Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mingo/linux...
[wrt350n-kernel.git] / drivers / md / raid1.c
blobff61b309129aa8ffa9dbd00987a71c1b6eb35bf5
1 /*
2 * raid1.c : Multiple Devices driver for Linux
4 * Copyright (C) 1999, 2000, 2001 Ingo Molnar, Red Hat
6 * Copyright (C) 1996, 1997, 1998 Ingo Molnar, Miguel de Icaza, Gadi Oxman
8 * RAID-1 management functions.
10 * Better read-balancing code written by Mika Kuoppala <miku@iki.fi>, 2000
12 * Fixes to reconstruction by Jakob Østergaard" <jakob@ostenfeld.dk>
13 * Various fixes by Neil Brown <neilb@cse.unsw.edu.au>
15 * Changes by Peter T. Breuer <ptb@it.uc3m.es> 31/1/2003 to support
16 * bitmapped intelligence in resync:
18 * - bitmap marked during normal i/o
19 * - bitmap used to skip nondirty blocks during sync
21 * Additions to bitmap code, (C) 2003-2004 Paul Clements, SteelEye Technology:
22 * - persistent bitmap code
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.
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.
34 #include "dm-bio-list.h"
35 #include <linux/raid/raid1.h>
36 #include <linux/raid/bitmap.h>
38 #define DEBUG 0
39 #if DEBUG
40 #define PRINTK(x...) printk(x)
41 #else
42 #define PRINTK(x...)
43 #endif
46 * Number of guaranteed r1bios in case of extreme VM load:
48 #define NR_RAID1_BIOS 256
51 static void unplug_slaves(mddev_t *mddev);
53 static void allow_barrier(conf_t *conf);
54 static void lower_barrier(conf_t *conf);
56 static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data)
58 struct pool_info *pi = data;
59 r1bio_t *r1_bio;
60 int size = offsetof(r1bio_t, bios[pi->raid_disks]);
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);
67 return r1_bio;
70 static void r1bio_pool_free(void *r1_bio, void *data)
72 kfree(r1_bio);
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)
81 static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data)
83 struct pool_info *pi = data;
84 struct page *page;
85 r1bio_t *r1_bio;
86 struct bio *bio;
87 int i, j;
89 r1_bio = r1bio_pool_alloc(gfp_flags, pi);
90 if (!r1_bio) {
91 unplug_slaves(pi->mddev);
92 return NULL;
96 * Allocate bios : 1 for reading, n-1 for writing
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;
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.
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;
121 bio->bi_io_vec[i].bv_page = page;
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;
132 r1_bio->master_bio = NULL;
134 return r1_bio;
136 out_free_pages:
137 for (i=0; i < RESYNC_PAGES ; i++)
138 for (j=0 ; j < pi->raid_disks; j++)
139 safe_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;
148 static void r1buf_pool_free(void *__r1_bio, void *data)
150 struct pool_info *pi = data;
151 int i,j;
152 r1bio_t *r1bio = __r1_bio;
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 safe_put_page(r1bio->bios[j]->bi_io_vec[i].bv_page);
161 for (i=0 ; i < pi->raid_disks; i++)
162 bio_put(r1bio->bios[i]);
164 r1bio_pool_free(r1bio, data);
167 static void put_all_bios(conf_t *conf, r1bio_t *r1_bio)
169 int i;
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;
179 static void free_r1bio(r1bio_t *r1_bio)
181 conf_t *conf = mddev_to_conf(r1_bio->mddev);
184 * Wake up any possible resync thread that waits for the device
185 * to go idle.
187 allow_barrier(conf);
189 put_all_bios(conf, r1_bio);
190 mempool_free(r1_bio, conf->r1bio_pool);
193 static void put_buf(r1bio_t *r1_bio)
195 conf_t *conf = mddev_to_conf(r1_bio->mddev);
196 int i;
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);
204 mempool_free(r1_bio, conf->r1buf_pool);
206 lower_barrier(conf);
209 static void reschedule_retry(r1bio_t *r1_bio)
211 unsigned long flags;
212 mddev_t *mddev = r1_bio->mddev;
213 conf_t *conf = mddev_to_conf(mddev);
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);
220 wake_up(&conf->wait_barrier);
221 md_wakeup_thread(mddev->thread);
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.
229 static void raid_end_bio_io(r1bio_t *r1_bio)
231 struct bio *bio = r1_bio->master_bio;
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);
241 bio_endio(bio,
242 test_bit(R1BIO_Uptodate, &r1_bio->state) ? 0 : -EIO);
244 free_r1bio(r1_bio);
248 * Update disk head position estimator based on IRQ completion info.
250 static inline void update_head_pos(int disk, r1bio_t *r1_bio)
252 conf_t *conf = mddev_to_conf(r1_bio->mddev);
254 conf->mirrors[disk].head_position =
255 r1_bio->sector + (r1_bio->sectors);
258 static void raid1_end_read_request(struct bio *bio, int error)
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);
265 mirror = r1_bio->read_disk;
267 * this branch is our 'one mirror IO has finished' event handler:
269 update_head_pos(mirror, r1_bio);
271 if (uptodate)
272 set_bit(R1BIO_Uptodate, &r1_bio->state);
273 else {
274 /* If all other devices have failed, we want to return
275 * the error upwards rather than fail the last device.
276 * Here we redefine "uptodate" to mean "Don't want to retry"
278 unsigned long flags;
279 spin_lock_irqsave(&conf->device_lock, flags);
280 if (r1_bio->mddev->degraded == conf->raid_disks ||
281 (r1_bio->mddev->degraded == conf->raid_disks-1 &&
282 !test_bit(Faulty, &conf->mirrors[mirror].rdev->flags)))
283 uptodate = 1;
284 spin_unlock_irqrestore(&conf->device_lock, flags);
287 if (uptodate)
288 raid_end_bio_io(r1_bio);
289 else {
291 * oops, read error:
293 char b[BDEVNAME_SIZE];
294 if (printk_ratelimit())
295 printk(KERN_ERR "raid1: %s: rescheduling sector %llu\n",
296 bdevname(conf->mirrors[mirror].rdev->bdev,b), (unsigned long long)r1_bio->sector);
297 reschedule_retry(r1_bio);
300 rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev);
303 static void raid1_end_write_request(struct bio *bio, int error)
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 struct bio *to_put = NULL;
312 for (mirror = 0; mirror < conf->raid_disks; mirror++)
313 if (r1_bio->bios[mirror] == bio)
314 break;
316 if (error == -EOPNOTSUPP && test_bit(R1BIO_Barrier, &r1_bio->state)) {
317 set_bit(BarriersNotsupp, &conf->mirrors[mirror].rdev->flags);
318 set_bit(R1BIO_BarrierRetry, &r1_bio->state);
319 r1_bio->mddev->barriers_work = 0;
320 /* Don't rdev_dec_pending in this branch - keep it for the retry */
321 } else {
323 * this branch is our 'one mirror IO has finished' event handler:
325 r1_bio->bios[mirror] = NULL;
326 to_put = bio;
327 if (!uptodate) {
328 md_error(r1_bio->mddev, conf->mirrors[mirror].rdev);
329 /* an I/O failed, we can't clear the bitmap */
330 set_bit(R1BIO_Degraded, &r1_bio->state);
331 } else
333 * Set R1BIO_Uptodate in our master bio, so that
334 * we will return a good error code for to the higher
335 * levels even if IO on some other mirrored buffer fails.
337 * The 'master' represents the composite IO operation to
338 * user-side. So if something waits for IO, then it will
339 * wait for the 'master' bio.
341 set_bit(R1BIO_Uptodate, &r1_bio->state);
343 update_head_pos(mirror, r1_bio);
345 if (behind) {
346 if (test_bit(WriteMostly, &conf->mirrors[mirror].rdev->flags))
347 atomic_dec(&r1_bio->behind_remaining);
349 /* In behind mode, we ACK the master bio once the I/O has safely
350 * reached all non-writemostly disks. Setting the Returned bit
351 * ensures that this gets done only once -- we don't ever want to
352 * return -EIO here, instead we'll wait */
354 if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) &&
355 test_bit(R1BIO_Uptodate, &r1_bio->state)) {
356 /* Maybe we can return now */
357 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
358 struct bio *mbio = r1_bio->master_bio;
359 PRINTK(KERN_DEBUG "raid1: behind end write sectors %llu-%llu\n",
360 (unsigned long long) mbio->bi_sector,
361 (unsigned long long) mbio->bi_sector +
362 (mbio->bi_size >> 9) - 1);
363 bio_endio(mbio, 0);
367 rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev);
371 * Let's see if all mirrored write operations have finished
372 * already.
