]>
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> |
bff61975 | 37 | #include <linux/seq_file.h> |
43b2e5d8 | 38 | #include "md.h" |
ef740c37 CH |
39 | #include "raid1.h" |
40 | #include "bitmap.h" | |
191ea9b2 N |
41 | |
42 | #define DEBUG 0 | |
43 | #if DEBUG | |
44 | #define PRINTK(x...) printk(x) | |
45 | #else | |
46 | #define PRINTK(x...) | |
47 | #endif | |
1da177e4 LT |
48 | |
49 | /* | |
50 | * Number of guaranteed r1bios in case of extreme VM load: | |
51 | */ | |
52 | #define NR_RAID1_BIOS 256 | |
53 | ||
1da177e4 | 54 | |
17999be4 N |
55 | static void allow_barrier(conf_t *conf); |
56 | static void lower_barrier(conf_t *conf); | |
1da177e4 | 57 | |
dd0fc66f | 58 | static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 LT |
59 | { |
60 | struct pool_info *pi = data; | |
1da177e4 LT |
61 | int size = offsetof(r1bio_t, bios[pi->raid_disks]); |
62 | ||
63 | /* allocate a r1bio with room for raid_disks entries in the bios array */ | |
7eaceacc | 64 | return kzalloc(size, gfp_flags); |
1da177e4 LT |
65 | } |
66 | ||
67 | static void r1bio_pool_free(void *r1_bio, void *data) | |
68 | { | |
69 | kfree(r1_bio); | |
70 | } | |
71 | ||
72 | #define RESYNC_BLOCK_SIZE (64*1024) | |
73 | //#define RESYNC_BLOCK_SIZE PAGE_SIZE | |
74 | #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9) | |
75 | #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE) | |
76 | #define RESYNC_WINDOW (2048*1024) | |
77 | ||
dd0fc66f | 78 | static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 LT |
79 | { |
80 | struct pool_info *pi = data; | |
81 | struct page *page; | |
82 | r1bio_t *r1_bio; | |
83 | struct bio *bio; | |
84 | int i, j; | |
85 | ||
86 | r1_bio = r1bio_pool_alloc(gfp_flags, pi); | |
7eaceacc | 87 | if (!r1_bio) |
1da177e4 | 88 | return NULL; |
1da177e4 LT |
89 | |
90 | /* | |
91 | * Allocate bios : 1 for reading, n-1 for writing | |
92 | */ | |
93 | for (j = pi->raid_disks ; j-- ; ) { | |
6746557f | 94 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); |
1da177e4 LT |
95 | if (!bio) |
96 | goto out_free_bio; | |
97 | r1_bio->bios[j] = bio; | |
98 | } | |
99 | /* | |
100 | * Allocate RESYNC_PAGES data pages and attach them to | |
d11c171e N |
101 | * the first bio. |
102 | * If this is a user-requested check/repair, allocate | |
103 | * RESYNC_PAGES for each bio. | |
1da177e4 | 104 | */ |
d11c171e N |
105 | if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) |
106 | j = pi->raid_disks; | |
107 | else | |
108 | j = 1; | |
109 | while(j--) { | |
110 | bio = r1_bio->bios[j]; | |
111 | for (i = 0; i < RESYNC_PAGES; i++) { | |
112 | page = alloc_page(gfp_flags); | |
113 | if (unlikely(!page)) | |
114 | goto out_free_pages; | |
115 | ||
116 | bio->bi_io_vec[i].bv_page = page; | |
303a0e11 | 117 | bio->bi_vcnt = i+1; |
d11c171e N |
118 | } |
119 | } | |
120 | /* If not user-requests, copy the page pointers to all bios */ | |
121 | if (!test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) { | |
122 | for (i=0; i<RESYNC_PAGES ; i++) | |
123 | for (j=1; j<pi->raid_disks; j++) | |
124 | r1_bio->bios[j]->bi_io_vec[i].bv_page = | |
125 | r1_bio->bios[0]->bi_io_vec[i].bv_page; | |
1da177e4 LT |
126 | } |
127 | ||
128 | r1_bio->master_bio = NULL; | |
129 | ||
130 | return r1_bio; | |
131 | ||
132 | out_free_pages: | |
303a0e11 N |
133 | for (j=0 ; j < pi->raid_disks; j++) |
134 | for (i=0; i < r1_bio->bios[j]->bi_vcnt ; i++) | |
135 | put_page(r1_bio->bios[j]->bi_io_vec[i].bv_page); | |
d11c171e | 136 | j = -1; |
1da177e4 LT |
137 | out_free_bio: |
138 | while ( ++j < pi->raid_disks ) | |
139 | bio_put(r1_bio->bios[j]); | |
140 | r1bio_pool_free(r1_bio, data); | |
141 | return NULL; | |
142 | } | |
143 | ||
144 | static void r1buf_pool_free(void *__r1_bio, void *data) | |
145 | { | |
146 | struct pool_info *pi = data; | |
d11c171e | 147 | int i,j; |
1da177e4 | 148 | r1bio_t *r1bio = __r1_bio; |
1da177e4 | 149 | |
d11c171e N |
150 | for (i = 0; i < RESYNC_PAGES; i++) |
151 | for (j = pi->raid_disks; j-- ;) { | |
152 | if (j == 0 || | |
153 | r1bio->bios[j]->bi_io_vec[i].bv_page != | |
154 | r1bio->bios[0]->bi_io_vec[i].bv_page) | |
1345b1d8 | 155 | safe_put_page(r1bio->bios[j]->bi_io_vec[i].bv_page); |
d11c171e | 156 | } |
1da177e4 LT |
157 | for (i=0 ; i < pi->raid_disks; i++) |
158 | bio_put(r1bio->bios[i]); | |
159 | ||
160 | r1bio_pool_free(r1bio, data); | |
161 | } | |
162 | ||
163 | static void put_all_bios(conf_t *conf, r1bio_t *r1_bio) | |
164 | { | |
165 | int i; | |
166 | ||
167 | for (i = 0; i < conf->raid_disks; i++) { | |
168 | struct bio **bio = r1_bio->bios + i; | |
cf30a473 | 169 | if (*bio && *bio != IO_BLOCKED) |
1da177e4 LT |
170 | bio_put(*bio); |
171 | *bio = NULL; | |
172 | } | |
173 | } | |
174 | ||
858119e1 | 175 | static void free_r1bio(r1bio_t *r1_bio) |
1da177e4 | 176 | { |
070ec55d | 177 | conf_t *conf = r1_bio->mddev->private; |
1da177e4 LT |
178 | |
179 | /* | |
180 | * Wake up any possible resync thread that waits for the device | |
181 | * to go idle. | |
182 | */ | |
17999be4 | 183 | allow_barrier(conf); |
1da177e4 LT |
184 | |
185 | put_all_bios(conf, r1_bio); | |
186 | mempool_free(r1_bio, conf->r1bio_pool); | |
187 | } | |
188 | ||
858119e1 | 189 | static void put_buf(r1bio_t *r1_bio) |
1da177e4 | 190 | { |
070ec55d | 191 | conf_t *conf = r1_bio->mddev->private; |
3e198f78 N |
192 | int i; |
193 | ||
194 | for (i=0; i<conf->raid_disks; i++) { | |
195 | struct bio *bio = r1_bio->bios[i]; | |
196 | if (bio->bi_end_io) | |
197 | rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev); | |
198 | } | |
1da177e4 LT |
199 | |
200 | mempool_free(r1_bio, conf->r1buf_pool); | |
201 | ||
17999be4 | 202 | lower_barrier(conf); |
1da177e4 LT |
203 | } |
204 | ||
205 | static void reschedule_retry(r1bio_t *r1_bio) | |
206 | { | |
207 | unsigned long flags; | |
208 | mddev_t *mddev = r1_bio->mddev; | |
070ec55d | 209 | conf_t *conf = mddev->private; |
1da177e4 LT |
210 | |
211 | spin_lock_irqsave(&conf->device_lock, flags); | |
212 | list_add(&r1_bio->retry_list, &conf->retry_list); | |
ddaf22ab | 213 | conf->nr_queued ++; |
1da177e4 LT |
214 | spin_unlock_irqrestore(&conf->device_lock, flags); |
215 | ||
17999be4 | 216 | wake_up(&conf->wait_barrier); |
1da177e4 LT |
217 | md_wakeup_thread(mddev->thread); |
218 | } | |
219 | ||
220 | /* | |
221 | * raid_end_bio_io() is called when we have finished servicing a mirrored | |
222 | * operation and are ready to return a success/failure code to the buffer | |
223 | * cache layer. | |
224 | */ | |
225 | static void raid_end_bio_io(r1bio_t *r1_bio) | |
226 | { | |
227 | struct bio *bio = r1_bio->master_bio; | |
228 | ||
4b6d287f N |
229 | /* if nobody has done the final endio yet, do it now */ |
230 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { | |
231 | PRINTK(KERN_DEBUG "raid1: sync end %s on sectors %llu-%llu\n", | |
232 | (bio_data_dir(bio) == WRITE) ? "write" : "read", | |
233 | (unsigned long long) bio->bi_sector, | |
234 | (unsigned long long) bio->bi_sector + | |
235 | (bio->bi_size >> 9) - 1); | |
236 | ||
6712ecf8 | 237 | bio_endio(bio, |
4b6d287f N |
238 | test_bit(R1BIO_Uptodate, &r1_bio->state) ? 0 : -EIO); |
239 | } | |
1da177e4 LT |
240 | free_r1bio(r1_bio); |
241 | } | |
242 | ||
243 | /* | |
244 | * Update disk head position estimator based on IRQ completion info. | |
245 | */ | |
246 | static inline void update_head_pos(int disk, r1bio_t *r1_bio) | |
247 | { | |
070ec55d | 248 | conf_t *conf = r1_bio->mddev->private; |
1da177e4 LT |
249 | |
250 | conf->mirrors[disk].head_position = | |
251 | r1_bio->sector + (r1_bio->sectors); | |
252 | } | |
253 | ||
6712ecf8 | 254 | static void raid1_end_read_request(struct bio *bio, int error) |
1da177e4 LT |
255 | { |
256 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
7b92813c | 257 | r1bio_t *r1_bio = bio->bi_private; |
1da177e4 | 258 | int mirror; |
070ec55d | 259 | conf_t *conf = r1_bio->mddev->private; |
1da177e4 | 260 | |
1da177e4 LT |
261 | mirror = r1_bio->read_disk; |
262 | /* | |
263 | * this branch is our 'one mirror IO has finished' event handler: | |
264 | */ | |
ddaf22ab N |
265 | update_head_pos(mirror, r1_bio); |
266 | ||
dd00a99e N |
267 | if (uptodate) |
268 | set_bit(R1BIO_Uptodate, &r1_bio->state); | |
269 | else { | |
270 | /* If all other devices have failed, we want to return | |
271 | * the error upwards rather than fail the last device. | |
272 | * Here we redefine "uptodate" to mean "Don't want to retry" | |
1da177e4 | 273 | */ |
dd00a99e N |
274 | unsigned long flags; |
275 | spin_lock_irqsave(&conf->device_lock, flags); | |
276 | if (r1_bio->mddev->degraded == conf->raid_disks || | |
277 | (r1_bio->mddev->degraded == conf->raid_disks-1 && | |
278 | !test_bit(Faulty, &conf->mirrors[mirror].rdev->flags))) | |
279 | uptodate = 1; | |
280 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
281 | } | |
1da177e4 | 282 | |
dd00a99e | 283 | if (uptodate) |
1da177e4 | 284 | raid_end_bio_io(r1_bio); |
dd00a99e | 285 | else { |
1da177e4 LT |
286 | /* |
287 | * oops, read error: | |
288 | */ | |
289 | char b[BDEVNAME_SIZE]; | |
290 | if (printk_ratelimit()) | |
9dd1e2fa N |
291 | printk(KERN_ERR "md/raid1:%s: %s: rescheduling sector %llu\n", |
292 | mdname(conf->mddev), | |
1da177e4 LT |
293 | bdevname(conf->mirrors[mirror].rdev->bdev,b), (unsigned long long)r1_bio->sector); |
294 | reschedule_retry(r1_bio); | |
295 | } | |
296 | ||
297 | rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev); | |
1da177e4 LT |
298 | } |
299 | ||
4e78064f N |
300 | static void r1_bio_write_done(r1bio_t *r1_bio, int vcnt, struct bio_vec *bv, |
301 | int behind) | |
302 | { | |
303 | if (atomic_dec_and_test(&r1_bio->remaining)) | |
304 | { | |
305 | /* it really is the end of this request */ | |
306 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { | |
307 | /* free extra copy of the data pages */ | |
308 | int i = vcnt; | |
309 | while (i--) | |
310 | safe_put_page(bv[i].bv_page); | |
311 | } | |
312 | /* clear the bitmap if all writes complete successfully */ | |
313 | bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector, | |
314 | r1_bio->sectors, | |
315 | !test_bit(R1BIO_Degraded, &r1_bio->state), | |
316 | behind); | |
317 | md_write_end(r1_bio->mddev); | |
318 | raid_end_bio_io(r1_bio); | |
319 | } | |
320 | } | |
321 | ||
6712ecf8 | 322 | static void raid1_end_write_request(struct bio *bio, int error) |
1da177e4 LT |
323 | { |
324 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
7b92813c | 325 | r1bio_t *r1_bio = bio->bi_private; |
a9701a30 | 326 | int mirror, behind = test_bit(R1BIO_BehindIO, &r1_bio->state); |
070ec55d | 327 | conf_t *conf = r1_bio->mddev->private; |
