]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - drivers/md/raid1.c
projects / mirror_ubuntu-zesty-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
[PATCH] md: remove personality numbering from md
[mirror_ubuntu-zesty-kernel.git] / drivers / md / raid1.c
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 *
12 * Fixes to reconstruction by Jakob Østergaard" <jakob@ostenfeld.dk>
13 * Various fixes by Neil Brown <neilb@cse.unsw.edu.au>
14 *
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 *
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
34 #include "dm-bio-list.h"
35 #include <linux/raid/raid1.h>
36 #include <linux/raid/bitmap.h>
37
38 #define DEBUG 0
39 #if DEBUG
40 #define PRINTK(x...) printk(x)
41 #else
42 #define PRINTK(x...)
43 #endif
44
45 /*
46 * Number of guaranteed r1bios in case of extreme VM load:
47 */
48 #define NR_RAID1_BIOS 256
49
50
51 static void unplug_slaves(mddev_t *mddev);
52
53 static void allow_barrier(conf_t *conf);
54 static void lower_barrier(conf_t *conf);
55
56 static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data)
57 {
58 struct pool_info *pi = data;
59 r1bio_t *r1_bio;
60 int size = offsetof(r1bio_t, bios[pi->raid_disks]);
61
62 /* allocate a r1bio with room for raid_disks entries in the bios array */
63 r1_bio = kzalloc(size, gfp_flags);
64 if (!r1_bio)
65 unplug_slaves(pi->mddev);
66
67 return r1_bio;
68 }
69
70 static void r1bio_pool_free(void *r1_bio, void *data)
71 {
72 kfree(r1_bio);
73 }
74
75 #define RESYNC_BLOCK_SIZE (64*1024)
76 //#define RESYNC_BLOCK_SIZE PAGE_SIZE
77 #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9)
78 #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE)
79 #define RESYNC_WINDOW (2048*1024)
80
81 static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data)
82 {
83 struct pool_info *pi = data;
84 struct page *page;
85 r1bio_t *r1_bio;
86 struct bio *bio;
87 int i, j;
88
89 r1_bio = r1bio_pool_alloc(gfp_flags, pi);
90 if (!r1_bio) {
91 unplug_slaves(pi->mddev);
92 return NULL;
93 }
94
95 /*
96 * Allocate bios : 1 for reading, n-1 for writing
97 */
98 for (j = pi->raid_disks ; j-- ; ) {
99 bio = bio_alloc(gfp_flags, RESYNC_PAGES);
100 if (!bio)
101 goto out_free_bio;
102 r1_bio->bios[j] = bio;
103 }
104 /*
105 * Allocate RESYNC_PAGES data pages and attach them to
106 * the first bio.
107 * If this is a user-requested check/repair, allocate
108 * RESYNC_PAGES for each bio.
109 */
110 if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery))
111 j = pi->raid_disks;
112 else
113 j = 1;
114 while(j--) {
115 bio = r1_bio->bios[j];
116 for (i = 0; i < RESYNC_PAGES; i++) {
117 page = alloc_page(gfp_flags);
118 if (unlikely(!page))
119 goto out_free_pages;
120
121 bio->bi_io_vec[i].bv_page = page;
122 }
123 }
124 /* If not user-requests, copy the page pointers to all bios */
125 if (!test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) {
126 for (i=0; i<RESYNC_PAGES ; i++)
127 for (j=1; j<pi->raid_disks; j++)
128 r1_bio->bios[j]->bi_io_vec[i].bv_page =
129 r1_bio->bios[0]->bi_io_vec[i].bv_page;
130 }
131
132 r1_bio->master_bio = NULL;
133
134 return r1_bio;
135
136 out_free_pages:
137 for (i=0; i < RESYNC_PAGES ; i++)
138 for (j=0 ; j < pi->raid_disks; j++)
139 put_page(r1_bio->bios[j]->bi_io_vec[i].bv_page);
140 j = -1;
141 out_free_bio:
142 while ( ++j < pi->raid_disks )
143 bio_put(r1_bio->bios[j]);
144 r1bio_pool_free(r1_bio, data);
145 return NULL;
146 }
147
148 static void r1buf_pool_free(void *__r1_bio, void *data)
149 {
150 struct pool_info *pi = data;
151 int i,j;
152 r1bio_t *r1bio = __r1_bio;
153
154 for (i = 0; i < RESYNC_PAGES; i++)
155 for (j = pi->raid_disks; j-- ;) {
156 if (j == 0 ||
157 r1bio->bios[j]->bi_io_vec[i].bv_page !=
158 r1bio->bios[0]->bi_io_vec[i].bv_page)
159 put_page(r1bio->bios[j]->bi_io_vec[i].bv_page);
160 }
161 for (i=0 ; i < pi->raid_disks; i++)
162 bio_put(r1bio->bios[i]);
163
164 r1bio_pool_free(r1bio, data);
165 }
166
167 static void put_all_bios(conf_t *conf, r1bio_t *r1_bio)
168 {
169 int i;
170
171 for (i = 0; i < conf->raid_disks; i++) {
172 struct bio **bio = r1_bio->bios + i;
173 if (*bio && *bio != IO_BLOCKED)
174 bio_put(*bio);
175 *bio = NULL;
176 }
177 }
178
179 static inline void free_r1bio(r1bio_t *r1_bio)
180 {
181 conf_t *conf = mddev_to_conf(r1_bio->mddev);
182
183 /*
184 * Wake up any possible resync thread that waits for the device
185 * to go idle.
186 */
187 allow_barrier(conf);
188
189 put_all_bios(conf, r1_bio);
190 mempool_free(r1_bio, conf->r1bio_pool);
191 }
192
193 static inline void put_buf(r1bio_t *r1_bio)
194 {
195 conf_t *conf = mddev_to_conf(r1_bio->mddev);
196 int i;
197
198 for (i=0; i<conf->raid_disks; i++) {
199 struct bio *bio = r1_bio->bios[i];
200 if (bio->bi_end_io)
201 rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev);
202 }
203
204 mempool_free(r1_bio, conf->r1buf_pool);
205
206 lower_barrier(conf);
207 }
208
209 static void reschedule_retry(r1bio_t *r1_bio)
210 {
211 unsigned long flags;
212 mddev_t *mddev = r1_bio->mddev;
213 conf_t *conf = mddev_to_conf(mddev);
214
215 spin_lock_irqsave(&conf->device_lock, flags);
216 list_add(&r1_bio->retry_list, &conf->retry_list);
217 conf->nr_queued ++;
218 spin_unlock_irqrestore(&conf->device_lock, flags);
219
220 wake_up(&conf->wait_barrier);
221 md_wakeup_thread(mddev->thread);
222 }
223
224 /*
225 * raid_end_bio_io() is called when we have finished servicing a mirrored
226 * operation and are ready to return a success/failure code to the buffer
227 * cache layer.
228 */
229 static void raid_end_bio_io(r1bio_t *r1_bio)
230 {
231 struct bio *bio = r1_bio->master_bio;
232
233 /* if nobody has done the final endio yet, do it now */
234 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
235 PRINTK(KERN_DEBUG "raid1: sync end %s on sectors %llu-%llu\n",
236 (bio_data_dir(bio) == WRITE) ? "write" : "read",
237 (unsigned long long) bio->bi_sector,
238 (unsigned long long) bio->bi_sector +
239 (bio->bi_size >> 9) - 1);
240
241 bio_endio(bio, bio->bi_size,
242 test_bit(R1BIO_Uptodate, &r1_bio->state) ? 0 : -EIO);
243 }
244 free_r1bio(r1_bio);
245 }
246
247 /*
248 * Update disk head position estimator based on IRQ completion info.
249 */
250 static inline void update_head_pos(int disk, r1bio_t *r1_bio)
251 {
252 conf_t *conf = mddev_to_conf(r1_bio->mddev);
253
254 conf->mirrors[disk].head_position =
255 r1_bio->sector + (r1_bio->sectors);
256 }
257
258 static int raid1_end_read_request(struct bio *bio, unsigned int bytes_done, int error)
259 {
260 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
261 r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
262 int mirror;
263 conf_t *conf = mddev_to_conf(r1_bio->mddev);
264
265 if (bio->bi_size)
266 return 1;
267
268 mirror = r1_bio->read_disk;
269 /*
270 * this branch is our 'one mirror IO has finished' event handler:
271 */
272 update_head_pos(mirror, r1_bio);
273
274 if (uptodate || conf->working_disks <= 1) {
275 /*
276 * Set R1BIO_Uptodate in our master bio, so that
277 * we will return a good error code for to the higher
278 * levels even if IO on some other mirrored buffer fails.
279 *
280 * The 'master' represents the composite IO operation to
281 * user-side. So if something waits for IO, then it will
282 * wait for the 'master' bio.