374 if (atomic_dec_and_test(&r1_bio->remaining)) {
375 if (test_bit(R1BIO_BarrierRetry, &r1_bio->state))
376 reschedule_retry(r1_bio);
377 else {
378 /* it really is the end of this request */
379 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
380 /* free extra copy of the data pages */
381 int i = bio->bi_vcnt;
382 while (i--)
383 safe_put_page(bio->bi_io_vec[i].bv_page);
385 /* clear the bitmap if all writes complete successfully */
386 bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector,
387 r1_bio->sectors,
388 !test_bit(R1BIO_Degraded, &r1_bio->state),
389 behind);
390 md_write_end(r1_bio->mddev);
391 raid_end_bio_io(r1_bio);
395 if (to_put)
396 bio_put(to_put);
401 * This routine returns the disk from which the requested read should
402 * be done. There is a per-array 'next expected sequential IO' sector
403 * number - if this matches on the next IO then we use the last disk.
404 * There is also a per-disk 'last know head position' sector that is
405 * maintained from IRQ contexts, both the normal and the resync IO
406 * completion handlers update this position correctly. If there is no
407 * perfect sequential match then we pick the disk whose head is closest.
409 * If there are 2 mirrors in the same 2 devices, performance degrades
410 * because position is mirror, not device based.
412 * The rdev for the device selected will have nr_pending incremented.
414 static int read_balance(conf_t *conf, r1bio_t *r1_bio)
416 const unsigned long this_sector = r1_bio->sector;
417 int new_disk = conf->last_used, disk = new_disk;
418 int wonly_disk = -1;
419 const int sectors = r1_bio->sectors;
420 sector_t new_distance, current_distance;
421 mdk_rdev_t *rdev;
423 rcu_read_lock();
425 * Check if we can balance. We can balance on the whole
426 * device if no resync is going on, or below the resync window.
427 * We take the first readable disk when above the resync window.
429 retry:
430 if (conf->mddev->recovery_cp < MaxSector &&
431 (this_sector + sectors >= conf->next_resync)) {
432 /* Choose the first operation device, for consistancy */
433 new_disk = 0;
435 for (rdev = rcu_dereference(conf->mirrors[new_disk].rdev);
436 r1_bio->bios[new_disk] == IO_BLOCKED ||
437 !rdev || !test_bit(In_sync, &rdev->flags)
438 || test_bit(WriteMostly, &rdev->flags);
439 rdev = rcu_dereference(conf->mirrors[++new_disk].rdev)) {
441 if (rdev && test_bit(In_sync, &rdev->flags) &&
442 r1_bio->bios[new_disk] != IO_BLOCKED)
443 wonly_disk = new_disk;
445 if (new_disk == conf->raid_disks - 1) {
446 new_disk = wonly_disk;
447 break;
450 goto rb_out;
454 /* make sure the disk is operational */
455 for (rdev = rcu_dereference(conf->mirrors[new_disk].rdev);
456 r1_bio->bios[new_disk] == IO_BLOCKED ||
457 !rdev || !test_bit(In_sync, &rdev->flags) ||
458 test_bit(WriteMostly, &rdev->flags);
459 rdev = rcu_dereference(conf->mirrors[new_disk].rdev)) {
461 if (rdev && test_bit(In_sync, &rdev->flags) &&
462 r1_bio->bios[new_disk] != IO_BLOCKED)
463 wonly_disk = new_disk;
465 if (new_disk <= 0)
466 new_disk = conf->raid_disks;
467 new_disk--;
468 if (new_disk == disk) {
469 new_disk = wonly_disk;
470 break;
474 if (new_disk < 0)
475 goto rb_out;
477 disk = new_disk;
478 /* now disk == new_disk == starting point for search */
481 * Don't change to another disk for sequential reads:
483 if (conf->next_seq_sect == this_sector)
484 goto rb_out;
485 if (this_sector == conf->mirrors[new_disk].head_position)
486 goto rb_out;
488 current_distance = abs(this_sector - conf->mirrors[disk].head_position);
490 /* Find the disk whose head is closest */
492 do {
493 if (disk <= 0)
494 disk = conf->raid_disks;
495 disk--;
497 rdev = rcu_dereference(conf->mirrors[disk].rdev);
499 if (!rdev || r1_bio->bios[disk] == IO_BLOCKED ||
500 !test_bit(In_sync, &rdev->flags) ||
501 test_bit(WriteMostly, &rdev->flags))
502 continue;
504 if (!atomic_read(&rdev->nr_pending)) {
505 new_disk = disk;
506 break;
508 new_distance = abs(this_sector - conf->mirrors[disk].head_position);
509 if (new_distance < current_distance) {
510 current_distance = new_distance;
511 new_disk = disk;
513 } while (disk != conf->last_used);
515 rb_out:
518 if (new_disk >= 0) {
519 rdev = rcu_dereference(conf->mirrors[new_disk].rdev);
520 if (!rdev)
521 goto retry;
522 atomic_inc(&rdev->nr_pending);
523 if (!test_bit(In_sync, &rdev->flags)) {
524 /* cannot risk returning a device that failed
525 * before we inc'ed nr_pending
527 rdev_dec_pending(rdev, conf->mddev);
528 goto retry;
530 conf->next_seq_sect = this_sector + sectors;
531 conf->last_used = new_disk;
533 rcu_read_unlock();
535 return new_disk;
538 static void unplug_slaves(mddev_t *mddev)
540 conf_t *conf = mddev_to_conf(mddev);
541 int i;
543 rcu_read_lock();
544 for (i=0; i<mddev->raid_disks; i++) {
545 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
546 if (rdev && !test_bit(Faulty, &rdev->flags) && atomic_read(&rdev->nr_pending)) {
547 struct request_queue *r_queue = bdev_get_queue(rdev->bdev);
549 atomic_inc(&rdev->nr_pending);
550 rcu_read_unlock();
552 blk_unplug(r_queue);
554 rdev_dec_pending(rdev, mddev);
555 rcu_read_lock();
558 rcu_read_unlock();
561 static void raid1_unplug(struct request_queue *q)
563 mddev_t *mddev = q->queuedata;
565 unplug_slaves(mddev);
566 md_wakeup_thread(mddev->thread);
569 static int raid1_congested(void *data, int bits)
571 mddev_t *mddev = data;
572 conf_t *conf = mddev_to_conf(mddev);
573 int i, ret = 0;
575 rcu_read_lock();
576 for (i = 0; i < mddev->raid_disks; i++) {
577 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
578 if (rdev && !test_bit(Faulty, &rdev->flags)) {
579 struct request_queue *q = bdev_get_queue(rdev->bdev);
581 /* Note the '|| 1' - when read_balance prefers
582 * non-congested targets, it can be removed
584 if ((bits & (1<<BDI_write_congested)) || 1)
585 ret |= bdi_congested(&q->backing_dev_info, bits);
586 else
587 ret &= bdi_congested(&q->backing_dev_info, bits);
590 rcu_read_unlock();
591 return ret;
595 static int flush_pending_writes(conf_t *conf)
597 /* Any writes that have been queued but are awaiting
598 * bitmap updates get flushed here.
599 * We return 1 if any requests were actually submitted.
601 int rv = 0;
603 spin_lock_irq(&conf->device_lock);
605 if (conf->pending_bio_list.head) {
606 struct bio *bio;
607 bio = bio_list_get(&conf->pending_bio_list);
608 blk_remove_plug(conf->mddev->queue);
609 spin_unlock_irq(&conf->device_lock);
610 /* flush any pending bitmap writes to
611 * disk before proceeding w/ I/O */
612 bitmap_unplug(conf->mddev->bitmap);
614 while (bio) { /* submit pending writes */
615 struct bio *next = bio->bi_next;
616 bio->bi_next = NULL;
617 generic_make_request(bio);
618 bio = next;
620 rv = 1;
621 } else
622 spin_unlock_irq(&conf->device_lock);
623 return rv;
626 /* Barriers....
627 * Sometimes we need to suspend IO while we do something else,
628 * either some resync/recovery, or reconfigure the array.
629 * To do this we raise a 'barrier'.
630 * The 'barrier' is a counter that can be raised multiple times
631 * to count how many activities are happening which preclude
632 * normal IO.
633 * We can only raise the barrier if there is no pending IO.
634 * i.e. if nr_pending == 0.
635 * We choose only to raise the barrier if no-one is waiting for the
636 * barrier to go down. This means that as soon as an IO request
637 * is ready, no other operations which require a barrier will start
638 * until the IO request has had a chance.
640 * So: regular IO calls 'wait_barrier'. When that returns there
641 * is no backgroup IO happening, It must arrange to call
642 * allow_barrier when it has finished its IO.
643 * backgroup IO calls must call raise_barrier. Once that returns
644 * there is no normal IO happeing. It must arrange to call
645 * lower_barrier when the particular background IO completes.
