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 <linux/slab.h>
35 #include <linux/delay.h>
36 #include <linux/blkdev.h>
37 #include <linux/module.h>
38 #include <linux/seq_file.h>
39 #include <linux/ratelimit.h>
40 #include <trace/events/block.h>
46 * Number of guaranteed r1bios in case of extreme VM load:
48 #define NR_RAID1_BIOS 256
50 /* when we get a read error on a read-only array, we redirect to another
51 * device without failing the first device, or trying to over-write to
52 * correct the read error. To keep track of bad blocks on a per-bio
53 * level, we store IO_BLOCKED in the appropriate 'bios' pointer
55 #define IO_BLOCKED ((struct bio *)1)
56 /* When we successfully write to a known bad-block, we need to remove the
57 * bad-block marking which must be done from process context. So we record
58 * the success by setting devs[n].bio to IO_MADE_GOOD
60 #define IO_MADE_GOOD ((struct bio *)2)
62 #define BIO_SPECIAL(bio) ((unsigned long)bio <= 2)
64 /* When there are this many requests queue to be written by
65 * the raid1 thread, we become 'congested' to provide back-pressure
68 static int max_queued_requests = 1024;
70 static void allow_barrier(struct r1conf *conf, sector_t start_next_window,
72 static void lower_barrier(struct r1conf *conf);
74 #define raid1_log(md, fmt, args...) \
75 do { if ((md)->queue) blk_add_trace_msg((md)->queue, "raid1 " fmt, ##args); } while (0)
77 static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data)
79 struct pool_info *pi = data;
80 int size = offsetof(struct r1bio, bios[pi->raid_disks]);
82 /* allocate a r1bio with room for raid_disks entries in the bios array */
83 return kzalloc(size, gfp_flags);
86 static void r1bio_pool_free(void *r1_bio, void *data)
91 #define RESYNC_BLOCK_SIZE (64*1024)
92 #define RESYNC_DEPTH 32
93 #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9)
94 #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE)
95 #define RESYNC_WINDOW (RESYNC_BLOCK_SIZE * RESYNC_DEPTH)
96 #define RESYNC_WINDOW_SECTORS (RESYNC_WINDOW >> 9)
97 #define CLUSTER_RESYNC_WINDOW (16 * RESYNC_WINDOW)
98 #define CLUSTER_RESYNC_WINDOW_SECTORS (CLUSTER_RESYNC_WINDOW >> 9)
99 #define NEXT_NORMALIO_DISTANCE (3 * RESYNC_WINDOW_SECTORS)
101 static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data)
103 struct pool_info *pi = data;
104 struct r1bio *r1_bio;
109 r1_bio = r1bio_pool_alloc(gfp_flags, pi);
114 * Allocate bios : 1 for reading, n-1 for writing
116 for (j = pi->raid_disks ; j-- ; ) {
117 bio = bio_kmalloc(gfp_flags, RESYNC_PAGES);
120 r1_bio->bios[j] = bio;
123 * Allocate RESYNC_PAGES data pages and attach them to
125 * If this is a user-requested check/repair, allocate
126 * RESYNC_PAGES for each bio.
128 if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery))
129 need_pages = pi->raid_disks;
132 for (j = 0; j < need_pages; j++) {
133 bio = r1_bio->bios[j];
134 bio->bi_vcnt = RESYNC_PAGES;
136 if (bio_alloc_pages(bio, gfp_flags))
139 /* If not user-requests, copy the page pointers to all bios */
140 if (!test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) {
141 for (i=0; i<RESYNC_PAGES ; i++)
142 for (j=1; j<pi->raid_disks; j++)
143 r1_bio->bios[j]->bi_io_vec[i].bv_page =
144 r1_bio->bios[0]->bi_io_vec[i].bv_page;
147 r1_bio->master_bio = NULL;
153 bio_free_pages(r1_bio->bios[j]);
156 while (++j < pi->raid_disks)
157 bio_put(r1_bio->bios[j]);
158 r1bio_pool_free(r1_bio, data);
162 static void r1buf_pool_free(void *__r1_bio, void *data)
164 struct pool_info *pi = data;
166 struct r1bio *r1bio = __r1_bio;
168 for (i = 0; i < RESYNC_PAGES; i++)
169 for (j = pi->raid_disks; j-- ;) {
171 r1bio->bios[j]->bi_io_vec[i].bv_page !=
172 r1bio->bios[0]->bi_io_vec[i].bv_page)
173 safe_put_page(r1bio->bios[j]->bi_io_vec[i].bv_page);
175 for (i=0 ; i < pi->raid_disks; i++)
176 bio_put(r1bio->bios[i]);
178 r1bio_pool_free(r1bio, data);
181 static void put_all_bios(struct r1conf *conf, struct r1bio *r1_bio)
185 for (i = 0; i < conf->raid_disks * 2; i++) {
186 struct bio **bio = r1_bio->bios + i;
187 if (!BIO_SPECIAL(*bio))
193 static void free_r1bio(struct r1bio *r1_bio)
195 struct r1conf *conf = r1_bio->mddev->private;
197 put_all_bios(conf, r1_bio);
198 mempool_free(r1_bio, conf->r1bio_pool);
201 static void put_buf(struct r1bio *r1_bio)
203 struct r1conf *conf = r1_bio->mddev->private;
206 for (i = 0; i < conf->raid_disks * 2; i++) {
207 struct bio *bio = r1_bio->bios[i];
209 rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev);
212 mempool_free(r1_bio, conf->r1buf_pool);
217 static void reschedule_retry(struct r1bio *r1_bio)
220 struct mddev *mddev = r1_bio->mddev;
221 struct r1conf *conf = mddev->private;
223 spin_lock_irqsave(&conf->device_lock, flags);
224 list_add(&r1_bio->retry_list, &conf->retry_list);
226 spin_unlock_irqrestore(&conf->device_lock, flags);
228 wake_up(&conf->wait_barrier);
229 md_wakeup_thread(mddev->thread);
233 * raid_end_bio_io() is called when we have finished servicing a mirrored
234 * operation and are ready to return a success/failure code to the buffer
237 static void call_bio_endio(struct r1bio *r1_bio)
239 struct bio *bio = r1_bio->master_bio;
241 struct r1conf *conf = r1_bio->mddev->private;
242 sector_t start_next_window = r1_bio->start_next_window;
243 sector_t bi_sector = bio->bi_iter.bi_sector;
245 if (bio->bi_phys_segments) {
247 spin_lock_irqsave(&conf->device_lock, flags);
248 bio->bi_phys_segments--;
249 done = (bio->bi_phys_segments == 0);
250 spin_unlock_irqrestore(&conf->device_lock, flags);
252 * make_request() might be waiting for
253 * bi_phys_segments to decrease
255 wake_up(&conf->wait_barrier);
259 if (!test_bit(R1BIO_Uptodate, &r1_bio->state))
260 bio->bi_error = -EIO;
265 * Wake up any possible resync thread that waits for the device
268 allow_barrier(conf, start_next_window, bi_sector);
272 static void raid_end_bio_io(struct r1bio *r1_bio)
274 struct bio *bio = r1_bio->master_bio;
276 /* if nobody has done the final endio yet, do it now */
277 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
278 pr_debug("raid1: sync end %s on sectors %llu-%llu\n",
279 (bio_data_dir(bio) == WRITE) ? "write" : "read",
280 (unsigned long long) bio->bi_iter.bi_sector,
281 (unsigned long long) bio_end_sector(bio) - 1);
283 call_bio_endio(r1_bio);
289 * Update disk head position estimator based on IRQ completion info.
291 static inline void update_head_pos(int disk, struct r1bio *r1_bio)
293 struct r1conf *conf = r1_bio->mddev->private;
295 conf->mirrors[disk].head_position =
296 r1_bio->sector + (r1_bio->sectors);
300 * Find the disk number which triggered given bio
302 static int find_bio_disk(struct r1bio *r1_bio, struct bio *bio)
305 struct r1conf *conf = r1_bio->mddev->private;
306 int raid_disks = conf->raid_disks;
308 for (mirror = 0; mirror < raid_disks * 2; mirror++)
309 if (r1_bio->bios[mirror] == bio)
312 BUG_ON(mirror == raid_disks * 2);
313 update_head_pos(mirror, r1_bio);
318 static void raid1_end_read_request(struct bio *bio)
320 int uptodate = !bio->bi_error;
321 struct r1bio *r1_bio = bio->bi_private;
322 struct r1conf *conf = r1_bio->mddev->private;
323 struct md_rdev *rdev = conf->mirrors[r1_bio->read_disk].rdev;
326 * this branch is our 'one mirror IO has finished' event handler:
328 update_head_pos(r1_bio->read_disk, r1_bio);
331 set_bit(R1BIO_Uptodate, &r1_bio->state);
332 else if (test_bit(FailFast, &rdev->flags) &&
333 test_bit(R1BIO_FailFast, &r1_bio->state))
334 /* This was a fail-fast read so we definitely
338 /* If all other devices have failed, we want to return
339 * the error upwards rather than fail the last device.
340 * Here we redefine "uptodate" to mean "Don't want to retry"
343 spin_lock_irqsave(&conf->device_lock, flags);
344 if (r1_bio->mddev->degraded == conf->raid_disks ||
345 (r1_bio->mddev->degraded == conf->raid_disks-1 &&
346 test_bit(In_sync, &rdev->flags)))
348 spin_unlock_irqrestore(&conf->device_lock, flags);
352 raid_end_bio_io(r1_bio);
353 rdev_dec_pending(rdev, conf->mddev);
358 char b[BDEVNAME_SIZE];
359 pr_err_ratelimited("md/raid1:%s: %s: rescheduling sector %llu\n",
361 bdevname(rdev->bdev, b),
362 (unsigned long long)r1_bio->sector);
363 set_bit(R1BIO_ReadError, &r1_bio->state);
364 reschedule_retry(r1_bio);
365 /* don't drop the reference on read_disk yet */
369 static void close_write(struct r1bio *r1_bio)
371 /* it really is the end of this request */
372 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
373 /* free extra copy of the data pages */
374 int i = r1_bio->behind_page_count;
376 safe_put_page(r1_bio->behind_bvecs[i].bv_page);
377 kfree(r1_bio->behind_bvecs);
378 r1_bio->behind_bvecs = NULL;
380 /* clear the bitmap if all writes complete successfully */
381 bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector,
383 !test_bit(R1BIO_Degraded, &r1_bio->state),
384 test_bit(R1BIO_BehindIO, &r1_bio->state));
385 md_write_end(r1_bio->mddev);
388 static void r1_bio_write_done(struct r1bio *r1_bio)
390 if (!atomic_dec_and_test(&r1_bio->remaining))
393 if (test_bit(R1BIO_WriteError, &r1_bio->state))
394 reschedule_retry(r1_bio);
397 if (test_bit(R1BIO_MadeGood, &r1_bio->state))
398 reschedule_retry(r1_bio);
400 raid_end_bio_io(r1_bio);
404 static void raid1_end_write_request(struct bio *bio)
406 struct r1bio *r1_bio = bio->bi_private;
407 int behind = test_bit(R1BIO_BehindIO, &r1_bio->state);
408 struct r1conf *conf = r1_bio->mddev->private;
409 struct bio *to_put = NULL;
410 int mirror = find_bio_disk(r1_bio, bio);
411 struct md_rdev *rdev = conf->mirrors[mirror].rdev;
414 discard_error = bio->bi_error && bio_op(bio) == REQ_OP_DISCARD;
417 * 'one mirror IO has finished' event handler:
419 if (bio->bi_error && !discard_error) {
420 set_bit(WriteErrorSeen, &rdev->flags);
421 if (!test_and_set_bit(WantReplacement, &rdev->flags))
422 set_bit(MD_RECOVERY_NEEDED, &
423 conf->mddev->recovery);
425 set_bit(R1BIO_WriteError, &r1_bio->state);
428 * Set R1BIO_Uptodate in our master bio, so that we
429 * will return a good error code for to the higher
430 * levels even if IO on some other mirrored buffer
433 * The 'master' represents the composite IO operation
434 * to user-side. So if something waits for IO, then it
435 * will wait for the 'master' bio.