283 */
284 if (uptodate)
285 set_bit(R1BIO_Uptodate, &r1_bio->state);
286
287 raid_end_bio_io(r1_bio);
288 } else {
289 /*
290 * oops, read error:
291 */
292 char b[BDEVNAME_SIZE];
293 if (printk_ratelimit())
294 printk(KERN_ERR "raid1: %s: rescheduling sector %llu\n",
295 bdevname(conf->mirrors[mirror].rdev->bdev,b), (unsigned long long)r1_bio->sector);
296 reschedule_retry(r1_bio);
297 }
298
299 rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev);
300 return 0;
301 }
302
303 static int raid1_end_write_request(struct bio *bio, unsigned int bytes_done, int error)
304 {
305 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
306 r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
307 int mirror, behind = test_bit(R1BIO_BehindIO, &r1_bio->state);
308 conf_t *conf = mddev_to_conf(r1_bio->mddev);
309
310 if (bio->bi_size)
311 return 1;
312
313 for (mirror = 0; mirror < conf->raid_disks; mirror++)
314 if (r1_bio->bios[mirror] == bio)
315 break;
316
317 if (error == -ENOTSUPP && test_bit(R1BIO_Barrier, &r1_bio->state)) {
318 set_bit(BarriersNotsupp, &conf->mirrors[mirror].rdev->flags);
319 set_bit(R1BIO_BarrierRetry, &r1_bio->state);
320 r1_bio->mddev->barriers_work = 0;
321 } else {
322 /*
323 * this branch is our 'one mirror IO has finished' event handler:
324 */
325 r1_bio->bios[mirror] = NULL;
326 if (!uptodate) {
327 md_error(r1_bio->mddev, conf->mirrors[mirror].rdev);
328 /* an I/O failed, we can't clear the bitmap */
329 set_bit(R1BIO_Degraded, &r1_bio->state);
330 } else
331 /*
332 * Set R1BIO_Uptodate in our master bio, so that
333 * we will return a good error code for to the higher
334 * levels even if IO on some other mirrored buffer fails.
335 *
336 * The 'master' represents the composite IO operation to
337 * user-side. So if something waits for IO, then it will
338 * wait for the 'master' bio.
339 */
340 set_bit(R1BIO_Uptodate, &r1_bio->state);
341
342 update_head_pos(mirror, r1_bio);
343
344 if (behind) {
345 if (test_bit(WriteMostly, &conf->mirrors[mirror].rdev->flags))
346 atomic_dec(&r1_bio->behind_remaining);
347
348 /* In behind mode, we ACK the master bio once the I/O has safely
349 * reached all non-writemostly disks. Setting the Returned bit
350 * ensures that this gets done only once -- we don't ever want to
351 * return -EIO here, instead we'll wait */
352
353 if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) &&
354 test_bit(R1BIO_Uptodate, &r1_bio->state)) {
355 /* Maybe we can return now */
356 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
357 struct bio *mbio = r1_bio->master_bio;
358 PRINTK(KERN_DEBUG "raid1: behind end write sectors %llu-%llu\n",
359 (unsigned long long) mbio->bi_sector,
360 (unsigned long long) mbio->bi_sector +
361 (mbio->bi_size >> 9) - 1);
362 bio_endio(mbio, mbio->bi_size, 0);
363 }
364 }
365 }
366 }
367 /*
368 *
369 * Let's see if all mirrored write operations have finished
370 * already.
371 */
372 if (atomic_dec_and_test(&r1_bio->remaining)) {
373 if (test_bit(R1BIO_BarrierRetry, &r1_bio->state)) {
374 reschedule_retry(r1_bio);
375 /* Don't dec_pending yet, we want to hold
376 * the reference over the retry
377 */
378 return 0;
379 }
380 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
381 /* free extra copy of the data pages */
382 int i = bio->bi_vcnt;
383 while (i--)
384 put_page(bio->bi_io_vec[i].bv_page);
385 }
386 /* clear the bitmap if all writes complete successfully */
387 bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector,
388 r1_bio->sectors,
389 !test_bit(R1BIO_Degraded, &r1_bio->state),
390 behind);
391 md_write_end(r1_bio->mddev);
392 raid_end_bio_io(r1_bio);
393 }
394
395 if (r1_bio->bios[mirror]==NULL)
396 bio_put(bio);
397
398 rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev);
399 return 0;
400 }
401
402
403 /*
404 * This routine returns the disk from which the requested read should
405 * be done. There is a per-array 'next expected sequential IO' sector
406 * number - if this matches on the next IO then we use the last disk.
407 * There is also a per-disk 'last know head position' sector that is
408 * maintained from IRQ contexts, both the normal and the resync IO
409 * completion handlers update this position correctly. If there is no
410 * perfect sequential match then we pick the disk whose head is closest.
411 *
412 * If there are 2 mirrors in the same 2 devices, performance degrades
413 * because position is mirror, not device based.
414 *
415 * The rdev for the device selected will have nr_pending incremented.
416 */
417 static int read_balance(conf_t *conf, r1bio_t *r1_bio)
418 {
419 const unsigned long this_sector = r1_bio->sector;
420 int new_disk = conf->last_used, disk = new_disk;
421 int wonly_disk = -1;
422 const int sectors = r1_bio->sectors;
423 sector_t new_distance, current_distance;
424 mdk_rdev_t *rdev;
425
426 rcu_read_lock();
427 /*
428 * Check if we can balance. We can balance on the whole
429 * device if no resync is going on, or below the resync window.
430 * We take the first readable disk when above the resync window.
431 */
432 retry:
433 if (conf->mddev->recovery_cp < MaxSector &&
434 (this_sector + sectors >= conf->next_resync)) {
435 /* Choose the first operation device, for consistancy */
436 new_disk = 0;
437
438 for (rdev = rcu_dereference(conf->mirrors[new_disk].rdev);
439 r1_bio->bios[new_disk] == IO_BLOCKED ||
440 !rdev || !test_bit(In_sync, &rdev->flags)
441 || test_bit(WriteMostly, &rdev->flags);
442 rdev = rcu_dereference(conf->mirrors[++new_disk].rdev)) {
443
444 if (rdev && test_bit(In_sync, &rdev->flags) &&
445 r1_bio->bios[new_disk] != IO_BLOCKED)
446 wonly_disk = new_disk;
447
448 if (new_disk == conf->raid_disks - 1) {
449 new_disk = wonly_disk;
450 break;
451 }
452 }
453 goto rb_out;
454 }
455
456
457 /* make sure the disk is operational */
458 for (rdev = rcu_dereference(conf->mirrors[new_disk].rdev);
459 r1_bio->bios[new_disk] == IO_BLOCKED ||
460 !rdev || !test_bit(In_sync, &rdev->flags) ||
461 test_bit(WriteMostly, &rdev->flags);
462 rdev = rcu_dereference(conf->mirrors[new_disk].rdev)) {
463
464 if (rdev && test_bit(In_sync, &rdev->flags) &&
465 r1_bio->bios[new_disk] != IO_BLOCKED)
466 wonly_disk = new_disk;
467
468 if (new_disk <= 0)
469 new_disk = conf->raid_disks;
470 new_disk--;
471 if (new_disk == disk) {
472 new_disk = wonly_disk;
473 break;
474 }
475 }
476
477 if (new_disk < 0)
478 goto rb_out;
479
480 disk = new_disk;
481 /* now disk == new_disk == starting point for search */
482
483 /*
484 * Don't change to another disk for sequential reads:
485 */
486 if (conf->next_seq_sect == this_sector)
487 goto rb_out;
488 if (this_sector == conf->mirrors[new_disk].head_position)
489 goto rb_out;
490
491 current_distance = abs(this_sector - conf->mirrors[disk].head_position);
492
493 /* Find the disk whose head is closest */
494
495 do {
496 if (disk <= 0)
497 disk = conf->raid_disks;
498 disk--;
499
500 rdev = rcu_dereference(conf->mirrors[disk].rdev);
501
502 if (!rdev || r1_bio->bios[disk] == IO_BLOCKED ||
503 !test_bit(In_sync, &rdev->flags) ||
504 test_bit(WriteMostly, &rdev->flags))
505 continue;
506
507 if (!atomic_read(&rdev->nr_pending)) {
508 new_disk = disk;
509 break;
510 }
511 new_distance = abs(this_sector - conf->mirrors[disk].head_position);
512 if (new_distance < current_distance) {
513 current_distance = new_distance;
514 new_disk = disk;
515 }
516 } while (disk != conf->last_used);
517
518 rb_out:
519
520
521 if (new_disk >= 0) {
522 rdev = rcu_dereference(conf->mirrors[new_disk].rdev);
523 if (!rdev)
524 goto retry;
525 atomic_inc(&rdev->nr_pending);
526 if (!test_bit(In_sync, &rdev->flags)) {
527 /* cannot risk returning a device that failed
528 * before we inc'ed nr_pending
529 */
530 atomic_dec(&rdev->nr_pending);
531 goto retry;
532 }
533 conf->next_seq_sect = this_sector + sectors;
534 conf->last_used = new_disk;
535 }
536 rcu_read_unlock();
537
538 return new_disk;
539 }
540
541 static void unplug_slaves(mddev_t *mddev)
542 {
543 conf_t *conf = mddev_to_conf(mddev);
544 int i;
545
546 rcu_read_lock();
547 for (i=0; i<mddev->raid_disks; i++) {
548 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
549 if (rdev && !test_bit(Faulty, &rdev->flags) && atomic_read(&rdev->nr_pending)) {
550 request_queue_t *r_queue = bdev_get_queue(rdev->bdev);
551
552 atomic_inc(&rdev->nr_pending);
553 rcu_read_unlock();
554
555 if (r_queue->unplug_fn)
556 r_queue->unplug_fn(r_queue);
557
558 rdev_dec_pending(rdev, mddev);
559 rcu_read_lock();
560 }
561 }
562 rcu_read_unlock();
563 }
564
565 static void raid1_unplug(request_queue_t *q)
566 {
567 mddev_t *mddev = q->queuedata;
568
569 unplug_slaves(mddev);
570 md_wakeup_thread(mddev->thread);
571 }
572
573 static int raid1_issue_flush(request_queue_t *q, struct gendisk *disk,
574 sector_t *error_sector)
575 {
576 mddev_t *mddev = q->queuedata;
577 conf_t *conf = mddev_to_conf(mddev);
578 int i, ret = 0;
579
580 rcu_read_lock();
581 for (i=0; i<mddev->raid_disks && ret == 0; i++) {
582 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
583 if (rdev && !test_bit(Faulty, &rdev->flags)) {
584 struct block_device *bdev = rdev->bdev;
585 request_queue_t *r_queue = bdev_get_queue(bdev);
586
587 if (!r_queue->issue_flush_fn)
588 ret = -EOPNOTSUPP;
589 else {
590 atomic_inc(&rdev->nr_pending);
591 rcu_read_unlock();
592 ret = r_queue->issue_flush_fn(r_queue, bdev->bd_disk,
593 error_sector);
594 rdev_dec_pending(rdev, mddev);
595 rcu_read_lock();
596 }
597 }
598 }
599 rcu_read_unlock();
600 return ret;
601 }
602
603 /* Barriers....