647 #define RESYNC_DEPTH 32
649 static void raise_barrier(conf_t *conf)
651 spin_lock_irq(&conf->resync_lock);
653 /* Wait until no block IO is waiting */
654 wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting,
655 conf->resync_lock,
656 raid1_unplug(conf->mddev->queue));
658 /* block any new IO from starting */
659 conf->barrier++;
661 /* No wait for all pending IO to complete */
662 wait_event_lock_irq(conf->wait_barrier,
663 !conf->nr_pending && conf->barrier < RESYNC_DEPTH,
664 conf->resync_lock,
665 raid1_unplug(conf->mddev->queue));
667 spin_unlock_irq(&conf->resync_lock);
670 static void lower_barrier(conf_t *conf)
672 unsigned long flags;
673 spin_lock_irqsave(&conf->resync_lock, flags);
674 conf->barrier--;
675 spin_unlock_irqrestore(&conf->resync_lock, flags);
676 wake_up(&conf->wait_barrier);
679 static void wait_barrier(conf_t *conf)
681 spin_lock_irq(&conf->resync_lock);
682 if (conf->barrier) {
683 conf->nr_waiting++;
684 wait_event_lock_irq(conf->wait_barrier, !conf->barrier,
685 conf->resync_lock,
686 raid1_unplug(conf->mddev->queue));
687 conf->nr_waiting--;
689 conf->nr_pending++;
690 spin_unlock_irq(&conf->resync_lock);
693 static void allow_barrier(conf_t *conf)
695 unsigned long flags;
696 spin_lock_irqsave(&conf->resync_lock, flags);
697 conf->nr_pending--;
698 spin_unlock_irqrestore(&conf->resync_lock, flags);
699 wake_up(&conf->wait_barrier);
702 static void freeze_array(conf_t *conf)
704 /* stop syncio and normal IO and wait for everything to
705 * go quite.
706 * We increment barrier and nr_waiting, and then
707 * wait until nr_pending match nr_queued+1
708 * This is called in the context of one normal IO request
709 * that has failed. Thus any sync request that might be pending
710 * will be blocked by nr_pending, and we need to wait for
711 * pending IO requests to complete or be queued for re-try.
712 * Thus the number queued (nr_queued) plus this request (1)
713 * must match the number of pending IOs (nr_pending) before
714 * we continue.
716 spin_lock_irq(&conf->resync_lock);
717 conf->barrier++;
718 conf->nr_waiting++;
719 wait_event_lock_irq(conf->wait_barrier,
720 conf->nr_pending == conf->nr_queued+1,
721 conf->resync_lock,
722 ({ flush_pending_writes(conf);
723 raid1_unplug(conf->mddev->queue); }));
724 spin_unlock_irq(&conf->resync_lock);
726 static void unfreeze_array(conf_t *conf)
728 /* reverse the effect of the freeze */
729 spin_lock_irq(&conf->resync_lock);
730 conf->barrier--;
731 conf->nr_waiting--;
732 wake_up(&conf->wait_barrier);
733 spin_unlock_irq(&conf->resync_lock);
737 /* duplicate the data pages for behind I/O */
738 static struct page **alloc_behind_pages(struct bio *bio)
740 int i;
741 struct bio_vec *bvec;
742 struct page **pages = kzalloc(bio->bi_vcnt * sizeof(struct page *),
743 GFP_NOIO);
744 if (unlikely(!pages))
745 goto do_sync_io;
747 bio_for_each_segment(bvec, bio, i) {
748 pages[i] = alloc_page(GFP_NOIO);
749 if (unlikely(!pages[i]))
750 goto do_sync_io;
751 memcpy(kmap(pages[i]) + bvec->bv_offset,
752 kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len);
753 kunmap(pages[i]);
754 kunmap(bvec->bv_page);
757 return pages;
759 do_sync_io:
760 if (pages)
761 for (i = 0; i < bio->bi_vcnt && pages[i]; i++)
762 put_page(pages[i]);
763 kfree(pages);
764 PRINTK("%dB behind alloc failed, doing sync I/O\n", bio->bi_size);
765 return NULL;
768 static int make_request(struct request_queue *q, struct bio * bio)
770 mddev_t *mddev = q->queuedata;
771 conf_t *conf = mddev_to_conf(mddev);
772 mirror_info_t *mirror;
773 r1bio_t *r1_bio;
774 struct bio *read_bio;
775 int i, targets = 0, disks;
776 mdk_rdev_t *rdev;
777 struct bitmap *bitmap = mddev->bitmap;
778 unsigned long flags;
779 struct bio_list bl;
780 struct page **behind_pages = NULL;
781 const int rw = bio_data_dir(bio);
782 const int do_sync = bio_sync(bio);
783 int do_barriers;
786 * Register the new request and wait if the reconstruction
787 * thread has put up a bar for new requests.
788 * Continue immediately if no resync is active currently.
789 * We test barriers_work *after* md_write_start as md_write_start
790 * may cause the first superblock write, and that will check out
791 * if barriers work.
794 md_write_start(mddev, bio); /* wait on superblock update early */
796 if (unlikely(!mddev->barriers_work && bio_barrier(bio))) {
797 if (rw == WRITE)
798 md_write_end(mddev);
799 bio_endio(bio, -EOPNOTSUPP);
800 return 0;
803 wait_barrier(conf);
805 disk_stat_inc(mddev->gendisk, ios[rw]);
806 disk_stat_add(mddev->gendisk, sectors[rw], bio_sectors(bio));
809 * make_request() can abort the operation when READA is being
810 * used and no empty request is available.
813 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
815 r1_bio->master_bio = bio;
816 r1_bio->sectors = bio->bi_size >> 9;
817 r1_bio->state = 0;
818 r1_bio->mddev = mddev;
819 r1_bio->sector = bio->bi_sector;
821 if (rw == READ) {
823 * read balancing logic:
825 int rdisk = read_balance(conf, r1_bio);
827 if (rdisk < 0) {
828 /* couldn't find anywhere to read from */
829 raid_end_bio_io(r1_bio);
830 return 0;
832 mirror = conf->mirrors + rdisk;
834 r1_bio->read_disk = rdisk;
836 read_bio = bio_clone(bio, GFP_NOIO);
838 r1_bio->bios[rdisk] = read_bio;
840 read_bio->bi_sector = r1_bio->sector + mirror->rdev->data_offset;
841 read_bio->bi_bdev = mirror->rdev->bdev;
842 read_bio->bi_end_io = raid1_end_read_request;
843 read_bio->bi_rw = READ | do_sync;
844 read_bio->bi_private = r1_bio;
846 generic_make_request(read_bio);
847 return 0;
851 * WRITE:
853 /* first select target devices under spinlock and
854 * inc refcount on their rdev. Record them by setting
855 * bios[x] to bio
857 disks = conf->raid_disks;
858 #if 0
859 { static int first=1;
860 if (first) printk("First Write sector %llu disks %d\n",
861 (unsigned long long)r1_bio->sector, disks);
862 first = 0;
864 #endif
865 rcu_read_lock();
866 for (i = 0; i < disks; i++) {
867 if ((rdev=rcu_dereference(conf->mirrors[i].rdev)) != NULL &&
868 !test_bit(Faulty, &rdev->flags)) {
869 atomic_inc(&rdev->nr_pending);
870 if (test_bit(Faulty, &rdev->flags)) {
871 rdev_dec_pending(rdev, mddev);
872 r1_bio->bios[i] = NULL;
873 } else
874 r1_bio->bios[i] = bio;
875 targets++;
876 } else
877 r1_bio->bios[i] = NULL;
879 rcu_read_unlock();
881 BUG_ON(targets == 0); /* we never fail the last device */
883 if (targets < conf->raid_disks) {
884 /* array is degraded, we will not clear the bitmap
885 * on I/O completion (see raid1_end_write_request) */
886 set_bit(R1BIO_Degraded, &r1_bio->state);
889 /* do behind I/O ? */
890 if (bitmap &&
891 atomic_read(&bitmap->behind_writes) < bitmap->max_write_behind &&
892 (behind_pages = alloc_behind_pages(bio)) != NULL)
893 set_bit(R1BIO_BehindIO, &r1_bio->state);
895 atomic_set(&r1_bio->remaining, 0);
896 atomic_set(&r1_bio->behind_remaining, 0);
898 do_barriers = bio_barrier(bio);
899 if (do_barriers)
900 set_bit(R1BIO_Barrier, &r1_bio->state);
902 bio_list_init(&bl);
903 for (i = 0; i < disks; i++) {
904 struct bio *mbio;
905 if (!r1_bio->bios[i])
906 continue;
908 mbio = bio_clone(bio, GFP_NOIO);
909 r1_bio->bios[i] = mbio;
911 mbio->bi_sector = r1_bio->sector + conf->mirrors[i].rdev->data_offset;
912 mbio->bi_bdev = conf->mirrors[i].rdev->bdev;
913 mbio->bi_end_io = raid1_end_write_request;
914 mbio->bi_rw = WRITE | do_barriers | do_sync;
915 mbio->bi_private = r1_bio;
917 if (behind_pages) {
918 struct bio_vec *bvec;
919 int j;
921 /* Yes, I really want the '__' version so that
922 * we clear any unused pointer in the io_vec, rather
923 * than leave them unchanged. This is important
924 * because when we come to free the pages, we won't
925 * know the originial bi_idx, so we just free
926 * them all
928 __bio_for_each_segment(bvec, mbio, j, 0)
929 bvec->bv_page = behind_pages[j];
930 if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags))
931 atomic_inc(&r1_bio->behind_remaining);
934 atomic_inc(&r1_bio->remaining);
936 bio_list_add(&bl, mbio);
938 kfree(behind_pages); /* the behind pages are attached to the bios now */
940 bitmap_startwrite(bitmap, bio->bi_sector, r1_bio->sectors,
941 test_bit(R1BIO_BehindIO, &r1_bio->state));
942 spin_lock_irqsave(&conf->device_lock, flags);
943 bio_list_merge(&conf->pending_bio_list, &bl);
944 bio_list_init(&bl);
946 blk_plug_device(mddev->queue);
947 spin_unlock_irqrestore(&conf->device_lock, flags);
949 /* In case raid1d snuck into freeze_array */
950 wake_up(&conf->wait_barrier);
952 if (do_sync)
953 md_wakeup_thread(mddev->thread);
954 #if 0
955 while ((bio = bio_list_pop(&bl)) != NULL)
956 generic_make_request(bio);
957 #endif
959 return 0;
962 static void status(struct seq_file *seq, mddev_t *mddev)
964 conf_t *conf = mddev_to_conf(mddev);
965 int i;
967 seq_printf(seq, " [%d/%d] [", conf->raid_disks,
968 conf->raid_disks - mddev->degraded);
969 rcu_read_lock();
970 for (i = 0; i < conf->raid_disks; i++) {
971 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
972 seq_printf(seq, "%s",
973 rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_");
975 rcu_read_unlock();
976 seq_printf(seq, "]");
980 static void error(mddev_t *mddev, mdk_rdev_t *rdev)
982 char b[BDEVNAME_SIZE];
983 conf_t *conf = mddev_to_conf(mddev);
986 * If it is not operational, then we have already marked it as dead
987 * else if it is the last working disks, ignore the error, let the
988 * next level up know.