04b857f7 | 328 | struct bio *to_put = NULL; |
1da177e4 | 329 | |
1da177e4 LT |
330 | |
331 | for (mirror = 0; mirror < conf->raid_disks; mirror++) | |
332 | if (r1_bio->bios[mirror] == bio) | |
333 | break; | |
334 | ||
e9c7469b TH |
335 | /* |
336 | * 'one mirror IO has finished' event handler: | |
337 | */ | |
338 | r1_bio->bios[mirror] = NULL; | |
339 | to_put = bio; | |
340 | if (!uptodate) { | |
341 | md_error(r1_bio->mddev, conf->mirrors[mirror].rdev); | |
342 | /* an I/O failed, we can't clear the bitmap */ | |
343 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
344 | } else | |
1da177e4 | 345 | /* |
e9c7469b TH |
346 | * Set R1BIO_Uptodate in our master bio, so that we |
347 | * will return a good error code for to the higher | |
348 | * levels even if IO on some other mirrored buffer | |
349 | * fails. | |
350 | * | |
351 | * The 'master' represents the composite IO operation | |
352 | * to user-side. So if something waits for IO, then it | |
353 | * will wait for the 'master' bio. | |
1da177e4 | 354 | */ |
e9c7469b TH |
355 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
356 | ||
357 | update_head_pos(mirror, r1_bio); | |
358 | ||
359 | if (behind) { | |
360 | if (test_bit(WriteMostly, &conf->mirrors[mirror].rdev->flags)) | |
361 | atomic_dec(&r1_bio->behind_remaining); | |
362 | ||
363 | /* | |
364 | * In behind mode, we ACK the master bio once the I/O | |
365 | * has safely reached all non-writemostly | |
366 | * disks. Setting the Returned bit ensures that this | |
367 | * gets done only once -- we don't ever want to return | |
368 | * -EIO here, instead we'll wait | |
369 | */ | |
370 | if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) && | |
371 | test_bit(R1BIO_Uptodate, &r1_bio->state)) { | |
372 | /* Maybe we can return now */ | |
373 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { | |
374 | struct bio *mbio = r1_bio->master_bio; | |
375 | PRINTK(KERN_DEBUG "raid1: behind end write sectors %llu-%llu\n", | |
376 | (unsigned long long) mbio->bi_sector, | |
377 | (unsigned long long) mbio->bi_sector + | |
378 | (mbio->bi_size >> 9) - 1); | |
379 | bio_endio(mbio, 0); | |
4b6d287f N |
380 | } |
381 | } | |
382 | } | |
e9c7469b TH |
383 | rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev); |
384 | ||
1da177e4 | 385 | /* |
1da177e4 LT |
386 | * Let's see if all mirrored write operations have finished |
387 | * already. | |
388 | */ | |
4e78064f | 389 | r1_bio_write_done(r1_bio, bio->bi_vcnt, bio->bi_io_vec, behind); |
c70810b3 | 390 | |
04b857f7 N |
391 | if (to_put) |
392 | bio_put(to_put); | |
1da177e4 LT |
393 | } |
394 | ||
395 | ||
396 | /* | |
397 | * This routine returns the disk from which the requested read should | |
398 | * be done. There is a per-array 'next expected sequential IO' sector | |
399 | * number - if this matches on the next IO then we use the last disk. | |
400 | * There is also a per-disk 'last know head position' sector that is | |
401 | * maintained from IRQ contexts, both the normal and the resync IO | |
402 | * completion handlers update this position correctly. If there is no | |
403 | * perfect sequential match then we pick the disk whose head is closest. | |
404 | * | |
405 | * If there are 2 mirrors in the same 2 devices, performance degrades | |
406 | * because position is mirror, not device based. | |
407 | * | |
408 | * The rdev for the device selected will have nr_pending incremented. | |
409 | */ | |
410 | static int read_balance(conf_t *conf, r1bio_t *r1_bio) | |
411 | { | |
af3a2cd6 | 412 | const sector_t this_sector = r1_bio->sector; |
1da177e4 | 413 | const int sectors = r1_bio->sectors; |
f3ac8bf7 N |
414 | int new_disk = -1; |
415 | int start_disk; | |
416 | int i; | |
1da177e4 | 417 | sector_t new_distance, current_distance; |
8ddf9efe | 418 | mdk_rdev_t *rdev; |
f3ac8bf7 | 419 | int choose_first; |
1da177e4 LT |
420 | |
421 | rcu_read_lock(); | |
422 | /* | |
8ddf9efe | 423 | * Check if we can balance. We can balance on the whole |
1da177e4 LT |
424 | * device if no resync is going on, or below the resync window. |
425 | * We take the first readable disk when above the resync window. | |
426 | */ | |
427 | retry: | |
428 | if (conf->mddev->recovery_cp < MaxSector && | |
429 | (this_sector + sectors >= conf->next_resync)) { | |
f3ac8bf7 N |
430 | choose_first = 1; |
431 | start_disk = 0; | |
432 | } else { | |
433 | choose_first = 0; | |
434 | start_disk = conf->last_used; | |
1da177e4 LT |
435 | } |
436 | ||
1da177e4 | 437 | /* make sure the disk is operational */ |
f3ac8bf7 N |
438 | for (i = 0 ; i < conf->raid_disks ; i++) { |
439 | int disk = start_disk + i; | |
440 | if (disk >= conf->raid_disks) | |
441 | disk -= conf->raid_disks; | |
442 | ||
443 | rdev = rcu_dereference(conf->mirrors[disk].rdev); | |
444 | if (r1_bio->bios[disk] == IO_BLOCKED | |
445 | || rdev == NULL | |
446 | || !test_bit(In_sync, &rdev->flags)) | |
447 | continue; | |
448 | ||
449 | new_disk = disk; | |
450 | if (!test_bit(WriteMostly, &rdev->flags)) | |
8ddf9efe | 451 | break; |
1da177e4 | 452 | } |
8ddf9efe | 453 | |
f3ac8bf7 | 454 | if (new_disk < 0 || choose_first) |
8ddf9efe N |
455 | goto rb_out; |
456 | ||
1da177e4 LT |
457 | /* |
458 | * Don't change to another disk for sequential reads: | |
459 | */ | |
460 | if (conf->next_seq_sect == this_sector) | |
461 | goto rb_out; | |
462 | if (this_sector == conf->mirrors[new_disk].head_position) | |
463 | goto rb_out; | |
464 | ||
f3ac8bf7 N |
465 | current_distance = abs(this_sector |
466 | - conf->mirrors[new_disk].head_position); | |
1da177e4 | 467 | |
f3ac8bf7 N |
468 | /* look for a better disk - i.e. head is closer */ |
469 | start_disk = new_disk; | |
470 | for (i = 1; i < conf->raid_disks; i++) { | |
471 | int disk = start_disk + 1; | |
472 | if (disk >= conf->raid_disks) | |
473 | disk -= conf->raid_disks; | |
1da177e4 | 474 | |
d6065f7b | 475 | rdev = rcu_dereference(conf->mirrors[disk].rdev); |
f3ac8bf7 N |
476 | if (r1_bio->bios[disk] == IO_BLOCKED |
477 | || rdev == NULL | |
478 | || !test_bit(In_sync, &rdev->flags) | |
479 | || test_bit(WriteMostly, &rdev->flags)) | |
1da177e4 LT |
480 | continue; |
481 | ||
482 | if (!atomic_read(&rdev->nr_pending)) { | |
483 | new_disk = disk; | |
1da177e4 LT |
484 | break; |
485 | } | |
486 | new_distance = abs(this_sector - conf->mirrors[disk].head_position); | |
487 | if (new_distance < current_distance) { | |
488 | current_distance = new_distance; | |
489 | new_disk = disk; | |
1da177e4 | 490 | } |
f3ac8bf7 | 491 | } |
1da177e4 | 492 | |
8ddf9efe | 493 | rb_out: |
1da177e4 | 494 | if (new_disk >= 0) { |
d6065f7b | 495 | rdev = rcu_dereference(conf->mirrors[new_disk].rdev); |
8ddf9efe N |
496 | if (!rdev) |
497 | goto retry; | |
498 | atomic_inc(&rdev->nr_pending); | |
b2d444d7 | 499 | if (!test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
500 | /* cannot risk returning a device that failed |
501 | * before we inc'ed nr_pending | |
502 | */ | |
03c902e1 | 503 | rdev_dec_pending(rdev, conf->mddev); |
1da177e4 LT |
504 | goto retry; |
505 | } | |
8ddf9efe N |
506 | conf->next_seq_sect = this_sector + sectors; |
507 | conf->last_used = new_disk; | |
1da177e4 LT |
508 | } |
509 | rcu_read_unlock(); | |
510 | ||
511 | return new_disk; | |
512 | } | |
513 | ||
0d129228 N |
514 | static int raid1_congested(void *data, int bits) |
515 | { | |
516 | mddev_t *mddev = data; | |
070ec55d | 517 | conf_t *conf = mddev->private; |
0d129228 N |
518 | int i, ret = 0; |
519 | ||
3fa841d7 N |
520 | if (mddev_congested(mddev, bits)) |
521 | return 1; | |
522 | ||
0d129228 N |
523 | rcu_read_lock(); |
524 | for (i = 0; i < mddev->raid_disks; i++) { | |
525 | mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev); | |
526 | if (rdev && !test_bit(Faulty, &rdev->flags)) { | |
165125e1 | 527 | struct request_queue *q = bdev_get_queue(rdev->bdev); |
0d129228 N |
528 | |
529 | /* Note the '|| 1' - when read_balance prefers | |
530 | * non-congested targets, it can be removed | |
531 | */ | |
91a9e99d | 532 | if ((bits & (1<<BDI_async_congested)) || 1) |
0d129228 N |
533 | ret |= bdi_congested(&q->backing_dev_info, bits); |
534 | else | |
535 | ret &= bdi_congested(&q->backing_dev_info, bits); | |
536 | } | |
537 | } | |
538 | rcu_read_unlock(); | |
539 | return ret; | |
540 | } | |
541 | ||
542 | ||
7eaceacc | 543 | static void flush_pending_writes(conf_t *conf) |
a35e63ef N |
544 | { |
545 | /* Any writes that have been queued but are awaiting | |
546 | * bitmap updates get flushed here. | |
a35e63ef | 547 | */ |
a35e63ef N |
548 | spin_lock_irq(&conf->device_lock); |
549 | ||
550 | if (conf->pending_bio_list.head) { | |
551 | struct bio *bio; | |
552 | bio = bio_list_get(&conf->pending_bio_list); | |
a35e63ef N |
553 | spin_unlock_irq(&conf->device_lock); |
554 | /* flush any pending bitmap writes to | |
555 | * disk before proceeding w/ I/O */ | |
556 | bitmap_unplug(conf->mddev->bitmap); | |
557 | ||
558 | while (bio) { /* submit pending writes */ | |
559 | struct bio *next = bio->bi_next; | |
560 | bio->bi_next = NULL; | |
561 | generic_make_request(bio); | |
562 | bio = next; | |
563 | } | |
a35e63ef N |
564 | } else |
565 | spin_unlock_irq(&conf->device_lock); | |
7eaceacc JA |
566 | } |
567 | ||
17999be4 N |
568 | /* Barriers.... |
569 | * Sometimes we need to suspend IO while we do something else, | |
570 | * either some resync/recovery, or reconfigure the array. | |
571 | * To do this we raise a 'barrier'. | |
572 | * The 'barrier' is a counter that can be raised multiple times | |
573 | * to count how many activities are happening which preclude | |
574 | * normal IO. | |
575 | * We can only raise the barrier if there is no pending IO. | |
576 | * i.e. if nr_pending == 0. | |
577 | * We choose only to raise the barrier if no-one is waiting for the | |
578 | * barrier to go down. This means that as soon as an IO request | |
579 | * is ready, no other operations which require a barrier will start | |
580 | * until the IO request has had a chance. | |
581 | * | |
582 | * So: regular IO calls 'wait_barrier'. When that returns there | |
583 | * is no backgroup IO happening, It must arrange to call | |
584 | * allow_barrier when it has finished its IO. | |
585 | * backgroup IO calls must call raise_barrier. Once that returns | |
586 | * there is no normal IO happeing. It must arrange to call | |
587 | * lower_barrier when the particular background IO completes. | |
1da177e4 LT |
588 | */ |
589 | #define RESYNC_DEPTH 32 | |
590 | ||
17999be4 | 591 | static void raise_barrier(conf_t *conf) |
1da177e4 LT |
592 | { |
593 | spin_lock_irq(&conf->resync_lock); | |
17999be4 N |
594 | |
595 | /* Wait until no block IO is waiting */ | |
