]>
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 IM |
40 | #include <linux/sched/signal.h> |
41 | ||
109e3765 | 42 | #include <trace/events/block.h> |
3f07c014 | 43 | |
43b2e5d8 | 44 | #include "md.h" |
ef740c37 CH |
45 | #include "raid1.h" |
46 | #include "bitmap.h" | |
191ea9b2 | 47 | |
394ed8e4 SL |
48 | #define UNSUPPORTED_MDDEV_FLAGS \ |
49 | ((1L << MD_HAS_JOURNAL) | \ | |
ea0213e0 AP |
50 | (1L << MD_JOURNAL_CLEAN) | \ |
51 | (1L << MD_HAS_PPL)) | |
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 | ||
98d30c58 ML |
84 | /* |
85 | * 'strct resync_pages' stores actual pages used for doing the resync | |
86 | * IO, and it is per-bio, so make .bi_private points to it. | |
87 | */ | |
88 | static inline struct resync_pages *get_resync_pages(struct bio *bio) | |
89 | { | |
90 | return bio->bi_private; | |
91 | } | |
92 | ||
93 | /* | |
94 | * for resync bio, r1bio pointer can be retrieved from the per-bio | |
95 | * 'struct resync_pages'. | |
96 | */ | |
97 | static inline struct r1bio *get_resync_r1bio(struct bio *bio) | |
98 | { | |
99 | return get_resync_pages(bio)->raid_bio; | |
100 | } | |
101 | ||
dd0fc66f | 102 | static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 LT |
103 | { |
104 | struct pool_info *pi = data; | |
9f2c9d12 | 105 | int size = offsetof(struct r1bio, bios[pi->raid_disks]); |
1da177e4 LT |
106 | |
107 | /* allocate a r1bio with room for raid_disks entries in the bios array */ | |
7eaceacc | 108 | return kzalloc(size, gfp_flags); |
1da177e4 LT |
109 | } |
110 | ||
111 | static void r1bio_pool_free(void *r1_bio, void *data) | |
112 | { | |
113 | kfree(r1_bio); | |
114 | } | |
115 | ||
8e005f7c | 116 | #define RESYNC_DEPTH 32 |
1da177e4 | 117 | #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9) |
8e005f7c | 118 | #define RESYNC_WINDOW (RESYNC_BLOCK_SIZE * RESYNC_DEPTH) |
119 | #define RESYNC_WINDOW_SECTORS (RESYNC_WINDOW >> 9) | |
c40f341f GR |
120 | #define CLUSTER_RESYNC_WINDOW (16 * RESYNC_WINDOW) |
121 | #define CLUSTER_RESYNC_WINDOW_SECTORS (CLUSTER_RESYNC_WINDOW >> 9) | |
1da177e4 | 122 | |
dd0fc66f | 123 | static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 LT |
124 | { |
125 | struct pool_info *pi = data; | |
9f2c9d12 | 126 | struct r1bio *r1_bio; |
1da177e4 | 127 | struct bio *bio; |
da1aab3d | 128 | int need_pages; |
98d30c58 ML |
129 | int j; |
130 | struct resync_pages *rps; | |
1da177e4 LT |
131 | |
132 | r1_bio = r1bio_pool_alloc(gfp_flags, pi); | |
7eaceacc | 133 | if (!r1_bio) |
1da177e4 | 134 | return NULL; |
1da177e4 | 135 | |
98d30c58 ML |
136 | rps = kmalloc(sizeof(struct resync_pages) * pi->raid_disks, |
137 | gfp_flags); | |
138 | if (!rps) | |
139 | goto out_free_r1bio; | |
140 | ||
1da177e4 LT |
141 | /* |
142 | * Allocate bios : 1 for reading, n-1 for writing | |
143 | */ | |
144 | for (j = pi->raid_disks ; j-- ; ) { | |
6746557f | 145 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); |
1da177e4 LT |
146 | if (!bio) |
147 | goto out_free_bio; | |
148 | r1_bio->bios[j] = bio; | |
149 | } | |
150 | /* | |
151 | * Allocate RESYNC_PAGES data pages and attach them to | |
d11c171e N |
152 | * the first bio. |
153 | * If this is a user-requested check/repair, allocate | |
154 | * RESYNC_PAGES for each bio. | |
1da177e4 | 155 | */ |
d11c171e | 156 | if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) |
da1aab3d | 157 | need_pages = pi->raid_disks; |
d11c171e | 158 | else |
da1aab3d | 159 | need_pages = 1; |
98d30c58 ML |
160 | for (j = 0; j < pi->raid_disks; j++) { |
161 | struct resync_pages *rp = &rps[j]; | |
162 | ||
d11c171e | 163 | bio = r1_bio->bios[j]; |
d11c171e | 164 | |
98d30c58 ML |
165 | if (j < need_pages) { |
166 | if (resync_alloc_pages(rp, gfp_flags)) | |
167 | goto out_free_pages; | |
168 | } else { | |
169 | memcpy(rp, &rps[0], sizeof(*rp)); | |
170 | resync_get_all_pages(rp); | |
171 | } | |
172 | ||
173 | rp->idx = 0; | |
174 | rp->raid_bio = r1_bio; | |
175 | bio->bi_private = rp; | |
1da177e4 LT |
176 | } |
177 | ||
178 | r1_bio->master_bio = NULL; | |
179 | ||
180 | return r1_bio; | |
181 | ||
da1aab3d | 182 | out_free_pages: |
491221f8 | 183 | while (--j >= 0) |
98d30c58 | 184 | resync_free_pages(&rps[j]); |
da1aab3d | 185 | |
1da177e4 | 186 | out_free_bio: |
8f19ccb2 | 187 | while (++j < pi->raid_disks) |
1da177e4 | 188 | bio_put(r1_bio->bios[j]); |
98d30c58 ML |
189 | kfree(rps); |
190 | ||
191 | out_free_r1bio: | |
1da177e4 LT |
192 | r1bio_pool_free(r1_bio, data); |
193 | return NULL; | |
194 | } | |
195 | ||
196 | static void r1buf_pool_free(void *__r1_bio, void *data) | |
197 | { | |
198 | struct pool_info *pi = data; | |
98d30c58 | 199 | int i; |
9f2c9d12 | 200 | struct r1bio *r1bio = __r1_bio; |
98d30c58 | 201 | struct resync_pages *rp = NULL; |
1da177e4 | 202 | |
98d30c58 ML |
203 | for (i = pi->raid_disks; i--; ) { |
204 | rp = get_resync_pages(r1bio->bios[i]); | |
205 | resync_free_pages(rp); | |
1da177e4 | 206 | bio_put(r1bio->bios[i]); |
98d30c58 ML |
207 | } |
208 | ||
209 | /* resync pages array stored in the 1st bio's .bi_private */ | |
210 | kfree(rp); | |
1da177e4 LT |
211 | |
212 | r1bio_pool_free(r1bio, data); | |
213 | } | |
214 | ||
e8096360 | 215 | static void put_all_bios(struct r1conf *conf, struct r1bio *r1_bio) |
1da177e4 LT |
216 | { |
217 | int i; | |
218 | ||
8f19ccb2 | 219 | for (i = 0; i < conf->raid_disks * 2; i++) { |
1da177e4 | 220 | struct bio **bio = r1_bio->bios + i; |
4367af55 | 221 | if (!BIO_SPECIAL(*bio)) |
1da177e4 LT |
222 | bio_put(*bio); |
223 | *bio = NULL; | |
224 | } | |
225 | } | |
226 | ||
9f2c9d12 | 227 | static void free_r1bio(struct r1bio *r1_bio) |
1da177e4 | 228 | { |
e8096360 | 229 | struct r1conf *conf = r1_bio->mddev->private; |
1da177e4 | 230 | |
1da177e4 LT |
231 | put_all_bios(conf, r1_bio); |
232 | mempool_free(r1_bio, conf->r1bio_pool); | |
233 | } | |
234 | ||
9f2c9d12 | 235 | static void put_buf(struct r1bio *r1_bio) |
1da177e4 | 236 | { |
e8096360 | 237 | struct r1conf *conf = r1_bio->mddev->private; |
af5f42a7 | 238 | sector_t sect = r1_bio->sector; |
3e198f78 N |
239 | int i; |
240 | ||
8f19ccb2 | 241 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3e198f78 N |
242 | struct bio *bio = r1_bio->bios[i]; |
243 | if (bio->bi_end_io) | |
244 | rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev); | |
245 | } | |
1da177e4 LT |
246 | |
247 | mempool_free(r1_bio, conf->r1buf_pool); | |
248 | ||
af5f42a7 | 249 | lower_barrier(conf, sect); |
1da177e4 LT |
250 | } |
251 | ||
9f2c9d12 | 252 | static void reschedule_retry(struct r1bio *r1_bio) |
1da177e4 LT |
253 | { |
254 | unsigned long flags; | |
fd01b88c | 255 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 256 | struct r1conf *conf = mddev->private; |
fd76863e | 257 | int idx; |
1da177e4 | 258 | |
fd76863e | 259 | idx = sector_to_idx(r1_bio->sector); |
1da177e4 LT |
260 | spin_lock_irqsave(&conf->device_lock, flags); |
261 | list_add(&r1_bio->retry_list, &conf->retry_list); | |
824e47da | 262 | atomic_inc(&conf->nr_queued[idx]); |
1da177e4 LT |
263 | spin_unlock_irqrestore(&conf->device_lock, flags); |
264 | ||
17999be4 | 265 | wake_up(&conf->wait_barrier); |
1da177e4 LT |
266 | md_wakeup_thread(mddev->thread); |
267 | } | |
268 | ||
269 | /* | |
270 | * raid_end_bio_io() is called when we have finished servicing a mirrored | |
271 | * operation and are ready to return a success/failure code to the buffer | |
272 | * cache layer. | |
273 | */ | |
9f2c9d12 | 274 | static void call_bio_endio(struct r1bio *r1_bio) |
d2eb35ac N |
275 | { |
276 | struct bio *bio = r1_bio->master_bio; | |
e8096360 | 277 | struct r1conf *conf = r1_bio->mddev->private; |
d2eb35ac N |
278 | |
279 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) | |
4246a0b6 CH |
280 | bio->bi_error = -EIO; |
281 | ||
37011e3a N |
282 | bio_endio(bio); |
283 | /* | |
284 | * Wake up any possible resync thread that waits for the device | |
285 | * to go idle. | |
286 | */ | |
287 | allow_barrier(conf, r1_bio->sector); | |
d2eb35ac N |
288 | } |
289 | ||
9f2c9d12 | 290 | static void raid_end_bio_io(struct r1bio *r1_bio) |
1da177e4 LT |
291 | { |
292 | struct bio *bio = r1_bio->master_bio; | |
293 | ||
4b6d287f N |
294 | /* if nobody has done the final endio yet, do it now */ |
295 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { | |
36a4e1fe N |
296 | pr_debug("raid1: sync end %s on sectors %llu-%llu\n", |
297 | (bio_data_dir(bio) == WRITE) ? "write" : "read", | |
4f024f37 KO |
298 | (unsigned long long) bio->bi_iter.bi_sector, |
299 | (unsigned long long) bio_end_sector(bio) - 1); | |
4b6d287f | 300 | |
d2eb35ac | 301 | call_bio_endio(r1_bio); |
4b6d287f | 302 | } |
1da177e4 LT |
303 | free_r1bio(r1_bio); |
304 | } | |
305 | ||
306 | /* | |
307 | * Update disk head position estimator based on IRQ completion info. | |
308 | */ | |
9f2c9d12 | 309 | static inline void update_head_pos(int disk, struct r1bio *r1_bio) |
1da177e4 | 310 | { |
e8096360 | 311 | struct r1conf *conf = r1_bio->mddev->private; |
1da177e4 LT |
312 | |
313 | conf->mirrors[disk].head_position = | |
314 | r1_bio->sector + (r1_bio->sectors); | |
315 | } | |
316 | ||
ba3ae3be NK |
317 | /* |
318 | * Find the disk number which triggered given bio | |
319 | */ | |
9f2c9d12 | 320 | static int find_bio_disk(struct r1bio *r1_bio, struct bio *bio) |
ba3ae3be NK |
321 | { |
322 | int mirror; | |
30194636 N |
323 | struct r1conf *conf = r1_bio->mddev->private; |
324 | int raid_disks = conf->raid_disks; | |
ba3ae3be | 325 | |
8f19ccb2 | 326 | for (mirror = 0; mirror < raid_disks * 2; mirror++) |
ba3ae3be NK |
327 | if (r1_bio->bios[mirror] == bio) |
328 | break; | |
329 | ||
8f19ccb2 | 330 | BUG_ON(mirror == raid_disks * 2); |
ba3ae3be NK |
331 | update_head_pos(mirror, r1_bio); |
332 | ||
333 | return mirror; | |
334 | } | |
335 | ||
4246a0b6 | 336 | static void raid1_end_read_request(struct bio *bio) |
1da177e4 | 337 | { |
4246a0b6 | 338 | int uptodate = !bio->bi_error; |
9f2c9d12 | 339 | struct r1bio *r1_bio = bio->bi_private; |
e8096360 | 340 | struct r1conf *conf = r1_bio->mddev->private; |
e5872d58 | 341 | struct md_rdev *rdev = conf->mirrors[r1_bio->read_disk].rdev; |
1da177e4 | 342 | |
1da177e4 LT |
343 | /* |
344 | * this branch is our 'one mirror IO has finished' event handler: | |
345 | */ | |
e5872d58 | 346 | update_head_pos(r1_bio->read_disk, r1_bio); |
ddaf22ab | 347 | |
dd00a99e N |
348 | if (uptodate) |
349 | set_bit(R1BIO_Uptodate, &r1_bio->state); | |
2e52d449 N |
350 | else if (test_bit(FailFast, &rdev->flags) && |
351 | test_bit(R1BIO_FailFast, &r1_bio->state)) | |
352 | /* This was a fail-fast read so we definitely | |
353 | * want to retry */ | |
354 | ; | |
dd00a99e N |
355 | else { |
356 | /* If all other devices have failed, we want to return | |
357 | * the error upwards rather than fail the last device. | |
358 | * Here we redefine "uptodate" to mean "Don't want to retry" | |
1da177e4 | 359 | */ |
dd00a99e N |
360 | unsigned long flags; |
361 | spin_lock_irqsave(&conf->device_lock, flags); | |
362 | if (r1_bio->mddev->degraded == conf->raid_disks || | |
363 | (r1_bio->mddev->degraded == conf->raid_disks-1 && | |
e5872d58 | 364 | test_bit(In_sync, &rdev->flags))) |
dd00a99e N |
365 | uptodate = 1; |
366 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
367 | } | |
1da177e4 | 368 | |
7ad4d4a6 | 369 | if (uptodate) { |
1da177e4 | 370 | raid_end_bio_io(r1_bio); |
e5872d58 | 371 | rdev_dec_pending(rdev, conf->mddev); |
7ad4d4a6 | 372 | } else { |
1da177e4 LT |
373 | /* |
374 | * oops, read error: | |
375 | */ | |
376 | char b[BDEVNAME_SIZE]; | |
1d41c216 N |
377 | pr_err_ratelimited("md/raid1:%s: %s: rescheduling sector %llu\n", |
378 | mdname(conf->mddev), | |
379 | bdevname(rdev->bdev, b), | |
380 | (unsigned long long)r1_bio->sector); | |
d2eb35ac | 381 | set_bit(R1BIO_ReadError, &r1_bio->state); |
1da177e4 | 382 | reschedule_retry(r1_bio); |
7ad4d4a6 | 383 | /* don't drop the reference on read_disk yet */ |
1da177e4 | 384 | } |
1da177e4 LT |
385 | } |
386 | ||
9f2c9d12 | 387 | static void close_write(struct r1bio *r1_bio) |
cd5ff9a1 N |
388 | { |
389 | /* it really is the end of this request */ | |
390 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { | |
841c1316 ML |
391 | bio_free_pages(r1_bio->behind_master_bio); |
392 | bio_put(r1_bio->behind_master_bio); | |
393 | r1_bio->behind_master_bio = NULL; | |
cd5ff9a1 N |
394 | } |
395 | /* clear the bitmap if all writes complete successfully */ | |
396 | bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector, | |
397 | r1_bio->sectors, | |
398 | !test_bit(R1BIO_Degraded, &r1_bio->state), | |
399 | test_bit(R1BIO_BehindIO, &r1_bio->state)); | |
400 | md_write_end(r1_bio->mddev); | |
401 | } | |
402 | ||
9f2c9d12 | 403 | static void r1_bio_write_done(struct r1bio *r1_bio) |
4e78064f | 404 | { |
cd5ff9a1 N |
405 | if (!atomic_dec_and_test(&r1_bio->remaining)) |
406 | return; | |
407 | ||
408 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
409 | reschedule_retry(r1_bio); | |
410 | else { | |
411 | close_write(r1_bio); | |
4367af55 N |
412 | if (test_bit(R1BIO_MadeGood, &r1_bio->state)) |
413 | reschedule_retry(r1_bio); | |
414 | else | |
415 | raid_end_bio_io(r1_bio); | |
4e78064f N |
416 | } |
417 | } | |
418 | ||
4246a0b6 | 419 | static void raid1_end_write_request(struct bio *bio) |
1da177e4 | 420 | { |
9f2c9d12 | 421 | struct r1bio *r1_bio = bio->bi_private; |
e5872d58 | 422 | int behind = test_bit(R1BIO_BehindIO, &r1_bio->state); |
e8096360 | 423 | struct r1conf *conf = r1_bio->mddev->private; |
04b857f7 | 424 | struct bio *to_put = NULL; |
e5872d58 N |
425 | int mirror = find_bio_disk(r1_bio, bio); |
426 | struct md_rdev *rdev = conf->mirrors[mirror].rdev; | |
e3f948cd SL |
427 | bool discard_error; |
428 | ||
429 | discard_error = bio->bi_error && bio_op(bio) == REQ_OP_DISCARD; | |
1da177e4 | 430 | |
e9c7469b TH |
431 | /* |
432 | * 'one mirror IO has finished' event handler: | |
433 | */ | |
e3f948cd | 434 | if (bio->bi_error && !discard_error) { |
e5872d58 N |
435 | set_bit(WriteErrorSeen, &rdev->flags); |
436 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) | |
19d67169 N |
437 | set_bit(MD_RECOVERY_NEEDED, & |
438 | conf->mddev->recovery); | |
439 | ||
212e7eb7 N |
440 | if (test_bit(FailFast, &rdev->flags) && |
441 | (bio->bi_opf & MD_FAILFAST) && | |
442 | /* We never try FailFast to WriteMostly devices */ | |
443 | !test_bit(WriteMostly, &rdev->flags)) { | |
444 | md_error(r1_bio->mddev, rdev); | |
445 | if (!test_bit(Faulty, &rdev->flags)) | |
446 | /* This is the only remaining device, | |
447 | * We need to retry the write without | |
448 | * FailFast | |
449 | */ | |
450 | set_bit(R1BIO_WriteError, &r1_bio->state); | |
451 | else { | |
452 | /* Finished with this branch */ | |
453 | r1_bio->bios[mirror] = NULL; | |
454 | to_put = bio; | |
455 | } | |
456 | } else | |
457 | set_bit(R1BIO_WriteError, &r1_bio->state); | |
4367af55 | 458 | } else { |
1da177e4 | 459 | /* |
e9c7469b TH |
460 | * Set R1BIO_Uptodate in our master bio, so that we |
461 | * will return a good error code for to the higher | |
462 | * levels even if IO on some other mirrored buffer | |
463 | * fails. | |
464 | * | |
465 | * The 'master' represents the composite IO operation | |
466 | * to user-side. So if something waits for IO, then it | |
467 | * will wait for the 'master' bio. | |
1da177e4 | 468 | */ |
4367af55 N |
469 | sector_t first_bad; |
470 | int bad_sectors; | |
471 | ||
cd5ff9a1 N |
472 | r1_bio->bios[mirror] = NULL; |
473 | to_put = bio; | |
3056e3ae AL |
474 | /* |
475 | * Do not set R1BIO_Uptodate if the current device is | |
476 | * rebuilding or Faulty. This is because we cannot use | |
477 | * such device for properly reading the data back (we could | |
478 | * potentially use it, if the current write would have felt | |
479 | * before rdev->recovery_offset, but for simplicity we don't | |
480 | * check this here. | |
481 | */ | |
e5872d58 N |
482 | if (test_bit(In_sync, &rdev->flags) && |
483 | !test_bit(Faulty, &rdev->flags)) | |
3056e3ae | 484 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
e9c7469b | 485 | |
4367af55 | 486 | /* Maybe we can clear some bad blocks. */ |
e5872d58 | 487 | if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors, |
e3f948cd | 488 | &first_bad, &bad_sectors) && !discard_error) { |
4367af55 N |
489 | r1_bio->bios[mirror] = IO_MADE_GOOD; |
490 | set_bit(R1BIO_MadeGood, &r1_bio->state); | |
491 | } | |
