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)
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>
40 #define PRINTK(x...) printk(x)
46 * Number of guaranteed r1bios in case of extreme VM load:
48 #define NR_RAID1_BIOS 256
50 static mdk_personality_t raid1_personality;
52 static void unplug_slaves(mddev_t *mddev);
54 static void allow_barrier(conf_t *conf);
55 static void lower_barrier(conf_t *conf);
57 static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data)
59 struct pool_info *pi = data;
61 int size = offsetof(r1bio_t, bios[pi->raid_disks]);
63 /* allocate a r1bio with room for raid_disks entries in the bios array */
64 r1_bio = kmalloc(size, gfp_flags);
66 memset(r1_bio, 0, size);
68 unplug_slaves(pi->mddev);
73 static void r1bio_pool_free(void *r1_bio, void *data)
78 #define RESYNC_BLOCK_SIZE (64*1024)
79 //#define RESYNC_BLOCK_SIZE PAGE_SIZE
80 #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9)
81 #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE)
82 #define RESYNC_WINDOW (2048*1024)
84 static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data)
86 struct pool_info *pi = data;
92 r1_bio = r1bio_pool_alloc(gfp_flags, pi);
94 unplug_slaves(pi->mddev);
99 * Allocate bios : 1 for reading, n-1 for writing
101 for (j = pi->raid_disks ; j-- ; ) {
102 bio = bio_alloc(gfp_flags, RESYNC_PAGES);
105 r1_bio->bios[j] = bio;
108 * Allocate RESYNC_PAGES data pages and attach them to
111 bio = r1_bio->bios[0];
112 for (i = 0; i < RESYNC_PAGES; i++) {
113 page = alloc_page(gfp_flags);
117 bio->bi_io_vec[i].bv_page = page;
120 r1_bio->master_bio = NULL;
126 __free_page(bio->bi_io_vec[i-1].bv_page);
128 while ( ++j < pi->raid_disks )
129 bio_put(r1_bio->bios[j]);
130 r1bio_pool_free(r1_bio, data);
134 static void r1buf_pool_free(void *__r1_bio, void *data)
136 struct pool_info *pi = data;
138 r1bio_t *r1bio = __r1_bio;
139 struct bio *bio = r1bio->bios[0];
141 for (i = 0; i < RESYNC_PAGES; i++) {
142 __free_page(bio->bi_io_vec[i].bv_page);
143 bio->bi_io_vec[i].bv_page = NULL;
145 for (i=0 ; i < pi->raid_disks; i++)
146 bio_put(r1bio->bios[i]);
148 r1bio_pool_free(r1bio, data);
151 static void put_all_bios(conf_t *conf, r1bio_t *r1_bio)
155 for (i = 0; i < conf->raid_disks; i++) {
156 struct bio **bio = r1_bio->bios + i;
163 static inline void free_r1bio(r1bio_t *r1_bio)
165 conf_t *conf = mddev_to_conf(r1_bio->mddev);
168 * Wake up any possible resync thread that waits for the device
173 put_all_bios(conf, r1_bio);
174 mempool_free(r1_bio, conf->r1bio_pool);
177 static inline void put_buf(r1bio_t *r1_bio)
179 conf_t *conf = mddev_to_conf(r1_bio->mddev);
182 for (i=0; i<conf->raid_disks; i++) {
183 struct bio *bio = r1_bio->bios[i];
185 rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev);
188 mempool_free(r1_bio, conf->r1buf_pool);
193 static void reschedule_retry(r1bio_t *r1_bio)
196 mddev_t *mddev = r1_bio->mddev;
197 conf_t *conf = mddev_to_conf(mddev);
199 spin_lock_irqsave(&conf->device_lock, flags);
200 list_add(&r1_bio->retry_list, &conf->retry_list);
202 spin_unlock_irqrestore(&conf->device_lock, flags);
204 wake_up(&conf->wait_barrier);
205 md_wakeup_thread(mddev->thread);
209 * raid_end_bio_io() is called when we have finished servicing a mirrored
210 * operation and are ready to return a success/failure code to the buffer
213 static void raid_end_bio_io(r1bio_t *r1_bio)
215 struct bio *bio = r1_bio->master_bio;
217 /* if nobody has done the final endio yet, do it now */
218 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
219 PRINTK(KERN_DEBUG "raid1: sync end %s on sectors %llu-%llu\n",
220 (bio_data_dir(bio) == WRITE) ? "write" : "read",
221 (unsigned long long) bio->bi_sector,
222 (unsigned long long) bio->bi_sector +
223 (bio->bi_size >> 9) - 1);
225 bio_endio(bio, bio->bi_size,
226 test_bit(R1BIO_Uptodate, &r1_bio->state) ? 0 : -EIO);
232 * Update disk head position estimator based on IRQ completion info.
234 static inline void update_head_pos(int disk, r1bio_t *r1_bio)
236 conf_t *conf = mddev_to_conf(r1_bio->mddev);
238 conf->mirrors[disk].head_position =
239 r1_bio->sector + (r1_bio->sectors);
242 static int raid1_end_read_request(struct bio *bio, unsigned int bytes_done, int error)
244 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
245 r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
247 conf_t *conf = mddev_to_conf(r1_bio->mddev);
252 mirror = r1_bio->read_disk;
254 * this branch is our 'one mirror IO has finished' event handler:
256 update_head_pos(mirror, r1_bio);
258 if (uptodate || conf->working_disks <= 1) {
260 * Set R1BIO_Uptodate in our master bio, so that
261 * we will return a good error code for to the higher
262 * levels even if IO on some other mirrored buffer fails.
264 * The 'master' represents the composite IO operation to
265 * user-side. So if something waits for IO, then it will
266 * wait for the 'master' bio.
