]>
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> |
43b2e5d8 | 40 | #include "md.h" |
ef740c37 CH |
41 | #include "raid1.h" |
42 | #include "bitmap.h" | |
191ea9b2 | 43 | |
1da177e4 LT |
44 | /* |
45 | * Number of guaranteed r1bios in case of extreme VM load: | |
46 | */ | |
47 | #define NR_RAID1_BIOS 256 | |
48 | ||
473e87ce JB |
49 | /* when we get a read error on a read-only array, we redirect to another |
50 | * device without failing the first device, or trying to over-write to | |
51 | * correct the read error. To keep track of bad blocks on a per-bio | |
52 | * level, we store IO_BLOCKED in the appropriate 'bios' pointer | |
53 | */ | |
54 | #define IO_BLOCKED ((struct bio *)1) | |
55 | /* When we successfully write to a known bad-block, we need to remove the | |
56 | * bad-block marking which must be done from process context. So we record | |
57 | * the success by setting devs[n].bio to IO_MADE_GOOD | |
58 | */ | |
59 | #define IO_MADE_GOOD ((struct bio *)2) | |
60 | ||
61 | #define BIO_SPECIAL(bio) ((unsigned long)bio <= 2) | |
62 | ||
34db0cd6 N |
63 | /* When there are this many requests queue to be written by |
64 | * the raid1 thread, we become 'congested' to provide back-pressure | |
65 | * for writeback. | |
66 | */ | |
67 | static int max_queued_requests = 1024; | |
1da177e4 | 68 | |
e8096360 N |
69 | static void allow_barrier(struct r1conf *conf); |
70 | static void lower_barrier(struct r1conf *conf); | |
1da177e4 | 71 | |
dd0fc66f | 72 | static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 LT |
73 | { |
74 | struct pool_info *pi = data; | |
9f2c9d12 | 75 | int size = offsetof(struct r1bio, bios[pi->raid_disks]); |
1da177e4 LT |
76 | |
77 | /* allocate a r1bio with room for raid_disks entries in the bios array */ | |
7eaceacc | 78 | return kzalloc(size, gfp_flags); |
1da177e4 LT |
79 | } |
80 | ||
81 | static void r1bio_pool_free(void *r1_bio, void *data) | |
82 | { | |
83 | kfree(r1_bio); | |
84 | } | |
85 | ||
86 | #define RESYNC_BLOCK_SIZE (64*1024) | |
87 | //#define RESYNC_BLOCK_SIZE PAGE_SIZE | |
88 | #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9) | |
89 | #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE) | |
90 | #define RESYNC_WINDOW (2048*1024) | |
91 | ||
dd0fc66f | 92 | static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 LT |
93 | { |
94 | struct pool_info *pi = data; | |
9f2c9d12 | 95 | struct r1bio *r1_bio; |
1da177e4 LT |
96 | struct bio *bio; |
97 | int i, j; | |
98 | ||
99 | r1_bio = r1bio_pool_alloc(gfp_flags, pi); | |
7eaceacc | 100 | if (!r1_bio) |
1da177e4 | 101 | return NULL; |
1da177e4 LT |
102 | |
103 | /* | |
104 | * Allocate bios : 1 for reading, n-1 for writing | |
105 | */ | |
106 | for (j = pi->raid_disks ; j-- ; ) { | |
6746557f | 107 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); |
1da177e4 LT |
108 | if (!bio) |
109 | goto out_free_bio; | |
110 | r1_bio->bios[j] = bio; | |
111 | } | |
112 | /* | |
113 | * Allocate RESYNC_PAGES data pages and attach them to | |
d11c171e N |
114 | * the first bio. |
115 | * If this is a user-requested check/repair, allocate | |
116 | * RESYNC_PAGES for each bio. | |
1da177e4 | 117 | */ |
d11c171e N |
118 | if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) |
119 | j = pi->raid_disks; | |
120 | else | |
121 | j = 1; | |
122 | while(j--) { | |
123 | bio = r1_bio->bios[j]; | |
a0787606 KO |
124 | bio->bi_vcnt = RESYNC_PAGES; |
125 | ||
126 | if (bio_alloc_pages(bio, gfp_flags)) | |
127 | goto out_free_bio; | |
d11c171e N |
128 | } |
129 | /* If not user-requests, copy the page pointers to all bios */ | |
130 | if (!test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) { | |
131 | for (i=0; i<RESYNC_PAGES ; i++) | |
132 | for (j=1; j<pi->raid_disks; j++) | |
133 | r1_bio->bios[j]->bi_io_vec[i].bv_page = | |
134 | r1_bio->bios[0]->bi_io_vec[i].bv_page; | |
1da177e4 LT |
135 | } |
136 | ||
137 | r1_bio->master_bio = NULL; | |
138 | ||
139 | return r1_bio; | |
140 | ||
1da177e4 | 141 | out_free_bio: |
8f19ccb2 | 142 | while (++j < pi->raid_disks) |
1da177e4 LT |
143 | bio_put(r1_bio->bios[j]); |
144 | r1bio_pool_free(r1_bio, data); | |
145 | return NULL; | |
146 | } | |
147 | ||
148 | static void r1buf_pool_free(void *__r1_bio, void *data) | |
149 | { | |
150 | struct pool_info *pi = data; | |
d11c171e | 151 | int i,j; |
9f2c9d12 | 152 | struct r1bio *r1bio = __r1_bio; |
1da177e4 | 153 | |
d11c171e N |
154 | for (i = 0; i < RESYNC_PAGES; i++) |
155 | for (j = pi->raid_disks; j-- ;) { | |
156 | if (j == 0 || | |
157 | r1bio->bios[j]->bi_io_vec[i].bv_page != | |
158 | r1bio->bios[0]->bi_io_vec[i].bv_page) | |
1345b1d8 | 159 | safe_put_page(r1bio->bios[j]->bi_io_vec[i].bv_page); |
d11c171e | 160 | } |
1da177e4 LT |
161 | for (i=0 ; i < pi->raid_disks; i++) |
162 | bio_put(r1bio->bios[i]); | |
163 | ||
164 | r1bio_pool_free(r1bio, data); | |
165 | } | |
166 | ||
e8096360 | 167 | static void put_all_bios(struct r1conf *conf, struct r1bio *r1_bio) |
1da177e4 LT |
168 | { |
169 | int i; | |
170 | ||
8f19ccb2 | 171 | for (i = 0; i < conf->raid_disks * 2; i++) { |
1da177e4 | 172 | struct bio **bio = r1_bio->bios + i; |
4367af55 | 173 | if (!BIO_SPECIAL(*bio)) |
1da177e4 LT |
174 | bio_put(*bio); |
175 | *bio = NULL; | |
176 | } | |
177 | } | |
178 | ||
9f2c9d12 | 179 | static void free_r1bio(struct r1bio *r1_bio) |
1da177e4 | 180 | { |
e8096360 | 181 | struct r1conf *conf = r1_bio->mddev->private; |
1da177e4 | 182 | |
1da177e4 LT |
183 | put_all_bios(conf, r1_bio); |
184 | mempool_free(r1_bio, conf->r1bio_pool); | |
185 | } | |
186 | ||
9f2c9d12 | 187 | static void put_buf(struct r1bio *r1_bio) |
1da177e4 | 188 | { |
e8096360 | 189 | struct r1conf *conf = r1_bio->mddev->private; |
3e198f78 N |
190 | int i; |
191 | ||
8f19ccb2 | 192 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3e198f78 N |
193 | struct bio *bio = r1_bio->bios[i]; |
194 | if (bio->bi_end_io) | |
195 | rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev); | |
196 | } | |
1da177e4 LT |
197 | |
198 | mempool_free(r1_bio, conf->r1buf_pool); | |
199 | ||
17999be4 | 200 | lower_barrier(conf); |
1da177e4 LT |
201 | } |
202 | ||
9f2c9d12 | 203 | static void reschedule_retry(struct r1bio *r1_bio) |
1da177e4 LT |
204 | { |
205 | unsigned long flags; | |
fd01b88c | 206 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 207 | struct r1conf *conf = mddev->private; |
1da177e4 LT |
208 | |
209 | spin_lock_irqsave(&conf->device_lock, flags); | |
210 | list_add(&r1_bio->retry_list, &conf->retry_list); | |
ddaf22ab | 211 | conf->nr_queued ++; |
1da177e4 LT |
212 | spin_unlock_irqrestore(&conf->device_lock, flags); |
213 | ||
17999be4 | 214 | wake_up(&conf->wait_barrier); |
1da177e4 LT |
215 | md_wakeup_thread(mddev->thread); |
216 | } | |
217 | ||
218 | /* | |
219 | * raid_end_bio_io() is called when we have finished servicing a mirrored | |
220 | * operation and are ready to return a success/failure code to the buffer | |
221 | * cache layer. | |
222 | */ | |
9f2c9d12 | 223 | static void call_bio_endio(struct r1bio *r1_bio) |
d2eb35ac N |
224 | { |
225 | struct bio *bio = r1_bio->master_bio; | |
226 | int done; | |
e8096360 | 227 | struct r1conf *conf = r1_bio->mddev->private; |
d2eb35ac N |
228 | |
229 | if (bio->bi_phys_segments) { | |
230 | unsigned long flags; | |
231 | spin_lock_irqsave(&conf->device_lock, flags); | |
232 | bio->bi_phys_segments--; | |
233 | done = (bio->bi_phys_segments == 0); | |
234 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
235 | } else | |
236 | done = 1; | |
237 | ||
238 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) | |
239 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
240 | if (done) { | |
241 | bio_endio(bio, 0); | |
242 | /* | |
243 | * Wake up any possible resync thread that waits for the device | |
244 | * to go idle. | |
245 | */ | |
246 | allow_barrier(conf); | |
247 | } | |
248 | } | |
249 | ||
9f2c9d12 | 250 | static void raid_end_bio_io(struct r1bio *r1_bio) |
1da177e4 LT |
251 | { |
252 | struct bio *bio = r1_bio->master_bio; | |
253 | ||
4b6d287f N |
254 | /* if nobody has done the final endio yet, do it now */ |
255 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { | |
36a4e1fe N |
256 | pr_debug("raid1: sync end %s on sectors %llu-%llu\n", |
257 | (bio_data_dir(bio) == WRITE) ? "write" : "read", | |
258 | (unsigned long long) bio->bi_sector, | |
259 | (unsigned long long) bio->bi_sector + | |
aa8b57aa | 260 | bio_sectors(bio) - 1); |
4b6d287f | 261 | |
d2eb35ac | 262 | call_bio_endio(r1_bio); |
4b6d287f | 263 | } |
1da177e4 LT |
264 | free_r1bio(r1_bio); |
265 | } | |
266 | ||
267 | /* | |
268 | * Update disk head position estimator based on IRQ completion info. | |
269 | */ | |
9f2c9d12 | 270 | static inline void update_head_pos(int disk, struct r1bio *r1_bio) |
1da177e4 | 271 | { |
e8096360 | 272 | struct r1conf *conf = r1_bio->mddev->private; |
1da177e4 LT |
273 | |
274 | conf->mirrors[disk].head_position = | |
275 | r1_bio->sector + (r1_bio->sectors); | |
276 | } | |
277 | ||
ba3ae3be NK |
278 | /* |
279 | * Find the disk number which triggered given bio | |
280 | */ | |
9f2c9d12 | 281 | static int find_bio_disk(struct r1bio *r1_bio, struct bio *bio) |
ba3ae3be NK |
282 | { |
283 | int mirror; | |
30194636 N |
284 | struct r1conf *conf = r1_bio->mddev->private; |
285 | int raid_disks = conf->raid_disks; | |
ba3ae3be | 286 | |
8f19ccb2 | 287 | for (mirror = 0; mirror < raid_disks * 2; mirror++) |
ba3ae3be NK |
288 | if (r1_bio->bios[mirror] == bio) |
289 | break; | |
290 | ||
8f19ccb2 | 291 | BUG_ON(mirror == raid_disks * 2); |
ba3ae3be NK |
292 | update_head_pos(mirror, r1_bio); |
293 | ||
294 | return mirror; | |
295 | } | |
296 | ||
6712ecf8 | 297 | static void raid1_end_read_request(struct bio *bio, int error) |
1da177e4 LT |
298 | { |
299 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
9f2c9d12 | 300 | struct r1bio *r1_bio = bio->bi_private; |
1da177e4 | 301 | int mirror; |
e8096360 | 302 | struct r1conf *conf = r1_bio->mddev->private; |
1da177e4 | 303 | |
1da177e4 LT |
304 | mirror = r1_bio->read_disk; |
305 | /* | |
306 | * this branch is our 'one mirror IO has finished' event handler: | |
307 | */ | |
ddaf22ab N |
308 | update_head_pos(mirror, r1_bio); |
309 | ||
dd00a99e N |
310 | if (uptodate) |
311 | set_bit(R1BIO_Uptodate, &r1_bio->state); | |
312 | else { | |
313 | /* If all other devices have failed, we want to return | |
314 | * the error upwards rather than fail the last device. | |
315 | * Here we redefine "uptodate" to mean "Don't want to retry" | |
1da177e4 | 316 | */ |
dd00a99e N |
317 | unsigned long flags; |
318 | spin_lock_irqsave(&conf->device_lock, flags); | |
319 | if (r1_bio->mddev->degraded == conf->raid_disks || | |
320 | (r1_bio->mddev->degraded == conf->raid_disks-1 && | |
321 | !test_bit(Faulty, &conf->mirrors[mirror].rdev->flags))) | |
322 | uptodate = 1; | |
323 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
324 | } | |
1da177e4 | 325 | |
7ad4d4a6 | 326 | if (uptodate) { |
1da177e4 | 327 | raid_end_bio_io(r1_bio); |
7ad4d4a6 N |
328 | rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev); |
329 | } else { | |
1da177e4 LT |
330 | /* |
331 | * oops, read error: | |
332 | */ | |
333 | char b[BDEVNAME_SIZE]; | |
8bda470e CD |
334 | printk_ratelimited( |
335 | KERN_ERR "md/raid1:%s: %s: " | |
336 | "rescheduling sector %llu\n", | |
337 | mdname(conf->mddev), | |
338 | bdevname(conf->mirrors[mirror].rdev->bdev, | |
339 | b), | |
340 | (unsigned long long)r1_bio->sector); | |
d2eb35ac | 341 | set_bit(R1BIO_ReadError, &r1_bio->state); |
1da177e4 | 342 | reschedule_retry(r1_bio); |
7ad4d4a6 | 343 | /* don't drop the reference on read_disk yet */ |
1da177e4 | 344 | } |
1da177e4 LT |
345 | } |
346 | ||
9f2c9d12 | 347 | static void close_write(struct r1bio *r1_bio) |
cd5ff9a1 N |
348 | { |
349 | /* it really is the end of this request */ | |
350 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { | |
351 | /* free extra copy of the data pages */ | |
352 | int i = r1_bio->behind_page_count; | |
353 | while (i--) | |
354 | safe_put_page(r1_bio->behind_bvecs[i].bv_page); | |
355 | kfree(r1_bio->behind_bvecs); | |
356 | r1_bio->behind_bvecs = NULL; | |
357 | } | |
358 | /* clear the bitmap if all writes complete successfully */ | |
359 | bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector, | |
360 | r1_bio->sectors, | |
361 | !test_bit(R1BIO_Degraded, &r1_bio->state), | |
362 | test_bit(R1BIO_BehindIO, &r1_bio->state)); | |
363 | md_write_end(r1_bio->mddev); | |
364 | } | |
365 | ||
9f2c9d12 | 366 | static void r1_bio_write_done(struct r1bio *r1_bio) |
4e78064f | 367 | { |
cd5ff9a1 N |
368 | if (!atomic_dec_and_test(&r1_bio->remaining)) |
369 | return; | |
370 | ||
371 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
372 | reschedule_retry(r1_bio); | |
373 | else { | |
374 | close_write(r1_bio); | |
4367af55 N |
375 | if (test_bit(R1BIO_MadeGood, &r1_bio->state)) |
376 | reschedule_retry(r1_bio); | |
377 | else | |
378 | raid_end_bio_io(r1_bio); | |
4e78064f N |
379 | } |
380 | } | |
381 | ||
6712ecf8 | 382 | static void raid1_end_write_request(struct bio *bio, int error) |
1da177e4 LT |
383 | { |
384 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
9f2c9d12 | 385 | struct r1bio *r1_bio = bio->bi_private; |
a9701a30 | 386 | int mirror, behind = test_bit(R1BIO_BehindIO, &r1_bio->state); |
e8096360 | 387 | struct r1conf *conf = r1_bio->mddev->private; |
04b857f7 | 388 | struct bio *to_put = NULL; |
1da177e4 | 389 | |
ba3ae3be | 390 | mirror = find_bio_disk(r1_bio, bio); |
1da177e4 | 391 | |
e9c7469b TH |
392 | /* |
393 | * 'one mirror IO has finished' event handler: | |
394 | */ | |
e9c7469b | 395 | if (!uptodate) { |
cd5ff9a1 N |
396 | set_bit(WriteErrorSeen, |
397 | &conf->mirrors[mirror].rdev->flags); | |
19d67169 N |
398 | if (!test_and_set_bit(WantReplacement, |
399 | &conf->mirrors[mirror].rdev->flags)) | |
400 | set_bit(MD_RECOVERY_NEEDED, & | |
401 | conf->mddev->recovery); | |
402 | ||
cd5ff9a1 | 403 | set_bit(R1BIO_WriteError, &r1_bio->state); |
4367af55 | 404 | } else { |
1da177e4 | 405 | /* |
e9c7469b TH |
406 | * Set R1BIO_Uptodate in our master bio, so that we |
407 | * will return a good error code for to the higher | |
408 | * levels even if IO on some other mirrored buffer | |
409 | * fails. | |
410 | * | |
411 | * The 'master' represents the composite IO operation | |
412 | * to user-side. So if something waits for IO, then it | |
413 | * will wait for the 'master' bio. | |
1da177e4 | 414 | */ |
4367af55 N |
415 | sector_t first_bad; |
416 | int bad_sectors; | |
417 | ||
cd5ff9a1 N |
418 | r1_bio->bios[mirror] = NULL; |
419 | to_put = bio; | |
e9c7469b TH |
420 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
421 | ||
4367af55 N |
422 | /* Maybe we can clear some bad blocks. */ |
423 | if (is_badblock(conf->mirrors[mirror].rdev, | |
424 | r1_bio->sector, r1_bio->sectors, | |
425 | &first_bad, &bad_sectors)) { | |
426 | r1_bio->bios[mirror] = IO_MADE_GOOD; | |