492 | } | |
493 | ||
e9c7469b | 494 | if (behind) { |
841c1316 ML |
495 | /* we release behind master bio when all write are done */ |
496 | if (r1_bio->behind_master_bio == bio) | |
497 | to_put = NULL; | |
498 | ||
e5872d58 | 499 | if (test_bit(WriteMostly, &rdev->flags)) |
e9c7469b TH |
500 | atomic_dec(&r1_bio->behind_remaining); |
501 | ||
502 | /* | |
503 | * In behind mode, we ACK the master bio once the I/O | |
504 | * has safely reached all non-writemostly | |
505 | * disks. Setting the Returned bit ensures that this | |
506 | * gets done only once -- we don't ever want to return | |
507 | * -EIO here, instead we'll wait | |
508 | */ | |
509 | if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) && | |
510 | test_bit(R1BIO_Uptodate, &r1_bio->state)) { | |
511 | /* Maybe we can return now */ | |
512 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { | |
513 | struct bio *mbio = r1_bio->master_bio; | |
36a4e1fe N |
514 | pr_debug("raid1: behind end write sectors" |
515 | " %llu-%llu\n", | |
4f024f37 KO |
516 | (unsigned long long) mbio->bi_iter.bi_sector, |
517 | (unsigned long long) bio_end_sector(mbio) - 1); | |
d2eb35ac | 518 | call_bio_endio(r1_bio); |
4b6d287f N |
519 | } |
520 | } | |
521 | } | |
4367af55 | 522 | if (r1_bio->bios[mirror] == NULL) |
e5872d58 | 523 | rdev_dec_pending(rdev, conf->mddev); |
e9c7469b | 524 | |
1da177e4 | 525 | /* |
1da177e4 LT |
526 | * Let's see if all mirrored write operations have finished |
527 | * already. | |
528 | */ | |
af6d7b76 | 529 | r1_bio_write_done(r1_bio); |
c70810b3 | 530 | |
04b857f7 N |
531 | if (to_put) |
532 | bio_put(to_put); | |
1da177e4 LT |
533 | } |
534 | ||
fd76863e | 535 | static sector_t align_to_barrier_unit_end(sector_t start_sector, |
536 | sector_t sectors) | |
537 | { | |
538 | sector_t len; | |
539 | ||
540 | WARN_ON(sectors == 0); | |
541 | /* | |
542 | * len is the number of sectors from start_sector to end of the | |
543 | * barrier unit which start_sector belongs to. | |
544 | */ | |
545 | len = round_up(start_sector + 1, BARRIER_UNIT_SECTOR_SIZE) - | |
546 | start_sector; | |
547 | ||
548 | if (len > sectors) | |
549 | len = sectors; | |
550 | ||
551 | return len; | |
552 | } | |
553 | ||
1da177e4 LT |
554 | /* |
555 | * This routine returns the disk from which the requested read should | |
556 | * be done. There is a per-array 'next expected sequential IO' sector | |
557 | * number - if this matches on the next IO then we use the last disk. | |
558 | * There is also a per-disk 'last know head position' sector that is | |
559 | * maintained from IRQ contexts, both the normal and the resync IO | |
560 | * completion handlers update this position correctly. If there is no | |
561 | * perfect sequential match then we pick the disk whose head is closest. | |
562 | * | |
563 | * If there are 2 mirrors in the same 2 devices, performance degrades | |
564 | * because position is mirror, not device based. | |
565 | * | |
566 | * The rdev for the device selected will have nr_pending incremented. | |
567 | */ | |
e8096360 | 568 | static int read_balance(struct r1conf *conf, struct r1bio *r1_bio, int *max_sectors) |
1da177e4 | 569 | { |
af3a2cd6 | 570 | const sector_t this_sector = r1_bio->sector; |
d2eb35ac N |
571 | int sectors; |
572 | int best_good_sectors; | |
9dedf603 SL |
573 | int best_disk, best_dist_disk, best_pending_disk; |
574 | int has_nonrot_disk; | |
be4d3280 | 575 | int disk; |
76073054 | 576 | sector_t best_dist; |
9dedf603 | 577 | unsigned int min_pending; |
3cb03002 | 578 | struct md_rdev *rdev; |
f3ac8bf7 | 579 | int choose_first; |
12cee5a8 | 580 | int choose_next_idle; |
1da177e4 LT |
581 | |
582 | rcu_read_lock(); | |
583 | /* | |
8ddf9efe | 584 | * Check if we can balance. We can balance on the whole |
1da177e4 LT |
585 | * device if no resync is going on, or below the resync window. |
586 | * We take the first readable disk when above the resync window. | |
587 | */ | |
588 | retry: | |
d2eb35ac | 589 | sectors = r1_bio->sectors; |
76073054 | 590 | best_disk = -1; |
9dedf603 | 591 | best_dist_disk = -1; |
76073054 | 592 | best_dist = MaxSector; |
9dedf603 SL |
593 | best_pending_disk = -1; |
594 | min_pending = UINT_MAX; | |
d2eb35ac | 595 | best_good_sectors = 0; |
9dedf603 | 596 | has_nonrot_disk = 0; |
12cee5a8 | 597 | choose_next_idle = 0; |
2e52d449 | 598 | clear_bit(R1BIO_FailFast, &r1_bio->state); |
d2eb35ac | 599 | |
7d49ffcf GR |
600 | if ((conf->mddev->recovery_cp < this_sector + sectors) || |
601 | (mddev_is_clustered(conf->mddev) && | |
90382ed9 | 602 | md_cluster_ops->area_resyncing(conf->mddev, READ, this_sector, |
7d49ffcf GR |
603 | this_sector + sectors))) |
604 | choose_first = 1; | |
605 | else | |
606 | choose_first = 0; | |
1da177e4 | 607 | |
be4d3280 | 608 | for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) { |
76073054 | 609 | sector_t dist; |
d2eb35ac N |
610 | sector_t first_bad; |
611 | int bad_sectors; | |
9dedf603 | 612 | unsigned int pending; |
12cee5a8 | 613 | bool nonrot; |
d2eb35ac | 614 | |
f3ac8bf7 N |
615 | rdev = rcu_dereference(conf->mirrors[disk].rdev); |
616 | if (r1_bio->bios[disk] == IO_BLOCKED | |
617 | || rdev == NULL | |
76073054 | 618 | || test_bit(Faulty, &rdev->flags)) |
f3ac8bf7 | 619 | continue; |
76073054 N |
620 | if (!test_bit(In_sync, &rdev->flags) && |
621 | rdev->recovery_offset < this_sector + sectors) | |
1da177e4 | 622 | continue; |
76073054 N |
623 | if (test_bit(WriteMostly, &rdev->flags)) { |
624 | /* Don't balance among write-mostly, just | |
625 | * use the first as a last resort */ | |
d1901ef0 | 626 | if (best_dist_disk < 0) { |
307729c8 N |
627 | if (is_badblock(rdev, this_sector, sectors, |
628 | &first_bad, &bad_sectors)) { | |
816b0acf | 629 | if (first_bad <= this_sector) |
307729c8 N |
630 | /* Cannot use this */ |
631 | continue; | |
632 | best_good_sectors = first_bad - this_sector; | |
633 | } else | |
634 | best_good_sectors = sectors; | |
d1901ef0 TH |
635 | best_dist_disk = disk; |
636 | best_pending_disk = disk; | |
307729c8 | 637 | } |
76073054 N |
638 | continue; |
639 | } | |
640 | /* This is a reasonable device to use. It might | |
641 | * even be best. | |
642 | */ | |
d2eb35ac N |
643 | if (is_badblock(rdev, this_sector, sectors, |
644 | &first_bad, &bad_sectors)) { | |
645 | if (best_dist < MaxSector) | |
646 | /* already have a better device */ | |
647 | continue; | |
648 | if (first_bad <= this_sector) { | |
649 | /* cannot read here. If this is the 'primary' | |
650 | * device, then we must not read beyond | |
651 | * bad_sectors from another device.. | |
652 | */ | |
653 | bad_sectors -= (this_sector - first_bad); | |
654 | if (choose_first && sectors > bad_sectors) | |
655 | sectors = bad_sectors; | |
656 | if (best_good_sectors > sectors) | |
657 | best_good_sectors = sectors; | |
658 | ||
659 | } else { | |
660 | sector_t good_sectors = first_bad - this_sector; | |
661 | if (good_sectors > best_good_sectors) { | |
662 | best_good_sectors = good_sectors; | |
663 | best_disk = disk; | |
664 | } | |
665 | if (choose_first) | |
666 | break; | |
667 | } | |
668 | continue; | |
d82dd0e3 TM |
669 | } else { |
670 | if ((sectors > best_good_sectors) && (best_disk >= 0)) | |
671 | best_disk = -1; | |
d2eb35ac | 672 | best_good_sectors = sectors; |
d82dd0e3 | 673 | } |
d2eb35ac | 674 | |
2e52d449 N |
675 | if (best_disk >= 0) |
676 | /* At least two disks to choose from so failfast is OK */ | |
677 | set_bit(R1BIO_FailFast, &r1_bio->state); | |
678 | ||
12cee5a8 SL |
679 | nonrot = blk_queue_nonrot(bdev_get_queue(rdev->bdev)); |
680 | has_nonrot_disk |= nonrot; | |
9dedf603 | 681 | pending = atomic_read(&rdev->nr_pending); |
76073054 | 682 | dist = abs(this_sector - conf->mirrors[disk].head_position); |
12cee5a8 | 683 | if (choose_first) { |
76073054 | 684 | best_disk = disk; |
1da177e4 LT |
685 | break; |
686 | } | |
12cee5a8 SL |
687 | /* Don't change to another disk for sequential reads */ |
688 | if (conf->mirrors[disk].next_seq_sect == this_sector | |
689 | || dist == 0) { | |
690 | int opt_iosize = bdev_io_opt(rdev->bdev) >> 9; | |
691 | struct raid1_info *mirror = &conf->mirrors[disk]; | |
692 | ||
693 | best_disk = disk; | |
694 | /* | |
695 | * If buffered sequential IO size exceeds optimal | |
696 | * iosize, check if there is idle disk. If yes, choose | |
697 | * the idle disk. read_balance could already choose an | |
698 | * idle disk before noticing it's a sequential IO in | |
699 | * this disk. This doesn't matter because this disk | |
700 | * will idle, next time it will be utilized after the | |
701 | * first disk has IO size exceeds optimal iosize. In | |
702 | * this way, iosize of the first disk will be optimal | |
703 | * iosize at least. iosize of the second disk might be | |
704 | * small, but not a big deal since when the second disk | |
705 | * starts IO, the first disk is likely still busy. | |
706 | */ | |
707 | if (nonrot && opt_iosize > 0 && | |
708 | mirror->seq_start != MaxSector && | |
709 | mirror->next_seq_sect > opt_iosize && | |
710 | mirror->next_seq_sect - opt_iosize >= | |
711 | mirror->seq_start) { | |
712 | choose_next_idle = 1; | |
713 | continue; | |
714 | } | |
715 | break; | |
716 | } | |
12cee5a8 SL |
717 | |
718 | if (choose_next_idle) | |
719 | continue; | |
9dedf603 SL |
720 | |
721 | if (min_pending > pending) { | |
722 | min_pending = pending; | |
723 | best_pending_disk = disk; | |
724 | } | |
725 | ||
76073054 N |
726 | if (dist < best_dist) { |
727 | best_dist = dist; | |
9dedf603 | 728 | best_dist_disk = disk; |
1da177e4 | 729 | } |
f3ac8bf7 | 730 | } |
1da177e4 | 731 | |
9dedf603 SL |
732 | /* |
733 | * If all disks are rotational, choose the closest disk. If any disk is | |
734 | * non-rotational, choose the disk with less pending request even the | |
735 | * disk is rotational, which might/might not be optimal for raids with | |
736 | * mixed ratation/non-rotational disks depending on workload. | |
737 | */ | |
738 | if (best_disk == -1) { | |
2e52d449 | 739 | if (has_nonrot_disk || min_pending == 0) |
9dedf603 SL |
740 | best_disk = best_pending_disk; |
741 | else | |
742 | best_disk = best_dist_disk; | |
743 | } | |
744 | ||
76073054 N |
745 | if (best_disk >= 0) { |
746 | rdev = rcu_dereference(conf->mirrors[best_disk].rdev); | |
8ddf9efe N |
747 | if (!rdev) |
748 | goto retry; | |
749 | atomic_inc(&rdev->nr_pending); | |
d2eb35ac | 750 | sectors = best_good_sectors; |
12cee5a8 SL |
751 | |
752 | if (conf->mirrors[best_disk].next_seq_sect != this_sector) | |
753 | conf->mirrors[best_disk].seq_start = this_sector; | |
754 | ||
be4d3280 | 755 | conf->mirrors[best_disk].next_seq_sect = this_sector + sectors; |
1da177e4 LT |
756 | } |
757 | rcu_read_unlock(); | |
d2eb35ac | 758 | *max_sectors = sectors; |
1da177e4 | 759 | |
76073054 | 760 | return best_disk; |
1da177e4 LT |
761 | } |
762 | ||
5c675f83 | 763 | static int raid1_congested(struct mddev *mddev, int bits) |
0d129228 | 764 | { |
e8096360 | 765 | struct r1conf *conf = mddev->private; |
0d129228 N |
766 | int i, ret = 0; |
767 | ||
4452226e | 768 | if ((bits & (1 << WB_async_congested)) && |
34db0cd6 N |
769 | conf->pending_count >= max_queued_requests) |
770 | return 1; | |
771 | ||
0d129228 | 772 | rcu_read_lock(); |
f53e29fc | 773 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3cb03002 | 774 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
0d129228 | 775 | if (rdev && !test_bit(Faulty, &rdev->flags)) { |
165125e1 | 776 | struct request_queue *q = bdev_get_queue(rdev->bdev); |
0d129228 | 777 | |
1ed7242e JB |
778 | BUG_ON(!q); |
779 | ||
0d129228 N |
780 | /* Note the '|| 1' - when read_balance prefers |
781 | * non-congested targets, it can be removed | |
782 | */ | |
4452226e | 783 | if ((bits & (1 << WB_async_congested)) || 1) |
dc3b17cc | 784 | ret |= bdi_congested(q->backing_dev_info, bits); |
0d129228 | 785 | else |
dc3b17cc | 786 | ret &= bdi_congested(q->backing_dev_info, bits); |
0d129228 N |
787 | } |
788 | } | |
789 | rcu_read_unlock(); | |
790 | return ret; | |
791 | } | |
0d129228 | 792 | |
673ca68d N |
793 | static void flush_bio_list(struct r1conf *conf, struct bio *bio) |
794 | { | |
795 | /* flush any pending bitmap writes to disk before proceeding w/ I/O */ | |
796 | bitmap_unplug(conf->mddev->bitmap); | |
797 | wake_up(&conf->wait_barrier); | |
798 | ||
799 | while (bio) { /* submit pending writes */ | |
800 | struct bio *next = bio->bi_next; | |
801 | struct md_rdev *rdev = (void*)bio->bi_bdev; | |
802 | bio->bi_next = NULL; | |
803 | bio->bi_bdev = rdev->bdev; | |
804 | if (test_bit(Faulty, &rdev->flags)) { | |
805 | bio->bi_error = -EIO; | |
806 | bio_endio(bio); | |
807 | } else if (unlikely((bio_op(bio) == REQ_OP_DISCARD) && | |
808 | !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) | |
809 | /* Just ignore it */ | |
810 | bio_endio(bio); | |
811 | else | |
812 | generic_make_request(bio); | |
813 | bio = next; | |
814 | } | |
815 | } | |
816 | ||
e8096360 | 817 | static void flush_pending_writes(struct r1conf *conf) |
a35e63ef N |
818 | { |
819 | /* Any writes that have been queued but are awaiting | |
820 | * bitmap updates get flushed here. | |
a35e63ef | 821 | */ |
a35e63ef N |
822 | spin_lock_irq(&conf->device_lock); |
823 | ||
824 | if (conf->pending_bio_list.head) { | |
825 | struct bio *bio; | |
826 | bio = bio_list_get(&conf->pending_bio_list); | |
34db0cd6 | 827 | conf->pending_count = 0; |
a35e63ef | 828 | spin_unlock_irq(&conf->device_lock); |
673ca68d | 829 | flush_bio_list(conf, bio); |
a35e63ef N |
830 | } else |
831 | spin_unlock_irq(&conf->device_lock); | |
7eaceacc JA |
832 | } |
833 | ||
17999be4 N |
834 | /* Barriers.... |
835 | * Sometimes we need to suspend IO while we do something else, | |
836 | * either some resync/recovery, or reconfigure the array. | |
837 | * To do this we raise a 'barrier'. | |
838 | * The 'barrier' is a counter that can be raised multiple times | |
839 | * to count how many activities are happening which preclude | |
840 | * normal IO. | |
841 | * We can only raise the barrier if there is no pending IO. | |
842 | * i.e. if nr_pending == 0. | |
843 | * We choose only to raise the barrier if no-one is waiting for the | |
844 | * barrier to go down. This means that as soon as an IO request | |
845 | * is ready, no other operations which require a barrier will start | |
846 | * until the IO request has had a chance. | |
847 | * | |
848 | * So: regular IO calls 'wait_barrier'. When that returns there | |
849 | * is no backgroup IO happening, It must arrange to call | |
850 | * allow_barrier when it has finished its IO. | |
851 | * backgroup IO calls must call raise_barrier. Once that returns | |
852 | * there is no normal IO happeing. It must arrange to call | |
853 | * lower_barrier when the particular background IO completes. | |
1da177e4 | 854 | */ |
c2fd4c94 | 855 | static void raise_barrier(struct r1conf *conf, sector_t sector_nr) |
1da177e4 | 856 | { |
fd76863e | 857 | int idx = sector_to_idx(sector_nr); |
858 | ||
1da177e4 | 859 | spin_lock_irq(&conf->resync_lock); |
17999be4 N |
860 | |
861 | /* Wait until no block IO is waiting */ | |
824e47da | 862 | wait_event_lock_irq(conf->wait_barrier, |
863 | !atomic_read(&conf->nr_waiting[idx]), | |
eed8c02e | 864 | conf->resync_lock); |
17999be4 N |
865 | |
866 | /* block any new IO from starting */ | |
824e47da | 867 | atomic_inc(&conf->barrier[idx]); |
868 | /* | |
869 | * In raise_barrier() we firstly increase conf->barrier[idx] then | |
870 | * check conf->nr_pending[idx]. In _wait_barrier() we firstly | |
871 | * increase conf->nr_pending[idx] then check conf->barrier[idx]. | |
872 | * A memory barrier here to make sure conf->nr_pending[idx] won't | |
873 | * be fetched before conf->barrier[idx] is increased. Otherwise | |
874 | * there will be a race between raise_barrier() and _wait_barrier(). | |
875 | */ | |
876 | smp_mb__after_atomic(); | |
17999be4 | 877 | |
79ef3a8a | 878 | /* For these conditions we must wait: |
879 | * A: while the array is in frozen state | |
fd76863e | 880 | * B: while conf->nr_pending[idx] is not 0, meaning regular I/O |
881 | * existing in corresponding I/O barrier bucket. | |
882 | * C: while conf->barrier[idx] >= RESYNC_DEPTH, meaning reaches | |
883 | * max resync count which allowed on current I/O barrier bucket. | |
79ef3a8a | 884 | */ |
17999be4 | 885 | wait_event_lock_irq(conf->wait_barrier, |
b364e3d0 | 886 | !conf->array_frozen && |