440 r1_bio->bios[mirror] = NULL;
443 * Do not set R1BIO_Uptodate if the current device is
444 * rebuilding or Faulty. This is because we cannot use
445 * such device for properly reading the data back (we could
446 * potentially use it, if the current write would have felt
447 * before rdev->recovery_offset, but for simplicity we don't
450 if (test_bit(In_sync, &rdev->flags) &&
451 !test_bit(Faulty, &rdev->flags))
452 set_bit(R1BIO_Uptodate, &r1_bio->state);
454 /* Maybe we can clear some bad blocks. */
455 if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors,
456 &first_bad, &bad_sectors) && !discard_error) {
457 r1_bio->bios[mirror] = IO_MADE_GOOD;
458 set_bit(R1BIO_MadeGood, &r1_bio->state);
463 if (test_bit(WriteMostly, &rdev->flags))
464 atomic_dec(&r1_bio->behind_remaining);
467 * In behind mode, we ACK the master bio once the I/O
468 * has safely reached all non-writemostly
469 * disks. Setting the Returned bit ensures that this
470 * gets done only once -- we don't ever want to return
471 * -EIO here, instead we'll wait
473 if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) &&
474 test_bit(R1BIO_Uptodate, &r1_bio->state)) {
475 /* Maybe we can return now */
476 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
477 struct bio *mbio = r1_bio->master_bio;
478 pr_debug("raid1: behind end write sectors"
480 (unsigned long long) mbio->bi_iter.bi_sector,
481 (unsigned long long) bio_end_sector(mbio) - 1);
482 call_bio_endio(r1_bio);
486 if (r1_bio->bios[mirror] == NULL)
487 rdev_dec_pending(rdev, conf->mddev);
490 * Let's see if all mirrored write operations have finished
493 r1_bio_write_done(r1_bio);
500 * This routine returns the disk from which the requested read should
501 * be done. There is a per-array 'next expected sequential IO' sector
502 * number - if this matches on the next IO then we use the last disk.
503 * There is also a per-disk 'last know head position' sector that is
504 * maintained from IRQ contexts, both the normal and the resync IO
505 * completion handlers update this position correctly. If there is no
506 * perfect sequential match then we pick the disk whose head is closest.
508 * If there are 2 mirrors in the same 2 devices, performance degrades
509 * because position is mirror, not device based.
511 * The rdev for the device selected will have nr_pending incremented.
513 static int read_balance(struct r1conf *conf, struct r1bio *r1_bio, int *max_sectors)
515 const sector_t this_sector = r1_bio->sector;
517 int best_good_sectors;
518 int best_disk, best_dist_disk, best_pending_disk;
522 unsigned int min_pending;
523 struct md_rdev *rdev;
525 int choose_next_idle;
529 * Check if we can balance. We can balance on the whole
530 * device if no resync is going on, or below the resync window.
531 * We take the first readable disk when above the resync window.
534 sectors = r1_bio->sectors;
537 best_dist = MaxSector;
538 best_pending_disk = -1;
539 min_pending = UINT_MAX;
540 best_good_sectors = 0;
542 choose_next_idle = 0;
543 clear_bit(R1BIO_FailFast, &r1_bio->state);
545 if ((conf->mddev->recovery_cp < this_sector + sectors) ||
546 (mddev_is_clustered(conf->mddev) &&
547 md_cluster_ops->area_resyncing(conf->mddev, READ, this_sector,
548 this_sector + sectors)))
553 for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) {
557 unsigned int pending;
560 rdev = rcu_dereference(conf->mirrors[disk].rdev);
561 if (r1_bio->bios[disk] == IO_BLOCKED
563 || test_bit(Faulty, &rdev->flags))
565 if (!test_bit(In_sync, &rdev->flags) &&
566 rdev->recovery_offset < this_sector + sectors)
568 if (test_bit(WriteMostly, &rdev->flags)) {
569 /* Don't balance among write-mostly, just
570 * use the first as a last resort */
571 if (best_dist_disk < 0) {
572 if (is_badblock(rdev, this_sector, sectors,
573 &first_bad, &bad_sectors)) {
574 if (first_bad <= this_sector)
575 /* Cannot use this */
577 best_good_sectors = first_bad - this_sector;
579 best_good_sectors = sectors;
580 best_dist_disk = disk;
581 best_pending_disk = disk;
585 /* This is a reasonable device to use. It might
588 if (is_badblock(rdev, this_sector, sectors,
589 &first_bad, &bad_sectors)) {
590 if (best_dist < MaxSector)
591 /* already have a better device */
593 if (first_bad <= this_sector) {
594 /* cannot read here. If this is the 'primary'
595 * device, then we must not read beyond
596 * bad_sectors from another device..
598 bad_sectors -= (this_sector - first_bad);
599 if (choose_first && sectors > bad_sectors)
600 sectors = bad_sectors;
601 if (best_good_sectors > sectors)
602 best_good_sectors = sectors;
605 sector_t good_sectors = first_bad - this_sector;
606 if (good_sectors > best_good_sectors) {
607 best_good_sectors = good_sectors;
615 best_good_sectors = sectors;
618 /* At least two disks to choose from so failfast is OK */
619 set_bit(R1BIO_FailFast, &r1_bio->state);
621 nonrot = blk_queue_nonrot(bdev_get_queue(rdev->bdev));
622 has_nonrot_disk |= nonrot;
623 pending = atomic_read(&rdev->nr_pending);
624 dist = abs(this_sector - conf->mirrors[disk].head_position);
629 /* Don't change to another disk for sequential reads */
630 if (conf->mirrors[disk].next_seq_sect == this_sector
632 int opt_iosize = bdev_io_opt(rdev->bdev) >> 9;
633 struct raid1_info *mirror = &conf->mirrors[disk];
637 * If buffered sequential IO size exceeds optimal
638 * iosize, check if there is idle disk. If yes, choose
639 * the idle disk. read_balance could already choose an
640 * idle disk before noticing it's a sequential IO in
641 * this disk. This doesn't matter because this disk
642 * will idle, next time it will be utilized after the
643 * first disk has IO size exceeds optimal iosize. In
644 * this way, iosize of the first disk will be optimal
645 * iosize at least. iosize of the second disk might be
646 * small, but not a big deal since when the second disk
647 * starts IO, the first disk is likely still busy.
649 if (nonrot && opt_iosize > 0 &&
650 mirror->seq_start != MaxSector &&
651 mirror->next_seq_sect > opt_iosize &&
652 mirror->next_seq_sect - opt_iosize >=
654 choose_next_idle = 1;
660 if (choose_next_idle)
663 if (min_pending > pending) {
664 min_pending = pending;
665 best_pending_disk = disk;
668 if (dist < best_dist) {
670 best_dist_disk = disk;
675 * If all disks are rotational, choose the closest disk. If any disk is
676 * non-rotational, choose the disk with less pending request even the
677 * disk is rotational, which might/might not be optimal for raids with
678 * mixed ratation/non-rotational disks depending on workload.
680 if (best_disk == -1) {
681 if (has_nonrot_disk || min_pending == 0)
682 best_disk = best_pending_disk;
684 best_disk = best_dist_disk;
687 if (best_disk >= 0) {
688 rdev = rcu_dereference(conf->mirrors[best_disk].rdev);
691 atomic_inc(&rdev->nr_pending);
692 sectors = best_good_sectors;
694 if (conf->mirrors[best_disk].next_seq_sect != this_sector)
695 conf->mirrors[best_disk].seq_start = this_sector;
697 conf->mirrors[best_disk].next_seq_sect = this_sector + sectors;
700 *max_sectors = sectors;
705 static int raid1_congested(struct mddev *mddev, int bits)
707 struct r1conf *conf = mddev->private;
710 if ((bits & (1 << WB_async_congested)) &&
711 conf->pending_count >= max_queued_requests)
715 for (i = 0; i < conf->raid_disks * 2; i++) {
716 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
717 if (rdev && !test_bit(Faulty, &rdev->flags)) {
718 struct request_queue *q = bdev_get_queue(rdev->bdev);
722 /* Note the '|| 1' - when read_balance prefers
723 * non-congested targets, it can be removed
725 if ((bits & (1 << WB_async_congested)) || 1)
726 ret |= bdi_congested(&q->backing_dev_info, bits);
728 ret &= bdi_congested(&q->backing_dev_info, bits);
735 static void flush_pending_writes(struct r1conf *conf)
737 /* Any writes that have been queued but are awaiting
738 * bitmap updates get flushed here.
740 spin_lock_irq(&conf->device_lock);
742 if (conf->pending_bio_list.head) {
744 bio = bio_list_get(&conf->pending_bio_list);
745 conf->pending_count = 0;
746 spin_unlock_irq(&conf->device_lock);
747 /* flush any pending bitmap writes to
748 * disk before proceeding w/ I/O */
749 bitmap_unplug(conf->mddev->bitmap);
750 wake_up(&conf->wait_barrier);
752 while (bio) { /* submit pending writes */
753 struct bio *next = bio->bi_next;
754 struct md_rdev *rdev = (void*)bio->bi_bdev;
756 bio->bi_bdev = rdev->bdev;
757 if (test_bit(Faulty, &rdev->flags)) {
758 bio->bi_error = -EIO;
760 } else if (unlikely((bio_op(bio) == REQ_OP_DISCARD) &&
761 !blk_queue_discard(bdev_get_queue(bio->bi_bdev))))
765 generic_make_request(bio);
769 spin_unlock_irq(&conf->device_lock);
773 * Sometimes we need to suspend IO while we do something else,
774 * either some resync/recovery, or reconfigure the array.
775 * To do this we raise a 'barrier'.
776 * The 'barrier' is a counter that can be raised multiple times
777 * to count how many activities are happening which preclude
779 * We can only raise the barrier if there is no pending IO.
780 * i.e. if nr_pending == 0.
781 * We choose only to raise the barrier if no-one is waiting for the
782 * barrier to go down. This means that as soon as an IO request
783 * is ready, no other operations which require a barrier will start
784 * until the IO request has had a chance.
786 * So: regular IO calls 'wait_barrier'. When that returns there
787 * is no backgroup IO happening, It must arrange to call
788 * allow_barrier when it has finished its IO.
789 * backgroup IO calls must call raise_barrier. Once that returns
790 * there is no normal IO happeing. It must arrange to call
791 * lower_barrier when the particular background IO completes.
793 static void raise_barrier(struct r1conf *conf, sector_t sector_nr)
795 spin_lock_irq(&conf->resync_lock);
797 /* Wait until no block IO is waiting */
798 wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting,
801 /* block any new IO from starting */
803 conf->next_resync = sector_nr;
805 /* For these conditions we must wait:
806 * A: while the array is in frozen state
807 * B: while barrier >= RESYNC_DEPTH, meaning resync reach
808 * the max count which allowed.
809 * C: next_resync + RESYNC_SECTORS > start_next_window, meaning
810 * next resync will reach to the window which normal bios are
812 * D: while there are any active requests in the current window.
814 wait_event_lock_irq(conf->wait_barrier,
815 !conf->array_frozen &&
816 conf->barrier < RESYNC_DEPTH &&
817 conf->current_window_requests == 0 &&
818 (conf->start_next_window >=
819 conf->next_resync + RESYNC_SECTORS),
823 spin_unlock_irq(&conf->resync_lock);
826 static void lower_barrier(struct r1conf *conf)
829 BUG_ON(conf->barrier <= 0);
830 spin_lock_irqsave(&conf->resync_lock, flags);
833 spin_unlock_irqrestore(&conf->resync_lock, flags);
834 wake_up(&conf->wait_barrier);
837 static bool need_to_wait_for_sync(struct r1conf *conf, struct bio *bio)
841 if (conf->array_frozen || !bio)
843 else if (conf->barrier && bio_data_dir(bio) == WRITE) {
844 if ((conf->mddev->curr_resync_completed
845 >= bio_end_sector(bio)) ||
846 (conf->start_next_window + NEXT_NORMALIO_DISTANCE
847 <= bio->bi_iter.bi_sector))
856 static sector_t wait_barrier(struct r1conf *conf, struct bio *bio)
860 spin_lock_irq(&conf->resync_lock);
861 if (need_to_wait_for_sync(conf, bio)) {
863 /* Wait for the barrier to drop.