427 | set_bit(R1BIO_MadeGood, &r1_bio->state); | |
428 | } | |
429 | } | |
430 | ||
e9c7469b TH |
431 | if (behind) { |
432 | if (test_bit(WriteMostly, &conf->mirrors[mirror].rdev->flags)) | |
433 | atomic_dec(&r1_bio->behind_remaining); | |
434 | ||
435 | /* | |
436 | * In behind mode, we ACK the master bio once the I/O | |
437 | * has safely reached all non-writemostly | |
438 | * disks. Setting the Returned bit ensures that this | |
439 | * gets done only once -- we don't ever want to return | |
440 | * -EIO here, instead we'll wait | |
441 | */ | |
442 | if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) && | |
443 | test_bit(R1BIO_Uptodate, &r1_bio->state)) { | |
444 | /* Maybe we can return now */ | |
445 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { | |
446 | struct bio *mbio = r1_bio->master_bio; | |
36a4e1fe N |
447 | pr_debug("raid1: behind end write sectors" |
448 | " %llu-%llu\n", | |
449 | (unsigned long long) mbio->bi_sector, | |
450 | (unsigned long long) mbio->bi_sector + | |
aa8b57aa | 451 | bio_sectors(mbio) - 1); |
d2eb35ac | 452 | call_bio_endio(r1_bio); |
4b6d287f N |
453 | } |
454 | } | |
455 | } | |
4367af55 N |
456 | if (r1_bio->bios[mirror] == NULL) |
457 | rdev_dec_pending(conf->mirrors[mirror].rdev, | |
458 | conf->mddev); | |
e9c7469b | 459 | |
1da177e4 | 460 | /* |
1da177e4 LT |
461 | * Let's see if all mirrored write operations have finished |
462 | * already. | |
463 | */ | |
af6d7b76 | 464 | r1_bio_write_done(r1_bio); |
c70810b3 | 465 | |
04b857f7 N |
466 | if (to_put) |
467 | bio_put(to_put); | |
1da177e4 LT |
468 | } |
469 | ||
470 | ||
471 | /* | |
472 | * This routine returns the disk from which the requested read should | |
473 | * be done. There is a per-array 'next expected sequential IO' sector | |
474 | * number - if this matches on the next IO then we use the last disk. | |
475 | * There is also a per-disk 'last know head position' sector that is | |
476 | * maintained from IRQ contexts, both the normal and the resync IO | |
477 | * completion handlers update this position correctly. If there is no | |
478 | * perfect sequential match then we pick the disk whose head is closest. | |
479 | * | |
480 | * If there are 2 mirrors in the same 2 devices, performance degrades | |
481 | * because position is mirror, not device based. | |
482 | * | |
483 | * The rdev for the device selected will have nr_pending incremented. | |
484 | */ | |
e8096360 | 485 | static int read_balance(struct r1conf *conf, struct r1bio *r1_bio, int *max_sectors) |
1da177e4 | 486 | { |
af3a2cd6 | 487 | const sector_t this_sector = r1_bio->sector; |
d2eb35ac N |
488 | int sectors; |
489 | int best_good_sectors; | |
9dedf603 SL |
490 | int best_disk, best_dist_disk, best_pending_disk; |
491 | int has_nonrot_disk; | |
be4d3280 | 492 | int disk; |
76073054 | 493 | sector_t best_dist; |
9dedf603 | 494 | unsigned int min_pending; |
3cb03002 | 495 | struct md_rdev *rdev; |
f3ac8bf7 | 496 | int choose_first; |
12cee5a8 | 497 | int choose_next_idle; |
1da177e4 LT |
498 | |
499 | rcu_read_lock(); | |
500 | /* | |
8ddf9efe | 501 | * Check if we can balance. We can balance on the whole |
1da177e4 LT |
502 | * device if no resync is going on, or below the resync window. |
503 | * We take the first readable disk when above the resync window. | |
504 | */ | |
505 | retry: | |
d2eb35ac | 506 | sectors = r1_bio->sectors; |
76073054 | 507 | best_disk = -1; |
9dedf603 | 508 | best_dist_disk = -1; |
76073054 | 509 | best_dist = MaxSector; |
9dedf603 SL |
510 | best_pending_disk = -1; |
511 | min_pending = UINT_MAX; | |
d2eb35ac | 512 | best_good_sectors = 0; |
9dedf603 | 513 | has_nonrot_disk = 0; |
12cee5a8 | 514 | choose_next_idle = 0; |
d2eb35ac | 515 | |
1da177e4 | 516 | if (conf->mddev->recovery_cp < MaxSector && |
be4d3280 | 517 | (this_sector + sectors >= conf->next_resync)) |
f3ac8bf7 | 518 | choose_first = 1; |
be4d3280 | 519 | else |
f3ac8bf7 | 520 | choose_first = 0; |
1da177e4 | 521 | |
be4d3280 | 522 | for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) { |
76073054 | 523 | sector_t dist; |
d2eb35ac N |
524 | sector_t first_bad; |
525 | int bad_sectors; | |
9dedf603 | 526 | unsigned int pending; |
12cee5a8 | 527 | bool nonrot; |
d2eb35ac | 528 | |
f3ac8bf7 N |
529 | rdev = rcu_dereference(conf->mirrors[disk].rdev); |
530 | if (r1_bio->bios[disk] == IO_BLOCKED | |
531 | || rdev == NULL | |
6b740b8d | 532 | || test_bit(Unmerged, &rdev->flags) |
76073054 | 533 | || test_bit(Faulty, &rdev->flags)) |
f3ac8bf7 | 534 | continue; |
76073054 N |
535 | if (!test_bit(In_sync, &rdev->flags) && |
536 | rdev->recovery_offset < this_sector + sectors) | |
1da177e4 | 537 | continue; |
76073054 N |
538 | if (test_bit(WriteMostly, &rdev->flags)) { |
539 | /* Don't balance among write-mostly, just | |
540 | * use the first as a last resort */ | |
307729c8 N |
541 | if (best_disk < 0) { |
542 | if (is_badblock(rdev, this_sector, sectors, | |
543 | &first_bad, &bad_sectors)) { | |
544 | if (first_bad < this_sector) | |
545 | /* Cannot use this */ | |
546 | continue; | |
547 | best_good_sectors = first_bad - this_sector; | |
548 | } else | |
549 | best_good_sectors = sectors; | |
76073054 | 550 | best_disk = disk; |
307729c8 | 551 | } |
76073054 N |
552 | continue; |
553 | } | |
554 | /* This is a reasonable device to use. It might | |
555 | * even be best. | |
556 | */ | |
d2eb35ac N |
557 | if (is_badblock(rdev, this_sector, sectors, |
558 | &first_bad, &bad_sectors)) { | |
559 | if (best_dist < MaxSector) | |
560 | /* already have a better device */ | |
561 | continue; | |
562 | if (first_bad <= this_sector) { | |
563 | /* cannot read here. If this is the 'primary' | |
564 | * device, then we must not read beyond | |
565 | * bad_sectors from another device.. | |
566 | */ | |
567 | bad_sectors -= (this_sector - first_bad); | |
568 | if (choose_first && sectors > bad_sectors) | |
569 | sectors = bad_sectors; | |
570 | if (best_good_sectors > sectors) | |
571 | best_good_sectors = sectors; | |
572 | ||
573 | } else { | |
574 | sector_t good_sectors = first_bad - this_sector; | |
575 | if (good_sectors > best_good_sectors) { | |
576 | best_good_sectors = good_sectors; | |
577 | best_disk = disk; | |
578 | } | |
579 | if (choose_first) | |
580 | break; | |
581 | } | |
582 | continue; | |
583 | } else | |
584 | best_good_sectors = sectors; | |
585 | ||
12cee5a8 SL |
586 | nonrot = blk_queue_nonrot(bdev_get_queue(rdev->bdev)); |
587 | has_nonrot_disk |= nonrot; | |
9dedf603 | 588 | pending = atomic_read(&rdev->nr_pending); |
76073054 | 589 | dist = abs(this_sector - conf->mirrors[disk].head_position); |
12cee5a8 | 590 | if (choose_first) { |
76073054 | 591 | best_disk = disk; |
1da177e4 LT |
592 | break; |
593 | } | |
12cee5a8 SL |
594 | /* Don't change to another disk for sequential reads */ |
595 | if (conf->mirrors[disk].next_seq_sect == this_sector | |
596 | || dist == 0) { | |
597 | int opt_iosize = bdev_io_opt(rdev->bdev) >> 9; | |
598 | struct raid1_info *mirror = &conf->mirrors[disk]; | |
599 | ||
600 | best_disk = disk; | |
601 | /* | |
602 | * If buffered sequential IO size exceeds optimal | |
603 | * iosize, check if there is idle disk. If yes, choose | |
604 | * the idle disk. read_balance could already choose an | |
605 | * idle disk before noticing it's a sequential IO in | |
606 | * this disk. This doesn't matter because this disk | |
607 | * will idle, next time it will be utilized after the | |
608 | * first disk has IO size exceeds optimal iosize. In | |
609 | * this way, iosize of the first disk will be optimal | |
610 | * iosize at least. iosize of the second disk might be | |
611 | * small, but not a big deal since when the second disk | |
612 | * starts IO, the first disk is likely still busy. | |
613 | */ | |
614 | if (nonrot && opt_iosize > 0 && | |
615 | mirror->seq_start != MaxSector && | |
616 | mirror->next_seq_sect > opt_iosize && | |
617 | mirror->next_seq_sect - opt_iosize >= | |
618 | mirror->seq_start) { | |
619 | choose_next_idle = 1; | |
620 | continue; | |
621 | } | |
622 | break; | |
623 | } | |
624 | /* If device is idle, use it */ | |
625 | if (pending == 0) { | |
626 | best_disk = disk; | |
627 | break; | |
628 | } | |
629 | ||
630 | if (choose_next_idle) | |
631 | continue; | |
9dedf603 SL |
632 | |
633 | if (min_pending > pending) { | |
634 | min_pending = pending; | |
635 | best_pending_disk = disk; | |
636 | } | |
637 | ||
76073054 N |
638 | if (dist < best_dist) { |
639 | best_dist = dist; | |
9dedf603 | 640 | best_dist_disk = disk; |
1da177e4 | 641 | } |
f3ac8bf7 | 642 | } |
1da177e4 | 643 | |
9dedf603 SL |
644 | /* |
645 | * If all disks are rotational, choose the closest disk. If any disk is | |
646 | * non-rotational, choose the disk with less pending request even the | |
647 | * disk is rotational, which might/might not be optimal for raids with | |
648 | * mixed ratation/non-rotational disks depending on workload. | |
649 | */ | |
650 | if (best_disk == -1) { | |
651 | if (has_nonrot_disk) | |
652 | best_disk = best_pending_disk; | |
653 | else | |
654 | best_disk = best_dist_disk; | |
655 | } | |
656 | ||
76073054 N |
657 | if (best_disk >= 0) { |
658 | rdev = rcu_dereference(conf->mirrors[best_disk].rdev); | |
8ddf9efe N |
659 | if (!rdev) |
660 | goto retry; | |
661 | atomic_inc(&rdev->nr_pending); | |
76073054 | 662 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
663 | /* cannot risk returning a device that failed |
664 | * before we inc'ed nr_pending | |
665 | */ | |
03c902e1 | 666 | rdev_dec_pending(rdev, conf->mddev); |
1da177e4 LT |
667 | goto retry; |
668 | } | |
d2eb35ac | 669 | sectors = best_good_sectors; |
12cee5a8 SL |
670 | |
671 | if (conf->mirrors[best_disk].next_seq_sect != this_sector) | |
672 | conf->mirrors[best_disk].seq_start = this_sector; | |
673 | ||
be4d3280 | 674 | conf->mirrors[best_disk].next_seq_sect = this_sector + sectors; |
1da177e4 LT |
675 | } |
676 | rcu_read_unlock(); | |
d2eb35ac | 677 | *max_sectors = sectors; |
1da177e4 | 678 | |
76073054 | 679 | return best_disk; |
1da177e4 LT |
680 | } |
681 | ||
6b740b8d N |
682 | static int raid1_mergeable_bvec(struct request_queue *q, |
683 | struct bvec_merge_data *bvm, | |
684 | struct bio_vec *biovec) | |
685 | { | |
686 | struct mddev *mddev = q->queuedata; | |
687 | struct r1conf *conf = mddev->private; | |
688 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); | |
689 | int max = biovec->bv_len; | |
690 | ||
691 | if (mddev->merge_check_needed) { | |
692 | int disk; | |
693 | rcu_read_lock(); | |
694 | for (disk = 0; disk < conf->raid_disks * 2; disk++) { | |
695 | struct md_rdev *rdev = rcu_dereference( | |
696 | conf->mirrors[disk].rdev); | |
697 | if (rdev && !test_bit(Faulty, &rdev->flags)) { | |
698 | struct request_queue *q = | |
699 | bdev_get_queue(rdev->bdev); | |
700 | if (q->merge_bvec_fn) { | |
701 | bvm->bi_sector = sector + | |
702 | rdev->data_offset; | |
703 | bvm->bi_bdev = rdev->bdev; | |
704 | max = min(max, q->merge_bvec_fn( | |
705 | q, bvm, biovec)); | |
706 | } | |
707 | } | |
708 | } | |
709 | rcu_read_unlock(); | |
710 | } | |
711 | return max; | |
712 | ||
713 | } | |
714 | ||
fd01b88c | 715 | int md_raid1_congested(struct mddev *mddev, int bits) |
0d129228 | 716 | { |
e8096360 | 717 | struct r1conf *conf = mddev->private; |
0d129228 N |
718 | int i, ret = 0; |
719 | ||
34db0cd6 N |
720 | if ((bits & (1 << BDI_async_congested)) && |
721 | conf->pending_count >= max_queued_requests) | |
722 | return 1; | |
723 | ||
0d129228 | 724 | rcu_read_lock(); |
f53e29fc | 725 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3cb03002 | 726 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
0d129228 | 727 | if (rdev && !test_bit(Faulty, &rdev->flags)) { |
165125e1 | 728 | struct request_queue *q = bdev_get_queue(rdev->bdev); |
0d129228 | 729 | |
1ed7242e JB |
730 | BUG_ON(!q); |
731 | ||
0d129228 N |
732 | /* Note the '|| 1' - when read_balance prefers |
733 | * non-congested targets, it can be removed | |
734 | */ | |
91a9e99d | 735 | if ((bits & (1<<BDI_async_congested)) || 1) |
0d129228 N |
736 | ret |= bdi_congested(&q->backing_dev_info, bits); |
737 | else | |
738 | ret &= bdi_congested(&q->backing_dev_info, bits); | |
739 | } | |
740 | } | |
741 | rcu_read_unlock(); | |
742 | return ret; | |
743 | } | |
1ed7242e | 744 | EXPORT_SYMBOL_GPL(md_raid1_congested); |
0d129228 | 745 | |
1ed7242e JB |
746 | static int raid1_congested(void *data, int bits) |
747 | { | |
fd01b88c | 748 | struct mddev *mddev = data; |
1ed7242e JB |
749 | |
750 | return mddev_congested(mddev, bits) || | |
751 | md_raid1_congested(mddev, bits); | |
752 | } | |
0d129228 | 753 | |
e8096360 | 754 | static void flush_pending_writes(struct r1conf *conf) |
a35e63ef N |
755 | { |
756 | /* Any writes that have been queued but are awaiting | |
757 | * bitmap updates get flushed here. | |
a35e63ef | 758 | */ |
a35e63ef N |
759 | spin_lock_irq(&conf->device_lock); |
760 | ||
761 | if (conf->pending_bio_list.head) { | |
762 | struct bio *bio; | |
763 | bio = bio_list_get(&conf->pending_bio_list); | |
34db0cd6 | 764 | conf->pending_count = 0; |
a35e63ef N |
765 | spin_unlock_irq(&conf->device_lock); |
766 | /* flush any pending bitmap writes to | |
767 | * disk before proceeding w/ I/O */ | |
768 | bitmap_unplug(conf->mddev->bitmap); | |
34db0cd6 | 769 | wake_up(&conf->wait_barrier); |
a35e63ef N |
770 | |
771 | while (bio) { /* submit pending writes */ | |
772 | struct bio *next = bio->bi_next; | |
773 | bio->bi_next = NULL; | |
2ff8cc2c SL |
774 | if (unlikely((bio->bi_rw & REQ_DISCARD) && |
775 | !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) | |
776 | /* Just ignore it */ | |
777 | bio_endio(bio, 0); | |
778 | else | |
779 | generic_make_request(bio); | |
a35e63ef N |
780 | bio = next; |
781 | } | |
a35e63ef N |
782 | } else |
783 | spin_unlock_irq(&conf->device_lock); | |
7eaceacc JA |
784 | } |
785 | ||
17999be4 N |
786 | /* Barriers.... |
787 | * Sometimes we need to suspend IO while we do something else, | |
788 | * either some resync/recovery, or reconfigure the array. | |
789 | * To do this we raise a 'barrier'. | |
790 | * The 'barrier' is a counter that can be raised multiple times | |
791 | * to count how many activities are happening which preclude | |
792 | * normal IO. | |
793 | * We can only raise the barrier if there is no pending IO. | |
794 | * i.e. if nr_pending == 0. | |
795 | * We choose only to raise the barrier if no-one is waiting for the | |
796 | * barrier to go down. This means that as soon as an IO request | |
797 | * is ready, no other operations which require a barrier will start | |
798 | * until the IO request has had a chance. | |
799 | * | |
800 | * So: regular IO calls 'wait_barrier'. When that returns there | |
801 | * is no backgroup IO happening, It must arrange to call | |
802 | * allow_barrier when it has finished its IO. | |
803 | * backgroup IO calls must call raise_barrier. Once that returns | |
