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 = kzalloc(size, gfp_flags);
66 unplug_slaves(pi->mddev);
71 static void r1bio_pool_free(void *r1_bio, void *data)
76 #define RESYNC_BLOCK_SIZE (64*1024)
77 //#define RESYNC_BLOCK_SIZE PAGE_SIZE
78 #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9)
79 #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE)
80 #define RESYNC_WINDOW (2048*1024)
82 static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data)
84 struct pool_info *pi = data;
90 r1_bio = r1bio_pool_alloc(gfp_flags, pi);
92 unplug_slaves(pi->mddev);
97 * Allocate bios : 1 for reading, n-1 for writing
99 for (j = pi->raid_disks ; j-- ; ) {
100 bio = bio_alloc(gfp_flags, RESYNC_PAGES);
103 r1_bio->bios[j] = bio;
106 * Allocate RESYNC_PAGES data pages and attach them to
108 * If this is a user-requested check/repair, allocate
109 * RESYNC_PAGES for each bio.
111 if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery))
116 bio = r1_bio->bios[j];
117 for (i = 0; i < RESYNC_PAGES; i++) {
118 page = alloc_page(gfp_flags);
122 bio->bi_io_vec[i].bv_page = page;
125 /* If not user-requests, copy the page pointers to all bios */
126 if (!test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) {
127 for (i=0; i<RESYNC_PAGES ; i++)
128 for (j=1; j<pi->raid_disks; j++)
129 r1_bio->bios[j]->bi_io_vec[i].bv_page =
130 r1_bio->bios[0]->bi_io_vec[i].bv_page;
133 r1_bio->master_bio = NULL;
138 for (i=0; i < RESYNC_PAGES ; i++)
139 for (j=0 ; j < pi->raid_disks; j++)
140 put_page(r1_bio->bios[j]->bi_io_vec[i].bv_page);
143 while ( ++j < pi->raid_disks )
144 bio_put(r1_bio->bios[j]);
145 r1bio_pool_free(r1_bio, data);
149 static void r1buf_pool_free(void *__r1_bio, void *data)
151 struct pool_info *pi = data;
153 r1bio_t *r1bio = __r1_bio;
155 for (i = 0; i < RESYNC_PAGES; i++)
156 for (j = pi->raid_disks; j-- ;) {
158 r1bio->bios[j]->bi_io_vec[i].bv_page !=
159 r1bio->bios[0]->bi_io_vec[i].bv_page)
160 put_page(r1bio->bios[j]->bi_io_vec[i].bv_page);
162 for (i=0 ; i < pi->raid_disks; i++)
163 bio_put(r1bio->bios[i]);
165 r1bio_pool_free(r1bio, data);
168 static void put_all_bios(conf_t *conf, r1bio_t *r1_bio)
172 for (i = 0; i < conf->raid_disks; i++) {
173 struct bio **bio = r1_bio->bios + i;
174 if (*bio && *bio != IO_BLOCKED)
180 static inline void free_r1bio(r1bio_t *r1_bio)
182 conf_t *conf = mddev_to_conf(r1_bio->mddev);
185 * Wake up any possible resync thread that waits for the device
190 put_all_bios(conf, r1_bio);
191 mempool_free(r1_bio, conf->r1bio_pool);
194 static inline void put_buf(r1bio_t *r1_bio)
196 conf_t *conf = mddev_to_conf(r1_bio->mddev);
199 for (i=0; i<conf->raid_disks; i++) {
200 struct bio *bio = r1_bio->bios[i];
202 rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev);
205 mempool_free(r1_bio, conf->r1buf_pool);
210 static void reschedule_retry(r1bio_t *r1_bio)
213 mddev_t *mddev = r1_bio->mddev;
214 conf_t *conf = mddev_to_conf(mddev);
216 spin_lock_irqsave(&conf->device_lock, flags);
217 list_add(&r1_bio->retry_list, &conf->retry_list);
219 spin_unlock_irqrestore(&conf->device_lock, flags);
221 wake_up(&conf->wait_barrier);
222 md_wakeup_thread(mddev->thread);
226 * raid_end_bio_io() is called when we have finished servicing a mirrored
227 * operation and are ready to return a success/failure code to the buffer
230 static void raid_end_bio_io(r1bio_t *r1_bio)
232 struct bio *bio = r1_bio->master_bio;
234 /* if nobody has done the final endio yet, do it now */
235 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
236 PRINTK(KERN_DEBUG "raid1: sync end %s on sectors %llu-%llu\n",
237 (bio_data_dir(bio) == WRITE) ? "write" : "read",
238 (unsigned long long) bio->bi_sector,
239 (unsigned long long) bio->bi_sector +
240 (bio->bi_size >> 9) - 1);
242 bio_endio(bio, bio->bi_size,
243 test_bit(R1BIO_Uptodate, &r1_bio->state) ? 0 : -EIO);
249 * Update disk head position estimator based on IRQ completion info.
251 static inline void update_head_pos(int disk, r1bio_t *r1_bio)
253 conf_t *conf = mddev_to_conf(r1_bio->mddev);
255 conf->mirrors[disk].head_position =
256 r1_bio->sector + (r1_bio->sectors);
259 static int raid1_end_read_request(struct bio *bio, unsigned int bytes_done, int error)
261 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
262 r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
264 conf_t *conf = mddev_to_conf(r1_bio->mddev);
269 mirror = r1_bio->read_disk;
271 * this branch is our 'one mirror IO has finished' event handler:
273 update_head_pos(mirror, r1_bio);
275 if (uptodate || conf->working_disks <= 1) {
277 * Set R1BIO_Uptodate in our master bio, so that
278 * we will return a good error code for to the higher
279 * levels even if IO on some other mirrored buffer fails.
281 * The 'master' represents the composite IO operation to
282 * user-side. So if something waits for IO, then it will
283 * wait for the 'master' bio.
286 set_bit(R1BIO_Uptodate, &r1_bio->state);
288 raid_end_bio_io(r1_bio);
293 char b[BDEVNAME_SIZE];
294 if (printk_ratelimit())
295 printk(KERN_ERR "raid1: %s: rescheduling sector %llu\n",
296 bdevname(conf->mirrors[mirror].rdev->bdev,b), (unsigned long long)r1_bio->sector);
297 reschedule_retry(r1_bio);
300 rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev);
304 static int raid1_end_write_request(struct bio *bio, unsigned int bytes_done, int error)
306 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
307 r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
308 int mirror, behind = test_bit(R1BIO_BehindIO, &r1_bio->state);
309 conf_t *conf = mddev_to_conf(r1_bio->mddev);
314 for (mirror = 0; mirror < conf->raid_disks; mirror++)
315 if (r1_bio->bios[mirror] == bio)
318 if (error == -ENOTSUPP && test_bit(R1BIO_Barrier, &r1_bio->state)) {
319 set_bit(BarriersNotsupp, &conf->mirrors[mirror].rdev->flags);
320 set_bit(R1BIO_BarrierRetry, &r1_bio->state);
321 r1_bio->mddev->barriers_work = 0;
324 * this branch is our 'one mirror IO has finished' event handler:
326 r1_bio->bios[mirror] = NULL;
328 md_error(r1_bio->mddev, conf->mirrors[mirror].rdev);
329 /* an I/O failed, we can't clear the bitmap */
330 set_bit(R1BIO_Degraded, &r1_bio->state);
333 * Set R1BIO_Uptodate in our master bio, so that
334 * we will return a good error code for to the higher
335 * levels even if IO on some other mirrored buffer fails.