824e47da | 887 | !atomic_read(&conf->nr_pending[idx]) && |
888 | atomic_read(&conf->barrier[idx]) < RESYNC_DEPTH, | |
eed8c02e | 889 | conf->resync_lock); |
17999be4 | 890 | |
43ac9b84 | 891 | atomic_inc(&conf->nr_sync_pending); |
17999be4 N |
892 | spin_unlock_irq(&conf->resync_lock); |
893 | } | |
894 | ||
fd76863e | 895 | static void lower_barrier(struct r1conf *conf, sector_t sector_nr) |
17999be4 | 896 | { |
fd76863e | 897 | int idx = sector_to_idx(sector_nr); |
898 | ||
824e47da | 899 | BUG_ON(atomic_read(&conf->barrier[idx]) <= 0); |
fd76863e | 900 | |
824e47da | 901 | atomic_dec(&conf->barrier[idx]); |
43ac9b84 | 902 | atomic_dec(&conf->nr_sync_pending); |
17999be4 N |
903 | wake_up(&conf->wait_barrier); |
904 | } | |
905 | ||
fd76863e | 906 | static void _wait_barrier(struct r1conf *conf, int idx) |
17999be4 | 907 | { |
824e47da | 908 | /* |
909 | * We need to increase conf->nr_pending[idx] very early here, | |
910 | * then raise_barrier() can be blocked when it waits for | |
911 | * conf->nr_pending[idx] to be 0. Then we can avoid holding | |
912 | * conf->resync_lock when there is no barrier raised in same | |
913 | * barrier unit bucket. Also if the array is frozen, I/O | |
914 | * should be blocked until array is unfrozen. | |
915 | */ | |
916 | atomic_inc(&conf->nr_pending[idx]); | |
917 | /* | |
918 | * In _wait_barrier() we firstly increase conf->nr_pending[idx], then | |
919 | * check conf->barrier[idx]. In raise_barrier() we firstly increase | |
920 | * conf->barrier[idx], then check conf->nr_pending[idx]. A memory | |
921 | * barrier is necessary here to make sure conf->barrier[idx] won't be | |
922 | * fetched before conf->nr_pending[idx] is increased. Otherwise there | |
923 | * will be a race between _wait_barrier() and raise_barrier(). | |
924 | */ | |
925 | smp_mb__after_atomic(); | |
79ef3a8a | 926 | |
824e47da | 927 | /* |
928 | * Don't worry about checking two atomic_t variables at same time | |
929 | * here. If during we check conf->barrier[idx], the array is | |
930 | * frozen (conf->array_frozen is 1), and chonf->barrier[idx] is | |
931 | * 0, it is safe to return and make the I/O continue. Because the | |
932 | * array is frozen, all I/O returned here will eventually complete | |
933 | * or be queued, no race will happen. See code comment in | |
934 | * frozen_array(). | |
935 | */ | |
936 | if (!READ_ONCE(conf->array_frozen) && | |
937 | !atomic_read(&conf->barrier[idx])) | |
938 | return; | |
79ef3a8a | 939 | |
824e47da | 940 | /* |
941 | * After holding conf->resync_lock, conf->nr_pending[idx] | |
942 | * should be decreased before waiting for barrier to drop. | |
943 | * Otherwise, we may encounter a race condition because | |
944 | * raise_barrer() might be waiting for conf->nr_pending[idx] | |
945 | * to be 0 at same time. | |
946 | */ | |
947 | spin_lock_irq(&conf->resync_lock); | |
948 | atomic_inc(&conf->nr_waiting[idx]); | |
949 | atomic_dec(&conf->nr_pending[idx]); | |
950 | /* | |
951 | * In case freeze_array() is waiting for | |
952 | * get_unqueued_pending() == extra | |
953 | */ | |
954 | wake_up(&conf->wait_barrier); | |
955 | /* Wait for the barrier in same barrier unit bucket to drop. */ | |
956 | wait_event_lock_irq(conf->wait_barrier, | |
957 | !conf->array_frozen && | |
958 | !atomic_read(&conf->barrier[idx]), | |
959 | conf->resync_lock); | |
960 | atomic_inc(&conf->nr_pending[idx]); | |
961 | atomic_dec(&conf->nr_waiting[idx]); | |
fd76863e | 962 | spin_unlock_irq(&conf->resync_lock); |
79ef3a8a | 963 | } |
964 | ||
fd76863e | 965 | static void wait_read_barrier(struct r1conf *conf, sector_t sector_nr) |
79ef3a8a | 966 | { |
fd76863e | 967 | int idx = sector_to_idx(sector_nr); |
79ef3a8a | 968 | |
824e47da | 969 | /* |
970 | * Very similar to _wait_barrier(). The difference is, for read | |
971 | * I/O we don't need wait for sync I/O, but if the whole array | |
972 | * is frozen, the read I/O still has to wait until the array is | |
973 | * unfrozen. Since there is no ordering requirement with | |
974 | * conf->barrier[idx] here, memory barrier is unnecessary as well. | |
975 | */ | |
976 | atomic_inc(&conf->nr_pending[idx]); | |
79ef3a8a | 977 | |
824e47da | 978 | if (!READ_ONCE(conf->array_frozen)) |
979 | return; | |
980 | ||
981 | spin_lock_irq(&conf->resync_lock); | |
982 | atomic_inc(&conf->nr_waiting[idx]); | |
983 | atomic_dec(&conf->nr_pending[idx]); | |
984 | /* | |
985 | * In case freeze_array() is waiting for | |
986 | * get_unqueued_pending() == extra | |
987 | */ | |
988 | wake_up(&conf->wait_barrier); | |
989 | /* Wait for array to be unfrozen */ | |
990 | wait_event_lock_irq(conf->wait_barrier, | |
991 | !conf->array_frozen, | |
992 | conf->resync_lock); | |
993 | atomic_inc(&conf->nr_pending[idx]); | |
994 | atomic_dec(&conf->nr_waiting[idx]); | |
1da177e4 LT |
995 | spin_unlock_irq(&conf->resync_lock); |
996 | } | |
997 | ||
fd76863e | 998 | static void wait_barrier(struct r1conf *conf, sector_t sector_nr) |
17999be4 | 999 | { |
fd76863e | 1000 | int idx = sector_to_idx(sector_nr); |
79ef3a8a | 1001 | |
fd76863e | 1002 | _wait_barrier(conf, idx); |
1003 | } | |
1004 | ||
1005 | static void wait_all_barriers(struct r1conf *conf) | |
1006 | { | |
1007 | int idx; | |
1008 | ||
1009 | for (idx = 0; idx < BARRIER_BUCKETS_NR; idx++) | |
1010 | _wait_barrier(conf, idx); | |
1011 | } | |
1012 | ||
1013 | static void _allow_barrier(struct r1conf *conf, int idx) | |
17999be4 | 1014 | { |
824e47da | 1015 | atomic_dec(&conf->nr_pending[idx]); |
17999be4 N |
1016 | wake_up(&conf->wait_barrier); |
1017 | } | |
1018 | ||
fd76863e | 1019 | static void allow_barrier(struct r1conf *conf, sector_t sector_nr) |
1020 | { | |
1021 | int idx = sector_to_idx(sector_nr); | |
1022 | ||
1023 | _allow_barrier(conf, idx); | |
1024 | } | |
1025 | ||
1026 | static void allow_all_barriers(struct r1conf *conf) | |
1027 | { | |
1028 | int idx; | |
1029 | ||
1030 | for (idx = 0; idx < BARRIER_BUCKETS_NR; idx++) | |
1031 | _allow_barrier(conf, idx); | |
1032 | } | |
1033 | ||
1034 | /* conf->resync_lock should be held */ | |
1035 | static int get_unqueued_pending(struct r1conf *conf) | |
1036 | { | |
1037 | int idx, ret; | |
1038 | ||
43ac9b84 XN |
1039 | ret = atomic_read(&conf->nr_sync_pending); |
1040 | for (idx = 0; idx < BARRIER_BUCKETS_NR; idx++) | |
824e47da | 1041 | ret += atomic_read(&conf->nr_pending[idx]) - |
1042 | atomic_read(&conf->nr_queued[idx]); | |
fd76863e | 1043 | |
1044 | return ret; | |
1045 | } | |
1046 | ||
e2d59925 | 1047 | static void freeze_array(struct r1conf *conf, int extra) |
ddaf22ab | 1048 | { |
fd76863e | 1049 | /* Stop sync I/O and normal I/O and wait for everything to |
11353b9d | 1050 | * go quiet. |
fd76863e | 1051 | * This is called in two situations: |
1052 | * 1) management command handlers (reshape, remove disk, quiesce). | |
1053 | * 2) one normal I/O request failed. | |
1054 | ||
1055 | * After array_frozen is set to 1, new sync IO will be blocked at | |
1056 | * raise_barrier(), and new normal I/O will blocked at _wait_barrier() | |
1057 | * or wait_read_barrier(). The flying I/Os will either complete or be | |
1058 | * queued. When everything goes quite, there are only queued I/Os left. | |
1059 | ||
1060 | * Every flying I/O contributes to a conf->nr_pending[idx], idx is the | |
1061 | * barrier bucket index which this I/O request hits. When all sync and | |
1062 | * normal I/O are queued, sum of all conf->nr_pending[] will match sum | |
1063 | * of all conf->nr_queued[]. But normal I/O failure is an exception, | |
1064 | * in handle_read_error(), we may call freeze_array() before trying to | |
1065 | * fix the read error. In this case, the error read I/O is not queued, | |
1066 | * so get_unqueued_pending() == 1. | |
1067 | * | |
1068 | * Therefore before this function returns, we need to wait until | |
1069 | * get_unqueued_pendings(conf) gets equal to extra. For | |
1070 | * normal I/O context, extra is 1, in rested situations extra is 0. | |
ddaf22ab N |
1071 | */ |
1072 | spin_lock_irq(&conf->resync_lock); | |
b364e3d0 | 1073 | conf->array_frozen = 1; |
578b54ad | 1074 | raid1_log(conf->mddev, "wait freeze"); |
fd76863e | 1075 | wait_event_lock_irq_cmd( |
1076 | conf->wait_barrier, | |
1077 | get_unqueued_pending(conf) == extra, | |
1078 | conf->resync_lock, | |
1079 | flush_pending_writes(conf)); | |
ddaf22ab N |
1080 | spin_unlock_irq(&conf->resync_lock); |
1081 | } | |
e8096360 | 1082 | static void unfreeze_array(struct r1conf *conf) |
ddaf22ab N |
1083 | { |
1084 | /* reverse the effect of the freeze */ | |
1085 | spin_lock_irq(&conf->resync_lock); | |
b364e3d0 | 1086 | conf->array_frozen = 0; |
ddaf22ab | 1087 | spin_unlock_irq(&conf->resync_lock); |
824e47da | 1088 | wake_up(&conf->wait_barrier); |
ddaf22ab N |
1089 | } |
1090 | ||
841c1316 | 1091 | static struct bio *alloc_behind_master_bio(struct r1bio *r1_bio, |
cb83efcf | 1092 | struct bio *bio) |
4b6d287f | 1093 | { |
cb83efcf | 1094 | int size = bio->bi_iter.bi_size; |
841c1316 ML |
1095 | unsigned vcnt = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; |
1096 | int i = 0; | |
1097 | struct bio *behind_bio = NULL; | |
1098 | ||
1099 | behind_bio = bio_alloc_mddev(GFP_NOIO, vcnt, r1_bio->mddev); | |
1100 | if (!behind_bio) | |
1101 | goto fail; | |
4b6d287f | 1102 | |
41743c1f SL |
1103 | /* discard op, we don't support writezero/writesame yet */ |
1104 | if (!bio_has_data(bio)) | |
1105 | goto skip_copy; | |
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: |
841c1316 | 1123 | r1_bio->behind_master_bio = behind_bio;; |
af6d7b76 | 1124 | set_bit(R1BIO_BehindIO, &r1_bio->state); |
4b6d287f | 1125 | |
841c1316 ML |
1126 | return behind_bio; |
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 ML |
1131 | bio_free_pages(behind_bio); |
1132 | fail: | |
1133 | return behind_bio; | |
4b6d287f N |
1134 | } |
1135 | ||
f54a9d0e N |
1136 | struct raid1_plug_cb { |
1137 | struct blk_plug_cb cb; | |
1138 | struct bio_list pending; | |
1139 | int pending_cnt; | |
1140 | }; | |
1141 | ||
1142 | static void raid1_unplug(struct blk_plug_cb *cb, bool from_schedule) | |
1143 | { | |
1144 | struct raid1_plug_cb *plug = container_of(cb, struct raid1_plug_cb, | |
1145 | cb); | |
1146 | struct mddev *mddev = plug->cb.data; | |
1147 | struct r1conf *conf = mddev->private; | |
1148 | struct bio *bio; | |
1149 | ||
874807a8 | 1150 | if (from_schedule || current->bio_list) { |
f54a9d0e N |
1151 | spin_lock_irq(&conf->device_lock); |
1152 | bio_list_merge(&conf->pending_bio_list, &plug->pending); | |
1153 | conf->pending_count += plug->pending_cnt; | |
1154 | spin_unlock_irq(&conf->device_lock); | |
ee0b0244 | 1155 | wake_up(&conf->wait_barrier); |
f54a9d0e N |
1156 | md_wakeup_thread(mddev->thread); |
1157 | kfree(plug); | |
1158 | return; | |
1159 | } | |
1160 | ||
1161 | /* we aren't scheduling, so we can do the write-out directly. */ | |
1162 | bio = bio_list_get(&plug->pending); | |
673ca68d | 1163 | flush_bio_list(conf, bio); |
f54a9d0e N |
1164 | kfree(plug); |
1165 | } | |
1166 | ||
689389a0 N |
1167 | static void init_r1bio(struct r1bio *r1_bio, struct mddev *mddev, struct bio *bio) |
1168 | { | |
1169 | r1_bio->master_bio = bio; | |
1170 | r1_bio->sectors = bio_sectors(bio); | |
1171 | r1_bio->state = 0; | |
1172 | r1_bio->mddev = mddev; | |
1173 | r1_bio->sector = bio->bi_iter.bi_sector; | |
1174 | } | |
1175 | ||
fd76863e | 1176 | static inline struct r1bio * |
689389a0 | 1177 | alloc_r1bio(struct mddev *mddev, struct bio *bio) |
fd76863e | 1178 | { |
1179 | struct r1conf *conf = mddev->private; | |
1180 | struct r1bio *r1_bio; | |
1181 | ||
1182 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
689389a0 N |
1183 | /* Ensure no bio records IO_BLOCKED */ |
1184 | memset(r1_bio->bios, 0, conf->raid_disks * sizeof(r1_bio->bios[0])); | |
1185 | init_r1bio(r1_bio, mddev, bio); | |
fd76863e | 1186 | return r1_bio; |
1187 | } | |
1188 | ||
c230e7e5 | 1189 | static void raid1_read_request(struct mddev *mddev, struct bio *bio, |
689389a0 | 1190 | int max_read_sectors, struct r1bio *r1_bio) |
1da177e4 | 1191 | { |
e8096360 | 1192 | struct r1conf *conf = mddev->private; |
0eaf822c | 1193 | struct raid1_info *mirror; |
1da177e4 | 1194 | struct bio *read_bio; |
3b046a97 RL |
1195 | struct bitmap *bitmap = mddev->bitmap; |
1196 | const int op = bio_op(bio); | |
1197 | const unsigned long do_sync = (bio->bi_opf & REQ_SYNC); | |
3b046a97 RL |
1198 | int max_sectors; |
1199 | int rdisk; | |
689389a0 N |
1200 | bool print_msg = !!r1_bio; |
1201 | char b[BDEVNAME_SIZE]; | |
3b046a97 | 1202 | |
fd76863e | 1203 | /* |
689389a0 N |
1204 | * If r1_bio is set, we are blocking the raid1d thread |
1205 | * so there is a tiny risk of deadlock. So ask for | |
1206 | * emergency memory if needed. | |
fd76863e | 1207 | */ |
689389a0 | 1208 | gfp_t gfp = r1_bio ? (GFP_NOIO | __GFP_HIGH) : GFP_NOIO; |
fd76863e | 1209 | |
689389a0 N |
1210 | if (print_msg) { |
1211 | /* Need to get the block device name carefully */ | |
1212 | struct md_rdev *rdev; | |
1213 | rcu_read_lock(); | |
1214 | rdev = rcu_dereference(conf->mirrors[r1_bio->read_disk].rdev); | |
1215 | if (rdev) | |
1216 | bdevname(rdev->bdev, b); | |
1217 | else | |
1218 | strcpy(b, "???"); | |
1219 | rcu_read_unlock(); | |
1220 | } | |
3b046a97 | 1221 | |
fd76863e | 1222 | /* |
fd76863e | 1223 | * Still need barrier for READ in case that whole |
1224 | * array is frozen. | |
fd76863e | 1225 | */ |
fd76863e | 1226 | wait_read_barrier(conf, bio->bi_iter.bi_sector); |
1227 | ||
689389a0 N |
1228 | if (!r1_bio) |
1229 | r1_bio = alloc_r1bio(mddev, bio); | |
1230 | else | |
1231 | init_r1bio(r1_bio, mddev, bio); | |
c230e7e5 | 1232 | r1_bio->sectors = max_read_sectors; |
fd76863e | 1233 | |
1234 | /* | |
1235 | * make_request() can abort the operation when read-ahead is being | |
1236 | * used and no empty request is available. | |
1237 | */ | |
3b046a97 RL |
1238 | rdisk = read_balance(conf, r1_bio, &max_sectors); |
1239 | ||
1240 | if (rdisk < 0) { | |
1241 | /* couldn't find anywhere to read from */ | |
689389a0 N |
1242 | if (print_msg) { |
1243 | pr_crit_ratelimited("md/raid1:%s: %s: unrecoverable I/O read error for block %llu\n", | |
1244 | mdname(mddev), | |
1245 | b, | |
1246 | (unsigned long long)r1_bio->sector); | |
1247 | } | |
3b046a97 RL |
1248 | raid_end_bio_io(r1_bio); |
1249 | return; | |
1250 | } | |
1251 | mirror = conf->mirrors + rdisk; | |
1252 | ||
689389a0 N |
1253 | if (print_msg) |
1254 | pr_info_ratelimited("md/raid1:%s: redirecting sector %llu to other mirror: %s\n", | |
1255 | mdname(mddev), | |
1256 | (unsigned long long)r1_bio->sector, | |
1257 | bdevname(mirror->rdev->bdev, b)); | |
1258 | ||
3b046a97 RL |
1259 | if (test_bit(WriteMostly, &mirror->rdev->flags) && |
1260 | bitmap) { | |
1261 | /* | |
1262 | * Reading from a write-mostly device must take care not to | |
1263 | * over-take any writes that are 'behind' | |
1264 | */ | |
1265 | raid1_log(mddev, "wait behind writes"); | |
1266 | wait_event(bitmap->behind_wait, | |
1267 | atomic_read(&bitmap->behind_writes) == 0); | |
1268 | } | |
c230e7e5 N |
1269 | |
1270 | if (max_sectors < bio_sectors(bio)) { | |
1271 | struct bio *split = bio_split(bio, max_sectors, | |
689389a0 | 1272 | gfp, conf->bio_split); |
c230e7e5 N |
1273 | bio_chain(split, bio); |
1274 | generic_make_request(bio); | |
1275 | bio = split; | |
1276 | r1_bio->master_bio = bio; | |
1277 | r1_bio->sectors = max_sectors; | |
1278 | } | |
1279 | ||
3b046a97 | 1280 | r1_bio->read_disk = rdisk; |
3b046a97 | 1281 | |
689389a0 | 1282 | read_bio = bio_clone_fast(bio, gfp, mddev->bio_set); |
3b046a97 RL |
1283 | |
1284 | r1_bio->bios[rdisk] = read_bio; | |
1285 | ||
1286 | read_bio->bi_iter.bi_sector = r1_bio->sector + | |
1287 | mirror->rdev->data_offset; | |
1288 | read_bio->bi_bdev = mirror->rdev->bdev; | |
1289 | read_bio->bi_end_io = raid1_end_read_request; | |
1290 | bio_set_op_attrs(read_bio, op, do_sync); | |
1291 | if (test_bit(FailFast, &mirror->rdev->flags) && | |
1292 | test_bit(R1BIO_FailFast, &r1_bio->state)) | |
1293 | read_bio->bi_opf |= MD_FAILFAST; | |
1294 | read_bio->bi_private = r1_bio; | |
1295 | ||
1296 | if (mddev->gendisk) | |
1297 | trace_block_bio_remap(bdev_get_queue(read_bio->bi_bdev), | |
1298 | read_bio, disk_devt(mddev->gendisk), | |
1299 | r1_bio->sector); | |
1300 | ||
c230e7e5 | 1301 | generic_make_request(read_bio); |
3b046a97 RL |
1302 | } |
