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
110 * If this is a user-requested check/repair, allocate
111 * RESYNC_PAGES for each bio.
113 if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery))
118 bio = r1_bio->bios[j];
119 for (i = 0; i < RESYNC_PAGES; i++) {
120 page = alloc_page(gfp_flags);
124 bio->bi_io_vec[i].bv_page = page;
127 /* If not user-requests, copy the page pointers to all bios */
128 if (!test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) {
129 for (i=0; i<RESYNC_PAGES ; i++)
130 for (j=1; j<pi->raid_disks; j++)
131 r1_bio->bios[j]->bi_io_vec[i].bv_page =
132 r1_bio->bios[0]->bi_io_vec[i].bv_page;
135 r1_bio->master_bio = NULL;
140 for (i=0; i < RESYNC_PAGES ; i++)
141 for (j=0 ; j < pi->raid_disks; j++)
142 __free_page(r1_bio->bios[j]->bi_io_vec[i].bv_page);
145 while ( ++j < pi->raid_disks )
146 bio_put(r1_bio->bios[j]);
147 r1bio_pool_free(r1_bio, data);
151 static void r1buf_pool_free(void *__r1_bio, void *data)
153 struct pool_info *pi = data;
155 r1bio_t *r1bio = __r1_bio;
157 for (i = 0; i < RESYNC_PAGES; i++)
158 for (j = pi->raid_disks; j-- ;) {
160 r1bio->bios[j]->bi_io_vec[i].bv_page !=
161 r1bio->bios[0]->bi_io_vec[i].bv_page)
162 __free_page(r1bio->bios[j]->bi_io_vec[i].bv_page);
164 for (i=0 ; i < pi->raid_disks; i++)
165 bio_put(r1bio->bios[i]);
167 r1bio_pool_free(r1bio, data);
170 static void put_all_bios(conf_t *conf, r1bio_t *r1_bio)
174 for (i = 0; i < conf->raid_disks; i++) {
175 struct bio **bio = r1_bio->bios + i;
176 if (*bio && *bio != IO_BLOCKED)
182 static inline void free_r1bio(r1bio_t *r1_bio)
184 conf_t *conf = mddev_to_conf(r1_bio->mddev);
187 * Wake up any possible resync thread that waits for the device
192 put_all_bios(conf, r1_bio);
193 mempool_free(r1_bio, conf->r1bio_pool);
196 static inline void put_buf(r1bio_t *r1_bio)
198 conf_t *conf = mddev_to_conf(r1_bio->mddev);
201 for (i=0; i<conf->raid_disks; i++) {
202 struct bio *bio = r1_bio->bios[i];
204 rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev);
207 mempool_free(r1_bio, conf->r1buf_pool);
212 static void reschedule_retry(r1bio_t *r1_bio)
215 mddev_t *mddev = r1_bio->mddev;
216 conf_t *conf = mddev_to_conf(mddev);
218 spin_lock_irqsave(&conf->device_lock, flags);
219 list_add(&r1_bio->retry_list, &conf->retry_list);
221 spin_unlock_irqrestore(&conf->device_lock, flags);
223 wake_up(&conf->wait_barrier);
224 md_wakeup_thread(mddev->thread);
228 * raid_end_bio_io() is called when we have finished servicing a mirrored
229 * operation and are ready to return a success/failure code to the buffer
232 static void raid_end_bio_io(r1bio_t *r1_bio)
234 struct bio *bio = r1_bio->master_bio;
236 /* if nobody has done the final endio yet, do it now */
237 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
238 PRINTK(KERN_DEBUG "raid1: sync end %s on sectors %llu-%llu\n",
239 (bio_data_dir(bio) == WRITE) ? "write" : "read",
240 (unsigned long long) bio->bi_sector,
241 (unsigned long long) bio->bi_sector +
242 (bio->bi_size >> 9) - 1);
244 bio_endio(bio, bio->bi_size,
245 test_bit(R1BIO_Uptodate, &r1_bio->state) ? 0 : -EIO);
251 * Update disk head position estimator based on IRQ completion info.
253 static inline void update_head_pos(int disk, r1bio_t *r1_bio)
255 conf_t *conf = mddev_to_conf(r1_bio->mddev);
257 conf->mirrors[disk].head_position =
258 r1_bio->sector + (r1_bio->sectors);
261 static int raid1_end_read_request(struct bio *bio, unsigned int bytes_done, int error)
263 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
264 r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
266 conf_t *conf = mddev_to_conf(r1_bio->mddev);
271 mirror = r1_bio->read_disk;
273 * this branch is our 'one mirror IO has finished' event handler:
275 update_head_pos(mirror, r1_bio);
277 if (uptodate || conf->working_disks <= 1) {
279 * Set R1BIO_Uptodate in our master bio, so that
280 * we will return a good error code for to the higher
281 * levels even if IO on some other mirrored buffer fails.
283 * The 'master' represents the composite IO operation to
284 * user-side. So if something waits for IO, then it will
285 * wait for the 'master' bio.
288 set_bit(R1BIO_Uptodate, &r1_bio->state);
290 raid_end_bio_io(r1_bio);
295 char b[BDEVNAME_SIZE];
296 if (printk_ratelimit())
297 printk(KERN_ERR "raid1: %s: rescheduling sector %llu\n",
298 bdevname(conf->mirrors[mirror].rdev->bdev,b), (unsigned long long)r1_bio->sector);
299 reschedule_retry(r1_bio);
302 rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev);
306 static int raid1_end_write_request(struct bio *bio, unsigned int bytes_done, int error)
308 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
309 r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
310 int mirror, behind = test_bit(R1BIO_BehindIO, &r1_bio->state);
311 conf_t *conf = mddev_to_conf(r1_bio->mddev);
316 for (mirror = 0; mirror < conf->raid_disks; mirror++)
317 if (r1_bio->bios[mirror] == bio)
320 if (error == -ENOTSUPP && test_bit(R1BIO_Barrier, &r1_bio->state)) {
321 set_bit(BarriersNotsupp, &conf->mirrors[mirror].rdev->flags);
322 set_bit(R1BIO_BarrierRetry, &r1_bio->state);
323 r1_bio->mddev->barriers_work = 0;
326 * this branch is our 'one mirror IO has finished' event handler:
328 r1_bio->bios[mirror] = NULL;
330 md_error(r1_bio->mddev, conf->mirrors[mirror].rdev);
331 /* an I/O failed, we can't clear the bitmap */
332 set_bit(R1BIO_Degraded, &r1_bio->state);
335 * Set R1BIO_Uptodate in our master bio, so that
336 * we will return a good error code for to the higher
337 * levels even if IO on some other mirrored buffer fails.
339 * The 'master' represents the composite IO operation to
340 * user-side. So if something waits for IO, then it will
341 * wait for the 'master' bio.