337 * The 'master' represents the composite IO operation to
338 * user-side. So if something waits for IO, then it will
339 * wait for the 'master' bio.
341 set_bit(R1BIO_Uptodate, &r1_bio->state);
343 update_head_pos(mirror, r1_bio);
346 if (test_bit(WriteMostly, &conf->mirrors[mirror].rdev->flags))
347 atomic_dec(&r1_bio->behind_remaining);
349 /* In behind mode, we ACK the master bio once the I/O has safely
350 * reached all non-writemostly disks. Setting the Returned bit
351 * ensures that this gets done only once -- we don't ever want to
352 * return -EIO here, instead we'll wait */
354 if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) &&
355 test_bit(R1BIO_Uptodate, &r1_bio->state)) {
356 /* Maybe we can return now */
357 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
358 struct bio *mbio = r1_bio->master_bio;
359 PRINTK(KERN_DEBUG "raid1: behind end write sectors %llu-%llu\n",
360 (unsigned long long) mbio->bi_sector,
361 (unsigned long long) mbio->bi_sector +
362 (mbio->bi_size >> 9) - 1);
363 bio_endio(mbio, mbio->bi_size, 0);
370 * Let's see if all mirrored write operations have finished
373 if (atomic_dec_and_test(&r1_bio->remaining)) {
374 if (test_bit(R1BIO_BarrierRetry, &r1_bio->state)) {
375 reschedule_retry(r1_bio);
376 /* Don't dec_pending yet, we want to hold
377 * the reference over the retry
381 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
382 /* free extra copy of the data pages */
383 int i = bio->bi_vcnt;
385 put_page(bio->bi_io_vec[i].bv_page);
387 /* clear the bitmap if all writes complete successfully */
388 bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector,
390 !test_bit(R1BIO_Degraded, &r1_bio->state),
392 md_write_end(r1_bio->mddev);
393 raid_end_bio_io(r1_bio);
396 if (r1_bio->bios[mirror]==NULL)
399 rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev);
405 * This routine returns the disk from which the requested read should
406 * be done. There is a per-array 'next expected sequential IO' sector
407 * number - if this matches on the next IO then we use the last disk.
408 * There is also a per-disk 'last know head position' sector that is
409 * maintained from IRQ contexts, both the normal and the resync IO
410 * completion handlers update this position correctly. If there is no
411 * perfect sequential match then we pick the disk whose head is closest.
413 * If there are 2 mirrors in the same 2 devices, performance degrades
414 * because position is mirror, not device based.
416 * The rdev for the device selected will have nr_pending incremented.
418 static int read_balance(conf_t *conf, r1bio_t *r1_bio)
420 const unsigned long this_sector = r1_bio->sector;
421 int new_disk = conf->last_used, disk = new_disk;
423 const int sectors = r1_bio->sectors;
424 sector_t new_distance, current_distance;
429 * Check if we can balance. We can balance on the whole
430 * device if no resync is going on, or below the resync window.
431 * We take the first readable disk when above the resync window.
434 if (conf->mddev->recovery_cp < MaxSector &&
435 (this_sector + sectors >= conf->next_resync)) {
436 /* Choose the first operation device, for consistancy */
439 for (rdev = rcu_dereference(conf->mirrors[new_disk].rdev);
440 r1_bio->bios[new_disk] == IO_BLOCKED ||
441 !rdev || !test_bit(In_sync, &rdev->flags)
442 || test_bit(WriteMostly, &rdev->flags);
443 rdev = rcu_dereference(conf->mirrors[++new_disk].rdev)) {
445 if (rdev && test_bit(In_sync, &rdev->flags) &&
446 r1_bio->bios[new_disk] != IO_BLOCKED)
447 wonly_disk = new_disk;
449 if (new_disk == conf->raid_disks - 1) {
450 new_disk = wonly_disk;
458 /* make sure the disk is operational */
459 for (rdev = rcu_dereference(conf->mirrors[new_disk].rdev);
460 r1_bio->bios[new_disk] == IO_BLOCKED ||
461 !rdev || !test_bit(In_sync, &rdev->flags) ||
462 test_bit(WriteMostly, &rdev->flags);
463 rdev = rcu_dereference(conf->mirrors[new_disk].rdev)) {
465 if (rdev && test_bit(In_sync, &rdev->flags) &&
466 r1_bio->bios[new_disk] != IO_BLOCKED)
467 wonly_disk = new_disk;
470 new_disk = conf->raid_disks;
472 if (new_disk == disk) {
473 new_disk = wonly_disk;
482 /* now disk == new_disk == starting point for search */
485 * Don't change to another disk for sequential reads:
487 if (conf->next_seq_sect == this_sector)
489 if (this_sector == conf->mirrors[new_disk].head_position)
492 current_distance = abs(this_sector - conf->mirrors[disk].head_position);
494 /* Find the disk whose head is closest */
498 disk = conf->raid_disks;
501 rdev = rcu_dereference(conf->mirrors[disk].rdev);
503 if (!rdev || r1_bio->bios[disk] == IO_BLOCKED ||
504 !test_bit(In_sync, &rdev->flags) ||
505 test_bit(WriteMostly, &rdev->flags))
508 if (!atomic_read(&rdev->nr_pending)) {
512 new_distance = abs(this_sector - conf->mirrors[disk].head_position);
513 if (new_distance < current_distance) {
514 current_distance = new_distance;
517 } while (disk != conf->last_used);
523 rdev = rcu_dereference(conf->mirrors[new_disk].rdev);
526 atomic_inc(&rdev->nr_pending);
527 if (!test_bit(In_sync, &rdev->flags)) {
528 /* cannot risk returning a device that failed
529 * before we inc'ed nr_pending
531 atomic_dec(&rdev->nr_pending);
534 conf->next_seq_sect = this_sector + sectors;
535 conf->last_used = new_disk;
542 static void unplug_slaves(mddev_t *mddev)
544 conf_t *conf = mddev_to_conf(mddev);
548 for (i=0; i<mddev->raid_disks; i++) {
549 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
550 if (rdev && !test_bit(Faulty, &rdev->flags) && atomic_read(&rdev->nr_pending)) {
551 request_queue_t *r_queue = bdev_get_queue(rdev->bdev);
553 atomic_inc(&rdev->nr_pending);
556 if (r_queue->unplug_fn)
557 r_queue->unplug_fn(r_queue);
559 rdev_dec_pending(rdev, mddev);
566 static void raid1_unplug(request_queue_t *q)
568 mddev_t *mddev = q->queuedata;
570 unplug_slaves(mddev);
571 md_wakeup_thread(mddev->thread);
574 static int raid1_issue_flush(request_queue_t *q, struct gendisk *disk,
575 sector_t *error_sector)
577 mddev_t *mddev = q->queuedata;
578 conf_t *conf = mddev_to_conf(mddev);
582 for (i=0; i<mddev->raid_disks && ret == 0; i++) {
583 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
584 if (rdev && !test_bit(Faulty, &rdev->flags)) {
585 struct block_device *bdev = rdev->bdev;
586 request_queue_t *r_queue = bdev_get_queue(bdev);
588 if (!r_queue->issue_flush_fn)
591 atomic_inc(&rdev->nr_pending);
593 ret = r_queue->issue_flush_fn(r_queue, bdev->bd_disk,
595 rdev_dec_pending(rdev, mddev);
605 * Sometimes we need to suspend IO while we do something else,
606 * either some resync/recovery, or reconfigure the array.
