]>
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 | ||
25570727 | 34 | #include <linux/delay.h> |
bff61975 | 35 | #include <linux/blkdev.h> |
bff61975 | 36 | #include <linux/seq_file.h> |
43b2e5d8 | 37 | #include "md.h" |
ef740c37 CH |
38 | #include "raid1.h" |
39 | #include "bitmap.h" | |
191ea9b2 N |
40 | |
41 | #define DEBUG 0 | |
42 | #if DEBUG | |
43 | #define PRINTK(x...) printk(x) | |
44 | #else | |
45 | #define PRINTK(x...) | |
46 | #endif | |
1da177e4 LT |
47 | |
48 | /* | |
49 | * Number of guaranteed r1bios in case of extreme VM load: | |
50 | */ | |
51 | #define NR_RAID1_BIOS 256 | |
52 | ||
1da177e4 LT |
53 | |
54 | static void unplug_slaves(mddev_t *mddev); | |
55 | ||
17999be4 N |
56 | static void allow_barrier(conf_t *conf); |
57 | static void lower_barrier(conf_t *conf); | |
1da177e4 | 58 | |
dd0fc66f | 59 | static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 LT |
60 | { |
61 | struct pool_info *pi = data; | |
62 | r1bio_t *r1_bio; | |
63 | int size = offsetof(r1bio_t, bios[pi->raid_disks]); | |
64 | ||
65 | /* allocate a r1bio with room for raid_disks entries in the bios array */ | |
9ffae0cf | 66 | r1_bio = kzalloc(size, gfp_flags); |
ed9bfdf1 | 67 | if (!r1_bio && pi->mddev) |
1da177e4 LT |
68 | unplug_slaves(pi->mddev); |
69 | ||
70 | return r1_bio; | |
71 | } | |
72 | ||
73 | static void r1bio_pool_free(void *r1_bio, void *data) | |
74 | { | |
75 | kfree(r1_bio); | |
76 | } | |
77 | ||
78 | #define RESYNC_BLOCK_SIZE (64*1024) | |
79 | //#define RESYNC_BLOCK_SIZE PAGE_SIZE | |
80 | #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9) | |
81 | #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE) | |
82 | #define RESYNC_WINDOW (2048*1024) | |
83 | ||
dd0fc66f | 84 | static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 LT |
85 | { |
86 | struct pool_info *pi = data; | |
87 | struct page *page; | |
88 | r1bio_t *r1_bio; | |
89 | struct bio *bio; | |
90 | int i, j; | |
91 | ||
92 | r1_bio = r1bio_pool_alloc(gfp_flags, pi); | |
93 | if (!r1_bio) { | |
94 | unplug_slaves(pi->mddev); | |
95 | return NULL; | |
96 | } | |
97 | ||
98 | /* | |
99 | * Allocate bios : 1 for reading, n-1 for writing | |
100 | */ | |
101 | for (j = pi->raid_disks ; j-- ; ) { | |
102 | bio = bio_alloc(gfp_flags, RESYNC_PAGES); | |
103 | if (!bio) | |
104 | goto out_free_bio; | |
105 | r1_bio->bios[j] = bio; | |
106 | } | |
107 | /* | |
108 | * Allocate RESYNC_PAGES data pages and attach them to | |
d11c171e N |
109 | * the first bio. |
110 | * If this is a user-requested check/repair, allocate | |
111 | * RESYNC_PAGES for each bio. | |
1da177e4 | 112 | */ |
d11c171e N |
113 | if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) |
114 | j = pi->raid_disks; | |
115 | else | |
116 | j = 1; | |
117 | while(j--) { | |
118 | bio = r1_bio->bios[j]; | |
119 | for (i = 0; i < RESYNC_PAGES; i++) { | |
120 | page = alloc_page(gfp_flags); | |
121 | if (unlikely(!page)) | |
122 | goto out_free_pages; | |
123 | ||
124 | bio->bi_io_vec[i].bv_page = page; | |
303a0e11 | 125 | bio->bi_vcnt = i+1; |
d11c171e N |
126 | } |
127 | } | |
128 | /* If not user-requests, copy the page pointers to all bios */ | |
129 | if (!test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) { | |
130 | for (i=0; i<RESYNC_PAGES ; i++) | |
131 | for (j=1; j<pi->raid_disks; j++) | |
132 | r1_bio->bios[j]->bi_io_vec[i].bv_page = | |
133 | r1_bio->bios[0]->bi_io_vec[i].bv_page; | |
1da177e4 LT |
134 | } |
135 | ||
136 | r1_bio->master_bio = NULL; | |
137 | ||
138 | return r1_bio; | |
139 | ||
140 | out_free_pages: | |
303a0e11 N |
141 | for (j=0 ; j < pi->raid_disks; j++) |
142 | for (i=0; i < r1_bio->bios[j]->bi_vcnt ; i++) | |
143 | put_page(r1_bio->bios[j]->bi_io_vec[i].bv_page); | |
d11c171e | 144 | j = -1; |
1da177e4 LT |
145 | out_free_bio: |
146 | while ( ++j < pi->raid_disks ) | |
147 | bio_put(r1_bio->bios[j]); | |
148 | r1bio_pool_free(r1_bio, data); | |
149 | return NULL; | |
150 | } | |
151 | ||
152 | static void r1buf_pool_free(void *__r1_bio, void *data) | |
153 | { | |
154 | struct pool_info *pi = data; | |
d11c171e | 155 | int i,j; |
1da177e4 | 156 | r1bio_t *r1bio = __r1_bio; |
1da177e4 | 157 | |
d11c171e N |
158 | for (i = 0; i < RESYNC_PAGES; i++) |
159 | for (j = pi->raid_disks; j-- ;) { | |
160 | if (j == 0 || | |
161 | r1bio->bios[j]->bi_io_vec[i].bv_page != | |
162 | r1bio->bios[0]->bi_io_vec[i].bv_page) | |
1345b1d8 | 163 | safe_put_page(r1bio->bios[j]->bi_io_vec[i].bv_page); |
d11c171e | 164 | } |
1da177e4 LT |
165 | for (i=0 ; i < pi->raid_disks; i++) |
166 | bio_put(r1bio->bios[i]); | |
167 | ||
168 | r1bio_pool_free(r1bio, data); | |
169 | } | |
170 | ||
171 | static void put_all_bios(conf_t *conf, r1bio_t *r1_bio) | |
172 | { | |
173 | int i; | |
174 | ||
175 | for (i = 0; i < conf->raid_disks; i++) { | |
176 | struct bio **bio = r1_bio->bios + i; | |
cf30a473 | 177 | if (*bio && *bio != IO_BLOCKED) |
1da177e4 LT |
178 | bio_put(*bio); |
179 | *bio = NULL; | |
180 | } | |
181 | } | |
182 | ||
858119e1 | 183 | static void free_r1bio(r1bio_t *r1_bio) |
1da177e4 | 184 | { |
070ec55d | 185 | conf_t *conf = r1_bio->mddev->private; |
1da177e4 LT |
186 | |
187 | /* | |
188 | * Wake up any possible resync thread that waits for the device | |
189 | * to go idle. | |
190 | */ | |
17999be4 | 191 | allow_barrier(conf); |
1da177e4 LT |
192 | |
193 | put_all_bios(conf, r1_bio); | |
194 | mempool_free(r1_bio, conf->r1bio_pool); | |
195 | } | |
196 | ||
858119e1 | 197 | static void put_buf(r1bio_t *r1_bio) |
1da177e4 | 198 | { |
070ec55d | 199 | conf_t *conf = r1_bio->mddev->private; |
3e198f78 N |
200 | int i; |
201 | ||
202 | for (i=0; i<conf->raid_disks; i++) { | |
203 | struct bio *bio = r1_bio->bios[i]; | |
204 | if (bio->bi_end_io) | |
205 | rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev); | |
206 | } | |
1da177e4 LT |
207 | |
208 | mempool_free(r1_bio, conf->r1buf_pool); | |
209 | ||
17999be4 | 210 | lower_barrier(conf); |
1da177e4 LT |
211 | } |
212 | ||
213 | static void reschedule_retry(r1bio_t *r1_bio) | |
214 | { | |
215 | unsigned long flags; | |
216 | mddev_t *mddev = r1_bio->mddev; | |
070ec55d | 217 | conf_t *conf = mddev->private; |
1da177e4 LT |
218 | |
219 | spin_lock_irqsave(&conf->device_lock, flags); | |
220 | list_add(&r1_bio->retry_list, &conf->retry_list); | |
ddaf22ab | 221 | conf->nr_queued ++; |
1da177e4 LT |
222 | spin_unlock_irqrestore(&conf->device_lock, flags); |
223 | ||
17999be4 | 224 | wake_up(&conf->wait_barrier); |
1da177e4 LT |
225 | md_wakeup_thread(mddev->thread); |
226 | } | |
227 | ||
228 | /* | |
229 | * raid_end_bio_io() is called when we have finished servicing a mirrored | |
230 | * operation and are ready to return a success/failure code to the buffer | |
231 | * cache layer. | |
232 | */ | |
233 | static void raid_end_bio_io(r1bio_t *r1_bio) | |
234 | { | |
235 | struct bio *bio = r1_bio->master_bio; | |
236 | ||
4b6d287f N |
237 | /* if nobody has done the final endio yet, do it now */ |
238 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { | |
239 | PRINTK(KERN_DEBUG "raid1: sync end %s on sectors %llu-%llu\n", | |
240 | (bio_data_dir(bio) == WRITE) ? "write" : "read", | |
241 | (unsigned long long) bio->bi_sector, | |
242 | (unsigned long long) bio->bi_sector + | |
243 | (bio->bi_size >> 9) - 1); | |
244 | ||
6712ecf8 | 245 | bio_endio(bio, |
4b6d287f N |
246 | test_bit(R1BIO_Uptodate, &r1_bio->state) ? 0 : -EIO); |
247 | } | |
1da177e4 LT |
248 | free_r1bio(r1_bio); |
249 | } | |
250 | ||
251 | /* | |
252 | * Update disk head position estimator based on IRQ completion info. | |
253 | */ | |
254 | static inline void update_head_pos(int disk, r1bio_t *r1_bio) | |
255 | { | |
070ec55d | 256 | conf_t *conf = r1_bio->mddev->private; |
1da177e4 LT |
257 | |
258 | conf->mirrors[disk].head_position = | |
259 | r1_bio->sector + (r1_bio->sectors); | |
260 | } | |
261 | ||
6712ecf8 | 262 | static void raid1_end_read_request(struct bio *bio, int error) |
1da177e4 LT |
263 | { |
264 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
265 | r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private); | |
266 | int mirror; | |
070ec55d | 267 | conf_t *conf = r1_bio->mddev->private; |
1da177e4 | 268 | |
1da177e4 LT |
269 | mirror = r1_bio->read_disk; |
270 | /* | |
271 | * this branch is our 'one mirror IO has finished' event handler: | |
272 | */ | |
ddaf22ab N |
273 | update_head_pos(mirror, r1_bio); |
274 | ||
dd00a99e N |
275 | if (uptodate) |
276 | set_bit(R1BIO_Uptodate, &r1_bio->state); | |
277 | else { | |
278 | /* If all other devices have failed, we want to return | |
279 | * the error upwards rather than fail the last device. | |
280 | * Here we redefine "uptodate" to mean "Don't want to retry" | |
1da177e4 | 281 | */ |
dd00a99e N |
282 | unsigned long flags; |
283 | spin_lock_irqsave(&conf->device_lock, flags); | |
284 | if (r1_bio->mddev->degraded == conf->raid_disks || | |
285 | (r1_bio->mddev->degraded == conf->raid_disks-1 && | |
286 | !test_bit(Faulty, &conf->mirrors[mirror].rdev->flags))) | |
287 | uptodate = 1; | |
288 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
289 | } | |
1da177e4 | 290 | |
dd00a99e | 291 | if (uptodate) |
1da177e4 | 292 | raid_end_bio_io(r1_bio); |
dd00a99e | 293 | else { |
1da177e4 LT |
294 | /* |
295 | * oops, read error: | |
296 | */ | |
297 | char b[BDEVNAME_SIZE]; | |
298 | if (printk_ratelimit()) | |
299 | printk(KERN_ERR "raid1: %s: rescheduling sector %llu\n", | |
300 | bdevname(conf->mirrors[mirror].rdev->bdev,b), (unsigned long long)r1_bio->sector); | |
301 | reschedule_retry(r1_bio); | |
302 | } | |
303 | ||
304 | rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev); | |
1da177e4 LT |
305 | } |
306 | ||
6712ecf8 | 307 | static void raid1_end_write_request(struct bio *bio, int error) |
1da177e4 LT |
308 | { |
309 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
310 | r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private); | |
a9701a30 | 311 | int mirror, behind = test_bit(R1BIO_BehindIO, &r1_bio->state); |
070ec55d | 312 | conf_t *conf = r1_bio->mddev->private; |
04b857f7 | 313 | struct bio *to_put = NULL; |
1da177e4 | 314 | |
1da177e4 LT |
315 | |
316 | for (mirror = 0; mirror < conf->raid_disks; mirror++) | |
317 | if (r1_bio->bios[mirror] == bio) | |
318 | break; | |
319 | ||
bea27718 | 320 | if (error == -EOPNOTSUPP && test_bit(R1BIO_Barrier, &r1_bio->state)) { |
a9701a30 N |
321 | set_bit(BarriersNotsupp, &conf->mirrors[mirror].rdev->flags); |
322 | set_bit(R1BIO_BarrierRetry, &r1_bio->state); | |
323 | r1_bio->mddev->barriers_work = 0; | |
5e7dd2ab | 324 | /* Don't rdev_dec_pending in this branch - keep it for the retry */ |
a9701a30 | 325 | } else { |
1da177e4 | 326 | /* |
a9701a30 | 327 | * this branch is our 'one mirror IO has finished' event handler: |
1da177e4 | 328 | */ |
a9701a30 | 329 | r1_bio->bios[mirror] = NULL; |
04b857f7 | 330 | to_put = bio; |
a9701a30 N |
331 | if (!uptodate) { |
332 | md_error(r1_bio->mddev, conf->mirrors[mirror].rdev); | |
333 | /* an I/O failed, we can't clear the bitmap */ | |
334 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
335 | } else | |
336 | /* | |
337 | * Set R1BIO_Uptodate in our master bio, so that | |