343 set_bit(R1BIO_Uptodate, &r1_bio->state);
345 update_head_pos(mirror, r1_bio);
348 if (test_bit(WriteMostly, &conf->mirrors[mirror].rdev->flags))
349 atomic_dec(&r1_bio->behind_remaining);
351 /* In behind mode, we ACK the master bio once the I/O has safely
352 * reached all non-writemostly disks. Setting the Returned bit
353 * ensures that this gets done only once -- we don't ever want to
354 * return -EIO here, instead we'll wait */
356 if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) &&
357 test_bit(R1BIO_Uptodate, &r1_bio->state)) {
358 /* Maybe we can return now */
359 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
360 struct bio *mbio = r1_bio->master_bio;
361 PRINTK(KERN_DEBUG "raid1: behind end write sectors %llu-%llu\n",
362 (unsigned long long) mbio->bi_sector,
363 (unsigned long long) mbio->bi_sector +
364 (mbio->bi_size >> 9) - 1);
365 bio_endio(mbio, mbio->bi_size, 0);
372 * Let's see if all mirrored write operations have finished
375 if (atomic_dec_and_test(&r1_bio->remaining)) {
376 if (test_bit(R1BIO_BarrierRetry, &r1_bio->state)) {
377 reschedule_retry(r1_bio);
378 /* Don't dec_pending yet, we want to hold
379 * the reference over the retry
383 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
384 /* free extra copy of the data pages */
385 int i = bio->bi_vcnt;
387 __free_page(bio->bi_io_vec[i].bv_page);
389 /* clear the bitmap if all writes complete successfully */
390 bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector,
392 !test_bit(R1BIO_Degraded, &r1_bio->state),
394 md_write_end(r1_bio->mddev);
395 raid_end_bio_io(r1_bio);
398 if (r1_bio->bios[mirror]==NULL)
401 rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev);
407 * This routine returns the disk from which the requested read should
408 * be done. There is a per-array 'next expected sequential IO' sector
409 * number - if this matches on the next IO then we use the last disk.
410 * There is also a per-disk 'last know head position' sector that is
411 * maintained from IRQ contexts, both the normal and the resync IO
412 * completion handlers update this position correctly. If there is no
413 * perfect sequential match then we pick the disk whose head is closest.
415 * If there are 2 mirrors in the same 2 devices, performance degrades
416 * because position is mirror, not device based.
418 * The rdev for the device selected will have nr_pending incremented.
420 static int read_balance(conf_t *conf, r1bio_t *r1_bio)
422 const unsigned long this_sector = r1_bio->sector;
423 int new_disk = conf->last_used, disk = new_disk;
425 const int sectors = r1_bio->sectors;
426 sector_t new_distance, current_distance;
431 * Check if we can balance. We can balance on the whole
432 * device if no resync is going on, or below the resync window.
433 * We take the first readable disk when above the resync window.
436 if (conf->mddev->recovery_cp < MaxSector &&
437 (this_sector + sectors >= conf->next_resync)) {
438 /* Choose the first operation device, for consistancy */
441 for (rdev = rcu_dereference(conf->mirrors[new_disk].rdev);
442 r1_bio->bios[new_disk] == IO_BLOCKED ||
443 !rdev || !test_bit(In_sync, &rdev->flags)
444 || test_bit(WriteMostly, &rdev->flags);
445 rdev = rcu_dereference(conf->mirrors[++new_disk].rdev)) {
447 if (rdev && test_bit(In_sync, &rdev->flags) &&
448 r1_bio->bios[new_disk] != IO_BLOCKED)
449 wonly_disk = new_disk;
451 if (new_disk == conf->raid_disks - 1) {
452 new_disk = wonly_disk;
460 /* make sure the disk is operational */
461 for (rdev = rcu_dereference(conf->mirrors[new_disk].rdev);
462 r1_bio->bios[new_disk] == IO_BLOCKED ||
463 !rdev || !test_bit(In_sync, &rdev->flags) ||
464 test_bit(WriteMostly, &rdev->flags);
465 rdev = rcu_dereference(conf->mirrors[new_disk].rdev)) {
467 if (rdev && test_bit(In_sync, &rdev->flags) &&
468 r1_bio->bios[new_disk] != IO_BLOCKED)
469 wonly_disk = new_disk;
472 new_disk = conf->raid_disks;
474 if (new_disk == disk) {
475 new_disk = wonly_disk;
484 /* now disk == new_disk == starting point for search */
487 * Don't change to another disk for sequential reads:
489 if (conf->next_seq_sect == this_sector)
491 if (this_sector == conf->mirrors[new_disk].head_position)
494 current_distance = abs(this_sector - conf->mirrors[disk].head_position);
496 /* Find the disk whose head is closest */
500 disk = conf->raid_disks;
503 rdev = rcu_dereference(conf->mirrors[disk].rdev);
505 if (!rdev || r1_bio->bios[disk] == IO_BLOCKED ||
506 !test_bit(In_sync, &rdev->flags) ||
507 test_bit(WriteMostly, &rdev->flags))
510 if (!atomic_read(&rdev->nr_pending)) {
514 new_distance = abs(this_sector - conf->mirrors[disk].head_position);
515 if (new_distance < current_distance) {
516 current_distance = new_distance;
519 } while (disk != conf->last_used);
525 rdev = rcu_dereference(conf->mirrors[new_disk].rdev);
528 atomic_inc(&rdev->nr_pending);
529 if (!test_bit(In_sync, &rdev->flags)) {
530 /* cannot risk returning a device that failed
531 * before we inc'ed nr_pending
533 atomic_dec(&rdev->nr_pending);
536 conf->next_seq_sect = this_sector + sectors;
537 conf->last_used = new_disk;
544 static void unplug_slaves(mddev_t *mddev)
546 conf_t *conf = mddev_to_conf(mddev);
550 for (i=0; i<mddev->raid_disks; i++) {
551 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
552 if (rdev && !test_bit(Faulty, &rdev->flags) && atomic_read(&rdev->nr_pending)) {
553 request_queue_t *r_queue = bdev_get_queue(rdev->bdev);
555 atomic_inc(&rdev->nr_pending);
558 if (r_queue->unplug_fn)
559 r_queue->unplug_fn(r_queue);
561 rdev_dec_pending(rdev, mddev);
568 static void raid1_unplug(request_queue_t *q)
570 mddev_t *mddev = q->queuedata;
572 unplug_slaves(mddev);
573 md_wakeup_thread(mddev->thread);
576 static int raid1_issue_flush(request_queue_t *q, struct gendisk *disk,
577 sector_t *error_sector)
579 mddev_t *mddev = q->queuedata;
580 conf_t *conf = mddev_to_conf(mddev);
584 for (i=0; i<mddev->raid_disks && ret == 0; i++) {
585 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
586 if (rdev && !test_bit(Faulty, &rdev->flags)) {
587 struct block_device *bdev = rdev->bdev;
588 request_queue_t *r_queue = bdev_get_queue(bdev);
590 if (!r_queue->issue_flush_fn)
593 atomic_inc(&rdev->nr_pending);
595 ret = r_queue->issue_flush_fn(r_queue, bdev->bd_disk,
597 rdev_dec_pending(rdev, mddev);
607 * Sometimes we need to suspend IO while we do something else,
608 * either some resync/recovery, or reconfigure the array.