607 * To do this we raise a 'barrier'.
608 * The 'barrier' is a counter that can be raised multiple times
609 * to count how many activities are happening which preclude
611 * We can only raise the barrier if there is no pending IO.
612 * i.e. if nr_pending == 0.
613 * We choose only to raise the barrier if no-one is waiting for the
614 * barrier to go down. This means that as soon as an IO request
615 * is ready, no other operations which require a barrier will start
616 * until the IO request has had a chance.
618 * So: regular IO calls 'wait_barrier'. When that returns there
619 * is no backgroup IO happening, It must arrange to call
620 * allow_barrier when it has finished its IO.
621 * backgroup IO calls must call raise_barrier. Once that returns
622 * there is no normal IO happeing. It must arrange to call
623 * lower_barrier when the particular background IO completes.
625 #define RESYNC_DEPTH 32
627 static void raise_barrier(conf_t *conf)
629 spin_lock_irq(&conf->resync_lock);
631 /* Wait until no block IO is waiting */
632 wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting,
634 raid1_unplug(conf->mddev->queue));
636 /* block any new IO from starting */
639 /* No wait for all pending IO to complete */
640 wait_event_lock_irq(conf->wait_barrier,
641 !conf->nr_pending && conf->barrier < RESYNC_DEPTH,
643 raid1_unplug(conf->mddev->queue));
645 spin_unlock_irq(&conf->resync_lock);
648 static void lower_barrier(conf_t *conf)
651 spin_lock_irqsave(&conf->resync_lock, flags);
653 spin_unlock_irqrestore(&conf->resync_lock, flags);
654 wake_up(&conf->wait_barrier);
657 static void wait_barrier(conf_t *conf)
659 spin_lock_irq(&conf->resync_lock);
662 wait_event_lock_irq(conf->wait_barrier, !conf->barrier,
664 raid1_unplug(conf->mddev->queue));
668 spin_unlock_irq(&conf->resync_lock);
671 static void allow_barrier(conf_t *conf)
674 spin_lock_irqsave(&conf->resync_lock, flags);
676 spin_unlock_irqrestore(&conf->resync_lock, flags);
677 wake_up(&conf->wait_barrier);
680 static void freeze_array(conf_t *conf)
682 /* stop syncio and normal IO and wait for everything to
684 * We increment barrier and nr_waiting, and then
685 * wait until barrier+nr_pending match nr_queued+2
687 spin_lock_irq(&conf->resync_lock);
690 wait_event_lock_irq(conf->wait_barrier,
691 conf->barrier+conf->nr_pending == conf->nr_queued+2,
693 raid1_unplug(conf->mddev->queue));
694 spin_unlock_irq(&conf->resync_lock);
696 static void unfreeze_array(conf_t *conf)
698 /* reverse the effect of the freeze */
699 spin_lock_irq(&conf->resync_lock);
702 wake_up(&conf->wait_barrier);
703 spin_unlock_irq(&conf->resync_lock);
707 /* duplicate the data pages for behind I/O */
708 static struct page **alloc_behind_pages(struct bio *bio)
711 struct bio_vec *bvec;
712 struct page **pages = kzalloc(bio->bi_vcnt * sizeof(struct page *),
714 if (unlikely(!pages))
717 bio_for_each_segment(bvec, bio, i) {
718 pages[i] = alloc_page(GFP_NOIO);
719 if (unlikely(!pages[i]))
721 memcpy(kmap(pages[i]) + bvec->bv_offset,
722 kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len);
724 kunmap(bvec->bv_page);
731 for (i = 0; i < bio->bi_vcnt && pages[i]; i++)
734 PRINTK("%dB behind alloc failed, doing sync I/O\n", bio->bi_size);
738 static int make_request(request_queue_t *q, struct bio * bio)
740 mddev_t *mddev = q->queuedata;
741 conf_t *conf = mddev_to_conf(mddev);
742 mirror_info_t *mirror;
744 struct bio *read_bio;
745 int i, targets = 0, disks;
747 struct bitmap *bitmap = mddev->bitmap;
750 struct page **behind_pages = NULL;
751 const int rw = bio_data_dir(bio);
754 if (unlikely(!mddev->barriers_work && bio_barrier(bio))) {
755 bio_endio(bio, bio->bi_size, -EOPNOTSUPP);
760 * Register the new request and wait if the reconstruction
761 * thread has put up a bar for new requests.
762 * Continue immediately if no resync is active currently.
764 md_write_start(mddev, bio); /* wait on superblock update early */
768 disk_stat_inc(mddev->gendisk, ios[rw]);
769 disk_stat_add(mddev->gendisk, sectors[rw], bio_sectors(bio));
772 * make_request() can abort the operation when READA is being
773 * used and no empty request is available.