338 | * we will return a good error code for to the higher | |
339 | * levels even if IO on some other mirrored buffer fails. | |
340 | * | |
341 | * The 'master' represents the composite IO operation to | |
342 | * user-side. So if something waits for IO, then it will | |
343 | * wait for the 'master' bio. | |
344 | */ | |
345 | set_bit(R1BIO_Uptodate, &r1_bio->state); | |
346 | ||
347 | update_head_pos(mirror, r1_bio); | |
348 | ||
349 | if (behind) { | |
350 | if (test_bit(WriteMostly, &conf->mirrors[mirror].rdev->flags)) | |
351 | atomic_dec(&r1_bio->behind_remaining); | |
352 | ||
353 | /* In behind mode, we ACK the master bio once the I/O has safely | |
354 | * reached all non-writemostly disks. Setting the Returned bit | |
355 | * ensures that this gets done only once -- we don't ever want to | |
356 | * return -EIO here, instead we'll wait */ | |
357 | ||
358 | if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) && | |
359 | test_bit(R1BIO_Uptodate, &r1_bio->state)) { | |
360 | /* Maybe we can return now */ | |
361 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { | |
362 | struct bio *mbio = r1_bio->master_bio; | |
363 | PRINTK(KERN_DEBUG "raid1: behind end write sectors %llu-%llu\n", | |
364 | (unsigned long long) mbio->bi_sector, | |
365 | (unsigned long long) mbio->bi_sector + | |
366 | (mbio->bi_size >> 9) - 1); | |
6712ecf8 | 367 | bio_endio(mbio, 0); |
a9701a30 | 368 | } |
4b6d287f N |
369 | } |
370 | } | |
5e7dd2ab | 371 | rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev); |
4b6d287f | 372 | } |
1da177e4 LT |
373 | /* |
374 | * | |
375 | * Let's see if all mirrored write operations have finished | |
376 | * already. | |
377 | */ | |
378 | if (atomic_dec_and_test(&r1_bio->remaining)) { | |
c70810b3 | 379 | if (test_bit(R1BIO_BarrierRetry, &r1_bio->state)) |
a9701a30 | 380 | reschedule_retry(r1_bio); |
c70810b3 N |
381 | else { |
382 | /* it really is the end of this request */ | |
383 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { | |
384 | /* free extra copy of the data pages */ | |
385 | int i = bio->bi_vcnt; | |
386 | while (i--) | |
387 | safe_put_page(bio->bi_io_vec[i].bv_page); | |
388 | } | |
389 | /* clear the bitmap if all writes complete successfully */ | |
390 | bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector, | |
391 | r1_bio->sectors, | |
392 | !test_bit(R1BIO_Degraded, &r1_bio->state), | |
393 | behind); | |
394 | md_write_end(r1_bio->mddev); | |
395 | raid_end_bio_io(r1_bio); | |
4b6d287f | 396 | } |
1da177e4 | 397 | } |
c70810b3 | 398 | |
04b857f7 N |
399 | if (to_put) |
400 | bio_put(to_put); | |
1da177e4 LT |
401 | } |
402 | ||
403 | ||
404 | /* | |
405 | * This routine returns the disk from which the requested read should | |
406 | * be done. There is a per-array 'next expected sequential IO' sector | |
407 | * number - if this matches on the next IO then we use the last disk. | |
408 | * There is also a per-disk 'last know head position' sector that is | |
409 | * maintained from IRQ contexts, both the normal and the resync IO | |
410 | * completion handlers update this position correctly. If there is no | |
411 | * perfect sequential match then we pick the disk whose head is closest. | |
412 | * | |
413 | * If there are 2 mirrors in the same 2 devices, performance degrades | |
414 | * because position is mirror, not device based. | |
415 | * | |
416 | * The rdev for the device selected will have nr_pending incremented. | |
417 | */ | |
418 | static int read_balance(conf_t *conf, r1bio_t *r1_bio) | |
419 | { | |
420 | const unsigned long this_sector = r1_bio->sector; | |
421 | int new_disk = conf->last_used, disk = new_disk; | |
8ddf9efe | 422 | int wonly_disk = -1; |
1da177e4 LT |
423 | const int sectors = r1_bio->sectors; |
424 | sector_t new_distance, current_distance; | |
8ddf9efe | 425 | mdk_rdev_t *rdev; |
1da177e4 LT |
426 | |
427 | rcu_read_lock(); | |
428 | /* | |
8ddf9efe | 429 | * Check if we can balance. We can balance on the whole |
1da177e4 LT |
430 | * device if no resync is going on, or below the resync window. |
431 | * We take the first readable disk when above the resync window. | |
432 | */ | |
433 | retry: | |
434 | if (conf->mddev->recovery_cp < MaxSector && | |
435 | (this_sector + sectors >= conf->next_resync)) { | |
436 | /* Choose the first operation device, for consistancy */ | |
437 | new_disk = 0; | |
438 | ||
d6065f7b | 439 | for (rdev = rcu_dereference(conf->mirrors[new_disk].rdev); |
cf30a473 | 440 | r1_bio->bios[new_disk] == IO_BLOCKED || |
b2d444d7 | 441 | !rdev || !test_bit(In_sync, &rdev->flags) |
8ddf9efe | 442 | || test_bit(WriteMostly, &rdev->flags); |
d6065f7b | 443 | rdev = rcu_dereference(conf->mirrors[++new_disk].rdev)) { |
8ddf9efe | 444 | |
cf30a473 N |
445 | if (rdev && test_bit(In_sync, &rdev->flags) && |
446 | r1_bio->bios[new_disk] != IO_BLOCKED) | |
8ddf9efe N |
447 | wonly_disk = new_disk; |
448 | ||
449 | if (new_disk == conf->raid_disks - 1) { | |
450 | new_disk = wonly_disk; | |
1da177e4 LT |
451 | break; |
452 | } | |
453 | } | |
454 | goto rb_out; | |
455 | } | |
456 | ||
457 | ||
458 | /* make sure the disk is operational */ | |
d6065f7b | 459 | for (rdev = rcu_dereference(conf->mirrors[new_disk].rdev); |
cf30a473 | 460 | r1_bio->bios[new_disk] == IO_BLOCKED || |
b2d444d7 | 461 | !rdev || !test_bit(In_sync, &rdev->flags) || |
8ddf9efe | 462 | test_bit(WriteMostly, &rdev->flags); |
d6065f7b | 463 | rdev = rcu_dereference(conf->mirrors[new_disk].rdev)) { |
8ddf9efe | 464 | |
cf30a473 N |
465 | if (rdev && test_bit(In_sync, &rdev->flags) && |
466 | r1_bio->bios[new_disk] != IO_BLOCKED) | |
8ddf9efe N |
467 | wonly_disk = new_disk; |
468 | ||
1da177e4 LT |
469 | if (new_disk <= 0) |
470 | new_disk = conf->raid_disks; | |
471 | new_disk--; | |
472 | if (new_disk == disk) { | |
8ddf9efe N |
473 | new_disk = wonly_disk; |
474 | break; | |
1da177e4 LT |
475 | } |
476 | } | |
8ddf9efe N |
477 | |
478 | if (new_disk < 0) | |
479 | goto rb_out; | |
480 | ||
1da177e4 LT |
481 | disk = new_disk; |
482 | /* now disk == new_disk == starting point for search */ | |
483 | ||
484 | /* | |
485 | * Don't change to another disk for sequential reads: | |
486 | */ | |
487 | if (conf->next_seq_sect == this_sector) | |
488 | goto rb_out; | |
489 | if (this_sector == conf->mirrors[new_disk].head_position) | |
490 | goto rb_out; | |
491 | ||
492 | current_distance = abs(this_sector - conf->mirrors[disk].head_position); | |
493 | ||
494 | /* Find the disk whose head is closest */ | |
495 | ||
496 | do { | |
497 | if (disk <= 0) | |
498 | disk = conf->raid_disks; | |
499 | disk--; | |
500 | ||
d6065f7b | 501 | rdev = rcu_dereference(conf->mirrors[disk].rdev); |
8ddf9efe | 502 | |
cf30a473 | 503 | if (!rdev || r1_bio->bios[disk] == IO_BLOCKED || |
b2d444d7 | 504 | !test_bit(In_sync, &rdev->flags) || |
8ddf9efe | 505 | test_bit(WriteMostly, &rdev->flags)) |
1da177e4 LT |
506 | continue; |
507 | ||
508 | if (!atomic_read(&rdev->nr_pending)) { | |
509 | new_disk = disk; | |
1da177e4 LT |
510 | break; |
511 | } | |
512 | new_distance = abs(this_sector - conf->mirrors[disk].head_position); | |
513 | if (new_distance < current_distance) { | |
514 | current_distance = new_distance; | |
515 | new_disk = disk; | |
1da177e4 LT |
516 | } |
517 | } while (disk != conf->last_used); | |
518 | ||
8ddf9efe | 519 | rb_out: |
1da177e4 LT |
520 | |
521 | ||
522 | if (new_disk >= 0) { | |
d6065f7b | 523 | rdev = rcu_dereference(conf->mirrors[new_disk].rdev); |
8ddf9efe N |
524 | if (!rdev) |
525 | goto retry; | |
526 | atomic_inc(&rdev->nr_pending); | |
b2d444d7 | 527 | if (!test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
528 | /* cannot risk returning a device that failed |
529 | * before we inc'ed nr_pending | |
530 | */ | |
03c902e1 | 531 | rdev_dec_pending(rdev, conf->mddev); |
1da177e4 LT |
532 | goto retry; |
533 | } | |
8ddf9efe N |
534 | conf->next_seq_sect = this_sector + sectors; |
535 | conf->last_used = new_disk; | |
1da177e4 LT |
536 | } |
537 | rcu_read_unlock(); | |
538 | ||
539 | return new_disk; | |
540 | } | |
541 | ||
542 | static void unplug_slaves(mddev_t *mddev) | |
543 | { | |
070ec55d | 544 | conf_t *conf = mddev->private; |
1da177e4 LT |
545 | int i; |
546 | ||
547 | rcu_read_lock(); | |
548 | for (i=0; i<mddev->raid_disks; i++) { | |
d6065f7b | 549 | mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev); |
b2d444d7 | 550 | if (rdev && !test_bit(Faulty, &rdev->flags) && atomic_read(&rdev->nr_pending)) { |
165125e1 | 551 | struct request_queue *r_queue = bdev_get_queue(rdev->bdev); |
1da177e4 LT |
552 | |
553 | atomic_inc(&rdev->nr_pending); | |
554 | rcu_read_unlock(); | |
555 | ||
2ad8b1ef | 556 | blk_unplug(r_queue); |
1da177e4 LT |
557 | |
558 | rdev_dec_pending(rdev, mddev); | |
559 | rcu_read_lock(); | |
560 | } | |
561 | } | |
562 | rcu_read_unlock(); | |
563 | } | |
564 | ||
165125e1 | 565 | static void raid1_unplug(struct request_queue *q) |
1da177e4 | 566 | { |
191ea9b2 N |
567 | mddev_t *mddev = q->queuedata; |
568 | ||
569 | unplug_slaves(mddev); | |
570 | md_wakeup_thread(mddev->thread); | |
1da177e4 LT |
571 | } |
572 | ||
0d129228 N |
573 | static int raid1_congested(void *data, int bits) |
574 | { | |
575 | mddev_t *mddev = data; | |
070ec55d | 576 | conf_t *conf = mddev->private; |
0d129228 N |
577 | int i, ret = 0; |
578 | ||
3fa841d7 N |
579 | if (mddev_congested(mddev, bits)) |
580 | return 1; | |
581 | ||
0d129228 N |
582 | rcu_read_lock(); |
583 | for (i = 0; i < mddev->raid_disks; i++) { | |
584 | mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev); | |
585 | if (rdev && !test_bit(Faulty, &rdev->flags)) { | |
165125e1 | 586 | struct request_queue *q = bdev_get_queue(rdev->bdev); |
0d129228 N |
587 | |
588 | /* Note the '|| 1' - when read_balance prefers | |
589 | * non-congested targets, it can be removed | |
590 | */ | |
91a9e99d | 591 | if ((bits & (1<<BDI_async_congested)) || 1) |
0d129228 N |
592 | ret |= bdi_congested(&q->backing_dev_info, bits); |
593 | else | |
594 | ret &= bdi_congested(&q->backing_dev_info, bits); | |
595 | } | |
596 | } | |
597 | rcu_read_unlock(); | |
598 | return ret; | |
599 | } | |
600 | ||
601 | ||
a35e63ef N |
602 | static int flush_pending_writes(conf_t *conf) |
603 | { | |
604 | /* Any writes that have been queued but are awaiting | |
605 | * bitmap updates get flushed here. | |
606 | * We return 1 if any requests were actually submitted. | |
607 | */ | |
608 | int rv = 0; | |
609 | ||
610 | spin_lock_irq(&conf->device_lock); | |
611 | ||
612 | if (conf->pending_bio_list.head) { | |
613 | struct bio *bio; | |
614 | bio = bio_list_get(&conf->pending_bio_list); | |
615 | blk_remove_plug(conf->mddev->queue); | |
616 | spin_unlock_irq(&conf->device_lock); | |
617 | /* flush any pending bitmap writes to | |
618 | * disk before proceeding w/ I/O */ | |
619 | bitmap_unplug(conf->mddev->bitmap); | |
620 | ||
621 | while (bio) { /* submit pending writes */ | |
622 | struct bio *next = bio->bi_next; | |
623 | bio->bi_next = NULL; | |
624 | generic_make_request(bio); | |
625 | bio = next; | |
626 | } | |
627 | rv = 1; | |
628 | } else | |
629 | spin_unlock_irq(&conf->device_lock); | |