864 * However if there are already pending
865 * requests (preventing the barrier from
866 * rising completely), and the
867 * per-process bio queue isn't empty,
868 * then don't wait, as we need to empty
869 * that queue to allow conf->start_next_window
872 raid1_log(conf->mddev, "wait barrier");
873 wait_event_lock_irq(conf->wait_barrier,
874 !conf->array_frozen &&
876 ((conf->start_next_window <
877 conf->next_resync + RESYNC_SECTORS) &&
879 !bio_list_empty(current->bio_list))),
884 if (bio && bio_data_dir(bio) == WRITE) {
885 if (bio->bi_iter.bi_sector >= conf->next_resync) {
886 if (conf->start_next_window == MaxSector)
887 conf->start_next_window =
889 NEXT_NORMALIO_DISTANCE;
891 if ((conf->start_next_window + NEXT_NORMALIO_DISTANCE)
892 <= bio->bi_iter.bi_sector)
893 conf->next_window_requests++;
895 conf->current_window_requests++;
896 sector = conf->start_next_window;
901 spin_unlock_irq(&conf->resync_lock);
905 static void allow_barrier(struct r1conf *conf, sector_t start_next_window,
910 spin_lock_irqsave(&conf->resync_lock, flags);
912 if (start_next_window) {
913 if (start_next_window == conf->start_next_window) {
914 if (conf->start_next_window + NEXT_NORMALIO_DISTANCE
916 conf->next_window_requests--;
918 conf->current_window_requests--;
920 conf->current_window_requests--;
922 if (!conf->current_window_requests) {
923 if (conf->next_window_requests) {
924 conf->current_window_requests =
925 conf->next_window_requests;
926 conf->next_window_requests = 0;
927 conf->start_next_window +=
928 NEXT_NORMALIO_DISTANCE;
930 conf->start_next_window = MaxSector;
933 spin_unlock_irqrestore(&conf->resync_lock, flags);
934 wake_up(&conf->wait_barrier);
937 static void freeze_array(struct r1conf *conf, int extra)
939 /* stop syncio and normal IO and wait for everything to
941 * We wait until nr_pending match nr_queued+extra
942 * This is called in the context of one normal IO request
943 * that has failed. Thus any sync request that might be pending
944 * will be blocked by nr_pending, and we need to wait for
945 * pending IO requests to complete or be queued for re-try.
946 * Thus the number queued (nr_queued) plus this request (extra)
947 * must match the number of pending IOs (nr_pending) before
950 spin_lock_irq(&conf->resync_lock);
951 conf->array_frozen = 1;
952 raid1_log(conf->mddev, "wait freeze");
953 wait_event_lock_irq_cmd(conf->wait_barrier,
954 conf->nr_pending == conf->nr_queued+extra,
956 flush_pending_writes(conf));
957 spin_unlock_irq(&conf->resync_lock);
959 static void unfreeze_array(struct r1conf *conf)
961 /* reverse the effect of the freeze */
962 spin_lock_irq(&conf->resync_lock);
963 conf->array_frozen = 0;
964 wake_up(&conf->wait_barrier);
965 spin_unlock_irq(&conf->resync_lock);
968 /* duplicate the data pages for behind I/O
970 static void alloc_behind_pages(struct bio *bio, struct r1bio *r1_bio)
973 struct bio_vec *bvec;
974 struct bio_vec *bvecs = kzalloc(bio->bi_vcnt * sizeof(struct bio_vec),
976 if (unlikely(!bvecs))
979 bio_for_each_segment_all(bvec, bio, i) {
981 bvecs[i].bv_page = alloc_page(GFP_NOIO);
982 if (unlikely(!bvecs[i].bv_page))
984 memcpy(kmap(bvecs[i].bv_page) + bvec->bv_offset,
985 kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len);
986 kunmap(bvecs[i].bv_page);
987 kunmap(bvec->bv_page);
989 r1_bio->behind_bvecs = bvecs;
990 r1_bio->behind_page_count = bio->bi_vcnt;
991 set_bit(R1BIO_BehindIO, &r1_bio->state);
995 for (i = 0; i < bio->bi_vcnt; i++)
996 if (bvecs[i].bv_page)
997 put_page(bvecs[i].bv_page);
999 pr_debug("%dB behind alloc failed, doing sync I/O\n",
1000 bio->bi_iter.bi_size);
1003 struct raid1_plug_cb {
1004 struct blk_plug_cb cb;
1005 struct bio_list pending;
1009 static void raid1_unplug(struct blk_plug_cb *cb, bool from_schedule)
1011 struct raid1_plug_cb *plug = container_of(cb, struct raid1_plug_cb,
1013 struct mddev *mddev = plug->cb.data;
1014 struct r1conf *conf = mddev->private;
1017 if (from_schedule || current->bio_list) {
1018 spin_lock_irq(&conf->device_lock);
1019 bio_list_merge(&conf->pending_bio_list, &plug->pending);
1020 conf->pending_count += plug->pending_cnt;
1021 spin_unlock_irq(&conf->device_lock);
1022 wake_up(&conf->wait_barrier);
1023 md_wakeup_thread(mddev->thread);
1028 /* we aren't scheduling, so we can do the write-out directly. */
1029 bio = bio_list_get(&plug->pending);
1030 bitmap_unplug(mddev->bitmap);
1031 wake_up(&conf->wait_barrier);
1033 while (bio) { /* submit pending writes */
1034 struct bio *next = bio->bi_next;
1035 struct md_rdev *rdev = (void*)bio->bi_bdev;
1036 bio->bi_next = NULL;
1037 bio->bi_bdev = rdev->bdev;
1038 if (test_bit(Faulty, &rdev->flags)) {
1039 bio->bi_error = -EIO;
1041 } else if (unlikely((bio_op(bio) == REQ_OP_DISCARD) &&
1042 !blk_queue_discard(bdev_get_queue(bio->bi_bdev))))
1043 /* Just ignore it */
1046 generic_make_request(bio);
1052 static void raid1_make_request(struct mddev *mddev, struct bio * bio)
1054 struct r1conf *conf = mddev->private;
1055 struct raid1_info *mirror;
1056 struct r1bio *r1_bio;
1057 struct bio *read_bio;
1059 struct bitmap *bitmap;
1060 unsigned long flags;
1061 const int op = bio_op(bio);
1062 const int rw = bio_data_dir(bio);
1063 const unsigned long do_sync = (bio->bi_opf & REQ_SYNC);
1064 const unsigned long do_flush_fua = (bio->bi_opf &
1065 (REQ_PREFLUSH | REQ_FUA));
1066 struct md_rdev *blocked_rdev;
1067 struct blk_plug_cb *cb;
1068 struct raid1_plug_cb *plug = NULL;
1070 int sectors_handled;
1072 sector_t start_next_window;
1075 * Register the new request and wait if the reconstruction
1076 * thread has put up a bar for new requests.
1077 * Continue immediately if no resync is active currently.
1080 md_write_start(mddev, bio); /* wait on superblock update early */
1082 if (bio_data_dir(bio) == WRITE &&
1083 ((bio_end_sector(bio) > mddev->suspend_lo &&
1084 bio->bi_iter.bi_sector < mddev->suspend_hi) ||
1085 (mddev_is_clustered(mddev) &&
1086 md_cluster_ops->area_resyncing(mddev, WRITE,
1087 bio->bi_iter.bi_sector, bio_end_sector(bio))))) {
1088 /* As the suspend_* range is controlled by
1089 * userspace, we want an interruptible
1094 flush_signals(current);
1095 prepare_to_wait(&conf->wait_barrier,
1096 &w, TASK_INTERRUPTIBLE);
1097 if (bio_end_sector(bio) <= mddev->suspend_lo ||
1098 bio->bi_iter.bi_sector >= mddev->suspend_hi ||
1099 (mddev_is_clustered(mddev) &&
1100 !md_cluster_ops->area_resyncing(mddev, WRITE,
1101 bio->bi_iter.bi_sector, bio_end_sector(bio))))
1105 finish_wait(&conf->wait_barrier, &w);
1108 start_next_window = wait_barrier(conf, bio);
1110 bitmap = mddev->bitmap;
1113 * make_request() can abort the operation when read-ahead is being
1114 * used and no empty request is available.
1117 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
1119 r1_bio->master_bio = bio;
1120 r1_bio->sectors = bio_sectors(bio);
1122 r1_bio->mddev = mddev;
1123 r1_bio->sector = bio->bi_iter.bi_sector;
1125 /* We might need to issue multiple reads to different
1126 * devices if there are bad blocks around, so we keep
1127 * track of the number of reads in bio->bi_phys_segments.
1128 * If this is 0, there is only one r1_bio and no locking
1129 * will be needed when requests complete. If it is
1130 * non-zero, then it is the number of not-completed requests.
1132 bio->bi_phys_segments = 0;
1133 bio_clear_flag(bio, BIO_SEG_VALID);
1137 * read balancing logic:
1142 rdisk = read_balance(conf, r1_bio, &max_sectors);
1145 /* couldn't find anywhere to read from */
1146 raid_end_bio_io(r1_bio);
1149 mirror = conf->mirrors + rdisk;
1151 if (test_bit(WriteMostly, &mirror->rdev->flags) &&
1153 /* Reading from a write-mostly device must
1154 * take care not to over-take any writes
1157 raid1_log(mddev, "wait behind writes");
1158 wait_event(bitmap->behind_wait,
1159 atomic_read(&bitmap->behind_writes) == 0);
1161 r1_bio->read_disk = rdisk;
1162 r1_bio->start_next_window = 0;
1164 read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev);
1165 bio_trim(read_bio, r1_bio->sector - bio->bi_iter.bi_sector,
1168 r1_bio->bios[rdisk] = read_bio;
1170 read_bio->bi_iter.bi_sector = r1_bio->sector +
1171 mirror->rdev->data_offset;
1172 read_bio->bi_bdev = mirror->rdev->bdev;
1173 read_bio->bi_end_io = raid1_end_read_request;
1174 bio_set_op_attrs(read_bio, op, do_sync);
1175 if (test_bit(FailFast, &mirror->rdev->flags) &&
1176 test_bit(R1BIO_FailFast, &r1_bio->state))
1177 read_bio->bi_opf |= MD_FAILFAST;
1178 read_bio->bi_private = r1_bio;
1181 trace_block_bio_remap(bdev_get_queue(read_bio->bi_bdev),
1182 read_bio, disk_devt(mddev->gendisk),
1185 if (max_sectors < r1_bio->sectors) {
1186 /* could not read all from this device, so we will
1187 * need another r1_bio.
1190 sectors_handled = (r1_bio->sector + max_sectors
1191 - bio->bi_iter.bi_sector);
1192 r1_bio->sectors = max_sectors;
1193 spin_lock_irq(&conf->device_lock);
1194 if (bio->bi_phys_segments == 0)
1195 bio->bi_phys_segments = 2;
1197 bio->bi_phys_segments++;
1198 spin_unlock_irq(&conf->device_lock);
1199 /* Cannot call generic_make_request directly
1200 * as that will be queued in __make_request
1201 * and subsequent mempool_alloc might block waiting
1202 * for it. So hand bio over to raid1d.
1204 reschedule_retry(r1_bio);
1206 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
1208 r1_bio->master_bio = bio;
1209 r1_bio->sectors = bio_sectors(bio) - sectors_handled;
1211 r1_bio->mddev = mddev;
1212 r1_bio->sector = bio->bi_iter.bi_sector +
1216 generic_make_request(read_bio);
1223 if (conf->pending_count >= max_queued_requests) {
1224 md_wakeup_thread(mddev->thread);
1225 raid1_log(mddev, "wait queued");
1226 wait_event(conf->wait_barrier,
1227 conf->pending_count < max_queued_requests);
1229 /* first select target devices under rcu_lock and
1230 * inc refcount on their rdev. Record them by setting
1232 * If there are known/acknowledged bad blocks on any device on
1233 * which we have seen a write error, we want to avoid writing those
1235 * This potentially requires several writes to write around
1236 * the bad blocks. Each set of writes gets it's own r1bio
1237 * with a set of bios attached.