1303 | ||
c230e7e5 N |
1304 | static void raid1_write_request(struct mddev *mddev, struct bio *bio, |
1305 | int max_write_sectors) | |
3b046a97 RL |
1306 | { |
1307 | struct r1conf *conf = mddev->private; | |
fd76863e | 1308 | struct r1bio *r1_bio; |
1f68f0c4 | 1309 | int i, disks; |
3b046a97 | 1310 | struct bitmap *bitmap = mddev->bitmap; |
191ea9b2 | 1311 | unsigned long flags; |
3cb03002 | 1312 | struct md_rdev *blocked_rdev; |
f54a9d0e N |
1313 | struct blk_plug_cb *cb; |
1314 | struct raid1_plug_cb *plug = NULL; | |
1f68f0c4 | 1315 | int first_clone; |
1f68f0c4 | 1316 | int max_sectors; |
191ea9b2 | 1317 | |
1da177e4 LT |
1318 | /* |
1319 | * Register the new request and wait if the reconstruction | |
1320 | * thread has put up a bar for new requests. | |
1321 | * Continue immediately if no resync is active currently. | |
1322 | */ | |
62de608d | 1323 | |
3d310eb7 N |
1324 | md_write_start(mddev, bio); /* wait on superblock update early */ |
1325 | ||
3b046a97 | 1326 | if ((bio_end_sector(bio) > mddev->suspend_lo && |
589a1c49 GR |
1327 | bio->bi_iter.bi_sector < mddev->suspend_hi) || |
1328 | (mddev_is_clustered(mddev) && | |
90382ed9 | 1329 | md_cluster_ops->area_resyncing(mddev, WRITE, |
3b046a97 RL |
1330 | bio->bi_iter.bi_sector, bio_end_sector(bio)))) { |
1331 | ||
1332 | /* | |
1333 | * As the suspend_* range is controlled by userspace, we want | |
1334 | * an interruptible wait. | |
6eef4b21 N |
1335 | */ |
1336 | DEFINE_WAIT(w); | |
1337 | for (;;) { | |
1338 | flush_signals(current); | |
1339 | prepare_to_wait(&conf->wait_barrier, | |
1340 | &w, TASK_INTERRUPTIBLE); | |
f73a1c7d | 1341 | if (bio_end_sector(bio) <= mddev->suspend_lo || |
589a1c49 GR |
1342 | bio->bi_iter.bi_sector >= mddev->suspend_hi || |
1343 | (mddev_is_clustered(mddev) && | |
90382ed9 | 1344 | !md_cluster_ops->area_resyncing(mddev, WRITE, |
3b046a97 RL |
1345 | bio->bi_iter.bi_sector, |
1346 | bio_end_sector(bio)))) | |
6eef4b21 N |
1347 | break; |
1348 | schedule(); | |
1349 | } | |
1350 | finish_wait(&conf->wait_barrier, &w); | |
1351 | } | |
fd76863e | 1352 | wait_barrier(conf, bio->bi_iter.bi_sector); |
1353 | ||
689389a0 | 1354 | r1_bio = alloc_r1bio(mddev, bio); |
c230e7e5 | 1355 | r1_bio->sectors = max_write_sectors; |
1da177e4 | 1356 | |
34db0cd6 N |
1357 | if (conf->pending_count >= max_queued_requests) { |
1358 | md_wakeup_thread(mddev->thread); | |
578b54ad | 1359 | raid1_log(mddev, "wait queued"); |
34db0cd6 N |
1360 | wait_event(conf->wait_barrier, |
1361 | conf->pending_count < max_queued_requests); | |
1362 | } | |
1f68f0c4 | 1363 | /* first select target devices under rcu_lock and |
1da177e4 LT |
1364 | * inc refcount on their rdev. Record them by setting |
1365 | * bios[x] to bio | |
1f68f0c4 N |
1366 | * If there are known/acknowledged bad blocks on any device on |
1367 | * which we have seen a write error, we want to avoid writing those | |
1368 | * blocks. | |
1369 | * This potentially requires several writes to write around | |
1370 | * the bad blocks. Each set of writes gets it's own r1bio | |
1371 | * with a set of bios attached. | |
1da177e4 | 1372 | */ |
c3b328ac | 1373 | |
8f19ccb2 | 1374 | disks = conf->raid_disks * 2; |
6bfe0b49 DW |
1375 | retry_write: |
1376 | blocked_rdev = NULL; | |
1da177e4 | 1377 | rcu_read_lock(); |
1f68f0c4 | 1378 | max_sectors = r1_bio->sectors; |
1da177e4 | 1379 | for (i = 0; i < disks; i++) { |
3cb03002 | 1380 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
6bfe0b49 DW |
1381 | if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { |
1382 | atomic_inc(&rdev->nr_pending); | |
1383 | blocked_rdev = rdev; | |
1384 | break; | |
1385 | } | |
1f68f0c4 | 1386 | r1_bio->bios[i] = NULL; |
8ae12666 | 1387 | if (!rdev || test_bit(Faulty, &rdev->flags)) { |
8f19ccb2 N |
1388 | if (i < conf->raid_disks) |
1389 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
1f68f0c4 N |
1390 | continue; |
1391 | } | |
1392 | ||
1393 | atomic_inc(&rdev->nr_pending); | |
1394 | if (test_bit(WriteErrorSeen, &rdev->flags)) { | |
1395 | sector_t first_bad; | |
1396 | int bad_sectors; | |
1397 | int is_bad; | |
1398 | ||
3b046a97 | 1399 | is_bad = is_badblock(rdev, r1_bio->sector, max_sectors, |
1f68f0c4 N |
1400 | &first_bad, &bad_sectors); |
1401 | if (is_bad < 0) { | |
1402 | /* mustn't write here until the bad block is | |
1403 | * acknowledged*/ | |
1404 | set_bit(BlockedBadBlocks, &rdev->flags); | |
1405 | blocked_rdev = rdev; | |
1406 | break; | |
1407 | } | |
1408 | if (is_bad && first_bad <= r1_bio->sector) { | |
1409 | /* Cannot write here at all */ | |
1410 | bad_sectors -= (r1_bio->sector - first_bad); | |
1411 | if (bad_sectors < max_sectors) | |
1412 | /* mustn't write more than bad_sectors | |
1413 | * to other devices yet | |
1414 | */ | |
1415 | max_sectors = bad_sectors; | |
03c902e1 | 1416 | rdev_dec_pending(rdev, mddev); |
1f68f0c4 N |
1417 | /* We don't set R1BIO_Degraded as that |
1418 | * only applies if the disk is | |
1419 | * missing, so it might be re-added, | |
1420 | * and we want to know to recover this | |
1421 | * chunk. | |
1422 | * In this case the device is here, | |
1423 | * and the fact that this chunk is not | |
1424 | * in-sync is recorded in the bad | |
1425 | * block log | |
1426 | */ | |
1427 | continue; | |
964147d5 | 1428 | } |
1f68f0c4 N |
1429 | if (is_bad) { |
1430 | int good_sectors = first_bad - r1_bio->sector; | |
1431 | if (good_sectors < max_sectors) | |
1432 | max_sectors = good_sectors; | |
1433 | } | |
1434 | } | |
1435 | r1_bio->bios[i] = bio; | |
1da177e4 LT |
1436 | } |
1437 | rcu_read_unlock(); | |
1438 | ||
6bfe0b49 DW |
1439 | if (unlikely(blocked_rdev)) { |
1440 | /* Wait for this device to become unblocked */ | |
1441 | int j; | |
1442 | ||
1443 | for (j = 0; j < i; j++) | |
1444 | if (r1_bio->bios[j]) | |
1445 | rdev_dec_pending(conf->mirrors[j].rdev, mddev); | |
1f68f0c4 | 1446 | r1_bio->state = 0; |
fd76863e | 1447 | allow_barrier(conf, bio->bi_iter.bi_sector); |
578b54ad | 1448 | raid1_log(mddev, "wait rdev %d blocked", blocked_rdev->raid_disk); |
6bfe0b49 | 1449 | md_wait_for_blocked_rdev(blocked_rdev, mddev); |
fd76863e | 1450 | wait_barrier(conf, bio->bi_iter.bi_sector); |
6bfe0b49 DW |
1451 | goto retry_write; |
1452 | } | |
1453 | ||
c230e7e5 N |
1454 | if (max_sectors < bio_sectors(bio)) { |
1455 | struct bio *split = bio_split(bio, max_sectors, | |
1456 | GFP_NOIO, conf->bio_split); | |
1457 | bio_chain(split, bio); | |
1458 | generic_make_request(bio); | |
1459 | bio = split; | |
1460 | r1_bio->master_bio = bio; | |
1f68f0c4 | 1461 | r1_bio->sectors = max_sectors; |
191ea9b2 | 1462 | } |
4b6d287f | 1463 | |
4e78064f | 1464 | atomic_set(&r1_bio->remaining, 1); |
4b6d287f | 1465 | atomic_set(&r1_bio->behind_remaining, 0); |
06d91a5f | 1466 | |
1f68f0c4 | 1467 | first_clone = 1; |
d8c84c4f | 1468 | |
1da177e4 | 1469 | for (i = 0; i < disks; i++) { |
8e58e327 | 1470 | struct bio *mbio = NULL; |
1da177e4 LT |
1471 | if (!r1_bio->bios[i]) |
1472 | continue; | |
1473 | ||
1f68f0c4 N |
1474 | |
1475 | if (first_clone) { | |
1476 | /* do behind I/O ? | |
1477 | * Not if there are too many, or cannot | |
1478 | * allocate memory, or a reader on WriteMostly | |
1479 | * is waiting for behind writes to flush */ | |
1480 | if (bitmap && | |
1481 | (atomic_read(&bitmap->behind_writes) | |
1482 | < mddev->bitmap_info.max_write_behind) && | |
8e58e327 | 1483 | !waitqueue_active(&bitmap->behind_wait)) { |
cb83efcf | 1484 | mbio = alloc_behind_master_bio(r1_bio, bio); |
8e58e327 | 1485 | } |
1f68f0c4 N |
1486 | |
1487 | bitmap_startwrite(bitmap, r1_bio->sector, | |
1488 | r1_bio->sectors, | |
1489 | test_bit(R1BIO_BehindIO, | |
1490 | &r1_bio->state)); | |
1491 | first_clone = 0; | |
1492 | } | |
8e58e327 ML |
1493 | |
1494 | if (!mbio) { | |
841c1316 ML |
1495 | if (r1_bio->behind_master_bio) |
1496 | mbio = bio_clone_fast(r1_bio->behind_master_bio, | |
1497 | GFP_NOIO, | |
1498 | mddev->bio_set); | |
c230e7e5 | 1499 | else |
1ec49223 | 1500 | mbio = bio_clone_fast(bio, GFP_NOIO, mddev->bio_set); |
8e58e327 ML |
1501 | } |
1502 | ||
841c1316 | 1503 | if (r1_bio->behind_master_bio) { |
4b6d287f N |
1504 | if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags)) |
1505 | atomic_inc(&r1_bio->behind_remaining); | |
1506 | } | |
1507 | ||
1f68f0c4 N |
1508 | r1_bio->bios[i] = mbio; |
1509 | ||
4f024f37 | 1510 | mbio->bi_iter.bi_sector = (r1_bio->sector + |
1f68f0c4 | 1511 | conf->mirrors[i].rdev->data_offset); |
109e3765 | 1512 | mbio->bi_bdev = conf->mirrors[i].rdev->bdev; |
1f68f0c4 | 1513 | mbio->bi_end_io = raid1_end_write_request; |
a682e003 | 1514 | mbio->bi_opf = bio_op(bio) | (bio->bi_opf & (REQ_SYNC | REQ_FUA)); |
212e7eb7 N |
1515 | if (test_bit(FailFast, &conf->mirrors[i].rdev->flags) && |
1516 | !test_bit(WriteMostly, &conf->mirrors[i].rdev->flags) && | |
1517 | conf->raid_disks - mddev->degraded > 1) | |
1518 | mbio->bi_opf |= MD_FAILFAST; | |
1f68f0c4 N |
1519 | mbio->bi_private = r1_bio; |
1520 | ||
1da177e4 | 1521 | atomic_inc(&r1_bio->remaining); |
f54a9d0e | 1522 | |
109e3765 N |
1523 | if (mddev->gendisk) |
1524 | trace_block_bio_remap(bdev_get_queue(mbio->bi_bdev), | |
1525 | mbio, disk_devt(mddev->gendisk), | |
1526 | r1_bio->sector); | |
1527 | /* flush_pending_writes() needs access to the rdev so...*/ | |
1528 | mbio->bi_bdev = (void*)conf->mirrors[i].rdev; | |
1529 | ||
f54a9d0e N |
1530 | cb = blk_check_plugged(raid1_unplug, mddev, sizeof(*plug)); |
1531 | if (cb) | |
1532 | plug = container_of(cb, struct raid1_plug_cb, cb); | |
1533 | else | |
1534 | plug = NULL; | |
f54a9d0e N |
1535 | if (plug) { |
1536 | bio_list_add(&plug->pending, mbio); | |
1537 | plug->pending_cnt++; | |
1538 | } else { | |
23b245c0 | 1539 | spin_lock_irqsave(&conf->device_lock, flags); |
f54a9d0e N |
1540 | bio_list_add(&conf->pending_bio_list, mbio); |
1541 | conf->pending_count++; | |
23b245c0 | 1542 | spin_unlock_irqrestore(&conf->device_lock, flags); |
b357f04a | 1543 | md_wakeup_thread(mddev->thread); |
23b245c0 | 1544 | } |
1da177e4 | 1545 | } |
1f68f0c4 | 1546 | |
079fa166 N |
1547 | r1_bio_write_done(r1_bio); |
1548 | ||
1549 | /* In case raid1d snuck in to freeze_array */ | |
1550 | wake_up(&conf->wait_barrier); | |
1da177e4 LT |
1551 | } |
1552 | ||
3b046a97 RL |
1553 | static void raid1_make_request(struct mddev *mddev, struct bio *bio) |
1554 | { | |
fd76863e | 1555 | sector_t sectors; |
3b046a97 | 1556 | |
aff8da09 SL |
1557 | if (unlikely(bio->bi_opf & REQ_PREFLUSH)) { |
1558 | md_flush_request(mddev, bio); | |
1559 | return; | |
1560 | } | |
3b046a97 | 1561 | |
c230e7e5 N |
1562 | /* |
1563 | * There is a limit to the maximum size, but | |
1564 | * the read/write handler might find a lower limit | |
1565 | * due to bad blocks. To avoid multiple splits, | |
1566 | * we pass the maximum number of sectors down | |
1567 | * and let the lower level perform the split. | |
1568 | */ | |
1569 | sectors = align_to_barrier_unit_end( | |
1570 | bio->bi_iter.bi_sector, bio_sectors(bio)); | |
61eb2b43 | 1571 | |
c230e7e5 | 1572 | if (bio_data_dir(bio) == READ) |
689389a0 | 1573 | raid1_read_request(mddev, bio, sectors, NULL); |
c230e7e5 N |
1574 | else |
1575 | raid1_write_request(mddev, bio, sectors); | |
3b046a97 RL |
1576 | } |
1577 | ||
849674e4 | 1578 | static void raid1_status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 1579 | { |
e8096360 | 1580 | struct r1conf *conf = mddev->private; |
1da177e4 LT |
1581 | int i; |
1582 | ||
1583 | seq_printf(seq, " [%d/%d] [", conf->raid_disks, | |
11ce99e6 | 1584 | conf->raid_disks - mddev->degraded); |
ddac7c7e N |
1585 | rcu_read_lock(); |
1586 | for (i = 0; i < conf->raid_disks; i++) { | |
3cb03002 | 1587 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
1da177e4 | 1588 | seq_printf(seq, "%s", |
ddac7c7e N |
1589 | rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); |
1590 | } | |
1591 | rcu_read_unlock(); | |
1da177e4 LT |
1592 | seq_printf(seq, "]"); |
1593 | } | |
1594 | ||
849674e4 | 1595 | static void raid1_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1596 | { |
1597 | char b[BDEVNAME_SIZE]; | |
e8096360 | 1598 | struct r1conf *conf = mddev->private; |
423f04d6 | 1599 | unsigned long flags; |
1da177e4 LT |
1600 | |
1601 | /* | |
1602 | * If it is not operational, then we have already marked it as dead | |
1603 | * else if it is the last working disks, ignore the error, let the | |
1604 | * next level up know. | |
1605 | * else mark the drive as failed | |
1606 | */ | |
2e52d449 | 1607 | spin_lock_irqsave(&conf->device_lock, flags); |
b2d444d7 | 1608 | if (test_bit(In_sync, &rdev->flags) |
4044ba58 | 1609 | && (conf->raid_disks - mddev->degraded) == 1) { |
1da177e4 LT |
1610 | /* |
1611 | * Don't fail the drive, act as though we were just a | |
4044ba58 N |
1612 | * normal single drive. |
1613 | * However don't try a recovery from this drive as | |
1614 | * it is very likely to fail. | |
1da177e4 | 1615 | */ |
5389042f | 1616 | conf->recovery_disabled = mddev->recovery_disabled; |
2e52d449 | 1617 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 1618 | return; |
4044ba58 | 1619 | } |
de393cde | 1620 | set_bit(Blocked, &rdev->flags); |
c04be0aa | 1621 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
1da177e4 | 1622 | mddev->degraded++; |
dd00a99e | 1623 | set_bit(Faulty, &rdev->flags); |
dd00a99e N |
1624 | } else |
1625 | set_bit(Faulty, &rdev->flags); | |
423f04d6 | 1626 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2446dba0 N |
1627 | /* |
1628 | * if recovery is running, make sure it aborts. | |
1629 | */ | |
1630 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2953079c SL |
1631 | set_mask_bits(&mddev->sb_flags, 0, |
1632 | BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING)); | |
1d41c216 N |
1633 | pr_crit("md/raid1:%s: Disk failure on %s, disabling device.\n" |
1634 | "md/raid1:%s: Operation continuing on %d devices.\n", | |
1635 | mdname(mddev), bdevname(rdev->bdev, b), | |
1636 | mdname(mddev), conf->raid_disks - mddev->degraded); | |
1da177e4 LT |
1637 | } |
1638 | ||
e8096360 | 1639 | static void print_conf(struct r1conf *conf) |
1da177e4 LT |
1640 | { |
1641 | int i; | |
1da177e4 | 1642 | |
1d41c216 | 1643 | pr_debug("RAID1 conf printout:\n"); |
1da177e4 | 1644 | if (!conf) { |
1d41c216 | 1645 | pr_debug("(!conf)\n"); |
1da177e4 LT |
1646 | return; |
1647 | } | |
1d41c216 N |
1648 | pr_debug(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded, |
1649 | conf->raid_disks); | |
1da177e4 | 1650 | |
ddac7c7e | 1651 | rcu_read_lock(); |
1da177e4 LT |
1652 | for (i = 0; i < conf->raid_disks; i++) { |
1653 | char b[BDEVNAME_SIZE]; | |
3cb03002 | 1654 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
ddac7c7e | 1655 | if (rdev) |
1d41c216 N |
1656 | pr_debug(" disk %d, wo:%d, o:%d, dev:%s\n", |
1657 | i, !test_bit(In_sync, &rdev->flags), | |
1658 | !test_bit(Faulty, &rdev->flags), | |
1659 | bdevname(rdev->bdev,b)); | |
1da177e4 | 1660 | } |
ddac7c7e | 1661 | rcu_read_unlock(); |
1da177e4 LT |
1662 | } |
1663 | ||
e8096360 | 1664 | static void close_sync(struct r1conf *conf) |
1da177e4 | 1665 | { |
fd76863e | 1666 | wait_all_barriers(conf); |
1667 | allow_all_barriers(conf); | |
1da177e4 LT |
1668 | |
1669 | mempool_destroy(conf->r1buf_pool); | |
1670 | conf->r1buf_pool = NULL; | |
1671 | } | |
1672 | ||
fd01b88c | 1673 | static int raid1_spare_active(struct mddev *mddev) |
1da177e4 LT |
1674 | { |
1675 | int i; | |
e8096360 | 1676 | struct r1conf *conf = mddev->private; |
6b965620 N |
1677 | int count = 0; |
1678 | unsigned long flags; | |
1da177e4 LT |
1679 | |
1680 | /* | |
f72ffdd6 | 1681 | * Find all failed disks within the RAID1 configuration |
ddac7c7e N |
1682 | * and mark them readable. |
1683 | * Called under mddev lock, so rcu protection not needed. | |
423f04d6 N |
1684 | * device_lock used to avoid races with raid1_end_read_request |
1685 | * which expects 'In_sync' flags and ->degraded to be consistent. | |
1da177e4 | 1686 | */ |
423f04d6 | 1687 | spin_lock_irqsave(&conf->device_lock, flags); |
1da177e4 | 1688 | for (i = 0; i < conf->raid_disks; i++) { |
3cb03002 | 1689 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
8c7a2c2b N |
1690 | struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev; |
1691 | if (repl | |
1aee41f6 | 1692 | && !test_bit(Candidate, &repl->flags) |
8c7a2c2b N |
1693 | && repl->recovery_offset == MaxSector |
1694 | && !test_bit(Faulty, &repl->flags) | |