776 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
778 r1_bio->master_bio = bio;
779 r1_bio->sectors = bio->bi_size >> 9;
781 r1_bio->mddev = mddev;
782 r1_bio->sector = bio->bi_sector;
786 * read balancing logic:
788 int rdisk = read_balance(conf, r1_bio);
791 /* couldn't find anywhere to read from */
792 raid_end_bio_io(r1_bio);
795 mirror = conf->mirrors + rdisk;
797 r1_bio->read_disk = rdisk;
799 read_bio = bio_clone(bio, GFP_NOIO);
801 r1_bio->bios[rdisk] = read_bio;
803 read_bio->bi_sector = r1_bio->sector + mirror->rdev->data_offset;
804 read_bio->bi_bdev = mirror->rdev->bdev;
805 read_bio->bi_end_io = raid1_end_read_request;
806 read_bio->bi_rw = READ;
807 read_bio->bi_private = r1_bio;
809 generic_make_request(read_bio);
816 /* first select target devices under spinlock and
817 * inc refcount on their rdev. Record them by setting
820 disks = conf->raid_disks;
822 { static int first=1;
823 if (first) printk("First Write sector %llu disks %d\n",
824 (unsigned long long)r1_bio->sector, disks);
829 for (i = 0; i < disks; i++) {
830 if ((rdev=rcu_dereference(conf->mirrors[i].rdev)) != NULL &&
831 !test_bit(Faulty, &rdev->flags)) {
832 atomic_inc(&rdev->nr_pending);
833 if (test_bit(Faulty, &rdev->flags)) {
834 atomic_dec(&rdev->nr_pending);
835 r1_bio->bios[i] = NULL;
837 r1_bio->bios[i] = bio;
840 r1_bio->bios[i] = NULL;
844 BUG_ON(targets == 0); /* we never fail the last device */
846 if (targets < conf->raid_disks) {
847 /* array is degraded, we will not clear the bitmap
848 * on I/O completion (see raid1_end_write_request) */
849 set_bit(R1BIO_Degraded, &r1_bio->state);
852 /* do behind I/O ? */
854 atomic_read(&bitmap->behind_writes) < bitmap->max_write_behind &&
855 (behind_pages = alloc_behind_pages(bio)) != NULL)
856 set_bit(R1BIO_BehindIO, &r1_bio->state);
858 atomic_set(&r1_bio->remaining, 0);
859 atomic_set(&r1_bio->behind_remaining, 0);
861 do_barriers = bio->bi_rw & BIO_RW_BARRIER;
863 set_bit(R1BIO_Barrier, &r1_bio->state);
866 for (i = 0; i < disks; i++) {
868 if (!r1_bio->bios[i])
871 mbio = bio_clone(bio, GFP_NOIO);
872 r1_bio->bios[i] = mbio;
874 mbio->bi_sector = r1_bio->sector + conf->mirrors[i].rdev->data_offset;
875 mbio->bi_bdev = conf->mirrors[i].rdev->bdev;
876 mbio->bi_end_io = raid1_end_write_request;
877 mbio->bi_rw = WRITE | do_barriers;
878 mbio->bi_private = r1_bio;
881 struct bio_vec *bvec;
884 /* Yes, I really want the '__' version so that
885 * we clear any unused pointer in the io_vec, rather
886 * than leave them unchanged. This is important
887 * because when we come to free the pages, we won't
888 * know the originial bi_idx, so we just free
891 __bio_for_each_segment(bvec, mbio, j, 0)
892 bvec->bv_page = behind_pages[j];
893 if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags))
894 atomic_inc(&r1_bio->behind_remaining);
897 atomic_inc(&r1_bio->remaining);
899 bio_list_add(&bl, mbio);
901 kfree(behind_pages); /* the behind pages are attached to the bios now */
903 bitmap_startwrite(bitmap, bio->bi_sector, r1_bio->sectors,
904 test_bit(R1BIO_BehindIO, &r1_bio->state));
905 spin_lock_irqsave(&conf->device_lock, flags);
906 bio_list_merge(&conf->pending_bio_list, &bl);
909 blk_plug_device(mddev->queue);
910 spin_unlock_irqrestore(&conf->device_lock, flags);
913 while ((bio = bio_list_pop(&bl)) != NULL)
914 generic_make_request(bio);
920 static void status(struct seq_file *seq, mddev_t *mddev)
922 conf_t *conf = mddev_to_conf(mddev);
925 seq_printf(seq, " [%d/%d] [", conf->raid_disks,
926 conf->working_disks);
927 for (i = 0; i < conf->raid_disks; i++)
928 seq_printf(seq, "%s",
929 conf->mirrors[i].rdev &&
930 test_bit(In_sync, &conf->mirrors[i].rdev->flags) ? "U" : "_");
931 seq_printf(seq, "]");
935 static void error(mddev_t *mddev, mdk_rdev_t *rdev)
937 char b[BDEVNAME_SIZE];
938 conf_t *conf = mddev_to_conf(mddev);
941 * If it is not operational, then we have already marked it as dead
942 * else if it is the last working disks, ignore the error, let the
943 * next level up know.
944 * else mark the drive as failed
946 if (test_bit(In_sync, &rdev->flags)
947 && conf->working_disks == 1)
949 * Don't fail the drive, act as though we were just a
950 * normal single drive
953 if (test_bit(In_sync, &rdev->flags)) {
955 conf->working_disks--;
957 * if recovery is running, make sure it aborts.
959 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
961 clear_bit(In_sync, &rdev->flags);
962 set_bit(Faulty, &rdev->flags);
964 printk(KERN_ALERT "raid1: Disk failure on %s, disabling device. \n"
965 " Operation continuing on %d devices\n",
966 bdevname(rdev->bdev,b), conf->working_disks);
969 static void print_conf(conf_t *conf)
974 printk("RAID1 conf printout:\n");
979 printk(" --- wd:%d rd:%d\n", conf->working_disks,
982 for (i = 0; i < conf->raid_disks; i++) {
983 char b[BDEVNAME_SIZE];
984 tmp = conf->mirrors + i;
986 printk(" disk %d, wo:%d, o:%d, dev:%s\n",
987 i, !test_bit(In_sync, &tmp->rdev->flags), !test_bit(Faulty, &tmp->rdev->flags),
988 bdevname(tmp->rdev->bdev,b));
992 static void close_sync(conf_t *conf)
997 mempool_destroy(conf->r1buf_pool);
998 conf->r1buf_pool = NULL;
1001 static int raid1_spare_active(mddev_t *mddev)
1004 conf_t *conf = mddev->private;
1008 * Find all failed disks within the RAID1 configuration
1009 * and mark them readable
1011 for (i = 0; i < conf->raid_disks; i++) {
1012 tmp = conf->mirrors + i;
1014 && !test_bit(Faulty, &tmp->rdev->flags)
1015 && !test_bit(In_sync, &tmp->rdev->flags)) {
1016 conf->working_disks++;
1018 set_bit(In_sync, &tmp->rdev->flags);
1027 static int raid1_add_disk(mddev_t *mddev, mdk_rdev_t *rdev)
1029 conf_t *conf = mddev->private;
1034 for (mirror=0; mirror < mddev->raid_disks; mirror++)
1035 if ( !(p=conf->mirrors+mirror)->rdev) {
1037 blk_queue_stack_limits(mddev->queue,
1038 rdev->bdev->bd_disk->queue);
1039 /* as we don't honour merge_bvec_fn, we must never risk
1040 * violating it, so limit ->max_sector to one PAGE, as
1041 * a one page request is never in violation.