989 * else mark the drive as failed
991 if (test_bit(In_sync, &rdev->flags)
992 && (conf->raid_disks - mddev->degraded) == 1)
994 * Don't fail the drive, act as though we were just a
995 * normal single drive
997 return;
998 if (test_and_clear_bit(In_sync, &rdev->flags)) {
999 unsigned long flags;
1000 spin_lock_irqsave(&conf->device_lock, flags);
1001 mddev->degraded++;
1002 set_bit(Faulty, &rdev->flags);
1003 spin_unlock_irqrestore(&conf->device_lock, flags);
1005 * if recovery is running, make sure it aborts.
1007 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
1008 } else
1009 set_bit(Faulty, &rdev->flags);
1010 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1011 printk(KERN_ALERT "raid1: Disk failure on %s, disabling device. \n"
1012 " Operation continuing on %d devices\n",
1013 bdevname(rdev->bdev,b), conf->raid_disks - mddev->degraded);
1016 static void print_conf(conf_t *conf)
1018 int i;
1020 printk("RAID1 conf printout:\n");
1021 if (!conf) {
1022 printk("(!conf)\n");
1023 return;
1025 printk(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded,
1026 conf->raid_disks);
1028 rcu_read_lock();
1029 for (i = 0; i < conf->raid_disks; i++) {
1030 char b[BDEVNAME_SIZE];
1031 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
1032 if (rdev)
1033 printk(" disk %d, wo:%d, o:%d, dev:%s\n",
1034 i, !test_bit(In_sync, &rdev->flags),
1035 !test_bit(Faulty, &rdev->flags),
1036 bdevname(rdev->bdev,b));
1038 rcu_read_unlock();
1041 static void close_sync(conf_t *conf)
1043 wait_barrier(conf);
1044 allow_barrier(conf);
1046 mempool_destroy(conf->r1buf_pool);
1047 conf->r1buf_pool = NULL;
1050 static int raid1_spare_active(mddev_t *mddev)
1052 int i;
1053 conf_t *conf = mddev->private;
1056 * Find all failed disks within the RAID1 configuration
1057 * and mark them readable.
1058 * Called under mddev lock, so rcu protection not needed.
1060 for (i = 0; i < conf->raid_disks; i++) {
1061 mdk_rdev_t *rdev = conf->mirrors[i].rdev;
1062 if (rdev
1063 && !test_bit(Faulty, &rdev->flags)
1064 && !test_and_set_bit(In_sync, &rdev->flags)) {
1065 unsigned long flags;
1066 spin_lock_irqsave(&conf->device_lock, flags);
1067 mddev->degraded--;
1068 spin_unlock_irqrestore(&conf->device_lock, flags);
1072 print_conf(conf);
1073 return 0;
1077 static int raid1_add_disk(mddev_t *mddev, mdk_rdev_t *rdev)
1079 conf_t *conf = mddev->private;
1080 int found = 0;
1081 int mirror = 0;
1082 mirror_info_t *p;
1084 for (mirror=0; mirror < mddev->raid_disks; mirror++)
1085 if ( !(p=conf->mirrors+mirror)->rdev) {
1087 blk_queue_stack_limits(mddev->queue,
1088 rdev->bdev->bd_disk->queue);
1089 /* as we don't honour merge_bvec_fn, we must never risk
1090 * violating it, so limit ->max_sector to one PAGE, as
1091 * a one page request is never in violation.
1093 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
1094 mddev->queue->max_sectors > (PAGE_SIZE>>9))
1095 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
1097 p->head_position = 0;
1098 rdev->raid_disk = mirror;
1099 found = 1;
1100 /* As all devices are equivalent, we don't need a full recovery
1101 * if this was recently any drive of the array
1103 if (rdev->saved_raid_disk < 0)
1104 conf->fullsync = 1;
1105 rcu_assign_pointer(p->rdev, rdev);
1106 break;
1109 print_conf(conf);
1110 return found;
1113 static int raid1_remove_disk(mddev_t *mddev, int number)
1115 conf_t *conf = mddev->private;
1116 int err = 0;
1117 mdk_rdev_t *rdev;
1118 mirror_info_t *p = conf->mirrors+ number;
1120 print_conf(conf);
1121 rdev = p->rdev;
1122 if (rdev) {
1123 if (test_bit(In_sync, &rdev->flags) ||
1124 atomic_read(&rdev->nr_pending)) {
1125 err = -EBUSY;
1126 goto abort;
1128 p->rdev = NULL;
1129 synchronize_rcu();
1130 if (atomic_read(&rdev->nr_pending)) {
1131 /* lost the race, try later */
1132 err = -EBUSY;
1133 p->rdev = rdev;
1136 abort:
1138 print_conf(conf);
1139 return err;
1143 static void end_sync_read(struct bio *bio, int error)
1145 r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
1146 int i;
1148 for (i=r1_bio->mddev->raid_disks; i--; )
1149 if (r1_bio->bios[i] == bio)
1150 break;
1151 BUG_ON(i < 0);
1152 update_head_pos(i, r1_bio);
1154 * we have read a block, now it needs to be re-written,
1155 * or re-read if the read failed.
1156 * We don't do much here, just schedule handling by raid1d
1158 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
1159 set_bit(R1BIO_Uptodate, &r1_bio->state);
1161 if (atomic_dec_and_test(&r1_bio->remaining))
1162 reschedule_retry(r1_bio);
1165 static void end_sync_write(struct bio *bio, int error)
1167 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1168 r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
1169 mddev_t *mddev = r1_bio->mddev;
1170 conf_t *conf = mddev_to_conf(mddev);
1171 int i;
1172 int mirror=0;
1174 for (i = 0; i < conf->raid_disks; i++)
1175 if (r1_bio->bios[i] == bio) {
1176 mirror = i;
1177 break;
1179 if (!uptodate) {
1180 int sync_blocks = 0;
1181 sector_t s = r1_bio->sector;
1182 long sectors_to_go = r1_bio->sectors;
1183 /* make sure these bits doesn't get cleared. */
1184 do {
1185 bitmap_end_sync(mddev->bitmap, s,
1186 &sync_blocks, 1);
1187 s += sync_blocks;
1188 sectors_to_go -= sync_blocks;
1189 } while (sectors_to_go > 0);
1190 md_error(mddev, conf->mirrors[mirror].rdev);
1193 update_head_pos(mirror, r1_bio);
1195 if (atomic_dec_and_test(&r1_bio->remaining)) {
1196 md_done_sync(mddev, r1_bio->sectors, uptodate);
1197 put_buf(r1_bio);
1201 static void sync_request_write(mddev_t *mddev, r1bio_t *r1_bio)
1203 conf_t *conf = mddev_to_conf(mddev);
1204 int i;
1205 int disks = conf->raid_disks;
1206 struct bio *bio, *wbio;
1208 bio = r1_bio->bios[r1_bio->read_disk];
1211 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
1212 /* We have read all readable devices. If we haven't
1213 * got the block, then there is no hope left.
