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
51 static void unplug_slaves(mddev_t
*mddev
);
53 static void allow_barrier(conf_t
*conf
);
54 static void lower_barrier(conf_t
*conf
);
56 static void * r1bio_pool_alloc(gfp_t gfp_flags
, void *data
)
58 struct pool_info
*pi
= data
;
60 int size
= offsetof(r1bio_t
, bios
[pi
->raid_disks
]);
62 /* allocate a r1bio with room for raid_disks entries in the bios array */
63 r1_bio
= kzalloc(size
, gfp_flags
);
65 unplug_slaves(pi
->mddev
);
70 static void r1bio_pool_free(void *r1_bio
, void *data
)
75 #define RESYNC_BLOCK_SIZE (64*1024)
76 //#define RESYNC_BLOCK_SIZE PAGE_SIZE
77 #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9)
78 #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE)
79 #define RESYNC_WINDOW (2048*1024)
81 static void * r1buf_pool_alloc(gfp_t gfp_flags
, void *data
)
83 struct pool_info
*pi
= data
;
89 r1_bio
= r1bio_pool_alloc(gfp_flags
, pi
);
91 unplug_slaves(pi
->mddev
);
96 * Allocate bios : 1 for reading, n-1 for writing
98 for (j
= pi
->raid_disks
; j
-- ; ) {
99 bio
= bio_alloc(gfp_flags
, RESYNC_PAGES
);
102 r1_bio
->bios
[j
] = bio
;
105 * Allocate RESYNC_PAGES data pages and attach them to
107 * If this is a user-requested check/repair, allocate
108 * RESYNC_PAGES for each bio.
110 if (test_bit(MD_RECOVERY_REQUESTED
, &pi
->mddev
->recovery
))
115 bio
= r1_bio
->bios
[j
];
116 for (i
= 0; i
< RESYNC_PAGES
; i
++) {
117 page
= alloc_page(gfp_flags
);
121 bio
->bi_io_vec
[i
].bv_page
= page
;
124 /* If not user-requests, copy the page pointers to all bios */
125 if (!test_bit(MD_RECOVERY_REQUESTED
, &pi
->mddev
->recovery
)) {
126 for (i
=0; i
<RESYNC_PAGES
; i
++)
127 for (j
=1; j
<pi
->raid_disks
; j
++)
128 r1_bio
->bios
[j
]->bi_io_vec
[i
].bv_page
=
129 r1_bio
->bios
[0]->bi_io_vec
[i
].bv_page
;
132 r1_bio
->master_bio
= NULL
;
137 for (i
=0; i
< RESYNC_PAGES
; i
++)
138 for (j
=0 ; j
< pi
->raid_disks
; j
++)
139 safe_put_page(r1_bio
->bios
[j
]->bi_io_vec
[i
].bv_page
);
142 while ( ++j
< pi
->raid_disks
)
143 bio_put(r1_bio
->bios
[j
]);
144 r1bio_pool_free(r1_bio
, data
);
148 static void r1buf_pool_free(void *__r1_bio
, void *data
)
150 struct pool_info
*pi
= data
;
152 r1bio_t
*r1bio
= __r1_bio
;
154 for (i
= 0; i
< RESYNC_PAGES
; i
++)
155 for (j
= pi
->raid_disks
; j
-- ;) {
157 r1bio
->bios
[j
]->bi_io_vec
[i
].bv_page
!=
158 r1bio
->bios
[0]->bi_io_vec
[i
].bv_page
)
159 safe_put_page(r1bio
->bios
[j
]->bi_io_vec
[i
].bv_page
);
161 for (i
=0 ; i
< pi
->raid_disks
; i
++)
162 bio_put(r1bio
->bios
[i
]);
164 r1bio_pool_free(r1bio
, data
);
167 static void put_all_bios(conf_t
*conf
, r1bio_t
*r1_bio
)
171 for (i
= 0; i
< conf
->raid_disks
; i
++) {
172 struct bio
**bio
= r1_bio
->bios
+ i
;
173 if (*bio
&& *bio
!= IO_BLOCKED
)
179 static inline void free_r1bio(r1bio_t
*r1_bio
)
181 conf_t
*conf
= mddev_to_conf(r1_bio
->mddev
);
184 * Wake up any possible resync thread that waits for the device
189 put_all_bios(conf
, r1_bio
);
190 mempool_free(r1_bio
, conf
->r1bio_pool
);
193 static inline void put_buf(r1bio_t
*r1_bio
)
195 conf_t
*conf
= mddev_to_conf(r1_bio
->mddev
);
198 for (i
=0; i
<conf
->raid_disks
; i
++) {
199 struct bio
*bio
= r1_bio
->bios
[i
];
201 rdev_dec_pending(conf
->mirrors
[i
].rdev
, r1_bio
->mddev
);
204 mempool_free(r1_bio
, conf
->r1buf_pool
);
209 static void reschedule_retry(r1bio_t
*r1_bio
)
212 mddev_t
*mddev
= r1_bio
->mddev
;
213 conf_t
*conf
= mddev_to_conf(mddev
);
215 spin_lock_irqsave(&conf
->device_lock
, flags
);
216 list_add(&r1_bio
->retry_list
, &conf
->retry_list
);
218 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
220 wake_up(&conf
->wait_barrier
);
221 md_wakeup_thread(mddev
->thread
);
225 * raid_end_bio_io() is called when we have finished servicing a mirrored
226 * operation and are ready to return a success/failure code to the buffer
229 static void raid_end_bio_io(r1bio_t
*r1_bio
)
231 struct bio
*bio
= r1_bio
->master_bio
;
233 /* if nobody has done the final endio yet, do it now */
234 if (!test_and_set_bit(R1BIO_Returned
, &r1_bio
->state
)) {
235 PRINTK(KERN_DEBUG
"raid1: sync end %s on sectors %llu-%llu\n",
236 (bio_data_dir(bio
) == WRITE
) ? "write" : "read",
237 (unsigned long long) bio
->bi_sector
,
238 (unsigned long long) bio
->bi_sector
+
239 (bio
->bi_size
>> 9) - 1);
241 bio_endio(bio
, bio
->bi_size
,
242 test_bit(R1BIO_Uptodate
, &r1_bio
->state
) ? 0 : -EIO
);
248 * Update disk head position estimator based on IRQ completion info.
250 static inline void update_head_pos(int disk
, r1bio_t
*r1_bio
)
252 conf_t
*conf
= mddev_to_conf(r1_bio
->mddev
);
254 conf
->mirrors
[disk
].head_position
=
255 r1_bio
->sector
+ (r1_bio
->sectors
);
258 static int raid1_end_read_request(struct bio
*bio
, unsigned int bytes_done
, int error
)
260 int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
261 r1bio_t
* r1_bio
= (r1bio_t
*)(bio
->bi_private
);
263 conf_t
*conf
= mddev_to_conf(r1_bio
->mddev
);
268 mirror
= r1_bio
->read_disk
;
270 * this branch is our 'one mirror IO has finished' event handler:
272 update_head_pos(mirror
, r1_bio
);
274 if (uptodate
|| conf
->working_disks
<= 1) {
276 * Set R1BIO_Uptodate in our master bio, so that
277 * we will return a good error code for to the higher
278 * levels even if IO on some other mirrored buffer fails.
280 * The 'master' represents the composite IO operation to
281 * user-side. So if something waits for IO, then it will
282 * wait for the 'master' bio.