268 set_bit(R1BIO_Uptodate, &r1_bio->state);
270 raid_end_bio_io(r1_bio);
275 char b[BDEVNAME_SIZE];
276 if (printk_ratelimit())
277 printk(KERN_ERR "raid1: %s: rescheduling sector %llu\n",
278 bdevname(conf->mirrors[mirror].rdev->bdev,b), (unsigned long long)r1_bio->sector);
279 reschedule_retry(r1_bio);
282 rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev);
286 static int raid1_end_write_request(struct bio *bio, unsigned int bytes_done, int error)
288 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
289 r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
290 int mirror, behind = test_bit(R1BIO_BehindIO, &r1_bio->state);
291 conf_t *conf = mddev_to_conf(r1_bio->mddev);
296 for (mirror = 0; mirror < conf->raid_disks; mirror++)
297 if (r1_bio->bios[mirror] == bio)
300 if (error == -ENOTSUPP && test_bit(R1BIO_Barrier, &r1_bio->state)) {
301 set_bit(BarriersNotsupp, &conf->mirrors[mirror].rdev->flags);
302 set_bit(R1BIO_BarrierRetry, &r1_bio->state);
303 r1_bio->mddev->barriers_work = 0;
306 * this branch is our 'one mirror IO has finished' event handler:
308 r1_bio->bios[mirror] = NULL;
310 md_error(r1_bio->mddev, conf->mirrors[mirror].rdev);
311 /* an I/O failed, we can't clear the bitmap */
312 set_bit(R1BIO_Degraded, &r1_bio->state);
315 * Set R1BIO_Uptodate in our master bio, so that
316 * we will return a good error code for to the higher
317 * levels even if IO on some other mirrored buffer fails.
319 * The 'master' represents the composite IO operation to
320 * user-side. So if something waits for IO, then it will
321 * wait for the 'master' bio.
323 set_bit(R1BIO_Uptodate, &r1_bio->state);
325 update_head_pos(mirror, r1_bio);
328 if (test_bit(WriteMostly, &conf->mirrors[mirror].rdev->flags))
329 atomic_dec(&r1_bio->behind_remaining);
331 /* In behind mode, we ACK the master bio once the I/O has safely
332 * reached all non-writemostly disks. Setting the Returned bit
333 * ensures that this gets done only once -- we don't ever want to
334 * return -EIO here, instead we'll wait */
336 if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) &&
337 test_bit(R1BIO_Uptodate, &r1_bio->state)) {
338 /* Maybe we can return now */
339 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
340 struct bio *mbio = r1_bio->master_bio;
341 PRINTK(KERN_DEBUG "raid1: behind end write sectors %llu-%llu\n",
342 (unsigned long long) mbio->bi_sector,
343 (unsigned long long) mbio->bi_sector +
344 (mbio->bi_size >> 9) - 1);
345 bio_endio(mbio, mbio->bi_size, 0);
352 * Let's see if all mirrored write operations have finished
355 if (atomic_dec_and_test(&r1_bio->remaining)) {
356 if (test_bit(R1BIO_BarrierRetry, &r1_bio->state)) {
357 reschedule_retry(r1_bio);
358 /* Don't dec_pending yet, we want to hold
359 * the reference over the retry
363 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
364 /* free extra copy of the data pages */
365 int i = bio->bi_vcnt;
367 __free_page(bio->bi_io_vec[i].bv_page);
369 /* clear the bitmap if all writes complete successfully */
370 bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector,
372 !test_bit(R1BIO_Degraded, &r1_bio->state),
374 md_write_end(r1_bio->mddev);
375 raid_end_bio_io(r1_bio);
378 if (r1_bio->bios[mirror]==NULL)
381 rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev);
387 * This routine returns the disk from which the requested read should
388 * be done. There is a per-array 'next expected sequential IO' sector
389 * number - if this matches on the next IO then we use the last disk.
390 * There is also a per-disk 'last know head position' sector that is
391 * maintained from IRQ contexts, both the normal and the resync IO
392 * completion handlers update this position correctly. If there is no
393 * perfect sequential match then we pick the disk whose head is closest.
395 * If there are 2 mirrors in the same 2 devices, performance degrades
396 * because position is mirror, not device based.
398 * The rdev for the device selected will have nr_pending incremented.
400 static int read_balance(conf_t *conf, r1bio_t *r1_bio)
402 const unsigned long this_sector = r1_bio->sector;
403 int new_disk = conf->last_used, disk = new_disk;
405 const int sectors = r1_bio->sectors;
406 sector_t new_distance, current_distance;
411 * Check if we can balance. We can balance on the whole
412 * device if no resync is going on, or below the resync window.
413 * We take the first readable disk when above the resync window.
416 if (conf->mddev->recovery_cp < MaxSector &&
417 (this_sector + sectors >= conf->next_resync)) {
418 /* Choose the first operation device, for consistancy */
421 for (rdev = rcu_dereference(conf->mirrors[new_disk].rdev);
422 !rdev || !test_bit(In_sync, &rdev->flags)
423 || test_bit(WriteMostly, &rdev->flags);
424 rdev = rcu_dereference(conf->mirrors[++new_disk].rdev)) {
426 if (rdev && test_bit(In_sync, &rdev->flags))
427 wonly_disk = new_disk;
429 if (new_disk == conf->raid_disks - 1) {
430 new_disk = wonly_disk;
438 /* make sure the disk is operational */
439 for (rdev = rcu_dereference(conf->mirrors[new_disk].rdev);
440 !rdev || !test_bit(In_sync, &rdev->flags) ||
441 test_bit(WriteMostly, &rdev->flags);
442 rdev = rcu_dereference(conf->mirrors[new_disk].rdev)) {
444 if (rdev && test_bit(In_sync, &rdev->flags))
445 wonly_disk = new_disk;
448 new_disk = conf->raid_disks;
450 if (new_disk == disk) {
451 new_disk = wonly_disk;
460 /* now disk == new_disk == starting point for search */
463 * Don't change to another disk for sequential reads:
465 if (conf->next_seq_sect == this_sector)
467 if (this_sector == conf->mirrors[new_disk].head_position)
470 current_distance = abs(this_sector - conf->mirrors[disk].head_position);
472 /* Find the disk whose head is closest */
476 disk = conf->raid_disks;
479 rdev = rcu_dereference(conf->mirrors[disk].rdev);
482 !test_bit(In_sync, &rdev->flags) ||
483 test_bit(WriteMostly, &rdev->flags))
486 if (!atomic_read(&rdev->nr_pending)) {
490 new_distance = abs(this_sector - conf->mirrors[disk].head_position);
491 if (new_distance < current_distance) {
492 current_distance = new_distance;
495 } while (disk != conf->last_used);
501 rdev = rcu_dereference(conf->mirrors[new_disk].rdev);
504 atomic_inc(&rdev->nr_pending);
505 if (!test_bit(In_sync, &rdev->flags)) {
506 /* cannot risk returning a device that failed
507 * before we inc'ed nr_pending
509 atomic_dec(&rdev->nr_pending);
512 conf->next_seq_sect = this_sector + sectors;
513 conf->last_used = new_disk;
520 static void unplug_slaves(mddev_t *mddev)
522 conf_t *conf = mddev_to_conf(mddev);
526 for (i=0; i<mddev->raid_disks; i++) {
527 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
528 if (rdev && !test_bit(Faulty, &rdev->flags) && atomic_read(&rdev->nr_pending)) {
529 request_queue_t *r_queue = bdev_get_queue(rdev->bdev);
531 atomic_inc(&rdev->nr_pending);
534 if (r_queue->unplug_fn)
535 r_queue->unplug_fn(r_queue);
537 rdev_dec_pending(rdev, mddev);
544 static void raid1_unplug(request_queue_t *q)
546 mddev_t *mddev = q->queuedata;
548 unplug_slaves(mddev);
549 md_wakeup_thread(mddev->thread);
552 static int raid1_issue_flush(request_queue_t *q, struct gendisk *disk,
553 sector_t *error_sector)
555 mddev_t *mddev = q->queuedata;
556 conf_t *conf = mddev_to_conf(mddev);
560 for (i=0; i<mddev->raid_disks && ret == 0; i++) {
561 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
562 if (rdev && !test_bit(Faulty, &rdev->flags)) {
563 struct block_device *bdev = rdev->bdev;
564 request_queue_t *r_queue = bdev_get_queue(bdev);
566 if (!r_queue->issue_flush_fn)
569 atomic_inc(&rdev->nr_pending);
571 ret = r_queue->issue_flush_fn(r_queue, bdev->bd_disk,
573 rdev_dec_pending(rdev, mddev);
583 * Sometimes we need to suspend IO while we do something else,
584 * either some resync/recovery, or reconfigure the array.
