]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * raid1.c : Multiple Devices driver for Linux | |
3 | * | |
4 | * Copyright (C) 1999, 2000, 2001 Ingo Molnar, Red Hat | |
5 | * | |
6 | * Copyright (C) 1996, 1997, 1998 Ingo Molnar, Miguel de Icaza, Gadi Oxman | |
7 | * | |
8 | * RAID-1 management functions. | |
9 | * | |
10 | * Better read-balancing code written by Mika Kuoppala <miku@iki.fi>, 2000 | |
11 | * | |
96de0e25 | 12 | * Fixes to reconstruction by Jakob Østergaard" <jakob@ostenfeld.dk> |
1da177e4 LT |
13 | * Various fixes by Neil Brown <neilb@cse.unsw.edu.au> |
14 | * | |
191ea9b2 N |
15 | * Changes by Peter T. Breuer <ptb@it.uc3m.es> 31/1/2003 to support |
16 | * bitmapped intelligence in resync: | |
17 | * | |
18 | * - bitmap marked during normal i/o | |
19 | * - bitmap used to skip nondirty blocks during sync | |
20 | * | |
21 | * Additions to bitmap code, (C) 2003-2004 Paul Clements, SteelEye Technology: | |
22 | * - persistent bitmap code | |
23 | * | |
1da177e4 LT |
24 | * This program is free software; you can redistribute it and/or modify |
25 | * it under the terms of the GNU General Public License as published by | |
26 | * the Free Software Foundation; either version 2, or (at your option) | |
27 | * any later version. | |
28 | * | |
29 | * You should have received a copy of the GNU General Public License | |
30 | * (for example /usr/src/linux/COPYING); if not, write to the Free | |
31 | * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
32 | */ | |
33 | ||
5a0e3ad6 | 34 | #include <linux/slab.h> |
25570727 | 35 | #include <linux/delay.h> |
bff61975 | 36 | #include <linux/blkdev.h> |
056075c7 | 37 | #include <linux/module.h> |
bff61975 | 38 | #include <linux/seq_file.h> |
8bda470e | 39 | #include <linux/ratelimit.h> |
3f07c014 IM |
40 | #include <linux/sched/signal.h> |
41 | ||
109e3765 | 42 | #include <trace/events/block.h> |
3f07c014 | 43 | |
43b2e5d8 | 44 | #include "md.h" |
ef740c37 CH |
45 | #include "raid1.h" |
46 | #include "bitmap.h" | |
191ea9b2 | 47 | |
394ed8e4 SL |
48 | #define UNSUPPORTED_MDDEV_FLAGS \ |
49 | ((1L << MD_HAS_JOURNAL) | \ | |
50 | (1L << MD_JOURNAL_CLEAN)) | |
51 | ||
1da177e4 LT |
52 | /* |
53 | * Number of guaranteed r1bios in case of extreme VM load: | |
54 | */ | |
55 | #define NR_RAID1_BIOS 256 | |
56 | ||
473e87ce JB |
57 | /* when we get a read error on a read-only array, we redirect to another |
58 | * device without failing the first device, or trying to over-write to | |
59 | * correct the read error. To keep track of bad blocks on a per-bio | |
60 | * level, we store IO_BLOCKED in the appropriate 'bios' pointer | |
61 | */ | |
62 | #define IO_BLOCKED ((struct bio *)1) | |
63 | /* When we successfully write to a known bad-block, we need to remove the | |
64 | * bad-block marking which must be done from process context. So we record | |
65 | * the success by setting devs[n].bio to IO_MADE_GOOD | |
66 | */ | |
67 | #define IO_MADE_GOOD ((struct bio *)2) | |
68 | ||
69 | #define BIO_SPECIAL(bio) ((unsigned long)bio <= 2) | |
70 | ||
34db0cd6 N |
71 | /* When there are this many requests queue to be written by |
72 | * the raid1 thread, we become 'congested' to provide back-pressure | |
73 | * for writeback. | |
74 | */ | |
75 | static int max_queued_requests = 1024; | |
1da177e4 | 76 | |
fd76863e | 77 | static void allow_barrier(struct r1conf *conf, sector_t sector_nr); |
78 | static void lower_barrier(struct r1conf *conf, sector_t sector_nr); | |
1da177e4 | 79 | |
578b54ad N |
80 | #define raid1_log(md, fmt, args...) \ |
81 | do { if ((md)->queue) blk_add_trace_msg((md)->queue, "raid1 " fmt, ##args); } while (0) | |
82 | ||
dd0fc66f | 83 | static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 LT |
84 | { |
85 | struct pool_info *pi = data; | |
9f2c9d12 | 86 | int size = offsetof(struct r1bio, bios[pi->raid_disks]); |
1da177e4 LT |
87 | |
88 | /* allocate a r1bio with room for raid_disks entries in the bios array */ | |
7eaceacc | 89 | return kzalloc(size, gfp_flags); |
1da177e4 LT |
90 | } |
91 | ||
92 | static void r1bio_pool_free(void *r1_bio, void *data) | |
93 | { | |
94 | kfree(r1_bio); | |
95 | } | |
96 | ||
97 | #define RESYNC_BLOCK_SIZE (64*1024) | |
8e005f7c | 98 | #define RESYNC_DEPTH 32 |
1da177e4 LT |
99 | #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9) |
100 | #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE) | |
8e005f7c | 101 | #define RESYNC_WINDOW (RESYNC_BLOCK_SIZE * RESYNC_DEPTH) |
102 | #define RESYNC_WINDOW_SECTORS (RESYNC_WINDOW >> 9) | |
c40f341f GR |
103 | #define CLUSTER_RESYNC_WINDOW (16 * RESYNC_WINDOW) |
104 | #define CLUSTER_RESYNC_WINDOW_SECTORS (CLUSTER_RESYNC_WINDOW >> 9) | |
1da177e4 | 105 | |
dd0fc66f | 106 | static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 LT |
107 | { |
108 | struct pool_info *pi = data; | |
9f2c9d12 | 109 | struct r1bio *r1_bio; |
1da177e4 | 110 | struct bio *bio; |
da1aab3d | 111 | int need_pages; |
1da177e4 LT |
112 | int i, j; |
113 | ||
114 | r1_bio = r1bio_pool_alloc(gfp_flags, pi); | |
7eaceacc | 115 | if (!r1_bio) |
1da177e4 | 116 | return NULL; |
1da177e4 LT |
117 | |
118 | /* | |
119 | * Allocate bios : 1 for reading, n-1 for writing | |
120 | */ | |
121 | for (j = pi->raid_disks ; j-- ; ) { | |
6746557f | 122 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); |
1da177e4 LT |
123 | if (!bio) |
124 | goto out_free_bio; | |
125 | r1_bio->bios[j] = bio; | |
126 | } | |
127 | /* | |
128 | * Allocate RESYNC_PAGES data pages and attach them to | |
d11c171e N |
129 | * the first bio. |
130 | * If this is a user-requested check/repair, allocate | |
131 | * RESYNC_PAGES for each bio. | |
1da177e4 | 132 | */ |
d11c171e | 133 | if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) |
da1aab3d | 134 | need_pages = pi->raid_disks; |
d11c171e | 135 | else |
da1aab3d N |
136 | need_pages = 1; |
137 | for (j = 0; j < need_pages; j++) { | |
d11c171e | 138 | bio = r1_bio->bios[j]; |
a0787606 | 139 | bio->bi_vcnt = RESYNC_PAGES; |
d11c171e | 140 | |
a0787606 | 141 | if (bio_alloc_pages(bio, gfp_flags)) |
da1aab3d | 142 | goto out_free_pages; |
d11c171e N |
143 | } |
144 | /* If not user-requests, copy the page pointers to all bios */ | |
145 | if (!test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) { | |
146 | for (i=0; i<RESYNC_PAGES ; i++) | |
147 | for (j=1; j<pi->raid_disks; j++) | |
148 | r1_bio->bios[j]->bi_io_vec[i].bv_page = | |
149 | r1_bio->bios[0]->bi_io_vec[i].bv_page; | |
1da177e4 LT |
150 | } |
151 | ||
152 | r1_bio->master_bio = NULL; | |
153 | ||
154 | return r1_bio; | |
155 | ||
da1aab3d | 156 | out_free_pages: |
491221f8 GJ |
157 | while (--j >= 0) |
158 | bio_free_pages(r1_bio->bios[j]); | |
da1aab3d | 159 | |
1da177e4 | 160 | out_free_bio: |
8f19ccb2 | 161 | while (++j < pi->raid_disks) |
1da177e4 LT |
162 | bio_put(r1_bio->bios[j]); |
163 | r1bio_pool_free(r1_bio, data); | |
164 | return NULL; | |
165 | } | |
166 | ||
167 | static void r1buf_pool_free(void *__r1_bio, void *data) | |
168 | { | |
169 | struct pool_info *pi = data; | |
d11c171e | 170 | int i,j; |
9f2c9d12 | 171 | struct r1bio *r1bio = __r1_bio; |
1da177e4 | 172 | |
d11c171e N |
173 | for (i = 0; i < RESYNC_PAGES; i++) |
174 | for (j = pi->raid_disks; j-- ;) { | |
175 | if (j == 0 || | |
176 | r1bio->bios[j]->bi_io_vec[i].bv_page != | |
177 | r1bio->bios[0]->bi_io_vec[i].bv_page) | |
1345b1d8 | 178 | safe_put_page(r1bio->bios[j]->bi_io_vec[i].bv_page); |
d11c171e | 179 | } |
1da177e4 LT |
180 | for (i=0 ; i < pi->raid_disks; i++) |
181 | bio_put(r1bio->bios[i]); | |
182 | ||
183 | r1bio_pool_free(r1bio, data); | |
184 | } | |
185 | ||
e8096360 | 186 | static void put_all_bios(struct r1conf *conf, struct r1bio *r1_bio) |
1da177e4 LT |
187 | { |
188 | int i; | |
189 | ||
8f19ccb2 | 190 | for (i = 0; i < conf->raid_disks * 2; i++) { |
1da177e4 | 191 | struct bio **bio = r1_bio->bios + i; |
4367af55 | 192 | if (!BIO_SPECIAL(*bio)) |
1da177e4 LT |
193 | bio_put(*bio); |
194 | *bio = NULL; | |
195 | } | |
196 | } | |
197 | ||
9f2c9d12 | 198 | static void free_r1bio(struct r1bio *r1_bio) |
1da177e4 | 199 | { |
e8096360 | 200 | struct r1conf *conf = r1_bio->mddev->private; |
1da177e4 | 201 | |
1da177e4 LT |
202 | put_all_bios(conf, r1_bio); |
203 | mempool_free(r1_bio, conf->r1bio_pool); | |
204 | } | |
205 | ||
9f2c9d12 | 206 | static void put_buf(struct r1bio *r1_bio) |
1da177e4 | 207 | { |
e8096360 | 208 | struct r1conf *conf = r1_bio->mddev->private; |
af5f42a7 | 209 | sector_t sect = r1_bio->sector; |
3e198f78 N |
210 | int i; |
211 | ||
8f19ccb2 | 212 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3e198f78 N |
213 | struct bio *bio = r1_bio->bios[i]; |
214 | if (bio->bi_end_io) | |
215 | rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev); | |
216 | } | |
1da177e4 LT |
217 | |
218 | mempool_free(r1_bio, conf->r1buf_pool); | |
219 | ||
af5f42a7 | 220 | lower_barrier(conf, sect); |
1da177e4 LT |
221 | } |
222 | ||
9f2c9d12 | 223 | static void reschedule_retry(struct r1bio *r1_bio) |
1da177e4 LT |
224 | { |
225 | unsigned long flags; | |
fd01b88c | 226 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 227 | struct r1conf *conf = mddev->private; |
fd76863e | 228 | int idx; |
1da177e4 | 229 | |
fd76863e | 230 | idx = sector_to_idx(r1_bio->sector); |
1da177e4 LT |
231 | spin_lock_irqsave(&conf->device_lock, flags); |
232 | list_add(&r1_bio->retry_list, &conf->retry_list); | |
824e47da | 233 | atomic_inc(&conf->nr_queued[idx]); |
1da177e4 LT |
234 | spin_unlock_irqrestore(&conf->device_lock, flags); |
235 | ||
17999be4 | 236 | wake_up(&conf->wait_barrier); |
1da177e4 LT |
237 | md_wakeup_thread(mddev->thread); |
238 | } | |
239 | ||
240 | /* | |
241 | * raid_end_bio_io() is called when we have finished servicing a mirrored | |
242 | * operation and are ready to return a success/failure code to the buffer | |
243 | * cache layer. | |
244 | */ | |
9f2c9d12 | 245 | static void call_bio_endio(struct r1bio *r1_bio) |
d2eb35ac N |
246 | { |
247 | struct bio *bio = r1_bio->master_bio; | |
248 | int done; | |
e8096360 | 249 | struct r1conf *conf = r1_bio->mddev->private; |
4f024f37 | 250 | sector_t bi_sector = bio->bi_iter.bi_sector; |
d2eb35ac N |
251 | |
252 | if (bio->bi_phys_segments) { | |
253 | unsigned long flags; | |
254 | spin_lock_irqsave(&conf->device_lock, flags); | |
255 | bio->bi_phys_segments--; | |
256 | done = (bio->bi_phys_segments == 0); | |
257 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
79ef3a8a | 258 | /* |
259 | * make_request() might be waiting for | |
260 | * bi_phys_segments to decrease | |
261 | */ | |
262 | wake_up(&conf->wait_barrier); | |
d2eb35ac N |
263 | } else |
264 | done = 1; | |
265 | ||
266 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) | |
4246a0b6 CH |
267 | bio->bi_error = -EIO; |
268 | ||
d2eb35ac | 269 | if (done) { |
4246a0b6 | 270 | bio_endio(bio); |
d2eb35ac N |
271 | /* |
272 | * Wake up any possible resync thread that waits for the device | |
273 | * to go idle. | |
274 | */ | |
fd76863e | 275 | allow_barrier(conf, bi_sector); |
d2eb35ac N |
276 | } |
277 | } | |
278 | ||
9f2c9d12 | 279 | static void raid_end_bio_io(struct r1bio *r1_bio) |
1da177e4 LT |
280 | { |
281 | struct bio *bio = r1_bio->master_bio; | |
282 | ||
4b6d287f N |
283 | /* if nobody has done the final endio yet, do it now */ |
284 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { | |
36a4e1fe N |
285 | pr_debug("raid1: sync end %s on sectors %llu-%llu\n", |
286 | (bio_data_dir(bio) == WRITE) ? "write" : "read", | |
4f024f37 KO |
287 | (unsigned long long) bio->bi_iter.bi_sector, |
288 | (unsigned long long) bio_end_sector(bio) - 1); | |
4b6d287f | 289 | |
d2eb35ac | 290 | call_bio_endio(r1_bio); |
4b6d287f | 291 | } |
1da177e4 LT |
292 | free_r1bio(r1_bio); |
293 | } | |
294 | ||
295 | /* | |
296 | * Update disk head position estimator based on IRQ completion info. | |
297 | */ | |
9f2c9d12 | 298 | static inline void update_head_pos(int disk, struct r1bio *r1_bio) |
1da177e4 | 299 | { |
e8096360 | 300 | struct r1conf *conf = r1_bio->mddev->private; |
1da177e4 LT |
301 | |
302 | conf->mirrors[disk].head_position = | |
303 | r1_bio->sector + (r1_bio->sectors); | |
304 | } | |
305 | ||
ba3ae3be NK |
306 | /* |
307 | * Find the disk number which triggered given bio | |
308 | */ | |
9f2c9d12 | 309 | static int find_bio_disk(struct r1bio *r1_bio, struct bio *bio) |
ba3ae3be NK |
310 | { |
311 | int mirror; | |
30194636 N |
312 | struct r1conf *conf = r1_bio->mddev->private; |
313 | int raid_disks = conf->raid_disks; | |
ba3ae3be | 314 | |
8f19ccb2 | 315 | for (mirror = 0; mirror < raid_disks * 2; mirror++) |
ba3ae3be NK |
316 | if (r1_bio->bios[mirror] == bio) |
317 | break; | |
318 | ||
8f19ccb2 | 319 | BUG_ON(mirror == raid_disks * 2); |
ba3ae3be NK |
320 | update_head_pos(mirror, r1_bio); |
321 | ||
322 | return mirror; | |
323 | } | |
324 | ||
4246a0b6 | 325 | static void raid1_end_read_request(struct bio *bio) |
1da177e4 | 326 | { |
4246a0b6 | 327 | int uptodate = !bio->bi_error; |
9f2c9d12 | 328 | struct r1bio *r1_bio = bio->bi_private; |
e8096360 | 329 | struct r1conf *conf = r1_bio->mddev->private; |
e5872d58 | 330 | struct md_rdev *rdev = conf->mirrors[r1_bio->read_disk].rdev; |
1da177e4 | 331 | |
1da177e4 LT |
332 | /* |
333 | * this branch is our 'one mirror IO has finished' event handler: | |
334 | */ | |
e5872d58 | 335 | update_head_pos(r1_bio->read_disk, r1_bio); |
ddaf22ab | 336 | |
dd00a99e N |
337 | if (uptodate) |
338 | set_bit(R1BIO_Uptodate, &r1_bio->state); | |
2e52d449 N |
339 | else if (test_bit(FailFast, &rdev->flags) && |
340 | test_bit(R1BIO_FailFast, &r1_bio->state)) | |
341 | /* This was a fail-fast read so we definitely | |
342 | * want to retry */ | |
343 | ; | |
dd00a99e N |
344 | else { |
345 | /* If all other devices have failed, we want to return | |
346 | * the error upwards rather than fail the last device. | |
347 | * Here we redefine "uptodate" to mean "Don't want to retry" | |
1da177e4 | 348 | */ |
dd00a99e N |
349 | unsigned long flags; |
350 | spin_lock_irqsave(&conf->device_lock, flags); | |
351 | if (r1_bio->mddev->degraded == conf->raid_disks || | |
352 | (r1_bio->mddev->degraded == conf->raid_disks-1 && | |
e5872d58 | 353 | test_bit(In_sync, &rdev->flags))) |
dd00a99e N |
354 | uptodate = 1; |
355 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
356 | } | |
1da177e4 | 357 | |
7ad4d4a6 | 358 | if (uptodate) { |
1da177e4 | 359 | raid_end_bio_io(r1_bio); |
e5872d58 | 360 | rdev_dec_pending(rdev, conf->mddev); |
7ad4d4a6 | 361 | } else { |
1da177e4 LT |
362 | /* |
363 | * oops, read error: | |
364 | */ | |
365 | char b[BDEVNAME_SIZE]; | |
1d41c216 N |
366 | pr_err_ratelimited("md/raid1:%s: %s: rescheduling sector %llu\n", |
367 | mdname(conf->mddev), | |
368 | bdevname(rdev->bdev, b), | |
369 | (unsigned long long)r1_bio->sector); | |
d2eb35ac | 370 | set_bit(R1BIO_ReadError, &r1_bio->state); |
1da177e4 | 371 | reschedule_retry(r1_bio); |
7ad4d4a6 | 372 | /* don't drop the reference on read_disk yet */ |
1da177e4 | 373 | } |
1da177e4 LT |
374 | } |
375 | ||
9f2c9d12 | 376 | static void close_write(struct r1bio *r1_bio) |
cd5ff9a1 N |
377 | { |
378 | /* it really is the end of this request */ | |
379 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { | |
380 | /* free extra copy of the data pages */ | |
381 | int i = r1_bio->behind_page_count; | |
382 | while (i--) | |
383 | safe_put_page(r1_bio->behind_bvecs[i].bv_page); | |
384 | kfree(r1_bio->behind_bvecs); | |
385 | r1_bio->behind_bvecs = NULL; | |
386 | } | |
387 | /* clear the bitmap if all writes complete successfully */ | |
388 | bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector, | |
389 | r1_bio->sectors, | |
390 | !test_bit(R1BIO_Degraded, &r1_bio->state), | |
391 | test_bit(R1BIO_BehindIO, &r1_bio->state)); | |
392 | md_write_end(r1_bio->mddev); | |
393 | } | |
394 | ||
9f2c9d12 | 395 | static void r1_bio_write_done(struct r1bio *r1_bio) |
4e78064f | 396 | { |
cd5ff9a1 N |
397 | if (!atomic_dec_and_test(&r1_bio->remaining)) |
398 | return; | |
399 | ||
400 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
401 | reschedule_retry(r1_bio); | |
402 | else { | |
403 | close_write(r1_bio); | |
4367af55 N |
404 | if (test_bit(R1BIO_MadeGood, &r1_bio->state)) |
405 | reschedule_retry(r1_bio); | |
406 | else | |
407 | raid_end_bio_io(r1_bio); | |
4e78064f N |
408 | } |
409 | } | |
410 | ||
4246a0b6 | 411 | static void raid1_end_write_request(struct bio *bio) |
1da177e4 | 412 | { |
9f2c9d12 | 413 | struct r1bio *r1_bio = bio->bi_private; |
e5872d58 | 414 | int behind = test_bit(R1BIO_BehindIO, &r1_bio->state); |
e8096360 | 415 | struct r1conf *conf = r1_bio->mddev->private; |
04b857f7 | 416 | struct bio *to_put = NULL; |
e5872d58 N |
417 | int mirror = find_bio_disk(r1_bio, bio); |
418 | struct md_rdev *rdev = conf->mirrors[mirror].rdev; | |
e3f948cd SL |
419 | bool discard_error; |
420 | ||
421 | discard_error = bio->bi_error && bio_op(bio) == REQ_OP_DISCARD; | |
1da177e4 | 422 | |
e9c7469b TH |
423 | /* |
424 | * 'one mirror IO has finished' event handler: | |
425 | */ | |
e3f948cd | 426 | if (bio->bi_error && !discard_error) { |
e5872d58 N |
427 | set_bit(WriteErrorSeen, &rdev->flags); |
428 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) | |
19d67169 N |
429 | set_bit(MD_RECOVERY_NEEDED, & |
430 | conf->mddev->recovery); | |
431 | ||
212e7eb7 N |
432 | if (test_bit(FailFast, &rdev->flags) && |
433 | (bio->bi_opf & MD_FAILFAST) && | |
434 | /* We never try FailFast to WriteMostly devices */ | |
435 | !test_bit(WriteMostly, &rdev->flags)) { | |
436 | md_error(r1_bio->mddev, rdev); | |
437 | if (!test_bit(Faulty, &rdev->flags)) | |
438 | /* This is the only remaining device, | |