596 | wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting, | |
c3b328ac | 597 | conf->resync_lock, ); |
17999be4 N |
598 | |
599 | /* block any new IO from starting */ | |
600 | conf->barrier++; | |
601 | ||
046abeed | 602 | /* Now wait for all pending IO to complete */ |
17999be4 N |
603 | wait_event_lock_irq(conf->wait_barrier, |
604 | !conf->nr_pending && conf->barrier < RESYNC_DEPTH, | |
c3b328ac | 605 | conf->resync_lock, ); |
17999be4 N |
606 | |
607 | spin_unlock_irq(&conf->resync_lock); | |
608 | } | |
609 | ||
610 | static void lower_barrier(conf_t *conf) | |
611 | { | |
612 | unsigned long flags; | |
709ae487 | 613 | BUG_ON(conf->barrier <= 0); |
17999be4 N |
614 | spin_lock_irqsave(&conf->resync_lock, flags); |
615 | conf->barrier--; | |
616 | spin_unlock_irqrestore(&conf->resync_lock, flags); | |
617 | wake_up(&conf->wait_barrier); | |
618 | } | |
619 | ||
620 | static void wait_barrier(conf_t *conf) | |
621 | { | |
622 | spin_lock_irq(&conf->resync_lock); | |
623 | if (conf->barrier) { | |
624 | conf->nr_waiting++; | |
625 | wait_event_lock_irq(conf->wait_barrier, !conf->barrier, | |
626 | conf->resync_lock, | |
c3b328ac | 627 | ); |
17999be4 | 628 | conf->nr_waiting--; |
1da177e4 | 629 | } |
17999be4 | 630 | conf->nr_pending++; |
1da177e4 LT |
631 | spin_unlock_irq(&conf->resync_lock); |
632 | } | |
633 | ||
17999be4 N |
634 | static void allow_barrier(conf_t *conf) |
635 | { | |
636 | unsigned long flags; | |
637 | spin_lock_irqsave(&conf->resync_lock, flags); | |
638 | conf->nr_pending--; | |
639 | spin_unlock_irqrestore(&conf->resync_lock, flags); | |
640 | wake_up(&conf->wait_barrier); | |
641 | } | |
642 | ||
ddaf22ab N |
643 | static void freeze_array(conf_t *conf) |
644 | { | |
645 | /* stop syncio and normal IO and wait for everything to | |
646 | * go quite. | |
647 | * We increment barrier and nr_waiting, and then | |
1c830532 N |
648 | * wait until nr_pending match nr_queued+1 |
649 | * This is called in the context of one normal IO request | |
650 | * that has failed. Thus any sync request that might be pending | |
651 | * will be blocked by nr_pending, and we need to wait for | |
652 | * pending IO requests to complete or be queued for re-try. | |
653 | * Thus the number queued (nr_queued) plus this request (1) | |
654 | * must match the number of pending IOs (nr_pending) before | |
655 | * we continue. | |
ddaf22ab N |
656 | */ |
657 | spin_lock_irq(&conf->resync_lock); | |
658 | conf->barrier++; | |
659 | conf->nr_waiting++; | |
660 | wait_event_lock_irq(conf->wait_barrier, | |
1c830532 | 661 | conf->nr_pending == conf->nr_queued+1, |
ddaf22ab | 662 | conf->resync_lock, |
c3b328ac | 663 | flush_pending_writes(conf)); |
ddaf22ab N |
664 | spin_unlock_irq(&conf->resync_lock); |
665 | } | |
666 | static void unfreeze_array(conf_t *conf) | |
667 | { | |
668 | /* reverse the effect of the freeze */ | |
669 | spin_lock_irq(&conf->resync_lock); | |
670 | conf->barrier--; | |
671 | conf->nr_waiting--; | |
672 | wake_up(&conf->wait_barrier); | |
673 | spin_unlock_irq(&conf->resync_lock); | |
674 | } | |
675 | ||
17999be4 | 676 | |
4e78064f N |
677 | /* duplicate the data pages for behind I/O |
678 | * We return a list of bio_vec rather than just page pointers | |
679 | * as it makes freeing easier | |
680 | */ | |
681 | static struct bio_vec *alloc_behind_pages(struct bio *bio) | |
4b6d287f N |
682 | { |
683 | int i; | |
684 | struct bio_vec *bvec; | |
4e78064f | 685 | struct bio_vec *pages = kzalloc(bio->bi_vcnt * sizeof(struct bio_vec), |
4b6d287f N |
686 | GFP_NOIO); |
687 | if (unlikely(!pages)) | |
688 | goto do_sync_io; | |
689 | ||
4b6d287f | 690 | bio_for_each_segment(bvec, bio, i) { |
4e78064f N |
691 | pages[i].bv_page = alloc_page(GFP_NOIO); |
692 | if (unlikely(!pages[i].bv_page)) | |
4b6d287f | 693 | goto do_sync_io; |
4e78064f | 694 | memcpy(kmap(pages[i].bv_page) + bvec->bv_offset, |
4b6d287f | 695 | kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len); |
4e78064f | 696 | kunmap(pages[i].bv_page); |
4b6d287f N |
697 | kunmap(bvec->bv_page); |
698 | } | |
699 | ||
700 | return pages; | |
701 | ||
702 | do_sync_io: | |
703 | if (pages) | |
4e78064f N |
704 | for (i = 0; i < bio->bi_vcnt && pages[i].bv_page; i++) |
705 | put_page(pages[i].bv_page); | |
4b6d287f N |
706 | kfree(pages); |
707 | PRINTK("%dB behind alloc failed, doing sync I/O\n", bio->bi_size); | |
708 | return NULL; | |
709 | } | |
710 | ||
21a52c6d | 711 | static int make_request(mddev_t *mddev, struct bio * bio) |
1da177e4 | 712 | { |
070ec55d | 713 | conf_t *conf = mddev->private; |
1da177e4 LT |
714 | mirror_info_t *mirror; |
715 | r1bio_t *r1_bio; | |
716 | struct bio *read_bio; | |
191ea9b2 | 717 | int i, targets = 0, disks; |
84255d10 | 718 | struct bitmap *bitmap; |
191ea9b2 | 719 | unsigned long flags; |
4e78064f | 720 | struct bio_vec *behind_pages = NULL; |
a362357b | 721 | const int rw = bio_data_dir(bio); |
2c7d46ec | 722 | const unsigned long do_sync = (bio->bi_rw & REQ_SYNC); |
e9c7469b | 723 | const unsigned long do_flush_fua = (bio->bi_rw & (REQ_FLUSH | REQ_FUA)); |
6bfe0b49 | 724 | mdk_rdev_t *blocked_rdev; |
c3b328ac | 725 | int plugged; |
191ea9b2 | 726 | |
1da177e4 LT |
727 | /* |
728 | * Register the new request and wait if the reconstruction | |
729 | * thread has put up a bar for new requests. | |
730 | * Continue immediately if no resync is active currently. | |
731 | */ | |
62de608d | 732 | |
3d310eb7 N |
733 | md_write_start(mddev, bio); /* wait on superblock update early */ |
734 | ||
6eef4b21 N |
735 | if (bio_data_dir(bio) == WRITE && |
736 | bio->bi_sector + bio->bi_size/512 > mddev->suspend_lo && | |
737 | bio->bi_sector < mddev->suspend_hi) { | |
738 | /* As the suspend_* range is controlled by | |
739 | * userspace, we want an interruptible | |
740 | * wait. | |
741 | */ | |
742 | DEFINE_WAIT(w); | |
743 | for (;;) { | |
744 | flush_signals(current); | |
745 | prepare_to_wait(&conf->wait_barrier, | |
746 | &w, TASK_INTERRUPTIBLE); | |
747 | if (bio->bi_sector + bio->bi_size/512 <= mddev->suspend_lo || | |
748 | bio->bi_sector >= mddev->suspend_hi) | |
749 | break; | |
750 | schedule(); | |
751 | } | |
752 | finish_wait(&conf->wait_barrier, &w); | |
753 | } | |
62de608d | 754 | |
17999be4 | 755 | wait_barrier(conf); |
1da177e4 | 756 | |
84255d10 N |
757 | bitmap = mddev->bitmap; |
758 | ||
1da177e4 LT |
759 | /* |
760 | * make_request() can abort the operation when READA is being | |
761 | * used and no empty request is available. | |
762 | * | |
763 | */ | |
764 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
765 | ||
766 | r1_bio->master_bio = bio; | |
767 | r1_bio->sectors = bio->bi_size >> 9; | |
191ea9b2 | 768 | r1_bio->state = 0; |
1da177e4 LT |
769 | r1_bio->mddev = mddev; |
770 | r1_bio->sector = bio->bi_sector; | |
771 | ||
a362357b | 772 | if (rw == READ) { |
1da177e4 LT |
773 | /* |
774 | * read balancing logic: | |
775 | */ | |
776 | int rdisk = read_balance(conf, r1_bio); | |
777 | ||
778 | if (rdisk < 0) { | |
779 | /* couldn't find anywhere to read from */ | |
780 | raid_end_bio_io(r1_bio); | |
781 | return 0; | |
782 | } | |
783 | mirror = conf->mirrors + rdisk; | |
784 | ||
e555190d N |
785 | if (test_bit(WriteMostly, &mirror->rdev->flags) && |
786 | bitmap) { | |
787 | /* Reading from a write-mostly device must | |
788 | * take care not to over-take any writes | |
789 | * that are 'behind' | |
790 | */ | |
791 | wait_event(bitmap->behind_wait, | |
792 | atomic_read(&bitmap->behind_writes) == 0); | |
793 | } | |
1da177e4 LT |
794 | r1_bio->read_disk = rdisk; |
795 | ||
a167f663 | 796 | read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev); |
1da177e4 LT |
797 | |
798 | r1_bio->bios[rdisk] = read_bio; | |
799 | ||
800 | read_bio->bi_sector = r1_bio->sector + mirror->rdev->data_offset; | |
801 | read_bio->bi_bdev = mirror->rdev->bdev; | |
802 | read_bio->bi_end_io = raid1_end_read_request; | |
7b6d91da | 803 | read_bio->bi_rw = READ | do_sync; |
1da177e4 LT |
804 | read_bio->bi_private = r1_bio; |
805 | ||
806 | generic_make_request(read_bio); | |
807 | return 0; | |
808 | } | |
809 | ||
810 | /* | |
811 | * WRITE: | |
812 | */ | |
813 | /* first select target devices under spinlock and | |
814 | * inc refcount on their rdev. Record them by setting | |
815 | * bios[x] to bio | |
816 | */ | |
c3b328ac N |
817 | plugged = mddev_check_plugged(mddev); |
818 | ||
1da177e4 | 819 | disks = conf->raid_disks; |
6bfe0b49 DW |
820 | retry_write: |
821 | blocked_rdev = NULL; | |
1da177e4 LT |
822 | rcu_read_lock(); |
823 | for (i = 0; i < disks; i++) { | |
6bfe0b49 DW |
824 | mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev); |
825 | if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { | |
826 | atomic_inc(&rdev->nr_pending); | |
827 | blocked_rdev = rdev; | |
828 | break; | |
829 | } | |
830 | if (rdev && !test_bit(Faulty, &rdev->flags)) { | |
1da177e4 | 831 | atomic_inc(&rdev->nr_pending); |
b2d444d7 | 832 | if (test_bit(Faulty, &rdev->flags)) { |
03c902e1 | 833 | rdev_dec_pending(rdev, mddev); |
1da177e4 | 834 | r1_bio->bios[i] = NULL; |
964147d5 | 835 | } else { |
1da177e4 | 836 | r1_bio->bios[i] = bio; |
964147d5 N |
837 | targets++; |
838 | } | |
1da177e4 LT |
839 | } else |
840 | r1_bio->bios[i] = NULL; | |
841 | } | |
842 | rcu_read_unlock(); | |
843 | ||
6bfe0b49 DW |
844 | if (unlikely(blocked_rdev)) { |
845 | /* Wait for this device to become unblocked */ | |
846 | int j; | |
847 | ||
848 | for (j = 0; j < i; j++) | |
849 | if (r1_bio->bios[j]) | |
850 | rdev_dec_pending(conf->mirrors[j].rdev, mddev); | |
851 | ||
852 | allow_barrier(conf); | |
853 | md_wait_for_blocked_rdev(blocked_rdev, mddev); | |
854 | wait_barrier(conf); | |
855 | goto retry_write; | |
856 | } | |
857 | ||
4b6d287f N |
858 | BUG_ON(targets == 0); /* we never fail the last device */ |
859 | ||
191ea9b2 N |
860 | if (targets < conf->raid_disks) { |
861 | /* array is degraded, we will not clear the bitmap | |
862 | * on I/O completion (see raid1_end_write_request) */ | |
863 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
864 | } | |
865 | ||
e555190d N |
866 | /* do behind I/O ? |
867 | * Not if there are too many, or cannot allocate memory, | |
868 | * or a reader on WriteMostly is waiting for behind writes | |
869 | * to flush */ | |
4b6d287f | 870 | if (bitmap && |
42a04b50 N |
871 | (atomic_read(&bitmap->behind_writes) |
872 | < mddev->bitmap_info.max_write_behind) && | |
e555190d | 873 | !waitqueue_active(&bitmap->behind_wait) && |
4b6d287f N |
874 | (behind_pages = alloc_behind_pages(bio)) != NULL) |
875 | set_bit(R1BIO_BehindIO, &r1_bio->state); | |
876 | ||