585 * To do this we raise a 'barrier'.
586 * The 'barrier' is a counter that can be raised multiple times
587 * to count how many activities are happening which preclude
589 * We can only raise the barrier if there is no pending IO.
590 * i.e. if nr_pending == 0.
591 * We choose only to raise the barrier if no-one is waiting for the
592 * barrier to go down. This means that as soon as an IO request
593 * is ready, no other operations which require a barrier will start
594 * until the IO request has had a chance.
596 * So: regular IO calls 'wait_barrier'. When that returns there
597 * is no backgroup IO happening, It must arrange to call
598 * allow_barrier when it has finished its IO.
599 * backgroup IO calls must call raise_barrier. Once that returns
600 * there is no normal IO happeing. It must arrange to call
601 * lower_barrier when the particular background IO completes.
603 #define RESYNC_DEPTH 32
605 static void raise_barrier(conf_t *conf)
607 spin_lock_irq(&conf->resync_lock);
609 /* Wait until no block IO is waiting */
610 wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting,
612 raid1_unplug(conf->mddev->queue));
614 /* block any new IO from starting */
617 /* No wait for all pending IO to complete */
618 wait_event_lock_irq(conf->wait_barrier,
619 !conf->nr_pending && conf->barrier < RESYNC_DEPTH,
621 raid1_unplug(conf->mddev->queue));
623 spin_unlock_irq(&conf->resync_lock);
626 static void lower_barrier(conf_t *conf)
629 spin_lock_irqsave(&conf->resync_lock, flags);
631 spin_unlock_irqrestore(&conf->resync_lock, flags);
632 wake_up(&conf->wait_barrier);
635 static void wait_barrier(conf_t *conf)
637 spin_lock_irq(&conf->resync_lock);
640 wait_event_lock_irq(conf->wait_barrier, !conf->barrier,
642 raid1_unplug(conf->mddev->queue));
646 spin_unlock_irq(&conf->resync_lock);
649 static void allow_barrier(conf_t *conf)
652 spin_lock_irqsave(&conf->resync_lock, flags);
654 spin_unlock_irqrestore(&conf->resync_lock, flags);
655 wake_up(&conf->wait_barrier);
658 static void freeze_array(conf_t *conf)
660 /* stop syncio and normal IO and wait for everything to
662 * We increment barrier and nr_waiting, and then
663 * wait until barrier+nr_pending match nr_queued+2
665 spin_lock_irq(&conf->resync_lock);
668 wait_event_lock_irq(conf->wait_barrier,
669 conf->barrier+conf->nr_pending == conf->nr_queued+2,
671 raid1_unplug(conf->mddev->queue));
672 spin_unlock_irq(&conf->resync_lock);
674 static void unfreeze_array(conf_t *conf)
676 /* reverse the effect of the freeze */
677 spin_lock_irq(&conf->resync_lock);
680 wake_up(&conf->wait_barrier);
681 spin_unlock_irq(&conf->resync_lock);
685 /* duplicate the data pages for behind I/O */
686 static struct page **alloc_behind_pages(struct bio *bio)
689 struct bio_vec *bvec;
690 struct page **pages = kmalloc(bio->bi_vcnt * sizeof(struct page *),
692 if (unlikely(!pages))
695 memset(pages, 0, bio->bi_vcnt * sizeof(struct page *));
697 bio_for_each_segment(bvec, bio, i) {
698 pages[i] = alloc_page(GFP_NOIO);
699 if (unlikely(!pages[i]))
701 memcpy(kmap(pages[i]) + bvec->bv_offset,
702 kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len);
704 kunmap(bvec->bv_page);
711 for (i = 0; i < bio->bi_vcnt && pages[i]; i++)
712 __free_page(pages[i]);
714 PRINTK("%dB behind alloc failed, doing sync I/O\n", bio->bi_size);
718 static int make_request(request_queue_t *q, struct bio * bio)
720 mddev_t *mddev = q->queuedata;
721 conf_t *conf = mddev_to_conf(mddev);
722 mirror_info_t *mirror;
724 struct bio *read_bio;
725 int i, targets = 0, disks;
727 struct bitmap *bitmap = mddev->bitmap;
730 struct page **behind_pages = NULL;
731 const int rw = bio_data_dir(bio);
734 if (unlikely(!mddev->barriers_work && bio_barrier(bio))) {
735 bio_endio(bio, bio->bi_size, -EOPNOTSUPP);
740 * Register the new request and wait if the reconstruction
741 * thread has put up a bar for new requests.
742 * Continue immediately if no resync is active currently.