1043 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
1044 mddev->queue->max_sectors > (PAGE_SIZE>>9))
1045 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
1047 p->head_position = 0;
1048 rdev->raid_disk = mirror;
1050 /* As all devices are equivalent, we don't need a full recovery
1051 * if this was recently any drive of the array
1053 if (rdev->saved_raid_disk < 0)
1055 rcu_assign_pointer(p->rdev, rdev);
1063 static int raid1_remove_disk(mddev_t *mddev, int number)
1065 conf_t *conf = mddev->private;
1068 mirror_info_t *p = conf->mirrors+ number;
1073 if (test_bit(In_sync, &rdev->flags) ||
1074 atomic_read(&rdev->nr_pending)) {
1080 if (atomic_read(&rdev->nr_pending)) {
1081 /* lost the race, try later */
1093 static int end_sync_read(struct bio *bio, unsigned int bytes_done, int error)
1095 r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
1101 for (i=r1_bio->mddev->raid_disks; i--; )
1102 if (r1_bio->bios[i] == bio)
1105 update_head_pos(i, r1_bio);
1107 * we have read a block, now it needs to be re-written,
1108 * or re-read if the read failed.
1109 * We don't do much here, just schedule handling by raid1d
1111 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
1112 set_bit(R1BIO_Uptodate, &r1_bio->state);
1114 if (atomic_dec_and_test(&r1_bio->remaining))
1115 reschedule_retry(r1_bio);
1119 static int end_sync_write(struct bio *bio, unsigned int bytes_done, int error)
1121 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1122 r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
1123 mddev_t *mddev = r1_bio->mddev;
1124 conf_t *conf = mddev_to_conf(mddev);
1131 for (i = 0; i < conf->raid_disks; i++)
1132 if (r1_bio->bios[i] == bio) {
1137 md_error(mddev, conf->mirrors[mirror].rdev);
1139 update_head_pos(mirror, r1_bio);
1141 if (atomic_dec_and_test(&r1_bio->remaining)) {
1142 md_done_sync(mddev, r1_bio->sectors, uptodate);
1148 static void sync_request_write(mddev_t *mddev, r1bio_t *r1_bio)
1150 conf_t *conf = mddev_to_conf(mddev);
1152 int disks = conf->raid_disks;
1153 struct bio *bio, *wbio;
1155 bio = r1_bio->bios[r1_bio->read_disk];
1158 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
1159 /* We have read all readable devices. If we haven't
1160 * got the block, then there is no hope left.
1161 * If we have, then we want to do a comparison
1162 * and skip the write if everything is the same.
1163 * If any blocks failed to read, then we need to
1164 * attempt an over-write
1167 if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) {
1168 for (i=0; i<mddev->raid_disks; i++)
1169 if (r1_bio->bios[i]->bi_end_io == end_sync_read)
1170 md_error(mddev, conf->mirrors[i].rdev);
1172 md_done_sync(mddev, r1_bio->sectors, 1);
1176 for (primary=0; primary<mddev->raid_disks; primary++)
1177 if (r1_bio->bios[primary]->bi_end_io == end_sync_read &&
1178 test_bit(BIO_UPTODATE, &r1_bio->bios[primary]->bi_flags)) {
1179 r1_bio->bios[primary]->bi_end_io = NULL;
1182 r1_bio->read_disk = primary;
1183 for (i=0; i<mddev->raid_disks; i++)
1184 if (r1_bio->bios[i]->bi_end_io == end_sync_read &&
1185 test_bit(BIO_UPTODATE, &r1_bio->bios[i]->bi_flags)) {
1187 int vcnt = r1_bio->sectors >> (PAGE_SHIFT- 9);
1188 struct bio *pbio = r1_bio->bios[primary];
1189 struct bio *sbio = r1_bio->bios[i];
1190 for (j = vcnt; j-- ; )
1191 if (memcmp(page_address(pbio->bi_io_vec[j].bv_page),
1192 page_address(sbio->bi_io_vec[j].bv_page),
1196 mddev->resync_mismatches += r1_bio->sectors;
1197 if (j < 0 || test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
1198 sbio->bi_end_io = NULL;
1200 /* fixup the bio for reuse */
1201 sbio->bi_vcnt = vcnt;
1202 sbio->bi_size = r1_bio->sectors << 9;
1204 sbio->bi_phys_segments = 0;
1205 sbio->bi_hw_segments = 0;
1206 sbio->bi_hw_front_size = 0;
1207 sbio->bi_hw_back_size = 0;
1208 sbio->bi_flags &= ~(BIO_POOL_MASK - 1);
1209 sbio->bi_flags |= 1 << BIO_UPTODATE;
1210 sbio->bi_next = NULL;
1211 sbio->bi_sector = r1_bio->sector +
1212 conf->mirrors[i].rdev->data_offset;
1213 sbio->bi_bdev = conf->mirrors[i].rdev->bdev;
1217 if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) {
1218 /* ouch - failed to read all of that.
1219 * Try some synchronous reads of other devices to get
1220 * good data, much like with normal read errors. Only
1221 * read into the pages we already have so they we don't
1222 * need to re-issue the read request.
1223 * We don't need to freeze the array, because being in an
1224 * active sync request, there is no normal IO, and
1225 * no overlapping syncs.
1227 sector_t sect = r1_bio->sector;
1228 int sectors = r1_bio->sectors;
1233 int d = r1_bio->read_disk;
1237 if (s > (PAGE_SIZE>>9))
1240 if (r1_bio->bios[d]->bi_end_io == end_sync_read) {
1241 rdev = conf->mirrors[d].rdev;
1242 if (sync_page_io(rdev->bdev,
1243 sect + rdev->data_offset,
1245 bio->bi_io_vec[idx].bv_page,
1252 if (d == conf->raid_disks)
1254 } while (!success && d != r1_bio->read_disk);
1257 /* write it back and re-read */
1258 set_bit(R1BIO_Uptodate, &r1_bio->state);
1259 while (d != r1_bio->read_disk) {
1261 d = conf->raid_disks;
1263 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1265 rdev = conf->mirrors[d].rdev;
1266 if (sync_page_io(rdev->bdev,
1267 sect + rdev->data_offset,
1269 bio->bi_io_vec[idx].bv_page,
1271 sync_page_io(rdev->bdev,
1272 sect + rdev->data_offset,
1274 bio->bi_io_vec[idx].bv_page,
1276 md_error(mddev, rdev);
1280 char b[BDEVNAME_SIZE];
1281 /* Cannot read from anywhere, array is toast */
1282 md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev);
1283 printk(KERN_ALERT "raid1: %s: unrecoverable I/O read error"
1284 " for block %llu\n",
1285 bdevname(bio->bi_bdev,b),
1286 (unsigned long long)r1_bio->sector);
1287 md_done_sync(mddev, r1_bio->sectors, 0);
1300 atomic_set(&r1_bio->remaining, 1);
1301 for (i = 0; i < disks ; i++) {
1302 wbio = r1_bio->bios[i];
1303 if (wbio->bi_end_io == NULL ||
1304 (wbio->bi_end_io == end_sync_read &&
1305 (i == r1_bio->read_disk ||
1306 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery))))
1309 wbio->bi_rw = WRITE;
1310 wbio->bi_end_io = end_sync_write;
1311 atomic_inc(&r1_bio->remaining);
1312 md_sync_acct(conf->mirrors[i].rdev->bdev, wbio->bi_size >> 9);
1314 generic_make_request(wbio);
1317 if (atomic_dec_and_test(&r1_bio->remaining)) {
1318 /* if we're here, all write(s) have completed, so clean up */
1319 md_done_sync(mddev, r1_bio->sectors, 1);
1325 * This is a kernel thread which:
1327 * 1. Retries failed read operations on working mirrors.