604 * Sometimes we need to suspend IO while we do something else,
605 * either some resync/recovery, or reconfigure the array.
606 * To do this we raise a 'barrier'.
607 * The 'barrier' is a counter that can be raised multiple times
608 * to count how many activities are happening which preclude
609 * normal IO.
610 * We can only raise the barrier if there is no pending IO.
611 * i.e. if nr_pending == 0.
612 * We choose only to raise the barrier if no-one is waiting for the
613 * barrier to go down. This means that as soon as an IO request
614 * is ready, no other operations which require a barrier will start
615 * until the IO request has had a chance.
616 *
617 * So: regular IO calls 'wait_barrier'. When that returns there
618 * is no backgroup IO happening, It must arrange to call
619 * allow_barrier when it has finished its IO.
620 * backgroup IO calls must call raise_barrier. Once that returns
621 * there is no normal IO happeing. It must arrange to call
622 * lower_barrier when the particular background IO completes.
623 */
624 #define RESYNC_DEPTH 32
625
626 static void raise_barrier(conf_t *conf)
627 {
628 spin_lock_irq(&conf->resync_lock);
629
630 /* Wait until no block IO is waiting */
631 wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting,
632 conf->resync_lock,
633 raid1_unplug(conf->mddev->queue));
634
635 /* block any new IO from starting */
636 conf->barrier++;
637
638 /* No wait for all pending IO to complete */
639 wait_event_lock_irq(conf->wait_barrier,
640 !conf->nr_pending && conf->barrier < RESYNC_DEPTH,
641 conf->resync_lock,
642 raid1_unplug(conf->mddev->queue));
643
644 spin_unlock_irq(&conf->resync_lock);
645 }
646
647 static void lower_barrier(conf_t *conf)
648 {
649 unsigned long flags;
650 spin_lock_irqsave(&conf->resync_lock, flags);
651 conf->barrier--;
652 spin_unlock_irqrestore(&conf->resync_lock, flags);
653 wake_up(&conf->wait_barrier);
654 }
655
656 static void wait_barrier(conf_t *conf)
657 {
658 spin_lock_irq(&conf->resync_lock);
659 if (conf->barrier) {
660 conf->nr_waiting++;
661 wait_event_lock_irq(conf->wait_barrier, !conf->barrier,
662 conf->resync_lock,
663 raid1_unplug(conf->mddev->queue));
664 conf->nr_waiting--;
665 }
666 conf->nr_pending++;
667 spin_unlock_irq(&conf->resync_lock);
668 }
669
670 static void allow_barrier(conf_t *conf)
671 {
672 unsigned long flags;
673 spin_lock_irqsave(&conf->resync_lock, flags);
674 conf->nr_pending--;
675 spin_unlock_irqrestore(&conf->resync_lock, flags);
676 wake_up(&conf->wait_barrier);
677 }
678
679 static void freeze_array(conf_t *conf)
680 {
681 /* stop syncio and normal IO and wait for everything to
682 * go quite.
683 * We increment barrier and nr_waiting, and then
684 * wait until barrier+nr_pending match nr_queued+2
685 */
686 spin_lock_irq(&conf->resync_lock);
687 conf->barrier++;
688 conf->nr_waiting++;
689 wait_event_lock_irq(conf->wait_barrier,
690 conf->barrier+conf->nr_pending == conf->nr_queued+2,
691 conf->resync_lock,
692 raid1_unplug(conf->mddev->queue));
693 spin_unlock_irq(&conf->resync_lock);
694 }
695 static void unfreeze_array(conf_t *conf)
696 {
697 /* reverse the effect of the freeze */
698 spin_lock_irq(&conf->resync_lock);
699 conf->barrier--;
700 conf->nr_waiting--;
701 wake_up(&conf->wait_barrier);
702 spin_unlock_irq(&conf->resync_lock);
703 }
704
705
706 /* duplicate the data pages for behind I/O */
707 static struct page **alloc_behind_pages(struct bio *bio)
708 {
709 int i;
710 struct bio_vec *bvec;
711 struct page **pages = kzalloc(bio->bi_vcnt * sizeof(struct page *),
712 GFP_NOIO);
713 if (unlikely(!pages))
714 goto do_sync_io;
715
716 bio_for_each_segment(bvec, bio, i) {
717 pages[i] = alloc_page(GFP_NOIO);
718 if (unlikely(!pages[i]))
719 goto do_sync_io;
720 memcpy(kmap(pages[i]) + bvec->bv_offset,
721 kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len);
722 kunmap(pages[i]);
723 kunmap(bvec->bv_page);
724 }
725
726 return pages;
727
728 do_sync_io:
729 if (pages)
730 for (i = 0; i < bio->bi_vcnt && pages[i]; i++)
731 put_page(pages[i]);
732 kfree(pages);
733 PRINTK("%dB behind alloc failed, doing sync I/O\n", bio->bi_size);
734 return NULL;
735 }
736
737 static int make_request(request_queue_t *q, struct bio * bio)
738 {
739 mddev_t *mddev = q->queuedata;
740 conf_t *conf = mddev_to_conf(mddev);
741 mirror_info_t *mirror;
742 r1bio_t *r1_bio;
743 struct bio *read_bio;
744 int i, targets = 0, disks;
745 mdk_rdev_t *rdev;
746 struct bitmap *bitmap = mddev->bitmap;
747 unsigned long flags;
748 struct bio_list bl;
749 struct page **behind_pages = NULL;
750 const int rw = bio_data_dir(bio);
751 int do_barriers;
752
753 if (unlikely(!mddev->barriers_work && bio_barrier(bio))) {
754 bio_endio(bio, bio->bi_size, -EOPNOTSUPP);
755 return 0;
756 }
757
758 /*
759 * Register the new request and wait if the reconstruction
760 * thread has put up a bar for new requests.
761 * Continue immediately if no resync is active currently.
762 */
763 md_write_start(mddev, bio); /* wait on superblock update early */
764
765 wait_barrier(conf);
766
767 disk_stat_inc(mddev->gendisk, ios[rw]);
768 disk_stat_add(mddev->gendisk, sectors[rw], bio_sectors(bio));
769
770 /*
771 * make_request() can abort the operation when READA is being
772 * used and no empty request is available.