630 | return rv; | |
631 | } | |
632 | ||
17999be4 N |
633 | /* Barriers.... |
634 | * Sometimes we need to suspend IO while we do something else, | |
635 | * either some resync/recovery, or reconfigure the array. | |
636 | * To do this we raise a 'barrier'. | |
637 | * The 'barrier' is a counter that can be raised multiple times | |
638 | * to count how many activities are happening which preclude | |
639 | * normal IO. | |
640 | * We can only raise the barrier if there is no pending IO. | |
641 | * i.e. if nr_pending == 0. | |
642 | * We choose only to raise the barrier if no-one is waiting for the | |
643 | * barrier to go down. This means that as soon as an IO request | |
644 | * is ready, no other operations which require a barrier will start | |
645 | * until the IO request has had a chance. | |
646 | * | |
647 | * So: regular IO calls 'wait_barrier'. When that returns there | |
648 | * is no backgroup IO happening, It must arrange to call | |
649 | * allow_barrier when it has finished its IO. | |
650 | * backgroup IO calls must call raise_barrier. Once that returns | |
651 | * there is no normal IO happeing. It must arrange to call | |
652 | * lower_barrier when the particular background IO completes. | |
1da177e4 LT |
653 | */ |
654 | #define RESYNC_DEPTH 32 | |
655 | ||
17999be4 | 656 | static void raise_barrier(conf_t *conf) |
1da177e4 LT |
657 | { |
658 | spin_lock_irq(&conf->resync_lock); | |
17999be4 N |
659 | |
660 | /* Wait until no block IO is waiting */ | |
661 | wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting, | |
662 | conf->resync_lock, | |
663 | raid1_unplug(conf->mddev->queue)); | |
664 | ||
665 | /* block any new IO from starting */ | |
666 | conf->barrier++; | |
667 | ||
668 | /* No wait for all pending IO to complete */ | |
669 | wait_event_lock_irq(conf->wait_barrier, | |
670 | !conf->nr_pending && conf->barrier < RESYNC_DEPTH, | |
671 | conf->resync_lock, | |
672 | raid1_unplug(conf->mddev->queue)); | |
673 | ||
674 | spin_unlock_irq(&conf->resync_lock); | |
675 | } | |
676 | ||
677 | static void lower_barrier(conf_t *conf) | |
678 | { | |
679 | unsigned long flags; | |
709ae487 | 680 | BUG_ON(conf->barrier <= 0); |
17999be4 N |
681 | spin_lock_irqsave(&conf->resync_lock, flags); |
682 | conf->barrier--; | |
683 | spin_unlock_irqrestore(&conf->resync_lock, flags); | |
684 | wake_up(&conf->wait_barrier); | |
685 | } | |
686 | ||
687 | static void wait_barrier(conf_t *conf) | |
688 | { | |
689 | spin_lock_irq(&conf->resync_lock); | |
690 | if (conf->barrier) { | |
691 | conf->nr_waiting++; | |
692 | wait_event_lock_irq(conf->wait_barrier, !conf->barrier, | |
693 | conf->resync_lock, | |
694 | raid1_unplug(conf->mddev->queue)); | |
695 | conf->nr_waiting--; | |
1da177e4 | 696 | } |
17999be4 | 697 | conf->nr_pending++; |
1da177e4 LT |
698 | spin_unlock_irq(&conf->resync_lock); |
699 | } | |
700 | ||
17999be4 N |
701 | static void allow_barrier(conf_t *conf) |
702 | { | |
703 | unsigned long flags; | |
704 | spin_lock_irqsave(&conf->resync_lock, flags); | |
705 | conf->nr_pending--; | |
706 | spin_unlock_irqrestore(&conf->resync_lock, flags); | |
707 | wake_up(&conf->wait_barrier); | |
708 | } | |
709 | ||
ddaf22ab N |
710 | static void freeze_array(conf_t *conf) |
711 | { | |
712 | /* stop syncio and normal IO and wait for everything to | |
713 | * go quite. | |
714 | * We increment barrier and nr_waiting, and then | |
1c830532 N |
715 | * wait until nr_pending match nr_queued+1 |
716 | * This is called in the context of one normal IO request | |
717 | * that has failed. Thus any sync request that might be pending | |
718 | * will be blocked by nr_pending, and we need to wait for | |
719 | * pending IO requests to complete or be queued for re-try. | |
720 | * Thus the number queued (nr_queued) plus this request (1) | |
721 | * must match the number of pending IOs (nr_pending) before | |
722 | * we continue. | |
ddaf22ab N |
723 | */ |
724 | spin_lock_irq(&conf->resync_lock); | |
725 | conf->barrier++; | |
726 | conf->nr_waiting++; | |
727 | wait_event_lock_irq(conf->wait_barrier, | |
1c830532 | 728 | conf->nr_pending == conf->nr_queued+1, |
ddaf22ab | 729 | conf->resync_lock, |
a35e63ef N |
730 | ({ flush_pending_writes(conf); |
731 | raid1_unplug(conf->mddev->queue); })); | |
ddaf22ab N |
732 | spin_unlock_irq(&conf->resync_lock); |
733 | } | |
734 | static void unfreeze_array(conf_t *conf) | |
735 | { | |
736 | /* reverse the effect of the freeze */ | |
737 | spin_lock_irq(&conf->resync_lock); | |
738 | conf->barrier--; | |
739 | conf->nr_waiting--; | |
740 | wake_up(&conf->wait_barrier); | |
741 | spin_unlock_irq(&conf->resync_lock); | |
742 | } | |
743 | ||
17999be4 | 744 | |
4b6d287f N |
745 | /* duplicate the data pages for behind I/O */ |
746 | static struct page **alloc_behind_pages(struct bio *bio) | |
747 | { | |
748 | int i; | |
749 | struct bio_vec *bvec; | |
9ffae0cf | 750 | struct page **pages = kzalloc(bio->bi_vcnt * sizeof(struct page *), |
4b6d287f N |
751 | GFP_NOIO); |
752 | if (unlikely(!pages)) | |
753 | goto do_sync_io; | |
754 | ||
4b6d287f N |
755 | bio_for_each_segment(bvec, bio, i) { |
756 | pages[i] = alloc_page(GFP_NOIO); | |
757 | if (unlikely(!pages[i])) | |
758 | goto do_sync_io; | |
759 | memcpy(kmap(pages[i]) + bvec->bv_offset, | |
760 | kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len); | |
761 | kunmap(pages[i]); | |
762 | kunmap(bvec->bv_page); | |
763 | } | |
764 | ||
765 | return pages; | |
766 | ||
767 | do_sync_io: | |
768 | if (pages) | |
769 | for (i = 0; i < bio->bi_vcnt && pages[i]; i++) | |
2d1f3b5d | 770 | put_page(pages[i]); |
4b6d287f N |
771 | kfree(pages); |
772 | PRINTK("%dB behind alloc failed, doing sync I/O\n", bio->bi_size); | |
773 | return NULL; | |
774 | } | |
775 | ||
165125e1 | 776 | static int make_request(struct request_queue *q, struct bio * bio) |
1da177e4 LT |
777 | { |
778 | mddev_t *mddev = q->queuedata; | |
070ec55d | 779 | conf_t *conf = mddev->private; |
1da177e4 LT |
780 | mirror_info_t *mirror; |
781 | r1bio_t *r1_bio; | |
782 | struct bio *read_bio; | |
191ea9b2 | 783 | int i, targets = 0, disks; |
84255d10 | 784 | struct bitmap *bitmap; |
191ea9b2 N |
785 | unsigned long flags; |
786 | struct bio_list bl; | |
4b6d287f | 787 | struct page **behind_pages = NULL; |
a362357b | 788 | const int rw = bio_data_dir(bio); |
1f98a13f JA |
789 | const bool do_sync = bio_rw_flagged(bio, BIO_RW_SYNCIO); |
790 | int cpu; | |
791 | bool do_barriers; | |
6bfe0b49 | 792 | mdk_rdev_t *blocked_rdev; |
191ea9b2 | 793 | |
1da177e4 LT |
794 | /* |
795 | * Register the new request and wait if the reconstruction | |
796 | * thread has put up a bar for new requests. | |
797 | * Continue immediately if no resync is active currently. | |
62de608d N |
798 | * We test barriers_work *after* md_write_start as md_write_start |
799 | * may cause the first superblock write, and that will check out | |
800 | * if barriers work. | |
1da177e4 | 801 | */ |
62de608d | 802 | |
3d310eb7 N |
803 | md_write_start(mddev, bio); /* wait on superblock update early */ |
804 | ||
6eef4b21 N |
805 | if (bio_data_dir(bio) == WRITE && |
806 | bio->bi_sector + bio->bi_size/512 > mddev->suspend_lo && | |
807 | bio->bi_sector < mddev->suspend_hi) { | |
808 | /* As the suspend_* range is controlled by | |
809 | * userspace, we want an interruptible | |
810 | * wait. | |
811 | */ | |
812 | DEFINE_WAIT(w); | |
813 | for (;;) { | |
814 | flush_signals(current); | |
815 | prepare_to_wait(&conf->wait_barrier, | |
816 | &w, TASK_INTERRUPTIBLE); | |
817 | if (bio->bi_sector + bio->bi_size/512 <= mddev->suspend_lo || | |
818 | bio->bi_sector >= mddev->suspend_hi) | |
819 | break; | |
820 | schedule(); | |
821 | } | |
822 | finish_wait(&conf->wait_barrier, &w); | |
823 | } | |
1f98a13f JA |
824 | if (unlikely(!mddev->barriers_work && |
825 | bio_rw_flagged(bio, BIO_RW_BARRIER))) { | |
62de608d N |
826 | if (rw == WRITE) |
827 | md_write_end(mddev); | |
6712ecf8 | 828 | bio_endio(bio, -EOPNOTSUPP); |
62de608d N |
829 | return 0; |
830 | } | |
831 | ||
17999be4 | 832 | wait_barrier(conf); |
1da177e4 | 833 | |
84255d10 N |
834 | bitmap = mddev->bitmap; |
835 | ||
074a7aca TH |
836 | cpu = part_stat_lock(); |
837 | part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]); | |
838 | part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], | |
839 | bio_sectors(bio)); | |
840 | part_stat_unlock(); | |
1da177e4 LT |
841 | |
842 | /* | |
843 | * make_request() can abort the operation when READA is being | |
844 | * used and no empty request is available. | |
845 | * | |
846 | */ | |
847 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
848 | ||
849 | r1_bio->master_bio = bio; | |
850 | r1_bio->sectors = bio->bi_size >> 9; | |
191ea9b2 | 851 | r1_bio->state = 0; |
1da177e4 LT |
852 | r1_bio->mddev = mddev; |
853 | r1_bio->sector = bio->bi_sector; | |
854 | ||
a362357b | 855 | if (rw == READ) { |
1da177e4 LT |
856 | /* |
857 | * read balancing logic: | |
858 | */ | |
859 | int rdisk = read_balance(conf, r1_bio); | |
860 | ||
861 | if (rdisk < 0) { | |
862 | /* couldn't find anywhere to read from */ | |
863 | raid_end_bio_io(r1_bio); | |
864 | return 0; | |
865 | } | |
866 | mirror = conf->mirrors + rdisk; | |
867 | ||
868 | r1_bio->read_disk = rdisk; | |
869 | ||
870 | read_bio = bio_clone(bio, GFP_NOIO); | |
871 | ||
872 | r1_bio->bios[rdisk] = read_bio; | |
873 | ||
874 | read_bio->bi_sector = r1_bio->sector + mirror->rdev->data_offset; | |
875 | read_bio->bi_bdev = mirror->rdev->bdev; | |
876 | read_bio->bi_end_io = raid1_end_read_request; | |
1ef04fef | 877 | read_bio->bi_rw = READ | (do_sync << BIO_RW_SYNCIO); |
1da177e4 LT |
878 | read_bio->bi_private = r1_bio; |
879 | ||
880 | generic_make_request(read_bio); | |
881 | return 0; | |
882 | } | |
883 | ||
884 | /* | |
885 | * WRITE: | |
886 | */ | |
887 | /* first select target devices under spinlock and | |
888 | * inc refcount on their rdev. Record them by setting | |
889 | * bios[x] to bio | |
890 | */ | |
891 | disks = conf->raid_disks; | |
191ea9b2 N |
892 | #if 0 |
893 | { static int first=1; | |
894 | if (first) printk("First Write sector %llu disks %d\n", | |
895 | (unsigned long long)r1_bio->sector, disks); | |
896 | first = 0; | |
897 | } | |
898 | #endif | |
6bfe0b49 DW |
899 | retry_write: |
900 | blocked_rdev = NULL; | |
1da177e4 LT |
901 | rcu_read_lock(); |
902 | for (i = 0; i < disks; i++) { | |
6bfe0b49 DW |
903 | mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev); |
904 | if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { | |
905 | atomic_inc(&rdev->nr_pending); | |
906 | blocked_rdev = rdev; | |
907 | break; | |
908 | } | |
909 | if (rdev && !test_bit(Faulty, &rdev->flags)) { | |
1da177e4 | 910 | atomic_inc(&rdev->nr_pending); |
b2d444d7 | 911 | if (test_bit(Faulty, &rdev->flags)) { |
03c902e1 | 912 | rdev_dec_pending(rdev, mddev); |
1da177e4 LT |
913 | r1_bio->bios[i] = NULL; |
914 | } else | |
915 | r1_bio->bios[i] = bio; | |
191ea9b2 | 916 | targets++; |
1da177e4 LT |
917 | } else |
918 | r1_bio->bios[i] = NULL; | |
919 | } | |
920 | rcu_read_unlock(); | |
921 | ||
6bfe0b49 DW |
922 | if (unlikely(blocked_rdev)) { |
