1d41c216 N |
369 | pr_err_ratelimited("md/raid1:%s: %s: rescheduling sector %llu\n", |
370 | mdname(conf->mddev), | |
371 | bdevname(rdev->bdev, b), | |
372 | (unsigned long long)r1_bio->sector); | |
d2eb35ac | 373 | set_bit(R1BIO_ReadError, &r1_bio->state); |
1da177e4 | 374 | reschedule_retry(r1_bio); |
7ad4d4a6 | 375 | /* don't drop the reference on read_disk yet */ |
1da177e4 | 376 | } |
1da177e4 LT |
377 | } |
378 | ||
9f2c9d12 | 379 | static void close_write(struct r1bio *r1_bio) |
cd5ff9a1 N |
380 | { |
381 | /* it really is the end of this request */ | |
382 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { | |
841c1316 ML |
383 | bio_free_pages(r1_bio->behind_master_bio); |
384 | bio_put(r1_bio->behind_master_bio); | |
385 | r1_bio->behind_master_bio = NULL; | |
cd5ff9a1 N |
386 | } |
387 | /* clear the bitmap if all writes complete successfully */ | |
e64e4018 AS |
388 | md_bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector, |
389 | r1_bio->sectors, | |
390 | !test_bit(R1BIO_Degraded, &r1_bio->state), | |
391 | test_bit(R1BIO_BehindIO, &r1_bio->state)); | |
cd5ff9a1 N |
392 | md_write_end(r1_bio->mddev); |
393 | } | |
394 | ||
9f2c9d12 | 395 | static void r1_bio_write_done(struct r1bio *r1_bio) |
4e78064f | 396 | { |
cd5ff9a1 N |
397 | if (!atomic_dec_and_test(&r1_bio->remaining)) |
398 | return; | |
399 | ||
400 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
401 | reschedule_retry(r1_bio); | |
402 | else { | |
403 | close_write(r1_bio); | |
4367af55 N |
404 | if (test_bit(R1BIO_MadeGood, &r1_bio->state)) |
405 | reschedule_retry(r1_bio); | |
406 | else | |
407 | raid_end_bio_io(r1_bio); | |
4e78064f N |
408 | } |
409 | } | |
410 | ||
4246a0b6 | 411 | static void raid1_end_write_request(struct bio *bio) |
1da177e4 | 412 | { |
9f2c9d12 | 413 | struct r1bio *r1_bio = bio->bi_private; |
e5872d58 | 414 | int behind = test_bit(R1BIO_BehindIO, &r1_bio->state); |
e8096360 | 415 | struct r1conf *conf = r1_bio->mddev->private; |
04b857f7 | 416 | struct bio *to_put = NULL; |
e5872d58 N |
417 | int mirror = find_bio_disk(r1_bio, bio); |
418 | struct md_rdev *rdev = conf->mirrors[mirror].rdev; | |
e3f948cd SL |
419 | bool discard_error; |
420 | ||
4e4cbee9 | 421 | discard_error = bio->bi_status && bio_op(bio) == REQ_OP_DISCARD; |
1da177e4 | 422 | |
e9c7469b TH |
423 | /* |
424 | * 'one mirror IO has finished' event handler: | |
425 | */ | |
4e4cbee9 | 426 | if (bio->bi_status && !discard_error) { |
e5872d58 N |
427 | set_bit(WriteErrorSeen, &rdev->flags); |
428 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) | |
19d67169 N |
429 | set_bit(MD_RECOVERY_NEEDED, & |
430 | conf->mddev->recovery); | |
431 | ||
212e7eb7 N |
432 | if (test_bit(FailFast, &rdev->flags) && |
433 | (bio->bi_opf & MD_FAILFAST) && | |
434 | /* We never try FailFast to WriteMostly devices */ | |
435 | !test_bit(WriteMostly, &rdev->flags)) { | |
436 | md_error(r1_bio->mddev, rdev); | |
437 | if (!test_bit(Faulty, &rdev->flags)) | |
438 | /* This is the only remaining device, | |
439 | * We need to retry the write without | |
440 | * FailFast | |
441 | */ | |
442 | set_bit(R1BIO_WriteError, &r1_bio->state); | |
443 | else { | |
444 | /* Finished with this branch */ | |
445 | r1_bio->bios[mirror] = NULL; | |
446 | to_put = bio; | |
447 | } | |
448 | } else | |
449 | set_bit(R1BIO_WriteError, &r1_bio->state); | |
4367af55 | 450 | } else { |
1da177e4 | 451 | /* |
e9c7469b TH |
452 | * Set R1BIO_Uptodate in our master bio, so that we |
453 | * will return a good error code for to the higher | |
454 | * levels even if IO on some other mirrored buffer | |
455 | * fails. | |
456 | * | |
457 | * The 'master' represents the composite IO operation | |
458 | * to user-side. So if something waits for IO, then it | |
459 | * will wait for the 'master' bio. | |
1da177e4 | 460 | */ |
4367af55 N |
461 | sector_t first_bad; |
462 | int bad_sectors; | |
463 | ||
cd5ff9a1 N |
464 | r1_bio->bios[mirror] = NULL; |
465 | to_put = bio; | |
3056e3ae AL |
466 | /* |
467 | * Do not set R1BIO_Uptodate if the current device is | |
468 | * rebuilding or Faulty. This is because we cannot use | |
469 | * such device for properly reading the data back (we could | |
470 | * potentially use it, if the current write would have felt | |
471 | * before rdev->recovery_offset, but for simplicity we don't | |
472 | * check this here. | |
473 | */ | |
e5872d58 N |
474 | if (test_bit(In_sync, &rdev->flags) && |
475 | !test_bit(Faulty, &rdev->flags)) | |
3056e3ae | 476 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
e9c7469b | 477 | |
4367af55 | 478 | /* Maybe we can clear some bad blocks. */ |
e5872d58 | 479 | if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors, |
e3f948cd | 480 | &first_bad, &bad_sectors) && !discard_error) { |
4367af55 N |
481 | r1_bio->bios[mirror] = IO_MADE_GOOD; |
482 | set_bit(R1BIO_MadeGood, &r1_bio->state); | |
483 | } | |
484 | } | |
485 | ||
e9c7469b | 486 | if (behind) { |
e5872d58 | 487 | if (test_bit(WriteMostly, &rdev->flags)) |
e9c7469b TH |
488 | atomic_dec(&r1_bio->behind_remaining); |
489 | ||
490 | /* | |
491 | * In behind mode, we ACK the master bio once the I/O | |
492 | * has safely reached all non-writemostly | |
493 | * disks. Setting the Returned bit ensures that this | |
494 | * gets done only once -- we don't ever want to return | |
495 | * -EIO here, instead we'll wait | |
496 | */ | |
497 | if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) && | |
498 | test_bit(R1BIO_Uptodate, &r1_bio->state)) { | |
499 | /* Maybe we can return now */ | |
500 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { | |
501 | struct bio *mbio = r1_bio->master_bio; | |
36a4e1fe N |
502 | pr_debug("raid1: behind end write sectors" |
503 | " %llu-%llu\n", | |
4f024f37 KO |
504 | (unsigned long long) mbio->bi_iter.bi_sector, |
505 | (unsigned long long) bio_end_sector(mbio) - 1); | |
d2eb35ac | 506 | call_bio_endio(r1_bio); |
4b6d287f N |
507 | } |
508 | } | |
509 | } | |
4367af55 | 510 | if (r1_bio->bios[mirror] == NULL) |
e5872d58 | 511 | rdev_dec_pending(rdev, conf->mddev); |
e9c7469b | 512 | |
1da177e4 | 513 | /* |
1da177e4 LT |
514 | * Let's see if all mirrored write operations have finished |
515 | * already. | |
516 | */ | |
af6d7b76 | 517 | r1_bio_write_done(r1_bio); |
c70810b3 | 518 | |
04b857f7 N |
519 | if (to_put) |
520 | bio_put(to_put); | |
1da177e4 LT |
521 | } |
522 | ||
fd76863e | 523 | static sector_t align_to_barrier_unit_end(sector_t start_sector, |
524 | sector_t sectors) | |
525 | { | |
526 | sector_t len; | |
527 | ||
528 | WARN_ON(sectors == 0); | |
529 | /* | |
530 | * len is the number of sectors from start_sector to end of the | |
531 | * barrier unit which start_sector belongs to. | |
532 | */ | |
533 | len = round_up(start_sector + 1, BARRIER_UNIT_SECTOR_SIZE) - | |
534 | start_sector; | |
535 | ||
536 | if (len > sectors) | |
537 | len = sectors; | |
538 | ||
539 | return len; | |
540 | } | |
541 | ||
1da177e4 LT |
542 | /* |
543 | * This routine returns the disk from which the requested read should | |
544 | * be done. There is a per-array 'next expected sequential IO' sector | |
545 | * number - if this matches on the next IO then we use the last disk. | |
546 | * There is also a per-disk 'last know head position' sector that is | |
547 | * maintained from IRQ contexts, both the normal and the resync IO | |
548 | * completion handlers update this position correctly. If there is no | |
549 | * perfect sequential match then we pick the disk whose head is closest. | |
550 | * | |
551 | * If there are 2 mirrors in the same 2 devices, performance degrades | |
552 | * because position is mirror, not device based. | |
553 | * | |
554 | * The rdev for the device selected will have nr_pending incremented. | |
555 | */ | |
e8096360 | 556 | static int read_balance(struct r1conf *conf, struct r1bio *r1_bio, int *max_sectors) |
1da177e4 | 557 | { |
af3a2cd6 | 558 | const sector_t this_sector = r1_bio->sector; |
d2eb35ac N |
559 | int sectors; |
560 | int best_good_sectors; | |
9dedf603 SL |
561 | int best_disk, best_dist_disk, best_pending_disk; |
562 | int has_nonrot_disk; | |
be4d3280 | 563 | int disk; |
76073054 | 564 | sector_t best_dist; |
9dedf603 | 565 | unsigned int min_pending; |
3cb03002 | 566 | struct md_rdev *rdev; |
f3ac8bf7 | 567 | int choose_first; |
12cee5a8 | 568 | int choose_next_idle; |
1da177e4 LT |
569 | |
570 | rcu_read_lock(); | |
571 | /* | |
8ddf9efe | 572 | * Check if we can balance. We can balance on the whole |
1da177e4 LT |
573 | * device if no resync is going on, or below the resync window. |
574 | * We take the first readable disk when above the resync window. | |
575 | */ | |
576 | retry: | |
d2eb35ac | 577 | sectors = r1_bio->sectors; |
76073054 | 578 | best_disk = -1; |
9dedf603 | 579 | best_dist_disk = -1; |
76073054 | 580 | best_dist = MaxSector; |
9dedf603 SL |
581 | best_pending_disk = -1; |
582 | min_pending = UINT_MAX; | |
d2eb35ac | 583 | best_good_sectors = 0; |
9dedf603 | 584 | has_nonrot_disk = 0; |
12cee5a8 | 585 | choose_next_idle = 0; |
2e52d449 | 586 | clear_bit(R1BIO_FailFast, &r1_bio->state); |
d2eb35ac | 587 | |
7d49ffcf GR |
588 | if ((conf->mddev->recovery_cp < this_sector + sectors) || |
589 | (mddev_is_clustered(conf->mddev) && | |
90382ed9 | 590 | md_cluster_ops->area_resyncing(conf->mddev, READ, this_sector, |
7d49ffcf GR |
591 | this_sector + sectors))) |
592 | choose_first = 1; | |
593 | else | |
594 | choose_first = 0; | |
1da177e4 | 595 | |
be4d3280 | 596 | for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) { |
76073054 | 597 | sector_t dist; |
d2eb35ac N |
598 | sector_t first_bad; |
599 | int bad_sectors; | |
9dedf603 | 600 | unsigned int pending; |
12cee5a8 | 601 | bool nonrot; |
d2eb35ac | 602 | |
f3ac8bf7 N |
603 | rdev = rcu_dereference(conf->mirrors[disk].rdev); |
604 | if (r1_bio->bios[disk] == IO_BLOCKED | |
605 | || rdev == NULL | |
76073054 | 606 | || test_bit(Faulty, &rdev->flags)) |
f3ac8bf7 | 607 | continue; |
76073054 N |
608 | if (!test_bit(In_sync, &rdev->flags) && |
609 | rdev->recovery_offset < this_sector + sectors) | |
1da177e4 | 610 | continue; |
76073054 N |
611 | if (test_bit(WriteMostly, &rdev->flags)) { |
612 | /* Don't balance among write-mostly, just | |
613 | * use the first as a last resort */ | |
d1901ef0 | 614 | if (best_dist_disk < 0) { |
307729c8 N |
615 | if (is_badblock(rdev, this_sector, sectors, |
616 | &first_bad, &bad_sectors)) { | |
816b0acf | 617 | if (first_bad <= this_sector) |
307729c8 N |
618 | /* Cannot use this */ |
619 | continue; | |
620 | best_good_sectors = first_bad - this_sector; | |
621 | } else | |
622 | best_good_sectors = sectors; | |
d1901ef0 TH |
623 | best_dist_disk = disk; |
624 | best_pending_disk = disk; | |
307729c8 | 625 | } |
76073054 N |
626 | continue; |
627 | } | |
628 | /* This is a reasonable device to use. It might | |
629 | * even be best. | |
630 | */ | |
d2eb35ac N |
631 | if (is_badblock(rdev, this_sector, sectors, |
632 | &first_bad, &bad_sectors)) { | |
633 | if (best_dist < MaxSector) | |
634 | /* already have a better device */ | |
635 | continue; | |
636 | if (first_bad <= this_sector) { | |
637 | /* cannot read here. If this is the 'primary' | |
638 | * device, then we must not read beyond | |
639 | * bad_sectors from another device.. | |
640 | */ | |
641 | bad_sectors -= (this_sector - first_bad); | |
642 | if (choose_first && sectors > bad_sectors) | |
643 | sectors = bad_sectors; | |
644 | if (best_good_sectors > sectors) | |
645 | best_good_sectors = sectors; | |
646 | ||
647 | } else { | |
648 | sector_t good_sectors = first_bad - this_sector; | |
649 | if (good_sectors > best_good_sectors) { | |
650 | best_good_sectors = good_sectors; | |
651 | best_disk = disk; | |
652 | } | |
653 | if (choose_first) | |
654 | break; | |
655 | } | |
656 | continue; | |
d82dd0e3 TM |
657 | } else { |
658 | if ((sectors > best_good_sectors) && (best_disk >= 0)) | |
659 | best_disk = -1; | |
d2eb35ac | 660 | best_good_sectors = sectors; |
d82dd0e3 | 661 | } |
d2eb35ac | 662 | |
2e52d449 N |
663 | if (best_disk >= 0) |
664 | /* At least two disks to choose from so failfast is OK */ | |
665 | set_bit(R1BIO_FailFast, &r1_bio->state); | |
666 | ||
12cee5a8 SL |
667 | nonrot = blk_queue_nonrot(bdev_get_queue(rdev->bdev)); |
668 | has_nonrot_disk |= nonrot; | |
9dedf603 | 669 | pending = atomic_read(&rdev->nr_pending); |
76073054 | 670 | dist = abs(this_sector - conf->mirrors[disk].head_position); |
12cee5a8 | 671 | if (choose_first) { |
76073054 | 672 | best_disk = disk; |
1da177e4 LT |
673 | break; |
674 | } | |
12cee5a8 SL |
675 | /* Don't change to another disk for sequential reads */ |
676 | if (conf->mirrors[disk].next_seq_sect == this_sector | |
677 | || dist == 0) { | |
678 | int opt_iosize = bdev_io_opt(rdev->bdev) >> 9; | |
679 | struct raid1_info *mirror = &conf->mirrors[disk]; | |
680 | ||
681 | best_disk = disk; | |
682 | /* | |
683 | * If buffered sequential IO size exceeds optimal | |
684 | * iosize, check if there is idle disk. If yes, choose | |
685 | * the idle disk. read_balance could already choose an | |
686 | * idle disk before noticing it's a sequential IO in | |
687 | * this disk. This doesn't matter because this disk | |
688 | * will idle, next time it will be utilized after the | |
689 | * first disk has IO size exceeds optimal iosize. In | |
690 | * this way, iosize of the first disk will be optimal | |
691 | * iosize at least. iosize of the second disk might be | |
692 | * small, but not a big deal since when the second disk | |
693 | * starts IO, the first disk is likely still busy. | |
694 | */ | |
695 | if (nonrot && opt_iosize > 0 && | |
696 | mirror->seq_start != MaxSector && | |
697 | mirror->next_seq_sect > opt_iosize && | |
698 | mirror->next_seq_sect - opt_iosize >= | |
699 | mirror->seq_start) { | |
700 | choose_next_idle = 1; | |
701 | continue; | |
702 | } | |
703 | break; | |
704 | } | |
12cee5a8 SL |
705 | |
706 | if (choose_next_idle) | |
707 | continue; | |
9dedf603 SL |
708 | |
709 | if (min_pending > pending) { | |
710 | min_pending = pending; | |
711 | best_pending_disk = disk; | |
712 | } | |
713 | ||
76073054 N |
714 | if (dist < best_dist) { |
715 | best_dist = dist; | |
9dedf603 | 716 | best_dist_disk = disk; |
1da177e4 | 717 | } |
f3ac8bf7 | 718 | } |
1da177e4 | 719 | |
9dedf603 SL |
720 | /* |
721 | * If all disks are rotational, choose the closest disk. If any disk is | |
722 | * non-rotational, choose the disk with less pending request even the | |
723 | * disk is rotational, which might/might not be optimal for raids with | |
724 | * mixed ratation/non-rotational disks depending on workload. | |
725 | */ | |
726 | if (best_disk == -1) { | |
2e52d449 | 727 | if (has_nonrot_disk || min_pending == 0) |
9dedf603 SL |
728 | best_disk = best_pending_disk; |
729 | else | |
730 | best_disk = best_dist_disk; | |
731 | } | |
732 | ||
76073054 N |
733 | if (best_disk >= 0) { |
734 | rdev = rcu_dereference(conf->mirrors[best_disk].rdev); | |
8ddf9efe N |
735 | if (!rdev) |
736 | goto retry; | |
737 | atomic_inc(&rdev->nr_pending); | |
d2eb35ac | 738 | sectors = best_good_sectors; |
12cee5a8 SL |
739 | |
740 | if (conf->mirrors[best_disk].next_seq_sect != this_sector) | |
741 | conf->mirrors[best_disk].seq_start = this_sector; | |
742 | ||
be4d3280 | 743 | conf->mirrors[best_disk].next_seq_sect = this_sector + sectors; |