773 *
774 */
775 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
776
777 r1_bio->master_bio = bio;
778 r1_bio->sectors = bio->bi_size >> 9;
779 r1_bio->state = 0;
780 r1_bio->mddev = mddev;
781 r1_bio->sector = bio->bi_sector;
782
783 if (rw == READ) {
784 /*
785 * read balancing logic:
786 */
787 int rdisk = read_balance(conf, r1_bio);
788
789 if (rdisk < 0) {
790 /* couldn't find anywhere to read from */
791 raid_end_bio_io(r1_bio);
792 return 0;
793 }
794 mirror = conf->mirrors + rdisk;
795
796 r1_bio->read_disk = rdisk;
797
798 read_bio = bio_clone(bio, GFP_NOIO);
799
800 r1_bio->bios[rdisk] = read_bio;
801
802 read_bio->bi_sector = r1_bio->sector + mirror->rdev->data_offset;
803 read_bio->bi_bdev = mirror->rdev->bdev;
804 read_bio->bi_end_io = raid1_end_read_request;
805 read_bio->bi_rw = READ;
806 read_bio->bi_private = r1_bio;
807
808 generic_make_request(read_bio);
809 return 0;
810 }
811
812 /*
813 * WRITE:
814 */
815 /* first select target devices under spinlock and
816 * inc refcount on their rdev. Record them by setting
817 * bios[x] to bio
818 */
819 disks = conf->raid_disks;
820 #if 0
821 { static int first=1;
822 if (first) printk("First Write sector %llu disks %d\n",
823 (unsigned long long)r1_bio->sector, disks);
824 first = 0;
825 }
826 #endif
827 rcu_read_lock();
828 for (i = 0; i < disks; i++) {
829 if ((rdev=rcu_dereference(conf->mirrors[i].rdev)) != NULL &&
830 !test_bit(Faulty, &rdev->flags)) {
831 atomic_inc(&rdev->nr_pending);
832 if (test_bit(Faulty, &rdev->flags)) {
833 atomic_dec(&rdev->nr_pending);
834 r1_bio->bios[i] = NULL;
835 } else
836 r1_bio->bios[i] = bio;
837 targets++;
838 } else
839 r1_bio->bios[i] = NULL;
840 }
841 rcu_read_unlock();
842
843 BUG_ON(targets == 0); /* we never fail the last device */
844
845 if (targets < conf->raid_disks) {
846 /* array is degraded, we will not clear the bitmap
847 * on I/O completion (see raid1_end_write_request) */
848 set_bit(R1BIO_Degraded, &r1_bio->state);
849 }
850
851 /* do behind I/O ? */
852 if (bitmap &&
853 atomic_read(&bitmap->behind_writes) < bitmap->max_write_behind &&
854 (behind_pages = alloc_behind_pages(bio)) != NULL)
855 set_bit(R1BIO_BehindIO, &r1_bio->state);
856
857 atomic_set(&r1_bio->remaining, 0);
858 atomic_set(&r1_bio->behind_remaining, 0);
859
860 do_barriers = bio->bi_rw & BIO_RW_BARRIER;
861 if (do_barriers)
862 set_bit(R1BIO_Barrier, &r1_bio->state);
863
864 bio_list_init(&bl);
865 for (i = 0; i < disks; i++) {
866 struct bio *mbio;
867 if (!r1_bio->bios[i])
868 continue;
869
870 mbio = bio_clone(bio, GFP_NOIO);
871 r1_bio->bios[i] = mbio;
872
873 mbio->bi_sector = r1_bio->sector + conf->mirrors[i].rdev->data_offset;
874 mbio->bi_bdev = conf->mirrors[i].rdev->bdev;
875 mbio->bi_end_io = raid1_end_write_request;
876 mbio->bi_rw = WRITE | do_barriers;
877 mbio->bi_private = r1_bio;
878
879 if (behind_pages) {
880 struct bio_vec *bvec;
881 int j;
882
883 /* Yes, I really want the '__' version so that
884 * we clear any unused pointer in the io_vec, rather
885 * than leave them unchanged. This is important
886 * because when we come to free the pages, we won't
887 * know the originial bi_idx, so we just free
888 * them all
889 */
890 __bio_for_each_segment(bvec, mbio, j, 0)
891 bvec->bv_page = behind_pages[j];
892 if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags))
893 atomic_inc(&r1_bio->behind_remaining);
894 }
895
896 atomic_inc(&r1_bio->remaining);
897
898 bio_list_add(&bl, mbio);
899 }
900 kfree(behind_pages); /* the behind pages are attached to the bios now */
901
902 bitmap_startwrite(bitmap, bio->bi_sector, r1_bio->sectors,
903 test_bit(R1BIO_BehindIO, &r1_bio->state));
904 spin_lock_irqsave(&conf->device_lock, flags);
905 bio_list_merge(&conf->pending_bio_list, &bl);
906 bio_list_init(&bl);
907
908 blk_plug_device(mddev->queue);
909 spin_unlock_irqrestore(&conf->device_lock, flags);
910
911 #if 0
912 while ((bio = bio_list_pop(&bl)) != NULL)
913 generic_make_request(bio);
914 #endif
915
916 return 0;
917 }
918
919 static void status(struct seq_file *seq, mddev_t *mddev)
920 {
921 conf_t *conf = mddev_to_conf(mddev);
922 int i;
923
924 seq_printf(seq, " [%d/%d] [", conf->raid_disks,
925 conf->working_disks);
926 for (i = 0; i < conf->raid_disks; i++)
927 seq_printf(seq, "%s",
928 conf->mirrors[i].rdev &&
929 test_bit(In_sync, &conf->mirrors[i].rdev->flags) ? "U" : "_");
930 seq_printf(seq, "]");
931 }
932
933
934 static void error(mddev_t *mddev, mdk_rdev_t *rdev)
935 {
936 char b[BDEVNAME_SIZE];
937 conf_t *conf = mddev_to_conf(mddev);
938
939 /*
940 * If it is not operational, then we have already marked it as dead
941 * else if it is the last working disks, ignore the error, let the
942 * next level up know.
943 * else mark the drive as failed
944 */
945 if (test_bit(In_sync, &rdev->flags)
946 && conf->working_disks == 1)
947 /*
948 * Don't fail the drive, act as though we were just a
949 * normal single drive
950 */
951 return;
952 if (test_bit(In_sync, &rdev->flags)) {
953 mddev->degraded++;
954 conf->working_disks--;
955 /*
956 * if recovery is running, make sure it aborts.
957 */
958 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
959 }
960 clear_bit(In_sync, &rdev->flags);
961 set_bit(Faulty, &rdev->flags);
962 mddev->sb_dirty = 1;
963 printk(KERN_ALERT "raid1: Disk failure on %s, disabling device. \n"
964 " Operation continuing on %d devices\n",
965 bdevname(rdev->bdev,b), conf->working_disks);
966 }
967
968 static void print_conf(conf_t *conf)
969 {
970 int i;
971 mirror_info_t *tmp;
972
973 printk("RAID1 conf printout:\n");
974 if (!conf) {
975 printk("(!conf)\n");
976 return;
977 }
978 printk(" --- wd:%d rd:%d\n", conf->working_disks,
979 conf->raid_disks);
980
981 for (i = 0; i < conf->raid_disks; i++) {
982 char b[BDEVNAME_SIZE];
983 tmp = conf->mirrors + i;
984 if (tmp->rdev)
985 printk(" disk %d, wo:%d, o:%d, dev:%s\n",
986 i, !test_bit(In_sync, &tmp->rdev->flags), !test_bit(Faulty, &tmp->rdev->flags),
987 bdevname(tmp->rdev->bdev,b));
988 }
989 }
990
991 static void close_sync(conf_t *conf)
992 {
993 wait_barrier(conf);
994 allow_barrier(conf);
995
996 mempool_destroy(conf->r1buf_pool);
997 conf->r1buf_pool = NULL;
998 }
999
1000 static int raid1_spare_active(mddev_t *mddev)
1001 {
1002 int i;
1003 conf_t *conf = mddev->private;
1004 mirror_info_t *tmp;
1005
1006 /*
1007 * Find all failed disks within the RAID1 configuration
1008 * and mark them readable
1009 */
1010 for (i = 0; i < conf->raid_disks; i++) {
1011 tmp = conf->mirrors + i;
1012 if (tmp->rdev
1013 && !test_bit(Faulty, &tmp->rdev->flags)
1014 && !test_bit(In_sync, &tmp->rdev->flags)) {
1015 conf->working_disks++;
1016 mddev->degraded--;
1017 set_bit(In_sync, &tmp->rdev->flags);
1018 }
1019 }
1020
1021 print_conf(conf);
1022 return 0;
1023 }
1024
1025
1026 static int raid1_add_disk(mddev_t *mddev, mdk_rdev_t *rdev)
1027 {
1028 conf_t *conf = mddev->private;
1029 int found = 0;
1030 int mirror = 0;
1031 mirror_info_t *p;
1032
1033 for (mirror=0; mirror < mddev->raid_disks; mirror++)
1034 if ( !(p=conf->mirrors+mirror)->rdev) {
1035
1036 blk_queue_stack_limits(mddev->queue,
1037 rdev->bdev->bd_disk->queue);
1038 /* as we don't honour merge_bvec_fn, we must never risk
1039 * violating it, so limit ->max_sector to one PAGE, as
1040 * a one page request is never in violation.