1240 disks = conf->raid_disks * 2;
1242 r1_bio->start_next_window = start_next_window;
1243 blocked_rdev = NULL;
1245 max_sectors = r1_bio->sectors;
1246 for (i = 0; i < disks; i++) {
1247 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
1248 if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
1249 atomic_inc(&rdev->nr_pending);
1250 blocked_rdev = rdev;
1253 r1_bio->bios[i] = NULL;
1254 if (!rdev || test_bit(Faulty, &rdev->flags)) {
1255 if (i < conf->raid_disks)
1256 set_bit(R1BIO_Degraded, &r1_bio->state);
1260 atomic_inc(&rdev->nr_pending);
1261 if (test_bit(WriteErrorSeen, &rdev->flags)) {
1266 is_bad = is_badblock(rdev, r1_bio->sector,
1268 &first_bad, &bad_sectors);
1270 /* mustn't write here until the bad block is
1272 set_bit(BlockedBadBlocks, &rdev->flags);
1273 blocked_rdev = rdev;
1276 if (is_bad && first_bad <= r1_bio->sector) {
1277 /* Cannot write here at all */
1278 bad_sectors -= (r1_bio->sector - first_bad);
1279 if (bad_sectors < max_sectors)
1280 /* mustn't write more than bad_sectors
1281 * to other devices yet
1283 max_sectors = bad_sectors;
1284 rdev_dec_pending(rdev, mddev);
1285 /* We don't set R1BIO_Degraded as that
1286 * only applies if the disk is
1287 * missing, so it might be re-added,
1288 * and we want to know to recover this
1290 * In this case the device is here,
1291 * and the fact that this chunk is not
1292 * in-sync is recorded in the bad
1298 int good_sectors = first_bad - r1_bio->sector;
1299 if (good_sectors < max_sectors)
1300 max_sectors = good_sectors;
1303 r1_bio->bios[i] = bio;
1307 if (unlikely(blocked_rdev)) {
1308 /* Wait for this device to become unblocked */
1310 sector_t old = start_next_window;
1312 for (j = 0; j < i; j++)
1313 if (r1_bio->bios[j])
1314 rdev_dec_pending(conf->mirrors[j].rdev, mddev);
1316 allow_barrier(conf, start_next_window, bio->bi_iter.bi_sector);
1317 raid1_log(mddev, "wait rdev %d blocked", blocked_rdev->raid_disk);
1318 md_wait_for_blocked_rdev(blocked_rdev, mddev);
1319 start_next_window = wait_barrier(conf, bio);
1321 * We must make sure the multi r1bios of bio have
1322 * the same value of bi_phys_segments
1324 if (bio->bi_phys_segments && old &&
1325 old != start_next_window)
1326 /* Wait for the former r1bio(s) to complete */
1327 wait_event(conf->wait_barrier,
1328 bio->bi_phys_segments == 1);
1332 if (max_sectors < r1_bio->sectors) {
1333 /* We are splitting this write into multiple parts, so
1334 * we need to prepare for allocating another r1_bio.
1336 r1_bio->sectors = max_sectors;
1337 spin_lock_irq(&conf->device_lock);
1338 if (bio->bi_phys_segments == 0)
1339 bio->bi_phys_segments = 2;
1341 bio->bi_phys_segments++;
1342 spin_unlock_irq(&conf->device_lock);
1344 sectors_handled = r1_bio->sector + max_sectors - bio->bi_iter.bi_sector;
1346 atomic_set(&r1_bio->remaining, 1);
1347 atomic_set(&r1_bio->behind_remaining, 0);
1350 for (i = 0; i < disks; i++) {
1352 if (!r1_bio->bios[i])
1355 mbio = bio_clone_mddev(bio, GFP_NOIO, mddev);
1356 bio_trim(mbio, r1_bio->sector - bio->bi_iter.bi_sector, max_sectors);
1360 * Not if there are too many, or cannot
1361 * allocate memory, or a reader on WriteMostly
1362 * is waiting for behind writes to flush */
1364 (atomic_read(&bitmap->behind_writes)
1365 < mddev->bitmap_info.max_write_behind) &&
1366 !waitqueue_active(&bitmap->behind_wait))
1367 alloc_behind_pages(mbio, r1_bio);
1369 bitmap_startwrite(bitmap, r1_bio->sector,
1371 test_bit(R1BIO_BehindIO,
1375 if (r1_bio->behind_bvecs) {
1376 struct bio_vec *bvec;
1380 * We trimmed the bio, so _all is legit
1382 bio_for_each_segment_all(bvec, mbio, j)
1383 bvec->bv_page = r1_bio->behind_bvecs[j].bv_page;
1384 if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags))
1385 atomic_inc(&r1_bio->behind_remaining);
1388 r1_bio->bios[i] = mbio;
1390 mbio->bi_iter.bi_sector = (r1_bio->sector +
1391 conf->mirrors[i].rdev->data_offset);
1392 mbio->bi_bdev = conf->mirrors[i].rdev->bdev;
1393 mbio->bi_end_io = raid1_end_write_request;
1394 bio_set_op_attrs(mbio, op, do_flush_fua | do_sync);
1395 mbio->bi_private = r1_bio;
1397 atomic_inc(&r1_bio->remaining);
1400 trace_block_bio_remap(bdev_get_queue(mbio->bi_bdev),
1401 mbio, disk_devt(mddev->gendisk),
1403 /* flush_pending_writes() needs access to the rdev so...*/
1404 mbio->bi_bdev = (void*)conf->mirrors[i].rdev;
1406 cb = blk_check_plugged(raid1_unplug, mddev, sizeof(*plug));
1408 plug = container_of(cb, struct raid1_plug_cb, cb);
1411 spin_lock_irqsave(&conf->device_lock, flags);
1413 bio_list_add(&plug->pending, mbio);
1414 plug->pending_cnt++;
1416 bio_list_add(&conf->pending_bio_list, mbio);
1417 conf->pending_count++;
1419 spin_unlock_irqrestore(&conf->device_lock, flags);
1421 md_wakeup_thread(mddev->thread);
1423 /* Mustn't call r1_bio_write_done before this next test,
1424 * as it could result in the bio being freed.
1426 if (sectors_handled < bio_sectors(bio)) {
1427 r1_bio_write_done(r1_bio);
1428 /* We need another r1_bio. It has already been counted
1429 * in bio->bi_phys_segments
1431 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
1432 r1_bio->master_bio = bio;
1433 r1_bio->sectors = bio_sectors(bio) - sectors_handled;
1435 r1_bio->mddev = mddev;
1436 r1_bio->sector = bio->bi_iter.bi_sector + sectors_handled;
1440 r1_bio_write_done(r1_bio);
1442 /* In case raid1d snuck in to freeze_array */
1443 wake_up(&conf->wait_barrier);
1446 static void raid1_status(struct seq_file *seq, struct mddev *mddev)
1448 struct r1conf *conf = mddev->private;
1451 seq_printf(seq, " [%d/%d] [", conf->raid_disks,
1452 conf->raid_disks - mddev->degraded);
1454 for (i = 0; i < conf->raid_disks; i++) {
1455 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
1456 seq_printf(seq, "%s",
1457 rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_");
1460 seq_printf(seq, "]");
1463 static void raid1_error(struct mddev *mddev, struct md_rdev *rdev)
1465 char b[BDEVNAME_SIZE];
1466 struct r1conf *conf = mddev->private;
1467 unsigned long flags;
1470 * If it is not operational, then we have already marked it as dead
1471 * else if it is the last working disks, ignore the error, let the
1472 * next level up know.
1473 * else mark the drive as failed
1475 spin_lock_irqsave(&conf->device_lock, flags);
1476 if (test_bit(In_sync, &rdev->flags)
1477 && (conf->raid_disks - mddev->degraded) == 1) {
1479 * Don't fail the drive, act as though we were just a
1480 * normal single drive.
1481 * However don't try a recovery from this drive as
1482 * it is very likely to fail.
1484 conf->recovery_disabled = mddev->recovery_disabled;
1485 spin_unlock_irqrestore(&conf->device_lock, flags);
1488 set_bit(Blocked, &rdev->flags);
1489 if (test_and_clear_bit(In_sync, &rdev->flags)) {
1491 set_bit(Faulty, &rdev->flags);
1493 set_bit(Faulty, &rdev->flags);
1494 spin_unlock_irqrestore(&conf->device_lock, flags);
1496 * if recovery is running, make sure it aborts.
1498 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1499 set_mask_bits(&mddev->flags, 0,
1500 BIT(MD_CHANGE_DEVS) | BIT(MD_CHANGE_PENDING));
1501 pr_crit("md/raid1:%s: Disk failure on %s, disabling device.\n"
1502 "md/raid1:%s: Operation continuing on %d devices.\n",
1503 mdname(mddev), bdevname(rdev->bdev, b),
1504 mdname(mddev), conf->raid_disks - mddev->degraded);
1507 static void print_conf(struct r1conf *conf)
1511 pr_debug("RAID1 conf printout:\n");
1513 pr_debug("(!conf)\n");
1516 pr_debug(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded,
1520 for (i = 0; i < conf->raid_disks; i++) {
1521 char b[BDEVNAME_SIZE];
1522 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
1524 pr_debug(" disk %d, wo:%d, o:%d, dev:%s\n",
1525 i, !test_bit(In_sync, &rdev->flags),
1526 !test_bit(Faulty, &rdev->flags),
1527 bdevname(rdev->bdev,b));
1532 static void close_sync(struct r1conf *conf)
1534 wait_barrier(conf, NULL);
1535 allow_barrier(conf, 0, 0);
1537 mempool_destroy(conf->r1buf_pool);
1538 conf->r1buf_pool = NULL;
1540 spin_lock_irq(&conf->resync_lock);
1541 conf->next_resync = MaxSector - 2 * NEXT_NORMALIO_DISTANCE;
1542 conf->start_next_window = MaxSector;
1543 conf->current_window_requests +=
1544 conf->next_window_requests;
1545 conf->next_window_requests = 0;
1546 spin_unlock_irq(&conf->resync_lock);
1549 static int raid1_spare_active(struct mddev *mddev)
1552 struct r1conf *conf = mddev->private;
1554 unsigned long flags;
1557 * Find all failed disks within the RAID1 configuration
1558 * and mark them readable.
1559 * Called under mddev lock, so rcu protection not needed.
1560 * device_lock used to avoid races with raid1_end_read_request
1561 * which expects 'In_sync' flags and ->degraded to be consistent.
