2 * multipath.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 * MULTIPATH management functions.
10 * derived from raid1.c.
12 * This program is free software; you can redistribute it and/or modify
13 * it under the terms of the GNU General Public License as published by
14 * the Free Software Foundation; either version 2, or (at your option)
17 * You should have received a copy of the GNU General Public License
18 * (for example /usr/src/linux/COPYING); if not, write to the Free
19 * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
22 #include <linux/module.h>
23 #include <linux/slab.h>
24 #include <linux/spinlock.h>
25 #include <linux/raid/multipath.h>
26 #include <linux/buffer_head.h>
27 #include <asm/atomic.h>
29 #define MAJOR_NR MD_MAJOR
31 #define MD_PERSONALITY
33 #define MAX_WORK_PER_DISK 128
35 #define NR_RESERVED_BUFS 32
38 static int multipath_map (multipath_conf_t
*conf
)
40 int i
, disks
= conf
->raid_disks
;
43 * Later we do read balancing on the read side
44 * now we use the first available disk.
48 for (i
= 0; i
< disks
; i
++) {
49 mdk_rdev_t
*rdev
= rcu_dereference(conf
->multipaths
[i
].rdev
);
50 if (rdev
&& test_bit(In_sync
, &rdev
->flags
)) {
51 atomic_inc(&rdev
->nr_pending
);
58 printk(KERN_ERR
"multipath_map(): no more operational IO paths?\n");
62 static void multipath_reschedule_retry (struct multipath_bh
*mp_bh
)
65 mddev_t
*mddev
= mp_bh
->mddev
;
66 multipath_conf_t
*conf
= mddev_to_conf(mddev
);
68 spin_lock_irqsave(&conf
->device_lock
, flags
);
69 list_add(&mp_bh
->retry_list
, &conf
->retry_list
);
70 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
71 md_wakeup_thread(mddev
->thread
);
76 * multipath_end_bh_io() is called when we have finished servicing a multipathed
77 * operation and are ready to return a success/failure code to the buffer
80 static void multipath_end_bh_io (struct multipath_bh
*mp_bh
, int err
)
82 struct bio
*bio
= mp_bh
->master_bio
;
83 multipath_conf_t
*conf
= mddev_to_conf(mp_bh
->mddev
);
85 bio_endio(bio
, bio
->bi_size
, err
);
86 mempool_free(mp_bh
, conf
->pool
);
89 static int multipath_end_request(struct bio
*bio
, unsigned int bytes_done
,
92 int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
93 struct multipath_bh
* mp_bh
= (struct multipath_bh
*)(bio
->bi_private
);
94 multipath_conf_t
*conf
= mddev_to_conf(mp_bh
->mddev
);
95 mdk_rdev_t
*rdev
= conf
->multipaths
[mp_bh
->path
].rdev
;
101 multipath_end_bh_io(mp_bh
, 0);
102 else if (!bio_rw_ahead(bio
)) {
106 char b
[BDEVNAME_SIZE
];
107 md_error (mp_bh
->mddev
, rdev
);
108 printk(KERN_ERR
"multipath: %s: rescheduling sector %llu\n",
109 bdevname(rdev
->bdev
,b
),
110 (unsigned long long)bio
->bi_sector
);
111 multipath_reschedule_retry(mp_bh
);
113 multipath_end_bh_io(mp_bh
, error
);
114 rdev_dec_pending(rdev
, conf
->mddev
);
118 static void unplug_slaves(mddev_t
*mddev
)
120 multipath_conf_t
*conf
= mddev_to_conf(mddev
);
124 for (i
=0; i
<mddev
->raid_disks
; i
++) {
125 mdk_rdev_t
*rdev
= rcu_dereference(conf
->multipaths
[i
].rdev
);
126 if (rdev
&& !test_bit(Faulty
, &rdev
->flags
)
127 && atomic_read(&rdev
->nr_pending
)) {
128 request_queue_t
*r_queue
= bdev_get_queue(rdev
->bdev
);
130 atomic_inc(&rdev
->nr_pending
);
133 if (r_queue
->unplug_fn
)
134 r_queue
->unplug_fn(r_queue
);
136 rdev_dec_pending(rdev
, mddev
);
143 static void multipath_unplug(request_queue_t
*q
)
145 unplug_slaves(q
->queuedata
);
149 static int multipath_make_request (request_queue_t
*q
, struct bio
* bio
)
151 mddev_t
*mddev
= q
->queuedata
;
152 multipath_conf_t
*conf
= mddev_to_conf(mddev
);
153 struct multipath_bh
* mp_bh
;
154 struct multipath_info
*multipath
;
155 const int rw
= bio_data_dir(bio
);
157 if (unlikely(bio_barrier(bio
))) {
158 bio_endio(bio
, bio
->bi_size
, -EOPNOTSUPP
);
162 mp_bh
= mempool_alloc(conf
->pool
, GFP_NOIO
);
164 mp_bh
->master_bio
= bio
;
165 mp_bh
->mddev
= mddev
;
167 disk_stat_inc(mddev
->gendisk
, ios
[rw
]);
168 disk_stat_add(mddev
->gendisk
, sectors
[rw
], bio_sectors(bio
));
170 mp_bh
->path