285 set_bit(R1BIO_Uptodate
, &r1_bio
->state
);
287 raid_end_bio_io(r1_bio
);
292 char b
[BDEVNAME_SIZE
];
293 if (printk_ratelimit())
294 printk(KERN_ERR
"raid1: %s: rescheduling sector %llu\n",
295 bdevname(conf
->mirrors
[mirror
].rdev
->bdev
,b
), (unsigned long long)r1_bio
->sector
);
296 reschedule_retry(r1_bio
);
299 rdev_dec_pending(conf
->mirrors
[mirror
].rdev
, conf
->mddev
);
303 static int raid1_end_write_request(struct bio
*bio
, unsigned int bytes_done
, int error
)
305 int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
306 r1bio_t
* r1_bio
= (r1bio_t
*)(bio
->bi_private
);
307 int mirror
, behind
= test_bit(R1BIO_BehindIO
, &r1_bio
->state
);
308 conf_t
*conf
= mddev_to_conf(r1_bio
->mddev
);
313 for (mirror
= 0; mirror
< conf
->raid_disks
; mirror
++)
314 if (r1_bio
->bios
[mirror
] == bio
)
317 if (error
== -ENOTSUPP
&& test_bit(R1BIO_Barrier
, &r1_bio
->state
)) {
318 set_bit(BarriersNotsupp
, &conf
->mirrors
[mirror
].rdev
->flags
);
319 set_bit(R1BIO_BarrierRetry
, &r1_bio
->state
);
320 r1_bio
->mddev
->barriers_work
= 0;
323 * this branch is our 'one mirror IO has finished' event handler:
325 r1_bio
->bios
[mirror
] = NULL
;
327 md_error(r1_bio
->mddev
, conf
->mirrors
[mirror
].rdev
);
328 /* an I/O failed, we can't clear the bitmap */
329 set_bit(R1BIO_Degraded
, &r1_bio
->state
);
332 * Set R1BIO_Uptodate in our master bio, so that
333 * we will return a good error code for to the higher
334 * levels even if IO on some other mirrored buffer fails.
336 * The 'master' represents the composite IO operation to
337 * user-side. So if something waits for IO, then it will
338 * wait for the 'master' bio.
340 set_bit(R1BIO_Uptodate
, &r1_bio
->state
);
342 update_head_pos(mirror
, r1_bio
);
345 if (test_bit(WriteMostly
, &conf
->mirrors
[mirror
].rdev
->flags
))
346 atomic_dec(&r1_bio
->behind_remaining
);
348 /* In behind mode, we ACK the master bio once the I/O has safely
349 * reached all non-writemostly disks. Setting the Returned bit
350 * ensures that this gets done only once -- we don't ever want to
351 * return -EIO here, instead we'll wait */
353 if (atomic_read(&r1_bio
->behind_remaining
) >= (atomic_read(&r1_bio
->remaining
)-1) &&
354 test_bit(R1BIO_Uptodate
, &r1_bio
->state
)) {
355 /* Maybe we can return now */
356 if (!test_and_set_bit(R1BIO_Returned
, &r1_bio
->state
)) {
357 struct bio
*mbio
= r1_bio
->master_bio
;
358 PRINTK(KERN_DEBUG
"raid1: behind end write sectors %llu-%llu\n",
359 (unsigned long long) mbio
->bi_sector
,
360 (unsigned long long) mbio
->bi_sector
+
361 (mbio
->bi_size
>> 9) - 1);
362 bio_endio(mbio
, mbio
->bi_size
, 0);
369 * Let's see if all mirrored write operations have finished
372 if (atomic_dec_and_test(&r1_bio
->remaining
)) {
373 if (test_bit(R1BIO_BarrierRetry
, &r1_bio
->state
)) {
374 reschedule_retry(r1_bio
);
375 /* Don't dec_pending yet, we want to hold
376 * the reference over the retry
380 if (test_bit(R1BIO_BehindIO
, &r1_bio
->state
)) {
381 /* free extra copy of the data pages */
382 int i
= bio
->bi_vcnt
;
384 safe_put_page(bio
->bi_io_vec
[i
].bv_page
);
386 /* clear the bitmap if all writes complete successfully */
387 bitmap_endwrite(r1_bio
->mddev
->bitmap
, r1_bio
->sector
,
389 !test_bit(R1BIO_Degraded
, &r1_bio
->state
),
391 md_write_end(r1_bio
->mddev
);
392 raid_end_bio_io(r1_bio
);
395 if (r1_bio
->bios
[mirror
]==NULL
)
398 rdev_dec_pending(conf
->mirrors
[mirror
].rdev
, conf
->mddev
);
404 * This routine returns the disk from which the requested read should
405 * be done. There is a per-array 'next expected sequential IO' sector
406 * number - if this matches on the next IO then we use the last disk.
407 * There is also a per-disk 'last know head position' sector that is
408 * maintained from IRQ contexts, both the normal and the resync IO
409 * completion handlers update this position correctly. If there is no
410 * perfect sequential match then we pick the disk whose head is closest.
412 * If there are 2 mirrors in the same 2 devices, performance degrades
413 * because position is mirror, not device based.
415 * The rdev for the device selected will have nr_pending incremented.
417 static int read_balance(conf_t
*conf
, r1bio_t
*r1_bio
)
419 const unsigned long this_sector
= r1_bio
->sector
;
420 int new_disk
= conf
->last_used
, disk
= new_disk
;
422 const int sectors
= r1_bio
->sectors
;
423 sector_t new_distance
, current_distance
;
428 * Check if we can balance. We can balance on the whole
429 * device if no resync is going on, or below the resync window.
430 * We take the first readable disk when above the resync window.
433 if (conf
->mddev
->recovery_cp
< MaxSector
&&
434 (this_sector
+ sectors
>= conf
->next_resync
)) {
435 /* Choose the first operation device, for consistancy */
438 for (rdev
= rcu_dereference(conf
->mirrors
[new_disk
].rdev
);
439 r1_bio
->bios
[new_disk
] == IO_BLOCKED
||
440 !rdev
|| !test_bit(In_sync
, &rdev
->flags
)
441 || test_bit(WriteMostly
, &rdev
->flags
);
442 rdev
= rcu_dereference(conf
->mirrors
[++new_disk
].rdev
)) {
444 if (rdev
&& test_bit(In_sync
, &rdev
->flags
) &&
445 r1_bio
->bios
[new_disk
] != IO_BLOCKED
)
446 wonly_disk
= new_disk
;
448 if (new_disk
== conf
->raid_disks
- 1) {
449 new_disk
= wonly_disk
;
457 /* make sure the disk is operational */
458 for (rdev
= rcu_dereference(conf
->mirrors
[new_disk
].rdev
);
459 r1_bio
->bios
[new_disk
] == IO_BLOCKED
||
460 !rdev
|| !test_bit(In_sync
, &rdev
->flags
) ||
461 test_bit(WriteMostly
, &rdev
->flags
);
462 rdev
= rcu_dereference(conf
->mirrors
[new_disk
].rdev
)) {
464 if (rdev
&& test_bit(In_sync
, &rdev
->flags
) &&
465 r1_bio
->bios
[new_disk
] != IO_BLOCKED
)
466 wonly_disk
= new_disk
;
469 new_disk
= conf
->raid_disks
;
471 if (new_disk
== disk
) {
472 new_disk
= wonly_disk
;
481 /* now disk == new_disk == starting point for search */
484 * Don't change to another disk for sequential reads:
486 if (conf
->next_seq_sect
== this_sector
)
488 if (this_sector
== conf
->mirrors
[new_disk
].head_position
)
491 current_distance
= abs(this_sector
- conf
->mirrors
[disk
].head_position
);
493 /* Find the disk whose head is closest */
497 disk
= conf
->raid_disks
;
500 rdev
= rcu_dereference(conf
->mirrors
[disk
].rdev
);
502 if (!rdev
|| r1_bio
->bios
[disk
] == IO_BLOCKED
||
503 !test_bit(In_sync
, &rdev
->flags
) ||
504 test_bit(WriteMostly
, &rdev
->flags
))
507 if (!atomic_read(&rdev
->nr_pending
)) {
511 new_distance
= abs(this_sector
- conf
->mirrors
[disk
].head_position
);
512 if (new_distance
< current_distance
) {
513 current_distance
= new_distance
;
516 } while (disk
!= conf
->last_used
);
522 rdev
= rcu_dereference(conf
->mirrors
[new_disk
].rdev
);
525 atomic_inc(&rdev
->nr_pending
);
526 if (!test_bit(In_sync
, &rdev
->flags
)) {
527 /* cannot risk returning a device that failed
528 * before we inc'ed nr_pending
530 rdev_dec_pending(rdev
, conf
->mddev
);
533 conf
->next_seq_sect
= this_sector
+ sectors
;
534 conf
->last_used
= new_disk
;
541 static void unplug_slaves(mddev_t
*mddev
)
543 conf_t
*conf
= mddev_to_conf(mddev
);
547 for (i
=0; i
<mddev
->raid_disks
; i
++) {
548 mdk_rdev_t
*rdev
= rcu_dereference(conf
->mirrors
[i
].rdev
);
549 if (rdev
&& !test_bit(Faulty
, &rdev
->flags
) && atomic_read(&rdev
->nr_pending
)) {
550 request_queue_t
*r_queue
= bdev_get_queue(rdev
->bdev
);
552 atomic_inc(&rdev
->nr_pending
);
555 if (r_queue
->unplug_fn
)
556 r_queue
->unplug_fn(r_queue
);
558 rdev_dec_pending(rdev
, mddev
);
565 static void raid1_unplug(request_queue_t
*q
)
567 mddev_t
*mddev
= q
->queuedata
;
569 unplug_slaves(mddev
);
570 md_wakeup_thread(mddev
->thread
);
573 static int raid1_issue_flush(request_queue_t
*q
, struct gendisk
*disk
,
574 sector_t
*error_sector
)
576 mddev_t
*mddev
= q
->queuedata
;
577 conf_t
*conf
= mddev_to_conf(mddev
);
581 for (i
=0; i
<mddev
->raid_disks
&& ret
== 0; i
++) {
582 mdk_rdev_t
*rdev
= rcu_dereference(conf
->mirrors
[i
].rdev
);
583 if (rdev
&& !test_bit(Faulty
, &rdev
->flags
)) {
584 struct block_device
*bdev
= rdev
->bdev
;
585 request_queue_t
*r_queue
= bdev_get_queue(bdev
);
587 if (!r_queue
->issue_flush_fn
)
590 atomic_inc(&rdev
->nr_pending
);
592 ret
= r_queue
->issue_flush_fn(r_queue
, bdev
->bd_disk
,
594 rdev_dec_pending(rdev
, mddev
);
604 * Sometimes we need to suspend IO while we do something else,
605 * either some resync/recovery, or reconfigure the array.