609 * To do this we raise a 'barrier'.
610 * The 'barrier' is a counter that can be raised multiple times
611 * to count how many activities are happening which preclude
613 * We can only raise the barrier if there is no pending IO.
614 * i.e. if nr_pending == 0.
615 * We choose only to raise the barrier if no-one is waiting for the
616 * barrier to go down. This means that as soon as an IO request
617 * is ready, no other operations which require a barrier will start
618 * until the IO request has had a chance.
620 * So: regular IO calls 'wait_barrier'. When that returns there
621 * is no backgroup IO happening, It must arrange to call
622 * allow_barrier when it has finished its IO.
623 * backgroup IO calls must call raise_barrier. Once that returns
624 * there is no normal IO happeing. It must arrange to call
625 * lower_barrier when the particular background IO completes.
627 #define RESYNC_DEPTH 32
629 static void raise_barrier(conf_t *conf)
631 spin_lock_irq(&conf->resync_lock);
633 /* Wait until no block IO is waiting */
634 wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting,
636 raid1_unplug(conf->mddev->queue));
638 /* block any new IO from starting */
641 /* No wait for all pending IO to complete */
642 wait_event_lock_irq(conf->wait_barrier,
643 !conf->nr_pending && conf->barrier < RESYNC_DEPTH,
645 raid1_unplug(conf->mddev->queue));
647 spin_unlock_irq(&conf->resync_lock);
650 static void lower_barrier(conf_t *conf)
653 spin_lock_irqsave(&conf->resync_lock, flags);
655 spin_unlock_irqrestore(&conf->resync_lock, flags);
656 wake_up(&conf->wait_barrier);
659 static void wait_barrier(conf_t *conf)
661 spin_lock_irq(&conf->resync_lock);
664 wait_event_lock_irq(conf->wait_barrier, !conf->barrier,
666 raid1_unplug(conf->mddev->queue));
670 spin_unlock_irq(&conf->resync_lock);
673 static void allow_barrier(conf_t *conf)
676 spin_lock_irqsave(&conf->resync_lock, flags);
678 spin_unlock_irqrestore(&conf->resync_lock, flags);
679 wake_up(&conf->wait_barrier);
682 static void freeze_array(conf_t *conf)
684 /* stop syncio and normal IO and wait for everything to
686 * We increment barrier and nr_waiting, and then
687 * wait until barrier+nr_pending match nr_queued+2
689 spin_lock_irq(&conf->resync_lock);
692 wait_event_lock_irq(conf->wait_barrier,
693 conf->barrier+conf->nr_pending == conf->nr_queued+2,
695 raid1_unplug(conf->mddev->queue));
696 spin_unlock_irq(&conf->resync_lock);
698 static void unfreeze_array(conf_t *conf)
700 /* reverse the effect of the freeze */
701 spin_lock_irq(&conf->resync_lock);
704 wake_up(&conf->wait_barrier);
705 spin_unlock_irq(&conf->resync_lock);
709 /* duplicate the data pages for behind I/O */
710 static struct page **alloc_behind_pages(struct bio *bio)
713 struct bio_vec *bvec;
714 struct page **pages = kmalloc(bio->bi_vcnt * sizeof(struct page *),
716 if (unlikely(!pages))
719 memset(pages, 0, bio->bi_vcnt * sizeof(struct page *));
721 bio_for_each_segment(bvec, bio, i) {
722 pages[i] = alloc_page(GFP_NOIO);
723 if (unlikely(!pages[i]))
725 memcpy(kmap(pages[i]) + bvec->bv_offset,
726 kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len);
728 kunmap(bvec->bv_page);
735 for (i = 0; i < bio->bi_vcnt && pages[i]; i++)
736 __free_page(pages[i]);
738 PRINTK("%dB behind alloc failed, doing sync I/O\n", bio->bi_size);
742 static int make_request(request_queue_t *q, struct bio * bio)
744 mddev_t *mddev = q->queuedata;
745 conf_t *conf = mddev_to_conf(mddev);
746 mirror_info_t *mirror;
748 struct bio *read_bio;
749 int i, targets = 0, disks;
751 struct bitmap *bitmap = mddev->bitmap;
754 struct page **behind_pages = NULL;
755 const int rw = bio_data_dir(bio);
758 if (unlikely(!mddev->barriers_work && bio_barrier(bio))) {
759 bio_endio(bio, bio->bi_size, -EOPNOTSUPP);
764 * Register the new request and wait if the reconstruction
765 * thread has put up a bar for new requests.
766 * Continue immediately if no resync is active currently.
768 md_write_start(mddev, bio); /* wait on superblock update early */
772 disk_stat_inc(mddev->gendisk, ios[rw]);
773 disk_stat_add(mddev->gendisk, sectors[rw], bio_sectors(bio));
776 * make_request() can abort the operation when READA is being
777 * used and no empty request is available.