1da177e4 LT |
744 | } |
745 | rcu_read_unlock(); | |
d2eb35ac | 746 | *max_sectors = sectors; |
1da177e4 | 747 | |
76073054 | 748 | return best_disk; |
1da177e4 LT |
749 | } |
750 | ||
5c675f83 | 751 | static int raid1_congested(struct mddev *mddev, int bits) |
0d129228 | 752 | { |
e8096360 | 753 | struct r1conf *conf = mddev->private; |
0d129228 N |
754 | int i, ret = 0; |
755 | ||
4452226e | 756 | if ((bits & (1 << WB_async_congested)) && |
34db0cd6 N |
757 | conf->pending_count >= max_queued_requests) |
758 | return 1; | |
759 | ||
0d129228 | 760 | rcu_read_lock(); |
f53e29fc | 761 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3cb03002 | 762 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
0d129228 | 763 | if (rdev && !test_bit(Faulty, &rdev->flags)) { |
165125e1 | 764 | struct request_queue *q = bdev_get_queue(rdev->bdev); |
0d129228 | 765 | |
1ed7242e JB |
766 | BUG_ON(!q); |
767 | ||
0d129228 N |
768 | /* Note the '|| 1' - when read_balance prefers |
769 | * non-congested targets, it can be removed | |
770 | */ | |
4452226e | 771 | if ((bits & (1 << WB_async_congested)) || 1) |
dc3b17cc | 772 | ret |= bdi_congested(q->backing_dev_info, bits); |
0d129228 | 773 | else |
dc3b17cc | 774 | ret &= bdi_congested(q->backing_dev_info, bits); |
0d129228 N |
775 | } |
776 | } | |
777 | rcu_read_unlock(); | |
778 | return ret; | |
779 | } | |
0d129228 | 780 | |
673ca68d N |
781 | static void flush_bio_list(struct r1conf *conf, struct bio *bio) |
782 | { | |
783 | /* flush any pending bitmap writes to disk before proceeding w/ I/O */ | |
e64e4018 | 784 | md_bitmap_unplug(conf->mddev->bitmap); |
673ca68d N |
785 | wake_up(&conf->wait_barrier); |
786 | ||
787 | while (bio) { /* submit pending writes */ | |
788 | struct bio *next = bio->bi_next; | |
74d46992 | 789 | struct md_rdev *rdev = (void *)bio->bi_disk; |
673ca68d | 790 | bio->bi_next = NULL; |
74d46992 | 791 | bio_set_dev(bio, rdev->bdev); |
673ca68d | 792 | if (test_bit(Faulty, &rdev->flags)) { |
6308d8e3 | 793 | bio_io_error(bio); |
673ca68d | 794 | } else if (unlikely((bio_op(bio) == REQ_OP_DISCARD) && |
74d46992 | 795 | !blk_queue_discard(bio->bi_disk->queue))) |
673ca68d N |
796 | /* Just ignore it */ |
797 | bio_endio(bio); | |
798 | else | |
799 | generic_make_request(bio); | |
800 | bio = next; | |
801 | } | |
802 | } | |
803 | ||
e8096360 | 804 | static void flush_pending_writes(struct r1conf *conf) |
a35e63ef N |
805 | { |
806 | /* Any writes that have been queued but are awaiting | |
807 | * bitmap updates get flushed here. | |
a35e63ef | 808 | */ |
a35e63ef N |
809 | spin_lock_irq(&conf->device_lock); |
810 | ||
811 | if (conf->pending_bio_list.head) { | |
18022a1b | 812 | struct blk_plug plug; |
a35e63ef | 813 | struct bio *bio; |
18022a1b | 814 | |
a35e63ef | 815 | bio = bio_list_get(&conf->pending_bio_list); |
34db0cd6 | 816 | conf->pending_count = 0; |
a35e63ef | 817 | spin_unlock_irq(&conf->device_lock); |
474beb57 N |
818 | |
819 | /* | |
820 | * As this is called in a wait_event() loop (see freeze_array), | |
821 | * current->state might be TASK_UNINTERRUPTIBLE which will | |
822 | * cause a warning when we prepare to wait again. As it is | |
823 | * rare that this path is taken, it is perfectly safe to force | |
824 | * us to go around the wait_event() loop again, so the warning | |
825 | * is a false-positive. Silence the warning by resetting | |
826 | * thread state | |
827 | */ | |
828 | __set_current_state(TASK_RUNNING); | |
18022a1b | 829 | blk_start_plug(&plug); |
673ca68d | 830 | flush_bio_list(conf, bio); |
18022a1b | 831 | blk_finish_plug(&plug); |
a35e63ef N |
832 | } else |
833 | spin_unlock_irq(&conf->device_lock); | |
7eaceacc JA |
834 | } |
835 | ||
17999be4 N |
836 | /* Barriers.... |
837 | * Sometimes we need to suspend IO while we do something else, | |
838 | * either some resync/recovery, or reconfigure the array. | |
839 | * To do this we raise a 'barrier'. | |
840 | * The 'barrier' is a counter that can be raised multiple times | |
841 | * to count how many activities are happening which preclude | |
842 | * normal IO. | |
843 | * We can only raise the barrier if there is no pending IO. | |
844 | * i.e. if nr_pending == 0. | |
845 | * We choose only to raise the barrier if no-one is waiting for the | |
846 | * barrier to go down. This means that as soon as an IO request | |
847 | * is ready, no other operations which require a barrier will start | |
848 | * until the IO request has had a chance. | |
849 | * | |
850 | * So: regular IO calls 'wait_barrier'. When that returns there | |
851 | * is no backgroup IO happening, It must arrange to call | |
852 | * allow_barrier when it has finished its IO. | |
853 | * backgroup IO calls must call raise_barrier. Once that returns | |
854 | * there is no normal IO happeing. It must arrange to call | |
855 | * lower_barrier when the particular background IO completes. | |
1da177e4 | 856 | */ |
8c242593 | 857 | static sector_t raise_barrier(struct r1conf *conf, sector_t sector_nr) |
1da177e4 | 858 | { |
fd76863e | 859 | int idx = sector_to_idx(sector_nr); |
860 | ||
1da177e4 | 861 | spin_lock_irq(&conf->resync_lock); |
17999be4 N |
862 | |
863 | /* Wait until no block IO is waiting */ | |
824e47da | 864 | wait_event_lock_irq(conf->wait_barrier, |
865 | !atomic_read(&conf->nr_waiting[idx]), | |
eed8c02e | 866 | conf->resync_lock); |
17999be4 N |
867 | |
868 | /* block any new IO from starting */ | |
824e47da | 869 | atomic_inc(&conf->barrier[idx]); |
870 | /* | |
871 | * In raise_barrier() we firstly increase conf->barrier[idx] then | |
872 | * check conf->nr_pending[idx]. In _wait_barrier() we firstly | |
873 | * increase conf->nr_pending[idx] then check conf->barrier[idx]. | |
874 | * A memory barrier here to make sure conf->nr_pending[idx] won't | |
875 | * be fetched before conf->barrier[idx] is increased. Otherwise | |
876 | * there will be a race between raise_barrier() and _wait_barrier(). | |
877 | */ | |
878 | smp_mb__after_atomic(); | |
17999be4 | 879 | |
79ef3a8a | 880 | /* For these conditions we must wait: |
881 | * A: while the array is in frozen state | |
fd76863e | 882 | * B: while conf->nr_pending[idx] is not 0, meaning regular I/O |
883 | * existing in corresponding I/O barrier bucket. | |
884 | * C: while conf->barrier[idx] >= RESYNC_DEPTH, meaning reaches | |
885 | * max resync count which allowed on current I/O barrier bucket. | |
79ef3a8a | 886 | */ |
17999be4 | 887 | wait_event_lock_irq(conf->wait_barrier, |
8c242593 | 888 | (!conf->array_frozen && |
824e47da | 889 | !atomic_read(&conf->nr_pending[idx]) && |
8c242593 YY |
890 | atomic_read(&conf->barrier[idx]) < RESYNC_DEPTH) || |
891 | test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery), | |
eed8c02e | 892 | conf->resync_lock); |
17999be4 | 893 | |
8c242593 YY |
894 | if (test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { |
895 | atomic_dec(&conf->barrier[idx]); | |
896 | spin_unlock_irq(&conf->resync_lock); | |
897 | wake_up(&conf->wait_barrier); | |
898 | return -EINTR; | |
899 | } | |
900 | ||
43ac9b84 | 901 | atomic_inc(&conf->nr_sync_pending); |
17999be4 | 902 | spin_unlock_irq(&conf->resync_lock); |
8c242593 YY |
903 | |
904 | return 0; | |
17999be4 N |
905 | } |
906 | ||
fd76863e | 907 | static void lower_barrier(struct r1conf *conf, sector_t sector_nr) |
17999be4 | 908 | { |
fd76863e | 909 | int idx = sector_to_idx(sector_nr); |
910 | ||
824e47da | 911 | BUG_ON(atomic_read(&conf->barrier[idx]) <= 0); |
fd76863e | 912 | |
824e47da | 913 | atomic_dec(&conf->barrier[idx]); |
43ac9b84 | 914 | atomic_dec(&conf->nr_sync_pending); |
17999be4 N |
915 | wake_up(&conf->wait_barrier); |
916 | } | |
917 | ||
fd76863e | 918 | static void _wait_barrier(struct r1conf *conf, int idx) |
17999be4 | 919 | { |
824e47da | 920 | /* |
921 | * We need to increase conf->nr_pending[idx] very early here, | |
922 | * then raise_barrier() can be blocked when it waits for | |
923 | * conf->nr_pending[idx] to be 0. Then we can avoid holding | |
924 | * conf->resync_lock when there is no barrier raised in same | |
925 | * barrier unit bucket. Also if the array is frozen, I/O | |
926 | * should be blocked until array is unfrozen. | |
927 | */ | |
928 | atomic_inc(&conf->nr_pending[idx]); | |
929 | /* | |
930 | * In _wait_barrier() we firstly increase conf->nr_pending[idx], then | |
931 | * check conf->barrier[idx]. In raise_barrier() we firstly increase | |
932 | * conf->barrier[idx], then check conf->nr_pending[idx]. A memory | |
933 | * barrier is necessary here to make sure conf->barrier[idx] won't be | |
934 | * fetched before conf->nr_pending[idx] is increased. Otherwise there | |
935 | * will be a race between _wait_barrier() and raise_barrier(). | |
936 | */ | |
937 | smp_mb__after_atomic(); | |
79ef3a8a | 938 | |
824e47da | 939 | /* |
940 | * Don't worry about checking two atomic_t variables at same time | |
941 | * here. If during we check conf->barrier[idx], the array is | |
942 | * frozen (conf->array_frozen is 1), and chonf->barrier[idx] is | |
943 | * 0, it is safe to return and make the I/O continue. Because the | |
944 | * array is frozen, all I/O returned here will eventually complete | |
945 | * or be queued, no race will happen. See code comment in | |
946 | * frozen_array(). | |
947 | */ | |
948 | if (!READ_ONCE(conf->array_frozen) && | |
949 | !atomic_read(&conf->barrier[idx])) | |
950 | return; | |
79ef3a8a | 951 | |
824e47da | 952 | /* |
953 | * After holding conf->resync_lock, conf->nr_pending[idx] | |
954 | * should be decreased before waiting for barrier to drop. | |
955 | * Otherwise, we may encounter a race condition because | |
956 | * raise_barrer() might be waiting for conf->nr_pending[idx] | |
957 | * to be 0 at same time. | |
958 | */ | |
959 | spin_lock_irq(&conf->resync_lock); | |
960 | atomic_inc(&conf->nr_waiting[idx]); | |
961 | atomic_dec(&conf->nr_pending[idx]); | |
962 | /* | |
963 | * In case freeze_array() is waiting for | |
964 | * get_unqueued_pending() == extra | |
965 | */ | |
966 | wake_up(&conf->wait_barrier); | |
967 | /* Wait for the barrier in same barrier unit bucket to drop. */ | |
968 | wait_event_lock_irq(conf->wait_barrier, | |
969 | !conf->array_frozen && | |
970 | !atomic_read(&conf->barrier[idx]), | |
971 | conf->resync_lock); | |
972 | atomic_inc(&conf->nr_pending[idx]); | |
973 | atomic_dec(&conf->nr_waiting[idx]); | |
fd76863e | 974 | spin_unlock_irq(&conf->resync_lock); |
79ef3a8a | 975 | } |
976 | ||
fd76863e | 977 | static void wait_read_barrier(struct r1conf *conf, sector_t sector_nr) |
79ef3a8a | 978 | { |
fd76863e | 979 | int idx = sector_to_idx(sector_nr); |
79ef3a8a | 980 | |
824e47da | 981 | /* |
982 | * Very similar to _wait_barrier(). The difference is, for read | |
983 | * I/O we don't need wait for sync I/O, but if the whole array | |
984 | * is frozen, the read I/O still has to wait until the array is | |
985 | * unfrozen. Since there is no ordering requirement with | |
986 | * conf->barrier[idx] here, memory barrier is unnecessary as well. | |
987 | */ | |
988 | atomic_inc(&conf->nr_pending[idx]); | |
79ef3a8a | 989 | |
824e47da | 990 | if (!READ_ONCE(conf->array_frozen)) |
991 | return; | |
992 | ||
993 | spin_lock_irq(&conf->resync_lock); | |
994 | atomic_inc(&conf->nr_waiting[idx]); | |
995 | atomic_dec(&conf->nr_pending[idx]); | |
996 | /* | |
997 | * In case freeze_array() is waiting for | |
998 | * get_unqueued_pending() == extra | |
999 | */ | |
1000 | wake_up(&conf->wait_barrier); | |
1001 | /* Wait for array to be unfrozen */ | |
1002 | wait_event_lock_irq(conf->wait_barrier, | |
1003 | !conf->array_frozen, | |
1004 | conf->resync_lock); | |
1005 | atomic_inc(&conf->nr_pending[idx]); | |
1006 | atomic_dec(&conf->nr_waiting[idx]); | |
1da177e4 LT |
1007 | spin_unlock_irq(&conf->resync_lock); |
1008 | } | |
1009 | ||
fd76863e | 1010 | static void wait_barrier(struct r1conf *conf, sector_t sector_nr) |
17999be4 | 1011 | { |
fd76863e | 1012 | int idx = sector_to_idx(sector_nr); |
79ef3a8a | 1013 | |
fd76863e | 1014 | _wait_barrier(conf, idx); |
1015 | } | |
1016 | ||
fd76863e | 1017 | static void _allow_barrier(struct r1conf *conf, int idx) |
17999be4 | 1018 | { |
824e47da | 1019 | atomic_dec(&conf->nr_pending[idx]); |
17999be4 N |
1020 | wake_up(&conf->wait_barrier); |
1021 | } | |
1022 | ||
fd76863e | 1023 | static void allow_barrier(struct r1conf *conf, sector_t sector_nr) |
1024 | { | |
1025 | int idx = sector_to_idx(sector_nr); | |
1026 | ||
1027 | _allow_barrier(conf, idx); | |
1028 | } | |
1029 | ||
fd76863e | 1030 | /* conf->resync_lock should be held */ |
1031 | static int get_unqueued_pending(struct r1conf *conf) | |
1032 | { | |
1033 | int idx, ret; | |
1034 | ||
43ac9b84 XN |
1035 | ret = atomic_read(&conf->nr_sync_pending); |
1036 | for (idx = 0; idx < BARRIER_BUCKETS_NR; idx++) | |
824e47da | 1037 | ret += atomic_read(&conf->nr_pending[idx]) - |
1038 | atomic_read(&conf->nr_queued[idx]); | |
fd76863e | 1039 | |
1040 | return ret; | |
1041 | } | |
1042 | ||
e2d59925 | 1043 | static void freeze_array(struct r1conf *conf, int extra) |
ddaf22ab | 1044 | { |
fd76863e | 1045 | /* Stop sync I/O and normal I/O and wait for everything to |
11353b9d | 1046 | * go quiet. |
fd76863e | 1047 | * This is called in two situations: |
1048 | * 1) management command handlers (reshape, remove disk, quiesce). | |
1049 | * 2) one normal I/O request failed. | |
1050 | ||
1051 | * After array_frozen is set to 1, new sync IO will be blocked at | |
1052 | * raise_barrier(), and new normal I/O will blocked at _wait_barrier() | |
1053 | * or wait_read_barrier(). The flying I/Os will either complete or be | |
1054 | * queued. When everything goes quite, there are only queued I/Os left. | |
1055 | ||
1056 | * Every flying I/O contributes to a conf->nr_pending[idx], idx is the | |
1057 | * barrier bucket index which this I/O request hits. When all sync and | |
1058 | * normal I/O are queued, sum of all conf->nr_pending[] will match sum | |
1059 | * of all conf->nr_queued[]. But normal I/O failure is an exception, | |
1060 | * in handle_read_error(), we may call freeze_array() before trying to | |
1061 | * fix the read error. In this case, the error read I/O is not queued, | |
1062 | * so get_unqueued_pending() == 1. | |
1063 | * | |
1064 | * Therefore before this function returns, we need to wait until | |
1065 | * get_unqueued_pendings(conf) gets equal to extra. For | |
1066 | * normal I/O context, extra is 1, in rested situations extra is 0. | |
ddaf22ab N |
1067 | */ |
1068 | spin_lock_irq(&conf->resync_lock); | |
b364e3d0 | 1069 | conf->array_frozen = 1; |
578b54ad | 1070 | raid1_log(conf->mddev, "wait freeze"); |
fd76863e | 1071 | wait_event_lock_irq_cmd( |
1072 | conf->wait_barrier, | |
1073 | get_unqueued_pending(conf) == extra, | |
1074 | conf->resync_lock, | |
1075 | flush_pending_writes(conf)); | |
ddaf22ab N |
1076 | spin_unlock_irq(&conf->resync_lock); |
1077 | } | |
e8096360 | 1078 | static void unfreeze_array(struct r1conf *conf) |
ddaf22ab N |
1079 | { |
1080 | /* reverse the effect of the freeze */ | |
1081 | spin_lock_irq(&conf->resync_lock); | |
b364e3d0 | 1082 | conf->array_frozen = 0; |
ddaf22ab | 1083 | spin_unlock_irq(&conf->resync_lock); |
824e47da | 1084 | wake_up(&conf->wait_barrier); |
ddaf22ab N |
1085 | } |
1086 | ||
16d56e2f | 1087 | static void alloc_behind_master_bio(struct r1bio *r1_bio, |
cb83efcf | 1088 | struct bio *bio) |