439 | * We need to retry the write without | |
440 | * FailFast | |
441 | */ | |
442 | set_bit(R1BIO_WriteError, &r1_bio->state); | |
443 | else { | |
444 | /* Finished with this branch */ | |
445 | r1_bio->bios[mirror] = NULL; | |
446 | to_put = bio; | |
447 | } | |
448 | } else | |
449 | set_bit(R1BIO_WriteError, &r1_bio->state); | |
4367af55 | 450 | } else { |
1da177e4 | 451 | /* |
e9c7469b TH |
452 | * Set R1BIO_Uptodate in our master bio, so that we |
453 | * will return a good error code for to the higher | |
454 | * levels even if IO on some other mirrored buffer | |
455 | * fails. | |
456 | * | |
457 | * The 'master' represents the composite IO operation | |
458 | * to user-side. So if something waits for IO, then it | |
459 | * will wait for the 'master' bio. | |
1da177e4 | 460 | */ |
4367af55 N |
461 | sector_t first_bad; |
462 | int bad_sectors; | |
463 | ||
cd5ff9a1 N |
464 | r1_bio->bios[mirror] = NULL; |
465 | to_put = bio; | |
3056e3ae AL |
466 | /* |
467 | * Do not set R1BIO_Uptodate if the current device is | |
468 | * rebuilding or Faulty. This is because we cannot use | |
469 | * such device for properly reading the data back (we could | |
470 | * potentially use it, if the current write would have felt | |
471 | * before rdev->recovery_offset, but for simplicity we don't | |
472 | * check this here. | |
473 | */ | |
e5872d58 N |
474 | if (test_bit(In_sync, &rdev->flags) && |
475 | !test_bit(Faulty, &rdev->flags)) | |
3056e3ae | 476 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
e9c7469b | 477 | |
4367af55 | 478 | /* Maybe we can clear some bad blocks. */ |
e5872d58 | 479 | if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors, |
e3f948cd | 480 | &first_bad, &bad_sectors) && !discard_error) { |
4367af55 N |
481 | r1_bio->bios[mirror] = IO_MADE_GOOD; |
482 | set_bit(R1BIO_MadeGood, &r1_bio->state); | |
483 | } | |
484 | } | |
485 | ||
e9c7469b | 486 | if (behind) { |
e5872d58 | 487 | if (test_bit(WriteMostly, &rdev->flags)) |
e9c7469b TH |
488 | atomic_dec(&r1_bio->behind_remaining); |
489 | ||
490 | /* | |
491 | * In behind mode, we ACK the master bio once the I/O | |
492 | * has safely reached all non-writemostly | |
493 | * disks. Setting the Returned bit ensures that this | |
494 | * gets done only once -- we don't ever want to return | |
495 | * -EIO here, instead we'll wait | |
496 | */ | |
497 | if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) && | |
498 | test_bit(R1BIO_Uptodate, &r1_bio->state)) { | |
499 | /* Maybe we can return now */ | |
500 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { | |
501 | struct bio *mbio = r1_bio->master_bio; | |
36a4e1fe N |
502 | pr_debug("raid1: behind end write sectors" |
503 | " %llu-%llu\n", | |
4f024f37 KO |
504 | (unsigned long long) mbio->bi_iter.bi_sector, |
505 | (unsigned long long) bio_end_sector(mbio) - 1); | |
d2eb35ac | 506 | call_bio_endio(r1_bio); |
4b6d287f N |
507 | } |
508 | } | |
509 | } | |
4367af55 | 510 | if (r1_bio->bios[mirror] == NULL) |
e5872d58 | 511 | rdev_dec_pending(rdev, conf->mddev); |
e9c7469b | 512 | |
1da177e4 | 513 | /* |
1da177e4 LT |
514 | * Let's see if all mirrored write operations have finished |
515 | * already. | |
516 | */ | |
af6d7b76 | 517 | r1_bio_write_done(r1_bio); |
c70810b3 | 518 | |
04b857f7 N |
519 | if (to_put) |
520 | bio_put(to_put); | |
1da177e4 LT |
521 | } |
522 | ||
fd76863e | 523 | static sector_t align_to_barrier_unit_end(sector_t start_sector, |
524 | sector_t sectors) | |
525 | { | |
526 | sector_t len; | |
527 | ||
528 | WARN_ON(sectors == 0); | |
529 | /* | |
530 | * len is the number of sectors from start_sector to end of the | |
531 | * barrier unit which start_sector belongs to. | |
532 | */ | |
533 | len = round_up(start_sector + 1, BARRIER_UNIT_SECTOR_SIZE) - | |
534 | start_sector; | |
535 | ||
536 | if (len > sectors) | |
537 | len = sectors; | |
538 | ||
539 | return len; | |
540 | } | |
541 | ||
1da177e4 LT |
542 | /* |
543 | * This routine returns the disk from which the requested read should | |
544 | * be done. There is a per-array 'next expected sequential IO' sector | |
545 | * number - if this matches on the next IO then we use the last disk. | |
546 | * There is also a per-disk 'last know head position' sector that is | |
547 | * maintained from IRQ contexts, both the normal and the resync IO | |
548 | * completion handlers update this position correctly. If there is no | |
549 | * perfect sequential match then we pick the disk whose head is closest. | |
550 | * | |
551 | * If there are 2 mirrors in the same 2 devices, performance degrades | |
552 | * because position is mirror, not device based. | |
553 | * | |
554 | * The rdev for the device selected will have nr_pending incremented. | |
555 | */ | |
e8096360 | 556 | static int read_balance(struct r1conf *conf, struct r1bio *r1_bio, int *max_sectors) |
1da177e4 | 557 | { |
af3a2cd6 | 558 | const sector_t this_sector = r1_bio->sector; |
d2eb35ac N |
559 | int sectors; |
560 | int best_good_sectors; | |
9dedf603 SL |
561 | int best_disk, best_dist_disk, best_pending_disk; |
562 | int has_nonrot_disk; | |
be4d3280 | 563 | int disk; |
76073054 | 564 | sector_t best_dist; |
9dedf603 | 565 | unsigned int min_pending; |
3cb03002 | 566 | struct md_rdev *rdev; |
f3ac8bf7 | 567 | int choose_first; |
12cee5a8 | 568 | int choose_next_idle; |
1da177e4 LT |
569 | |
570 | rcu_read_lock(); | |
571 | /* | |
8ddf9efe | 572 | * Check if we can balance. We can balance on the whole |
1da177e4 LT |
573 | * device if no resync is going on, or below the resync window. |
574 | * We take the first readable disk when above the resync window. | |
575 | */ | |
576 | retry: | |
d2eb35ac | 577 | sectors = r1_bio->sectors; |
76073054 | 578 | best_disk = -1; |
9dedf603 | 579 | best_dist_disk = -1; |
76073054 | 580 | best_dist = MaxSector; |
9dedf603 SL |
581 | best_pending_disk = -1; |
582 | min_pending = UINT_MAX; | |
d2eb35ac | 583 | best_good_sectors = 0; |
9dedf603 | 584 | has_nonrot_disk = 0; |
12cee5a8 | 585 | choose_next_idle = 0; |
2e52d449 | 586 | clear_bit(R1BIO_FailFast, &r1_bio->state); |
d2eb35ac | 587 | |
7d49ffcf GR |
588 | if ((conf->mddev->recovery_cp < this_sector + sectors) || |
589 | (mddev_is_clustered(conf->mddev) && | |
90382ed9 | 590 | md_cluster_ops->area_resyncing(conf->mddev, READ, this_sector, |
7d49ffcf GR |
591 | this_sector + sectors))) |
592 | choose_first = 1; | |
593 | else | |
594 | choose_first = 0; | |
1da177e4 | 595 | |
be4d3280 | 596 | for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) { |
76073054 | 597 | sector_t dist; |
d2eb35ac N |
598 | sector_t first_bad; |
599 | int bad_sectors; | |
9dedf603 | 600 | unsigned int pending; |
12cee5a8 | 601 | bool nonrot; |
d2eb35ac | 602 | |
f3ac8bf7 N |
603 | rdev = rcu_dereference(conf->mirrors[disk].rdev); |
604 | if (r1_bio->bios[disk] == IO_BLOCKED | |
605 | || rdev == NULL | |
76073054 | 606 | || test_bit(Faulty, &rdev->flags)) |
f3ac8bf7 | 607 | continue; |
76073054 N |
608 | if (!test_bit(In_sync, &rdev->flags) && |
609 | rdev->recovery_offset < this_sector + sectors) | |
1da177e4 | 610 | continue; |
76073054 N |
611 | if (test_bit(WriteMostly, &rdev->flags)) { |
612 | /* Don't balance among write-mostly, just | |
613 | * use the first as a last resort */ | |
d1901ef0 | 614 | if (best_dist_disk < 0) { |
307729c8 N |
615 | if (is_badblock(rdev, this_sector, sectors, |
616 | &first_bad, &bad_sectors)) { | |
816b0acf | 617 | if (first_bad <= this_sector) |
307729c8 N |
618 | /* Cannot use this */ |
619 | continue; | |
620 | best_good_sectors = first_bad - this_sector; | |
621 | } else | |
622 | best_good_sectors = sectors; | |
d1901ef0 TH |
623 | best_dist_disk = disk; |
624 | best_pending_disk = disk; | |
307729c8 | 625 | } |
76073054 N |
626 | continue; |
627 | } | |
628 | /* This is a reasonable device to use. It might | |
629 | * even be best. | |
630 | */ | |
d2eb35ac N |
631 | if (is_badblock(rdev, this_sector, sectors, |
632 | &first_bad, &bad_sectors)) { | |
633 | if (best_dist < MaxSector) | |
634 | /* already have a better device */ | |
635 | continue; | |
636 | if (first_bad <= this_sector) { | |
637 | /* cannot read here. If this is the 'primary' | |
638 | * device, then we must not read beyond | |
639 | * bad_sectors from another device.. | |
640 | */ | |
641 | bad_sectors -= (this_sector - first_bad); | |
642 | if (choose_first && sectors > bad_sectors) | |
643 | sectors = bad_sectors; | |
644 | if (best_good_sectors > sectors) | |
645 | best_good_sectors = sectors; | |
646 | ||
647 | } else { | |
648 | sector_t good_sectors = first_bad - this_sector; | |
649 | if (good_sectors > best_good_sectors) { | |
650 | best_good_sectors = good_sectors; | |
651 | best_disk = disk; | |
652 | } | |
653 | if (choose_first) | |
654 | break; | |
655 | } | |
656 | continue; | |
657 | } else | |
658 | best_good_sectors = sectors; | |
659 | ||
2e52d449 N |
660 | if (best_disk >= 0) |
661 | /* At least two disks to choose from so failfast is OK */ | |
662 | set_bit(R1BIO_FailFast, &r1_bio->state); | |
663 | ||
12cee5a8 SL |
664 | nonrot = blk_queue_nonrot(bdev_get_queue(rdev->bdev)); |
665 | has_nonrot_disk |= nonrot; | |
9dedf603 | 666 | pending = atomic_read(&rdev->nr_pending); |
76073054 | 667 | dist = abs(this_sector - conf->mirrors[disk].head_position); |
12cee5a8 | 668 | if (choose_first) { |
76073054 | 669 | best_disk = disk; |
1da177e4 LT |
670 | break; |
671 | } | |
12cee5a8 SL |
672 | /* Don't change to another disk for sequential reads */ |
673 | if (conf->mirrors[disk].next_seq_sect == this_sector | |
674 | || dist == 0) { | |
675 | int opt_iosize = bdev_io_opt(rdev->bdev) >> 9; | |
676 | struct raid1_info *mirror = &conf->mirrors[disk]; | |
677 | ||
678 | best_disk = disk; | |
679 | /* | |
680 | * If buffered sequential IO size exceeds optimal | |
681 | * iosize, check if there is idle disk. If yes, choose | |
682 | * the idle disk. read_balance could already choose an | |
683 | * idle disk before noticing it's a sequential IO in | |
684 | * this disk. This doesn't matter because this disk | |
685 | * will idle, next time it will be utilized after the | |
686 | * first disk has IO size exceeds optimal iosize. In | |
687 | * this way, iosize of the first disk will be optimal | |
688 | * iosize at least. iosize of the second disk might be | |
689 | * small, but not a big deal since when the second disk | |
690 | * starts IO, the first disk is likely still busy. | |
691 | */ | |
692 | if (nonrot && opt_iosize > 0 && | |
693 | mirror->seq_start != MaxSector && | |
694 | mirror->next_seq_sect > opt_iosize && | |
695 | mirror->next_seq_sect - opt_iosize >= | |
696 | mirror->seq_start) { | |
697 | choose_next_idle = 1; | |
698 | continue; | |
699 | } | |
700 | break; | |
701 | } | |
12cee5a8 SL |
702 | |
703 | if (choose_next_idle) | |
704 | continue; | |
9dedf603 SL |
705 | |
706 | if (min_pending > pending) { | |
707 | min_pending = pending; | |
708 | best_pending_disk = disk; | |
709 | } | |
710 | ||
76073054 N |
711 | if (dist < best_dist) { |
712 | best_dist = dist; | |
9dedf603 | 713 | best_dist_disk = disk; |
1da177e4 | 714 | } |
f3ac8bf7 | 715 | } |
1da177e4 | 716 | |
9dedf603 SL |
717 | /* |
718 | * If all disks are rotational, choose the closest disk. If any disk is | |
719 | * non-rotational, choose the disk with less pending request even the | |
720 | * disk is rotational, which might/might not be optimal for raids with | |
721 | * mixed ratation/non-rotational disks depending on workload. | |
722 | */ | |
723 | if (best_disk == -1) { | |
2e52d449 | 724 | if (has_nonrot_disk || min_pending == 0) |
9dedf603 SL |
725 | best_disk = best_pending_disk; |
726 | else | |
727 | best_disk = best_dist_disk; | |
728 | } | |
729 | ||
76073054 N |
730 | if (best_disk >= 0) { |
731 | rdev = rcu_dereference(conf->mirrors[best_disk].rdev); | |
8ddf9efe N |
732 | if (!rdev) |
733 | goto retry; | |
734 | atomic_inc(&rdev->nr_pending); | |
d2eb35ac | 735 | sectors = best_good_sectors; |
12cee5a8 SL |
736 | |
737 | if (conf->mirrors[best_disk].next_seq_sect != this_sector) | |
738 | conf->mirrors[best_disk].seq_start = this_sector; | |
739 | ||
be4d3280 | 740 | conf->mirrors[best_disk].next_seq_sect = this_sector + sectors; |
1da177e4 LT |
741 | } |
742 | rcu_read_unlock(); | |
d2eb35ac | 743 | *max_sectors = sectors; |
1da177e4 | 744 | |
76073054 | 745 | return best_disk; |
1da177e4 LT |
746 | } |
747 | ||
5c675f83 | 748 | static int raid1_congested(struct mddev *mddev, int bits) |
0d129228 | 749 | { |
e8096360 | 750 | struct r1conf *conf = mddev->private; |
0d129228 N |
751 | int i, ret = 0; |
752 | ||
4452226e | 753 | if ((bits & (1 << WB_async_congested)) && |
34db0cd6 N |
754 | conf->pending_count >= max_queued_requests) |
755 | return 1; | |
756 | ||
0d129228 | 757 | rcu_read_lock(); |
f53e29fc | 758 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3cb03002 | 759 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
0d129228 | 760 | if (rdev && !test_bit(Faulty, &rdev->flags)) { |
165125e1 | 761 | struct request_queue *q = bdev_get_queue(rdev->bdev); |
0d129228 | 762 | |
1ed7242e JB |
763 | BUG_ON(!q); |
764 | ||
0d129228 N |
765 | /* Note the '|| 1' - when read_balance prefers |
766 | * non-congested targets, it can be removed | |
767 | */ | |
4452226e | 768 | if ((bits & (1 << WB_async_congested)) || 1) |
dc3b17cc | 769 | ret |= bdi_congested(q->backing_dev_info, bits); |
0d129228 | 770 | else |
dc3b17cc | 771 | ret &= bdi_congested(q->backing_dev_info, bits); |
0d129228 N |
772 | } |
773 | } | |
774 | rcu_read_unlock(); | |
775 | return ret; | |
776 | } | |
0d129228 | 777 | |
e8096360 | 778 | static void flush_pending_writes(struct r1conf *conf) |
a35e63ef N |
779 | { |
780 | /* Any writes that have been queued but are awaiting | |
781 | * bitmap updates get flushed here. | |
a35e63ef | 782 | */ |
a35e63ef N |
783 | spin_lock_irq(&conf->device_lock); |
784 | ||
785 | if (conf->pending_bio_list.head) { | |
786 | struct bio *bio; | |
787 | bio = bio_list_get(&conf->pending_bio_list); | |
34db0cd6 | 788 | conf->pending_count = 0; |
a35e63ef N |
789 | spin_unlock_irq(&conf->device_lock); |
790 | /* flush any pending bitmap writes to | |
791 | * disk before proceeding w/ I/O */ | |
792 | bitmap_unplug(conf->mddev->bitmap); | |
34db0cd6 | 793 | wake_up(&conf->wait_barrier); |
a35e63ef N |
794 | |
795 | while (bio) { /* submit pending writes */ | |
796 | struct bio *next = bio->bi_next; | |
5e2c7a36 | 797 | struct md_rdev *rdev = (void*)bio->bi_bdev; |
a35e63ef | 798 | bio->bi_next = NULL; |
5e2c7a36 N |
799 | bio->bi_bdev = rdev->bdev; |
800 | if (test_bit(Faulty, &rdev->flags)) { | |
801 | bio->bi_error = -EIO; | |
802 | bio_endio(bio); | |
803 | } else if (unlikely((bio_op(bio) == REQ_OP_DISCARD) && | |
804 | !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) | |
2ff8cc2c | 805 | /* Just ignore it */ |
4246a0b6 | 806 | bio_endio(bio); |
2ff8cc2c SL |
807 | else |
808 | generic_make_request(bio); | |
a35e63ef N |
809 | bio = next; |
810 | } | |
a35e63ef N |
811 | } else |
812 | spin_unlock_irq(&conf->device_lock); | |
7eaceacc JA |
813 | } |
814 | ||
17999be4 N |
815 | /* Barriers.... |
816 | * Sometimes we need to suspend IO while we do something else, | |
817 | * either some resync/recovery, or reconfigure the array. | |
818 | * To do this we raise a 'barrier'. | |
819 | * The 'barrier' is a counter that can be raised multiple times | |
820 | * to count how many activities are happening which preclude | |
821 | * normal IO. | |
822 | * We can only raise the barrier if there is no pending IO. | |
823 | * i.e. if nr_pending == 0. | |
824 | * We choose only to raise the barrier if no-one is waiting for the | |
825 | * barrier to go down. This means that as soon as an IO request | |
826 | * is ready, no other operations which require a barrier will start | |
827 | * until the IO request has had a chance. | |
828 | * | |
829 | * So: regular IO calls 'wait_barrier'. When that returns there | |
830 | * is no backgroup IO happening, It must arrange to call | |
831 | * allow_barrier when it has finished its IO. | |
832 | * backgroup IO calls must call raise_barrier. Once that returns | |
833 | * there is no normal IO happeing. It must arrange to call | |
834 | * lower_barrier when the particular background IO completes. | |
1da177e4 | 835 | */ |
c2fd4c94 | 836 | static void raise_barrier(struct r1conf *conf, sector_t sector_nr) |
1da177e4 | 837 | { |
fd76863e | 838 | int idx = sector_to_idx(sector_nr); |
839 | ||
1da177e4 | 840 | spin_lock_irq(&conf->resync_lock); |
17999be4 N |
841 | |
842 | /* Wait until no block IO is waiting */ | |
824e47da | 843 | wait_event_lock_irq(conf->wait_barrier, |
844 | !atomic_read(&conf->nr_waiting[idx]), | |
eed8c02e | 845 | conf->resync_lock); |
17999be4 N |
846 | |
847 | /* block any new IO from starting */ | |
824e47da | 848 | atomic_inc(&conf->barrier[idx]); |
849 | /* | |
850 | * In raise_barrier() we firstly increase conf->barrier[idx] then | |
851 | * check conf->nr_pending[idx]. In _wait_barrier() we firstly | |
852 | * increase conf->nr_pending[idx] then check conf->barrier[idx]. | |
853 | * A memory barrier here to make sure conf->nr_pending[idx] won't | |
854 | * be fetched before conf->barrier[idx] is increased. Otherwise | |
855 | * there will be a race between raise_barrier() and _wait_barrier(). | |
856 | */ | |
857 | smp_mb__after_atomic(); | |
17999be4 | 858 | |
79ef3a8a | 859 | /* For these conditions we must wait: |
860 | * A: while the array is in frozen state | |
fd76863e | 861 | * B: while conf->nr_pending[idx] is not 0, meaning regular I/O |
862 | * existing in corresponding I/O barrier bucket. | |