780 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
782 r1_bio->master_bio = bio;
783 r1_bio->sectors = bio->bi_size >> 9;
785 r1_bio->mddev = mddev;
786 r1_bio->sector = bio->bi_sector;
790 * read balancing logic:
792 int rdisk = read_balance(conf, r1_bio);
795 /* couldn't find anywhere to read from */
796 raid_end_bio_io(r1_bio);
799 mirror = conf->mirrors + rdisk;
801 r1_bio->read_disk = rdisk;
803 read_bio = bio_clone(bio, GFP_NOIO);
805 r1_bio->bios[rdisk] = read_bio;
807 read_bio->bi_sector = r1_bio->sector + mirror->rdev->data_offset;
808 read_bio->bi_bdev = mirror->rdev->bdev;
809 read_bio->bi_end_io = raid1_end_read_request;
810 read_bio->bi_rw = READ;
811 read_bio->bi_private = r1_bio;
813 generic_make_request(read_bio);
820 /* first select target devices under spinlock and
821 * inc refcount on their rdev. Record them by setting
824 disks = conf->raid_disks;
826 { static int first=1;
827 if (first) printk("First Write sector %llu disks %d\n",
828 (unsigned long long)r1_bio->sector, disks);
833 for (i = 0; i < disks; i++) {
834 if ((rdev=rcu_dereference(conf->mirrors[i].rdev)) != NULL &&
835 !test_bit(Faulty, &rdev->flags)) {
836 atomic_inc(&rdev->nr_pending);
837 if (test_bit(Faulty, &rdev->flags)) {
838 atomic_dec(&rdev->nr_pending);
839 r1_bio->bios[i] = NULL;
841 r1_bio->bios[i] = bio;
844 r1_bio->bios[i] = NULL;
848 BUG_ON(targets == 0); /* we never fail the last device */
850 if (targets < conf->raid_disks) {
851 /* array is degraded, we will not clear the bitmap
852 * on I/O completion (see raid1_end_write_request) */
853 set_bit(R1BIO_Degraded, &r1_bio->state);
856 /* do behind I/O ? */
858 atomic_read(&bitmap->behind_writes) < bitmap->max_write_behind &&
859 (behind_pages = alloc_behind_pages(bio)) != NULL)
860 set_bit(R1BIO_BehindIO, &r1_bio->state);
862 atomic_set(&r1_bio->remaining, 0);
863 atomic_set(&r1_bio->behind_remaining, 0);
865 do_barriers = bio->bi_rw & BIO_RW_BARRIER;
867 set_bit(R1BIO_Barrier, &r1_bio->state);
870 for (i = 0; i < disks; i++) {
872 if (!r1_bio->bios[i])
875 mbio = bio_clone(bio, GFP_NOIO);
876 r1_bio->bios[i] = mbio;
878 mbio->bi_sector = r1_bio->sector + conf->mirrors[i].rdev->data_offset;
879 mbio->bi_bdev = conf->mirrors[i].rdev->bdev;
880 mbio->bi_end_io = raid1_end_write_request;
881 mbio->bi_rw = WRITE | do_barriers;
882 mbio->bi_private = r1_bio;
885 struct bio_vec *bvec;
888 /* Yes, I really want the '__' version so that
889 * we clear any unused pointer in the io_vec, rather
890 * than leave them unchanged. This is important
891 * because when we come to free the pages, we won't
892 * know the originial bi_idx, so we just free
895 __bio_for_each_segment(bvec, mbio, j, 0)
896 bvec->bv_page = behind_pages[j];
897 if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags))
898 atomic_inc(&r1_bio->behind_remaining);
901 atomic_inc(&r1_bio->remaining);
903 bio_list_add(&bl, mbio);
905 kfree(behind_pages); /* the behind pages are attached to the bios now */
907 bitmap_startwrite(bitmap, bio->bi_sector, r1_bio->sectors,
908 test_bit(R1BIO_BehindIO, &r1_bio->state));
909 spin_lock_irqsave(&conf->device_lock, flags);
910 bio_list_merge(&conf->pending_bio_list, &bl);
913 blk_plug_device(mddev->queue);
914 spin_unlock_irqrestore(&conf->device_lock, flags);
917 while ((bio = bio_list_pop(&bl)) != NULL)
918 generic_make_request(bio);
924 static void status(struct seq_file *seq, mddev_t *mddev)
926 conf_t *conf = mddev_to_conf(mddev);
929 seq_printf(seq, " [%d/%d] [", conf->raid_disks,
930 conf->working_disks);
931 for (i = 0; i < conf->raid_disks; i++)
932 seq_printf(seq, "%s",
933 conf->mirrors[i].rdev &&
934 test_bit(In_sync, &conf->mirrors[i].rdev->flags) ? "U" : "_");
935 seq_printf(seq, "]");
939 static void error(mddev_t *mddev, mdk_rdev_t *rdev)
941 char b[BDEVNAME_SIZE];
942 conf_t *conf = mddev_to_conf(mddev);
945 * If it is not operational, then we have already marked it as dead
946 * else if it is the last working disks, ignore the error, let the
947 * next level up know.
948 * else mark the drive as failed
950 if (test_bit(In_sync, &rdev->flags)
951 && conf->working_disks == 1)
953 * Don't fail the drive, act as though we were just a
954 * normal single drive
957 if (test_bit(In_sync, &rdev->flags)) {
959 conf->working_disks--;
961 * if recovery is running, make sure it aborts.
963 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
965 clear_bit(In_sync, &rdev->flags);
966 set_bit(Faulty, &rdev->flags);
968 printk(KERN_ALERT "raid1: Disk failure on %s, disabling device. \n"
969 " Operation continuing on %d devices\n",
970 bdevname(rdev->bdev,b), conf->working_disks);
973 static void print_conf(conf_t *conf)
978 printk("RAID1 conf printout:\n");
983 printk(" --- wd:%d rd:%d\n", conf->working_disks,
986 for (i = 0; i < conf->raid_disks; i++) {
987 char b[BDEVNAME_SIZE];
988 tmp = conf->mirrors + i;
990 printk(" disk %d, wo:%d, o:%d, dev:%s\n",
991 i, !test_bit(In_sync, &tmp->rdev->flags), !test_bit(Faulty, &tmp->rdev->flags),
992 bdevname(tmp->rdev->bdev,b));
996 static void close_sync(conf_t *conf)
1001 mempool_destroy(conf->r1buf_pool);
1002 conf->r1buf_pool = NULL;
1005 static int raid1_spare_active(mddev_t *mddev)
1008 conf_t *conf = mddev->private;
1012 * Find all failed disks within the RAID1 configuration
1013 * and mark them readable
1015 for (i = 0; i < conf->raid_disks; i++) {
1016 tmp = conf->mirrors + i;
1018 && !test_bit(Faulty, &tmp->rdev->flags)
1019 && !test_bit(In_sync, &tmp->rdev->flags)) {
1020 conf->working_disks++;
1022 set_bit(In_sync, &tmp->rdev->flags);
1031 static int raid1_add_disk(mddev_t *mddev, mdk_rdev_t *rdev)
1033 conf_t *conf = mddev->private;
1038 for (mirror=0; mirror < mddev->raid_disks; mirror++)
1039 if ( !(p=conf->mirrors+mirror)->rdev) {
1041 blk_queue_stack_limits(mddev->queue,
1042 rdev->bdev->bd_disk->queue);
1043 /* as we don't honour merge_bvec_fn, we must never risk
1044 * violating it, so limit ->max_sector to one PAGE, as
1045 * a one page request is never in violation.
1047 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
1048 mddev->queue->max_sectors > (PAGE_SIZE>>9))
1049 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
1051 p->head_position = 0;
1052 rdev->raid_disk = mirror;
1054 /* As all devices are equivalent, we don't need a full recovery
1055 * if this was recently any drive of the array
1057 if (rdev->saved_raid_disk < 0)
1059 rcu_assign_pointer(p->rdev, rdev);
1067 static int raid1_remove_disk(mddev_t *mddev, int number)
1069 conf_t *conf = mddev->private;
1072 mirror_info_t *p = conf->mirrors+ number;
1077 if (test_bit(In_sync, &rdev->flags) ||
1078 atomic_read(&rdev->nr_pending)) {
1084 if (atomic_read(&rdev->nr_pending)) {
1085 /* lost the race, try later */
1097 static int end_sync_read(struct bio *bio, unsigned int bytes_done, int error)
1099 r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
1105 for (i=r1_bio->mddev->raid_disks; i--; )
1106 if (r1_bio->bios[i] == bio)
1109 update_head_pos(i, r1_bio);
1111 * we have read a block, now it needs to be re-written,
1112 * or re-read if the read failed.