1695 | && !test_and_set_bit(In_sync, &repl->flags)) { | |
1696 | /* replacement has just become active */ | |
1697 | if (!rdev || | |
1698 | !test_and_clear_bit(In_sync, &rdev->flags)) | |
1699 | count++; | |
1700 | if (rdev) { | |
1701 | /* Replaced device not technically | |
1702 | * faulty, but we need to be sure | |
1703 | * it gets removed and never re-added | |
1704 | */ | |
1705 | set_bit(Faulty, &rdev->flags); | |
1706 | sysfs_notify_dirent_safe( | |
1707 | rdev->sysfs_state); | |
1708 | } | |
1709 | } | |
ddac7c7e | 1710 | if (rdev |
61e4947c | 1711 | && rdev->recovery_offset == MaxSector |
ddac7c7e | 1712 | && !test_bit(Faulty, &rdev->flags) |
c04be0aa | 1713 | && !test_and_set_bit(In_sync, &rdev->flags)) { |
6b965620 | 1714 | count++; |
654e8b5a | 1715 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
1716 | } |
1717 | } | |
6b965620 N |
1718 | mddev->degraded -= count; |
1719 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 LT |
1720 | |
1721 | print_conf(conf); | |
6b965620 | 1722 | return count; |
1da177e4 LT |
1723 | } |
1724 | ||
fd01b88c | 1725 | static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1726 | { |
e8096360 | 1727 | struct r1conf *conf = mddev->private; |
199050ea | 1728 | int err = -EEXIST; |
41158c7e | 1729 | int mirror = 0; |
0eaf822c | 1730 | struct raid1_info *p; |
6c2fce2e | 1731 | int first = 0; |
30194636 | 1732 | int last = conf->raid_disks - 1; |
1da177e4 | 1733 | |
5389042f N |
1734 | if (mddev->recovery_disabled == conf->recovery_disabled) |
1735 | return -EBUSY; | |
1736 | ||
1501efad DW |
1737 | if (md_integrity_add_rdev(rdev, mddev)) |
1738 | return -ENXIO; | |
1739 | ||
6c2fce2e NB |
1740 | if (rdev->raid_disk >= 0) |
1741 | first = last = rdev->raid_disk; | |
1742 | ||
70bcecdb GR |
1743 | /* |
1744 | * find the disk ... but prefer rdev->saved_raid_disk | |
1745 | * if possible. | |
1746 | */ | |
1747 | if (rdev->saved_raid_disk >= 0 && | |
1748 | rdev->saved_raid_disk >= first && | |
1749 | conf->mirrors[rdev->saved_raid_disk].rdev == NULL) | |
1750 | first = last = rdev->saved_raid_disk; | |
1751 | ||
7ef449d1 N |
1752 | for (mirror = first; mirror <= last; mirror++) { |
1753 | p = conf->mirrors+mirror; | |
1754 | if (!p->rdev) { | |
1da177e4 | 1755 | |
9092c02d JB |
1756 | if (mddev->gendisk) |
1757 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
1758 | rdev->data_offset << 9); | |
1da177e4 LT |
1759 | |
1760 | p->head_position = 0; | |
1761 | rdev->raid_disk = mirror; | |
199050ea | 1762 | err = 0; |
6aea114a N |
1763 | /* As all devices are equivalent, we don't need a full recovery |
1764 | * if this was recently any drive of the array | |
1765 | */ | |
1766 | if (rdev->saved_raid_disk < 0) | |
41158c7e | 1767 | conf->fullsync = 1; |
d6065f7b | 1768 | rcu_assign_pointer(p->rdev, rdev); |
1da177e4 LT |
1769 | break; |
1770 | } | |
7ef449d1 N |
1771 | if (test_bit(WantReplacement, &p->rdev->flags) && |
1772 | p[conf->raid_disks].rdev == NULL) { | |
1773 | /* Add this device as a replacement */ | |
1774 | clear_bit(In_sync, &rdev->flags); | |
1775 | set_bit(Replacement, &rdev->flags); | |
1776 | rdev->raid_disk = mirror; | |
1777 | err = 0; | |
1778 | conf->fullsync = 1; | |
1779 | rcu_assign_pointer(p[conf->raid_disks].rdev, rdev); | |
1780 | break; | |
1781 | } | |
1782 | } | |
9092c02d | 1783 | if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev))) |
2ff8cc2c | 1784 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue); |
1da177e4 | 1785 | print_conf(conf); |
199050ea | 1786 | return err; |
1da177e4 LT |
1787 | } |
1788 | ||
b8321b68 | 1789 | static int raid1_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1790 | { |
e8096360 | 1791 | struct r1conf *conf = mddev->private; |
1da177e4 | 1792 | int err = 0; |
b8321b68 | 1793 | int number = rdev->raid_disk; |
0eaf822c | 1794 | struct raid1_info *p = conf->mirrors + number; |
1da177e4 | 1795 | |
b014f14c N |
1796 | if (rdev != p->rdev) |
1797 | p = conf->mirrors + conf->raid_disks + number; | |
1798 | ||
1da177e4 | 1799 | print_conf(conf); |
b8321b68 | 1800 | if (rdev == p->rdev) { |
b2d444d7 | 1801 | if (test_bit(In_sync, &rdev->flags) || |
1da177e4 LT |
1802 | atomic_read(&rdev->nr_pending)) { |
1803 | err = -EBUSY; | |
1804 | goto abort; | |
1805 | } | |
046abeed | 1806 | /* Only remove non-faulty devices if recovery |
dfc70645 N |
1807 | * is not possible. |
1808 | */ | |
1809 | if (!test_bit(Faulty, &rdev->flags) && | |
5389042f | 1810 | mddev->recovery_disabled != conf->recovery_disabled && |
dfc70645 N |
1811 | mddev->degraded < conf->raid_disks) { |
1812 | err = -EBUSY; | |
1813 | goto abort; | |
1814 | } | |
1da177e4 | 1815 | p->rdev = NULL; |
d787be40 N |
1816 | if (!test_bit(RemoveSynchronized, &rdev->flags)) { |
1817 | synchronize_rcu(); | |
1818 | if (atomic_read(&rdev->nr_pending)) { | |
1819 | /* lost the race, try later */ | |
1820 | err = -EBUSY; | |
1821 | p->rdev = rdev; | |
1822 | goto abort; | |
1823 | } | |
1824 | } | |
1825 | if (conf->mirrors[conf->raid_disks + number].rdev) { | |
8c7a2c2b N |
1826 | /* We just removed a device that is being replaced. |
1827 | * Move down the replacement. We drain all IO before | |
1828 | * doing this to avoid confusion. | |
1829 | */ | |
1830 | struct md_rdev *repl = | |
1831 | conf->mirrors[conf->raid_disks + number].rdev; | |
e2d59925 | 1832 | freeze_array(conf, 0); |
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 | */ | |
4246a0b6 | 1859 | if (!bio->bi_error) |
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 | { |
4246a0b6 | 1868 | int uptodate = !bio->bi_error; |
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 { | |
5e3db645 | 1882 | bitmap_end_sync(mddev->bitmap, s, |
6b1117d5 N |
1883 | &sync_blocks, 1); |
1884 | s += sync_blocks; | |
1885 | sectors_to_go -= sync_blocks; | |
1886 | } while (sectors_to_go > 0); | |
854abd75 N |
1887 | set_bit(WriteErrorSeen, &rdev->flags); |
1888 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) | |
19d67169 N |
1889 | set_bit(MD_RECOVERY_NEEDED, & |
1890 | mddev->recovery); | |
d8f05d29 | 1891 | set_bit(R1BIO_WriteError, &r1_bio->state); |
854abd75 | 1892 | } else if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors, |
3a9f28a5 N |
1893 | &first_bad, &bad_sectors) && |
1894 | !is_badblock(conf->mirrors[r1_bio->read_disk].rdev, | |
1895 | r1_bio->sector, | |
1896 | r1_bio->sectors, | |
1897 | &first_bad, &bad_sectors) | |
1898 | ) | |
4367af55 | 1899 | set_bit(R1BIO_MadeGood, &r1_bio->state); |
e3b9703e | 1900 | |
1da177e4 | 1901 | if (atomic_dec_and_test(&r1_bio->remaining)) { |
4367af55 | 1902 | int s = r1_bio->sectors; |
d8f05d29 N |
1903 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
1904 | test_bit(R1BIO_WriteError, &r1_bio->state)) | |
4367af55 N |
1905 | reschedule_retry(r1_bio); |
1906 | else { | |
1907 | put_buf(r1_bio); | |
1908 | md_done_sync(mddev, s, uptodate); | |
1909 | } | |
1da177e4 | 1910 | } |
1da177e4 LT |
1911 | } |
1912 | ||
3cb03002 | 1913 | static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector, |
d8f05d29 N |
1914 | int sectors, struct page *page, int rw) |
1915 | { | |
796a5cf0 | 1916 | if (sync_page_io(rdev, sector, sectors << 9, page, rw, 0, false)) |
d8f05d29 N |
1917 | /* success */ |
1918 | return 1; | |
19d67169 | 1919 | if (rw == WRITE) { |
d8f05d29 | 1920 | set_bit(WriteErrorSeen, &rdev->flags); |
19d67169 N |
1921 | if (!test_and_set_bit(WantReplacement, |
1922 | &rdev->flags)) | |
1923 | set_bit(MD_RECOVERY_NEEDED, & | |
1924 | rdev->mddev->recovery); | |
1925 | } | |
d8f05d29 N |
1926 | /* need to record an error - either for the block or the device */ |
1927 | if (!rdev_set_badblocks(rdev, sector, sectors, 0)) | |
1928 | md_error(rdev->mddev, rdev); | |
1929 | return 0; | |
1930 | } | |
1931 | ||
9f2c9d12 | 1932 | static int fix_sync_read_error(struct r1bio *r1_bio) |
1da177e4 | 1933 | { |
a68e5870 N |
1934 | /* Try some synchronous reads of other devices to get |
1935 | * good data, much like with normal read errors. Only | |
1936 | * read into the pages we already have so we don't | |
1937 | * need to re-issue the read request. | |
1938 | * We don't need to freeze the array, because being in an | |
1939 | * active sync request, there is no normal IO, and | |
1940 | * no overlapping syncs. | |
06f60385 N |
1941 | * We don't need to check is_badblock() again as we |
1942 | * made sure that anything with a bad block in range | |
1943 | * will have bi_end_io clear. | |
a68e5870 | 1944 | */ |
fd01b88c | 1945 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 1946 | struct r1conf *conf = mddev->private; |
a68e5870 | 1947 | struct bio *bio = r1_bio->bios[r1_bio->read_disk]; |
44cf0f4d | 1948 | struct page **pages = get_resync_pages(bio)->pages; |
a68e5870 N |
1949 | sector_t sect = r1_bio->sector; |
1950 | int sectors = r1_bio->sectors; | |
1951 | int idx = 0; | |
2e52d449 N |
1952 | struct md_rdev *rdev; |
1953 | ||
1954 | rdev = conf->mirrors[r1_bio->read_disk].rdev; | |
1955 | if (test_bit(FailFast, &rdev->flags)) { | |
1956 | /* Don't try recovering from here - just fail it | |
1957 | * ... unless it is the last working device of course */ | |
1958 | md_error(mddev, rdev); | |
1959 | if (test_bit(Faulty, &rdev->flags)) | |
1960 | /* Don't try to read from here, but make sure | |
1961 | * put_buf does it's thing | |
1962 | */ | |
1963 | bio->bi_end_io = end_sync_write; | |
1964 | } | |
a68e5870 N |
1965 | |
1966 | while(sectors) { | |
1967 | int s = sectors; | |
1968 | int d = r1_bio->read_disk; | |
1969 | int success = 0; | |
78d7f5f7 | 1970 | int start; |
a68e5870 N |
1971 | |
1972 | if (s > (PAGE_SIZE>>9)) | |
1973 | s = PAGE_SIZE >> 9; | |
1974 | do { | |
1975 | if (r1_bio->bios[d]->bi_end_io == end_sync_read) { | |
1976 | /* No rcu protection needed here devices | |
1977 | * can only be removed when no resync is | |
1978 | * active, and resync is currently active | |
1979 | */ | |
1980 | rdev = conf->mirrors[d].rdev; | |
9d3d8011 | 1981 | if (sync_page_io(rdev, sect, s<<9, |
44cf0f4d | 1982 | pages[idx], |
796a5cf0 | 1983 | REQ_OP_READ, 0, false)) { |
a68e5870 N |
1984 | success = 1; |
1985 | break; | |
1986 | } | |
1987 | } | |
1988 | d++; | |
8f19ccb2 | 1989 | if (d == conf->raid_disks * 2) |
a68e5870 N |
1990 | d = 0; |
1991 | } while (!success && d != r1_bio->read_disk); | |
1992 | ||
78d7f5f7 | 1993 | if (!success) { |
a68e5870 | 1994 | char b[BDEVNAME_SIZE]; |
3a9f28a5 N |
1995 | int abort = 0; |
1996 | /* Cannot read from anywhere, this block is lost. | |
1997 | * Record a bad block on each device. If that doesn't | |
1998 | * work just disable and interrupt the recovery. | |
1999 | * Don't fail devices as that won't really help. | |
2000 | */ | |
1d41c216 N |
2001 | pr_crit_ratelimited("md/raid1:%s: %s: unrecoverable I/O read error for block %llu\n", |
2002 | mdname(mddev), | |
2003 | bdevname(bio->bi_bdev, b), | |
2004 | (unsigned long long)r1_bio->sector); | |
8f19ccb2 | 2005 | for (d = 0; d < conf->raid_disks * 2; d++) { |
3a9f28a5 N |
2006 | rdev = conf->mirrors[d].rdev; |
2007 | if (!rdev || test_bit(Faulty, &rdev->flags)) | |
2008 | continue; | |
2009 | if (!rdev_set_badblocks(rdev, sect, s, 0)) | |
2010 | abort = 1; | |
2011 | } | |
2012 | if (abort) { | |
d890fa2b N |
2013 | conf->recovery_disabled = |
2014 | mddev->recovery_disabled; | |
3a9f28a5 N |
2015 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
2016 | md_done_sync(mddev, r1_bio->sectors, 0); | |
2017 | put_buf(r1_bio); | |
2018 | return 0; | |
2019 | } | |
2020 | /* Try next page */ | |
2021 | sectors -= s; | |
2022 | sect += s; | |
2023 | idx++; | |
2024 | continue; | |
d11c171e | 2025 | } |
78d7f5f7 N |
2026 | |
2027 | start = d; | |
2028 | /* write it back and re-read */ | |
2029 | while (d != r1_bio->read_disk) { | |
2030 | if (d == 0) | |
8f19ccb2 | 2031 | d = conf->raid_disks * 2; |
78d7f5f7 N |
2032 | d--; |
2033 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) | |
2034 | continue; | |
2035 | rdev = conf->mirrors[d].rdev; | |
d8f05d29 | 2036 | if (r1_sync_page_io(rdev, sect, s, |
44cf0f4d | 2037 | pages[idx], |
d8f05d29 | 2038 | WRITE) == 0) { |
78d7f5f7 N |
2039 | r1_bio->bios[d]->bi_end_io = NULL; |
2040 | rdev_dec_pending(rdev, mddev); | |
9d3d8011 | 2041 | } |
78d7f5f7 N |
2042 | } |
2043 | d = start; | |
2044 | while (d != r1_bio->read_disk) { | |
2045 | if (d == 0) | |
8f19ccb2 | 2046 | d = conf->raid_disks * 2; |
78d7f5f7 N |
2047 | d--; |
2048 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) | |
2049 | continue; | |
2050 | rdev = conf->mirrors[d].rdev; | |
d8f05d29 | 2051 | if (r1_sync_page_io(rdev, sect, s, |
44cf0f4d | 2052 | pages[idx], |
d8f05d29 | 2053 | READ) != 0) |
9d3d8011 | 2054 | atomic_add(s, &rdev->corrected_errors); |
78d7f5f7 | 2055 | } |
a68e5870 N |
2056 | sectors -= s; |
2057 | sect += s; | |
2058 | idx ++; | |
2059 | } | |
78d7f5f7 | 2060 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
4246a0b6 | 2061 | bio->bi_error = 0; |
a68e5870 N |
2062 | return 1; |
2063 | } | |
2064 | ||
c95e6385 | 2065 | static void process_checks(struct r1bio *r1_bio) |
a68e5870 N |
2066 | { |
2067 | /* We have read all readable devices. If we haven't | |
2068 | * got the block, then there is no hope left. | |
2069 | * If we have, then we want to do a comparison | |
2070 | * and skip the write if everything is the same. | |
2071 | * If any blocks failed to read, then we need to | |
2072 | * attempt an over-write | |
2073 | */ | |
fd01b88c | 2074 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 2075 | struct r1conf *conf = mddev->private; |
a68e5870 N |
2076 | int primary; |
2077 | int i; | |
f4380a91 | 2078 | int vcnt; |
a68e5870 | 2079 | |
30bc9b53 N |
2080 | /* Fix variable parts of all bios */ |
2081 | vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9); | |
2082 | for (i = 0; i < conf->raid_disks * 2; i++) { | |
2083 | int j; | |
2084 | int size; | |
4246a0b6 | 2085 | int error; |
60928a91 | 2086 | struct bio_vec *bi; |
30bc9b53 | 2087 | struct bio *b = r1_bio->bios[i]; |
98d30c58 | 2088 | struct resync_pages *rp = get_resync_pages(b); |
30bc9b53 N |
2089 | if (b->bi_end_io != end_sync_read) |
2090 | continue; | |
4246a0b6 CH |
2091 | /* fixup the bio for reuse, but preserve errno */ |
2092 | error = b->bi_error; | |
30bc9b53 | 2093 | bio_reset(b); |
4246a0b6 | 2094 | b->bi_error = error; |
30bc9b53 | 2095 | b->bi_vcnt = vcnt; |
4f024f37 KO |
2096 | b->bi_iter.bi_size = r1_bio->sectors << 9; |
2097 | b->bi_iter.bi_sector = r1_bio->sector + | |
30bc9b53 N |
2098 | conf->mirrors[i].rdev->data_offset; |
2099 | b->bi_bdev = conf->mirrors[i].rdev->bdev; | |
2100 | b->bi_end_io = end_sync_read; | |
98d30c58 ML |
2101 | rp->raid_bio = r1_bio; |
2102 | b->bi_private = rp; | |
30bc9b53 | 2103 | |
4f024f37 | 2104 | size = b->bi_iter.bi_size; |
60928a91 | 2105 | bio_for_each_segment_all(bi, b, j) { |
30bc9b53 N |
2106 | bi->bv_offset = 0; |
2107 | if (size > PAGE_SIZE) | |
2108 | bi->bv_len = PAGE_SIZE; | |
2109 | else | |
2110 | bi->bv_len = size; | |
2111 | size -= PAGE_SIZE; | |
2112 | } | |
2113 | } | |
8f19ccb2 | 2114 | for (primary = 0; primary < conf->raid_disks * 2; primary++) |
a68e5870 | 2115 | if (r1_bio->bios[primary]->bi_end_io == end_sync_read && |
4246a0b6 | 2116 | !r1_bio->bios[primary]->bi_error) { |
a68e5870 N |
2117 | r1_bio->bios[primary]->bi_end_io = NULL; |
2118 | rdev_dec_pending(conf->mirrors[primary].rdev, mddev); | |
2119 | break; | |
2120 | } | |
2121 | r1_bio->read_disk = primary; | |
8f19ccb2 | 2122 | for (i = 0; i < conf->raid_disks * 2; i++) { |
78d7f5f7 | 2123 | int j; |
78d7f5f7 N |
2124 | struct bio *pbio = r1_bio->bios[primary]; |
2125 | struct bio *sbio = r1_bio->bios[i]; | |
4246a0b6 | 2126 | int error = sbio->bi_error; |
44cf0f4d ML |
2127 | struct page **ppages = get_resync_pages(pbio)->pages; |
2128 | struct page **spages = get_resync_pages(sbio)->pages; | |
60928a91 | 2129 | struct bio_vec *bi; |
8fc04e6e | 2130 | int page_len[RESYNC_PAGES] = { 0 }; |
a68e5870 | 2131 | |
2aabaa65 | 2132 | if (sbio->bi_end_io != end_sync_read) |
78d7f5f7 | 2133 | continue; |
4246a0b6 CH |
2134 | /* Now we can 'fixup' the error value */ |
2135 | sbio->bi_error = 0; | |
78d7f5f7 | 2136 | |
60928a91 ML |
2137 | bio_for_each_segment_all(bi, sbio, j) |
2138 | page_len[j] = bi->bv_len; | |
2139 | ||
4246a0b6 | 2140 | if (!error) { |
78d7f5f7 | 2141 | for (j = vcnt; j-- ; ) { |
44cf0f4d ML |
2142 | if (memcmp(page_address(ppages[j]), |
2143 | page_address(spages[j]), | |