4b6d287f | 1089 | { |
cb83efcf | 1090 | int size = bio->bi_iter.bi_size; |
841c1316 ML |
1091 | unsigned vcnt = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; |
1092 | int i = 0; | |
1093 | struct bio *behind_bio = NULL; | |
1094 | ||
1095 | behind_bio = bio_alloc_mddev(GFP_NOIO, vcnt, r1_bio->mddev); | |
1096 | if (!behind_bio) | |
16d56e2f | 1097 | return; |
4b6d287f | 1098 | |
41743c1f | 1099 | /* discard op, we don't support writezero/writesame yet */ |
16d56e2f SL |
1100 | if (!bio_has_data(bio)) { |
1101 | behind_bio->bi_iter.bi_size = size; | |
41743c1f | 1102 | goto skip_copy; |
16d56e2f | 1103 | } |
41743c1f | 1104 | |
dba40d46 MD |
1105 | behind_bio->bi_write_hint = bio->bi_write_hint; |
1106 | ||
841c1316 ML |
1107 | while (i < vcnt && size) { |
1108 | struct page *page; | |
1109 | int len = min_t(int, PAGE_SIZE, size); | |
1110 | ||
1111 | page = alloc_page(GFP_NOIO); | |
1112 | if (unlikely(!page)) | |
1113 | goto free_pages; | |
1114 | ||
1115 | bio_add_page(behind_bio, page, len, 0); | |
1116 | ||
1117 | size -= len; | |
1118 | i++; | |
4b6d287f | 1119 | } |
841c1316 | 1120 | |
cb83efcf | 1121 | bio_copy_data(behind_bio, bio); |
41743c1f | 1122 | skip_copy: |
56a64c17 | 1123 | r1_bio->behind_master_bio = behind_bio; |
af6d7b76 | 1124 | set_bit(R1BIO_BehindIO, &r1_bio->state); |
4b6d287f | 1125 | |
16d56e2f | 1126 | return; |
841c1316 ML |
1127 | |
1128 | free_pages: | |
4f024f37 KO |
1129 | pr_debug("%dB behind alloc failed, doing sync I/O\n", |
1130 | bio->bi_iter.bi_size); | |
841c1316 | 1131 | bio_free_pages(behind_bio); |
16d56e2f | 1132 | bio_put(behind_bio); |
4b6d287f N |
1133 | } |
1134 | ||
f54a9d0e N |
1135 | struct raid1_plug_cb { |
1136 | struct blk_plug_cb cb; | |
1137 | struct bio_list pending; | |
1138 | int pending_cnt; | |
1139 | }; | |
1140 | ||
1141 | static void raid1_unplug(struct blk_plug_cb *cb, bool from_schedule) | |
1142 | { | |
1143 | struct raid1_plug_cb *plug = container_of(cb, struct raid1_plug_cb, | |
1144 | cb); | |
1145 | struct mddev *mddev = plug->cb.data; | |
1146 | struct r1conf *conf = mddev->private; | |
1147 | struct bio *bio; | |
1148 | ||
874807a8 | 1149 | if (from_schedule || current->bio_list) { |
f54a9d0e N |
1150 | spin_lock_irq(&conf->device_lock); |
1151 | bio_list_merge(&conf->pending_bio_list, &plug->pending); | |
1152 | conf->pending_count += plug->pending_cnt; | |
1153 | spin_unlock_irq(&conf->device_lock); | |
ee0b0244 | 1154 | wake_up(&conf->wait_barrier); |
f54a9d0e N |
1155 | md_wakeup_thread(mddev->thread); |
1156 | kfree(plug); | |
1157 | return; | |
1158 | } | |
1159 | ||
1160 | /* we aren't scheduling, so we can do the write-out directly. */ | |
1161 | bio = bio_list_get(&plug->pending); | |
673ca68d | 1162 | flush_bio_list(conf, bio); |
f54a9d0e N |
1163 | kfree(plug); |
1164 | } | |
1165 | ||
689389a0 N |
1166 | static void init_r1bio(struct r1bio *r1_bio, struct mddev *mddev, struct bio *bio) |
1167 | { | |
1168 | r1_bio->master_bio = bio; | |
1169 | r1_bio->sectors = bio_sectors(bio); | |
1170 | r1_bio->state = 0; | |
1171 | r1_bio->mddev = mddev; | |
1172 | r1_bio->sector = bio->bi_iter.bi_sector; | |
1173 | } | |
1174 | ||
fd76863e | 1175 | static inline struct r1bio * |
689389a0 | 1176 | alloc_r1bio(struct mddev *mddev, struct bio *bio) |
fd76863e | 1177 | { |
1178 | struct r1conf *conf = mddev->private; | |
1179 | struct r1bio *r1_bio; | |
1180 | ||
afeee514 | 1181 | r1_bio = mempool_alloc(&conf->r1bio_pool, GFP_NOIO); |
689389a0 N |
1182 | /* Ensure no bio records IO_BLOCKED */ |
1183 | memset(r1_bio->bios, 0, conf->raid_disks * sizeof(r1_bio->bios[0])); | |
1184 | init_r1bio(r1_bio, mddev, bio); | |
fd76863e | 1185 | return r1_bio; |
1186 | } | |
1187 | ||
c230e7e5 | 1188 | static void raid1_read_request(struct mddev *mddev, struct bio *bio, |
689389a0 | 1189 | int max_read_sectors, struct r1bio *r1_bio) |
1da177e4 | 1190 | { |
e8096360 | 1191 | struct r1conf *conf = mddev->private; |
0eaf822c | 1192 | struct raid1_info *mirror; |
1da177e4 | 1193 | struct bio *read_bio; |
3b046a97 RL |
1194 | struct bitmap *bitmap = mddev->bitmap; |
1195 | const int op = bio_op(bio); | |
1196 | const unsigned long do_sync = (bio->bi_opf & REQ_SYNC); | |
3b046a97 RL |
1197 | int max_sectors; |
1198 | int rdisk; | |
689389a0 N |
1199 | bool print_msg = !!r1_bio; |
1200 | char b[BDEVNAME_SIZE]; | |
3b046a97 | 1201 | |
fd76863e | 1202 | /* |
689389a0 N |
1203 | * If r1_bio is set, we are blocking the raid1d thread |
1204 | * so there is a tiny risk of deadlock. So ask for | |
1205 | * emergency memory if needed. | |
fd76863e | 1206 | */ |
689389a0 | 1207 | gfp_t gfp = r1_bio ? (GFP_NOIO | __GFP_HIGH) : GFP_NOIO; |
fd76863e | 1208 | |
689389a0 N |
1209 | if (print_msg) { |
1210 | /* Need to get the block device name carefully */ | |
1211 | struct md_rdev *rdev; | |
1212 | rcu_read_lock(); | |
1213 | rdev = rcu_dereference(conf->mirrors[r1_bio->read_disk].rdev); | |
1214 | if (rdev) | |
1215 | bdevname(rdev->bdev, b); | |
1216 | else | |
1217 | strcpy(b, "???"); | |
1218 | rcu_read_unlock(); | |
1219 | } | |
3b046a97 | 1220 | |
fd76863e | 1221 | /* |
fd76863e | 1222 | * Still need barrier for READ in case that whole |
1223 | * array is frozen. | |
fd76863e | 1224 | */ |
fd76863e | 1225 | wait_read_barrier(conf, bio->bi_iter.bi_sector); |
1226 | ||
689389a0 N |
1227 | if (!r1_bio) |
1228 | r1_bio = alloc_r1bio(mddev, bio); | |
1229 | else | |
1230 | init_r1bio(r1_bio, mddev, bio); | |
c230e7e5 | 1231 | r1_bio->sectors = max_read_sectors; |
fd76863e | 1232 | |
1233 | /* | |
1234 | * make_request() can abort the operation when read-ahead is being | |
1235 | * used and no empty request is available. | |
1236 | */ | |
3b046a97 RL |
1237 | rdisk = read_balance(conf, r1_bio, &max_sectors); |
1238 | ||
1239 | if (rdisk < 0) { | |
1240 | /* couldn't find anywhere to read from */ | |
689389a0 N |
1241 | if (print_msg) { |
1242 | pr_crit_ratelimited("md/raid1:%s: %s: unrecoverable I/O read error for block %llu\n", | |
1243 | mdname(mddev), | |
1244 | b, | |
1245 | (unsigned long long)r1_bio->sector); | |
1246 | } | |
3b046a97 RL |
1247 | raid_end_bio_io(r1_bio); |
1248 | return; | |
1249 | } | |
1250 | mirror = conf->mirrors + rdisk; | |
1251 | ||
689389a0 N |
1252 | if (print_msg) |
1253 | pr_info_ratelimited("md/raid1:%s: redirecting sector %llu to other mirror: %s\n", | |
1254 | mdname(mddev), | |
1255 | (unsigned long long)r1_bio->sector, | |
1256 | bdevname(mirror->rdev->bdev, b)); | |
1257 | ||
3b046a97 RL |
1258 | if (test_bit(WriteMostly, &mirror->rdev->flags) && |
1259 | bitmap) { | |
1260 | /* | |
1261 | * Reading from a write-mostly device must take care not to | |
1262 | * over-take any writes that are 'behind' | |
1263 | */ | |
1264 | raid1_log(mddev, "wait behind writes"); | |
1265 | wait_event(bitmap->behind_wait, | |
1266 | atomic_read(&bitmap->behind_writes) == 0); | |
1267 | } | |
c230e7e5 N |
1268 | |
1269 | if (max_sectors < bio_sectors(bio)) { | |
1270 | struct bio *split = bio_split(bio, max_sectors, | |
afeee514 | 1271 | gfp, &conf->bio_split); |
c230e7e5 N |
1272 | bio_chain(split, bio); |
1273 | generic_make_request(bio); | |
1274 | bio = split; | |
1275 | r1_bio->master_bio = bio; | |
1276 | r1_bio->sectors = max_sectors; | |
1277 | } | |
1278 | ||
3b046a97 | 1279 | r1_bio->read_disk = rdisk; |
3b046a97 | 1280 | |
afeee514 | 1281 | read_bio = bio_clone_fast(bio, gfp, &mddev->bio_set); |
3b046a97 RL |
1282 | |
1283 | r1_bio->bios[rdisk] = read_bio; | |
1284 | ||
1285 | read_bio->bi_iter.bi_sector = r1_bio->sector + | |
1286 | mirror->rdev->data_offset; | |
74d46992 | 1287 | bio_set_dev(read_bio, mirror->rdev->bdev); |
3b046a97 RL |
1288 | read_bio->bi_end_io = raid1_end_read_request; |
1289 | bio_set_op_attrs(read_bio, op, do_sync); | |
1290 | if (test_bit(FailFast, &mirror->rdev->flags) && | |
1291 | test_bit(R1BIO_FailFast, &r1_bio->state)) | |
1292 | read_bio->bi_opf |= MD_FAILFAST; | |
1293 | read_bio->bi_private = r1_bio; | |
1294 | ||
1295 | if (mddev->gendisk) | |
74d46992 CH |
1296 | trace_block_bio_remap(read_bio->bi_disk->queue, read_bio, |
1297 | disk_devt(mddev->gendisk), r1_bio->sector); | |
3b046a97 | 1298 | |
c230e7e5 | 1299 | generic_make_request(read_bio); |
3b046a97 RL |
1300 | } |
1301 | ||
c230e7e5 N |
1302 | static void raid1_write_request(struct mddev *mddev, struct bio *bio, |
1303 | int max_write_sectors) | |
3b046a97 RL |
1304 | { |
1305 | struct r1conf *conf = mddev->private; | |
fd76863e | 1306 | struct r1bio *r1_bio; |
1f68f0c4 | 1307 | int i, disks; |
3b046a97 | 1308 | struct bitmap *bitmap = mddev->bitmap; |
191ea9b2 | 1309 | unsigned long flags; |
3cb03002 | 1310 | struct md_rdev *blocked_rdev; |
f54a9d0e N |
1311 | struct blk_plug_cb *cb; |
1312 | struct raid1_plug_cb *plug = NULL; | |
1f68f0c4 | 1313 | int first_clone; |
1f68f0c4 | 1314 | int max_sectors; |
191ea9b2 | 1315 | |
b3143b9a | 1316 | if (mddev_is_clustered(mddev) && |
90382ed9 | 1317 | md_cluster_ops->area_resyncing(mddev, WRITE, |
b3143b9a | 1318 | bio->bi_iter.bi_sector, bio_end_sector(bio))) { |
3b046a97 | 1319 | |
6eef4b21 N |
1320 | DEFINE_WAIT(w); |
1321 | for (;;) { | |
6eef4b21 | 1322 | prepare_to_wait(&conf->wait_barrier, |
ae89fd3d | 1323 | &w, TASK_IDLE); |
f81f7302 | 1324 | if (!md_cluster_ops->area_resyncing(mddev, WRITE, |
385f4d7f | 1325 | bio->bi_iter.bi_sector, |
b3143b9a | 1326 | bio_end_sector(bio))) |
6eef4b21 N |
1327 | break; |
1328 | schedule(); | |
1329 | } | |
1330 | finish_wait(&conf->wait_barrier, &w); | |
1331 | } | |
f81f7302 GJ |
1332 | |
1333 | /* | |
1334 | * Register the new request and wait if the reconstruction | |
1335 | * thread has put up a bar for new requests. | |
1336 | * Continue immediately if no resync is active currently. | |
1337 | */ | |
fd76863e | 1338 | wait_barrier(conf, bio->bi_iter.bi_sector); |
1339 | ||
689389a0 | 1340 | r1_bio = alloc_r1bio(mddev, bio); |
c230e7e5 | 1341 | r1_bio->sectors = max_write_sectors; |
1da177e4 | 1342 | |
34db0cd6 N |
1343 | if (conf->pending_count >= max_queued_requests) { |
1344 | md_wakeup_thread(mddev->thread); | |
578b54ad | 1345 | raid1_log(mddev, "wait queued"); |
34db0cd6 N |
1346 | wait_event(conf->wait_barrier, |
1347 | conf->pending_count < max_queued_requests); | |
1348 | } | |
1f68f0c4 | 1349 | /* first select target devices under rcu_lock and |
1da177e4 LT |
1350 | * inc refcount on their rdev. Record them by setting |
1351 | * bios[x] to bio | |
1f68f0c4 N |
1352 | * If there are known/acknowledged bad blocks on any device on |
1353 | * which we have seen a write error, we want to avoid writing those | |
1354 | * blocks. | |
1355 | * This potentially requires several writes to write around | |
1356 | * the bad blocks. Each set of writes gets it's own r1bio | |
1357 | * with a set of bios attached. | |
1da177e4 | 1358 | */ |
c3b328ac | 1359 | |
8f19ccb2 | 1360 | disks = conf->raid_disks * 2; |
6bfe0b49 DW |
1361 | retry_write: |
1362 | blocked_rdev = NULL; | |
1da177e4 | 1363 | rcu_read_lock(); |
1f68f0c4 | 1364 | max_sectors = r1_bio->sectors; |
1da177e4 | 1365 | for (i = 0; i < disks; i++) { |
3cb03002 | 1366 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
6bfe0b49 DW |
1367 | if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { |
1368 | atomic_inc(&rdev->nr_pending); | |
1369 | blocked_rdev = rdev; | |
1370 | break; | |
1371 | } | |
1f68f0c4 | 1372 | r1_bio->bios[i] = NULL; |
8ae12666 | 1373 | if (!rdev || test_bit(Faulty, &rdev->flags)) { |
8f19ccb2 N |
1374 | if (i < conf->raid_disks) |
1375 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
1f68f0c4 N |
1376 | continue; |
1377 | } | |
1378 | ||
1379 | atomic_inc(&rdev->nr_pending); | |
1380 | if (test_bit(WriteErrorSeen, &rdev->flags)) { | |
1381 | sector_t first_bad; | |
1382 | int bad_sectors; | |
1383 | int is_bad; | |
1384 | ||
3b046a97 | 1385 | is_bad = is_badblock(rdev, r1_bio->sector, max_sectors, |
1f68f0c4 N |
1386 | &first_bad, &bad_sectors); |
1387 | if (is_bad < 0) { | |
1388 | /* mustn't write here until the bad block is | |
1389 | * acknowledged*/ | |
1390 | set_bit(BlockedBadBlocks, &rdev->flags); | |
1391 | blocked_rdev = rdev; | |
1392 | break; | |
1393 | } | |
1394 | if (is_bad && first_bad <= r1_bio->sector) { | |
1395 | /* Cannot write here at all */ | |
1396 | bad_sectors -= (r1_bio->sector - first_bad); | |
1397 | if (bad_sectors < max_sectors) | |
1398 | /* mustn't write more than bad_sectors | |
1399 | * to other devices yet | |
1400 | */ | |
1401 | max_sectors = bad_sectors; | |
03c902e1 | 1402 | rdev_dec_pending(rdev, mddev); |
1f68f0c4 N |
1403 | /* We don't set R1BIO_Degraded as that |
1404 | * only applies if the disk is | |
1405 | * missing, so it might be re-added, | |
1406 | * and we want to know to recover this | |
1407 | * chunk. | |
1408 | * In this case the device is here, | |
1409 | * and the fact that this chunk is not | |
1410 | * in-sync is recorded in the bad | |
1411 | * block log | |
1412 | */ | |
1413 | continue; | |
964147d5 | 1414 | } |
1f68f0c4 N |
1415 | if (is_bad) { |
1416 | int good_sectors = first_bad - r1_bio->sector; | |
1417 | if (good_sectors < max_sectors) | |
1418 | max_sectors = good_sectors; | |
1419 | } | |
1420 | } | |
1421 | r1_bio->bios[i] = bio; | |
1da177e4 LT |
1422 | } |
1423 | rcu_read_unlock(); | |
1424 | ||
6bfe0b49 DW |
1425 | if (unlikely(blocked_rdev)) { |
1426 | /* Wait for this device to become unblocked */ | |
1427 | int j; | |
1428 | ||
1429 | for (j = 0; j < i; j++) | |
1430 | if (r1_bio->bios[j]) | |
1431 | rdev_dec_pending(conf->mirrors[j].rdev, mddev); | |
1f68f0c4 | 1432 | r1_bio->state = 0; |
fd76863e | 1433 | allow_barrier(conf, bio->bi_iter.bi_sector); |
578b54ad | 1434 | raid1_log(mddev, "wait rdev %d blocked", blocked_rdev->raid_disk); |
6bfe0b49 | 1435 | md_wait_for_blocked_rdev(blocked_rdev, mddev); |
fd76863e | 1436 | wait_barrier(conf, bio->bi_iter.bi_sector); |
6bfe0b49 DW |
1437 | goto retry_write; |
1438 | } | |
1439 | ||
c230e7e5 N |
1440 | if (max_sectors < bio_sectors(bio)) { |
1441 | struct bio *split = bio_split(bio, max_sectors, | |
afeee514 | 1442 | GFP_NOIO, &conf->bio_split); |
c230e7e5 N |
1443 | bio_chain(split, bio); |
1444 | generic_make_request(bio); | |
1445 | bio = split; | |
1446 | r1_bio->master_bio = bio; | |
1f68f0c4 | 1447 | r1_bio->sectors = max_sectors; |
191ea9b2 | 1448 | } |
4b6d287f | 1449 | |
4e78064f | 1450 | atomic_set(&r1_bio->remaining, 1); |
4b6d287f | 1451 | atomic_set(&r1_bio->behind_remaining, 0); |
06d91a5f | 1452 | |
1f68f0c4 | 1453 | first_clone = 1; |
d8c84c4f | 1454 | |
1da177e4 | 1455 | for (i = 0; i < disks; i++) { |
8e58e327 | 1456 | struct bio *mbio = NULL; |
1da177e4 LT |
1457 | if (!r1_bio->bios[i]) |
1458 | continue; | |
1459 | ||
1f68f0c4 N |
1460 | |
1461 | if (first_clone) { | |
1462 | /* do behind I/O ? | |
1463 | * Not if there are too many, or cannot | |
1464 | * allocate memory, or a reader on WriteMostly | |
1465 | * is waiting for behind writes to flush */ | |
1466 | if (bitmap && | |
1467 | (atomic_read(&bitmap->behind_writes) | |
1468 | < mddev->bitmap_info.max_write_behind) && | |
8e58e327 | 1469 | !waitqueue_active(&bitmap->behind_wait)) { |
16d56e2f | 1470 | alloc_behind_master_bio(r1_bio, bio); |
8e58e327 | 1471 | } |
1f68f0c4 | 1472 | |
e64e4018 AS |
1473 | md_bitmap_startwrite(bitmap, r1_bio->sector, r1_bio->sectors, |
1474 | test_bit(R1BIO_BehindIO, &r1_bio->state)); | |
1f68f0c4 N |
1475 | first_clone = 0; |