744 md_write_start(mddev, bio); /* wait on superblock update early */
748 disk_stat_inc(mddev->gendisk, ios[rw]);
749 disk_stat_add(mddev->gendisk, sectors[rw], bio_sectors(bio));
752 * make_request() can abort the operation when READA is being
753 * used and no empty request is available.
756 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
758 r1_bio->master_bio = bio;
759 r1_bio->sectors = bio->bi_size >> 9;
761 r1_bio->mddev = mddev;
762 r1_bio->sector = bio->bi_sector;
766 * read balancing logic:
768 int rdisk = read_balance(conf, r1_bio);
771 /* couldn't find anywhere to read from */
772 raid_end_bio_io(r1_bio);
775 mirror = conf->mirrors + rdisk;
777 r1_bio->read_disk = rdisk;
779 read_bio = bio_clone(bio, GFP_NOIO);
781 r1_bio->bios[rdisk] = read_bio;
783 read_bio->bi_sector = r1_bio->sector + mirror->rdev->data_offset;
784 read_bio->bi_bdev = mirror->rdev->bdev;
785 read_bio->bi_end_io = raid1_end_read_request;
786 read_bio->bi_rw = READ;
787 read_bio->bi_private = r1_bio;
789 generic_make_request(read_bio);
796 /* first select target devices under spinlock and
797 * inc refcount on their rdev. Record them by setting
800 disks = conf->raid_disks;
802 { static int first=1;
803 if (first) printk("First Write sector %llu disks %d\n",
804 (unsigned long long)r1_bio->sector, disks);
809 for (i = 0; i < disks; i++) {
810 if ((rdev=rcu_dereference(conf->mirrors[i].rdev)) != NULL &&
811 !test_bit(Faulty, &rdev->flags)) {
812 atomic_inc(&rdev->nr_pending);
813 if (test_bit(Faulty, &rdev->flags)) {
814 atomic_dec(&rdev->nr_pending);
815 r1_bio->bios[i] = NULL;
817 r1_bio->bios[i] = bio;
820 r1_bio->bios[i] = NULL;
824 BUG_ON(targets == 0); /* we never fail the last device */
826 if (targets < conf->raid_disks) {
827 /* array is degraded, we will not clear the bitmap
828 * on I/O completion (see raid1_end_write_request) */
829 set_bit(R1BIO_Degraded, &r1_bio->state);
832 /* do behind I/O ? */
834 atomic_read(&bitmap->behind_writes) < bitmap->max_write_behind &&
835 (behind_pages = alloc_behind_pages(bio)) != NULL)
836 set_bit(R1BIO_BehindIO, &r1_bio->state);
838 atomic_set(&r1_bio->remaining, 0);
839 atomic_set(&r1_bio->behind_remaining, 0);
841 do_barriers = bio->bi_rw & BIO_RW_BARRIER;
843 set_bit(R1BIO_Barrier, &r1_bio->state);
846 for (i = 0; i < disks; i++) {
848 if (!r1_bio->bios[i])
851 mbio = bio_clone(bio, GFP_NOIO);
852 r1_bio->bios[i] = mbio;
854 mbio->bi_sector = r1_bio->sector + conf->mirrors[i].rdev->data_offset;
855 mbio->bi_bdev = conf->mirrors[i].rdev->bdev;
856 mbio->bi_end_io = raid1_end_write_request;
857 mbio->bi_rw = WRITE | do_barriers;
858 mbio->bi_private = r1_bio;
861 struct bio_vec *bvec;
864 /* Yes, I really want the '__' version so that
865 * we clear any unused pointer in the io_vec, rather
866 * than leave them unchanged. This is important
867 * because when we come to free the pages, we won't
868 * know the originial bi_idx, so we just free
871 __bio_for_each_segment(bvec, mbio, j, 0)
872 bvec->bv_page = behind_pages[j];
873 if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags))
874 atomic_inc(&r1_bio->behind_remaining);
877 atomic_inc(&r1_bio->remaining);
879 bio_list_add(&bl, mbio);
881 kfree(behind_pages); /* the behind pages are attached to the bios now */
883 bitmap_startwrite(bitmap, bio->bi_sector, r1_bio->sectors,
884 test_bit(R1BIO_BehindIO, &r1_bio->state));
885 spin_lock_irqsave(&conf->device_lock, flags);
886 bio_list_merge(&conf->pending_bio_list, &bl);
889 blk_plug_device(mddev->queue);
890 spin_unlock_irqrestore(&conf->device_lock, flags);
893 while ((bio = bio_list_pop(&bl)) != NULL)
894 generic_make_request(bio);
900 static void status(struct seq_file *seq, mddev_t *mddev)
902 conf_t *conf = mddev_to_conf(mddev);
905 seq_printf(seq, " [%d/%d] [", conf->raid_disks,
906 conf->working_disks);
907 for (i = 0; i < conf->raid_disks; i++)
908 seq_printf(seq, "%s",
909 conf->mirrors[i].rdev &&
910 test_bit(In_sync, &conf->mirrors[i].rdev->flags) ? "U" : "_");
911 seq_printf(seq, "]");
915 static void error(mddev_t *mddev, mdk_rdev_t *rdev)
917 char b[BDEVNAME_SIZE];
918 conf_t *conf = mddev_to_conf(mddev);
921 * If it is not operational, then we have already marked it as dead
922 * else if it is the last working disks, ignore the error, let the
923 * next level up know.
924 * else mark the drive as failed
926 if (test_bit(In_sync, &rdev->flags)
927 && conf->working_disks == 1)
929 * Don't fail the drive, act as though we were just a
930 * normal single drive
933 if (test_bit(In_sync, &rdev->flags)) {
935 conf->working_disks--;
937 * if recovery is running, make sure it aborts.