1563 spin_lock_irqsave(&conf->device_lock, flags);
1564 for (i = 0; i < conf->raid_disks; i++) {
1565 struct md_rdev *rdev = conf->mirrors[i].rdev;
1566 struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev;
1568 && !test_bit(Candidate, &repl->flags)
1569 && repl->recovery_offset == MaxSector
1570 && !test_bit(Faulty, &repl->flags)
1571 && !test_and_set_bit(In_sync, &repl->flags)) {
1572 /* replacement has just become active */
1574 !test_and_clear_bit(In_sync, &rdev->flags))
1577 /* Replaced device not technically
1578 * faulty, but we need to be sure
1579 * it gets removed and never re-added
1581 set_bit(Faulty, &rdev->flags);
1582 sysfs_notify_dirent_safe(
1587 && rdev->recovery_offset == MaxSector
1588 && !test_bit(Faulty, &rdev->flags)
1589 && !test_and_set_bit(In_sync, &rdev->flags)) {
1591 sysfs_notify_dirent_safe(rdev->sysfs_state);
1594 mddev->degraded -= count;
1595 spin_unlock_irqrestore(&conf->device_lock, flags);
1601 static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev)
1603 struct r1conf *conf = mddev->private;
1606 struct raid1_info *p;
1608 int last = conf->raid_disks - 1;
1610 if (mddev->recovery_disabled == conf->recovery_disabled)
1613 if (md_integrity_add_rdev(rdev, mddev))
1616 if (rdev->raid_disk >= 0)
1617 first = last = rdev->raid_disk;
1620 * find the disk ... but prefer rdev->saved_raid_disk
1623 if (rdev->saved_raid_disk >= 0 &&
1624 rdev->saved_raid_disk >= first &&
1625 conf->mirrors[rdev->saved_raid_disk].rdev == NULL)
1626 first = last = rdev->saved_raid_disk;
1628 for (mirror = first; mirror <= last; mirror++) {
1629 p = conf->mirrors+mirror;
1633 disk_stack_limits(mddev->gendisk, rdev->bdev,
1634 rdev->data_offset << 9);
1636 p->head_position = 0;
1637 rdev->raid_disk = mirror;
1639 /* As all devices are equivalent, we don't need a full recovery
1640 * if this was recently any drive of the array
1642 if (rdev->saved_raid_disk < 0)
1644 rcu_assign_pointer(p->rdev, rdev);
1647 if (test_bit(WantReplacement, &p->rdev->flags) &&
1648 p[conf->raid_disks].rdev == NULL) {
1649 /* Add this device as a replacement */
1650 clear_bit(In_sync, &rdev->flags);
1651 set_bit(Replacement, &rdev->flags);
1652 rdev->raid_disk = mirror;
1655 rcu_assign_pointer(p[conf->raid_disks].rdev, rdev);
1659 if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev)))
1660 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
1665 static int raid1_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
1667 struct r1conf *conf = mddev->private;
1669 int number = rdev->raid_disk;
1670 struct raid1_info *p = conf->mirrors + number;
1672 if (rdev != p->rdev)
1673 p = conf->mirrors + conf->raid_disks + number;
1676 if (rdev == p->rdev) {
1677 if (test_bit(In_sync, &rdev->flags) ||
1678 atomic_read(&rdev->nr_pending)) {
1682 /* Only remove non-faulty devices if recovery
1685 if (!test_bit(Faulty, &rdev->flags) &&
1686 mddev->recovery_disabled != conf->recovery_disabled &&
1687 mddev->degraded < conf->raid_disks) {
1692 if (!test_bit(RemoveSynchronized, &rdev->flags)) {
1694 if (atomic_read(&rdev->nr_pending)) {
1695 /* lost the race, try later */
1701 if (conf->mirrors[conf->raid_disks + number].rdev) {
1702 /* We just removed a device that is being replaced.
1703 * Move down the replacement. We drain all IO before
1704 * doing this to avoid confusion.
1706 struct md_rdev *repl =
1707 conf->mirrors[conf->raid_disks + number].rdev;
1708 freeze_array(conf, 0);
1709 clear_bit(Replacement, &repl->flags);
1711 conf->mirrors[conf->raid_disks + number].rdev = NULL;
1712 unfreeze_array(conf);
1713 clear_bit(WantReplacement, &rdev->flags);
1715 clear_bit(WantReplacement, &rdev->flags);
1716 err = md_integrity_register(mddev);
1724 static void end_sync_read(struct bio *bio)
1726 struct r1bio *r1_bio = bio->bi_private;
1728 update_head_pos(r1_bio->read_disk, r1_bio);
1731 * we have read a block, now it needs to be re-written,
1732 * or re-read if the read failed.
1733 * We don't do much here, just schedule handling by raid1d
1736 set_bit(R1BIO_Uptodate, &r1_bio->state);
1738 if (atomic_dec_and_test(&r1_bio->remaining))
1739 reschedule_retry(r1_bio);
1742 static void end_sync_write(struct bio *bio)
1744 int uptodate = !bio->bi_error;
1745 struct r1bio *r1_bio = bio->bi_private;
1746 struct mddev *mddev = r1_bio->mddev;
1747 struct r1conf *conf = mddev->private;
1750 struct md_rdev *rdev = conf->mirrors[find_bio_disk(r1_bio, bio)].rdev;
1753 sector_t sync_blocks = 0;
1754 sector_t s = r1_bio->sector;
1755 long sectors_to_go = r1_bio->sectors;
1756 /* make sure these bits doesn't get cleared. */
1758 bitmap_end_sync(mddev->bitmap, s,
1761 sectors_to_go -= sync_blocks;
1762 } while (sectors_to_go > 0);
1763 set_bit(WriteErrorSeen, &rdev->flags);
1764 if (!test_and_set_bit(WantReplacement, &rdev->flags))
1765 set_bit(MD_RECOVERY_NEEDED, &
1767 set_bit(R1BIO_WriteError, &r1_bio->state);
1768 } else if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors,
1769 &first_bad, &bad_sectors) &&
1770 !is_badblock(conf->mirrors[r1_bio->read_disk].rdev,
1773 &first_bad, &bad_sectors)
1775 set_bit(R1BIO_MadeGood, &r1_bio->state);
1777 if (atomic_dec_and_test(&r1_bio->remaining)) {
1778 int s = r1_bio->sectors;
1779 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
1780 test_bit(R1BIO_WriteError, &r1_bio->state))
1781 reschedule_retry(r1_bio);
1784 md_done_sync(mddev, s, uptodate);
1789 static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector,
1790 int sectors, struct page *page, int rw)
1792 if (sync_page_io(rdev, sector, sectors << 9, page, rw, 0, false))
1796 set_bit(WriteErrorSeen, &rdev->flags);
1797 if (!test_and_set_bit(WantReplacement,
1799 set_bit(MD_RECOVERY_NEEDED, &
1800 rdev->mddev->recovery);
1802 /* need to record an error - either for the block or the device */
1803 if (!rdev_set_badblocks(rdev, sector, sectors, 0))
1804 md_error(rdev->mddev, rdev);
1808 static int fix_sync_read_error(struct r1bio *r1_bio)
1810 /* Try some synchronous reads of other devices to get
1811 * good data, much like with normal read errors. Only
1812 * read into the pages we already have so we don't
1813 * need to re-issue the read request.
1814 * We don't need to freeze the array, because being in an
1815 * active sync request, there is no normal IO, and
1816 * no overlapping syncs.
1817 * We don't need to check is_badblock() again as we
1818 * made sure that anything with a bad block in range
1819 * will have bi_end_io clear.
1821 struct mddev *mddev = r1_bio->mddev;
1822 struct r1conf *conf = mddev->private;
1823 struct bio *bio = r1_bio->bios[r1_bio->read_disk];
1824 sector_t sect = r1_bio->sector;
1825 int sectors = r1_bio->sectors;
1827 struct md_rdev *rdev;
1829 rdev = conf->mirrors[r1_bio->read_disk].rdev;
1830 if (test_bit(FailFast, &rdev->flags)) {
1831 /* Don't try recovering from here - just fail it
1832 * ... unless it is the last working device of course */
1833 md_error(mddev, rdev);
1834 if (test_bit(Faulty, &rdev->flags))
1835 /* Don't try to read from here, but make sure
1836 * put_buf does it's thing
1838 bio->bi_end_io = end_sync_write;
1843 int d = r1_bio->read_disk;
1847 if (s > (PAGE_SIZE>>9))
1850 if (r1_bio->bios[d]->bi_end_io == end_sync_read) {
1851 /* No rcu protection needed here devices
1852 * can only be removed when no resync is
1853 * active, and resync is currently active
1855 rdev = conf->mirrors[d].rdev;
1856 if (sync_page_io(rdev, sect, s<<9,
1857 bio->bi_io_vec[idx].bv_page,
1858 REQ_OP_READ, 0, false)) {
1864 if (d == conf->raid_disks * 2)
1866 } while (!success && d != r1_bio->read_disk);
1869 char b[BDEVNAME_SIZE];
1871 /* Cannot read from anywhere, this block is lost.
1872 * Record a bad block on each device. If that doesn't
1873 * work just disable and interrupt the recovery.
1874 * Don't fail devices as that won't really help.
1876 pr_crit_ratelimited("md/raid1:%s: %s: unrecoverable I/O read error for block %llu\n",
1878 bdevname(bio->bi_bdev, b),
1879 (unsigned long long)r1_bio->sector);
1880 for (d = 0; d < conf->raid_disks * 2; d++) {
1881 rdev = conf->mirrors[d].rdev;
1882 if (!rdev || test_bit(Faulty, &rdev->flags))
1884 if (!rdev_set_badblocks(rdev, sect, s, 0))
1888 conf->recovery_disabled =
1889 mddev->recovery_disabled;
1890 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1891 md_done_sync(mddev, r1_bio->sectors, 0);
1903 /* write it back and re-read */
1904 while (d != r1_bio->read_disk) {
1906 d = conf->raid_disks * 2;
1908 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1910 rdev = conf->mirrors[d].rdev;
1911 if (r1_sync_page_io(rdev, sect, s,
1912 bio->bi_io_vec[idx].bv_page,
1914 r1_bio->bios[d]->bi_end_io = NULL;
1915 rdev_dec_pending(rdev, mddev);
1919 while (d != r1_bio->read_disk) {
1921 d = conf->raid_disks * 2;
1923 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1925 rdev = conf->mirrors[d].rdev;
1926 if (r1_sync_page_io(rdev, sect, s,
1927 bio->bi_io_vec[idx].bv_page,
1929 atomic_add(s, &rdev->corrected_errors);
1935 set_bit(R1BIO_Uptodate, &r1_bio->state);
1940 static void process_checks(struct r1bio *r1_bio)
1942 /* We have read all readable devices. If we haven't
1943 * got the block, then there is no hope left.
1944 * If we have, then we want to do a comparison
1945 * and skip the write if everything is the same.
1946 * If any blocks failed to read, then we need to
1947 * attempt an over-write
1949 struct mddev *mddev = r1_bio->mddev;
1950 struct r1conf *conf = mddev->private;
1955 /* Fix variable parts of all bios */
1956 vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9);
1957 for (i = 0; i < conf->raid_disks * 2; i++) {
1961 struct bio *b = r1_bio->bios[i];
1962 if (b->bi_end_io != end_sync_read)
1964 /* fixup the bio for reuse, but preserve errno */
1965 error = b->bi_error;
1967 b->bi_error = error;
1969 b->bi_iter.bi_size = r1_bio->sectors << 9;
1970 b->bi_iter.bi_sector = r1_bio->sector +
1971 conf->mirrors[i].rdev->data_offset;
1972 b->bi_bdev = conf->mirrors[i].rdev->bdev;
1973 b->bi_end_io = end_sync_read;
1974 b->bi_private = r1_bio;
1976 size = b->bi_iter.bi_size;
1977 for (j = 0; j < vcnt ; j++) {
1979 bi = &b->bi_io_vec[j];
1981 if (size > PAGE_SIZE)
1982 bi->bv_len = PAGE_SIZE;
1988 for (primary = 0; primary < conf->raid_disks * 2; primary++)
1989 if (r1_bio->bios[primary]->bi_end_io == end_sync_read &&
1990 !r1_bio->bios[primary]->bi_error) {
1991 r1_bio->bios[primary]->bi_end_io = NULL;
1992 rdev_dec_pending(conf->mirrors[primary].rdev, mddev);
1995 r1_bio->read_disk = primary;
1996 for (i = 0; i < conf->raid_disks * 2; i++) {
1998 struct bio *pbio = r1_bio->bios[primary];
1999 struct bio *sbio = r1_bio->bios[i];
2000 int error = sbio->bi_error;
2002 if (sbio->bi_end_io != end_sync_read)
2004 /* Now we can 'fixup' the error value */
2008 for (j = vcnt; j-- ; ) {
2010 p = pbio->bi_io_vec[j].bv_page;
2011 s = sbio->bi_io_vec[j].bv_page;
2012 if (memcmp(page_address(p),
2014 sbio->bi_io_vec[j].bv_len))
2020 atomic64_add(r1_bio->sectors, &mddev->resync_mismatches);
2021 if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)
2023 /* No need to write to this device. */
2024 sbio->bi_end_io = NULL;
2025 rdev_dec_pending(conf->mirrors[i].rdev, mddev);
2029 bio_copy_data(sbio, pbio);
2033 static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio)
2035 struct r1conf *conf = mddev->private;
2037 int disks = conf->raid_disks * 2;
2038 struct bio *bio, *wbio;
2040 bio = r1_bio->bios[r1_bio->read_disk];
2042 if (!test_bit(R1BIO_Uptodate, &r1_bio->state))
2043 /* ouch - failed to read all of that. */
2044 if (!fix_sync_read_error(r1_bio))
2047 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2048 process_checks(r1_bio);
2053 atomic_set(&r1_bio->remaining, 1);
2054 for (i = 0; i < disks ; i++) {
2055 wbio = r1_bio->bios[i];
2056 if (wbio->bi_end_io == NULL ||
2057 (wbio->bi_end_io == end_sync_read &&
2058 (i == r1_bio->read_disk ||
2059 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery))))
2062 bio_set_op_attrs(wbio, REQ_OP_WRITE, 0);
2063 wbio->bi_end_io = end_sync_write;
2064 atomic_inc(&r1_bio->remaining);
2065 md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio));
2067 generic_make_request(wbio);
2070 if (atomic_dec_and_test(&r1_bio->remaining)) {
2071 /* if we're here, all write(s) have completed, so clean up */
2072 int s = r1_bio->sectors;
2073 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
2074 test_bit(R1BIO_WriteError, &r1_bio->state))
2075 reschedule_retry(r1_bio);
2078 md_done_sync(mddev, s, 1);
2084 * This is a kernel thread which:
2086 * 1. Retries failed read operations on working mirrors.
