]>
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> |
109e3765 | 40 | #include <trace/events/block.h> |
43b2e5d8 | 41 | #include "md.h" |
ef740c37 CH |
42 | #include "raid1.h" |
43 | #include "bitmap.h" | |
191ea9b2 | 44 | |
1da177e4 LT |
45 | /* |
46 | * Number of guaranteed r1bios in case of extreme VM load: | |
47 | */ | |
48 | #define NR_RAID1_BIOS 256 | |
49 | ||
473e87ce JB |
50 | /* when we get a read error on a read-only array, we redirect to another |
51 | * device without failing the first device, or trying to over-write to | |
52 | * correct the read error. To keep track of bad blocks on a per-bio | |
53 | * level, we store IO_BLOCKED in the appropriate 'bios' pointer | |
54 | */ | |
55 | #define IO_BLOCKED ((struct bio *)1) | |
56 | /* When we successfully write to a known bad-block, we need to remove the | |
57 | * bad-block marking which must be done from process context. So we record | |
58 | * the success by setting devs[n].bio to IO_MADE_GOOD | |
59 | */ | |
60 | #define IO_MADE_GOOD ((struct bio *)2) | |
61 | ||
62 | #define BIO_SPECIAL(bio) ((unsigned long)bio <= 2) | |
63 | ||
34db0cd6 N |
64 | /* When there are this many requests queue to be written by |
65 | * the raid1 thread, we become 'congested' to provide back-pressure | |
66 | * for writeback. | |
67 | */ | |
68 | static int max_queued_requests = 1024; | |
1da177e4 | 69 | |
79ef3a8a | 70 | static void allow_barrier(struct r1conf *conf, sector_t start_next_window, |
71 | sector_t bi_sector); | |
e8096360 | 72 | static void lower_barrier(struct r1conf *conf); |
1da177e4 | 73 | |
578b54ad N |
74 | #define raid1_log(md, fmt, args...) \ |
75 | do { if ((md)->queue) blk_add_trace_msg((md)->queue, "raid1 " fmt, ##args); } while (0) | |
76 | ||
dd0fc66f | 77 | static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 LT |
78 | { |
79 | struct pool_info *pi = data; | |
9f2c9d12 | 80 | int size = offsetof(struct r1bio, bios[pi->raid_disks]); |
1da177e4 LT |
81 | |
82 | /* allocate a r1bio with room for raid_disks entries in the bios array */ | |
7eaceacc | 83 | return kzalloc(size, gfp_flags); |
1da177e4 LT |
84 | } |
85 | ||
86 | static void r1bio_pool_free(void *r1_bio, void *data) | |
87 | { | |
88 | kfree(r1_bio); | |
89 | } | |
90 | ||
91 | #define RESYNC_BLOCK_SIZE (64*1024) | |
8e005f7c | 92 | #define RESYNC_DEPTH 32 |
1da177e4 LT |
93 | #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9) |
94 | #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE) | |
8e005f7c | 95 | #define RESYNC_WINDOW (RESYNC_BLOCK_SIZE * RESYNC_DEPTH) |
96 | #define RESYNC_WINDOW_SECTORS (RESYNC_WINDOW >> 9) | |
c40f341f GR |
97 | #define CLUSTER_RESYNC_WINDOW (16 * RESYNC_WINDOW) |
98 | #define CLUSTER_RESYNC_WINDOW_SECTORS (CLUSTER_RESYNC_WINDOW >> 9) | |
8e005f7c | 99 | #define NEXT_NORMALIO_DISTANCE (3 * RESYNC_WINDOW_SECTORS) |
1da177e4 | 100 | |
dd0fc66f | 101 | static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 LT |
102 | { |
103 | struct pool_info *pi = data; | |
9f2c9d12 | 104 | struct r1bio *r1_bio; |
1da177e4 | 105 | struct bio *bio; |
da1aab3d | 106 | int need_pages; |
1da177e4 LT |
107 | int i, j; |
108 | ||
109 | r1_bio = r1bio_pool_alloc(gfp_flags, pi); | |
7eaceacc | 110 | if (!r1_bio) |
1da177e4 | 111 | return NULL; |
1da177e4 LT |
112 | |
113 | /* | |
114 | * Allocate bios : 1 for reading, n-1 for writing | |
115 | */ | |
116 | for (j = pi->raid_disks ; j-- ; ) { | |
6746557f | 117 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); |
1da177e4 LT |
118 | if (!bio) |
119 | goto out_free_bio; | |
120 | r1_bio->bios[j] = bio; | |
121 | } | |
122 | /* | |
123 | * Allocate RESYNC_PAGES data pages and attach them to | |
d11c171e N |
124 | * the first bio. |
125 | * If this is a user-requested check/repair, allocate | |
126 | * RESYNC_PAGES for each bio. | |
1da177e4 | 127 | */ |
d11c171e | 128 | if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) |
da1aab3d | 129 | need_pages = pi->raid_disks; |
d11c171e | 130 | else |
da1aab3d N |
131 | need_pages = 1; |
132 | for (j = 0; j < need_pages; j++) { | |
d11c171e | 133 | bio = r1_bio->bios[j]; |
a0787606 | 134 | bio->bi_vcnt = RESYNC_PAGES; |
d11c171e | 135 | |
a0787606 | 136 | if (bio_alloc_pages(bio, gfp_flags)) |
da1aab3d | 137 | goto out_free_pages; |
d11c171e N |
138 | } |
139 | /* If not user-requests, copy the page pointers to all bios */ | |
140 | if (!test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) { | |
141 | for (i=0; i<RESYNC_PAGES ; i++) | |
142 | for (j=1; j<pi->raid_disks; j++) | |
143 | r1_bio->bios[j]->bi_io_vec[i].bv_page = | |
144 | r1_bio->bios[0]->bi_io_vec[i].bv_page; | |
1da177e4 LT |
145 | } |
146 | ||
147 | r1_bio->master_bio = NULL; | |
148 | ||
149 | return r1_bio; | |
150 | ||
da1aab3d | 151 | out_free_pages: |
491221f8 GJ |
152 | while (--j >= 0) |
153 | bio_free_pages(r1_bio->bios[j]); | |
da1aab3d | 154 | |
1da177e4 | 155 | out_free_bio: |
8f19ccb2 | 156 | while (++j < pi->raid_disks) |
1da177e4 LT |
157 | bio_put(r1_bio->bios[j]); |
158 | r1bio_pool_free(r1_bio, data); | |
159 | return NULL; | |
160 | } | |
161 | ||
162 | static void r1buf_pool_free(void *__r1_bio, void *data) | |
163 | { | |
164 | struct pool_info *pi = data; | |
d11c171e | 165 | int i,j; |
9f2c9d12 | 166 | struct r1bio *r1bio = __r1_bio; |
1da177e4 | 167 | |
d11c171e N |
168 | for (i = 0; i < RESYNC_PAGES; i++) |
169 | for (j = pi->raid_disks; j-- ;) { | |
170 | if (j == 0 || | |
171 | r1bio->bios[j]->bi_io_vec[i].bv_page != | |
172 | r1bio->bios[0]->bi_io_vec[i].bv_page) | |
1345b1d8 | 173 | safe_put_page(r1bio->bios[j]->bi_io_vec[i].bv_page); |
d11c171e | 174 | } |
1da177e4 LT |
175 | for (i=0 ; i < pi->raid_disks; i++) |
176 | bio_put(r1bio->bios[i]); | |
177 | ||
178 | r1bio_pool_free(r1bio, data); | |
179 | } | |
180 | ||
e8096360 | 181 | static void put_all_bios(struct r1conf *conf, struct r1bio *r1_bio) |
1da177e4 LT |
182 | { |
183 | int i; | |
184 | ||
8f19ccb2 | 185 | for (i = 0; i < conf->raid_disks * 2; i++) { |
1da177e4 | 186 | struct bio **bio = r1_bio->bios + i; |
4367af55 | 187 | if (!BIO_SPECIAL(*bio)) |
1da177e4 LT |
188 | bio_put(*bio); |
189 | *bio = NULL; | |
190 | } | |
191 | } | |
192 | ||
9f2c9d12 | 193 | static void free_r1bio(struct r1bio *r1_bio) |
1da177e4 | 194 | { |
e8096360 | 195 | struct r1conf *conf = r1_bio->mddev->private; |
1da177e4 | 196 | |
1da177e4 LT |
197 | put_all_bios(conf, r1_bio); |
198 | mempool_free(r1_bio, conf->r1bio_pool); | |
199 | } | |
200 | ||
9f2c9d12 | 201 | static void put_buf(struct r1bio *r1_bio) |
1da177e4 | 202 | { |
e8096360 | 203 | struct r1conf *conf = r1_bio->mddev->private; |
3e198f78 N |
204 | int i; |
205 | ||
8f19ccb2 | 206 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3e198f78 N |
207 | struct bio *bio = r1_bio->bios[i]; |
208 | if (bio->bi_end_io) | |
209 | rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev); | |
210 | } | |
1da177e4 LT |
211 | |
212 | mempool_free(r1_bio, conf->r1buf_pool); | |
213 | ||
17999be4 | 214 | lower_barrier(conf); |
1da177e4 LT |
215 | } |
216 | ||
9f2c9d12 | 217 | static void reschedule_retry(struct r1bio *r1_bio) |
1da177e4 LT |
218 | { |
219 | unsigned long flags; | |
fd01b88c | 220 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 221 | struct r1conf *conf = mddev->private; |
1da177e4 LT |
222 | |
223 | spin_lock_irqsave(&conf->device_lock, flags); | |
224 | list_add(&r1_bio->retry_list, &conf->retry_list); | |
ddaf22ab | 225 | conf->nr_queued ++; |
1da177e4 LT |
226 | spin_unlock_irqrestore(&conf->device_lock, flags); |
227 | ||
17999be4 | 228 | wake_up(&conf->wait_barrier); |
1da177e4 LT |
229 | md_wakeup_thread(mddev->thread); |
230 | } | |
231 | ||
232 | /* | |
233 | * raid_end_bio_io() is called when we have finished servicing a mirrored | |
234 | * operation and are ready to return a success/failure code to the buffer | |
235 | * cache layer. | |
236 | */ | |
9f2c9d12 | 237 | static void call_bio_endio(struct r1bio *r1_bio) |
d2eb35ac N |
238 | { |
239 | struct bio *bio = r1_bio->master_bio; | |
240 | int done; | |
e8096360 | 241 | struct r1conf *conf = r1_bio->mddev->private; |
79ef3a8a | 242 | sector_t start_next_window = r1_bio->start_next_window; |
4f024f37 | 243 | sector_t bi_sector = bio->bi_iter.bi_sector; |
d2eb35ac N |
244 | |
245 | if (bio->bi_phys_segments) { | |
246 | unsigned long flags; | |
247 | spin_lock_irqsave(&conf->device_lock, flags); | |
248 | bio->bi_phys_segments--; | |
249 | done = (bio->bi_phys_segments == 0); | |
250 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
79ef3a8a | 251 | /* |
252 | * make_request() might be waiting for | |
253 | * bi_phys_segments to decrease | |
254 | */ | |
255 | wake_up(&conf->wait_barrier); | |
d2eb35ac N |
256 | } else |
257 | done = 1; | |
258 | ||
259 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) | |
4246a0b6 CH |
260 | bio->bi_error = -EIO; |
261 | ||
d2eb35ac | 262 | if (done) { |
4246a0b6 | 263 | bio_endio(bio); |
d2eb35ac N |
264 | /* |
265 | * Wake up any possible resync thread that waits for the device | |
266 | * to go idle. | |
267 | */ | |
79ef3a8a | 268 | allow_barrier(conf, start_next_window, bi_sector); |
d2eb35ac N |
269 | } |
270 | } | |
271 | ||
9f2c9d12 | 272 | static void raid_end_bio_io(struct r1bio *r1_bio) |
1da177e4 LT |
273 | { |
274 | struct bio *bio = r1_bio->master_bio; | |
275 | ||
4b6d287f N |
276 | /* if nobody has done the final endio yet, do it now */ |
277 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { | |
36a4e1fe N |
278 | pr_debug("raid1: sync end %s on sectors %llu-%llu\n", |
279 | (bio_data_dir(bio) == WRITE) ? "write" : "read", | |
4f024f37 KO |
280 | (unsigned long long) bio->bi_iter.bi_sector, |
281 | (unsigned long long) bio_end_sector(bio) - 1); | |
4b6d287f | 282 | |
d2eb35ac | 283 | call_bio_endio(r1_bio); |
4b6d287f | 284 | } |
1da177e4 LT |
285 | free_r1bio(r1_bio); |
286 | } | |
287 | ||
288 | /* | |
289 | * Update disk head position estimator based on IRQ completion info. | |
290 | */ | |
9f2c9d12 | 291 | static inline void update_head_pos(int disk, struct r1bio *r1_bio) |
1da177e4 | 292 | { |
e8096360 | 293 | struct r1conf *conf = r1_bio->mddev->private; |
1da177e4 LT |
294 | |
295 | conf->mirrors[disk].head_position = | |
296 | r1_bio->sector + (r1_bio->sectors); | |
297 | } | |
298 | ||
ba3ae3be NK |
299 | /* |
300 | * Find the disk number which triggered given bio | |
301 | */ | |
9f2c9d12 | 302 | static int find_bio_disk(struct r1bio *r1_bio, struct bio *bio) |
ba3ae3be NK |
303 | { |
304 | int mirror; | |
30194636 N |
305 | struct r1conf *conf = r1_bio->mddev->private; |
306 | int raid_disks = conf->raid_disks; | |
ba3ae3be | 307 | |
8f19ccb2 | 308 | for (mirror = 0; mirror < raid_disks * 2; mirror++) |
ba3ae3be NK |
309 | if (r1_bio->bios[mirror] == bio) |
310 | break; | |
311 | ||
8f19ccb2 | 312 | BUG_ON(mirror == raid_disks * 2); |
ba3ae3be NK |
313 | update_head_pos(mirror, r1_bio); |
314 | ||
315 | return mirror; | |
316 | } | |
317 | ||
4246a0b6 | 318 | static void raid1_end_read_request(struct bio *bio) |
1da177e4 | 319 | { |
4246a0b6 | 320 | int uptodate = !bio->bi_error; |
9f2c9d12 | 321 | struct r1bio *r1_bio = bio->bi_private; |
e8096360 | 322 | struct r1conf *conf = r1_bio->mddev->private; |
e5872d58 | 323 | struct md_rdev *rdev = conf->mirrors[r1_bio->read_disk].rdev; |
1da177e4 | 324 | |
1da177e4 LT |
325 | /* |
326 | * this branch is our 'one mirror IO has finished' event handler: | |
327 | */ | |
e5872d58 | 328 | update_head_pos(r1_bio->read_disk, r1_bio); |
ddaf22ab | 329 | |
dd00a99e N |
330 | if (uptodate) |
331 | set_bit(R1BIO_Uptodate, &r1_bio->state); | |
2e52d449 N |
332 | else if (test_bit(FailFast, &rdev->flags) && |
333 | test_bit(R1BIO_FailFast, &r1_bio->state)) | |
334 | /* This was a fail-fast read so we definitely | |
335 | * want to retry */ | |
336 | ; | |
dd00a99e N |
337 | else { |
338 | /* If all other devices have failed, we want to return | |
339 | * the error upwards rather than fail the last device. | |
340 | * Here we redefine "uptodate" to mean "Don't want to retry" | |
1da177e4 | 341 | */ |
dd00a99e N |
342 | unsigned long flags; |
343 | spin_lock_irqsave(&conf->device_lock, flags); | |
344 | if (r1_bio->mddev->degraded == conf->raid_disks || | |
345 | (r1_bio->mddev->degraded == conf->raid_disks-1 && | |
e5872d58 | 346 | test_bit(In_sync, &rdev->flags))) |
dd00a99e N |
347 | uptodate = 1; |
348 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
349 | } | |
1da177e4 | 350 | |
7ad4d4a6 | 351 | if (uptodate) { |
1da177e4 | 352 | raid_end_bio_io(r1_bio); |
e5872d58 | 353 | rdev_dec_pending(rdev, conf->mddev); |
7ad4d4a6 | 354 | } else { |
1da177e4 LT |
355 | /* |
356 | * oops, read error: | |
357 | */ | |
358 | char b[BDEVNAME_SIZE]; | |
1d41c216 N |
359 | pr_err_ratelimited("md/raid1:%s: %s: rescheduling sector %llu\n", |
360 | mdname(conf->mddev), | |
361 | bdevname(rdev->bdev, b), | |
362 | (unsigned long long)r1_bio->sector); | |
d2eb35ac | 363 | set_bit(R1BIO_ReadError, &r1_bio->state); |
1da177e4 | 364 | reschedule_retry(r1_bio); |
7ad4d4a6 | 365 | /* don't drop the reference on read_disk yet */ |
1da177e4 | 366 | } |
1da177e4 LT |
367 | } |
368 | ||
9f2c9d12 | 369 | static void close_write(struct r1bio *r1_bio) |
cd5ff9a1 N |
370 | { |
371 | /* it really is the end of this request */ | |
372 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { | |
373 | /* free extra copy of the data pages */ | |
374 | int i = r1_bio->behind_page_count; | |
375 | while (i--) | |
376 | safe_put_page(r1_bio->behind_bvecs[i].bv_page); | |
377 | kfree(r1_bio->behind_bvecs); | |
378 | r1_bio->behind_bvecs = NULL; | |
379 | } | |
380 | /* clear the bitmap if all writes complete successfully */ | |
381 | bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector, | |
382 | r1_bio->sectors, | |
383 | !test_bit(R1BIO_Degraded, &r1_bio->state), | |
384 | test_bit(R1BIO_BehindIO, &r1_bio->state)); | |
385 | md_write_end(r1_bio->mddev); | |
386 | } | |
387 | ||
9f2c9d12 | 388 | static void r1_bio_write_done(struct r1bio *r1_bio) |
4e78064f | 389 | { |
cd5ff9a1 N |
390 | if (!atomic_dec_and_test(&r1_bio->remaining)) |
391 | return; | |
392 | ||
393 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
394 | reschedule_retry(r1_bio); | |
395 | else { | |
396 | close_write(r1_bio); | |
4367af55 N |
397 | if (test_bit(R1BIO_MadeGood, &r1_bio->state)) |
398 | reschedule_retry(r1_bio); | |
399 | else | |
400 | raid_end_bio_io(r1_bio); | |
4e78064f N |
401 | } |
402 | } | |
403 | ||
4246a0b6 | 404 | static void raid1_end_write_request(struct bio *bio) |
1da177e4 | 405 | { |
9f2c9d12 | 406 | struct r1bio *r1_bio = bio->bi_private; |
e5872d58 | 407 | int behind = test_bit(R1BIO_BehindIO, &r1_bio->state); |
e8096360 | 408 | struct r1conf *conf = r1_bio->mddev->private; |
04b857f7 | 409 | struct bio *to_put = NULL; |
e5872d58 N |
410 | int mirror = find_bio_disk(r1_bio, bio); |
411 | struct md_rdev *rdev = conf->mirrors[mirror].rdev; | |
e3f948cd SL |
412 | bool discard_error; |
413 | ||
414 | discard_error = bio->bi_error && bio_op(bio) == REQ_OP_DISCARD; | |
1da177e4 | 415 | |
e9c7469b TH |
416 | /* |
417 | * 'one mirror IO has finished' event handler: | |
418 | */ | |
e3f948cd | 419 | if (bio->bi_error && !discard_error) { |
e5872d58 N |
420 | set_bit(WriteErrorSeen, &rdev->flags); |
421 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) | |
19d67169 N |
422 | set_bit(MD_RECOVERY_NEEDED, & |
423 | conf->mddev->recovery); | |
424 | ||
212e7eb7 N |
425 | if (test_bit(FailFast, &rdev->flags) && |
426 | (bio->bi_opf & MD_FAILFAST) && | |
427 | /* We never try FailFast to WriteMostly devices */ | |
428 | !test_bit(WriteMostly, &rdev->flags)) { | |
429 | md_error(r1_bio->mddev, rdev); | |
430 | if (!test_bit(Faulty, &rdev->flags)) | |
431 | /* This is the only remaining device, | |
432 | * We need to retry the write without | |
433 | * FailFast | |
434 | */ | |
435 | set_bit(R1BIO_WriteError, &r1_bio->state); | |
436 | else { | |
437 | /* Finished with this branch */ | |
438 | r1_bio->bios[mirror] = NULL; | |
439 | to_put = bio; | |
440 | } | |
441 | } else | |
442 | set_bit(R1BIO_WriteError, &r1_bio->state); | |
4367af55 | 443 | } else { |
1da177e4 | 444 | /* |
e9c7469b TH |
445 | * Set R1BIO_Uptodate in our master bio, so that we |
446 | * will return a good error code for to the higher | |
447 | * levels even if IO on some other mirrored buffer | |
448 | * fails. | |
449 | * | |
450 | * The 'master' represents the composite IO operation | |
451 | * to user-side. So if something waits for IO, then it | |
452 | * will wait for the 'master' bio. | |