939 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
941 clear_bit(In_sync, &rdev->flags);
942 set_bit(Faulty, &rdev->flags);
944 printk(KERN_ALERT "raid1: Disk failure on %s, disabling device. \n"
945 " Operation continuing on %d devices\n",
946 bdevname(rdev->bdev,b), conf->working_disks);
949 static void print_conf(conf_t *conf)
954 printk("RAID1 conf printout:\n");
959 printk(" --- wd:%d rd:%d\n", conf->working_disks,
962 for (i = 0; i < conf->raid_disks; i++) {
963 char b[BDEVNAME_SIZE];
964 tmp = conf->mirrors + i;
966 printk(" disk %d, wo:%d, o:%d, dev:%s\n",
967 i, !test_bit(In_sync, &tmp->rdev->flags), !test_bit(Faulty, &tmp->rdev->flags),
968 bdevname(tmp->rdev->bdev,b));
972 static void close_sync(conf_t *conf)
977 mempool_destroy(conf->r1buf_pool);
978 conf->r1buf_pool = NULL;
981 static int raid1_spare_active(mddev_t *mddev)
984 conf_t *conf = mddev->private;
988 * Find all failed disks within the RAID1 configuration
989 * and mark them readable
991 for (i = 0; i < conf->raid_disks; i++) {
992 tmp = conf->mirrors + i;
994 && !test_bit(Faulty, &tmp->rdev->flags)
995 && !test_bit(In_sync, &tmp->rdev->flags)) {
996 conf->working_disks++;
998 set_bit(In_sync, &tmp->rdev->flags);
1007 static int raid1_add_disk(mddev_t *mddev, mdk_rdev_t *rdev)
1009 conf_t *conf = mddev->private;
1014 for (mirror=0; mirror < mddev->raid_disks; mirror++)
1015 if ( !(p=conf->mirrors+mirror)->rdev) {
1017 blk_queue_stack_limits(mddev->queue,
1018 rdev->bdev->bd_disk->queue);
1019 /* as we don't honour merge_bvec_fn, we must never risk
1020 * violating it, so limit ->max_sector to one PAGE, as
1021 * a one page request is never in violation.
1023 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
1024 mddev->queue->max_sectors > (PAGE_SIZE>>9))
1025 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
1027 p->head_position = 0;
1028 rdev->raid_disk = mirror;
1030 /* As all devices are equivalent, we don't need a full recovery
1031 * if this was recently any drive of the array
1033 if (rdev->saved_raid_disk < 0)
1035 rcu_assign_pointer(p->rdev, rdev);
1043 static int raid1_remove_disk(mddev_t *mddev, int number)
1045 conf_t *conf = mddev->private;
1048 mirror_info_t *p = conf->mirrors+ number;
1053 if (test_bit(In_sync, &rdev->flags) ||
1054 atomic_read(&rdev->nr_pending)) {
1060 if (atomic_read(&rdev->nr_pending)) {
1061 /* lost the race, try later */
1073 static int end_sync_read(struct bio *bio, unsigned int bytes_done, int error)
1075 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1076 r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
1077 conf_t *conf = mddev_to_conf(r1_bio->mddev);
1082 if (r1_bio->bios[r1_bio->read_disk] != bio)
1084 update_head_pos(r1_bio->read_disk, r1_bio);
1086 * we have read a block, now it needs to be re-written,
1087 * or re-read if the read failed.
1088 * We don't do much here, just schedule handling by raid1d
1091 md_error(r1_bio->mddev,
1092 conf->mirrors[r1_bio->read_disk].rdev);
1094 set_bit(R1BIO_Uptodate, &r1_bio->state);
1095 reschedule_retry(r1_bio);
1099 static int end_sync_write(struct bio *bio, unsigned int bytes_done, int error)
1101 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1102 r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
1103 mddev_t *mddev = r1_bio->mddev;
1104 conf_t *conf = mddev_to_conf(mddev);
1111 for (i = 0; i < conf->raid_disks; i++)
1112 if (r1_bio->bios[i] == bio) {
1117 md_error(mddev, conf->mirrors[mirror].rdev);
1119 update_head_pos(mirror, r1_bio);
1121 if (atomic_dec_and_test(&r1_bio->remaining)) {
1122 md_done_sync(mddev, r1_bio->sectors, uptodate);
1128 static void sync_request_write(mddev_t *mddev, r1bio_t *r1_bio)
1130 conf_t *conf = mddev_to_conf(mddev);
1132 int disks = conf->raid_disks;
1133 struct bio *bio, *wbio;
1135 bio = r1_bio->bios[r1_bio->read_disk];
1138 if (r1_bio->sector == 0) printk("First sync write startss\n");
1143 if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) {
1145 * There is no point trying a read-for-reconstruct as
1146 * reconstruct is about to be aborted
1148 char b[BDEVNAME_SIZE];
1149 printk(KERN_ALERT "raid1: %s: unrecoverable I/O read error"
1150 " for block %llu\n",
1151 bdevname(bio->bi_bdev,b),
1152 (unsigned long long)r1_bio->sector);
1153 md_done_sync(mddev, r1_bio->sectors, 0);
1158 atomic_set(&r1_bio->remaining, 1);
1159 for (i = 0; i < disks ; i++) {
1160 wbio = r1_bio->bios[i];
1161 if (wbio->bi_end_io == NULL ||
1162 (wbio->bi_end_io == end_sync_read &&
1163 (i == r1_bio->read_disk ||
1164 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery))))
1167 wbio->bi_rw = WRITE;
1168 wbio->bi_end_io = end_sync_write;
1169 atomic_inc(&r1_bio->remaining);
1170 md_sync_acct(conf->mirrors[i].rdev->bdev, wbio->bi_size >> 9);
1172 generic_make_request(wbio);
1175 if (atomic_dec_and_test(&r1_bio->remaining)) {
1176 /* if we're here, all write(s) have completed, so clean up */
1177 md_done_sync(mddev, r1_bio->sectors, 1);
1183 * This is a kernel thread which:
1185 * 1. Retries failed read operations on working mirrors.
1186 * 2. Updates the raid superblock when problems encounter.
1187 * 3. Performs writes following reads for array syncronising.
1190 static void raid1d(mddev_t *mddev)
1194 unsigned long flags;
1195 conf_t *conf = mddev_to_conf(mddev);
1196 struct list_head *head = &conf->retry_list;
1200 md_check_recovery(mddev);
1203 char b[BDEVNAME_SIZE];
1204 spin_lock_irqsave(&conf->device_lock, flags);
1206 if (conf->pending_bio_list.head) {
1207 bio = bio_list_get(&conf->pending_bio_list);
1208 blk_remove_plug(mddev->queue);
1209 spin_unlock_irqrestore(&conf->device_lock, flags);
1210 /* flush any pending bitmap writes to disk before proceeding w/ I/O */
1211 if (bitmap_unplug(mddev->bitmap) != 0)
1212 printk("%s: bitmap file write failed!\n", mdname(mddev));
1214 while (bio) { /* submit pending writes */
1215 struct bio *next = bio->bi_next;
1216 bio->bi_next = NULL;
1217 generic_make_request(bio);
1225 if (list_empty(head))
1227 r1_bio = list_entry(head->prev, r1bio_t, retry_list);
1228 list_del(head->prev);
1230 spin_unlock_irqrestore(&conf->device_lock, flags);
1232 mddev = r1_bio->mddev;
1233 conf = mddev_to_conf(mddev);
1234 if (test_bit(R1BIO_IsSync, &r1_bio->state)) {
1235 sync_request_write(mddev, r1_bio);
1237 } else if (test_bit(R1BIO_BarrierRetry, &r1_bio->state)) {
1238 /* some requests in the r1bio were BIO_RW_BARRIER
1239 * requests which failed with -ENOTSUPP. Hohumm..