863 | * C: while conf->barrier[idx] >= RESYNC_DEPTH, meaning reaches | |
864 | * max resync count which allowed on current I/O barrier bucket. | |
79ef3a8a | 865 | */ |
17999be4 | 866 | wait_event_lock_irq(conf->wait_barrier, |
b364e3d0 | 867 | !conf->array_frozen && |
824e47da | 868 | !atomic_read(&conf->nr_pending[idx]) && |
869 | atomic_read(&conf->barrier[idx]) < RESYNC_DEPTH, | |
eed8c02e | 870 | conf->resync_lock); |
17999be4 | 871 | |
824e47da | 872 | atomic_inc(&conf->nr_pending[idx]); |
17999be4 N |
873 | spin_unlock_irq(&conf->resync_lock); |
874 | } | |
875 | ||
fd76863e | 876 | static void lower_barrier(struct r1conf *conf, sector_t sector_nr) |
17999be4 | 877 | { |
fd76863e | 878 | int idx = sector_to_idx(sector_nr); |
879 | ||
824e47da | 880 | BUG_ON(atomic_read(&conf->barrier[idx]) <= 0); |
fd76863e | 881 | |
824e47da | 882 | atomic_dec(&conf->barrier[idx]); |
883 | atomic_dec(&conf->nr_pending[idx]); | |
17999be4 N |
884 | wake_up(&conf->wait_barrier); |
885 | } | |
886 | ||
fd76863e | 887 | static void _wait_barrier(struct r1conf *conf, int idx) |
17999be4 | 888 | { |
824e47da | 889 | /* |
890 | * We need to increase conf->nr_pending[idx] very early here, | |
891 | * then raise_barrier() can be blocked when it waits for | |
892 | * conf->nr_pending[idx] to be 0. Then we can avoid holding | |
893 | * conf->resync_lock when there is no barrier raised in same | |
894 | * barrier unit bucket. Also if the array is frozen, I/O | |
895 | * should be blocked until array is unfrozen. | |
896 | */ | |
897 | atomic_inc(&conf->nr_pending[idx]); | |
898 | /* | |
899 | * In _wait_barrier() we firstly increase conf->nr_pending[idx], then | |
900 | * check conf->barrier[idx]. In raise_barrier() we firstly increase | |
901 | * conf->barrier[idx], then check conf->nr_pending[idx]. A memory | |
902 | * barrier is necessary here to make sure conf->barrier[idx] won't be | |
903 | * fetched before conf->nr_pending[idx] is increased. Otherwise there | |
904 | * will be a race between _wait_barrier() and raise_barrier(). | |
905 | */ | |
906 | smp_mb__after_atomic(); | |
79ef3a8a | 907 | |
824e47da | 908 | /* |
909 | * Don't worry about checking two atomic_t variables at same time | |
910 | * here. If during we check conf->barrier[idx], the array is | |
911 | * frozen (conf->array_frozen is 1), and chonf->barrier[idx] is | |
912 | * 0, it is safe to return and make the I/O continue. Because the | |
913 | * array is frozen, all I/O returned here will eventually complete | |
914 | * or be queued, no race will happen. See code comment in | |
915 | * frozen_array(). | |
916 | */ | |
917 | if (!READ_ONCE(conf->array_frozen) && | |
918 | !atomic_read(&conf->barrier[idx])) | |
919 | return; | |
79ef3a8a | 920 | |
824e47da | 921 | /* |
922 | * After holding conf->resync_lock, conf->nr_pending[idx] | |
923 | * should be decreased before waiting for barrier to drop. | |
924 | * Otherwise, we may encounter a race condition because | |
925 | * raise_barrer() might be waiting for conf->nr_pending[idx] | |
926 | * to be 0 at same time. | |
927 | */ | |
928 | spin_lock_irq(&conf->resync_lock); | |
929 | atomic_inc(&conf->nr_waiting[idx]); | |
930 | atomic_dec(&conf->nr_pending[idx]); | |
931 | /* | |
932 | * In case freeze_array() is waiting for | |
933 | * get_unqueued_pending() == extra | |
934 | */ | |
935 | wake_up(&conf->wait_barrier); | |
936 | /* Wait for the barrier in same barrier unit bucket to drop. */ | |
937 | wait_event_lock_irq(conf->wait_barrier, | |
938 | !conf->array_frozen && | |
939 | !atomic_read(&conf->barrier[idx]), | |
940 | conf->resync_lock); | |
941 | atomic_inc(&conf->nr_pending[idx]); | |
942 | atomic_dec(&conf->nr_waiting[idx]); | |
fd76863e | 943 | spin_unlock_irq(&conf->resync_lock); |
79ef3a8a | 944 | } |
945 | ||
fd76863e | 946 | static void wait_read_barrier(struct r1conf *conf, sector_t sector_nr) |
79ef3a8a | 947 | { |
fd76863e | 948 | int idx = sector_to_idx(sector_nr); |
79ef3a8a | 949 | |
824e47da | 950 | /* |
951 | * Very similar to _wait_barrier(). The difference is, for read | |
952 | * I/O we don't need wait for sync I/O, but if the whole array | |
953 | * is frozen, the read I/O still has to wait until the array is | |
954 | * unfrozen. Since there is no ordering requirement with | |
955 | * conf->barrier[idx] here, memory barrier is unnecessary as well. | |
956 | */ | |
957 | atomic_inc(&conf->nr_pending[idx]); | |
79ef3a8a | 958 | |
824e47da | 959 | if (!READ_ONCE(conf->array_frozen)) |
960 | return; | |
961 | ||
962 | spin_lock_irq(&conf->resync_lock); | |
963 | atomic_inc(&conf->nr_waiting[idx]); | |
964 | atomic_dec(&conf->nr_pending[idx]); | |
965 | /* | |
966 | * In case freeze_array() is waiting for | |
967 | * get_unqueued_pending() == extra | |
968 | */ | |
969 | wake_up(&conf->wait_barrier); | |
970 | /* Wait for array to be unfrozen */ | |
971 | wait_event_lock_irq(conf->wait_barrier, | |
972 | !conf->array_frozen, | |
973 | conf->resync_lock); | |
974 | atomic_inc(&conf->nr_pending[idx]); | |
975 | atomic_dec(&conf->nr_waiting[idx]); | |
1da177e4 LT |
976 | spin_unlock_irq(&conf->resync_lock); |
977 | } | |
978 | ||
fd76863e | 979 | static void wait_barrier(struct r1conf *conf, sector_t sector_nr) |
17999be4 | 980 | { |
fd76863e | 981 | int idx = sector_to_idx(sector_nr); |
79ef3a8a | 982 | |
fd76863e | 983 | _wait_barrier(conf, idx); |
984 | } | |
985 | ||
986 | static void wait_all_barriers(struct r1conf *conf) | |
987 | { | |
988 | int idx; | |
989 | ||
990 | for (idx = 0; idx < BARRIER_BUCKETS_NR; idx++) | |
991 | _wait_barrier(conf, idx); | |
992 | } | |
993 | ||
994 | static void _allow_barrier(struct r1conf *conf, int idx) | |
17999be4 | 995 | { |
824e47da | 996 | atomic_dec(&conf->nr_pending[idx]); |
17999be4 N |
997 | wake_up(&conf->wait_barrier); |
998 | } | |
999 | ||
fd76863e | 1000 | static void allow_barrier(struct r1conf *conf, sector_t sector_nr) |
1001 | { | |
1002 | int idx = sector_to_idx(sector_nr); | |
1003 | ||
1004 | _allow_barrier(conf, idx); | |
1005 | } | |
1006 | ||
1007 | static void allow_all_barriers(struct r1conf *conf) | |
1008 | { | |
1009 | int idx; | |
1010 | ||
1011 | for (idx = 0; idx < BARRIER_BUCKETS_NR; idx++) | |
1012 | _allow_barrier(conf, idx); | |
1013 | } | |
1014 | ||
1015 | /* conf->resync_lock should be held */ | |
1016 | static int get_unqueued_pending(struct r1conf *conf) | |
1017 | { | |
1018 | int idx, ret; | |
1019 | ||
1020 | for (ret = 0, idx = 0; idx < BARRIER_BUCKETS_NR; idx++) | |
824e47da | 1021 | ret += atomic_read(&conf->nr_pending[idx]) - |
1022 | atomic_read(&conf->nr_queued[idx]); | |
fd76863e | 1023 | |
1024 | return ret; | |
1025 | } | |
1026 | ||
e2d59925 | 1027 | static void freeze_array(struct r1conf *conf, int extra) |
ddaf22ab | 1028 | { |
fd76863e | 1029 | /* Stop sync I/O and normal I/O and wait for everything to |
11353b9d | 1030 | * go quiet. |
fd76863e | 1031 | * This is called in two situations: |
1032 | * 1) management command handlers (reshape, remove disk, quiesce). | |
1033 | * 2) one normal I/O request failed. | |
1034 | ||
1035 | * After array_frozen is set to 1, new sync IO will be blocked at | |
1036 | * raise_barrier(), and new normal I/O will blocked at _wait_barrier() | |
1037 | * or wait_read_barrier(). The flying I/Os will either complete or be | |
1038 | * queued. When everything goes quite, there are only queued I/Os left. | |
1039 | ||
1040 | * Every flying I/O contributes to a conf->nr_pending[idx], idx is the | |
1041 | * barrier bucket index which this I/O request hits. When all sync and | |
1042 | * normal I/O are queued, sum of all conf->nr_pending[] will match sum | |
1043 | * of all conf->nr_queued[]. But normal I/O failure is an exception, | |
1044 | * in handle_read_error(), we may call freeze_array() before trying to | |
1045 | * fix the read error. In this case, the error read I/O is not queued, | |
1046 | * so get_unqueued_pending() == 1. | |
1047 | * | |
1048 | * Therefore before this function returns, we need to wait until | |
1049 | * get_unqueued_pendings(conf) gets equal to extra. For | |
1050 | * normal I/O context, extra is 1, in rested situations extra is 0. | |
ddaf22ab N |
1051 | */ |
1052 | spin_lock_irq(&conf->resync_lock); | |
b364e3d0 | 1053 | conf->array_frozen = 1; |
578b54ad | 1054 | raid1_log(conf->mddev, "wait freeze"); |
fd76863e | 1055 | wait_event_lock_irq_cmd( |
1056 | conf->wait_barrier, | |
1057 | get_unqueued_pending(conf) == extra, | |
1058 | conf->resync_lock, | |
1059 | flush_pending_writes(conf)); | |
ddaf22ab N |
1060 | spin_unlock_irq(&conf->resync_lock); |
1061 | } | |
e8096360 | 1062 | static void unfreeze_array(struct r1conf *conf) |
ddaf22ab N |
1063 | { |
1064 | /* reverse the effect of the freeze */ | |
1065 | spin_lock_irq(&conf->resync_lock); | |
b364e3d0 | 1066 | conf->array_frozen = 0; |
ddaf22ab | 1067 | spin_unlock_irq(&conf->resync_lock); |
824e47da | 1068 | wake_up(&conf->wait_barrier); |
ddaf22ab N |
1069 | } |
1070 | ||
f72ffdd6 | 1071 | /* duplicate the data pages for behind I/O |
4e78064f | 1072 | */ |
9f2c9d12 | 1073 | static void alloc_behind_pages(struct bio *bio, struct r1bio *r1_bio) |
4b6d287f N |
1074 | { |
1075 | int i; | |
1076 | struct bio_vec *bvec; | |
2ca68f5e | 1077 | struct bio_vec *bvecs = kzalloc(bio->bi_vcnt * sizeof(struct bio_vec), |
4b6d287f | 1078 | GFP_NOIO); |
2ca68f5e | 1079 | if (unlikely(!bvecs)) |
af6d7b76 | 1080 | return; |
4b6d287f | 1081 | |
cb34e057 | 1082 | bio_for_each_segment_all(bvec, bio, i) { |
2ca68f5e N |
1083 | bvecs[i] = *bvec; |
1084 | bvecs[i].bv_page = alloc_page(GFP_NOIO); | |
1085 | if (unlikely(!bvecs[i].bv_page)) | |
4b6d287f | 1086 | goto do_sync_io; |
2ca68f5e N |
1087 | memcpy(kmap(bvecs[i].bv_page) + bvec->bv_offset, |
1088 | kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len); | |
1089 | kunmap(bvecs[i].bv_page); | |
4b6d287f N |
1090 | kunmap(bvec->bv_page); |
1091 | } | |
2ca68f5e | 1092 | r1_bio->behind_bvecs = bvecs; |
af6d7b76 N |
1093 | r1_bio->behind_page_count = bio->bi_vcnt; |
1094 | set_bit(R1BIO_BehindIO, &r1_bio->state); | |
1095 | return; | |
4b6d287f N |
1096 | |
1097 | do_sync_io: | |
af6d7b76 | 1098 | for (i = 0; i < bio->bi_vcnt; i++) |
2ca68f5e N |
1099 | if (bvecs[i].bv_page) |
1100 | put_page(bvecs[i].bv_page); | |
1101 | kfree(bvecs); | |
4f024f37 KO |
1102 | pr_debug("%dB behind alloc failed, doing sync I/O\n", |
1103 | bio->bi_iter.bi_size); | |
4b6d287f N |
1104 | } |
1105 | ||
f54a9d0e N |
1106 | struct raid1_plug_cb { |
1107 | struct blk_plug_cb cb; | |
1108 | struct bio_list pending; | |
1109 | int pending_cnt; | |
1110 | }; | |
1111 | ||
1112 | static void raid1_unplug(struct blk_plug_cb *cb, bool from_schedule) | |
1113 | { | |
1114 | struct raid1_plug_cb *plug = container_of(cb, struct raid1_plug_cb, | |
1115 | cb); | |
1116 | struct mddev *mddev = plug->cb.data; | |
1117 | struct r1conf *conf = mddev->private; | |
1118 | struct bio *bio; | |
1119 | ||
874807a8 | 1120 | if (from_schedule || current->bio_list) { |
f54a9d0e N |
1121 | spin_lock_irq(&conf->device_lock); |
1122 | bio_list_merge(&conf->pending_bio_list, &plug->pending); | |
1123 | conf->pending_count += plug->pending_cnt; | |
1124 | spin_unlock_irq(&conf->device_lock); | |
ee0b0244 | 1125 | wake_up(&conf->wait_barrier); |
f54a9d0e N |
1126 | md_wakeup_thread(mddev->thread); |
1127 | kfree(plug); | |
1128 | return; | |
1129 | } | |
1130 | ||
1131 | /* we aren't scheduling, so we can do the write-out directly. */ | |
1132 | bio = bio_list_get(&plug->pending); | |
1133 | bitmap_unplug(mddev->bitmap); | |
1134 | wake_up(&conf->wait_barrier); | |
1135 | ||
1136 | while (bio) { /* submit pending writes */ | |
1137 | struct bio *next = bio->bi_next; | |
5e2c7a36 | 1138 | struct md_rdev *rdev = (void*)bio->bi_bdev; |
f54a9d0e | 1139 | bio->bi_next = NULL; |
5e2c7a36 N |
1140 | bio->bi_bdev = rdev->bdev; |
1141 | if (test_bit(Faulty, &rdev->flags)) { | |
1142 | bio->bi_error = -EIO; | |
1143 | bio_endio(bio); | |
1144 | } else if (unlikely((bio_op(bio) == REQ_OP_DISCARD) && | |
1145 | !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) | |
32f9f570 | 1146 | /* Just ignore it */ |
4246a0b6 | 1147 | bio_endio(bio); |
32f9f570 SL |
1148 | else |
1149 | generic_make_request(bio); | |
f54a9d0e N |
1150 | bio = next; |
1151 | } | |
1152 | kfree(plug); | |
1153 | } | |
1154 | ||
fd76863e | 1155 | static inline struct r1bio * |
1156 | alloc_r1bio(struct mddev *mddev, struct bio *bio, sector_t sectors_handled) | |
1157 | { | |
1158 | struct r1conf *conf = mddev->private; | |
1159 | struct r1bio *r1_bio; | |
1160 | ||
1161 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
1162 | ||
1163 | r1_bio->master_bio = bio; | |
1164 | r1_bio->sectors = bio_sectors(bio) - sectors_handled; | |
1165 | r1_bio->state = 0; | |
1166 | r1_bio->mddev = mddev; | |
1167 | r1_bio->sector = bio->bi_iter.bi_sector + sectors_handled; | |
1168 | ||
1169 | return r1_bio; | |
1170 | } | |
1171 | ||
1172 | static void raid1_read_request(struct mddev *mddev, struct bio *bio) | |
1da177e4 | 1173 | { |
e8096360 | 1174 | struct r1conf *conf = mddev->private; |
0eaf822c | 1175 | struct raid1_info *mirror; |
fd76863e | 1176 | struct r1bio *r1_bio; |
1da177e4 | 1177 | struct bio *read_bio; |
3b046a97 RL |
1178 | struct bitmap *bitmap = mddev->bitmap; |
1179 | const int op = bio_op(bio); | |
1180 | const unsigned long do_sync = (bio->bi_opf & REQ_SYNC); | |
1181 | int sectors_handled; | |
1182 | int max_sectors; | |
1183 | int rdisk; | |
1184 | ||
fd76863e | 1185 | /* |
1186 | * Still need barrier for READ in case that whole | |
1187 | * array is frozen. | |
1188 | */ | |
1189 | wait_read_barrier(conf, bio->bi_iter.bi_sector); | |
1190 | ||
1191 | r1_bio = alloc_r1bio(mddev, bio, 0); | |
3b046a97 | 1192 | |
fd76863e | 1193 | /* |
1194 | * We might need to issue multiple reads to different | |
1195 | * devices if there are bad blocks around, so we keep | |
1196 | * track of the number of reads in bio->bi_phys_segments. | |
1197 | * If this is 0, there is only one r1_bio and no locking | |
1198 | * will be needed when requests complete. If it is | |
1199 | * non-zero, then it is the number of not-completed requests. | |
1200 | */ | |
1201 | bio->bi_phys_segments = 0; | |
1202 | bio_clear_flag(bio, BIO_SEG_VALID); | |
1203 | ||
1204 | /* | |
1205 | * make_request() can abort the operation when read-ahead is being | |
1206 | * used and no empty request is available. | |
1207 | */ | |
3b046a97 RL |
1208 | read_again: |
1209 | rdisk = read_balance(conf, r1_bio, &max_sectors); | |
1210 | ||
1211 | if (rdisk < 0) { | |
1212 | /* couldn't find anywhere to read from */ | |
1213 | raid_end_bio_io(r1_bio); | |
1214 | return; | |
1215 | } | |
1216 | mirror = conf->mirrors + rdisk; | |
1217 | ||
1218 | if (test_bit(WriteMostly, &mirror->rdev->flags) && | |
1219 | bitmap) { | |
1220 | /* | |
1221 | * Reading from a write-mostly device must take care not to | |
1222 | * over-take any writes that are 'behind' | |
1223 | */ | |
1224 | raid1_log(mddev, "wait behind writes"); | |
1225 | wait_event(bitmap->behind_wait, | |
1226 | atomic_read(&bitmap->behind_writes) == 0); | |
1227 | } | |
1228 | r1_bio->read_disk = rdisk; | |
3b046a97 | 1229 | |
d7a10308 | 1230 | read_bio = bio_clone_fast(bio, GFP_NOIO, mddev->bio_set); |
3b046a97 RL |
1231 | bio_trim(read_bio, r1_bio->sector - bio->bi_iter.bi_sector, |
1232 | max_sectors); | |
1233 | ||
1234 | r1_bio->bios[rdisk] = read_bio; | |
1235 | ||
1236 | read_bio->bi_iter.bi_sector = r1_bio->sector + | |
1237 | mirror->rdev->data_offset; | |
1238 | read_bio->bi_bdev = mirror->rdev->bdev; | |
1239 | read_bio->bi_end_io = raid1_end_read_request; | |
1240 | bio_set_op_attrs(read_bio, op, do_sync); | |
1241 | if (test_bit(FailFast, &mirror->rdev->flags) && | |
1242 | test_bit(R1BIO_FailFast, &r1_bio->state)) | |
1243 | read_bio->bi_opf |= MD_FAILFAST; | |
1244 | read_bio->bi_private = r1_bio; | |
1245 | ||
1246 | if (mddev->gendisk) | |
1247 | trace_block_bio_remap(bdev_get_queue(read_bio->bi_bdev), | |
1248 | read_bio, disk_devt(mddev->gendisk), | |
1249 | r1_bio->sector); | |
1250 | ||
1251 | if (max_sectors < r1_bio->sectors) { | |
1252 | /* | |
1253 | * could not read all from this device, so we will need another | |
1254 | * r1_bio. | |
1255 | */ | |
1256 | sectors_handled = (r1_bio->sector + max_sectors | |
1257 | - bio->bi_iter.bi_sector); | |
1258 | r1_bio->sectors = max_sectors; | |
1259 | spin_lock_irq(&conf->device_lock); | |
1260 | if (bio->bi_phys_segments == 0) | |
1261 | bio->bi_phys_segments = 2; | |
1262 | else | |
1263 | bio->bi_phys_segments++; | |
1264 | spin_unlock_irq(&conf->device_lock); | |
1265 | ||
1266 | /* | |
1267 | * Cannot call generic_make_request directly as that will be | |
1268 | * queued in __make_request and subsequent mempool_alloc might | |
1269 | * block waiting for it. So hand bio over to raid1d. | |
1270 | */ | |
1271 | reschedule_retry(r1_bio); | |
1272 | ||
fd76863e | 1273 | r1_bio = alloc_r1bio(mddev, bio, sectors_handled); |
3b046a97 RL |
1274 | goto read_again; |
1275 | } else | |
1276 | generic_make_request(read_bio); | |
1277 | } | |
1278 | ||
fd76863e | 1279 | static void raid1_write_request(struct mddev *mddev, struct bio *bio) |
3b046a97 RL |
1280 | { |
1281 | struct r1conf *conf = mddev->private; | |
fd76863e | 1282 | struct r1bio *r1_bio; |
1f68f0c4 | 1283 | int i, disks; |
3b046a97 | 1284 | struct bitmap *bitmap = mddev->bitmap; |
191ea9b2 | 1285 | unsigned long flags; |
3cb03002 | 1286 | struct md_rdev *blocked_rdev; |
f54a9d0e N |
1287 | struct blk_plug_cb *cb; |
1288 | struct raid1_plug_cb *plug = NULL; | |
1f68f0c4 N |
1289 | int first_clone; |
1290 | int sectors_handled; | |