804 | * there is no normal IO happeing. It must arrange to call | |
805 | * lower_barrier when the particular background IO completes. | |
1da177e4 LT |
806 | */ |
807 | #define RESYNC_DEPTH 32 | |
808 | ||
e8096360 | 809 | static void raise_barrier(struct r1conf *conf) |
1da177e4 LT |
810 | { |
811 | spin_lock_irq(&conf->resync_lock); | |
17999be4 N |
812 | |
813 | /* Wait until no block IO is waiting */ | |
814 | wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting, | |
eed8c02e | 815 | conf->resync_lock); |
17999be4 N |
816 | |
817 | /* block any new IO from starting */ | |
818 | conf->barrier++; | |
819 | ||
046abeed | 820 | /* Now wait for all pending IO to complete */ |
17999be4 N |
821 | wait_event_lock_irq(conf->wait_barrier, |
822 | !conf->nr_pending && conf->barrier < RESYNC_DEPTH, | |
eed8c02e | 823 | conf->resync_lock); |
17999be4 N |
824 | |
825 | spin_unlock_irq(&conf->resync_lock); | |
826 | } | |
827 | ||
e8096360 | 828 | static void lower_barrier(struct r1conf *conf) |
17999be4 N |
829 | { |
830 | unsigned long flags; | |
709ae487 | 831 | BUG_ON(conf->barrier <= 0); |
17999be4 N |
832 | spin_lock_irqsave(&conf->resync_lock, flags); |
833 | conf->barrier--; | |
834 | spin_unlock_irqrestore(&conf->resync_lock, flags); | |
835 | wake_up(&conf->wait_barrier); | |
836 | } | |
837 | ||
e8096360 | 838 | static void wait_barrier(struct r1conf *conf) |
17999be4 N |
839 | { |
840 | spin_lock_irq(&conf->resync_lock); | |
841 | if (conf->barrier) { | |
842 | conf->nr_waiting++; | |
d6b42dcb N |
843 | /* Wait for the barrier to drop. |
844 | * However if there are already pending | |
845 | * requests (preventing the barrier from | |
846 | * rising completely), and the | |
847 | * pre-process bio queue isn't empty, | |
848 | * then don't wait, as we need to empty | |
849 | * that queue to get the nr_pending | |
850 | * count down. | |
851 | */ | |
852 | wait_event_lock_irq(conf->wait_barrier, | |
853 | !conf->barrier || | |
854 | (conf->nr_pending && | |
855 | current->bio_list && | |
856 | !bio_list_empty(current->bio_list)), | |
eed8c02e | 857 | conf->resync_lock); |
17999be4 | 858 | conf->nr_waiting--; |
1da177e4 | 859 | } |
17999be4 | 860 | conf->nr_pending++; |
1da177e4 LT |
861 | spin_unlock_irq(&conf->resync_lock); |
862 | } | |
863 | ||
e8096360 | 864 | static void allow_barrier(struct r1conf *conf) |
17999be4 N |
865 | { |
866 | unsigned long flags; | |
867 | spin_lock_irqsave(&conf->resync_lock, flags); | |
868 | conf->nr_pending--; | |
869 | spin_unlock_irqrestore(&conf->resync_lock, flags); | |
870 | wake_up(&conf->wait_barrier); | |
871 | } | |
872 | ||
e8096360 | 873 | static void freeze_array(struct r1conf *conf) |
ddaf22ab N |
874 | { |
875 | /* stop syncio and normal IO and wait for everything to | |
876 | * go quite. | |
877 | * We increment barrier and nr_waiting, and then | |
1c830532 N |
878 | * wait until nr_pending match nr_queued+1 |
879 | * This is called in the context of one normal IO request | |
880 | * that has failed. Thus any sync request that might be pending | |
881 | * will be blocked by nr_pending, and we need to wait for | |
882 | * pending IO requests to complete or be queued for re-try. | |
883 | * Thus the number queued (nr_queued) plus this request (1) | |
884 | * must match the number of pending IOs (nr_pending) before | |
885 | * we continue. | |
ddaf22ab N |
886 | */ |
887 | spin_lock_irq(&conf->resync_lock); | |
888 | conf->barrier++; | |
889 | conf->nr_waiting++; | |
eed8c02e LC |
890 | wait_event_lock_irq_cmd(conf->wait_barrier, |
891 | conf->nr_pending == conf->nr_queued+1, | |
892 | conf->resync_lock, | |
893 | flush_pending_writes(conf)); | |
ddaf22ab N |
894 | spin_unlock_irq(&conf->resync_lock); |
895 | } | |
e8096360 | 896 | static void unfreeze_array(struct r1conf *conf) |
ddaf22ab N |
897 | { |
898 | /* reverse the effect of the freeze */ | |
899 | spin_lock_irq(&conf->resync_lock); | |
900 | conf->barrier--; | |
901 | conf->nr_waiting--; | |
902 | wake_up(&conf->wait_barrier); | |
903 | spin_unlock_irq(&conf->resync_lock); | |
904 | } | |
905 | ||
17999be4 | 906 | |
4e78064f | 907 | /* duplicate the data pages for behind I/O |
4e78064f | 908 | */ |
9f2c9d12 | 909 | static void alloc_behind_pages(struct bio *bio, struct r1bio *r1_bio) |
4b6d287f N |
910 | { |
911 | int i; | |
912 | struct bio_vec *bvec; | |
2ca68f5e | 913 | struct bio_vec *bvecs = kzalloc(bio->bi_vcnt * sizeof(struct bio_vec), |
4b6d287f | 914 | GFP_NOIO); |
2ca68f5e | 915 | if (unlikely(!bvecs)) |
af6d7b76 | 916 | return; |
4b6d287f | 917 | |
cb34e057 | 918 | bio_for_each_segment_all(bvec, bio, i) { |
2ca68f5e N |
919 | bvecs[i] = *bvec; |
920 | bvecs[i].bv_page = alloc_page(GFP_NOIO); | |
921 | if (unlikely(!bvecs[i].bv_page)) | |
4b6d287f | 922 | goto do_sync_io; |
2ca68f5e N |
923 | memcpy(kmap(bvecs[i].bv_page) + bvec->bv_offset, |
924 | kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len); | |
925 | kunmap(bvecs[i].bv_page); | |
4b6d287f N |
926 | kunmap(bvec->bv_page); |
927 | } | |
2ca68f5e | 928 | r1_bio->behind_bvecs = bvecs; |
af6d7b76 N |
929 | r1_bio->behind_page_count = bio->bi_vcnt; |
930 | set_bit(R1BIO_BehindIO, &r1_bio->state); | |
931 | return; | |
4b6d287f N |
932 | |
933 | do_sync_io: | |
af6d7b76 | 934 | for (i = 0; i < bio->bi_vcnt; i++) |
2ca68f5e N |
935 | if (bvecs[i].bv_page) |
936 | put_page(bvecs[i].bv_page); | |
937 | kfree(bvecs); | |
36a4e1fe | 938 | pr_debug("%dB behind alloc failed, doing sync I/O\n", bio->bi_size); |
4b6d287f N |
939 | } |
940 | ||
f54a9d0e N |
941 | struct raid1_plug_cb { |
942 | struct blk_plug_cb cb; | |
943 | struct bio_list pending; | |
944 | int pending_cnt; | |
945 | }; | |
946 | ||
947 | static void raid1_unplug(struct blk_plug_cb *cb, bool from_schedule) | |
948 | { | |
949 | struct raid1_plug_cb *plug = container_of(cb, struct raid1_plug_cb, | |
950 | cb); | |
951 | struct mddev *mddev = plug->cb.data; | |
952 | struct r1conf *conf = mddev->private; | |
953 | struct bio *bio; | |
954 | ||
874807a8 | 955 | if (from_schedule || current->bio_list) { |
f54a9d0e N |
956 | spin_lock_irq(&conf->device_lock); |
957 | bio_list_merge(&conf->pending_bio_list, &plug->pending); | |
958 | conf->pending_count += plug->pending_cnt; | |
959 | spin_unlock_irq(&conf->device_lock); | |
ee0b0244 | 960 | wake_up(&conf->wait_barrier); |
f54a9d0e N |
961 | md_wakeup_thread(mddev->thread); |
962 | kfree(plug); | |
963 | return; | |
964 | } | |
965 | ||
966 | /* we aren't scheduling, so we can do the write-out directly. */ | |
967 | bio = bio_list_get(&plug->pending); | |
968 | bitmap_unplug(mddev->bitmap); | |
969 | wake_up(&conf->wait_barrier); | |
970 | ||
971 | while (bio) { /* submit pending writes */ | |
972 | struct bio *next = bio->bi_next; | |
973 | bio->bi_next = NULL; | |
974 | generic_make_request(bio); | |
975 | bio = next; | |
976 | } | |
977 | kfree(plug); | |
978 | } | |
979 | ||
b4fdcb02 | 980 | static void make_request(struct mddev *mddev, struct bio * bio) |
1da177e4 | 981 | { |
e8096360 | 982 | struct r1conf *conf = mddev->private; |
0eaf822c | 983 | struct raid1_info *mirror; |
9f2c9d12 | 984 | struct r1bio *r1_bio; |
1da177e4 | 985 | struct bio *read_bio; |
1f68f0c4 | 986 | int i, disks; |
84255d10 | 987 | struct bitmap *bitmap; |
191ea9b2 | 988 | unsigned long flags; |
a362357b | 989 | const int rw = bio_data_dir(bio); |
2c7d46ec | 990 | const unsigned long do_sync = (bio->bi_rw & REQ_SYNC); |
e9c7469b | 991 | const unsigned long do_flush_fua = (bio->bi_rw & (REQ_FLUSH | REQ_FUA)); |
2ff8cc2c SL |
992 | const unsigned long do_discard = (bio->bi_rw |
993 | & (REQ_DISCARD | REQ_SECURE)); | |
c8dc9c65 | 994 | const unsigned long do_same = (bio->bi_rw & REQ_WRITE_SAME); |
3cb03002 | 995 | struct md_rdev *blocked_rdev; |
f54a9d0e N |
996 | struct blk_plug_cb *cb; |
997 | struct raid1_plug_cb *plug = NULL; | |
1f68f0c4 N |
998 | int first_clone; |
999 | int sectors_handled; | |
1000 | int max_sectors; | |
191ea9b2 | 1001 | |
1da177e4 LT |
1002 | /* |
1003 | * Register the new request and wait if the reconstruction | |
1004 | * thread has put up a bar for new requests. | |
1005 | * Continue immediately if no resync is active currently. | |
1006 | */ | |
62de608d | 1007 | |
3d310eb7 N |
1008 | md_write_start(mddev, bio); /* wait on superblock update early */ |
1009 | ||
6eef4b21 | 1010 | if (bio_data_dir(bio) == WRITE && |
f73a1c7d | 1011 | bio_end_sector(bio) > mddev->suspend_lo && |
6eef4b21 N |
1012 | bio->bi_sector < mddev->suspend_hi) { |
1013 | /* As the suspend_* range is controlled by | |
1014 | * userspace, we want an interruptible | |
1015 | * wait. | |
1016 | */ | |
1017 | DEFINE_WAIT(w); | |
1018 | for (;;) { | |
1019 | flush_signals(current); | |
1020 | prepare_to_wait(&conf->wait_barrier, | |
1021 | &w, TASK_INTERRUPTIBLE); | |
f73a1c7d | 1022 | if (bio_end_sector(bio) <= mddev->suspend_lo || |
6eef4b21 N |
1023 | bio->bi_sector >= mddev->suspend_hi) |
1024 | break; | |
1025 | schedule(); | |
1026 | } | |
1027 | finish_wait(&conf->wait_barrier, &w); | |
1028 | } | |
62de608d | 1029 | |
17999be4 | 1030 | wait_barrier(conf); |
1da177e4 | 1031 | |
84255d10 N |
1032 | bitmap = mddev->bitmap; |
1033 | ||
1da177e4 LT |
1034 | /* |
1035 | * make_request() can abort the operation when READA is being | |
1036 | * used and no empty request is available. | |
1037 | * | |
1038 | */ | |
1039 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
1040 | ||
1041 | r1_bio->master_bio = bio; | |
aa8b57aa | 1042 | r1_bio->sectors = bio_sectors(bio); |
191ea9b2 | 1043 | r1_bio->state = 0; |
1da177e4 LT |
1044 | r1_bio->mddev = mddev; |
1045 | r1_bio->sector = bio->bi_sector; | |
1046 | ||
d2eb35ac N |
1047 | /* We might need to issue multiple reads to different |
1048 | * devices if there are bad blocks around, so we keep | |
1049 | * track of the number of reads in bio->bi_phys_segments. | |
1050 | * If this is 0, there is only one r1_bio and no locking | |
1051 | * will be needed when requests complete. If it is | |
1052 | * non-zero, then it is the number of not-completed requests. | |
1053 | */ | |
1054 | bio->bi_phys_segments = 0; | |
1055 | clear_bit(BIO_SEG_VALID, &bio->bi_flags); | |
1056 | ||
a362357b | 1057 | if (rw == READ) { |
1da177e4 LT |
1058 | /* |
1059 | * read balancing logic: | |
1060 | */ | |
d2eb35ac N |
1061 | int rdisk; |
1062 | ||
1063 | read_again: | |
1064 | rdisk = read_balance(conf, r1_bio, &max_sectors); | |
1da177e4 LT |
1065 | |
1066 | if (rdisk < 0) { | |
1067 | /* couldn't find anywhere to read from */ | |
1068 | raid_end_bio_io(r1_bio); | |
5a7bbad2 | 1069 | return; |
1da177e4 LT |
1070 | } |
1071 | mirror = conf->mirrors + rdisk; | |
1072 | ||
e555190d N |
1073 | if (test_bit(WriteMostly, &mirror->rdev->flags) && |
1074 | bitmap) { | |
1075 | /* Reading from a write-mostly device must | |
1076 | * take care not to over-take any writes | |
1077 | * that are 'behind' | |
1078 | */ | |
1079 | wait_event(bitmap->behind_wait, | |
1080 | atomic_read(&bitmap->behind_writes) == 0); | |
1081 | } | |
1da177e4 LT |
1082 | r1_bio->read_disk = rdisk; |
1083 | ||
a167f663 | 1084 | read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev); |
d2eb35ac N |
1085 | md_trim_bio(read_bio, r1_bio->sector - bio->bi_sector, |
1086 | max_sectors); | |
1da177e4 LT |
1087 | |
1088 | r1_bio->bios[rdisk] = read_bio; | |
1089 | ||
1090 | read_bio->bi_sector = r1_bio->sector + mirror->rdev->data_offset; | |
1091 | read_bio->bi_bdev = mirror->rdev->bdev; | |
1092 | read_bio->bi_end_io = raid1_end_read_request; | |
7b6d91da | 1093 | read_bio->bi_rw = READ | do_sync; |
1da177e4 LT |
1094 | read_bio->bi_private = r1_bio; |
1095 | ||
d2eb35ac N |
1096 | if (max_sectors < r1_bio->sectors) { |
1097 | /* could not read all from this device, so we will | |
1098 | * need another r1_bio. | |
1099 | */ | |
d2eb35ac N |
1100 | |
1101 | sectors_handled = (r1_bio->sector + max_sectors | |
1102 | - bio->bi_sector); | |
1103 | r1_bio->sectors = max_sectors; | |
1104 | spin_lock_irq(&conf->device_lock); | |
1105 | if (bio->bi_phys_segments == 0) | |
1106 | bio->bi_phys_segments = 2; | |
1107 | else | |
1108 | bio->bi_phys_segments++; | |
1109 | spin_unlock_irq(&conf->device_lock); | |
1110 | /* Cannot call generic_make_request directly | |
1111 | * as that will be queued in __make_request | |
1112 | * and subsequent mempool_alloc might block waiting | |
1113 | * for it. So hand bio over to raid1d. | |
1114 | */ | |
1115 | reschedule_retry(r1_bio); | |
1116 | ||
1117 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
1118 | ||
1119 | r1_bio->master_bio = bio; | |
aa8b57aa | 1120 | r1_bio->sectors = bio_sectors(bio) - sectors_handled; |
d2eb35ac N |
1121 | r1_bio->state = 0; |
1122 | r1_bio->mddev = mddev; | |
1123 | r1_bio->sector = bio->bi_sector + sectors_handled; | |
1124 | goto read_again; | |
1125 | } else | |
1126 | generic_make_request(read_bio); | |
5a7bbad2 | 1127 | return; |
1da177e4 LT |
1128 | } |
1129 | ||
1130 | /* | |
1131 | * WRITE: | |
1132 | */ | |
34db0cd6 N |
1133 | if (conf->pending_count >= max_queued_requests) { |
1134 | md_wakeup_thread(mddev->thread); | |
1135 | wait_event(conf->wait_barrier, | |
1136 | conf->pending_count < max_queued_requests); | |
1137 | } | |
1f68f0c4 | 1138 | /* first select target devices under rcu_lock and |
1da177e4 LT |
1139 | * inc refcount on their rdev. Record them by setting |
1140 | * bios[x] to bio | |
1f68f0c4 N |
1141 | * If there are known/acknowledged bad blocks on any device on |
1142 | * which we have seen a write error, we want to avoid writing those | |
1143 | * blocks. | |
1144 | * This potentially requires several writes to write around | |
1145 | * the bad blocks. Each set of writes gets it's own r1bio | |
1146 | * with a set of bios attached. | |
1da177e4 | 1147 | */ |
c3b328ac | 1148 | |
8f19ccb2 | 1149 | disks = conf->raid_disks * 2; |
6bfe0b49 DW |
1150 | retry_write: |
1151 | blocked_rdev = NULL; | |
1da177e4 | 1152 | rcu_read_lock(); |
1f68f0c4 | 1153 | max_sectors = r1_bio->sectors; |
1da177e4 | 1154 | for (i = 0; i < disks; i++) { |
3cb03002 | 1155 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
6bfe0b49 DW |
1156 | if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { |
1157 | atomic_inc(&rdev->nr_pending); | |
1158 | blocked_rdev = rdev; | |
1159 | break; | |
1160 | } | |
1f68f0c4 | 1161 | r1_bio->bios[i] = NULL; |
6b740b8d N |
1162 | if (!rdev || test_bit(Faulty, &rdev->flags) |
1163 | || test_bit(Unmerged, &rdev->flags)) { | |
8f19ccb2 N |
1164 | if (i < conf->raid_disks) |
1165 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
1f68f0c4 N |
1166 | continue; |
1167 | } | |
1168 | ||
1169 | atomic_inc(&rdev->nr_pending); | |
1170 | if (test_bit(WriteErrorSeen, &rdev->flags)) { | |
1171 | sector_t first_bad; | |
1172 | int bad_sectors; | |
1173 | int is_bad; | |
1174 | ||
1175 | is_bad = is_badblock(rdev, r1_bio->sector, | |
1176 | max_sectors, | |
1177 | &first_bad, &bad_sectors); | |
1178 | if (is_bad < 0) { | |