923 | /* Wait for this device to become unblocked */ | |
924 | int j; | |
925 | ||
926 | for (j = 0; j < i; j++) | |
927 | if (r1_bio->bios[j]) | |
928 | rdev_dec_pending(conf->mirrors[j].rdev, mddev); | |
929 | ||
930 | allow_barrier(conf); | |
931 | md_wait_for_blocked_rdev(blocked_rdev, mddev); | |
932 | wait_barrier(conf); | |
933 | goto retry_write; | |
934 | } | |
935 | ||
4b6d287f N |
936 | BUG_ON(targets == 0); /* we never fail the last device */ |
937 | ||
191ea9b2 N |
938 | if (targets < conf->raid_disks) { |
939 | /* array is degraded, we will not clear the bitmap | |
940 | * on I/O completion (see raid1_end_write_request) */ | |
941 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
942 | } | |
943 | ||
4b6d287f N |
944 | /* do behind I/O ? */ |
945 | if (bitmap && | |
42a04b50 N |
946 | (atomic_read(&bitmap->behind_writes) |
947 | < mddev->bitmap_info.max_write_behind) && | |
4b6d287f N |
948 | (behind_pages = alloc_behind_pages(bio)) != NULL) |
949 | set_bit(R1BIO_BehindIO, &r1_bio->state); | |
950 | ||
191ea9b2 | 951 | atomic_set(&r1_bio->remaining, 0); |
4b6d287f | 952 | atomic_set(&r1_bio->behind_remaining, 0); |
06d91a5f | 953 | |
1f98a13f | 954 | do_barriers = bio_rw_flagged(bio, BIO_RW_BARRIER); |
a9701a30 N |
955 | if (do_barriers) |
956 | set_bit(R1BIO_Barrier, &r1_bio->state); | |
957 | ||
191ea9b2 | 958 | bio_list_init(&bl); |
1da177e4 LT |
959 | for (i = 0; i < disks; i++) { |
960 | struct bio *mbio; | |
961 | if (!r1_bio->bios[i]) | |
962 | continue; | |
963 | ||
964 | mbio = bio_clone(bio, GFP_NOIO); | |
965 | r1_bio->bios[i] = mbio; | |
966 | ||
967 | mbio->bi_sector = r1_bio->sector + conf->mirrors[i].rdev->data_offset; | |
968 | mbio->bi_bdev = conf->mirrors[i].rdev->bdev; | |
969 | mbio->bi_end_io = raid1_end_write_request; | |
1ef04fef DM |
970 | mbio->bi_rw = WRITE | (do_barriers << BIO_RW_BARRIER) | |
971 | (do_sync << BIO_RW_SYNCIO); | |
1da177e4 LT |
972 | mbio->bi_private = r1_bio; |
973 | ||
4b6d287f N |
974 | if (behind_pages) { |
975 | struct bio_vec *bvec; | |
976 | int j; | |
977 | ||
978 | /* Yes, I really want the '__' version so that | |
979 | * we clear any unused pointer in the io_vec, rather | |
980 | * than leave them unchanged. This is important | |
981 | * because when we come to free the pages, we won't | |
982 | * know the originial bi_idx, so we just free | |
983 | * them all | |
984 | */ | |
985 | __bio_for_each_segment(bvec, mbio, j, 0) | |
986 | bvec->bv_page = behind_pages[j]; | |
987 | if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags)) | |
988 | atomic_inc(&r1_bio->behind_remaining); | |
989 | } | |
990 | ||
1da177e4 | 991 | atomic_inc(&r1_bio->remaining); |
1da177e4 | 992 | |
191ea9b2 | 993 | bio_list_add(&bl, mbio); |
1da177e4 | 994 | } |
4b6d287f | 995 | kfree(behind_pages); /* the behind pages are attached to the bios now */ |
1da177e4 | 996 | |
4b6d287f N |
997 | bitmap_startwrite(bitmap, bio->bi_sector, r1_bio->sectors, |
998 | test_bit(R1BIO_BehindIO, &r1_bio->state)); | |
191ea9b2 N |
999 | spin_lock_irqsave(&conf->device_lock, flags); |
1000 | bio_list_merge(&conf->pending_bio_list, &bl); | |
1001 | bio_list_init(&bl); | |
1002 | ||
1003 | blk_plug_device(mddev->queue); | |
1004 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1005 | ||
a35e63ef N |
1006 | /* In case raid1d snuck into freeze_array */ |
1007 | wake_up(&conf->wait_barrier); | |
1008 | ||
e3881a68 LE |
1009 | if (do_sync) |
1010 | md_wakeup_thread(mddev->thread); | |
191ea9b2 N |
1011 | #if 0 |
1012 | while ((bio = bio_list_pop(&bl)) != NULL) | |
1013 | generic_make_request(bio); | |
1014 | #endif | |
1015 | ||
1da177e4 LT |
1016 | return 0; |
1017 | } | |
1018 | ||
1019 | static void status(struct seq_file *seq, mddev_t *mddev) | |
1020 | { | |
070ec55d | 1021 | conf_t *conf = mddev->private; |
1da177e4 LT |
1022 | int i; |
1023 | ||
1024 | seq_printf(seq, " [%d/%d] [", conf->raid_disks, | |
11ce99e6 | 1025 | conf->raid_disks - mddev->degraded); |
ddac7c7e N |
1026 | rcu_read_lock(); |
1027 | for (i = 0; i < conf->raid_disks; i++) { | |
1028 | mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev); | |
1da177e4 | 1029 | seq_printf(seq, "%s", |
ddac7c7e N |
1030 | rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); |
1031 | } | |
1032 | rcu_read_unlock(); | |
1da177e4 LT |
1033 | seq_printf(seq, "]"); |
1034 | } | |
1035 | ||
1036 | ||
1037 | static void error(mddev_t *mddev, mdk_rdev_t *rdev) | |
1038 | { | |
1039 | char b[BDEVNAME_SIZE]; | |
070ec55d | 1040 | conf_t *conf = mddev->private; |
1da177e4 LT |
1041 | |
1042 | /* | |
1043 | * If it is not operational, then we have already marked it as dead | |
1044 | * else if it is the last working disks, ignore the error, let the | |
1045 | * next level up know. | |
1046 | * else mark the drive as failed | |
1047 | */ | |
b2d444d7 | 1048 | if (test_bit(In_sync, &rdev->flags) |
4044ba58 | 1049 | && (conf->raid_disks - mddev->degraded) == 1) { |
1da177e4 LT |
1050 | /* |
1051 | * Don't fail the drive, act as though we were just a | |
4044ba58 N |
1052 | * normal single drive. |
1053 | * However don't try a recovery from this drive as | |
1054 | * it is very likely to fail. | |
1da177e4 | 1055 | */ |
4044ba58 | 1056 | mddev->recovery_disabled = 1; |
1da177e4 | 1057 | return; |
4044ba58 | 1058 | } |
c04be0aa N |
1059 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
1060 | unsigned long flags; | |
1061 | spin_lock_irqsave(&conf->device_lock, flags); | |
1da177e4 | 1062 | mddev->degraded++; |
dd00a99e | 1063 | set_bit(Faulty, &rdev->flags); |
c04be0aa | 1064 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 LT |
1065 | /* |
1066 | * if recovery is running, make sure it aborts. | |
1067 | */ | |
dfc70645 | 1068 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
dd00a99e N |
1069 | } else |
1070 | set_bit(Faulty, &rdev->flags); | |
850b2b42 | 1071 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
d7a420c9 NA |
1072 | printk(KERN_ALERT "raid1: Disk failure on %s, disabling device.\n" |
1073 | "raid1: Operation continuing on %d devices.\n", | |
11ce99e6 | 1074 | bdevname(rdev->bdev,b), conf->raid_disks - mddev->degraded); |
1da177e4 LT |
1075 | } |
1076 | ||
1077 | static void print_conf(conf_t *conf) | |
1078 | { | |
1079 | int i; | |
1da177e4 LT |
1080 | |
1081 | printk("RAID1 conf printout:\n"); | |
1082 | if (!conf) { | |
1083 | printk("(!conf)\n"); | |
1084 | return; | |
1085 | } | |
11ce99e6 | 1086 | printk(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded, |
1da177e4 LT |
1087 | conf->raid_disks); |
1088 | ||
ddac7c7e | 1089 | rcu_read_lock(); |
1da177e4 LT |
1090 | for (i = 0; i < conf->raid_disks; i++) { |
1091 | char b[BDEVNAME_SIZE]; | |
ddac7c7e N |
1092 | mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev); |
1093 | if (rdev) | |
1da177e4 | 1094 | printk(" disk %d, wo:%d, o:%d, dev:%s\n", |
ddac7c7e N |
1095 | i, !test_bit(In_sync, &rdev->flags), |
1096 | !test_bit(Faulty, &rdev->flags), | |
1097 | bdevname(rdev->bdev,b)); | |
1da177e4 | 1098 | } |
ddac7c7e | 1099 | rcu_read_unlock(); |
1da177e4 LT |
1100 | } |
1101 | ||
1102 | static void close_sync(conf_t *conf) | |
1103 | { | |
17999be4 N |
1104 | wait_barrier(conf); |
1105 | allow_barrier(conf); | |
1da177e4 LT |
1106 | |
1107 | mempool_destroy(conf->r1buf_pool); | |
1108 | conf->r1buf_pool = NULL; | |
1109 | } | |
1110 | ||
1111 | static int raid1_spare_active(mddev_t *mddev) | |
1112 | { | |
1113 | int i; | |
1114 | conf_t *conf = mddev->private; | |
1da177e4 LT |
1115 | |
1116 | /* | |
1117 | * Find all failed disks within the RAID1 configuration | |
ddac7c7e N |
1118 | * and mark them readable. |
1119 | * Called under mddev lock, so rcu protection not needed. | |
1da177e4 LT |
1120 | */ |
1121 | for (i = 0; i < conf->raid_disks; i++) { | |
ddac7c7e N |
1122 | mdk_rdev_t *rdev = conf->mirrors[i].rdev; |
1123 | if (rdev | |
1124 | && !test_bit(Faulty, &rdev->flags) | |
c04be0aa N |
1125 | && !test_and_set_bit(In_sync, &rdev->flags)) { |
1126 | unsigned long flags; | |
1127 | spin_lock_irqsave(&conf->device_lock, flags); | |
1da177e4 | 1128 | mddev->degraded--; |
c04be0aa | 1129 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 LT |
1130 | } |
1131 | } | |
1132 | ||
1133 | print_conf(conf); | |
1134 | return 0; | |
1135 | } | |
1136 | ||
1137 | ||
1138 | static int raid1_add_disk(mddev_t *mddev, mdk_rdev_t *rdev) | |
1139 | { | |
1140 | conf_t *conf = mddev->private; | |
199050ea | 1141 | int err = -EEXIST; |
41158c7e | 1142 | int mirror = 0; |
1da177e4 | 1143 | mirror_info_t *p; |
6c2fce2e NB |
1144 | int first = 0; |
1145 | int last = mddev->raid_disks - 1; | |
1da177e4 | 1146 | |
6c2fce2e NB |
1147 | if (rdev->raid_disk >= 0) |
1148 | first = last = rdev->raid_disk; | |
1149 | ||
1150 | for (mirror = first; mirror <= last; mirror++) | |
1da177e4 LT |
1151 | if ( !(p=conf->mirrors+mirror)->rdev) { |
1152 | ||
8f6c2e4b MP |
1153 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
1154 | rdev->data_offset << 9); | |
1da177e4 LT |
1155 | /* as we don't honour merge_bvec_fn, we must never risk |
1156 | * violating it, so limit ->max_sector to one PAGE, as | |
1157 | * a one page request is never in violation. | |
1158 | */ | |
1159 | if (rdev->bdev->bd_disk->queue->merge_bvec_fn && | |
ae03bf63 | 1160 | queue_max_sectors(mddev->queue) > (PAGE_SIZE>>9)) |
086fa5ff | 1161 | blk_queue_max_hw_sectors(mddev->queue, PAGE_SIZE>>9); |
1da177e4 LT |
1162 | |
1163 | p->head_position = 0; | |
1164 | rdev->raid_disk = mirror; | |
199050ea | 1165 | err = 0; |
6aea114a N |
1166 | /* As all devices are equivalent, we don't need a full recovery |
1167 | * if this was recently any drive of the array | |
1168 | */ | |
1169 | if (rdev->saved_raid_disk < 0) | |
41158c7e | 1170 | conf->fullsync = 1; |
d6065f7b | 1171 | rcu_assign_pointer(p->rdev, rdev); |
1da177e4 LT |
1172 | break; |
1173 | } | |
ac5e7113 | 1174 | md_integrity_add_rdev(rdev, mddev); |
1da177e4 | 1175 | print_conf(conf); |
199050ea | 1176 | return err; |
1da177e4 LT |
1177 | } |
1178 | ||
1179 | static int raid1_remove_disk(mddev_t *mddev, int number) | |
1180 | { | |
1181 | conf_t *conf = mddev->private; | |
1182 | int err = 0; | |
1183 | mdk_rdev_t *rdev; | |
1184 | mirror_info_t *p = conf->mirrors+ number; | |
1185 | ||
1186 | print_conf(conf); | |
1187 | rdev = p->rdev; | |
1188 | if (rdev) { | |
b2d444d7 | 1189 | if (test_bit(In_sync, &rdev->flags) || |
1da177e4 LT |
1190 | atomic_read(&rdev->nr_pending)) { |
1191 | err = -EBUSY; | |
1192 | goto abort; | |
1193 | } | |
dfc70645 N |
1194 | /* Only remove non-faulty devices is recovery |
1195 | * is not possible. | |
1196 | */ | |
1197 | if (!test_bit(Faulty, &rdev->flags) && | |
1198 | mddev->degraded < conf->raid_disks) { | |
1199 | err = -EBUSY; | |
1200 | goto abort; | |
1201 | } | |
1da177e4 | 1202 | p->rdev = NULL; |
fbd568a3 | 1203 | synchronize_rcu(); |
1da177e4 LT |
1204 | if (atomic_read(&rdev->nr_pending)) { |
1205 | /* lost the race, try later */ | |
1206 | err = -EBUSY; | |
1207 | p->rdev = rdev; | |
ac5e7113 | 1208 | goto abort; |
1da177e4 | 1209 | } |
ac5e7113 | 1210 | md_integrity_register(mddev); |
1da177e4 LT |
1211 | } |
1212 | abort: | |