1476 | } | |
8e58e327 | 1477 | |
16d56e2f SL |
1478 | if (r1_bio->behind_master_bio) |
1479 | mbio = bio_clone_fast(r1_bio->behind_master_bio, | |
afeee514 | 1480 | GFP_NOIO, &mddev->bio_set); |
16d56e2f | 1481 | else |
afeee514 | 1482 | mbio = bio_clone_fast(bio, GFP_NOIO, &mddev->bio_set); |
8e58e327 | 1483 | |
841c1316 | 1484 | if (r1_bio->behind_master_bio) { |
4b6d287f N |
1485 | if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags)) |
1486 | atomic_inc(&r1_bio->behind_remaining); | |
1487 | } | |
1488 | ||
1f68f0c4 N |
1489 | r1_bio->bios[i] = mbio; |
1490 | ||
4f024f37 | 1491 | mbio->bi_iter.bi_sector = (r1_bio->sector + |
1f68f0c4 | 1492 | conf->mirrors[i].rdev->data_offset); |
74d46992 | 1493 | bio_set_dev(mbio, conf->mirrors[i].rdev->bdev); |
1f68f0c4 | 1494 | mbio->bi_end_io = raid1_end_write_request; |
a682e003 | 1495 | mbio->bi_opf = bio_op(bio) | (bio->bi_opf & (REQ_SYNC | REQ_FUA)); |
212e7eb7 N |
1496 | if (test_bit(FailFast, &conf->mirrors[i].rdev->flags) && |
1497 | !test_bit(WriteMostly, &conf->mirrors[i].rdev->flags) && | |
1498 | conf->raid_disks - mddev->degraded > 1) | |
1499 | mbio->bi_opf |= MD_FAILFAST; | |
1f68f0c4 N |
1500 | mbio->bi_private = r1_bio; |
1501 | ||
1da177e4 | 1502 | atomic_inc(&r1_bio->remaining); |
f54a9d0e | 1503 | |
109e3765 | 1504 | if (mddev->gendisk) |
74d46992 | 1505 | trace_block_bio_remap(mbio->bi_disk->queue, |
109e3765 N |
1506 | mbio, disk_devt(mddev->gendisk), |
1507 | r1_bio->sector); | |
1508 | /* flush_pending_writes() needs access to the rdev so...*/ | |
74d46992 | 1509 | mbio->bi_disk = (void *)conf->mirrors[i].rdev; |
109e3765 | 1510 | |
f54a9d0e N |
1511 | cb = blk_check_plugged(raid1_unplug, mddev, sizeof(*plug)); |
1512 | if (cb) | |
1513 | plug = container_of(cb, struct raid1_plug_cb, cb); | |
1514 | else | |
1515 | plug = NULL; | |
f54a9d0e N |
1516 | if (plug) { |
1517 | bio_list_add(&plug->pending, mbio); | |
1518 | plug->pending_cnt++; | |
1519 | } else { | |
23b245c0 | 1520 | spin_lock_irqsave(&conf->device_lock, flags); |
f54a9d0e N |
1521 | bio_list_add(&conf->pending_bio_list, mbio); |
1522 | conf->pending_count++; | |
23b245c0 | 1523 | spin_unlock_irqrestore(&conf->device_lock, flags); |
b357f04a | 1524 | md_wakeup_thread(mddev->thread); |
23b245c0 | 1525 | } |
1da177e4 | 1526 | } |
1f68f0c4 | 1527 | |
079fa166 N |
1528 | r1_bio_write_done(r1_bio); |
1529 | ||
1530 | /* In case raid1d snuck in to freeze_array */ | |
1531 | wake_up(&conf->wait_barrier); | |
1da177e4 LT |
1532 | } |
1533 | ||
cc27b0c7 | 1534 | static bool raid1_make_request(struct mddev *mddev, struct bio *bio) |
3b046a97 | 1535 | { |
fd76863e | 1536 | sector_t sectors; |
3b046a97 | 1537 | |
aff8da09 SL |
1538 | if (unlikely(bio->bi_opf & REQ_PREFLUSH)) { |
1539 | md_flush_request(mddev, bio); | |
cc27b0c7 | 1540 | return true; |
aff8da09 | 1541 | } |
3b046a97 | 1542 | |
c230e7e5 N |
1543 | /* |
1544 | * There is a limit to the maximum size, but | |
1545 | * the read/write handler might find a lower limit | |
1546 | * due to bad blocks. To avoid multiple splits, | |
1547 | * we pass the maximum number of sectors down | |
1548 | * and let the lower level perform the split. | |
1549 | */ | |
1550 | sectors = align_to_barrier_unit_end( | |
1551 | bio->bi_iter.bi_sector, bio_sectors(bio)); | |
61eb2b43 | 1552 | |
c230e7e5 | 1553 | if (bio_data_dir(bio) == READ) |
689389a0 | 1554 | raid1_read_request(mddev, bio, sectors, NULL); |
cc27b0c7 N |
1555 | else { |
1556 | if (!md_write_start(mddev,bio)) | |
1557 | return false; | |
c230e7e5 | 1558 | raid1_write_request(mddev, bio, sectors); |
cc27b0c7 N |
1559 | } |
1560 | return true; | |
3b046a97 RL |
1561 | } |
1562 | ||
849674e4 | 1563 | static void raid1_status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 1564 | { |
e8096360 | 1565 | struct r1conf *conf = mddev->private; |
1da177e4 LT |
1566 | int i; |
1567 | ||
1568 | seq_printf(seq, " [%d/%d] [", conf->raid_disks, | |
11ce99e6 | 1569 | conf->raid_disks - mddev->degraded); |
ddac7c7e N |
1570 | rcu_read_lock(); |
1571 | for (i = 0; i < conf->raid_disks; i++) { | |
3cb03002 | 1572 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
1da177e4 | 1573 | seq_printf(seq, "%s", |
ddac7c7e N |
1574 | rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); |
1575 | } | |
1576 | rcu_read_unlock(); | |
1da177e4 LT |
1577 | seq_printf(seq, "]"); |
1578 | } | |
1579 | ||
849674e4 | 1580 | static void raid1_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1581 | { |
1582 | char b[BDEVNAME_SIZE]; | |
e8096360 | 1583 | struct r1conf *conf = mddev->private; |
423f04d6 | 1584 | unsigned long flags; |
1da177e4 LT |
1585 | |
1586 | /* | |
1587 | * If it is not operational, then we have already marked it as dead | |
1588 | * else if it is the last working disks, ignore the error, let the | |
1589 | * next level up know. | |
1590 | * else mark the drive as failed | |
1591 | */ | |
2e52d449 | 1592 | spin_lock_irqsave(&conf->device_lock, flags); |
b2d444d7 | 1593 | if (test_bit(In_sync, &rdev->flags) |
4044ba58 | 1594 | && (conf->raid_disks - mddev->degraded) == 1) { |
1da177e4 LT |
1595 | /* |
1596 | * Don't fail the drive, act as though we were just a | |
4044ba58 N |
1597 | * normal single drive. |
1598 | * However don't try a recovery from this drive as | |
1599 | * it is very likely to fail. | |
1da177e4 | 1600 | */ |
5389042f | 1601 | conf->recovery_disabled = mddev->recovery_disabled; |
2e52d449 | 1602 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 1603 | return; |
4044ba58 | 1604 | } |
de393cde | 1605 | set_bit(Blocked, &rdev->flags); |
c04be0aa | 1606 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
1da177e4 | 1607 | mddev->degraded++; |
dd00a99e | 1608 | set_bit(Faulty, &rdev->flags); |
dd00a99e N |
1609 | } else |
1610 | set_bit(Faulty, &rdev->flags); | |
423f04d6 | 1611 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2446dba0 N |
1612 | /* |
1613 | * if recovery is running, make sure it aborts. | |
1614 | */ | |
1615 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2953079c SL |
1616 | set_mask_bits(&mddev->sb_flags, 0, |
1617 | BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING)); | |
1d41c216 N |
1618 | pr_crit("md/raid1:%s: Disk failure on %s, disabling device.\n" |
1619 | "md/raid1:%s: Operation continuing on %d devices.\n", | |
1620 | mdname(mddev), bdevname(rdev->bdev, b), | |
1621 | mdname(mddev), conf->raid_disks - mddev->degraded); | |
1da177e4 LT |
1622 | } |
1623 | ||
e8096360 | 1624 | static void print_conf(struct r1conf *conf) |
1da177e4 LT |
1625 | { |
1626 | int i; | |
1da177e4 | 1627 | |
1d41c216 | 1628 | pr_debug("RAID1 conf printout:\n"); |
1da177e4 | 1629 | if (!conf) { |
1d41c216 | 1630 | pr_debug("(!conf)\n"); |
1da177e4 LT |
1631 | return; |
1632 | } | |
1d41c216 N |
1633 | pr_debug(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded, |
1634 | conf->raid_disks); | |
1da177e4 | 1635 | |
ddac7c7e | 1636 | rcu_read_lock(); |
1da177e4 LT |
1637 | for (i = 0; i < conf->raid_disks; i++) { |
1638 | char b[BDEVNAME_SIZE]; | |
3cb03002 | 1639 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
ddac7c7e | 1640 | if (rdev) |
1d41c216 N |
1641 | pr_debug(" disk %d, wo:%d, o:%d, dev:%s\n", |
1642 | i, !test_bit(In_sync, &rdev->flags), | |
1643 | !test_bit(Faulty, &rdev->flags), | |
1644 | bdevname(rdev->bdev,b)); | |
1da177e4 | 1645 | } |
ddac7c7e | 1646 | rcu_read_unlock(); |
1da177e4 LT |
1647 | } |
1648 | ||
e8096360 | 1649 | static void close_sync(struct r1conf *conf) |
1da177e4 | 1650 | { |
f6eca2d4 ND |
1651 | int idx; |
1652 | ||
1653 | for (idx = 0; idx < BARRIER_BUCKETS_NR; idx++) { | |
1654 | _wait_barrier(conf, idx); | |
1655 | _allow_barrier(conf, idx); | |
1656 | } | |
1da177e4 | 1657 | |
afeee514 | 1658 | mempool_exit(&conf->r1buf_pool); |
1da177e4 LT |
1659 | } |
1660 | ||
fd01b88c | 1661 | static int raid1_spare_active(struct mddev *mddev) |
1da177e4 LT |
1662 | { |
1663 | int i; | |
e8096360 | 1664 | struct r1conf *conf = mddev->private; |
6b965620 N |
1665 | int count = 0; |
1666 | unsigned long flags; | |
1da177e4 LT |
1667 | |
1668 | /* | |
f72ffdd6 | 1669 | * Find all failed disks within the RAID1 configuration |
ddac7c7e N |
1670 | * and mark them readable. |
1671 | * Called under mddev lock, so rcu protection not needed. | |
423f04d6 N |
1672 | * device_lock used to avoid races with raid1_end_read_request |
1673 | * which expects 'In_sync' flags and ->degraded to be consistent. | |
1da177e4 | 1674 | */ |
423f04d6 | 1675 | spin_lock_irqsave(&conf->device_lock, flags); |
1da177e4 | 1676 | for (i = 0; i < conf->raid_disks; i++) { |
3cb03002 | 1677 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
8c7a2c2b N |
1678 | struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev; |
1679 | if (repl | |
1aee41f6 | 1680 | && !test_bit(Candidate, &repl->flags) |
8c7a2c2b N |
1681 | && repl->recovery_offset == MaxSector |
1682 | && !test_bit(Faulty, &repl->flags) | |
1683 | && !test_and_set_bit(In_sync, &repl->flags)) { | |
1684 | /* replacement has just become active */ | |
1685 | if (!rdev || | |
1686 | !test_and_clear_bit(In_sync, &rdev->flags)) | |
1687 | count++; | |
1688 | if (rdev) { | |
1689 | /* Replaced device not technically | |
1690 | * faulty, but we need to be sure | |
1691 | * it gets removed and never re-added | |
1692 | */ | |
1693 | set_bit(Faulty, &rdev->flags); | |
1694 | sysfs_notify_dirent_safe( | |
1695 | rdev->sysfs_state); | |
1696 | } | |
1697 | } | |
ddac7c7e | 1698 | if (rdev |
61e4947c | 1699 | && rdev->recovery_offset == MaxSector |
ddac7c7e | 1700 | && !test_bit(Faulty, &rdev->flags) |
c04be0aa | 1701 | && !test_and_set_bit(In_sync, &rdev->flags)) { |
6b965620 | 1702 | count++; |
654e8b5a | 1703 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
1704 | } |
1705 | } | |
6b965620 N |
1706 | mddev->degraded -= count; |
1707 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 LT |
1708 | |
1709 | print_conf(conf); | |
6b965620 | 1710 | return count; |
1da177e4 LT |
1711 | } |
1712 | ||
fd01b88c | 1713 | static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1714 | { |
e8096360 | 1715 | struct r1conf *conf = mddev->private; |
199050ea | 1716 | int err = -EEXIST; |
41158c7e | 1717 | int mirror = 0; |
0eaf822c | 1718 | struct raid1_info *p; |
6c2fce2e | 1719 | int first = 0; |
30194636 | 1720 | int last = conf->raid_disks - 1; |
1da177e4 | 1721 | |
5389042f N |
1722 | if (mddev->recovery_disabled == conf->recovery_disabled) |
1723 | return -EBUSY; | |
1724 | ||
1501efad DW |
1725 | if (md_integrity_add_rdev(rdev, mddev)) |
1726 | return -ENXIO; | |
1727 | ||
6c2fce2e NB |
1728 | if (rdev->raid_disk >= 0) |
1729 | first = last = rdev->raid_disk; | |
1730 | ||
70bcecdb GR |
1731 | /* |
1732 | * find the disk ... but prefer rdev->saved_raid_disk | |
1733 | * if possible. | |
1734 | */ | |
1735 | if (rdev->saved_raid_disk >= 0 && | |
1736 | rdev->saved_raid_disk >= first && | |
9e753ba9 | 1737 | rdev->saved_raid_disk < conf->raid_disks && |
70bcecdb GR |
1738 | conf->mirrors[rdev->saved_raid_disk].rdev == NULL) |
1739 | first = last = rdev->saved_raid_disk; | |
1740 | ||
7ef449d1 N |
1741 | for (mirror = first; mirror <= last; mirror++) { |
1742 | p = conf->mirrors+mirror; | |
1743 | if (!p->rdev) { | |
1da177e4 | 1744 | |
9092c02d JB |
1745 | if (mddev->gendisk) |
1746 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
1747 | rdev->data_offset << 9); | |
1da177e4 LT |
1748 | |
1749 | p->head_position = 0; | |
1750 | rdev->raid_disk = mirror; | |
199050ea | 1751 | err = 0; |
6aea114a N |
1752 | /* As all devices are equivalent, we don't need a full recovery |
1753 | * if this was recently any drive of the array | |
1754 | */ | |
1755 | if (rdev->saved_raid_disk < 0) | |
41158c7e | 1756 | conf->fullsync = 1; |
d6065f7b | 1757 | rcu_assign_pointer(p->rdev, rdev); |
1da177e4 LT |
1758 | break; |
1759 | } | |
7ef449d1 N |
1760 | if (test_bit(WantReplacement, &p->rdev->flags) && |
1761 | p[conf->raid_disks].rdev == NULL) { | |
1762 | /* Add this device as a replacement */ | |
1763 | clear_bit(In_sync, &rdev->flags); | |
1764 | set_bit(Replacement, &rdev->flags); | |
1765 | rdev->raid_disk = mirror; | |
1766 | err = 0; | |
1767 | conf->fullsync = 1; | |
1768 | rcu_assign_pointer(p[conf->raid_disks].rdev, rdev); | |
1769 | break; | |
1770 | } | |
1771 | } | |
9092c02d | 1772 | if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev))) |
8b904b5b | 1773 | blk_queue_flag_set(QUEUE_FLAG_DISCARD, mddev->queue); |
1da177e4 | 1774 | print_conf(conf); |
199050ea | 1775 | return err; |
1da177e4 LT |
1776 | } |
1777 | ||
b8321b68 | 1778 | static int raid1_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1779 | { |
e8096360 | 1780 | struct r1conf *conf = mddev->private; |
1da177e4 | 1781 | int err = 0; |
b8321b68 | 1782 | int number = rdev->raid_disk; |
0eaf822c | 1783 | struct raid1_info *p = conf->mirrors + number; |
1da177e4 | 1784 | |
b014f14c N |
1785 | if (rdev != p->rdev) |
1786 | p = conf->mirrors + conf->raid_disks + number; | |
1787 | ||
1da177e4 | 1788 | print_conf(conf); |
b8321b68 | 1789 | if (rdev == p->rdev) { |
b2d444d7 | 1790 | if (test_bit(In_sync, &rdev->flags) || |
1da177e4 LT |
1791 | atomic_read(&rdev->nr_pending)) { |
1792 | err = -EBUSY; | |
1793 | goto abort; | |
1794 | } | |
046abeed | 1795 | /* Only remove non-faulty devices if recovery |
dfc70645 N |
1796 | * is not possible. |
1797 | */ | |
1798 | if (!test_bit(Faulty, &rdev->flags) && | |
5389042f | 1799 | mddev->recovery_disabled != conf->recovery_disabled && |
dfc70645 N |
1800 | mddev->degraded < conf->raid_disks) { |
1801 | err = -EBUSY; | |
1802 | goto abort; | |
1803 | } | |
1da177e4 | 1804 | p->rdev = NULL; |
d787be40 N |
1805 | if (!test_bit(RemoveSynchronized, &rdev->flags)) { |
1806 | synchronize_rcu(); | |
1807 | if (atomic_read(&rdev->nr_pending)) { | |
1808 | /* lost the race, try later */ | |
1809 | err = -EBUSY; | |
1810 | p->rdev = rdev; | |
1811 | goto abort; | |
1812 | } | |
1813 | } | |
1814 | if (conf->mirrors[conf->raid_disks + number].rdev) { | |
8c7a2c2b N |
1815 | /* We just removed a device that is being replaced. |
1816 | * Move down the replacement. We drain all IO before | |
1817 | * doing this to avoid confusion. | |
1818 | */ | |
1819 | struct md_rdev *repl = | |
1820 | conf->mirrors[conf->raid_disks + number].rdev; | |
e2d59925 | 1821 | freeze_array(conf, 0); |
3de59bb9 YY |
1822 | if (atomic_read(&repl->nr_pending)) { |
1823 | /* It means that some queued IO of retry_list | |
1824 | * hold repl. Thus, we cannot set replacement | |
1825 | * as NULL, avoiding rdev NULL pointer | |
1826 | * dereference in sync_request_write and | |
1827 | * handle_write_finished. | |
1828 | */ | |
1829 | err = -EBUSY; | |
1830 | unfreeze_array(conf); | |
1831 | goto abort; | |
1832 | } | |
8c7a2c2b N |
1833 | clear_bit(Replacement, &repl->flags); |
1834 | p->rdev = repl; | |
1835 | conf->mirrors[conf->raid_disks + number].rdev = NULL; | |
e2d59925 | 1836 | unfreeze_array(conf); |
e5bc9c3c GJ |
1837 | } |
1838 | ||
1839 | clear_bit(WantReplacement, &rdev->flags); | |
a91a2785 | 1840 | err = md_integrity_register(mddev); |
1da177e4 LT |
1841 | } |
1842 | abort: | |
1843 | ||
1844 | print_conf(conf); | |
1845 | return err; | |
1846 | } | |
1847 | ||
4246a0b6 | 1848 | static void end_sync_read(struct bio *bio) |
1da177e4 | 1849 | { |
98d30c58 | 1850 | struct r1bio *r1_bio = get_resync_r1bio(bio); |