1291 | int max_sectors; | |
191ea9b2 | 1292 | |
1da177e4 LT |
1293 | /* |
1294 | * Register the new request and wait if the reconstruction | |
1295 | * thread has put up a bar for new requests. | |
1296 | * Continue immediately if no resync is active currently. | |
1297 | */ | |
62de608d | 1298 | |
3d310eb7 N |
1299 | md_write_start(mddev, bio); /* wait on superblock update early */ |
1300 | ||
3b046a97 | 1301 | if ((bio_end_sector(bio) > mddev->suspend_lo && |
589a1c49 GR |
1302 | bio->bi_iter.bi_sector < mddev->suspend_hi) || |
1303 | (mddev_is_clustered(mddev) && | |
90382ed9 | 1304 | md_cluster_ops->area_resyncing(mddev, WRITE, |
3b046a97 RL |
1305 | bio->bi_iter.bi_sector, bio_end_sector(bio)))) { |
1306 | ||
1307 | /* | |
1308 | * As the suspend_* range is controlled by userspace, we want | |
1309 | * an interruptible wait. | |
6eef4b21 N |
1310 | */ |
1311 | DEFINE_WAIT(w); | |
1312 | for (;;) { | |
1313 | flush_signals(current); | |
1314 | prepare_to_wait(&conf->wait_barrier, | |
1315 | &w, TASK_INTERRUPTIBLE); | |
f73a1c7d | 1316 | if (bio_end_sector(bio) <= mddev->suspend_lo || |
589a1c49 GR |
1317 | bio->bi_iter.bi_sector >= mddev->suspend_hi || |
1318 | (mddev_is_clustered(mddev) && | |
90382ed9 | 1319 | !md_cluster_ops->area_resyncing(mddev, WRITE, |
3b046a97 RL |
1320 | bio->bi_iter.bi_sector, |
1321 | bio_end_sector(bio)))) | |
6eef4b21 N |
1322 | break; |
1323 | schedule(); | |
1324 | } | |
1325 | finish_wait(&conf->wait_barrier, &w); | |
1326 | } | |
fd76863e | 1327 | wait_barrier(conf, bio->bi_iter.bi_sector); |
1328 | ||
1329 | r1_bio = alloc_r1bio(mddev, bio, 0); | |
1330 | ||
1331 | /* We might need to issue multiple writes to different | |
1332 | * devices if there are bad blocks around, so we keep | |
1333 | * track of the number of writes in bio->bi_phys_segments. | |
1334 | * If this is 0, there is only one r1_bio and no locking | |
1335 | * will be needed when requests complete. If it is | |
1336 | * non-zero, then it is the number of not-completed requests. | |
1337 | */ | |
1338 | bio->bi_phys_segments = 0; | |
1339 | bio_clear_flag(bio, BIO_SEG_VALID); | |
1da177e4 | 1340 | |
34db0cd6 N |
1341 | if (conf->pending_count >= max_queued_requests) { |
1342 | md_wakeup_thread(mddev->thread); | |
578b54ad | 1343 | raid1_log(mddev, "wait queued"); |
34db0cd6 N |
1344 | wait_event(conf->wait_barrier, |
1345 | conf->pending_count < max_queued_requests); | |
1346 | } | |
1f68f0c4 | 1347 | /* first select target devices under rcu_lock and |
1da177e4 LT |
1348 | * inc refcount on their rdev. Record them by setting |
1349 | * bios[x] to bio | |
1f68f0c4 N |
1350 | * If there are known/acknowledged bad blocks on any device on |
1351 | * which we have seen a write error, we want to avoid writing those | |
1352 | * blocks. | |
1353 | * This potentially requires several writes to write around | |
1354 | * the bad blocks. Each set of writes gets it's own r1bio | |
1355 | * with a set of bios attached. | |
1da177e4 | 1356 | */ |
c3b328ac | 1357 | |
8f19ccb2 | 1358 | disks = conf->raid_disks * 2; |
6bfe0b49 DW |
1359 | retry_write: |
1360 | blocked_rdev = NULL; | |
1da177e4 | 1361 | rcu_read_lock(); |
1f68f0c4 | 1362 | max_sectors = r1_bio->sectors; |
1da177e4 | 1363 | for (i = 0; i < disks; i++) { |
3cb03002 | 1364 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
6bfe0b49 DW |
1365 | if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { |
1366 | atomic_inc(&rdev->nr_pending); | |
1367 | blocked_rdev = rdev; | |
1368 | break; | |
1369 | } | |
1f68f0c4 | 1370 | r1_bio->bios[i] = NULL; |
8ae12666 | 1371 | if (!rdev || test_bit(Faulty, &rdev->flags)) { |
8f19ccb2 N |
1372 | if (i < conf->raid_disks) |
1373 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
1f68f0c4 N |
1374 | continue; |
1375 | } | |
1376 | ||
1377 | atomic_inc(&rdev->nr_pending); | |
1378 | if (test_bit(WriteErrorSeen, &rdev->flags)) { | |
1379 | sector_t first_bad; | |
1380 | int bad_sectors; | |
1381 | int is_bad; | |
1382 | ||
3b046a97 | 1383 | is_bad = is_badblock(rdev, r1_bio->sector, max_sectors, |
1f68f0c4 N |
1384 | &first_bad, &bad_sectors); |
1385 | if (is_bad < 0) { | |
1386 | /* mustn't write here until the bad block is | |
1387 | * acknowledged*/ | |
1388 | set_bit(BlockedBadBlocks, &rdev->flags); | |
1389 | blocked_rdev = rdev; | |
1390 | break; | |
1391 | } | |
1392 | if (is_bad && first_bad <= r1_bio->sector) { | |
1393 | /* Cannot write here at all */ | |
1394 | bad_sectors -= (r1_bio->sector - first_bad); | |
1395 | if (bad_sectors < max_sectors) | |
1396 | /* mustn't write more than bad_sectors | |
1397 | * to other devices yet | |
1398 | */ | |
1399 | max_sectors = bad_sectors; | |
03c902e1 | 1400 | rdev_dec_pending(rdev, mddev); |
1f68f0c4 N |
1401 | /* We don't set R1BIO_Degraded as that |
1402 | * only applies if the disk is | |
1403 | * missing, so it might be re-added, | |
1404 | * and we want to know to recover this | |
1405 | * chunk. | |
1406 | * In this case the device is here, | |
1407 | * and the fact that this chunk is not | |
1408 | * in-sync is recorded in the bad | |
1409 | * block log | |
1410 | */ | |
1411 | continue; | |
964147d5 | 1412 | } |
1f68f0c4 N |
1413 | if (is_bad) { |
1414 | int good_sectors = first_bad - r1_bio->sector; | |
1415 | if (good_sectors < max_sectors) | |
1416 | max_sectors = good_sectors; | |
1417 | } | |
1418 | } | |
1419 | r1_bio->bios[i] = bio; | |
1da177e4 LT |
1420 | } |
1421 | rcu_read_unlock(); | |
1422 | ||
6bfe0b49 DW |
1423 | if (unlikely(blocked_rdev)) { |
1424 | /* Wait for this device to become unblocked */ | |
1425 | int j; | |
1426 | ||
1427 | for (j = 0; j < i; j++) | |
1428 | if (r1_bio->bios[j]) | |
1429 | rdev_dec_pending(conf->mirrors[j].rdev, mddev); | |
1f68f0c4 | 1430 | r1_bio->state = 0; |
fd76863e | 1431 | allow_barrier(conf, bio->bi_iter.bi_sector); |
578b54ad | 1432 | raid1_log(mddev, "wait rdev %d blocked", blocked_rdev->raid_disk); |
6bfe0b49 | 1433 | md_wait_for_blocked_rdev(blocked_rdev, mddev); |
fd76863e | 1434 | wait_barrier(conf, bio->bi_iter.bi_sector); |
6bfe0b49 DW |
1435 | goto retry_write; |
1436 | } | |
1437 | ||
1f68f0c4 N |
1438 | if (max_sectors < r1_bio->sectors) { |
1439 | /* We are splitting this write into multiple parts, so | |
1440 | * we need to prepare for allocating another r1_bio. | |
1441 | */ | |
1442 | r1_bio->sectors = max_sectors; | |
1443 | spin_lock_irq(&conf->device_lock); | |
1444 | if (bio->bi_phys_segments == 0) | |
1445 | bio->bi_phys_segments = 2; | |
1446 | else | |
1447 | bio->bi_phys_segments++; | |
1448 | spin_unlock_irq(&conf->device_lock); | |
191ea9b2 | 1449 | } |
4f024f37 | 1450 | sectors_handled = r1_bio->sector + max_sectors - bio->bi_iter.bi_sector; |
4b6d287f | 1451 | |
4e78064f | 1452 | atomic_set(&r1_bio->remaining, 1); |
4b6d287f | 1453 | atomic_set(&r1_bio->behind_remaining, 0); |
06d91a5f | 1454 | |
1f68f0c4 | 1455 | first_clone = 1; |
1da177e4 | 1456 | for (i = 0; i < disks; i++) { |
8e58e327 ML |
1457 | struct bio *mbio = NULL; |
1458 | sector_t offset; | |
1da177e4 LT |
1459 | if (!r1_bio->bios[i]) |
1460 | continue; | |
1461 | ||
8e58e327 | 1462 | offset = r1_bio->sector - bio->bi_iter.bi_sector; |
1f68f0c4 N |
1463 | |
1464 | if (first_clone) { | |
1465 | /* do behind I/O ? | |
1466 | * Not if there are too many, or cannot | |
1467 | * allocate memory, or a reader on WriteMostly | |
1468 | * is waiting for behind writes to flush */ | |
1469 | if (bitmap && | |
1470 | (atomic_read(&bitmap->behind_writes) | |
1471 | < mddev->bitmap_info.max_write_behind) && | |
8e58e327 ML |
1472 | !waitqueue_active(&bitmap->behind_wait)) { |
1473 | mbio = bio_clone_bioset_partial(bio, GFP_NOIO, | |
1474 | mddev->bio_set, | |
1ec49223 SL |
1475 | offset << 9, |
1476 | max_sectors << 9); | |
1f68f0c4 | 1477 | alloc_behind_pages(mbio, r1_bio); |
8e58e327 | 1478 | } |
1f68f0c4 N |
1479 | |
1480 | bitmap_startwrite(bitmap, r1_bio->sector, | |
1481 | r1_bio->sectors, | |
1482 | test_bit(R1BIO_BehindIO, | |
1483 | &r1_bio->state)); | |
1484 | first_clone = 0; | |
1485 | } | |
8e58e327 ML |
1486 | |
1487 | if (!mbio) { | |
1ec49223 SL |
1488 | if (r1_bio->behind_bvecs) |
1489 | mbio = bio_clone_bioset_partial(bio, GFP_NOIO, | |
1490 | mddev->bio_set, | |
1491 | offset << 9, | |
1492 | max_sectors << 9); | |
1493 | else { | |
1494 | mbio = bio_clone_fast(bio, GFP_NOIO, mddev->bio_set); | |
1495 | bio_trim(mbio, offset, max_sectors); | |
1496 | } | |
8e58e327 ML |
1497 | } |
1498 | ||
2ca68f5e | 1499 | if (r1_bio->behind_bvecs) { |
4b6d287f N |
1500 | struct bio_vec *bvec; |
1501 | int j; | |
1502 | ||
cb34e057 KO |
1503 | /* |
1504 | * We trimmed the bio, so _all is legit | |
4b6d287f | 1505 | */ |
d74c6d51 | 1506 | bio_for_each_segment_all(bvec, mbio, j) |
2ca68f5e | 1507 | bvec->bv_page = r1_bio->behind_bvecs[j].bv_page; |
4b6d287f N |
1508 | if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags)) |
1509 | atomic_inc(&r1_bio->behind_remaining); | |
1510 | } | |
1511 | ||
1f68f0c4 N |
1512 | r1_bio->bios[i] = mbio; |
1513 | ||
4f024f37 | 1514 | mbio->bi_iter.bi_sector = (r1_bio->sector + |
1f68f0c4 | 1515 | conf->mirrors[i].rdev->data_offset); |
109e3765 | 1516 | mbio->bi_bdev = conf->mirrors[i].rdev->bdev; |
1f68f0c4 | 1517 | mbio->bi_end_io = raid1_end_write_request; |
a682e003 | 1518 | mbio->bi_opf = bio_op(bio) | (bio->bi_opf & (REQ_SYNC | REQ_FUA)); |
212e7eb7 N |
1519 | if (test_bit(FailFast, &conf->mirrors[i].rdev->flags) && |
1520 | !test_bit(WriteMostly, &conf->mirrors[i].rdev->flags) && | |
1521 | conf->raid_disks - mddev->degraded > 1) | |
1522 | mbio->bi_opf |= MD_FAILFAST; | |
1f68f0c4 N |
1523 | mbio->bi_private = r1_bio; |
1524 | ||
1da177e4 | 1525 | atomic_inc(&r1_bio->remaining); |
f54a9d0e | 1526 | |
109e3765 N |
1527 | if (mddev->gendisk) |
1528 | trace_block_bio_remap(bdev_get_queue(mbio->bi_bdev), | |
1529 | mbio, disk_devt(mddev->gendisk), | |
1530 | r1_bio->sector); | |
1531 | /* flush_pending_writes() needs access to the rdev so...*/ | |
1532 | mbio->bi_bdev = (void*)conf->mirrors[i].rdev; | |
1533 | ||
f54a9d0e N |
1534 | cb = blk_check_plugged(raid1_unplug, mddev, sizeof(*plug)); |
1535 | if (cb) | |
1536 | plug = container_of(cb, struct raid1_plug_cb, cb); | |
1537 | else | |
1538 | plug = NULL; | |
4e78064f | 1539 | spin_lock_irqsave(&conf->device_lock, flags); |
f54a9d0e N |
1540 | if (plug) { |
1541 | bio_list_add(&plug->pending, mbio); | |
1542 | plug->pending_cnt++; | |
1543 | } else { | |
1544 | bio_list_add(&conf->pending_bio_list, mbio); | |
1545 | conf->pending_count++; | |
1546 | } | |
4e78064f | 1547 | spin_unlock_irqrestore(&conf->device_lock, flags); |
f54a9d0e | 1548 | if (!plug) |
b357f04a | 1549 | md_wakeup_thread(mddev->thread); |
1da177e4 | 1550 | } |
079fa166 N |
1551 | /* Mustn't call r1_bio_write_done before this next test, |
1552 | * as it could result in the bio being freed. | |
1553 | */ | |
aa8b57aa | 1554 | if (sectors_handled < bio_sectors(bio)) { |
079fa166 | 1555 | r1_bio_write_done(r1_bio); |
1f68f0c4 N |
1556 | /* We need another r1_bio. It has already been counted |
1557 | * in bio->bi_phys_segments | |
1558 | */ | |
fd76863e | 1559 | r1_bio = alloc_r1bio(mddev, bio, sectors_handled); |
1f68f0c4 N |
1560 | goto retry_write; |
1561 | } | |
1562 | ||
079fa166 N |
1563 | r1_bio_write_done(r1_bio); |
1564 | ||
1565 | /* In case raid1d snuck in to freeze_array */ | |
1566 | wake_up(&conf->wait_barrier); | |
1da177e4 LT |
1567 | } |
1568 | ||
3b046a97 RL |
1569 | static void raid1_make_request(struct mddev *mddev, struct bio *bio) |
1570 | { | |
fd76863e | 1571 | struct bio *split; |
1572 | sector_t sectors; | |
3b046a97 | 1573 | |
aff8da09 SL |
1574 | if (unlikely(bio->bi_opf & REQ_PREFLUSH)) { |
1575 | md_flush_request(mddev, bio); | |
1576 | return; | |
1577 | } | |
3b046a97 | 1578 | |
fd76863e | 1579 | /* if bio exceeds barrier unit boundary, split it */ |
1580 | do { | |
1581 | sectors = align_to_barrier_unit_end( | |
1582 | bio->bi_iter.bi_sector, bio_sectors(bio)); | |
1583 | if (sectors < bio_sectors(bio)) { | |
1584 | split = bio_split(bio, sectors, GFP_NOIO, fs_bio_set); | |
1585 | bio_chain(split, bio); | |
1586 | } else { | |
1587 | split = bio; | |
1588 | } | |
3b046a97 | 1589 | |
61eb2b43 | 1590 | if (bio_data_dir(split) == READ) { |
fd76863e | 1591 | raid1_read_request(mddev, split); |
61eb2b43 SL |
1592 | |
1593 | /* | |
1594 | * If a bio is splitted, the first part of bio will | |
1595 | * pass barrier but the bio is queued in | |
1596 | * current->bio_list (see generic_make_request). If | |
1597 | * there is a raise_barrier() called here, the second | |
1598 | * part of bio can't pass barrier. But since the first | |
1599 | * part bio isn't dispatched to underlaying disks yet, | |
1600 | * the barrier is never released, hence raise_barrier | |
1601 | * will alays wait. We have a deadlock. | |
1602 | * Note, this only happens in read path. For write | |
1603 | * path, the first part of bio is dispatched in a | |
1604 | * schedule() call (because of blk plug) or offloaded | |
1605 | * to raid10d. | |
1606 | * Quitting from the function immediately can change | |
1607 | * the bio order queued in bio_list and avoid the deadlock. | |
1608 | */ | |
1609 | if (split != bio) { | |
1610 | generic_make_request(bio); | |
1611 | break; | |
1612 | } | |
1613 | } else | |
fd76863e | 1614 | raid1_write_request(mddev, split); |
1615 | } while (split != bio); | |
3b046a97 RL |
1616 | } |
1617 | ||
849674e4 | 1618 | static void raid1_status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 1619 | { |
e8096360 | 1620 | struct r1conf *conf = mddev->private; |
1da177e4 LT |
1621 | int i; |
1622 | ||
1623 | seq_printf(seq, " [%d/%d] [", conf->raid_disks, | |
11ce99e6 | 1624 | conf->raid_disks - mddev->degraded); |
ddac7c7e N |
1625 | rcu_read_lock(); |
1626 | for (i = 0; i < conf->raid_disks; i++) { | |
3cb03002 | 1627 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
1da177e4 | 1628 | seq_printf(seq, "%s", |
ddac7c7e N |
1629 | rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); |
1630 | } | |
1631 | rcu_read_unlock(); | |
1da177e4 LT |
1632 | seq_printf(seq, "]"); |
1633 | } | |
1634 | ||
849674e4 | 1635 | static void raid1_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1636 | { |
1637 | char b[BDEVNAME_SIZE]; | |
e8096360 | 1638 | struct r1conf *conf = mddev->private; |
423f04d6 | 1639 | unsigned long flags; |
1da177e4 LT |
1640 | |
1641 | /* | |
1642 | * If it is not operational, then we have already marked it as dead | |
1643 | * else if it is the last working disks, ignore the error, let the | |
1644 | * next level up know. | |
1645 | * else mark the drive as failed | |
1646 | */ | |
2e52d449 | 1647 | spin_lock_irqsave(&conf->device_lock, flags); |
b2d444d7 | 1648 | if (test_bit(In_sync, &rdev->flags) |
4044ba58 | 1649 | && (conf->raid_disks - mddev->degraded) == 1) { |
1da177e4 LT |
1650 | /* |
1651 | * Don't fail the drive, act as though we were just a | |
4044ba58 N |
1652 | * normal single drive. |
1653 | * However don't try a recovery from this drive as | |
1654 | * it is very likely to fail. | |
1da177e4 | 1655 | */ |
5389042f | 1656 | conf->recovery_disabled = mddev->recovery_disabled; |
2e52d449 | 1657 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 1658 | return; |
4044ba58 | 1659 | } |
de393cde | 1660 | set_bit(Blocked, &rdev->flags); |
c04be0aa | 1661 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
1da177e4 | 1662 | mddev->degraded++; |
dd00a99e | 1663 | set_bit(Faulty, &rdev->flags); |
dd00a99e N |
1664 | } else |
1665 | set_bit(Faulty, &rdev->flags); | |
423f04d6 | 1666 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2446dba0 N |
1667 | /* |
1668 | * if recovery is running, make sure it aborts. | |
1669 | */ | |
1670 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2953079c SL |
1671 | set_mask_bits(&mddev->sb_flags, 0, |
1672 | BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING)); | |
1d41c216 N |
1673 | pr_crit("md/raid1:%s: Disk failure on %s, disabling device.\n" |
1674 | "md/raid1:%s: Operation continuing on %d devices.\n", | |
1675 | mdname(mddev), bdevname(rdev->bdev, b), | |
1676 | mdname(mddev), conf->raid_disks - mddev->degraded); | |
1da177e4 LT |
1677 | } |
1678 | ||
e8096360 | 1679 | static void print_conf(struct r1conf *conf) |
1da177e4 LT |
1680 | { |
1681 | int i; | |
1da177e4 | 1682 | |
1d41c216 | 1683 | pr_debug("RAID1 conf printout:\n"); |
1da177e4 | 1684 | if (!conf) { |
1d41c216 | 1685 | pr_debug("(!conf)\n"); |
1da177e4 LT |
1686 | return; |
1687 | } | |
1d41c216 N |
1688 | pr_debug(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded, |
1689 | conf->raid_disks); | |
1da177e4 | 1690 | |
ddac7c7e | 1691 | rcu_read_lock(); |
1da177e4 LT |
1692 | for (i = 0; i < conf->raid_disks; i++) { |
1693 | char b[BDEVNAME_SIZE]; | |
3cb03002 | 1694 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
ddac7c7e | 1695 | if (rdev) |
1d41c216 N |
1696 | pr_debug(" disk %d, wo:%d, o:%d, dev:%s\n", |
1697 | i, !test_bit(In_sync, &rdev->flags), | |
1698 | !test_bit(Faulty, &rdev->flags), | |
1699 | bdevname(rdev->bdev,b)); | |
1da177e4 | 1700 | } |
ddac7c7e | 1701 | rcu_read_unlock(); |
1da177e4 LT |
1702 | } |
1703 | ||
e8096360 | 1704 | static void close_sync(struct r1conf *conf) |
1da177e4 | 1705 | { |
fd76863e | 1706 | wait_all_barriers(conf); |
1707 | allow_all_barriers(conf); | |
1da177e4 LT |
1708 | |
1709 | mempool_destroy(conf->r1buf_pool); | |
1710 | conf->r1buf_pool = NULL; | |
1711 | } | |
1712 | ||
fd01b88c | 1713 | static int raid1_spare_active(struct mddev *mddev) |
1da177e4 LT |
1714 | { |
1715 | int i; | |
e8096360 | 1716 | struct r1conf *conf = mddev->private; |
6b965620 N |
1717 | int count = 0; |