1328 * 2. Updates the raid superblock when problems encounter.
1329 * 3. Performs writes following reads for array syncronising.
1332 static void raid1d(mddev_t *mddev)
1336 unsigned long flags;
1337 conf_t *conf = mddev_to_conf(mddev);
1338 struct list_head *head = &conf->retry_list;
1342 md_check_recovery(mddev);
1345 char b[BDEVNAME_SIZE];
1346 spin_lock_irqsave(&conf->device_lock, flags);
1348 if (conf->pending_bio_list.head) {
1349 bio = bio_list_get(&conf->pending_bio_list);
1350 blk_remove_plug(mddev->queue);
1351 spin_unlock_irqrestore(&conf->device_lock, flags);
1352 /* flush any pending bitmap writes to disk before proceeding w/ I/O */
1353 if (bitmap_unplug(mddev->bitmap) != 0)
1354 printk("%s: bitmap file write failed!\n", mdname(mddev));
1356 while (bio) { /* submit pending writes */
1357 struct bio *next = bio->bi_next;
1358 bio->bi_next = NULL;
1359 generic_make_request(bio);
1367 if (list_empty(head))
1369 r1_bio = list_entry(head->prev, r1bio_t, retry_list);
1370 list_del(head->prev);
1372 spin_unlock_irqrestore(&conf->device_lock, flags);
1374 mddev = r1_bio->mddev;
1375 conf = mddev_to_conf(mddev);
1376 if (test_bit(R1BIO_IsSync, &r1_bio->state)) {
1377 sync_request_write(mddev, r1_bio);
1379 } else if (test_bit(R1BIO_BarrierRetry, &r1_bio->state)) {
1380 /* some requests in the r1bio were BIO_RW_BARRIER
1381 * requests which failed with -ENOTSUPP. Hohumm..
1382 * Better resubmit without the barrier.
1383 * We know which devices to resubmit for, because
1384 * all others have had their bios[] entry cleared.
1387 clear_bit(R1BIO_BarrierRetry, &r1_bio->state);
1388 clear_bit(R1BIO_Barrier, &r1_bio->state);
1389 for (i=0; i < conf->raid_disks; i++)
1390 if (r1_bio->bios[i]) {
1391 struct bio_vec *bvec;
1394 bio = bio_clone(r1_bio->master_bio, GFP_NOIO);
1395 /* copy pages from the failed bio, as
1396 * this might be a write-behind device */
1397 __bio_for_each_segment(bvec, bio, j, 0)
1398 bvec->bv_page = bio_iovec_idx(r1_bio->bios[i], j)->bv_page;
1399 bio_put(r1_bio->bios[i]);
1400 bio->bi_sector = r1_bio->sector +
1401 conf->mirrors[i].rdev->data_offset;
1402 bio->bi_bdev = conf->mirrors[i].rdev->bdev;
1403 bio->bi_end_io = raid1_end_write_request;
1405 bio->bi_private = r1_bio;
1406 r1_bio->bios[i] = bio;
1407 generic_make_request(bio);
1412 /* we got a read error. Maybe the drive is bad. Maybe just
1413 * the block and we can fix it.
1414 * We freeze all other IO, and try reading the block from
1415 * other devices. When we find one, we re-write
1416 * and check it that fixes the read error.
1417 * This is all done synchronously while the array is
1420 sector_t sect = r1_bio->sector;
1421 int sectors = r1_bio->sectors;
1423 if (mddev->ro == 0) while(sectors) {
1425 int d = r1_bio->read_disk;
1428 if (s > (PAGE_SIZE>>9))
1432 rdev = conf->mirrors[d].rdev;
1434 test_bit(In_sync, &rdev->flags) &&
1435 sync_page_io(rdev->bdev,
1436 sect + rdev->data_offset,
1438 conf->tmppage, READ))
1442 if (d == conf->raid_disks)
1445 } while (!success && d != r1_bio->read_disk);
1448 /* write it back and re-read */
1449 while (d != r1_bio->read_disk) {
1451 d = conf->raid_disks;
1453 rdev = conf->mirrors[d].rdev;
1455 test_bit(In_sync, &rdev->flags)) {
1456 if (sync_page_io(rdev->bdev,
1457 sect + rdev->data_offset,
1458 s<<9, conf->tmppage, WRITE) == 0 ||
1459 sync_page_io(rdev->bdev,
1460 sect + rdev->data_offset,
1461 s<<9, conf->tmppage, READ) == 0) {
1462 /* Well, this device is dead */
1463 md_error(mddev, rdev);
1468 /* Cannot read from anywhere -- bye bye array */
1469 md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev);
1476 unfreeze_array(conf);
1478 bio = r1_bio->bios[r1_bio->read_disk];
1479 if ((disk=read_balance(conf, r1_bio)) == -1) {
1480 printk(KERN_ALERT "raid1: %s: unrecoverable I/O"
1481 " read error for block %llu\n",
1482 bdevname(bio->bi_bdev,b),
1483 (unsigned long long)r1_bio->sector);
1484 raid_end_bio_io(r1_bio);
1486 r1_bio->bios[r1_bio->read_disk] =
1487 mddev->ro ? IO_BLOCKED : NULL;
1488 r1_bio->read_disk = disk;
1490 bio = bio_clone(r1_bio->master_bio, GFP_NOIO);
1491 r1_bio->bios[r1_bio->read_disk] = bio;
1492 rdev = conf->mirrors[disk].rdev;
1493 if (printk_ratelimit())
1494 printk(KERN_ERR "raid1: %s: redirecting sector %llu to"
1495 " another mirror\n",
1496 bdevname(rdev->bdev,b),
1497 (unsigned long long)r1_bio->sector);
1498 bio->bi_sector = r1_bio->sector + rdev->data_offset;
1499 bio->bi_bdev = rdev->bdev;
1500 bio->bi_end_io = raid1_end_read_request;
1502 bio->bi_private = r1_bio;
1504 generic_make_request(bio);
1508 spin_unlock_irqrestore(&conf->device_lock, flags);
1510 unplug_slaves(mddev);
1514 static int init_resync(conf_t *conf)
1518 buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;
1519 if (conf->r1buf_pool)
1521 conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free,
1523 if (!conf->r1buf_pool)
1525 conf->next_resync = 0;
1530 * perform a "sync" on one "block"
1532 * We need to make sure that no normal I/O request - particularly write
1533 * requests - conflict with active sync requests.
1535 * This is achieved by tracking pending requests and a 'barrier' concept
1536 * that can be installed to exclude normal IO requests.