1041 */
1042 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
1043 mddev->queue->max_sectors > (PAGE_SIZE>>9))
1044 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
1045
1046 p->head_position = 0;
1047 rdev->raid_disk = mirror;
1048 found = 1;
1049 /* As all devices are equivalent, we don't need a full recovery
1050 * if this was recently any drive of the array
1051 */
1052 if (rdev->saved_raid_disk < 0)
1053 conf->fullsync = 1;
1054 rcu_assign_pointer(p->rdev, rdev);
1055 break;
1056 }
1057
1058 print_conf(conf);
1059 return found;
1060 }
1061
1062 static int raid1_remove_disk(mddev_t *mddev, int number)
1063 {
1064 conf_t *conf = mddev->private;
1065 int err = 0;
1066 mdk_rdev_t *rdev;
1067 mirror_info_t *p = conf->mirrors+ number;
1068
1069 print_conf(conf);
1070 rdev = p->rdev;
1071 if (rdev) {
1072 if (test_bit(In_sync, &rdev->flags) ||
1073 atomic_read(&rdev->nr_pending)) {
1074 err = -EBUSY;
1075 goto abort;
1076 }
1077 p->rdev = NULL;
1078 synchronize_rcu();
1079 if (atomic_read(&rdev->nr_pending)) {
1080 /* lost the race, try later */
1081 err = -EBUSY;
1082 p->rdev = rdev;
1083 }
1084 }
1085 abort:
1086
1087 print_conf(conf);
1088 return err;
1089 }
1090
1091
1092 static int end_sync_read(struct bio *bio, unsigned int bytes_done, int error)
1093 {
1094 r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
1095 int i;
1096
1097 if (bio->bi_size)
1098 return 1;
1099
1100 for (i=r1_bio->mddev->raid_disks; i--; )
1101 if (r1_bio->bios[i] == bio)
1102 break;
1103 BUG_ON(i < 0);
1104 update_head_pos(i, r1_bio);
1105 /*
1106 * we have read a block, now it needs to be re-written,
1107 * or re-read if the read failed.
1108 * We don't do much here, just schedule handling by raid1d
1109 */
1110 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
1111 set_bit(R1BIO_Uptodate, &r1_bio->state);
1112
1113 if (atomic_dec_and_test(&r1_bio->remaining))
1114 reschedule_retry(r1_bio);
1115 return 0;
1116 }
1117
1118 static int end_sync_write(struct bio *bio, unsigned int bytes_done, int error)
1119 {
1120 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1121 r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
1122 mddev_t *mddev = r1_bio->mddev;
1123 conf_t *conf = mddev_to_conf(mddev);
1124 int i;
1125 int mirror=0;
1126
1127 if (bio->bi_size)
1128 return 1;
1129
1130 for (i = 0; i < conf->raid_disks; i++)
1131 if (r1_bio->bios[i] == bio) {
1132 mirror = i;
1133 break;
1134 }
1135 if (!uptodate)
1136 md_error(mddev, conf->mirrors[mirror].rdev);
1137
1138 update_head_pos(mirror, r1_bio);
1139
1140 if (atomic_dec_and_test(&r1_bio->remaining)) {
1141 md_done_sync(mddev, r1_bio->sectors, uptodate);
1142 put_buf(r1_bio);
1143 }
1144 return 0;
1145 }
1146
1147 static void sync_request_write(mddev_t *mddev, r1bio_t *r1_bio)
1148 {
1149 conf_t *conf = mddev_to_conf(mddev);
1150 int i;
1151 int disks = conf->raid_disks;
1152 struct bio *bio, *wbio;
1153
1154 bio = r1_bio->bios[r1_bio->read_disk];
1155
1156
1157 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
1158 /* We have read all readable devices. If we haven't
1159 * got the block, then there is no hope left.
1160 * If we have, then we want to do a comparison
1161 * and skip the write if everything is the same.
1162 * If any blocks failed to read, then we need to
1163 * attempt an over-write
1164 */
1165 int primary;
1166 if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) {
1167 for (i=0; i<mddev->raid_disks; i++)
1168 if (r1_bio->bios[i]->bi_end_io == end_sync_read)
1169 md_error(mddev, conf->mirrors[i].rdev);
1170
1171 md_done_sync(mddev, r1_bio->sectors, 1);
1172 put_buf(r1_bio);
1173 return;
1174 }
1175 for (primary=0; primary<mddev->raid_disks; primary++)
1176 if (r1_bio->bios[primary]->bi_end_io == end_sync_read &&
1177 test_bit(BIO_UPTODATE, &r1_bio->bios[primary]->bi_flags)) {
1178 r1_bio->bios[primary]->bi_end_io = NULL;
1179 break;
1180 }
1181 r1_bio->read_disk = primary;
1182 for (i=0; i<mddev->raid_disks; i++)
1183 if (r1_bio->bios[i]->bi_end_io == end_sync_read &&
1184 test_bit(BIO_UPTODATE, &r1_bio->bios[i]->bi_flags)) {
1185 int j;
1186 int vcnt = r1_bio->sectors >> (PAGE_SHIFT- 9);
1187 struct bio *pbio = r1_bio->bios[primary];
1188 struct bio *sbio = r1_bio->bios[i];
1189 for (j = vcnt; j-- ; )
1190 if (memcmp(page_address(pbio->bi_io_vec[j].bv_page),
1191 page_address(sbio->bi_io_vec[j].bv_page),
1192 PAGE_SIZE))
1193 break;
1194 if (j >= 0)
1195 mddev->resync_mismatches += r1_bio->sectors;
1196 if (j < 0 || test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
1197 sbio->bi_end_io = NULL;
1198 else {
1199 /* fixup the bio for reuse */
1200 sbio->bi_vcnt = vcnt;
1201 sbio->bi_size = r1_bio->sectors << 9;
1202 sbio->bi_idx = 0;
1203 sbio->bi_phys_segments = 0;
1204 sbio->bi_hw_segments = 0;
1205 sbio->bi_hw_front_size = 0;
1206 sbio->bi_hw_back_size = 0;
1207 sbio->bi_flags &= ~(BIO_POOL_MASK - 1);
1208 sbio->bi_flags |= 1 << BIO_UPTODATE;
1209 sbio->bi_next = NULL;
1210 sbio->bi_sector = r1_bio->sector +
1211 conf->mirrors[i].rdev->data_offset;
1212 sbio->bi_bdev = conf->mirrors[i].rdev->bdev;
1213 }
1214 }
1215 }
1216 if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) {
1217 /* ouch - failed to read all of that.
1218 * Try some synchronous reads of other devices to get
1219 * good data, much like with normal read errors. Only
1220 * read into the pages we already have so they we don't
1221 * need to re-issue the read request.
1222 * We don't need to freeze the array, because being in an
1223 * active sync request, there is no normal IO, and
1224 * no overlapping syncs.
1225 */
1226 sector_t sect = r1_bio->sector;
1227 int sectors = r1_bio->sectors;
1228 int idx = 0;
1229
1230 while(sectors) {
1231 int s = sectors;
1232 int d = r1_bio->read_disk;
1233 int success = 0;
1234 mdk_rdev_t *rdev;
1235
1236 if (s > (PAGE_SIZE>>9))
1237 s = PAGE_SIZE >> 9;
1238 do {
1239 if (r1_bio->bios[d]->bi_end_io == end_sync_read) {
1240 rdev = conf->mirrors[d].rdev;
1241 if (sync_page_io(rdev->bdev,
1242 sect + rdev->data_offset,
1243 s<<9,
1244 bio->bi_io_vec[idx].bv_page,
1245 READ)) {
1246 success = 1;
1247 break;
1248 }
1249 }
1250 d++;
1251 if (d == conf->raid_disks)
1252 d = 0;
1253 } while (!success && d != r1_bio->read_disk);
1254
1255 if (success) {
1256 /* write it back and re-read */
1257 set_bit(R1BIO_Uptodate, &r1_bio->state);
1258 while (d != r1_bio->read_disk) {
1259 if (d == 0)
1260 d = conf->raid_disks;
1261 d--;
1262 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1263 continue;
1264 rdev = conf->mirrors[d].rdev;
1265 if (sync_page_io(rdev->bdev,
1266 sect + rdev->data_offset,
1267 s<<9,
1268 bio->bi_io_vec[idx].bv_page,
1269 WRITE) == 0 ||
1270 sync_page_io(rdev->bdev,
1271 sect + rdev->data_offset,
1272 s<<9,
1273 bio->bi_io_vec[idx].bv_page,
1274 READ) == 0) {
1275 md_error(mddev, rdev);
1276 }
1277 }
1278 } else {
1279 char b[BDEVNAME_SIZE];
1280 /* Cannot read from anywhere, array is toast */
1281 md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev);
1282 printk(KERN_ALERT "raid1: %s: unrecoverable I/O read error"
1283 " for block %llu\n",
1284 bdevname(bio->bi_bdev,b),
1285 (unsigned long long)r1_bio->sector);
1286 md_done_sync(mddev, r1_bio->sectors, 0);
1287 put_buf(r1_bio);
1288 return;
1289 }
1290 sectors -= s;
1291 sect += s;
1292 idx ++;
1293 }
1294 }
1295
1296 /*
1297 * schedule writes
1298 */
1299 atomic_set(&r1_bio->remaining, 1);
1300 for (i = 0; i < disks ; i++) {
1301 wbio = r1_bio->bios[i];
1302 if (wbio->bi_end_io == NULL ||
1303 (wbio->bi_end_io == end_sync_read &&
1304 (i == r1_bio->read_disk ||
1305 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery))))
1306 continue;
1307
1308 wbio->bi_rw = WRITE;
1309 wbio->bi_end_io = end_sync_write;
1310 atomic_inc(&r1_bio->remaining);
1311 md_sync_acct(conf->mirrors[i].rdev->bdev, wbio->bi_size >> 9);
1312
1313 generic_make_request(wbio);
1314 }
1315
1316 if (atomic_dec_and_test(&r1_bio->remaining)) {
1317 /* if we're here, all write(s) have completed, so clean up */
1318 md_done_sync(mddev, r1_bio->sectors, 1);
1319 put_buf(r1_bio);
1320 }
1321 }
1322
1323 /*
1324 * This is a kernel thread which:
1325 *
1326 * 1. Retries failed read operations on working mirrors.