1718 | unsigned long flags; | |
1da177e4 LT |
1719 | |
1720 | /* | |
f72ffdd6 | 1721 | * Find all failed disks within the RAID1 configuration |
ddac7c7e N |
1722 | * and mark them readable. |
1723 | * Called under mddev lock, so rcu protection not needed. | |
423f04d6 N |
1724 | * device_lock used to avoid races with raid1_end_read_request |
1725 | * which expects 'In_sync' flags and ->degraded to be consistent. | |
1da177e4 | 1726 | */ |
423f04d6 | 1727 | spin_lock_irqsave(&conf->device_lock, flags); |
1da177e4 | 1728 | for (i = 0; i < conf->raid_disks; i++) { |
3cb03002 | 1729 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
8c7a2c2b N |
1730 | struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev; |
1731 | if (repl | |
1aee41f6 | 1732 | && !test_bit(Candidate, &repl->flags) |
8c7a2c2b N |
1733 | && repl->recovery_offset == MaxSector |
1734 | && !test_bit(Faulty, &repl->flags) | |
1735 | && !test_and_set_bit(In_sync, &repl->flags)) { | |
1736 | /* replacement has just become active */ | |
1737 | if (!rdev || | |
1738 | !test_and_clear_bit(In_sync, &rdev->flags)) | |
1739 | count++; | |
1740 | if (rdev) { | |
1741 | /* Replaced device not technically | |
1742 | * faulty, but we need to be sure | |
1743 | * it gets removed and never re-added | |
1744 | */ | |
1745 | set_bit(Faulty, &rdev->flags); | |
1746 | sysfs_notify_dirent_safe( | |
1747 | rdev->sysfs_state); | |
1748 | } | |
1749 | } | |
ddac7c7e | 1750 | if (rdev |
61e4947c | 1751 | && rdev->recovery_offset == MaxSector |
ddac7c7e | 1752 | && !test_bit(Faulty, &rdev->flags) |
c04be0aa | 1753 | && !test_and_set_bit(In_sync, &rdev->flags)) { |
6b965620 | 1754 | count++; |
654e8b5a | 1755 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
1756 | } |
1757 | } | |
6b965620 N |
1758 | mddev->degraded -= count; |
1759 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 LT |
1760 | |
1761 | print_conf(conf); | |
6b965620 | 1762 | return count; |
1da177e4 LT |
1763 | } |
1764 | ||
fd01b88c | 1765 | static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1766 | { |
e8096360 | 1767 | struct r1conf *conf = mddev->private; |
199050ea | 1768 | int err = -EEXIST; |
41158c7e | 1769 | int mirror = 0; |
0eaf822c | 1770 | struct raid1_info *p; |
6c2fce2e | 1771 | int first = 0; |
30194636 | 1772 | int last = conf->raid_disks - 1; |
1da177e4 | 1773 | |
5389042f N |
1774 | if (mddev->recovery_disabled == conf->recovery_disabled) |
1775 | return -EBUSY; | |
1776 | ||
1501efad DW |
1777 | if (md_integrity_add_rdev(rdev, mddev)) |
1778 | return -ENXIO; | |
1779 | ||
6c2fce2e NB |
1780 | if (rdev->raid_disk >= 0) |
1781 | first = last = rdev->raid_disk; | |
1782 | ||
70bcecdb GR |
1783 | /* |
1784 | * find the disk ... but prefer rdev->saved_raid_disk | |
1785 | * if possible. | |
1786 | */ | |
1787 | if (rdev->saved_raid_disk >= 0 && | |
1788 | rdev->saved_raid_disk >= first && | |
1789 | conf->mirrors[rdev->saved_raid_disk].rdev == NULL) | |
1790 | first = last = rdev->saved_raid_disk; | |
1791 | ||
7ef449d1 N |
1792 | for (mirror = first; mirror <= last; mirror++) { |
1793 | p = conf->mirrors+mirror; | |
1794 | if (!p->rdev) { | |
1da177e4 | 1795 | |
9092c02d JB |
1796 | if (mddev->gendisk) |
1797 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
1798 | rdev->data_offset << 9); | |
1da177e4 LT |
1799 | |
1800 | p->head_position = 0; | |
1801 | rdev->raid_disk = mirror; | |
199050ea | 1802 | err = 0; |
6aea114a N |
1803 | /* As all devices are equivalent, we don't need a full recovery |
1804 | * if this was recently any drive of the array | |
1805 | */ | |
1806 | if (rdev->saved_raid_disk < 0) | |
41158c7e | 1807 | conf->fullsync = 1; |
d6065f7b | 1808 | rcu_assign_pointer(p->rdev, rdev); |
1da177e4 LT |
1809 | break; |
1810 | } | |
7ef449d1 N |
1811 | if (test_bit(WantReplacement, &p->rdev->flags) && |
1812 | p[conf->raid_disks].rdev == NULL) { | |
1813 | /* Add this device as a replacement */ | |
1814 | clear_bit(In_sync, &rdev->flags); | |
1815 | set_bit(Replacement, &rdev->flags); | |
1816 | rdev->raid_disk = mirror; | |
1817 | err = 0; | |
1818 | conf->fullsync = 1; | |
1819 | rcu_assign_pointer(p[conf->raid_disks].rdev, rdev); | |
1820 | break; | |
1821 | } | |
1822 | } | |
9092c02d | 1823 | if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev))) |
2ff8cc2c | 1824 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue); |
1da177e4 | 1825 | print_conf(conf); |
199050ea | 1826 | return err; |
1da177e4 LT |
1827 | } |
1828 | ||
b8321b68 | 1829 | static int raid1_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1830 | { |
e8096360 | 1831 | struct r1conf *conf = mddev->private; |
1da177e4 | 1832 | int err = 0; |
b8321b68 | 1833 | int number = rdev->raid_disk; |
0eaf822c | 1834 | struct raid1_info *p = conf->mirrors + number; |
1da177e4 | 1835 | |
b014f14c N |
1836 | if (rdev != p->rdev) |
1837 | p = conf->mirrors + conf->raid_disks + number; | |
1838 | ||
1da177e4 | 1839 | print_conf(conf); |
b8321b68 | 1840 | if (rdev == p->rdev) { |
b2d444d7 | 1841 | if (test_bit(In_sync, &rdev->flags) || |
1da177e4 LT |
1842 | atomic_read(&rdev->nr_pending)) { |
1843 | err = -EBUSY; | |
1844 | goto abort; | |
1845 | } | |
046abeed | 1846 | /* Only remove non-faulty devices if recovery |
dfc70645 N |
1847 | * is not possible. |
1848 | */ | |
1849 | if (!test_bit(Faulty, &rdev->flags) && | |
5389042f | 1850 | mddev->recovery_disabled != conf->recovery_disabled && |
dfc70645 N |
1851 | mddev->degraded < conf->raid_disks) { |
1852 | err = -EBUSY; | |
1853 | goto abort; | |
1854 | } | |
1da177e4 | 1855 | p->rdev = NULL; |
d787be40 N |
1856 | if (!test_bit(RemoveSynchronized, &rdev->flags)) { |
1857 | synchronize_rcu(); | |
1858 | if (atomic_read(&rdev->nr_pending)) { | |
1859 | /* lost the race, try later */ | |
1860 | err = -EBUSY; | |
1861 | p->rdev = rdev; | |
1862 | goto abort; | |
1863 | } | |
1864 | } | |
1865 | if (conf->mirrors[conf->raid_disks + number].rdev) { | |
8c7a2c2b N |
1866 | /* We just removed a device that is being replaced. |
1867 | * Move down the replacement. We drain all IO before | |
1868 | * doing this to avoid confusion. | |
1869 | */ | |
1870 | struct md_rdev *repl = | |
1871 | conf->mirrors[conf->raid_disks + number].rdev; | |
e2d59925 | 1872 | freeze_array(conf, 0); |
8c7a2c2b N |
1873 | clear_bit(Replacement, &repl->flags); |
1874 | p->rdev = repl; | |
1875 | conf->mirrors[conf->raid_disks + number].rdev = NULL; | |
e2d59925 | 1876 | unfreeze_array(conf); |
8c7a2c2b N |
1877 | clear_bit(WantReplacement, &rdev->flags); |
1878 | } else | |
b014f14c | 1879 | clear_bit(WantReplacement, &rdev->flags); |
a91a2785 | 1880 | err = md_integrity_register(mddev); |
1da177e4 LT |
1881 | } |
1882 | abort: | |
1883 | ||
1884 | print_conf(conf); | |
1885 | return err; | |
1886 | } | |
1887 | ||
4246a0b6 | 1888 | static void end_sync_read(struct bio *bio) |
1da177e4 | 1889 | { |
9f2c9d12 | 1890 | struct r1bio *r1_bio = bio->bi_private; |
1da177e4 | 1891 | |
0fc280f6 | 1892 | update_head_pos(r1_bio->read_disk, r1_bio); |
ba3ae3be | 1893 | |
1da177e4 LT |
1894 | /* |
1895 | * we have read a block, now it needs to be re-written, | |
1896 | * or re-read if the read failed. | |
1897 | * We don't do much here, just schedule handling by raid1d | |
1898 | */ | |
4246a0b6 | 1899 | if (!bio->bi_error) |
1da177e4 | 1900 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
d11c171e N |
1901 | |
1902 | if (atomic_dec_and_test(&r1_bio->remaining)) | |
1903 | reschedule_retry(r1_bio); | |
1da177e4 LT |
1904 | } |
1905 | ||
4246a0b6 | 1906 | static void end_sync_write(struct bio *bio) |
1da177e4 | 1907 | { |
4246a0b6 | 1908 | int uptodate = !bio->bi_error; |
9f2c9d12 | 1909 | struct r1bio *r1_bio = bio->bi_private; |
fd01b88c | 1910 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 1911 | struct r1conf *conf = mddev->private; |
4367af55 N |
1912 | sector_t first_bad; |
1913 | int bad_sectors; | |
854abd75 | 1914 | struct md_rdev *rdev = conf->mirrors[find_bio_disk(r1_bio, bio)].rdev; |
ba3ae3be | 1915 | |
6b1117d5 | 1916 | if (!uptodate) { |
57dab0bd | 1917 | sector_t sync_blocks = 0; |
6b1117d5 N |
1918 | sector_t s = r1_bio->sector; |
1919 | long sectors_to_go = r1_bio->sectors; | |
1920 | /* make sure these bits doesn't get cleared. */ | |
1921 | do { | |
5e3db645 | 1922 | bitmap_end_sync(mddev->bitmap, s, |
6b1117d5 N |
1923 | &sync_blocks, 1); |
1924 | s += sync_blocks; | |
1925 | sectors_to_go -= sync_blocks; | |
1926 | } while (sectors_to_go > 0); | |
854abd75 N |
1927 | set_bit(WriteErrorSeen, &rdev->flags); |
1928 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) | |
19d67169 N |
1929 | set_bit(MD_RECOVERY_NEEDED, & |
1930 | mddev->recovery); | |
d8f05d29 | 1931 | set_bit(R1BIO_WriteError, &r1_bio->state); |
854abd75 | 1932 | } else if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors, |
3a9f28a5 N |
1933 | &first_bad, &bad_sectors) && |
1934 | !is_badblock(conf->mirrors[r1_bio->read_disk].rdev, | |
1935 | r1_bio->sector, | |
1936 | r1_bio->sectors, | |
1937 | &first_bad, &bad_sectors) | |
1938 | ) | |
4367af55 | 1939 | set_bit(R1BIO_MadeGood, &r1_bio->state); |
e3b9703e | 1940 | |
1da177e4 | 1941 | if (atomic_dec_and_test(&r1_bio->remaining)) { |
4367af55 | 1942 | int s = r1_bio->sectors; |
d8f05d29 N |
1943 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
1944 | test_bit(R1BIO_WriteError, &r1_bio->state)) | |
4367af55 N |
1945 | reschedule_retry(r1_bio); |
1946 | else { | |
1947 | put_buf(r1_bio); | |
1948 | md_done_sync(mddev, s, uptodate); | |
1949 | } | |
1da177e4 | 1950 | } |
1da177e4 LT |
1951 | } |
1952 | ||
3cb03002 | 1953 | static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector, |
d8f05d29 N |
1954 | int sectors, struct page *page, int rw) |
1955 | { | |
796a5cf0 | 1956 | if (sync_page_io(rdev, sector, sectors << 9, page, rw, 0, false)) |
d8f05d29 N |
1957 | /* success */ |
1958 | return 1; | |
19d67169 | 1959 | if (rw == WRITE) { |
d8f05d29 | 1960 | set_bit(WriteErrorSeen, &rdev->flags); |
19d67169 N |
1961 | if (!test_and_set_bit(WantReplacement, |
1962 | &rdev->flags)) | |
1963 | set_bit(MD_RECOVERY_NEEDED, & | |
1964 | rdev->mddev->recovery); | |
1965 | } | |
d8f05d29 N |
1966 | /* need to record an error - either for the block or the device */ |
1967 | if (!rdev_set_badblocks(rdev, sector, sectors, 0)) | |
1968 | md_error(rdev->mddev, rdev); | |
1969 | return 0; | |
1970 | } | |
1971 | ||
9f2c9d12 | 1972 | static int fix_sync_read_error(struct r1bio *r1_bio) |
1da177e4 | 1973 | { |
a68e5870 N |
1974 | /* Try some synchronous reads of other devices to get |
1975 | * good data, much like with normal read errors. Only | |
1976 | * read into the pages we already have so we don't | |
1977 | * need to re-issue the read request. | |
1978 | * We don't need to freeze the array, because being in an | |
1979 | * active sync request, there is no normal IO, and | |
1980 | * no overlapping syncs. | |
06f60385 N |
1981 | * We don't need to check is_badblock() again as we |
1982 | * made sure that anything with a bad block in range | |
1983 | * will have bi_end_io clear. | |
a68e5870 | 1984 | */ |
fd01b88c | 1985 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 1986 | struct r1conf *conf = mddev->private; |
a68e5870 N |
1987 | struct bio *bio = r1_bio->bios[r1_bio->read_disk]; |
1988 | sector_t sect = r1_bio->sector; | |
1989 | int sectors = r1_bio->sectors; | |
1990 | int idx = 0; | |
2e52d449 N |
1991 | struct md_rdev *rdev; |
1992 | ||
1993 | rdev = conf->mirrors[r1_bio->read_disk].rdev; | |
1994 | if (test_bit(FailFast, &rdev->flags)) { | |
1995 | /* Don't try recovering from here - just fail it | |
1996 | * ... unless it is the last working device of course */ | |
1997 | md_error(mddev, rdev); | |
1998 | if (test_bit(Faulty, &rdev->flags)) | |
1999 | /* Don't try to read from here, but make sure | |
2000 | * put_buf does it's thing | |
2001 | */ | |
2002 | bio->bi_end_io = end_sync_write; | |
2003 | } | |
a68e5870 N |
2004 | |
2005 | while(sectors) { | |
2006 | int s = sectors; | |
2007 | int d = r1_bio->read_disk; | |
2008 | int success = 0; | |
78d7f5f7 | 2009 | int start; |
a68e5870 N |
2010 | |
2011 | if (s > (PAGE_SIZE>>9)) | |
2012 | s = PAGE_SIZE >> 9; | |
2013 | do { | |
2014 | if (r1_bio->bios[d]->bi_end_io == end_sync_read) { | |
2015 | /* No rcu protection needed here devices | |
2016 | * can only be removed when no resync is | |
2017 | * active, and resync is currently active | |
2018 | */ | |
2019 | rdev = conf->mirrors[d].rdev; | |
9d3d8011 | 2020 | if (sync_page_io(rdev, sect, s<<9, |
a68e5870 | 2021 | bio->bi_io_vec[idx].bv_page, |
796a5cf0 | 2022 | REQ_OP_READ, 0, false)) { |
a68e5870 N |
2023 | success = 1; |
2024 | break; | |
2025 | } | |
2026 | } | |
2027 | d++; | |
8f19ccb2 | 2028 | if (d == conf->raid_disks * 2) |
a68e5870 N |
2029 | d = 0; |
2030 | } while (!success && d != r1_bio->read_disk); | |
2031 | ||
78d7f5f7 | 2032 | if (!success) { |
a68e5870 | 2033 | char b[BDEVNAME_SIZE]; |
3a9f28a5 N |
2034 | int abort = 0; |
2035 | /* Cannot read from anywhere, this block is lost. | |
2036 | * Record a bad block on each device. If that doesn't | |
2037 | * work just disable and interrupt the recovery. | |
2038 | * Don't fail devices as that won't really help. | |
2039 | */ | |
1d41c216 N |
2040 | pr_crit_ratelimited("md/raid1:%s: %s: unrecoverable I/O read error for block %llu\n", |
2041 | mdname(mddev), | |
2042 | bdevname(bio->bi_bdev, b), | |
2043 | (unsigned long long)r1_bio->sector); | |
8f19ccb2 | 2044 | for (d = 0; d < conf->raid_disks * 2; d++) { |
3a9f28a5 N |
2045 | rdev = conf->mirrors[d].rdev; |
2046 | if (!rdev || test_bit(Faulty, &rdev->flags)) | |
2047 | continue; | |
2048 | if (!rdev_set_badblocks(rdev, sect, s, 0)) | |
2049 | abort = 1; | |
2050 | } | |
2051 | if (abort) { | |
d890fa2b N |
2052 | conf->recovery_disabled = |
2053 | mddev->recovery_disabled; | |
3a9f28a5 N |
2054 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
2055 | md_done_sync(mddev, r1_bio->sectors, 0); | |
2056 | put_buf(r1_bio); | |
2057 | return 0; | |
2058 | } | |
2059 | /* Try next page */ | |
2060 | sectors -= s; | |
2061 | sect += s; | |
2062 | idx++; | |
2063 | continue; | |
d11c171e | 2064 | } |
78d7f5f7 N |
2065 | |
2066 | start = d; | |
2067 | /* write it back and re-read */ | |
2068 | while (d != r1_bio->read_disk) { | |
2069 | if (d == 0) | |
8f19ccb2 | 2070 | d = conf->raid_disks * 2; |
78d7f5f7 N |
2071 | d--; |
2072 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) | |
2073 | continue; | |
2074 | rdev = conf->mirrors[d].rdev; | |
d8f05d29 N |
2075 | if (r1_sync_page_io(rdev, sect, s, |
2076 | bio->bi_io_vec[idx].bv_page, | |
2077 | WRITE) == 0) { | |
78d7f5f7 N |
2078 | r1_bio->bios[d]->bi_end_io = NULL; |
2079 | rdev_dec_pending(rdev, mddev); | |
9d3d8011 | 2080 | } |
78d7f5f7 N |
2081 | } |
2082 | d = start; | |
2083 | while (d != r1_bio->read_disk) { | |
2084 | if (d == 0) | |
8f19ccb2 | 2085 | d = conf->raid_disks * 2; |
78d7f5f7 N |
2086 | d--; |
2087 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) | |
2088 | continue; | |
2089 | rdev = conf->mirrors[d].rdev; | |
d8f05d29 N |
2090 | if (r1_sync_page_io(rdev, sect, s, |
2091 | bio->bi_io_vec[idx].bv_page, | |
2092 | READ) != 0) | |
9d3d8011 | 2093 | atomic_add(s, &rdev->corrected_errors); |
78d7f5f7 | 2094 | } |
a68e5870 N |
2095 | sectors -= s; |
2096 | sect += s; | |
2097 | idx ++; | |
2098 | } | |
78d7f5f7 | 2099 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
4246a0b6 | 2100 | bio->bi_error = 0; |
a68e5870 N |
2101 | return 1; |
2102 | } | |
2103 | ||
c95e6385 | 2104 | static void process_checks(struct r1bio *r1_bio) |
a68e5870 N |
2105 | { |
2106 | /* We have read all readable devices. If we haven't | |
2107 | * got the block, then there is no hope left. | |
2108 | * If we have, then we want to do a comparison | |
2109 | * and skip the write if everything is the same. | |
2110 | * If any blocks failed to read, then we need to | |
2111 | * attempt an over-write | |
2112 | */ | |
fd01b88c | 2113 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 2114 | struct r1conf *conf = mddev->private; |
a68e5870 N |
2115 | int primary; |
2116 | int i; | |
f4380a91 | 2117 | int vcnt; |
a68e5870 | 2118 | |
30bc9b53 N |
2119 | /* Fix variable parts of all bios */ |
2120 | vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9); | |
2121 | for (i = 0; i < conf->raid_disks * 2; i++) { | |
2122 | int j; | |
2123 | int size; | |
4246a0b6 | 2124 | int error; |
30bc9b53 N |
2125 | struct bio *b = r1_bio->bios[i]; |
2126 | if (b->bi_end_io != end_sync_read) | |
2127 | continue; | |
4246a0b6 CH |
2128 | /* fixup the bio for reuse, but preserve errno */ |
2129 | error = b->bi_error; | |
30bc9b53 | 2130 | bio_reset(b); |
4246a0b6 | 2131 | b->bi_error = error; |
30bc9b53 | 2132 | b->bi_vcnt = vcnt; |
4f024f37 KO |
2133 | b->bi_iter.bi_size = r1_bio->sectors << 9; |
2134 | b->bi_iter.bi_sector = r1_bio->sector + | |
30bc9b53 N |
2135 | conf->mirrors[i].rdev->data_offset; |
2136 | b->bi_bdev = conf->mirrors[i].rdev->bdev; | |
2137 | b->bi_end_io = end_sync_read; | |
2138 | b->bi_private = r1_bio; | |
2139 | ||
4f024f37 | 2140 | size = b->bi_iter.bi_size; |
30bc9b53 N |
2141 | for (j = 0; j < vcnt ; j++) { |
2142 | struct bio_vec *bi; | |
2143 | bi = &b->bi_io_vec[j]; | |
2144 | bi->bv_offset = 0; | |
2145 | if (size > PAGE_SIZE) | |
2146 | bi->bv_len = PAGE_SIZE; | |
2147 | else | |
2148 | bi->bv_len = size; | |
2149 | size -= PAGE_SIZE; | |
2150 | } | |
2151 | } | |
8f19ccb2 | 2152 | for (primary = 0; primary < conf->raid_disks * 2; primary++) |
a68e5870 | 2153 | if (r1_bio->bios[primary]->bi_end_io == end_sync_read && |
4246a0b6 | 2154 | !r1_bio->bios[primary]->bi_error) { |