606 * To do this we raise a 'barrier'.
607 * The 'barrier' is a counter that can be raised multiple times
608 * to count how many activities are happening which preclude
610 * We can only raise the barrier if there is no pending IO.
611 * i.e. if nr_pending == 0.
612 * We choose only to raise the barrier if no-one is waiting for the
613 * barrier to go down. This means that as soon as an IO request
614 * is ready, no other operations which require a barrier will start
615 * until the IO request has had a chance.
617 * So: regular IO calls 'wait_barrier'. When that returns there
618 * is no backgroup IO happening, It must arrange to call
619 * allow_barrier when it has finished its IO.
620 * backgroup IO calls must call raise_barrier. Once that returns
621 * there is no normal IO happeing. It must arrange to call
622 * lower_barrier when the particular background IO completes.
624 #define RESYNC_DEPTH 32
626 static void raise_barrier(conf_t
*conf
)
628 spin_lock_irq(&conf
->resync_lock
);
630 /* Wait until no block IO is waiting */
631 wait_event_lock_irq(conf
->wait_barrier
, !conf
->nr_waiting
,
633 raid1_unplug(conf
->mddev
->queue
));
635 /* block any new IO from starting */
638 /* No wait for all pending IO to complete */
639 wait_event_lock_irq(conf
->wait_barrier
,
640 !conf
->nr_pending
&& conf
->barrier
< RESYNC_DEPTH
,
642 raid1_unplug(conf
->mddev
->queue
));
644 spin_unlock_irq(&conf
->resync_lock
);
647 static void lower_barrier(conf_t
*conf
)
650 spin_lock_irqsave(&conf
->resync_lock
, flags
);
652 spin_unlock_irqrestore(&conf
->resync_lock
, flags
);
653 wake_up(&conf
->wait_barrier
);
656 static void wait_barrier(conf_t
*conf
)
658 spin_lock_irq(&conf
->resync_lock
);
661 wait_event_lock_irq(conf
->wait_barrier
, !conf
->barrier
,
663 raid1_unplug(conf
->mddev
->queue
));
667 spin_unlock_irq(&conf
->resync_lock
);
670 static void allow_barrier(conf_t
*conf
)
673 spin_lock_irqsave(&conf
->resync_lock
, flags
);
675 spin_unlock_irqrestore(&conf
->resync_lock
, flags
);
676 wake_up(&conf
->wait_barrier
);
679 static void freeze_array(conf_t
*conf
)
681 /* stop syncio and normal IO and wait for everything to
683 * We increment barrier and nr_waiting, and then
684 * wait until barrier+nr_pending match nr_queued+2
686 spin_lock_irq(&conf
->resync_lock
);
689 wait_event_lock_irq(conf
->wait_barrier
,
690 conf
->barrier
+conf
->nr_pending
== conf
->nr_queued
+2,
692 raid1_unplug(conf
->mddev
->queue
));
693 spin_unlock_irq(&conf
->resync_lock
);
695 static void unfreeze_array(conf_t
*conf
)
697 /* reverse the effect of the freeze */
698 spin_lock_irq(&conf
->resync_lock
);
701 wake_up(&conf
->wait_barrier
);
702 spin_unlock_irq(&conf
->resync_lock
);
706 /* duplicate the data pages for behind I/O */
707 static struct page
**alloc_behind_pages(struct bio
*bio
)
710 struct bio_vec
*bvec
;
711 struct page
**pages
= kzalloc(bio
->bi_vcnt
* sizeof(struct page
*),
713 if (unlikely(!pages
))
716 bio_for_each_segment(bvec
, bio
, i
) {
717 pages
[i
] = alloc_page(GFP_NOIO
);
718 if (unlikely(!pages
[i
]))
720 memcpy(kmap(pages
[i
]) + bvec
->bv_offset
,
721 kmap(bvec
->bv_page
) + bvec
->bv_offset
, bvec
->bv_len
);
723 kunmap(bvec
->bv_page
);
730 for (i
= 0; i
< bio
->bi_vcnt
&& pages
[i
]; i
++)
733 PRINTK("%dB behind alloc failed, doing sync I/O\n", bio
->bi_size
);
737 static int make_request(request_queue_t
*q
, struct bio
* bio
)
739 mddev_t
*mddev
= q
->queuedata
;
740 conf_t
*conf
= mddev_to_conf(mddev
);
741 mirror_info_t
*mirror
;
743 struct bio
*read_bio
;
744 int i
, targets
= 0, disks
;
746 struct bitmap
*bitmap
= mddev
->bitmap
;
749 struct page
**behind_pages
= NULL
;
750 const int rw
= bio_data_dir(bio
);
753 if (unlikely(!mddev
->barriers_work
&& bio_barrier(bio
))) {
754 bio_endio(bio
, bio
->bi_size
, -EOPNOTSUPP
);
759 * Register the new request and wait if the reconstruction
760 * thread has put up a bar for new requests.
761 * Continue immediately if no resync is active currently.
763 md_write_start(mddev
, bio
); /* wait on superblock update early */
767 disk_stat_inc(mddev
->gendisk
, ios
[rw
]);
768 disk_stat_add(mddev
->gendisk
, sectors
[rw
], bio_sectors(bio
));
771 * make_request() can abort the operation when READA is being
772 * used and no empty request is available.