1da177e4 | 453 | */ |
4367af55 N |
454 | sector_t first_bad; |
455 | int bad_sectors; | |
456 | ||
cd5ff9a1 N |
457 | r1_bio->bios[mirror] = NULL; |
458 | to_put = bio; | |
3056e3ae AL |
459 | /* |
460 | * Do not set R1BIO_Uptodate if the current device is | |
461 | * rebuilding or Faulty. This is because we cannot use | |
462 | * such device for properly reading the data back (we could | |
463 | * potentially use it, if the current write would have felt | |
464 | * before rdev->recovery_offset, but for simplicity we don't | |
465 | * check this here. | |
466 | */ | |
e5872d58 N |
467 | if (test_bit(In_sync, &rdev->flags) && |
468 | !test_bit(Faulty, &rdev->flags)) | |
3056e3ae | 469 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
e9c7469b | 470 | |
4367af55 | 471 | /* Maybe we can clear some bad blocks. */ |
e5872d58 | 472 | if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors, |
e3f948cd | 473 | &first_bad, &bad_sectors) && !discard_error) { |
4367af55 N |
474 | r1_bio->bios[mirror] = IO_MADE_GOOD; |
475 | set_bit(R1BIO_MadeGood, &r1_bio->state); | |
476 | } | |
477 | } | |
478 | ||
e9c7469b | 479 | if (behind) { |
e5872d58 | 480 | if (test_bit(WriteMostly, &rdev->flags)) |
e9c7469b TH |
481 | atomic_dec(&r1_bio->behind_remaining); |
482 | ||
483 | /* | |
484 | * In behind mode, we ACK the master bio once the I/O | |
485 | * has safely reached all non-writemostly | |
486 | * disks. Setting the Returned bit ensures that this | |
487 | * gets done only once -- we don't ever want to return | |
488 | * -EIO here, instead we'll wait | |
489 | */ | |
490 | if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) && | |
491 | test_bit(R1BIO_Uptodate, &r1_bio->state)) { | |
492 | /* Maybe we can return now */ | |
493 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { | |
494 | struct bio *mbio = r1_bio->master_bio; | |
36a4e1fe N |
495 | pr_debug("raid1: behind end write sectors" |
496 | " %llu-%llu\n", | |
4f024f37 KO |
497 | (unsigned long long) mbio->bi_iter.bi_sector, |
498 | (unsigned long long) bio_end_sector(mbio) - 1); | |
d2eb35ac | 499 | call_bio_endio(r1_bio); |
4b6d287f N |
500 | } |
501 | } | |
502 | } | |
4367af55 | 503 | if (r1_bio->bios[mirror] == NULL) |
e5872d58 | 504 | rdev_dec_pending(rdev, conf->mddev); |
e9c7469b | 505 | |
1da177e4 | 506 | /* |
1da177e4 LT |
507 | * Let's see if all mirrored write operations have finished |
508 | * already. | |
509 | */ | |
af6d7b76 | 510 | r1_bio_write_done(r1_bio); |
c70810b3 | 511 | |
04b857f7 N |
512 | if (to_put) |
513 | bio_put(to_put); | |
1da177e4 LT |
514 | } |
515 | ||
1da177e4 LT |
516 | /* |
517 | * This routine returns the disk from which the requested read should | |
518 | * be done. There is a per-array 'next expected sequential IO' sector | |
519 | * number - if this matches on the next IO then we use the last disk. | |
520 | * There is also a per-disk 'last know head position' sector that is | |
521 | * maintained from IRQ contexts, both the normal and the resync IO | |
522 | * completion handlers update this position correctly. If there is no | |
523 | * perfect sequential match then we pick the disk whose head is closest. | |
524 | * | |
525 | * If there are 2 mirrors in the same 2 devices, performance degrades | |
526 | * because position is mirror, not device based. | |
527 | * | |
528 | * The rdev for the device selected will have nr_pending incremented. | |
529 | */ | |
e8096360 | 530 | static int read_balance(struct r1conf *conf, struct r1bio *r1_bio, int *max_sectors) |
1da177e4 | 531 | { |
af3a2cd6 | 532 | const sector_t this_sector = r1_bio->sector; |
d2eb35ac N |
533 | int sectors; |
534 | int best_good_sectors; | |
9dedf603 SL |
535 | int best_disk, best_dist_disk, best_pending_disk; |
536 | int has_nonrot_disk; | |
be4d3280 | 537 | int disk; |
76073054 | 538 | sector_t best_dist; |
9dedf603 | 539 | unsigned int min_pending; |
3cb03002 | 540 | struct md_rdev *rdev; |
f3ac8bf7 | 541 | int choose_first; |
12cee5a8 | 542 | int choose_next_idle; |
1da177e4 LT |
543 | |
544 | rcu_read_lock(); | |
545 | /* | |
8ddf9efe | 546 | * Check if we can balance. We can balance on the whole |
1da177e4 LT |
547 | * device if no resync is going on, or below the resync window. |
548 | * We take the first readable disk when above the resync window. | |
549 | */ | |
550 | retry: | |
d2eb35ac | 551 | sectors = r1_bio->sectors; |
76073054 | 552 | best_disk = -1; |
9dedf603 | 553 | best_dist_disk = -1; |
76073054 | 554 | best_dist = MaxSector; |
9dedf603 SL |
555 | best_pending_disk = -1; |
556 | min_pending = UINT_MAX; | |
d2eb35ac | 557 | best_good_sectors = 0; |
9dedf603 | 558 | has_nonrot_disk = 0; |
12cee5a8 | 559 | choose_next_idle = 0; |
2e52d449 | 560 | clear_bit(R1BIO_FailFast, &r1_bio->state); |
d2eb35ac | 561 | |
7d49ffcf GR |
562 | if ((conf->mddev->recovery_cp < this_sector + sectors) || |
563 | (mddev_is_clustered(conf->mddev) && | |
90382ed9 | 564 | md_cluster_ops->area_resyncing(conf->mddev, READ, this_sector, |
7d49ffcf GR |
565 | this_sector + sectors))) |
566 | choose_first = 1; | |
567 | else | |
568 | choose_first = 0; | |
1da177e4 | 569 | |
be4d3280 | 570 | for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) { |
76073054 | 571 | sector_t dist; |
d2eb35ac N |
572 | sector_t first_bad; |
573 | int bad_sectors; | |
9dedf603 | 574 | unsigned int pending; |
12cee5a8 | 575 | bool nonrot; |
d2eb35ac | 576 | |
f3ac8bf7 N |
577 | rdev = rcu_dereference(conf->mirrors[disk].rdev); |
578 | if (r1_bio->bios[disk] == IO_BLOCKED | |
579 | || rdev == NULL | |
76073054 | 580 | || test_bit(Faulty, &rdev->flags)) |
f3ac8bf7 | 581 | continue; |
76073054 N |
582 | if (!test_bit(In_sync, &rdev->flags) && |
583 | rdev->recovery_offset < this_sector + sectors) | |
1da177e4 | 584 | continue; |
76073054 N |
585 | if (test_bit(WriteMostly, &rdev->flags)) { |
586 | /* Don't balance among write-mostly, just | |
587 | * use the first as a last resort */ | |
d1901ef0 | 588 | if (best_dist_disk < 0) { |
307729c8 N |
589 | if (is_badblock(rdev, this_sector, sectors, |
590 | &first_bad, &bad_sectors)) { | |
816b0acf | 591 | if (first_bad <= this_sector) |
307729c8 N |
592 | /* Cannot use this */ |
593 | continue; | |
594 | best_good_sectors = first_bad - this_sector; | |
595 | } else | |
596 | best_good_sectors = sectors; | |
d1901ef0 TH |
597 | best_dist_disk = disk; |
598 | best_pending_disk = disk; | |
307729c8 | 599 | } |
76073054 N |
600 | continue; |
601 | } | |
602 | /* This is a reasonable device to use. It might | |
603 | * even be best. | |
604 | */ | |
d2eb35ac N |
605 | if (is_badblock(rdev, this_sector, sectors, |
606 | &first_bad, &bad_sectors)) { | |
607 | if (best_dist < MaxSector) | |
608 | /* already have a better device */ | |
609 | continue; | |
610 | if (first_bad <= this_sector) { | |
611 | /* cannot read here. If this is the 'primary' | |
612 | * device, then we must not read beyond | |
613 | * bad_sectors from another device.. | |
614 | */ | |
615 | bad_sectors -= (this_sector - first_bad); | |
616 | if (choose_first && sectors > bad_sectors) | |
617 | sectors = bad_sectors; | |
618 | if (best_good_sectors > sectors) | |
619 | best_good_sectors = sectors; | |
620 | ||
621 | } else { | |
622 | sector_t good_sectors = first_bad - this_sector; | |
623 | if (good_sectors > best_good_sectors) { | |
624 | best_good_sectors = good_sectors; | |
625 | best_disk = disk; | |
626 | } | |
627 | if (choose_first) | |
628 | break; | |
629 | } | |
630 | continue; | |
631 | } else | |
632 | best_good_sectors = sectors; | |
633 | ||
2e52d449 N |
634 | if (best_disk >= 0) |
635 | /* At least two disks to choose from so failfast is OK */ | |
636 | set_bit(R1BIO_FailFast, &r1_bio->state); | |
637 | ||
12cee5a8 SL |
638 | nonrot = blk_queue_nonrot(bdev_get_queue(rdev->bdev)); |
639 | has_nonrot_disk |= nonrot; | |
9dedf603 | 640 | pending = atomic_read(&rdev->nr_pending); |
76073054 | 641 | dist = abs(this_sector - conf->mirrors[disk].head_position); |
12cee5a8 | 642 | if (choose_first) { |
76073054 | 643 | best_disk = disk; |
1da177e4 LT |
644 | break; |
645 | } | |
12cee5a8 SL |
646 | /* Don't change to another disk for sequential reads */ |
647 | if (conf->mirrors[disk].next_seq_sect == this_sector | |
648 | || dist == 0) { | |
649 | int opt_iosize = bdev_io_opt(rdev->bdev) >> 9; | |
650 | struct raid1_info *mirror = &conf->mirrors[disk]; | |
651 | ||
652 | best_disk = disk; | |
653 | /* | |
654 | * If buffered sequential IO size exceeds optimal | |
655 | * iosize, check if there is idle disk. If yes, choose | |
656 | * the idle disk. read_balance could already choose an | |
657 | * idle disk before noticing it's a sequential IO in | |
658 | * this disk. This doesn't matter because this disk | |
659 | * will idle, next time it will be utilized after the | |
660 | * first disk has IO size exceeds optimal iosize. In | |
661 | * this way, iosize of the first disk will be optimal | |
662 | * iosize at least. iosize of the second disk might be | |
663 | * small, but not a big deal since when the second disk | |
664 | * starts IO, the first disk is likely still busy. | |
665 | */ | |
666 | if (nonrot && opt_iosize > 0 && | |
667 | mirror->seq_start != MaxSector && | |
668 | mirror->next_seq_sect > opt_iosize && | |
669 | mirror->next_seq_sect - opt_iosize >= | |
670 | mirror->seq_start) { | |
671 | choose_next_idle = 1; | |
672 | continue; | |
673 | } | |
674 | break; | |
675 | } | |
12cee5a8 SL |
676 | |
677 | if (choose_next_idle) | |
678 | continue; | |
9dedf603 SL |
679 | |
680 | if (min_pending > pending) { | |
681 | min_pending = pending; | |
682 | best_pending_disk = disk; | |
683 | } | |
684 | ||
76073054 N |
685 | if (dist < best_dist) { |
686 | best_dist = dist; | |
9dedf603 | 687 | best_dist_disk = disk; |
1da177e4 | 688 | } |
f3ac8bf7 | 689 | } |
1da177e4 | 690 | |
9dedf603 SL |
691 | /* |
692 | * If all disks are rotational, choose the closest disk. If any disk is | |
693 | * non-rotational, choose the disk with less pending request even the | |
694 | * disk is rotational, which might/might not be optimal for raids with | |
695 | * mixed ratation/non-rotational disks depending on workload. | |
696 | */ | |
697 | if (best_disk == -1) { | |
2e52d449 | 698 | if (has_nonrot_disk || min_pending == 0) |
9dedf603 SL |
699 | best_disk = best_pending_disk; |
700 | else | |
701 | best_disk = best_dist_disk; | |
702 | } | |
703 | ||
76073054 N |
704 | if (best_disk >= 0) { |
705 | rdev = rcu_dereference(conf->mirrors[best_disk].rdev); | |
8ddf9efe N |
706 | if (!rdev) |
707 | goto retry; | |
708 | atomic_inc(&rdev->nr_pending); | |
d2eb35ac | 709 | sectors = best_good_sectors; |
12cee5a8 SL |
710 | |
711 | if (conf->mirrors[best_disk].next_seq_sect != this_sector) | |
712 | conf->mirrors[best_disk].seq_start = this_sector; | |
713 | ||
be4d3280 | 714 | conf->mirrors[best_disk].next_seq_sect = this_sector + sectors; |
1da177e4 LT |
715 | } |
716 | rcu_read_unlock(); | |
d2eb35ac | 717 | *max_sectors = sectors; |
1da177e4 | 718 | |
76073054 | 719 | return best_disk; |
1da177e4 LT |
720 | } |
721 | ||
5c675f83 | 722 | static int raid1_congested(struct mddev *mddev, int bits) |
0d129228 | 723 | { |
e8096360 | 724 | struct r1conf *conf = mddev->private; |
0d129228 N |
725 | int i, ret = 0; |
726 | ||
4452226e | 727 | if ((bits & (1 << WB_async_congested)) && |
34db0cd6 N |
728 | conf->pending_count >= max_queued_requests) |
729 | return 1; | |
730 | ||
0d129228 | 731 | rcu_read_lock(); |
f53e29fc | 732 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3cb03002 | 733 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
0d129228 | 734 | if (rdev && !test_bit(Faulty, &rdev->flags)) { |
165125e1 | 735 | struct request_queue *q = bdev_get_queue(rdev->bdev); |
0d129228 | 736 | |
1ed7242e JB |
737 | BUG_ON(!q); |
738 | ||
0d129228 N |
739 | /* Note the '|| 1' - when read_balance prefers |
740 | * non-congested targets, it can be removed | |
741 | */ | |
4452226e | 742 | if ((bits & (1 << WB_async_congested)) || 1) |
0d129228 N |
743 | ret |= bdi_congested(&q->backing_dev_info, bits); |
744 | else | |
745 | ret &= bdi_congested(&q->backing_dev_info, bits); | |
746 | } | |
747 | } | |
748 | rcu_read_unlock(); | |
749 | return ret; | |
750 | } | |
0d129228 | 751 | |
e8096360 | 752 | static void flush_pending_writes(struct r1conf *conf) |
a35e63ef N |
753 | { |
754 | /* Any writes that have been queued but are awaiting | |
755 | * bitmap updates get flushed here. | |
a35e63ef | 756 | */ |
a35e63ef N |
757 | spin_lock_irq(&conf->device_lock); |
758 | ||
759 | if (conf->pending_bio_list.head) { | |
760 | struct bio *bio; | |
761 | bio = bio_list_get(&conf->pending_bio_list); | |
34db0cd6 | 762 | conf->pending_count = 0; |
a35e63ef N |
763 | spin_unlock_irq(&conf->device_lock); |
764 | /* flush any pending bitmap writes to | |
765 | * disk before proceeding w/ I/O */ | |
766 | bitmap_unplug(conf->mddev->bitmap); | |
34db0cd6 | 767 | wake_up(&conf->wait_barrier); |
a35e63ef N |
768 | |
769 | while (bio) { /* submit pending writes */ | |
770 | struct bio *next = bio->bi_next; | |
5e2c7a36 | 771 | struct md_rdev *rdev = (void*)bio->bi_bdev; |
a35e63ef | 772 | bio->bi_next = NULL; |
5e2c7a36 N |
773 | bio->bi_bdev = rdev->bdev; |
774 | if (test_bit(Faulty, &rdev->flags)) { | |
775 | bio->bi_error = -EIO; | |
776 | bio_endio(bio); | |
777 | } else if (unlikely((bio_op(bio) == REQ_OP_DISCARD) && | |
778 | !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) | |
2ff8cc2c | 779 | /* Just ignore it */ |
4246a0b6 | 780 | bio_endio(bio); |
2ff8cc2c SL |
781 | else |
782 | generic_make_request(bio); | |
a35e63ef N |
783 | bio = next; |
784 | } | |
a35e63ef N |
785 | } else |
786 | spin_unlock_irq(&conf->device_lock); | |
7eaceacc JA |
787 | } |
788 | ||
17999be4 N |
789 | /* Barriers.... |
790 | * Sometimes we need to suspend IO while we do something else, | |
791 | * either some resync/recovery, or reconfigure the array. | |
792 | * To do this we raise a 'barrier'. | |
793 | * The 'barrier' is a counter that can be raised multiple times | |
794 | * to count how many activities are happening which preclude | |
795 | * normal IO. | |
796 | * We can only raise the barrier if there is no pending IO. | |
797 | * i.e. if nr_pending == 0. | |
798 | * We choose only to raise the barrier if no-one is waiting for the | |
799 | * barrier to go down. This means that as soon as an IO request | |
800 | * is ready, no other operations which require a barrier will start | |
801 | * until the IO request has had a chance. | |
802 | * | |
803 | * So: regular IO calls 'wait_barrier'. When that returns there | |
804 | * is no backgroup IO happening, It must arrange to call | |
805 | * allow_barrier when it has finished its IO. | |
806 | * backgroup IO calls must call raise_barrier. Once that returns | |
807 | * there is no normal IO happeing. It must arrange to call | |
808 | * lower_barrier when the particular background IO completes. | |
1da177e4 | 809 | */ |
c2fd4c94 | 810 | static void raise_barrier(struct r1conf *conf, sector_t sector_nr) |
1da177e4 LT |
811 | { |
812 | spin_lock_irq(&conf->resync_lock); | |
17999be4 N |
813 | |
814 | /* Wait until no block IO is waiting */ | |
815 | wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting, | |
eed8c02e | 816 | conf->resync_lock); |
17999be4 N |
817 | |
818 | /* block any new IO from starting */ | |
819 | conf->barrier++; | |
c2fd4c94 | 820 | conf->next_resync = sector_nr; |
17999be4 | 821 | |
79ef3a8a | 822 | /* For these conditions we must wait: |
823 | * A: while the array is in frozen state | |
824 | * B: while barrier >= RESYNC_DEPTH, meaning resync reach | |
825 | * the max count which allowed. | |
826 | * C: next_resync + RESYNC_SECTORS > start_next_window, meaning | |
827 | * next resync will reach to the window which normal bios are | |
828 | * handling. | |
2f73d3c5 | 829 | * D: while there are any active requests in the current window. |
79ef3a8a | 830 | */ |
17999be4 | 831 | wait_event_lock_irq(conf->wait_barrier, |
b364e3d0 | 832 | !conf->array_frozen && |
79ef3a8a | 833 | conf->barrier < RESYNC_DEPTH && |
2f73d3c5 | 834 | conf->current_window_requests == 0 && |
79ef3a8a | 835 | (conf->start_next_window >= |
836 | conf->next_resync + RESYNC_SECTORS), | |
eed8c02e | 837 | conf->resync_lock); |
17999be4 | 838 | |
34e97f17 | 839 | conf->nr_pending++; |
17999be4 N |
840 | spin_unlock_irq(&conf->resync_lock); |
841 | } | |
842 | ||
e8096360 | 843 | static void lower_barrier(struct r1conf *conf) |
17999be4 N |
844 | { |
845 | unsigned long flags; | |