a68e5870 N |
2155 | r1_bio->bios[primary]->bi_end_io = NULL; |
2156 | rdev_dec_pending(conf->mirrors[primary].rdev, mddev); | |
2157 | break; | |
2158 | } | |
2159 | r1_bio->read_disk = primary; | |
8f19ccb2 | 2160 | for (i = 0; i < conf->raid_disks * 2; i++) { |
78d7f5f7 | 2161 | int j; |
78d7f5f7 N |
2162 | struct bio *pbio = r1_bio->bios[primary]; |
2163 | struct bio *sbio = r1_bio->bios[i]; | |
4246a0b6 | 2164 | int error = sbio->bi_error; |
a68e5870 | 2165 | |
2aabaa65 | 2166 | if (sbio->bi_end_io != end_sync_read) |
78d7f5f7 | 2167 | continue; |
4246a0b6 CH |
2168 | /* Now we can 'fixup' the error value */ |
2169 | sbio->bi_error = 0; | |
78d7f5f7 | 2170 | |
4246a0b6 | 2171 | if (!error) { |
78d7f5f7 N |
2172 | for (j = vcnt; j-- ; ) { |
2173 | struct page *p, *s; | |
2174 | p = pbio->bi_io_vec[j].bv_page; | |
2175 | s = sbio->bi_io_vec[j].bv_page; | |
2176 | if (memcmp(page_address(p), | |
2177 | page_address(s), | |
5020ad7d | 2178 | sbio->bi_io_vec[j].bv_len)) |
78d7f5f7 | 2179 | break; |
69382e85 | 2180 | } |
78d7f5f7 N |
2181 | } else |
2182 | j = 0; | |
2183 | if (j >= 0) | |
7f7583d4 | 2184 | atomic64_add(r1_bio->sectors, &mddev->resync_mismatches); |
78d7f5f7 | 2185 | if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery) |
4246a0b6 | 2186 | && !error)) { |
78d7f5f7 N |
2187 | /* No need to write to this device. */ |
2188 | sbio->bi_end_io = NULL; | |
2189 | rdev_dec_pending(conf->mirrors[i].rdev, mddev); | |
2190 | continue; | |
2191 | } | |
d3b45c2a KO |
2192 | |
2193 | bio_copy_data(sbio, pbio); | |
78d7f5f7 | 2194 | } |
a68e5870 N |
2195 | } |
2196 | ||
9f2c9d12 | 2197 | static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio) |
a68e5870 | 2198 | { |
e8096360 | 2199 | struct r1conf *conf = mddev->private; |
a68e5870 | 2200 | int i; |
8f19ccb2 | 2201 | int disks = conf->raid_disks * 2; |
a68e5870 N |
2202 | struct bio *bio, *wbio; |
2203 | ||
2204 | bio = r1_bio->bios[r1_bio->read_disk]; | |
2205 | ||
a68e5870 N |
2206 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) |
2207 | /* ouch - failed to read all of that. */ | |
2208 | if (!fix_sync_read_error(r1_bio)) | |
2209 | return; | |
7ca78d57 N |
2210 | |
2211 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
c95e6385 N |
2212 | process_checks(r1_bio); |
2213 | ||
d11c171e N |
2214 | /* |
2215 | * schedule writes | |
2216 | */ | |
1da177e4 LT |
2217 | atomic_set(&r1_bio->remaining, 1); |
2218 | for (i = 0; i < disks ; i++) { | |
2219 | wbio = r1_bio->bios[i]; | |
3e198f78 N |
2220 | if (wbio->bi_end_io == NULL || |
2221 | (wbio->bi_end_io == end_sync_read && | |
2222 | (i == r1_bio->read_disk || | |
2223 | !test_bit(MD_RECOVERY_SYNC, &mddev->recovery)))) | |
1da177e4 LT |
2224 | continue; |
2225 | ||
796a5cf0 | 2226 | bio_set_op_attrs(wbio, REQ_OP_WRITE, 0); |
212e7eb7 N |
2227 | if (test_bit(FailFast, &conf->mirrors[i].rdev->flags)) |
2228 | wbio->bi_opf |= MD_FAILFAST; | |
2229 | ||
3e198f78 | 2230 | wbio->bi_end_io = end_sync_write; |
1da177e4 | 2231 | atomic_inc(&r1_bio->remaining); |
aa8b57aa | 2232 | md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio)); |
191ea9b2 | 2233 | |
1da177e4 LT |
2234 | generic_make_request(wbio); |
2235 | } | |
2236 | ||
2237 | if (atomic_dec_and_test(&r1_bio->remaining)) { | |
191ea9b2 | 2238 | /* if we're here, all write(s) have completed, so clean up */ |
58e94ae1 N |
2239 | int s = r1_bio->sectors; |
2240 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || | |
2241 | test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2242 | reschedule_retry(r1_bio); | |
2243 | else { | |
2244 | put_buf(r1_bio); | |
2245 | md_done_sync(mddev, s, 1); | |
2246 | } | |
1da177e4 LT |
2247 | } |
2248 | } | |
2249 | ||
2250 | /* | |
2251 | * This is a kernel thread which: | |
2252 | * | |
2253 | * 1. Retries failed read operations on working mirrors. | |
2254 | * 2. Updates the raid superblock when problems encounter. | |
d2eb35ac | 2255 | * 3. Performs writes following reads for array synchronising. |
1da177e4 LT |
2256 | */ |
2257 | ||
e8096360 | 2258 | static void fix_read_error(struct r1conf *conf, int read_disk, |
867868fb N |
2259 | sector_t sect, int sectors) |
2260 | { | |
fd01b88c | 2261 | struct mddev *mddev = conf->mddev; |
867868fb N |
2262 | while(sectors) { |
2263 | int s = sectors; | |
2264 | int d = read_disk; | |
2265 | int success = 0; | |
2266 | int start; | |
3cb03002 | 2267 | struct md_rdev *rdev; |
867868fb N |
2268 | |
2269 | if (s > (PAGE_SIZE>>9)) | |
2270 | s = PAGE_SIZE >> 9; | |
2271 | ||
2272 | do { | |
d2eb35ac N |
2273 | sector_t first_bad; |
2274 | int bad_sectors; | |
2275 | ||
707a6a42 N |
2276 | rcu_read_lock(); |
2277 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
867868fb | 2278 | if (rdev && |
da8840a7 | 2279 | (test_bit(In_sync, &rdev->flags) || |
2280 | (!test_bit(Faulty, &rdev->flags) && | |
2281 | rdev->recovery_offset >= sect + s)) && | |
d2eb35ac | 2282 | is_badblock(rdev, sect, s, |
707a6a42 N |
2283 | &first_bad, &bad_sectors) == 0) { |
2284 | atomic_inc(&rdev->nr_pending); | |
2285 | rcu_read_unlock(); | |
2286 | if (sync_page_io(rdev, sect, s<<9, | |
796a5cf0 | 2287 | conf->tmppage, REQ_OP_READ, 0, false)) |
707a6a42 N |
2288 | success = 1; |
2289 | rdev_dec_pending(rdev, mddev); | |
2290 | if (success) | |
2291 | break; | |
2292 | } else | |
2293 | rcu_read_unlock(); | |
2294 | d++; | |
2295 | if (d == conf->raid_disks * 2) | |
2296 | d = 0; | |
867868fb N |
2297 | } while (!success && d != read_disk); |
2298 | ||
2299 | if (!success) { | |
d8f05d29 | 2300 | /* Cannot read from anywhere - mark it bad */ |
3cb03002 | 2301 | struct md_rdev *rdev = conf->mirrors[read_disk].rdev; |
d8f05d29 N |
2302 | if (!rdev_set_badblocks(rdev, sect, s, 0)) |
2303 | md_error(mddev, rdev); | |
867868fb N |
2304 | break; |
2305 | } | |
2306 | /* write it back and re-read */ | |
2307 | start = d; | |
2308 | while (d != read_disk) { | |
2309 | if (d==0) | |
8f19ccb2 | 2310 | d = conf->raid_disks * 2; |
867868fb | 2311 | d--; |
707a6a42 N |
2312 | rcu_read_lock(); |
2313 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
867868fb | 2314 | if (rdev && |
707a6a42 N |
2315 | !test_bit(Faulty, &rdev->flags)) { |
2316 | atomic_inc(&rdev->nr_pending); | |
2317 | rcu_read_unlock(); | |
d8f05d29 N |
2318 | r1_sync_page_io(rdev, sect, s, |
2319 | conf->tmppage, WRITE); | |
707a6a42 N |
2320 | rdev_dec_pending(rdev, mddev); |
2321 | } else | |
2322 | rcu_read_unlock(); | |
867868fb N |
2323 | } |
2324 | d = start; | |
2325 | while (d != read_disk) { | |
2326 | char b[BDEVNAME_SIZE]; | |
2327 | if (d==0) | |
8f19ccb2 | 2328 | d = conf->raid_disks * 2; |
867868fb | 2329 | d--; |
707a6a42 N |
2330 | rcu_read_lock(); |
2331 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
867868fb | 2332 | if (rdev && |
b8cb6b4c | 2333 | !test_bit(Faulty, &rdev->flags)) { |
707a6a42 N |
2334 | atomic_inc(&rdev->nr_pending); |
2335 | rcu_read_unlock(); | |
d8f05d29 N |
2336 | if (r1_sync_page_io(rdev, sect, s, |
2337 | conf->tmppage, READ)) { | |
867868fb | 2338 | atomic_add(s, &rdev->corrected_errors); |
1d41c216 N |
2339 | pr_info("md/raid1:%s: read error corrected (%d sectors at %llu on %s)\n", |
2340 | mdname(mddev), s, | |
2341 | (unsigned long long)(sect + | |
2342 | rdev->data_offset), | |
2343 | bdevname(rdev->bdev, b)); | |
867868fb | 2344 | } |
707a6a42 N |
2345 | rdev_dec_pending(rdev, mddev); |
2346 | } else | |
2347 | rcu_read_unlock(); | |
867868fb N |
2348 | } |
2349 | sectors -= s; | |
2350 | sect += s; | |
2351 | } | |
2352 | } | |
2353 | ||
9f2c9d12 | 2354 | static int narrow_write_error(struct r1bio *r1_bio, int i) |
cd5ff9a1 | 2355 | { |
fd01b88c | 2356 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 2357 | struct r1conf *conf = mddev->private; |
3cb03002 | 2358 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
cd5ff9a1 N |
2359 | |
2360 | /* bio has the data to be written to device 'i' where | |
2361 | * we just recently had a write error. | |
2362 | * We repeatedly clone the bio and trim down to one block, | |
2363 | * then try the write. Where the write fails we record | |
2364 | * a bad block. | |
2365 | * It is conceivable that the bio doesn't exactly align with | |
2366 | * blocks. We must handle this somehow. | |
2367 | * | |
2368 | * We currently own a reference on the rdev. | |
2369 | */ | |
2370 | ||
2371 | int block_sectors; | |
2372 | sector_t sector; | |
2373 | int sectors; | |
2374 | int sect_to_write = r1_bio->sectors; | |
2375 | int ok = 1; | |
2376 | ||
2377 | if (rdev->badblocks.shift < 0) | |
2378 | return 0; | |
2379 | ||
ab713cdc ND |
2380 | block_sectors = roundup(1 << rdev->badblocks.shift, |
2381 | bdev_logical_block_size(rdev->bdev) >> 9); | |
cd5ff9a1 N |
2382 | sector = r1_bio->sector; |
2383 | sectors = ((sector + block_sectors) | |
2384 | & ~(sector_t)(block_sectors - 1)) | |
2385 | - sector; | |
2386 | ||
cd5ff9a1 N |
2387 | while (sect_to_write) { |
2388 | struct bio *wbio; | |
2389 | if (sectors > sect_to_write) | |
2390 | sectors = sect_to_write; | |
2391 | /* Write at 'sector' for 'sectors'*/ | |
2392 | ||
b783863f KO |
2393 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { |
2394 | unsigned vcnt = r1_bio->behind_page_count; | |
2395 | struct bio_vec *vec = r1_bio->behind_bvecs; | |
2396 | ||
2397 | while (!vec->bv_page) { | |
2398 | vec++; | |
2399 | vcnt--; | |
2400 | } | |
2401 | ||
2402 | wbio = bio_alloc_mddev(GFP_NOIO, vcnt, mddev); | |
2403 | memcpy(wbio->bi_io_vec, vec, vcnt * sizeof(struct bio_vec)); | |
2404 | ||
2405 | wbio->bi_vcnt = vcnt; | |
2406 | } else { | |
d7a10308 ML |
2407 | wbio = bio_clone_fast(r1_bio->master_bio, GFP_NOIO, |
2408 | mddev->bio_set); | |
b783863f KO |
2409 | } |
2410 | ||
796a5cf0 | 2411 | bio_set_op_attrs(wbio, REQ_OP_WRITE, 0); |
4f024f37 KO |
2412 | wbio->bi_iter.bi_sector = r1_bio->sector; |
2413 | wbio->bi_iter.bi_size = r1_bio->sectors << 9; | |
cd5ff9a1 | 2414 | |
6678d83f | 2415 | bio_trim(wbio, sector - r1_bio->sector, sectors); |
4f024f37 | 2416 | wbio->bi_iter.bi_sector += rdev->data_offset; |
cd5ff9a1 | 2417 | wbio->bi_bdev = rdev->bdev; |
4e49ea4a MC |
2418 | |
2419 | if (submit_bio_wait(wbio) < 0) | |
cd5ff9a1 N |
2420 | /* failure! */ |
2421 | ok = rdev_set_badblocks(rdev, sector, | |
2422 | sectors, 0) | |
2423 | && ok; | |
2424 | ||
2425 | bio_put(wbio); | |
2426 | sect_to_write -= sectors; | |
2427 | sector += sectors; | |
2428 | sectors = block_sectors; | |
2429 | } | |
2430 | return ok; | |
2431 | } | |
2432 | ||
e8096360 | 2433 | static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce N |
2434 | { |
2435 | int m; | |
2436 | int s = r1_bio->sectors; | |
8f19ccb2 | 2437 | for (m = 0; m < conf->raid_disks * 2 ; m++) { |
3cb03002 | 2438 | struct md_rdev *rdev = conf->mirrors[m].rdev; |
62096bce N |
2439 | struct bio *bio = r1_bio->bios[m]; |
2440 | if (bio->bi_end_io == NULL) | |
2441 | continue; | |
4246a0b6 | 2442 | if (!bio->bi_error && |
62096bce | 2443 | test_bit(R1BIO_MadeGood, &r1_bio->state)) { |
c6563a8c | 2444 | rdev_clear_badblocks(rdev, r1_bio->sector, s, 0); |
62096bce | 2445 | } |
4246a0b6 | 2446 | if (bio->bi_error && |
62096bce N |
2447 | test_bit(R1BIO_WriteError, &r1_bio->state)) { |
2448 | if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0)) | |
2449 | md_error(conf->mddev, rdev); | |
2450 | } | |
2451 | } | |
2452 | put_buf(r1_bio); | |
2453 | md_done_sync(conf->mddev, s, 1); | |
2454 | } | |
2455 | ||
e8096360 | 2456 | static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce | 2457 | { |
fd76863e | 2458 | int m, idx; |
55ce74d4 | 2459 | bool fail = false; |
fd76863e | 2460 | |
8f19ccb2 | 2461 | for (m = 0; m < conf->raid_disks * 2 ; m++) |
62096bce | 2462 | if (r1_bio->bios[m] == IO_MADE_GOOD) { |
3cb03002 | 2463 | struct md_rdev *rdev = conf->mirrors[m].rdev; |
62096bce N |
2464 | rdev_clear_badblocks(rdev, |
2465 | r1_bio->sector, | |
c6563a8c | 2466 | r1_bio->sectors, 0); |
62096bce N |
2467 | rdev_dec_pending(rdev, conf->mddev); |
2468 | } else if (r1_bio->bios[m] != NULL) { | |
2469 | /* This drive got a write error. We need to | |
2470 | * narrow down and record precise write | |
2471 | * errors. | |
2472 | */ | |
55ce74d4 | 2473 | fail = true; |
62096bce N |
2474 | if (!narrow_write_error(r1_bio, m)) { |
2475 | md_error(conf->mddev, | |
2476 | conf->mirrors[m].rdev); | |
2477 | /* an I/O failed, we can't clear the bitmap */ | |
2478 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
2479 | } | |
2480 | rdev_dec_pending(conf->mirrors[m].rdev, | |
2481 | conf->mddev); | |
2482 | } | |
55ce74d4 N |
2483 | if (fail) { |
2484 | spin_lock_irq(&conf->device_lock); | |
2485 | list_add(&r1_bio->retry_list, &conf->bio_end_io_list); | |
fd76863e | 2486 | idx = sector_to_idx(r1_bio->sector); |
824e47da | 2487 | atomic_inc(&conf->nr_queued[idx]); |
55ce74d4 | 2488 | spin_unlock_irq(&conf->device_lock); |
824e47da | 2489 | /* |
2490 | * In case freeze_array() is waiting for condition | |
2491 | * get_unqueued_pending() == extra to be true. | |
2492 | */ | |
2493 | wake_up(&conf->wait_barrier); | |
55ce74d4 | 2494 | md_wakeup_thread(conf->mddev->thread); |
bd8688a1 N |
2495 | } else { |
2496 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2497 | close_write(r1_bio); | |
55ce74d4 | 2498 | raid_end_bio_io(r1_bio); |
bd8688a1 | 2499 | } |
62096bce N |
2500 | } |
2501 | ||
e8096360 | 2502 | static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce N |
2503 | { |
2504 | int disk; | |
2505 | int max_sectors; | |
fd01b88c | 2506 | struct mddev *mddev = conf->mddev; |
62096bce N |
2507 | struct bio *bio; |
2508 | char b[BDEVNAME_SIZE]; | |
3cb03002 | 2509 | struct md_rdev *rdev; |
109e3765 N |
2510 | dev_t bio_dev; |
2511 | sector_t bio_sector; | |
62096bce N |
2512 | |
2513 | clear_bit(R1BIO_ReadError, &r1_bio->state); | |
2514 | /* we got a read error. Maybe the drive is bad. Maybe just | |
2515 | * the block and we can fix it. | |
2516 | * We freeze all other IO, and try reading the block from | |
2517 | * other devices. When we find one, we re-write | |
2518 | * and check it that fixes the read error. | |
2519 | * This is all done synchronously while the array is | |
2520 | * frozen | |
2521 | */ | |
7449f699 TM |
2522 | |
2523 | bio = r1_bio->bios[r1_bio->read_disk]; | |
2524 | bdevname(bio->bi_bdev, b); | |
109e3765 N |
2525 | bio_dev = bio->bi_bdev->bd_dev; |
2526 | bio_sector = conf->mirrors[r1_bio->read_disk].rdev->data_offset + r1_bio->sector; | |
7449f699 TM |
2527 | bio_put(bio); |
2528 | r1_bio->bios[r1_bio->read_disk] = NULL; | |
2529 | ||
2e52d449 N |
2530 | rdev = conf->mirrors[r1_bio->read_disk].rdev; |
2531 | if (mddev->ro == 0 | |
2532 | && !test_bit(FailFast, &rdev->flags)) { | |
e2d59925 | 2533 | freeze_array(conf, 1); |
62096bce N |
2534 | fix_read_error(conf, r1_bio->read_disk, |
2535 | r1_bio->sector, r1_bio->sectors); | |
2536 | unfreeze_array(conf); | |
7449f699 TM |
2537 | } else { |
2538 | r1_bio->bios[r1_bio->read_disk] = IO_BLOCKED; | |
2539 | } | |
2540 | ||
2e52d449 | 2541 | rdev_dec_pending(rdev, conf->mddev); |
62096bce | 2542 | |
62096bce N |
2543 | read_more: |
2544 | disk = read_balance(conf, r1_bio, &max_sectors); | |
2545 | if (disk == -1) { | |
1d41c216 N |
2546 | pr_crit_ratelimited("md/raid1:%s: %s: unrecoverable I/O read error for block %llu\n", |
2547 | mdname(mddev), b, (unsigned long long)r1_bio->sector); | |
62096bce N |
2548 | raid_end_bio_io(r1_bio); |
2549 | } else { | |
2550 | const unsigned long do_sync | |
1eff9d32 | 2551 | = r1_bio->master_bio->bi_opf & REQ_SYNC; |
62096bce | 2552 | r1_bio->read_disk = disk; |
d7a10308 ML |
2553 | bio = bio_clone_fast(r1_bio->master_bio, GFP_NOIO, |
2554 | mddev->bio_set); | |
4f024f37 KO |
2555 | bio_trim(bio, r1_bio->sector - bio->bi_iter.bi_sector, |
2556 | max_sectors); | |
62096bce N |
2557 | r1_bio->bios[r1_bio->read_disk] = bio; |
2558 | rdev = conf->mirrors[disk].rdev; | |
1d41c216 N |
2559 | pr_info_ratelimited("md/raid1:%s: redirecting sector %llu to other mirror: %s\n", |
2560 | mdname(mddev), | |
2561 | (unsigned long long)r1_bio->sector, | |
2562 | bdevname(rdev->bdev, b)); | |
4f024f37 | 2563 | bio->bi_iter.bi_sector = r1_bio->sector + rdev->data_offset; |
62096bce N |
2564 | bio->bi_bdev = rdev->bdev; |
2565 | bio->bi_end_io = raid1_end_read_request; | |
796a5cf0 | 2566 | bio_set_op_attrs(bio, REQ_OP_READ, do_sync); |
2e52d449 N |
2567 | if (test_bit(FailFast, &rdev->flags) && |
2568 | test_bit(R1BIO_FailFast, &r1_bio->state)) | |
2569 | bio->bi_opf |= MD_FAILFAST; | |
62096bce N |
2570 | bio->bi_private = r1_bio; |
2571 | if (max_sectors < r1_bio->sectors) { | |
2572 | /* Drat - have to split this up more */ | |
2573 | struct bio *mbio = r1_bio->master_bio; | |
2574 | int sectors_handled = (r1_bio->sector + max_sectors | |
4f024f37 | 2575 | - mbio->bi_iter.bi_sector); |
62096bce N |
2576 | r1_bio->sectors = max_sectors; |
2577 | spin_lock_irq(&conf->device_lock); | |
2578 | if (mbio->bi_phys_segments == 0) | |
2579 | mbio->bi_phys_segments = 2; | |
2580 | else | |
2581 | mbio->bi_phys_segments++; | |
2582 | spin_unlock_irq(&conf->device_lock); | |
109e3765 N |
2583 | trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), |
2584 | bio, bio_dev, bio_sector); | |
62096bce N |
2585 | generic_make_request(bio); |
2586 | bio = NULL; | |
2587 | ||
fd76863e | 2588 | r1_bio = alloc_r1bio(mddev, mbio, sectors_handled); |
62096bce | 2589 | set_bit(R1BIO_ReadError, &r1_bio->state); |
62096bce N |
2590 | |
2591 | goto read_more; | |
109e3765 N |
2592 | } else { |
2593 | trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), | |
2594 | bio, bio_dev, bio_sector); | |
62096bce | 2595 | generic_make_request(bio); |
109e3765 | 2596 | } |
62096bce N |
2597 | } |
2598 | } | |