60928a91 | 2144 | page_len[j])) |
78d7f5f7 | 2145 | break; |
69382e85 | 2146 | } |
78d7f5f7 N |
2147 | } else |
2148 | j = 0; | |
2149 | if (j >= 0) | |
7f7583d4 | 2150 | atomic64_add(r1_bio->sectors, &mddev->resync_mismatches); |
78d7f5f7 | 2151 | if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery) |
4246a0b6 | 2152 | && !error)) { |
78d7f5f7 N |
2153 | /* No need to write to this device. */ |
2154 | sbio->bi_end_io = NULL; | |
2155 | rdev_dec_pending(conf->mirrors[i].rdev, mddev); | |
2156 | continue; | |
2157 | } | |
d3b45c2a KO |
2158 | |
2159 | bio_copy_data(sbio, pbio); | |
78d7f5f7 | 2160 | } |
a68e5870 N |
2161 | } |
2162 | ||
9f2c9d12 | 2163 | static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio) |
a68e5870 | 2164 | { |
e8096360 | 2165 | struct r1conf *conf = mddev->private; |
a68e5870 | 2166 | int i; |
8f19ccb2 | 2167 | int disks = conf->raid_disks * 2; |
a68e5870 N |
2168 | struct bio *bio, *wbio; |
2169 | ||
2170 | bio = r1_bio->bios[r1_bio->read_disk]; | |
2171 | ||
a68e5870 N |
2172 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) |
2173 | /* ouch - failed to read all of that. */ | |
2174 | if (!fix_sync_read_error(r1_bio)) | |
2175 | return; | |
7ca78d57 N |
2176 | |
2177 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
c95e6385 N |
2178 | process_checks(r1_bio); |
2179 | ||
d11c171e N |
2180 | /* |
2181 | * schedule writes | |
2182 | */ | |
1da177e4 LT |
2183 | atomic_set(&r1_bio->remaining, 1); |
2184 | for (i = 0; i < disks ; i++) { | |
2185 | wbio = r1_bio->bios[i]; | |
3e198f78 N |
2186 | if (wbio->bi_end_io == NULL || |
2187 | (wbio->bi_end_io == end_sync_read && | |
2188 | (i == r1_bio->read_disk || | |
2189 | !test_bit(MD_RECOVERY_SYNC, &mddev->recovery)))) | |
1da177e4 | 2190 | continue; |
0c9d5b12 N |
2191 | if (test_bit(Faulty, &conf->mirrors[i].rdev->flags)) |
2192 | continue; | |
1da177e4 | 2193 | |
796a5cf0 | 2194 | bio_set_op_attrs(wbio, REQ_OP_WRITE, 0); |
212e7eb7 N |
2195 | if (test_bit(FailFast, &conf->mirrors[i].rdev->flags)) |
2196 | wbio->bi_opf |= MD_FAILFAST; | |
2197 | ||
3e198f78 | 2198 | wbio->bi_end_io = end_sync_write; |
1da177e4 | 2199 | atomic_inc(&r1_bio->remaining); |
aa8b57aa | 2200 | md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio)); |
191ea9b2 | 2201 | |
1da177e4 LT |
2202 | generic_make_request(wbio); |
2203 | } | |
2204 | ||
2205 | if (atomic_dec_and_test(&r1_bio->remaining)) { | |
191ea9b2 | 2206 | /* if we're here, all write(s) have completed, so clean up */ |
58e94ae1 N |
2207 | int s = r1_bio->sectors; |
2208 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || | |
2209 | test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2210 | reschedule_retry(r1_bio); | |
2211 | else { | |
2212 | put_buf(r1_bio); | |
2213 | md_done_sync(mddev, s, 1); | |
2214 | } | |
1da177e4 LT |
2215 | } |
2216 | } | |
2217 | ||
2218 | /* | |
2219 | * This is a kernel thread which: | |
2220 | * | |
2221 | * 1. Retries failed read operations on working mirrors. | |
2222 | * 2. Updates the raid superblock when problems encounter. | |
d2eb35ac | 2223 | * 3. Performs writes following reads for array synchronising. |
1da177e4 LT |
2224 | */ |
2225 | ||
e8096360 | 2226 | static void fix_read_error(struct r1conf *conf, int read_disk, |
867868fb N |
2227 | sector_t sect, int sectors) |
2228 | { | |
fd01b88c | 2229 | struct mddev *mddev = conf->mddev; |
867868fb N |
2230 | while(sectors) { |
2231 | int s = sectors; | |
2232 | int d = read_disk; | |
2233 | int success = 0; | |
2234 | int start; | |
3cb03002 | 2235 | struct md_rdev *rdev; |
867868fb N |
2236 | |
2237 | if (s > (PAGE_SIZE>>9)) | |
2238 | s = PAGE_SIZE >> 9; | |
2239 | ||
2240 | do { | |
d2eb35ac N |
2241 | sector_t first_bad; |
2242 | int bad_sectors; | |
2243 | ||
707a6a42 N |
2244 | rcu_read_lock(); |
2245 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
867868fb | 2246 | if (rdev && |
da8840a7 | 2247 | (test_bit(In_sync, &rdev->flags) || |
2248 | (!test_bit(Faulty, &rdev->flags) && | |
2249 | rdev->recovery_offset >= sect + s)) && | |
d2eb35ac | 2250 | is_badblock(rdev, sect, s, |
707a6a42 N |
2251 | &first_bad, &bad_sectors) == 0) { |
2252 | atomic_inc(&rdev->nr_pending); | |
2253 | rcu_read_unlock(); | |
2254 | if (sync_page_io(rdev, sect, s<<9, | |
796a5cf0 | 2255 | conf->tmppage, REQ_OP_READ, 0, false)) |
707a6a42 N |
2256 | success = 1; |
2257 | rdev_dec_pending(rdev, mddev); | |
2258 | if (success) | |
2259 | break; | |
2260 | } else | |
2261 | rcu_read_unlock(); | |
2262 | d++; | |
2263 | if (d == conf->raid_disks * 2) | |
2264 | d = 0; | |
867868fb N |
2265 | } while (!success && d != read_disk); |
2266 | ||
2267 | if (!success) { | |
d8f05d29 | 2268 | /* Cannot read from anywhere - mark it bad */ |
3cb03002 | 2269 | struct md_rdev *rdev = conf->mirrors[read_disk].rdev; |
d8f05d29 N |
2270 | if (!rdev_set_badblocks(rdev, sect, s, 0)) |
2271 | md_error(mddev, rdev); | |
867868fb N |
2272 | break; |
2273 | } | |
2274 | /* write it back and re-read */ | |
2275 | start = d; | |
2276 | while (d != read_disk) { | |
2277 | if (d==0) | |
8f19ccb2 | 2278 | d = conf->raid_disks * 2; |
867868fb | 2279 | d--; |
707a6a42 N |
2280 | rcu_read_lock(); |
2281 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
867868fb | 2282 | if (rdev && |
707a6a42 N |
2283 | !test_bit(Faulty, &rdev->flags)) { |
2284 | atomic_inc(&rdev->nr_pending); | |
2285 | rcu_read_unlock(); | |
d8f05d29 N |
2286 | r1_sync_page_io(rdev, sect, s, |
2287 | conf->tmppage, WRITE); | |
707a6a42 N |
2288 | rdev_dec_pending(rdev, mddev); |
2289 | } else | |
2290 | rcu_read_unlock(); | |
867868fb N |
2291 | } |
2292 | d = start; | |
2293 | while (d != read_disk) { | |
2294 | char b[BDEVNAME_SIZE]; | |
2295 | if (d==0) | |
8f19ccb2 | 2296 | d = conf->raid_disks * 2; |
867868fb | 2297 | d--; |
707a6a42 N |
2298 | rcu_read_lock(); |
2299 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
867868fb | 2300 | if (rdev && |
b8cb6b4c | 2301 | !test_bit(Faulty, &rdev->flags)) { |
707a6a42 N |
2302 | atomic_inc(&rdev->nr_pending); |
2303 | rcu_read_unlock(); | |
d8f05d29 N |
2304 | if (r1_sync_page_io(rdev, sect, s, |
2305 | conf->tmppage, READ)) { | |
867868fb | 2306 | atomic_add(s, &rdev->corrected_errors); |
1d41c216 N |
2307 | pr_info("md/raid1:%s: read error corrected (%d sectors at %llu on %s)\n", |
2308 | mdname(mddev), s, | |
2309 | (unsigned long long)(sect + | |
2310 | rdev->data_offset), | |
2311 | bdevname(rdev->bdev, b)); | |
867868fb | 2312 | } |
707a6a42 N |
2313 | rdev_dec_pending(rdev, mddev); |
2314 | } else | |
2315 | rcu_read_unlock(); | |
867868fb N |
2316 | } |
2317 | sectors -= s; | |
2318 | sect += s; | |
2319 | } | |
2320 | } | |
2321 | ||
9f2c9d12 | 2322 | static int narrow_write_error(struct r1bio *r1_bio, int i) |
cd5ff9a1 | 2323 | { |
fd01b88c | 2324 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 2325 | struct r1conf *conf = mddev->private; |
3cb03002 | 2326 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
cd5ff9a1 N |
2327 | |
2328 | /* bio has the data to be written to device 'i' where | |
2329 | * we just recently had a write error. | |
2330 | * We repeatedly clone the bio and trim down to one block, | |
2331 | * then try the write. Where the write fails we record | |
2332 | * a bad block. | |
2333 | * It is conceivable that the bio doesn't exactly align with | |
2334 | * blocks. We must handle this somehow. | |
2335 | * | |
2336 | * We currently own a reference on the rdev. | |
2337 | */ | |
2338 | ||
2339 | int block_sectors; | |
2340 | sector_t sector; | |
2341 | int sectors; | |
2342 | int sect_to_write = r1_bio->sectors; | |
2343 | int ok = 1; | |
2344 | ||
2345 | if (rdev->badblocks.shift < 0) | |
2346 | return 0; | |
2347 | ||
ab713cdc ND |
2348 | block_sectors = roundup(1 << rdev->badblocks.shift, |
2349 | bdev_logical_block_size(rdev->bdev) >> 9); | |
cd5ff9a1 N |
2350 | sector = r1_bio->sector; |
2351 | sectors = ((sector + block_sectors) | |
2352 | & ~(sector_t)(block_sectors - 1)) | |
2353 | - sector; | |
2354 | ||
cd5ff9a1 N |
2355 | while (sect_to_write) { |
2356 | struct bio *wbio; | |
2357 | if (sectors > sect_to_write) | |
2358 | sectors = sect_to_write; | |
2359 | /* Write at 'sector' for 'sectors'*/ | |
2360 | ||
b783863f | 2361 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { |
841c1316 ML |
2362 | wbio = bio_clone_fast(r1_bio->behind_master_bio, |
2363 | GFP_NOIO, | |
2364 | mddev->bio_set); | |
2365 | /* We really need a _all clone */ | |
2366 | wbio->bi_iter = (struct bvec_iter){ 0 }; | |
b783863f | 2367 | } else { |
d7a10308 ML |
2368 | wbio = bio_clone_fast(r1_bio->master_bio, GFP_NOIO, |
2369 | mddev->bio_set); | |
b783863f KO |
2370 | } |
2371 | ||
796a5cf0 | 2372 | bio_set_op_attrs(wbio, REQ_OP_WRITE, 0); |
4f024f37 KO |
2373 | wbio->bi_iter.bi_sector = r1_bio->sector; |
2374 | wbio->bi_iter.bi_size = r1_bio->sectors << 9; | |
cd5ff9a1 | 2375 | |
6678d83f | 2376 | bio_trim(wbio, sector - r1_bio->sector, sectors); |
4f024f37 | 2377 | wbio->bi_iter.bi_sector += rdev->data_offset; |
cd5ff9a1 | 2378 | wbio->bi_bdev = rdev->bdev; |
4e49ea4a MC |
2379 | |
2380 | if (submit_bio_wait(wbio) < 0) | |
cd5ff9a1 N |
2381 | /* failure! */ |
2382 | ok = rdev_set_badblocks(rdev, sector, | |
2383 | sectors, 0) | |
2384 | && ok; | |
2385 | ||
2386 | bio_put(wbio); | |
2387 | sect_to_write -= sectors; | |
2388 | sector += sectors; | |
2389 | sectors = block_sectors; | |
2390 | } | |
2391 | return ok; | |
2392 | } | |
2393 | ||
e8096360 | 2394 | static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce N |
2395 | { |
2396 | int m; | |
2397 | int s = r1_bio->sectors; | |
8f19ccb2 | 2398 | for (m = 0; m < conf->raid_disks * 2 ; m++) { |
3cb03002 | 2399 | struct md_rdev *rdev = conf->mirrors[m].rdev; |
62096bce N |
2400 | struct bio *bio = r1_bio->bios[m]; |
2401 | if (bio->bi_end_io == NULL) | |
2402 | continue; | |
4246a0b6 | 2403 | if (!bio->bi_error && |
62096bce | 2404 | test_bit(R1BIO_MadeGood, &r1_bio->state)) { |
c6563a8c | 2405 | rdev_clear_badblocks(rdev, r1_bio->sector, s, 0); |
62096bce | 2406 | } |
4246a0b6 | 2407 | if (bio->bi_error && |
62096bce N |
2408 | test_bit(R1BIO_WriteError, &r1_bio->state)) { |
2409 | if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0)) | |
2410 | md_error(conf->mddev, rdev); | |
2411 | } | |
2412 | } | |
2413 | put_buf(r1_bio); | |
2414 | md_done_sync(conf->mddev, s, 1); | |
2415 | } | |
2416 | ||
e8096360 | 2417 | static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce | 2418 | { |
fd76863e | 2419 | int m, idx; |
55ce74d4 | 2420 | bool fail = false; |
fd76863e | 2421 | |
8f19ccb2 | 2422 | for (m = 0; m < conf->raid_disks * 2 ; m++) |
62096bce | 2423 | if (r1_bio->bios[m] == IO_MADE_GOOD) { |
3cb03002 | 2424 | struct md_rdev *rdev = conf->mirrors[m].rdev; |
62096bce N |
2425 | rdev_clear_badblocks(rdev, |
2426 | r1_bio->sector, | |
c6563a8c | 2427 | r1_bio->sectors, 0); |
62096bce N |
2428 | rdev_dec_pending(rdev, conf->mddev); |
2429 | } else if (r1_bio->bios[m] != NULL) { | |
2430 | /* This drive got a write error. We need to | |
2431 | * narrow down and record precise write | |
2432 | * errors. | |
2433 | */ | |
55ce74d4 | 2434 | fail = true; |
62096bce N |
2435 | if (!narrow_write_error(r1_bio, m)) { |
2436 | md_error(conf->mddev, | |
2437 | conf->mirrors[m].rdev); | |
2438 | /* an I/O failed, we can't clear the bitmap */ | |
2439 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
2440 | } | |
2441 | rdev_dec_pending(conf->mirrors[m].rdev, | |
2442 | conf->mddev); | |
2443 | } | |
55ce74d4 N |
2444 | if (fail) { |
2445 | spin_lock_irq(&conf->device_lock); | |
2446 | list_add(&r1_bio->retry_list, &conf->bio_end_io_list); | |
fd76863e | 2447 | idx = sector_to_idx(r1_bio->sector); |
824e47da | 2448 | atomic_inc(&conf->nr_queued[idx]); |
55ce74d4 | 2449 | spin_unlock_irq(&conf->device_lock); |
824e47da | 2450 | /* |
2451 | * In case freeze_array() is waiting for condition | |
2452 | * get_unqueued_pending() == extra to be true. | |
2453 | */ | |
2454 | wake_up(&conf->wait_barrier); | |
55ce74d4 | 2455 | md_wakeup_thread(conf->mddev->thread); |
bd8688a1 N |
2456 | } else { |
2457 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2458 | close_write(r1_bio); | |
55ce74d4 | 2459 | raid_end_bio_io(r1_bio); |
bd8688a1 | 2460 | } |
62096bce N |
2461 | } |
2462 | ||
e8096360 | 2463 | static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce | 2464 | { |
fd01b88c | 2465 | struct mddev *mddev = conf->mddev; |
62096bce | 2466 | struct bio *bio; |
3cb03002 | 2467 | struct md_rdev *rdev; |
109e3765 N |
2468 | dev_t bio_dev; |
2469 | sector_t bio_sector; | |
62096bce N |
2470 | |
2471 | clear_bit(R1BIO_ReadError, &r1_bio->state); | |
2472 | /* we got a read error. Maybe the drive is bad. Maybe just | |
2473 | * the block and we can fix it. | |
2474 | * We freeze all other IO, and try reading the block from | |
2475 | * other devices. When we find one, we re-write | |
2476 | * and check it that fixes the read error. | |
2477 | * This is all done synchronously while the array is | |
2478 | * frozen | |
2479 | */ | |
7449f699 TM |
2480 | |
2481 | bio = r1_bio->bios[r1_bio->read_disk]; | |
109e3765 N |
2482 | bio_dev = bio->bi_bdev->bd_dev; |
2483 | bio_sector = conf->mirrors[r1_bio->read_disk].rdev->data_offset + r1_bio->sector; | |
7449f699 TM |
2484 | bio_put(bio); |
2485 | r1_bio->bios[r1_bio->read_disk] = NULL; | |
2486 | ||
2e52d449 N |
2487 | rdev = conf->mirrors[r1_bio->read_disk].rdev; |
2488 | if (mddev->ro == 0 | |
2489 | && !test_bit(FailFast, &rdev->flags)) { | |
e2d59925 | 2490 | freeze_array(conf, 1); |
62096bce N |
2491 | fix_read_error(conf, r1_bio->read_disk, |
2492 | r1_bio->sector, r1_bio->sectors); | |
2493 | unfreeze_array(conf); | |
7449f699 TM |
2494 | } else { |
2495 | r1_bio->bios[r1_bio->read_disk] = IO_BLOCKED; | |
2496 | } | |
2497 | ||
2e52d449 | 2498 | rdev_dec_pending(rdev, conf->mddev); |
689389a0 N |
2499 | allow_barrier(conf, r1_bio->sector); |
2500 | bio = r1_bio->master_bio; | |
62096bce | 2501 | |
689389a0 N |
2502 | /* Reuse the old r1_bio so that the IO_BLOCKED settings are preserved */ |
2503 | r1_bio->state = 0; | |
2504 | raid1_read_request(mddev, bio, r1_bio->sectors, r1_bio); | |
62096bce N |
2505 | } |
2506 | ||
4ed8731d | 2507 | static void raid1d(struct md_thread *thread) |
1da177e4 | 2508 | { |
4ed8731d | 2509 | struct mddev *mddev = thread->mddev; |
9f2c9d12 | 2510 | struct r1bio *r1_bio; |
1da177e4 | 2511 | unsigned long flags; |
e8096360 | 2512 | struct r1conf *conf = mddev->private; |
1da177e4 | 2513 | struct list_head *head = &conf->retry_list; |
e1dfa0a2 | 2514 | struct blk_plug plug; |
fd76863e | 2515 | int idx; |
1da177e4 LT |
2516 | |
2517 | md_check_recovery(mddev); | |
e1dfa0a2 | 2518 | |
55ce74d4 | 2519 | if (!list_empty_careful(&conf->bio_end_io_list) && |
2953079c | 2520 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { |
55ce74d4 N |
2521 | LIST_HEAD(tmp); |
2522 | spin_lock_irqsave(&conf->device_lock, flags); | |
fd76863e | 2523 | if (!test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) |
2524 | list_splice_init(&conf->bio_end_io_list, &tmp); | |
55ce74d4 N |
2525 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2526 | while (!list_empty(&tmp)) { | |
a452744b MP |
2527 | r1_bio = list_first_entry(&tmp, struct r1bio, |
2528 | retry_list); | |
55ce74d4 | 2529 | list_del(&r1_bio->retry_list); |
fd76863e | 2530 | idx = sector_to_idx(r1_bio->sector); |
824e47da | 2531 | atomic_dec(&conf->nr_queued[idx]); |
bd8688a1 N |
2532 | if (mddev->degraded) |
2533 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
2534 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2535 | close_write(r1_bio); | |
55ce74d4 N |
2536 | raid_end_bio_io(r1_bio); |
2537 | } | |
2538 | } | |
2539 | ||
e1dfa0a2 | 2540 | blk_start_plug(&plug); |
1da177e4 | 2541 | for (;;) { |
191ea9b2 | 2542 | |
0021b7bc | 2543 | flush_pending_writes(conf); |
191ea9b2 | 2544 | |
a35e63ef N |
2545 | spin_lock_irqsave(&conf->device_lock, flags); |
2546 | if (list_empty(head)) { | |
2547 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 | 2548 | break; |
a35e63ef | 2549 | } |
9f2c9d12 | 2550 | r1_bio = list_entry(head->prev, struct r1bio, retry_list); |
1da177e4 | 2551 | list_del(head->prev); |
fd76863e | 2552 | idx = sector_to_idx(r1_bio->sector); |
824e47da | 2553 | atomic_dec(&conf->nr_queued[idx]); |
1da177e4 LT |
2554 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2555 | ||