1214 * If we have, then we want to do a comparison
1215 * and skip the write if everything is the same.
1216 * If any blocks failed to read, then we need to
1217 * attempt an over-write
1219 int primary;
1220 if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) {
1221 for (i=0; i<mddev->raid_disks; i++)
1222 if (r1_bio->bios[i]->bi_end_io == end_sync_read)
1223 md_error(mddev, conf->mirrors[i].rdev);
1225 md_done_sync(mddev, r1_bio->sectors, 1);
1226 put_buf(r1_bio);
1227 return;
1229 for (primary=0; primary<mddev->raid_disks; primary++)
1230 if (r1_bio->bios[primary]->bi_end_io == end_sync_read &&
1231 test_bit(BIO_UPTODATE, &r1_bio->bios[primary]->bi_flags)) {
1232 r1_bio->bios[primary]->bi_end_io = NULL;
1233 rdev_dec_pending(conf->mirrors[primary].rdev, mddev);
1234 break;
1236 r1_bio->read_disk = primary;
1237 for (i=0; i<mddev->raid_disks; i++)
1238 if (r1_bio->bios[i]->bi_end_io == end_sync_read) {
1239 int j;
1240 int vcnt = r1_bio->sectors >> (PAGE_SHIFT- 9);
1241 struct bio *pbio = r1_bio->bios[primary];
1242 struct bio *sbio = r1_bio->bios[i];
1244 if (test_bit(BIO_UPTODATE, &sbio->bi_flags)) {
1245 for (j = vcnt; j-- ; ) {
1246 struct page *p, *s;
1247 p = pbio->bi_io_vec[j].bv_page;
1248 s = sbio->bi_io_vec[j].bv_page;
1249 if (memcmp(page_address(p),
1250 page_address(s),
1251 PAGE_SIZE))
1252 break;
1254 } else
1255 j = 0;
1256 if (j >= 0)
1257 mddev->resync_mismatches += r1_bio->sectors;
1258 if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)
1259 && test_bit(BIO_UPTODATE, &sbio->bi_flags))) {
1260 sbio->bi_end_io = NULL;
1261 rdev_dec_pending(conf->mirrors[i].rdev, mddev);
1262 } else {
1263 /* fixup the bio for reuse */
1264 sbio->bi_vcnt = vcnt;
1265 sbio->bi_size = r1_bio->sectors << 9;
1266 sbio->bi_idx = 0;
1267 sbio->bi_phys_segments = 0;
1268 sbio->bi_hw_segments = 0;
1269 sbio->bi_hw_front_size = 0;
1270 sbio->bi_hw_back_size = 0;
1271 sbio->bi_flags &= ~(BIO_POOL_MASK - 1);
1272 sbio->bi_flags |= 1 << BIO_UPTODATE;
1273 sbio->bi_next = NULL;
1274 sbio->bi_sector = r1_bio->sector +
1275 conf->mirrors[i].rdev->data_offset;
1276 sbio->bi_bdev = conf->mirrors[i].rdev->bdev;
1277 for (j = 0; j < vcnt ; j++)
1278 memcpy(page_address(sbio->bi_io_vec[j].bv_page),
1279 page_address(pbio->bi_io_vec[j].bv_page),
1280 PAGE_SIZE);
1285 if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) {
1286 /* ouch - failed to read all of that.
1287 * Try some synchronous reads of other devices to get
1288 * good data, much like with normal read errors. Only
1289 * read into the pages we already have so we don't
1290 * need to re-issue the read request.
1291 * We don't need to freeze the array, because being in an
1292 * active sync request, there is no normal IO, and
1293 * no overlapping syncs.
1295 sector_t sect = r1_bio->sector;
1296 int sectors = r1_bio->sectors;
1297 int idx = 0;
1299 while(sectors) {
1300 int s = sectors;
1301 int d = r1_bio->read_disk;
1302 int success = 0;
1303 mdk_rdev_t *rdev;
1305 if (s > (PAGE_SIZE>>9))
1306 s = PAGE_SIZE >> 9;
1307 do {
1308 if (r1_bio->bios[d]->bi_end_io == end_sync_read) {
1309 /* No rcu protection needed here devices
1310 * can only be removed when no resync is
1311 * active, and resync is currently active
1313 rdev = conf->mirrors[d].rdev;
1314 if (sync_page_io(rdev->bdev,
1315 sect + rdev->data_offset,
1316 s<<9,
1317 bio->bi_io_vec[idx].bv_page,
1318 READ)) {
1319 success = 1;
1320 break;
1323 d++;
1324 if (d == conf->raid_disks)
1325 d = 0;
1326 } while (!success && d != r1_bio->read_disk);
1328 if (success) {
1329 int start = d;
1330 /* write it back and re-read */
1331 set_bit(R1BIO_Uptodate, &r1_bio->state);
1332 while (d != r1_bio->read_disk) {
1333 if (d == 0)
1334 d = conf->raid_disks;
1335 d--;
1336 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1337 continue;
1338 rdev = conf->mirrors[d].rdev;
1339 atomic_add(s, &rdev->corrected_errors);
1340 if (sync_page_io(rdev->bdev,
1341 sect + rdev->data_offset,
1342 s<<9,
1343 bio->bi_io_vec[idx].bv_page,
1344 WRITE) == 0)
1345 md_error(mddev, rdev);
1347 d = start;
1348 while (d != r1_bio->read_disk) {
1349 if (d == 0)
1350 d = conf->raid_disks;
1351 d--;
1352 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1353 continue;
1354 rdev = conf->mirrors[d].rdev;
1355 if (sync_page_io(rdev->bdev,
1356 sect + rdev->data_offset,
1357 s<<9,
1358 bio->bi_io_vec[idx].bv_page,
1359 READ) == 0)
1360 md_error(mddev, rdev);
1362 } else {
1363 char b[BDEVNAME_SIZE];
1364 /* Cannot read from anywhere, array is toast */
1365 md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev);
1366 printk(KERN_ALERT "raid1: %s: unrecoverable I/O read error"
1367 " for block %llu\n",
1368 bdevname(bio->bi_bdev,b),
1369 (unsigned long long)r1_bio->sector);
1370 md_done_sync(mddev, r1_bio->sectors, 0);
1371 put_buf(r1_bio);
1372 return;
1374 sectors -= s;
1375 sect += s;
1376 idx ++;
1381 * schedule writes
1383 atomic_set(&r1_bio->remaining, 1);
1384 for (i = 0; i < disks ; i++) {
1385 wbio = r1_bio->bios[i];
1386 if (wbio->bi_end_io == NULL ||
1387 (wbio->bi_end_io == end_sync_read &&
1388 (i == r1_bio->read_disk ||
1389 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery))))
1390 continue;
1392 wbio->bi_rw = WRITE;
1393 wbio->bi_end_io = end_sync_write;
1394 atomic_inc(&r1_bio->remaining);
1395 md_sync_acct(conf->mirrors[i].rdev->bdev, wbio->bi_size >> 9);
1397 generic_make_request(wbio);
1400 if (atomic_dec_and_test(&r1_bio->remaining)) {
1401 /* if we're here, all write(s) have completed, so clean up */
1402 md_done_sync(mddev, r1_bio->sectors, 1);
1403 put_buf(r1_bio);
1408 * This is a kernel thread which:
1410 * 1. Retries failed read operations on working mirrors.
1411 * 2. Updates the raid superblock when problems encounter.
1412 * 3. Performs writes following reads for array syncronising.
1415 static void fix_read_error(conf_t *conf, int read_disk,
1416 sector_t sect, int sectors)
1418 mddev_t *mddev = conf->mddev;
1419 while(sectors) {
1420 int s = sectors;
1421 int d = read_disk;
1422 int success = 0;
1423 int start;
1424 mdk_rdev_t *rdev;
1426 if (s > (PAGE_SIZE>>9))
1427 s = PAGE_SIZE >> 9;
1429 do {
1430 /* Note: no rcu protection needed here
1431 * as this is synchronous in the raid1d thread
1432 * which is the thread that might remove
1433 * a device. If raid1d ever becomes multi-threaded....