1240 * Better resubmit without the barrier.
1241 * We know which devices to resubmit for, because
1242 * all others have had their bios[] entry cleared.
1245 clear_bit(R1BIO_BarrierRetry, &r1_bio->state);
1246 clear_bit(R1BIO_Barrier, &r1_bio->state);
1247 for (i=0; i < conf->raid_disks; i++)
1248 if (r1_bio->bios[i]) {
1249 struct bio_vec *bvec;
1252 bio = bio_clone(r1_bio->master_bio, GFP_NOIO);
1253 /* copy pages from the failed bio, as
1254 * this might be a write-behind device */
1255 __bio_for_each_segment(bvec, bio, j, 0)
1256 bvec->bv_page = bio_iovec_idx(r1_bio->bios[i], j)->bv_page;
1257 bio_put(r1_bio->bios[i]);
1258 bio->bi_sector = r1_bio->sector +
1259 conf->mirrors[i].rdev->data_offset;
1260 bio->bi_bdev = conf->mirrors[i].rdev->bdev;
1261 bio->bi_end_io = raid1_end_write_request;
1263 bio->bi_private = r1_bio;
1264 r1_bio->bios[i] = bio;
1265 generic_make_request(bio);
1270 /* we got a read error. Maybe the drive is bad. Maybe just
1271 * the block and we can fix it.
1272 * We freeze all other IO, and try reading the block from
1273 * other devices. When we find one, we re-write
1274 * and check it that fixes the read error.
1275 * This is all done synchronously while the array is
1278 sector_t sect = r1_bio->sector;
1279 int sectors = r1_bio->sectors;
1283 int d = r1_bio->read_disk;
1286 if (s > (PAGE_SIZE>>9))
1290 rdev = conf->mirrors[d].rdev;
1292 test_bit(In_sync, &rdev->flags) &&
1293 sync_page_io(rdev->bdev,
1294 sect + rdev->data_offset,
1296 conf->tmppage, READ))
1300 if (d == conf->raid_disks)
1303 } while (!success && d != r1_bio->read_disk);
1306 /* write it back and re-read */
1307 while (d != r1_bio->read_disk) {
1309 d = conf->raid_disks;
1311 rdev = conf->mirrors[d].rdev;
1313 test_bit(In_sync, &rdev->flags)) {
1314 if (sync_page_io(rdev->bdev,
1315 sect + rdev->data_offset,
1316 s<<9, conf->tmppage, WRITE) == 0 ||
1317 sync_page_io(rdev->bdev,
1318 sect + rdev->data_offset,
1319 s<<9, conf->tmppage, READ) == 0) {
1320 /* Well, this device is dead */
1321 md_error(mddev, rdev);
1326 /* Cannot read from anywhere -- bye bye array */
1327 md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev);
1335 unfreeze_array(conf);
1337 bio = r1_bio->bios[r1_bio->read_disk];
1338 if ((disk=read_balance(conf, r1_bio)) == -1) {
1339 printk(KERN_ALERT "raid1: %s: unrecoverable I/O"
1340 " read error for block %llu\n",
1341 bdevname(bio->bi_bdev,b),
1342 (unsigned long long)r1_bio->sector);
1343 raid_end_bio_io(r1_bio);
1345 r1_bio->bios[r1_bio->read_disk] = NULL;
1346 r1_bio->read_disk = disk;
1348 bio = bio_clone(r1_bio->master_bio, GFP_NOIO);
1349 r1_bio->bios[r1_bio->read_disk] = bio;
1350 rdev = conf->mirrors[disk].rdev;
1351 if (printk_ratelimit())
1352 printk(KERN_ERR "raid1: %s: redirecting sector %llu to"
1353 " another mirror\n",
1354 bdevname(rdev->bdev,b),
1355 (unsigned long long)r1_bio->sector);
1356 bio->bi_sector = r1_bio->sector + rdev->data_offset;
1357 bio->bi_bdev = rdev->bdev;
1358 bio->bi_end_io = raid1_end_read_request;
1360 bio->bi_private = r1_bio;
1362 generic_make_request(bio);
1366 spin_unlock_irqrestore(&conf->device_lock, flags);
1368 unplug_slaves(mddev);
1372 static int init_resync(conf_t *conf)
1376 buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;
1377 if (conf->r1buf_pool)
1379 conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free,
1381 if (!conf->r1buf_pool)
1383 conf->next_resync = 0;
1388 * perform a "sync" on one "block"
1390 * We need to make sure that no normal I/O request - particularly write
1391 * requests - conflict with active sync requests.
1393 * This is achieved by tracking pending requests and a 'barrier' concept
1394 * that can be installed to exclude normal IO requests.
1397 static sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster)
1399 conf_t *conf = mddev_to_conf(mddev);
1402 sector_t max_sector, nr_sectors;
1406 int write_targets = 0, read_targets = 0;
1408 int still_degraded = 0;
1410 if (!conf->r1buf_pool)
1413 printk("sync start - bitmap %p\n", mddev->bitmap);
1415 if (init_resync(conf))
1419 max_sector = mddev->size << 1;
1420 if (sector_nr >= max_sector) {
1421 /* If we aborted, we need to abort the
1422 * sync on the 'current' bitmap chunk (there will
1423 * only be one in raid1 resync.
1424 * We can find the current addess in mddev->curr_resync
1426 if (mddev->curr_resync < max_sector) /* aborted */
1427 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
1429 else /* completed sync */
1432 bitmap_close_sync(mddev->bitmap);
1437 /* before building a request, check if we can skip these blocks..
1438 * This call the bitmap_start_sync doesn't actually record anything
1440 if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
1441 !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
1442 /* We can skip this block, and probably several more */
1447 * If there is non-resync activity waiting for a turn,
1448 * and resync is going fast enough,
1449 * then let it though before starting on this new sync request.