1179 | /* mustn't write here until the bad block is | |
1180 | * acknowledged*/ | |
1181 | set_bit(BlockedBadBlocks, &rdev->flags); | |
1182 | blocked_rdev = rdev; | |
1183 | break; | |
1184 | } | |
1185 | if (is_bad && first_bad <= r1_bio->sector) { | |
1186 | /* Cannot write here at all */ | |
1187 | bad_sectors -= (r1_bio->sector - first_bad); | |
1188 | if (bad_sectors < max_sectors) | |
1189 | /* mustn't write more than bad_sectors | |
1190 | * to other devices yet | |
1191 | */ | |
1192 | max_sectors = bad_sectors; | |
03c902e1 | 1193 | rdev_dec_pending(rdev, mddev); |
1f68f0c4 N |
1194 | /* We don't set R1BIO_Degraded as that |
1195 | * only applies if the disk is | |
1196 | * missing, so it might be re-added, | |
1197 | * and we want to know to recover this | |
1198 | * chunk. | |
1199 | * In this case the device is here, | |
1200 | * and the fact that this chunk is not | |
1201 | * in-sync is recorded in the bad | |
1202 | * block log | |
1203 | */ | |
1204 | continue; | |
964147d5 | 1205 | } |
1f68f0c4 N |
1206 | if (is_bad) { |
1207 | int good_sectors = first_bad - r1_bio->sector; | |
1208 | if (good_sectors < max_sectors) | |
1209 | max_sectors = good_sectors; | |
1210 | } | |
1211 | } | |
1212 | r1_bio->bios[i] = bio; | |
1da177e4 LT |
1213 | } |
1214 | rcu_read_unlock(); | |
1215 | ||
6bfe0b49 DW |
1216 | if (unlikely(blocked_rdev)) { |
1217 | /* Wait for this device to become unblocked */ | |
1218 | int j; | |
1219 | ||
1220 | for (j = 0; j < i; j++) | |
1221 | if (r1_bio->bios[j]) | |
1222 | rdev_dec_pending(conf->mirrors[j].rdev, mddev); | |
1f68f0c4 | 1223 | r1_bio->state = 0; |
6bfe0b49 DW |
1224 | allow_barrier(conf); |
1225 | md_wait_for_blocked_rdev(blocked_rdev, mddev); | |
1226 | wait_barrier(conf); | |
1227 | goto retry_write; | |
1228 | } | |
1229 | ||
1f68f0c4 N |
1230 | if (max_sectors < r1_bio->sectors) { |
1231 | /* We are splitting this write into multiple parts, so | |
1232 | * we need to prepare for allocating another r1_bio. | |
1233 | */ | |
1234 | r1_bio->sectors = max_sectors; | |
1235 | spin_lock_irq(&conf->device_lock); | |
1236 | if (bio->bi_phys_segments == 0) | |
1237 | bio->bi_phys_segments = 2; | |
1238 | else | |
1239 | bio->bi_phys_segments++; | |
1240 | spin_unlock_irq(&conf->device_lock); | |
191ea9b2 | 1241 | } |
1f68f0c4 | 1242 | sectors_handled = r1_bio->sector + max_sectors - bio->bi_sector; |
4b6d287f | 1243 | |
4e78064f | 1244 | atomic_set(&r1_bio->remaining, 1); |
4b6d287f | 1245 | atomic_set(&r1_bio->behind_remaining, 0); |
06d91a5f | 1246 | |
1f68f0c4 | 1247 | first_clone = 1; |
1da177e4 LT |
1248 | for (i = 0; i < disks; i++) { |
1249 | struct bio *mbio; | |
1250 | if (!r1_bio->bios[i]) | |
1251 | continue; | |
1252 | ||
a167f663 | 1253 | mbio = bio_clone_mddev(bio, GFP_NOIO, mddev); |
1f68f0c4 N |
1254 | md_trim_bio(mbio, r1_bio->sector - bio->bi_sector, max_sectors); |
1255 | ||
1256 | if (first_clone) { | |
1257 | /* do behind I/O ? | |
1258 | * Not if there are too many, or cannot | |
1259 | * allocate memory, or a reader on WriteMostly | |
1260 | * is waiting for behind writes to flush */ | |
1261 | if (bitmap && | |
1262 | (atomic_read(&bitmap->behind_writes) | |
1263 | < mddev->bitmap_info.max_write_behind) && | |
1264 | !waitqueue_active(&bitmap->behind_wait)) | |
1265 | alloc_behind_pages(mbio, r1_bio); | |
1266 | ||
1267 | bitmap_startwrite(bitmap, r1_bio->sector, | |
1268 | r1_bio->sectors, | |
1269 | test_bit(R1BIO_BehindIO, | |
1270 | &r1_bio->state)); | |
1271 | first_clone = 0; | |
1272 | } | |
2ca68f5e | 1273 | if (r1_bio->behind_bvecs) { |
4b6d287f N |
1274 | struct bio_vec *bvec; |
1275 | int j; | |
1276 | ||
cb34e057 KO |
1277 | /* |
1278 | * We trimmed the bio, so _all is legit | |
4b6d287f | 1279 | */ |
d74c6d51 | 1280 | bio_for_each_segment_all(bvec, mbio, j) |
2ca68f5e | 1281 | bvec->bv_page = r1_bio->behind_bvecs[j].bv_page; |
4b6d287f N |
1282 | if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags)) |
1283 | atomic_inc(&r1_bio->behind_remaining); | |
1284 | } | |
1285 | ||
1f68f0c4 N |
1286 | r1_bio->bios[i] = mbio; |
1287 | ||
1288 | mbio->bi_sector = (r1_bio->sector + | |
1289 | conf->mirrors[i].rdev->data_offset); | |
1290 | mbio->bi_bdev = conf->mirrors[i].rdev->bdev; | |
1291 | mbio->bi_end_io = raid1_end_write_request; | |
c8dc9c65 JL |
1292 | mbio->bi_rw = |
1293 | WRITE | do_flush_fua | do_sync | do_discard | do_same; | |
1f68f0c4 N |
1294 | mbio->bi_private = r1_bio; |
1295 | ||
1da177e4 | 1296 | atomic_inc(&r1_bio->remaining); |
f54a9d0e N |
1297 | |
1298 | cb = blk_check_plugged(raid1_unplug, mddev, sizeof(*plug)); | |
1299 | if (cb) | |
1300 | plug = container_of(cb, struct raid1_plug_cb, cb); | |
1301 | else | |
1302 | plug = NULL; | |
4e78064f | 1303 | spin_lock_irqsave(&conf->device_lock, flags); |
f54a9d0e N |
1304 | if (plug) { |
1305 | bio_list_add(&plug->pending, mbio); | |
1306 | plug->pending_cnt++; | |
1307 | } else { | |
1308 | bio_list_add(&conf->pending_bio_list, mbio); | |
1309 | conf->pending_count++; | |
1310 | } | |
4e78064f | 1311 | spin_unlock_irqrestore(&conf->device_lock, flags); |
f54a9d0e | 1312 | if (!plug) |
b357f04a | 1313 | md_wakeup_thread(mddev->thread); |
1da177e4 | 1314 | } |
079fa166 N |
1315 | /* Mustn't call r1_bio_write_done before this next test, |
1316 | * as it could result in the bio being freed. | |
1317 | */ | |
aa8b57aa | 1318 | if (sectors_handled < bio_sectors(bio)) { |
079fa166 | 1319 | r1_bio_write_done(r1_bio); |
1f68f0c4 N |
1320 | /* We need another r1_bio. It has already been counted |
1321 | * in bio->bi_phys_segments | |
1322 | */ | |
1323 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
1324 | r1_bio->master_bio = bio; | |
aa8b57aa | 1325 | r1_bio->sectors = bio_sectors(bio) - sectors_handled; |
1f68f0c4 N |
1326 | r1_bio->state = 0; |
1327 | r1_bio->mddev = mddev; | |
1328 | r1_bio->sector = bio->bi_sector + sectors_handled; | |
1329 | goto retry_write; | |
1330 | } | |
1331 | ||
079fa166 N |
1332 | r1_bio_write_done(r1_bio); |
1333 | ||
1334 | /* In case raid1d snuck in to freeze_array */ | |
1335 | wake_up(&conf->wait_barrier); | |
1da177e4 LT |
1336 | } |
1337 | ||
fd01b88c | 1338 | static void status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 1339 | { |
e8096360 | 1340 | struct r1conf *conf = mddev->private; |
1da177e4 LT |
1341 | int i; |
1342 | ||
1343 | seq_printf(seq, " [%d/%d] [", conf->raid_disks, | |
11ce99e6 | 1344 | conf->raid_disks - mddev->degraded); |
ddac7c7e N |
1345 | rcu_read_lock(); |
1346 | for (i = 0; i < conf->raid_disks; i++) { | |
3cb03002 | 1347 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
1da177e4 | 1348 | seq_printf(seq, "%s", |
ddac7c7e N |
1349 | rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); |
1350 | } | |
1351 | rcu_read_unlock(); | |
1da177e4 LT |
1352 | seq_printf(seq, "]"); |
1353 | } | |
1354 | ||
1355 | ||
fd01b88c | 1356 | static void error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1357 | { |
1358 | char b[BDEVNAME_SIZE]; | |
e8096360 | 1359 | struct r1conf *conf = mddev->private; |
1da177e4 LT |
1360 | |
1361 | /* | |
1362 | * If it is not operational, then we have already marked it as dead | |
1363 | * else if it is the last working disks, ignore the error, let the | |
1364 | * next level up know. | |
1365 | * else mark the drive as failed | |
1366 | */ | |
b2d444d7 | 1367 | if (test_bit(In_sync, &rdev->flags) |
4044ba58 | 1368 | && (conf->raid_disks - mddev->degraded) == 1) { |
1da177e4 LT |
1369 | /* |
1370 | * Don't fail the drive, act as though we were just a | |
4044ba58 N |
1371 | * normal single drive. |
1372 | * However don't try a recovery from this drive as | |
1373 | * it is very likely to fail. | |
1da177e4 | 1374 | */ |
5389042f | 1375 | conf->recovery_disabled = mddev->recovery_disabled; |
1da177e4 | 1376 | return; |
4044ba58 | 1377 | } |
de393cde | 1378 | set_bit(Blocked, &rdev->flags); |
c04be0aa N |
1379 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
1380 | unsigned long flags; | |
1381 | spin_lock_irqsave(&conf->device_lock, flags); | |
1da177e4 | 1382 | mddev->degraded++; |
dd00a99e | 1383 | set_bit(Faulty, &rdev->flags); |
c04be0aa | 1384 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 LT |
1385 | /* |
1386 | * if recovery is running, make sure it aborts. | |
1387 | */ | |
dfc70645 | 1388 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
dd00a99e N |
1389 | } else |
1390 | set_bit(Faulty, &rdev->flags); | |
850b2b42 | 1391 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
067032bc JP |
1392 | printk(KERN_ALERT |
1393 | "md/raid1:%s: Disk failure on %s, disabling device.\n" | |
1394 | "md/raid1:%s: Operation continuing on %d devices.\n", | |
9dd1e2fa N |
1395 | mdname(mddev), bdevname(rdev->bdev, b), |
1396 | mdname(mddev), conf->raid_disks - mddev->degraded); | |
1da177e4 LT |
1397 | } |
1398 | ||
e8096360 | 1399 | static void print_conf(struct r1conf *conf) |
1da177e4 LT |
1400 | { |
1401 | int i; | |
1da177e4 | 1402 | |
9dd1e2fa | 1403 | printk(KERN_DEBUG "RAID1 conf printout:\n"); |
1da177e4 | 1404 | if (!conf) { |
9dd1e2fa | 1405 | printk(KERN_DEBUG "(!conf)\n"); |
1da177e4 LT |
1406 | return; |
1407 | } | |
9dd1e2fa | 1408 | printk(KERN_DEBUG " --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded, |
1da177e4 LT |
1409 | conf->raid_disks); |
1410 | ||
ddac7c7e | 1411 | rcu_read_lock(); |
1da177e4 LT |
1412 | for (i = 0; i < conf->raid_disks; i++) { |
1413 | char b[BDEVNAME_SIZE]; | |
3cb03002 | 1414 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
ddac7c7e | 1415 | if (rdev) |
9dd1e2fa | 1416 | printk(KERN_DEBUG " disk %d, wo:%d, o:%d, dev:%s\n", |
ddac7c7e N |
1417 | i, !test_bit(In_sync, &rdev->flags), |
1418 | !test_bit(Faulty, &rdev->flags), | |
1419 | bdevname(rdev->bdev,b)); | |
1da177e4 | 1420 | } |
ddac7c7e | 1421 | rcu_read_unlock(); |
1da177e4 LT |
1422 | } |
1423 | ||
e8096360 | 1424 | static void close_sync(struct r1conf *conf) |
1da177e4 | 1425 | { |
17999be4 N |
1426 | wait_barrier(conf); |
1427 | allow_barrier(conf); | |
1da177e4 LT |
1428 | |
1429 | mempool_destroy(conf->r1buf_pool); | |
1430 | conf->r1buf_pool = NULL; | |
1431 | } | |
1432 | ||
fd01b88c | 1433 | static int raid1_spare_active(struct mddev *mddev) |
1da177e4 LT |
1434 | { |
1435 | int i; | |
e8096360 | 1436 | struct r1conf *conf = mddev->private; |
6b965620 N |
1437 | int count = 0; |
1438 | unsigned long flags; | |
1da177e4 LT |
1439 | |
1440 | /* | |
1441 | * Find all failed disks within the RAID1 configuration | |
ddac7c7e N |
1442 | * and mark them readable. |
1443 | * Called under mddev lock, so rcu protection not needed. | |
1da177e4 LT |
1444 | */ |
1445 | for (i = 0; i < conf->raid_disks; i++) { | |
3cb03002 | 1446 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
8c7a2c2b N |
1447 | struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev; |
1448 | if (repl | |
1449 | && repl->recovery_offset == MaxSector | |
1450 | && !test_bit(Faulty, &repl->flags) | |
1451 | && !test_and_set_bit(In_sync, &repl->flags)) { | |
1452 | /* replacement has just become active */ | |
1453 | if (!rdev || | |
1454 | !test_and_clear_bit(In_sync, &rdev->flags)) | |
1455 | count++; | |
1456 | if (rdev) { | |
1457 | /* Replaced device not technically | |
1458 | * faulty, but we need to be sure | |
1459 | * it gets removed and never re-added | |
1460 | */ | |
1461 | set_bit(Faulty, &rdev->flags); | |
1462 | sysfs_notify_dirent_safe( | |
1463 | rdev->sysfs_state); | |
1464 | } | |
1465 | } | |
ddac7c7e N |
1466 | if (rdev |
1467 | && !test_bit(Faulty, &rdev->flags) | |
c04be0aa | 1468 | && !test_and_set_bit(In_sync, &rdev->flags)) { |
6b965620 | 1469 | count++; |
654e8b5a | 1470 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
1471 | } |
1472 | } | |
6b965620 N |
1473 | spin_lock_irqsave(&conf->device_lock, flags); |
1474 | mddev->degraded -= count; | |
1475 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 LT |
1476 | |
1477 | print_conf(conf); | |
6b965620 | 1478 | return count; |
1da177e4 LT |
1479 | } |
1480 | ||
1481 | ||
fd01b88c | 1482 | static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1483 | { |
e8096360 | 1484 | struct r1conf *conf = mddev->private; |
199050ea | 1485 | int err = -EEXIST; |
41158c7e | 1486 | int mirror = 0; |
0eaf822c | 1487 | struct raid1_info *p; |
6c2fce2e | 1488 | int first = 0; |
30194636 | 1489 | int last = conf->raid_disks - 1; |
6b740b8d | 1490 | struct request_queue *q = bdev_get_queue(rdev->bdev); |
1da177e4 | 1491 | |
5389042f N |
1492 | if (mddev->recovery_disabled == conf->recovery_disabled) |
1493 | return -EBUSY; | |
1494 | ||
6c2fce2e NB |
1495 | if (rdev->raid_disk >= 0) |
1496 | first = last = rdev->raid_disk; | |
1497 | ||
6b740b8d N |
1498 | if (q->merge_bvec_fn) { |
1499 | set_bit(Unmerged, &rdev->flags); | |
1500 | mddev->merge_check_needed = 1; | |
1501 | } | |
1502 | ||
7ef449d1 N |
1503 | for (mirror = first; mirror <= last; mirror++) { |
1504 | p = conf->mirrors+mirror; | |
1505 | if (!p->rdev) { | |
1da177e4 | 1506 | |
8f6c2e4b MP |
1507 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
1508 | rdev->data_offset << 9); | |
1da177e4 LT |
1509 | |
1510 | p->head_position = 0; | |
1511 | rdev->raid_disk = mirror; | |
199050ea | 1512 | err = 0; |
6aea114a N |
1513 | /* As all devices are equivalent, we don't need a full recovery |
1514 | * if this was recently any drive of the array | |
1515 | */ | |
1516 | if (rdev->saved_raid_disk < 0) | |
41158c7e | 1517 | conf->fullsync = 1; |
d6065f7b | 1518 | rcu_assign_pointer(p->rdev, rdev); |
1da177e4 LT |
1519 | break; |
1520 | } | |
7ef449d1 N |
1521 | if (test_bit(WantReplacement, &p->rdev->flags) && |
1522 | p[conf->raid_disks].rdev == NULL) { | |
1523 | /* Add this device as a replacement */ | |
1524 | clear_bit(In_sync, &rdev->flags); | |
1525 | set_bit(Replacement, &rdev->flags); | |
1526 | rdev->raid_disk = mirror; | |
1527 | err = 0; | |
1528 | conf->fullsync = 1; | |
1529 | rcu_assign_pointer(p[conf->raid_disks].rdev, rdev); | |
1530 | break; | |
1531 | } | |
1532 | } | |
6b740b8d N |
1533 | if (err == 0 && test_bit(Unmerged, &rdev->flags)) { |
1534 | /* Some requests might not have seen this new | |
1535 | * merge_bvec_fn. We must wait for them to complete | |
1536 | * before merging the device fully. | |
1537 | * First we make sure any code which has tested | |
1538 | * our function has submitted the request, then | |
1539 | * we wait for all outstanding requests to complete. | |
1540 | */ | |
1541 | synchronize_sched(); | |
1542 | raise_barrier(conf); | |
1543 | lower_barrier(conf); | |
1544 | clear_bit(Unmerged, &rdev->flags); | |
1545 | } | |
ac5e7113 | 1546 | md_integrity_add_rdev(rdev, mddev); |
2ff8cc2c SL |
1547 | if (blk_queue_discard(bdev_get_queue(rdev->bdev))) |
1548 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue); | |
1da177e4 | 1549 | print_conf(conf); |
199050ea | 1550 | return err; |
1da177e4 LT |
1551 | } |
1552 | ||
b8321b68 | 1553 | static int raid1_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1554 | { |