709ae487 | 846 | BUG_ON(conf->barrier <= 0); |
17999be4 N |
847 | spin_lock_irqsave(&conf->resync_lock, flags); |
848 | conf->barrier--; | |
34e97f17 | 849 | conf->nr_pending--; |
17999be4 N |
850 | spin_unlock_irqrestore(&conf->resync_lock, flags); |
851 | wake_up(&conf->wait_barrier); | |
852 | } | |
853 | ||
79ef3a8a | 854 | static bool need_to_wait_for_sync(struct r1conf *conf, struct bio *bio) |
17999be4 | 855 | { |
79ef3a8a | 856 | bool wait = false; |
857 | ||
858 | if (conf->array_frozen || !bio) | |
859 | wait = true; | |
860 | else if (conf->barrier && bio_data_dir(bio) == WRITE) { | |
23554960 N |
861 | if ((conf->mddev->curr_resync_completed |
862 | >= bio_end_sector(bio)) || | |
f2c771a6 | 863 | (conf->start_next_window + NEXT_NORMALIO_DISTANCE |
23554960 | 864 | <= bio->bi_iter.bi_sector)) |
79ef3a8a | 865 | wait = false; |
866 | else | |
867 | wait = true; | |
868 | } | |
869 | ||
870 | return wait; | |
871 | } | |
872 | ||
873 | static sector_t wait_barrier(struct r1conf *conf, struct bio *bio) | |
874 | { | |
875 | sector_t sector = 0; | |
876 | ||
17999be4 | 877 | spin_lock_irq(&conf->resync_lock); |
79ef3a8a | 878 | if (need_to_wait_for_sync(conf, bio)) { |
17999be4 | 879 | conf->nr_waiting++; |
d6b42dcb N |
880 | /* Wait for the barrier to drop. |
881 | * However if there are already pending | |
882 | * requests (preventing the barrier from | |
883 | * rising completely), and the | |
5965b642 | 884 | * per-process bio queue isn't empty, |
d6b42dcb | 885 | * then don't wait, as we need to empty |
5965b642 N |
886 | * that queue to allow conf->start_next_window |
887 | * to increase. | |
d6b42dcb | 888 | */ |
578b54ad | 889 | raid1_log(conf->mddev, "wait barrier"); |
d6b42dcb | 890 | wait_event_lock_irq(conf->wait_barrier, |
b364e3d0 | 891 | !conf->array_frozen && |
892 | (!conf->barrier || | |
5965b642 N |
893 | ((conf->start_next_window < |
894 | conf->next_resync + RESYNC_SECTORS) && | |
895 | current->bio_list && | |
896 | !bio_list_empty(current->bio_list))), | |
eed8c02e | 897 | conf->resync_lock); |
17999be4 | 898 | conf->nr_waiting--; |
1da177e4 | 899 | } |
79ef3a8a | 900 | |
901 | if (bio && bio_data_dir(bio) == WRITE) { | |
e8ff8bf0 | 902 | if (bio->bi_iter.bi_sector >= conf->next_resync) { |
79ef3a8a | 903 | if (conf->start_next_window == MaxSector) |
904 | conf->start_next_window = | |
905 | conf->next_resync + | |
906 | NEXT_NORMALIO_DISTANCE; | |
907 | ||
908 | if ((conf->start_next_window + NEXT_NORMALIO_DISTANCE) | |
4f024f37 | 909 | <= bio->bi_iter.bi_sector) |
79ef3a8a | 910 | conf->next_window_requests++; |
911 | else | |
912 | conf->current_window_requests++; | |
79ef3a8a | 913 | sector = conf->start_next_window; |
41a336e0 | 914 | } |
79ef3a8a | 915 | } |
916 | ||
17999be4 | 917 | conf->nr_pending++; |
1da177e4 | 918 | spin_unlock_irq(&conf->resync_lock); |
79ef3a8a | 919 | return sector; |
1da177e4 LT |
920 | } |
921 | ||
79ef3a8a | 922 | static void allow_barrier(struct r1conf *conf, sector_t start_next_window, |
923 | sector_t bi_sector) | |
17999be4 N |
924 | { |
925 | unsigned long flags; | |
79ef3a8a | 926 | |
17999be4 N |
927 | spin_lock_irqsave(&conf->resync_lock, flags); |
928 | conf->nr_pending--; | |
79ef3a8a | 929 | if (start_next_window) { |
930 | if (start_next_window == conf->start_next_window) { | |
931 | if (conf->start_next_window + NEXT_NORMALIO_DISTANCE | |
932 | <= bi_sector) | |
933 | conf->next_window_requests--; | |
934 | else | |
935 | conf->current_window_requests--; | |
936 | } else | |
937 | conf->current_window_requests--; | |
938 | ||
939 | if (!conf->current_window_requests) { | |
940 | if (conf->next_window_requests) { | |
941 | conf->current_window_requests = | |
942 | conf->next_window_requests; | |
943 | conf->next_window_requests = 0; | |
944 | conf->start_next_window += | |
945 | NEXT_NORMALIO_DISTANCE; | |
946 | } else | |
947 | conf->start_next_window = MaxSector; | |
948 | } | |
949 | } | |
17999be4 N |
950 | spin_unlock_irqrestore(&conf->resync_lock, flags); |
951 | wake_up(&conf->wait_barrier); | |
952 | } | |
953 | ||
e2d59925 | 954 | static void freeze_array(struct r1conf *conf, int extra) |
ddaf22ab N |
955 | { |
956 | /* stop syncio and normal IO and wait for everything to | |
957 | * go quite. | |
b364e3d0 | 958 | * We wait until nr_pending match nr_queued+extra |
1c830532 N |
959 | * This is called in the context of one normal IO request |
960 | * that has failed. Thus any sync request that might be pending | |
961 | * will be blocked by nr_pending, and we need to wait for | |
962 | * pending IO requests to complete or be queued for re-try. | |
e2d59925 | 963 | * Thus the number queued (nr_queued) plus this request (extra) |
1c830532 N |
964 | * must match the number of pending IOs (nr_pending) before |
965 | * we continue. | |
ddaf22ab N |
966 | */ |
967 | spin_lock_irq(&conf->resync_lock); | |
b364e3d0 | 968 | conf->array_frozen = 1; |
578b54ad | 969 | raid1_log(conf->mddev, "wait freeze"); |
eed8c02e | 970 | wait_event_lock_irq_cmd(conf->wait_barrier, |
e2d59925 | 971 | conf->nr_pending == conf->nr_queued+extra, |
eed8c02e LC |
972 | conf->resync_lock, |
973 | flush_pending_writes(conf)); | |
ddaf22ab N |
974 | spin_unlock_irq(&conf->resync_lock); |
975 | } | |
e8096360 | 976 | static void unfreeze_array(struct r1conf *conf) |
ddaf22ab N |
977 | { |
978 | /* reverse the effect of the freeze */ | |
979 | spin_lock_irq(&conf->resync_lock); | |
b364e3d0 | 980 | conf->array_frozen = 0; |
ddaf22ab N |
981 | wake_up(&conf->wait_barrier); |
982 | spin_unlock_irq(&conf->resync_lock); | |
983 | } | |
984 | ||
f72ffdd6 | 985 | /* duplicate the data pages for behind I/O |
4e78064f | 986 | */ |
9f2c9d12 | 987 | static void alloc_behind_pages(struct bio *bio, struct r1bio *r1_bio) |
4b6d287f N |
988 | { |
989 | int i; | |
990 | struct bio_vec *bvec; | |
2ca68f5e | 991 | struct bio_vec *bvecs = kzalloc(bio->bi_vcnt * sizeof(struct bio_vec), |
4b6d287f | 992 | GFP_NOIO); |
2ca68f5e | 993 | if (unlikely(!bvecs)) |
af6d7b76 | 994 | return; |
4b6d287f | 995 | |
cb34e057 | 996 | bio_for_each_segment_all(bvec, bio, i) { |
2ca68f5e N |
997 | bvecs[i] = *bvec; |
998 | bvecs[i].bv_page = alloc_page(GFP_NOIO); | |
999 | if (unlikely(!bvecs[i].bv_page)) | |
4b6d287f | 1000 | goto do_sync_io; |
2ca68f5e N |
1001 | memcpy(kmap(bvecs[i].bv_page) + bvec->bv_offset, |
1002 | kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len); | |
1003 | kunmap(bvecs[i].bv_page); | |
4b6d287f N |
1004 | kunmap(bvec->bv_page); |
1005 | } | |
2ca68f5e | 1006 | r1_bio->behind_bvecs = bvecs; |
af6d7b76 N |
1007 | r1_bio->behind_page_count = bio->bi_vcnt; |
1008 | set_bit(R1BIO_BehindIO, &r1_bio->state); | |
1009 | return; | |
4b6d287f N |
1010 | |
1011 | do_sync_io: | |
af6d7b76 | 1012 | for (i = 0; i < bio->bi_vcnt; i++) |
2ca68f5e N |
1013 | if (bvecs[i].bv_page) |
1014 | put_page(bvecs[i].bv_page); | |
1015 | kfree(bvecs); | |
4f024f37 KO |
1016 | pr_debug("%dB behind alloc failed, doing sync I/O\n", |
1017 | bio->bi_iter.bi_size); | |
4b6d287f N |
1018 | } |
1019 | ||
f54a9d0e N |
1020 | struct raid1_plug_cb { |
1021 | struct blk_plug_cb cb; | |
1022 | struct bio_list pending; | |
1023 | int pending_cnt; | |
1024 | }; | |
1025 | ||
1026 | static void raid1_unplug(struct blk_plug_cb *cb, bool from_schedule) | |
1027 | { | |
1028 | struct raid1_plug_cb *plug = container_of(cb, struct raid1_plug_cb, | |
1029 | cb); | |
1030 | struct mddev *mddev = plug->cb.data; | |
1031 | struct r1conf *conf = mddev->private; | |
1032 | struct bio *bio; | |
1033 | ||
874807a8 | 1034 | if (from_schedule || current->bio_list) { |
f54a9d0e N |
1035 | spin_lock_irq(&conf->device_lock); |
1036 | bio_list_merge(&conf->pending_bio_list, &plug->pending); | |
1037 | conf->pending_count += plug->pending_cnt; | |
1038 | spin_unlock_irq(&conf->device_lock); | |
ee0b0244 | 1039 | wake_up(&conf->wait_barrier); |
f54a9d0e N |
1040 | md_wakeup_thread(mddev->thread); |
1041 | kfree(plug); | |
1042 | return; | |
1043 | } | |
1044 | ||
1045 | /* we aren't scheduling, so we can do the write-out directly. */ | |
1046 | bio = bio_list_get(&plug->pending); | |
1047 | bitmap_unplug(mddev->bitmap); | |
1048 | wake_up(&conf->wait_barrier); | |
1049 | ||
1050 | while (bio) { /* submit pending writes */ | |
1051 | struct bio *next = bio->bi_next; | |
5e2c7a36 | 1052 | struct md_rdev *rdev = (void*)bio->bi_bdev; |
f54a9d0e | 1053 | bio->bi_next = NULL; |
5e2c7a36 N |
1054 | bio->bi_bdev = rdev->bdev; |
1055 | if (test_bit(Faulty, &rdev->flags)) { | |
1056 | bio->bi_error = -EIO; | |
1057 | bio_endio(bio); | |
1058 | } else if (unlikely((bio_op(bio) == REQ_OP_DISCARD) && | |
1059 | !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) | |
32f9f570 | 1060 | /* Just ignore it */ |
4246a0b6 | 1061 | bio_endio(bio); |
32f9f570 SL |
1062 | else |
1063 | generic_make_request(bio); | |
f54a9d0e N |
1064 | bio = next; |
1065 | } | |
1066 | kfree(plug); | |
1067 | } | |
1068 | ||
849674e4 | 1069 | static void raid1_make_request(struct mddev *mddev, struct bio * bio) |
1da177e4 | 1070 | { |
e8096360 | 1071 | struct r1conf *conf = mddev->private; |
0eaf822c | 1072 | struct raid1_info *mirror; |
9f2c9d12 | 1073 | struct r1bio *r1_bio; |
1da177e4 | 1074 | struct bio *read_bio; |
1f68f0c4 | 1075 | int i, disks; |
84255d10 | 1076 | struct bitmap *bitmap; |
191ea9b2 | 1077 | unsigned long flags; |
796a5cf0 | 1078 | const int op = bio_op(bio); |
a362357b | 1079 | const int rw = bio_data_dir(bio); |
1eff9d32 JA |
1080 | const unsigned long do_sync = (bio->bi_opf & REQ_SYNC); |
1081 | const unsigned long do_flush_fua = (bio->bi_opf & | |
28a8f0d3 | 1082 | (REQ_PREFLUSH | REQ_FUA)); |
3cb03002 | 1083 | struct md_rdev *blocked_rdev; |
f54a9d0e N |
1084 | struct blk_plug_cb *cb; |
1085 | struct raid1_plug_cb *plug = NULL; | |
1f68f0c4 N |
1086 | int first_clone; |
1087 | int sectors_handled; | |
1088 | int max_sectors; | |
79ef3a8a | 1089 | sector_t start_next_window; |
191ea9b2 | 1090 | |
1da177e4 LT |
1091 | /* |
1092 | * Register the new request and wait if the reconstruction | |
1093 | * thread has put up a bar for new requests. | |
1094 | * Continue immediately if no resync is active currently. | |
1095 | */ | |
62de608d | 1096 | |
3d310eb7 N |
1097 | md_write_start(mddev, bio); /* wait on superblock update early */ |
1098 | ||
6eef4b21 | 1099 | if (bio_data_dir(bio) == WRITE && |
589a1c49 GR |
1100 | ((bio_end_sector(bio) > mddev->suspend_lo && |
1101 | bio->bi_iter.bi_sector < mddev->suspend_hi) || | |
1102 | (mddev_is_clustered(mddev) && | |
90382ed9 GR |
1103 | md_cluster_ops->area_resyncing(mddev, WRITE, |
1104 | bio->bi_iter.bi_sector, bio_end_sector(bio))))) { | |
6eef4b21 N |
1105 | /* As the suspend_* range is controlled by |
1106 | * userspace, we want an interruptible | |
1107 | * wait. | |
1108 | */ | |
1109 | DEFINE_WAIT(w); | |
1110 | for (;;) { | |
1111 | flush_signals(current); | |
1112 | prepare_to_wait(&conf->wait_barrier, | |
1113 | &w, TASK_INTERRUPTIBLE); | |
f73a1c7d | 1114 | if (bio_end_sector(bio) <= mddev->suspend_lo || |
589a1c49 GR |
1115 | bio->bi_iter.bi_sector >= mddev->suspend_hi || |
1116 | (mddev_is_clustered(mddev) && | |
90382ed9 | 1117 | !md_cluster_ops->area_resyncing(mddev, WRITE, |
589a1c49 | 1118 | bio->bi_iter.bi_sector, bio_end_sector(bio)))) |
6eef4b21 N |
1119 | break; |
1120 | schedule(); | |
1121 | } | |
1122 | finish_wait(&conf->wait_barrier, &w); | |
1123 | } | |
62de608d | 1124 | |
79ef3a8a | 1125 | start_next_window = wait_barrier(conf, bio); |
1da177e4 | 1126 | |
84255d10 N |
1127 | bitmap = mddev->bitmap; |
1128 | ||
1da177e4 | 1129 | /* |
70246286 | 1130 | * make_request() can abort the operation when read-ahead is being |
1da177e4 LT |
1131 | * used and no empty request is available. |
1132 | * | |
1133 | */ | |
1134 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
1135 | ||
1136 | r1_bio->master_bio = bio; | |
aa8b57aa | 1137 | r1_bio->sectors = bio_sectors(bio); |
191ea9b2 | 1138 | r1_bio->state = 0; |
1da177e4 | 1139 | r1_bio->mddev = mddev; |
4f024f37 | 1140 | r1_bio->sector = bio->bi_iter.bi_sector; |
1da177e4 | 1141 | |
d2eb35ac N |
1142 | /* We might need to issue multiple reads to different |
1143 | * devices if there are bad blocks around, so we keep | |
1144 | * track of the number of reads in bio->bi_phys_segments. | |
1145 | * If this is 0, there is only one r1_bio and no locking | |
1146 | * will be needed when requests complete. If it is | |
1147 | * non-zero, then it is the number of not-completed requests. | |
1148 | */ | |
1149 | bio->bi_phys_segments = 0; | |
b7c44ed9 | 1150 | bio_clear_flag(bio, BIO_SEG_VALID); |
d2eb35ac | 1151 | |
a362357b | 1152 | if (rw == READ) { |
1da177e4 LT |
1153 | /* |
1154 | * read balancing logic: | |
1155 | */ | |
d2eb35ac N |
1156 | int rdisk; |
1157 | ||
1158 | read_again: | |
1159 | rdisk = read_balance(conf, r1_bio, &max_sectors); | |
1da177e4 LT |
1160 | |
1161 | if (rdisk < 0) { | |
1162 | /* couldn't find anywhere to read from */ | |
1163 | raid_end_bio_io(r1_bio); | |
5a7bbad2 | 1164 | return; |
1da177e4 LT |
1165 | } |
1166 | mirror = conf->mirrors + rdisk; | |
1167 | ||
e555190d N |
1168 | if (test_bit(WriteMostly, &mirror->rdev->flags) && |
1169 | bitmap) { | |
1170 | /* Reading from a write-mostly device must | |
1171 | * take care not to over-take any writes | |
1172 | * that are 'behind' | |
1173 | */ | |
578b54ad | 1174 | raid1_log(mddev, "wait behind writes"); |
e555190d N |
1175 | wait_event(bitmap->behind_wait, |
1176 | atomic_read(&bitmap->behind_writes) == 0); | |
1177 | } | |
1da177e4 | 1178 | r1_bio->read_disk = rdisk; |
f0cc9a05 | 1179 | r1_bio->start_next_window = 0; |
1da177e4 | 1180 | |
a167f663 | 1181 | read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev); |
4f024f37 | 1182 | bio_trim(read_bio, r1_bio->sector - bio->bi_iter.bi_sector, |
6678d83f | 1183 | max_sectors); |
1da177e4 LT |
1184 | |
1185 | r1_bio->bios[rdisk] = read_bio; | |
1186 | ||
4f024f37 KO |
1187 | read_bio->bi_iter.bi_sector = r1_bio->sector + |
1188 | mirror->rdev->data_offset; | |
1da177e4 LT |
1189 | read_bio->bi_bdev = mirror->rdev->bdev; |
1190 | read_bio->bi_end_io = raid1_end_read_request; | |
796a5cf0 | 1191 | bio_set_op_attrs(read_bio, op, do_sync); |
2e52d449 N |
1192 | if (test_bit(FailFast, &mirror->rdev->flags) && |
1193 | test_bit(R1BIO_FailFast, &r1_bio->state)) | |
1194 | read_bio->bi_opf |= MD_FAILFAST; | |
1da177e4 LT |
1195 | read_bio->bi_private = r1_bio; |
1196 | ||
109e3765 N |
1197 | if (mddev->gendisk) |
1198 | trace_block_bio_remap(bdev_get_queue(read_bio->bi_bdev), | |
1199 | read_bio, disk_devt(mddev->gendisk), | |
1200 | r1_bio->sector); | |
1201 | ||
d2eb35ac N |
1202 | if (max_sectors < r1_bio->sectors) { |
1203 | /* could not read all from this device, so we will | |
1204 | * need another r1_bio. | |
1205 | */ | |
d2eb35ac N |
1206 | |
1207 | sectors_handled = (r1_bio->sector + max_sectors | |
4f024f37 | 1208 | - bio->bi_iter.bi_sector); |
d2eb35ac N |
1209 | r1_bio->sectors = max_sectors; |
1210 | spin_lock_irq(&conf->device_lock); | |
1211 | if (bio->bi_phys_segments == 0) | |
1212 | bio->bi_phys_segments = 2; | |
1213 | else | |
1214 | bio->bi_phys_segments++; | |
1215 | spin_unlock_irq(&conf->device_lock); | |
1216 | /* Cannot call generic_make_request directly | |
1217 | * as that will be queued in __make_request | |
1218 | * and subsequent mempool_alloc might block waiting | |
1219 | * for it. So hand bio over to raid1d. | |
1220 | */ | |
1221 | reschedule_retry(r1_bio); | |
1222 | ||
1223 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
1224 | ||
1225 | r1_bio->master_bio = bio; | |
aa8b57aa | 1226 | r1_bio->sectors = bio_sectors(bio) - sectors_handled; |
d2eb35ac N |
1227 | r1_bio->state = 0; |
1228 | r1_bio->mddev = mddev; | |
4f024f37 KO |
1229 | r1_bio->sector = bio->bi_iter.bi_sector + |
1230 | sectors_handled; | |
d2eb35ac N |
1231 | goto read_again; |
1232 | } else | |
1233 | generic_make_request(read_bio); | |
5a7bbad2 | 1234 | return; |
1da177e4 LT |
1235 | } |
1236 | ||
1237 | /* | |
1238 | * WRITE: | |
1239 | */ | |
34db0cd6 N |
1240 | if (conf->pending_count >= max_queued_requests) { |
1241 | md_wakeup_thread(mddev->thread); | |
578b54ad | 1242 | raid1_log(mddev, "wait queued"); |
34db0cd6 N |
1243 | wait_event(conf->wait_barrier, |
1244 | conf->pending_count < max_queued_requests); | |
1245 | } | |
1f68f0c4 | 1246 | /* first select target devices under rcu_lock and |
1da177e4 LT |
1247 | * inc refcount on their rdev. Record them by setting |
1248 | * bios[x] to bio | |
1f68f0c4 N |