775 r1_bio
= mempool_alloc(conf
->r1bio_pool
, GFP_NOIO
);
777 r1_bio
->master_bio
= bio
;
778 r1_bio
->sectors
= bio
->bi_size
>> 9;
780 r1_bio
->mddev
= mddev
;
781 r1_bio
->sector
= bio
->bi_sector
;
785 * read balancing logic:
787 int rdisk
= read_balance(conf
, r1_bio
);
790 /* couldn't find anywhere to read from */
791 raid_end_bio_io(r1_bio
);
794 mirror
= conf
->mirrors
+ rdisk
;
796 r1_bio
->read_disk
= rdisk
;
798 read_bio
= bio_clone(bio
, GFP_NOIO
);
800 r1_bio
->bios
[rdisk
] = read_bio
;
802 read_bio
->bi_sector
= r1_bio
->sector
+ mirror
->rdev
->data_offset
;
803 read_bio
->bi_bdev
= mirror
->rdev
->bdev
;
804 read_bio
->bi_end_io
= raid1_end_read_request
;
805 read_bio
->bi_rw
= READ
;
806 read_bio
->bi_private
= r1_bio
;
808 generic_make_request(read_bio
);
815 /* first select target devices under spinlock and
816 * inc refcount on their rdev. Record them by setting
819 disks
= conf
->raid_disks
;
821 { static int first
=1;
822 if (first
) printk("First Write sector %llu disks %d\n",
823 (unsigned long long)r1_bio
->sector
, disks
);
828 for (i
= 0; i
< disks
; i
++) {
829 if ((rdev
=rcu_dereference(conf
->mirrors
[i
].rdev
)) != NULL
&&
830 !test_bit(Faulty
, &rdev
->flags
)) {
831 atomic_inc(&rdev
->nr_pending
);
832 if (test_bit(Faulty
, &rdev
->flags
)) {
833 rdev_dec_pending(rdev
, mddev
);
834 r1_bio
->bios
[i
] = NULL
;
836 r1_bio
->bios
[i
] = bio
;
839 r1_bio
->bios
[i
] = NULL
;
843 BUG_ON(targets
== 0); /* we never fail the last device */
845 if (targets
< conf
->raid_disks
) {
846 /* array is degraded, we will not clear the bitmap
847 * on I/O completion (see raid1_end_write_request) */
848 set_bit(R1BIO_Degraded
, &r1_bio
->state
);
851 /* do behind I/O ? */
853 atomic_read(&bitmap
->behind_writes
) < bitmap
->max_write_behind
&&
854 (behind_pages
= alloc_behind_pages(bio
)) != NULL
)
855 set_bit(R1BIO_BehindIO
, &r1_bio
->state
);
857 atomic_set(&r1_bio
->remaining
, 0);
858 atomic_set(&r1_bio
->behind_remaining
, 0);
860 do_barriers
= bio
->bi_rw
& BIO_RW_BARRIER
;
862 set_bit(R1BIO_Barrier
, &r1_bio
->state
);
865 for (i
= 0; i
< disks
; i
++) {
867 if (!r1_bio
->bios
[i
])
870 mbio
= bio_clone(bio
, GFP_NOIO
);
871 r1_bio
->bios
[i
] = mbio
;
873 mbio
->bi_sector
= r1_bio
->sector
+ conf
->mirrors
[i
].rdev
->data_offset
;
874 mbio
->bi_bdev
= conf
->mirrors
[i
].rdev
->bdev
;
875 mbio
->bi_end_io
= raid1_end_write_request
;
876 mbio
->bi_rw
= WRITE
| do_barriers
;
877 mbio
->bi_private
= r1_bio
;
880 struct bio_vec
*bvec
;
883 /* Yes, I really want the '__' version so that
884 * we clear any unused pointer in the io_vec, rather
885 * than leave them unchanged. This is important
886 * because when we come to free the pages, we won't
887 * know the originial bi_idx, so we just free
890 __bio_for_each_segment(bvec
, mbio
, j
, 0)
891 bvec
->bv_page
= behind_pages
[j
];
892 if (test_bit(WriteMostly
, &conf
->mirrors
[i
].rdev
->flags
))
893 atomic_inc(&r1_bio
->behind_remaining
);
896 atomic_inc(&r1_bio
->remaining
);
898 bio_list_add(&bl
, mbio
);
900 kfree(behind_pages
); /* the behind pages are attached to the bios now */
902 bitmap_startwrite(bitmap
, bio
->bi_sector
, r1_bio
->sectors
,
903 test_bit(R1BIO_BehindIO
, &r1_bio
->state
));
904 spin_lock_irqsave(&conf
->device_lock
, flags
);
905 bio_list_merge(&conf
->pending_bio_list
, &bl
);
908 blk_plug_device(mddev
->queue
);
909 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
912 while ((bio
= bio_list_pop(&bl
)) != NULL
)
913 generic_make_request(bio
);
919 static void status(struct seq_file
*seq
, mddev_t
*mddev
)
921 conf_t
*conf
= mddev_to_conf(mddev
);
924 seq_printf(seq
, " [%d/%d] [", conf
->raid_disks
,
925 conf
->working_disks
);
926 for (i
= 0; i
< conf
->raid_disks
; i
++)
927 seq_printf(seq
, "%s",
928 conf
->mirrors
[i
].rdev
&&
929 test_bit(In_sync
, &conf
->mirrors
[i
].rdev
->flags
) ? "U" : "_");
930 seq_printf(seq
, "]");
934 static void error(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
936 char b
[BDEVNAME_SIZE
];
937 conf_t
*conf
= mddev_to_conf(mddev
);
940 * If it is not operational, then we have already marked it as dead
941 * else if it is the last working disks, ignore the error, let the
942 * next level up know.
943 * else mark the drive as failed
945 if (test_bit(In_sync
, &rdev
->flags
)
946 && conf
->working_disks
== 1)
948 * Don't fail the drive, act as though we were just a
949 * normal single drive
952 if (test_bit(In_sync
, &rdev
->flags
)) {
954 conf
->working_disks
--;
956 * if recovery is running, make sure it aborts.
958 set_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
960 clear_bit(In_sync
, &rdev
->flags
);
961 set_bit(Faulty
, &rdev
->flags
);
963 printk(KERN_ALERT
"raid1: Disk failure on %s, disabling device. \n"
964 " Operation continuing on %d devices\n",
965 bdevname(rdev
->bdev
,b
), conf
->working_disks
);
968 static void print_conf(conf_t
*conf
)
973 printk("RAID1 conf printout:\n");
978 printk(" --- wd:%d rd:%d\n", conf
->working_disks
,
981 for (i
= 0; i
< conf
->raid_disks
; i
++) {
982 char b
[BDEVNAME_SIZE
];
983 tmp
= conf
->mirrors
+ i
;
985 printk(" disk %d, wo:%d, o:%d, dev:%s\n",
986 i
, !test_bit(In_sync
, &tmp
->rdev
->flags
), !test_bit(Faulty
, &tmp
->rdev
->flags
),
987 bdevname(tmp
->rdev
->bdev
,b
));
991 static void close_sync(conf_t
*conf
)
996 mempool_destroy(conf
->r1buf_pool
);
997 conf
->r1buf_pool
= NULL
;
1000 static int raid1_spare_active(mddev_t
*mddev
)
1003 conf_t
*conf
= mddev
->private;
1007 * Find all failed disks within the RAID1 configuration
1008 * and mark them readable
1010 for (i
= 0; i
< conf
->raid_disks
; i
++) {
1011 tmp
= conf
->mirrors
+ i
;
1013 && !test_bit(Faulty
, &tmp
->rdev
->flags
)
1014 && !test_bit(In_sync
, &tmp
->rdev
->flags
)) {
1015 conf
->working_disks
++;
1017 set_bit(In_sync
, &tmp
->rdev
->flags
);
1026 static int raid1_add_disk(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1028 conf_t
*conf
= mddev
->private;
1033 for (mirror
=0; mirror
< mddev
->raid_disks
; mirror
++)
1034 if ( !(p
=conf
->mirrors
+mirror
)->rdev
) {
1036 blk_queue_stack_limits(mddev
->queue
,
1037 rdev
->bdev
->bd_disk
->queue
);
1038 /* as we don't honour merge_bvec_fn, we must never risk
1039 * violating it, so limit ->max_sector to one PAGE, as
1040 * a one page request is never in violation.