= multipath_map(conf
);
171 if (mp_bh
->path
< 0) {
172 bio_endio(bio
, bio
->bi_size
, -EIO
);
173 mempool_free(mp_bh
, conf
->pool
);
176 multipath
= conf
->multipaths
+ mp_bh
->path
;
179 mp_bh
->bio
.bi_sector
+= multipath
->rdev
->data_offset
;
180 mp_bh
->bio
.bi_bdev
= multipath
->rdev
->bdev
;
181 mp_bh
->bio
.bi_rw
|= (1 << BIO_RW_FAILFAST
);
182 mp_bh
->bio
.bi_end_io
= multipath_end_request
;
183 mp_bh
->bio
.bi_private
= mp_bh
;
184 generic_make_request(&mp_bh
->bio
);
188 static void multipath_status (struct seq_file
*seq
, mddev_t
*mddev
)
190 multipath_conf_t
*conf
= mddev_to_conf(mddev
);
193 seq_printf (seq
, " [%d/%d] [", conf
->raid_disks
,
194 conf
->working_disks
);
195 for (i
= 0; i
< conf
->raid_disks
; i
++)
196 seq_printf (seq
, "%s",
197 conf
->multipaths
[i
].rdev
&&
198 test_bit(In_sync
, &conf
->multipaths
[i
].rdev
->flags
) ? "U" : "_");
199 seq_printf (seq
, "]");
202 static int multipath_issue_flush(request_queue_t
*q
, struct gendisk
*disk
,
203 sector_t
*error_sector
)
205 mddev_t
*mddev
= q
->queuedata
;
206 multipath_conf_t
*conf
= mddev_to_conf(mddev
);
210 for (i
=0; i
<mddev
->raid_disks
&& ret
== 0; i
++) {
211 mdk_rdev_t
*rdev
= rcu_dereference(conf
->multipaths
[i
].rdev
);
212 if (rdev
&& !test_bit(Faulty
, &rdev
->flags
)) {
213 struct block_device
*bdev
= rdev
->bdev
;
214 request_queue_t
*r_queue
= bdev_get_queue(bdev
);
216 if (!r_queue
->issue_flush_fn
)
219 atomic_inc(&rdev
->nr_pending
);
221 ret
= r_queue
->issue_flush_fn(r_queue
, bdev
->bd_disk
,
223 rdev_dec_pending(rdev
, mddev
);
231 static int multipath_congested(void *data
, int bits
)
233 mddev_t
*mddev
= data
;
234 multipath_conf_t
*conf
= mddev_to_conf(mddev
);
238 for (i
= 0; i
< mddev
->raid_disks
; i
++) {
239 mdk_rdev_t
*rdev
= rcu_dereference(conf
->multipaths
[i
].rdev
);
240 if (rdev
&& !test_bit(Faulty
, &rdev
->flags
)) {
241 request_queue_t
*q
= bdev_get_queue(rdev
->bdev
);
243 ret
|= bdi_congested(&q
->backing_dev_info
, bits
);
244 /* Just like multipath_map, we just check the
245 * first available device
255 * Careful, this can execute in IRQ contexts as well!
257 static void multipath_error (mddev_t
*mddev
, mdk_rdev_t
*rdev
)
259 multipath_conf_t
*conf
= mddev_to_conf(mddev
);
261 if (conf
->working_disks
<= 1) {
263 * Uh oh, we can do nothing if this is our last path, but
264 * first check if this is a queued request for a device
265 * which has just failed.
268 "multipath: only one IO path left and IO error.\n");
269 /* leave it active... it's all we have */
272 * Mark disk as unusable
274 if (!test_bit(Faulty
, &rdev
->flags
)) {
275 char b
[BDEVNAME_SIZE
];
276 clear_bit(In_sync
, &rdev
->flags
);
277 set_bit(Faulty
, &rdev
->flags
);
278 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
279 conf
->working_disks
--;
280 printk(KERN_ALERT
"multipath: IO failure on %s,"
281 " disabling IO path. \n Operation continuing"
282 " on %d IO paths.\n",
283 bdevname (rdev
->bdev
,b
),
284 conf
->working_disks
);
289 static void print_multipath_conf (multipath_conf_t
*conf
)
292 struct multipath_info
*tmp
;
294 printk("MULTIPATH conf printout:\n");
296 printk("(conf==NULL)\n");
299 printk(" --- wd:%d rd:%d\n", conf
->working_disks
,
302 for (i
= 0; i
< conf
->raid_disks
; i
++) {
303 char b
[BDEVNAME_SIZE
];
304 tmp
= conf
->multipaths
+ i
;
306 printk(" disk%d, o:%d, dev:%s\n",
307 i
,!test_bit(Faulty
, &tmp
->rdev
->flags
),
308 bdevname(tmp
->rdev
->bdev
,b
));
313 static int multipath_add_disk(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
315 multipath_conf_t
*conf
= mddev
->private;
316 struct request_queue
*q
;
319 struct multipath_info
*p
;
321 print_multipath_conf(conf
);
323 for (path
=0; path
<mddev
->raid_disks
; path
++)
324 if ((p
=conf
->multipaths
+path
)->rdev
== NULL
) {
325 q
= rdev
->bdev
->bd_disk
->queue
;
326 blk_queue_stack_limits(mddev
->queue
, q
);
328 /* as we don't honour merge_bvec_fn, we must never risk
329 * violating it, so limit ->max_sector to one PAGE, as
330 * a one page request is never in violation.