e8096360 | 1555 | struct r1conf *conf = mddev->private; |
1da177e4 | 1556 | int err = 0; |
b8321b68 | 1557 | int number = rdev->raid_disk; |
0eaf822c | 1558 | struct raid1_info *p = conf->mirrors + number; |
1da177e4 | 1559 | |
b014f14c N |
1560 | if (rdev != p->rdev) |
1561 | p = conf->mirrors + conf->raid_disks + number; | |
1562 | ||
1da177e4 | 1563 | print_conf(conf); |
b8321b68 | 1564 | if (rdev == p->rdev) { |
b2d444d7 | 1565 | if (test_bit(In_sync, &rdev->flags) || |
1da177e4 LT |
1566 | atomic_read(&rdev->nr_pending)) { |
1567 | err = -EBUSY; | |
1568 | goto abort; | |
1569 | } | |
046abeed | 1570 | /* Only remove non-faulty devices if recovery |
dfc70645 N |
1571 | * is not possible. |
1572 | */ | |
1573 | if (!test_bit(Faulty, &rdev->flags) && | |
5389042f | 1574 | mddev->recovery_disabled != conf->recovery_disabled && |
dfc70645 N |
1575 | mddev->degraded < conf->raid_disks) { |
1576 | err = -EBUSY; | |
1577 | goto abort; | |
1578 | } | |
1da177e4 | 1579 | p->rdev = NULL; |
fbd568a3 | 1580 | synchronize_rcu(); |
1da177e4 LT |
1581 | if (atomic_read(&rdev->nr_pending)) { |
1582 | /* lost the race, try later */ | |
1583 | err = -EBUSY; | |
1584 | p->rdev = rdev; | |
ac5e7113 | 1585 | goto abort; |
8c7a2c2b N |
1586 | } else if (conf->mirrors[conf->raid_disks + number].rdev) { |
1587 | /* We just removed a device that is being replaced. | |
1588 | * Move down the replacement. We drain all IO before | |
1589 | * doing this to avoid confusion. | |
1590 | */ | |
1591 | struct md_rdev *repl = | |
1592 | conf->mirrors[conf->raid_disks + number].rdev; | |
1593 | raise_barrier(conf); | |
1594 | clear_bit(Replacement, &repl->flags); | |
1595 | p->rdev = repl; | |
1596 | conf->mirrors[conf->raid_disks + number].rdev = NULL; | |
1597 | lower_barrier(conf); | |
1598 | clear_bit(WantReplacement, &rdev->flags); | |
1599 | } else | |
b014f14c | 1600 | clear_bit(WantReplacement, &rdev->flags); |
a91a2785 | 1601 | err = md_integrity_register(mddev); |
1da177e4 LT |
1602 | } |
1603 | abort: | |
1604 | ||
1605 | print_conf(conf); | |
1606 | return err; | |
1607 | } | |
1608 | ||
1609 | ||
6712ecf8 | 1610 | static void end_sync_read(struct bio *bio, int error) |
1da177e4 | 1611 | { |
9f2c9d12 | 1612 | struct r1bio *r1_bio = bio->bi_private; |
1da177e4 | 1613 | |
0fc280f6 | 1614 | update_head_pos(r1_bio->read_disk, r1_bio); |
ba3ae3be | 1615 | |
1da177e4 LT |
1616 | /* |
1617 | * we have read a block, now it needs to be re-written, | |
1618 | * or re-read if the read failed. | |
1619 | * We don't do much here, just schedule handling by raid1d | |
1620 | */ | |
69382e85 | 1621 | if (test_bit(BIO_UPTODATE, &bio->bi_flags)) |
1da177e4 | 1622 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
d11c171e N |
1623 | |
1624 | if (atomic_dec_and_test(&r1_bio->remaining)) | |
1625 | reschedule_retry(r1_bio); | |
1da177e4 LT |
1626 | } |
1627 | ||
6712ecf8 | 1628 | static void end_sync_write(struct bio *bio, int error) |
1da177e4 LT |
1629 | { |
1630 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
9f2c9d12 | 1631 | struct r1bio *r1_bio = bio->bi_private; |
fd01b88c | 1632 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 1633 | struct r1conf *conf = mddev->private; |
1da177e4 | 1634 | int mirror=0; |
4367af55 N |
1635 | sector_t first_bad; |
1636 | int bad_sectors; | |
1da177e4 | 1637 | |
ba3ae3be NK |
1638 | mirror = find_bio_disk(r1_bio, bio); |
1639 | ||
6b1117d5 | 1640 | if (!uptodate) { |
57dab0bd | 1641 | sector_t sync_blocks = 0; |
6b1117d5 N |
1642 | sector_t s = r1_bio->sector; |
1643 | long sectors_to_go = r1_bio->sectors; | |
1644 | /* make sure these bits doesn't get cleared. */ | |
1645 | do { | |
5e3db645 | 1646 | bitmap_end_sync(mddev->bitmap, s, |
6b1117d5 N |
1647 | &sync_blocks, 1); |
1648 | s += sync_blocks; | |
1649 | sectors_to_go -= sync_blocks; | |
1650 | } while (sectors_to_go > 0); | |
d8f05d29 N |
1651 | set_bit(WriteErrorSeen, |
1652 | &conf->mirrors[mirror].rdev->flags); | |
19d67169 N |
1653 | if (!test_and_set_bit(WantReplacement, |
1654 | &conf->mirrors[mirror].rdev->flags)) | |
1655 | set_bit(MD_RECOVERY_NEEDED, & | |
1656 | mddev->recovery); | |
d8f05d29 | 1657 | set_bit(R1BIO_WriteError, &r1_bio->state); |
4367af55 N |
1658 | } else if (is_badblock(conf->mirrors[mirror].rdev, |
1659 | r1_bio->sector, | |
1660 | r1_bio->sectors, | |
3a9f28a5 N |
1661 | &first_bad, &bad_sectors) && |
1662 | !is_badblock(conf->mirrors[r1_bio->read_disk].rdev, | |
1663 | r1_bio->sector, | |
1664 | r1_bio->sectors, | |
1665 | &first_bad, &bad_sectors) | |
1666 | ) | |
4367af55 | 1667 | set_bit(R1BIO_MadeGood, &r1_bio->state); |
e3b9703e | 1668 | |
1da177e4 | 1669 | if (atomic_dec_and_test(&r1_bio->remaining)) { |
4367af55 | 1670 | int s = r1_bio->sectors; |
d8f05d29 N |
1671 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
1672 | test_bit(R1BIO_WriteError, &r1_bio->state)) | |
4367af55 N |
1673 | reschedule_retry(r1_bio); |
1674 | else { | |
1675 | put_buf(r1_bio); | |
1676 | md_done_sync(mddev, s, uptodate); | |
1677 | } | |
1da177e4 | 1678 | } |
1da177e4 LT |
1679 | } |
1680 | ||
3cb03002 | 1681 | static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector, |
d8f05d29 N |
1682 | int sectors, struct page *page, int rw) |
1683 | { | |
1684 | if (sync_page_io(rdev, sector, sectors << 9, page, rw, false)) | |
1685 | /* success */ | |
1686 | return 1; | |
19d67169 | 1687 | if (rw == WRITE) { |
d8f05d29 | 1688 | set_bit(WriteErrorSeen, &rdev->flags); |
19d67169 N |
1689 | if (!test_and_set_bit(WantReplacement, |
1690 | &rdev->flags)) | |
1691 | set_bit(MD_RECOVERY_NEEDED, & | |
1692 | rdev->mddev->recovery); | |
1693 | } | |
d8f05d29 N |
1694 | /* need to record an error - either for the block or the device */ |
1695 | if (!rdev_set_badblocks(rdev, sector, sectors, 0)) | |
1696 | md_error(rdev->mddev, rdev); | |
1697 | return 0; | |
1698 | } | |
1699 | ||
9f2c9d12 | 1700 | static int fix_sync_read_error(struct r1bio *r1_bio) |
1da177e4 | 1701 | { |
a68e5870 N |
1702 | /* Try some synchronous reads of other devices to get |
1703 | * good data, much like with normal read errors. Only | |
1704 | * read into the pages we already have so we don't | |
1705 | * need to re-issue the read request. | |
1706 | * We don't need to freeze the array, because being in an | |
1707 | * active sync request, there is no normal IO, and | |
1708 | * no overlapping syncs. | |
06f60385 N |
1709 | * We don't need to check is_badblock() again as we |
1710 | * made sure that anything with a bad block in range | |
1711 | * will have bi_end_io clear. | |
a68e5870 | 1712 | */ |
fd01b88c | 1713 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 1714 | struct r1conf *conf = mddev->private; |
a68e5870 N |
1715 | struct bio *bio = r1_bio->bios[r1_bio->read_disk]; |
1716 | sector_t sect = r1_bio->sector; | |
1717 | int sectors = r1_bio->sectors; | |
1718 | int idx = 0; | |
1719 | ||
1720 | while(sectors) { | |
1721 | int s = sectors; | |
1722 | int d = r1_bio->read_disk; | |
1723 | int success = 0; | |
3cb03002 | 1724 | struct md_rdev *rdev; |
78d7f5f7 | 1725 | int start; |
a68e5870 N |
1726 | |
1727 | if (s > (PAGE_SIZE>>9)) | |
1728 | s = PAGE_SIZE >> 9; | |
1729 | do { | |
1730 | if (r1_bio->bios[d]->bi_end_io == end_sync_read) { | |
1731 | /* No rcu protection needed here devices | |
1732 | * can only be removed when no resync is | |
1733 | * active, and resync is currently active | |
1734 | */ | |
1735 | rdev = conf->mirrors[d].rdev; | |
9d3d8011 | 1736 | if (sync_page_io(rdev, sect, s<<9, |
a68e5870 N |
1737 | bio->bi_io_vec[idx].bv_page, |
1738 | READ, false)) { | |
1739 | success = 1; | |
1740 | break; | |
1741 | } | |
1742 | } | |
1743 | d++; | |
8f19ccb2 | 1744 | if (d == conf->raid_disks * 2) |
a68e5870 N |
1745 | d = 0; |
1746 | } while (!success && d != r1_bio->read_disk); | |
1747 | ||
78d7f5f7 | 1748 | if (!success) { |
a68e5870 | 1749 | char b[BDEVNAME_SIZE]; |
3a9f28a5 N |
1750 | int abort = 0; |
1751 | /* Cannot read from anywhere, this block is lost. | |
1752 | * Record a bad block on each device. If that doesn't | |
1753 | * work just disable and interrupt the recovery. | |
1754 | * Don't fail devices as that won't really help. | |
1755 | */ | |
a68e5870 N |
1756 | printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O read error" |
1757 | " for block %llu\n", | |
1758 | mdname(mddev), | |
1759 | bdevname(bio->bi_bdev, b), | |
1760 | (unsigned long long)r1_bio->sector); | |
8f19ccb2 | 1761 | for (d = 0; d < conf->raid_disks * 2; d++) { |
3a9f28a5 N |
1762 | rdev = conf->mirrors[d].rdev; |
1763 | if (!rdev || test_bit(Faulty, &rdev->flags)) | |
1764 | continue; | |
1765 | if (!rdev_set_badblocks(rdev, sect, s, 0)) | |
1766 | abort = 1; | |
1767 | } | |
1768 | if (abort) { | |
d890fa2b N |
1769 | conf->recovery_disabled = |
1770 | mddev->recovery_disabled; | |
3a9f28a5 N |
1771 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1772 | md_done_sync(mddev, r1_bio->sectors, 0); | |
1773 | put_buf(r1_bio); | |
1774 | return 0; | |
1775 | } | |
1776 | /* Try next page */ | |
1777 | sectors -= s; | |
1778 | sect += s; | |
1779 | idx++; | |
1780 | continue; | |
d11c171e | 1781 | } |
78d7f5f7 N |
1782 | |
1783 | start = d; | |
1784 | /* write it back and re-read */ | |
1785 | while (d != r1_bio->read_disk) { | |
1786 | if (d == 0) | |
8f19ccb2 | 1787 | d = conf->raid_disks * 2; |
78d7f5f7 N |
1788 | d--; |
1789 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) | |
1790 | continue; | |
1791 | rdev = conf->mirrors[d].rdev; | |
d8f05d29 N |
1792 | if (r1_sync_page_io(rdev, sect, s, |
1793 | bio->bi_io_vec[idx].bv_page, | |
1794 | WRITE) == 0) { | |
78d7f5f7 N |
1795 | r1_bio->bios[d]->bi_end_io = NULL; |
1796 | rdev_dec_pending(rdev, mddev); | |
9d3d8011 | 1797 | } |
78d7f5f7 N |
1798 | } |
1799 | d = start; | |
1800 | while (d != r1_bio->read_disk) { | |
1801 | if (d == 0) | |
8f19ccb2 | 1802 | d = conf->raid_disks * 2; |
78d7f5f7 N |
1803 | d--; |
1804 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) | |
1805 | continue; | |
1806 | rdev = conf->mirrors[d].rdev; | |
d8f05d29 N |
1807 | if (r1_sync_page_io(rdev, sect, s, |
1808 | bio->bi_io_vec[idx].bv_page, | |
1809 | READ) != 0) | |
9d3d8011 | 1810 | atomic_add(s, &rdev->corrected_errors); |
78d7f5f7 | 1811 | } |
a68e5870 N |
1812 | sectors -= s; |
1813 | sect += s; | |
1814 | idx ++; | |
1815 | } | |
78d7f5f7 | 1816 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
7ca78d57 | 1817 | set_bit(BIO_UPTODATE, &bio->bi_flags); |
a68e5870 N |
1818 | return 1; |
1819 | } | |
1820 | ||
9f2c9d12 | 1821 | static int process_checks(struct r1bio *r1_bio) |
a68e5870 N |
1822 | { |
1823 | /* We have read all readable devices. If we haven't | |
1824 | * got the block, then there is no hope left. | |
1825 | * If we have, then we want to do a comparison | |
1826 | * and skip the write if everything is the same. | |
1827 | * If any blocks failed to read, then we need to | |
1828 | * attempt an over-write | |
1829 | */ | |
fd01b88c | 1830 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 1831 | struct r1conf *conf = mddev->private; |
a68e5870 N |
1832 | int primary; |
1833 | int i; | |
f4380a91 | 1834 | int vcnt; |
a68e5870 | 1835 | |
8f19ccb2 | 1836 | for (primary = 0; primary < conf->raid_disks * 2; primary++) |
a68e5870 N |
1837 | if (r1_bio->bios[primary]->bi_end_io == end_sync_read && |
1838 | test_bit(BIO_UPTODATE, &r1_bio->bios[primary]->bi_flags)) { | |
1839 | r1_bio->bios[primary]->bi_end_io = NULL; | |
1840 | rdev_dec_pending(conf->mirrors[primary].rdev, mddev); | |
1841 | break; | |
1842 | } | |
1843 | r1_bio->read_disk = primary; | |
f4380a91 | 1844 | vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9); |
8f19ccb2 | 1845 | for (i = 0; i < conf->raid_disks * 2; i++) { |
78d7f5f7 | 1846 | int j; |
78d7f5f7 N |
1847 | struct bio *pbio = r1_bio->bios[primary]; |
1848 | struct bio *sbio = r1_bio->bios[i]; | |
1849 | int size; | |
a68e5870 | 1850 | |
2aabaa65 | 1851 | if (sbio->bi_end_io != end_sync_read) |
78d7f5f7 N |
1852 | continue; |
1853 | ||
1854 | if (test_bit(BIO_UPTODATE, &sbio->bi_flags)) { | |
1855 | for (j = vcnt; j-- ; ) { | |
1856 | struct page *p, *s; | |
1857 | p = pbio->bi_io_vec[j].bv_page; | |
1858 | s = sbio->bi_io_vec[j].bv_page; | |
1859 | if (memcmp(page_address(p), | |
1860 | page_address(s), | |
5020ad7d | 1861 | sbio->bi_io_vec[j].bv_len)) |
78d7f5f7 | 1862 | break; |
69382e85 | 1863 | } |
78d7f5f7 N |
1864 | } else |
1865 | j = 0; | |
1866 | if (j >= 0) | |
7f7583d4 | 1867 | atomic64_add(r1_bio->sectors, &mddev->resync_mismatches); |
78d7f5f7 N |
1868 | if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery) |
1869 | && test_bit(BIO_UPTODATE, &sbio->bi_flags))) { | |
1870 | /* No need to write to this device. */ | |
1871 | sbio->bi_end_io = NULL; | |
1872 | rdev_dec_pending(conf->mirrors[i].rdev, mddev); | |
1873 | continue; | |
1874 | } | |
1875 | /* fixup the bio for reuse */ | |
2aabaa65 | 1876 | bio_reset(sbio); |
78d7f5f7 N |
1877 | sbio->bi_vcnt = vcnt; |
1878 | sbio->bi_size = r1_bio->sectors << 9; | |
78d7f5f7 N |
1879 | sbio->bi_sector = r1_bio->sector + |
1880 | conf->mirrors[i].rdev->data_offset; | |
1881 | sbio->bi_bdev = conf->mirrors[i].rdev->bdev; | |
2aabaa65 KO |
1882 | sbio->bi_end_io = end_sync_read; |
1883 | sbio->bi_private = r1_bio; | |
1884 | ||
78d7f5f7 N |
1885 | size = sbio->bi_size; |
1886 | for (j = 0; j < vcnt ; j++) { | |
1887 | struct bio_vec *bi; | |
1888 | bi = &sbio->bi_io_vec[j]; | |
1889 | bi->bv_offset = 0; | |
1890 | if (size > PAGE_SIZE) | |
1891 | bi->bv_len = PAGE_SIZE; | |
1892 | else | |
1893 | bi->bv_len = size; | |
1894 | size -= PAGE_SIZE; | |
69382e85 | 1895 | } |
d3b45c2a KO |
1896 | |
1897 | bio_copy_data(sbio, pbio); | |
78d7f5f7 | 1898 | } |
a68e5870 N |
1899 | return 0; |
1900 | } | |
1901 | ||
9f2c9d12 | 1902 | static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio) |
a68e5870 | 1903 | { |
e8096360 | 1904 | struct r1conf *conf = mddev->private; |
a68e5870 | 1905 | int i; |
8f19ccb2 | 1906 | int disks = conf->raid_disks * 2; |
a68e5870 N |
1907 | struct bio *bio, *wbio; |
1908 | ||
1909 | bio = r1_bio->bios[r1_bio->read_disk]; | |
1910 | ||
a68e5870 N |
1911 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) |
1912 | /* ouch - failed to read all of that. */ | |
1913 | if (!fix_sync_read_error(r1_bio)) | |
1914 | return; | |
7ca78d57 N |
1915 | |
1916 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
1917 | if (process_checks(r1_bio) < 0) | |
1918 | return; | |
d11c171e N |
1919 | /* |
1920 | * schedule writes | |
1921 | */ | |
1da177e4 LT |
1922 | atomic_set(&r1_bio->remaining, 1); |
1923 | for (i = 0; i < disks ; i++) { | |
1924 | wbio = r1_bio->bios[i]; | |
3e198f78 N |
1925 | if (wbio->bi_end_io == NULL || |
1926 | (wbio->bi_end_io == end_sync_read && | |
1927 | (i == r1_bio->read_disk || | |
1928 | !test_bit(MD_RECOVERY_SYNC, &mddev->recovery)))) | |
1da177e4 LT |
1929 | continue; |