2599 | ||
4ed8731d | 2600 | static void raid1d(struct md_thread *thread) |
1da177e4 | 2601 | { |
4ed8731d | 2602 | struct mddev *mddev = thread->mddev; |
9f2c9d12 | 2603 | struct r1bio *r1_bio; |
1da177e4 | 2604 | unsigned long flags; |
e8096360 | 2605 | struct r1conf *conf = mddev->private; |
1da177e4 | 2606 | struct list_head *head = &conf->retry_list; |
e1dfa0a2 | 2607 | struct blk_plug plug; |
fd76863e | 2608 | int idx; |
1da177e4 LT |
2609 | |
2610 | md_check_recovery(mddev); | |
e1dfa0a2 | 2611 | |
55ce74d4 | 2612 | if (!list_empty_careful(&conf->bio_end_io_list) && |
2953079c | 2613 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { |
55ce74d4 N |
2614 | LIST_HEAD(tmp); |
2615 | spin_lock_irqsave(&conf->device_lock, flags); | |
fd76863e | 2616 | if (!test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) |
2617 | list_splice_init(&conf->bio_end_io_list, &tmp); | |
55ce74d4 N |
2618 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2619 | while (!list_empty(&tmp)) { | |
a452744b MP |
2620 | r1_bio = list_first_entry(&tmp, struct r1bio, |
2621 | retry_list); | |
55ce74d4 | 2622 | list_del(&r1_bio->retry_list); |
fd76863e | 2623 | idx = sector_to_idx(r1_bio->sector); |
824e47da | 2624 | atomic_dec(&conf->nr_queued[idx]); |
bd8688a1 N |
2625 | if (mddev->degraded) |
2626 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
2627 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2628 | close_write(r1_bio); | |
55ce74d4 N |
2629 | raid_end_bio_io(r1_bio); |
2630 | } | |
2631 | } | |
2632 | ||
e1dfa0a2 | 2633 | blk_start_plug(&plug); |
1da177e4 | 2634 | for (;;) { |
191ea9b2 | 2635 | |
0021b7bc | 2636 | flush_pending_writes(conf); |
191ea9b2 | 2637 | |
a35e63ef N |
2638 | spin_lock_irqsave(&conf->device_lock, flags); |
2639 | if (list_empty(head)) { | |
2640 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 | 2641 | break; |
a35e63ef | 2642 | } |
9f2c9d12 | 2643 | r1_bio = list_entry(head->prev, struct r1bio, retry_list); |
1da177e4 | 2644 | list_del(head->prev); |
fd76863e | 2645 | idx = sector_to_idx(r1_bio->sector); |
824e47da | 2646 | atomic_dec(&conf->nr_queued[idx]); |
1da177e4 LT |
2647 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2648 | ||
2649 | mddev = r1_bio->mddev; | |
070ec55d | 2650 | conf = mddev->private; |
4367af55 | 2651 | if (test_bit(R1BIO_IsSync, &r1_bio->state)) { |
d8f05d29 | 2652 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
62096bce N |
2653 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
2654 | handle_sync_write_finished(conf, r1_bio); | |
2655 | else | |
4367af55 | 2656 | sync_request_write(mddev, r1_bio); |
cd5ff9a1 | 2657 | } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
62096bce N |
2658 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
2659 | handle_write_finished(conf, r1_bio); | |
2660 | else if (test_bit(R1BIO_ReadError, &r1_bio->state)) | |
2661 | handle_read_error(conf, r1_bio); | |
2662 | else | |
d2eb35ac N |
2663 | /* just a partial read to be scheduled from separate |
2664 | * context | |
2665 | */ | |
2666 | generic_make_request(r1_bio->bios[r1_bio->read_disk]); | |
62096bce | 2667 | |
1d9d5241 | 2668 | cond_resched(); |
2953079c | 2669 | if (mddev->sb_flags & ~(1<<MD_SB_CHANGE_PENDING)) |
de393cde | 2670 | md_check_recovery(mddev); |
1da177e4 | 2671 | } |
e1dfa0a2 | 2672 | blk_finish_plug(&plug); |
1da177e4 LT |
2673 | } |
2674 | ||
e8096360 | 2675 | static int init_resync(struct r1conf *conf) |
1da177e4 LT |
2676 | { |
2677 | int buffs; | |
2678 | ||
2679 | buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; | |
9e77c485 | 2680 | BUG_ON(conf->r1buf_pool); |
1da177e4 LT |
2681 | conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free, |
2682 | conf->poolinfo); | |
2683 | if (!conf->r1buf_pool) | |
2684 | return -ENOMEM; | |
1da177e4 LT |
2685 | return 0; |
2686 | } | |
2687 | ||
2688 | /* | |
2689 | * perform a "sync" on one "block" | |
2690 | * | |
2691 | * We need to make sure that no normal I/O request - particularly write | |
2692 | * requests - conflict with active sync requests. | |
2693 | * | |
2694 | * This is achieved by tracking pending requests and a 'barrier' concept | |
2695 | * that can be installed to exclude normal IO requests. | |
2696 | */ | |
2697 | ||
849674e4 SL |
2698 | static sector_t raid1_sync_request(struct mddev *mddev, sector_t sector_nr, |
2699 | int *skipped) | |
1da177e4 | 2700 | { |
e8096360 | 2701 | struct r1conf *conf = mddev->private; |
9f2c9d12 | 2702 | struct r1bio *r1_bio; |
1da177e4 LT |
2703 | struct bio *bio; |
2704 | sector_t max_sector, nr_sectors; | |
3e198f78 | 2705 | int disk = -1; |
1da177e4 | 2706 | int i; |
3e198f78 N |
2707 | int wonly = -1; |
2708 | int write_targets = 0, read_targets = 0; | |
57dab0bd | 2709 | sector_t sync_blocks; |
e3b9703e | 2710 | int still_degraded = 0; |
06f60385 N |
2711 | int good_sectors = RESYNC_SECTORS; |
2712 | int min_bad = 0; /* number of sectors that are bad in all devices */ | |
fd76863e | 2713 | int idx = sector_to_idx(sector_nr); |
1da177e4 LT |
2714 | |
2715 | if (!conf->r1buf_pool) | |
2716 | if (init_resync(conf)) | |
57afd89f | 2717 | return 0; |
1da177e4 | 2718 | |
58c0fed4 | 2719 | max_sector = mddev->dev_sectors; |
1da177e4 | 2720 | if (sector_nr >= max_sector) { |
191ea9b2 N |
2721 | /* If we aborted, we need to abort the |
2722 | * sync on the 'current' bitmap chunk (there will | |
2723 | * only be one in raid1 resync. | |
2724 | * We can find the current addess in mddev->curr_resync | |
2725 | */ | |
6a806c51 N |
2726 | if (mddev->curr_resync < max_sector) /* aborted */ |
2727 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
191ea9b2 | 2728 | &sync_blocks, 1); |
6a806c51 | 2729 | else /* completed sync */ |
191ea9b2 | 2730 | conf->fullsync = 0; |
6a806c51 N |
2731 | |
2732 | bitmap_close_sync(mddev->bitmap); | |
1da177e4 | 2733 | close_sync(conf); |
c40f341f GR |
2734 | |
2735 | if (mddev_is_clustered(mddev)) { | |
2736 | conf->cluster_sync_low = 0; | |
2737 | conf->cluster_sync_high = 0; | |
c40f341f | 2738 | } |
1da177e4 LT |
2739 | return 0; |
2740 | } | |
2741 | ||
07d84d10 N |
2742 | if (mddev->bitmap == NULL && |
2743 | mddev->recovery_cp == MaxSector && | |
6394cca5 | 2744 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
07d84d10 N |
2745 | conf->fullsync == 0) { |
2746 | *skipped = 1; | |
2747 | return max_sector - sector_nr; | |
2748 | } | |
6394cca5 N |
2749 | /* before building a request, check if we can skip these blocks.. |
2750 | * This call the bitmap_start_sync doesn't actually record anything | |
2751 | */ | |
e3b9703e | 2752 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
e5de485f | 2753 | !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { |
191ea9b2 N |
2754 | /* We can skip this block, and probably several more */ |
2755 | *skipped = 1; | |
2756 | return sync_blocks; | |
2757 | } | |
17999be4 | 2758 | |
7ac50447 TM |
2759 | /* |
2760 | * If there is non-resync activity waiting for a turn, then let it | |
2761 | * though before starting on this new sync request. | |
2762 | */ | |
824e47da | 2763 | if (atomic_read(&conf->nr_waiting[idx])) |
7ac50447 TM |
2764 | schedule_timeout_uninterruptible(1); |
2765 | ||
c40f341f GR |
2766 | /* we are incrementing sector_nr below. To be safe, we check against |
2767 | * sector_nr + two times RESYNC_SECTORS | |
2768 | */ | |
2769 | ||
2770 | bitmap_cond_end_sync(mddev->bitmap, sector_nr, | |
2771 | mddev_is_clustered(mddev) && (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high)); | |
1c4588e9 | 2772 | r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO); |
17999be4 | 2773 | |
c2fd4c94 | 2774 | raise_barrier(conf, sector_nr); |
1da177e4 | 2775 | |
3e198f78 | 2776 | rcu_read_lock(); |
1da177e4 | 2777 | /* |
3e198f78 N |
2778 | * If we get a correctably read error during resync or recovery, |
2779 | * we might want to read from a different device. So we | |
2780 | * flag all drives that could conceivably be read from for READ, | |
2781 | * and any others (which will be non-In_sync devices) for WRITE. | |
2782 | * If a read fails, we try reading from something else for which READ | |
2783 | * is OK. | |
1da177e4 | 2784 | */ |
1da177e4 | 2785 | |
1da177e4 LT |
2786 | r1_bio->mddev = mddev; |
2787 | r1_bio->sector = sector_nr; | |
191ea9b2 | 2788 | r1_bio->state = 0; |
1da177e4 | 2789 | set_bit(R1BIO_IsSync, &r1_bio->state); |
fd76863e | 2790 | /* make sure good_sectors won't go across barrier unit boundary */ |
2791 | good_sectors = align_to_barrier_unit_end(sector_nr, good_sectors); | |
1da177e4 | 2792 | |
8f19ccb2 | 2793 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3cb03002 | 2794 | struct md_rdev *rdev; |
1da177e4 | 2795 | bio = r1_bio->bios[i]; |
2aabaa65 | 2796 | bio_reset(bio); |
1da177e4 | 2797 | |
3e198f78 N |
2798 | rdev = rcu_dereference(conf->mirrors[i].rdev); |
2799 | if (rdev == NULL || | |
06f60385 | 2800 | test_bit(Faulty, &rdev->flags)) { |
8f19ccb2 N |
2801 | if (i < conf->raid_disks) |
2802 | still_degraded = 1; | |
3e198f78 | 2803 | } else if (!test_bit(In_sync, &rdev->flags)) { |
796a5cf0 | 2804 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
1da177e4 LT |
2805 | bio->bi_end_io = end_sync_write; |
2806 | write_targets ++; | |
3e198f78 N |
2807 | } else { |
2808 | /* may need to read from here */ | |
06f60385 N |
2809 | sector_t first_bad = MaxSector; |
2810 | int bad_sectors; | |
2811 | ||
2812 | if (is_badblock(rdev, sector_nr, good_sectors, | |
2813 | &first_bad, &bad_sectors)) { | |
2814 | if (first_bad > sector_nr) | |
2815 | good_sectors = first_bad - sector_nr; | |
2816 | else { | |
2817 | bad_sectors -= (sector_nr - first_bad); | |
2818 | if (min_bad == 0 || | |
2819 | min_bad > bad_sectors) | |
2820 | min_bad = bad_sectors; | |
2821 | } | |
2822 | } | |
2823 | if (sector_nr < first_bad) { | |
2824 | if (test_bit(WriteMostly, &rdev->flags)) { | |
2825 | if (wonly < 0) | |
2826 | wonly = i; | |
2827 | } else { | |
2828 | if (disk < 0) | |
2829 | disk = i; | |
2830 | } | |
796a5cf0 | 2831 | bio_set_op_attrs(bio, REQ_OP_READ, 0); |
06f60385 N |
2832 | bio->bi_end_io = end_sync_read; |
2833 | read_targets++; | |
d57368af AL |
2834 | } else if (!test_bit(WriteErrorSeen, &rdev->flags) && |
2835 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && | |
2836 | !test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { | |
2837 | /* | |
2838 | * The device is suitable for reading (InSync), | |
2839 | * but has bad block(s) here. Let's try to correct them, | |
2840 | * if we are doing resync or repair. Otherwise, leave | |
2841 | * this device alone for this sync request. | |
2842 | */ | |
796a5cf0 | 2843 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
d57368af AL |
2844 | bio->bi_end_io = end_sync_write; |
2845 | write_targets++; | |
3e198f78 | 2846 | } |
3e198f78 | 2847 | } |
06f60385 N |
2848 | if (bio->bi_end_io) { |
2849 | atomic_inc(&rdev->nr_pending); | |
4f024f37 | 2850 | bio->bi_iter.bi_sector = sector_nr + rdev->data_offset; |
06f60385 N |
2851 | bio->bi_bdev = rdev->bdev; |
2852 | bio->bi_private = r1_bio; | |
2e52d449 N |
2853 | if (test_bit(FailFast, &rdev->flags)) |
2854 | bio->bi_opf |= MD_FAILFAST; | |
06f60385 | 2855 | } |
1da177e4 | 2856 | } |
3e198f78 N |
2857 | rcu_read_unlock(); |
2858 | if (disk < 0) | |
2859 | disk = wonly; | |
2860 | r1_bio->read_disk = disk; | |
191ea9b2 | 2861 | |
06f60385 N |
2862 | if (read_targets == 0 && min_bad > 0) { |
2863 | /* These sectors are bad on all InSync devices, so we | |
2864 | * need to mark them bad on all write targets | |
2865 | */ | |
2866 | int ok = 1; | |
8f19ccb2 | 2867 | for (i = 0 ; i < conf->raid_disks * 2 ; i++) |
06f60385 | 2868 | if (r1_bio->bios[i]->bi_end_io == end_sync_write) { |
a42f9d83 | 2869 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
06f60385 N |
2870 | ok = rdev_set_badblocks(rdev, sector_nr, |
2871 | min_bad, 0 | |
2872 | ) && ok; | |
2873 | } | |
2953079c | 2874 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
06f60385 N |
2875 | *skipped = 1; |
2876 | put_buf(r1_bio); | |
2877 | ||
2878 | if (!ok) { | |
2879 | /* Cannot record the badblocks, so need to | |
2880 | * abort the resync. | |
2881 | * If there are multiple read targets, could just | |
2882 | * fail the really bad ones ??? | |
2883 | */ | |
2884 | conf->recovery_disabled = mddev->recovery_disabled; | |
2885 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2886 | return 0; | |
2887 | } else | |
2888 | return min_bad; | |
2889 | ||
2890 | } | |
2891 | if (min_bad > 0 && min_bad < good_sectors) { | |
2892 | /* only resync enough to reach the next bad->good | |
2893 | * transition */ | |
2894 | good_sectors = min_bad; | |
2895 | } | |
2896 | ||
3e198f78 N |
2897 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0) |
2898 | /* extra read targets are also write targets */ | |
2899 | write_targets += read_targets-1; | |
2900 | ||
2901 | if (write_targets == 0 || read_targets == 0) { | |
1da177e4 LT |
2902 | /* There is nowhere to write, so all non-sync |
2903 | * drives must be failed - so we are finished | |
2904 | */ | |
b7219ccb N |
2905 | sector_t rv; |
2906 | if (min_bad > 0) | |
2907 | max_sector = sector_nr + min_bad; | |
2908 | rv = max_sector - sector_nr; | |
57afd89f | 2909 | *skipped = 1; |
1da177e4 | 2910 | put_buf(r1_bio); |
1da177e4 LT |
2911 | return rv; |
2912 | } | |
2913 | ||
c6207277 N |
2914 | if (max_sector > mddev->resync_max) |
2915 | max_sector = mddev->resync_max; /* Don't do IO beyond here */ | |
06f60385 N |
2916 | if (max_sector > sector_nr + good_sectors) |
2917 | max_sector = sector_nr + good_sectors; | |
1da177e4 | 2918 | nr_sectors = 0; |
289e99e8 | 2919 | sync_blocks = 0; |
1da177e4 LT |
2920 | do { |
2921 | struct page *page; | |
2922 | int len = PAGE_SIZE; | |
2923 | if (sector_nr + (len>>9) > max_sector) | |
2924 | len = (max_sector - sector_nr) << 9; | |
2925 | if (len == 0) | |
2926 | break; | |
6a806c51 N |
2927 | if (sync_blocks == 0) { |
2928 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, | |
e5de485f N |
2929 | &sync_blocks, still_degraded) && |
2930 | !conf->fullsync && | |
2931 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6a806c51 | 2932 | break; |
7571ae88 | 2933 | if ((len >> 9) > sync_blocks) |
6a806c51 | 2934 | len = sync_blocks<<9; |
ab7a30c7 | 2935 | } |
191ea9b2 | 2936 | |
8f19ccb2 | 2937 | for (i = 0 ; i < conf->raid_disks * 2; i++) { |
1da177e4 LT |
2938 | bio = r1_bio->bios[i]; |
2939 | if (bio->bi_end_io) { | |
d11c171e | 2940 | page = bio->bi_io_vec[bio->bi_vcnt].bv_page; |
1da177e4 LT |
2941 | if (bio_add_page(bio, page, len, 0) == 0) { |
2942 | /* stop here */ | |
d11c171e | 2943 | bio->bi_io_vec[bio->bi_vcnt].bv_page = page; |
1da177e4 LT |
2944 | while (i > 0) { |
2945 | i--; | |
2946 | bio = r1_bio->bios[i]; | |
6a806c51 N |
2947 | if (bio->bi_end_io==NULL) |
2948 | continue; | |
1da177e4 LT |
2949 | /* remove last page from this bio */ |
2950 | bio->bi_vcnt--; | |
4f024f37 | 2951 | bio->bi_iter.bi_size -= len; |
b7c44ed9 | 2952 | bio_clear_flag(bio, BIO_SEG_VALID); |
1da177e4 LT |
2953 | } |
2954 | goto bio_full; | |
2955 | } | |
2956 | } | |
2957 | } | |
2958 | nr_sectors += len>>9; | |
2959 | sector_nr += len>>9; | |
191ea9b2 | 2960 | sync_blocks -= (len>>9); |
1da177e4 LT |
2961 | } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES); |
2962 | bio_full: | |
1da177e4 LT |
2963 | r1_bio->sectors = nr_sectors; |
2964 | ||
c40f341f GR |
2965 | if (mddev_is_clustered(mddev) && |
2966 | conf->cluster_sync_high < sector_nr + nr_sectors) { | |
2967 | conf->cluster_sync_low = mddev->curr_resync_completed; | |
2968 | conf->cluster_sync_high = conf->cluster_sync_low + CLUSTER_RESYNC_WINDOW_SECTORS; | |
2969 | /* Send resync message */ | |
2970 | md_cluster_ops->resync_info_update(mddev, | |
2971 | conf->cluster_sync_low, | |
2972 | conf->cluster_sync_high); | |
2973 | } | |
2974 | ||
d11c171e N |
2975 | /* For a user-requested sync, we read all readable devices and do a |
2976 | * compare | |
2977 | */ | |
2978 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
2979 | atomic_set(&r1_bio->remaining, read_targets); | |
2d4f4f33 | 2980 | for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) { |
d11c171e N |
2981 | bio = r1_bio->bios[i]; |
2982 | if (bio->bi_end_io == end_sync_read) { | |
2d4f4f33 | 2983 | read_targets--; |
ddac7c7e | 2984 | md_sync_acct(bio->bi_bdev, nr_sectors); |
2e52d449 N |
2985 | if (read_targets == 1) |
2986 | bio->bi_opf &= ~MD_FAILFAST; | |
d11c171e N |
2987 | generic_make_request(bio); |
2988 | } | |
2989 | } | |
2990 | } else { | |
2991 | atomic_set(&r1_bio->remaining, 1); | |
2992 | bio = r1_bio->bios[r1_bio->read_disk]; | |
ddac7c7e | 2993 | md_sync_acct(bio->bi_bdev, nr_sectors); |
2e52d449 N |
2994 | if (read_targets == 1) |
2995 | bio->bi_opf &= ~MD_FAILFAST; | |
d11c171e | 2996 | generic_make_request(bio); |
1da177e4 | 2997 | |
d11c171e | 2998 | } |
1da177e4 LT |
2999 | return nr_sectors; |
3000 | } | |
3001 | ||
fd01b88c | 3002 | static sector_t raid1_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce DW |
3003 | { |
3004 | if (sectors) | |
3005 | return sectors; | |
3006 | ||
3007 | return mddev->dev_sectors; | |
3008 | } | |
3009 | ||
e8096360 | 3010 | static struct r1conf *setup_conf(struct mddev *mddev) |
1da177e4 | 3011 | { |
e8096360 | 3012 | struct r1conf *conf; |
709ae487 | 3013 | int i; |
0eaf822c | 3014 | struct raid1_info *disk; |
3cb03002 | 3015 | struct md_rdev *rdev; |
709ae487 | 3016 | int err = -ENOMEM; |
1da177e4 | 3017 | |
e8096360 | 3018 | conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL); |
1da177e4 | 3019 | if (!conf) |
709ae487 | 3020 | goto abort; |
1da177e4 | 3021 | |
fd76863e | 3022 | conf->nr_pending = kcalloc(BARRIER_BUCKETS_NR, |