1113 * We don't do much here, just schedule handling by raid1d
1115 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
1116 set_bit(R1BIO_Uptodate, &r1_bio->state);
1118 if (atomic_dec_and_test(&r1_bio->remaining))
1119 reschedule_retry(r1_bio);
1123 static int end_sync_write(struct bio *bio, unsigned int bytes_done, int error)
1125 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1126 r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
1127 mddev_t *mddev = r1_bio->mddev;
1128 conf_t *conf = mddev_to_conf(mddev);
1135 for (i = 0; i < conf->raid_disks; i++)
1136 if (r1_bio->bios[i] == bio) {
1141 md_error(mddev, conf->mirrors[mirror].rdev);
1143 update_head_pos(mirror, r1_bio);
1145 if (atomic_dec_and_test(&r1_bio->remaining)) {
1146 md_done_sync(mddev, r1_bio->sectors, uptodate);
1152 static void sync_request_write(mddev_t *mddev, r1bio_t *r1_bio)
1154 conf_t *conf = mddev_to_conf(mddev);
1156 int disks = conf->raid_disks;
1157 struct bio *bio, *wbio;
1159 bio = r1_bio->bios[r1_bio->read_disk];
1162 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
1163 /* We have read all readable devices. If we haven't
1164 * got the block, then there is no hope left.
1165 * If we have, then we want to do a comparison
1166 * and skip the write if everything is the same.
1167 * If any blocks failed to read, then we need to
1168 * attempt an over-write
1171 if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) {
1172 for (i=0; i<mddev->raid_disks; i++)
1173 if (r1_bio->bios[i]->bi_end_io == end_sync_read)
1174 md_error(mddev, conf->mirrors[i].rdev);
1176 md_done_sync(mddev, r1_bio->sectors, 1);
1180 for (primary=0; primary<mddev->raid_disks; primary++)
1181 if (r1_bio->bios[primary]->bi_end_io == end_sync_read &&
1182 test_bit(BIO_UPTODATE, &r1_bio->bios[primary]->bi_flags)) {
1183 r1_bio->bios[primary]->bi_end_io = NULL;
1186 r1_bio->read_disk = primary;
1187 for (i=0; i<mddev->raid_disks; i++)
1188 if (r1_bio->bios[i]->bi_end_io == end_sync_read &&
1189 test_bit(BIO_UPTODATE, &r1_bio->bios[i]->bi_flags)) {
1191 int vcnt = r1_bio->sectors >> (PAGE_SHIFT- 9);
1192 struct bio *pbio = r1_bio->bios[primary];
1193 struct bio *sbio = r1_bio->bios[i];
1194 for (j = vcnt; j-- ; )
1195 if (memcmp(page_address(pbio->bi_io_vec[j].bv_page),
1196 page_address(sbio->bi_io_vec[j].bv_page),
1200 mddev->resync_mismatches += r1_bio->sectors;
1201 if (j < 0 || test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
1202 sbio->bi_end_io = NULL;
1204 /* fixup the bio for reuse */
1205 sbio->bi_vcnt = vcnt;
1206 sbio->bi_size = r1_bio->sectors << 9;
1208 sbio->bi_phys_segments = 0;
1209 sbio->bi_hw_segments = 0;
1210 sbio->bi_hw_front_size = 0;
1211 sbio->bi_hw_back_size = 0;
1212 sbio->bi_flags &= ~(BIO_POOL_MASK - 1);
1213 sbio->bi_flags |= 1 << BIO_UPTODATE;
1214 sbio->bi_next = NULL;
1215 sbio->bi_sector = r1_bio->sector +
1216 conf->mirrors[i].rdev->data_offset;
1217 sbio->bi_bdev = conf->mirrors[i].rdev->bdev;
1221 if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) {
1222 /* ouch - failed to read all of that.
1223 * Try some synchronous reads of other devices to get
1224 * good data, much like with normal read errors. Only
1225 * read into the pages we already have so they we don't
1226 * need to re-issue the read request.
1227 * We don't need to freeze the array, because being in an
1228 * active sync request, there is no normal IO, and
1229 * no overlapping syncs.
1231 sector_t sect = r1_bio->sector;
1232 int sectors = r1_bio->sectors;
1237 int d = r1_bio->read_disk;
1241 if (s > (PAGE_SIZE>>9))
1244 if (r1_bio->bios[d]->bi_end_io == end_sync_read) {
1245 rdev = conf->mirrors[d].rdev;
1246 if (sync_page_io(rdev->bdev,
1247 sect + rdev->data_offset,
1249 bio->bi_io_vec[idx].bv_page,
1256 if (d == conf->raid_disks)
1258 } while (!success && d != r1_bio->read_disk);
1261 /* write it back and re-read */
1262 set_bit(R1BIO_Uptodate, &r1_bio->state);
1263 while (d != r1_bio->read_disk) {
1265 d = conf->raid_disks;
1267 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1269 rdev = conf->mirrors[d].rdev;
1270 if (sync_page_io(rdev->bdev,
1271 sect + rdev->data_offset,
1273 bio->bi_io_vec[idx].bv_page,
1275 sync_page_io(rdev->bdev,
1276 sect + rdev->data_offset,
1278 bio->bi_io_vec[idx].bv_page,
1280 md_error(mddev, rdev);
1284 char b[BDEVNAME_SIZE];
1285 /* Cannot read from anywhere, array is toast */
1286 md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev);
1287 printk(KERN_ALERT "raid1: %s: unrecoverable I/O read error"
1288 " for block %llu\n",
1289 bdevname(bio->bi_bdev,b),
1290 (unsigned long long)r1_bio->sector);
1291 md_done_sync(mddev, r1_bio->sectors, 0);
1304 atomic_set(&r1_bio->remaining, 1);
1305 for (i = 0; i < disks ; i++) {
1306 wbio = r1_bio->bios[i];
1307 if (wbio->bi_end_io == NULL ||
1308 (wbio->bi_end_io == end_sync_read &&
1309 (i == r1_bio->read_disk ||
1310 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery))))
1313 wbio->bi_rw = WRITE;
1314 wbio->bi_end_io = end_sync_write;
1315 atomic_inc(&r1_bio->remaining);
1316 md_sync_acct(conf->mirrors[i].rdev->bdev, wbio->bi_size >> 9);
1318 generic_make_request(wbio);
1321 if (atomic_dec_and_test(&r1_bio->remaining)) {
1322 /* if we're here, all write(s) have completed, so clean up */
1323 md_done_sync(mddev, r1_bio->sectors, 1);
1329 * This is a kernel thread which:
1331 * 1. Retries failed read operations on working mirrors.
1332 * 2. Updates the raid superblock when problems encounter.
1333 * 3. Performs writes following reads for array syncronising.