1327 * 2. Updates the raid superblock when problems encounter.
1328 * 3. Performs writes following reads for array syncronising.
1329 */
1330
1331 static void raid1d(mddev_t *mddev)
1332 {
1333 r1bio_t *r1_bio;
1334 struct bio *bio;
1335 unsigned long flags;
1336 conf_t *conf = mddev_to_conf(mddev);
1337 struct list_head *head = &conf->retry_list;
1338 int unplug=0;
1339 mdk_rdev_t *rdev;
1340
1341 md_check_recovery(mddev);
1342
1343 for (;;) {
1344 char b[BDEVNAME_SIZE];
1345 spin_lock_irqsave(&conf->device_lock, flags);
1346
1347 if (conf->pending_bio_list.head) {
1348 bio = bio_list_get(&conf->pending_bio_list);
1349 blk_remove_plug(mddev->queue);
1350 spin_unlock_irqrestore(&conf->device_lock, flags);
1351 /* flush any pending bitmap writes to disk before proceeding w/ I/O */
1352 if (bitmap_unplug(mddev->bitmap) != 0)
1353 printk("%s: bitmap file write failed!\n", mdname(mddev));
1354
1355 while (bio) { /* submit pending writes */
1356 struct bio *next = bio->bi_next;
1357 bio->bi_next = NULL;
1358 generic_make_request(bio);
1359 bio = next;
1360 }
1361 unplug = 1;
1362
1363 continue;
1364 }
1365
1366 if (list_empty(head))
1367 break;
1368 r1_bio = list_entry(head->prev, r1bio_t, retry_list);
1369 list_del(head->prev);
1370 conf->nr_queued--;
1371 spin_unlock_irqrestore(&conf->device_lock, flags);
1372
1373 mddev = r1_bio->mddev;
1374 conf = mddev_to_conf(mddev);
1375 if (test_bit(R1BIO_IsSync, &r1_bio->state)) {
1376 sync_request_write(mddev, r1_bio);
1377 unplug = 1;
1378 } else if (test_bit(R1BIO_BarrierRetry, &r1_bio->state)) {
1379 /* some requests in the r1bio were BIO_RW_BARRIER
1380 * requests which failed with -ENOTSUPP. Hohumm..
1381 * Better resubmit without the barrier.
1382 * We know which devices to resubmit for, because
1383 * all others have had their bios[] entry cleared.
1384 */
1385 int i;
1386 clear_bit(R1BIO_BarrierRetry, &r1_bio->state);
1387 clear_bit(R1BIO_Barrier, &r1_bio->state);
1388 for (i=0; i < conf->raid_disks; i++)
1389 if (r1_bio->bios[i]) {
1390 struct bio_vec *bvec;
1391 int j;
1392
1393 bio = bio_clone(r1_bio->master_bio, GFP_NOIO);
1394 /* copy pages from the failed bio, as
1395 * this might be a write-behind device */
1396 __bio_for_each_segment(bvec, bio, j, 0)
1397 bvec->bv_page = bio_iovec_idx(r1_bio->bios[i], j)->bv_page;
1398 bio_put(r1_bio->bios[i]);
1399 bio->bi_sector = r1_bio->sector +
1400 conf->mirrors[i].rdev->data_offset;
1401 bio->bi_bdev = conf->mirrors[i].rdev->bdev;
1402 bio->bi_end_io = raid1_end_write_request;
1403 bio->bi_rw = WRITE;
1404 bio->bi_private = r1_bio;
1405 r1_bio->bios[i] = bio;
1406 generic_make_request(bio);
1407 }
1408 } else {
1409 int disk;
1410
1411 /* we got a read error. Maybe the drive is bad. Maybe just
1412 * the block and we can fix it.
1413 * We freeze all other IO, and try reading the block from
1414 * other devices. When we find one, we re-write
1415 * and check it that fixes the read error.
1416 * This is all done synchronously while the array is
1417 * frozen
1418 */
1419 sector_t sect = r1_bio->sector;
1420 int sectors = r1_bio->sectors;
1421 freeze_array(conf);
1422 if (mddev->ro == 0) while(sectors) {
1423 int s = sectors;
1424 int d = r1_bio->read_disk;
1425 int success = 0;
1426
1427 if (s > (PAGE_SIZE>>9))
1428 s = PAGE_SIZE >> 9;
1429
1430 do {
1431 rdev = conf->mirrors[d].rdev;
1432 if (rdev &&
1433 test_bit(In_sync, &rdev->flags) &&
1434 sync_page_io(rdev->bdev,
1435 sect + rdev->data_offset,
1436 s<<9,
1437 conf->tmppage, READ))
1438 success = 1;
1439 else {
1440 d++;
1441 if (d == conf->raid_disks)
1442 d = 0;
1443 }
1444 } while (!success && d != r1_bio->read_disk);
1445
1446 if (success) {
1447 /* write it back and re-read */
1448 while (d != r1_bio->read_disk) {
1449 if (d==0)
1450 d = conf->raid_disks;
1451 d--;
1452 rdev = conf->mirrors[d].rdev;
1453 if (rdev &&
1454 test_bit(In_sync, &rdev->flags)) {
1455 if (sync_page_io(rdev->bdev,
1456 sect + rdev->data_offset,
1457 s<<9, conf->tmppage, WRITE) == 0 ||
1458 sync_page_io(rdev->bdev,
1459 sect + rdev->data_offset,
1460 s<<9, conf->tmppage, READ) == 0) {
1461 /* Well, this device is dead */
1462 md_error(mddev, rdev);
1463 }
1464 }
1465 }
1466 } else {
1467 /* Cannot read from anywhere -- bye bye array */
1468 md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev);
1469 break;
1470 }
1471 sectors -= s;
1472 sect += s;
1473 }
1474
1475 unfreeze_array(conf);
1476
1477 bio = r1_bio->bios[r1_bio->read_disk];
1478 if ((disk=read_balance(conf, r1_bio)) == -1) {
1479 printk(KERN_ALERT "raid1: %s: unrecoverable I/O"
1480 " read error for block %llu\n",
1481 bdevname(bio->bi_bdev,b),
1482 (unsigned long long)r1_bio->sector);
1483 raid_end_bio_io(r1_bio);
1484 } else {
1485 r1_bio->bios[r1_bio->read_disk] =
1486 mddev->ro ? IO_BLOCKED : NULL;
1487 r1_bio->read_disk = disk;
1488 bio_put(bio);
1489 bio = bio_clone(r1_bio->master_bio, GFP_NOIO);
1490 r1_bio->bios[r1_bio->read_disk] = bio;
1491 rdev = conf->mirrors[disk].rdev;
1492 if (printk_ratelimit())
1493 printk(KERN_ERR "raid1: %s: redirecting sector %llu to"
1494 " another mirror\n",
1495 bdevname(rdev->bdev,b),
1496 (unsigned long long)r1_bio->sector);
1497 bio->bi_sector = r1_bio->sector + rdev->data_offset;
1498 bio->bi_bdev = rdev->bdev;
1499 bio->bi_end_io = raid1_end_read_request;
1500 bio->bi_rw = READ;
1501 bio->bi_private = r1_bio;
1502 unplug = 1;
1503 generic_make_request(bio);
1504 }
1505 }
1506 }
1507 spin_unlock_irqrestore(&conf->device_lock, flags);
1508 if (unplug)
1509 unplug_slaves(mddev);
1510 }
1511
1512
1513 static int init_resync(conf_t *conf)
1514 {
1515 int buffs;
1516
1517 buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;
1518 if (conf->r1buf_pool)
1519 BUG();
1520 conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free,
1521 conf->poolinfo);
1522 if (!conf->r1buf_pool)
1523 return -ENOMEM;
1524 conf->next_resync = 0;
1525 return 0;
1526 }
1527
1528 /*
1529 * perform a "sync" on one "block"
1530 *
1531 * We need to make sure that no normal I/O request - particularly write
1532 * requests - conflict with active sync requests.
1533 *
1534 * This is achieved by tracking pending requests and a 'barrier' concept
1535 * that can be installed to exclude normal IO requests.
1536 */
1537
1538 static sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster)
1539 {
1540 conf_t *conf = mddev_to_conf(mddev);
1541 r1bio_t *r1_bio;
1542 struct bio *bio;
1543 sector_t max_sector, nr_sectors;
1544 int disk = -1;
1545 int i;
1546 int wonly = -1;
1547 int write_targets = 0, read_targets = 0;
1548 int sync_blocks;
1549 int still_degraded = 0;
1550
1551 if (!conf->r1buf_pool)
1552 {
1553 /*
1554 printk("sync start - bitmap %p\n", mddev->bitmap);
1555 */
1556 if (init_resync(conf))
1557 return 0;
1558 }
1559
1560 max_sector = mddev->size << 1;
1561 if (sector_nr >= max_sector) {
1562 /* If we aborted, we need to abort the
1563 * sync on the 'current' bitmap chunk (there will
1564 * only be one in raid1 resync.