1da177e4 | 1851 | |
0fc280f6 | 1852 | update_head_pos(r1_bio->read_disk, r1_bio); |
ba3ae3be | 1853 | |
1da177e4 LT |
1854 | /* |
1855 | * we have read a block, now it needs to be re-written, | |
1856 | * or re-read if the read failed. | |
1857 | * We don't do much here, just schedule handling by raid1d | |
1858 | */ | |
4e4cbee9 | 1859 | if (!bio->bi_status) |
1da177e4 | 1860 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
d11c171e N |
1861 | |
1862 | if (atomic_dec_and_test(&r1_bio->remaining)) | |
1863 | reschedule_retry(r1_bio); | |
1da177e4 LT |
1864 | } |
1865 | ||
4246a0b6 | 1866 | static void end_sync_write(struct bio *bio) |
1da177e4 | 1867 | { |
4e4cbee9 | 1868 | int uptodate = !bio->bi_status; |
98d30c58 | 1869 | struct r1bio *r1_bio = get_resync_r1bio(bio); |
fd01b88c | 1870 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 1871 | struct r1conf *conf = mddev->private; |
4367af55 N |
1872 | sector_t first_bad; |
1873 | int bad_sectors; | |
854abd75 | 1874 | struct md_rdev *rdev = conf->mirrors[find_bio_disk(r1_bio, bio)].rdev; |
ba3ae3be | 1875 | |
6b1117d5 | 1876 | if (!uptodate) { |
57dab0bd | 1877 | sector_t sync_blocks = 0; |
6b1117d5 N |
1878 | sector_t s = r1_bio->sector; |
1879 | long sectors_to_go = r1_bio->sectors; | |
1880 | /* make sure these bits doesn't get cleared. */ | |
1881 | do { | |
e64e4018 | 1882 | md_bitmap_end_sync(mddev->bitmap, s, &sync_blocks, 1); |
6b1117d5 N |
1883 | s += sync_blocks; |
1884 | sectors_to_go -= sync_blocks; | |
1885 | } while (sectors_to_go > 0); | |
854abd75 N |
1886 | set_bit(WriteErrorSeen, &rdev->flags); |
1887 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) | |
19d67169 N |
1888 | set_bit(MD_RECOVERY_NEEDED, & |
1889 | mddev->recovery); | |
d8f05d29 | 1890 | set_bit(R1BIO_WriteError, &r1_bio->state); |
854abd75 | 1891 | } else if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors, |
3a9f28a5 N |
1892 | &first_bad, &bad_sectors) && |
1893 | !is_badblock(conf->mirrors[r1_bio->read_disk].rdev, | |
1894 | r1_bio->sector, | |
1895 | r1_bio->sectors, | |
1896 | &first_bad, &bad_sectors) | |
1897 | ) | |
4367af55 | 1898 | set_bit(R1BIO_MadeGood, &r1_bio->state); |
e3b9703e | 1899 | |
1da177e4 | 1900 | if (atomic_dec_and_test(&r1_bio->remaining)) { |
4367af55 | 1901 | int s = r1_bio->sectors; |
d8f05d29 N |
1902 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
1903 | test_bit(R1BIO_WriteError, &r1_bio->state)) | |
4367af55 N |
1904 | reschedule_retry(r1_bio); |
1905 | else { | |
1906 | put_buf(r1_bio); | |
1907 | md_done_sync(mddev, s, uptodate); | |
1908 | } | |
1da177e4 | 1909 | } |
1da177e4 LT |
1910 | } |
1911 | ||
3cb03002 | 1912 | static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector, |
d8f05d29 N |
1913 | int sectors, struct page *page, int rw) |
1914 | { | |
796a5cf0 | 1915 | if (sync_page_io(rdev, sector, sectors << 9, page, rw, 0, false)) |
d8f05d29 N |
1916 | /* success */ |
1917 | return 1; | |
19d67169 | 1918 | if (rw == WRITE) { |
d8f05d29 | 1919 | set_bit(WriteErrorSeen, &rdev->flags); |
19d67169 N |
1920 | if (!test_and_set_bit(WantReplacement, |
1921 | &rdev->flags)) | |
1922 | set_bit(MD_RECOVERY_NEEDED, & | |
1923 | rdev->mddev->recovery); | |
1924 | } | |
d8f05d29 N |
1925 | /* need to record an error - either for the block or the device */ |
1926 | if (!rdev_set_badblocks(rdev, sector, sectors, 0)) | |
1927 | md_error(rdev->mddev, rdev); | |
1928 | return 0; | |
1929 | } | |
1930 | ||
9f2c9d12 | 1931 | static int fix_sync_read_error(struct r1bio *r1_bio) |
1da177e4 | 1932 | { |
a68e5870 N |
1933 | /* Try some synchronous reads of other devices to get |
1934 | * good data, much like with normal read errors. Only | |
1935 | * read into the pages we already have so we don't | |
1936 | * need to re-issue the read request. | |
1937 | * We don't need to freeze the array, because being in an | |
1938 | * active sync request, there is no normal IO, and | |
1939 | * no overlapping syncs. | |
06f60385 N |
1940 | * We don't need to check is_badblock() again as we |
1941 | * made sure that anything with a bad block in range | |
1942 | * will have bi_end_io clear. | |
a68e5870 | 1943 | */ |
fd01b88c | 1944 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 1945 | struct r1conf *conf = mddev->private; |
a68e5870 | 1946 | struct bio *bio = r1_bio->bios[r1_bio->read_disk]; |
44cf0f4d | 1947 | struct page **pages = get_resync_pages(bio)->pages; |
a68e5870 N |
1948 | sector_t sect = r1_bio->sector; |
1949 | int sectors = r1_bio->sectors; | |
1950 | int idx = 0; | |
2e52d449 N |
1951 | struct md_rdev *rdev; |
1952 | ||
1953 | rdev = conf->mirrors[r1_bio->read_disk].rdev; | |
1954 | if (test_bit(FailFast, &rdev->flags)) { | |
1955 | /* Don't try recovering from here - just fail it | |
1956 | * ... unless it is the last working device of course */ | |
1957 | md_error(mddev, rdev); | |
1958 | if (test_bit(Faulty, &rdev->flags)) | |
1959 | /* Don't try to read from here, but make sure | |
1960 | * put_buf does it's thing | |
1961 | */ | |
1962 | bio->bi_end_io = end_sync_write; | |
1963 | } | |
a68e5870 N |
1964 | |
1965 | while(sectors) { | |
1966 | int s = sectors; | |
1967 | int d = r1_bio->read_disk; | |
1968 | int success = 0; | |
78d7f5f7 | 1969 | int start; |
a68e5870 N |
1970 | |
1971 | if (s > (PAGE_SIZE>>9)) | |
1972 | s = PAGE_SIZE >> 9; | |
1973 | do { | |
1974 | if (r1_bio->bios[d]->bi_end_io == end_sync_read) { | |
1975 | /* No rcu protection needed here devices | |
1976 | * can only be removed when no resync is | |
1977 | * active, and resync is currently active | |
1978 | */ | |
1979 | rdev = conf->mirrors[d].rdev; | |
9d3d8011 | 1980 | if (sync_page_io(rdev, sect, s<<9, |
44cf0f4d | 1981 | pages[idx], |
796a5cf0 | 1982 | REQ_OP_READ, 0, false)) { |
a68e5870 N |
1983 | success = 1; |
1984 | break; | |
1985 | } | |
1986 | } | |
1987 | d++; | |
8f19ccb2 | 1988 | if (d == conf->raid_disks * 2) |
a68e5870 N |
1989 | d = 0; |
1990 | } while (!success && d != r1_bio->read_disk); | |
1991 | ||
78d7f5f7 | 1992 | if (!success) { |
a68e5870 | 1993 | char b[BDEVNAME_SIZE]; |
3a9f28a5 N |
1994 | int abort = 0; |
1995 | /* Cannot read from anywhere, this block is lost. | |
1996 | * Record a bad block on each device. If that doesn't | |
1997 | * work just disable and interrupt the recovery. | |
1998 | * Don't fail devices as that won't really help. | |
1999 | */ | |
1d41c216 | 2000 | pr_crit_ratelimited("md/raid1:%s: %s: unrecoverable I/O read error for block %llu\n", |
74d46992 | 2001 | mdname(mddev), bio_devname(bio, b), |
1d41c216 | 2002 | (unsigned long long)r1_bio->sector); |
8f19ccb2 | 2003 | for (d = 0; d < conf->raid_disks * 2; d++) { |
3a9f28a5 N |
2004 | rdev = conf->mirrors[d].rdev; |
2005 | if (!rdev || test_bit(Faulty, &rdev->flags)) | |
2006 | continue; | |
2007 | if (!rdev_set_badblocks(rdev, sect, s, 0)) | |
2008 | abort = 1; | |
2009 | } | |
2010 | if (abort) { | |
d890fa2b N |
2011 | conf->recovery_disabled = |
2012 | mddev->recovery_disabled; | |
3a9f28a5 N |
2013 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
2014 | md_done_sync(mddev, r1_bio->sectors, 0); | |
2015 | put_buf(r1_bio); | |
2016 | return 0; | |
2017 | } | |
2018 | /* Try next page */ | |
2019 | sectors -= s; | |
2020 | sect += s; | |
2021 | idx++; | |
2022 | continue; | |
d11c171e | 2023 | } |
78d7f5f7 N |
2024 | |
2025 | start = d; | |
2026 | /* write it back and re-read */ | |
2027 | while (d != r1_bio->read_disk) { | |
2028 | if (d == 0) | |
8f19ccb2 | 2029 | d = conf->raid_disks * 2; |
78d7f5f7 N |
2030 | d--; |
2031 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) | |
2032 | continue; | |
2033 | rdev = conf->mirrors[d].rdev; | |
d8f05d29 | 2034 | if (r1_sync_page_io(rdev, sect, s, |
44cf0f4d | 2035 | pages[idx], |
d8f05d29 | 2036 | WRITE) == 0) { |
78d7f5f7 N |
2037 | r1_bio->bios[d]->bi_end_io = NULL; |
2038 | rdev_dec_pending(rdev, mddev); | |
9d3d8011 | 2039 | } |
78d7f5f7 N |
2040 | } |
2041 | d = start; | |
2042 | while (d != r1_bio->read_disk) { | |
2043 | if (d == 0) | |
8f19ccb2 | 2044 | d = conf->raid_disks * 2; |
78d7f5f7 N |
2045 | d--; |
2046 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) | |
2047 | continue; | |
2048 | rdev = conf->mirrors[d].rdev; | |
d8f05d29 | 2049 | if (r1_sync_page_io(rdev, sect, s, |
44cf0f4d | 2050 | pages[idx], |
d8f05d29 | 2051 | READ) != 0) |
9d3d8011 | 2052 | atomic_add(s, &rdev->corrected_errors); |
78d7f5f7 | 2053 | } |
a68e5870 N |
2054 | sectors -= s; |
2055 | sect += s; | |
2056 | idx ++; | |
2057 | } | |
78d7f5f7 | 2058 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
4e4cbee9 | 2059 | bio->bi_status = 0; |
a68e5870 N |
2060 | return 1; |
2061 | } | |
2062 | ||
c95e6385 | 2063 | static void process_checks(struct r1bio *r1_bio) |
a68e5870 N |
2064 | { |
2065 | /* We have read all readable devices. If we haven't | |
2066 | * got the block, then there is no hope left. | |
2067 | * If we have, then we want to do a comparison | |
2068 | * and skip the write if everything is the same. | |
2069 | * If any blocks failed to read, then we need to | |
2070 | * attempt an over-write | |
2071 | */ | |
fd01b88c | 2072 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 2073 | struct r1conf *conf = mddev->private; |
a68e5870 N |
2074 | int primary; |
2075 | int i; | |
f4380a91 | 2076 | int vcnt; |
a68e5870 | 2077 | |
30bc9b53 N |
2078 | /* Fix variable parts of all bios */ |
2079 | vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9); | |
2080 | for (i = 0; i < conf->raid_disks * 2; i++) { | |
4e4cbee9 | 2081 | blk_status_t status; |
30bc9b53 | 2082 | struct bio *b = r1_bio->bios[i]; |
98d30c58 | 2083 | struct resync_pages *rp = get_resync_pages(b); |
30bc9b53 N |
2084 | if (b->bi_end_io != end_sync_read) |
2085 | continue; | |
4246a0b6 | 2086 | /* fixup the bio for reuse, but preserve errno */ |
4e4cbee9 | 2087 | status = b->bi_status; |
30bc9b53 | 2088 | bio_reset(b); |
4e4cbee9 | 2089 | b->bi_status = status; |
4f024f37 | 2090 | b->bi_iter.bi_sector = r1_bio->sector + |
30bc9b53 | 2091 | conf->mirrors[i].rdev->data_offset; |
74d46992 | 2092 | bio_set_dev(b, conf->mirrors[i].rdev->bdev); |
30bc9b53 | 2093 | b->bi_end_io = end_sync_read; |
98d30c58 ML |
2094 | rp->raid_bio = r1_bio; |
2095 | b->bi_private = rp; | |
30bc9b53 | 2096 | |
fb0eb5df ML |
2097 | /* initialize bvec table again */ |
2098 | md_bio_reset_resync_pages(b, rp, r1_bio->sectors << 9); | |
30bc9b53 | 2099 | } |
8f19ccb2 | 2100 | for (primary = 0; primary < conf->raid_disks * 2; primary++) |
a68e5870 | 2101 | if (r1_bio->bios[primary]->bi_end_io == end_sync_read && |
4e4cbee9 | 2102 | !r1_bio->bios[primary]->bi_status) { |
a68e5870 N |
2103 | r1_bio->bios[primary]->bi_end_io = NULL; |
2104 | rdev_dec_pending(conf->mirrors[primary].rdev, mddev); | |
2105 | break; | |
2106 | } | |
2107 | r1_bio->read_disk = primary; | |
8f19ccb2 | 2108 | for (i = 0; i < conf->raid_disks * 2; i++) { |
78d7f5f7 | 2109 | int j; |
78d7f5f7 N |
2110 | struct bio *pbio = r1_bio->bios[primary]; |
2111 | struct bio *sbio = r1_bio->bios[i]; | |
4e4cbee9 | 2112 | blk_status_t status = sbio->bi_status; |
44cf0f4d ML |
2113 | struct page **ppages = get_resync_pages(pbio)->pages; |
2114 | struct page **spages = get_resync_pages(sbio)->pages; | |
60928a91 | 2115 | struct bio_vec *bi; |
8fc04e6e | 2116 | int page_len[RESYNC_PAGES] = { 0 }; |
a68e5870 | 2117 | |
2aabaa65 | 2118 | if (sbio->bi_end_io != end_sync_read) |
78d7f5f7 | 2119 | continue; |
4246a0b6 | 2120 | /* Now we can 'fixup' the error value */ |
4e4cbee9 | 2121 | sbio->bi_status = 0; |
78d7f5f7 | 2122 | |
60928a91 ML |
2123 | bio_for_each_segment_all(bi, sbio, j) |
2124 | page_len[j] = bi->bv_len; | |
2125 | ||
4e4cbee9 | 2126 | if (!status) { |
78d7f5f7 | 2127 | for (j = vcnt; j-- ; ) { |
44cf0f4d ML |
2128 | if (memcmp(page_address(ppages[j]), |
2129 | page_address(spages[j]), | |
60928a91 | 2130 | page_len[j])) |
78d7f5f7 | 2131 | break; |
69382e85 | 2132 | } |
78d7f5f7 N |
2133 | } else |
2134 | j = 0; | |
2135 | if (j >= 0) | |
7f7583d4 | 2136 | atomic64_add(r1_bio->sectors, &mddev->resync_mismatches); |
78d7f5f7 | 2137 | if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery) |
4e4cbee9 | 2138 | && !status)) { |
78d7f5f7 N |
2139 | /* No need to write to this device. */ |
2140 | sbio->bi_end_io = NULL; | |
2141 | rdev_dec_pending(conf->mirrors[i].rdev, mddev); | |
2142 | continue; | |
2143 | } | |
d3b45c2a KO |
2144 | |
2145 | bio_copy_data(sbio, pbio); | |
78d7f5f7 | 2146 | } |
a68e5870 N |
2147 | } |
2148 | ||
9f2c9d12 | 2149 | static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio) |
a68e5870 | 2150 | { |
e8096360 | 2151 | struct r1conf *conf = mddev->private; |
a68e5870 | 2152 | int i; |
8f19ccb2 | 2153 | int disks = conf->raid_disks * 2; |
037d2ff6 | 2154 | struct bio *wbio; |
a68e5870 | 2155 | |
a68e5870 N |
2156 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) |
2157 | /* ouch - failed to read all of that. */ | |
2158 | if (!fix_sync_read_error(r1_bio)) | |
2159 | return; | |
7ca78d57 N |
2160 | |
2161 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
c95e6385 N |
2162 | process_checks(r1_bio); |
2163 | ||
d11c171e N |
2164 | /* |
2165 | * schedule writes | |
2166 | */ | |
1da177e4 LT |
2167 | atomic_set(&r1_bio->remaining, 1); |
2168 | for (i = 0; i < disks ; i++) { | |
2169 | wbio = r1_bio->bios[i]; | |
3e198f78 N |
2170 | if (wbio->bi_end_io == NULL || |
2171 | (wbio->bi_end_io == end_sync_read && | |
2172 | (i == r1_bio->read_disk || | |
2173 | !test_bit(MD_RECOVERY_SYNC, &mddev->recovery)))) | |
1da177e4 | 2174 | continue; |
0c9d5b12 N |
2175 | if (test_bit(Faulty, &conf->mirrors[i].rdev->flags)) |
2176 | continue; | |
1da177e4 | 2177 | |
796a5cf0 | 2178 | bio_set_op_attrs(wbio, REQ_OP_WRITE, 0); |
212e7eb7 N |
2179 | if (test_bit(FailFast, &conf->mirrors[i].rdev->flags)) |
2180 | wbio->bi_opf |= MD_FAILFAST; | |
2181 | ||
3e198f78 | 2182 | wbio->bi_end_io = end_sync_write; |
1da177e4 | 2183 | atomic_inc(&r1_bio->remaining); |
aa8b57aa | 2184 | md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio)); |
191ea9b2 | 2185 | |
1da177e4 LT |
2186 | generic_make_request(wbio); |
2187 | } | |
2188 | ||
2189 | if (atomic_dec_and_test(&r1_bio->remaining)) { | |
191ea9b2 | 2190 | /* if we're here, all write(s) have completed, so clean up */ |
58e94ae1 N |
2191 | int s = r1_bio->sectors; |
2192 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || | |
2193 | test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2194 | reschedule_retry(r1_bio); | |
2195 | else { | |
2196 | put_buf(r1_bio); | |
2197 | md_done_sync(mddev, s, 1); | |
2198 | } | |
1da177e4 LT |
2199 | } |
2200 | } | |
2201 | ||
2202 | /* | |
2203 | * This is a kernel thread which: | |
2204 | * | |
2205 | * 1. Retries failed read operations on working mirrors. | |
2206 | * 2. Updates the raid superblock when problems encounter. | |
d2eb35ac | 2207 | * 3. Performs writes following reads for array synchronising. |
1da177e4 LT |
2208 | */ |
2209 | ||
e8096360 | 2210 | static void fix_read_error(struct r1conf *conf, int read_disk, |
867868fb N |
2211 | sector_t sect, int sectors) |
2212 | { | |
fd01b88c | 2213 | struct mddev *mddev = conf->mddev; |
867868fb N |
2214 | while(sectors) { |
2215 | int s = sectors; | |
2216 | int d = read_disk; | |
2217 | int success = 0; | |
2218 | int start; | |
3cb03002 | 2219 | struct md_rdev *rdev; |
867868fb N |
2220 | |