1213 | ||
1214 | print_conf(conf); | |
1215 | return err; | |
1216 | } | |
1217 | ||
1218 | ||
6712ecf8 | 1219 | static void end_sync_read(struct bio *bio, int error) |
1da177e4 | 1220 | { |
1da177e4 | 1221 | r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private); |
d11c171e | 1222 | int i; |
1da177e4 | 1223 | |
d11c171e N |
1224 | for (i=r1_bio->mddev->raid_disks; i--; ) |
1225 | if (r1_bio->bios[i] == bio) | |
1226 | break; | |
1227 | BUG_ON(i < 0); | |
1228 | update_head_pos(i, r1_bio); | |
1da177e4 LT |
1229 | /* |
1230 | * we have read a block, now it needs to be re-written, | |
1231 | * or re-read if the read failed. | |
1232 | * We don't do much here, just schedule handling by raid1d | |
1233 | */ | |
69382e85 | 1234 | if (test_bit(BIO_UPTODATE, &bio->bi_flags)) |
1da177e4 | 1235 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
d11c171e N |
1236 | |
1237 | if (atomic_dec_and_test(&r1_bio->remaining)) | |
1238 | reschedule_retry(r1_bio); | |
1da177e4 LT |
1239 | } |
1240 | ||
6712ecf8 | 1241 | static void end_sync_write(struct bio *bio, int error) |
1da177e4 LT |
1242 | { |
1243 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
1244 | r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private); | |
1245 | mddev_t *mddev = r1_bio->mddev; | |
070ec55d | 1246 | conf_t *conf = mddev->private; |
1da177e4 LT |
1247 | int i; |
1248 | int mirror=0; | |
1249 | ||
1da177e4 LT |
1250 | for (i = 0; i < conf->raid_disks; i++) |
1251 | if (r1_bio->bios[i] == bio) { | |
1252 | mirror = i; | |
1253 | break; | |
1254 | } | |
6b1117d5 N |
1255 | if (!uptodate) { |
1256 | int sync_blocks = 0; | |
1257 | sector_t s = r1_bio->sector; | |
1258 | long sectors_to_go = r1_bio->sectors; | |
1259 | /* make sure these bits doesn't get cleared. */ | |
1260 | do { | |
5e3db645 | 1261 | bitmap_end_sync(mddev->bitmap, s, |
6b1117d5 N |
1262 | &sync_blocks, 1); |
1263 | s += sync_blocks; | |
1264 | sectors_to_go -= sync_blocks; | |
1265 | } while (sectors_to_go > 0); | |
1da177e4 | 1266 | md_error(mddev, conf->mirrors[mirror].rdev); |
6b1117d5 | 1267 | } |
e3b9703e | 1268 | |
1da177e4 LT |
1269 | update_head_pos(mirror, r1_bio); |
1270 | ||
1271 | if (atomic_dec_and_test(&r1_bio->remaining)) { | |
73d5c38a | 1272 | sector_t s = r1_bio->sectors; |
1da177e4 | 1273 | put_buf(r1_bio); |
73d5c38a | 1274 | md_done_sync(mddev, s, uptodate); |
1da177e4 | 1275 | } |
1da177e4 LT |
1276 | } |
1277 | ||
1278 | static void sync_request_write(mddev_t *mddev, r1bio_t *r1_bio) | |
1279 | { | |
070ec55d | 1280 | conf_t *conf = mddev->private; |
1da177e4 LT |
1281 | int i; |
1282 | int disks = conf->raid_disks; | |
1283 | struct bio *bio, *wbio; | |
1284 | ||
1285 | bio = r1_bio->bios[r1_bio->read_disk]; | |
1286 | ||
69382e85 | 1287 | |
d11c171e N |
1288 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { |
1289 | /* We have read all readable devices. If we haven't | |
1290 | * got the block, then there is no hope left. | |
1291 | * If we have, then we want to do a comparison | |
1292 | * and skip the write if everything is the same. | |
1293 | * If any blocks failed to read, then we need to | |
1294 | * attempt an over-write | |
1295 | */ | |
1296 | int primary; | |
1297 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) { | |
1298 | for (i=0; i<mddev->raid_disks; i++) | |
1299 | if (r1_bio->bios[i]->bi_end_io == end_sync_read) | |
1300 | md_error(mddev, conf->mirrors[i].rdev); | |
1301 | ||
1302 | md_done_sync(mddev, r1_bio->sectors, 1); | |
1303 | put_buf(r1_bio); | |
1304 | return; | |
1305 | } | |
1306 | for (primary=0; primary<mddev->raid_disks; primary++) | |
1307 | if (r1_bio->bios[primary]->bi_end_io == end_sync_read && | |
1308 | test_bit(BIO_UPTODATE, &r1_bio->bios[primary]->bi_flags)) { | |
1309 | r1_bio->bios[primary]->bi_end_io = NULL; | |
03c902e1 | 1310 | rdev_dec_pending(conf->mirrors[primary].rdev, mddev); |
d11c171e N |
1311 | break; |
1312 | } | |
1313 | r1_bio->read_disk = primary; | |
1314 | for (i=0; i<mddev->raid_disks; i++) | |
ed456662 | 1315 | if (r1_bio->bios[i]->bi_end_io == end_sync_read) { |
d11c171e N |
1316 | int j; |
1317 | int vcnt = r1_bio->sectors >> (PAGE_SHIFT- 9); | |
1318 | struct bio *pbio = r1_bio->bios[primary]; | |
1319 | struct bio *sbio = r1_bio->bios[i]; | |
ed456662 MA |
1320 | |
1321 | if (test_bit(BIO_UPTODATE, &sbio->bi_flags)) { | |
1322 | for (j = vcnt; j-- ; ) { | |
1323 | struct page *p, *s; | |
1324 | p = pbio->bi_io_vec[j].bv_page; | |
1325 | s = sbio->bi_io_vec[j].bv_page; | |
1326 | if (memcmp(page_address(p), | |
1327 | page_address(s), | |
1328 | PAGE_SIZE)) | |
1329 | break; | |
1330 | } | |
1331 | } else | |
1332 | j = 0; | |
d11c171e N |
1333 | if (j >= 0) |
1334 | mddev->resync_mismatches += r1_bio->sectors; | |
cf7a4416 N |
1335 | if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery) |
1336 | && test_bit(BIO_UPTODATE, &sbio->bi_flags))) { | |
d11c171e | 1337 | sbio->bi_end_io = NULL; |
03c902e1 N |
1338 | rdev_dec_pending(conf->mirrors[i].rdev, mddev); |
1339 | } else { | |
d11c171e | 1340 | /* fixup the bio for reuse */ |
698b18c1 | 1341 | int size; |
d11c171e N |
1342 | sbio->bi_vcnt = vcnt; |
1343 | sbio->bi_size = r1_bio->sectors << 9; | |
1344 | sbio->bi_idx = 0; | |
1345 | sbio->bi_phys_segments = 0; | |
d11c171e N |
1346 | sbio->bi_flags &= ~(BIO_POOL_MASK - 1); |
1347 | sbio->bi_flags |= 1 << BIO_UPTODATE; | |
1348 | sbio->bi_next = NULL; | |
1349 | sbio->bi_sector = r1_bio->sector + | |
1350 | conf->mirrors[i].rdev->data_offset; | |
1351 | sbio->bi_bdev = conf->mirrors[i].rdev->bdev; | |
698b18c1 N |
1352 | size = sbio->bi_size; |
1353 | for (j = 0; j < vcnt ; j++) { | |
1354 | struct bio_vec *bi; | |
1355 | bi = &sbio->bi_io_vec[j]; | |
1356 | bi->bv_offset = 0; | |
1357 | if (size > PAGE_SIZE) | |
1358 | bi->bv_len = PAGE_SIZE; | |
1359 | else | |
1360 | bi->bv_len = size; | |
1361 | size -= PAGE_SIZE; | |
1362 | memcpy(page_address(bi->bv_page), | |
3eda22d1 N |
1363 | page_address(pbio->bi_io_vec[j].bv_page), |
1364 | PAGE_SIZE); | |
698b18c1 | 1365 | } |
3eda22d1 | 1366 | |
d11c171e N |
1367 | } |
1368 | } | |
1369 | } | |
1da177e4 | 1370 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) { |
69382e85 N |
1371 | /* ouch - failed to read all of that. |
1372 | * Try some synchronous reads of other devices to get | |
1373 | * good data, much like with normal read errors. Only | |
ddac7c7e | 1374 | * read into the pages we already have so we don't |
69382e85 N |
1375 | * need to re-issue the read request. |
1376 | * We don't need to freeze the array, because being in an | |
1377 | * active sync request, there is no normal IO, and | |
1378 | * no overlapping syncs. | |
1da177e4 | 1379 | */ |
69382e85 N |
1380 | sector_t sect = r1_bio->sector; |
1381 | int sectors = r1_bio->sectors; | |
1382 | int idx = 0; | |
1383 | ||
1384 | while(sectors) { | |
1385 | int s = sectors; | |
1386 | int d = r1_bio->read_disk; | |
1387 | int success = 0; | |
1388 | mdk_rdev_t *rdev; | |
1389 | ||
1390 | if (s > (PAGE_SIZE>>9)) | |
1391 | s = PAGE_SIZE >> 9; | |
1392 | do { | |
1393 | if (r1_bio->bios[d]->bi_end_io == end_sync_read) { | |
ddac7c7e N |
1394 | /* No rcu protection needed here devices |
1395 | * can only be removed when no resync is | |
1396 | * active, and resync is currently active | |
1397 | */ | |
69382e85 N |
1398 | rdev = conf->mirrors[d].rdev; |
1399 | if (sync_page_io(rdev->bdev, | |
1400 | sect + rdev->data_offset, | |
1401 | s<<9, | |
1402 | bio->bi_io_vec[idx].bv_page, | |
1403 | READ)) { | |
1404 | success = 1; | |
1405 | break; | |
1406 | } | |
1407 | } | |
1408 | d++; | |
1409 | if (d == conf->raid_disks) | |
1410 | d = 0; | |
1411 | } while (!success && d != r1_bio->read_disk); | |
1412 | ||
1413 | if (success) { | |
097426f6 | 1414 | int start = d; |
69382e85 N |
1415 | /* write it back and re-read */ |
1416 | set_bit(R1BIO_Uptodate, &r1_bio->state); | |
1417 | while (d != r1_bio->read_disk) { | |
1418 | if (d == 0) | |
1419 | d = conf->raid_disks; | |
1420 | d--; | |
1421 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) | |
1422 | continue; | |
1423 | rdev = conf->mirrors[d].rdev; | |
4dbcdc75 | 1424 | atomic_add(s, &rdev->corrected_errors); |
69382e85 N |
1425 | if (sync_page_io(rdev->bdev, |
1426 | sect + rdev->data_offset, | |
1427 | s<<9, | |
1428 | bio->bi_io_vec[idx].bv_page, | |
097426f6 N |
1429 | WRITE) == 0) |
1430 | md_error(mddev, rdev); | |
1431 | } | |
1432 | d = start; | |
1433 | while (d != r1_bio->read_disk) { | |
1434 | if (d == 0) | |
1435 | d = conf->raid_disks; | |
1436 | d--; | |
1437 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) | |
1438 | continue; | |
1439 | rdev = conf->mirrors[d].rdev; | |
1440 | if (sync_page_io(rdev->bdev, | |
69382e85 N |
1441 | sect + rdev->data_offset, |
1442 | s<<9, | |
1443 | bio->bi_io_vec[idx].bv_page, | |
097426f6 | 1444 | READ) == 0) |
69382e85 | 1445 | md_error(mddev, rdev); |
69382e85 N |
1446 | } |
1447 | } else { | |
1448 | char b[BDEVNAME_SIZE]; | |
1449 | /* Cannot read from anywhere, array is toast */ | |
1450 | md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev); | |
1451 | printk(KERN_ALERT "raid1: %s: unrecoverable I/O read error" | |
1452 | " for block %llu\n", | |
1453 | bdevname(bio->bi_bdev,b), | |
1454 | (unsigned long long)r1_bio->sector); | |
1455 | md_done_sync(mddev, r1_bio->sectors, 0); | |
1456 | put_buf(r1_bio); | |
1457 | return; | |
1458 | } | |
1459 | sectors -= s; | |
1460 | sect += s; | |
1461 | idx ++; | |
1462 | } | |
1da177e4 | 1463 | } |
d11c171e N |
1464 | |
1465 | /* | |
1466 | * schedule writes | |
1467 | */ | |
1da177e4 LT |
1468 | atomic_set(&r1_bio->remaining, 1); |
1469 | for (i = 0; i < disks ; i++) { | |
1470 | wbio = r1_bio->bios[i]; | |
3e198f78 N |
1471 | if (wbio->bi_end_io == NULL || |
1472 | (wbio->bi_end_io == end_sync_read && | |
1473 | (i == r1_bio->read_disk || | |
1474 | !test_bit(MD_RECOVERY_SYNC, &mddev->recovery)))) | |
1da177e4 LT |
1475 | continue; |
1476 | ||
3e198f78 N |
1477 | wbio->bi_rw = WRITE; |
1478 | wbio->bi_end_io = end_sync_write; | |
1da177e4 LT |
1479 | atomic_inc(&r1_bio->remaining); |
1480 | md_sync_acct(conf->mirrors[i].rdev->bdev, wbio->bi_size >> 9); | |
191ea9b2 | 1481 | |
1da177e4 LT |
1482 | generic_make_request(wbio); |
1483 | } | |
1484 | ||
1485 | if (atomic_dec_and_test(&r1_bio->remaining)) { | |
191ea9b2 | 1486 | /* if we're here, all write(s) have completed, so clean up */ |
1da177e4 LT |
1487 | md_done_sync(mddev, r1_bio->sectors, 1); |
1488 | put_buf(r1_bio); | |
1489 | } | |
1490 | } | |
1491 | ||
1492 | /* | |
1493 | * This is a kernel thread which: | |
1494 | * | |
1495 | * 1. Retries failed read operations on working mirrors. | |
1496 | * 2. Updates the raid superblock when problems encounter. | |
1497 | * 3. Performs writes following reads for array syncronising. | |
1498 | */ | |
1499 | ||
867868fb N |
1500 | static void fix_read_error(conf_t *conf, int read_disk, |
1501 | sector_t sect, int sectors) | |
1502 | { | |
1503 | mddev_t *mddev = conf->mddev; | |
1504 | while(sectors) { | |
1505 | int s = sectors; | |
1506 | int d = read_disk; | |
1507 | int success = 0; | |
1508 | int start; | |
1509 | mdk_rdev_t *rdev; | |