1042 if (rdev
->bdev
->bd_disk
->queue
->merge_bvec_fn
&&
1043 mddev
->queue
->max_sectors
> (PAGE_SIZE
>>9))
1044 blk_queue_max_sectors(mddev
->queue
, PAGE_SIZE
>>9);
1046 p
->head_position
= 0;
1047 rdev
->raid_disk
= mirror
;
1049 /* As all devices are equivalent, we don't need a full recovery
1050 * if this was recently any drive of the array
1052 if (rdev
->saved_raid_disk
< 0)
1054 rcu_assign_pointer(p
->rdev
, rdev
);
1062 static int raid1_remove_disk(mddev_t
*mddev
, int number
)
1064 conf_t
*conf
= mddev
->private;
1067 mirror_info_t
*p
= conf
->mirrors
+ number
;
1072 if (test_bit(In_sync
, &rdev
->flags
) ||
1073 atomic_read(&rdev
->nr_pending
)) {
1079 if (atomic_read(&rdev
->nr_pending
)) {
1080 /* lost the race, try later */
1092 static int end_sync_read(struct bio
*bio
, unsigned int bytes_done
, int error
)
1094 r1bio_t
* r1_bio
= (r1bio_t
*)(bio
->bi_private
);
1100 for (i
=r1_bio
->mddev
->raid_disks
; i
--; )
1101 if (r1_bio
->bios
[i
] == bio
)
1104 update_head_pos(i
, r1_bio
);
1106 * we have read a block, now it needs to be re-written,
1107 * or re-read if the read failed.
1108 * We don't do much here, just schedule handling by raid1d
1110 if (test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
1111 set_bit(R1BIO_Uptodate
, &r1_bio
->state
);
1113 if (atomic_dec_and_test(&r1_bio
->remaining
))
1114 reschedule_retry(r1_bio
);
1118 static int end_sync_write(struct bio
*bio
, unsigned int bytes_done
, int error
)
1120 int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1121 r1bio_t
* r1_bio
= (r1bio_t
*)(bio
->bi_private
);
1122 mddev_t
*mddev
= r1_bio
->mddev
;
1123 conf_t
*conf
= mddev_to_conf(mddev
);
1130 for (i
= 0; i
< conf
->raid_disks
; i
++)
1131 if (r1_bio
->bios
[i
] == bio
) {
1136 md_error(mddev
, conf
->mirrors
[mirror
].rdev
);
1138 update_head_pos(mirror
, r1_bio
);
1140 if (atomic_dec_and_test(&r1_bio
->remaining
)) {
1141 md_done_sync(mddev
, r1_bio
->sectors
, uptodate
);
1147 static void sync_request_write(mddev_t
*mddev
, r1bio_t
*r1_bio
)
1149 conf_t
*conf
= mddev_to_conf(mddev
);
1151 int disks
= conf
->raid_disks
;
1152 struct bio
*bio
, *wbio
;
1154 bio
= r1_bio
->bios
[r1_bio
->read_disk
];
1157 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
1158 /* We have read all readable devices. If we haven't
1159 * got the block, then there is no hope left.
1160 * If we have, then we want to do a comparison
1161 * and skip the write if everything is the same.
1162 * If any blocks failed to read, then we need to
1163 * attempt an over-write
1166 if (!test_bit(R1BIO_Uptodate
, &r1_bio
->state
)) {
1167 for (i
=0; i
<mddev
->raid_disks
; i
++)
1168 if (r1_bio
->bios
[i
]->bi_end_io
== end_sync_read
)
1169 md_error(mddev
, conf
->mirrors
[i
].rdev
);
1171 md_done_sync(mddev
, r1_bio
->sectors
, 1);
1175 for (primary
=0; primary
<mddev
->raid_disks
; primary
++)
1176 if (r1_bio
->bios
[primary
]->bi_end_io
== end_sync_read
&&
1177 test_bit(BIO_UPTODATE
, &r1_bio
->bios
[primary
]->bi_flags
)) {
1178 r1_bio
->bios
[primary
]->bi_end_io
= NULL
;
1179 rdev_dec_pending(conf
->mirrors
[primary
].rdev
, mddev
);
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
;
1199 rdev_dec_pending(conf
->mirrors
[i
].rdev
, mddev
);
1201 /* fixup the bio for reuse */
1202 sbio
->bi_vcnt
= vcnt
;
1203 sbio
->bi_size
= r1_bio
->sectors
<< 9;
1205 sbio
->bi_phys_segments
= 0;
1206 sbio
->bi_hw_segments
= 0;
1207 sbio
->bi_hw_front_size
= 0;
1208 sbio
->bi_hw_back_size
= 0;
1209 sbio
->bi_flags
&= ~(BIO_POOL_MASK
- 1);
1210 sbio
->bi_flags
|= 1 << BIO_UPTODATE
;
1211 sbio
->bi_next
= NULL
;
1212 sbio
->bi_sector
= r1_bio
->sector
+
1213 conf
->mirrors
[i
].rdev
->data_offset
;
1214 sbio
->bi_bdev
= conf
->mirrors
[i
].rdev
->bdev
;
1218 if (!test_bit(R1BIO_Uptodate
, &r1_bio
->state
)) {
1219 /* ouch - failed to read all of that.
1220 * Try some synchronous reads of other devices to get
1221 * good data, much like with normal read errors. Only
1222 * read into the pages we already have so they we don't
1223 * need to re-issue the read request.
1224 * We don't need to freeze the array, because being in an
1225 * active sync request, there is no normal IO, and
1226 * no overlapping syncs.
1228 sector_t sect
= r1_bio
->sector
;
1229 int sectors
= r1_bio
->sectors
;
1234 int d
= r1_bio
->read_disk
;
1238 if (s
> (PAGE_SIZE
>>9))
1241 if (r1_bio
->bios
[d
]->bi_end_io
== end_sync_read
) {
1242 rdev
= conf
->mirrors
[d
].rdev
;
1243 if (sync_page_io(rdev
->bdev
,
1244 sect
+ rdev
->data_offset
,
1246 bio
->bi_io_vec
[idx
].bv_page
,
1253 if (d
== conf
->raid_disks
)
1255 } while (!success
&& d
!= r1_bio
->read_disk
);
1259 /* write it back and re-read */
1260 set_bit(R1BIO_Uptodate
, &r1_bio
->state
);
1261 while (d
!= r1_bio
->read_disk
) {
1263 d
= conf
->raid_disks
;
1265 if (r1_bio
->bios
[d
]->bi_end_io
!= end_sync_read
)
1267 rdev
= conf
->mirrors
[d
].rdev
;
1268 if (sync_page_io(rdev
->bdev
,
1269 sect
+ rdev
->data_offset
,
1271 bio
->bi_io_vec
[idx
].bv_page
,
1273 md_error(mddev
, rdev
);
1276 while (d
!= r1_bio
->read_disk
) {
1278 d
= conf
->raid_disks
;
1280 if (r1_bio
->bios
[d
]->bi_end_io
!= end_sync_read
)
1282 rdev
= conf
->mirrors
[d
].rdev
;
1283 if (sync_page_io(rdev
->bdev
,
1284 sect
+ rdev
->data_offset
,
1286 bio
->bi_io_vec
[idx
].bv_page
,
1288 md_error(mddev
, rdev
);
1291 char b
[BDEVNAME_SIZE
];
1292 /* Cannot read from anywhere, array is toast */
1293 md_error(mddev
, conf
->mirrors
[r1_bio
->read_disk
].rdev
);
1294 printk(KERN_ALERT
"raid1: %s: unrecoverable I/O read error"
1295 " for block %llu\n",
1296 bdevname(bio
->bi_bdev
,b
),
1297 (unsigned long long)r1_bio
->sector
);
1298 md_done_sync(mddev
, r1_bio
->sectors
, 0);
1311 atomic_set(&r1_bio
->remaining
, 1);
1312 for (i
= 0; i
< disks
; i
++) {
1313 wbio
= r1_bio
->bios
[i
];
1314 if (wbio
->bi_end_io
== NULL
||
1315 (wbio
->bi_end_io
== end_sync_read
&&
1316 (i
== r1_bio
->read_disk
||
1317 !test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))))
1320 wbio
->bi_rw
= WRITE
;
1321 wbio
->bi_end_io
= end_sync_write
;
1322 atomic_inc(&r1_bio
->remaining
);
1323 md_sync_acct(conf
->mirrors
[i
].rdev
->bdev
, wbio
->bi_size
>> 9);
1325 generic_make_request(wbio
);
1328 if (atomic_dec_and_test(&r1_bio
->remaining
)) {
1329 /* if we're here, all write(s) have completed, so clean up */
1330 md_done_sync(mddev
, r1_bio
->sectors
, 1);
1336 * This is a kernel thread which:
1338 * 1. Retries failed read operations on working mirrors.