1435 rdev = conf->mirrors[d].rdev;
1436 if (rdev &&
1437 test_bit(In_sync, &rdev->flags) &&
1438 sync_page_io(rdev->bdev,
1439 sect + rdev->data_offset,
1440 s<<9,
1441 conf->tmppage, READ))
1442 success = 1;
1443 else {
1444 d++;
1445 if (d == conf->raid_disks)
1446 d = 0;
1448 } while (!success && d != read_disk);
1450 if (!success) {
1451 /* Cannot read from anywhere -- bye bye array */
1452 md_error(mddev, conf->mirrors[read_disk].rdev);
1453 break;
1455 /* write it back and re-read */
1456 start = d;
1457 while (d != read_disk) {
1458 if (d==0)
1459 d = conf->raid_disks;
1460 d--;
1461 rdev = conf->mirrors[d].rdev;
1462 if (rdev &&
1463 test_bit(In_sync, &rdev->flags)) {
1464 if (sync_page_io(rdev->bdev,
1465 sect + rdev->data_offset,
1466 s<<9, conf->tmppage, WRITE)
1467 == 0)
1468 /* Well, this device is dead */
1469 md_error(mddev, rdev);
1472 d = start;
1473 while (d != read_disk) {
1474 char b[BDEVNAME_SIZE];
1475 if (d==0)
1476 d = conf->raid_disks;
1477 d--;
1478 rdev = conf->mirrors[d].rdev;
1479 if (rdev &&
1480 test_bit(In_sync, &rdev->flags)) {
1481 if (sync_page_io(rdev->bdev,
1482 sect + rdev->data_offset,
1483 s<<9, conf->tmppage, READ)
1484 == 0)
1485 /* Well, this device is dead */
1486 md_error(mddev, rdev);
1487 else {
1488 atomic_add(s, &rdev->corrected_errors);
1489 printk(KERN_INFO
1490 "raid1:%s: read error corrected "
1491 "(%d sectors at %llu on %s)\n",
1492 mdname(mddev), s,
1493 (unsigned long long)(sect +
1494 rdev->data_offset),
1495 bdevname(rdev->bdev, b));
1499 sectors -= s;
1500 sect += s;
1504 static void raid1d(mddev_t *mddev)
1506 r1bio_t *r1_bio;
1507 struct bio *bio;
1508 unsigned long flags;
1509 conf_t *conf = mddev_to_conf(mddev);
1510 struct list_head *head = &conf->retry_list;
1511 int unplug=0;
1512 mdk_rdev_t *rdev;
1514 md_check_recovery(mddev);
1516 for (;;) {
1517 char b[BDEVNAME_SIZE];
1519 unplug += flush_pending_writes(conf);
1521 spin_lock_irqsave(&conf->device_lock, flags);
1522 if (list_empty(head)) {
1523 spin_unlock_irqrestore(&conf->device_lock, flags);
1524 break;
1526 r1_bio = list_entry(head->prev, r1bio_t, retry_list);
1527 list_del(head->prev);
1528 conf->nr_queued--;
1529 spin_unlock_irqrestore(&conf->device_lock, flags);
1531 mddev = r1_bio->mddev;
1532 conf = mddev_to_conf(mddev);
1533 if (test_bit(R1BIO_IsSync, &r1_bio->state)) {
1534 sync_request_write(mddev, r1_bio);
1535 unplug = 1;
1536 } else if (test_bit(R1BIO_BarrierRetry, &r1_bio->state)) {
1537 /* some requests in the r1bio were BIO_RW_BARRIER
1538 * requests which failed with -EOPNOTSUPP. Hohumm..
1539 * Better resubmit without the barrier.
1540 * We know which devices to resubmit for, because
1541 * all others have had their bios[] entry cleared.
1542 * We already have a nr_pending reference on these rdevs.
1544 int i;
1545 const int do_sync = bio_sync(r1_bio->master_bio);
1546 clear_bit(R1BIO_BarrierRetry, &r1_bio->state);
1547 clear_bit(R1BIO_Barrier, &r1_bio->state);
1548 for (i=0; i < conf->raid_disks; i++)
1549 if (r1_bio->bios[i])
1550 atomic_inc(&r1_bio->remaining);
1551 for (i=0; i < conf->raid_disks; i++)
1552 if (r1_bio->bios[i]) {
1553 struct bio_vec *bvec;
1554 int j;
1556 bio = bio_clone(r1_bio->master_bio, GFP_NOIO);
1557 /* copy pages from the failed bio, as
1558 * this might be a write-behind device */
1559 __bio_for_each_segment(bvec, bio, j, 0)
1560 bvec->bv_page = bio_iovec_idx(r1_bio->bios[i], j)->bv_page;
1561 bio_put(r1_bio->bios[i]);
1562 bio->bi_sector = r1_bio->sector +
1563 conf->mirrors[i].rdev->data_offset;
1564 bio->bi_bdev = conf->mirrors[i].rdev->bdev;
1565 bio->bi_end_io = raid1_end_write_request;
1566 bio->bi_rw = WRITE | do_sync;
1567 bio->bi_private = r1_bio;
1568 r1_bio->bios[i] = bio;
1569 generic_make_request(bio);
1571 } else {
1572 int disk;
1574 /* we got a read error. Maybe the drive is bad. Maybe just
1575 * the block and we can fix it.
1576 * We freeze all other IO, and try reading the block from
1577 * other devices. When we find one, we re-write
1578 * and check it that fixes the read error.
1579 * This is all done synchronously while the array is
1580 * frozen
1582 if (mddev->ro == 0) {
1583 freeze_array(conf);
1584 fix_read_error(conf, r1_bio->read_disk,
1585 r1_bio->sector,
1586 r1_bio->sectors);
1587 unfreeze_array(conf);
1590 bio = r1_bio->bios[r1_bio->read_disk];
1591 if ((disk=read_balance(conf, r1_bio)) == -1) {
1592 printk(KERN_ALERT "raid1: %s: unrecoverable I/O"
1593 " read error for block %llu\n",
1594 bdevname(bio->bi_bdev,b),
1595 (unsigned long long)r1_bio->sector);
1596 raid_end_bio_io(r1_bio);
1597 } else {
1598 const int do_sync = bio_sync(r1_bio->master_bio);
1599 r1_bio->bios[r1_bio->read_disk] =
1600 mddev->ro ? IO_BLOCKED : NULL;
1601 r1_bio->read_disk = disk;
1602 bio_put(bio);
1603 bio = bio_clone(r1_bio->master_bio, GFP_NOIO);
1604 r1_bio->bios[r1_bio->read_disk] = bio;
1605 rdev = conf->mirrors[disk].rdev;
1606 if (printk_ratelimit())
1607 printk(KERN_ERR "raid1: %s: redirecting sector %llu to"
1608 " another mirror\n",
1609 bdevname(rdev->bdev,b),
1610 (unsigned long long)r1_bio->sector);
1611 bio->bi_sector = r1_bio->sector + rdev->data_offset;
1612 bio->bi_bdev = rdev->bdev;
1613 bio->bi_end_io = raid1_end_read_request;
1614 bio->bi_rw = READ | do_sync;
1615 bio->bi_private = r1_bio;
1616 unplug = 1;
1617 generic_make_request(bio);
1621 if (unplug)
1622 unplug_slaves(mddev);
1626 static int init_resync(conf_t *conf)
1628 int buffs;
1630 buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;
1631 BUG_ON(conf->r1buf_pool);
1632 conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free,
1633 conf->poolinfo);
1634 if (!conf->r1buf_pool)
1635 return -ENOMEM;
1636 conf->next_resync = 0;
1637 return 0;
1641 * perform a "sync" on one "block"
1643 * We need to make sure that no normal I/O request - particularly write
1644 * requests - conflict with active sync requests.
1646 * This is achieved by tracking pending requests and a 'barrier' concept
1647 * that can be installed to exclude normal IO requests.
1650 static sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster)
1652 conf_t *conf = mddev_to_conf(mddev);
1653 r1bio_t *r1_bio;
1654 struct bio *bio;
1655 sector_t max_sector, nr_sectors;
1656 int disk = -1;
1657 int i;
1658 int wonly = -1;
1659 int write_targets = 0, read_targets = 0;
1660 int sync_blocks;
1661 int still_degraded = 0;
1663 if (!conf->r1buf_pool)
1666 printk("sync start - bitmap %p\n", mddev->bitmap);
1668 if (init_resync(conf))
1669 return 0;
1672 max_sector = mddev->size << 1;
1673 if (sector_nr >= max_sector) {
1674 /* If we aborted, we need to abort the
1675 * sync on the 'current' bitmap chunk (there will
1676 * only be one in raid1 resync.
1677 * We can find the current addess in mddev->curr_resync
1679 if (mddev->curr_resync < max_sector) /* aborted */
1680 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
1681 &sync_blocks, 1);
1682 else /* completed sync */
1683 conf->fullsync = 0;
1685 bitmap_close_sync(mddev->bitmap);
1686 close_sync(conf);
1687 return 0;
1690 if (mddev->bitmap == NULL &&
1691 mddev->recovery_cp == MaxSector &&
1692 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
1693 conf->fullsync == 0) {
1694 *skipped = 1;
1695 return max_sector - sector_nr;
1697 /* before building a request, check if we can skip these blocks..
1698 * This call the bitmap_start_sync doesn't actually record anything
1700 if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
1701 !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
1702 /* We can skip this block, and probably several more */
1703 *skipped = 1;
1704 return sync_blocks;
1707 * If there is non-resync activity waiting for a turn,
1708 * and resync is going fast enough,
1709 * then let it though before starting on this new sync request.