1510 | ||
1511 | if (s > (PAGE_SIZE>>9)) | |
1512 | s = PAGE_SIZE >> 9; | |
1513 | ||
1514 | do { | |
1515 | /* Note: no rcu protection needed here | |
1516 | * as this is synchronous in the raid1d thread | |
1517 | * which is the thread that might remove | |
1518 | * a device. If raid1d ever becomes multi-threaded.... | |
1519 | */ | |
1520 | rdev = conf->mirrors[d].rdev; | |
1521 | if (rdev && | |
1522 | test_bit(In_sync, &rdev->flags) && | |
1523 | sync_page_io(rdev->bdev, | |
1524 | sect + rdev->data_offset, | |
1525 | s<<9, | |
1526 | conf->tmppage, READ)) | |
1527 | success = 1; | |
1528 | else { | |
1529 | d++; | |
1530 | if (d == conf->raid_disks) | |
1531 | d = 0; | |
1532 | } | |
1533 | } while (!success && d != read_disk); | |
1534 | ||
1535 | if (!success) { | |
1536 | /* Cannot read from anywhere -- bye bye array */ | |
1537 | md_error(mddev, conf->mirrors[read_disk].rdev); | |
1538 | break; | |
1539 | } | |
1540 | /* write it back and re-read */ | |
1541 | start = d; | |
1542 | while (d != read_disk) { | |
1543 | if (d==0) | |
1544 | d = conf->raid_disks; | |
1545 | d--; | |
1546 | rdev = conf->mirrors[d].rdev; | |
1547 | if (rdev && | |
1548 | test_bit(In_sync, &rdev->flags)) { | |
1549 | if (sync_page_io(rdev->bdev, | |
1550 | sect + rdev->data_offset, | |
1551 | s<<9, conf->tmppage, WRITE) | |
1552 | == 0) | |
1553 | /* Well, this device is dead */ | |
1554 | md_error(mddev, rdev); | |
1555 | } | |
1556 | } | |
1557 | d = start; | |
1558 | while (d != read_disk) { | |
1559 | char b[BDEVNAME_SIZE]; | |
1560 | if (d==0) | |
1561 | d = conf->raid_disks; | |
1562 | d--; | |
1563 | rdev = conf->mirrors[d].rdev; | |
1564 | if (rdev && | |
1565 | test_bit(In_sync, &rdev->flags)) { | |
1566 | if (sync_page_io(rdev->bdev, | |
1567 | sect + rdev->data_offset, | |
1568 | s<<9, conf->tmppage, READ) | |
1569 | == 0) | |
1570 | /* Well, this device is dead */ | |
1571 | md_error(mddev, rdev); | |
1572 | else { | |
1573 | atomic_add(s, &rdev->corrected_errors); | |
1574 | printk(KERN_INFO | |
1575 | "raid1:%s: read error corrected " | |
1576 | "(%d sectors at %llu on %s)\n", | |
1577 | mdname(mddev), s, | |
969b755a RD |
1578 | (unsigned long long)(sect + |
1579 | rdev->data_offset), | |
867868fb N |
1580 | bdevname(rdev->bdev, b)); |
1581 | } | |
1582 | } | |
1583 | } | |
1584 | sectors -= s; | |
1585 | sect += s; | |
1586 | } | |
1587 | } | |
1588 | ||
1da177e4 LT |
1589 | static void raid1d(mddev_t *mddev) |
1590 | { | |
1591 | r1bio_t *r1_bio; | |
1592 | struct bio *bio; | |
1593 | unsigned long flags; | |
070ec55d | 1594 | conf_t *conf = mddev->private; |
1da177e4 LT |
1595 | struct list_head *head = &conf->retry_list; |
1596 | int unplug=0; | |
1597 | mdk_rdev_t *rdev; | |
1598 | ||
1599 | md_check_recovery(mddev); | |
1da177e4 LT |
1600 | |
1601 | for (;;) { | |
1602 | char b[BDEVNAME_SIZE]; | |
191ea9b2 | 1603 | |
a35e63ef | 1604 | unplug += flush_pending_writes(conf); |
191ea9b2 | 1605 | |
a35e63ef N |
1606 | spin_lock_irqsave(&conf->device_lock, flags); |
1607 | if (list_empty(head)) { | |
1608 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 | 1609 | break; |
a35e63ef | 1610 | } |
1da177e4 LT |
1611 | r1_bio = list_entry(head->prev, r1bio_t, retry_list); |
1612 | list_del(head->prev); | |
ddaf22ab | 1613 | conf->nr_queued--; |
1da177e4 LT |
1614 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1615 | ||
1616 | mddev = r1_bio->mddev; | |
070ec55d | 1617 | conf = mddev->private; |
1da177e4 LT |
1618 | if (test_bit(R1BIO_IsSync, &r1_bio->state)) { |
1619 | sync_request_write(mddev, r1_bio); | |
1620 | unplug = 1; | |
a9701a30 N |
1621 | } else if (test_bit(R1BIO_BarrierRetry, &r1_bio->state)) { |
1622 | /* some requests in the r1bio were BIO_RW_BARRIER | |
bea27718 | 1623 | * requests which failed with -EOPNOTSUPP. Hohumm.. |
a9701a30 N |
1624 | * Better resubmit without the barrier. |
1625 | * We know which devices to resubmit for, because | |
1626 | * all others have had their bios[] entry cleared. | |
5e7dd2ab | 1627 | * We already have a nr_pending reference on these rdevs. |
a9701a30 N |
1628 | */ |
1629 | int i; | |
1f98a13f | 1630 | const bool do_sync = bio_rw_flagged(r1_bio->master_bio, BIO_RW_SYNCIO); |
a9701a30 N |
1631 | clear_bit(R1BIO_BarrierRetry, &r1_bio->state); |
1632 | clear_bit(R1BIO_Barrier, &r1_bio->state); | |
2f889129 N |
1633 | for (i=0; i < conf->raid_disks; i++) |
1634 | if (r1_bio->bios[i]) | |
1635 | atomic_inc(&r1_bio->remaining); | |
a9701a30 N |
1636 | for (i=0; i < conf->raid_disks; i++) |
1637 | if (r1_bio->bios[i]) { | |
1638 | struct bio_vec *bvec; | |
1639 | int j; | |
1640 | ||
1641 | bio = bio_clone(r1_bio->master_bio, GFP_NOIO); | |
1642 | /* copy pages from the failed bio, as | |
1643 | * this might be a write-behind device */ | |
1644 | __bio_for_each_segment(bvec, bio, j, 0) | |
1645 | bvec->bv_page = bio_iovec_idx(r1_bio->bios[i], j)->bv_page; | |
1646 | bio_put(r1_bio->bios[i]); | |
1647 | bio->bi_sector = r1_bio->sector + | |
1648 | conf->mirrors[i].rdev->data_offset; | |
1649 | bio->bi_bdev = conf->mirrors[i].rdev->bdev; | |
1650 | bio->bi_end_io = raid1_end_write_request; | |
1ef04fef DM |
1651 | bio->bi_rw = WRITE | |
1652 | (do_sync << BIO_RW_SYNCIO); | |
a9701a30 N |
1653 | bio->bi_private = r1_bio; |
1654 | r1_bio->bios[i] = bio; | |
1655 | generic_make_request(bio); | |
1656 | } | |
1da177e4 LT |
1657 | } else { |
1658 | int disk; | |
ddaf22ab N |
1659 | |
1660 | /* we got a read error. Maybe the drive is bad. Maybe just | |
1661 | * the block and we can fix it. | |
1662 | * We freeze all other IO, and try reading the block from | |
1663 | * other devices. When we find one, we re-write | |
1664 | * and check it that fixes the read error. | |
1665 | * This is all done synchronously while the array is | |
1666 | * frozen | |
1667 | */ | |
867868fb N |
1668 | if (mddev->ro == 0) { |
1669 | freeze_array(conf); | |
1670 | fix_read_error(conf, r1_bio->read_disk, | |
1671 | r1_bio->sector, | |
1672 | r1_bio->sectors); | |
1673 | unfreeze_array(conf); | |
d0e26078 N |
1674 | } else |
1675 | md_error(mddev, | |
1676 | conf->mirrors[r1_bio->read_disk].rdev); | |
ddaf22ab | 1677 | |
1da177e4 | 1678 | bio = r1_bio->bios[r1_bio->read_disk]; |
d0e26078 | 1679 | if ((disk=read_balance(conf, r1_bio)) == -1) { |
1da177e4 LT |
1680 | printk(KERN_ALERT "raid1: %s: unrecoverable I/O" |
1681 | " read error for block %llu\n", | |
1682 | bdevname(bio->bi_bdev,b), | |
1683 | (unsigned long long)r1_bio->sector); | |
1684 | raid_end_bio_io(r1_bio); | |
1685 | } else { | |
1f98a13f | 1686 | const bool do_sync = bio_rw_flagged(r1_bio->master_bio, BIO_RW_SYNCIO); |
cf30a473 N |
1687 | r1_bio->bios[r1_bio->read_disk] = |
1688 | mddev->ro ? IO_BLOCKED : NULL; | |
1da177e4 LT |
1689 | r1_bio->read_disk = disk; |
1690 | bio_put(bio); | |
1691 | bio = bio_clone(r1_bio->master_bio, GFP_NOIO); | |
1692 | r1_bio->bios[r1_bio->read_disk] = bio; | |
1693 | rdev = conf->mirrors[disk].rdev; | |
1694 | if (printk_ratelimit()) | |
1695 | printk(KERN_ERR "raid1: %s: redirecting sector %llu to" | |
1696 | " another mirror\n", | |
1697 | bdevname(rdev->bdev,b), | |
1698 | (unsigned long long)r1_bio->sector); | |
1699 | bio->bi_sector = r1_bio->sector + rdev->data_offset; | |
1700 | bio->bi_bdev = rdev->bdev; | |
1701 | bio->bi_end_io = raid1_end_read_request; | |
1ef04fef | 1702 | bio->bi_rw = READ | (do_sync << BIO_RW_SYNCIO); |
1da177e4 LT |
1703 | bio->bi_private = r1_bio; |
1704 | unplug = 1; | |
1705 | generic_make_request(bio); | |
1706 | } | |
1707 | } | |
1d9d5241 | 1708 | cond_resched(); |
1da177e4 | 1709 | } |
1da177e4 LT |
1710 | if (unplug) |
1711 | unplug_slaves(mddev); | |
1712 | } | |
1713 | ||
1714 | ||
1715 | static int init_resync(conf_t *conf) | |
1716 | { | |
1717 | int buffs; | |
1718 | ||
1719 | buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; | |
9e77c485 | 1720 | BUG_ON(conf->r1buf_pool); |
1da177e4 LT |
1721 | conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free, |
1722 | conf->poolinfo); | |
1723 | if (!conf->r1buf_pool) | |
1724 | return -ENOMEM; | |
1725 | conf->next_resync = 0; | |
1726 | return 0; | |
1727 | } | |
1728 | ||
1729 | /* | |
1730 | * perform a "sync" on one "block" | |
1731 | * | |
1732 | * We need to make sure that no normal I/O request - particularly write | |
1733 | * requests - conflict with active sync requests. | |
1734 | * | |
1735 | * This is achieved by tracking pending requests and a 'barrier' concept | |
1736 | * that can be installed to exclude normal IO requests. | |
1737 | */ | |
1738 | ||
57afd89f | 1739 | static sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster) |
1da177e4 | 1740 | { |
070ec55d | 1741 | conf_t *conf = mddev->private; |
1da177e4 LT |
1742 | r1bio_t *r1_bio; |
1743 | struct bio *bio; | |
1744 | sector_t max_sector, nr_sectors; | |
3e198f78 | 1745 | int disk = -1; |
1da177e4 | 1746 | int i; |
3e198f78 N |
1747 | int wonly = -1; |
1748 | int write_targets = 0, read_targets = 0; | |
191ea9b2 | 1749 | int sync_blocks; |
e3b9703e | 1750 | int still_degraded = 0; |
1da177e4 LT |
1751 | |
1752 | if (!conf->r1buf_pool) | |
191ea9b2 N |
1753 | { |
1754 | /* | |
1755 | printk("sync start - bitmap %p\n", mddev->bitmap); | |
1756 | */ | |
1da177e4 | 1757 | if (init_resync(conf)) |
57afd89f | 1758 | return 0; |
191ea9b2 | 1759 | } |
1da177e4 | 1760 | |
58c0fed4 | 1761 | max_sector = mddev->dev_sectors; |
1da177e4 | 1762 | if (sector_nr >= max_sector) { |
191ea9b2 N |
1763 | /* If we aborted, we need to abort the |
1764 | * sync on the 'current' bitmap chunk (there will | |
1765 | * only be one in raid1 resync. | |
1766 | * We can find the current addess in mddev->curr_resync | |
1767 | */ | |
6a806c51 N |
1768 | if (mddev->curr_resync < max_sector) /* aborted */ |
1769 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
191ea9b2 | 1770 | &sync_blocks, 1); |
6a806c51 | 1771 | else /* completed sync */ |
191ea9b2 | 1772 | conf->fullsync = 0; |
6a806c51 N |
1773 | |
1774 | bitmap_close_sync(mddev->bitmap); | |
1da177e4 LT |
1775 | close_sync(conf); |
1776 | return 0; | |
1777 | } | |
1778 | ||
07d84d10 N |
1779 | if (mddev->bitmap == NULL && |
1780 | mddev->recovery_cp == MaxSector && | |
6394cca5 | 1781 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
07d84d10 N |
1782 | conf->fullsync == 0) { |
1783 | *skipped = 1; | |
1784 | return max_sector - sector_nr; | |
1785 | } | |
6394cca5 N |
1786 | /* before building a request, check if we can skip these blocks.. |
1787 | * This call the bitmap_start_sync doesn't actually record anything | |
1788 | */ | |
e3b9703e | 1789 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
e5de485f | 1790 | !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { |
191ea9b2 N |
1791 | /* We can skip this block, and probably several more */ |
1792 | *skipped = 1; | |
1793 | return sync_blocks; | |
1794 | } | |
1da177e4 | 1795 | /* |
17999be4 N |
1796 | * If there is non-resync activity waiting for a turn, |
1797 | * and resync is going fast enough, | |
1798 | * then let it though before starting on this new sync request. | |
1da177e4 | 1799 | */ |
17999be4 | 1800 | if (!go_faster && conf->nr_waiting) |