1539 static sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster)
1541 conf_t *conf = mddev_to_conf(mddev);
1544 sector_t max_sector, nr_sectors;
1548 int write_targets = 0, read_targets = 0;
1550 int still_degraded = 0;
1552 if (!conf->r1buf_pool)
1555 printk("sync start - bitmap %p\n", mddev->bitmap);
1557 if (init_resync(conf))
1561 max_sector = mddev->size << 1;
1562 if (sector_nr >= max_sector) {
1563 /* If we aborted, we need to abort the
1564 * sync on the 'current' bitmap chunk (there will
1565 * only be one in raid1 resync.
1566 * We can find the current addess in mddev->curr_resync
1568 if (mddev->curr_resync < max_sector) /* aborted */
1569 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
1571 else /* completed sync */
1574 bitmap_close_sync(mddev->bitmap);
1579 /* before building a request, check if we can skip these blocks..
1580 * This call the bitmap_start_sync doesn't actually record anything
1582 if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
1583 !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
1584 /* We can skip this block, and probably several more */
1589 * If there is non-resync activity waiting for a turn,
1590 * and resync is going fast enough,
1591 * then let it though before starting on this new sync request.
1593 if (!go_faster && conf->nr_waiting)
1594 msleep_interruptible(1000);
1596 raise_barrier(conf);
1598 conf->next_resync = sector_nr;
1600 r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO);
1603 * If we get a correctably read error during resync or recovery,
1604 * we might want to read from a different device. So we
1605 * flag all drives that could conceivably be read from for READ,
1606 * and any others (which will be non-In_sync devices) for WRITE.
1607 * If a read fails, we try reading from something else for which READ
1611 r1_bio->mddev = mddev;
1612 r1_bio->sector = sector_nr;
1614 set_bit(R1BIO_IsSync, &r1_bio->state);
1616 for (i=0; i < conf->raid_disks; i++) {
1618 bio = r1_bio->bios[i];
1620 /* take from bio_init */
1621 bio->bi_next = NULL;
1622 bio->bi_flags |= 1 << BIO_UPTODATE;
1626 bio->bi_phys_segments = 0;
1627 bio->bi_hw_segments = 0;
1629 bio->bi_end_io = NULL;
1630 bio->bi_private = NULL;
1632 rdev = rcu_dereference(conf->mirrors[i].rdev);
1634 test_bit(Faulty, &rdev->flags)) {
1637 } else if (!test_bit(In_sync, &rdev->flags)) {
1639 bio->bi_end_io = end_sync_write;
1642 /* may need to read from here */
1644 bio->bi_end_io = end_sync_read;
1645 if (test_bit(WriteMostly, &rdev->flags)) {
1654 atomic_inc(&rdev->nr_pending);
1655 bio->bi_sector = sector_nr + rdev->data_offset;
1656 bio->bi_bdev = rdev->bdev;
1657 bio->bi_private = r1_bio;
1662 r1_bio->read_disk = disk;
1664 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0)
1665 /* extra read targets are also write targets */
1666 write_targets += read_targets-1;
1668 if (write_targets == 0 || read_targets == 0) {
1669 /* There is nowhere to write, so all non-sync
1670 * drives must be failed - so we are finished
1672 sector_t rv = max_sector - sector_nr;
1682 int len = PAGE_SIZE;
1683 if (sector_nr + (len>>9) > max_sector)
1684 len = (max_sector - sector_nr) << 9;
1687 if (sync_blocks == 0) {
1688 if (!bitmap_start_sync(mddev->bitmap, sector_nr,
1689 &sync_blocks, still_degraded) &&
1691 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1693 if (sync_blocks < (PAGE_SIZE>>9))
1695 if (len > (sync_blocks<<9))
1696 len = sync_blocks<<9;
1699 for (i=0 ; i < conf->raid_disks; i++) {
1700 bio = r1_bio->bios[i];
1701 if (bio->bi_end_io) {
1702 page = bio->bi_io_vec[bio->bi_vcnt].bv_page;
1703 if (bio_add_page(bio, page, len, 0) == 0) {
1705 bio->bi_io_vec[bio->bi_vcnt].bv_page = page;
1708 bio = r1_bio->bios[i];
1709 if (bio->bi_end_io==NULL)
1711 /* remove last page from this bio */
1713 bio->bi_size -= len;
1714 bio->bi_flags &= ~(1<< BIO_SEG_VALID);
1720 nr_sectors += len>>9;
1721 sector_nr += len>>9;
1722 sync_blocks -= (len>>9);
1723 } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES);
1725 r1_bio->sectors = nr_sectors;
1727 /* For a user-requested sync, we read all readable devices and do a
1730 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
1731 atomic_set(&r1_bio->remaining, read_targets);
1732 for (i=0; i<conf->raid_disks; i++) {
1733 bio = r1_bio->bios[i];
1734 if (bio->bi_end_io == end_sync_read) {
1735 md_sync_acct(conf->mirrors[i].rdev->bdev, nr_sectors);
1736 generic_make_request(bio);
1740 atomic_set(&r1_bio->remaining, 1);
1741 bio = r1_bio->bios[r1_bio->read_disk];
1742 md_sync_acct(conf->mirrors[r1_bio->read_disk].rdev->bdev,
1744 generic_make_request(bio);
1751 static int run(mddev_t *mddev)
1755 mirror_info_t *disk;
1757 struct list_head *tmp;
1759 if (mddev->level != 1) {
1760 printk("raid1: %s: raid level not set to mirroring (%d)\n",
1761 mdname(mddev), mddev->level);
1765 * copy the already verified devices into our private RAID1
1766 * bookkeeping area. [whatever we allocate in run(),
1767 * should be freed in stop()]
1769 conf = kzalloc(sizeof(conf_t), GFP_KERNEL);
1770 mddev->private = conf;
1774 conf->mirrors = kzalloc(sizeof(struct mirror_info)*mddev->raid_disks,
1779 conf->tmppage = alloc_page(GFP_KERNEL);
1783 conf->poolinfo = kmalloc(sizeof(*conf->poolinfo), GFP_KERNEL);
1784 if (!conf->poolinfo)
1786 conf->poolinfo->mddev = mddev;
1787 conf->poolinfo->raid_disks = mddev->raid_disks;
1788 conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
1791 if (!conf->r1bio_pool)
1794 ITERATE_RDEV(mddev, rdev, tmp) {
1795 disk_idx = rdev->raid_disk;
1796 if (disk_idx >= mddev->raid_disks
1799 disk = conf->mirrors + disk_idx;
1803 blk_queue_stack_limits(mddev->queue,
1804 rdev->bdev->bd_disk->queue);
1805 /* as we don't honour merge_bvec_fn, we must never risk
1806 * violating it, so limit ->max_sector to one PAGE, as
1807 * a one page request is never in violation.