4e78064f | 877 | atomic_set(&r1_bio->remaining, 1); |
4b6d287f | 878 | atomic_set(&r1_bio->behind_remaining, 0); |
06d91a5f | 879 | |
4e78064f N |
880 | bitmap_startwrite(bitmap, bio->bi_sector, r1_bio->sectors, |
881 | test_bit(R1BIO_BehindIO, &r1_bio->state)); | |
1da177e4 LT |
882 | for (i = 0; i < disks; i++) { |
883 | struct bio *mbio; | |
884 | if (!r1_bio->bios[i]) | |
885 | continue; | |
886 | ||
a167f663 | 887 | mbio = bio_clone_mddev(bio, GFP_NOIO, mddev); |
1da177e4 LT |
888 | r1_bio->bios[i] = mbio; |
889 | ||
890 | mbio->bi_sector = r1_bio->sector + conf->mirrors[i].rdev->data_offset; | |
891 | mbio->bi_bdev = conf->mirrors[i].rdev->bdev; | |
892 | mbio->bi_end_io = raid1_end_write_request; | |
e9c7469b | 893 | mbio->bi_rw = WRITE | do_flush_fua | do_sync; |
1da177e4 LT |
894 | mbio->bi_private = r1_bio; |
895 | ||
4b6d287f N |
896 | if (behind_pages) { |
897 | struct bio_vec *bvec; | |
898 | int j; | |
899 | ||
900 | /* Yes, I really want the '__' version so that | |
901 | * we clear any unused pointer in the io_vec, rather | |
902 | * than leave them unchanged. This is important | |
903 | * because when we come to free the pages, we won't | |
046abeed | 904 | * know the original bi_idx, so we just free |
4b6d287f N |
905 | * them all |
906 | */ | |
907 | __bio_for_each_segment(bvec, mbio, j, 0) | |
4e78064f | 908 | bvec->bv_page = behind_pages[j].bv_page; |
4b6d287f N |
909 | if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags)) |
910 | atomic_inc(&r1_bio->behind_remaining); | |
911 | } | |
912 | ||
1da177e4 | 913 | atomic_inc(&r1_bio->remaining); |
4e78064f N |
914 | spin_lock_irqsave(&conf->device_lock, flags); |
915 | bio_list_add(&conf->pending_bio_list, mbio); | |
4e78064f | 916 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 917 | } |
4e78064f | 918 | r1_bio_write_done(r1_bio, bio->bi_vcnt, behind_pages, behind_pages != NULL); |
4b6d287f | 919 | kfree(behind_pages); /* the behind pages are attached to the bios now */ |
1da177e4 | 920 | |
4e78064f | 921 | /* In case raid1d snuck in to freeze_array */ |
a35e63ef N |
922 | wake_up(&conf->wait_barrier); |
923 | ||
c3b328ac | 924 | if (do_sync || !bitmap || !plugged) |
e3881a68 | 925 | md_wakeup_thread(mddev->thread); |
191ea9b2 | 926 | |
1da177e4 LT |
927 | return 0; |
928 | } | |
929 | ||
930 | static void status(struct seq_file *seq, mddev_t *mddev) | |
931 | { | |
070ec55d | 932 | conf_t *conf = mddev->private; |
1da177e4 LT |
933 | int i; |
934 | ||
935 | seq_printf(seq, " [%d/%d] [", conf->raid_disks, | |
11ce99e6 | 936 | conf->raid_disks - mddev->degraded); |
ddac7c7e N |
937 | rcu_read_lock(); |
938 | for (i = 0; i < conf->raid_disks; i++) { | |
939 | mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev); | |
1da177e4 | 940 | seq_printf(seq, "%s", |
ddac7c7e N |
941 | rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); |
942 | } | |
943 | rcu_read_unlock(); | |
1da177e4 LT |
944 | seq_printf(seq, "]"); |
945 | } | |
946 | ||
947 | ||
948 | static void error(mddev_t *mddev, mdk_rdev_t *rdev) | |
949 | { | |
950 | char b[BDEVNAME_SIZE]; | |
070ec55d | 951 | conf_t *conf = mddev->private; |
1da177e4 LT |
952 | |
953 | /* | |
954 | * If it is not operational, then we have already marked it as dead | |
955 | * else if it is the last working disks, ignore the error, let the | |
956 | * next level up know. | |
957 | * else mark the drive as failed | |
958 | */ | |
b2d444d7 | 959 | if (test_bit(In_sync, &rdev->flags) |
4044ba58 | 960 | && (conf->raid_disks - mddev->degraded) == 1) { |
1da177e4 LT |
961 | /* |
962 | * Don't fail the drive, act as though we were just a | |
4044ba58 N |
963 | * normal single drive. |
964 | * However don't try a recovery from this drive as | |
965 | * it is very likely to fail. | |
1da177e4 | 966 | */ |
4044ba58 | 967 | mddev->recovery_disabled = 1; |
1da177e4 | 968 | return; |
4044ba58 | 969 | } |
c04be0aa N |
970 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
971 | unsigned long flags; | |
972 | spin_lock_irqsave(&conf->device_lock, flags); | |
1da177e4 | 973 | mddev->degraded++; |
dd00a99e | 974 | set_bit(Faulty, &rdev->flags); |
c04be0aa | 975 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 LT |
976 | /* |
977 | * if recovery is running, make sure it aborts. | |
978 | */ | |
dfc70645 | 979 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
dd00a99e N |
980 | } else |
981 | set_bit(Faulty, &rdev->flags); | |
850b2b42 | 982 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
067032bc JP |
983 | printk(KERN_ALERT |
984 | "md/raid1:%s: Disk failure on %s, disabling device.\n" | |
985 | "md/raid1:%s: Operation continuing on %d devices.\n", | |
9dd1e2fa N |
986 | mdname(mddev), bdevname(rdev->bdev, b), |
987 | mdname(mddev), conf->raid_disks - mddev->degraded); | |
1da177e4 LT |
988 | } |
989 | ||
990 | static void print_conf(conf_t *conf) | |
991 | { | |
992 | int i; | |
1da177e4 | 993 | |
9dd1e2fa | 994 | printk(KERN_DEBUG "RAID1 conf printout:\n"); |
1da177e4 | 995 | if (!conf) { |
9dd1e2fa | 996 | printk(KERN_DEBUG "(!conf)\n"); |
1da177e4 LT |
997 | return; |
998 | } | |
9dd1e2fa | 999 | printk(KERN_DEBUG " --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded, |
1da177e4 LT |
1000 | conf->raid_disks); |
1001 | ||
ddac7c7e | 1002 | rcu_read_lock(); |
1da177e4 LT |
1003 | for (i = 0; i < conf->raid_disks; i++) { |
1004 | char b[BDEVNAME_SIZE]; | |
ddac7c7e N |
1005 | mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev); |
1006 | if (rdev) | |
9dd1e2fa | 1007 | printk(KERN_DEBUG " disk %d, wo:%d, o:%d, dev:%s\n", |
ddac7c7e N |
1008 | i, !test_bit(In_sync, &rdev->flags), |
1009 | !test_bit(Faulty, &rdev->flags), | |
1010 | bdevname(rdev->bdev,b)); | |
1da177e4 | 1011 | } |
ddac7c7e | 1012 | rcu_read_unlock(); |
1da177e4 LT |
1013 | } |
1014 | ||
1015 | static void close_sync(conf_t *conf) | |
1016 | { | |
17999be4 N |
1017 | wait_barrier(conf); |
1018 | allow_barrier(conf); | |
1da177e4 LT |
1019 | |
1020 | mempool_destroy(conf->r1buf_pool); | |
1021 | conf->r1buf_pool = NULL; | |
1022 | } | |
1023 | ||
1024 | static int raid1_spare_active(mddev_t *mddev) | |
1025 | { | |
1026 | int i; | |
1027 | conf_t *conf = mddev->private; | |
6b965620 N |
1028 | int count = 0; |
1029 | unsigned long flags; | |
1da177e4 LT |
1030 | |
1031 | /* | |
1032 | * Find all failed disks within the RAID1 configuration | |
ddac7c7e N |
1033 | * and mark them readable. |
1034 | * Called under mddev lock, so rcu protection not needed. | |
1da177e4 LT |
1035 | */ |
1036 | for (i = 0; i < conf->raid_disks; i++) { | |
ddac7c7e N |
1037 | mdk_rdev_t *rdev = conf->mirrors[i].rdev; |
1038 | if (rdev | |
1039 | && !test_bit(Faulty, &rdev->flags) | |
c04be0aa | 1040 | && !test_and_set_bit(In_sync, &rdev->flags)) { |
6b965620 | 1041 | count++; |
e6ffbcb6 | 1042 | sysfs_notify_dirent(rdev->sysfs_state); |
1da177e4 LT |
1043 | } |
1044 | } | |
6b965620 N |
1045 | spin_lock_irqsave(&conf->device_lock, flags); |
1046 | mddev->degraded -= count; | |
1047 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 LT |
1048 | |
1049 | print_conf(conf); | |
6b965620 | 1050 | return count; |
1da177e4 LT |
1051 | } |
1052 | ||
1053 | ||
1054 | static int raid1_add_disk(mddev_t *mddev, mdk_rdev_t *rdev) | |
1055 | { | |
1056 | conf_t *conf = mddev->private; | |
199050ea | 1057 | int err = -EEXIST; |
41158c7e | 1058 | int mirror = 0; |
1da177e4 | 1059 | mirror_info_t *p; |
6c2fce2e NB |
1060 | int first = 0; |
1061 | int last = mddev->raid_disks - 1; | |
1da177e4 | 1062 | |
6c2fce2e NB |
1063 | if (rdev->raid_disk >= 0) |
1064 | first = last = rdev->raid_disk; | |
1065 | ||
1066 | for (mirror = first; mirror <= last; mirror++) | |
1da177e4 LT |
1067 | if ( !(p=conf->mirrors+mirror)->rdev) { |
1068 | ||
8f6c2e4b MP |
1069 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
1070 | rdev->data_offset << 9); | |
627a2d3c N |
1071 | /* as we don't honour merge_bvec_fn, we must |
1072 | * never risk violating it, so limit | |
1073 | * ->max_segments to one lying with a single | |
1074 | * page, as a one page request is never in | |
1075 | * violation. | |
1da177e4 | 1076 | */ |
627a2d3c N |
1077 | if (rdev->bdev->bd_disk->queue->merge_bvec_fn) { |
1078 | blk_queue_max_segments(mddev->queue, 1); | |
1079 | blk_queue_segment_boundary(mddev->queue, | |
1080 | PAGE_CACHE_SIZE - 1); | |
1081 | } | |
1da177e4 LT |
1082 | |
1083 | p->head_position = 0; | |
1084 | rdev->raid_disk = mirror; | |
199050ea | 1085 | err = 0; |
6aea114a N |
1086 | /* As all devices are equivalent, we don't need a full recovery |
1087 | * if this was recently any drive of the array | |
1088 | */ | |
1089 | if (rdev->saved_raid_disk < 0) | |
41158c7e | 1090 | conf->fullsync = 1; |
d6065f7b | 1091 | rcu_assign_pointer(p->rdev, rdev); |
1da177e4 LT |
1092 | break; |
1093 | } | |
ac5e7113 | 1094 | md_integrity_add_rdev(rdev, mddev); |
1da177e4 | 1095 | print_conf(conf); |
199050ea | 1096 | return err; |
1da177e4 LT |
1097 | } |
1098 | ||
1099 | static int raid1_remove_disk(mddev_t *mddev, int number) | |
1100 | { | |
1101 | conf_t *conf = mddev->private; | |
1102 | int err = 0; | |
1103 | mdk_rdev_t *rdev; | |
1104 | mirror_info_t *p = conf->mirrors+ number; | |
1105 | ||
1106 | print_conf(conf); | |
1107 | rdev = p->rdev; | |
1108 | if (rdev) { | |
b2d444d7 | 1109 | if (test_bit(In_sync, &rdev->flags) || |
1da177e4 LT |
1110 | atomic_read(&rdev->nr_pending)) { |
1111 | err = -EBUSY; | |
1112 | goto abort; | |
1113 | } | |
046abeed | 1114 | /* Only remove non-faulty devices if recovery |
dfc70645 N |
1115 | * is not possible. |
1116 | */ | |
1117 | if (!test_bit(Faulty, &rdev->flags) && | |
8f9e0ee3 | 1118 | !mddev->recovery_disabled && |
dfc70645 N |
1119 | mddev->degraded < conf->raid_disks) { |
1120 | err = -EBUSY; | |
1121 | goto abort; | |
1122 | } | |
1da177e4 | 1123 | p->rdev = NULL; |
fbd568a3 | 1124 | synchronize_rcu(); |
1da177e4 LT |
1125 | if (atomic_read(&rdev->nr_pending)) { |
1126 | /* lost the race, try later */ | |
1127 | err = -EBUSY; | |
1128 | p->rdev = rdev; | |
ac5e7113 | 1129 | goto abort; |
1da177e4 | 1130 | } |
a91a2785 | 1131 | err = md_integrity_register(mddev); |
1da177e4 LT |
1132 | } |
1133 | abort: | |
1134 | ||
1135 | print_conf(conf); | |
1136 | return err; | |
1137 | } | |
1138 | ||
1139 | ||
6712ecf8 | 1140 | static void end_sync_read(struct bio *bio, int error) |
1da177e4 | 1141 | { |
7b92813c | 1142 | r1bio_t *r1_bio = bio->bi_private; |
d11c171e | 1143 | int i; |
1da177e4 | 1144 | |
d11c171e N |
1145 | for (i=r1_bio->mddev->raid_disks; i--; ) |
1146 | if (r1_bio->bios[i] == bio) | |
1147 | break; | |
1148 | BUG_ON(i < 0); | |
1149 | update_head_pos(i, r1_bio); | |
1da177e4 LT |
1150 | /* |