1da177e4 | 1801 | msleep_interruptible(1000); |
17999be4 | 1802 | |
b47490c9 | 1803 | bitmap_cond_end_sync(mddev->bitmap, sector_nr); |
17999be4 N |
1804 | raise_barrier(conf); |
1805 | ||
1806 | conf->next_resync = sector_nr; | |
1da177e4 | 1807 | |
3e198f78 N |
1808 | r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO); |
1809 | rcu_read_lock(); | |
1da177e4 | 1810 | /* |
3e198f78 N |
1811 | * If we get a correctably read error during resync or recovery, |
1812 | * we might want to read from a different device. So we | |
1813 | * flag all drives that could conceivably be read from for READ, | |
1814 | * and any others (which will be non-In_sync devices) for WRITE. | |
1815 | * If a read fails, we try reading from something else for which READ | |
1816 | * is OK. | |
1da177e4 | 1817 | */ |
1da177e4 | 1818 | |
1da177e4 LT |
1819 | r1_bio->mddev = mddev; |
1820 | r1_bio->sector = sector_nr; | |
191ea9b2 | 1821 | r1_bio->state = 0; |
1da177e4 | 1822 | set_bit(R1BIO_IsSync, &r1_bio->state); |
1da177e4 LT |
1823 | |
1824 | for (i=0; i < conf->raid_disks; i++) { | |
3e198f78 | 1825 | mdk_rdev_t *rdev; |
1da177e4 LT |
1826 | bio = r1_bio->bios[i]; |
1827 | ||
1828 | /* take from bio_init */ | |
1829 | bio->bi_next = NULL; | |
1830 | bio->bi_flags |= 1 << BIO_UPTODATE; | |
802ba064 | 1831 | bio->bi_rw = READ; |
1da177e4 LT |
1832 | bio->bi_vcnt = 0; |
1833 | bio->bi_idx = 0; | |
1834 | bio->bi_phys_segments = 0; | |
1da177e4 LT |
1835 | bio->bi_size = 0; |
1836 | bio->bi_end_io = NULL; | |
1837 | bio->bi_private = NULL; | |
1838 | ||
3e198f78 N |
1839 | rdev = rcu_dereference(conf->mirrors[i].rdev); |
1840 | if (rdev == NULL || | |
1841 | test_bit(Faulty, &rdev->flags)) { | |
e3b9703e N |
1842 | still_degraded = 1; |
1843 | continue; | |
3e198f78 | 1844 | } else if (!test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
1845 | bio->bi_rw = WRITE; |
1846 | bio->bi_end_io = end_sync_write; | |
1847 | write_targets ++; | |
3e198f78 N |
1848 | } else { |
1849 | /* may need to read from here */ | |
1850 | bio->bi_rw = READ; | |
1851 | bio->bi_end_io = end_sync_read; | |
1852 | if (test_bit(WriteMostly, &rdev->flags)) { | |
1853 | if (wonly < 0) | |
1854 | wonly = i; | |
1855 | } else { | |
1856 | if (disk < 0) | |
1857 | disk = i; | |
1858 | } | |
1859 | read_targets++; | |
1860 | } | |
1861 | atomic_inc(&rdev->nr_pending); | |
1862 | bio->bi_sector = sector_nr + rdev->data_offset; | |
1863 | bio->bi_bdev = rdev->bdev; | |
1da177e4 LT |
1864 | bio->bi_private = r1_bio; |
1865 | } | |
3e198f78 N |
1866 | rcu_read_unlock(); |
1867 | if (disk < 0) | |
1868 | disk = wonly; | |
1869 | r1_bio->read_disk = disk; | |
191ea9b2 | 1870 | |
3e198f78 N |
1871 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0) |
1872 | /* extra read targets are also write targets */ | |
1873 | write_targets += read_targets-1; | |
1874 | ||
1875 | if (write_targets == 0 || read_targets == 0) { | |
1da177e4 LT |
1876 | /* There is nowhere to write, so all non-sync |
1877 | * drives must be failed - so we are finished | |
1878 | */ | |
57afd89f N |
1879 | sector_t rv = max_sector - sector_nr; |
1880 | *skipped = 1; | |
1da177e4 | 1881 | put_buf(r1_bio); |
1da177e4 LT |
1882 | return rv; |
1883 | } | |
1884 | ||
c6207277 N |
1885 | if (max_sector > mddev->resync_max) |
1886 | max_sector = mddev->resync_max; /* Don't do IO beyond here */ | |
1da177e4 | 1887 | nr_sectors = 0; |
289e99e8 | 1888 | sync_blocks = 0; |
1da177e4 LT |
1889 | do { |
1890 | struct page *page; | |
1891 | int len = PAGE_SIZE; | |
1892 | if (sector_nr + (len>>9) > max_sector) | |
1893 | len = (max_sector - sector_nr) << 9; | |
1894 | if (len == 0) | |
1895 | break; | |
6a806c51 N |
1896 | if (sync_blocks == 0) { |
1897 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, | |
e5de485f N |
1898 | &sync_blocks, still_degraded) && |
1899 | !conf->fullsync && | |
1900 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6a806c51 | 1901 | break; |
9e77c485 | 1902 | BUG_ON(sync_blocks < (PAGE_SIZE>>9)); |
6a806c51 N |
1903 | if (len > (sync_blocks<<9)) |
1904 | len = sync_blocks<<9; | |
ab7a30c7 | 1905 | } |
191ea9b2 | 1906 | |
1da177e4 LT |
1907 | for (i=0 ; i < conf->raid_disks; i++) { |
1908 | bio = r1_bio->bios[i]; | |
1909 | if (bio->bi_end_io) { | |
d11c171e | 1910 | page = bio->bi_io_vec[bio->bi_vcnt].bv_page; |
1da177e4 LT |
1911 | if (bio_add_page(bio, page, len, 0) == 0) { |
1912 | /* stop here */ | |
d11c171e | 1913 | bio->bi_io_vec[bio->bi_vcnt].bv_page = page; |
1da177e4 LT |
1914 | while (i > 0) { |
1915 | i--; | |
1916 | bio = r1_bio->bios[i]; | |
6a806c51 N |
1917 | if (bio->bi_end_io==NULL) |
1918 | continue; | |
1da177e4 LT |
1919 | /* remove last page from this bio */ |
1920 | bio->bi_vcnt--; | |
1921 | bio->bi_size -= len; | |
1922 | bio->bi_flags &= ~(1<< BIO_SEG_VALID); | |
1923 | } | |
1924 | goto bio_full; | |
1925 | } | |
1926 | } | |
1927 | } | |
1928 | nr_sectors += len>>9; | |
1929 | sector_nr += len>>9; | |
191ea9b2 | 1930 | sync_blocks -= (len>>9); |
1da177e4 LT |
1931 | } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES); |
1932 | bio_full: | |
1da177e4 LT |
1933 | r1_bio->sectors = nr_sectors; |
1934 | ||
d11c171e N |
1935 | /* For a user-requested sync, we read all readable devices and do a |
1936 | * compare | |
1937 | */ | |
1938 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
1939 | atomic_set(&r1_bio->remaining, read_targets); | |
1940 | for (i=0; i<conf->raid_disks; i++) { | |
1941 | bio = r1_bio->bios[i]; | |
1942 | if (bio->bi_end_io == end_sync_read) { | |
ddac7c7e | 1943 | md_sync_acct(bio->bi_bdev, nr_sectors); |
d11c171e N |
1944 | generic_make_request(bio); |
1945 | } | |
1946 | } | |
1947 | } else { | |
1948 | atomic_set(&r1_bio->remaining, 1); | |
1949 | bio = r1_bio->bios[r1_bio->read_disk]; | |
ddac7c7e | 1950 | md_sync_acct(bio->bi_bdev, nr_sectors); |
d11c171e | 1951 | generic_make_request(bio); |
1da177e4 | 1952 | |
d11c171e | 1953 | } |
1da177e4 LT |
1954 | return nr_sectors; |
1955 | } | |
1956 | ||
80c3a6ce DW |
1957 | static sector_t raid1_size(mddev_t *mddev, sector_t sectors, int raid_disks) |
1958 | { | |
1959 | if (sectors) | |
1960 | return sectors; | |
1961 | ||
1962 | return mddev->dev_sectors; | |
1963 | } | |
1964 | ||
709ae487 | 1965 | static conf_t *setup_conf(mddev_t *mddev) |
1da177e4 LT |
1966 | { |
1967 | conf_t *conf; | |
709ae487 | 1968 | int i; |
1da177e4 LT |
1969 | mirror_info_t *disk; |
1970 | mdk_rdev_t *rdev; | |
709ae487 | 1971 | int err = -ENOMEM; |
1da177e4 | 1972 | |
9ffae0cf | 1973 | conf = kzalloc(sizeof(conf_t), GFP_KERNEL); |
1da177e4 | 1974 | if (!conf) |
709ae487 | 1975 | goto abort; |
1da177e4 | 1976 | |
9ffae0cf | 1977 | conf->mirrors = kzalloc(sizeof(struct mirror_info)*mddev->raid_disks, |
1da177e4 LT |
1978 | GFP_KERNEL); |
1979 | if (!conf->mirrors) | |
709ae487 | 1980 | goto abort; |
1da177e4 | 1981 | |
ddaf22ab N |
1982 | conf->tmppage = alloc_page(GFP_KERNEL); |
1983 | if (!conf->tmppage) | |
709ae487 | 1984 | goto abort; |
ddaf22ab | 1985 | |
709ae487 | 1986 | conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL); |
1da177e4 | 1987 | if (!conf->poolinfo) |
709ae487 | 1988 | goto abort; |
1da177e4 LT |
1989 | conf->poolinfo->raid_disks = mddev->raid_disks; |
1990 | conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, | |
1991 | r1bio_pool_free, | |
1992 | conf->poolinfo); | |
1993 | if (!conf->r1bio_pool) | |
709ae487 N |
1994 | goto abort; |
1995 | ||
ed9bfdf1 | 1996 | conf->poolinfo->mddev = mddev; |
1da177e4 | 1997 | |
e7e72bf6 | 1998 | spin_lock_init(&conf->device_lock); |
159ec1fc | 1999 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
709ae487 | 2000 | int disk_idx = rdev->raid_disk; |
1da177e4 LT |
2001 | if (disk_idx >= mddev->raid_disks |
2002 | || disk_idx < 0) | |
2003 | continue; | |
2004 | disk = conf->mirrors + disk_idx; | |
2005 | ||
2006 | disk->rdev = rdev; | |
1da177e4 LT |
2007 | |
2008 | disk->head_position = 0; | |
1da177e4 LT |
2009 | } |
2010 | conf->raid_disks = mddev->raid_disks; | |
2011 | conf->mddev = mddev; | |
1da177e4 | 2012 | INIT_LIST_HEAD(&conf->retry_list); |
1da177e4 LT |
2013 | |
2014 | spin_lock_init(&conf->resync_lock); | |
17999be4 | 2015 | init_waitqueue_head(&conf->wait_barrier); |
1da177e4 | 2016 | |
191ea9b2 N |
2017 | bio_list_init(&conf->pending_bio_list); |
2018 | bio_list_init(&conf->flushing_bio_list); | |
2019 | ||
709ae487 | 2020 | conf->last_used = -1; |
1da177e4 LT |
2021 | for (i = 0; i < conf->raid_disks; i++) { |
2022 | ||
2023 | disk = conf->mirrors + i; | |
2024 | ||
5fd6c1dc N |
2025 | if (!disk->rdev || |
2026 | !test_bit(In_sync, &disk->rdev->flags)) { | |
1da177e4 | 2027 | disk->head_position = 0; |
918f0238 N |
2028 | if (disk->rdev) |
2029 | conf->fullsync = 1; | |
709ae487 N |
2030 | } else if (conf->last_used < 0) |
2031 | /* | |
2032 | * The first working device is used as a | |
2033 | * starting point to read balancing. | |
2034 | */ | |
2035 | conf->last_used = i; | |
1da177e4 | 2036 | } |
709ae487 N |
2037 | |
2038 | err = -EIO; | |
2039 | if (conf->last_used < 0) { | |
11ce99e6 | 2040 | printk(KERN_ERR "raid1: no operational mirrors for %s\n", |
709ae487 N |
2041 | mdname(mddev)); |
2042 | goto abort; | |
2043 | } | |
2044 | err = -ENOMEM; | |
2045 | conf->thread = md_register_thread(raid1d, mddev, NULL); | |
2046 | if (!conf->thread) { | |
2047 | printk(KERN_ERR | |
2048 | "raid1: couldn't allocate thread for %s\n", | |
2049 | mdname(mddev)); | |
2050 | goto abort; | |
11ce99e6 | 2051 | } |
1da177e4 | 2052 | |
709ae487 N |
2053 | return conf; |
2054 | ||
2055 | abort: | |
2056 | if (conf) { | |
2057 | if (conf->r1bio_pool) | |
2058 | mempool_destroy(conf->r1bio_pool); | |
2059 | kfree(conf->mirrors); | |
2060 | safe_put_page(conf->tmppage); | |
2061 | kfree(conf->poolinfo); | |
2062 | kfree(conf); | |
2063 | } | |
2064 | return ERR_PTR(err); | |
2065 | } | |
2066 | ||
2067 | static int run(mddev_t *mddev) | |
2068 | { | |
2069 | conf_t *conf; | |
2070 | int i; | |
2071 | mdk_rdev_t *rdev; | |
2072 | ||
2073 | if (mddev->level != 1) { | |
2074 | printk("raid1: %s: raid level not set to mirroring (%d)\n", | |
2075 | mdname(mddev), mddev->level); | |
2076 | return -EIO; | |
2077 | } | |
2078 | if (mddev->reshape_position != MaxSector) { | |
2079 | printk("raid1: %s: reshape_position set but not supported\n", | |
2080 | mdname(mddev)); | |
2081 | return -EIO; | |
2082 | } | |
1da177e4 | 2083 | /* |
709ae487 N |
2084 | * copy the already verified devices into our private RAID1 |
2085 | * bookkeeping area. [whatever we allocate in run(), | |
2086 | * should be freed in stop()] | |
1da177e4 | 2087 | */ |
709ae487 N |
2088 | if (mddev->private == NULL) |
2089 | conf = setup_conf(mddev); | |
2090 | else | |
2091 | conf = mddev->private; | |
1da177e4 | 2092 | |
709ae487 N |
2093 | if (IS_ERR(conf)) |
2094 | return PTR_ERR(conf); | |
1da177e4 | 2095 | |
709ae487 N |
2096 | mddev->queue->queue_lock = &conf->device_lock; |
2097 | list_for_each_entry(rdev, &mddev->disks, same_set) { | |
2098 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
2099 | rdev->data_offset << 9); | |
2100 | /* as we don't honour merge_bvec_fn, we must never risk | |