2221 | if (s > (PAGE_SIZE>>9)) | |
2222 | s = PAGE_SIZE >> 9; | |
2223 | ||
2224 | do { | |
d2eb35ac N |
2225 | sector_t first_bad; |
2226 | int bad_sectors; | |
2227 | ||
707a6a42 N |
2228 | rcu_read_lock(); |
2229 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
867868fb | 2230 | if (rdev && |
da8840a7 | 2231 | (test_bit(In_sync, &rdev->flags) || |
2232 | (!test_bit(Faulty, &rdev->flags) && | |
2233 | rdev->recovery_offset >= sect + s)) && | |
d2eb35ac | 2234 | is_badblock(rdev, sect, s, |
707a6a42 N |
2235 | &first_bad, &bad_sectors) == 0) { |
2236 | atomic_inc(&rdev->nr_pending); | |
2237 | rcu_read_unlock(); | |
2238 | if (sync_page_io(rdev, sect, s<<9, | |
796a5cf0 | 2239 | conf->tmppage, REQ_OP_READ, 0, false)) |
707a6a42 N |
2240 | success = 1; |
2241 | rdev_dec_pending(rdev, mddev); | |
2242 | if (success) | |
2243 | break; | |
2244 | } else | |
2245 | rcu_read_unlock(); | |
2246 | d++; | |
2247 | if (d == conf->raid_disks * 2) | |
2248 | d = 0; | |
867868fb N |
2249 | } while (!success && d != read_disk); |
2250 | ||
2251 | if (!success) { | |
d8f05d29 | 2252 | /* Cannot read from anywhere - mark it bad */ |
3cb03002 | 2253 | struct md_rdev *rdev = conf->mirrors[read_disk].rdev; |
d8f05d29 N |
2254 | if (!rdev_set_badblocks(rdev, sect, s, 0)) |
2255 | md_error(mddev, rdev); | |
867868fb N |
2256 | break; |
2257 | } | |
2258 | /* write it back and re-read */ | |
2259 | start = d; | |
2260 | while (d != read_disk) { | |
2261 | if (d==0) | |
8f19ccb2 | 2262 | d = conf->raid_disks * 2; |
867868fb | 2263 | d--; |
707a6a42 N |
2264 | rcu_read_lock(); |
2265 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
867868fb | 2266 | if (rdev && |
707a6a42 N |
2267 | !test_bit(Faulty, &rdev->flags)) { |
2268 | atomic_inc(&rdev->nr_pending); | |
2269 | rcu_read_unlock(); | |
d8f05d29 N |
2270 | r1_sync_page_io(rdev, sect, s, |
2271 | conf->tmppage, WRITE); | |
707a6a42 N |
2272 | rdev_dec_pending(rdev, mddev); |
2273 | } else | |
2274 | rcu_read_unlock(); | |
867868fb N |
2275 | } |
2276 | d = start; | |
2277 | while (d != read_disk) { | |
2278 | char b[BDEVNAME_SIZE]; | |
2279 | if (d==0) | |
8f19ccb2 | 2280 | d = conf->raid_disks * 2; |
867868fb | 2281 | d--; |
707a6a42 N |
2282 | rcu_read_lock(); |
2283 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
867868fb | 2284 | if (rdev && |
b8cb6b4c | 2285 | !test_bit(Faulty, &rdev->flags)) { |
707a6a42 N |
2286 | atomic_inc(&rdev->nr_pending); |
2287 | rcu_read_unlock(); | |
d8f05d29 N |
2288 | if (r1_sync_page_io(rdev, sect, s, |
2289 | conf->tmppage, READ)) { | |
867868fb | 2290 | atomic_add(s, &rdev->corrected_errors); |
1d41c216 N |
2291 | pr_info("md/raid1:%s: read error corrected (%d sectors at %llu on %s)\n", |
2292 | mdname(mddev), s, | |
2293 | (unsigned long long)(sect + | |
2294 | rdev->data_offset), | |
2295 | bdevname(rdev->bdev, b)); | |
867868fb | 2296 | } |
707a6a42 N |
2297 | rdev_dec_pending(rdev, mddev); |
2298 | } else | |
2299 | rcu_read_unlock(); | |
867868fb N |
2300 | } |
2301 | sectors -= s; | |
2302 | sect += s; | |
2303 | } | |
2304 | } | |
2305 | ||
9f2c9d12 | 2306 | static int narrow_write_error(struct r1bio *r1_bio, int i) |
cd5ff9a1 | 2307 | { |
fd01b88c | 2308 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 2309 | struct r1conf *conf = mddev->private; |
3cb03002 | 2310 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
cd5ff9a1 N |
2311 | |
2312 | /* bio has the data to be written to device 'i' where | |
2313 | * we just recently had a write error. | |
2314 | * We repeatedly clone the bio and trim down to one block, | |
2315 | * then try the write. Where the write fails we record | |
2316 | * a bad block. | |
2317 | * It is conceivable that the bio doesn't exactly align with | |
2318 | * blocks. We must handle this somehow. | |
2319 | * | |
2320 | * We currently own a reference on the rdev. | |
2321 | */ | |
2322 | ||
2323 | int block_sectors; | |
2324 | sector_t sector; | |
2325 | int sectors; | |
2326 | int sect_to_write = r1_bio->sectors; | |
2327 | int ok = 1; | |
2328 | ||
2329 | if (rdev->badblocks.shift < 0) | |
2330 | return 0; | |
2331 | ||
ab713cdc ND |
2332 | block_sectors = roundup(1 << rdev->badblocks.shift, |
2333 | bdev_logical_block_size(rdev->bdev) >> 9); | |
cd5ff9a1 N |
2334 | sector = r1_bio->sector; |
2335 | sectors = ((sector + block_sectors) | |
2336 | & ~(sector_t)(block_sectors - 1)) | |
2337 | - sector; | |
2338 | ||
cd5ff9a1 N |
2339 | while (sect_to_write) { |
2340 | struct bio *wbio; | |
2341 | if (sectors > sect_to_write) | |
2342 | sectors = sect_to_write; | |
2343 | /* Write at 'sector' for 'sectors'*/ | |
2344 | ||
b783863f | 2345 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { |
841c1316 ML |
2346 | wbio = bio_clone_fast(r1_bio->behind_master_bio, |
2347 | GFP_NOIO, | |
afeee514 | 2348 | &mddev->bio_set); |
b783863f | 2349 | } else { |
d7a10308 | 2350 | wbio = bio_clone_fast(r1_bio->master_bio, GFP_NOIO, |
afeee514 | 2351 | &mddev->bio_set); |
b783863f KO |
2352 | } |
2353 | ||
796a5cf0 | 2354 | bio_set_op_attrs(wbio, REQ_OP_WRITE, 0); |
4f024f37 KO |
2355 | wbio->bi_iter.bi_sector = r1_bio->sector; |
2356 | wbio->bi_iter.bi_size = r1_bio->sectors << 9; | |
cd5ff9a1 | 2357 | |
6678d83f | 2358 | bio_trim(wbio, sector - r1_bio->sector, sectors); |
4f024f37 | 2359 | wbio->bi_iter.bi_sector += rdev->data_offset; |
74d46992 | 2360 | bio_set_dev(wbio, rdev->bdev); |
4e49ea4a MC |
2361 | |
2362 | if (submit_bio_wait(wbio) < 0) | |
cd5ff9a1 N |
2363 | /* failure! */ |
2364 | ok = rdev_set_badblocks(rdev, sector, | |
2365 | sectors, 0) | |
2366 | && ok; | |
2367 | ||
2368 | bio_put(wbio); | |
2369 | sect_to_write -= sectors; | |
2370 | sector += sectors; | |
2371 | sectors = block_sectors; | |
2372 | } | |
2373 | return ok; | |
2374 | } | |
2375 | ||
e8096360 | 2376 | static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce N |
2377 | { |
2378 | int m; | |
2379 | int s = r1_bio->sectors; | |
8f19ccb2 | 2380 | for (m = 0; m < conf->raid_disks * 2 ; m++) { |
3cb03002 | 2381 | struct md_rdev *rdev = conf->mirrors[m].rdev; |
62096bce N |
2382 | struct bio *bio = r1_bio->bios[m]; |
2383 | if (bio->bi_end_io == NULL) | |
2384 | continue; | |
4e4cbee9 | 2385 | if (!bio->bi_status && |
62096bce | 2386 | test_bit(R1BIO_MadeGood, &r1_bio->state)) { |
c6563a8c | 2387 | rdev_clear_badblocks(rdev, r1_bio->sector, s, 0); |
62096bce | 2388 | } |
4e4cbee9 | 2389 | if (bio->bi_status && |
62096bce N |
2390 | test_bit(R1BIO_WriteError, &r1_bio->state)) { |
2391 | if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0)) | |
2392 | md_error(conf->mddev, rdev); | |
2393 | } | |
2394 | } | |
2395 | put_buf(r1_bio); | |
2396 | md_done_sync(conf->mddev, s, 1); | |
2397 | } | |
2398 | ||
e8096360 | 2399 | static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce | 2400 | { |
fd76863e | 2401 | int m, idx; |
55ce74d4 | 2402 | bool fail = false; |
fd76863e | 2403 | |
8f19ccb2 | 2404 | for (m = 0; m < conf->raid_disks * 2 ; m++) |
62096bce | 2405 | if (r1_bio->bios[m] == IO_MADE_GOOD) { |
3cb03002 | 2406 | struct md_rdev *rdev = conf->mirrors[m].rdev; |
62096bce N |
2407 | rdev_clear_badblocks(rdev, |
2408 | r1_bio->sector, | |
c6563a8c | 2409 | r1_bio->sectors, 0); |
62096bce N |
2410 | rdev_dec_pending(rdev, conf->mddev); |
2411 | } else if (r1_bio->bios[m] != NULL) { | |
2412 | /* This drive got a write error. We need to | |
2413 | * narrow down and record precise write | |
2414 | * errors. | |
2415 | */ | |
55ce74d4 | 2416 | fail = true; |
62096bce N |
2417 | if (!narrow_write_error(r1_bio, m)) { |
2418 | md_error(conf->mddev, | |
2419 | conf->mirrors[m].rdev); | |
2420 | /* an I/O failed, we can't clear the bitmap */ | |
2421 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
2422 | } | |
2423 | rdev_dec_pending(conf->mirrors[m].rdev, | |
2424 | conf->mddev); | |
2425 | } | |
55ce74d4 N |
2426 | if (fail) { |
2427 | spin_lock_irq(&conf->device_lock); | |
2428 | list_add(&r1_bio->retry_list, &conf->bio_end_io_list); | |
fd76863e | 2429 | idx = sector_to_idx(r1_bio->sector); |
824e47da | 2430 | atomic_inc(&conf->nr_queued[idx]); |
55ce74d4 | 2431 | spin_unlock_irq(&conf->device_lock); |
824e47da | 2432 | /* |
2433 | * In case freeze_array() is waiting for condition | |
2434 | * get_unqueued_pending() == extra to be true. | |
2435 | */ | |
2436 | wake_up(&conf->wait_barrier); | |
55ce74d4 | 2437 | md_wakeup_thread(conf->mddev->thread); |
bd8688a1 N |
2438 | } else { |
2439 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2440 | close_write(r1_bio); | |
55ce74d4 | 2441 | raid_end_bio_io(r1_bio); |
bd8688a1 | 2442 | } |
62096bce N |
2443 | } |
2444 | ||
e8096360 | 2445 | static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce | 2446 | { |
fd01b88c | 2447 | struct mddev *mddev = conf->mddev; |
62096bce | 2448 | struct bio *bio; |
3cb03002 | 2449 | struct md_rdev *rdev; |
62096bce N |
2450 | |
2451 | clear_bit(R1BIO_ReadError, &r1_bio->state); | |
2452 | /* we got a read error. Maybe the drive is bad. Maybe just | |
2453 | * the block and we can fix it. | |
2454 | * We freeze all other IO, and try reading the block from | |
2455 | * other devices. When we find one, we re-write | |
2456 | * and check it that fixes the read error. | |
2457 | * This is all done synchronously while the array is | |
2458 | * frozen | |
2459 | */ | |
7449f699 TM |
2460 | |
2461 | bio = r1_bio->bios[r1_bio->read_disk]; | |
7449f699 TM |
2462 | bio_put(bio); |
2463 | r1_bio->bios[r1_bio->read_disk] = NULL; | |
2464 | ||
2e52d449 N |
2465 | rdev = conf->mirrors[r1_bio->read_disk].rdev; |
2466 | if (mddev->ro == 0 | |
2467 | && !test_bit(FailFast, &rdev->flags)) { | |
e2d59925 | 2468 | freeze_array(conf, 1); |
62096bce N |
2469 | fix_read_error(conf, r1_bio->read_disk, |
2470 | r1_bio->sector, r1_bio->sectors); | |
2471 | unfreeze_array(conf); | |
b33d1062 GK |
2472 | } else if (mddev->ro == 0 && test_bit(FailFast, &rdev->flags)) { |
2473 | md_error(mddev, rdev); | |
7449f699 TM |
2474 | } else { |
2475 | r1_bio->bios[r1_bio->read_disk] = IO_BLOCKED; | |
2476 | } | |
2477 | ||
2e52d449 | 2478 | rdev_dec_pending(rdev, conf->mddev); |
689389a0 N |
2479 | allow_barrier(conf, r1_bio->sector); |
2480 | bio = r1_bio->master_bio; | |
62096bce | 2481 | |
689389a0 N |
2482 | /* Reuse the old r1_bio so that the IO_BLOCKED settings are preserved */ |
2483 | r1_bio->state = 0; | |
2484 | raid1_read_request(mddev, bio, r1_bio->sectors, r1_bio); | |
62096bce N |
2485 | } |
2486 | ||
4ed8731d | 2487 | static void raid1d(struct md_thread *thread) |
1da177e4 | 2488 | { |
4ed8731d | 2489 | struct mddev *mddev = thread->mddev; |
9f2c9d12 | 2490 | struct r1bio *r1_bio; |
1da177e4 | 2491 | unsigned long flags; |
e8096360 | 2492 | struct r1conf *conf = mddev->private; |
1da177e4 | 2493 | struct list_head *head = &conf->retry_list; |
e1dfa0a2 | 2494 | struct blk_plug plug; |
fd76863e | 2495 | int idx; |
1da177e4 LT |
2496 | |
2497 | md_check_recovery(mddev); | |
e1dfa0a2 | 2498 | |
55ce74d4 | 2499 | if (!list_empty_careful(&conf->bio_end_io_list) && |
2953079c | 2500 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { |
55ce74d4 N |
2501 | LIST_HEAD(tmp); |
2502 | spin_lock_irqsave(&conf->device_lock, flags); | |
fd76863e | 2503 | if (!test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) |
2504 | list_splice_init(&conf->bio_end_io_list, &tmp); | |
55ce74d4 N |
2505 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2506 | while (!list_empty(&tmp)) { | |
a452744b MP |
2507 | r1_bio = list_first_entry(&tmp, struct r1bio, |
2508 | retry_list); | |
55ce74d4 | 2509 | list_del(&r1_bio->retry_list); |
fd76863e | 2510 | idx = sector_to_idx(r1_bio->sector); |
824e47da | 2511 | atomic_dec(&conf->nr_queued[idx]); |
bd8688a1 N |
2512 | if (mddev->degraded) |
2513 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
2514 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2515 | close_write(r1_bio); | |
55ce74d4 N |
2516 | raid_end_bio_io(r1_bio); |
2517 | } | |
2518 | } | |
2519 | ||
e1dfa0a2 | 2520 | blk_start_plug(&plug); |
1da177e4 | 2521 | for (;;) { |
191ea9b2 | 2522 | |
0021b7bc | 2523 | flush_pending_writes(conf); |
191ea9b2 | 2524 | |
a35e63ef N |
2525 | spin_lock_irqsave(&conf->device_lock, flags); |
2526 | if (list_empty(head)) { | |
2527 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 | 2528 | break; |
a35e63ef | 2529 | } |
9f2c9d12 | 2530 | r1_bio = list_entry(head->prev, struct r1bio, retry_list); |
1da177e4 | 2531 | list_del(head->prev); |
fd76863e | 2532 | idx = sector_to_idx(r1_bio->sector); |
824e47da | 2533 | atomic_dec(&conf->nr_queued[idx]); |
1da177e4 LT |
2534 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2535 | ||
2536 | mddev = r1_bio->mddev; | |
070ec55d | 2537 | conf = mddev->private; |
4367af55 | 2538 | if (test_bit(R1BIO_IsSync, &r1_bio->state)) { |
d8f05d29 | 2539 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
62096bce N |
2540 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
2541 | handle_sync_write_finished(conf, r1_bio); | |
2542 | else | |
4367af55 | 2543 | sync_request_write(mddev, r1_bio); |
cd5ff9a1 | 2544 | } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
62096bce N |
2545 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
2546 | handle_write_finished(conf, r1_bio); | |
2547 | else if (test_bit(R1BIO_ReadError, &r1_bio->state)) | |
2548 | handle_read_error(conf, r1_bio); | |
2549 | else | |
c230e7e5 | 2550 | WARN_ON_ONCE(1); |
62096bce | 2551 | |
1d9d5241 | 2552 | cond_resched(); |
2953079c | 2553 | if (mddev->sb_flags & ~(1<<MD_SB_CHANGE_PENDING)) |
de393cde | 2554 | md_check_recovery(mddev); |
1da177e4 | 2555 | } |
e1dfa0a2 | 2556 | blk_finish_plug(&plug); |
1da177e4 LT |
2557 | } |
2558 | ||
e8096360 | 2559 | static int init_resync(struct r1conf *conf) |
1da177e4 LT |
2560 | { |
2561 | int buffs; | |
2562 | ||
2563 | buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; | |
afeee514 KO |
2564 | BUG_ON(mempool_initialized(&conf->r1buf_pool)); |
2565 | ||
2566 | return mempool_init(&conf->r1buf_pool, buffs, r1buf_pool_alloc, | |
2567 | r1buf_pool_free, conf->poolinfo); | |
1da177e4 LT |
2568 | } |
2569 | ||
208410b5 SL |
2570 | static struct r1bio *raid1_alloc_init_r1buf(struct r1conf *conf) |
2571 | { | |
afeee514 | 2572 | struct r1bio *r1bio = mempool_alloc(&conf->r1buf_pool, GFP_NOIO); |
208410b5 SL |
2573 | struct resync_pages *rps; |
2574 | struct bio *bio; | |
2575 | int i; | |
2576 | ||
2577 | for (i = conf->poolinfo->raid_disks; i--; ) { | |
2578 | bio = r1bio->bios[i]; | |
2579 | rps = bio->bi_private; | |
2580 | bio_reset(bio); | |
2581 | bio->bi_private = rps; | |
2582 | } | |
2583 | r1bio->master_bio = NULL; | |
2584 | return r1bio; | |
2585 | } | |
2586 | ||
1da177e4 LT |
2587 | /* |
2588 | * perform a "sync" on one "block" | |
2589 | * | |
2590 | * We need to make sure that no normal I/O request - particularly write | |
2591 | * requests - conflict with active sync requests. | |
2592 | * | |
2593 | * This is achieved by tracking pending requests and a 'barrier' concept | |
2594 | * that can be installed to exclude normal IO requests. | |
2595 | */ | |
2596 | ||
849674e4 SL |
2597 | static sector_t raid1_sync_request(struct mddev *mddev, sector_t sector_nr, |
2598 | int *skipped) | |
1da177e4 | 2599 | { |
e8096360 | 2600 | struct r1conf *conf = mddev->private; |
9f2c9d12 | 2601 | struct r1bio *r1_bio; |
1da177e4 LT |
2602 | struct bio *bio; |
2603 | sector_t max_sector, nr_sectors; | |