1151 | * we have read a block, now it needs to be re-written, | |
1152 | * or re-read if the read failed. | |
1153 | * We don't do much here, just schedule handling by raid1d | |
1154 | */ | |
69382e85 | 1155 | if (test_bit(BIO_UPTODATE, &bio->bi_flags)) |
1da177e4 | 1156 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
d11c171e N |
1157 | |
1158 | if (atomic_dec_and_test(&r1_bio->remaining)) | |
1159 | reschedule_retry(r1_bio); | |
1da177e4 LT |
1160 | } |
1161 | ||
6712ecf8 | 1162 | static void end_sync_write(struct bio *bio, int error) |
1da177e4 LT |
1163 | { |
1164 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
7b92813c | 1165 | r1bio_t *r1_bio = bio->bi_private; |
1da177e4 | 1166 | mddev_t *mddev = r1_bio->mddev; |
070ec55d | 1167 | conf_t *conf = mddev->private; |
1da177e4 LT |
1168 | int i; |
1169 | int mirror=0; | |
1170 | ||
1da177e4 LT |
1171 | for (i = 0; i < conf->raid_disks; i++) |
1172 | if (r1_bio->bios[i] == bio) { | |
1173 | mirror = i; | |
1174 | break; | |
1175 | } | |
6b1117d5 | 1176 | if (!uptodate) { |
57dab0bd | 1177 | sector_t sync_blocks = 0; |
6b1117d5 N |
1178 | sector_t s = r1_bio->sector; |
1179 | long sectors_to_go = r1_bio->sectors; | |
1180 | /* make sure these bits doesn't get cleared. */ | |
1181 | do { | |
5e3db645 | 1182 | bitmap_end_sync(mddev->bitmap, s, |
6b1117d5 N |
1183 | &sync_blocks, 1); |
1184 | s += sync_blocks; | |
1185 | sectors_to_go -= sync_blocks; | |
1186 | } while (sectors_to_go > 0); | |
1da177e4 | 1187 | md_error(mddev, conf->mirrors[mirror].rdev); |
6b1117d5 | 1188 | } |
e3b9703e | 1189 | |
1da177e4 LT |
1190 | update_head_pos(mirror, r1_bio); |
1191 | ||
1192 | if (atomic_dec_and_test(&r1_bio->remaining)) { | |
73d5c38a | 1193 | sector_t s = r1_bio->sectors; |
1da177e4 | 1194 | put_buf(r1_bio); |
73d5c38a | 1195 | md_done_sync(mddev, s, uptodate); |
1da177e4 | 1196 | } |
1da177e4 LT |
1197 | } |
1198 | ||
1199 | static void sync_request_write(mddev_t *mddev, r1bio_t *r1_bio) | |
1200 | { | |
070ec55d | 1201 | conf_t *conf = mddev->private; |
1da177e4 LT |
1202 | int i; |
1203 | int disks = conf->raid_disks; | |
1204 | struct bio *bio, *wbio; | |
1205 | ||
1206 | bio = r1_bio->bios[r1_bio->read_disk]; | |
1207 | ||
69382e85 | 1208 | |
d11c171e N |
1209 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { |
1210 | /* We have read all readable devices. If we haven't | |
1211 | * got the block, then there is no hope left. | |
1212 | * If we have, then we want to do a comparison | |
1213 | * and skip the write if everything is the same. | |
1214 | * If any blocks failed to read, then we need to | |
1215 | * attempt an over-write | |
1216 | */ | |
1217 | int primary; | |
1218 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) { | |
1219 | for (i=0; i<mddev->raid_disks; i++) | |
1220 | if (r1_bio->bios[i]->bi_end_io == end_sync_read) | |
1221 | md_error(mddev, conf->mirrors[i].rdev); | |
1222 | ||
1223 | md_done_sync(mddev, r1_bio->sectors, 1); | |
1224 | put_buf(r1_bio); | |
1225 | return; | |
1226 | } | |
1227 | for (primary=0; primary<mddev->raid_disks; primary++) | |
1228 | if (r1_bio->bios[primary]->bi_end_io == end_sync_read && | |
1229 | test_bit(BIO_UPTODATE, &r1_bio->bios[primary]->bi_flags)) { | |
1230 | r1_bio->bios[primary]->bi_end_io = NULL; | |
03c902e1 | 1231 | rdev_dec_pending(conf->mirrors[primary].rdev, mddev); |
d11c171e N |
1232 | break; |
1233 | } | |
1234 | r1_bio->read_disk = primary; | |
1235 | for (i=0; i<mddev->raid_disks; i++) | |
ed456662 | 1236 | if (r1_bio->bios[i]->bi_end_io == end_sync_read) { |
d11c171e N |
1237 | int j; |
1238 | int vcnt = r1_bio->sectors >> (PAGE_SHIFT- 9); | |
1239 | struct bio *pbio = r1_bio->bios[primary]; | |
1240 | struct bio *sbio = r1_bio->bios[i]; | |
ed456662 MA |
1241 | |
1242 | if (test_bit(BIO_UPTODATE, &sbio->bi_flags)) { | |
1243 | for (j = vcnt; j-- ; ) { | |
1244 | struct page *p, *s; | |
1245 | p = pbio->bi_io_vec[j].bv_page; | |
1246 | s = sbio->bi_io_vec[j].bv_page; | |
1247 | if (memcmp(page_address(p), | |
1248 | page_address(s), | |
1249 | PAGE_SIZE)) | |
1250 | break; | |
1251 | } | |
1252 | } else | |
1253 | j = 0; | |
d11c171e N |
1254 | if (j >= 0) |
1255 | mddev->resync_mismatches += r1_bio->sectors; | |
cf7a4416 N |
1256 | if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery) |
1257 | && test_bit(BIO_UPTODATE, &sbio->bi_flags))) { | |
d11c171e | 1258 | sbio->bi_end_io = NULL; |
03c902e1 N |
1259 | rdev_dec_pending(conf->mirrors[i].rdev, mddev); |
1260 | } else { | |
d11c171e | 1261 | /* fixup the bio for reuse */ |
698b18c1 | 1262 | int size; |
d11c171e N |
1263 | sbio->bi_vcnt = vcnt; |
1264 | sbio->bi_size = r1_bio->sectors << 9; | |
1265 | sbio->bi_idx = 0; | |
1266 | sbio->bi_phys_segments = 0; | |
d11c171e N |
1267 | sbio->bi_flags &= ~(BIO_POOL_MASK - 1); |
1268 | sbio->bi_flags |= 1 << BIO_UPTODATE; | |
1269 | sbio->bi_next = NULL; | |
1270 | sbio->bi_sector = r1_bio->sector + | |
1271 | conf->mirrors[i].rdev->data_offset; | |
1272 | sbio->bi_bdev = conf->mirrors[i].rdev->bdev; | |
698b18c1 N |
1273 | size = sbio->bi_size; |
1274 | for (j = 0; j < vcnt ; j++) { | |
1275 | struct bio_vec *bi; | |
1276 | bi = &sbio->bi_io_vec[j]; | |
1277 | bi->bv_offset = 0; | |
1278 | if (size > PAGE_SIZE) | |
1279 | bi->bv_len = PAGE_SIZE; | |
1280 | else | |
1281 | bi->bv_len = size; | |
1282 | size -= PAGE_SIZE; | |
1283 | memcpy(page_address(bi->bv_page), | |
3eda22d1 N |
1284 | page_address(pbio->bi_io_vec[j].bv_page), |
1285 | PAGE_SIZE); | |
698b18c1 | 1286 | } |
3eda22d1 | 1287 | |
d11c171e N |
1288 | } |
1289 | } | |
1290 | } | |
1da177e4 | 1291 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) { |
69382e85 N |
1292 | /* ouch - failed to read all of that. |
1293 | * Try some synchronous reads of other devices to get | |
1294 | * good data, much like with normal read errors. Only | |
ddac7c7e | 1295 | * read into the pages we already have so we don't |
69382e85 N |
1296 | * need to re-issue the read request. |
1297 | * We don't need to freeze the array, because being in an | |
1298 | * active sync request, there is no normal IO, and | |
1299 | * no overlapping syncs. | |
1da177e4 | 1300 | */ |
69382e85 N |
1301 | sector_t sect = r1_bio->sector; |
1302 | int sectors = r1_bio->sectors; | |
1303 | int idx = 0; | |
1304 | ||
1305 | while(sectors) { | |
1306 | int s = sectors; | |
1307 | int d = r1_bio->read_disk; | |
1308 | int success = 0; | |
1309 | mdk_rdev_t *rdev; | |
1310 | ||
1311 | if (s > (PAGE_SIZE>>9)) | |
1312 | s = PAGE_SIZE >> 9; | |
1313 | do { | |
1314 | if (r1_bio->bios[d]->bi_end_io == end_sync_read) { | |
ddac7c7e N |
1315 | /* No rcu protection needed here devices |
1316 | * can only be removed when no resync is | |
1317 | * active, and resync is currently active | |
1318 | */ | |
69382e85 | 1319 | rdev = conf->mirrors[d].rdev; |
2b193363 | 1320 | if (sync_page_io(rdev, |
ccebd4c4 | 1321 | sect, |
69382e85 N |
1322 | s<<9, |
1323 | bio->bi_io_vec[idx].bv_page, | |
ccebd4c4 | 1324 | READ, false)) { |
69382e85 N |
1325 | success = 1; |
1326 | break; | |
1327 | } | |
1328 | } | |
1329 | d++; | |
1330 | if (d == conf->raid_disks) | |
1331 | d = 0; | |
1332 | } while (!success && d != r1_bio->read_disk); | |
1333 | ||
1334 | if (success) { | |
097426f6 | 1335 | int start = d; |
69382e85 N |
1336 | /* write it back and re-read */ |
1337 | set_bit(R1BIO_Uptodate, &r1_bio->state); | |
1338 | while (d != r1_bio->read_disk) { | |
1339 | if (d == 0) | |
1340 | d = conf->raid_disks; | |
1341 | d--; | |
1342 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) | |
1343 | continue; | |
1344 | rdev = conf->mirrors[d].rdev; | |
4dbcdc75 | 1345 | atomic_add(s, &rdev->corrected_errors); |
2b193363 | 1346 | if (sync_page_io(rdev, |
ccebd4c4 | 1347 | sect, |
69382e85 N |
1348 | s<<9, |
1349 | bio->bi_io_vec[idx].bv_page, | |
ccebd4c4 | 1350 | WRITE, false) == 0) |
097426f6 N |
1351 | md_error(mddev, rdev); |
1352 | } | |
1353 | d = start; | |
1354 | while (d != r1_bio->read_disk) { | |
1355 | if (d == 0) | |
1356 | d = conf->raid_disks; | |
1357 | d--; | |
1358 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) | |
1359 | continue; | |
1360 | rdev = conf->mirrors[d].rdev; | |
2b193363 | 1361 | if (sync_page_io(rdev, |
ccebd4c4 | 1362 | sect, |
69382e85 N |
1363 | s<<9, |
1364 | bio->bi_io_vec[idx].bv_page, | |
ccebd4c4 | 1365 | READ, false) == 0) |
69382e85 | 1366 | md_error(mddev, rdev); |
69382e85 N |
1367 | } |
1368 | } else { | |
1369 | char b[BDEVNAME_SIZE]; | |
1370 | /* Cannot read from anywhere, array is toast */ | |
1371 | md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev); | |
9dd1e2fa | 1372 | printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O read error" |
69382e85 | 1373 | " for block %llu\n", |
9dd1e2fa N |
1374 | mdname(mddev), |
1375 | bdevname(bio->bi_bdev, b), | |
69382e85 N |
1376 | (unsigned long long)r1_bio->sector); |
1377 | md_done_sync(mddev, r1_bio->sectors, 0); | |
1378 | put_buf(r1_bio); | |
1379 | return; | |
1380 | } | |
1381 | sectors -= s; | |
1382 | sect += s; | |
1383 | idx ++; | |
1384 | } | |
1da177e4 | 1385 | } |
d11c171e N |
1386 | |
1387 | /* | |
1388 | * schedule writes | |
1389 | */ | |
1da177e4 LT |
1390 | atomic_set(&r1_bio->remaining, 1); |
1391 | for (i = 0; i < disks ; i++) { | |
1392 | wbio = r1_bio->bios[i]; | |
3e198f78 N |
1393 | if (wbio->bi_end_io == NULL || |
1394 | (wbio->bi_end_io == end_sync_read && | |
1395 | (i == r1_bio->read_disk || | |
1396 | !test_bit(MD_RECOVERY_SYNC, &mddev->recovery)))) | |
1da177e4 LT |
1397 | continue; |
1398 | ||
3e198f78 N |
1399 | wbio->bi_rw = WRITE; |
1400 | wbio->bi_end_io = end_sync_write; | |
1da177e4 LT |
1401 | atomic_inc(&r1_bio->remaining); |
1402 | md_sync_acct(conf->mirrors[i].rdev->bdev, wbio->bi_size >> 9); | |
191ea9b2 | 1403 | |
1da177e4 LT |
1404 | generic_make_request(wbio); |
1405 | } | |
1406 | ||
1407 | if (atomic_dec_and_test(&r1_bio->remaining)) { | |
191ea9b2 | 1408 | /* if we're here, all write(s) have completed, so clean up */ |
1da177e4 LT |
1409 | md_done_sync(mddev, r1_bio->sectors, 1); |
1410 | put_buf(r1_bio); | |
1411 | } | |
1412 | } | |
1413 | ||
1414 | /* | |
1415 | * This is a kernel thread which: | |
1416 | * | |
1417 | * 1. Retries failed read operations on working mirrors. | |
1418 | * 2. Updates the raid superblock when problems encounter. | |
1419 | * 3. Performs writes following reads for array syncronising. | |
1420 | */ | |
1421 | ||
867868fb N |
1422 | static void fix_read_error(conf_t *conf, int read_disk, |
1423 | sector_t sect, int sectors) | |
1424 | { | |
1425 | mddev_t *mddev = conf->mddev; | |
1426 | while(sectors) { | |
1427 | int s = sectors; | |