2556 | mddev = r1_bio->mddev; | |
070ec55d | 2557 | conf = mddev->private; |
4367af55 | 2558 | if (test_bit(R1BIO_IsSync, &r1_bio->state)) { |
d8f05d29 | 2559 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
62096bce N |
2560 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
2561 | handle_sync_write_finished(conf, r1_bio); | |
2562 | else | |
4367af55 | 2563 | sync_request_write(mddev, r1_bio); |
cd5ff9a1 | 2564 | } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
62096bce N |
2565 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
2566 | handle_write_finished(conf, r1_bio); | |
2567 | else if (test_bit(R1BIO_ReadError, &r1_bio->state)) | |
2568 | handle_read_error(conf, r1_bio); | |
2569 | else | |
c230e7e5 | 2570 | WARN_ON_ONCE(1); |
62096bce | 2571 | |
1d9d5241 | 2572 | cond_resched(); |
2953079c | 2573 | if (mddev->sb_flags & ~(1<<MD_SB_CHANGE_PENDING)) |
de393cde | 2574 | md_check_recovery(mddev); |
1da177e4 | 2575 | } |
e1dfa0a2 | 2576 | blk_finish_plug(&plug); |
1da177e4 LT |
2577 | } |
2578 | ||
e8096360 | 2579 | static int init_resync(struct r1conf *conf) |
1da177e4 LT |
2580 | { |
2581 | int buffs; | |
2582 | ||
2583 | buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; | |
9e77c485 | 2584 | BUG_ON(conf->r1buf_pool); |
1da177e4 LT |
2585 | conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free, |
2586 | conf->poolinfo); | |
2587 | if (!conf->r1buf_pool) | |
2588 | return -ENOMEM; | |
1da177e4 LT |
2589 | return 0; |
2590 | } | |
2591 | ||
2592 | /* | |
2593 | * perform a "sync" on one "block" | |
2594 | * | |
2595 | * We need to make sure that no normal I/O request - particularly write | |
2596 | * requests - conflict with active sync requests. | |
2597 | * | |
2598 | * This is achieved by tracking pending requests and a 'barrier' concept | |
2599 | * that can be installed to exclude normal IO requests. | |
2600 | */ | |
2601 | ||
849674e4 SL |
2602 | static sector_t raid1_sync_request(struct mddev *mddev, sector_t sector_nr, |
2603 | int *skipped) | |
1da177e4 | 2604 | { |
e8096360 | 2605 | struct r1conf *conf = mddev->private; |
9f2c9d12 | 2606 | struct r1bio *r1_bio; |
1da177e4 LT |
2607 | struct bio *bio; |
2608 | sector_t max_sector, nr_sectors; | |
3e198f78 | 2609 | int disk = -1; |
1da177e4 | 2610 | int i; |
3e198f78 N |
2611 | int wonly = -1; |
2612 | int write_targets = 0, read_targets = 0; | |
57dab0bd | 2613 | sector_t sync_blocks; |
e3b9703e | 2614 | int still_degraded = 0; |
06f60385 N |
2615 | int good_sectors = RESYNC_SECTORS; |
2616 | int min_bad = 0; /* number of sectors that are bad in all devices */ | |
fd76863e | 2617 | int idx = sector_to_idx(sector_nr); |
1da177e4 LT |
2618 | |
2619 | if (!conf->r1buf_pool) | |
2620 | if (init_resync(conf)) | |
57afd89f | 2621 | return 0; |
1da177e4 | 2622 | |
58c0fed4 | 2623 | max_sector = mddev->dev_sectors; |
1da177e4 | 2624 | if (sector_nr >= max_sector) { |
191ea9b2 N |
2625 | /* If we aborted, we need to abort the |
2626 | * sync on the 'current' bitmap chunk (there will | |
2627 | * only be one in raid1 resync. | |
2628 | * We can find the current addess in mddev->curr_resync | |
2629 | */ | |
6a806c51 N |
2630 | if (mddev->curr_resync < max_sector) /* aborted */ |
2631 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
191ea9b2 | 2632 | &sync_blocks, 1); |
6a806c51 | 2633 | else /* completed sync */ |
191ea9b2 | 2634 | conf->fullsync = 0; |
6a806c51 N |
2635 | |
2636 | bitmap_close_sync(mddev->bitmap); | |
1da177e4 | 2637 | close_sync(conf); |
c40f341f GR |
2638 | |
2639 | if (mddev_is_clustered(mddev)) { | |
2640 | conf->cluster_sync_low = 0; | |
2641 | conf->cluster_sync_high = 0; | |
c40f341f | 2642 | } |
1da177e4 LT |
2643 | return 0; |
2644 | } | |
2645 | ||
07d84d10 N |
2646 | if (mddev->bitmap == NULL && |
2647 | mddev->recovery_cp == MaxSector && | |
6394cca5 | 2648 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
07d84d10 N |
2649 | conf->fullsync == 0) { |
2650 | *skipped = 1; | |
2651 | return max_sector - sector_nr; | |
2652 | } | |
6394cca5 N |
2653 | /* before building a request, check if we can skip these blocks.. |
2654 | * This call the bitmap_start_sync doesn't actually record anything | |
2655 | */ | |
e3b9703e | 2656 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
e5de485f | 2657 | !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { |
191ea9b2 N |
2658 | /* We can skip this block, and probably several more */ |
2659 | *skipped = 1; | |
2660 | return sync_blocks; | |
2661 | } | |
17999be4 | 2662 | |
7ac50447 TM |
2663 | /* |
2664 | * If there is non-resync activity waiting for a turn, then let it | |
2665 | * though before starting on this new sync request. | |
2666 | */ | |
824e47da | 2667 | if (atomic_read(&conf->nr_waiting[idx])) |
7ac50447 TM |
2668 | schedule_timeout_uninterruptible(1); |
2669 | ||
c40f341f GR |
2670 | /* we are incrementing sector_nr below. To be safe, we check against |
2671 | * sector_nr + two times RESYNC_SECTORS | |
2672 | */ | |
2673 | ||
2674 | bitmap_cond_end_sync(mddev->bitmap, sector_nr, | |
2675 | mddev_is_clustered(mddev) && (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high)); | |
1c4588e9 | 2676 | r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO); |
17999be4 | 2677 | |
c2fd4c94 | 2678 | raise_barrier(conf, sector_nr); |
1da177e4 | 2679 | |
3e198f78 | 2680 | rcu_read_lock(); |
1da177e4 | 2681 | /* |
3e198f78 N |
2682 | * If we get a correctably read error during resync or recovery, |
2683 | * we might want to read from a different device. So we | |
2684 | * flag all drives that could conceivably be read from for READ, | |
2685 | * and any others (which will be non-In_sync devices) for WRITE. | |
2686 | * If a read fails, we try reading from something else for which READ | |
2687 | * is OK. | |
1da177e4 | 2688 | */ |
1da177e4 | 2689 | |
1da177e4 LT |
2690 | r1_bio->mddev = mddev; |
2691 | r1_bio->sector = sector_nr; | |
191ea9b2 | 2692 | r1_bio->state = 0; |
1da177e4 | 2693 | set_bit(R1BIO_IsSync, &r1_bio->state); |
fd76863e | 2694 | /* make sure good_sectors won't go across barrier unit boundary */ |
2695 | good_sectors = align_to_barrier_unit_end(sector_nr, good_sectors); | |
1da177e4 | 2696 | |
8f19ccb2 | 2697 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3cb03002 | 2698 | struct md_rdev *rdev; |
1da177e4 | 2699 | bio = r1_bio->bios[i]; |
1da177e4 | 2700 | |
3e198f78 N |
2701 | rdev = rcu_dereference(conf->mirrors[i].rdev); |
2702 | if (rdev == NULL || | |
06f60385 | 2703 | test_bit(Faulty, &rdev->flags)) { |
8f19ccb2 N |
2704 | if (i < conf->raid_disks) |
2705 | still_degraded = 1; | |
3e198f78 | 2706 | } else if (!test_bit(In_sync, &rdev->flags)) { |
796a5cf0 | 2707 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
1da177e4 LT |
2708 | bio->bi_end_io = end_sync_write; |
2709 | write_targets ++; | |
3e198f78 N |
2710 | } else { |
2711 | /* may need to read from here */ | |
06f60385 N |
2712 | sector_t first_bad = MaxSector; |
2713 | int bad_sectors; | |
2714 | ||
2715 | if (is_badblock(rdev, sector_nr, good_sectors, | |
2716 | &first_bad, &bad_sectors)) { | |
2717 | if (first_bad > sector_nr) | |
2718 | good_sectors = first_bad - sector_nr; | |
2719 | else { | |
2720 | bad_sectors -= (sector_nr - first_bad); | |
2721 | if (min_bad == 0 || | |
2722 | min_bad > bad_sectors) | |
2723 | min_bad = bad_sectors; | |
2724 | } | |
2725 | } | |
2726 | if (sector_nr < first_bad) { | |
2727 | if (test_bit(WriteMostly, &rdev->flags)) { | |
2728 | if (wonly < 0) | |
2729 | wonly = i; | |
2730 | } else { | |
2731 | if (disk < 0) | |
2732 | disk = i; | |
2733 | } | |
796a5cf0 | 2734 | bio_set_op_attrs(bio, REQ_OP_READ, 0); |
06f60385 N |
2735 | bio->bi_end_io = end_sync_read; |
2736 | read_targets++; | |
d57368af AL |
2737 | } else if (!test_bit(WriteErrorSeen, &rdev->flags) && |
2738 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && | |
2739 | !test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { | |
2740 | /* | |
2741 | * The device is suitable for reading (InSync), | |
2742 | * but has bad block(s) here. Let's try to correct them, | |
2743 | * if we are doing resync or repair. Otherwise, leave | |
2744 | * this device alone for this sync request. | |
2745 | */ | |
796a5cf0 | 2746 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
d57368af AL |
2747 | bio->bi_end_io = end_sync_write; |
2748 | write_targets++; | |
3e198f78 | 2749 | } |
3e198f78 | 2750 | } |
06f60385 N |
2751 | if (bio->bi_end_io) { |
2752 | atomic_inc(&rdev->nr_pending); | |
4f024f37 | 2753 | bio->bi_iter.bi_sector = sector_nr + rdev->data_offset; |
06f60385 | 2754 | bio->bi_bdev = rdev->bdev; |
2e52d449 N |
2755 | if (test_bit(FailFast, &rdev->flags)) |
2756 | bio->bi_opf |= MD_FAILFAST; | |
06f60385 | 2757 | } |
1da177e4 | 2758 | } |
3e198f78 N |
2759 | rcu_read_unlock(); |
2760 | if (disk < 0) | |
2761 | disk = wonly; | |
2762 | r1_bio->read_disk = disk; | |
191ea9b2 | 2763 | |
06f60385 N |
2764 | if (read_targets == 0 && min_bad > 0) { |
2765 | /* These sectors are bad on all InSync devices, so we | |
2766 | * need to mark them bad on all write targets | |
2767 | */ | |
2768 | int ok = 1; | |
8f19ccb2 | 2769 | for (i = 0 ; i < conf->raid_disks * 2 ; i++) |
06f60385 | 2770 | if (r1_bio->bios[i]->bi_end_io == end_sync_write) { |
a42f9d83 | 2771 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
06f60385 N |
2772 | ok = rdev_set_badblocks(rdev, sector_nr, |
2773 | min_bad, 0 | |
2774 | ) && ok; | |
2775 | } | |
2953079c | 2776 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
06f60385 N |
2777 | *skipped = 1; |
2778 | put_buf(r1_bio); | |
2779 | ||
2780 | if (!ok) { | |
2781 | /* Cannot record the badblocks, so need to | |
2782 | * abort the resync. | |
2783 | * If there are multiple read targets, could just | |
2784 | * fail the really bad ones ??? | |
2785 | */ | |
2786 | conf->recovery_disabled = mddev->recovery_disabled; | |
2787 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2788 | return 0; | |
2789 | } else | |
2790 | return min_bad; | |
2791 | ||
2792 | } | |
2793 | if (min_bad > 0 && min_bad < good_sectors) { | |
2794 | /* only resync enough to reach the next bad->good | |
2795 | * transition */ | |
2796 | good_sectors = min_bad; | |
2797 | } | |
2798 | ||
3e198f78 N |
2799 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0) |
2800 | /* extra read targets are also write targets */ | |
2801 | write_targets += read_targets-1; | |
2802 | ||
2803 | if (write_targets == 0 || read_targets == 0) { | |
1da177e4 LT |
2804 | /* There is nowhere to write, so all non-sync |
2805 | * drives must be failed - so we are finished | |
2806 | */ | |
b7219ccb N |
2807 | sector_t rv; |
2808 | if (min_bad > 0) | |
2809 | max_sector = sector_nr + min_bad; | |
2810 | rv = max_sector - sector_nr; | |
57afd89f | 2811 | *skipped = 1; |
1da177e4 | 2812 | put_buf(r1_bio); |
1da177e4 LT |
2813 | return rv; |
2814 | } | |
2815 | ||
c6207277 N |
2816 | if (max_sector > mddev->resync_max) |
2817 | max_sector = mddev->resync_max; /* Don't do IO beyond here */ | |
06f60385 N |
2818 | if (max_sector > sector_nr + good_sectors) |
2819 | max_sector = sector_nr + good_sectors; | |
1da177e4 | 2820 | nr_sectors = 0; |
289e99e8 | 2821 | sync_blocks = 0; |
1da177e4 LT |
2822 | do { |
2823 | struct page *page; | |
2824 | int len = PAGE_SIZE; | |
2825 | if (sector_nr + (len>>9) > max_sector) | |
2826 | len = (max_sector - sector_nr) << 9; | |
2827 | if (len == 0) | |
2828 | break; | |
6a806c51 N |
2829 | if (sync_blocks == 0) { |
2830 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, | |
e5de485f N |
2831 | &sync_blocks, still_degraded) && |
2832 | !conf->fullsync && | |
2833 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6a806c51 | 2834 | break; |
7571ae88 | 2835 | if ((len >> 9) > sync_blocks) |
6a806c51 | 2836 | len = sync_blocks<<9; |
ab7a30c7 | 2837 | } |
191ea9b2 | 2838 | |
8f19ccb2 | 2839 | for (i = 0 ; i < conf->raid_disks * 2; i++) { |
98d30c58 ML |
2840 | struct resync_pages *rp; |
2841 | ||
1da177e4 | 2842 | bio = r1_bio->bios[i]; |
98d30c58 | 2843 | rp = get_resync_pages(bio); |
1da177e4 | 2844 | if (bio->bi_end_io) { |
98d30c58 | 2845 | page = resync_fetch_page(rp, rp->idx++); |
c85ba149 ML |
2846 | |
2847 | /* | |
2848 | * won't fail because the vec table is big | |
2849 | * enough to hold all these pages | |
2850 | */ | |
2851 | bio_add_page(bio, page, len, 0); | |
1da177e4 LT |
2852 | } |
2853 | } | |
2854 | nr_sectors += len>>9; | |
2855 | sector_nr += len>>9; | |
191ea9b2 | 2856 | sync_blocks -= (len>>9); |
98d30c58 ML |
2857 | } while (get_resync_pages(r1_bio->bios[disk]->bi_private)->idx < RESYNC_PAGES); |
2858 | ||
1da177e4 LT |
2859 | r1_bio->sectors = nr_sectors; |
2860 | ||
c40f341f GR |
2861 | if (mddev_is_clustered(mddev) && |
2862 | conf->cluster_sync_high < sector_nr + nr_sectors) { | |
2863 | conf->cluster_sync_low = mddev->curr_resync_completed; | |
2864 | conf->cluster_sync_high = conf->cluster_sync_low + CLUSTER_RESYNC_WINDOW_SECTORS; | |
2865 | /* Send resync message */ | |
2866 | md_cluster_ops->resync_info_update(mddev, | |
2867 | conf->cluster_sync_low, | |
2868 | conf->cluster_sync_high); | |
2869 | } | |
2870 | ||
d11c171e N |
2871 | /* For a user-requested sync, we read all readable devices and do a |
2872 | * compare | |
2873 | */ | |
2874 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
2875 | atomic_set(&r1_bio->remaining, read_targets); | |
2d4f4f33 | 2876 | for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) { |
d11c171e N |
2877 | bio = r1_bio->bios[i]; |
2878 | if (bio->bi_end_io == end_sync_read) { | |
2d4f4f33 | 2879 | read_targets--; |
ddac7c7e | 2880 | md_sync_acct(bio->bi_bdev, nr_sectors); |
2e52d449 N |
2881 | if (read_targets == 1) |
2882 | bio->bi_opf &= ~MD_FAILFAST; | |
d11c171e N |
2883 | generic_make_request(bio); |
2884 | } | |
2885 | } | |
2886 | } else { | |
2887 | atomic_set(&r1_bio->remaining, 1); | |
2888 | bio = r1_bio->bios[r1_bio->read_disk]; | |
ddac7c7e | 2889 | md_sync_acct(bio->bi_bdev, nr_sectors); |
2e52d449 N |
2890 | if (read_targets == 1) |
2891 | bio->bi_opf &= ~MD_FAILFAST; | |
d11c171e | 2892 | generic_make_request(bio); |
1da177e4 | 2893 | |
d11c171e | 2894 | } |
1da177e4 LT |
2895 | return nr_sectors; |
2896 | } | |
2897 | ||
fd01b88c | 2898 | static sector_t raid1_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce DW |
2899 | { |
2900 | if (sectors) | |
2901 | return sectors; | |
2902 | ||
2903 | return mddev->dev_sectors; | |
2904 | } | |
2905 | ||
e8096360 | 2906 | static struct r1conf *setup_conf(struct mddev *mddev) |
1da177e4 | 2907 | { |
e8096360 | 2908 | struct r1conf *conf; |
709ae487 | 2909 | int i; |
0eaf822c | 2910 | struct raid1_info *disk; |
3cb03002 | 2911 | struct md_rdev *rdev; |
709ae487 | 2912 | int err = -ENOMEM; |
1da177e4 | 2913 | |
e8096360 | 2914 | conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL); |
1da177e4 | 2915 | if (!conf) |
709ae487 | 2916 | goto abort; |
1da177e4 | 2917 | |
fd76863e | 2918 | conf->nr_pending = kcalloc(BARRIER_BUCKETS_NR, |
824e47da | 2919 | sizeof(atomic_t), GFP_KERNEL); |
fd76863e | 2920 | if (!conf->nr_pending) |
2921 | goto abort; | |
2922 | ||
2923 | conf->nr_waiting = kcalloc(BARRIER_BUCKETS_NR, | |
824e47da | 2924 | sizeof(atomic_t), GFP_KERNEL); |
fd76863e | 2925 | if (!conf->nr_waiting) |
2926 | goto abort; | |
2927 | ||
2928 | conf->nr_queued = kcalloc(BARRIER_BUCKETS_NR, | |
824e47da | 2929 | sizeof(atomic_t), GFP_KERNEL); |
fd76863e | 2930 | if (!conf->nr_queued) |
2931 | goto abort; | |
2932 | ||
2933 | conf->barrier = kcalloc(BARRIER_BUCKETS_NR, | |
824e47da | 2934 | sizeof(atomic_t), GFP_KERNEL); |
fd76863e | 2935 | if (!conf->barrier) |
2936 | goto abort; | |
2937 | ||
0eaf822c | 2938 | conf->mirrors = kzalloc(sizeof(struct raid1_info) |
8f19ccb2 | 2939 | * mddev->raid_disks * 2, |
1da177e4 LT |
2940 | GFP_KERNEL); |
2941 | if (!conf->mirrors) | |
709ae487 | 2942 | goto abort; |
1da177e4 | 2943 | |
ddaf22ab N |
2944 | conf->tmppage = alloc_page(GFP_KERNEL); |
2945 | if (!conf->tmppage) | |
709ae487 | 2946 | goto abort; |
ddaf22ab | 2947 | |
709ae487 | 2948 | conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL); |
1da177e4 | 2949 | if (!conf->poolinfo) |
709ae487 | 2950 | goto abort; |
8f19ccb2 | 2951 | conf->poolinfo->raid_disks = mddev->raid_disks * 2; |
1da177e4 LT |
2952 | conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, |
2953 | r1bio_pool_free, | |