3e198f78 | 2604 | int disk = -1; |
1da177e4 | 2605 | int i; |
3e198f78 N |
2606 | int wonly = -1; |
2607 | int write_targets = 0, read_targets = 0; | |
57dab0bd | 2608 | sector_t sync_blocks; |
e3b9703e | 2609 | int still_degraded = 0; |
06f60385 N |
2610 | int good_sectors = RESYNC_SECTORS; |
2611 | int min_bad = 0; /* number of sectors that are bad in all devices */ | |
fd76863e | 2612 | int idx = sector_to_idx(sector_nr); |
022e510f | 2613 | int page_idx = 0; |
1da177e4 | 2614 | |
afeee514 | 2615 | if (!mempool_initialized(&conf->r1buf_pool)) |
1da177e4 | 2616 | if (init_resync(conf)) |
57afd89f | 2617 | return 0; |
1da177e4 | 2618 | |
58c0fed4 | 2619 | max_sector = mddev->dev_sectors; |
1da177e4 | 2620 | if (sector_nr >= max_sector) { |
191ea9b2 N |
2621 | /* If we aborted, we need to abort the |
2622 | * sync on the 'current' bitmap chunk (there will | |
2623 | * only be one in raid1 resync. | |
2624 | * We can find the current addess in mddev->curr_resync | |
2625 | */ | |
6a806c51 | 2626 | if (mddev->curr_resync < max_sector) /* aborted */ |
e64e4018 AS |
2627 | md_bitmap_end_sync(mddev->bitmap, mddev->curr_resync, |
2628 | &sync_blocks, 1); | |
6a806c51 | 2629 | else /* completed sync */ |
191ea9b2 | 2630 | conf->fullsync = 0; |
6a806c51 | 2631 | |
e64e4018 | 2632 | md_bitmap_close_sync(mddev->bitmap); |
1da177e4 | 2633 | close_sync(conf); |
c40f341f GR |
2634 | |
2635 | if (mddev_is_clustered(mddev)) { | |
2636 | conf->cluster_sync_low = 0; | |
2637 | conf->cluster_sync_high = 0; | |
c40f341f | 2638 | } |
1da177e4 LT |
2639 | return 0; |
2640 | } | |
2641 | ||
07d84d10 N |
2642 | if (mddev->bitmap == NULL && |
2643 | mddev->recovery_cp == MaxSector && | |
6394cca5 | 2644 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
07d84d10 N |
2645 | conf->fullsync == 0) { |
2646 | *skipped = 1; | |
2647 | return max_sector - sector_nr; | |
2648 | } | |
6394cca5 N |
2649 | /* before building a request, check if we can skip these blocks.. |
2650 | * This call the bitmap_start_sync doesn't actually record anything | |
2651 | */ | |
e64e4018 | 2652 | if (!md_bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
e5de485f | 2653 | !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { |
191ea9b2 N |
2654 | /* We can skip this block, and probably several more */ |
2655 | *skipped = 1; | |
2656 | return sync_blocks; | |
2657 | } | |
17999be4 | 2658 | |
7ac50447 TM |
2659 | /* |
2660 | * If there is non-resync activity waiting for a turn, then let it | |
2661 | * though before starting on this new sync request. | |
2662 | */ | |
824e47da | 2663 | if (atomic_read(&conf->nr_waiting[idx])) |
7ac50447 TM |
2664 | schedule_timeout_uninterruptible(1); |
2665 | ||
c40f341f GR |
2666 | /* we are incrementing sector_nr below. To be safe, we check against |
2667 | * sector_nr + two times RESYNC_SECTORS | |
2668 | */ | |
2669 | ||
e64e4018 | 2670 | md_bitmap_cond_end_sync(mddev->bitmap, sector_nr, |
c40f341f | 2671 | mddev_is_clustered(mddev) && (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high)); |
17999be4 | 2672 | |
8c242593 YY |
2673 | |
2674 | if (raise_barrier(conf, sector_nr)) | |
2675 | return 0; | |
2676 | ||
2677 | r1_bio = raid1_alloc_init_r1buf(conf); | |
1da177e4 | 2678 | |
3e198f78 | 2679 | rcu_read_lock(); |
1da177e4 | 2680 | /* |
3e198f78 N |
2681 | * If we get a correctably read error during resync or recovery, |
2682 | * we might want to read from a different device. So we | |
2683 | * flag all drives that could conceivably be read from for READ, | |
2684 | * and any others (which will be non-In_sync devices) for WRITE. | |
2685 | * If a read fails, we try reading from something else for which READ | |
2686 | * is OK. | |
1da177e4 | 2687 | */ |
1da177e4 | 2688 | |
1da177e4 LT |
2689 | r1_bio->mddev = mddev; |
2690 | r1_bio->sector = sector_nr; | |
191ea9b2 | 2691 | r1_bio->state = 0; |
1da177e4 | 2692 | set_bit(R1BIO_IsSync, &r1_bio->state); |
fd76863e | 2693 | /* make sure good_sectors won't go across barrier unit boundary */ |
2694 | good_sectors = align_to_barrier_unit_end(sector_nr, good_sectors); | |
1da177e4 | 2695 | |
8f19ccb2 | 2696 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3cb03002 | 2697 | struct md_rdev *rdev; |
1da177e4 | 2698 | bio = r1_bio->bios[i]; |
1da177e4 | 2699 | |
3e198f78 N |
2700 | rdev = rcu_dereference(conf->mirrors[i].rdev); |
2701 | if (rdev == NULL || | |
06f60385 | 2702 | test_bit(Faulty, &rdev->flags)) { |
8f19ccb2 N |
2703 | if (i < conf->raid_disks) |
2704 | still_degraded = 1; | |
3e198f78 | 2705 | } else if (!test_bit(In_sync, &rdev->flags)) { |
796a5cf0 | 2706 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
1da177e4 LT |
2707 | bio->bi_end_io = end_sync_write; |
2708 | write_targets ++; | |
3e198f78 N |
2709 | } else { |
2710 | /* may need to read from here */ | |
06f60385 N |
2711 | sector_t first_bad = MaxSector; |
2712 | int bad_sectors; | |
2713 | ||
2714 | if (is_badblock(rdev, sector_nr, good_sectors, | |
2715 | &first_bad, &bad_sectors)) { | |
2716 | if (first_bad > sector_nr) | |
2717 | good_sectors = first_bad - sector_nr; | |
2718 | else { | |
2719 | bad_sectors -= (sector_nr - first_bad); | |
2720 | if (min_bad == 0 || | |
2721 | min_bad > bad_sectors) | |
2722 | min_bad = bad_sectors; | |
2723 | } | |
2724 | } | |
2725 | if (sector_nr < first_bad) { | |
2726 | if (test_bit(WriteMostly, &rdev->flags)) { | |
2727 | if (wonly < 0) | |
2728 | wonly = i; | |
2729 | } else { | |
2730 | if (disk < 0) | |
2731 | disk = i; | |
2732 | } | |
796a5cf0 | 2733 | bio_set_op_attrs(bio, REQ_OP_READ, 0); |
06f60385 N |
2734 | bio->bi_end_io = end_sync_read; |
2735 | read_targets++; | |
d57368af AL |
2736 | } else if (!test_bit(WriteErrorSeen, &rdev->flags) && |
2737 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && | |
2738 | !test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { | |
2739 | /* | |
2740 | * The device is suitable for reading (InSync), | |
2741 | * but has bad block(s) here. Let's try to correct them, | |
2742 | * if we are doing resync or repair. Otherwise, leave | |
2743 | * this device alone for this sync request. | |
2744 | */ | |
796a5cf0 | 2745 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
d57368af AL |
2746 | bio->bi_end_io = end_sync_write; |
2747 | write_targets++; | |
3e198f78 | 2748 | } |
3e198f78 | 2749 | } |
06f60385 N |
2750 | if (bio->bi_end_io) { |
2751 | atomic_inc(&rdev->nr_pending); | |
4f024f37 | 2752 | bio->bi_iter.bi_sector = sector_nr + rdev->data_offset; |
74d46992 | 2753 | bio_set_dev(bio, rdev->bdev); |
2e52d449 N |
2754 | if (test_bit(FailFast, &rdev->flags)) |
2755 | bio->bi_opf |= MD_FAILFAST; | |
06f60385 | 2756 | } |
1da177e4 | 2757 | } |
3e198f78 N |
2758 | rcu_read_unlock(); |
2759 | if (disk < 0) | |
2760 | disk = wonly; | |
2761 | r1_bio->read_disk = disk; | |
191ea9b2 | 2762 | |
06f60385 N |
2763 | if (read_targets == 0 && min_bad > 0) { |
2764 | /* These sectors are bad on all InSync devices, so we | |
2765 | * need to mark them bad on all write targets | |
2766 | */ | |
2767 | int ok = 1; | |
8f19ccb2 | 2768 | for (i = 0 ; i < conf->raid_disks * 2 ; i++) |
06f60385 | 2769 | if (r1_bio->bios[i]->bi_end_io == end_sync_write) { |
a42f9d83 | 2770 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
06f60385 N |
2771 | ok = rdev_set_badblocks(rdev, sector_nr, |
2772 | min_bad, 0 | |
2773 | ) && ok; | |
2774 | } | |
2953079c | 2775 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
06f60385 N |
2776 | *skipped = 1; |
2777 | put_buf(r1_bio); | |
2778 | ||
2779 | if (!ok) { | |
2780 | /* Cannot record the badblocks, so need to | |
2781 | * abort the resync. | |
2782 | * If there are multiple read targets, could just | |
2783 | * fail the really bad ones ??? | |
2784 | */ | |
2785 | conf->recovery_disabled = mddev->recovery_disabled; | |
2786 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2787 | return 0; | |
2788 | } else | |
2789 | return min_bad; | |
2790 | ||
2791 | } | |
2792 | if (min_bad > 0 && min_bad < good_sectors) { | |
2793 | /* only resync enough to reach the next bad->good | |
2794 | * transition */ | |
2795 | good_sectors = min_bad; | |
2796 | } | |
2797 | ||
3e198f78 N |
2798 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0) |
2799 | /* extra read targets are also write targets */ | |
2800 | write_targets += read_targets-1; | |
2801 | ||
2802 | if (write_targets == 0 || read_targets == 0) { | |
1da177e4 LT |
2803 | /* There is nowhere to write, so all non-sync |
2804 | * drives must be failed - so we are finished | |
2805 | */ | |
b7219ccb N |
2806 | sector_t rv; |
2807 | if (min_bad > 0) | |
2808 | max_sector = sector_nr + min_bad; | |
2809 | rv = max_sector - sector_nr; | |
57afd89f | 2810 | *skipped = 1; |
1da177e4 | 2811 | put_buf(r1_bio); |
1da177e4 LT |
2812 | return rv; |
2813 | } | |
2814 | ||
c6207277 N |
2815 | if (max_sector > mddev->resync_max) |
2816 | max_sector = mddev->resync_max; /* Don't do IO beyond here */ | |
06f60385 N |
2817 | if (max_sector > sector_nr + good_sectors) |
2818 | max_sector = sector_nr + good_sectors; | |
1da177e4 | 2819 | nr_sectors = 0; |
289e99e8 | 2820 | sync_blocks = 0; |
1da177e4 LT |
2821 | do { |
2822 | struct page *page; | |
2823 | int len = PAGE_SIZE; | |
2824 | if (sector_nr + (len>>9) > max_sector) | |
2825 | len = (max_sector - sector_nr) << 9; | |
2826 | if (len == 0) | |
2827 | break; | |
6a806c51 | 2828 | if (sync_blocks == 0) { |
e64e4018 AS |
2829 | if (!md_bitmap_start_sync(mddev->bitmap, sector_nr, |
2830 | &sync_blocks, still_degraded) && | |
e5de485f N |
2831 | !conf->fullsync && |
2832 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6a806c51 | 2833 | break; |
7571ae88 | 2834 | if ((len >> 9) > sync_blocks) |
6a806c51 | 2835 | len = sync_blocks<<9; |
ab7a30c7 | 2836 | } |
191ea9b2 | 2837 | |
8f19ccb2 | 2838 | for (i = 0 ; i < conf->raid_disks * 2; i++) { |
98d30c58 ML |
2839 | struct resync_pages *rp; |
2840 | ||
1da177e4 | 2841 | bio = r1_bio->bios[i]; |
98d30c58 | 2842 | rp = get_resync_pages(bio); |
1da177e4 | 2843 | if (bio->bi_end_io) { |
022e510f | 2844 | page = resync_fetch_page(rp, page_idx); |
c85ba149 ML |
2845 | |
2846 | /* | |
2847 | * won't fail because the vec table is big | |
2848 | * enough to hold all these pages | |
2849 | */ | |
2850 | bio_add_page(bio, page, len, 0); | |
1da177e4 LT |
2851 | } |
2852 | } | |
2853 | nr_sectors += len>>9; | |
2854 | sector_nr += len>>9; | |
191ea9b2 | 2855 | sync_blocks -= (len>>9); |
022e510f | 2856 | } while (++page_idx < RESYNC_PAGES); |
98d30c58 | 2857 | |
1da177e4 LT |
2858 | r1_bio->sectors = nr_sectors; |
2859 | ||
c40f341f GR |
2860 | if (mddev_is_clustered(mddev) && |
2861 | conf->cluster_sync_high < sector_nr + nr_sectors) { | |
2862 | conf->cluster_sync_low = mddev->curr_resync_completed; | |
2863 | conf->cluster_sync_high = conf->cluster_sync_low + CLUSTER_RESYNC_WINDOW_SECTORS; | |
2864 | /* Send resync message */ | |
2865 | md_cluster_ops->resync_info_update(mddev, | |
2866 | conf->cluster_sync_low, | |
2867 | conf->cluster_sync_high); | |
2868 | } | |
2869 | ||
d11c171e N |
2870 | /* For a user-requested sync, we read all readable devices and do a |
2871 | * compare | |
2872 | */ | |
2873 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
2874 | atomic_set(&r1_bio->remaining, read_targets); | |
2d4f4f33 | 2875 | for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) { |
d11c171e N |
2876 | bio = r1_bio->bios[i]; |
2877 | if (bio->bi_end_io == end_sync_read) { | |
2d4f4f33 | 2878 | read_targets--; |
74d46992 | 2879 | md_sync_acct_bio(bio, nr_sectors); |
2e52d449 N |
2880 | if (read_targets == 1) |
2881 | bio->bi_opf &= ~MD_FAILFAST; | |
d11c171e N |
2882 | generic_make_request(bio); |
2883 | } | |
2884 | } | |
2885 | } else { | |
2886 | atomic_set(&r1_bio->remaining, 1); | |
2887 | bio = r1_bio->bios[r1_bio->read_disk]; | |
74d46992 | 2888 | md_sync_acct_bio(bio, nr_sectors); |
2e52d449 N |
2889 | if (read_targets == 1) |
2890 | bio->bi_opf &= ~MD_FAILFAST; | |
d11c171e | 2891 | generic_make_request(bio); |
1da177e4 | 2892 | |
d11c171e | 2893 | } |
1da177e4 LT |
2894 | return nr_sectors; |
2895 | } | |
2896 | ||
fd01b88c | 2897 | static sector_t raid1_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce DW |
2898 | { |
2899 | if (sectors) | |
2900 | return sectors; | |
2901 | ||
2902 | return mddev->dev_sectors; | |
2903 | } | |
2904 | ||
e8096360 | 2905 | static struct r1conf *setup_conf(struct mddev *mddev) |
1da177e4 | 2906 | { |
e8096360 | 2907 | struct r1conf *conf; |
709ae487 | 2908 | int i; |
0eaf822c | 2909 | struct raid1_info *disk; |
3cb03002 | 2910 | struct md_rdev *rdev; |
709ae487 | 2911 | int err = -ENOMEM; |
1da177e4 | 2912 | |
e8096360 | 2913 | conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL); |
1da177e4 | 2914 | if (!conf) |
709ae487 | 2915 | goto abort; |
1da177e4 | 2916 | |
fd76863e | 2917 | conf->nr_pending = kcalloc(BARRIER_BUCKETS_NR, |
824e47da | 2918 | sizeof(atomic_t), GFP_KERNEL); |
fd76863e | 2919 | if (!conf->nr_pending) |
2920 | goto abort; | |
2921 | ||
2922 | conf->nr_waiting = kcalloc(BARRIER_BUCKETS_NR, | |
824e47da | 2923 | sizeof(atomic_t), GFP_KERNEL); |
fd76863e | 2924 | if (!conf->nr_waiting) |
2925 | goto abort; | |
2926 | ||
2927 | conf->nr_queued = kcalloc(BARRIER_BUCKETS_NR, | |
824e47da | 2928 | sizeof(atomic_t), GFP_KERNEL); |
fd76863e | 2929 | if (!conf->nr_queued) |
2930 | goto abort; | |
2931 | ||
2932 | conf->barrier = kcalloc(BARRIER_BUCKETS_NR, | |
824e47da | 2933 | sizeof(atomic_t), GFP_KERNEL); |
fd76863e | 2934 | if (!conf->barrier) |
2935 | goto abort; | |
2936 | ||
6396bb22 KC |
2937 | conf->mirrors = kzalloc(array3_size(sizeof(struct raid1_info), |
2938 | mddev->raid_disks, 2), | |
2939 | GFP_KERNEL); | |
1da177e4 | 2940 | if (!conf->mirrors) |
709ae487 | 2941 | goto abort; |
1da177e4 | 2942 | |
ddaf22ab N |
2943 | conf->tmppage = alloc_page(GFP_KERNEL); |
2944 | if (!conf->tmppage) | |
709ae487 | 2945 | goto abort; |
ddaf22ab | 2946 | |
709ae487 | 2947 | conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL); |
1da177e4 | 2948 | if (!conf->poolinfo) |
709ae487 | 2949 | goto abort; |
8f19ccb2 | 2950 | conf->poolinfo->raid_disks = mddev->raid_disks * 2; |
afeee514 KO |
2951 | err = mempool_init(&conf->r1bio_pool, NR_RAID1_BIOS, r1bio_pool_alloc, |
2952 | r1bio_pool_free, conf->poolinfo); | |
2953 | if (err) | |
709ae487 N |
2954 | goto abort; |
2955 | ||
afeee514 KO |
2956 | err = bioset_init(&conf->bio_split, BIO_POOL_SIZE, 0, 0); |
2957 | if (err) | |
c230e7e5 N |
2958 | goto abort; |
2959 | ||
ed9bfdf1 | 2960 | conf->poolinfo->mddev = mddev; |
1da177e4 | 2961 | |
c19d5798 | 2962 | err = -EINVAL; |
e7e72bf6 | 2963 | spin_lock_init(&conf->device_lock); |
dafb20fa | 2964 | rdev_for_each(rdev, mddev) { |
709ae487 | 2965 | int disk_idx = rdev->raid_disk; |
1da177e4 LT |
2966 | if (disk_idx >= mddev->raid_disks |
2967 | || disk_idx < 0) | |
2968 | continue; | |
c19d5798 | 2969 | if (test_bit(Replacement, &rdev->flags)) |
02b898f2 | 2970 | disk = conf->mirrors + mddev->raid_disks + disk_idx; |
c19d5798 N |
2971 | else |
2972 | disk = conf->mirrors + disk_idx; | |
1da177e4 | 2973 | |
c19d5798 N |
2974 | if (disk->rdev) |
2975 | goto abort; | |
1da177e4 | 2976 | disk->rdev = rdev; |
1da177e4 | 2977 | disk->head_position = 0; |
12cee5a8 | 2978 | disk->seq_start = MaxSector; |
1da177e4 LT |
2979 | } |
2980 | conf->raid_disks = mddev->raid_disks; | |