1930 | ||
3e198f78 N |
1931 | wbio->bi_rw = WRITE; |
1932 | wbio->bi_end_io = end_sync_write; | |
1da177e4 | 1933 | atomic_inc(&r1_bio->remaining); |
aa8b57aa | 1934 | md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio)); |
191ea9b2 | 1935 | |
1da177e4 LT |
1936 | generic_make_request(wbio); |
1937 | } | |
1938 | ||
1939 | if (atomic_dec_and_test(&r1_bio->remaining)) { | |
191ea9b2 | 1940 | /* if we're here, all write(s) have completed, so clean up */ |
58e94ae1 N |
1941 | int s = r1_bio->sectors; |
1942 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || | |
1943 | test_bit(R1BIO_WriteError, &r1_bio->state)) | |
1944 | reschedule_retry(r1_bio); | |
1945 | else { | |
1946 | put_buf(r1_bio); | |
1947 | md_done_sync(mddev, s, 1); | |
1948 | } | |
1da177e4 LT |
1949 | } |
1950 | } | |
1951 | ||
1952 | /* | |
1953 | * This is a kernel thread which: | |
1954 | * | |
1955 | * 1. Retries failed read operations on working mirrors. | |
1956 | * 2. Updates the raid superblock when problems encounter. | |
d2eb35ac | 1957 | * 3. Performs writes following reads for array synchronising. |
1da177e4 LT |
1958 | */ |
1959 | ||
e8096360 | 1960 | static void fix_read_error(struct r1conf *conf, int read_disk, |
867868fb N |
1961 | sector_t sect, int sectors) |
1962 | { | |
fd01b88c | 1963 | struct mddev *mddev = conf->mddev; |
867868fb N |
1964 | while(sectors) { |
1965 | int s = sectors; | |
1966 | int d = read_disk; | |
1967 | int success = 0; | |
1968 | int start; | |
3cb03002 | 1969 | struct md_rdev *rdev; |
867868fb N |
1970 | |
1971 | if (s > (PAGE_SIZE>>9)) | |
1972 | s = PAGE_SIZE >> 9; | |
1973 | ||
1974 | do { | |
1975 | /* Note: no rcu protection needed here | |
1976 | * as this is synchronous in the raid1d thread | |
1977 | * which is the thread that might remove | |
1978 | * a device. If raid1d ever becomes multi-threaded.... | |
1979 | */ | |
d2eb35ac N |
1980 | sector_t first_bad; |
1981 | int bad_sectors; | |
1982 | ||
867868fb N |
1983 | rdev = conf->mirrors[d].rdev; |
1984 | if (rdev && | |
da8840a7 | 1985 | (test_bit(In_sync, &rdev->flags) || |
1986 | (!test_bit(Faulty, &rdev->flags) && | |
1987 | rdev->recovery_offset >= sect + s)) && | |
d2eb35ac N |
1988 | is_badblock(rdev, sect, s, |
1989 | &first_bad, &bad_sectors) == 0 && | |
ccebd4c4 JB |
1990 | sync_page_io(rdev, sect, s<<9, |
1991 | conf->tmppage, READ, false)) | |
867868fb N |
1992 | success = 1; |
1993 | else { | |
1994 | d++; | |
8f19ccb2 | 1995 | if (d == conf->raid_disks * 2) |
867868fb N |
1996 | d = 0; |
1997 | } | |
1998 | } while (!success && d != read_disk); | |
1999 | ||
2000 | if (!success) { | |
d8f05d29 | 2001 | /* Cannot read from anywhere - mark it bad */ |
3cb03002 | 2002 | struct md_rdev *rdev = conf->mirrors[read_disk].rdev; |
d8f05d29 N |
2003 | if (!rdev_set_badblocks(rdev, sect, s, 0)) |
2004 | md_error(mddev, rdev); | |
867868fb N |
2005 | break; |
2006 | } | |
2007 | /* write it back and re-read */ | |
2008 | start = d; | |
2009 | while (d != read_disk) { | |
2010 | if (d==0) | |
8f19ccb2 | 2011 | d = conf->raid_disks * 2; |
867868fb N |
2012 | d--; |
2013 | rdev = conf->mirrors[d].rdev; | |
2014 | if (rdev && | |
d8f05d29 N |
2015 | test_bit(In_sync, &rdev->flags)) |
2016 | r1_sync_page_io(rdev, sect, s, | |
2017 | conf->tmppage, WRITE); | |
867868fb N |
2018 | } |
2019 | d = start; | |
2020 | while (d != read_disk) { | |
2021 | char b[BDEVNAME_SIZE]; | |
2022 | if (d==0) | |
8f19ccb2 | 2023 | d = conf->raid_disks * 2; |
867868fb N |
2024 | d--; |
2025 | rdev = conf->mirrors[d].rdev; | |
2026 | if (rdev && | |
2027 | test_bit(In_sync, &rdev->flags)) { | |
d8f05d29 N |
2028 | if (r1_sync_page_io(rdev, sect, s, |
2029 | conf->tmppage, READ)) { | |
867868fb N |
2030 | atomic_add(s, &rdev->corrected_errors); |
2031 | printk(KERN_INFO | |
9dd1e2fa | 2032 | "md/raid1:%s: read error corrected " |
867868fb N |
2033 | "(%d sectors at %llu on %s)\n", |
2034 | mdname(mddev), s, | |
969b755a RD |
2035 | (unsigned long long)(sect + |
2036 | rdev->data_offset), | |
867868fb N |
2037 | bdevname(rdev->bdev, b)); |
2038 | } | |
2039 | } | |
2040 | } | |
2041 | sectors -= s; | |
2042 | sect += s; | |
2043 | } | |
2044 | } | |
2045 | ||
9f2c9d12 | 2046 | static int narrow_write_error(struct r1bio *r1_bio, int i) |
cd5ff9a1 | 2047 | { |
fd01b88c | 2048 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 2049 | struct r1conf *conf = mddev->private; |
3cb03002 | 2050 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
cd5ff9a1 N |
2051 | |
2052 | /* bio has the data to be written to device 'i' where | |
2053 | * we just recently had a write error. | |
2054 | * We repeatedly clone the bio and trim down to one block, | |
2055 | * then try the write. Where the write fails we record | |
2056 | * a bad block. | |
2057 | * It is conceivable that the bio doesn't exactly align with | |
2058 | * blocks. We must handle this somehow. | |
2059 | * | |
2060 | * We currently own a reference on the rdev. | |
2061 | */ | |
2062 | ||
2063 | int block_sectors; | |
2064 | sector_t sector; | |
2065 | int sectors; | |
2066 | int sect_to_write = r1_bio->sectors; | |
2067 | int ok = 1; | |
2068 | ||
2069 | if (rdev->badblocks.shift < 0) | |
2070 | return 0; | |
2071 | ||
2072 | block_sectors = 1 << rdev->badblocks.shift; | |
2073 | sector = r1_bio->sector; | |
2074 | sectors = ((sector + block_sectors) | |
2075 | & ~(sector_t)(block_sectors - 1)) | |
2076 | - sector; | |
2077 | ||
cd5ff9a1 N |
2078 | while (sect_to_write) { |
2079 | struct bio *wbio; | |
2080 | if (sectors > sect_to_write) | |
2081 | sectors = sect_to_write; | |
2082 | /* Write at 'sector' for 'sectors'*/ | |
2083 | ||
b783863f KO |
2084 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { |
2085 | unsigned vcnt = r1_bio->behind_page_count; | |
2086 | struct bio_vec *vec = r1_bio->behind_bvecs; | |
2087 | ||
2088 | while (!vec->bv_page) { | |
2089 | vec++; | |
2090 | vcnt--; | |
2091 | } | |
2092 | ||
2093 | wbio = bio_alloc_mddev(GFP_NOIO, vcnt, mddev); | |
2094 | memcpy(wbio->bi_io_vec, vec, vcnt * sizeof(struct bio_vec)); | |
2095 | ||
2096 | wbio->bi_vcnt = vcnt; | |
2097 | } else { | |
2098 | wbio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev); | |
2099 | } | |
2100 | ||
cd5ff9a1 | 2101 | wbio->bi_rw = WRITE; |
b783863f | 2102 | wbio->bi_sector = r1_bio->sector; |
cd5ff9a1 | 2103 | wbio->bi_size = r1_bio->sectors << 9; |
cd5ff9a1 N |
2104 | |
2105 | md_trim_bio(wbio, sector - r1_bio->sector, sectors); | |
2106 | wbio->bi_sector += rdev->data_offset; | |
2107 | wbio->bi_bdev = rdev->bdev; | |
2108 | if (submit_bio_wait(WRITE, wbio) == 0) | |
2109 | /* failure! */ | |
2110 | ok = rdev_set_badblocks(rdev, sector, | |
2111 | sectors, 0) | |
2112 | && ok; | |
2113 | ||
2114 | bio_put(wbio); | |
2115 | sect_to_write -= sectors; | |
2116 | sector += sectors; | |
2117 | sectors = block_sectors; | |
2118 | } | |
2119 | return ok; | |
2120 | } | |
2121 | ||
e8096360 | 2122 | static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce N |
2123 | { |
2124 | int m; | |
2125 | int s = r1_bio->sectors; | |
8f19ccb2 | 2126 | for (m = 0; m < conf->raid_disks * 2 ; m++) { |
3cb03002 | 2127 | struct md_rdev *rdev = conf->mirrors[m].rdev; |
62096bce N |
2128 | struct bio *bio = r1_bio->bios[m]; |
2129 | if (bio->bi_end_io == NULL) | |
2130 | continue; | |
2131 | if (test_bit(BIO_UPTODATE, &bio->bi_flags) && | |
2132 | test_bit(R1BIO_MadeGood, &r1_bio->state)) { | |
c6563a8c | 2133 | rdev_clear_badblocks(rdev, r1_bio->sector, s, 0); |
62096bce N |
2134 | } |
2135 | if (!test_bit(BIO_UPTODATE, &bio->bi_flags) && | |
2136 | test_bit(R1BIO_WriteError, &r1_bio->state)) { | |
2137 | if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0)) | |
2138 | md_error(conf->mddev, rdev); | |
2139 | } | |
2140 | } | |
2141 | put_buf(r1_bio); | |
2142 | md_done_sync(conf->mddev, s, 1); | |
2143 | } | |
2144 | ||
e8096360 | 2145 | static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce N |
2146 | { |
2147 | int m; | |
8f19ccb2 | 2148 | for (m = 0; m < conf->raid_disks * 2 ; m++) |
62096bce | 2149 | if (r1_bio->bios[m] == IO_MADE_GOOD) { |
3cb03002 | 2150 | struct md_rdev *rdev = conf->mirrors[m].rdev; |
62096bce N |
2151 | rdev_clear_badblocks(rdev, |
2152 | r1_bio->sector, | |
c6563a8c | 2153 | r1_bio->sectors, 0); |
62096bce N |
2154 | rdev_dec_pending(rdev, conf->mddev); |
2155 | } else if (r1_bio->bios[m] != NULL) { | |
2156 | /* This drive got a write error. We need to | |
2157 | * narrow down and record precise write | |
2158 | * errors. | |
2159 | */ | |
2160 | if (!narrow_write_error(r1_bio, m)) { | |
2161 | md_error(conf->mddev, | |
2162 | conf->mirrors[m].rdev); | |
2163 | /* an I/O failed, we can't clear the bitmap */ | |
2164 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
2165 | } | |
2166 | rdev_dec_pending(conf->mirrors[m].rdev, | |
2167 | conf->mddev); | |
2168 | } | |
2169 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2170 | close_write(r1_bio); | |
2171 | raid_end_bio_io(r1_bio); | |
2172 | } | |
2173 | ||
e8096360 | 2174 | static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce N |
2175 | { |
2176 | int disk; | |
2177 | int max_sectors; | |
fd01b88c | 2178 | struct mddev *mddev = conf->mddev; |
62096bce N |
2179 | struct bio *bio; |
2180 | char b[BDEVNAME_SIZE]; | |
3cb03002 | 2181 | struct md_rdev *rdev; |
62096bce N |
2182 | |
2183 | clear_bit(R1BIO_ReadError, &r1_bio->state); | |
2184 | /* we got a read error. Maybe the drive is bad. Maybe just | |
2185 | * the block and we can fix it. | |
2186 | * We freeze all other IO, and try reading the block from | |
2187 | * other devices. When we find one, we re-write | |
2188 | * and check it that fixes the read error. | |
2189 | * This is all done synchronously while the array is | |
2190 | * frozen | |
2191 | */ | |
2192 | if (mddev->ro == 0) { | |
2193 | freeze_array(conf); | |
2194 | fix_read_error(conf, r1_bio->read_disk, | |
2195 | r1_bio->sector, r1_bio->sectors); | |
2196 | unfreeze_array(conf); | |
2197 | } else | |
2198 | md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev); | |
7ad4d4a6 | 2199 | rdev_dec_pending(conf->mirrors[r1_bio->read_disk].rdev, conf->mddev); |
62096bce N |
2200 | |
2201 | bio = r1_bio->bios[r1_bio->read_disk]; | |
2202 | bdevname(bio->bi_bdev, b); | |
2203 | read_more: | |
2204 | disk = read_balance(conf, r1_bio, &max_sectors); | |
2205 | if (disk == -1) { | |
2206 | printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O" | |
2207 | " read error for block %llu\n", | |
2208 | mdname(mddev), b, (unsigned long long)r1_bio->sector); | |
2209 | raid_end_bio_io(r1_bio); | |
2210 | } else { | |
2211 | const unsigned long do_sync | |
2212 | = r1_bio->master_bio->bi_rw & REQ_SYNC; | |
2213 | if (bio) { | |
2214 | r1_bio->bios[r1_bio->read_disk] = | |
2215 | mddev->ro ? IO_BLOCKED : NULL; | |
2216 | bio_put(bio); | |
2217 | } | |
2218 | r1_bio->read_disk = disk; | |
2219 | bio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev); | |
2220 | md_trim_bio(bio, r1_bio->sector - bio->bi_sector, max_sectors); | |
2221 | r1_bio->bios[r1_bio->read_disk] = bio; | |
2222 | rdev = conf->mirrors[disk].rdev; | |
2223 | printk_ratelimited(KERN_ERR | |
2224 | "md/raid1:%s: redirecting sector %llu" | |
2225 | " to other mirror: %s\n", | |
2226 | mdname(mddev), | |
2227 | (unsigned long long)r1_bio->sector, | |
2228 | bdevname(rdev->bdev, b)); | |
2229 | bio->bi_sector = r1_bio->sector + rdev->data_offset; | |
2230 | bio->bi_bdev = rdev->bdev; | |
2231 | bio->bi_end_io = raid1_end_read_request; | |
2232 | bio->bi_rw = READ | do_sync; | |
2233 | bio->bi_private = r1_bio; | |
2234 | if (max_sectors < r1_bio->sectors) { | |
2235 | /* Drat - have to split this up more */ | |
2236 | struct bio *mbio = r1_bio->master_bio; | |
2237 | int sectors_handled = (r1_bio->sector + max_sectors | |
2238 | - mbio->bi_sector); | |
2239 | r1_bio->sectors = max_sectors; | |
2240 | spin_lock_irq(&conf->device_lock); | |
2241 | if (mbio->bi_phys_segments == 0) | |
2242 | mbio->bi_phys_segments = 2; | |
2243 | else | |
2244 | mbio->bi_phys_segments++; | |
2245 | spin_unlock_irq(&conf->device_lock); | |
2246 | generic_make_request(bio); | |
2247 | bio = NULL; | |
2248 | ||
2249 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
2250 | ||
2251 | r1_bio->master_bio = mbio; | |
aa8b57aa | 2252 | r1_bio->sectors = bio_sectors(mbio) - sectors_handled; |
62096bce N |
2253 | r1_bio->state = 0; |
2254 | set_bit(R1BIO_ReadError, &r1_bio->state); | |
2255 | r1_bio->mddev = mddev; | |
2256 | r1_bio->sector = mbio->bi_sector + sectors_handled; | |
2257 | ||
2258 | goto read_more; | |
2259 | } else | |
2260 | generic_make_request(bio); | |
2261 | } | |
2262 | } | |
2263 | ||
4ed8731d | 2264 | static void raid1d(struct md_thread *thread) |
1da177e4 | 2265 | { |
4ed8731d | 2266 | struct mddev *mddev = thread->mddev; |
9f2c9d12 | 2267 | struct r1bio *r1_bio; |
1da177e4 | 2268 | unsigned long flags; |
e8096360 | 2269 | struct r1conf *conf = mddev->private; |
1da177e4 | 2270 | struct list_head *head = &conf->retry_list; |
e1dfa0a2 | 2271 | struct blk_plug plug; |
1da177e4 LT |
2272 | |
2273 | md_check_recovery(mddev); | |
e1dfa0a2 N |
2274 | |
2275 | blk_start_plug(&plug); | |
1da177e4 | 2276 | for (;;) { |
191ea9b2 | 2277 | |
0021b7bc | 2278 | flush_pending_writes(conf); |
191ea9b2 | 2279 | |
a35e63ef N |
2280 | spin_lock_irqsave(&conf->device_lock, flags); |
2281 | if (list_empty(head)) { | |
2282 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 | 2283 | break; |
a35e63ef | 2284 | } |
9f2c9d12 | 2285 | r1_bio = list_entry(head->prev, struct r1bio, retry_list); |
1da177e4 | 2286 | list_del(head->prev); |
ddaf22ab | 2287 | conf->nr_queued--; |
1da177e4 LT |
2288 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2289 | ||
2290 | mddev = r1_bio->mddev; | |
070ec55d | 2291 | conf = mddev->private; |
4367af55 | 2292 | if (test_bit(R1BIO_IsSync, &r1_bio->state)) { |
d8f05d29 | 2293 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
62096bce N |
2294 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
2295 | handle_sync_write_finished(conf, r1_bio); | |
2296 | else | |
4367af55 | 2297 | sync_request_write(mddev, r1_bio); |
cd5ff9a1 | 2298 | } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
62096bce N |
2299 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
2300 | handle_write_finished(conf, r1_bio); | |
2301 | else if (test_bit(R1BIO_ReadError, &r1_bio->state)) | |
2302 | handle_read_error(conf, r1_bio); | |
2303 | else | |
d2eb35ac N |
2304 | /* just a partial read to be scheduled from separate |
2305 | * context | |
2306 | */ | |
2307 | generic_make_request(r1_bio->bios[r1_bio->read_disk]); | |
62096bce | 2308 | |
1d9d5241 | 2309 | cond_resched(); |
de393cde N |
2310 | if (mddev->flags & ~(1<<MD_CHANGE_PENDING)) |
2311 | md_check_recovery(mddev); | |
1da177e4 | 2312 | } |
e1dfa0a2 | 2313 | blk_finish_plug(&plug); |
1da177e4 LT |
2314 | } |
2315 | ||
2316 | ||