1336 static void raid1d(mddev_t *mddev)
1340 unsigned long flags;
1341 conf_t *conf = mddev_to_conf(mddev);
1342 struct list_head *head = &conf->retry_list;
1346 md_check_recovery(mddev);
1349 char b[BDEVNAME_SIZE];
1350 spin_lock_irqsave(&conf->device_lock, flags);
1352 if (conf->pending_bio_list.head) {
1353 bio = bio_list_get(&conf->pending_bio_list);
1354 blk_remove_plug(mddev->queue);
1355 spin_unlock_irqrestore(&conf->device_lock, flags);
1356 /* flush any pending bitmap writes to disk before proceeding w/ I/O */
1357 if (bitmap_unplug(mddev->bitmap) != 0)
1358 printk("%s: bitmap file write failed!\n", mdname(mddev));
1360 while (bio) { /* submit pending writes */
1361 struct bio *next = bio->bi_next;
1362 bio->bi_next = NULL;
1363 generic_make_request(bio);
1371 if (list_empty(head))
1373 r1_bio = list_entry(head->prev, r1bio_t, retry_list);
1374 list_del(head->prev);
1376 spin_unlock_irqrestore(&conf->device_lock, flags);
1378 mddev = r1_bio->mddev;
1379 conf = mddev_to_conf(mddev);
1380 if (test_bit(R1BIO_IsSync, &r1_bio->state)) {
1381 sync_request_write(mddev, r1_bio);
1383 } else if (test_bit(R1BIO_BarrierRetry, &r1_bio->state)) {
1384 /* some requests in the r1bio were BIO_RW_BARRIER
1385 * requests which failed with -ENOTSUPP. Hohumm..
1386 * Better resubmit without the barrier.
1387 * We know which devices to resubmit for, because
1388 * all others have had their bios[] entry cleared.
1391 clear_bit(R1BIO_BarrierRetry, &r1_bio->state);
1392 clear_bit(R1BIO_Barrier, &r1_bio->state);
1393 for (i=0; i < conf->raid_disks; i++)
1394 if (r1_bio->bios[i]) {
1395 struct bio_vec *bvec;
1398 bio = bio_clone(r1_bio->master_bio, GFP_NOIO);
1399 /* copy pages from the failed bio, as
1400 * this might be a write-behind device */
1401 __bio_for_each_segment(bvec, bio, j, 0)
1402 bvec->bv_page = bio_iovec_idx(r1_bio->bios[i], j)->bv_page;
1403 bio_put(r1_bio->bios[i]);
1404 bio->bi_sector = r1_bio->sector +
1405 conf->mirrors[i].rdev->data_offset;
1406 bio->bi_bdev = conf->mirrors[i].rdev->bdev;
1407 bio->bi_end_io = raid1_end_write_request;
1409 bio->bi_private = r1_bio;
1410 r1_bio->bios[i] = bio;
1411 generic_make_request(bio);
1416 /* we got a read error. Maybe the drive is bad. Maybe just
1417 * the block and we can fix it.
1418 * We freeze all other IO, and try reading the block from
1419 * other devices. When we find one, we re-write
1420 * and check it that fixes the read error.
1421 * This is all done synchronously while the array is
1424 sector_t sect = r1_bio->sector;
1425 int sectors = r1_bio->sectors;
1427 if (mddev->ro == 0) while(sectors) {
1429 int d = r1_bio->read_disk;
1432 if (s > (PAGE_SIZE>>9))
1436 rdev = conf->mirrors[d].rdev;
1438 test_bit(In_sync, &rdev->flags) &&
1439 sync_page_io(rdev->bdev,
1440 sect + rdev->data_offset,
1442 conf->tmppage, READ))
1446 if (d == conf->raid_disks)
1449 } while (!success && d != r1_bio->read_disk);
1452 /* write it back and re-read */
1453 while (d != r1_bio->read_disk) {
1455 d = conf->raid_disks;
1457 rdev = conf->mirrors[d].rdev;
1459 test_bit(In_sync, &rdev->flags)) {
1460 if (sync_page_io(rdev->bdev,
1461 sect + rdev->data_offset,
1462 s<<9, conf->tmppage, WRITE) == 0 ||
1463 sync_page_io(rdev->bdev,
1464 sect + rdev->data_offset,
1465 s<<9, conf->tmppage, READ) == 0) {
1466 /* Well, this device is dead */
1467 md_error(mddev, rdev);
1472 /* Cannot read from anywhere -- bye bye array */
1473 md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev);
1480 unfreeze_array(conf);
1482 bio = r1_bio->bios[r1_bio->read_disk];
1483 if ((disk=read_balance(conf, r1_bio)) == -1) {
1484 printk(KERN_ALERT "raid1: %s: unrecoverable I/O"
1485 " read error for block %llu\n",
1486 bdevname(bio->bi_bdev,b),
1487 (unsigned long long)r1_bio->sector);
1488 raid_end_bio_io(r1_bio);
1490 r1_bio->bios[r1_bio->read_disk] =
1491 mddev->ro ? IO_BLOCKED : NULL;
1492 r1_bio->read_disk = disk;
1494 bio = bio_clone(r1_bio->master_bio, GFP_NOIO);
1495 r1_bio->bios[r1_bio->read_disk] = bio;
1496 rdev = conf->mirrors[disk].rdev;
1497 if (printk_ratelimit())
1498 printk(KERN_ERR "raid1: %s: redirecting sector %llu to"
1499 " another mirror\n",
1500 bdevname(rdev->bdev,b),
1501 (unsigned long long)r1_bio->sector);
1502 bio->bi_sector = r1_bio->sector + rdev->data_offset;
1503 bio->bi_bdev = rdev->bdev;
1504 bio->bi_end_io = raid1_end_read_request;
1506 bio->bi_private = r1_bio;
1508 generic_make_request(bio);
1512 spin_unlock_irqrestore(&conf->device_lock, flags);
1514 unplug_slaves(mddev);
1518 static int init_resync(conf_t *conf)
1522 buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;
1523 if (conf->r1buf_pool)
1525 conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free,
1527 if (!conf->r1buf_pool)
1529 conf->next_resync = 0;
1534 * perform a "sync" on one "block"
1536 * We need to make sure that no normal I/O request - particularly write
1537 * requests - conflict with active sync requests.
1539 * This is achieved by tracking pending requests and a 'barrier' concept
1540 * that can be installed to exclude normal IO requests.
1543 static sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster)
1545 conf_t *conf = mddev_to_conf(mddev);
1548 sector_t max_sector, nr_sectors;
1552 int write_targets = 0, read_targets = 0;
1554 int still_degraded = 0;
1556 if (!conf->r1buf_pool)
1559 printk("sync start - bitmap %p\n", mddev->bitmap);
1561 if (init_resync(conf))
1565 max_sector = mddev->size << 1;
1566 if (sector_nr >= max_sector) {
1567 /* If we aborted, we need to abort the
1568 * sync on the 'current' bitmap chunk (there will
1569 * only be one in raid1 resync.