2981 | conf->mddev = mddev; | |
1da177e4 | 2982 | INIT_LIST_HEAD(&conf->retry_list); |
55ce74d4 | 2983 | INIT_LIST_HEAD(&conf->bio_end_io_list); |
1da177e4 LT |
2984 | |
2985 | spin_lock_init(&conf->resync_lock); | |
17999be4 | 2986 | init_waitqueue_head(&conf->wait_barrier); |
1da177e4 | 2987 | |
191ea9b2 | 2988 | bio_list_init(&conf->pending_bio_list); |
34db0cd6 | 2989 | conf->pending_count = 0; |
d890fa2b | 2990 | conf->recovery_disabled = mddev->recovery_disabled - 1; |
191ea9b2 | 2991 | |
c19d5798 | 2992 | err = -EIO; |
8f19ccb2 | 2993 | for (i = 0; i < conf->raid_disks * 2; i++) { |
1da177e4 LT |
2994 | |
2995 | disk = conf->mirrors + i; | |
2996 | ||
c19d5798 N |
2997 | if (i < conf->raid_disks && |
2998 | disk[conf->raid_disks].rdev) { | |
2999 | /* This slot has a replacement. */ | |
3000 | if (!disk->rdev) { | |
3001 | /* No original, just make the replacement | |
3002 | * a recovering spare | |
3003 | */ | |
3004 | disk->rdev = | |
3005 | disk[conf->raid_disks].rdev; | |
3006 | disk[conf->raid_disks].rdev = NULL; | |
3007 | } else if (!test_bit(In_sync, &disk->rdev->flags)) | |
3008 | /* Original is not in_sync - bad */ | |
3009 | goto abort; | |
3010 | } | |
3011 | ||
5fd6c1dc N |
3012 | if (!disk->rdev || |
3013 | !test_bit(In_sync, &disk->rdev->flags)) { | |
1da177e4 | 3014 | disk->head_position = 0; |
4f0a5e01 JB |
3015 | if (disk->rdev && |
3016 | (disk->rdev->saved_raid_disk < 0)) | |
918f0238 | 3017 | conf->fullsync = 1; |
be4d3280 | 3018 | } |
1da177e4 | 3019 | } |
709ae487 | 3020 | |
709ae487 | 3021 | err = -ENOMEM; |
0232605d | 3022 | conf->thread = md_register_thread(raid1d, mddev, "raid1"); |
1d41c216 | 3023 | if (!conf->thread) |
709ae487 | 3024 | goto abort; |
1da177e4 | 3025 | |
709ae487 N |
3026 | return conf; |
3027 | ||
3028 | abort: | |
3029 | if (conf) { | |
afeee514 | 3030 | mempool_exit(&conf->r1bio_pool); |
709ae487 N |
3031 | kfree(conf->mirrors); |
3032 | safe_put_page(conf->tmppage); | |
3033 | kfree(conf->poolinfo); | |
fd76863e | 3034 | kfree(conf->nr_pending); |
3035 | kfree(conf->nr_waiting); | |
3036 | kfree(conf->nr_queued); | |
3037 | kfree(conf->barrier); | |
afeee514 | 3038 | bioset_exit(&conf->bio_split); |
709ae487 N |
3039 | kfree(conf); |
3040 | } | |
3041 | return ERR_PTR(err); | |
3042 | } | |
3043 | ||
afa0f557 | 3044 | static void raid1_free(struct mddev *mddev, void *priv); |
849674e4 | 3045 | static int raid1_run(struct mddev *mddev) |
709ae487 | 3046 | { |
e8096360 | 3047 | struct r1conf *conf; |
709ae487 | 3048 | int i; |
3cb03002 | 3049 | struct md_rdev *rdev; |
5220ea1e | 3050 | int ret; |
2ff8cc2c | 3051 | bool discard_supported = false; |
709ae487 N |
3052 | |
3053 | if (mddev->level != 1) { | |
1d41c216 N |
3054 | pr_warn("md/raid1:%s: raid level not set to mirroring (%d)\n", |
3055 | mdname(mddev), mddev->level); | |
709ae487 N |
3056 | return -EIO; |
3057 | } | |
3058 | if (mddev->reshape_position != MaxSector) { | |
1d41c216 N |
3059 | pr_warn("md/raid1:%s: reshape_position set but not supported\n", |
3060 | mdname(mddev)); | |
709ae487 N |
3061 | return -EIO; |
3062 | } | |
a415c0f1 N |
3063 | if (mddev_init_writes_pending(mddev) < 0) |
3064 | return -ENOMEM; | |
1da177e4 | 3065 | /* |
709ae487 N |
3066 | * copy the already verified devices into our private RAID1 |
3067 | * bookkeeping area. [whatever we allocate in run(), | |
afa0f557 | 3068 | * should be freed in raid1_free()] |
1da177e4 | 3069 | */ |
709ae487 N |
3070 | if (mddev->private == NULL) |
3071 | conf = setup_conf(mddev); | |
3072 | else | |
3073 | conf = mddev->private; | |
1da177e4 | 3074 | |
709ae487 N |
3075 | if (IS_ERR(conf)) |
3076 | return PTR_ERR(conf); | |
1da177e4 | 3077 | |
3deff1a7 | 3078 | if (mddev->queue) { |
5026d7a9 | 3079 | blk_queue_max_write_same_sectors(mddev->queue, 0); |
3deff1a7 CH |
3080 | blk_queue_max_write_zeroes_sectors(mddev->queue, 0); |
3081 | } | |
5026d7a9 | 3082 | |
dafb20fa | 3083 | rdev_for_each(rdev, mddev) { |
1ed7242e JB |
3084 | if (!mddev->gendisk) |
3085 | continue; | |
709ae487 N |
3086 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
3087 | rdev->data_offset << 9); | |
2ff8cc2c SL |
3088 | if (blk_queue_discard(bdev_get_queue(rdev->bdev))) |
3089 | discard_supported = true; | |
1da177e4 | 3090 | } |
191ea9b2 | 3091 | |
709ae487 N |
3092 | mddev->degraded = 0; |
3093 | for (i=0; i < conf->raid_disks; i++) | |
3094 | if (conf->mirrors[i].rdev == NULL || | |
3095 | !test_bit(In_sync, &conf->mirrors[i].rdev->flags) || | |
3096 | test_bit(Faulty, &conf->mirrors[i].rdev->flags)) | |
3097 | mddev->degraded++; | |
3098 | ||
3099 | if (conf->raid_disks - mddev->degraded == 1) | |
3100 | mddev->recovery_cp = MaxSector; | |
3101 | ||
8c6ac868 | 3102 | if (mddev->recovery_cp != MaxSector) |
1d41c216 N |
3103 | pr_info("md/raid1:%s: not clean -- starting background reconstruction\n", |
3104 | mdname(mddev)); | |
3105 | pr_info("md/raid1:%s: active with %d out of %d mirrors\n", | |
f72ffdd6 | 3106 | mdname(mddev), mddev->raid_disks - mddev->degraded, |
1da177e4 | 3107 | mddev->raid_disks); |
709ae487 | 3108 | |
1da177e4 LT |
3109 | /* |
3110 | * Ok, everything is just fine now | |
3111 | */ | |
709ae487 N |
3112 | mddev->thread = conf->thread; |
3113 | conf->thread = NULL; | |
3114 | mddev->private = conf; | |
46533ff7 | 3115 | set_bit(MD_FAILFAST_SUPPORTED, &mddev->flags); |
709ae487 | 3116 | |
1f403624 | 3117 | md_set_array_sectors(mddev, raid1_size(mddev, 0, 0)); |
1da177e4 | 3118 | |
1ed7242e | 3119 | if (mddev->queue) { |
2ff8cc2c | 3120 | if (discard_supported) |
8b904b5b | 3121 | blk_queue_flag_set(QUEUE_FLAG_DISCARD, |
2ff8cc2c SL |
3122 | mddev->queue); |
3123 | else | |
8b904b5b | 3124 | blk_queue_flag_clear(QUEUE_FLAG_DISCARD, |
2ff8cc2c | 3125 | mddev->queue); |
1ed7242e | 3126 | } |
5220ea1e | 3127 | |
3128 | ret = md_integrity_register(mddev); | |
5aa61f42 N |
3129 | if (ret) { |
3130 | md_unregister_thread(&mddev->thread); | |
afa0f557 | 3131 | raid1_free(mddev, conf); |
5aa61f42 | 3132 | } |
5220ea1e | 3133 | return ret; |
1da177e4 LT |
3134 | } |
3135 | ||
afa0f557 | 3136 | static void raid1_free(struct mddev *mddev, void *priv) |
1da177e4 | 3137 | { |
afa0f557 | 3138 | struct r1conf *conf = priv; |
409c57f3 | 3139 | |
afeee514 | 3140 | mempool_exit(&conf->r1bio_pool); |
990a8baf | 3141 | kfree(conf->mirrors); |
0fea7ed8 | 3142 | safe_put_page(conf->tmppage); |
990a8baf | 3143 | kfree(conf->poolinfo); |
fd76863e | 3144 | kfree(conf->nr_pending); |
3145 | kfree(conf->nr_waiting); | |
3146 | kfree(conf->nr_queued); | |
3147 | kfree(conf->barrier); | |
afeee514 | 3148 | bioset_exit(&conf->bio_split); |
1da177e4 | 3149 | kfree(conf); |
1da177e4 LT |
3150 | } |
3151 | ||
fd01b88c | 3152 | static int raid1_resize(struct mddev *mddev, sector_t sectors) |
1da177e4 LT |
3153 | { |
3154 | /* no resync is happening, and there is enough space | |
3155 | * on all devices, so we can resize. | |
3156 | * We need to make sure resync covers any new space. | |
3157 | * If the array is shrinking we should possibly wait until | |
3158 | * any io in the removed space completes, but it hardly seems | |
3159 | * worth it. | |
3160 | */ | |
a4a6125a N |
3161 | sector_t newsize = raid1_size(mddev, sectors, 0); |
3162 | if (mddev->external_size && | |
3163 | mddev->array_sectors > newsize) | |
b522adcd | 3164 | return -EINVAL; |
a4a6125a | 3165 | if (mddev->bitmap) { |
e64e4018 | 3166 | int ret = md_bitmap_resize(mddev->bitmap, newsize, 0, 0); |
a4a6125a N |
3167 | if (ret) |
3168 | return ret; | |
3169 | } | |
3170 | md_set_array_sectors(mddev, newsize); | |
b522adcd | 3171 | if (sectors > mddev->dev_sectors && |
b098636c | 3172 | mddev->recovery_cp > mddev->dev_sectors) { |
58c0fed4 | 3173 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
3174 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
3175 | } | |
b522adcd | 3176 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 3177 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
3178 | return 0; |
3179 | } | |
3180 | ||
fd01b88c | 3181 | static int raid1_reshape(struct mddev *mddev) |
1da177e4 LT |
3182 | { |
3183 | /* We need to: | |
3184 | * 1/ resize the r1bio_pool | |
3185 | * 2/ resize conf->mirrors | |
3186 | * | |
3187 | * We allocate a new r1bio_pool if we can. | |
3188 | * Then raise a device barrier and wait until all IO stops. | |
3189 | * Then resize conf->mirrors and swap in the new r1bio pool. | |
6ea9c07c N |
3190 | * |
3191 | * At the same time, we "pack" the devices so that all the missing | |
3192 | * devices have the higher raid_disk numbers. | |
1da177e4 | 3193 | */ |
afeee514 | 3194 | mempool_t newpool, oldpool; |
1da177e4 | 3195 | struct pool_info *newpoolinfo; |
0eaf822c | 3196 | struct raid1_info *newmirrors; |
e8096360 | 3197 | struct r1conf *conf = mddev->private; |
63c70c4f | 3198 | int cnt, raid_disks; |
c04be0aa | 3199 | unsigned long flags; |
2214c260 | 3200 | int d, d2; |
afeee514 KO |
3201 | int ret; |
3202 | ||
3203 | memset(&newpool, 0, sizeof(newpool)); | |
3204 | memset(&oldpool, 0, sizeof(oldpool)); | |
1da177e4 | 3205 | |
63c70c4f | 3206 | /* Cannot change chunk_size, layout, or level */ |
664e7c41 | 3207 | if (mddev->chunk_sectors != mddev->new_chunk_sectors || |
63c70c4f N |
3208 | mddev->layout != mddev->new_layout || |
3209 | mddev->level != mddev->new_level) { | |
664e7c41 | 3210 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
63c70c4f N |
3211 | mddev->new_layout = mddev->layout; |
3212 | mddev->new_level = mddev->level; | |
3213 | return -EINVAL; | |
3214 | } | |
3215 | ||
2214c260 AP |
3216 | if (!mddev_is_clustered(mddev)) |
3217 | md_allow_write(mddev); | |
2a2275d6 | 3218 | |
63c70c4f N |
3219 | raid_disks = mddev->raid_disks + mddev->delta_disks; |
3220 | ||
6ea9c07c N |
3221 | if (raid_disks < conf->raid_disks) { |
3222 | cnt=0; | |
3223 | for (d= 0; d < conf->raid_disks; d++) | |
3224 | if (conf->mirrors[d].rdev) | |
3225 | cnt++; | |
3226 | if (cnt > raid_disks) | |
1da177e4 | 3227 | return -EBUSY; |
6ea9c07c | 3228 | } |
1da177e4 LT |
3229 | |
3230 | newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL); | |
3231 | if (!newpoolinfo) | |
3232 | return -ENOMEM; | |
3233 | newpoolinfo->mddev = mddev; | |
8f19ccb2 | 3234 | newpoolinfo->raid_disks = raid_disks * 2; |
1da177e4 | 3235 | |
afeee514 KO |
3236 | ret = mempool_init(&newpool, NR_RAID1_BIOS, r1bio_pool_alloc, |
3237 | r1bio_pool_free, newpoolinfo); | |
3238 | if (ret) { | |
1da177e4 | 3239 | kfree(newpoolinfo); |
afeee514 | 3240 | return ret; |
1da177e4 | 3241 | } |
6396bb22 KC |
3242 | newmirrors = kzalloc(array3_size(sizeof(struct raid1_info), |
3243 | raid_disks, 2), | |
8f19ccb2 | 3244 | GFP_KERNEL); |
1da177e4 LT |
3245 | if (!newmirrors) { |
3246 | kfree(newpoolinfo); | |
afeee514 | 3247 | mempool_exit(&newpool); |
1da177e4 LT |
3248 | return -ENOMEM; |
3249 | } | |
1da177e4 | 3250 | |
e2d59925 | 3251 | freeze_array(conf, 0); |
1da177e4 LT |
3252 | |
3253 | /* ok, everything is stopped */ | |
3254 | oldpool = conf->r1bio_pool; | |
3255 | conf->r1bio_pool = newpool; | |
6ea9c07c | 3256 | |
a88aa786 | 3257 | for (d = d2 = 0; d < conf->raid_disks; d++) { |
3cb03002 | 3258 | struct md_rdev *rdev = conf->mirrors[d].rdev; |
a88aa786 | 3259 | if (rdev && rdev->raid_disk != d2) { |
36fad858 | 3260 | sysfs_unlink_rdev(mddev, rdev); |
a88aa786 | 3261 | rdev->raid_disk = d2; |
36fad858 NK |
3262 | sysfs_unlink_rdev(mddev, rdev); |
3263 | if (sysfs_link_rdev(mddev, rdev)) | |
1d41c216 N |
3264 | pr_warn("md/raid1:%s: cannot register rd%d\n", |
3265 | mdname(mddev), rdev->raid_disk); | |
6ea9c07c | 3266 | } |
a88aa786 N |
3267 | if (rdev) |
3268 | newmirrors[d2++].rdev = rdev; | |
3269 | } | |
1da177e4 LT |
3270 | kfree(conf->mirrors); |
3271 | conf->mirrors = newmirrors; | |
3272 | kfree(conf->poolinfo); | |
3273 | conf->poolinfo = newpoolinfo; | |
3274 | ||
c04be0aa | 3275 | spin_lock_irqsave(&conf->device_lock, flags); |
1da177e4 | 3276 | mddev->degraded += (raid_disks - conf->raid_disks); |
c04be0aa | 3277 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 3278 | conf->raid_disks = mddev->raid_disks = raid_disks; |
63c70c4f | 3279 | mddev->delta_disks = 0; |
1da177e4 | 3280 | |
e2d59925 | 3281 | unfreeze_array(conf); |
1da177e4 | 3282 | |
985ca973 | 3283 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
1da177e4 LT |
3284 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
3285 | md_wakeup_thread(mddev->thread); | |
3286 | ||
afeee514 | 3287 | mempool_exit(&oldpool); |
1da177e4 LT |
3288 | return 0; |
3289 | } | |
3290 | ||
b03e0ccb | 3291 | static void raid1_quiesce(struct mddev *mddev, int quiesce) |
36fa3063 | 3292 | { |
e8096360 | 3293 | struct r1conf *conf = mddev->private; |
36fa3063 | 3294 | |
b03e0ccb | 3295 | if (quiesce) |
07169fd4 | 3296 | freeze_array(conf, 0); |
b03e0ccb | 3297 | else |
07169fd4 | 3298 | unfreeze_array(conf); |
36fa3063 N |
3299 | } |
3300 | ||
fd01b88c | 3301 | static void *raid1_takeover(struct mddev *mddev) |
709ae487 N |
3302 | { |
3303 | /* raid1 can take over: | |
3304 | * raid5 with 2 devices, any layout or chunk size | |
3305 | */ | |
3306 | if (mddev->level == 5 && mddev->raid_disks == 2) { | |
e8096360 | 3307 | struct r1conf *conf; |
709ae487 N |
3308 | mddev->new_level = 1; |
3309 | mddev->new_layout = 0; | |
3310 | mddev->new_chunk_sectors = 0; | |
3311 | conf = setup_conf(mddev); | |
6995f0b2 | 3312 | if (!IS_ERR(conf)) { |
07169fd4 | 3313 | /* Array must appear to be quiesced */ |
3314 | conf->array_frozen = 1; | |
394ed8e4 SL |
3315 | mddev_clear_unsupported_flags(mddev, |
3316 | UNSUPPORTED_MDDEV_FLAGS); | |
6995f0b2 | 3317 | } |
709ae487 N |
3318 | return conf; |
3319 | } | |
3320 | return ERR_PTR(-EINVAL); | |
3321 | } | |
1da177e4 | 3322 | |
84fc4b56 | 3323 | static struct md_personality raid1_personality = |
1da177e4 LT |
3324 | { |
3325 | .name = "raid1", | |
2604b703 | 3326 | .level = 1, |
1da177e4 | 3327 | .owner = THIS_MODULE, |
849674e4 SL |
3328 | .make_request = raid1_make_request, |
3329 | .run = raid1_run, | |
afa0f557 | 3330 | .free = raid1_free, |
849674e4 SL |
3331 | .status = raid1_status, |
3332 | .error_handler = raid1_error, | |
1da177e4 LT |
3333 | .hot_add_disk = raid1_add_disk, |
3334 | .hot_remove_disk= raid1_remove_disk, | |
3335 | .spare_active = raid1_spare_active, | |
849674e4 | 3336 | .sync_request = raid1_sync_request, |
1da177e4 | 3337 | .resize = raid1_resize, |
80c3a6ce | 3338 | .size = raid1_size, |
63c70c4f | 3339 | .check_reshape = raid1_reshape, |
36fa3063 | 3340 | .quiesce = raid1_quiesce, |
709ae487 | 3341 | .takeover = raid1_takeover, |
5c675f83 | 3342 | .congested = raid1_congested, |
1da177e4 LT |
3343 | }; |
3344 | ||
3345 | static int __init raid_init(void) | |
3346 | { | |
2604b703 | 3347 | return register_md_personality(&raid1_personality); |
1da177e4 LT |
3348 | } |
3349 | ||
3350 | static void raid_exit(void) | |
3351 | { | |
2604b703 | 3352 | unregister_md_personality(&raid1_personality); |
1da177e4 LT |
3353 | } |
3354 | ||
3355 | module_init(raid_init); | |
3356 | module_exit(raid_exit); | |
3357 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 3358 | MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD"); |
1da177e4 | 3359 | MODULE_ALIAS("md-personality-3"); /* RAID1 */ |
d9d166c2 | 3360 | MODULE_ALIAS("md-raid1"); |
2604b703 | 3361 | MODULE_ALIAS("md-level-1"); |
34db0cd6 N |
3362 | |
3363 | module_param(max_queued_requests, int, S_IRUGO|S_IWUSR); |