1249 | * If there are known/acknowledged bad blocks on any device on |
1250 | * which we have seen a write error, we want to avoid writing those | |
1251 | * blocks. | |
1252 | * This potentially requires several writes to write around | |
1253 | * the bad blocks. Each set of writes gets it's own r1bio | |
1254 | * with a set of bios attached. | |
1da177e4 | 1255 | */ |
c3b328ac | 1256 | |
8f19ccb2 | 1257 | disks = conf->raid_disks * 2; |
6bfe0b49 | 1258 | retry_write: |
79ef3a8a | 1259 | r1_bio->start_next_window = start_next_window; |
6bfe0b49 | 1260 | blocked_rdev = NULL; |
1da177e4 | 1261 | rcu_read_lock(); |
1f68f0c4 | 1262 | max_sectors = r1_bio->sectors; |
1da177e4 | 1263 | for (i = 0; i < disks; i++) { |
3cb03002 | 1264 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
6bfe0b49 DW |
1265 | if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { |
1266 | atomic_inc(&rdev->nr_pending); | |
1267 | blocked_rdev = rdev; | |
1268 | break; | |
1269 | } | |
1f68f0c4 | 1270 | r1_bio->bios[i] = NULL; |
8ae12666 | 1271 | if (!rdev || test_bit(Faulty, &rdev->flags)) { |
8f19ccb2 N |
1272 | if (i < conf->raid_disks) |
1273 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
1f68f0c4 N |
1274 | continue; |
1275 | } | |
1276 | ||
1277 | atomic_inc(&rdev->nr_pending); | |
1278 | if (test_bit(WriteErrorSeen, &rdev->flags)) { | |
1279 | sector_t first_bad; | |
1280 | int bad_sectors; | |
1281 | int is_bad; | |
1282 | ||
1283 | is_bad = is_badblock(rdev, r1_bio->sector, | |
1284 | max_sectors, | |
1285 | &first_bad, &bad_sectors); | |
1286 | if (is_bad < 0) { | |
1287 | /* mustn't write here until the bad block is | |
1288 | * acknowledged*/ | |
1289 | set_bit(BlockedBadBlocks, &rdev->flags); | |
1290 | blocked_rdev = rdev; | |
1291 | break; | |
1292 | } | |
1293 | if (is_bad && first_bad <= r1_bio->sector) { | |
1294 | /* Cannot write here at all */ | |
1295 | bad_sectors -= (r1_bio->sector - first_bad); | |
1296 | if (bad_sectors < max_sectors) | |
1297 | /* mustn't write more than bad_sectors | |
1298 | * to other devices yet | |
1299 | */ | |
1300 | max_sectors = bad_sectors; | |
03c902e1 | 1301 | rdev_dec_pending(rdev, mddev); |
1f68f0c4 N |
1302 | /* We don't set R1BIO_Degraded as that |
1303 | * only applies if the disk is | |
1304 | * missing, so it might be re-added, | |
1305 | * and we want to know to recover this | |
1306 | * chunk. | |
1307 | * In this case the device is here, | |
1308 | * and the fact that this chunk is not | |
1309 | * in-sync is recorded in the bad | |
1310 | * block log | |
1311 | */ | |
1312 | continue; | |
964147d5 | 1313 | } |
1f68f0c4 N |
1314 | if (is_bad) { |
1315 | int good_sectors = first_bad - r1_bio->sector; | |
1316 | if (good_sectors < max_sectors) | |
1317 | max_sectors = good_sectors; | |
1318 | } | |
1319 | } | |
1320 | r1_bio->bios[i] = bio; | |
1da177e4 LT |
1321 | } |
1322 | rcu_read_unlock(); | |
1323 | ||
6bfe0b49 DW |
1324 | if (unlikely(blocked_rdev)) { |
1325 | /* Wait for this device to become unblocked */ | |
1326 | int j; | |
79ef3a8a | 1327 | sector_t old = start_next_window; |
6bfe0b49 DW |
1328 | |
1329 | for (j = 0; j < i; j++) | |
1330 | if (r1_bio->bios[j]) | |
1331 | rdev_dec_pending(conf->mirrors[j].rdev, mddev); | |
1f68f0c4 | 1332 | r1_bio->state = 0; |
4f024f37 | 1333 | allow_barrier(conf, start_next_window, bio->bi_iter.bi_sector); |
578b54ad | 1334 | raid1_log(mddev, "wait rdev %d blocked", blocked_rdev->raid_disk); |
6bfe0b49 | 1335 | md_wait_for_blocked_rdev(blocked_rdev, mddev); |
79ef3a8a | 1336 | start_next_window = wait_barrier(conf, bio); |
1337 | /* | |
1338 | * We must make sure the multi r1bios of bio have | |
1339 | * the same value of bi_phys_segments | |
1340 | */ | |
1341 | if (bio->bi_phys_segments && old && | |
1342 | old != start_next_window) | |
1343 | /* Wait for the former r1bio(s) to complete */ | |
1344 | wait_event(conf->wait_barrier, | |
1345 | bio->bi_phys_segments == 1); | |
6bfe0b49 DW |
1346 | goto retry_write; |
1347 | } | |
1348 | ||
1f68f0c4 N |
1349 | if (max_sectors < r1_bio->sectors) { |
1350 | /* We are splitting this write into multiple parts, so | |
1351 | * we need to prepare for allocating another r1_bio. | |
1352 | */ | |
1353 | r1_bio->sectors = max_sectors; | |
1354 | spin_lock_irq(&conf->device_lock); | |
1355 | if (bio->bi_phys_segments == 0) | |
1356 | bio->bi_phys_segments = 2; | |
1357 | else | |
1358 | bio->bi_phys_segments++; | |
1359 | spin_unlock_irq(&conf->device_lock); | |
191ea9b2 | 1360 | } |
4f024f37 | 1361 | sectors_handled = r1_bio->sector + max_sectors - bio->bi_iter.bi_sector; |
4b6d287f | 1362 | |
4e78064f | 1363 | atomic_set(&r1_bio->remaining, 1); |
4b6d287f | 1364 | atomic_set(&r1_bio->behind_remaining, 0); |
06d91a5f | 1365 | |
1f68f0c4 | 1366 | first_clone = 1; |
1da177e4 LT |
1367 | for (i = 0; i < disks; i++) { |
1368 | struct bio *mbio; | |
1369 | if (!r1_bio->bios[i]) | |
1370 | continue; | |
1371 | ||
a167f663 | 1372 | mbio = bio_clone_mddev(bio, GFP_NOIO, mddev); |
4f024f37 | 1373 | bio_trim(mbio, r1_bio->sector - bio->bi_iter.bi_sector, max_sectors); |
1f68f0c4 N |
1374 | |
1375 | if (first_clone) { | |
1376 | /* do behind I/O ? | |
1377 | * Not if there are too many, or cannot | |
1378 | * allocate memory, or a reader on WriteMostly | |
1379 | * is waiting for behind writes to flush */ | |
1380 | if (bitmap && | |
1381 | (atomic_read(&bitmap->behind_writes) | |
1382 | < mddev->bitmap_info.max_write_behind) && | |
1383 | !waitqueue_active(&bitmap->behind_wait)) | |
1384 | alloc_behind_pages(mbio, r1_bio); | |
1385 | ||
1386 | bitmap_startwrite(bitmap, r1_bio->sector, | |
1387 | r1_bio->sectors, | |
1388 | test_bit(R1BIO_BehindIO, | |
1389 | &r1_bio->state)); | |
1390 | first_clone = 0; | |
1391 | } | |
2ca68f5e | 1392 | if (r1_bio->behind_bvecs) { |
4b6d287f N |
1393 | struct bio_vec *bvec; |
1394 | int j; | |
1395 | ||
cb34e057 KO |
1396 | /* |
1397 | * We trimmed the bio, so _all is legit | |
4b6d287f | 1398 | */ |
d74c6d51 | 1399 | bio_for_each_segment_all(bvec, mbio, j) |
2ca68f5e | 1400 | bvec->bv_page = r1_bio->behind_bvecs[j].bv_page; |
4b6d287f N |
1401 | if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags)) |
1402 | atomic_inc(&r1_bio->behind_remaining); | |
1403 | } | |
1404 | ||
1f68f0c4 N |
1405 | r1_bio->bios[i] = mbio; |
1406 | ||
4f024f37 | 1407 | mbio->bi_iter.bi_sector = (r1_bio->sector + |
1f68f0c4 | 1408 | conf->mirrors[i].rdev->data_offset); |
109e3765 | 1409 | mbio->bi_bdev = conf->mirrors[i].rdev->bdev; |
1f68f0c4 | 1410 | mbio->bi_end_io = raid1_end_write_request; |
288dab8a | 1411 | bio_set_op_attrs(mbio, op, do_flush_fua | do_sync); |
212e7eb7 N |
1412 | if (test_bit(FailFast, &conf->mirrors[i].rdev->flags) && |
1413 | !test_bit(WriteMostly, &conf->mirrors[i].rdev->flags) && | |
1414 | conf->raid_disks - mddev->degraded > 1) | |
1415 | mbio->bi_opf |= MD_FAILFAST; | |
1f68f0c4 N |
1416 | mbio->bi_private = r1_bio; |
1417 | ||
1da177e4 | 1418 | atomic_inc(&r1_bio->remaining); |
f54a9d0e | 1419 | |
109e3765 N |
1420 | if (mddev->gendisk) |
1421 | trace_block_bio_remap(bdev_get_queue(mbio->bi_bdev), | |
1422 | mbio, disk_devt(mddev->gendisk), | |
1423 | r1_bio->sector); | |
1424 | /* flush_pending_writes() needs access to the rdev so...*/ | |
1425 | mbio->bi_bdev = (void*)conf->mirrors[i].rdev; | |
1426 | ||
f54a9d0e N |
1427 | cb = blk_check_plugged(raid1_unplug, mddev, sizeof(*plug)); |
1428 | if (cb) | |
1429 | plug = container_of(cb, struct raid1_plug_cb, cb); | |
1430 | else | |
1431 | plug = NULL; | |
4e78064f | 1432 | spin_lock_irqsave(&conf->device_lock, flags); |
f54a9d0e N |
1433 | if (plug) { |
1434 | bio_list_add(&plug->pending, mbio); | |
1435 | plug->pending_cnt++; | |
1436 | } else { | |
1437 | bio_list_add(&conf->pending_bio_list, mbio); | |
1438 | conf->pending_count++; | |
1439 | } | |
4e78064f | 1440 | spin_unlock_irqrestore(&conf->device_lock, flags); |
f54a9d0e | 1441 | if (!plug) |
b357f04a | 1442 | md_wakeup_thread(mddev->thread); |
1da177e4 | 1443 | } |
079fa166 N |
1444 | /* Mustn't call r1_bio_write_done before this next test, |
1445 | * as it could result in the bio being freed. | |
1446 | */ | |
aa8b57aa | 1447 | if (sectors_handled < bio_sectors(bio)) { |
079fa166 | 1448 | r1_bio_write_done(r1_bio); |
1f68f0c4 N |
1449 | /* We need another r1_bio. It has already been counted |
1450 | * in bio->bi_phys_segments | |
1451 | */ | |
1452 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
1453 | r1_bio->master_bio = bio; | |
aa8b57aa | 1454 | r1_bio->sectors = bio_sectors(bio) - sectors_handled; |
1f68f0c4 N |
1455 | r1_bio->state = 0; |
1456 | r1_bio->mddev = mddev; | |
4f024f37 | 1457 | r1_bio->sector = bio->bi_iter.bi_sector + sectors_handled; |
1f68f0c4 N |
1458 | goto retry_write; |
1459 | } | |
1460 | ||
079fa166 N |
1461 | r1_bio_write_done(r1_bio); |
1462 | ||
1463 | /* In case raid1d snuck in to freeze_array */ | |
1464 | wake_up(&conf->wait_barrier); | |
1da177e4 LT |
1465 | } |
1466 | ||
849674e4 | 1467 | static void raid1_status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 1468 | { |
e8096360 | 1469 | struct r1conf *conf = mddev->private; |
1da177e4 LT |
1470 | int i; |
1471 | ||
1472 | seq_printf(seq, " [%d/%d] [", conf->raid_disks, | |
11ce99e6 | 1473 | conf->raid_disks - mddev->degraded); |
ddac7c7e N |
1474 | rcu_read_lock(); |
1475 | for (i = 0; i < conf->raid_disks; i++) { | |
3cb03002 | 1476 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
1da177e4 | 1477 | seq_printf(seq, "%s", |
ddac7c7e N |
1478 | rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); |
1479 | } | |
1480 | rcu_read_unlock(); | |
1da177e4 LT |
1481 | seq_printf(seq, "]"); |
1482 | } | |
1483 | ||
849674e4 | 1484 | static void raid1_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1485 | { |
1486 | char b[BDEVNAME_SIZE]; | |
e8096360 | 1487 | struct r1conf *conf = mddev->private; |
423f04d6 | 1488 | unsigned long flags; |
1da177e4 LT |
1489 | |
1490 | /* | |
1491 | * If it is not operational, then we have already marked it as dead | |
1492 | * else if it is the last working disks, ignore the error, let the | |
1493 | * next level up know. | |
1494 | * else mark the drive as failed | |
1495 | */ | |
2e52d449 | 1496 | spin_lock_irqsave(&conf->device_lock, flags); |
b2d444d7 | 1497 | if (test_bit(In_sync, &rdev->flags) |
4044ba58 | 1498 | && (conf->raid_disks - mddev->degraded) == 1) { |
1da177e4 LT |
1499 | /* |
1500 | * Don't fail the drive, act as though we were just a | |
4044ba58 N |
1501 | * normal single drive. |
1502 | * However don't try a recovery from this drive as | |
1503 | * it is very likely to fail. | |
1da177e4 | 1504 | */ |
5389042f | 1505 | conf->recovery_disabled = mddev->recovery_disabled; |
2e52d449 | 1506 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 1507 | return; |
4044ba58 | 1508 | } |
de393cde | 1509 | set_bit(Blocked, &rdev->flags); |
c04be0aa | 1510 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
1da177e4 | 1511 | mddev->degraded++; |
dd00a99e | 1512 | set_bit(Faulty, &rdev->flags); |
dd00a99e N |
1513 | } else |
1514 | set_bit(Faulty, &rdev->flags); | |
423f04d6 | 1515 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2446dba0 N |
1516 | /* |
1517 | * if recovery is running, make sure it aborts. | |
1518 | */ | |
1519 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
85ad1d13 GJ |
1520 | set_mask_bits(&mddev->flags, 0, |
1521 | BIT(MD_CHANGE_DEVS) | BIT(MD_CHANGE_PENDING)); | |
1d41c216 N |
1522 | pr_crit("md/raid1:%s: Disk failure on %s, disabling device.\n" |
1523 | "md/raid1:%s: Operation continuing on %d devices.\n", | |
1524 | mdname(mddev), bdevname(rdev->bdev, b), | |
1525 | mdname(mddev), conf->raid_disks - mddev->degraded); | |
1da177e4 LT |
1526 | } |
1527 | ||
e8096360 | 1528 | static void print_conf(struct r1conf *conf) |
1da177e4 LT |
1529 | { |
1530 | int i; | |
1da177e4 | 1531 | |
1d41c216 | 1532 | pr_debug("RAID1 conf printout:\n"); |
1da177e4 | 1533 | if (!conf) { |
1d41c216 | 1534 | pr_debug("(!conf)\n"); |
1da177e4 LT |
1535 | return; |
1536 | } | |
1d41c216 N |
1537 | pr_debug(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded, |
1538 | conf->raid_disks); | |
1da177e4 | 1539 | |
ddac7c7e | 1540 | rcu_read_lock(); |
1da177e4 LT |
1541 | for (i = 0; i < conf->raid_disks; i++) { |
1542 | char b[BDEVNAME_SIZE]; | |
3cb03002 | 1543 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
ddac7c7e | 1544 | if (rdev) |
1d41c216 N |
1545 | pr_debug(" disk %d, wo:%d, o:%d, dev:%s\n", |
1546 | i, !test_bit(In_sync, &rdev->flags), | |
1547 | !test_bit(Faulty, &rdev->flags), | |
1548 | bdevname(rdev->bdev,b)); | |
1da177e4 | 1549 | } |
ddac7c7e | 1550 | rcu_read_unlock(); |
1da177e4 LT |
1551 | } |
1552 | ||
e8096360 | 1553 | static void close_sync(struct r1conf *conf) |
1da177e4 | 1554 | { |
79ef3a8a | 1555 | wait_barrier(conf, NULL); |
1556 | allow_barrier(conf, 0, 0); | |
1da177e4 LT |
1557 | |
1558 | mempool_destroy(conf->r1buf_pool); | |
1559 | conf->r1buf_pool = NULL; | |
79ef3a8a | 1560 | |
669cc7ba | 1561 | spin_lock_irq(&conf->resync_lock); |
e8ff8bf0 | 1562 | conf->next_resync = MaxSector - 2 * NEXT_NORMALIO_DISTANCE; |
79ef3a8a | 1563 | conf->start_next_window = MaxSector; |
669cc7ba N |
1564 | conf->current_window_requests += |
1565 | conf->next_window_requests; | |
1566 | conf->next_window_requests = 0; | |
1567 | spin_unlock_irq(&conf->resync_lock); | |
1da177e4 LT |
1568 | } |
1569 | ||
fd01b88c | 1570 | static int raid1_spare_active(struct mddev *mddev) |
1da177e4 LT |
1571 | { |
1572 | int i; | |
e8096360 | 1573 | struct r1conf *conf = mddev->private; |
6b965620 N |
1574 | int count = 0; |
1575 | unsigned long flags; | |
1da177e4 LT |
1576 | |
1577 | /* | |
f72ffdd6 | 1578 | * Find all failed disks within the RAID1 configuration |
ddac7c7e N |
1579 | * and mark them readable. |
1580 | * Called under mddev lock, so rcu protection not needed. | |
423f04d6 N |
1581 | * device_lock used to avoid races with raid1_end_read_request |
1582 | * which expects 'In_sync' flags and ->degraded to be consistent. | |
1da177e4 | 1583 | */ |
423f04d6 | 1584 | spin_lock_irqsave(&conf->device_lock, flags); |
1da177e4 | 1585 | for (i = 0; i < conf->raid_disks; i++) { |
3cb03002 | 1586 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
8c7a2c2b N |
1587 | struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev; |
1588 | if (repl | |
1aee41f6 | 1589 | && !test_bit(Candidate, &repl->flags) |
8c7a2c2b N |
1590 | && repl->recovery_offset == MaxSector |
1591 | && !test_bit(Faulty, &repl->flags) | |
1592 | && !test_and_set_bit(In_sync, &repl->flags)) { | |
1593 | /* replacement has just become active */ | |
1594 | if (!rdev || | |
1595 | !test_and_clear_bit(In_sync, &rdev->flags)) | |
1596 | count++; | |
1597 | if (rdev) { | |
1598 | /* Replaced device not technically | |
1599 | * faulty, but we need to be sure | |
1600 | * it gets removed and never re-added | |
1601 | */ | |
1602 | set_bit(Faulty, &rdev->flags); | |
1603 | sysfs_notify_dirent_safe( | |
1604 | rdev->sysfs_state); | |
1605 | } | |
1606 | } | |
ddac7c7e | 1607 | if (rdev |
61e4947c | 1608 | && rdev->recovery_offset == MaxSector |
ddac7c7e | 1609 | && !test_bit(Faulty, &rdev->flags) |
c04be0aa | 1610 | && !test_and_set_bit(In_sync, &rdev->flags)) { |
6b965620 | 1611 | count++; |
654e8b5a | 1612 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
1613 | } |
1614 | } | |
6b965620 N |
1615 | mddev->degraded -= count; |
1616 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 LT |
1617 | |
1618 | print_conf(conf); | |
6b965620 | 1619 | return count; |
1da177e4 LT |
1620 | } |
1621 | ||
fd01b88c | 1622 | static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1623 | { |
e8096360 | 1624 | struct r1conf *conf = mddev->private; |
199050ea | 1625 | int err = -EEXIST; |
41158c7e | 1626 | int mirror = 0; |
0eaf822c | 1627 | struct raid1_info *p; |
6c2fce2e | 1628 | int first = 0; |
30194636 | 1629 | int last = conf->raid_disks - 1; |
1da177e4 | 1630 | |
5389042f N |
1631 | if (mddev->recovery_disabled == conf->recovery_disabled) |
1632 | return -EBUSY; | |
1633 | ||
1501efad DW |
1634 | if (md_integrity_add_rdev(rdev, mddev)) |
1635 | return -ENXIO; | |
1636 | ||
6c2fce2e NB |
1637 | if (rdev->raid_disk >= 0) |
1638 | first = last = rdev->raid_disk; | |
1639 | ||
70bcecdb GR |
1640 | /* |
1641 | * find the disk ... but prefer rdev->saved_raid_disk | |
1642 | * if possible. | |
1643 | */ | |
1644 | if (rdev->saved_raid_disk >= 0 && | |
1645 | rdev->saved_raid_disk >= first && | |
1646 | conf->mirrors[rdev->saved_raid_disk].rdev == NULL) | |