1565 * We can find the current addess in mddev->curr_resync
1566 */
1567 if (mddev->curr_resync < max_sector) /* aborted */
1568 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
1569 &sync_blocks, 1);
1570 else /* completed sync */
1571 conf->fullsync = 0;
1572
1573 bitmap_close_sync(mddev->bitmap);
1574 close_sync(conf);
1575 return 0;
1576 }
1577
1578 /* before building a request, check if we can skip these blocks..
1579 * This call the bitmap_start_sync doesn't actually record anything
1580 */
1581 if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
1582 !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
1583 /* We can skip this block, and probably several more */
1584 *skipped = 1;
1585 return sync_blocks;
1586 }
1587 /*
1588 * If there is non-resync activity waiting for a turn,
1589 * and resync is going fast enough,
1590 * then let it though before starting on this new sync request.
1591 */
1592 if (!go_faster && conf->nr_waiting)
1593 msleep_interruptible(1000);
1594
1595 raise_barrier(conf);
1596
1597 conf->next_resync = sector_nr;
1598
1599 r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO);
1600 rcu_read_lock();
1601 /*
1602 * If we get a correctably read error during resync or recovery,
1603 * we might want to read from a different device. So we
1604 * flag all drives that could conceivably be read from for READ,
1605 * and any others (which will be non-In_sync devices) for WRITE.
1606 * If a read fails, we try reading from something else for which READ
1607 * is OK.
1608 */
1609
1610 r1_bio->mddev = mddev;
1611 r1_bio->sector = sector_nr;
1612 r1_bio->state = 0;
1613 set_bit(R1BIO_IsSync, &r1_bio->state);
1614
1615 for (i=0; i < conf->raid_disks; i++) {
1616 mdk_rdev_t *rdev;
1617 bio = r1_bio->bios[i];
1618
1619 /* take from bio_init */
1620 bio->bi_next = NULL;
1621 bio->bi_flags |= 1 << BIO_UPTODATE;
1622 bio->bi_rw = 0;
1623 bio->bi_vcnt = 0;
1624 bio->bi_idx = 0;
1625 bio->bi_phys_segments = 0;
1626 bio->bi_hw_segments = 0;
1627 bio->bi_size = 0;
1628 bio->bi_end_io = NULL;
1629 bio->bi_private = NULL;
1630
1631 rdev = rcu_dereference(conf->mirrors[i].rdev);
1632 if (rdev == NULL ||
1633 test_bit(Faulty, &rdev->flags)) {
1634 still_degraded = 1;
1635 continue;
1636 } else if (!test_bit(In_sync, &rdev->flags)) {
1637 bio->bi_rw = WRITE;
1638 bio->bi_end_io = end_sync_write;
1639 write_targets ++;
1640 } else {
1641 /* may need to read from here */
1642 bio->bi_rw = READ;
1643 bio->bi_end_io = end_sync_read;
1644 if (test_bit(WriteMostly, &rdev->flags)) {
1645 if (wonly < 0)
1646 wonly = i;
1647 } else {
1648 if (disk < 0)
1649 disk = i;
1650 }
1651 read_targets++;
1652 }
1653 atomic_inc(&rdev->nr_pending);
1654 bio->bi_sector = sector_nr + rdev->data_offset;
1655 bio->bi_bdev = rdev->bdev;
1656 bio->bi_private = r1_bio;
1657 }
1658 rcu_read_unlock();
1659 if (disk < 0)
1660 disk = wonly;
1661 r1_bio->read_disk = disk;
1662
1663 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0)
1664 /* extra read targets are also write targets */
1665 write_targets += read_targets-1;
1666
1667 if (write_targets == 0 || read_targets == 0) {
1668 /* There is nowhere to write, so all non-sync
1669 * drives must be failed - so we are finished
1670 */
1671 sector_t rv = max_sector - sector_nr;
1672 *skipped = 1;
1673 put_buf(r1_bio);
1674 return rv;
1675 }
1676
1677 nr_sectors = 0;
1678 sync_blocks = 0;
1679 do {
1680 struct page *page;
1681 int len = PAGE_SIZE;
1682 if (sector_nr + (len>>9) > max_sector)
1683 len = (max_sector - sector_nr) << 9;
1684 if (len == 0)
1685 break;
1686 if (sync_blocks == 0) {
1687 if (!bitmap_start_sync(mddev->bitmap, sector_nr,
1688 &sync_blocks, still_degraded) &&
1689 !conf->fullsync &&
1690 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1691 break;
1692 if (sync_blocks < (PAGE_SIZE>>9))
1693 BUG();
1694 if (len > (sync_blocks<<9))
1695 len = sync_blocks<<9;
1696 }
1697
1698 for (i=0 ; i < conf->raid_disks; i++) {
1699 bio = r1_bio->bios[i];
1700 if (bio->bi_end_io) {
1701 page = bio->bi_io_vec[bio->bi_vcnt].bv_page;
1702 if (bio_add_page(bio, page, len, 0) == 0) {
1703 /* stop here */
1704 bio->bi_io_vec[bio->bi_vcnt].bv_page = page;
1705 while (i > 0) {
1706 i--;
1707 bio = r1_bio->bios[i];
1708 if (bio->bi_end_io==NULL)
1709 continue;
1710 /* remove last page from this bio */
1711 bio->bi_vcnt--;
1712 bio->bi_size -= len;
1713 bio->bi_flags &= ~(1<< BIO_SEG_VALID);
1714 }
1715 goto bio_full;
1716 }
1717 }
1718 }
1719 nr_sectors += len>>9;
1720 sector_nr += len>>9;
1721 sync_blocks -= (len>>9);
1722 } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES);
1723 bio_full:
1724 r1_bio->sectors = nr_sectors;
1725
1726 /* For a user-requested sync, we read all readable devices and do a
1727 * compare
1728 */
1729 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
1730 atomic_set(&r1_bio->remaining, read_targets);
1731 for (i=0; i<conf->raid_disks; i++) {
1732 bio = r1_bio->bios[i];
1733 if (bio->bi_end_io == end_sync_read) {
1734 md_sync_acct(conf->mirrors[i].rdev->bdev, nr_sectors);
1735 generic_make_request(bio);
1736 }
1737 }
1738 } else {
1739 atomic_set(&r1_bio->remaining, 1);
1740 bio = r1_bio->bios[r1_bio->read_disk];
1741 md_sync_acct(conf->mirrors[r1_bio->read_disk].rdev->bdev,
1742 nr_sectors);
1743 generic_make_request(bio);
1744
1745 }
1746
1747 return nr_sectors;
1748 }
1749
1750 static int run(mddev_t *mddev)
1751 {
1752 conf_t *conf;
1753 int i, j, disk_idx;
1754 mirror_info_t *disk;
1755 mdk_rdev_t *rdev;
1756 struct list_head *tmp;
1757
1758 if (mddev->level != 1) {
1759 printk("raid1: %s: raid level not set to mirroring (%d)\n",
1760 mdname(mddev), mddev->level);
1761 goto out;
1762 }
1763 /*
1764 * copy the already verified devices into our private RAID1
1765 * bookkeeping area. [whatever we allocate in run(),
1766 * should be freed in stop()]
1767 */
1768 conf = kzalloc(sizeof(conf_t), GFP_KERNEL);
1769 mddev->private = conf;
1770 if (!conf)
1771 goto out_no_mem;
1772
1773 conf->mirrors = kzalloc(sizeof(struct mirror_info)*mddev->raid_disks,
1774 GFP_KERNEL);
1775 if (!conf->mirrors)
1776 goto out_no_mem;
1777
1778 conf->tmppage = alloc_page(GFP_KERNEL);
1779 if (!conf->tmppage)
1780 goto out_no_mem;
1781
1782 conf->poolinfo = kmalloc(sizeof(*conf->poolinfo), GFP_KERNEL);
1783 if (!conf->poolinfo)
1784 goto out_no_mem;
1785 conf->poolinfo->mddev = mddev;
1786 conf->poolinfo->raid_disks = mddev->raid_disks;
1787 conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
1788 r1bio_pool_free,
1789 conf->poolinfo);
1790 if (!conf->r1bio_pool)
1791 goto out_no_mem;
1792
1793 ITERATE_RDEV(mddev, rdev, tmp) {
1794 disk_idx = rdev->raid_disk;
1795 if (disk_idx >= mddev->raid_disks
1796 || disk_idx < 0)
1797 continue;
1798 disk = conf->mirrors + disk_idx;
1799
1800 disk->rdev = rdev;
1801
1802 blk_queue_stack_limits(mddev->queue,
1803 rdev->bdev->bd_disk->queue);
1804 /* as we don't honour merge_bvec_fn, we must never risk
1805 * violating it, so limit ->max_sector to one PAGE, as
1806 * a one page request is never in violation.