1428 | int d = read_disk; | |
1429 | int success = 0; | |
1430 | int start; | |
1431 | mdk_rdev_t *rdev; | |
1432 | ||
1433 | if (s > (PAGE_SIZE>>9)) | |
1434 | s = PAGE_SIZE >> 9; | |
1435 | ||
1436 | do { | |
1437 | /* Note: no rcu protection needed here | |
1438 | * as this is synchronous in the raid1d thread | |
1439 | * which is the thread that might remove | |
1440 | * a device. If raid1d ever becomes multi-threaded.... | |
1441 | */ | |
1442 | rdev = conf->mirrors[d].rdev; | |
1443 | if (rdev && | |
1444 | test_bit(In_sync, &rdev->flags) && | |
ccebd4c4 JB |
1445 | sync_page_io(rdev, sect, s<<9, |
1446 | conf->tmppage, READ, false)) | |
867868fb N |
1447 | success = 1; |
1448 | else { | |
1449 | d++; | |
1450 | if (d == conf->raid_disks) | |
1451 | d = 0; | |
1452 | } | |
1453 | } while (!success && d != read_disk); | |
1454 | ||
1455 | if (!success) { | |
1456 | /* Cannot read from anywhere -- bye bye array */ | |
1457 | md_error(mddev, conf->mirrors[read_disk].rdev); | |
1458 | break; | |
1459 | } | |
1460 | /* write it back and re-read */ | |
1461 | start = d; | |
1462 | while (d != read_disk) { | |
1463 | if (d==0) | |
1464 | d = conf->raid_disks; | |
1465 | d--; | |
1466 | rdev = conf->mirrors[d].rdev; | |
1467 | if (rdev && | |
1468 | test_bit(In_sync, &rdev->flags)) { | |
ccebd4c4 JB |
1469 | if (sync_page_io(rdev, sect, s<<9, |
1470 | conf->tmppage, WRITE, false) | |
867868fb N |
1471 | == 0) |
1472 | /* Well, this device is dead */ | |
1473 | md_error(mddev, rdev); | |
1474 | } | |
1475 | } | |
1476 | d = start; | |
1477 | while (d != read_disk) { | |
1478 | char b[BDEVNAME_SIZE]; | |
1479 | if (d==0) | |
1480 | d = conf->raid_disks; | |
1481 | d--; | |
1482 | rdev = conf->mirrors[d].rdev; | |
1483 | if (rdev && | |
1484 | test_bit(In_sync, &rdev->flags)) { | |
ccebd4c4 JB |
1485 | if (sync_page_io(rdev, sect, s<<9, |
1486 | conf->tmppage, READ, false) | |
867868fb N |
1487 | == 0) |
1488 | /* Well, this device is dead */ | |
1489 | md_error(mddev, rdev); | |
1490 | else { | |
1491 | atomic_add(s, &rdev->corrected_errors); | |
1492 | printk(KERN_INFO | |
9dd1e2fa | 1493 | "md/raid1:%s: read error corrected " |
867868fb N |
1494 | "(%d sectors at %llu on %s)\n", |
1495 | mdname(mddev), s, | |
969b755a RD |
1496 | (unsigned long long)(sect + |
1497 | rdev->data_offset), | |
867868fb N |
1498 | bdevname(rdev->bdev, b)); |
1499 | } | |
1500 | } | |
1501 | } | |
1502 | sectors -= s; | |
1503 | sect += s; | |
1504 | } | |
1505 | } | |
1506 | ||
1da177e4 LT |
1507 | static void raid1d(mddev_t *mddev) |
1508 | { | |
1509 | r1bio_t *r1_bio; | |
1510 | struct bio *bio; | |
1511 | unsigned long flags; | |
070ec55d | 1512 | conf_t *conf = mddev->private; |
1da177e4 | 1513 | struct list_head *head = &conf->retry_list; |
1da177e4 | 1514 | mdk_rdev_t *rdev; |
e1dfa0a2 | 1515 | struct blk_plug plug; |
1da177e4 LT |
1516 | |
1517 | md_check_recovery(mddev); | |
e1dfa0a2 N |
1518 | |
1519 | blk_start_plug(&plug); | |
1da177e4 LT |
1520 | for (;;) { |
1521 | char b[BDEVNAME_SIZE]; | |
191ea9b2 | 1522 | |
c3b328ac N |
1523 | if (atomic_read(&mddev->plug_cnt) == 0) |
1524 | flush_pending_writes(conf); | |
191ea9b2 | 1525 | |
a35e63ef N |
1526 | spin_lock_irqsave(&conf->device_lock, flags); |
1527 | if (list_empty(head)) { | |
1528 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 | 1529 | break; |
a35e63ef | 1530 | } |
1da177e4 LT |
1531 | r1_bio = list_entry(head->prev, r1bio_t, retry_list); |
1532 | list_del(head->prev); | |
ddaf22ab | 1533 | conf->nr_queued--; |
1da177e4 LT |
1534 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1535 | ||
1536 | mddev = r1_bio->mddev; | |
070ec55d | 1537 | conf = mddev->private; |
7eaceacc | 1538 | if (test_bit(R1BIO_IsSync, &r1_bio->state)) |
1da177e4 | 1539 | sync_request_write(mddev, r1_bio); |
7eaceacc | 1540 | else { |
1da177e4 | 1541 | int disk; |
ddaf22ab N |
1542 | |
1543 | /* we got a read error. Maybe the drive is bad. Maybe just | |
1544 | * the block and we can fix it. | |
1545 | * We freeze all other IO, and try reading the block from | |
1546 | * other devices. When we find one, we re-write | |
1547 | * and check it that fixes the read error. | |
1548 | * This is all done synchronously while the array is | |
1549 | * frozen | |
1550 | */ | |
867868fb N |
1551 | if (mddev->ro == 0) { |
1552 | freeze_array(conf); | |
1553 | fix_read_error(conf, r1_bio->read_disk, | |
1554 | r1_bio->sector, | |
1555 | r1_bio->sectors); | |
1556 | unfreeze_array(conf); | |
d0e26078 N |
1557 | } else |
1558 | md_error(mddev, | |
1559 | conf->mirrors[r1_bio->read_disk].rdev); | |
ddaf22ab | 1560 | |
1da177e4 | 1561 | bio = r1_bio->bios[r1_bio->read_disk]; |
d0e26078 | 1562 | if ((disk=read_balance(conf, r1_bio)) == -1) { |
9dd1e2fa | 1563 | printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O" |
1da177e4 | 1564 | " read error for block %llu\n", |
9dd1e2fa | 1565 | mdname(mddev), |
1da177e4 LT |
1566 | bdevname(bio->bi_bdev,b), |
1567 | (unsigned long long)r1_bio->sector); | |
1568 | raid_end_bio_io(r1_bio); | |
1569 | } else { | |
2c7d46ec | 1570 | const unsigned long do_sync = r1_bio->master_bio->bi_rw & REQ_SYNC; |
cf30a473 N |
1571 | r1_bio->bios[r1_bio->read_disk] = |
1572 | mddev->ro ? IO_BLOCKED : NULL; | |
1da177e4 LT |
1573 | r1_bio->read_disk = disk; |
1574 | bio_put(bio); | |
a167f663 N |
1575 | bio = bio_clone_mddev(r1_bio->master_bio, |
1576 | GFP_NOIO, mddev); | |
1da177e4 LT |
1577 | r1_bio->bios[r1_bio->read_disk] = bio; |
1578 | rdev = conf->mirrors[disk].rdev; | |
1579 | if (printk_ratelimit()) | |
9dd1e2fa | 1580 | printk(KERN_ERR "md/raid1:%s: redirecting sector %llu to" |
d754c5ae | 1581 | " other mirror: %s\n", |
9dd1e2fa | 1582 | mdname(mddev), |
d754c5ae N |
1583 | (unsigned long long)r1_bio->sector, |
1584 | bdevname(rdev->bdev,b)); | |
1da177e4 LT |
1585 | bio->bi_sector = r1_bio->sector + rdev->data_offset; |
1586 | bio->bi_bdev = rdev->bdev; | |
1587 | bio->bi_end_io = raid1_end_read_request; | |
7b6d91da | 1588 | bio->bi_rw = READ | do_sync; |
1da177e4 | 1589 | bio->bi_private = r1_bio; |
1da177e4 LT |
1590 | generic_make_request(bio); |
1591 | } | |
1592 | } | |
1d9d5241 | 1593 | cond_resched(); |
1da177e4 | 1594 | } |
e1dfa0a2 | 1595 | blk_finish_plug(&plug); |
1da177e4 LT |
1596 | } |
1597 | ||
1598 | ||
1599 | static int init_resync(conf_t *conf) | |
1600 | { | |
1601 | int buffs; | |
1602 | ||
1603 | buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; | |
9e77c485 | 1604 | BUG_ON(conf->r1buf_pool); |
1da177e4 LT |
1605 | conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free, |
1606 | conf->poolinfo); | |
1607 | if (!conf->r1buf_pool) | |
1608 | return -ENOMEM; | |
1609 | conf->next_resync = 0; | |
1610 | return 0; | |
1611 | } | |
1612 | ||
1613 | /* | |
1614 | * perform a "sync" on one "block" | |
1615 | * | |
1616 | * We need to make sure that no normal I/O request - particularly write | |
1617 | * requests - conflict with active sync requests. | |
1618 | * | |
1619 | * This is achieved by tracking pending requests and a 'barrier' concept | |
1620 | * that can be installed to exclude normal IO requests. | |
1621 | */ | |
1622 | ||
57afd89f | 1623 | static sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster) |
1da177e4 | 1624 | { |
070ec55d | 1625 | conf_t *conf = mddev->private; |
1da177e4 LT |
1626 | r1bio_t *r1_bio; |
1627 | struct bio *bio; | |
1628 | sector_t max_sector, nr_sectors; | |
3e198f78 | 1629 | int disk = -1; |
1da177e4 | 1630 | int i; |
3e198f78 N |
1631 | int wonly = -1; |
1632 | int write_targets = 0, read_targets = 0; | |
57dab0bd | 1633 | sector_t sync_blocks; |
e3b9703e | 1634 | int still_degraded = 0; |
1da177e4 LT |
1635 | |
1636 | if (!conf->r1buf_pool) | |
1637 | if (init_resync(conf)) | |
57afd89f | 1638 | return 0; |
1da177e4 | 1639 | |
58c0fed4 | 1640 | max_sector = mddev->dev_sectors; |
1da177e4 | 1641 | if (sector_nr >= max_sector) { |
191ea9b2 N |
1642 | /* If we aborted, we need to abort the |
1643 | * sync on the 'current' bitmap chunk (there will | |
1644 | * only be one in raid1 resync. | |
1645 | * We can find the current addess in mddev->curr_resync | |
1646 | */ | |
6a806c51 N |
1647 | if (mddev->curr_resync < max_sector) /* aborted */ |
1648 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
191ea9b2 | 1649 | &sync_blocks, 1); |
6a806c51 | 1650 | else /* completed sync */ |
191ea9b2 | 1651 | conf->fullsync = 0; |
6a806c51 N |
1652 | |
1653 | bitmap_close_sync(mddev->bitmap); | |
1da177e4 LT |
1654 | close_sync(conf); |
1655 | return 0; | |
1656 | } | |
1657 | ||
07d84d10 N |
1658 | if (mddev->bitmap == NULL && |
1659 | mddev->recovery_cp == MaxSector && | |
6394cca5 | 1660 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
07d84d10 N |
1661 | conf->fullsync == 0) { |
1662 | *skipped = 1; | |
1663 | return max_sector - sector_nr; | |
1664 | } | |
6394cca5 N |
1665 | /* before building a request, check if we can skip these blocks.. |
1666 | * This call the bitmap_start_sync doesn't actually record anything | |
1667 | */ | |
e3b9703e | 1668 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
e5de485f | 1669 | !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { |
191ea9b2 N |
1670 | /* We can skip this block, and probably several more */ |
1671 | *skipped = 1; | |
1672 | return sync_blocks; | |
1673 | } | |
1da177e4 | 1674 | /* |
17999be4 N |
1675 | * If there is non-resync activity waiting for a turn, |
1676 | * and resync is going fast enough, | |
1677 | * then let it though before starting on this new sync request. | |
1da177e4 | 1678 | */ |
17999be4 | 1679 | if (!go_faster && conf->nr_waiting) |
1da177e4 | 1680 | msleep_interruptible(1000); |
17999be4 | 1681 | |
b47490c9 | 1682 | bitmap_cond_end_sync(mddev->bitmap, sector_nr); |
1c4588e9 | 1683 | r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO); |
17999be4 N |
1684 | raise_barrier(conf); |
1685 | ||
1686 | conf->next_resync = sector_nr; | |
1da177e4 | 1687 | |
3e198f78 | 1688 | rcu_read_lock(); |
1da177e4 | 1689 | /* |
3e198f78 N |
1690 | * If we get a correctably read error during resync or recovery, |
1691 | * we might want to read from a different device. So we | |
1692 | * flag all drives that could conceivably be read from for READ, | |
1693 | * and any others (which will be non-In_sync devices) for WRITE. | |
1694 | * If a read fails, we try reading from something else for which READ | |
1695 | * is OK. | |
1da177e4 | 1696 | */ |
1da177e4 | 1697 | |
1da177e4 LT |
1698 | r1_bio->mddev = mddev; |
1699 | r1_bio->sector = sector_nr; | |
191ea9b2 | 1700 | r1_bio->state = 0; |
1da177e4 | 1701 | set_bit(R1BIO_IsSync, &r1_bio->state); |
1da177e4 LT |
1702 | |