1647 | first = last = rdev->saved_raid_disk; | |
1648 | ||
7ef449d1 N |
1649 | for (mirror = first; mirror <= last; mirror++) { |
1650 | p = conf->mirrors+mirror; | |
1651 | if (!p->rdev) { | |
1da177e4 | 1652 | |
9092c02d JB |
1653 | if (mddev->gendisk) |
1654 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
1655 | rdev->data_offset << 9); | |
1da177e4 LT |
1656 | |
1657 | p->head_position = 0; | |
1658 | rdev->raid_disk = mirror; | |
199050ea | 1659 | err = 0; |
6aea114a N |
1660 | /* As all devices are equivalent, we don't need a full recovery |
1661 | * if this was recently any drive of the array | |
1662 | */ | |
1663 | if (rdev->saved_raid_disk < 0) | |
41158c7e | 1664 | conf->fullsync = 1; |
d6065f7b | 1665 | rcu_assign_pointer(p->rdev, rdev); |
1da177e4 LT |
1666 | break; |
1667 | } | |
7ef449d1 N |
1668 | if (test_bit(WantReplacement, &p->rdev->flags) && |
1669 | p[conf->raid_disks].rdev == NULL) { | |
1670 | /* Add this device as a replacement */ | |
1671 | clear_bit(In_sync, &rdev->flags); | |
1672 | set_bit(Replacement, &rdev->flags); | |
1673 | rdev->raid_disk = mirror; | |
1674 | err = 0; | |
1675 | conf->fullsync = 1; | |
1676 | rcu_assign_pointer(p[conf->raid_disks].rdev, rdev); | |
1677 | break; | |
1678 | } | |
1679 | } | |
9092c02d | 1680 | if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev))) |
2ff8cc2c | 1681 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue); |
1da177e4 | 1682 | print_conf(conf); |
199050ea | 1683 | return err; |
1da177e4 LT |
1684 | } |
1685 | ||
b8321b68 | 1686 | static int raid1_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1687 | { |
e8096360 | 1688 | struct r1conf *conf = mddev->private; |
1da177e4 | 1689 | int err = 0; |
b8321b68 | 1690 | int number = rdev->raid_disk; |
0eaf822c | 1691 | struct raid1_info *p = conf->mirrors + number; |
1da177e4 | 1692 | |
b014f14c N |
1693 | if (rdev != p->rdev) |
1694 | p = conf->mirrors + conf->raid_disks + number; | |
1695 | ||
1da177e4 | 1696 | print_conf(conf); |
b8321b68 | 1697 | if (rdev == p->rdev) { |
b2d444d7 | 1698 | if (test_bit(In_sync, &rdev->flags) || |
1da177e4 LT |
1699 | atomic_read(&rdev->nr_pending)) { |
1700 | err = -EBUSY; | |
1701 | goto abort; | |
1702 | } | |
046abeed | 1703 | /* Only remove non-faulty devices if recovery |
dfc70645 N |
1704 | * is not possible. |
1705 | */ | |
1706 | if (!test_bit(Faulty, &rdev->flags) && | |
5389042f | 1707 | mddev->recovery_disabled != conf->recovery_disabled && |
dfc70645 N |
1708 | mddev->degraded < conf->raid_disks) { |
1709 | err = -EBUSY; | |
1710 | goto abort; | |
1711 | } | |
1da177e4 | 1712 | p->rdev = NULL; |
d787be40 N |
1713 | if (!test_bit(RemoveSynchronized, &rdev->flags)) { |
1714 | synchronize_rcu(); | |
1715 | if (atomic_read(&rdev->nr_pending)) { | |
1716 | /* lost the race, try later */ | |
1717 | err = -EBUSY; | |
1718 | p->rdev = rdev; | |
1719 | goto abort; | |
1720 | } | |
1721 | } | |
1722 | if (conf->mirrors[conf->raid_disks + number].rdev) { | |
8c7a2c2b N |
1723 | /* We just removed a device that is being replaced. |
1724 | * Move down the replacement. We drain all IO before | |
1725 | * doing this to avoid confusion. | |
1726 | */ | |
1727 | struct md_rdev *repl = | |
1728 | conf->mirrors[conf->raid_disks + number].rdev; | |
e2d59925 | 1729 | freeze_array(conf, 0); |
8c7a2c2b N |
1730 | clear_bit(Replacement, &repl->flags); |
1731 | p->rdev = repl; | |
1732 | conf->mirrors[conf->raid_disks + number].rdev = NULL; | |
e2d59925 | 1733 | unfreeze_array(conf); |
8c7a2c2b N |
1734 | clear_bit(WantReplacement, &rdev->flags); |
1735 | } else | |
b014f14c | 1736 | clear_bit(WantReplacement, &rdev->flags); |
a91a2785 | 1737 | err = md_integrity_register(mddev); |
1da177e4 LT |
1738 | } |
1739 | abort: | |
1740 | ||
1741 | print_conf(conf); | |
1742 | return err; | |
1743 | } | |
1744 | ||
4246a0b6 | 1745 | static void end_sync_read(struct bio *bio) |
1da177e4 | 1746 | { |
9f2c9d12 | 1747 | struct r1bio *r1_bio = bio->bi_private; |
1da177e4 | 1748 | |
0fc280f6 | 1749 | update_head_pos(r1_bio->read_disk, r1_bio); |
ba3ae3be | 1750 | |
1da177e4 LT |
1751 | /* |
1752 | * we have read a block, now it needs to be re-written, | |
1753 | * or re-read if the read failed. | |
1754 | * We don't do much here, just schedule handling by raid1d | |
1755 | */ | |
4246a0b6 | 1756 | if (!bio->bi_error) |
1da177e4 | 1757 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
d11c171e N |
1758 | |
1759 | if (atomic_dec_and_test(&r1_bio->remaining)) | |
1760 | reschedule_retry(r1_bio); | |
1da177e4 LT |
1761 | } |
1762 | ||
4246a0b6 | 1763 | static void end_sync_write(struct bio *bio) |
1da177e4 | 1764 | { |
4246a0b6 | 1765 | int uptodate = !bio->bi_error; |
9f2c9d12 | 1766 | struct r1bio *r1_bio = bio->bi_private; |
fd01b88c | 1767 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 1768 | struct r1conf *conf = mddev->private; |
4367af55 N |
1769 | sector_t first_bad; |
1770 | int bad_sectors; | |
854abd75 | 1771 | struct md_rdev *rdev = conf->mirrors[find_bio_disk(r1_bio, bio)].rdev; |
ba3ae3be | 1772 | |
6b1117d5 | 1773 | if (!uptodate) { |
57dab0bd | 1774 | sector_t sync_blocks = 0; |
6b1117d5 N |
1775 | sector_t s = r1_bio->sector; |
1776 | long sectors_to_go = r1_bio->sectors; | |
1777 | /* make sure these bits doesn't get cleared. */ | |
1778 | do { | |
5e3db645 | 1779 | bitmap_end_sync(mddev->bitmap, s, |
6b1117d5 N |
1780 | &sync_blocks, 1); |
1781 | s += sync_blocks; | |
1782 | sectors_to_go -= sync_blocks; | |
1783 | } while (sectors_to_go > 0); | |
854abd75 N |
1784 | set_bit(WriteErrorSeen, &rdev->flags); |
1785 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) | |
19d67169 N |
1786 | set_bit(MD_RECOVERY_NEEDED, & |
1787 | mddev->recovery); | |
d8f05d29 | 1788 | set_bit(R1BIO_WriteError, &r1_bio->state); |
854abd75 | 1789 | } else if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors, |
3a9f28a5 N |
1790 | &first_bad, &bad_sectors) && |
1791 | !is_badblock(conf->mirrors[r1_bio->read_disk].rdev, | |
1792 | r1_bio->sector, | |
1793 | r1_bio->sectors, | |
1794 | &first_bad, &bad_sectors) | |
1795 | ) | |
4367af55 | 1796 | set_bit(R1BIO_MadeGood, &r1_bio->state); |
e3b9703e | 1797 | |
1da177e4 | 1798 | if (atomic_dec_and_test(&r1_bio->remaining)) { |
4367af55 | 1799 | int s = r1_bio->sectors; |
d8f05d29 N |
1800 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
1801 | test_bit(R1BIO_WriteError, &r1_bio->state)) | |
4367af55 N |
1802 | reschedule_retry(r1_bio); |
1803 | else { | |
1804 | put_buf(r1_bio); | |
1805 | md_done_sync(mddev, s, uptodate); | |
1806 | } | |
1da177e4 | 1807 | } |
1da177e4 LT |
1808 | } |
1809 | ||
3cb03002 | 1810 | static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector, |
d8f05d29 N |
1811 | int sectors, struct page *page, int rw) |
1812 | { | |
796a5cf0 | 1813 | if (sync_page_io(rdev, sector, sectors << 9, page, rw, 0, false)) |
d8f05d29 N |
1814 | /* success */ |
1815 | return 1; | |
19d67169 | 1816 | if (rw == WRITE) { |
d8f05d29 | 1817 | set_bit(WriteErrorSeen, &rdev->flags); |
19d67169 N |
1818 | if (!test_and_set_bit(WantReplacement, |
1819 | &rdev->flags)) | |
1820 | set_bit(MD_RECOVERY_NEEDED, & | |
1821 | rdev->mddev->recovery); | |
1822 | } | |
d8f05d29 N |
1823 | /* need to record an error - either for the block or the device */ |
1824 | if (!rdev_set_badblocks(rdev, sector, sectors, 0)) | |
1825 | md_error(rdev->mddev, rdev); | |
1826 | return 0; | |
1827 | } | |
1828 | ||
9f2c9d12 | 1829 | static int fix_sync_read_error(struct r1bio *r1_bio) |
1da177e4 | 1830 | { |
a68e5870 N |
1831 | /* Try some synchronous reads of other devices to get |
1832 | * good data, much like with normal read errors. Only | |
1833 | * read into the pages we already have so we don't | |
1834 | * need to re-issue the read request. | |
1835 | * We don't need to freeze the array, because being in an | |
1836 | * active sync request, there is no normal IO, and | |
1837 | * no overlapping syncs. | |
06f60385 N |
1838 | * We don't need to check is_badblock() again as we |
1839 | * made sure that anything with a bad block in range | |
1840 | * will have bi_end_io clear. | |
a68e5870 | 1841 | */ |
fd01b88c | 1842 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 1843 | struct r1conf *conf = mddev->private; |
a68e5870 N |
1844 | struct bio *bio = r1_bio->bios[r1_bio->read_disk]; |
1845 | sector_t sect = r1_bio->sector; | |
1846 | int sectors = r1_bio->sectors; | |
1847 | int idx = 0; | |
2e52d449 N |
1848 | struct md_rdev *rdev; |
1849 | ||
1850 | rdev = conf->mirrors[r1_bio->read_disk].rdev; | |
1851 | if (test_bit(FailFast, &rdev->flags)) { | |
1852 | /* Don't try recovering from here - just fail it | |
1853 | * ... unless it is the last working device of course */ | |
1854 | md_error(mddev, rdev); | |
1855 | if (test_bit(Faulty, &rdev->flags)) | |
1856 | /* Don't try to read from here, but make sure | |
1857 | * put_buf does it's thing | |
1858 | */ | |
1859 | bio->bi_end_io = end_sync_write; | |
1860 | } | |
a68e5870 N |
1861 | |
1862 | while(sectors) { | |
1863 | int s = sectors; | |
1864 | int d = r1_bio->read_disk; | |
1865 | int success = 0; | |
78d7f5f7 | 1866 | int start; |
a68e5870 N |
1867 | |
1868 | if (s > (PAGE_SIZE>>9)) | |
1869 | s = PAGE_SIZE >> 9; | |
1870 | do { | |
1871 | if (r1_bio->bios[d]->bi_end_io == end_sync_read) { | |
1872 | /* No rcu protection needed here devices | |
1873 | * can only be removed when no resync is | |
1874 | * active, and resync is currently active | |
1875 | */ | |
1876 | rdev = conf->mirrors[d].rdev; | |
9d3d8011 | 1877 | if (sync_page_io(rdev, sect, s<<9, |
a68e5870 | 1878 | bio->bi_io_vec[idx].bv_page, |
796a5cf0 | 1879 | REQ_OP_READ, 0, false)) { |
a68e5870 N |
1880 | success = 1; |
1881 | break; | |
1882 | } | |
1883 | } | |
1884 | d++; | |
8f19ccb2 | 1885 | if (d == conf->raid_disks * 2) |
a68e5870 N |
1886 | d = 0; |
1887 | } while (!success && d != r1_bio->read_disk); | |
1888 | ||
78d7f5f7 | 1889 | if (!success) { |
a68e5870 | 1890 | char b[BDEVNAME_SIZE]; |
3a9f28a5 N |
1891 | int abort = 0; |
1892 | /* Cannot read from anywhere, this block is lost. | |
1893 | * Record a bad block on each device. If that doesn't | |
1894 | * work just disable and interrupt the recovery. | |
1895 | * Don't fail devices as that won't really help. | |
1896 | */ | |
1d41c216 N |
1897 | pr_crit_ratelimited("md/raid1:%s: %s: unrecoverable I/O read error for block %llu\n", |
1898 | mdname(mddev), | |
1899 | bdevname(bio->bi_bdev, b), | |
1900 | (unsigned long long)r1_bio->sector); | |
8f19ccb2 | 1901 | for (d = 0; d < conf->raid_disks * 2; d++) { |
3a9f28a5 N |
1902 | rdev = conf->mirrors[d].rdev; |
1903 | if (!rdev || test_bit(Faulty, &rdev->flags)) | |
1904 | continue; | |
1905 | if (!rdev_set_badblocks(rdev, sect, s, 0)) | |
1906 | abort = 1; | |
1907 | } | |
1908 | if (abort) { | |
d890fa2b N |
1909 | conf->recovery_disabled = |
1910 | mddev->recovery_disabled; | |
3a9f28a5 N |
1911 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1912 | md_done_sync(mddev, r1_bio->sectors, 0); | |
1913 | put_buf(r1_bio); | |
1914 | return 0; | |
1915 | } | |
1916 | /* Try next page */ | |
1917 | sectors -= s; | |
1918 | sect += s; | |
1919 | idx++; | |
1920 | continue; | |
d11c171e | 1921 | } |
78d7f5f7 N |
1922 | |
1923 | start = d; | |
1924 | /* write it back and re-read */ | |
1925 | while (d != r1_bio->read_disk) { | |
1926 | if (d == 0) | |
8f19ccb2 | 1927 | d = conf->raid_disks * 2; |
78d7f5f7 N |
1928 | d--; |
1929 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) | |
1930 | continue; | |
1931 | rdev = conf->mirrors[d].rdev; | |
d8f05d29 N |
1932 | if (r1_sync_page_io(rdev, sect, s, |
1933 | bio->bi_io_vec[idx].bv_page, | |
1934 | WRITE) == 0) { | |
78d7f5f7 N |
1935 | r1_bio->bios[d]->bi_end_io = NULL; |
1936 | rdev_dec_pending(rdev, mddev); | |
9d3d8011 | 1937 | } |
78d7f5f7 N |
1938 | } |
1939 | d = start; | |
1940 | while (d != r1_bio->read_disk) { | |
1941 | if (d == 0) | |
8f19ccb2 | 1942 | d = conf->raid_disks * 2; |
78d7f5f7 N |
1943 | d--; |
1944 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) | |
1945 | continue; | |
1946 | rdev = conf->mirrors[d].rdev; | |
d8f05d29 N |
1947 | if (r1_sync_page_io(rdev, sect, s, |
1948 | bio->bi_io_vec[idx].bv_page, | |
1949 | READ) != 0) | |
9d3d8011 | 1950 | atomic_add(s, &rdev->corrected_errors); |
78d7f5f7 | 1951 | } |
a68e5870 N |
1952 | sectors -= s; |
1953 | sect += s; | |
1954 | idx ++; | |
1955 | } | |
78d7f5f7 | 1956 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
4246a0b6 | 1957 | bio->bi_error = 0; |
a68e5870 N |
1958 | return 1; |
1959 | } | |
1960 | ||
c95e6385 | 1961 | static void process_checks(struct r1bio *r1_bio) |
a68e5870 N |
1962 | { |
1963 | /* We have read all readable devices. If we haven't | |
1964 | * got the block, then there is no hope left. | |
1965 | * If we have, then we want to do a comparison | |
1966 | * and skip the write if everything is the same. | |
1967 | * If any blocks failed to read, then we need to | |
1968 | * attempt an over-write | |
1969 | */ | |
fd01b88c | 1970 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 1971 | struct r1conf *conf = mddev->private; |
a68e5870 N |
1972 | int primary; |
1973 | int i; | |
f4380a91 | 1974 | int vcnt; |
a68e5870 | 1975 | |
30bc9b53 N |
1976 | /* Fix variable parts of all bios */ |
1977 | vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9); | |
1978 | for (i = 0; i < conf->raid_disks * 2; i++) { | |
1979 | int j; | |
1980 | int size; | |
4246a0b6 | 1981 | int error; |
30bc9b53 N |
1982 | struct bio *b = r1_bio->bios[i]; |
1983 | if (b->bi_end_io != end_sync_read) | |
1984 | continue; | |
4246a0b6 CH |
1985 | /* fixup the bio for reuse, but preserve errno */ |
1986 | error = b->bi_error; | |
30bc9b53 | 1987 | bio_reset(b); |
4246a0b6 | 1988 | b->bi_error = error; |
30bc9b53 | 1989 | b->bi_vcnt = vcnt; |
4f024f37 KO |
1990 | b->bi_iter.bi_size = r1_bio->sectors << 9; |
1991 | b->bi_iter.bi_sector = r1_bio->sector + | |
30bc9b53 N |
1992 | conf->mirrors[i].rdev->data_offset; |
1993 | b->bi_bdev = conf->mirrors[i].rdev->bdev; | |
1994 | b->bi_end_io = end_sync_read; | |
1995 | b->bi_private = r1_bio; | |
1996 | ||
4f024f37 | 1997 | size = b->bi_iter.bi_size; |
30bc9b53 N |
1998 | for (j = 0; j < vcnt ; j++) { |
1999 | struct bio_vec *bi; | |
2000 | bi = &b->bi_io_vec[j]; | |
2001 | bi->bv_offset = 0; | |
2002 | if (size > PAGE_SIZE) | |
2003 | bi->bv_len = PAGE_SIZE; | |
2004 | else | |
2005 | bi->bv_len = size; | |
2006 | size -= PAGE_SIZE; | |
2007 | } | |
2008 | } | |
8f19ccb2 | 2009 | for (primary = 0; primary < conf->raid_disks * 2; primary++) |
a68e5870 | 2010 | if (r1_bio->bios[primary]->bi_end_io == end_sync_read && |
4246a0b6 | 2011 | !r1_bio->bios[primary]->bi_error) { |
a68e5870 N |
2012 | r1_bio->bios[primary]->bi_end_io = NULL; |
2013 | rdev_dec_pending(conf->mirrors[primary].rdev, mddev); | |
2014 | break; | |
2015 | } | |
2016 | r1_bio->read_disk = primary; | |
8f19ccb2 | 2017 | for (i = 0; i < conf->raid_disks * 2; i++) { |
78d7f5f7 | 2018 | int j; |
78d7f5f7 N |
2019 | struct bio *pbio = r1_bio->bios[primary]; |
2020 | struct bio *sbio = r1_bio->bios[i]; | |
4246a0b6 | 2021 | int error = sbio->bi_error; |
a68e5870 | 2022 | |
2aabaa65 | 2023 | if (sbio->bi_end_io != end_sync_read) |
78d7f5f7 | 2024 | continue; |
4246a0b6 CH |
2025 | /* Now we can 'fixup' the error value */ |
2026 | sbio->bi_error = 0; | |
78d7f5f7 | 2027 | |
4246a0b6 | 2028 | if (!error) { |
78d7f5f7 N |
2029 | for (j = vcnt; j-- ; ) { |
2030 | struct page *p, *s; | |
2031 | p = pbio->bi_io_vec[j].bv_page; | |
2032 | s = sbio->bi_io_vec[j].bv_page; | |
2033 | if (memcmp(page_address(p), | |
2034 | page_address(s), | |
5020ad7d | 2035 | sbio->bi_io_vec[j].bv_len)) |
78d7f5f7 | 2036 | break; |
69382e85 | 2037 | } |
78d7f5f7 N |
2038 | } else |
2039 | j = 0; | |
2040 | if (j >= 0) | |
7f7583d4 | 2041 | atomic64_add(r1_bio->sectors, &mddev->resync_mismatches); |
78d7f5f7 | 2042 | if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery) |
4246a0b6 | 2043 | && !error)) { |
78d7f5f7 N |
2044 | /* No need to write to this device. */ |
2045 | sbio->bi_end_io = NULL; | |
2046 | rdev_dec_pending(conf->mirrors[i].rdev, mddev); | |
2047 | continue; | |
2048 | } | |
d3b45c2a KO |
2049 | |
2050 | bio_copy_data(sbio, pbio); | |
78d7f5f7 | 2051 | } |
a68e5870 N |
2052 | } |
2053 | ||
9f2c9d12 | 2054 | static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio) |
a68e5870 | 2055 | { |
e8096360 | 2056 | struct r1conf *conf = mddev->private; |
a68e5870 | 2057 | int i; |
8f19ccb2 | 2058 | int disks = conf->raid_disks * 2; |