1451 if (!go_faster && conf->nr_waiting)
1452 msleep_interruptible(1000);
1454 raise_barrier(conf);
1456 conf->next_resync = sector_nr;
1458 r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO);
1461 * If we get a correctably read error during resync or recovery,
1462 * we might want to read from a different device. So we
1463 * flag all drives that could conceivably be read from for READ,
1464 * and any others (which will be non-In_sync devices) for WRITE.
1465 * If a read fails, we try reading from something else for which READ
1469 r1_bio->mddev = mddev;
1470 r1_bio->sector = sector_nr;
1472 set_bit(R1BIO_IsSync, &r1_bio->state);
1474 for (i=0; i < conf->raid_disks; i++) {
1476 bio = r1_bio->bios[i];
1478 /* take from bio_init */
1479 bio->bi_next = NULL;
1480 bio->bi_flags |= 1 << BIO_UPTODATE;
1484 bio->bi_phys_segments = 0;
1485 bio->bi_hw_segments = 0;
1487 bio->bi_end_io = NULL;
1488 bio->bi_private = NULL;
1490 rdev = rcu_dereference(conf->mirrors[i].rdev);
1492 test_bit(Faulty, &rdev->flags)) {
1495 } else if (!test_bit(In_sync, &rdev->flags)) {
1497 bio->bi_end_io = end_sync_write;
1500 /* may need to read from here */
1502 bio->bi_end_io = end_sync_read;
1503 if (test_bit(WriteMostly, &rdev->flags)) {
1512 atomic_inc(&rdev->nr_pending);
1513 bio->bi_sector = sector_nr + rdev->data_offset;
1514 bio->bi_bdev = rdev->bdev;
1515 bio->bi_private = r1_bio;
1520 r1_bio->read_disk = disk;
1522 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0)
1523 /* extra read targets are also write targets */
1524 write_targets += read_targets-1;
1526 if (write_targets == 0 || read_targets == 0) {
1527 /* There is nowhere to write, so all non-sync
1528 * drives must be failed - so we are finished
1530 sector_t rv = max_sector - sector_nr;
1540 int len = PAGE_SIZE;
1541 if (sector_nr + (len>>9) > max_sector)
1542 len = (max_sector - sector_nr) << 9;
1545 if (sync_blocks == 0) {
1546 if (!bitmap_start_sync(mddev->bitmap, sector_nr,
1547 &sync_blocks, still_degraded) &&
1549 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1551 if (sync_blocks < (PAGE_SIZE>>9))
1553 if (len > (sync_blocks<<9))
1554 len = sync_blocks<<9;
1557 for (i=0 ; i < conf->raid_disks; i++) {
1558 bio = r1_bio->bios[i];
1559 if (bio->bi_end_io) {
1560 page = r1_bio->bios[0]->bi_io_vec[bio->bi_vcnt].bv_page;
1561 if (bio_add_page(bio, page, len, 0) == 0) {
1563 r1_bio->bios[0]->bi_io_vec[bio->bi_vcnt].bv_page = page;
1566 bio = r1_bio->bios[i];
1567 if (bio->bi_end_io==NULL)
1569 /* remove last page from this bio */
1571 bio->bi_size -= len;
1572 bio->bi_flags &= ~(1<< BIO_SEG_VALID);
1578 nr_sectors += len>>9;
1579 sector_nr += len>>9;
1580 sync_blocks -= (len>>9);
1581 } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES);
1583 bio = r1_bio->bios[r1_bio->read_disk];
1584 r1_bio->sectors = nr_sectors;
1586 md_sync_acct(conf->mirrors[r1_bio->read_disk].rdev->bdev, nr_sectors);
1588 generic_make_request(bio);
1593 static int run(mddev_t *mddev)
1597 mirror_info_t *disk;
1599 struct list_head *tmp;
1601 if (mddev->level != 1) {
1602 printk("raid1: %s: raid level not set to mirroring (%d)\n",
1603 mdname(mddev), mddev->level);
1607 * copy the already verified devices into our private RAID1
1608 * bookkeeping area. [whatever we allocate in run(),
1609 * should be freed in stop()]
1611 conf = kmalloc(sizeof(conf_t), GFP_KERNEL);
1612 mddev->private = conf;
1616 memset(conf, 0, sizeof(*conf));
1617 conf->mirrors = kmalloc(sizeof(struct mirror_info)*mddev->raid_disks,
1622 memset(conf->mirrors, 0, sizeof(struct mirror_info)*mddev->raid_disks);
1624 conf->tmppage = alloc_page(GFP_KERNEL);
1628 conf->poolinfo = kmalloc(sizeof(*conf->poolinfo), GFP_KERNEL);
1629 if (!conf->poolinfo)
1631 conf->poolinfo->mddev = mddev;
1632 conf->poolinfo->raid_disks = mddev->raid_disks;
1633 conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
1636 if (!conf->r1bio_pool)
1639 ITERATE_RDEV(mddev, rdev, tmp) {
1640 disk_idx = rdev->raid_disk;
1641 if (disk_idx >= mddev->raid_disks
1644 disk = conf->mirrors + disk_idx;
1648 blk_queue_stack_limits(mddev->queue,
1649 rdev->bdev->bd_disk->queue);
1650 /* as we don't honour merge_bvec_fn, we must never risk
1651 * violating it, so limit ->max_sector to one PAGE, as
1652 * a one page request is never in violation.
1654 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
1655 mddev->queue->max_sectors > (PAGE_SIZE>>9))
1656 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
1658 disk->head_position = 0;
1659 if (!test_bit(Faulty, &rdev->flags) && test_bit(In_sync, &rdev->flags))
1660 conf->working_disks++;
1662 conf->raid_disks = mddev->raid_disks;
1663 conf->mddev = mddev;
1664 spin_lock_init(&conf->device_lock);
1665 INIT_LIST_HEAD(&conf->retry_list);
1666 if (conf->working_disks == 1)
1667 mddev->recovery_cp = MaxSector;
1669 spin_lock_init(&conf->resync_lock);
1670 init_waitqueue_head(&conf->wait_barrier);
1672 bio_list_init(&conf->pending_bio_list);
1673 bio_list_init(&conf->flushing_bio_list);
1675 if (!conf->working_disks) {
1676 printk(KERN_ERR "raid1: no operational mirrors for %s\n",
1681 mddev->degraded = 0;
1682 for (i = 0; i < conf->raid_disks; i++) {
1684 disk = conf->mirrors + i;
1687 disk->head_position = 0;
1693 * find the first working one and use it as a starting point
1694 * to read balancing.