1711 if (!go_faster && conf->nr_waiting)
1712 msleep_interruptible(1000);
1714 bitmap_cond_end_sync(mddev->bitmap, sector_nr);
1715 raise_barrier(conf);
1717 conf->next_resync = sector_nr;
1719 r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO);
1720 rcu_read_lock();
1722 * If we get a correctably read error during resync or recovery,
1723 * we might want to read from a different device. So we
1724 * flag all drives that could conceivably be read from for READ,
1725 * and any others (which will be non-In_sync devices) for WRITE.
1726 * If a read fails, we try reading from something else for which READ
1727 * is OK.
1730 r1_bio->mddev = mddev;
1731 r1_bio->sector = sector_nr;
1732 r1_bio->state = 0;
1733 set_bit(R1BIO_IsSync, &r1_bio->state);
1735 for (i=0; i < conf->raid_disks; i++) {
1736 mdk_rdev_t *rdev;
1737 bio = r1_bio->bios[i];
1739 /* take from bio_init */
1740 bio->bi_next = NULL;
1741 bio->bi_flags |= 1 << BIO_UPTODATE;
1742 bio->bi_rw = READ;
1743 bio->bi_vcnt = 0;
1744 bio->bi_idx = 0;
1745 bio->bi_phys_segments = 0;
1746 bio->bi_hw_segments = 0;
1747 bio->bi_size = 0;
1748 bio->bi_end_io = NULL;
1749 bio->bi_private = NULL;
1751 rdev = rcu_dereference(conf->mirrors[i].rdev);
1752 if (rdev == NULL ||
1753 test_bit(Faulty, &rdev->flags)) {
1754 still_degraded = 1;
1755 continue;
1756 } else if (!test_bit(In_sync, &rdev->flags)) {
1757 bio->bi_rw = WRITE;
1758 bio->bi_end_io = end_sync_write;
1759 write_targets ++;
1760 } else {
1761 /* may need to read from here */
1762 bio->bi_rw = READ;
1763 bio->bi_end_io = end_sync_read;
1764 if (test_bit(WriteMostly, &rdev->flags)) {
1765 if (wonly < 0)
1766 wonly = i;
1767 } else {
1768 if (disk < 0)
1769 disk = i;
1771 read_targets++;
1773 atomic_inc(&rdev->nr_pending);
1774 bio->bi_sector = sector_nr + rdev->data_offset;
1775 bio->bi_bdev = rdev->bdev;
1776 bio->bi_private = r1_bio;
1778 rcu_read_unlock();
1779 if (disk < 0)
1780 disk = wonly;
1781 r1_bio->read_disk = disk;
1783 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0)
1784 /* extra read targets are also write targets */
1785 write_targets += read_targets-1;
1787 if (write_targets == 0 || read_targets == 0) {
1788 /* There is nowhere to write, so all non-sync
1789 * drives must be failed - so we are finished
1791 sector_t rv = max_sector - sector_nr;
1792 *skipped = 1;
1793 put_buf(r1_bio);
1794 return rv;
1797 if (max_sector > mddev->resync_max)
1798 max_sector = mddev->resync_max; /* Don't do IO beyond here */
1799 nr_sectors = 0;
1800 sync_blocks = 0;
1801 do {
1802 struct page *page;
1803 int len = PAGE_SIZE;
1804 if (sector_nr + (len>>9) > max_sector)
1805 len = (max_sector - sector_nr) << 9;
1806 if (len == 0)
1807 break;
1808 if (sync_blocks == 0) {
1809 if (!bitmap_start_sync(mddev->bitmap, sector_nr,
1810 &sync_blocks, still_degraded) &&
1811 !conf->fullsync &&
1812 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1813 break;
1814 BUG_ON(sync_blocks < (PAGE_SIZE>>9));
1815 if (len > (sync_blocks<<9))
1816 len = sync_blocks<<9;
1819 for (i=0 ; i < conf->raid_disks; i++) {
1820 bio = r1_bio->bios[i];
1821 if (bio->bi_end_io) {
1822 page = bio->bi_io_vec[bio->bi_vcnt].bv_page;
1823 if (bio_add_page(bio, page, len, 0) == 0) {
1824 /* stop here */
1825 bio->bi_io_vec[bio->bi_vcnt].bv_page = page;
1826 while (i > 0) {
1827 i--;
1828 bio = r1_bio->bios[i];
1829 if (bio->bi_end_io==NULL)
1830 continue;
1831 /* remove last page from this bio */
1832 bio->bi_vcnt--;
1833 bio->bi_size -= len;
1834 bio->bi_flags &= ~(1<< BIO_SEG_VALID);
1836 goto bio_full;
1840 nr_sectors += len>>9;
1841 sector_nr += len>>9;
1842 sync_blocks -= (len>>9);
1843 } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES);
1844 bio_full:
1845 r1_bio->sectors = nr_sectors;
1847 /* For a user-requested sync, we read all readable devices and do a
1848 * compare
1850 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
1851 atomic_set(&r1_bio->remaining, read_targets);
1852 for (i=0; i<conf->raid_disks; i++) {
1853 bio = r1_bio->bios[i];
1854 if (bio->bi_end_io == end_sync_read) {
1855 md_sync_acct(bio->bi_bdev, nr_sectors);
1856 generic_make_request(bio);
1859 } else {
1860 atomic_set(&r1_bio->remaining, 1);
1861 bio = r1_bio->bios[r1_bio->read_disk];
1862 md_sync_acct(bio->bi_bdev, nr_sectors);
1863 generic_make_request(bio);
1866 return nr_sectors;
1869 static int run(mddev_t *mddev)
1871 conf_t *conf;
1872 int i, j, disk_idx;
1873 mirror_info_t *disk;
1874 mdk_rdev_t *rdev;
1875 struct list_head *tmp;
1877 if (mddev->level != 1) {
1878 printk("raid1: %s: raid level not set to mirroring (%d)\n",
1879 mdname(mddev), mddev->level);
1880 goto out;
1882 if (mddev->reshape_position != MaxSector) {
1883 printk("raid1: %s: reshape_position set but not supported\n",
1884 mdname(mddev));
1885 goto out;
1888 * copy the already verified devices into our private RAID1
1889 * bookkeeping area. [whatever we allocate in run(),
1890 * should be freed in stop()]
1892 conf = kzalloc(sizeof(conf_t), GFP_KERNEL);
1893 mddev->private = conf;
1894 if (!conf)
1895 goto out_no_mem;
1897 conf->mirrors = kzalloc(sizeof(struct mirror_info)*mddev->raid_disks,
1898 GFP_KERNEL);
1899 if (!conf->mirrors)
1900 goto out_no_mem;
1902 conf->tmppage = alloc_page(GFP_KERNEL);
1903 if (!conf->tmppage)
1904 goto out_no_mem;
1906 conf->poolinfo = kmalloc(sizeof(*conf->poolinfo), GFP_KERNEL);
1907 if (!conf->poolinfo)
1908 goto out_no_mem;
1909 conf->poolinfo->mddev = mddev;
1910 conf->poolinfo->raid_disks = mddev->raid_disks;
1911 conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
1912 r1bio_pool_free,
1913 conf->poolinfo);
1914 if (!conf->r1bio_pool)
1915 goto out_no_mem;
1917 rdev_for_each(rdev, tmp, mddev) {
1918 disk_idx = rdev->raid_disk;
1919 if (disk_idx >= mddev->raid_disks
1920 || disk_idx < 0)
1921 continue;
1922 disk = conf->mirrors + disk_idx;
1924 disk->rdev = rdev;
1926 blk_queue_stack_limits(mddev->queue,
1927 rdev->bdev->bd_disk->queue);
1928 /* as we don't honour merge_bvec_fn, we must never risk
1929 * violating it, so limit ->max_sector to one PAGE, as
1930 * a one page request is never in violation.
1932 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
1933 mddev->queue->max_sectors > (PAGE_SIZE>>9))
1934 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
1936 disk->head_position = 0;
1938 conf->raid_disks = mddev->raid_disks;
1939 conf->mddev = mddev;
1940 spin_lock_init(&conf->device_lock);
1941 INIT_LIST_HEAD(&conf->retry_list);
1943 spin_lock_init(&conf->resync_lock);
1944 init_waitqueue_head(&conf->wait_barrier);
1946 bio_list_init(&conf->pending_bio_list);
1947 bio_list_init(&conf->flushing_bio_list);
1950 mddev->degraded = 0;
1951 for (i = 0; i < conf->raid_disks; i++) {
1953 disk = conf->mirrors + i;
1955 if (!disk->rdev ||
1956 !test_bit(In_sync, &disk->rdev->flags)) {
1957 disk->head_position = 0;
1958 mddev->degraded++;
1959 if (disk->rdev)
1960 conf->fullsync = 1;
1963 if (mddev->degraded == conf->raid_disks) {
1964 printk(KERN_ERR "raid1: no operational mirrors for %s\n",
1965 mdname(mddev));
1966 goto out_free_conf;
1968 if (conf->raid_disks - mddev->degraded == 1)
1969 mddev->recovery_cp = MaxSector;
1972 * find the first working one and use it as a starting point
1973 * to read balancing.