a68e5870 N |
2059 | struct bio *bio, *wbio; |
2060 | ||
2061 | bio = r1_bio->bios[r1_bio->read_disk]; | |
2062 | ||
a68e5870 N |
2063 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) |
2064 | /* ouch - failed to read all of that. */ | |
2065 | if (!fix_sync_read_error(r1_bio)) | |
2066 | return; | |
7ca78d57 N |
2067 | |
2068 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
c95e6385 N |
2069 | process_checks(r1_bio); |
2070 | ||
d11c171e N |
2071 | /* |
2072 | * schedule writes | |
2073 | */ | |
1da177e4 LT |
2074 | atomic_set(&r1_bio->remaining, 1); |
2075 | for (i = 0; i < disks ; i++) { | |
2076 | wbio = r1_bio->bios[i]; | |
3e198f78 N |
2077 | if (wbio->bi_end_io == NULL || |
2078 | (wbio->bi_end_io == end_sync_read && | |
2079 | (i == r1_bio->read_disk || | |
2080 | !test_bit(MD_RECOVERY_SYNC, &mddev->recovery)))) | |
1da177e4 LT |
2081 | continue; |
2082 | ||
796a5cf0 | 2083 | bio_set_op_attrs(wbio, REQ_OP_WRITE, 0); |
212e7eb7 N |
2084 | if (test_bit(FailFast, &conf->mirrors[i].rdev->flags)) |
2085 | wbio->bi_opf |= MD_FAILFAST; | |
2086 | ||
3e198f78 | 2087 | wbio->bi_end_io = end_sync_write; |
1da177e4 | 2088 | atomic_inc(&r1_bio->remaining); |
aa8b57aa | 2089 | md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio)); |
191ea9b2 | 2090 | |
1da177e4 LT |
2091 | generic_make_request(wbio); |
2092 | } | |
2093 | ||
2094 | if (atomic_dec_and_test(&r1_bio->remaining)) { | |
191ea9b2 | 2095 | /* if we're here, all write(s) have completed, so clean up */ |
58e94ae1 N |
2096 | int s = r1_bio->sectors; |
2097 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || | |
2098 | test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2099 | reschedule_retry(r1_bio); | |
2100 | else { | |
2101 | put_buf(r1_bio); | |
2102 | md_done_sync(mddev, s, 1); | |
2103 | } | |
1da177e4 LT |
2104 | } |
2105 | } | |
2106 | ||
2107 | /* | |
2108 | * This is a kernel thread which: | |
2109 | * | |
2110 | * 1. Retries failed read operations on working mirrors. | |
2111 | * 2. Updates the raid superblock when problems encounter. | |
d2eb35ac | 2112 | * 3. Performs writes following reads for array synchronising. |
1da177e4 LT |
2113 | */ |
2114 | ||
e8096360 | 2115 | static void fix_read_error(struct r1conf *conf, int read_disk, |
867868fb N |
2116 | sector_t sect, int sectors) |
2117 | { | |
fd01b88c | 2118 | struct mddev *mddev = conf->mddev; |
867868fb N |
2119 | while(sectors) { |
2120 | int s = sectors; | |
2121 | int d = read_disk; | |
2122 | int success = 0; | |
2123 | int start; | |
3cb03002 | 2124 | struct md_rdev *rdev; |
867868fb N |
2125 | |
2126 | if (s > (PAGE_SIZE>>9)) | |
2127 | s = PAGE_SIZE >> 9; | |
2128 | ||
2129 | do { | |
d2eb35ac N |
2130 | sector_t first_bad; |
2131 | int bad_sectors; | |
2132 | ||
707a6a42 N |
2133 | rcu_read_lock(); |
2134 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
867868fb | 2135 | if (rdev && |
da8840a7 | 2136 | (test_bit(In_sync, &rdev->flags) || |
2137 | (!test_bit(Faulty, &rdev->flags) && | |
2138 | rdev->recovery_offset >= sect + s)) && | |
d2eb35ac | 2139 | is_badblock(rdev, sect, s, |
707a6a42 N |
2140 | &first_bad, &bad_sectors) == 0) { |
2141 | atomic_inc(&rdev->nr_pending); | |
2142 | rcu_read_unlock(); | |
2143 | if (sync_page_io(rdev, sect, s<<9, | |
796a5cf0 | 2144 | conf->tmppage, REQ_OP_READ, 0, false)) |
707a6a42 N |
2145 | success = 1; |
2146 | rdev_dec_pending(rdev, mddev); | |
2147 | if (success) | |
2148 | break; | |
2149 | } else | |
2150 | rcu_read_unlock(); | |
2151 | d++; | |
2152 | if (d == conf->raid_disks * 2) | |
2153 | d = 0; | |
867868fb N |
2154 | } while (!success && d != read_disk); |
2155 | ||
2156 | if (!success) { | |
d8f05d29 | 2157 | /* Cannot read from anywhere - mark it bad */ |
3cb03002 | 2158 | struct md_rdev *rdev = conf->mirrors[read_disk].rdev; |
d8f05d29 N |
2159 | if (!rdev_set_badblocks(rdev, sect, s, 0)) |
2160 | md_error(mddev, rdev); | |
867868fb N |
2161 | break; |
2162 | } | |
2163 | /* write it back and re-read */ | |
2164 | start = d; | |
2165 | while (d != read_disk) { | |
2166 | if (d==0) | |
8f19ccb2 | 2167 | d = conf->raid_disks * 2; |
867868fb | 2168 | d--; |
707a6a42 N |
2169 | rcu_read_lock(); |
2170 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
867868fb | 2171 | if (rdev && |
707a6a42 N |
2172 | !test_bit(Faulty, &rdev->flags)) { |
2173 | atomic_inc(&rdev->nr_pending); | |
2174 | rcu_read_unlock(); | |
d8f05d29 N |
2175 | r1_sync_page_io(rdev, sect, s, |
2176 | conf->tmppage, WRITE); | |
707a6a42 N |
2177 | rdev_dec_pending(rdev, mddev); |
2178 | } else | |
2179 | rcu_read_unlock(); | |
867868fb N |
2180 | } |
2181 | d = start; | |
2182 | while (d != read_disk) { | |
2183 | char b[BDEVNAME_SIZE]; | |
2184 | if (d==0) | |
8f19ccb2 | 2185 | d = conf->raid_disks * 2; |
867868fb | 2186 | d--; |
707a6a42 N |
2187 | rcu_read_lock(); |
2188 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
867868fb | 2189 | if (rdev && |
b8cb6b4c | 2190 | !test_bit(Faulty, &rdev->flags)) { |
707a6a42 N |
2191 | atomic_inc(&rdev->nr_pending); |
2192 | rcu_read_unlock(); | |
d8f05d29 N |
2193 | if (r1_sync_page_io(rdev, sect, s, |
2194 | conf->tmppage, READ)) { | |
867868fb | 2195 | atomic_add(s, &rdev->corrected_errors); |
1d41c216 N |
2196 | pr_info("md/raid1:%s: read error corrected (%d sectors at %llu on %s)\n", |
2197 | mdname(mddev), s, | |
2198 | (unsigned long long)(sect + | |
2199 | rdev->data_offset), | |
2200 | bdevname(rdev->bdev, b)); | |
867868fb | 2201 | } |
707a6a42 N |
2202 | rdev_dec_pending(rdev, mddev); |
2203 | } else | |
2204 | rcu_read_unlock(); | |
867868fb N |
2205 | } |
2206 | sectors -= s; | |
2207 | sect += s; | |
2208 | } | |
2209 | } | |
2210 | ||
9f2c9d12 | 2211 | static int narrow_write_error(struct r1bio *r1_bio, int i) |
cd5ff9a1 | 2212 | { |
fd01b88c | 2213 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 2214 | struct r1conf *conf = mddev->private; |
3cb03002 | 2215 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
cd5ff9a1 N |
2216 | |
2217 | /* bio has the data to be written to device 'i' where | |
2218 | * we just recently had a write error. | |
2219 | * We repeatedly clone the bio and trim down to one block, | |
2220 | * then try the write. Where the write fails we record | |
2221 | * a bad block. | |
2222 | * It is conceivable that the bio doesn't exactly align with | |
2223 | * blocks. We must handle this somehow. | |
2224 | * | |
2225 | * We currently own a reference on the rdev. | |
2226 | */ | |
2227 | ||
2228 | int block_sectors; | |
2229 | sector_t sector; | |
2230 | int sectors; | |
2231 | int sect_to_write = r1_bio->sectors; | |
2232 | int ok = 1; | |
2233 | ||
2234 | if (rdev->badblocks.shift < 0) | |
2235 | return 0; | |
2236 | ||
ab713cdc ND |
2237 | block_sectors = roundup(1 << rdev->badblocks.shift, |
2238 | bdev_logical_block_size(rdev->bdev) >> 9); | |
cd5ff9a1 N |
2239 | sector = r1_bio->sector; |
2240 | sectors = ((sector + block_sectors) | |
2241 | & ~(sector_t)(block_sectors - 1)) | |
2242 | - sector; | |
2243 | ||
cd5ff9a1 N |
2244 | while (sect_to_write) { |
2245 | struct bio *wbio; | |
2246 | if (sectors > sect_to_write) | |
2247 | sectors = sect_to_write; | |
2248 | /* Write at 'sector' for 'sectors'*/ | |
2249 | ||
b783863f KO |
2250 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { |
2251 | unsigned vcnt = r1_bio->behind_page_count; | |
2252 | struct bio_vec *vec = r1_bio->behind_bvecs; | |
2253 | ||
2254 | while (!vec->bv_page) { | |
2255 | vec++; | |
2256 | vcnt--; | |
2257 | } | |
2258 | ||
2259 | wbio = bio_alloc_mddev(GFP_NOIO, vcnt, mddev); | |
2260 | memcpy(wbio->bi_io_vec, vec, vcnt * sizeof(struct bio_vec)); | |
2261 | ||
2262 | wbio->bi_vcnt = vcnt; | |
2263 | } else { | |
2264 | wbio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev); | |
2265 | } | |
2266 | ||
796a5cf0 | 2267 | bio_set_op_attrs(wbio, REQ_OP_WRITE, 0); |
4f024f37 KO |
2268 | wbio->bi_iter.bi_sector = r1_bio->sector; |
2269 | wbio->bi_iter.bi_size = r1_bio->sectors << 9; | |
cd5ff9a1 | 2270 | |
6678d83f | 2271 | bio_trim(wbio, sector - r1_bio->sector, sectors); |
4f024f37 | 2272 | wbio->bi_iter.bi_sector += rdev->data_offset; |
cd5ff9a1 | 2273 | wbio->bi_bdev = rdev->bdev; |
4e49ea4a MC |
2274 | |
2275 | if (submit_bio_wait(wbio) < 0) | |
cd5ff9a1 N |
2276 | /* failure! */ |
2277 | ok = rdev_set_badblocks(rdev, sector, | |
2278 | sectors, 0) | |
2279 | && ok; | |
2280 | ||
2281 | bio_put(wbio); | |
2282 | sect_to_write -= sectors; | |
2283 | sector += sectors; | |
2284 | sectors = block_sectors; | |
2285 | } | |
2286 | return ok; | |
2287 | } | |
2288 | ||
e8096360 | 2289 | static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce N |
2290 | { |
2291 | int m; | |
2292 | int s = r1_bio->sectors; | |
8f19ccb2 | 2293 | for (m = 0; m < conf->raid_disks * 2 ; m++) { |
3cb03002 | 2294 | struct md_rdev *rdev = conf->mirrors[m].rdev; |
62096bce N |
2295 | struct bio *bio = r1_bio->bios[m]; |
2296 | if (bio->bi_end_io == NULL) | |
2297 | continue; | |
4246a0b6 | 2298 | if (!bio->bi_error && |
62096bce | 2299 | test_bit(R1BIO_MadeGood, &r1_bio->state)) { |
c6563a8c | 2300 | rdev_clear_badblocks(rdev, r1_bio->sector, s, 0); |
62096bce | 2301 | } |
4246a0b6 | 2302 | if (bio->bi_error && |
62096bce N |
2303 | test_bit(R1BIO_WriteError, &r1_bio->state)) { |
2304 | if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0)) | |
2305 | md_error(conf->mddev, rdev); | |
2306 | } | |
2307 | } | |
2308 | put_buf(r1_bio); | |
2309 | md_done_sync(conf->mddev, s, 1); | |
2310 | } | |
2311 | ||
e8096360 | 2312 | static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce N |
2313 | { |
2314 | int m; | |
55ce74d4 | 2315 | bool fail = false; |
8f19ccb2 | 2316 | for (m = 0; m < conf->raid_disks * 2 ; m++) |
62096bce | 2317 | if (r1_bio->bios[m] == IO_MADE_GOOD) { |
3cb03002 | 2318 | struct md_rdev *rdev = conf->mirrors[m].rdev; |
62096bce N |
2319 | rdev_clear_badblocks(rdev, |
2320 | r1_bio->sector, | |
c6563a8c | 2321 | r1_bio->sectors, 0); |
62096bce N |
2322 | rdev_dec_pending(rdev, conf->mddev); |
2323 | } else if (r1_bio->bios[m] != NULL) { | |
2324 | /* This drive got a write error. We need to | |
2325 | * narrow down and record precise write | |
2326 | * errors. | |
2327 | */ | |
55ce74d4 | 2328 | fail = true; |
62096bce N |
2329 | if (!narrow_write_error(r1_bio, m)) { |
2330 | md_error(conf->mddev, | |
2331 | conf->mirrors[m].rdev); | |
2332 | /* an I/O failed, we can't clear the bitmap */ | |
2333 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
2334 | } | |
2335 | rdev_dec_pending(conf->mirrors[m].rdev, | |
2336 | conf->mddev); | |
2337 | } | |
55ce74d4 N |
2338 | if (fail) { |
2339 | spin_lock_irq(&conf->device_lock); | |
2340 | list_add(&r1_bio->retry_list, &conf->bio_end_io_list); | |
ccfc7bf1 | 2341 | conf->nr_queued++; |
55ce74d4 N |
2342 | spin_unlock_irq(&conf->device_lock); |
2343 | md_wakeup_thread(conf->mddev->thread); | |
bd8688a1 N |
2344 | } else { |
2345 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2346 | close_write(r1_bio); | |
55ce74d4 | 2347 | raid_end_bio_io(r1_bio); |
bd8688a1 | 2348 | } |
62096bce N |
2349 | } |
2350 | ||
e8096360 | 2351 | static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce N |
2352 | { |
2353 | int disk; | |
2354 | int max_sectors; | |
fd01b88c | 2355 | struct mddev *mddev = conf->mddev; |
62096bce N |
2356 | struct bio *bio; |
2357 | char b[BDEVNAME_SIZE]; | |
3cb03002 | 2358 | struct md_rdev *rdev; |
109e3765 N |
2359 | dev_t bio_dev; |
2360 | sector_t bio_sector; | |
62096bce N |
2361 | |
2362 | clear_bit(R1BIO_ReadError, &r1_bio->state); | |
2363 | /* we got a read error. Maybe the drive is bad. Maybe just | |
2364 | * the block and we can fix it. | |
2365 | * We freeze all other IO, and try reading the block from | |
2366 | * other devices. When we find one, we re-write | |
2367 | * and check it that fixes the read error. | |
2368 | * This is all done synchronously while the array is | |
2369 | * frozen | |
2370 | */ | |
7449f699 TM |
2371 | |
2372 | bio = r1_bio->bios[r1_bio->read_disk]; | |
2373 | bdevname(bio->bi_bdev, b); | |
109e3765 N |
2374 | bio_dev = bio->bi_bdev->bd_dev; |
2375 | bio_sector = conf->mirrors[r1_bio->read_disk].rdev->data_offset + r1_bio->sector; | |
7449f699 TM |
2376 | bio_put(bio); |
2377 | r1_bio->bios[r1_bio->read_disk] = NULL; | |
2378 | ||
2e52d449 N |
2379 | rdev = conf->mirrors[r1_bio->read_disk].rdev; |
2380 | if (mddev->ro == 0 | |
2381 | && !test_bit(FailFast, &rdev->flags)) { | |
e2d59925 | 2382 | freeze_array(conf, 1); |
62096bce N |
2383 | fix_read_error(conf, r1_bio->read_disk, |
2384 | r1_bio->sector, r1_bio->sectors); | |
2385 | unfreeze_array(conf); | |
7449f699 TM |
2386 | } else { |
2387 | r1_bio->bios[r1_bio->read_disk] = IO_BLOCKED; | |
2388 | } | |
2389 | ||
2e52d449 | 2390 | rdev_dec_pending(rdev, conf->mddev); |
62096bce | 2391 | |
62096bce N |
2392 | read_more: |
2393 | disk = read_balance(conf, r1_bio, &max_sectors); | |
2394 | if (disk == -1) { | |
1d41c216 N |
2395 | pr_crit_ratelimited("md/raid1:%s: %s: unrecoverable I/O read error for block %llu\n", |
2396 | mdname(mddev), b, (unsigned long long)r1_bio->sector); | |
62096bce N |
2397 | raid_end_bio_io(r1_bio); |
2398 | } else { | |
2399 | const unsigned long do_sync | |
1eff9d32 | 2400 | = r1_bio->master_bio->bi_opf & REQ_SYNC; |
62096bce N |
2401 | r1_bio->read_disk = disk; |
2402 | bio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev); | |
4f024f37 KO |
2403 | bio_trim(bio, r1_bio->sector - bio->bi_iter.bi_sector, |
2404 | max_sectors); | |
62096bce N |
2405 | r1_bio->bios[r1_bio->read_disk] = bio; |
2406 | rdev = conf->mirrors[disk].rdev; | |
1d41c216 N |
2407 | pr_info_ratelimited("md/raid1:%s: redirecting sector %llu to other mirror: %s\n", |
2408 | mdname(mddev), | |
2409 | (unsigned long long)r1_bio->sector, | |
2410 | bdevname(rdev->bdev, b)); | |
4f024f37 | 2411 | bio->bi_iter.bi_sector = r1_bio->sector + rdev->data_offset; |
62096bce N |
2412 | bio->bi_bdev = rdev->bdev; |
2413 | bio->bi_end_io = raid1_end_read_request; | |
796a5cf0 | 2414 | bio_set_op_attrs(bio, REQ_OP_READ, do_sync); |
2e52d449 N |
2415 | if (test_bit(FailFast, &rdev->flags) && |
2416 | test_bit(R1BIO_FailFast, &r1_bio->state)) | |
2417 | bio->bi_opf |= MD_FAILFAST; | |
62096bce N |
2418 | bio->bi_private = r1_bio; |
2419 | if (max_sectors < r1_bio->sectors) { | |
2420 | /* Drat - have to split this up more */ | |
2421 | struct bio *mbio = r1_bio->master_bio; | |
2422 | int sectors_handled = (r1_bio->sector + max_sectors | |
4f024f37 | 2423 | - mbio->bi_iter.bi_sector); |
62096bce N |
2424 | r1_bio->sectors = max_sectors; |
2425 | spin_lock_irq(&conf->device_lock); | |
2426 | if (mbio->bi_phys_segments == 0) | |
2427 | mbio->bi_phys_segments = 2; | |
2428 | else | |
2429 | mbio->bi_phys_segments++; | |
2430 | spin_unlock_irq(&conf->device_lock); | |
109e3765 N |
2431 | trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), |
2432 | bio, bio_dev, bio_sector); | |
62096bce N |
2433 | generic_make_request(bio); |
2434 | bio = NULL; | |
2435 | ||
2436 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
2437 | ||
2438 | r1_bio->master_bio = mbio; | |
aa8b57aa | 2439 | r1_bio->sectors = bio_sectors(mbio) - sectors_handled; |
62096bce N |
2440 | r1_bio->state = 0; |
2441 | set_bit(R1BIO_ReadError, &r1_bio->state); | |
2442 | r1_bio->mddev = mddev; | |
4f024f37 KO |
2443 | r1_bio->sector = mbio->bi_iter.bi_sector + |
2444 | sectors_handled; | |
62096bce N |
2445 | |
2446 | goto read_more; | |
109e3765 N |
2447 | } else { |
2448 | trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), | |
2449 | bio, bio_dev, bio_sector); | |
62096bce | 2450 | generic_make_request(bio); |
109e3765 | 2451 | } |
62096bce N |
2452 | } |
2453 | } | |
2454 | ||
4ed8731d | 2455 | static void raid1d(struct md_thread *thread) |
1da177e4 | 2456 | { |
4ed8731d | 2457 | struct mddev *mddev = thread->mddev; |
9f2c9d12 | 2458 | struct r1bio *r1_bio; |
1da177e4 | 2459 | unsigned long flags; |
e8096360 | 2460 | struct r1conf *conf = mddev->private; |
1da177e4 | 2461 | struct list_head *head = &conf->retry_list; |
e1dfa0a2 | 2462 | struct blk_plug plug; |
1da177e4 LT |
2463 | |
2464 | md_check_recovery(mddev); | |
e1dfa0a2 | 2465 | |
55ce74d4 N |
2466 | if (!list_empty_careful(&conf->bio_end_io_list) && |
2467 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)) { | |
2468 | LIST_HEAD(tmp); | |
2469 | spin_lock_irqsave(&conf->device_lock, flags); | |
2470 | if (!test_bit(MD_CHANGE_PENDING, &mddev->flags)) { | |
ccfc7bf1 ND |
2471 | while (!list_empty(&conf->bio_end_io_list)) { |