824e47da | 3023 | sizeof(atomic_t), GFP_KERNEL); |
fd76863e | 3024 | if (!conf->nr_pending) |
3025 | goto abort; | |
3026 | ||
3027 | conf->nr_waiting = kcalloc(BARRIER_BUCKETS_NR, | |
824e47da | 3028 | sizeof(atomic_t), GFP_KERNEL); |
fd76863e | 3029 | if (!conf->nr_waiting) |
3030 | goto abort; | |
3031 | ||
3032 | conf->nr_queued = kcalloc(BARRIER_BUCKETS_NR, | |
824e47da | 3033 | sizeof(atomic_t), GFP_KERNEL); |
fd76863e | 3034 | if (!conf->nr_queued) |
3035 | goto abort; | |
3036 | ||
3037 | conf->barrier = kcalloc(BARRIER_BUCKETS_NR, | |
824e47da | 3038 | sizeof(atomic_t), GFP_KERNEL); |
fd76863e | 3039 | if (!conf->barrier) |
3040 | goto abort; | |
3041 | ||
0eaf822c | 3042 | conf->mirrors = kzalloc(sizeof(struct raid1_info) |
8f19ccb2 | 3043 | * mddev->raid_disks * 2, |
1da177e4 LT |
3044 | GFP_KERNEL); |
3045 | if (!conf->mirrors) | |
709ae487 | 3046 | goto abort; |
1da177e4 | 3047 | |
ddaf22ab N |
3048 | conf->tmppage = alloc_page(GFP_KERNEL); |
3049 | if (!conf->tmppage) | |
709ae487 | 3050 | goto abort; |
ddaf22ab | 3051 | |
709ae487 | 3052 | conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL); |
1da177e4 | 3053 | if (!conf->poolinfo) |
709ae487 | 3054 | goto abort; |
8f19ccb2 | 3055 | conf->poolinfo->raid_disks = mddev->raid_disks * 2; |
1da177e4 LT |
3056 | conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, |
3057 | r1bio_pool_free, | |
3058 | conf->poolinfo); | |
3059 | if (!conf->r1bio_pool) | |
709ae487 N |
3060 | goto abort; |
3061 | ||
ed9bfdf1 | 3062 | conf->poolinfo->mddev = mddev; |
1da177e4 | 3063 | |
c19d5798 | 3064 | err = -EINVAL; |
e7e72bf6 | 3065 | spin_lock_init(&conf->device_lock); |
dafb20fa | 3066 | rdev_for_each(rdev, mddev) { |
aba336bd | 3067 | struct request_queue *q; |
709ae487 | 3068 | int disk_idx = rdev->raid_disk; |
1da177e4 LT |
3069 | if (disk_idx >= mddev->raid_disks |
3070 | || disk_idx < 0) | |
3071 | continue; | |
c19d5798 | 3072 | if (test_bit(Replacement, &rdev->flags)) |
02b898f2 | 3073 | disk = conf->mirrors + mddev->raid_disks + disk_idx; |
c19d5798 N |
3074 | else |
3075 | disk = conf->mirrors + disk_idx; | |
1da177e4 | 3076 | |
c19d5798 N |
3077 | if (disk->rdev) |
3078 | goto abort; | |
1da177e4 | 3079 | disk->rdev = rdev; |
aba336bd | 3080 | q = bdev_get_queue(rdev->bdev); |
1da177e4 LT |
3081 | |
3082 | disk->head_position = 0; | |
12cee5a8 | 3083 | disk->seq_start = MaxSector; |
1da177e4 LT |
3084 | } |
3085 | conf->raid_disks = mddev->raid_disks; | |
3086 | conf->mddev = mddev; | |
1da177e4 | 3087 | INIT_LIST_HEAD(&conf->retry_list); |
55ce74d4 | 3088 | INIT_LIST_HEAD(&conf->bio_end_io_list); |
1da177e4 LT |
3089 | |
3090 | spin_lock_init(&conf->resync_lock); | |
17999be4 | 3091 | init_waitqueue_head(&conf->wait_barrier); |
1da177e4 | 3092 | |
191ea9b2 | 3093 | bio_list_init(&conf->pending_bio_list); |
34db0cd6 | 3094 | conf->pending_count = 0; |
d890fa2b | 3095 | conf->recovery_disabled = mddev->recovery_disabled - 1; |
191ea9b2 | 3096 | |
c19d5798 | 3097 | err = -EIO; |
8f19ccb2 | 3098 | for (i = 0; i < conf->raid_disks * 2; i++) { |
1da177e4 LT |
3099 | |
3100 | disk = conf->mirrors + i; | |
3101 | ||
c19d5798 N |
3102 | if (i < conf->raid_disks && |
3103 | disk[conf->raid_disks].rdev) { | |
3104 | /* This slot has a replacement. */ | |
3105 | if (!disk->rdev) { | |
3106 | /* No original, just make the replacement | |
3107 | * a recovering spare | |
3108 | */ | |
3109 | disk->rdev = | |
3110 | disk[conf->raid_disks].rdev; | |
3111 | disk[conf->raid_disks].rdev = NULL; | |
3112 | } else if (!test_bit(In_sync, &disk->rdev->flags)) | |
3113 | /* Original is not in_sync - bad */ | |
3114 | goto abort; | |
3115 | } | |
3116 | ||
5fd6c1dc N |
3117 | if (!disk->rdev || |
3118 | !test_bit(In_sync, &disk->rdev->flags)) { | |
1da177e4 | 3119 | disk->head_position = 0; |
4f0a5e01 JB |
3120 | if (disk->rdev && |
3121 | (disk->rdev->saved_raid_disk < 0)) | |
918f0238 | 3122 | conf->fullsync = 1; |
be4d3280 | 3123 | } |
1da177e4 | 3124 | } |
709ae487 | 3125 | |
709ae487 | 3126 | err = -ENOMEM; |
0232605d | 3127 | conf->thread = md_register_thread(raid1d, mddev, "raid1"); |
1d41c216 | 3128 | if (!conf->thread) |
709ae487 | 3129 | goto abort; |
1da177e4 | 3130 | |
709ae487 N |
3131 | return conf; |
3132 | ||
3133 | abort: | |
3134 | if (conf) { | |
644df1a8 | 3135 | mempool_destroy(conf->r1bio_pool); |
709ae487 N |
3136 | kfree(conf->mirrors); |
3137 | safe_put_page(conf->tmppage); | |
3138 | kfree(conf->poolinfo); | |
fd76863e | 3139 | kfree(conf->nr_pending); |
3140 | kfree(conf->nr_waiting); | |
3141 | kfree(conf->nr_queued); | |
3142 | kfree(conf->barrier); | |
709ae487 N |
3143 | kfree(conf); |
3144 | } | |
3145 | return ERR_PTR(err); | |
3146 | } | |
3147 | ||
afa0f557 | 3148 | static void raid1_free(struct mddev *mddev, void *priv); |
849674e4 | 3149 | static int raid1_run(struct mddev *mddev) |
709ae487 | 3150 | { |
e8096360 | 3151 | struct r1conf *conf; |
709ae487 | 3152 | int i; |
3cb03002 | 3153 | struct md_rdev *rdev; |
5220ea1e | 3154 | int ret; |
2ff8cc2c | 3155 | bool discard_supported = false; |
709ae487 N |
3156 | |
3157 | if (mddev->level != 1) { | |
1d41c216 N |
3158 | pr_warn("md/raid1:%s: raid level not set to mirroring (%d)\n", |
3159 | mdname(mddev), mddev->level); | |
709ae487 N |
3160 | return -EIO; |
3161 | } | |
3162 | if (mddev->reshape_position != MaxSector) { | |
1d41c216 N |
3163 | pr_warn("md/raid1:%s: reshape_position set but not supported\n", |
3164 | mdname(mddev)); | |
709ae487 N |
3165 | return -EIO; |
3166 | } | |
1da177e4 | 3167 | /* |
709ae487 N |
3168 | * copy the already verified devices into our private RAID1 |
3169 | * bookkeeping area. [whatever we allocate in run(), | |
afa0f557 | 3170 | * should be freed in raid1_free()] |
1da177e4 | 3171 | */ |
709ae487 N |
3172 | if (mddev->private == NULL) |
3173 | conf = setup_conf(mddev); | |
3174 | else | |
3175 | conf = mddev->private; | |
1da177e4 | 3176 | |
709ae487 N |
3177 | if (IS_ERR(conf)) |
3178 | return PTR_ERR(conf); | |
1da177e4 | 3179 | |
3deff1a7 | 3180 | if (mddev->queue) { |
5026d7a9 | 3181 | blk_queue_max_write_same_sectors(mddev->queue, 0); |
3deff1a7 CH |
3182 | blk_queue_max_write_zeroes_sectors(mddev->queue, 0); |
3183 | } | |
5026d7a9 | 3184 | |
dafb20fa | 3185 | rdev_for_each(rdev, mddev) { |
1ed7242e JB |
3186 | if (!mddev->gendisk) |
3187 | continue; | |
709ae487 N |
3188 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
3189 | rdev->data_offset << 9); | |
2ff8cc2c SL |
3190 | if (blk_queue_discard(bdev_get_queue(rdev->bdev))) |
3191 | discard_supported = true; | |
1da177e4 | 3192 | } |
191ea9b2 | 3193 | |
709ae487 N |
3194 | mddev->degraded = 0; |
3195 | for (i=0; i < conf->raid_disks; i++) | |
3196 | if (conf->mirrors[i].rdev == NULL || | |
3197 | !test_bit(In_sync, &conf->mirrors[i].rdev->flags) || | |
3198 | test_bit(Faulty, &conf->mirrors[i].rdev->flags)) | |
3199 | mddev->degraded++; | |
3200 | ||
3201 | if (conf->raid_disks - mddev->degraded == 1) | |
3202 | mddev->recovery_cp = MaxSector; | |
3203 | ||
8c6ac868 | 3204 | if (mddev->recovery_cp != MaxSector) |
1d41c216 N |
3205 | pr_info("md/raid1:%s: not clean -- starting background reconstruction\n", |
3206 | mdname(mddev)); | |
3207 | pr_info("md/raid1:%s: active with %d out of %d mirrors\n", | |
f72ffdd6 | 3208 | mdname(mddev), mddev->raid_disks - mddev->degraded, |
1da177e4 | 3209 | mddev->raid_disks); |
709ae487 | 3210 | |
1da177e4 LT |
3211 | /* |
3212 | * Ok, everything is just fine now | |
3213 | */ | |
709ae487 N |
3214 | mddev->thread = conf->thread; |
3215 | conf->thread = NULL; | |
3216 | mddev->private = conf; | |
46533ff7 | 3217 | set_bit(MD_FAILFAST_SUPPORTED, &mddev->flags); |
709ae487 | 3218 | |
1f403624 | 3219 | md_set_array_sectors(mddev, raid1_size(mddev, 0, 0)); |
1da177e4 | 3220 | |
1ed7242e | 3221 | if (mddev->queue) { |
2ff8cc2c SL |
3222 | if (discard_supported) |
3223 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, | |
3224 | mddev->queue); | |
3225 | else | |
3226 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, | |
3227 | mddev->queue); | |
1ed7242e | 3228 | } |
5220ea1e | 3229 | |
3230 | ret = md_integrity_register(mddev); | |
5aa61f42 N |
3231 | if (ret) { |
3232 | md_unregister_thread(&mddev->thread); | |
afa0f557 | 3233 | raid1_free(mddev, conf); |
5aa61f42 | 3234 | } |
5220ea1e | 3235 | return ret; |
1da177e4 LT |
3236 | } |
3237 | ||
afa0f557 | 3238 | static void raid1_free(struct mddev *mddev, void *priv) |
1da177e4 | 3239 | { |
afa0f557 | 3240 | struct r1conf *conf = priv; |
409c57f3 | 3241 | |
644df1a8 | 3242 | mempool_destroy(conf->r1bio_pool); |
990a8baf | 3243 | kfree(conf->mirrors); |
0fea7ed8 | 3244 | safe_put_page(conf->tmppage); |
990a8baf | 3245 | kfree(conf->poolinfo); |
fd76863e | 3246 | kfree(conf->nr_pending); |
3247 | kfree(conf->nr_waiting); | |
3248 | kfree(conf->nr_queued); | |
3249 | kfree(conf->barrier); | |
1da177e4 | 3250 | kfree(conf); |
1da177e4 LT |
3251 | } |
3252 | ||
fd01b88c | 3253 | static int raid1_resize(struct mddev *mddev, sector_t sectors) |
1da177e4 LT |
3254 | { |
3255 | /* no resync is happening, and there is enough space | |
3256 | * on all devices, so we can resize. | |
3257 | * We need to make sure resync covers any new space. | |
3258 | * If the array is shrinking we should possibly wait until | |
3259 | * any io in the removed space completes, but it hardly seems | |
3260 | * worth it. | |
3261 | */ | |
a4a6125a N |
3262 | sector_t newsize = raid1_size(mddev, sectors, 0); |
3263 | if (mddev->external_size && | |
3264 | mddev->array_sectors > newsize) | |
b522adcd | 3265 | return -EINVAL; |
a4a6125a N |
3266 | if (mddev->bitmap) { |
3267 | int ret = bitmap_resize(mddev->bitmap, newsize, 0, 0); | |
3268 | if (ret) | |
3269 | return ret; | |
3270 | } | |
3271 | md_set_array_sectors(mddev, newsize); | |
b522adcd | 3272 | if (sectors > mddev->dev_sectors && |
b098636c | 3273 | mddev->recovery_cp > mddev->dev_sectors) { |
58c0fed4 | 3274 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
3275 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
3276 | } | |
b522adcd | 3277 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 3278 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
3279 | return 0; |
3280 | } | |
3281 | ||
fd01b88c | 3282 | static int raid1_reshape(struct mddev *mddev) |
1da177e4 LT |
3283 | { |
3284 | /* We need to: | |
3285 | * 1/ resize the r1bio_pool | |
3286 | * 2/ resize conf->mirrors | |
3287 | * | |
3288 | * We allocate a new r1bio_pool if we can. | |
3289 | * Then raise a device barrier and wait until all IO stops. | |
3290 | * Then resize conf->mirrors and swap in the new r1bio pool. | |
6ea9c07c N |
3291 | * |
3292 | * At the same time, we "pack" the devices so that all the missing | |
3293 | * devices have the higher raid_disk numbers. | |
1da177e4 LT |
3294 | */ |
3295 | mempool_t *newpool, *oldpool; | |
3296 | struct pool_info *newpoolinfo; | |
0eaf822c | 3297 | struct raid1_info *newmirrors; |
e8096360 | 3298 | struct r1conf *conf = mddev->private; |
63c70c4f | 3299 | int cnt, raid_disks; |
c04be0aa | 3300 | unsigned long flags; |
b5470dc5 | 3301 | int d, d2, err; |
1da177e4 | 3302 | |
63c70c4f | 3303 | /* Cannot change chunk_size, layout, or level */ |
664e7c41 | 3304 | if (mddev->chunk_sectors != mddev->new_chunk_sectors || |
63c70c4f N |
3305 | mddev->layout != mddev->new_layout || |
3306 | mddev->level != mddev->new_level) { | |
664e7c41 | 3307 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
63c70c4f N |
3308 | mddev->new_layout = mddev->layout; |
3309 | mddev->new_level = mddev->level; | |
3310 | return -EINVAL; | |
3311 | } | |
3312 | ||
28c1b9fd GR |
3313 | if (!mddev_is_clustered(mddev)) { |
3314 | err = md_allow_write(mddev); | |
3315 | if (err) | |
3316 | return err; | |
3317 | } | |
2a2275d6 | 3318 | |
63c70c4f N |
3319 | raid_disks = mddev->raid_disks + mddev->delta_disks; |
3320 | ||
6ea9c07c N |
3321 | if (raid_disks < conf->raid_disks) { |
3322 | cnt=0; | |
3323 | for (d= 0; d < conf->raid_disks; d++) | |
3324 | if (conf->mirrors[d].rdev) | |
3325 | cnt++; | |
3326 | if (cnt > raid_disks) | |
1da177e4 | 3327 | return -EBUSY; |
6ea9c07c | 3328 | } |
1da177e4 LT |
3329 | |
3330 | newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL); | |
3331 | if (!newpoolinfo) | |
3332 | return -ENOMEM; | |
3333 | newpoolinfo->mddev = mddev; | |
8f19ccb2 | 3334 | newpoolinfo->raid_disks = raid_disks * 2; |
1da177e4 LT |
3335 | |
3336 | newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, | |
3337 | r1bio_pool_free, newpoolinfo); | |
3338 | if (!newpool) { | |
3339 | kfree(newpoolinfo); | |
3340 | return -ENOMEM; | |
3341 | } | |
0eaf822c | 3342 | newmirrors = kzalloc(sizeof(struct raid1_info) * raid_disks * 2, |
8f19ccb2 | 3343 | GFP_KERNEL); |
1da177e4 LT |
3344 | if (!newmirrors) { |
3345 | kfree(newpoolinfo); | |
3346 | mempool_destroy(newpool); | |
3347 | return -ENOMEM; | |
3348 | } | |
1da177e4 | 3349 | |
e2d59925 | 3350 | freeze_array(conf, 0); |
1da177e4 LT |
3351 | |
3352 | /* ok, everything is stopped */ | |
3353 | oldpool = conf->r1bio_pool; | |
3354 | conf->r1bio_pool = newpool; | |
6ea9c07c | 3355 | |
a88aa786 | 3356 | for (d = d2 = 0; d < conf->raid_disks; d++) { |
3cb03002 | 3357 | struct md_rdev *rdev = conf->mirrors[d].rdev; |
a88aa786 | 3358 | if (rdev && rdev->raid_disk != d2) { |
36fad858 | 3359 | sysfs_unlink_rdev(mddev, rdev); |
a88aa786 | 3360 | rdev->raid_disk = d2; |
36fad858 NK |
3361 | sysfs_unlink_rdev(mddev, rdev); |
3362 | if (sysfs_link_rdev(mddev, rdev)) | |
1d41c216 N |
3363 | pr_warn("md/raid1:%s: cannot register rd%d\n", |
3364 | mdname(mddev), rdev->raid_disk); | |
6ea9c07c | 3365 | } |
a88aa786 N |
3366 | if (rdev) |
3367 | newmirrors[d2++].rdev = rdev; | |
3368 | } | |
1da177e4 LT |
3369 | kfree(conf->mirrors); |
3370 | conf->mirrors = newmirrors; | |
3371 | kfree(conf->poolinfo); | |
3372 | conf->poolinfo = newpoolinfo; | |
3373 | ||
c04be0aa | 3374 | spin_lock_irqsave(&conf->device_lock, flags); |
1da177e4 | 3375 | mddev->degraded += (raid_disks - conf->raid_disks); |
c04be0aa | 3376 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 3377 | conf->raid_disks = mddev->raid_disks = raid_disks; |
63c70c4f | 3378 | mddev->delta_disks = 0; |
1da177e4 | 3379 | |
e2d59925 | 3380 | unfreeze_array(conf); |
1da177e4 | 3381 | |
985ca973 | 3382 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
1da177e4 LT |
3383 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
3384 | md_wakeup_thread(mddev->thread); | |
3385 | ||
3386 | mempool_destroy(oldpool); | |
3387 | return 0; | |
3388 | } | |
3389 | ||
fd01b88c | 3390 | static void raid1_quiesce(struct mddev *mddev, int state) |
36fa3063 | 3391 | { |
e8096360 | 3392 | struct r1conf *conf = mddev->private; |
36fa3063 N |
3393 | |
3394 | switch(state) { | |
6eef4b21 N |
3395 | case 2: /* wake for suspend */ |
3396 | wake_up(&conf->wait_barrier); | |
3397 | break; | |
9e6603da | 3398 | case 1: |
07169fd4 | 3399 | freeze_array(conf, 0); |
36fa3063 | 3400 | break; |
9e6603da | 3401 | case 0: |
07169fd4 | 3402 | unfreeze_array(conf); |
36fa3063 N |
3403 | break; |
3404 | } | |
36fa3063 N |
3405 | } |
3406 | ||
fd01b88c | 3407 | static void *raid1_takeover(struct mddev *mddev) |
709ae487 N |
3408 | { |
3409 | /* raid1 can take over: | |
3410 | * raid5 with 2 devices, any layout or chunk size | |
3411 | */ | |
3412 | if (mddev->level == 5 && mddev->raid_disks == 2) { | |
e8096360 | 3413 | struct r1conf *conf; |
709ae487 N |
3414 | mddev->new_level = 1; |
3415 | mddev->new_layout = 0; | |
3416 | mddev->new_chunk_sectors = 0; | |
3417 | conf = setup_conf(mddev); | |
6995f0b2 | 3418 | if (!IS_ERR(conf)) { |
07169fd4 | 3419 | /* Array must appear to be quiesced */ |
3420 | conf->array_frozen = 1; | |
394ed8e4 SL |
3421 | mddev_clear_unsupported_flags(mddev, |
3422 | UNSUPPORTED_MDDEV_FLAGS); | |
6995f0b2 | 3423 | } |
709ae487 N |
3424 | return conf; |
3425 | } | |
3426 | return ERR_PTR(-EINVAL); | |
3427 | } | |
1da177e4 | 3428 | |
84fc4b56 | 3429 | static struct md_personality raid1_personality = |
1da177e4 LT |
3430 | { |
3431 | .name = "raid1", | |
2604b703 | 3432 | .level = 1, |
1da177e4 | 3433 | .owner = THIS_MODULE, |
849674e4 SL |
3434 | .make_request = raid1_make_request, |
3435 | .run = raid1_run, | |
afa0f557 | 3436 | .free = raid1_free, |
849674e4 SL |
3437 | .status = raid1_status, |
3438 | .error_handler = raid1_error, | |
1da177e4 LT |
3439 | .hot_add_disk = raid1_add_disk, |
3440 | .hot_remove_disk= raid1_remove_disk, | |
3441 | .spare_active = raid1_spare_active, | |
849674e4 | 3442 | .sync_request = raid1_sync_request, |
1da177e4 | 3443 | .resize = raid1_resize, |
80c3a6ce | 3444 | .size = raid1_size, |
63c70c4f | 3445 | .check_reshape = raid1_reshape, |
36fa3063 | 3446 | .quiesce = raid1_quiesce, |
709ae487 | 3447 | .takeover = raid1_takeover, |
5c675f83 | 3448 | .congested = raid1_congested, |
1da177e4 LT |
3449 | }; |
3450 | ||
3451 | static int __init raid_init(void) | |
3452 | { | |
2604b703 | 3453 | return register_md_personality(&raid1_personality); |
1da177e4 LT |
3454 | } |
3455 | ||
3456 | static void raid_exit(void) | |
3457 | { | |
2604b703 | 3458 | unregister_md_personality(&raid1_personality); |
1da177e4 LT |
3459 | } |
3460 | ||
3461 | module_init(raid_init); | |
3462 | module_exit(raid_exit); | |
3463 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 3464 | MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD"); |
1da177e4 | 3465 | MODULE_ALIAS("md-personality-3"); /* RAID1 */ |
d9d166c2 | 3466 | MODULE_ALIAS("md-raid1"); |
2604b703 | 3467 | MODULE_ALIAS("md-level-1"); |
34db0cd6 N |
3468 | |
3469 | module_param(max_queued_requests, int, S_IRUGO|S_IWUSR); |