2101 | * violating it, so limit ->max_sector to one PAGE, as | |
2102 | * a one page request is never in violation. | |
2103 | */ | |
2104 | if (rdev->bdev->bd_disk->queue->merge_bvec_fn && | |
2105 | queue_max_sectors(mddev->queue) > (PAGE_SIZE>>9)) | |
086fa5ff | 2106 | blk_queue_max_hw_sectors(mddev->queue, PAGE_SIZE>>9); |
1da177e4 | 2107 | } |
191ea9b2 | 2108 | |
709ae487 N |
2109 | mddev->degraded = 0; |
2110 | for (i=0; i < conf->raid_disks; i++) | |
2111 | if (conf->mirrors[i].rdev == NULL || | |
2112 | !test_bit(In_sync, &conf->mirrors[i].rdev->flags) || | |
2113 | test_bit(Faulty, &conf->mirrors[i].rdev->flags)) | |
2114 | mddev->degraded++; | |
2115 | ||
2116 | if (conf->raid_disks - mddev->degraded == 1) | |
2117 | mddev->recovery_cp = MaxSector; | |
2118 | ||
8c6ac868 AN |
2119 | if (mddev->recovery_cp != MaxSector) |
2120 | printk(KERN_NOTICE "raid1: %s is not clean" | |
2121 | " -- starting background reconstruction\n", | |
2122 | mdname(mddev)); | |
1da177e4 LT |
2123 | printk(KERN_INFO |
2124 | "raid1: raid set %s active with %d out of %d mirrors\n", | |
2125 | mdname(mddev), mddev->raid_disks - mddev->degraded, | |
2126 | mddev->raid_disks); | |
709ae487 | 2127 | |
1da177e4 LT |
2128 | /* |
2129 | * Ok, everything is just fine now | |
2130 | */ | |
709ae487 N |
2131 | mddev->thread = conf->thread; |
2132 | conf->thread = NULL; | |
2133 | mddev->private = conf; | |
2134 | ||
1f403624 | 2135 | md_set_array_sectors(mddev, raid1_size(mddev, 0, 0)); |
1da177e4 | 2136 | |
7a5febe9 | 2137 | mddev->queue->unplug_fn = raid1_unplug; |
0d129228 N |
2138 | mddev->queue->backing_dev_info.congested_fn = raid1_congested; |
2139 | mddev->queue->backing_dev_info.congested_data = mddev; | |
ac5e7113 | 2140 | md_integrity_register(mddev); |
1da177e4 | 2141 | return 0; |
1da177e4 LT |
2142 | } |
2143 | ||
2144 | static int stop(mddev_t *mddev) | |
2145 | { | |
070ec55d | 2146 | conf_t *conf = mddev->private; |
4b6d287f N |
2147 | struct bitmap *bitmap = mddev->bitmap; |
2148 | int behind_wait = 0; | |
2149 | ||
2150 | /* wait for behind writes to complete */ | |
2151 | while (bitmap && atomic_read(&bitmap->behind_writes) > 0) { | |
2152 | behind_wait++; | |
2153 | printk(KERN_INFO "raid1: behind writes in progress on device %s, waiting to stop (%d)\n", mdname(mddev), behind_wait); | |
2154 | set_current_state(TASK_UNINTERRUPTIBLE); | |
2155 | schedule_timeout(HZ); /* wait a second */ | |
2156 | /* need to kick something here to make sure I/O goes? */ | |
2157 | } | |
1da177e4 | 2158 | |
409c57f3 N |
2159 | raise_barrier(conf); |
2160 | lower_barrier(conf); | |
2161 | ||
1da177e4 LT |
2162 | md_unregister_thread(mddev->thread); |
2163 | mddev->thread = NULL; | |
2164 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ | |
2165 | if (conf->r1bio_pool) | |
2166 | mempool_destroy(conf->r1bio_pool); | |
990a8baf JJ |
2167 | kfree(conf->mirrors); |
2168 | kfree(conf->poolinfo); | |
1da177e4 LT |
2169 | kfree(conf); |
2170 | mddev->private = NULL; | |
2171 | return 0; | |
2172 | } | |
2173 | ||
2174 | static int raid1_resize(mddev_t *mddev, sector_t sectors) | |
2175 | { | |
2176 | /* no resync is happening, and there is enough space | |
2177 | * on all devices, so we can resize. | |
2178 | * We need to make sure resync covers any new space. | |
2179 | * If the array is shrinking we should possibly wait until | |
2180 | * any io in the removed space completes, but it hardly seems | |
2181 | * worth it. | |
2182 | */ | |
1f403624 | 2183 | md_set_array_sectors(mddev, raid1_size(mddev, sectors, 0)); |
b522adcd DW |
2184 | if (mddev->array_sectors > raid1_size(mddev, sectors, 0)) |
2185 | return -EINVAL; | |
f233ea5c | 2186 | set_capacity(mddev->gendisk, mddev->array_sectors); |
44ce6294 | 2187 | mddev->changed = 1; |
449aad3e | 2188 | revalidate_disk(mddev->gendisk); |
b522adcd | 2189 | if (sectors > mddev->dev_sectors && |
f233ea5c | 2190 | mddev->recovery_cp == MaxSector) { |
58c0fed4 | 2191 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
2192 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
2193 | } | |
b522adcd | 2194 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 2195 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
2196 | return 0; |
2197 | } | |
2198 | ||
63c70c4f | 2199 | static int raid1_reshape(mddev_t *mddev) |
1da177e4 LT |
2200 | { |
2201 | /* We need to: | |
2202 | * 1/ resize the r1bio_pool | |
2203 | * 2/ resize conf->mirrors | |
2204 | * | |
2205 | * We allocate a new r1bio_pool if we can. | |
2206 | * Then raise a device barrier and wait until all IO stops. | |
2207 | * Then resize conf->mirrors and swap in the new r1bio pool. | |
6ea9c07c N |
2208 | * |
2209 | * At the same time, we "pack" the devices so that all the missing | |
2210 | * devices have the higher raid_disk numbers. | |
1da177e4 LT |
2211 | */ |
2212 | mempool_t *newpool, *oldpool; | |
2213 | struct pool_info *newpoolinfo; | |
2214 | mirror_info_t *newmirrors; | |
070ec55d | 2215 | conf_t *conf = mddev->private; |
63c70c4f | 2216 | int cnt, raid_disks; |
c04be0aa | 2217 | unsigned long flags; |
b5470dc5 | 2218 | int d, d2, err; |
1da177e4 | 2219 | |
63c70c4f | 2220 | /* Cannot change chunk_size, layout, or level */ |
664e7c41 | 2221 | if (mddev->chunk_sectors != mddev->new_chunk_sectors || |
63c70c4f N |
2222 | mddev->layout != mddev->new_layout || |
2223 | mddev->level != mddev->new_level) { | |
664e7c41 | 2224 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
63c70c4f N |
2225 | mddev->new_layout = mddev->layout; |
2226 | mddev->new_level = mddev->level; | |
2227 | return -EINVAL; | |
2228 | } | |
2229 | ||
b5470dc5 DW |
2230 | err = md_allow_write(mddev); |
2231 | if (err) | |
2232 | return err; | |
2a2275d6 | 2233 | |
63c70c4f N |
2234 | raid_disks = mddev->raid_disks + mddev->delta_disks; |
2235 | ||
6ea9c07c N |
2236 | if (raid_disks < conf->raid_disks) { |
2237 | cnt=0; | |
2238 | for (d= 0; d < conf->raid_disks; d++) | |
2239 | if (conf->mirrors[d].rdev) | |
2240 | cnt++; | |
2241 | if (cnt > raid_disks) | |
1da177e4 | 2242 | return -EBUSY; |
6ea9c07c | 2243 | } |
1da177e4 LT |
2244 | |
2245 | newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL); | |
2246 | if (!newpoolinfo) | |
2247 | return -ENOMEM; | |
2248 | newpoolinfo->mddev = mddev; | |
2249 | newpoolinfo->raid_disks = raid_disks; | |
2250 | ||
2251 | newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, | |
2252 | r1bio_pool_free, newpoolinfo); | |
2253 | if (!newpool) { | |
2254 | kfree(newpoolinfo); | |
2255 | return -ENOMEM; | |
2256 | } | |
9ffae0cf | 2257 | newmirrors = kzalloc(sizeof(struct mirror_info) * raid_disks, GFP_KERNEL); |
1da177e4 LT |
2258 | if (!newmirrors) { |
2259 | kfree(newpoolinfo); | |
2260 | mempool_destroy(newpool); | |
2261 | return -ENOMEM; | |
2262 | } | |
1da177e4 | 2263 | |
17999be4 | 2264 | raise_barrier(conf); |
1da177e4 LT |
2265 | |
2266 | /* ok, everything is stopped */ | |
2267 | oldpool = conf->r1bio_pool; | |
2268 | conf->r1bio_pool = newpool; | |
6ea9c07c | 2269 | |
a88aa786 N |
2270 | for (d = d2 = 0; d < conf->raid_disks; d++) { |
2271 | mdk_rdev_t *rdev = conf->mirrors[d].rdev; | |
2272 | if (rdev && rdev->raid_disk != d2) { | |
2273 | char nm[20]; | |
2274 | sprintf(nm, "rd%d", rdev->raid_disk); | |
2275 | sysfs_remove_link(&mddev->kobj, nm); | |
2276 | rdev->raid_disk = d2; | |
2277 | sprintf(nm, "rd%d", rdev->raid_disk); | |
2278 | sysfs_remove_link(&mddev->kobj, nm); | |
2279 | if (sysfs_create_link(&mddev->kobj, | |
2280 | &rdev->kobj, nm)) | |
2281 | printk(KERN_WARNING | |
2282 | "md/raid1: cannot register " | |
2283 | "%s for %s\n", | |
2284 | nm, mdname(mddev)); | |
6ea9c07c | 2285 | } |
a88aa786 N |
2286 | if (rdev) |
2287 | newmirrors[d2++].rdev = rdev; | |
2288 | } | |
1da177e4 LT |
2289 | kfree(conf->mirrors); |
2290 | conf->mirrors = newmirrors; | |
2291 | kfree(conf->poolinfo); | |
2292 | conf->poolinfo = newpoolinfo; | |
2293 | ||
c04be0aa | 2294 | spin_lock_irqsave(&conf->device_lock, flags); |
1da177e4 | 2295 | mddev->degraded += (raid_disks - conf->raid_disks); |
c04be0aa | 2296 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 2297 | conf->raid_disks = mddev->raid_disks = raid_disks; |
63c70c4f | 2298 | mddev->delta_disks = 0; |
1da177e4 | 2299 | |
6ea9c07c | 2300 | conf->last_used = 0; /* just make sure it is in-range */ |
17999be4 | 2301 | lower_barrier(conf); |
1da177e4 LT |
2302 | |
2303 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
2304 | md_wakeup_thread(mddev->thread); | |
2305 | ||
2306 | mempool_destroy(oldpool); | |
2307 | return 0; | |
2308 | } | |
2309 | ||
500af87a | 2310 | static void raid1_quiesce(mddev_t *mddev, int state) |
36fa3063 | 2311 | { |
070ec55d | 2312 | conf_t *conf = mddev->private; |
36fa3063 N |
2313 | |
2314 | switch(state) { | |
6eef4b21 N |
2315 | case 2: /* wake for suspend */ |
2316 | wake_up(&conf->wait_barrier); | |
2317 | break; | |
9e6603da | 2318 | case 1: |
17999be4 | 2319 | raise_barrier(conf); |
36fa3063 | 2320 | break; |
9e6603da | 2321 | case 0: |
17999be4 | 2322 | lower_barrier(conf); |
36fa3063 N |
2323 | break; |
2324 | } | |
36fa3063 N |
2325 | } |
2326 | ||
709ae487 N |
2327 | static void *raid1_takeover(mddev_t *mddev) |
2328 | { | |
2329 | /* raid1 can take over: | |
2330 | * raid5 with 2 devices, any layout or chunk size | |
2331 | */ | |
2332 | if (mddev->level == 5 && mddev->raid_disks == 2) { | |
2333 | conf_t *conf; | |
2334 | mddev->new_level = 1; | |
2335 | mddev->new_layout = 0; | |
2336 | mddev->new_chunk_sectors = 0; | |
2337 | conf = setup_conf(mddev); | |
2338 | if (!IS_ERR(conf)) | |
2339 | conf->barrier = 1; | |
2340 | return conf; | |
2341 | } | |
2342 | return ERR_PTR(-EINVAL); | |
2343 | } | |
1da177e4 | 2344 | |
2604b703 | 2345 | static struct mdk_personality raid1_personality = |
1da177e4 LT |
2346 | { |
2347 | .name = "raid1", | |
2604b703 | 2348 | .level = 1, |
1da177e4 LT |
2349 | .owner = THIS_MODULE, |
2350 | .make_request = make_request, | |
2351 | .run = run, | |
2352 | .stop = stop, | |
2353 | .status = status, | |
2354 | .error_handler = error, | |
2355 | .hot_add_disk = raid1_add_disk, | |
2356 | .hot_remove_disk= raid1_remove_disk, | |
2357 | .spare_active = raid1_spare_active, | |
2358 | .sync_request = sync_request, | |
2359 | .resize = raid1_resize, | |
80c3a6ce | 2360 | .size = raid1_size, |
63c70c4f | 2361 | .check_reshape = raid1_reshape, |
36fa3063 | 2362 | .quiesce = raid1_quiesce, |
709ae487 | 2363 | .takeover = raid1_takeover, |
1da177e4 LT |
2364 | }; |
2365 | ||
2366 | static int __init raid_init(void) | |
2367 | { | |
2604b703 | 2368 | return register_md_personality(&raid1_personality); |
1da177e4 LT |
2369 | } |
2370 | ||
2371 | static void raid_exit(void) | |
2372 | { | |
2604b703 | 2373 | unregister_md_personality(&raid1_personality); |
1da177e4 LT |
2374 | } |
2375 | ||
2376 | module_init(raid_init); | |
2377 | module_exit(raid_exit); | |
2378 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 2379 | MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD"); |
1da177e4 | 2380 | MODULE_ALIAS("md-personality-3"); /* RAID1 */ |
d9d166c2 | 2381 | MODULE_ALIAS("md-raid1"); |
2604b703 | 2382 | MODULE_ALIAS("md-level-1"); |