1703 | for (i=0; i < conf->raid_disks; i++) { | |
3e198f78 | 1704 | mdk_rdev_t *rdev; |
1da177e4 LT |
1705 | bio = r1_bio->bios[i]; |
1706 | ||
1707 | /* take from bio_init */ | |
1708 | bio->bi_next = NULL; | |
db8d9d35 | 1709 | bio->bi_flags &= ~(BIO_POOL_MASK-1); |
1da177e4 | 1710 | bio->bi_flags |= 1 << BIO_UPTODATE; |
db8d9d35 | 1711 | bio->bi_comp_cpu = -1; |
802ba064 | 1712 | bio->bi_rw = READ; |
1da177e4 LT |
1713 | bio->bi_vcnt = 0; |
1714 | bio->bi_idx = 0; | |
1715 | bio->bi_phys_segments = 0; | |
1da177e4 LT |
1716 | bio->bi_size = 0; |
1717 | bio->bi_end_io = NULL; | |
1718 | bio->bi_private = NULL; | |
1719 | ||
3e198f78 N |
1720 | rdev = rcu_dereference(conf->mirrors[i].rdev); |
1721 | if (rdev == NULL || | |
1722 | test_bit(Faulty, &rdev->flags)) { | |
e3b9703e N |
1723 | still_degraded = 1; |
1724 | continue; | |
3e198f78 | 1725 | } else if (!test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
1726 | bio->bi_rw = WRITE; |
1727 | bio->bi_end_io = end_sync_write; | |
1728 | write_targets ++; | |
3e198f78 N |
1729 | } else { |
1730 | /* may need to read from here */ | |
1731 | bio->bi_rw = READ; | |
1732 | bio->bi_end_io = end_sync_read; | |
1733 | if (test_bit(WriteMostly, &rdev->flags)) { | |
1734 | if (wonly < 0) | |
1735 | wonly = i; | |
1736 | } else { | |
1737 | if (disk < 0) | |
1738 | disk = i; | |
1739 | } | |
1740 | read_targets++; | |
1741 | } | |
1742 | atomic_inc(&rdev->nr_pending); | |
1743 | bio->bi_sector = sector_nr + rdev->data_offset; | |
1744 | bio->bi_bdev = rdev->bdev; | |
1da177e4 LT |
1745 | bio->bi_private = r1_bio; |
1746 | } | |
3e198f78 N |
1747 | rcu_read_unlock(); |
1748 | if (disk < 0) | |
1749 | disk = wonly; | |
1750 | r1_bio->read_disk = disk; | |
191ea9b2 | 1751 | |
3e198f78 N |
1752 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0) |
1753 | /* extra read targets are also write targets */ | |
1754 | write_targets += read_targets-1; | |
1755 | ||
1756 | if (write_targets == 0 || read_targets == 0) { | |
1da177e4 LT |
1757 | /* There is nowhere to write, so all non-sync |
1758 | * drives must be failed - so we are finished | |
1759 | */ | |
57afd89f N |
1760 | sector_t rv = max_sector - sector_nr; |
1761 | *skipped = 1; | |
1da177e4 | 1762 | put_buf(r1_bio); |
1da177e4 LT |
1763 | return rv; |
1764 | } | |
1765 | ||
c6207277 N |
1766 | if (max_sector > mddev->resync_max) |
1767 | max_sector = mddev->resync_max; /* Don't do IO beyond here */ | |
1da177e4 | 1768 | nr_sectors = 0; |
289e99e8 | 1769 | sync_blocks = 0; |
1da177e4 LT |
1770 | do { |
1771 | struct page *page; | |
1772 | int len = PAGE_SIZE; | |
1773 | if (sector_nr + (len>>9) > max_sector) | |
1774 | len = (max_sector - sector_nr) << 9; | |
1775 | if (len == 0) | |
1776 | break; | |
6a806c51 N |
1777 | if (sync_blocks == 0) { |
1778 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, | |
e5de485f N |
1779 | &sync_blocks, still_degraded) && |
1780 | !conf->fullsync && | |
1781 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6a806c51 | 1782 | break; |
9e77c485 | 1783 | BUG_ON(sync_blocks < (PAGE_SIZE>>9)); |
7571ae88 | 1784 | if ((len >> 9) > sync_blocks) |
6a806c51 | 1785 | len = sync_blocks<<9; |
ab7a30c7 | 1786 | } |
191ea9b2 | 1787 | |
1da177e4 LT |
1788 | for (i=0 ; i < conf->raid_disks; i++) { |
1789 | bio = r1_bio->bios[i]; | |
1790 | if (bio->bi_end_io) { | |
d11c171e | 1791 | page = bio->bi_io_vec[bio->bi_vcnt].bv_page; |
1da177e4 LT |
1792 | if (bio_add_page(bio, page, len, 0) == 0) { |
1793 | /* stop here */ | |
d11c171e | 1794 | bio->bi_io_vec[bio->bi_vcnt].bv_page = page; |
1da177e4 LT |
1795 | while (i > 0) { |
1796 | i--; | |
1797 | bio = r1_bio->bios[i]; | |
6a806c51 N |
1798 | if (bio->bi_end_io==NULL) |
1799 | continue; | |
1da177e4 LT |
1800 | /* remove last page from this bio */ |
1801 | bio->bi_vcnt--; | |
1802 | bio->bi_size -= len; | |
1803 | bio->bi_flags &= ~(1<< BIO_SEG_VALID); | |
1804 | } | |
1805 | goto bio_full; | |
1806 | } | |
1807 | } | |
1808 | } | |
1809 | nr_sectors += len>>9; | |
1810 | sector_nr += len>>9; | |
191ea9b2 | 1811 | sync_blocks -= (len>>9); |
1da177e4 LT |
1812 | } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES); |
1813 | bio_full: | |
1da177e4 LT |
1814 | r1_bio->sectors = nr_sectors; |
1815 | ||
d11c171e N |
1816 | /* For a user-requested sync, we read all readable devices and do a |
1817 | * compare | |
1818 | */ | |
1819 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
1820 | atomic_set(&r1_bio->remaining, read_targets); | |
1821 | for (i=0; i<conf->raid_disks; i++) { | |
1822 | bio = r1_bio->bios[i]; | |
1823 | if (bio->bi_end_io == end_sync_read) { | |
ddac7c7e | 1824 | md_sync_acct(bio->bi_bdev, nr_sectors); |
d11c171e N |
1825 | generic_make_request(bio); |
1826 | } | |
1827 | } | |
1828 | } else { | |
1829 | atomic_set(&r1_bio->remaining, 1); | |
1830 | bio = r1_bio->bios[r1_bio->read_disk]; | |
ddac7c7e | 1831 | md_sync_acct(bio->bi_bdev, nr_sectors); |
d11c171e | 1832 | generic_make_request(bio); |
1da177e4 | 1833 | |
d11c171e | 1834 | } |
1da177e4 LT |
1835 | return nr_sectors; |
1836 | } | |
1837 | ||
80c3a6ce DW |
1838 | static sector_t raid1_size(mddev_t *mddev, sector_t sectors, int raid_disks) |
1839 | { | |
1840 | if (sectors) | |
1841 | return sectors; | |
1842 | ||
1843 | return mddev->dev_sectors; | |
1844 | } | |
1845 | ||
709ae487 | 1846 | static conf_t *setup_conf(mddev_t *mddev) |
1da177e4 LT |
1847 | { |
1848 | conf_t *conf; | |
709ae487 | 1849 | int i; |
1da177e4 LT |
1850 | mirror_info_t *disk; |
1851 | mdk_rdev_t *rdev; | |
709ae487 | 1852 | int err = -ENOMEM; |
1da177e4 | 1853 | |
9ffae0cf | 1854 | conf = kzalloc(sizeof(conf_t), GFP_KERNEL); |
1da177e4 | 1855 | if (!conf) |
709ae487 | 1856 | goto abort; |
1da177e4 | 1857 | |
9ffae0cf | 1858 | conf->mirrors = kzalloc(sizeof(struct mirror_info)*mddev->raid_disks, |
1da177e4 LT |
1859 | GFP_KERNEL); |
1860 | if (!conf->mirrors) | |
709ae487 | 1861 | goto abort; |
1da177e4 | 1862 | |
ddaf22ab N |
1863 | conf->tmppage = alloc_page(GFP_KERNEL); |
1864 | if (!conf->tmppage) | |
709ae487 | 1865 | goto abort; |
ddaf22ab | 1866 | |
709ae487 | 1867 | conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL); |
1da177e4 | 1868 | if (!conf->poolinfo) |
709ae487 | 1869 | goto abort; |
1da177e4 LT |
1870 | conf->poolinfo->raid_disks = mddev->raid_disks; |
1871 | conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, | |
1872 | r1bio_pool_free, | |
1873 | conf->poolinfo); | |
1874 | if (!conf->r1bio_pool) | |
709ae487 N |
1875 | goto abort; |
1876 | ||
ed9bfdf1 | 1877 | conf->poolinfo->mddev = mddev; |
1da177e4 | 1878 | |
e7e72bf6 | 1879 | spin_lock_init(&conf->device_lock); |
159ec1fc | 1880 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
709ae487 | 1881 | int disk_idx = rdev->raid_disk; |
1da177e4 LT |
1882 | if (disk_idx >= mddev->raid_disks |
1883 | || disk_idx < 0) | |
1884 | continue; | |
1885 | disk = conf->mirrors + disk_idx; | |
1886 | ||
1887 | disk->rdev = rdev; | |
1da177e4 LT |
1888 | |
1889 | disk->head_position = 0; | |
1da177e4 LT |
1890 | } |
1891 | conf->raid_disks = mddev->raid_disks; | |
1892 | conf->mddev = mddev; | |
1da177e4 | 1893 | INIT_LIST_HEAD(&conf->retry_list); |
1da177e4 LT |
1894 | |
1895 | spin_lock_init(&conf->resync_lock); | |
17999be4 | 1896 | init_waitqueue_head(&conf->wait_barrier); |
1da177e4 | 1897 | |
191ea9b2 | 1898 | bio_list_init(&conf->pending_bio_list); |
191ea9b2 | 1899 | |
709ae487 | 1900 | conf->last_used = -1; |
1da177e4 LT |
1901 | for (i = 0; i < conf->raid_disks; i++) { |
1902 | ||
1903 | disk = conf->mirrors + i; | |
1904 | ||
5fd6c1dc N |
1905 | if (!disk->rdev || |
1906 | !test_bit(In_sync, &disk->rdev->flags)) { | |
1da177e4 | 1907 | disk->head_position = 0; |
918f0238 N |
1908 | if (disk->rdev) |
1909 | conf->fullsync = 1; | |
709ae487 N |
1910 | } else if (conf->last_used < 0) |
1911 | /* | |
1912 | * The first working device is used as a | |
1913 | * starting point to read balancing. | |
1914 | */ | |
1915 | conf->last_used = i; | |
1da177e4 | 1916 | } |
709ae487 N |
1917 | |
1918 | err = -EIO; | |
1919 | if (conf->last_used < 0) { | |
9dd1e2fa | 1920 | printk(KERN_ERR "md/raid1:%s: no operational mirrors\n", |
709ae487 N |
1921 | mdname(mddev)); |
1922 | goto abort; | |
1923 | } | |
1924 | err = -ENOMEM; | |
1925 | conf->thread = md_register_thread(raid1d, mddev, NULL); | |
1926 | if (!conf->thread) { | |
1927 | printk(KERN_ERR | |
9dd1e2fa | 1928 | "md/raid1:%s: couldn't allocate thread\n", |
709ae487 N |
1929 | mdname(mddev)); |
1930 | goto abort; | |
11ce99e6 | 1931 | } |
1da177e4 | 1932 | |
709ae487 N |
1933 | return conf; |
1934 | ||
1935 | abort: | |
1936 | if (conf) { | |
1937 | if (conf->r1bio_pool) | |
1938 | mempool_destroy(conf->r1bio_pool); | |
1939 | kfree(conf->mirrors); | |
1940 | safe_put_page(conf->tmppage); | |
1941 | kfree(conf->poolinfo); | |
1942 | kfree(conf); | |
1943 | } | |
1944 | return ERR_PTR(err); | |
1945 | } | |
1946 | ||
1947 | static int run(mddev_t *mddev) | |
1948 | { | |
1949 | conf_t *conf; | |
1950 | int i; | |
1951 | mdk_rdev_t *rdev; | |
1952 | ||
1953 | if (mddev->level != 1) { | |
9dd1e2fa | 1954 | printk(KERN_ERR "md/raid1:%s: raid level not set to mirroring (%d)\n", |
709ae487 N |
1955 | mdname(mddev), mddev->level); |
1956 | return -EIO; | |
1957 | } | |
1958 | if (mddev->reshape_position != MaxSector) { | |
9dd1e2fa | 1959 | printk(KERN_ERR "md/raid1:%s: reshape_position set but not supported\n", |
709ae487 N |
1960 | mdname(mddev)); |
1961 | return -EIO; | |
1962 | } | |
1da177e4 | 1963 | /* |
709ae487 N |
1964 | * copy the already verified devices into our private RAID1 |
1965 | * bookkeeping area. [whatever we allocate in run(), | |
1966 | * should be freed in stop()] | |
1da177e4 | 1967 | */ |
709ae487 N |
1968 | if (mddev->private == NULL) |
1969 | conf = setup_conf(mddev); | |
1970 | else | |
1971 | conf = mddev->private; | |
1da177e4 | 1972 | |
709ae487 N |
1973 | if (IS_ERR(conf)) |
1974 | return PTR_ERR(conf); | |
1da177e4 | 1975 | |
709ae487 N |
1976 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1977 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
1978 | rdev->data_offset << 9); | |
1979 | /* as we don't honour merge_bvec_fn, we must never risk | |
627a2d3c N |
1980 | * violating it, so limit ->max_segments to 1 lying within |
1981 | * a single page, as a one page request is never in violation. | |
709ae487 | 1982 | */ |
627a2d3c N |
1983 | if (rdev->bdev->bd_disk->queue->merge_bvec_fn) { |
1984 | blk_queue_max_segments(mddev->queue, 1); | |
1985 | blk_queue_segment_boundary(mddev->queue, | |
1986 | PAGE_CACHE_SIZE - 1); | |
1987 | } | |
1da177e4 | 1988 | } |