1696 for (j = 0; j < conf->raid_disks &&
1697 (!conf->mirrors[j].rdev ||
1698 !test_bit(In_sync, &conf->mirrors[j].rdev->flags)) ; j++)
1700 conf->last_used = j;
1703 mddev->thread = md_register_thread(raid1d, mddev, "%s_raid1");
1704 if (!mddev->thread) {
1706 "raid1: couldn't allocate thread for %s\n",
1712 "raid1: raid set %s active with %d out of %d mirrors\n",
1713 mdname(mddev), mddev->raid_disks - mddev->degraded,
1716 * Ok, everything is just fine now
1718 mddev->array_size = mddev->size;
1720 mddev->queue->unplug_fn = raid1_unplug;
1721 mddev->queue->issue_flush_fn = raid1_issue_flush;
1726 printk(KERN_ERR "raid1: couldn't allocate memory for %s\n",
1731 if (conf->r1bio_pool)
1732 mempool_destroy(conf->r1bio_pool);
1733 kfree(conf->mirrors);
1734 __free_page(conf->tmppage);
1735 kfree(conf->poolinfo);
1737 mddev->private = NULL;
1743 static int stop(mddev_t *mddev)
1745 conf_t *conf = mddev_to_conf(mddev);
1746 struct bitmap *bitmap = mddev->bitmap;
1747 int behind_wait = 0;
1749 /* wait for behind writes to complete */
1750 while (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
1752 printk(KERN_INFO "raid1: behind writes in progress on device %s, waiting to stop (%d)\n", mdname(mddev), behind_wait);
1753 set_current_state(TASK_UNINTERRUPTIBLE);
1754 schedule_timeout(HZ); /* wait a second */
1755 /* need to kick something here to make sure I/O goes? */
1758 md_unregister_thread(mddev->thread);
1759 mddev->thread = NULL;
1760 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
1761 if (conf->r1bio_pool)
1762 mempool_destroy(conf->r1bio_pool);
1763 kfree(conf->mirrors);
1764 kfree(conf->poolinfo);
1766 mddev->private = NULL;
1770 static int raid1_resize(mddev_t *mddev, sector_t sectors)
1772 /* no resync is happening, and there is enough space
1773 * on all devices, so we can resize.
1774 * We need to make sure resync covers any new space.
1775 * If the array is shrinking we should possibly wait until
1776 * any io in the removed space completes, but it hardly seems
1779 mddev->array_size = sectors>>1;
1780 set_capacity(mddev->gendisk, mddev->array_size << 1);
1782 if (mddev->array_size > mddev->size && mddev->recovery_cp == MaxSector) {
1783 mddev->recovery_cp = mddev->size << 1;
1784 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1786 mddev->size = mddev->array_size;
1787 mddev->resync_max_sectors = sectors;
1791 static int raid1_reshape(mddev_t *mddev, int raid_disks)
1794 * 1/ resize the r1bio_pool
1795 * 2/ resize conf->mirrors
1797 * We allocate a new r1bio_pool if we can.
1798 * Then raise a device barrier and wait until all IO stops.
1799 * Then resize conf->mirrors and swap in the new r1bio pool.
1801 * At the same time, we "pack" the devices so that all the missing
1802 * devices have the higher raid_disk numbers.
1804 mempool_t *newpool, *oldpool;
1805 struct pool_info *newpoolinfo;
1806 mirror_info_t *newmirrors;
1807 conf_t *conf = mddev_to_conf(mddev);
1812 if (raid_disks < conf->raid_disks) {
1814 for (d= 0; d < conf->raid_disks; d++)
1815 if (conf->mirrors[d].rdev)
1817 if (cnt > raid_disks)
1821 newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL);
1824 newpoolinfo->mddev = mddev;
1825 newpoolinfo->raid_disks = raid_disks;
1827 newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
1828 r1bio_pool_free, newpoolinfo);
1833 newmirrors = kmalloc(sizeof(struct mirror_info) * raid_disks, GFP_KERNEL);
1836 mempool_destroy(newpool);
1839 memset(newmirrors, 0, sizeof(struct mirror_info)*raid_disks);
1841 raise_barrier(conf);
1843 /* ok, everything is stopped */
1844 oldpool = conf->r1bio_pool;
1845 conf->r1bio_pool = newpool;
1847 for (d=d2=0; d < conf->raid_disks; d++)
1848 if (conf->mirrors[d].rdev) {
1849 conf->mirrors[d].rdev->raid_disk = d2;
1850 newmirrors[d2++].rdev = conf->mirrors[d].rdev;
1852 kfree(conf->mirrors);
1853 conf->mirrors = newmirrors;
1854 kfree(conf->poolinfo);
1855 conf->poolinfo = newpoolinfo;
1857 mddev->degraded += (raid_disks - conf->raid_disks);
1858 conf->raid_disks = mddev->raid_disks = raid_disks;
1860 conf->last_used = 0; /* just make sure it is in-range */
1861 lower_barrier(conf);
1863 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1864 md_wakeup_thread(mddev->thread);
1866 mempool_destroy(oldpool);
1870 static void raid1_quiesce(mddev_t *mddev, int state)
1872 conf_t *conf = mddev_to_conf(mddev);
1876 raise_barrier(conf);
1879 lower_barrier(conf);
1885 static mdk_personality_t raid1_personality =
1888 .owner = THIS_MODULE,
1889 .make_request = make_request,
1893 .error_handler = error,
1894 .hot_add_disk = raid1_add_disk,
1895 .hot_remove_disk= raid1_remove_disk,
1896 .spare_active = raid1_spare_active,
1897 .sync_request = sync_request,
1898 .resize = raid1_resize,
1899 .reshape = raid1_reshape,
1900 .quiesce = raid1_quiesce,
1903 static int __init raid_init(void)
1905 return register_md_personality(RAID1, &raid1_personality);
1908 static void raid_exit(void)
1910 unregister_md_personality(RAID1);
1913 module_init(raid_init);
1914 module_exit(raid_exit);
1915 MODULE_LICENSE("GPL");
1916 MODULE_ALIAS("md-personality-3"); /* RAID1 */