1339 * 2. Updates the raid superblock when problems encounter.
1340 * 3. Performs writes following reads for array syncronising.
1343 static void raid1d(mddev_t
*mddev
)
1347 unsigned long flags
;
1348 conf_t
*conf
= mddev_to_conf(mddev
);
1349 struct list_head
*head
= &conf
->retry_list
;
1353 md_check_recovery(mddev
);
1356 char b
[BDEVNAME_SIZE
];
1357 spin_lock_irqsave(&conf
->device_lock
, flags
);
1359 if (conf
->pending_bio_list
.head
) {
1360 bio
= bio_list_get(&conf
->pending_bio_list
);
1361 blk_remove_plug(mddev
->queue
);
1362 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
1363 /* flush any pending bitmap writes to disk before proceeding w/ I/O */
1364 if (bitmap_unplug(mddev
->bitmap
) != 0)
1365 printk("%s: bitmap file write failed!\n", mdname(mddev
));
1367 while (bio
) { /* submit pending writes */
1368 struct bio
*next
= bio
->bi_next
;
1369 bio
->bi_next
= NULL
;
1370 generic_make_request(bio
);
1378 if (list_empty(head
))
1380 r1_bio
= list_entry(head
->prev
, r1bio_t
, retry_list
);
1381 list_del(head
->prev
);
1383 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
1385 mddev
= r1_bio
->mddev
;
1386 conf
= mddev_to_conf(mddev
);
1387 if (test_bit(R1BIO_IsSync
, &r1_bio
->state
)) {
1388 sync_request_write(mddev
, r1_bio
);
1390 } else if (test_bit(R1BIO_BarrierRetry
, &r1_bio
->state
)) {
1391 /* some requests in the r1bio were BIO_RW_BARRIER
1392 * requests which failed with -ENOTSUPP. Hohumm..
1393 * Better resubmit without the barrier.
1394 * We know which devices to resubmit for, because
1395 * all others have had their bios[] entry cleared.
1398 clear_bit(R1BIO_BarrierRetry
, &r1_bio
->state
);
1399 clear_bit(R1BIO_Barrier
, &r1_bio
->state
);
1400 for (i
=0; i
< conf
->raid_disks
; i
++)
1401 if (r1_bio
->bios
[i
]) {
1402 struct bio_vec
*bvec
;
1405 bio
= bio_clone(r1_bio
->master_bio
, GFP_NOIO
);
1406 /* copy pages from the failed bio, as
1407 * this might be a write-behind device */
1408 __bio_for_each_segment(bvec
, bio
, j
, 0)
1409 bvec
->bv_page
= bio_iovec_idx(r1_bio
->bios
[i
], j
)->bv_page
;
1410 bio_put(r1_bio
->bios
[i
]);
1411 bio
->bi_sector
= r1_bio
->sector
+
1412 conf
->mirrors
[i
].rdev
->data_offset
;
1413 bio
->bi_bdev
= conf
->mirrors
[i
].rdev
->bdev
;
1414 bio
->bi_end_io
= raid1_end_write_request
;
1416 bio
->bi_private
= r1_bio
;
1417 r1_bio
->bios
[i
] = bio
;
1418 generic_make_request(bio
);
1423 /* we got a read error. Maybe the drive is bad. Maybe just
1424 * the block and we can fix it.
1425 * We freeze all other IO, and try reading the block from
1426 * other devices. When we find one, we re-write
1427 * and check it that fixes the read error.
1428 * This is all done synchronously while the array is
1431 sector_t sect
= r1_bio
->sector
;
1432 int sectors
= r1_bio
->sectors
;
1434 if (mddev
->ro
== 0) while(sectors
) {
1436 int d
= r1_bio
->read_disk
;
1439 if (s
> (PAGE_SIZE
>>9))
1443 rdev
= conf
->mirrors
[d
].rdev
;
1445 test_bit(In_sync
, &rdev
->flags
) &&
1446 sync_page_io(rdev
->bdev
,
1447 sect
+ rdev
->data_offset
,
1449 conf
->tmppage
, READ
))
1453 if (d
== conf
->raid_disks
)
1456 } while (!success
&& d
!= r1_bio
->read_disk
);
1459 /* write it back and re-read */
1461 while (d
!= r1_bio
->read_disk
) {
1463 d
= conf
->raid_disks
;
1465 rdev
= conf
->mirrors
[d
].rdev
;
1467 test_bit(In_sync
, &rdev
->flags
)) {
1468 if (sync_page_io(rdev
->bdev
,
1469 sect
+ rdev
->data_offset
,
1470 s
<<9, conf
->tmppage
, WRITE
) == 0)
1471 /* Well, this device is dead */
1472 md_error(mddev
, rdev
);
1476 while (d
!= r1_bio
->read_disk
) {
1478 d
= conf
->raid_disks
;
1480 rdev
= conf
->mirrors
[d
].rdev
;
1482 test_bit(In_sync
, &rdev
->flags
)) {
1483 if (sync_page_io(rdev
->bdev
,
1484 sect
+ rdev
->data_offset
,
1485 s
<<9, conf
->tmppage
, READ
) == 0)
1486 /* Well, this device is dead */
1487 md_error(mddev
, rdev
);
1491 /* Cannot read from anywhere -- bye bye array */
1492 md_error(mddev
, conf
->mirrors
[r1_bio
->read_disk
].rdev
);
1499 unfreeze_array(conf
);
1501 bio
= r1_bio
->bios
[r1_bio
->read_disk
];
1502 if ((disk
=read_balance(conf
, r1_bio
)) == -1) {
1503 printk(KERN_ALERT
"raid1: %s: unrecoverable I/O"
1504 " read error for block %llu\n",
1505 bdevname(bio
->bi_bdev
,b
),
1506 (unsigned long long)r1_bio
->sector
);
1507 raid_end_bio_io(r1_bio
);
1509 r1_bio
->bios
[r1_bio
->read_disk
] =
1510 mddev
->ro
? IO_BLOCKED
: NULL
;
1511 r1_bio
->read_disk
= disk
;
1513 bio
= bio_clone(r1_bio
->master_bio
, GFP_NOIO
);
1514 r1_bio
->bios
[r1_bio
->read_disk
] = bio
;
1515 rdev
= conf
->mirrors
[disk
].rdev
;
1516 if (printk_ratelimit())
1517 printk(KERN_ERR
"raid1: %s: redirecting sector %llu to"
1518 " another mirror\n",
1519 bdevname(rdev
->bdev
,b
),
1520 (unsigned long long)r1_bio
->sector
);
1521 bio
->bi_sector
= r1_bio
->sector
+ rdev
->data_offset
;
1522 bio
->bi_bdev
= rdev
->bdev
;
1523 bio
->bi_end_io
= raid1_end_read_request
;
1525 bio
->bi_private
= r1_bio
;
1527 generic_make_request(bio
);
1531 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
1533 unplug_slaves(mddev
);
1537 static int init_resync(conf_t
*conf
)
1541 buffs
= RESYNC_WINDOW
/ RESYNC_BLOCK_SIZE
;
1542 if (conf
->r1buf_pool
)
1544 conf
->r1buf_pool
= mempool_create(buffs
, r1buf_pool_alloc
, r1buf_pool_free
,
1546 if (!conf
->r1buf_pool
)
1548 conf
->next_resync
= 0;
1553 * perform a "sync" on one "block"
1555 * We need to make sure that no normal I/O request - particularly write
1556 * requests - conflict with active sync requests.
1558 * This is achieved by tracking pending requests and a 'barrier' concept
1559 * that can be installed to exclude normal IO requests.
1562 static sector_t
sync_request(mddev_t
*mddev
, sector_t sector_nr
, int *skipped
, int go_faster
)
1564 conf_t
*conf
= mddev_to_conf(mddev
);
1567 sector_t max_sector
, nr_sectors
;
1571 int write_targets
= 0, read_targets
= 0;
1573 int still_degraded
= 0;
1575 if (!conf
->r1buf_pool
)
1578 printk("sync start - bitmap %p\n", mddev->bitmap);
1580 if (init_resync(conf
))
1584 max_sector
= mddev
->size
<< 1;
1585 if (sector_nr
>= max_sector
) {
1586 /* If we aborted, we need to abort the
1587 * sync on the 'current' bitmap chunk (there will
1588 * only be one in raid1 resync.