1807 */
1808 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
1809 mddev->queue->max_sectors > (PAGE_SIZE>>9))
1810 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
1811
1812 disk->head_position = 0;
1813 if (!test_bit(Faulty, &rdev->flags) && test_bit(In_sync, &rdev->flags))
1814 conf->working_disks++;
1815 }
1816 conf->raid_disks = mddev->raid_disks;
1817 conf->mddev = mddev;
1818 spin_lock_init(&conf->device_lock);
1819 INIT_LIST_HEAD(&conf->retry_list);
1820 if (conf->working_disks == 1)
1821 mddev->recovery_cp = MaxSector;
1822
1823 spin_lock_init(&conf->resync_lock);
1824 init_waitqueue_head(&conf->wait_barrier);
1825
1826 bio_list_init(&conf->pending_bio_list);
1827 bio_list_init(&conf->flushing_bio_list);
1828
1829 if (!conf->working_disks) {
1830 printk(KERN_ERR "raid1: no operational mirrors for %s\n",
1831 mdname(mddev));
1832 goto out_free_conf;
1833 }
1834
1835 mddev->degraded = 0;
1836 for (i = 0; i < conf->raid_disks; i++) {
1837
1838 disk = conf->mirrors + i;
1839
1840 if (!disk->rdev) {
1841 disk->head_position = 0;
1842 mddev->degraded++;
1843 }
1844 }
1845
1846 /*
1847 * find the first working one and use it as a starting point
1848 * to read balancing.
1849 */
1850 for (j = 0; j < conf->raid_disks &&
1851 (!conf->mirrors[j].rdev ||
1852 !test_bit(In_sync, &conf->mirrors[j].rdev->flags)) ; j++)
1853 /* nothing */;
1854 conf->last_used = j;
1855
1856
1857 mddev->thread = md_register_thread(raid1d, mddev, "%s_raid1");
1858 if (!mddev->thread) {
1859 printk(KERN_ERR
1860 "raid1: couldn't allocate thread for %s\n",
1861 mdname(mddev));
1862 goto out_free_conf;
1863 }
1864
1865 printk(KERN_INFO
1866 "raid1: raid set %s active with %d out of %d mirrors\n",
1867 mdname(mddev), mddev->raid_disks - mddev->degraded,
1868 mddev->raid_disks);
1869 /*
1870 * Ok, everything is just fine now
1871 */
1872 mddev->array_size = mddev->size;
1873
1874 mddev->queue->unplug_fn = raid1_unplug;
1875 mddev->queue->issue_flush_fn = raid1_issue_flush;
1876
1877 return 0;
1878
1879 out_no_mem:
1880 printk(KERN_ERR "raid1: couldn't allocate memory for %s\n",
1881 mdname(mddev));
1882
1883 out_free_conf:
1884 if (conf) {
1885 if (conf->r1bio_pool)
1886 mempool_destroy(conf->r1bio_pool);
1887 kfree(conf->mirrors);
1888 put_page(conf->tmppage);
1889 kfree(conf->poolinfo);
1890 kfree(conf);
1891 mddev->private = NULL;
1892 }
1893 out:
1894 return -EIO;
1895 }
1896
1897 static int stop(mddev_t *mddev)
1898 {
1899 conf_t *conf = mddev_to_conf(mddev);
1900 struct bitmap *bitmap = mddev->bitmap;
1901 int behind_wait = 0;
1902
1903 /* wait for behind writes to complete */
1904 while (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
1905 behind_wait++;
1906 printk(KERN_INFO "raid1: behind writes in progress on device %s, waiting to stop (%d)\n", mdname(mddev), behind_wait);
1907 set_current_state(TASK_UNINTERRUPTIBLE);
1908 schedule_timeout(HZ); /* wait a second */
1909 /* need to kick something here to make sure I/O goes? */
1910 }
1911
1912 md_unregister_thread(mddev->thread);
1913 mddev->thread = NULL;
1914 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
1915 if (conf->r1bio_pool)
1916 mempool_destroy(conf->r1bio_pool);
1917 kfree(conf->mirrors);
1918 kfree(conf->poolinfo);
1919 kfree(conf);
1920 mddev->private = NULL;
1921 return 0;
1922 }
1923
1924 static int raid1_resize(mddev_t *mddev, sector_t sectors)
1925 {
1926 /* no resync is happening, and there is enough space
1927 * on all devices, so we can resize.
1928 * We need to make sure resync covers any new space.
1929 * If the array is shrinking we should possibly wait until
1930 * any io in the removed space completes, but it hardly seems
1931 * worth it.
1932 */
1933 mddev->array_size = sectors>>1;
1934 set_capacity(mddev->gendisk, mddev->array_size << 1);
1935 mddev->changed = 1;
1936 if (mddev->array_size > mddev->size && mddev->recovery_cp == MaxSector) {
1937 mddev->recovery_cp = mddev->size << 1;
1938 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1939 }
1940 mddev->size = mddev->array_size;
1941 mddev->resync_max_sectors = sectors;
1942 return 0;
1943 }
1944
1945 static int raid1_reshape(mddev_t *mddev, int raid_disks)
1946 {
1947 /* We need to:
1948 * 1/ resize the r1bio_pool
1949 * 2/ resize conf->mirrors
1950 *
1951 * We allocate a new r1bio_pool if we can.
1952 * Then raise a device barrier and wait until all IO stops.
1953 * Then resize conf->mirrors and swap in the new r1bio pool.
1954 *
1955 * At the same time, we "pack" the devices so that all the missing
1956 * devices have the higher raid_disk numbers.
1957 */
1958 mempool_t *newpool, *oldpool;
1959 struct pool_info *newpoolinfo;
1960 mirror_info_t *newmirrors;
1961 conf_t *conf = mddev_to_conf(mddev);
1962 int cnt;
1963
1964 int d, d2;
1965
1966 if (raid_disks < conf->raid_disks) {
1967 cnt=0;
1968 for (d= 0; d < conf->raid_disks; d++)
1969 if (conf->mirrors[d].rdev)
1970 cnt++;
1971 if (cnt > raid_disks)
1972 return -EBUSY;
1973 }
1974
1975 newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL);
1976 if (!newpoolinfo)
1977 return -ENOMEM;
1978 newpoolinfo->mddev = mddev;
1979 newpoolinfo->raid_disks = raid_disks;
1980
1981 newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
1982 r1bio_pool_free, newpoolinfo);
1983 if (!newpool) {
1984 kfree(newpoolinfo);
1985 return -ENOMEM;
1986 }
1987 newmirrors = kzalloc(sizeof(struct mirror_info) * raid_disks, GFP_KERNEL);
1988 if (!newmirrors) {
1989 kfree(newpoolinfo);
1990 mempool_destroy(newpool);
1991 return -ENOMEM;
1992 }
1993
1994 raise_barrier(conf);
1995
1996 /* ok, everything is stopped */
1997 oldpool = conf->r1bio_pool;
1998 conf->r1bio_pool = newpool;
1999
2000 for (d=d2=0; d < conf->raid_disks; d++)
2001 if (conf->mirrors[d].rdev) {
2002 conf->mirrors[d].rdev->raid_disk = d2;
2003 newmirrors[d2++].rdev = conf->mirrors[d].rdev;
2004 }
2005 kfree(conf->mirrors);
2006 conf->mirrors = newmirrors;
2007 kfree(conf->poolinfo);
2008 conf->poolinfo = newpoolinfo;
2009
2010 mddev->degraded += (raid_disks - conf->raid_disks);
2011 conf->raid_disks = mddev->raid_disks = raid_disks;
2012
2013 conf->last_used = 0; /* just make sure it is in-range */
2014 lower_barrier(conf);
2015
2016 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2017 md_wakeup_thread(mddev->thread);
2018
2019 mempool_destroy(oldpool);
2020 return 0;
2021 }
2022
2023 static void raid1_quiesce(mddev_t *mddev, int state)
2024 {
2025 conf_t *conf = mddev_to_conf(mddev);
2026
2027 switch(state) {
2028 case 1:
2029 raise_barrier(conf);
2030 break;
2031 case 0:
2032 lower_barrier(conf);
2033 break;
2034 }
2035 }
2036
2037
2038 static struct mdk_personality raid1_personality =
2039 {
2040 .name = "raid1",
2041 .level = 1,
2042 .owner = THIS_MODULE,
2043 .make_request = make_request,
2044 .run = run,
2045 .stop = stop,
2046 .status = status,
2047 .error_handler = error,
2048 .hot_add_disk = raid1_add_disk,
2049 .hot_remove_disk= raid1_remove_disk,
2050 .spare_active = raid1_spare_active,
2051 .sync_request = sync_request,
2052 .resize = raid1_resize,
2053 .reshape = raid1_reshape,
2054 .quiesce = raid1_quiesce,
2055 };
2056
2057 static int __init raid_init(void)
2058 {
2059 return register_md_personality(&raid1_personality);
2060 }
2061
2062 static void raid_exit(void)
2063 {
2064 unregister_md_personality(&raid1_personality);
2065 }
2066
2067 module_init(raid_init);
2068 module_exit(raid_exit);
2069 MODULE_LICENSE("GPL");
2070 MODULE_ALIAS("md-personality-3"); /* RAID1 */
2071 MODULE_ALIAS("md-level-1");
ubuntu-zesty-kernel mirror