2472 | list_move(conf->bio_end_io_list.prev, &tmp); | |
2473 | conf->nr_queued--; | |
2474 | } | |
55ce74d4 N |
2475 | } |
2476 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
2477 | while (!list_empty(&tmp)) { | |
a452744b MP |
2478 | r1_bio = list_first_entry(&tmp, struct r1bio, |
2479 | retry_list); | |
55ce74d4 | 2480 | list_del(&r1_bio->retry_list); |
bd8688a1 N |
2481 | if (mddev->degraded) |
2482 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
2483 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2484 | close_write(r1_bio); | |
55ce74d4 N |
2485 | raid_end_bio_io(r1_bio); |
2486 | } | |
2487 | } | |
2488 | ||
e1dfa0a2 | 2489 | blk_start_plug(&plug); |
1da177e4 | 2490 | for (;;) { |
191ea9b2 | 2491 | |
0021b7bc | 2492 | flush_pending_writes(conf); |
191ea9b2 | 2493 | |
a35e63ef N |
2494 | spin_lock_irqsave(&conf->device_lock, flags); |
2495 | if (list_empty(head)) { | |
2496 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 | 2497 | break; |
a35e63ef | 2498 | } |
9f2c9d12 | 2499 | r1_bio = list_entry(head->prev, struct r1bio, retry_list); |
1da177e4 | 2500 | list_del(head->prev); |
ddaf22ab | 2501 | conf->nr_queued--; |
1da177e4 LT |
2502 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2503 | ||
2504 | mddev = r1_bio->mddev; | |
070ec55d | 2505 | conf = mddev->private; |
4367af55 | 2506 | if (test_bit(R1BIO_IsSync, &r1_bio->state)) { |
d8f05d29 | 2507 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
62096bce N |
2508 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
2509 | handle_sync_write_finished(conf, r1_bio); | |
2510 | else | |
4367af55 | 2511 | sync_request_write(mddev, r1_bio); |
cd5ff9a1 | 2512 | } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
62096bce N |
2513 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
2514 | handle_write_finished(conf, r1_bio); | |
2515 | else if (test_bit(R1BIO_ReadError, &r1_bio->state)) | |
2516 | handle_read_error(conf, r1_bio); | |
2517 | else | |
d2eb35ac N |
2518 | /* just a partial read to be scheduled from separate |
2519 | * context | |
2520 | */ | |
2521 | generic_make_request(r1_bio->bios[r1_bio->read_disk]); | |
62096bce | 2522 | |
1d9d5241 | 2523 | cond_resched(); |
de393cde N |
2524 | if (mddev->flags & ~(1<<MD_CHANGE_PENDING)) |
2525 | md_check_recovery(mddev); | |
1da177e4 | 2526 | } |
e1dfa0a2 | 2527 | blk_finish_plug(&plug); |
1da177e4 LT |
2528 | } |
2529 | ||
e8096360 | 2530 | static int init_resync(struct r1conf *conf) |
1da177e4 LT |
2531 | { |
2532 | int buffs; | |
2533 | ||
2534 | buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; | |
9e77c485 | 2535 | BUG_ON(conf->r1buf_pool); |
1da177e4 LT |
2536 | conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free, |
2537 | conf->poolinfo); | |
2538 | if (!conf->r1buf_pool) | |
2539 | return -ENOMEM; | |
2540 | conf->next_resync = 0; | |
2541 | return 0; | |
2542 | } | |
2543 | ||
2544 | /* | |
2545 | * perform a "sync" on one "block" | |
2546 | * | |
2547 | * We need to make sure that no normal I/O request - particularly write | |
2548 | * requests - conflict with active sync requests. | |
2549 | * | |
2550 | * This is achieved by tracking pending requests and a 'barrier' concept | |
2551 | * that can be installed to exclude normal IO requests. | |
2552 | */ | |
2553 | ||
849674e4 SL |
2554 | static sector_t raid1_sync_request(struct mddev *mddev, sector_t sector_nr, |
2555 | int *skipped) | |
1da177e4 | 2556 | { |
e8096360 | 2557 | struct r1conf *conf = mddev->private; |
9f2c9d12 | 2558 | struct r1bio *r1_bio; |
1da177e4 LT |
2559 | struct bio *bio; |
2560 | sector_t max_sector, nr_sectors; | |
3e198f78 | 2561 | int disk = -1; |
1da177e4 | 2562 | int i; |
3e198f78 N |
2563 | int wonly = -1; |
2564 | int write_targets = 0, read_targets = 0; | |
57dab0bd | 2565 | sector_t sync_blocks; |
e3b9703e | 2566 | int still_degraded = 0; |
06f60385 N |
2567 | int good_sectors = RESYNC_SECTORS; |
2568 | int min_bad = 0; /* number of sectors that are bad in all devices */ | |
1da177e4 LT |
2569 | |
2570 | if (!conf->r1buf_pool) | |
2571 | if (init_resync(conf)) | |
57afd89f | 2572 | return 0; |
1da177e4 | 2573 | |
58c0fed4 | 2574 | max_sector = mddev->dev_sectors; |
1da177e4 | 2575 | if (sector_nr >= max_sector) { |
191ea9b2 N |
2576 | /* If we aborted, we need to abort the |
2577 | * sync on the 'current' bitmap chunk (there will | |
2578 | * only be one in raid1 resync. | |
2579 | * We can find the current addess in mddev->curr_resync | |
2580 | */ | |
6a806c51 N |
2581 | if (mddev->curr_resync < max_sector) /* aborted */ |
2582 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
191ea9b2 | 2583 | &sync_blocks, 1); |
6a806c51 | 2584 | else /* completed sync */ |
191ea9b2 | 2585 | conf->fullsync = 0; |
6a806c51 N |
2586 | |
2587 | bitmap_close_sync(mddev->bitmap); | |
1da177e4 | 2588 | close_sync(conf); |
c40f341f GR |
2589 | |
2590 | if (mddev_is_clustered(mddev)) { | |
2591 | conf->cluster_sync_low = 0; | |
2592 | conf->cluster_sync_high = 0; | |
c40f341f | 2593 | } |
1da177e4 LT |
2594 | return 0; |
2595 | } | |
2596 | ||
07d84d10 N |
2597 | if (mddev->bitmap == NULL && |
2598 | mddev->recovery_cp == MaxSector && | |
6394cca5 | 2599 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
07d84d10 N |
2600 | conf->fullsync == 0) { |
2601 | *skipped = 1; | |
2602 | return max_sector - sector_nr; | |
2603 | } | |
6394cca5 N |
2604 | /* before building a request, check if we can skip these blocks.. |
2605 | * This call the bitmap_start_sync doesn't actually record anything | |
2606 | */ | |
e3b9703e | 2607 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
e5de485f | 2608 | !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { |
191ea9b2 N |
2609 | /* We can skip this block, and probably several more */ |
2610 | *skipped = 1; | |
2611 | return sync_blocks; | |
2612 | } | |
17999be4 | 2613 | |
7ac50447 TM |
2614 | /* |
2615 | * If there is non-resync activity waiting for a turn, then let it | |
2616 | * though before starting on this new sync request. | |
2617 | */ | |
2618 | if (conf->nr_waiting) | |
2619 | schedule_timeout_uninterruptible(1); | |
2620 | ||
c40f341f GR |
2621 | /* we are incrementing sector_nr below. To be safe, we check against |
2622 | * sector_nr + two times RESYNC_SECTORS | |
2623 | */ | |
2624 | ||
2625 | bitmap_cond_end_sync(mddev->bitmap, sector_nr, | |
2626 | mddev_is_clustered(mddev) && (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high)); | |
1c4588e9 | 2627 | r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO); |
17999be4 | 2628 | |
c2fd4c94 | 2629 | raise_barrier(conf, sector_nr); |
1da177e4 | 2630 | |
3e198f78 | 2631 | rcu_read_lock(); |
1da177e4 | 2632 | /* |
3e198f78 N |
2633 | * If we get a correctably read error during resync or recovery, |
2634 | * we might want to read from a different device. So we | |
2635 | * flag all drives that could conceivably be read from for READ, | |
2636 | * and any others (which will be non-In_sync devices) for WRITE. | |
2637 | * If a read fails, we try reading from something else for which READ | |
2638 | * is OK. | |
1da177e4 | 2639 | */ |
1da177e4 | 2640 | |
1da177e4 LT |
2641 | r1_bio->mddev = mddev; |
2642 | r1_bio->sector = sector_nr; | |
191ea9b2 | 2643 | r1_bio->state = 0; |
1da177e4 | 2644 | set_bit(R1BIO_IsSync, &r1_bio->state); |
1da177e4 | 2645 | |
8f19ccb2 | 2646 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3cb03002 | 2647 | struct md_rdev *rdev; |
1da177e4 | 2648 | bio = r1_bio->bios[i]; |
2aabaa65 | 2649 | bio_reset(bio); |
1da177e4 | 2650 | |
3e198f78 N |
2651 | rdev = rcu_dereference(conf->mirrors[i].rdev); |
2652 | if (rdev == NULL || | |
06f60385 | 2653 | test_bit(Faulty, &rdev->flags)) { |
8f19ccb2 N |
2654 | if (i < conf->raid_disks) |
2655 | still_degraded = 1; | |
3e198f78 | 2656 | } else if (!test_bit(In_sync, &rdev->flags)) { |
796a5cf0 | 2657 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
1da177e4 LT |
2658 | bio->bi_end_io = end_sync_write; |
2659 | write_targets ++; | |
3e198f78 N |
2660 | } else { |
2661 | /* may need to read from here */ | |
06f60385 N |
2662 | sector_t first_bad = MaxSector; |
2663 | int bad_sectors; | |
2664 | ||
2665 | if (is_badblock(rdev, sector_nr, good_sectors, | |
2666 | &first_bad, &bad_sectors)) { | |
2667 | if (first_bad > sector_nr) | |
2668 | good_sectors = first_bad - sector_nr; | |
2669 | else { | |
2670 | bad_sectors -= (sector_nr - first_bad); | |
2671 | if (min_bad == 0 || | |
2672 | min_bad > bad_sectors) | |
2673 | min_bad = bad_sectors; | |
2674 | } | |
2675 | } | |
2676 | if (sector_nr < first_bad) { | |
2677 | if (test_bit(WriteMostly, &rdev->flags)) { | |
2678 | if (wonly < 0) | |
2679 | wonly = i; | |
2680 | } else { | |
2681 | if (disk < 0) | |
2682 | disk = i; | |
2683 | } | |
796a5cf0 | 2684 | bio_set_op_attrs(bio, REQ_OP_READ, 0); |
06f60385 N |
2685 | bio->bi_end_io = end_sync_read; |
2686 | read_targets++; | |
d57368af AL |
2687 | } else if (!test_bit(WriteErrorSeen, &rdev->flags) && |
2688 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && | |
2689 | !test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { | |
2690 | /* | |
2691 | * The device is suitable for reading (InSync), | |
2692 | * but has bad block(s) here. Let's try to correct them, | |
2693 | * if we are doing resync or repair. Otherwise, leave | |
2694 | * this device alone for this sync request. | |
2695 | */ | |
796a5cf0 | 2696 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
d57368af AL |
2697 | bio->bi_end_io = end_sync_write; |
2698 | write_targets++; | |
3e198f78 | 2699 | } |
3e198f78 | 2700 | } |
06f60385 N |
2701 | if (bio->bi_end_io) { |
2702 | atomic_inc(&rdev->nr_pending); | |
4f024f37 | 2703 | bio->bi_iter.bi_sector = sector_nr + rdev->data_offset; |
06f60385 N |
2704 | bio->bi_bdev = rdev->bdev; |
2705 | bio->bi_private = r1_bio; | |
2e52d449 N |
2706 | if (test_bit(FailFast, &rdev->flags)) |
2707 | bio->bi_opf |= MD_FAILFAST; | |
06f60385 | 2708 | } |
1da177e4 | 2709 | } |
3e198f78 N |
2710 | rcu_read_unlock(); |
2711 | if (disk < 0) | |
2712 | disk = wonly; | |
2713 | r1_bio->read_disk = disk; | |
191ea9b2 | 2714 | |
06f60385 N |
2715 | if (read_targets == 0 && min_bad > 0) { |
2716 | /* These sectors are bad on all InSync devices, so we | |
2717 | * need to mark them bad on all write targets | |
2718 | */ | |
2719 | int ok = 1; | |
8f19ccb2 | 2720 | for (i = 0 ; i < conf->raid_disks * 2 ; i++) |
06f60385 | 2721 | if (r1_bio->bios[i]->bi_end_io == end_sync_write) { |
a42f9d83 | 2722 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
06f60385 N |
2723 | ok = rdev_set_badblocks(rdev, sector_nr, |
2724 | min_bad, 0 | |
2725 | ) && ok; | |
2726 | } | |
2727 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
2728 | *skipped = 1; | |
2729 | put_buf(r1_bio); | |
2730 | ||
2731 | if (!ok) { | |
2732 | /* Cannot record the badblocks, so need to | |
2733 | * abort the resync. | |
2734 | * If there are multiple read targets, could just | |
2735 | * fail the really bad ones ??? | |
2736 | */ | |
2737 | conf->recovery_disabled = mddev->recovery_disabled; | |
2738 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2739 | return 0; | |
2740 | } else | |
2741 | return min_bad; | |
2742 | ||
2743 | } | |
2744 | if (min_bad > 0 && min_bad < good_sectors) { | |
2745 | /* only resync enough to reach the next bad->good | |
2746 | * transition */ | |
2747 | good_sectors = min_bad; | |
2748 | } | |
2749 | ||
3e198f78 N |
2750 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0) |
2751 | /* extra read targets are also write targets */ | |
2752 | write_targets += read_targets-1; | |
2753 | ||
2754 | if (write_targets == 0 || read_targets == 0) { | |
1da177e4 LT |
2755 | /* There is nowhere to write, so all non-sync |
2756 | * drives must be failed - so we are finished | |
2757 | */ | |
b7219ccb N |
2758 | sector_t rv; |
2759 | if (min_bad > 0) | |
2760 | max_sector = sector_nr + min_bad; | |
2761 | rv = max_sector - sector_nr; | |
57afd89f | 2762 | *skipped = 1; |
1da177e4 | 2763 | put_buf(r1_bio); |
1da177e4 LT |
2764 | return rv; |
2765 | } | |
2766 | ||
c6207277 N |
2767 | if (max_sector > mddev->resync_max) |
2768 | max_sector = mddev->resync_max; /* Don't do IO beyond here */ | |
06f60385 N |
2769 | if (max_sector > sector_nr + good_sectors) |
2770 | max_sector = sector_nr + good_sectors; | |
1da177e4 | 2771 | nr_sectors = 0; |
289e99e8 | 2772 | sync_blocks = 0; |
1da177e4 LT |
2773 | do { |
2774 | struct page *page; | |
2775 | int len = PAGE_SIZE; | |
2776 | if (sector_nr + (len>>9) > max_sector) | |
2777 | len = (max_sector - sector_nr) << 9; | |
2778 | if (len == 0) | |
2779 | break; | |
6a806c51 N |
2780 | if (sync_blocks == 0) { |
2781 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, | |
e5de485f N |
2782 | &sync_blocks, still_degraded) && |
2783 | !conf->fullsync && | |
2784 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6a806c51 | 2785 | break; |
7571ae88 | 2786 | if ((len >> 9) > sync_blocks) |
6a806c51 | 2787 | len = sync_blocks<<9; |
ab7a30c7 | 2788 | } |
191ea9b2 | 2789 | |
8f19ccb2 | 2790 | for (i = 0 ; i < conf->raid_disks * 2; i++) { |
1da177e4 LT |
2791 | bio = r1_bio->bios[i]; |
2792 | if (bio->bi_end_io) { | |
d11c171e | 2793 | page = bio->bi_io_vec[bio->bi_vcnt].bv_page; |
1da177e4 LT |
2794 | if (bio_add_page(bio, page, len, 0) == 0) { |
2795 | /* stop here */ | |
d11c171e | 2796 | bio->bi_io_vec[bio->bi_vcnt].bv_page = page; |
1da177e4 LT |
2797 | while (i > 0) { |
2798 | i--; | |
2799 | bio = r1_bio->bios[i]; | |
6a806c51 N |
2800 | if (bio->bi_end_io==NULL) |
2801 | continue; | |
1da177e4 LT |
2802 | /* remove last page from this bio */ |
2803 | bio->bi_vcnt--; | |
4f024f37 | 2804 | bio->bi_iter.bi_size -= len; |
b7c44ed9 | 2805 | bio_clear_flag(bio, BIO_SEG_VALID); |
1da177e4 LT |
2806 | } |
2807 | goto bio_full; | |
2808 | } | |
2809 | } | |
2810 | } | |
2811 | nr_sectors += len>>9; | |
2812 | sector_nr += len>>9; | |
191ea9b2 | 2813 | sync_blocks -= (len>>9); |
1da177e4 LT |
2814 | } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES); |
2815 | bio_full: | |
1da177e4 LT |
2816 | r1_bio->sectors = nr_sectors; |
2817 | ||
c40f341f GR |
2818 | if (mddev_is_clustered(mddev) && |
2819 | conf->cluster_sync_high < sector_nr + nr_sectors) { | |
2820 | conf->cluster_sync_low = mddev->curr_resync_completed; | |
2821 | conf->cluster_sync_high = conf->cluster_sync_low + CLUSTER_RESYNC_WINDOW_SECTORS; | |
2822 | /* Send resync message */ | |
2823 | md_cluster_ops->resync_info_update(mddev, | |
2824 | conf->cluster_sync_low, | |
2825 | conf->cluster_sync_high); | |
2826 | } | |
2827 | ||
d11c171e N |
2828 | /* For a user-requested sync, we read all readable devices and do a |
2829 | * compare | |
2830 | */ | |
2831 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
2832 | atomic_set(&r1_bio->remaining, read_targets); | |
2d4f4f33 | 2833 | for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) { |
d11c171e N |
2834 | bio = r1_bio->bios[i]; |
2835 | if (bio->bi_end_io == end_sync_read) { | |
2d4f4f33 | 2836 | read_targets--; |
ddac7c7e | 2837 | md_sync_acct(bio->bi_bdev, nr_sectors); |
2e52d449 N |
2838 | if (read_targets == 1) |
2839 | bio->bi_opf &= ~MD_FAILFAST; | |
d11c171e N |
2840 | generic_make_request(bio); |
2841 | } | |
2842 | } | |
2843 | } else { | |
2844 | atomic_set(&r1_bio->remaining, 1); | |
2845 | bio = r1_bio->bios[r1_bio->read_disk]; | |
ddac7c7e | 2846 | md_sync_acct(bio->bi_bdev, nr_sectors); |
2e52d449 N |
2847 | if (read_targets == 1) |
2848 | bio->bi_opf &= ~MD_FAILFAST; | |
d11c171e | 2849 | generic_make_request(bio); |
1da177e4 | 2850 | |
d11c171e | 2851 | } |
1da177e4 LT |
2852 | return nr_sectors; |
2853 | } | |
2854 | ||
fd01b88c | 2855 | static sector_t raid1_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce DW |
2856 | { |
2857 | if (sectors) | |
2858 | return sectors; | |
2859 | ||
2860 | return mddev->dev_sectors; | |
2861 | } | |
2862 | ||
e8096360 | 2863 | static struct r1conf *setup_conf(struct mddev *mddev) |
1da177e4 | 2864 | { |
e8096360 | 2865 | struct r1conf *conf; |
709ae487 | 2866 | int i; |
0eaf822c | 2867 | struct raid1_info *disk; |
3cb03002 | 2868 | struct md_rdev *rdev; |
709ae487 | 2869 | int err = -ENOMEM; |
1da177e4 | 2870 | |
e8096360 | 2871 | conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL); |
1da177e4 | 2872 | if (!conf) |
709ae487 | 2873 | goto abort; |
1da177e4 | 2874 | |
0eaf822c | 2875 | conf->mirrors = kzalloc(sizeof(struct raid1_info) |
8f19ccb2 | 2876 | * mddev->raid_disks * 2, |
1da177e4 LT |
2877 | GFP_KERNEL); |