1570 * We can find the current addess in mddev->curr_resync
1572 if (mddev->curr_resync < max_sector) /* aborted */
1573 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
1575 else /* completed sync */
1578 bitmap_close_sync(mddev->bitmap);
1583 /* before building a request, check if we can skip these blocks..
1584 * This call the bitmap_start_sync doesn't actually record anything
1586 if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
1587 !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
1588 /* We can skip this block, and probably several more */
1593 * If there is non-resync activity waiting for a turn,
1594 * and resync is going fast enough,
1595 * then let it though before starting on this new sync request.
1597 if (!go_faster && conf->nr_waiting)
1598 msleep_interruptible(1000);
1600 raise_barrier(conf);
1602 conf->next_resync = sector_nr;
1604 r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO);
1607 * If we get a correctably read error during resync or recovery,
1608 * we might want to read from a different device. So we
1609 * flag all drives that could conceivably be read from for READ,
1610 * and any others (which will be non-In_sync devices) for WRITE.
1611 * If a read fails, we try reading from something else for which READ
1615 r1_bio->mddev = mddev;
1616 r1_bio->sector = sector_nr;
1618 set_bit(R1BIO_IsSync, &r1_bio->state);
1620 for (i=0; i < conf->raid_disks; i++) {
1622 bio = r1_bio->bios[i];
1624 /* take from bio_init */
1625 bio->bi_next = NULL;
1626 bio->bi_flags |= 1 << BIO_UPTODATE;
1630 bio->bi_phys_segments = 0;
1631 bio->bi_hw_segments = 0;
1633 bio->bi_end_io = NULL;
1634 bio->bi_private = NULL;
1636 rdev = rcu_dereference(conf->mirrors[i].rdev);
1638 test_bit(Faulty, &rdev->flags)) {
1641 } else if (!test_bit(In_sync, &rdev->flags)) {
1643 bio->bi_end_io = end_sync_write;
1646 /* may need to read from here */
1648 bio->bi_end_io = end_sync_read;
1649 if (test_bit(WriteMostly, &rdev->flags)) {
1658 atomic_inc(&rdev->nr_pending);
1659 bio->bi_sector = sector_nr + rdev->data_offset;
1660 bio->bi_bdev = rdev->bdev;
1661 bio->bi_private = r1_bio;
1666 r1_bio->read_disk = disk;
1668 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0)
1669 /* extra read targets are also write targets */
1670 write_targets += read_targets-1;
1672 if (write_targets == 0 || read_targets == 0) {
1673 /* There is nowhere to write, so all non-sync
1674 * drives must be failed - so we are finished
1676 sector_t rv = max_sector - sector_nr;
1686 int len = PAGE_SIZE;
1687 if (sector_nr + (len>>9) > max_sector)
1688 len = (max_sector - sector_nr) << 9;
1691 if (sync_blocks == 0) {
1692 if (!bitmap_start_sync(mddev->bitmap, sector_nr,
1693 &sync_blocks, still_degraded) &&
1695 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1697 if (sync_blocks < (PAGE_SIZE>>9))
1699 if (len > (sync_blocks<<9))
1700 len = sync_blocks<<9;
1703 for (i=0 ; i < conf->raid_disks; i++) {
1704 bio = r1_bio->bios[i];
1705 if (bio->bi_end_io) {
1706 page = bio->bi_io_vec[bio->bi_vcnt].bv_page;
1707 if (bio_add_page(bio, page, len, 0) == 0) {
1709 bio->bi_io_vec[bio->bi_vcnt].bv_page = page;
1712 bio = r1_bio->bios[i];
1713 if (bio->bi_end_io==NULL)
1715 /* remove last page from this bio */
1717 bio->bi_size -= len;
1718 bio->bi_flags &= ~(1<< BIO_SEG_VALID);
1724 nr_sectors += len>>9;
1725 sector_nr += len>>9;
1726 sync_blocks -= (len>>9);
1727 } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES);
1729 r1_bio->sectors = nr_sectors;
1731 /* For a user-requested sync, we read all readable devices and do a
1734 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
1735 atomic_set(&r1_bio->remaining, read_targets);
1736 for (i=0; i<conf->raid_disks; i++) {
1737 bio = r1_bio->bios[i];
1738 if (bio->bi_end_io == end_sync_read) {
1739 md_sync_acct(conf->mirrors[i].rdev->bdev, nr_sectors);
1740 generic_make_request(bio);
1744 atomic_set(&r1_bio->remaining, 1);
1745 bio = r1_bio->bios[r1_bio->read_disk];
1746 md_sync_acct(conf->mirrors[r1_bio->read_disk].rdev->bdev,
1748 generic_make_request(bio);
1755 static int run(mddev_t *mddev)
1759 mirror_info_t *disk;
1761 struct list_head *tmp;
1763 if (mddev->level != 1) {
1764 printk("raid1: %s: raid level not set to mirroring (%d)\n",
1765 mdname(mddev), mddev->level);
1769 * copy the already verified devices into our private RAID1
1770 * bookkeeping area. [whatever we allocate in run(),
1771 * should be freed in stop()]
1773 conf = kmalloc(sizeof(conf_t), GFP_KERNEL);
1774 mddev->private = conf;
1778 memset(conf, 0, sizeof(*conf));
1779 conf->mirrors = kmalloc(sizeof(struct mirror_info)*mddev->raid_disks,
1784 memset(conf->mirrors, 0, sizeof(struct mirror_info)*mddev->raid_disks);
1786 conf->tmppage = alloc_page(GFP_KERNEL);
1790 conf->poolinfo = kmalloc(sizeof(*conf->poolinfo), GFP_KERNEL);
1791 if (!conf->poolinfo)
1793 conf->poolinfo->mddev = mddev;
1794 conf->poolinfo->raid_disks = mddev->raid_disks;
1795 conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
1798 if (!conf->r1bio_pool)
1801 ITERATE_RDEV(mddev, rdev, tmp) {
1802 disk_idx = rdev->raid_disk;
1803 if (disk_idx >= mddev->raid_disks
1806 disk = conf->mirrors + disk_idx;
1810 blk_queue_stack_limits(mddev->queue,
1811 rdev->bdev->bd_disk->queue);
1812 /* as we don't honour merge_bvec_fn, we must never risk
1813 * violating it, so limit ->max_sector to one PAGE, as
1814 * a one page request is never in violation.