2954 | conf->poolinfo); | |
2955 | if (!conf->r1bio_pool) | |
709ae487 N |
2956 | goto abort; |
2957 | ||
c230e7e5 N |
2958 | conf->bio_split = bioset_create(BIO_POOL_SIZE, 0); |
2959 | if (!conf->bio_split) | |
2960 | goto abort; | |
2961 | ||
ed9bfdf1 | 2962 | conf->poolinfo->mddev = mddev; |
1da177e4 | 2963 | |
c19d5798 | 2964 | err = -EINVAL; |
e7e72bf6 | 2965 | spin_lock_init(&conf->device_lock); |
dafb20fa | 2966 | rdev_for_each(rdev, mddev) { |
709ae487 | 2967 | int disk_idx = rdev->raid_disk; |
1da177e4 LT |
2968 | if (disk_idx >= mddev->raid_disks |
2969 | || disk_idx < 0) | |
2970 | continue; | |
c19d5798 | 2971 | if (test_bit(Replacement, &rdev->flags)) |
02b898f2 | 2972 | disk = conf->mirrors + mddev->raid_disks + disk_idx; |
c19d5798 N |
2973 | else |
2974 | disk = conf->mirrors + disk_idx; | |
1da177e4 | 2975 | |
c19d5798 N |
2976 | if (disk->rdev) |
2977 | goto abort; | |
1da177e4 | 2978 | disk->rdev = rdev; |
1da177e4 | 2979 | disk->head_position = 0; |
12cee5a8 | 2980 | disk->seq_start = MaxSector; |
1da177e4 LT |
2981 | } |
2982 | conf->raid_disks = mddev->raid_disks; | |
2983 | conf->mddev = mddev; | |
1da177e4 | 2984 | INIT_LIST_HEAD(&conf->retry_list); |
55ce74d4 | 2985 | INIT_LIST_HEAD(&conf->bio_end_io_list); |
1da177e4 LT |
2986 | |
2987 | spin_lock_init(&conf->resync_lock); | |
17999be4 | 2988 | init_waitqueue_head(&conf->wait_barrier); |
1da177e4 | 2989 | |
191ea9b2 | 2990 | bio_list_init(&conf->pending_bio_list); |
34db0cd6 | 2991 | conf->pending_count = 0; |
d890fa2b | 2992 | conf->recovery_disabled = mddev->recovery_disabled - 1; |
191ea9b2 | 2993 | |
c19d5798 | 2994 | err = -EIO; |
8f19ccb2 | 2995 | for (i = 0; i < conf->raid_disks * 2; i++) { |
1da177e4 LT |
2996 | |
2997 | disk = conf->mirrors + i; | |
2998 | ||
c19d5798 N |
2999 | if (i < conf->raid_disks && |
3000 | disk[conf->raid_disks].rdev) { | |
3001 | /* This slot has a replacement. */ | |
3002 | if (!disk->rdev) { | |
3003 | /* No original, just make the replacement | |
3004 | * a recovering spare | |
3005 | */ | |
3006 | disk->rdev = | |
3007 | disk[conf->raid_disks].rdev; | |
3008 | disk[conf->raid_disks].rdev = NULL; | |
3009 | } else if (!test_bit(In_sync, &disk->rdev->flags)) | |
3010 | /* Original is not in_sync - bad */ | |
3011 | goto abort; | |
3012 | } | |
3013 | ||
5fd6c1dc N |
3014 | if (!disk->rdev || |
3015 | !test_bit(In_sync, &disk->rdev->flags)) { | |
1da177e4 | 3016 | disk->head_position = 0; |
4f0a5e01 JB |
3017 | if (disk->rdev && |
3018 | (disk->rdev->saved_raid_disk < 0)) | |
918f0238 | 3019 | conf->fullsync = 1; |
be4d3280 | 3020 | } |
1da177e4 | 3021 | } |
709ae487 | 3022 | |
709ae487 | 3023 | err = -ENOMEM; |
0232605d | 3024 | conf->thread = md_register_thread(raid1d, mddev, "raid1"); |
1d41c216 | 3025 | if (!conf->thread) |
709ae487 | 3026 | goto abort; |
1da177e4 | 3027 | |
709ae487 N |
3028 | return conf; |
3029 | ||
3030 | abort: | |
3031 | if (conf) { | |
644df1a8 | 3032 | mempool_destroy(conf->r1bio_pool); |
709ae487 N |
3033 | kfree(conf->mirrors); |
3034 | safe_put_page(conf->tmppage); | |
3035 | kfree(conf->poolinfo); | |
fd76863e | 3036 | kfree(conf->nr_pending); |
3037 | kfree(conf->nr_waiting); | |
3038 | kfree(conf->nr_queued); | |
3039 | kfree(conf->barrier); | |
c230e7e5 N |
3040 | if (conf->bio_split) |
3041 | bioset_free(conf->bio_split); | |
709ae487 N |
3042 | kfree(conf); |
3043 | } | |
3044 | return ERR_PTR(err); | |
3045 | } | |
3046 | ||
afa0f557 | 3047 | static void raid1_free(struct mddev *mddev, void *priv); |
849674e4 | 3048 | static int raid1_run(struct mddev *mddev) |
709ae487 | 3049 | { |
e8096360 | 3050 | struct r1conf *conf; |
709ae487 | 3051 | int i; |
3cb03002 | 3052 | struct md_rdev *rdev; |
5220ea1e | 3053 | int ret; |
2ff8cc2c | 3054 | bool discard_supported = false; |
709ae487 N |
3055 | |
3056 | if (mddev->level != 1) { | |
1d41c216 N |
3057 | pr_warn("md/raid1:%s: raid level not set to mirroring (%d)\n", |
3058 | mdname(mddev), mddev->level); | |
709ae487 N |
3059 | return -EIO; |
3060 | } | |
3061 | if (mddev->reshape_position != MaxSector) { | |
1d41c216 N |
3062 | pr_warn("md/raid1:%s: reshape_position set but not supported\n", |
3063 | mdname(mddev)); | |
709ae487 N |
3064 | return -EIO; |
3065 | } | |
1da177e4 | 3066 | /* |
709ae487 N |
3067 | * copy the already verified devices into our private RAID1 |
3068 | * bookkeeping area. [whatever we allocate in run(), | |
afa0f557 | 3069 | * should be freed in raid1_free()] |
1da177e4 | 3070 | */ |
709ae487 N |
3071 | if (mddev->private == NULL) |
3072 | conf = setup_conf(mddev); | |
3073 | else | |
3074 | conf = mddev->private; | |
1da177e4 | 3075 | |
709ae487 N |
3076 | if (IS_ERR(conf)) |
3077 | return PTR_ERR(conf); | |
1da177e4 | 3078 | |
3deff1a7 | 3079 | if (mddev->queue) { |
5026d7a9 | 3080 | blk_queue_max_write_same_sectors(mddev->queue, 0); |
3deff1a7 CH |
3081 | blk_queue_max_write_zeroes_sectors(mddev->queue, 0); |
3082 | } | |
5026d7a9 | 3083 | |
dafb20fa | 3084 | rdev_for_each(rdev, mddev) { |
1ed7242e JB |
3085 | if (!mddev->gendisk) |
3086 | continue; | |
709ae487 N |
3087 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
3088 | rdev->data_offset << 9); | |
2ff8cc2c SL |
3089 | if (blk_queue_discard(bdev_get_queue(rdev->bdev))) |
3090 | discard_supported = true; | |
1da177e4 | 3091 | } |
191ea9b2 | 3092 | |
709ae487 N |
3093 | mddev->degraded = 0; |
3094 | for (i=0; i < conf->raid_disks; i++) | |
3095 | if (conf->mirrors[i].rdev == NULL || | |
3096 | !test_bit(In_sync, &conf->mirrors[i].rdev->flags) || | |
3097 | test_bit(Faulty, &conf->mirrors[i].rdev->flags)) | |
3098 | mddev->degraded++; | |
3099 | ||
3100 | if (conf->raid_disks - mddev->degraded == 1) | |
3101 | mddev->recovery_cp = MaxSector; | |
3102 | ||
8c6ac868 | 3103 | if (mddev->recovery_cp != MaxSector) |
1d41c216 N |
3104 | pr_info("md/raid1:%s: not clean -- starting background reconstruction\n", |
3105 | mdname(mddev)); | |
3106 | pr_info("md/raid1:%s: active with %d out of %d mirrors\n", | |
f72ffdd6 | 3107 | mdname(mddev), mddev->raid_disks - mddev->degraded, |
1da177e4 | 3108 | mddev->raid_disks); |
709ae487 | 3109 | |
1da177e4 LT |
3110 | /* |
3111 | * Ok, everything is just fine now | |
3112 | */ | |
709ae487 N |
3113 | mddev->thread = conf->thread; |
3114 | conf->thread = NULL; | |
3115 | mddev->private = conf; | |
46533ff7 | 3116 | set_bit(MD_FAILFAST_SUPPORTED, &mddev->flags); |
709ae487 | 3117 | |
1f403624 | 3118 | md_set_array_sectors(mddev, raid1_size(mddev, 0, 0)); |
1da177e4 | 3119 | |
1ed7242e | 3120 | if (mddev->queue) { |
2ff8cc2c SL |
3121 | if (discard_supported) |
3122 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, | |
3123 | mddev->queue); | |
3124 | else | |
3125 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, | |
3126 | mddev->queue); | |
1ed7242e | 3127 | } |
5220ea1e | 3128 | |
3129 | ret = md_integrity_register(mddev); | |
5aa61f42 N |
3130 | if (ret) { |
3131 | md_unregister_thread(&mddev->thread); | |
afa0f557 | 3132 | raid1_free(mddev, conf); |
5aa61f42 | 3133 | } |
5220ea1e | 3134 | return ret; |
1da177e4 LT |
3135 | } |
3136 | ||
afa0f557 | 3137 | static void raid1_free(struct mddev *mddev, void *priv) |
1da177e4 | 3138 | { |
afa0f557 | 3139 | struct r1conf *conf = priv; |
409c57f3 | 3140 | |
644df1a8 | 3141 | mempool_destroy(conf->r1bio_pool); |
990a8baf | 3142 | kfree(conf->mirrors); |
0fea7ed8 | 3143 | safe_put_page(conf->tmppage); |
990a8baf | 3144 | kfree(conf->poolinfo); |
fd76863e | 3145 | kfree(conf->nr_pending); |
3146 | kfree(conf->nr_waiting); | |
3147 | kfree(conf->nr_queued); | |
3148 | kfree(conf->barrier); | |
c230e7e5 N |
3149 | if (conf->bio_split) |
3150 | bioset_free(conf->bio_split); | |
1da177e4 | 3151 | kfree(conf); |
1da177e4 LT |
3152 | } |
3153 | ||
fd01b88c | 3154 | static int raid1_resize(struct mddev *mddev, sector_t sectors) |
1da177e4 LT |
3155 | { |
3156 | /* no resync is happening, and there is enough space | |
3157 | * on all devices, so we can resize. | |
3158 | * We need to make sure resync covers any new space. | |
3159 | * If the array is shrinking we should possibly wait until | |
3160 | * any io in the removed space completes, but it hardly seems | |
3161 | * worth it. | |
3162 | */ | |
a4a6125a N |
3163 | sector_t newsize = raid1_size(mddev, sectors, 0); |
3164 | if (mddev->external_size && | |
3165 | mddev->array_sectors > newsize) | |
b522adcd | 3166 | return -EINVAL; |
a4a6125a N |
3167 | if (mddev->bitmap) { |
3168 | int ret = bitmap_resize(mddev->bitmap, newsize, 0, 0); | |
3169 | if (ret) | |
3170 | return ret; | |
3171 | } | |
3172 | md_set_array_sectors(mddev, newsize); | |
b522adcd | 3173 | if (sectors > mddev->dev_sectors && |
b098636c | 3174 | mddev->recovery_cp > mddev->dev_sectors) { |
58c0fed4 | 3175 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
3176 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
3177 | } | |
b522adcd | 3178 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 3179 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
3180 | return 0; |
3181 | } | |
3182 | ||
fd01b88c | 3183 | static int raid1_reshape(struct mddev *mddev) |
1da177e4 LT |
3184 | { |
3185 | /* We need to: | |
3186 | * 1/ resize the r1bio_pool | |
3187 | * 2/ resize conf->mirrors | |
3188 | * | |
3189 | * We allocate a new r1bio_pool if we can. | |
3190 | * Then raise a device barrier and wait until all IO stops. | |
3191 | * Then resize conf->mirrors and swap in the new r1bio pool. | |
6ea9c07c N |
3192 | * |
3193 | * At the same time, we "pack" the devices so that all the missing | |
3194 | * devices have the higher raid_disk numbers. | |
1da177e4 LT |
3195 | */ |
3196 | mempool_t *newpool, *oldpool; | |
3197 | struct pool_info *newpoolinfo; | |
0eaf822c | 3198 | struct raid1_info *newmirrors; |
e8096360 | 3199 | struct r1conf *conf = mddev->private; |
63c70c4f | 3200 | int cnt, raid_disks; |
c04be0aa | 3201 | unsigned long flags; |
2214c260 | 3202 | int d, d2; |
1da177e4 | 3203 | |
63c70c4f | 3204 | /* Cannot change chunk_size, layout, or level */ |
664e7c41 | 3205 | if (mddev->chunk_sectors != mddev->new_chunk_sectors || |
63c70c4f N |
3206 | mddev->layout != mddev->new_layout || |
3207 | mddev->level != mddev->new_level) { | |
664e7c41 | 3208 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
63c70c4f N |
3209 | mddev->new_layout = mddev->layout; |
3210 | mddev->new_level = mddev->level; | |
3211 | return -EINVAL; | |
3212 | } | |
3213 | ||
2214c260 AP |
3214 | if (!mddev_is_clustered(mddev)) |
3215 | md_allow_write(mddev); | |
2a2275d6 | 3216 | |
63c70c4f N |
3217 | raid_disks = mddev->raid_disks + mddev->delta_disks; |
3218 | ||
6ea9c07c N |
3219 | if (raid_disks < conf->raid_disks) { |
3220 | cnt=0; | |
3221 | for (d= 0; d < conf->raid_disks; d++) | |
3222 | if (conf->mirrors[d].rdev) | |
3223 | cnt++; | |
3224 | if (cnt > raid_disks) | |
1da177e4 | 3225 | return -EBUSY; |
6ea9c07c | 3226 | } |
1da177e4 LT |
3227 | |
3228 | newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL); | |
3229 | if (!newpoolinfo) | |
3230 | return -ENOMEM; | |
3231 | newpoolinfo->mddev = mddev; | |
8f19ccb2 | 3232 | newpoolinfo->raid_disks = raid_disks * 2; |
1da177e4 LT |
3233 | |
3234 | newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, | |
3235 | r1bio_pool_free, newpoolinfo); | |
3236 | if (!newpool) { | |
3237 | kfree(newpoolinfo); | |
3238 | return -ENOMEM; | |
3239 | } | |
0eaf822c | 3240 | newmirrors = kzalloc(sizeof(struct raid1_info) * raid_disks * 2, |
8f19ccb2 | 3241 | GFP_KERNEL); |
1da177e4 LT |
3242 | if (!newmirrors) { |
3243 | kfree(newpoolinfo); | |
3244 | mempool_destroy(newpool); | |
3245 | return -ENOMEM; | |
3246 | } | |
1da177e4 | 3247 | |
e2d59925 | 3248 | freeze_array(conf, 0); |
1da177e4 LT |
3249 | |
3250 | /* ok, everything is stopped */ | |
3251 | oldpool = conf->r1bio_pool; | |
3252 | conf->r1bio_pool = newpool; | |
6ea9c07c | 3253 | |
a88aa786 | 3254 | for (d = d2 = 0; d < conf->raid_disks; d++) { |
3cb03002 | 3255 | struct md_rdev *rdev = conf->mirrors[d].rdev; |
a88aa786 | 3256 | if (rdev && rdev->raid_disk != d2) { |
36fad858 | 3257 | sysfs_unlink_rdev(mddev, rdev); |
a88aa786 | 3258 | rdev->raid_disk = d2; |
36fad858 NK |
3259 | sysfs_unlink_rdev(mddev, rdev); |
3260 | if (sysfs_link_rdev(mddev, rdev)) | |
1d41c216 N |
3261 | pr_warn("md/raid1:%s: cannot register rd%d\n", |
3262 | mdname(mddev), rdev->raid_disk); | |
6ea9c07c | 3263 | } |
a88aa786 N |
3264 | if (rdev) |
3265 | newmirrors[d2++].rdev = rdev; | |
3266 | } | |
1da177e4 LT |
3267 | kfree(conf->mirrors); |
3268 | conf->mirrors = newmirrors; | |
3269 | kfree(conf->poolinfo); | |
3270 | conf->poolinfo = newpoolinfo; | |
3271 | ||
c04be0aa | 3272 | spin_lock_irqsave(&conf->device_lock, flags); |
1da177e4 | 3273 | mddev->degraded += (raid_disks - conf->raid_disks); |
c04be0aa | 3274 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 3275 | conf->raid_disks = mddev->raid_disks = raid_disks; |
63c70c4f | 3276 | mddev->delta_disks = 0; |
1da177e4 | 3277 | |
e2d59925 | 3278 | unfreeze_array(conf); |
1da177e4 | 3279 | |
985ca973 | 3280 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
1da177e4 LT |
3281 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
3282 | md_wakeup_thread(mddev->thread); | |
3283 | ||
3284 | mempool_destroy(oldpool); | |
3285 | return 0; | |
3286 | } | |
3287 | ||
fd01b88c | 3288 | static void raid1_quiesce(struct mddev *mddev, int state) |
36fa3063 | 3289 | { |
e8096360 | 3290 | struct r1conf *conf = mddev->private; |
36fa3063 N |
3291 | |
3292 | switch(state) { | |
6eef4b21 N |
3293 | case 2: /* wake for suspend */ |
3294 | wake_up(&conf->wait_barrier); | |
3295 | break; | |
9e6603da | 3296 | case 1: |
07169fd4 | 3297 | freeze_array(conf, 0); |
36fa3063 | 3298 | break; |
9e6603da | 3299 | case 0: |
07169fd4 | 3300 | unfreeze_array(conf); |
36fa3063 N |
3301 | break; |
3302 | } | |
36fa3063 N |
3303 | } |
3304 | ||
fd01b88c | 3305 | static void *raid1_takeover(struct mddev *mddev) |
709ae487 N |
3306 | { |
3307 | /* raid1 can take over: | |
3308 | * raid5 with 2 devices, any layout or chunk size | |
3309 | */ | |
3310 | if (mddev->level == 5 && mddev->raid_disks == 2) { | |
e8096360 | 3311 | struct r1conf *conf; |
709ae487 N |
3312 | mddev->new_level = 1; |
3313 | mddev->new_layout = 0; | |
3314 | mddev->new_chunk_sectors = 0; | |
3315 | conf = setup_conf(mddev); | |
6995f0b2 | 3316 | if (!IS_ERR(conf)) { |
07169fd4 | 3317 | /* Array must appear to be quiesced */ |
3318 | conf->array_frozen = 1; | |
394ed8e4 SL |
3319 | mddev_clear_unsupported_flags(mddev, |
3320 | UNSUPPORTED_MDDEV_FLAGS); | |
6995f0b2 | 3321 | } |
709ae487 N |
3322 | return conf; |
3323 | } | |
3324 | return ERR_PTR(-EINVAL); | |
3325 | } | |
1da177e4 | 3326 | |
84fc4b56 | 3327 | static struct md_personality raid1_personality = |
1da177e4 LT |
3328 | { |
3329 | .name = "raid1", | |
2604b703 | 3330 | .level = 1, |
1da177e4 | 3331 | .owner = THIS_MODULE, |
849674e4 SL |
3332 | .make_request = raid1_make_request, |
3333 | .run = raid1_run, | |
afa0f557 | 3334 | .free = raid1_free, |
849674e4 SL |
3335 | .status = raid1_status, |
3336 | .error_handler = raid1_error, | |
1da177e4 LT |
3337 | .hot_add_disk = raid1_add_disk, |
3338 | .hot_remove_disk= raid1_remove_disk, | |
3339 | .spare_active = raid1_spare_active, | |
849674e4 | 3340 | .sync_request = raid1_sync_request, |
1da177e4 | 3341 | .resize = raid1_resize, |
80c3a6ce | 3342 | .size = raid1_size, |
63c70c4f | 3343 | .check_reshape = raid1_reshape, |
36fa3063 | 3344 | .quiesce = raid1_quiesce, |
709ae487 | 3345 | .takeover = raid1_takeover, |
5c675f83 | 3346 | .congested = raid1_congested, |
1da177e4 LT |
3347 | }; |
3348 | ||
3349 | static int __init raid_init(void) | |
3350 | { | |
2604b703 | 3351 | return register_md_personality(&raid1_personality); |
1da177e4 LT |
3352 | } |
3353 | ||
3354 | static void raid_exit(void) | |
3355 | { | |
2604b703 | 3356 | unregister_md_personality(&raid1_personality); |
1da177e4 LT |
3357 | } |
3358 | ||
3359 | module_init(raid_init); | |
3360 | module_exit(raid_exit); | |
3361 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 3362 | MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD"); |
1da177e4 | 3363 | MODULE_ALIAS("md-personality-3"); /* RAID1 */ |
d9d166c2 | 3364 | MODULE_ALIAS("md-raid1"); |
2604b703 | 3365 | MODULE_ALIAS("md-level-1"); |
34db0cd6 N |
3366 | |
3367 | module_param(max_queued_requests, int, S_IRUGO|S_IWUSR); |