191ea9b2 | 1989 | |
709ae487 N |
1990 | mddev->degraded = 0; |
1991 | for (i=0; i < conf->raid_disks; i++) | |
1992 | if (conf->mirrors[i].rdev == NULL || | |
1993 | !test_bit(In_sync, &conf->mirrors[i].rdev->flags) || | |
1994 | test_bit(Faulty, &conf->mirrors[i].rdev->flags)) | |
1995 | mddev->degraded++; | |
1996 | ||
1997 | if (conf->raid_disks - mddev->degraded == 1) | |
1998 | mddev->recovery_cp = MaxSector; | |
1999 | ||
8c6ac868 | 2000 | if (mddev->recovery_cp != MaxSector) |
9dd1e2fa | 2001 | printk(KERN_NOTICE "md/raid1:%s: not clean" |
8c6ac868 AN |
2002 | " -- starting background reconstruction\n", |
2003 | mdname(mddev)); | |
1da177e4 | 2004 | printk(KERN_INFO |
9dd1e2fa | 2005 | "md/raid1:%s: active with %d out of %d mirrors\n", |
1da177e4 LT |
2006 | mdname(mddev), mddev->raid_disks - mddev->degraded, |
2007 | mddev->raid_disks); | |
709ae487 | 2008 | |
1da177e4 LT |
2009 | /* |
2010 | * Ok, everything is just fine now | |
2011 | */ | |
709ae487 N |
2012 | mddev->thread = conf->thread; |
2013 | conf->thread = NULL; | |
2014 | mddev->private = conf; | |
2015 | ||
1f403624 | 2016 | md_set_array_sectors(mddev, raid1_size(mddev, 0, 0)); |
1da177e4 | 2017 | |
0d129228 N |
2018 | mddev->queue->backing_dev_info.congested_fn = raid1_congested; |
2019 | mddev->queue->backing_dev_info.congested_data = mddev; | |
a91a2785 | 2020 | return md_integrity_register(mddev); |
1da177e4 LT |
2021 | } |
2022 | ||
2023 | static int stop(mddev_t *mddev) | |
2024 | { | |
070ec55d | 2025 | conf_t *conf = mddev->private; |
4b6d287f | 2026 | struct bitmap *bitmap = mddev->bitmap; |
4b6d287f N |
2027 | |
2028 | /* wait for behind writes to complete */ | |
e555190d | 2029 | if (bitmap && atomic_read(&bitmap->behind_writes) > 0) { |
9dd1e2fa N |
2030 | printk(KERN_INFO "md/raid1:%s: behind writes in progress - waiting to stop.\n", |
2031 | mdname(mddev)); | |
4b6d287f | 2032 | /* need to kick something here to make sure I/O goes? */ |
e555190d N |
2033 | wait_event(bitmap->behind_wait, |
2034 | atomic_read(&bitmap->behind_writes) == 0); | |
4b6d287f | 2035 | } |
1da177e4 | 2036 | |
409c57f3 N |
2037 | raise_barrier(conf); |
2038 | lower_barrier(conf); | |
2039 | ||
1da177e4 LT |
2040 | md_unregister_thread(mddev->thread); |
2041 | mddev->thread = NULL; | |
1da177e4 LT |
2042 | if (conf->r1bio_pool) |
2043 | mempool_destroy(conf->r1bio_pool); | |
990a8baf JJ |
2044 | kfree(conf->mirrors); |
2045 | kfree(conf->poolinfo); | |
1da177e4 LT |
2046 | kfree(conf); |
2047 | mddev->private = NULL; | |
2048 | return 0; | |
2049 | } | |
2050 | ||
2051 | static int raid1_resize(mddev_t *mddev, sector_t sectors) | |
2052 | { | |
2053 | /* no resync is happening, and there is enough space | |
2054 | * on all devices, so we can resize. | |
2055 | * We need to make sure resync covers any new space. | |
2056 | * If the array is shrinking we should possibly wait until | |
2057 | * any io in the removed space completes, but it hardly seems | |
2058 | * worth it. | |
2059 | */ | |
1f403624 | 2060 | md_set_array_sectors(mddev, raid1_size(mddev, sectors, 0)); |
b522adcd DW |
2061 | if (mddev->array_sectors > raid1_size(mddev, sectors, 0)) |
2062 | return -EINVAL; | |
f233ea5c | 2063 | set_capacity(mddev->gendisk, mddev->array_sectors); |
449aad3e | 2064 | revalidate_disk(mddev->gendisk); |
b522adcd | 2065 | if (sectors > mddev->dev_sectors && |
f233ea5c | 2066 | mddev->recovery_cp == MaxSector) { |
58c0fed4 | 2067 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
2068 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
2069 | } | |
b522adcd | 2070 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 2071 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
2072 | return 0; |
2073 | } | |
2074 | ||
63c70c4f | 2075 | static int raid1_reshape(mddev_t *mddev) |
1da177e4 LT |
2076 | { |
2077 | /* We need to: | |
2078 | * 1/ resize the r1bio_pool | |
2079 | * 2/ resize conf->mirrors | |
2080 | * | |
2081 | * We allocate a new r1bio_pool if we can. | |
2082 | * Then raise a device barrier and wait until all IO stops. | |
2083 | * Then resize conf->mirrors and swap in the new r1bio pool. | |
6ea9c07c N |
2084 | * |
2085 | * At the same time, we "pack" the devices so that all the missing | |
2086 | * devices have the higher raid_disk numbers. | |
1da177e4 LT |
2087 | */ |
2088 | mempool_t *newpool, *oldpool; | |
2089 | struct pool_info *newpoolinfo; | |
2090 | mirror_info_t *newmirrors; | |
070ec55d | 2091 | conf_t *conf = mddev->private; |
63c70c4f | 2092 | int cnt, raid_disks; |
c04be0aa | 2093 | unsigned long flags; |
b5470dc5 | 2094 | int d, d2, err; |
1da177e4 | 2095 | |
63c70c4f | 2096 | /* Cannot change chunk_size, layout, or level */ |
664e7c41 | 2097 | if (mddev->chunk_sectors != mddev->new_chunk_sectors || |
63c70c4f N |
2098 | mddev->layout != mddev->new_layout || |
2099 | mddev->level != mddev->new_level) { | |
664e7c41 | 2100 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
63c70c4f N |
2101 | mddev->new_layout = mddev->layout; |
2102 | mddev->new_level = mddev->level; | |
2103 | return -EINVAL; | |
2104 | } | |
2105 | ||
b5470dc5 DW |
2106 | err = md_allow_write(mddev); |
2107 | if (err) | |
2108 | return err; | |
2a2275d6 | 2109 | |
63c70c4f N |
2110 | raid_disks = mddev->raid_disks + mddev->delta_disks; |
2111 | ||
6ea9c07c N |
2112 | if (raid_disks < conf->raid_disks) { |
2113 | cnt=0; | |
2114 | for (d= 0; d < conf->raid_disks; d++) | |
2115 | if (conf->mirrors[d].rdev) | |
2116 | cnt++; | |
2117 | if (cnt > raid_disks) | |
1da177e4 | 2118 | return -EBUSY; |
6ea9c07c | 2119 | } |
1da177e4 LT |
2120 | |
2121 | newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL); | |
2122 | if (!newpoolinfo) | |
2123 | return -ENOMEM; | |
2124 | newpoolinfo->mddev = mddev; | |
2125 | newpoolinfo->raid_disks = raid_disks; | |
2126 | ||
2127 | newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, | |
2128 | r1bio_pool_free, newpoolinfo); | |
2129 | if (!newpool) { | |
2130 | kfree(newpoolinfo); | |
2131 | return -ENOMEM; | |
2132 | } | |
9ffae0cf | 2133 | newmirrors = kzalloc(sizeof(struct mirror_info) * raid_disks, GFP_KERNEL); |
1da177e4 LT |
2134 | if (!newmirrors) { |
2135 | kfree(newpoolinfo); | |
2136 | mempool_destroy(newpool); | |
2137 | return -ENOMEM; | |
2138 | } | |
1da177e4 | 2139 | |
17999be4 | 2140 | raise_barrier(conf); |
1da177e4 LT |
2141 | |
2142 | /* ok, everything is stopped */ | |
2143 | oldpool = conf->r1bio_pool; | |
2144 | conf->r1bio_pool = newpool; | |
6ea9c07c | 2145 | |
a88aa786 N |
2146 | for (d = d2 = 0; d < conf->raid_disks; d++) { |
2147 | mdk_rdev_t *rdev = conf->mirrors[d].rdev; | |
2148 | if (rdev && rdev->raid_disk != d2) { | |
2149 | char nm[20]; | |
2150 | sprintf(nm, "rd%d", rdev->raid_disk); | |
2151 | sysfs_remove_link(&mddev->kobj, nm); | |
2152 | rdev->raid_disk = d2; | |
2153 | sprintf(nm, "rd%d", rdev->raid_disk); | |
2154 | sysfs_remove_link(&mddev->kobj, nm); | |
2155 | if (sysfs_create_link(&mddev->kobj, | |
2156 | &rdev->kobj, nm)) | |
2157 | printk(KERN_WARNING | |
9dd1e2fa N |
2158 | "md/raid1:%s: cannot register " |
2159 | "%s\n", | |
2160 | mdname(mddev), nm); | |
6ea9c07c | 2161 | } |
a88aa786 N |
2162 | if (rdev) |
2163 | newmirrors[d2++].rdev = rdev; | |
2164 | } | |
1da177e4 LT |
2165 | kfree(conf->mirrors); |
2166 | conf->mirrors = newmirrors; | |
2167 | kfree(conf->poolinfo); | |
2168 | conf->poolinfo = newpoolinfo; | |
2169 | ||
c04be0aa | 2170 | spin_lock_irqsave(&conf->device_lock, flags); |
1da177e4 | 2171 | mddev->degraded += (raid_disks - conf->raid_disks); |
c04be0aa | 2172 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 2173 | conf->raid_disks = mddev->raid_disks = raid_disks; |
63c70c4f | 2174 | mddev->delta_disks = 0; |
1da177e4 | 2175 | |
6ea9c07c | 2176 | conf->last_used = 0; /* just make sure it is in-range */ |
17999be4 | 2177 | lower_barrier(conf); |
1da177e4 LT |
2178 | |
2179 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
2180 | md_wakeup_thread(mddev->thread); | |
2181 | ||
2182 | mempool_destroy(oldpool); | |
2183 | return 0; | |
2184 | } | |
2185 | ||
500af87a | 2186 | static void raid1_quiesce(mddev_t *mddev, int state) |
36fa3063 | 2187 | { |
070ec55d | 2188 | conf_t *conf = mddev->private; |
36fa3063 N |
2189 | |
2190 | switch(state) { | |
6eef4b21 N |
2191 | case 2: /* wake for suspend */ |
2192 | wake_up(&conf->wait_barrier); | |
2193 | break; | |
9e6603da | 2194 | case 1: |
17999be4 | 2195 | raise_barrier(conf); |
36fa3063 | 2196 | break; |
9e6603da | 2197 | case 0: |
17999be4 | 2198 | lower_barrier(conf); |
36fa3063 N |
2199 | break; |
2200 | } | |
36fa3063 N |
2201 | } |
2202 | ||
709ae487 N |
2203 | static void *raid1_takeover(mddev_t *mddev) |
2204 | { | |
2205 | /* raid1 can take over: | |
2206 | * raid5 with 2 devices, any layout or chunk size | |
2207 | */ | |
2208 | if (mddev->level == 5 && mddev->raid_disks == 2) { | |
2209 | conf_t *conf; | |
2210 | mddev->new_level = 1; | |
2211 | mddev->new_layout = 0; | |
2212 | mddev->new_chunk_sectors = 0; | |
2213 | conf = setup_conf(mddev); | |
2214 | if (!IS_ERR(conf)) | |
2215 | conf->barrier = 1; | |
2216 | return conf; | |
2217 | } | |
2218 | return ERR_PTR(-EINVAL); | |
2219 | } | |
1da177e4 | 2220 | |
2604b703 | 2221 | static struct mdk_personality raid1_personality = |
1da177e4 LT |
2222 | { |
2223 | .name = "raid1", | |
2604b703 | 2224 | .level = 1, |
1da177e4 LT |
2225 | .owner = THIS_MODULE, |
2226 | .make_request = make_request, | |
2227 | .run = run, | |
2228 | .stop = stop, | |
2229 | .status = status, | |
2230 | .error_handler = error, | |
2231 | .hot_add_disk = raid1_add_disk, | |
2232 | .hot_remove_disk= raid1_remove_disk, | |
2233 | .spare_active = raid1_spare_active, | |
2234 | .sync_request = sync_request, | |
2235 | .resize = raid1_resize, | |
80c3a6ce | 2236 | .size = raid1_size, |
63c70c4f | 2237 | .check_reshape = raid1_reshape, |
36fa3063 | 2238 | .quiesce = raid1_quiesce, |
709ae487 | 2239 | .takeover = raid1_takeover, |
1da177e4 LT |
2240 | }; |
2241 | ||
2242 | static int __init raid_init(void) | |
2243 | { | |
2604b703 | 2244 | return register_md_personality(&raid1_personality); |
1da177e4 LT |
2245 | } |
2246 | ||
2247 | static void raid_exit(void) | |
2248 | { | |
2604b703 | 2249 | unregister_md_personality(&raid1_personality); |
1da177e4 LT |
2250 | } |
2251 | ||
2252 | module_init(raid_init); | |
2253 | module_exit(raid_exit); | |
2254 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 2255 | MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD"); |
1da177e4 | 2256 | MODULE_ALIAS("md-personality-3"); /* RAID1 */ |
d9d166c2 | 2257 | MODULE_ALIAS("md-raid1"); |
2604b703 | 2258 | MODULE_ALIAS("md-level-1"); |