1816 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
1817 mddev->queue->max_sectors > (PAGE_SIZE>>9))
1818 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
1820 disk->head_position = 0;
1821 if (!test_bit(Faulty, &rdev->flags) && test_bit(In_sync, &rdev->flags))
1822 conf->working_disks++;
1824 conf->raid_disks = mddev->raid_disks;
1825 conf->mddev = mddev;
1826 spin_lock_init(&conf->device_lock);
1827 INIT_LIST_HEAD(&conf->retry_list);
1828 if (conf->working_disks == 1)
1829 mddev->recovery_cp = MaxSector;
1831 spin_lock_init(&conf->resync_lock);
1832 init_waitqueue_head(&conf->wait_barrier);
1834 bio_list_init(&conf->pending_bio_list);
1835 bio_list_init(&conf->flushing_bio_list);
1837 if (!conf->working_disks) {
1838 printk(KERN_ERR "raid1: no operational mirrors for %s\n",
1843 mddev->degraded = 0;
1844 for (i = 0; i < conf->raid_disks; i++) {
1846 disk = conf->mirrors + i;
1849 disk->head_position = 0;
1855 * find the first working one and use it as a starting point
1856 * to read balancing.
1858 for (j = 0; j < conf->raid_disks &&
1859 (!conf->mirrors[j].rdev ||
1860 !test_bit(In_sync, &conf->mirrors[j].rdev->flags)) ; j++)
1862 conf->last_used = j;
1865 mddev->thread = md_register_thread(raid1d, mddev, "%s_raid1");
1866 if (!mddev->thread) {
1868 "raid1: couldn't allocate thread for %s\n",
1874 "raid1: raid set %s active with %d out of %d mirrors\n",
1875 mdname(mddev), mddev->raid_disks - mddev->degraded,
1878 * Ok, everything is just fine now
1880 mddev->array_size = mddev->size;
1882 mddev->queue->unplug_fn = raid1_unplug;
1883 mddev->queue->issue_flush_fn = raid1_issue_flush;
1888 printk(KERN_ERR "raid1: couldn't allocate memory for %s\n",
1893 if (conf->r1bio_pool)
1894 mempool_destroy(conf->r1bio_pool);
1895 kfree(conf->mirrors);
1896 __free_page(conf->tmppage);
1897 kfree(conf->poolinfo);
1899 mddev->private = NULL;
1905 static int stop(mddev_t *mddev)
1907 conf_t *conf = mddev_to_conf(mddev);
1908 struct bitmap *bitmap = mddev->bitmap;
1909 int behind_wait = 0;
1911 /* wait for behind writes to complete */
1912 while (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
1914 printk(KERN_INFO "raid1: behind writes in progress on device %s, waiting to stop (%d)\n", mdname(mddev), behind_wait);
1915 set_current_state(TASK_UNINTERRUPTIBLE);
1916 schedule_timeout(HZ); /* wait a second */
1917 /* need to kick something here to make sure I/O goes? */
1920 md_unregister_thread(mddev->thread);
1921 mddev->thread = NULL;
1922 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
1923 if (conf->r1bio_pool)
1924 mempool_destroy(conf->r1bio_pool);
1925 kfree(conf->mirrors);
1926 kfree(conf->poolinfo);
1928 mddev->private = NULL;
1932 static int raid1_resize(mddev_t *mddev, sector_t sectors)
1934 /* no resync is happening, and there is enough space
1935 * on all devices, so we can resize.
1936 * We need to make sure resync covers any new space.
1937 * If the array is shrinking we should possibly wait until
1938 * any io in the removed space completes, but it hardly seems
1941 mddev->array_size = sectors>>1;
1942 set_capacity(mddev->gendisk, mddev->array_size << 1);
1944 if (mddev->array_size > mddev->size && mddev->recovery_cp == MaxSector) {
1945 mddev->recovery_cp = mddev->size << 1;
1946 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1948 mddev->size = mddev->array_size;
1949 mddev->resync_max_sectors = sectors;
1953 static int raid1_reshape(mddev_t *mddev, int raid_disks)
1956 * 1/ resize the r1bio_pool
1957 * 2/ resize conf->mirrors
1959 * We allocate a new r1bio_pool if we can.
1960 * Then raise a device barrier and wait until all IO stops.
1961 * Then resize conf->mirrors and swap in the new r1bio pool.
1963 * At the same time, we "pack" the devices so that all the missing
1964 * devices have the higher raid_disk numbers.
1966 mempool_t *newpool, *oldpool;
1967 struct pool_info *newpoolinfo;
1968 mirror_info_t *newmirrors;
1969 conf_t *conf = mddev_to_conf(mddev);
1974 if (raid_disks < conf->raid_disks) {
1976 for (d= 0; d < conf->raid_disks; d++)
1977 if (conf->mirrors[d].rdev)
1979 if (cnt > raid_disks)
1983 newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL);
1986 newpoolinfo->mddev = mddev;
1987 newpoolinfo->raid_disks = raid_disks;
1989 newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
1990 r1bio_pool_free, newpoolinfo);
1995 newmirrors = kmalloc(sizeof(struct mirror_info) * raid_disks, GFP_KERNEL);
1998 mempool_destroy(newpool);
2001 memset(newmirrors, 0, sizeof(struct mirror_info)*raid_disks);
2003 raise_barrier(conf);
2005 /* ok, everything is stopped */
2006 oldpool = conf->r1bio_pool;
2007 conf->r1bio_pool = newpool;
2009 for (d=d2=0; d < conf->raid_disks; d++)
2010 if (conf->mirrors[d].rdev) {
2011 conf->mirrors[d].rdev->raid_disk = d2;
2012 newmirrors[d2++].rdev = conf->mirrors[d].rdev;
2014 kfree(conf->mirrors);
2015 conf->mirrors = newmirrors;
2016 kfree(conf->poolinfo);
2017 conf->poolinfo = newpoolinfo;
2019 mddev->degraded += (raid_disks - conf->raid_disks);
2020 conf->raid_disks = mddev->raid_disks = raid_disks;
2022 conf->last_used = 0; /* just make sure it is in-range */
2023 lower_barrier(conf);
2025 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2026 md_wakeup_thread(mddev->thread);
2028 mempool_destroy(oldpool);
2032 static void raid1_quiesce(mddev_t *mddev, int state)
2034 conf_t *conf = mddev_to_conf(mddev);
2038 raise_barrier(conf);
2041 lower_barrier(conf);
2047 static mdk_personality_t raid1_personality =
2050 .owner = THIS_MODULE,
2051 .make_request = make_request,
2055 .error_handler = error,
2056 .hot_add_disk = raid1_add_disk,
2057 .hot_remove_disk= raid1_remove_disk,
2058 .spare_active = raid1_spare_active,
2059 .sync_request = sync_request,
2060 .resize = raid1_resize,
2061 .reshape = raid1_reshape,
2062 .quiesce = raid1_quiesce,
2065 static int __init raid_init(void)
2067 return register_md_personality(RAID1, &raid1_personality);
2070 static void raid_exit(void)
2072 unregister_md_personality(RAID1);
2075 module_init(raid_init);
2076 module_exit(raid_exit);
2077 MODULE_LICENSE("GPL");
2078 MODULE_ALIAS("md-personality-3"); /* RAID1 */