2087 * 2. Updates the raid superblock when problems encounter.
2088 * 3. Performs writes following reads for array synchronising.
2091 static void fix_read_error(struct r1conf *conf, int read_disk,
2092 sector_t sect, int sectors)
2094 struct mddev *mddev = conf->mddev;
2100 struct md_rdev *rdev;
2102 if (s > (PAGE_SIZE>>9))
2110 rdev = rcu_dereference(conf->mirrors[d].rdev);
2112 (test_bit(In_sync, &rdev->flags) ||
2113 (!test_bit(Faulty, &rdev->flags) &&
2114 rdev->recovery_offset >= sect + s)) &&
2115 is_badblock(rdev, sect, s,
2116 &first_bad, &bad_sectors) == 0) {
2117 atomic_inc(&rdev->nr_pending);
2119 if (sync_page_io(rdev, sect, s<<9,
2120 conf->tmppage, REQ_OP_READ, 0, false))
2122 rdev_dec_pending(rdev, mddev);
2128 if (d == conf->raid_disks * 2)
2130 } while (!success && d != read_disk);
2133 /* Cannot read from anywhere - mark it bad */
2134 struct md_rdev *rdev = conf->mirrors[read_disk].rdev;
2135 if (!rdev_set_badblocks(rdev, sect, s, 0))
2136 md_error(mddev, rdev);
2139 /* write it back and re-read */
2141 while (d != read_disk) {
2143 d = conf->raid_disks * 2;
2146 rdev = rcu_dereference(conf->mirrors[d].rdev);
2148 !test_bit(Faulty, &rdev->flags)) {
2149 atomic_inc(&rdev->nr_pending);
2151 r1_sync_page_io(rdev, sect, s,
2152 conf->tmppage, WRITE);
2153 rdev_dec_pending(rdev, mddev);
2158 while (d != read_disk) {
2159 char b[BDEVNAME_SIZE];
2161 d = conf->raid_disks * 2;
2164 rdev = rcu_dereference(conf->mirrors[d].rdev);
2166 !test_bit(Faulty, &rdev->flags)) {
2167 atomic_inc(&rdev->nr_pending);
2169 if (r1_sync_page_io(rdev, sect, s,
2170 conf->tmppage, READ)) {
2171 atomic_add(s, &rdev->corrected_errors);
2172 pr_info("md/raid1:%s: read error corrected (%d sectors at %llu on %s)\n",
2174 (unsigned long long)(sect +
2176 bdevname(rdev->bdev, b));
2178 rdev_dec_pending(rdev, mddev);
2187 static int narrow_write_error(struct r1bio *r1_bio, int i)
2189 struct mddev *mddev = r1_bio->mddev;
2190 struct r1conf *conf = mddev->private;
2191 struct md_rdev *rdev = conf->mirrors[i].rdev;
2193 /* bio has the data to be written to device 'i' where
2194 * we just recently had a write error.
2195 * We repeatedly clone the bio and trim down to one block,
2196 * then try the write. Where the write fails we record
2198 * It is conceivable that the bio doesn't exactly align with
2199 * blocks. We must handle this somehow.
2201 * We currently own a reference on the rdev.
2207 int sect_to_write = r1_bio->sectors;
2210 if (rdev->badblocks.shift < 0)
2213 block_sectors = roundup(1 << rdev->badblocks.shift,
2214 bdev_logical_block_size(rdev->bdev) >> 9);
2215 sector = r1_bio->sector;
2216 sectors = ((sector + block_sectors)
2217 & ~(sector_t)(block_sectors - 1))
2220 while (sect_to_write) {
2222 if (sectors > sect_to_write)
2223 sectors = sect_to_write;
2224 /* Write at 'sector' for 'sectors'*/
2226 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
2227 unsigned vcnt = r1_bio->behind_page_count;
2228 struct bio_vec *vec = r1_bio->behind_bvecs;
2230 while (!vec->bv_page) {
2235 wbio = bio_alloc_mddev(GFP_NOIO, vcnt, mddev);
2236 memcpy(wbio->bi_io_vec, vec, vcnt * sizeof(struct bio_vec));
2238 wbio->bi_vcnt = vcnt;
2240 wbio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev);
2243 bio_set_op_attrs(wbio, REQ_OP_WRITE, 0);
2244 wbio->bi_iter.bi_sector = r1_bio->sector;
2245 wbio->bi_iter.bi_size = r1_bio->sectors << 9;
2247 bio_trim(wbio, sector - r1_bio->sector, sectors);
2248 wbio->bi_iter.bi_sector += rdev->data_offset;
2249 wbio->bi_bdev = rdev->bdev;
2251 if (submit_bio_wait(wbio) < 0)
2253 ok = rdev_set_badblocks(rdev, sector,
2258 sect_to_write -= sectors;
2260 sectors = block_sectors;
2265 static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio)
2268 int s = r1_bio->sectors;
2269 for (m = 0; m < conf->raid_disks * 2 ; m++) {
2270 struct md_rdev *rdev = conf->mirrors[m].rdev;
2271 struct bio *bio = r1_bio->bios[m];
2272 if (bio->bi_end_io == NULL)
2274 if (!bio->bi_error &&
2275 test_bit(R1BIO_MadeGood, &r1_bio->state)) {
2276 rdev_clear_badblocks(rdev, r1_bio->sector, s, 0);
2278 if (bio->bi_error &&
2279 test_bit(R1BIO_WriteError, &r1_bio->state)) {
2280 if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0))
2281 md_error(conf->mddev, rdev);
2285 md_done_sync(conf->mddev, s, 1);
2288 static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio)
2292 for (m = 0; m < conf->raid_disks * 2 ; m++)
2293 if (r1_bio->bios[m] == IO_MADE_GOOD) {
2294 struct md_rdev *rdev = conf->mirrors[m].rdev;
2295 rdev_clear_badblocks(rdev,
2297 r1_bio->sectors, 0);
2298 rdev_dec_pending(rdev, conf->mddev);
2299 } else if (r1_bio->bios[m] != NULL) {
2300 /* This drive got a write error. We need to
2301 * narrow down and record precise write
2305 if (!narrow_write_error(r1_bio, m)) {
2306 md_error(conf->mddev,
2307 conf->mirrors[m].rdev);
2308 /* an I/O failed, we can't clear the bitmap */
2309 set_bit(R1BIO_Degraded, &r1_bio->state);
2311 rdev_dec_pending(conf->mirrors[m].rdev,
2315 spin_lock_irq(&conf->device_lock);
2316 list_add(&r1_bio->retry_list, &conf->bio_end_io_list);
2318 spin_unlock_irq(&conf->device_lock);
2319 md_wakeup_thread(conf->mddev->thread);
2321 if (test_bit(R1BIO_WriteError, &r1_bio->state))
2322 close_write(r1_bio);
2323 raid_end_bio_io(r1_bio);
2327 static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio)
2331 struct mddev *mddev = conf->mddev;
2333 char b[BDEVNAME_SIZE];
2334 struct md_rdev *rdev;
2336 sector_t bio_sector;
2338 clear_bit(R1BIO_ReadError, &r1_bio->state);
2339 /* we got a read error. Maybe the drive is bad. Maybe just
2340 * the block and we can fix it.
2341 * We freeze all other IO, and try reading the block from
2342 * other devices. When we find one, we re-write
2343 * and check it that fixes the read error.