1589 * We can find the current addess in mddev->curr_resync
1591 if (mddev
->curr_resync
< max_sector
) /* aborted */
1592 bitmap_end_sync(mddev
->bitmap
, mddev
->curr_resync
,
1594 else /* completed sync */
1597 bitmap_close_sync(mddev
->bitmap
);
1602 /* before building a request, check if we can skip these blocks..
1603 * This call the bitmap_start_sync doesn't actually record anything
1605 if (!bitmap_start_sync(mddev
->bitmap
, sector_nr
, &sync_blocks
, 1) &&
1606 !conf
->fullsync
&& !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
1607 /* We can skip this block, and probably several more */
1612 * If there is non-resync activity waiting for a turn,
1613 * and resync is going fast enough,
1614 * then let it though before starting on this new sync request.
1616 if (!go_faster
&& conf
->nr_waiting
)
1617 msleep_interruptible(1000);
1619 raise_barrier(conf
);
1621 conf
->next_resync
= sector_nr
;
1623 r1_bio
= mempool_alloc(conf
->r1buf_pool
, GFP_NOIO
);
1626 * If we get a correctably read error during resync or recovery,
1627 * we might want to read from a different device. So we
1628 * flag all drives that could conceivably be read from for READ,
1629 * and any others (which will be non-In_sync devices) for WRITE.
1630 * If a read fails, we try reading from something else for which READ
1634 r1_bio
->mddev
= mddev
;
1635 r1_bio
->sector
= sector_nr
;
1637 set_bit(R1BIO_IsSync
, &r1_bio
->state
);
1639 for (i
=0; i
< conf
->raid_disks
; i
++) {
1641 bio
= r1_bio
->bios
[i
];
1643 /* take from bio_init */
1644 bio
->bi_next
= NULL
;
1645 bio
->bi_flags
|= 1 << BIO_UPTODATE
;
1649 bio
->bi_phys_segments
= 0;
1650 bio
->bi_hw_segments
= 0;
1652 bio
->bi_end_io
= NULL
;
1653 bio
->bi_private
= NULL
;
1655 rdev
= rcu_dereference(conf
->mirrors
[i
].rdev
);
1657 test_bit(Faulty
, &rdev
->flags
)) {
1660 } else if (!test_bit(In_sync
, &rdev
->flags
)) {
1662 bio
->bi_end_io
= end_sync_write
;
1665 /* may need to read from here */
1667 bio
->bi_end_io
= end_sync_read
;
1668 if (test_bit(WriteMostly
, &rdev
->flags
)) {
1677 atomic_inc(&rdev
->nr_pending
);
1678 bio
->bi_sector
= sector_nr
+ rdev
->data_offset
;
1679 bio
->bi_bdev
= rdev
->bdev
;
1680 bio
->bi_private
= r1_bio
;
1685 r1_bio
->read_disk
= disk
;
1687 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) && read_targets
> 0)
1688 /* extra read targets are also write targets */
1689 write_targets
+= read_targets
-1;
1691 if (write_targets
== 0 || read_targets
== 0) {
1692 /* There is nowhere to write, so all non-sync
1693 * drives must be failed - so we are finished
1695 sector_t rv
= max_sector
- sector_nr
;
1705 int len
= PAGE_SIZE
;
1706 if (sector_nr
+ (len
>>9) > max_sector
)
1707 len
= (max_sector
- sector_nr
) << 9;
1710 if (sync_blocks
== 0) {
1711 if (!bitmap_start_sync(mddev
->bitmap
, sector_nr
,
1712 &sync_blocks
, still_degraded
) &&
1714 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
1716 if (sync_blocks
< (PAGE_SIZE
>>9))
1718 if (len
> (sync_blocks
<<9))
1719 len
= sync_blocks
<<9;
1722 for (i
=0 ; i
< conf
->raid_disks
; i
++) {
1723 bio
= r1_bio
->bios
[i
];
1724 if (bio
->bi_end_io
) {
1725 page
= bio
->bi_io_vec
[bio
->bi_vcnt
].bv_page
;
1726 if (bio_add_page(bio
, page
, len
, 0) == 0) {
1728 bio
->bi_io_vec
[bio
->bi_vcnt
].bv_page
= page
;
1731 bio
= r1_bio
->bios
[i
];
1732 if (bio
->bi_end_io
==NULL
)
1734 /* remove last page from this bio */
1736 bio
->bi_size
-= len
;
1737 bio
->bi_flags
&= ~(1<< BIO_SEG_VALID
);
1743 nr_sectors
+= len
>>9;
1744 sector_nr
+= len
>>9;
1745 sync_blocks
-= (len
>>9);
1746 } while (r1_bio
->bios
[disk
]->bi_vcnt
< RESYNC_PAGES
);
1748 r1_bio
->sectors
= nr_sectors
;
1750 /* For a user-requested sync, we read all readable devices and do a
1753 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
1754 atomic_set(&r1_bio
->remaining
, read_targets
);
1755 for (i
=0; i
<conf
->raid_disks
; i
++) {
1756 bio
= r1_bio
->bios
[i
];
1757 if (bio
->bi_end_io
== end_sync_read
) {
1758 md_sync_acct(conf
->mirrors
[i
].rdev
->bdev
, nr_sectors
);
1759 generic_make_request(bio
);
1763 atomic_set(&r1_bio
->remaining
, 1);
1764 bio
= r1_bio
->bios
[r1_bio
->read_disk
];
1765 md_sync_acct(conf
->mirrors
[r1_bio
->read_disk
].rdev
->bdev
,
1767 generic_make_request(bio
);
1774 static int run(mddev_t
*mddev
)
1778 mirror_info_t
*disk
;
1780 struct list_head
*tmp
;
1782 if (mddev
->level
!= 1) {
1783 printk("raid1: %s: raid level not set to mirroring (%d)\n",
1784 mdname(mddev
), mddev
->level
);
1788 * copy the already verified devices into our private RAID1
1789 * bookkeeping area. [whatever we allocate in run(),
1790 * should be freed in stop()]
1792 conf
= kzalloc(sizeof(conf_t
), GFP_KERNEL
);
1793 mddev
->private = conf
;
1797 conf
->mirrors
= kzalloc(sizeof(struct mirror_info
)*mddev
->raid_disks
,
1802 conf
->tmppage
= alloc_page(GFP_KERNEL
);
1806 conf
->poolinfo
= kmalloc(sizeof(*conf
->poolinfo
), GFP_KERNEL
);
1807 if (!conf
->poolinfo
)
1809 conf
->poolinfo
->mddev
= mddev
;
1810 conf
->poolinfo
->raid_disks
= mddev
->raid_disks
;
1811 conf
->r1bio_pool
= mempool_create(NR_RAID1_BIOS
, r1bio_pool_alloc
,
1814 if (!conf
->r1bio_pool
)
1817 ITERATE_RDEV(mddev
, rdev
, tmp
) {
1818 disk_idx
= rdev
->raid_disk
;
1819 if (disk_idx
>= mddev
->raid_disks
1822 disk
= conf
->mirrors
+ disk_idx
;
1826 blk_queue_stack_limits(mddev
->queue
,
1827 rdev
->bdev
->bd_disk
->queue
);
1828 /* as we don't honour merge_bvec_fn, we must never risk
1829 * violating it, so limit ->max_sector to one PAGE, as
1830 * a one page request is never in violation.