1975 for (j = 0; j < conf->raid_disks &&
1976 (!conf->mirrors[j].rdev ||
1977 !test_bit(In_sync, &conf->mirrors[j].rdev->flags)) ; j++)
1978 /* nothing */;
1979 conf->last_used = j;
1982 mddev->thread = md_register_thread(raid1d, mddev, "%s_raid1");
1983 if (!mddev->thread) {
1984 printk(KERN_ERR
1985 "raid1: couldn't allocate thread for %s\n",
1986 mdname(mddev));
1987 goto out_free_conf;
1990 printk(KERN_INFO
1991 "raid1: raid set %s active with %d out of %d mirrors\n",
1992 mdname(mddev), mddev->raid_disks - mddev->degraded,
1993 mddev->raid_disks);
1995 * Ok, everything is just fine now
1997 mddev->array_size = mddev->size;
1999 mddev->queue->unplug_fn = raid1_unplug;
2000 mddev->queue->backing_dev_info.congested_fn = raid1_congested;
2001 mddev->queue->backing_dev_info.congested_data = mddev;
2003 return 0;
2005 out_no_mem:
2006 printk(KERN_ERR "raid1: couldn't allocate memory for %s\n",
2007 mdname(mddev));
2009 out_free_conf:
2010 if (conf) {
2011 if (conf->r1bio_pool)
2012 mempool_destroy(conf->r1bio_pool);
2013 kfree(conf->mirrors);
2014 safe_put_page(conf->tmppage);
2015 kfree(conf->poolinfo);
2016 kfree(conf);
2017 mddev->private = NULL;
2019 out:
2020 return -EIO;
2023 static int stop(mddev_t *mddev)
2025 conf_t *conf = mddev_to_conf(mddev);
2026 struct bitmap *bitmap = mddev->bitmap;
2027 int behind_wait = 0;
2029 /* wait for behind writes to complete */
2030 while (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
2031 behind_wait++;
2032 printk(KERN_INFO "raid1: behind writes in progress on device %s, waiting to stop (%d)\n", mdname(mddev), behind_wait);
2033 set_current_state(TASK_UNINTERRUPTIBLE);
2034 schedule_timeout(HZ); /* wait a second */
2035 /* need to kick something here to make sure I/O goes? */
2038 md_unregister_thread(mddev->thread);
2039 mddev->thread = NULL;
2040 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
2041 if (conf->r1bio_pool)
2042 mempool_destroy(conf->r1bio_pool);
2043 kfree(conf->mirrors);
2044 kfree(conf->poolinfo);
2045 kfree(conf);
2046 mddev->private = NULL;
2047 return 0;
2050 static int raid1_resize(mddev_t *mddev, sector_t sectors)
2052 /* no resync is happening, and there is enough space
2053 * on all devices, so we can resize.
2054 * We need to make sure resync covers any new space.
2055 * If the array is shrinking we should possibly wait until
2056 * any io in the removed space completes, but it hardly seems
2057 * worth it.
2059 mddev->array_size = sectors>>1;
2060 set_capacity(mddev->gendisk, mddev->array_size << 1);
2061 mddev->changed = 1;
2062 if (mddev->array_size > mddev->size && mddev->recovery_cp == MaxSector) {
2063 mddev->recovery_cp = mddev->size << 1;
2064 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2066 mddev->size = mddev->array_size;
2067 mddev->resync_max_sectors = sectors;
2068 return 0;
2071 static int raid1_reshape(mddev_t *mddev)
2073 /* We need to:
2074 * 1/ resize the r1bio_pool
2075 * 2/ resize conf->mirrors
2077 * We allocate a new r1bio_pool if we can.
2078 * Then raise a device barrier and wait until all IO stops.
2079 * Then resize conf->mirrors and swap in the new r1bio pool.
2081 * At the same time, we "pack" the devices so that all the missing
2082 * devices have the higher raid_disk numbers.
2084 mempool_t *newpool, *oldpool;
2085 struct pool_info *newpoolinfo;
2086 mirror_info_t *newmirrors;
2087 conf_t *conf = mddev_to_conf(mddev);
2088 int cnt, raid_disks;
2089 unsigned long flags;
2090 int d, d2;
2092 /* Cannot change chunk_size, layout, or level */
2093 if (mddev->chunk_size != mddev->new_chunk ||
2094 mddev->layout != mddev->new_layout ||
2095 mddev->level != mddev->new_level) {
2096 mddev->new_chunk = mddev->chunk_size;
2097 mddev->new_layout = mddev->layout;
2098 mddev->new_level = mddev->level;
2099 return -EINVAL;
2102 md_allow_write(mddev);
2104 raid_disks = mddev->raid_disks + mddev->delta_disks;
2106 if (raid_disks < conf->raid_disks) {
2107 cnt=0;
2108 for (d= 0; d < conf->raid_disks; d++)
2109 if (conf->mirrors[d].rdev)
2110 cnt++;
2111 if (cnt > raid_disks)
2112 return -EBUSY;
2115 newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL);
2116 if (!newpoolinfo)
2117 return -ENOMEM;
2118 newpoolinfo->mddev = mddev;
2119 newpoolinfo->raid_disks = raid_disks;
2121 newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
2122 r1bio_pool_free, newpoolinfo);
2123 if (!newpool) {
2124 kfree(newpoolinfo);
2125 return -ENOMEM;
2127 newmirrors = kzalloc(sizeof(struct mirror_info) * raid_disks, GFP_KERNEL);
2128 if (!newmirrors) {
2129 kfree(newpoolinfo);
2130 mempool_destroy(newpool);
2131 return -ENOMEM;
2134 raise_barrier(conf);
2136 /* ok, everything is stopped */
2137 oldpool = conf->r1bio_pool;
2138 conf->r1bio_pool = newpool;
2140 for (d = d2 = 0; d < conf->raid_disks; d++) {
2141 mdk_rdev_t *rdev = conf->mirrors[d].rdev;
2142 if (rdev && rdev->raid_disk != d2) {
2143 char nm[20];
2144 sprintf(nm, "rd%d", rdev->raid_disk);
2145 sysfs_remove_link(&mddev->kobj, nm);
2146 rdev->raid_disk = d2;
2147 sprintf(nm, "rd%d", rdev->raid_disk);
2148 sysfs_remove_link(&mddev->kobj, nm);
2149 if (sysfs_create_link(&mddev->kobj,
2150 &rdev->kobj, nm))
2151 printk(KERN_WARNING
2152 "md/raid1: cannot register "
2153 "%s for %s\n",
2154 nm, mdname(mddev));
2156 if (rdev)
2157 newmirrors[d2++].rdev = rdev;
2159 kfree(conf->mirrors);
2160 conf->mirrors = newmirrors;
2161 kfree(conf->poolinfo);
2162 conf->poolinfo = newpoolinfo;
2164 spin_lock_irqsave(&conf->device_lock, flags);
2165 mddev->degraded += (raid_disks - conf->raid_disks);
2166 spin_unlock_irqrestore(&conf->device_lock, flags);
2167 conf->raid_disks = mddev->raid_disks = raid_disks;
2168 mddev->delta_disks = 0;
2170 conf->last_used = 0; /* just make sure it is in-range */
2171 lower_barrier(conf);
2173 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2174 md_wakeup_thread(mddev->thread);
2176 mempool_destroy(oldpool);
2177 return 0;
2180 static void raid1_quiesce(mddev_t *mddev, int state)
2182 conf_t *conf = mddev_to_conf(mddev);
2184 switch(state) {
2185 case 1:
2186 raise_barrier(conf);
2187 break;
2188 case 0:
2189 lower_barrier(conf);
2190 break;
2195 static struct mdk_personality raid1_personality =
2197 .name = "raid1",
2198 .level = 1,
2199 .owner = THIS_MODULE,
2200 .make_request = make_request,
2201 .run = run,
2202 .stop = stop,
2203 .status = status,
2204 .error_handler = error,
2205 .hot_add_disk = raid1_add_disk,
2206 .hot_remove_disk= raid1_remove_disk,
2207 .spare_active = raid1_spare_active,
2208 .sync_request = sync_request,
2209 .resize = raid1_resize,
2210 .check_reshape = raid1_reshape,
2211 .quiesce = raid1_quiesce,
2214 static int __init raid_init(void)
2216 return register_md_personality(&raid1_personality);
2219 static void raid_exit(void)
2221 unregister_md_personality(&raid1_personality);
2224 module_init(raid_init);
2225 module_exit(raid_exit);
2226 MODULE_LICENSE("GPL");
2227 MODULE_ALIAS("md-personality-3"); /* RAID1 */
2228 MODULE_ALIAS("md-raid1");
2229 MODULE_ALIAS("md-level-1");