2344 * This is all done synchronously while the array is
2348 bio = r1_bio->bios[r1_bio->read_disk];
2349 bdevname(bio->bi_bdev, b);
2350 bio_dev = bio->bi_bdev->bd_dev;
2351 bio_sector = conf->mirrors[r1_bio->read_disk].rdev->data_offset + r1_bio->sector;
2353 r1_bio->bios[r1_bio->read_disk] = NULL;
2355 rdev = conf->mirrors[r1_bio->read_disk].rdev;
2357 && !test_bit(FailFast, &rdev->flags)) {
2358 freeze_array(conf, 1);
2359 fix_read_error(conf, r1_bio->read_disk,
2360 r1_bio->sector, r1_bio->sectors);
2361 unfreeze_array(conf);
2363 r1_bio->bios[r1_bio->read_disk] = IO_BLOCKED;
2366 rdev_dec_pending(rdev, conf->mddev);
2369 disk = read_balance(conf, r1_bio, &max_sectors);
2371 pr_crit_ratelimited("md/raid1:%s: %s: unrecoverable I/O read error for block %llu\n",
2372 mdname(mddev), b, (unsigned long long)r1_bio->sector);
2373 raid_end_bio_io(r1_bio);
2375 const unsigned long do_sync
2376 = r1_bio->master_bio->bi_opf & REQ_SYNC;
2377 r1_bio->read_disk = disk;
2378 bio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev);
2379 bio_trim(bio, r1_bio->sector - bio->bi_iter.bi_sector,
2381 r1_bio->bios[r1_bio->read_disk] = bio;
2382 rdev = conf->mirrors[disk].rdev;
2383 pr_info_ratelimited("md/raid1:%s: redirecting sector %llu to other mirror: %s\n",
2385 (unsigned long long)r1_bio->sector,
2386 bdevname(rdev->bdev, b));
2387 bio->bi_iter.bi_sector = r1_bio->sector + rdev->data_offset;
2388 bio->bi_bdev = rdev->bdev;
2389 bio->bi_end_io = raid1_end_read_request;
2390 bio_set_op_attrs(bio, REQ_OP_READ, do_sync);
2391 if (test_bit(FailFast, &rdev->flags) &&
2392 test_bit(R1BIO_FailFast, &r1_bio->state))
2393 bio->bi_opf |= MD_FAILFAST;
2394 bio->bi_private = r1_bio;
2395 if (max_sectors < r1_bio->sectors) {
2396 /* Drat - have to split this up more */
2397 struct bio *mbio = r1_bio->master_bio;
2398 int sectors_handled = (r1_bio->sector + max_sectors
2399 - mbio->bi_iter.bi_sector);
2400 r1_bio->sectors = max_sectors;
2401 spin_lock_irq(&conf->device_lock);
2402 if (mbio->bi_phys_segments == 0)
2403 mbio->bi_phys_segments = 2;
2405 mbio->bi_phys_segments++;
2406 spin_unlock_irq(&conf->device_lock);
2407 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev),
2408 bio, bio_dev, bio_sector);
2409 generic_make_request(bio);
2412 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
2414 r1_bio->master_bio = mbio;
2415 r1_bio->sectors = bio_sectors(mbio) - sectors_handled;
2417 set_bit(R1BIO_ReadError, &r1_bio->state);
2418 r1_bio->mddev = mddev;
2419 r1_bio->sector = mbio->bi_iter.bi_sector +
2424 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev),
2425 bio, bio_dev, bio_sector);
2426 generic_make_request(bio);
2431 static void raid1d(struct md_thread *thread)
2433 struct mddev *mddev = thread->mddev;
2434 struct r1bio *r1_bio;
2435 unsigned long flags;
2436 struct r1conf *conf = mddev->private;
2437 struct list_head *head = &conf->retry_list;
2438 struct blk_plug plug;
2440 md_check_recovery(mddev);
2442 if (!list_empty_careful(&conf->bio_end_io_list) &&
2443 !test_bit(MD_CHANGE_PENDING, &mddev->flags)) {
2445 spin_lock_irqsave(&conf->device_lock, flags);
2446 if (!test_bit(MD_CHANGE_PENDING, &mddev->flags)) {
2447 while (!list_empty(&conf->bio_end_io_list)) {
2448 list_move(conf->bio_end_io_list.prev, &tmp);
2452 spin_unlock_irqrestore(&conf->device_lock, flags);
2453 while (!list_empty(&tmp)) {
2454 r1_bio = list_first_entry(&tmp, struct r1bio,
2456 list_del(&r1_bio->retry_list);
2457 if (mddev->degraded)
2458 set_bit(R1BIO_Degraded, &r1_bio->state);
2459 if (test_bit(R1BIO_WriteError, &r1_bio->state))
2460 close_write(r1_bio);
2461 raid_end_bio_io(r1_bio);
2465 blk_start_plug(&plug);
2468 flush_pending_writes(conf);
2470 spin_lock_irqsave(&conf->device_lock, flags);
2471 if (list_empty(head)) {
2472 spin_unlock_irqrestore(&conf->device_lock, flags);
2475 r1_bio = list_entry(head->prev, struct r1bio, retry_list);
2476 list_del(head->prev);
2478 spin_unlock_irqrestore(&conf->device_lock, flags);
2480 mddev = r1_bio->mddev;
2481 conf = mddev->private;
2482 if (test_bit(R1BIO_IsSync, &r1_bio->state)) {
2483 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
2484 test_bit(R1BIO_WriteError, &r1_bio->state))
2485 handle_sync_write_finished(conf, r1_bio);
2487 sync_request_write(mddev, r1_bio);
2488 } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
2489 test_bit(R1BIO_WriteError, &r1_bio->state))
2490 handle_write_finished(conf, r1_bio);
2491 else if (test_bit(R1BIO_ReadError, &r1_bio->state))
2492 handle_read_error(conf, r1_bio);
2494 /* just a partial read to be scheduled from separate
2497 generic_make_request(r1_bio->bios[r1_bio->read_disk]);
2500 if (mddev->flags & ~(1<<MD_CHANGE_PENDING))
2501 md_check_recovery(mddev);
2503 blk_finish_plug(&plug);
2506 static int init_resync(struct r1conf *conf)
2510 buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;
2511 BUG_ON(conf->r1buf_pool);
2512 conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free,
2514 if (!conf->r1buf_pool)
2516 conf->next_resync = 0;
2521 * perform a "sync" on one "block"
2523 * We need to make sure that no normal I/O request - particularly write
2524 * requests - conflict with active sync requests.
2526 * This is achieved by tracking pending requests and a 'barrier' concept
2527 * that can be installed to exclude normal IO requests.
2530 static sector_t raid1_sync_request(struct mddev *mddev, sector_t sector_nr,
2533 struct r1conf *conf = mddev->private;
2534 struct r1bio *r1_bio;
2536 sector_t max_sector, nr_sectors;
2540 int write_targets = 0, read_targets = 0;
2541 sector_t sync_blocks;
2542 int still_degraded = 0;
2543 int good_sectors = RESYNC_SECTORS;
2544 int min_bad = 0; /* number of sectors that are bad in all devices */
2546 if (!conf->r1buf_pool)
2547 if (init_resync(conf))
2550 max_sector = mddev->dev_sectors;
2551 if (sector_nr >= max_sector) {
2552 /* If we aborted, we need to abort the
2553 * sync on the 'current' bitmap chunk (there will
2554 * only be one in raid1 resync.
2555 * We can find the current addess in mddev->curr_resync
2557 if (mddev->curr_resync < max_sector) /* aborted */
2558 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
2560 else /* completed sync */
2563 bitmap_close_sync(mddev->bitmap);
2566 if (mddev_is_clustered(mddev)) {
2567 conf->cluster_sync_low = 0;
2568 conf->cluster_sync_high = 0;
2573 if (mddev->bitmap == NULL &&
2574 mddev->recovery_cp == MaxSector &&
2575 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
2576 conf->fullsync == 0) {
2578 return max_sector - sector_nr;
2580 /* before building a request, check if we can skip these blocks..
2581 * This call the bitmap_start_sync doesn't actually record anything
2583 if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
2584 !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
2585 /* We can skip this block, and probably several more */
2591 * If there is non-resync activity waiting for a turn, then let it
2592 * though before starting on this new sync request.
2594 if (conf->nr_waiting)
2595 schedule_timeout_uninterruptible(1);
2597 /* we are incrementing sector_nr below. To be safe, we check against
2598 * sector_nr + two times RESYNC_SECTORS
2601 bitmap_cond_end_sync(mddev->bitmap, sector_nr,
2602 mddev_is_clustered(mddev) && (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high));
2603 r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO);
2605 raise_barrier(conf, sector_nr);
2609 * If we get a correctably read error during resync or recovery,
2610 * we might want to read from a different device. So we
2611 * flag all drives that could conceivably be read from for READ,
2612 * and any others (which will be non-In_sync devices) for WRITE.
2613 * If a read fails, we try reading from something else for which READ
2617 r1_bio->mddev = mddev;
2618 r1_bio->sector = sector_nr;
2620 set_bit(R1BIO_IsSync, &r1_bio->state);
2622 for (i = 0; i < conf->raid_disks * 2; i++) {
2623 struct md_rdev *rdev;
2624 bio = r1_bio->bios[i];
2627 rdev = rcu_dereference(conf->mirrors[i].rdev);
2629 test_bit(Faulty, &rdev->flags)) {
2630 if (i < conf->raid_disks)
2632 } else if (!test_bit(In_sync, &rdev->flags)) {
2633 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
2634 bio->bi_end_io = end_sync_write;
2637 /* may need to read from here */
2638 sector_t first_bad = MaxSector;
2641 if (is_badblock(rdev, sector_nr, good_sectors,
2642 &first_bad, &bad_sectors)) {
2643 if (first_bad > sector_nr)
2644 good_sectors = first_bad - sector_nr;
2646 bad_sectors -= (sector_nr - first_bad);
2648 min_bad > bad_sectors)
2649 min_bad = bad_sectors;
2652 if (sector_nr < first_bad) {
2653 if (test_bit(WriteMostly, &rdev->flags)) {
2660 bio_set_op_attrs(bio, REQ_OP_READ, 0);
2661 bio->bi_end_io = end_sync_read;
2663 } else if (!test_bit(WriteErrorSeen, &rdev->flags) &&
2664 test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
2665 !test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
2667 * The device is suitable for reading (InSync),
2668 * but has bad block(s) here. Let's try to correct them,
2669 * if we are doing resync or repair. Otherwise, leave
2670 * this device alone for this sync request.
2672 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
2673 bio->bi_end_io = end_sync_write;
2677 if (bio->bi_end_io) {
2678 atomic_inc(&rdev->nr_pending);
2679 bio->bi_iter.bi_sector = sector_nr + rdev->data_offset;
2680 bio->bi_bdev = rdev->bdev;
2681 bio->bi_private = r1_bio;
2682 if (test_bit(FailFast, &rdev->flags))
2683 bio->bi_opf |= MD_FAILFAST;
2689 r1_bio->read_disk = disk;
2691 if (read_targets == 0 && min_bad > 0) {
2692 /* These sectors are bad on all InSync devices, so we
2693 * need to mark them bad on all write targets
2696 for (i = 0 ; i < conf->raid_disks * 2 ; i++)
2697 if (r1_bio->bios[i]->bi_end_io == end_sync_write) {
2698 struct md_rdev *rdev = conf->mirrors[i].rdev;
2699 ok = rdev_set_badblocks(rdev, sector_nr,
2703 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2708 /* Cannot record the badblocks, so need to
2710 * If there are multiple read targets, could just
2711 * fail the really bad ones ???