e8096360 | 2317 | static int init_resync(struct r1conf *conf) |
1da177e4 LT |
2318 | { |
2319 | int buffs; | |
2320 | ||
2321 | buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; | |
9e77c485 | 2322 | BUG_ON(conf->r1buf_pool); |
1da177e4 LT |
2323 | conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free, |
2324 | conf->poolinfo); | |
2325 | if (!conf->r1buf_pool) | |
2326 | return -ENOMEM; | |
2327 | conf->next_resync = 0; | |
2328 | return 0; | |
2329 | } | |
2330 | ||
2331 | /* | |
2332 | * perform a "sync" on one "block" | |
2333 | * | |
2334 | * We need to make sure that no normal I/O request - particularly write | |
2335 | * requests - conflict with active sync requests. | |
2336 | * | |
2337 | * This is achieved by tracking pending requests and a 'barrier' concept | |
2338 | * that can be installed to exclude normal IO requests. | |
2339 | */ | |
2340 | ||
fd01b88c | 2341 | static sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster) |
1da177e4 | 2342 | { |
e8096360 | 2343 | struct r1conf *conf = mddev->private; |
9f2c9d12 | 2344 | struct r1bio *r1_bio; |
1da177e4 LT |
2345 | struct bio *bio; |
2346 | sector_t max_sector, nr_sectors; | |
3e198f78 | 2347 | int disk = -1; |
1da177e4 | 2348 | int i; |
3e198f78 N |
2349 | int wonly = -1; |
2350 | int write_targets = 0, read_targets = 0; | |
57dab0bd | 2351 | sector_t sync_blocks; |
e3b9703e | 2352 | int still_degraded = 0; |
06f60385 N |
2353 | int good_sectors = RESYNC_SECTORS; |
2354 | int min_bad = 0; /* number of sectors that are bad in all devices */ | |
1da177e4 LT |
2355 | |
2356 | if (!conf->r1buf_pool) | |
2357 | if (init_resync(conf)) | |
57afd89f | 2358 | return 0; |
1da177e4 | 2359 | |
58c0fed4 | 2360 | max_sector = mddev->dev_sectors; |
1da177e4 | 2361 | if (sector_nr >= max_sector) { |
191ea9b2 N |
2362 | /* If we aborted, we need to abort the |
2363 | * sync on the 'current' bitmap chunk (there will | |
2364 | * only be one in raid1 resync. | |
2365 | * We can find the current addess in mddev->curr_resync | |
2366 | */ | |
6a806c51 N |
2367 | if (mddev->curr_resync < max_sector) /* aborted */ |
2368 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
191ea9b2 | 2369 | &sync_blocks, 1); |
6a806c51 | 2370 | else /* completed sync */ |
191ea9b2 | 2371 | conf->fullsync = 0; |
6a806c51 N |
2372 | |
2373 | bitmap_close_sync(mddev->bitmap); | |
1da177e4 LT |
2374 | close_sync(conf); |
2375 | return 0; | |
2376 | } | |
2377 | ||
07d84d10 N |
2378 | if (mddev->bitmap == NULL && |
2379 | mddev->recovery_cp == MaxSector && | |
6394cca5 | 2380 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
07d84d10 N |
2381 | conf->fullsync == 0) { |
2382 | *skipped = 1; | |
2383 | return max_sector - sector_nr; | |
2384 | } | |
6394cca5 N |
2385 | /* before building a request, check if we can skip these blocks.. |
2386 | * This call the bitmap_start_sync doesn't actually record anything | |
2387 | */ | |
e3b9703e | 2388 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
e5de485f | 2389 | !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { |
191ea9b2 N |
2390 | /* We can skip this block, and probably several more */ |
2391 | *skipped = 1; | |
2392 | return sync_blocks; | |
2393 | } | |
1da177e4 | 2394 | /* |
17999be4 N |
2395 | * If there is non-resync activity waiting for a turn, |
2396 | * and resync is going fast enough, | |
2397 | * then let it though before starting on this new sync request. | |
1da177e4 | 2398 | */ |
17999be4 | 2399 | if (!go_faster && conf->nr_waiting) |
1da177e4 | 2400 | msleep_interruptible(1000); |
17999be4 | 2401 | |
b47490c9 | 2402 | bitmap_cond_end_sync(mddev->bitmap, sector_nr); |
1c4588e9 | 2403 | r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO); |
17999be4 N |
2404 | raise_barrier(conf); |
2405 | ||
2406 | conf->next_resync = sector_nr; | |
1da177e4 | 2407 | |
3e198f78 | 2408 | rcu_read_lock(); |
1da177e4 | 2409 | /* |
3e198f78 N |
2410 | * If we get a correctably read error during resync or recovery, |
2411 | * we might want to read from a different device. So we | |
2412 | * flag all drives that could conceivably be read from for READ, | |
2413 | * and any others (which will be non-In_sync devices) for WRITE. | |
2414 | * If a read fails, we try reading from something else for which READ | |
2415 | * is OK. | |
1da177e4 | 2416 | */ |
1da177e4 | 2417 | |
1da177e4 LT |
2418 | r1_bio->mddev = mddev; |
2419 | r1_bio->sector = sector_nr; | |
191ea9b2 | 2420 | r1_bio->state = 0; |
1da177e4 | 2421 | set_bit(R1BIO_IsSync, &r1_bio->state); |
1da177e4 | 2422 | |
8f19ccb2 | 2423 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3cb03002 | 2424 | struct md_rdev *rdev; |
1da177e4 | 2425 | bio = r1_bio->bios[i]; |
2aabaa65 | 2426 | bio_reset(bio); |
1da177e4 | 2427 | |
3e198f78 N |
2428 | rdev = rcu_dereference(conf->mirrors[i].rdev); |
2429 | if (rdev == NULL || | |
06f60385 | 2430 | test_bit(Faulty, &rdev->flags)) { |
8f19ccb2 N |
2431 | if (i < conf->raid_disks) |
2432 | still_degraded = 1; | |
3e198f78 | 2433 | } else if (!test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
2434 | bio->bi_rw = WRITE; |
2435 | bio->bi_end_io = end_sync_write; | |
2436 | write_targets ++; | |
3e198f78 N |
2437 | } else { |
2438 | /* may need to read from here */ | |
06f60385 N |
2439 | sector_t first_bad = MaxSector; |
2440 | int bad_sectors; | |
2441 | ||
2442 | if (is_badblock(rdev, sector_nr, good_sectors, | |
2443 | &first_bad, &bad_sectors)) { | |
2444 | if (first_bad > sector_nr) | |
2445 | good_sectors = first_bad - sector_nr; | |
2446 | else { | |
2447 | bad_sectors -= (sector_nr - first_bad); | |
2448 | if (min_bad == 0 || | |
2449 | min_bad > bad_sectors) | |
2450 | min_bad = bad_sectors; | |
2451 | } | |
2452 | } | |
2453 | if (sector_nr < first_bad) { | |
2454 | if (test_bit(WriteMostly, &rdev->flags)) { | |
2455 | if (wonly < 0) | |
2456 | wonly = i; | |
2457 | } else { | |
2458 | if (disk < 0) | |
2459 | disk = i; | |
2460 | } | |
2461 | bio->bi_rw = READ; | |
2462 | bio->bi_end_io = end_sync_read; | |
2463 | read_targets++; | |
d57368af AL |
2464 | } else if (!test_bit(WriteErrorSeen, &rdev->flags) && |
2465 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && | |
2466 | !test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { | |
2467 | /* | |
2468 | * The device is suitable for reading (InSync), | |
2469 | * but has bad block(s) here. Let's try to correct them, | |
2470 | * if we are doing resync or repair. Otherwise, leave | |
2471 | * this device alone for this sync request. | |
2472 | */ | |
2473 | bio->bi_rw = WRITE; | |
2474 | bio->bi_end_io = end_sync_write; | |
2475 | write_targets++; | |
3e198f78 | 2476 | } |
3e198f78 | 2477 | } |
06f60385 N |
2478 | if (bio->bi_end_io) { |
2479 | atomic_inc(&rdev->nr_pending); | |
2480 | bio->bi_sector = sector_nr + rdev->data_offset; | |
2481 | bio->bi_bdev = rdev->bdev; | |
2482 | bio->bi_private = r1_bio; | |
2483 | } | |
1da177e4 | 2484 | } |
3e198f78 N |
2485 | rcu_read_unlock(); |
2486 | if (disk < 0) | |
2487 | disk = wonly; | |
2488 | r1_bio->read_disk = disk; | |
191ea9b2 | 2489 | |
06f60385 N |
2490 | if (read_targets == 0 && min_bad > 0) { |
2491 | /* These sectors are bad on all InSync devices, so we | |
2492 | * need to mark them bad on all write targets | |
2493 | */ | |
2494 | int ok = 1; | |
8f19ccb2 | 2495 | for (i = 0 ; i < conf->raid_disks * 2 ; i++) |
06f60385 | 2496 | if (r1_bio->bios[i]->bi_end_io == end_sync_write) { |
a42f9d83 | 2497 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
06f60385 N |
2498 | ok = rdev_set_badblocks(rdev, sector_nr, |
2499 | min_bad, 0 | |
2500 | ) && ok; | |
2501 | } | |
2502 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
2503 | *skipped = 1; | |
2504 | put_buf(r1_bio); | |
2505 | ||
2506 | if (!ok) { | |
2507 | /* Cannot record the badblocks, so need to | |
2508 | * abort the resync. | |
2509 | * If there are multiple read targets, could just | |
2510 | * fail the really bad ones ??? | |
2511 | */ | |
2512 | conf->recovery_disabled = mddev->recovery_disabled; | |
2513 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2514 | return 0; | |
2515 | } else | |
2516 | return min_bad; | |
2517 | ||
2518 | } | |
2519 | if (min_bad > 0 && min_bad < good_sectors) { | |
2520 | /* only resync enough to reach the next bad->good | |
2521 | * transition */ | |
2522 | good_sectors = min_bad; | |
2523 | } | |
2524 | ||
3e198f78 N |
2525 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0) |
2526 | /* extra read targets are also write targets */ | |
2527 | write_targets += read_targets-1; | |
2528 | ||
2529 | if (write_targets == 0 || read_targets == 0) { | |
1da177e4 LT |
2530 | /* There is nowhere to write, so all non-sync |
2531 | * drives must be failed - so we are finished | |
2532 | */ | |
b7219ccb N |
2533 | sector_t rv; |
2534 | if (min_bad > 0) | |
2535 | max_sector = sector_nr + min_bad; | |
2536 | rv = max_sector - sector_nr; | |
57afd89f | 2537 | *skipped = 1; |
1da177e4 | 2538 | put_buf(r1_bio); |
1da177e4 LT |
2539 | return rv; |
2540 | } | |
2541 | ||
c6207277 N |
2542 | if (max_sector > mddev->resync_max) |
2543 | max_sector = mddev->resync_max; /* Don't do IO beyond here */ | |
06f60385 N |
2544 | if (max_sector > sector_nr + good_sectors) |
2545 | max_sector = sector_nr + good_sectors; | |
1da177e4 | 2546 | nr_sectors = 0; |
289e99e8 | 2547 | sync_blocks = 0; |
1da177e4 LT |
2548 | do { |
2549 | struct page *page; | |
2550 | int len = PAGE_SIZE; | |
2551 | if (sector_nr + (len>>9) > max_sector) | |
2552 | len = (max_sector - sector_nr) << 9; | |
2553 | if (len == 0) | |
2554 | break; | |
6a806c51 N |
2555 | if (sync_blocks == 0) { |
2556 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, | |
e5de485f N |
2557 | &sync_blocks, still_degraded) && |
2558 | !conf->fullsync && | |
2559 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6a806c51 | 2560 | break; |
9e77c485 | 2561 | BUG_ON(sync_blocks < (PAGE_SIZE>>9)); |
7571ae88 | 2562 | if ((len >> 9) > sync_blocks) |
6a806c51 | 2563 | len = sync_blocks<<9; |
ab7a30c7 | 2564 | } |
191ea9b2 | 2565 | |
8f19ccb2 | 2566 | for (i = 0 ; i < conf->raid_disks * 2; i++) { |
1da177e4 LT |
2567 | bio = r1_bio->bios[i]; |
2568 | if (bio->bi_end_io) { | |
d11c171e | 2569 | page = bio->bi_io_vec[bio->bi_vcnt].bv_page; |
1da177e4 LT |
2570 | if (bio_add_page(bio, page, len, 0) == 0) { |
2571 | /* stop here */ | |
d11c171e | 2572 | bio->bi_io_vec[bio->bi_vcnt].bv_page = page; |
1da177e4 LT |
2573 | while (i > 0) { |
2574 | i--; | |
2575 | bio = r1_bio->bios[i]; | |
6a806c51 N |
2576 | if (bio->bi_end_io==NULL) |
2577 | continue; | |
1da177e4 LT |
2578 | /* remove last page from this bio */ |
2579 | bio->bi_vcnt--; | |
2580 | bio->bi_size -= len; | |
2581 | bio->bi_flags &= ~(1<< BIO_SEG_VALID); | |
2582 | } | |
2583 | goto bio_full; | |
2584 | } | |
2585 | } | |
2586 | } | |
2587 | nr_sectors += len>>9; | |
2588 | sector_nr += len>>9; | |
191ea9b2 | 2589 | sync_blocks -= (len>>9); |
1da177e4 LT |
2590 | } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES); |
2591 | bio_full: | |
1da177e4 LT |
2592 | r1_bio->sectors = nr_sectors; |
2593 | ||
d11c171e N |
2594 | /* For a user-requested sync, we read all readable devices and do a |
2595 | * compare | |
2596 | */ | |
2597 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
2598 | atomic_set(&r1_bio->remaining, read_targets); | |
2d4f4f33 | 2599 | for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) { |
d11c171e N |
2600 | bio = r1_bio->bios[i]; |
2601 | if (bio->bi_end_io == end_sync_read) { | |
2d4f4f33 | 2602 | read_targets--; |
ddac7c7e | 2603 | md_sync_acct(bio->bi_bdev, nr_sectors); |
d11c171e N |
2604 | generic_make_request(bio); |
2605 | } | |
2606 | } | |
2607 | } else { | |
2608 | atomic_set(&r1_bio->remaining, 1); | |
2609 | bio = r1_bio->bios[r1_bio->read_disk]; | |
ddac7c7e | 2610 | md_sync_acct(bio->bi_bdev, nr_sectors); |
d11c171e | 2611 | generic_make_request(bio); |
1da177e4 | 2612 | |
d11c171e | 2613 | } |
1da177e4 LT |
2614 | return nr_sectors; |
2615 | } | |
2616 | ||
fd01b88c | 2617 | static sector_t raid1_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce DW |
2618 | { |
2619 | if (sectors) | |
2620 | return sectors; | |
2621 | ||
2622 | return mddev->dev_sectors; | |
2623 | } | |
2624 | ||
e8096360 | 2625 | static struct r1conf *setup_conf(struct mddev *mddev) |
1da177e4 | 2626 | { |
e8096360 | 2627 | struct r1conf *conf; |
709ae487 | 2628 | int i; |
0eaf822c | 2629 | struct raid1_info *disk; |
3cb03002 | 2630 | struct md_rdev *rdev; |
709ae487 | 2631 | int err = -ENOMEM; |
1da177e4 | 2632 | |
e8096360 | 2633 | conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL); |
1da177e4 | 2634 | if (!conf) |
709ae487 | 2635 | goto abort; |
1da177e4 | 2636 | |
0eaf822c | 2637 | conf->mirrors = kzalloc(sizeof(struct raid1_info) |
8f19ccb2 | 2638 | * mddev->raid_disks * 2, |
1da177e4 LT |
2639 | GFP_KERNEL); |
2640 | if (!conf->mirrors) | |
709ae487 | 2641 | goto abort; |
1da177e4 | 2642 | |
ddaf22ab N |
2643 | conf->tmppage = alloc_page(GFP_KERNEL); |
2644 | if (!conf->tmppage) | |
709ae487 | 2645 | goto abort; |
ddaf22ab | 2646 | |
709ae487 | 2647 | conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL); |
1da177e4 | 2648 | if (!conf->poolinfo) |
709ae487 | 2649 | goto abort; |
8f19ccb2 | 2650 | conf->poolinfo->raid_disks = mddev->raid_disks * 2; |
1da177e4 LT |
2651 | conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, |
2652 | r1bio_pool_free, | |
2653 | conf->poolinfo); | |
2654 | if (!conf->r1bio_pool) | |
709ae487 N |
2655 | goto abort; |
2656 | ||
ed9bfdf1 | 2657 | conf->poolinfo->mddev = mddev; |
1da177e4 | 2658 | |
c19d5798 | 2659 | err = -EINVAL; |
e7e72bf6 | 2660 | spin_lock_init(&conf->device_lock); |
dafb20fa | 2661 | rdev_for_each(rdev, mddev) { |
aba336bd | 2662 | struct request_queue *q; |
709ae487 | 2663 | int disk_idx = rdev->raid_disk; |
1da177e4 LT |
2664 | if (disk_idx >= mddev->raid_disks |
2665 | || disk_idx < 0) | |
2666 | continue; | |
c19d5798 | 2667 | if (test_bit(Replacement, &rdev->flags)) |
02b898f2 | 2668 | disk = conf->mirrors + mddev->raid_disks + disk_idx; |
c19d5798 N |
2669 | else |
2670 | disk = conf->mirrors + disk_idx; | |
1da177e4 | 2671 | |
c19d5798 N |
2672 | if (disk->rdev) |
2673 | goto abort; | |
1da177e4 | 2674 | disk->rdev = rdev; |
aba336bd N |
2675 | q = bdev_get_queue(rdev->bdev); |
2676 | if (q->merge_bvec_fn) | |
2677 | mddev->merge_check_needed = 1; | |
1da177e4 LT |
2678 | |
2679 | disk->head_position = 0; | |
12cee5a8 | 2680 | disk->seq_start = MaxSector; |
1da177e4 LT |
2681 | } |
2682 | conf->raid_disks = mddev->raid_disks; | |
2683 | conf->mddev = mddev; | |
1da177e4 | 2684 | INIT_LIST_HEAD(&conf->retry_list); |
1da177e4 LT |
2685 | |
2686 | spin_lock_init(&conf->resync_lock); | |
17999be4 | 2687 | init_waitqueue_head(&conf->wait_barrier); |
1da177e4 | 2688 | |
191ea9b2 | 2689 | bio_list_init(&conf->pending_bio_list); |
34db0cd6 | 2690 | conf->pending_count = 0; |
d890fa2b | 2691 | conf->recovery_disabled = mddev->recovery_disabled - 1; |