2878 | if (!conf->mirrors) | |
709ae487 | 2879 | goto abort; |
1da177e4 | 2880 | |
ddaf22ab N |
2881 | conf->tmppage = alloc_page(GFP_KERNEL); |
2882 | if (!conf->tmppage) | |
709ae487 | 2883 | goto abort; |
ddaf22ab | 2884 | |
709ae487 | 2885 | conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL); |
1da177e4 | 2886 | if (!conf->poolinfo) |
709ae487 | 2887 | goto abort; |
8f19ccb2 | 2888 | conf->poolinfo->raid_disks = mddev->raid_disks * 2; |
1da177e4 LT |
2889 | conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, |
2890 | r1bio_pool_free, | |
2891 | conf->poolinfo); | |
2892 | if (!conf->r1bio_pool) | |
709ae487 N |
2893 | goto abort; |
2894 | ||
ed9bfdf1 | 2895 | conf->poolinfo->mddev = mddev; |
1da177e4 | 2896 | |
c19d5798 | 2897 | err = -EINVAL; |
e7e72bf6 | 2898 | spin_lock_init(&conf->device_lock); |
dafb20fa | 2899 | rdev_for_each(rdev, mddev) { |
aba336bd | 2900 | struct request_queue *q; |
709ae487 | 2901 | int disk_idx = rdev->raid_disk; |
1da177e4 LT |
2902 | if (disk_idx >= mddev->raid_disks |
2903 | || disk_idx < 0) | |
2904 | continue; | |
c19d5798 | 2905 | if (test_bit(Replacement, &rdev->flags)) |
02b898f2 | 2906 | disk = conf->mirrors + mddev->raid_disks + disk_idx; |
c19d5798 N |
2907 | else |
2908 | disk = conf->mirrors + disk_idx; | |
1da177e4 | 2909 | |
c19d5798 N |
2910 | if (disk->rdev) |
2911 | goto abort; | |
1da177e4 | 2912 | disk->rdev = rdev; |
aba336bd | 2913 | q = bdev_get_queue(rdev->bdev); |
1da177e4 LT |
2914 | |
2915 | disk->head_position = 0; | |
12cee5a8 | 2916 | disk->seq_start = MaxSector; |
1da177e4 LT |
2917 | } |
2918 | conf->raid_disks = mddev->raid_disks; | |
2919 | conf->mddev = mddev; | |
1da177e4 | 2920 | INIT_LIST_HEAD(&conf->retry_list); |
55ce74d4 | 2921 | INIT_LIST_HEAD(&conf->bio_end_io_list); |
1da177e4 LT |
2922 | |
2923 | spin_lock_init(&conf->resync_lock); | |
17999be4 | 2924 | init_waitqueue_head(&conf->wait_barrier); |
1da177e4 | 2925 | |
191ea9b2 | 2926 | bio_list_init(&conf->pending_bio_list); |
34db0cd6 | 2927 | conf->pending_count = 0; |
d890fa2b | 2928 | conf->recovery_disabled = mddev->recovery_disabled - 1; |
191ea9b2 | 2929 | |
79ef3a8a | 2930 | conf->start_next_window = MaxSector; |
2931 | conf->current_window_requests = conf->next_window_requests = 0; | |
2932 | ||
c19d5798 | 2933 | err = -EIO; |
8f19ccb2 | 2934 | for (i = 0; i < conf->raid_disks * 2; i++) { |
1da177e4 LT |
2935 | |
2936 | disk = conf->mirrors + i; | |
2937 | ||
c19d5798 N |
2938 | if (i < conf->raid_disks && |
2939 | disk[conf->raid_disks].rdev) { | |
2940 | /* This slot has a replacement. */ | |
2941 | if (!disk->rdev) { | |
2942 | /* No original, just make the replacement | |
2943 | * a recovering spare | |
2944 | */ | |
2945 | disk->rdev = | |
2946 | disk[conf->raid_disks].rdev; | |
2947 | disk[conf->raid_disks].rdev = NULL; | |
2948 | } else if (!test_bit(In_sync, &disk->rdev->flags)) | |
2949 | /* Original is not in_sync - bad */ | |
2950 | goto abort; | |
2951 | } | |
2952 | ||
5fd6c1dc N |
2953 | if (!disk->rdev || |
2954 | !test_bit(In_sync, &disk->rdev->flags)) { | |
1da177e4 | 2955 | disk->head_position = 0; |
4f0a5e01 JB |
2956 | if (disk->rdev && |
2957 | (disk->rdev->saved_raid_disk < 0)) | |
918f0238 | 2958 | conf->fullsync = 1; |
be4d3280 | 2959 | } |
1da177e4 | 2960 | } |
709ae487 | 2961 | |
709ae487 | 2962 | err = -ENOMEM; |
0232605d | 2963 | conf->thread = md_register_thread(raid1d, mddev, "raid1"); |
1d41c216 | 2964 | if (!conf->thread) |
709ae487 | 2965 | goto abort; |
1da177e4 | 2966 | |
709ae487 N |
2967 | return conf; |
2968 | ||
2969 | abort: | |
2970 | if (conf) { | |
644df1a8 | 2971 | mempool_destroy(conf->r1bio_pool); |
709ae487 N |
2972 | kfree(conf->mirrors); |
2973 | safe_put_page(conf->tmppage); | |
2974 | kfree(conf->poolinfo); | |
2975 | kfree(conf); | |
2976 | } | |
2977 | return ERR_PTR(err); | |
2978 | } | |
2979 | ||
afa0f557 | 2980 | static void raid1_free(struct mddev *mddev, void *priv); |
849674e4 | 2981 | static int raid1_run(struct mddev *mddev) |
709ae487 | 2982 | { |
e8096360 | 2983 | struct r1conf *conf; |
709ae487 | 2984 | int i; |
3cb03002 | 2985 | struct md_rdev *rdev; |
5220ea1e | 2986 | int ret; |
2ff8cc2c | 2987 | bool discard_supported = false; |
709ae487 N |
2988 | |
2989 | if (mddev->level != 1) { | |
1d41c216 N |
2990 | pr_warn("md/raid1:%s: raid level not set to mirroring (%d)\n", |
2991 | mdname(mddev), mddev->level); | |
709ae487 N |
2992 | return -EIO; |
2993 | } | |
2994 | if (mddev->reshape_position != MaxSector) { | |
1d41c216 N |
2995 | pr_warn("md/raid1:%s: reshape_position set but not supported\n", |
2996 | mdname(mddev)); | |
709ae487 N |
2997 | return -EIO; |
2998 | } | |
1da177e4 | 2999 | /* |
709ae487 N |
3000 | * copy the already verified devices into our private RAID1 |
3001 | * bookkeeping area. [whatever we allocate in run(), | |
afa0f557 | 3002 | * should be freed in raid1_free()] |
1da177e4 | 3003 | */ |
709ae487 N |
3004 | if (mddev->private == NULL) |
3005 | conf = setup_conf(mddev); | |
3006 | else | |
3007 | conf = mddev->private; | |
1da177e4 | 3008 | |
709ae487 N |
3009 | if (IS_ERR(conf)) |
3010 | return PTR_ERR(conf); | |
1da177e4 | 3011 | |
c8dc9c65 | 3012 | if (mddev->queue) |
5026d7a9 PA |
3013 | blk_queue_max_write_same_sectors(mddev->queue, 0); |
3014 | ||
dafb20fa | 3015 | rdev_for_each(rdev, mddev) { |
1ed7242e JB |
3016 | if (!mddev->gendisk) |
3017 | continue; | |
709ae487 N |
3018 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
3019 | rdev->data_offset << 9); | |
2ff8cc2c SL |
3020 | if (blk_queue_discard(bdev_get_queue(rdev->bdev))) |
3021 | discard_supported = true; | |
1da177e4 | 3022 | } |
191ea9b2 | 3023 | |
709ae487 N |
3024 | mddev->degraded = 0; |
3025 | for (i=0; i < conf->raid_disks; i++) | |
3026 | if (conf->mirrors[i].rdev == NULL || | |
3027 | !test_bit(In_sync, &conf->mirrors[i].rdev->flags) || | |
3028 | test_bit(Faulty, &conf->mirrors[i].rdev->flags)) | |
3029 | mddev->degraded++; | |
3030 | ||
3031 | if (conf->raid_disks - mddev->degraded == 1) | |
3032 | mddev->recovery_cp = MaxSector; | |
3033 | ||
8c6ac868 | 3034 | if (mddev->recovery_cp != MaxSector) |
1d41c216 N |
3035 | pr_info("md/raid1:%s: not clean -- starting background reconstruction\n", |
3036 | mdname(mddev)); | |
3037 | pr_info("md/raid1:%s: active with %d out of %d mirrors\n", | |
f72ffdd6 | 3038 | mdname(mddev), mddev->raid_disks - mddev->degraded, |
1da177e4 | 3039 | mddev->raid_disks); |
709ae487 | 3040 | |
1da177e4 LT |
3041 | /* |
3042 | * Ok, everything is just fine now | |
3043 | */ | |
709ae487 N |
3044 | mddev->thread = conf->thread; |
3045 | conf->thread = NULL; | |
3046 | mddev->private = conf; | |
46533ff7 | 3047 | set_bit(MD_FAILFAST_SUPPORTED, &mddev->flags); |
709ae487 | 3048 | |
1f403624 | 3049 | md_set_array_sectors(mddev, raid1_size(mddev, 0, 0)); |
1da177e4 | 3050 | |
1ed7242e | 3051 | if (mddev->queue) { |
2ff8cc2c SL |
3052 | if (discard_supported) |
3053 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, | |
3054 | mddev->queue); | |
3055 | else | |
3056 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, | |
3057 | mddev->queue); | |
1ed7242e | 3058 | } |
5220ea1e | 3059 | |
3060 | ret = md_integrity_register(mddev); | |
5aa61f42 N |
3061 | if (ret) { |
3062 | md_unregister_thread(&mddev->thread); | |
afa0f557 | 3063 | raid1_free(mddev, conf); |
5aa61f42 | 3064 | } |
5220ea1e | 3065 | return ret; |
1da177e4 LT |
3066 | } |
3067 | ||
afa0f557 | 3068 | static void raid1_free(struct mddev *mddev, void *priv) |
1da177e4 | 3069 | { |
afa0f557 | 3070 | struct r1conf *conf = priv; |
409c57f3 | 3071 | |
644df1a8 | 3072 | mempool_destroy(conf->r1bio_pool); |
990a8baf | 3073 | kfree(conf->mirrors); |
0fea7ed8 | 3074 | safe_put_page(conf->tmppage); |
990a8baf | 3075 | kfree(conf->poolinfo); |
1da177e4 | 3076 | kfree(conf); |
1da177e4 LT |
3077 | } |
3078 | ||
fd01b88c | 3079 | static int raid1_resize(struct mddev *mddev, sector_t sectors) |
1da177e4 LT |
3080 | { |
3081 | /* no resync is happening, and there is enough space | |
3082 | * on all devices, so we can resize. | |
3083 | * We need to make sure resync covers any new space. | |
3084 | * If the array is shrinking we should possibly wait until | |
3085 | * any io in the removed space completes, but it hardly seems | |
3086 | * worth it. | |
3087 | */ | |
a4a6125a N |
3088 | sector_t newsize = raid1_size(mddev, sectors, 0); |
3089 | if (mddev->external_size && | |
3090 | mddev->array_sectors > newsize) | |
b522adcd | 3091 | return -EINVAL; |
a4a6125a N |
3092 | if (mddev->bitmap) { |
3093 | int ret = bitmap_resize(mddev->bitmap, newsize, 0, 0); | |
3094 | if (ret) | |
3095 | return ret; | |
3096 | } | |
3097 | md_set_array_sectors(mddev, newsize); | |
f233ea5c | 3098 | set_capacity(mddev->gendisk, mddev->array_sectors); |
449aad3e | 3099 | revalidate_disk(mddev->gendisk); |
b522adcd | 3100 | if (sectors > mddev->dev_sectors && |
b098636c | 3101 | mddev->recovery_cp > mddev->dev_sectors) { |
58c0fed4 | 3102 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
3103 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
3104 | } | |
b522adcd | 3105 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 3106 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
3107 | return 0; |
3108 | } | |
3109 | ||
fd01b88c | 3110 | static int raid1_reshape(struct mddev *mddev) |
1da177e4 LT |
3111 | { |
3112 | /* We need to: | |
3113 | * 1/ resize the r1bio_pool | |
3114 | * 2/ resize conf->mirrors | |
3115 | * | |
3116 | * We allocate a new r1bio_pool if we can. | |
3117 | * Then raise a device barrier and wait until all IO stops. | |
3118 | * Then resize conf->mirrors and swap in the new r1bio pool. | |
6ea9c07c N |
3119 | * |
3120 | * At the same time, we "pack" the devices so that all the missing | |
3121 | * devices have the higher raid_disk numbers. | |
1da177e4 LT |
3122 | */ |
3123 | mempool_t *newpool, *oldpool; | |
3124 | struct pool_info *newpoolinfo; | |
0eaf822c | 3125 | struct raid1_info *newmirrors; |
e8096360 | 3126 | struct r1conf *conf = mddev->private; |
63c70c4f | 3127 | int cnt, raid_disks; |
c04be0aa | 3128 | unsigned long flags; |
b5470dc5 | 3129 | int d, d2, err; |
1da177e4 | 3130 | |
63c70c4f | 3131 | /* Cannot change chunk_size, layout, or level */ |
664e7c41 | 3132 | if (mddev->chunk_sectors != mddev->new_chunk_sectors || |
63c70c4f N |
3133 | mddev->layout != mddev->new_layout || |
3134 | mddev->level != mddev->new_level) { | |
664e7c41 | 3135 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
63c70c4f N |
3136 | mddev->new_layout = mddev->layout; |
3137 | mddev->new_level = mddev->level; | |
3138 | return -EINVAL; | |
3139 | } | |
3140 | ||
28c1b9fd GR |
3141 | if (!mddev_is_clustered(mddev)) { |
3142 | err = md_allow_write(mddev); | |
3143 | if (err) | |
3144 | return err; | |
3145 | } | |
2a2275d6 | 3146 | |
63c70c4f N |
3147 | raid_disks = mddev->raid_disks + mddev->delta_disks; |
3148 | ||
6ea9c07c N |
3149 | if (raid_disks < conf->raid_disks) { |
3150 | cnt=0; | |
3151 | for (d= 0; d < conf->raid_disks; d++) | |
3152 | if (conf->mirrors[d].rdev) | |
3153 | cnt++; | |
3154 | if (cnt > raid_disks) | |
1da177e4 | 3155 | return -EBUSY; |
6ea9c07c | 3156 | } |
1da177e4 LT |
3157 | |
3158 | newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL); | |
3159 | if (!newpoolinfo) | |
3160 | return -ENOMEM; | |
3161 | newpoolinfo->mddev = mddev; | |
8f19ccb2 | 3162 | newpoolinfo->raid_disks = raid_disks * 2; |
1da177e4 LT |
3163 | |
3164 | newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, | |
3165 | r1bio_pool_free, newpoolinfo); | |
3166 | if (!newpool) { | |
3167 | kfree(newpoolinfo); | |
3168 | return -ENOMEM; | |
3169 | } | |
0eaf822c | 3170 | newmirrors = kzalloc(sizeof(struct raid1_info) * raid_disks * 2, |
8f19ccb2 | 3171 | GFP_KERNEL); |
1da177e4 LT |
3172 | if (!newmirrors) { |
3173 | kfree(newpoolinfo); | |
3174 | mempool_destroy(newpool); | |
3175 | return -ENOMEM; | |
3176 | } | |
1da177e4 | 3177 | |
e2d59925 | 3178 | freeze_array(conf, 0); |
1da177e4 LT |
3179 | |
3180 | /* ok, everything is stopped */ | |
3181 | oldpool = conf->r1bio_pool; | |
3182 | conf->r1bio_pool = newpool; | |
6ea9c07c | 3183 | |
a88aa786 | 3184 | for (d = d2 = 0; d < conf->raid_disks; d++) { |
3cb03002 | 3185 | struct md_rdev *rdev = conf->mirrors[d].rdev; |
a88aa786 | 3186 | if (rdev && rdev->raid_disk != d2) { |
36fad858 | 3187 | sysfs_unlink_rdev(mddev, rdev); |
a88aa786 | 3188 | rdev->raid_disk = d2; |
36fad858 NK |
3189 | sysfs_unlink_rdev(mddev, rdev); |
3190 | if (sysfs_link_rdev(mddev, rdev)) | |
1d41c216 N |
3191 | pr_warn("md/raid1:%s: cannot register rd%d\n", |
3192 | mdname(mddev), rdev->raid_disk); | |
6ea9c07c | 3193 | } |
a88aa786 N |
3194 | if (rdev) |
3195 | newmirrors[d2++].rdev = rdev; | |
3196 | } | |
1da177e4 LT |
3197 | kfree(conf->mirrors); |
3198 | conf->mirrors = newmirrors; | |
3199 | kfree(conf->poolinfo); | |
3200 | conf->poolinfo = newpoolinfo; | |
3201 | ||
c04be0aa | 3202 | spin_lock_irqsave(&conf->device_lock, flags); |
1da177e4 | 3203 | mddev->degraded += (raid_disks - conf->raid_disks); |
c04be0aa | 3204 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 3205 | conf->raid_disks = mddev->raid_disks = raid_disks; |
63c70c4f | 3206 | mddev->delta_disks = 0; |
1da177e4 | 3207 | |
e2d59925 | 3208 | unfreeze_array(conf); |
1da177e4 | 3209 | |
985ca973 | 3210 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
1da177e4 LT |
3211 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
3212 | md_wakeup_thread(mddev->thread); | |
3213 | ||
3214 | mempool_destroy(oldpool); | |
3215 | return 0; | |
3216 | } | |
3217 | ||
fd01b88c | 3218 | static void raid1_quiesce(struct mddev *mddev, int state) |
36fa3063 | 3219 | { |
e8096360 | 3220 | struct r1conf *conf = mddev->private; |
36fa3063 N |
3221 | |
3222 | switch(state) { | |
6eef4b21 N |
3223 | case 2: /* wake for suspend */ |
3224 | wake_up(&conf->wait_barrier); | |
3225 | break; | |
9e6603da | 3226 | case 1: |
07169fd4 | 3227 | freeze_array(conf, 0); |
36fa3063 | 3228 | break; |
9e6603da | 3229 | case 0: |
07169fd4 | 3230 | unfreeze_array(conf); |
36fa3063 N |
3231 | break; |
3232 | } | |
36fa3063 N |
3233 | } |
3234 | ||
fd01b88c | 3235 | static void *raid1_takeover(struct mddev *mddev) |
709ae487 N |
3236 | { |
3237 | /* raid1 can take over: | |
3238 | * raid5 with 2 devices, any layout or chunk size | |
3239 | */ | |
3240 | if (mddev->level == 5 && mddev->raid_disks == 2) { | |
e8096360 | 3241 | struct r1conf *conf; |
709ae487 N |
3242 | mddev->new_level = 1; |
3243 | mddev->new_layout = 0; | |
3244 | mddev->new_chunk_sectors = 0; | |
3245 | conf = setup_conf(mddev); | |
3246 | if (!IS_ERR(conf)) | |
07169fd4 | 3247 | /* Array must appear to be quiesced */ |
3248 | conf->array_frozen = 1; | |
709ae487 N |
3249 | return conf; |
3250 | } | |
3251 | return ERR_PTR(-EINVAL); | |
3252 | } | |
1da177e4 | 3253 | |
84fc4b56 | 3254 | static struct md_personality raid1_personality = |
1da177e4 LT |
3255 | { |
3256 | .name = "raid1", | |
2604b703 | 3257 | .level = 1, |
1da177e4 | 3258 | .owner = THIS_MODULE, |
849674e4 SL |
3259 | .make_request = raid1_make_request, |
3260 | .run = raid1_run, | |
afa0f557 | 3261 | .free = raid1_free, |
849674e4 SL |
3262 | .status = raid1_status, |
3263 | .error_handler = raid1_error, | |
1da177e4 LT |
3264 | .hot_add_disk = raid1_add_disk, |
3265 | .hot_remove_disk= raid1_remove_disk, | |
3266 | .spare_active = raid1_spare_active, | |
849674e4 | 3267 | .sync_request = raid1_sync_request, |
1da177e4 | 3268 | .resize = raid1_resize, |
80c3a6ce | 3269 | .size = raid1_size, |
63c70c4f | 3270 | .check_reshape = raid1_reshape, |
36fa3063 | 3271 | .quiesce = raid1_quiesce, |
709ae487 | 3272 | .takeover = raid1_takeover, |
5c675f83 | 3273 | .congested = raid1_congested, |
1da177e4 LT |
3274 | }; |
3275 | ||
3276 | static int __init raid_init(void) | |
3277 | { | |
2604b703 | 3278 | return register_md_personality(&raid1_personality); |
1da177e4 LT |
3279 | } |
3280 | ||
3281 | static void raid_exit(void) | |
3282 | { | |
2604b703 | 3283 | unregister_md_personality(&raid1_personality); |
1da177e4 LT |
3284 | } |
3285 | ||
3286 | module_init(raid_init); | |
3287 | module_exit(raid_exit); | |
3288 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 3289 | MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD"); |
1da177e4 | 3290 | MODULE_ALIAS("md-personality-3"); /* RAID1 */ |
d9d166c2 | 3291 | MODULE_ALIAS("md-raid1"); |
2604b703 | 3292 | MODULE_ALIAS("md-level-1"); |
34db0cd6 N |
3293 | |
3294 | module_param(max_queued_requests, int, S_IRUGO|S_IWUSR); |