2713 conf->recovery_disabled = mddev->recovery_disabled;
2714 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2720 if (min_bad > 0 && min_bad < good_sectors) {
2721 /* only resync enough to reach the next bad->good
2723 good_sectors = min_bad;
2726 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0)
2727 /* extra read targets are also write targets */
2728 write_targets += read_targets-1;
2730 if (write_targets == 0 || read_targets == 0) {
2731 /* There is nowhere to write, so all non-sync
2732 * drives must be failed - so we are finished
2736 max_sector = sector_nr + min_bad;
2737 rv = max_sector - sector_nr;
2743 if (max_sector > mddev->resync_max)
2744 max_sector = mddev->resync_max; /* Don't do IO beyond here */
2745 if (max_sector > sector_nr + good_sectors)
2746 max_sector = sector_nr + good_sectors;
2751 int len = PAGE_SIZE;
2752 if (sector_nr + (len>>9) > max_sector)
2753 len = (max_sector - sector_nr) << 9;
2756 if (sync_blocks == 0) {
2757 if (!bitmap_start_sync(mddev->bitmap, sector_nr,
2758 &sync_blocks, still_degraded) &&
2760 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2762 if ((len >> 9) > sync_blocks)
2763 len = sync_blocks<<9;
2766 for (i = 0 ; i < conf->raid_disks * 2; i++) {
2767 bio = r1_bio->bios[i];
2768 if (bio->bi_end_io) {
2769 page = bio->bi_io_vec[bio->bi_vcnt].bv_page;
2770 if (bio_add_page(bio, page, len, 0) == 0) {
2772 bio->bi_io_vec[bio->bi_vcnt].bv_page = page;
2775 bio = r1_bio->bios[i];
2776 if (bio->bi_end_io==NULL)
2778 /* remove last page from this bio */
2780 bio->bi_iter.bi_size -= len;
2781 bio_clear_flag(bio, BIO_SEG_VALID);
2787 nr_sectors += len>>9;
2788 sector_nr += len>>9;
2789 sync_blocks -= (len>>9);
2790 } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES);
2792 r1_bio->sectors = nr_sectors;
2794 if (mddev_is_clustered(mddev) &&
2795 conf->cluster_sync_high < sector_nr + nr_sectors) {
2796 conf->cluster_sync_low = mddev->curr_resync_completed;
2797 conf->cluster_sync_high = conf->cluster_sync_low + CLUSTER_RESYNC_WINDOW_SECTORS;
2798 /* Send resync message */
2799 md_cluster_ops->resync_info_update(mddev,
2800 conf->cluster_sync_low,
2801 conf->cluster_sync_high);
2804 /* For a user-requested sync, we read all readable devices and do a
2807 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
2808 atomic_set(&r1_bio->remaining, read_targets);
2809 for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) {
2810 bio = r1_bio->bios[i];
2811 if (bio->bi_end_io == end_sync_read) {
2813 md_sync_acct(bio->bi_bdev, nr_sectors);
2814 if (read_targets == 1)
2815 bio->bi_opf &= ~MD_FAILFAST;
2816 generic_make_request(bio);
2820 atomic_set(&r1_bio->remaining, 1);
2821 bio = r1_bio->bios[r1_bio->read_disk];
2822 md_sync_acct(bio->bi_bdev, nr_sectors);
2823 if (read_targets == 1)
2824 bio->bi_opf &= ~MD_FAILFAST;
2825 generic_make_request(bio);
2831 static sector_t raid1_size(struct mddev *mddev, sector_t sectors, int raid_disks)
2836 return mddev->dev_sectors;
2839 static struct r1conf *setup_conf(struct mddev *mddev)
2841 struct r1conf *conf;
2843 struct raid1_info *disk;
2844 struct md_rdev *rdev;
2847 conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL);
2851 conf->mirrors = kzalloc(sizeof(struct raid1_info)
2852 * mddev->raid_disks * 2,
2857 conf->tmppage = alloc_page(GFP_KERNEL);
2861 conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL);
2862 if (!conf->poolinfo)
2864 conf->poolinfo->raid_disks = mddev->raid_disks * 2;
2865 conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
2868 if (!conf->r1bio_pool)
2871 conf->poolinfo->mddev = mddev;
2874 spin_lock_init(&conf->device_lock);
2875 rdev_for_each(rdev, mddev) {
2876 struct request_queue *q;
2877 int disk_idx = rdev->raid_disk;
2878 if (disk_idx >= mddev->raid_disks
2881 if (test_bit(Replacement, &rdev->flags))
2882 disk = conf->mirrors + mddev->raid_disks + disk_idx;
2884 disk = conf->mirrors + disk_idx;
2889 q = bdev_get_queue(rdev->bdev);
2891 disk->head_position = 0;
2892 disk->seq_start = MaxSector;
2894 conf->raid_disks = mddev->raid_disks;
2895 conf->mddev = mddev;
2896 INIT_LIST_HEAD(&conf->retry_list);
2897 INIT_LIST_HEAD(&conf->bio_end_io_list);
2899 spin_lock_init(&conf->resync_lock);
2900 init_waitqueue_head(&conf->wait_barrier);
2902 bio_list_init(&conf->pending_bio_list);
2903 conf->pending_count = 0;
2904 conf->recovery_disabled = mddev->recovery_disabled - 1;
2906 conf->start_next_window = MaxSector;
2907 conf->current_window_requests = conf->next_window_requests = 0;
2910 for (i = 0; i < conf->raid_disks * 2; i++) {
2912 disk = conf->mirrors + i;
2914 if (i < conf->raid_disks &&
2915 disk[conf->raid_disks].rdev) {
2916 /* This slot has a replacement. */
2918 /* No original, just make the replacement
2919 * a recovering spare
2922 disk[conf->raid_disks].rdev;
2923 disk[conf->raid_disks].rdev = NULL;
2924 } else if (!test_bit(In_sync, &disk->rdev->flags))
2925 /* Original is not in_sync - bad */
2930 !test_bit(In_sync, &disk->rdev->flags)) {
2931 disk->head_position = 0;
2933 (disk->rdev->saved_raid_disk < 0))
2939 conf->thread = md_register_thread(raid1d, mddev, "raid1");
2947 mempool_destroy(conf->r1bio_pool);
2948 kfree(conf->mirrors);
2949 safe_put_page(conf->tmppage);
2950 kfree(conf->poolinfo);
2953 return ERR_PTR(err);
2956 static void raid1_free(struct mddev *mddev, void *priv);
2957 static int raid1_run(struct mddev *mddev)
2959 struct r1conf *conf;
2961 struct md_rdev *rdev;
2963 bool discard_supported = false;
2965 if (mddev->level != 1) {
2966 pr_warn("md/raid1:%s: raid level not set to mirroring (%d)\n",
2967 mdname(mddev), mddev->level);
2970 if (mddev->reshape_position != MaxSector) {
2971 pr_warn("md/raid1:%s: reshape_position set but not supported\n",
2976 * copy the already verified devices into our private RAID1
2977 * bookkeeping area. [whatever we allocate in run(),
2978 * should be freed in raid1_free()]
2980 if (mddev->private == NULL)
2981 conf = setup_conf(mddev);
2983 conf = mddev->private;
2986 return PTR_ERR(conf);
2989 blk_queue_max_write_same_sectors(mddev->queue, 0);
2991 rdev_for_each(rdev, mddev) {
2992 if (!mddev->gendisk)
2994 disk_stack_limits(mddev->gendisk, rdev->bdev,
2995 rdev->data_offset << 9);
2996 if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
2997 discard_supported = true;
3000 mddev->degraded = 0;
3001 for (i=0; i < conf->raid_disks; i++)
3002 if (conf->mirrors[i].rdev == NULL ||
3003 !test_bit(In_sync, &conf->mirrors[i].rdev->flags) ||
3004 test_bit(Faulty, &conf->mirrors[i].rdev->flags))
3007 if (conf->raid_disks - mddev->degraded == 1)
3008 mddev->recovery_cp = MaxSector;
3010 if (mddev->recovery_cp != MaxSector)
3011 pr_info("md/raid1:%s: not clean -- starting background reconstruction\n",
3013 pr_info("md/raid1:%s: active with %d out of %d mirrors\n",
3014 mdname(mddev), mddev->raid_disks - mddev->degraded,
3018 * Ok, everything is just fine now
3020 mddev->thread = conf->thread;
3021 conf->thread = NULL;
3022 mddev->private = conf;
3023 set_bit(MD_FAILFAST_SUPPORTED, &mddev->flags);
3025 md_set_array_sectors(mddev, raid1_size(mddev, 0, 0));
3028 if (discard_supported)
3029 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD,
3032 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD,
3036 ret = md_integrity_register(mddev);
3038 md_unregister_thread(&mddev->thread);
3039 raid1_free(mddev, conf);
3044 static void raid1_free(struct mddev *mddev, void *priv)
3046 struct r1conf *conf = priv;
3048 mempool_destroy(conf->r1bio_pool);
3049 kfree(conf->mirrors);
3050 safe_put_page(conf->tmppage);
3051 kfree(conf->poolinfo);
3055 static int raid1_resize(struct mddev *mddev, sector_t sectors)
3057 /* no resync is happening, and there is enough space
3058 * on all devices, so we can resize.
3059 * We need to make sure resync covers any new space.
3060 * If the array is shrinking we should possibly wait until
3061 * any io in the removed space completes, but it hardly seems
3064 sector_t newsize = raid1_size(mddev, sectors, 0);
3065 if (mddev->external_size &&
3066 mddev->array_sectors > newsize)
3068 if (mddev->bitmap) {
3069 int ret = bitmap_resize(mddev->bitmap, newsize, 0, 0);
3073 md_set_array_sectors(mddev, newsize);
3074 set_capacity(mddev->gendisk, mddev->array_sectors);
3075 revalidate_disk(mddev->gendisk);
3076 if (sectors > mddev->dev_sectors &&
3077 mddev->recovery_cp > mddev->dev_sectors) {
3078 mddev->recovery_cp = mddev->dev_sectors;
3079 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3081 mddev->dev_sectors = sectors;
3082 mddev->resync_max_sectors = sectors;
3086 static int raid1_reshape(struct mddev *mddev)
3089 * 1/ resize the r1bio_pool
3090 * 2/ resize conf->mirrors
3092 * We allocate a new r1bio_pool if we can.
3093 * Then raise a device barrier and wait until all IO stops.
3094 * Then resize conf->mirrors and swap in the new r1bio pool.
3096 * At the same time, we "pack" the devices so that all the missing
3097 * devices have the higher raid_disk numbers.
3099 mempool_t *newpool, *oldpool;
3100 struct pool_info *newpoolinfo;
3101 struct raid1_info *newmirrors;
3102 struct r1conf *conf = mddev->private;
3103 int cnt, raid_disks;
3104 unsigned long flags;
3107 /* Cannot change chunk_size, layout, or level */
3108 if (mddev->chunk_sectors != mddev->new_chunk_sectors ||
3109 mddev->layout != mddev->new_layout ||
3110 mddev->level != mddev->new_level) {
3111 mddev->new_chunk_sectors = mddev->chunk_sectors;
3112 mddev->new_layout = mddev->layout;
3113 mddev->new_level = mddev->level;
3117 if (!mddev_is_clustered(mddev)) {
3118 err = md_allow_write(mddev);
3123 raid_disks = mddev->raid_disks + mddev->delta_disks;
3125 if (raid_disks < conf->raid_disks) {
3127 for (d= 0; d < conf->raid_disks; d++)
3128 if (conf->mirrors[d].rdev)
3130 if (cnt > raid_disks)
3134 newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL);
3137 newpoolinfo->mddev = mddev;
3138 newpoolinfo->raid_disks = raid_disks * 2;
3140 newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
3141 r1bio_pool_free, newpoolinfo);
3146 newmirrors = kzalloc(sizeof(struct raid1_info) * raid_disks * 2,
3150 mempool_destroy(newpool);
3154 freeze_array(conf, 0);
3156 /* ok, everything is stopped */
3157 oldpool = conf->r1bio_pool;
3158 conf->r1bio_pool = newpool;
3160 for (d = d2 = 0; d < conf->raid_disks; d++) {
3161 struct md_rdev *rdev = conf->mirrors[d].rdev;
3162 if (rdev && rdev->raid_disk != d2) {
3163 sysfs_unlink_rdev(mddev, rdev);
3164 rdev->raid_disk = d2;
3165 sysfs_unlink_rdev(mddev, rdev);
3166 if (sysfs_link_rdev(mddev, rdev))
3167 pr_warn("md/raid1:%s: cannot register rd%d\n",
3168 mdname(mddev), rdev->raid_disk);
3171 newmirrors[d2++].rdev = rdev;
3173 kfree(conf->mirrors);
3174 conf->mirrors = newmirrors;
3175 kfree(conf->poolinfo);
3176 conf->poolinfo = newpoolinfo;
3178 spin_lock_irqsave(&conf->device_lock, flags);
3179 mddev->degraded += (raid_disks - conf->raid_disks);
3180 spin_unlock_irqrestore(&conf->device_lock, flags);
3181 conf->raid_disks = mddev->raid_disks = raid_disks;
3182 mddev->delta_disks = 0;
3184 unfreeze_array(conf);
3186 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3187 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3188 md_wakeup_thread(mddev->thread);
3190 mempool_destroy(oldpool);
3194 static void raid1_quiesce(struct mddev *mddev, int state)
3196 struct r1conf *conf = mddev->private;
3199 case 2: /* wake for suspend */
3200 wake_up(&conf->wait_barrier);
3203 freeze_array(conf, 0);
3206 unfreeze_array(conf);
3211 static void *raid1_takeover(struct mddev *mddev)
3213 /* raid1 can take over:
3214 * raid5 with 2 devices, any layout or chunk size
3216 if (mddev->level == 5 && mddev->raid_disks == 2) {
3217 struct r1conf *conf;
3218 mddev->new_level = 1;
3219 mddev->new_layout = 0;
3220 mddev->new_chunk_sectors = 0;
3221 conf = setup_conf(mddev);
3223 /* Array must appear to be quiesced */
3224 conf->array_frozen = 1;
3227 return ERR_PTR(-EINVAL);
3230 static struct md_personality raid1_personality =
3234 .owner = THIS_MODULE,
3235 .make_request = raid1_make_request,
3238 .status = raid1_status,
3239 .error_handler = raid1_error,
3240 .hot_add_disk = raid1_add_disk,
3241 .hot_remove_disk= raid1_remove_disk,
3242 .spare_active = raid1_spare_active,
3243 .sync_request = raid1_sync_request,
3244 .resize = raid1_resize,
3246 .check_reshape = raid1_reshape,
3247 .quiesce = raid1_quiesce,
3248 .takeover = raid1_takeover,
3249 .congested = raid1_congested,
3252 static int __init raid_init(void)
3254 return register_md_personality(&raid1_personality);
3257 static void raid_exit(void)
3259 unregister_md_personality(&raid1_personality);
3262 module_init(raid_init);
3263 module_exit(raid_exit);
3264 MODULE_LICENSE("GPL");
3265 MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD");
3266 MODULE_ALIAS("md-personality-3"); /* RAID1 */
3267 MODULE_ALIAS("md-raid1");
3268 MODULE_ALIAS("md-level-1");
3270 module_param(max_queued_requests, int, S_IRUGO|S_IWUSR);