1832 if (rdev
->bdev
->bd_disk
->queue
->merge_bvec_fn
&&
1833 mddev
->queue
->max_sectors
> (PAGE_SIZE
>>9))
1834 blk_queue_max_sectors(mddev
->queue
, PAGE_SIZE
>>9);
1836 disk
->head_position
= 0;
1837 if (!test_bit(Faulty
, &rdev
->flags
) && test_bit(In_sync
, &rdev
->flags
))
1838 conf
->working_disks
++;
1840 conf
->raid_disks
= mddev
->raid_disks
;
1841 conf
->mddev
= mddev
;
1842 spin_lock_init(&conf
->device_lock
);
1843 INIT_LIST_HEAD(&conf
->retry_list
);
1844 if (conf
->working_disks
== 1)
1845 mddev
->recovery_cp
= MaxSector
;
1847 spin_lock_init(&conf
->resync_lock
);
1848 init_waitqueue_head(&conf
->wait_barrier
);
1850 bio_list_init(&conf
->pending_bio_list
);
1851 bio_list_init(&conf
->flushing_bio_list
);
1853 if (!conf
->working_disks
) {
1854 printk(KERN_ERR
"raid1: no operational mirrors for %s\n",
1859 mddev
->degraded
= 0;
1860 for (i
= 0; i
< conf
->raid_disks
; i
++) {
1862 disk
= conf
->mirrors
+ i
;
1865 disk
->head_position
= 0;
1871 * find the first working one and use it as a starting point
1872 * to read balancing.
1874 for (j
= 0; j
< conf
->raid_disks
&&
1875 (!conf
->mirrors
[j
].rdev
||
1876 !test_bit(In_sync
, &conf
->mirrors
[j
].rdev
->flags
)) ; j
++)
1878 conf
->last_used
= j
;
1881 mddev
->thread
= md_register_thread(raid1d
, mddev
, "%s_raid1");
1882 if (!mddev
->thread
) {
1884 "raid1: couldn't allocate thread for %s\n",
1890 "raid1: raid set %s active with %d out of %d mirrors\n",
1891 mdname(mddev
), mddev
->raid_disks
- mddev
->degraded
,
1894 * Ok, everything is just fine now
1896 mddev
->array_size
= mddev
->size
;
1898 mddev
->queue
->unplug_fn
= raid1_unplug
;
1899 mddev
->queue
->issue_flush_fn
= raid1_issue_flush
;
1904 printk(KERN_ERR
"raid1: couldn't allocate memory for %s\n",
1909 if (conf
->r1bio_pool
)
1910 mempool_destroy(conf
->r1bio_pool
);
1911 kfree(conf
->mirrors
);
1912 safe_put_page(conf
->tmppage
);
1913 kfree(conf
->poolinfo
);
1915 mddev
->private = NULL
;
1921 static int stop(mddev_t
*mddev
)
1923 conf_t
*conf
= mddev_to_conf(mddev
);
1924 struct bitmap
*bitmap
= mddev
->bitmap
;
1925 int behind_wait
= 0;
1927 /* wait for behind writes to complete */
1928 while (bitmap
&& atomic_read(&bitmap
->behind_writes
) > 0) {
1930 printk(KERN_INFO
"raid1: behind writes in progress on device %s, waiting to stop (%d)\n", mdname(mddev
), behind_wait
);
1931 set_current_state(TASK_UNINTERRUPTIBLE
);
1932 schedule_timeout(HZ
); /* wait a second */
1933 /* need to kick something here to make sure I/O goes? */
1936 md_unregister_thread(mddev
->thread
);
1937 mddev
->thread
= NULL
;
1938 blk_sync_queue(mddev
->queue
); /* the unplug fn references 'conf'*/
1939 if (conf
->r1bio_pool
)
1940 mempool_destroy(conf
->r1bio_pool
);
1941 kfree(conf
->mirrors
);
1942 kfree(conf
->poolinfo
);
1944 mddev
->private = NULL
;
1948 static int raid1_resize(mddev_t
*mddev
, sector_t sectors
)
1950 /* no resync is happening, and there is enough space
1951 * on all devices, so we can resize.
1952 * We need to make sure resync covers any new space.
1953 * If the array is shrinking we should possibly wait until
1954 * any io in the removed space completes, but it hardly seems
1957 mddev
->array_size
= sectors
>>1;
1958 set_capacity(mddev
->gendisk
, mddev
->array_size
<< 1);
1960 if (mddev
->array_size
> mddev
->size
&& mddev
->recovery_cp
== MaxSector
) {
1961 mddev
->recovery_cp
= mddev
->size
<< 1;
1962 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
1964 mddev
->size
= mddev
->array_size
;
1965 mddev
->resync_max_sectors
= sectors
;
1969 static int raid1_reshape(mddev_t
*mddev
, int raid_disks
)
1972 * 1/ resize the r1bio_pool
1973 * 2/ resize conf->mirrors
1975 * We allocate a new r1bio_pool if we can.
1976 * Then raise a device barrier and wait until all IO stops.
1977 * Then resize conf->mirrors and swap in the new r1bio pool.
1979 * At the same time, we "pack" the devices so that all the missing
1980 * devices have the higher raid_disk numbers.
1982 mempool_t
*newpool
, *oldpool
;
1983 struct pool_info
*newpoolinfo
;
1984 mirror_info_t
*newmirrors
;
1985 conf_t
*conf
= mddev_to_conf(mddev
);
1990 if (raid_disks
< conf
->raid_disks
) {
1992 for (d
= 0; d
< conf
->raid_disks
; d
++)
1993 if (conf
->mirrors
[d
].rdev
)
1995 if (cnt
> raid_disks
)
1999 newpoolinfo
= kmalloc(sizeof(*newpoolinfo
), GFP_KERNEL
);
2002 newpoolinfo
->mddev
= mddev
;
2003 newpoolinfo
->raid_disks
= raid_disks
;
2005 newpool
= mempool_create(NR_RAID1_BIOS
, r1bio_pool_alloc
,
2006 r1bio_pool_free
, newpoolinfo
);
2011 newmirrors
= kzalloc(sizeof(struct mirror_info
) * raid_disks
, GFP_KERNEL
);
2014 mempool_destroy(newpool
);
2018 raise_barrier(conf
);
2020 /* ok, everything is stopped */
2021 oldpool
= conf
->r1bio_pool
;
2022 conf
->r1bio_pool
= newpool
;
2024 for (d
=d2
=0; d
< conf
->raid_disks
; d
++)
2025 if (conf
->mirrors
[d
].rdev
) {
2026 conf
->mirrors
[d
].rdev
->raid_disk
= d2
;
2027 newmirrors
[d2
++].rdev
= conf
->mirrors
[d
].rdev
;
2029 kfree(conf
->mirrors
);
2030 conf
->mirrors
= newmirrors
;
2031 kfree(conf
->poolinfo
);
2032 conf
->poolinfo
= newpoolinfo
;
2034 mddev
->degraded
+= (raid_disks
- conf
->raid_disks
);
2035 conf
->raid_disks
= mddev
->raid_disks
= raid_disks
;
2037 conf
->last_used
= 0; /* just make sure it is in-range */
2038 lower_barrier(conf
);
2040 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2041 md_wakeup_thread(mddev
->thread
);
2043 mempool_destroy(oldpool
);
2047 static void raid1_quiesce(mddev_t
*mddev
, int state
)
2049 conf_t
*conf
= mddev_to_conf(mddev
);
2053 raise_barrier(conf
);
2056 lower_barrier(conf
);
2062 static struct mdk_personality raid1_personality
=
2066 .owner
= THIS_MODULE
,
2067 .make_request
= make_request
,
2071 .error_handler
= error
,
2072 .hot_add_disk
= raid1_add_disk
,
2073 .hot_remove_disk
= raid1_remove_disk
,
2074 .spare_active
= raid1_spare_active
,
2075 .sync_request
= sync_request
,
2076 .resize
= raid1_resize
,
2077 .reshape
= raid1_reshape
,
2078 .quiesce
= raid1_quiesce
,
2081 static int __init
raid_init(void)
2083 return register_md_personality(&raid1_personality
);
2086 static void raid_exit(void)
2088 unregister_md_personality(&raid1_personality
);
2091 module_init(raid_init
);
2092 module_exit(raid_exit
);
2093 MODULE_LICENSE("GPL");
2094 MODULE_ALIAS("md-personality-3"); /* RAID1 */
2095 MODULE_ALIAS("md-raid1");
2096 MODULE_ALIAS("md-level-1");