191ea9b2 | 2692 | |
c19d5798 | 2693 | err = -EIO; |
8f19ccb2 | 2694 | for (i = 0; i < conf->raid_disks * 2; i++) { |
1da177e4 LT |
2695 | |
2696 | disk = conf->mirrors + i; | |
2697 | ||
c19d5798 N |
2698 | if (i < conf->raid_disks && |
2699 | disk[conf->raid_disks].rdev) { | |
2700 | /* This slot has a replacement. */ | |
2701 | if (!disk->rdev) { | |
2702 | /* No original, just make the replacement | |
2703 | * a recovering spare | |
2704 | */ | |
2705 | disk->rdev = | |
2706 | disk[conf->raid_disks].rdev; | |
2707 | disk[conf->raid_disks].rdev = NULL; | |
2708 | } else if (!test_bit(In_sync, &disk->rdev->flags)) | |
2709 | /* Original is not in_sync - bad */ | |
2710 | goto abort; | |
2711 | } | |
2712 | ||
5fd6c1dc N |
2713 | if (!disk->rdev || |
2714 | !test_bit(In_sync, &disk->rdev->flags)) { | |
1da177e4 | 2715 | disk->head_position = 0; |
4f0a5e01 JB |
2716 | if (disk->rdev && |
2717 | (disk->rdev->saved_raid_disk < 0)) | |
918f0238 | 2718 | conf->fullsync = 1; |
be4d3280 | 2719 | } |
1da177e4 | 2720 | } |
709ae487 | 2721 | |
709ae487 | 2722 | err = -ENOMEM; |
0232605d | 2723 | conf->thread = md_register_thread(raid1d, mddev, "raid1"); |
709ae487 N |
2724 | if (!conf->thread) { |
2725 | printk(KERN_ERR | |
9dd1e2fa | 2726 | "md/raid1:%s: couldn't allocate thread\n", |
709ae487 N |
2727 | mdname(mddev)); |
2728 | goto abort; | |
11ce99e6 | 2729 | } |
1da177e4 | 2730 | |
709ae487 N |
2731 | return conf; |
2732 | ||
2733 | abort: | |
2734 | if (conf) { | |
2735 | if (conf->r1bio_pool) | |
2736 | mempool_destroy(conf->r1bio_pool); | |
2737 | kfree(conf->mirrors); | |
2738 | safe_put_page(conf->tmppage); | |
2739 | kfree(conf->poolinfo); | |
2740 | kfree(conf); | |
2741 | } | |
2742 | return ERR_PTR(err); | |
2743 | } | |
2744 | ||
5220ea1e | 2745 | static int stop(struct mddev *mddev); |
fd01b88c | 2746 | static int run(struct mddev *mddev) |
709ae487 | 2747 | { |
e8096360 | 2748 | struct r1conf *conf; |
709ae487 | 2749 | int i; |
3cb03002 | 2750 | struct md_rdev *rdev; |
5220ea1e | 2751 | int ret; |
2ff8cc2c | 2752 | bool discard_supported = false; |
709ae487 N |
2753 | |
2754 | if (mddev->level != 1) { | |
9dd1e2fa | 2755 | printk(KERN_ERR "md/raid1:%s: raid level not set to mirroring (%d)\n", |
709ae487 N |
2756 | mdname(mddev), mddev->level); |
2757 | return -EIO; | |
2758 | } | |
2759 | if (mddev->reshape_position != MaxSector) { | |
9dd1e2fa | 2760 | printk(KERN_ERR "md/raid1:%s: reshape_position set but not supported\n", |
709ae487 N |
2761 | mdname(mddev)); |
2762 | return -EIO; | |
2763 | } | |
1da177e4 | 2764 | /* |
709ae487 N |
2765 | * copy the already verified devices into our private RAID1 |
2766 | * bookkeeping area. [whatever we allocate in run(), | |
2767 | * should be freed in stop()] | |
1da177e4 | 2768 | */ |
709ae487 N |
2769 | if (mddev->private == NULL) |
2770 | conf = setup_conf(mddev); | |
2771 | else | |
2772 | conf = mddev->private; | |
1da177e4 | 2773 | |
709ae487 N |
2774 | if (IS_ERR(conf)) |
2775 | return PTR_ERR(conf); | |
1da177e4 | 2776 | |
c8dc9c65 JL |
2777 | if (mddev->queue) |
2778 | blk_queue_max_write_same_sectors(mddev->queue, | |
2779 | mddev->chunk_sectors); | |
dafb20fa | 2780 | rdev_for_each(rdev, mddev) { |
1ed7242e JB |
2781 | if (!mddev->gendisk) |
2782 | continue; | |
709ae487 N |
2783 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
2784 | rdev->data_offset << 9); | |
2ff8cc2c SL |
2785 | if (blk_queue_discard(bdev_get_queue(rdev->bdev))) |
2786 | discard_supported = true; | |
1da177e4 | 2787 | } |
191ea9b2 | 2788 | |
709ae487 N |
2789 | mddev->degraded = 0; |
2790 | for (i=0; i < conf->raid_disks; i++) | |
2791 | if (conf->mirrors[i].rdev == NULL || | |
2792 | !test_bit(In_sync, &conf->mirrors[i].rdev->flags) || | |
2793 | test_bit(Faulty, &conf->mirrors[i].rdev->flags)) | |
2794 | mddev->degraded++; | |
2795 | ||
2796 | if (conf->raid_disks - mddev->degraded == 1) | |
2797 | mddev->recovery_cp = MaxSector; | |
2798 | ||
8c6ac868 | 2799 | if (mddev->recovery_cp != MaxSector) |
9dd1e2fa | 2800 | printk(KERN_NOTICE "md/raid1:%s: not clean" |
8c6ac868 AN |
2801 | " -- starting background reconstruction\n", |
2802 | mdname(mddev)); | |
1da177e4 | 2803 | printk(KERN_INFO |
9dd1e2fa | 2804 | "md/raid1:%s: active with %d out of %d mirrors\n", |
1da177e4 LT |
2805 | mdname(mddev), mddev->raid_disks - mddev->degraded, |
2806 | mddev->raid_disks); | |
709ae487 | 2807 | |
1da177e4 LT |
2808 | /* |
2809 | * Ok, everything is just fine now | |
2810 | */ | |
709ae487 N |
2811 | mddev->thread = conf->thread; |
2812 | conf->thread = NULL; | |
2813 | mddev->private = conf; | |
2814 | ||
1f403624 | 2815 | md_set_array_sectors(mddev, raid1_size(mddev, 0, 0)); |
1da177e4 | 2816 | |
1ed7242e JB |
2817 | if (mddev->queue) { |
2818 | mddev->queue->backing_dev_info.congested_fn = raid1_congested; | |
2819 | mddev->queue->backing_dev_info.congested_data = mddev; | |
6b740b8d | 2820 | blk_queue_merge_bvec(mddev->queue, raid1_mergeable_bvec); |
2ff8cc2c SL |
2821 | |
2822 | if (discard_supported) | |
2823 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, | |
2824 | mddev->queue); | |
2825 | else | |
2826 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, | |
2827 | mddev->queue); | |
1ed7242e | 2828 | } |
5220ea1e | 2829 | |
2830 | ret = md_integrity_register(mddev); | |
2831 | if (ret) | |
2832 | stop(mddev); | |
2833 | return ret; | |
1da177e4 LT |
2834 | } |
2835 | ||
fd01b88c | 2836 | static int stop(struct mddev *mddev) |
1da177e4 | 2837 | { |
e8096360 | 2838 | struct r1conf *conf = mddev->private; |
4b6d287f | 2839 | struct bitmap *bitmap = mddev->bitmap; |
4b6d287f N |
2840 | |
2841 | /* wait for behind writes to complete */ | |
e555190d | 2842 | if (bitmap && atomic_read(&bitmap->behind_writes) > 0) { |
9dd1e2fa N |
2843 | printk(KERN_INFO "md/raid1:%s: behind writes in progress - waiting to stop.\n", |
2844 | mdname(mddev)); | |
4b6d287f | 2845 | /* need to kick something here to make sure I/O goes? */ |
e555190d N |
2846 | wait_event(bitmap->behind_wait, |
2847 | atomic_read(&bitmap->behind_writes) == 0); | |
4b6d287f | 2848 | } |
1da177e4 | 2849 | |
409c57f3 N |
2850 | raise_barrier(conf); |
2851 | lower_barrier(conf); | |
2852 | ||
01f96c0a | 2853 | md_unregister_thread(&mddev->thread); |
1da177e4 LT |
2854 | if (conf->r1bio_pool) |
2855 | mempool_destroy(conf->r1bio_pool); | |
990a8baf JJ |
2856 | kfree(conf->mirrors); |
2857 | kfree(conf->poolinfo); | |
1da177e4 LT |
2858 | kfree(conf); |
2859 | mddev->private = NULL; | |
2860 | return 0; | |
2861 | } | |
2862 | ||
fd01b88c | 2863 | static int raid1_resize(struct mddev *mddev, sector_t sectors) |
1da177e4 LT |
2864 | { |
2865 | /* no resync is happening, and there is enough space | |
2866 | * on all devices, so we can resize. | |
2867 | * We need to make sure resync covers any new space. | |
2868 | * If the array is shrinking we should possibly wait until | |
2869 | * any io in the removed space completes, but it hardly seems | |
2870 | * worth it. | |
2871 | */ | |
a4a6125a N |
2872 | sector_t newsize = raid1_size(mddev, sectors, 0); |
2873 | if (mddev->external_size && | |
2874 | mddev->array_sectors > newsize) | |
b522adcd | 2875 | return -EINVAL; |
a4a6125a N |
2876 | if (mddev->bitmap) { |
2877 | int ret = bitmap_resize(mddev->bitmap, newsize, 0, 0); | |
2878 | if (ret) | |
2879 | return ret; | |
2880 | } | |
2881 | md_set_array_sectors(mddev, newsize); | |
f233ea5c | 2882 | set_capacity(mddev->gendisk, mddev->array_sectors); |
449aad3e | 2883 | revalidate_disk(mddev->gendisk); |
b522adcd | 2884 | if (sectors > mddev->dev_sectors && |
b098636c | 2885 | mddev->recovery_cp > mddev->dev_sectors) { |
58c0fed4 | 2886 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
2887 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
2888 | } | |
b522adcd | 2889 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 2890 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
2891 | return 0; |
2892 | } | |
2893 | ||
fd01b88c | 2894 | static int raid1_reshape(struct mddev *mddev) |
1da177e4 LT |
2895 | { |
2896 | /* We need to: | |
2897 | * 1/ resize the r1bio_pool | |
2898 | * 2/ resize conf->mirrors | |
2899 | * | |
2900 | * We allocate a new r1bio_pool if we can. | |
2901 | * Then raise a device barrier and wait until all IO stops. | |
2902 | * Then resize conf->mirrors and swap in the new r1bio pool. | |
6ea9c07c N |
2903 | * |
2904 | * At the same time, we "pack" the devices so that all the missing | |
2905 | * devices have the higher raid_disk numbers. | |
1da177e4 LT |
2906 | */ |
2907 | mempool_t *newpool, *oldpool; | |
2908 | struct pool_info *newpoolinfo; | |
0eaf822c | 2909 | struct raid1_info *newmirrors; |
e8096360 | 2910 | struct r1conf *conf = mddev->private; |
63c70c4f | 2911 | int cnt, raid_disks; |
c04be0aa | 2912 | unsigned long flags; |
b5470dc5 | 2913 | int d, d2, err; |
1da177e4 | 2914 | |
63c70c4f | 2915 | /* Cannot change chunk_size, layout, or level */ |
664e7c41 | 2916 | if (mddev->chunk_sectors != mddev->new_chunk_sectors || |
63c70c4f N |
2917 | mddev->layout != mddev->new_layout || |
2918 | mddev->level != mddev->new_level) { | |
664e7c41 | 2919 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
63c70c4f N |
2920 | mddev->new_layout = mddev->layout; |
2921 | mddev->new_level = mddev->level; | |
2922 | return -EINVAL; | |
2923 | } | |
2924 | ||
b5470dc5 DW |
2925 | err = md_allow_write(mddev); |
2926 | if (err) | |
2927 | return err; | |
2a2275d6 | 2928 | |
63c70c4f N |
2929 | raid_disks = mddev->raid_disks + mddev->delta_disks; |
2930 | ||
6ea9c07c N |
2931 | if (raid_disks < conf->raid_disks) { |
2932 | cnt=0; | |
2933 | for (d= 0; d < conf->raid_disks; d++) | |
2934 | if (conf->mirrors[d].rdev) | |
2935 | cnt++; | |
2936 | if (cnt > raid_disks) | |
1da177e4 | 2937 | return -EBUSY; |
6ea9c07c | 2938 | } |
1da177e4 LT |
2939 | |
2940 | newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL); | |
2941 | if (!newpoolinfo) | |
2942 | return -ENOMEM; | |
2943 | newpoolinfo->mddev = mddev; | |
8f19ccb2 | 2944 | newpoolinfo->raid_disks = raid_disks * 2; |
1da177e4 LT |
2945 | |
2946 | newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, | |
2947 | r1bio_pool_free, newpoolinfo); | |
2948 | if (!newpool) { | |
2949 | kfree(newpoolinfo); | |
2950 | return -ENOMEM; | |
2951 | } | |
0eaf822c | 2952 | newmirrors = kzalloc(sizeof(struct raid1_info) * raid_disks * 2, |
8f19ccb2 | 2953 | GFP_KERNEL); |
1da177e4 LT |
2954 | if (!newmirrors) { |
2955 | kfree(newpoolinfo); | |
2956 | mempool_destroy(newpool); | |
2957 | return -ENOMEM; | |
2958 | } | |
1da177e4 | 2959 | |
17999be4 | 2960 | raise_barrier(conf); |
1da177e4 LT |
2961 | |
2962 | /* ok, everything is stopped */ | |
2963 | oldpool = conf->r1bio_pool; | |
2964 | conf->r1bio_pool = newpool; | |
6ea9c07c | 2965 | |
a88aa786 | 2966 | for (d = d2 = 0; d < conf->raid_disks; d++) { |
3cb03002 | 2967 | struct md_rdev *rdev = conf->mirrors[d].rdev; |
a88aa786 | 2968 | if (rdev && rdev->raid_disk != d2) { |
36fad858 | 2969 | sysfs_unlink_rdev(mddev, rdev); |
a88aa786 | 2970 | rdev->raid_disk = d2; |
36fad858 NK |
2971 | sysfs_unlink_rdev(mddev, rdev); |
2972 | if (sysfs_link_rdev(mddev, rdev)) | |
a88aa786 | 2973 | printk(KERN_WARNING |
36fad858 NK |
2974 | "md/raid1:%s: cannot register rd%d\n", |
2975 | mdname(mddev), rdev->raid_disk); | |
6ea9c07c | 2976 | } |
a88aa786 N |
2977 | if (rdev) |
2978 | newmirrors[d2++].rdev = rdev; | |
2979 | } | |
1da177e4 LT |
2980 | kfree(conf->mirrors); |
2981 | conf->mirrors = newmirrors; | |
2982 | kfree(conf->poolinfo); | |
2983 | conf->poolinfo = newpoolinfo; | |
2984 | ||
c04be0aa | 2985 | spin_lock_irqsave(&conf->device_lock, flags); |
1da177e4 | 2986 | mddev->degraded += (raid_disks - conf->raid_disks); |
c04be0aa | 2987 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 2988 | conf->raid_disks = mddev->raid_disks = raid_disks; |
63c70c4f | 2989 | mddev->delta_disks = 0; |
1da177e4 | 2990 | |
17999be4 | 2991 | lower_barrier(conf); |
1da177e4 LT |
2992 | |
2993 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
2994 | md_wakeup_thread(mddev->thread); | |
2995 | ||
2996 | mempool_destroy(oldpool); | |
2997 | return 0; | |
2998 | } | |
2999 | ||
fd01b88c | 3000 | static void raid1_quiesce(struct mddev *mddev, int state) |
36fa3063 | 3001 | { |
e8096360 | 3002 | struct r1conf *conf = mddev->private; |
36fa3063 N |
3003 | |
3004 | switch(state) { | |
6eef4b21 N |
3005 | case 2: /* wake for suspend */ |
3006 | wake_up(&conf->wait_barrier); | |
3007 | break; | |
9e6603da | 3008 | case 1: |
17999be4 | 3009 | raise_barrier(conf); |
36fa3063 | 3010 | break; |
9e6603da | 3011 | case 0: |
17999be4 | 3012 | lower_barrier(conf); |
36fa3063 N |
3013 | break; |
3014 | } | |
36fa3063 N |
3015 | } |
3016 | ||
fd01b88c | 3017 | static void *raid1_takeover(struct mddev *mddev) |
709ae487 N |
3018 | { |
3019 | /* raid1 can take over: | |
3020 | * raid5 with 2 devices, any layout or chunk size | |
3021 | */ | |
3022 | if (mddev->level == 5 && mddev->raid_disks == 2) { | |
e8096360 | 3023 | struct r1conf *conf; |
709ae487 N |
3024 | mddev->new_level = 1; |
3025 | mddev->new_layout = 0; | |
3026 | mddev->new_chunk_sectors = 0; | |
3027 | conf = setup_conf(mddev); | |
3028 | if (!IS_ERR(conf)) | |
3029 | conf->barrier = 1; | |
3030 | return conf; | |
3031 | } | |
3032 | return ERR_PTR(-EINVAL); | |
3033 | } | |
1da177e4 | 3034 | |
84fc4b56 | 3035 | static struct md_personality raid1_personality = |
1da177e4 LT |
3036 | { |
3037 | .name = "raid1", | |
2604b703 | 3038 | .level = 1, |
1da177e4 LT |
3039 | .owner = THIS_MODULE, |
3040 | .make_request = make_request, | |
3041 | .run = run, | |
3042 | .stop = stop, | |
3043 | .status = status, | |
3044 | .error_handler = error, | |
3045 | .hot_add_disk = raid1_add_disk, | |
3046 | .hot_remove_disk= raid1_remove_disk, | |
3047 | .spare_active = raid1_spare_active, | |
3048 | .sync_request = sync_request, | |
3049 | .resize = raid1_resize, | |
80c3a6ce | 3050 | .size = raid1_size, |
63c70c4f | 3051 | .check_reshape = raid1_reshape, |
36fa3063 | 3052 | .quiesce = raid1_quiesce, |
709ae487 | 3053 | .takeover = raid1_takeover, |
1da177e4 LT |
3054 | }; |
3055 | ||
3056 | static int __init raid_init(void) | |
3057 | { | |
2604b703 | 3058 | return register_md_personality(&raid1_personality); |
1da177e4 LT |
3059 | } |
3060 | ||
3061 | static void raid_exit(void) | |
3062 | { | |
2604b703 | 3063 | unregister_md_personality(&raid1_personality); |
1da177e4 LT |
3064 | } |
3065 | ||
3066 | module_init(raid_init); | |
3067 | module_exit(raid_exit); | |
3068 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 3069 | MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD"); |
1da177e4 | 3070 | MODULE_ALIAS("md-personality-3"); /* RAID1 */ |
d9d166c2 | 3071 | MODULE_ALIAS("md-raid1"); |
2604b703 | 3072 | MODULE_ALIAS("md-level-1"); |
34db0cd6 N |
3073 | |
3074 | module_param(max_queued_requests, int, S_IRUGO|S_IWUSR); |