1809 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
1810 mddev->queue->max_sectors > (PAGE_SIZE>>9))
1811 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
1813 disk->head_position = 0;
1814 if (!test_bit(Faulty, &rdev->flags) && test_bit(In_sync, &rdev->flags))
1815 conf->working_disks++;
1817 conf->raid_disks = mddev->raid_disks;
1818 conf->mddev = mddev;
1819 spin_lock_init(&conf->device_lock);
1820 INIT_LIST_HEAD(&conf->retry_list);
1821 if (conf->working_disks == 1)
1822 mddev->recovery_cp = MaxSector;
1824 spin_lock_init(&conf->resync_lock);
1825 init_waitqueue_head(&conf->wait_barrier);
1827 bio_list_init(&conf->pending_bio_list);
1828 bio_list_init(&conf->flushing_bio_list);
1830 if (!conf->working_disks) {
1831 printk(KERN_ERR "raid1: no operational mirrors for %s\n",
1836 mddev->degraded = 0;
1837 for (i = 0; i < conf->raid_disks; i++) {
1839 disk = conf->mirrors + i;
1842 disk->head_position = 0;
1848 * find the first working one and use it as a starting point
1849 * to read balancing.
1851 for (j = 0; j < conf->raid_disks &&
1852 (!conf->mirrors[j].rdev ||
1853 !test_bit(In_sync, &conf->mirrors[j].rdev->flags)) ; j++)
1855 conf->last_used = j;
1858 mddev->thread = md_register_thread(raid1d, mddev, "%s_raid1");
1859 if (!mddev->thread) {
1861 "raid1: couldn't allocate thread for %s\n",
1867 "raid1: raid set %s active with %d out of %d mirrors\n",
1868 mdname(mddev), mddev->raid_disks - mddev->degraded,
1871 * Ok, everything is just fine now
1873 mddev->array_size = mddev->size;
1875 mddev->queue->unplug_fn = raid1_unplug;
1876 mddev->queue->issue_flush_fn = raid1_issue_flush;
1881 printk(KERN_ERR "raid1: couldn't allocate memory for %s\n",
1886 if (conf->r1bio_pool)
1887 mempool_destroy(conf->r1bio_pool);
1888 kfree(conf->mirrors);
1889 put_page(conf->tmppage);
1890 kfree(conf->poolinfo);
1892 mddev->private = NULL;
1898 static int stop(mddev_t *mddev)
1900 conf_t *conf = mddev_to_conf(mddev);
1901 struct bitmap *bitmap = mddev->bitmap;
1902 int behind_wait = 0;
1904 /* wait for behind writes to complete */
1905 while (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
1907 printk(KERN_INFO "raid1: behind writes in progress on device %s, waiting to stop (%d)\n", mdname(mddev), behind_wait);
1908 set_current_state(TASK_UNINTERRUPTIBLE);
1909 schedule_timeout(HZ); /* wait a second */
1910 /* need to kick something here to make sure I/O goes? */
1913 md_unregister_thread(mddev->thread);
1914 mddev->thread = NULL;
1915 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
1916 if (conf->r1bio_pool)
1917 mempool_destroy(conf->r1bio_pool);
1918 kfree(conf->mirrors);
1919 kfree(conf->poolinfo);
1921 mddev->private = NULL;
1925 static int raid1_resize(mddev_t *mddev, sector_t sectors)
1927 /* no resync is happening, and there is enough space
1928 * on all devices, so we can resize.
1929 * We need to make sure resync covers any new space.
1930 * If the array is shrinking we should possibly wait until
1931 * any io in the removed space completes, but it hardly seems
1934 mddev->array_size = sectors>>1;
1935 set_capacity(mddev->gendisk, mddev->array_size << 1);
1937 if (mddev->array_size > mddev->size && mddev->recovery_cp == MaxSector) {
1938 mddev->recovery_cp = mddev->size << 1;
1939 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1941 mddev->size = mddev->array_size;
1942 mddev->resync_max_sectors = sectors;
1946 static int raid1_reshape(mddev_t *mddev, int raid_disks)
1949 * 1/ resize the r1bio_pool
1950 * 2/ resize conf->mirrors
1952 * We allocate a new r1bio_pool if we can.
1953 * Then raise a device barrier and wait until all IO stops.
1954 * Then resize conf->mirrors and swap in the new r1bio pool.
1956 * At the same time, we "pack" the devices so that all the missing
1957 * devices have the higher raid_disk numbers.
1959 mempool_t *newpool, *oldpool;
1960 struct pool_info *newpoolinfo;
1961 mirror_info_t *newmirrors;
1962 conf_t *conf = mddev_to_conf(mddev);
1967 if (raid_disks < conf->raid_disks) {
1969 for (d= 0; d < conf->raid_disks; d++)
1970 if (conf->mirrors[d].rdev)
1972 if (cnt > raid_disks)
1976 newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL);
1979 newpoolinfo->mddev = mddev;
1980 newpoolinfo->raid_disks = raid_disks;
1982 newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
1983 r1bio_pool_free, newpoolinfo);
1988 newmirrors = kzalloc(sizeof(struct mirror_info) * raid_disks, GFP_KERNEL);
1991 mempool_destroy(newpool);
1995 raise_barrier(conf);
1997 /* ok, everything is stopped */
1998 oldpool = conf->r1bio_pool;
1999 conf->r1bio_pool = newpool;
2001 for (d=d2=0; d < conf->raid_disks; d++)
2002 if (conf->mirrors[d].rdev) {
2003 conf->mirrors[d].rdev->raid_disk = d2;
2004 newmirrors[d2++].rdev = conf->mirrors[d].rdev;
2006 kfree(conf->mirrors);
2007 conf->mirrors = newmirrors;
2008 kfree(conf->poolinfo);
2009 conf->poolinfo = newpoolinfo;
2011 mddev->degraded += (raid_disks - conf->raid_disks);
2012 conf->raid_disks = mddev->raid_disks = raid_disks;
2014 conf->last_used = 0; /* just make sure it is in-range */
2015 lower_barrier(conf);
2017 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2018 md_wakeup_thread(mddev->thread);
2020 mempool_destroy(oldpool);
2024 static void raid1_quiesce(mddev_t *mddev, int state)
2026 conf_t *conf = mddev_to_conf(mddev);
2030 raise_barrier(conf);
2033 lower_barrier(conf);
2039 static mdk_personality_t raid1_personality =
2042 .owner = THIS_MODULE,
2043 .make_request = make_request,
2047 .error_handler = error,
2048 .hot_add_disk = raid1_add_disk,
2049 .hot_remove_disk= raid1_remove_disk,
2050 .spare_active = raid1_spare_active,
2051 .sync_request = sync_request,
2052 .resize = raid1_resize,
2053 .reshape = raid1_reshape,
2054 .quiesce = raid1_quiesce,
2057 static int __init raid_init(void)
2059 return register_md_personality(RAID1, &raid1_personality);
2062 static void raid_exit(void)
2064 unregister_md_personality(RAID1);
2067 module_init(raid_init);
2068 module_exit(raid_exit);
2069 MODULE_LICENSE("GPL");
2070 MODULE_ALIAS("md-personality-3"); /* RAID1 */