331 * (Note: it is very unlikely that a device with
332 * merge_bvec_fn will be involved in multipath.)
334 if (q
->merge_bvec_fn
&&
335 mddev
->queue
->max_sectors
> (PAGE_SIZE
>>9))
336 blk_queue_max_sectors(mddev
->queue
, PAGE_SIZE
>>9);
338 conf
->working_disks
++;
339 rdev
->raid_disk
= path
;
340 set_bit(In_sync
, &rdev
->flags
);
341 rcu_assign_pointer(p
->rdev
, rdev
);
345 print_multipath_conf(conf
);
349 static int multipath_remove_disk(mddev_t
*mddev
, int number
)
351 multipath_conf_t
*conf
= mddev
->private;
354 struct multipath_info
*p
= conf
->multipaths
+ number
;
356 print_multipath_conf(conf
);
360 if (test_bit(In_sync
, &rdev
->flags
) ||
361 atomic_read(&rdev
->nr_pending
)) {
362 printk(KERN_ERR
"hot-remove-disk, slot %d is identified" " but is still operational!\n", number
);
368 if (atomic_read(&rdev
->nr_pending
)) {
369 /* lost the race, try later */
376 print_multipath_conf(conf
);
383 * This is a kernel thread which:
385 * 1. Retries failed read operations on working multipaths.
386 * 2. Updates the raid superblock when problems encounter.
387 * 3. Performs writes following reads for array syncronising.
390 static void multipathd (mddev_t
*mddev
)
392 struct multipath_bh
*mp_bh
;
395 multipath_conf_t
*conf
= mddev_to_conf(mddev
);
396 struct list_head
*head
= &conf
->retry_list
;
398 md_check_recovery(mddev
);
400 char b
[BDEVNAME_SIZE
];
401 spin_lock_irqsave(&conf
->device_lock
, flags
);
402 if (list_empty(head
))
404 mp_bh
= list_entry(head
->prev
, struct multipath_bh
, retry_list
);
405 list_del(head
->prev
);
406 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
409 bio
->bi_sector
= mp_bh
->master_bio
->bi_sector
;
411 if ((mp_bh
->path
= multipath_map (conf
))<0) {
412 printk(KERN_ALERT
"multipath: %s: unrecoverable IO read"
413 " error for block %llu\n",
414 bdevname(bio
->bi_bdev
,b
),
415 (unsigned long long)bio
->bi_sector
);
416 multipath_end_bh_io(mp_bh
, -EIO
);
418 printk(KERN_ERR
"multipath: %s: redirecting sector %llu"
419 " to another IO path\n",
420 bdevname(bio
->bi_bdev
,b
),
421 (unsigned long long)bio
->bi_sector
);
422 *bio
= *(mp_bh
->master_bio
);
423 bio
->bi_sector
+= conf
->multipaths
[mp_bh
->path
].rdev
->data_offset
;
424 bio
->bi_bdev
= conf
->multipaths
[mp_bh
->path
].rdev
->bdev
;
425 bio
->bi_rw
|= (1 << BIO_RW_FAILFAST
);
426 bio
->bi_end_io
= multipath_end_request
;
427 bio
->bi_private
= mp_bh
;
428 generic_make_request(bio
);
431 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
434 static int multipath_run (mddev_t
*mddev
)
436 multipath_conf_t
*conf
;
438 struct multipath_info
*disk
;
440 struct list_head
*tmp
;
442 if (mddev
->level
!= LEVEL_MULTIPATH
) {
443 printk("multipath: %s: raid level not set to multipath IO (%d)\n",
444 mdname(mddev
), mddev
->level
);
448 * copy the already verified devices into our private MULTIPATH
449 * bookkeeping area. [whatever we allocate in multipath_run(),
450 * should be freed in multipath_stop()]
453 conf
= kzalloc(sizeof(multipath_conf_t
), GFP_KERNEL
);
454 mddev
->private = conf
;
457 "multipath: couldn't allocate memory for %s\n",
462 conf
->multipaths
= kzalloc(sizeof(struct multipath_info
)*mddev
->raid_disks
,
464 if (!conf
->multipaths
) {
466 "multipath: couldn't allocate memory for %s\n",
471 conf
->working_disks
= 0;
472 ITERATE_RDEV(mddev
,rdev
,tmp
) {
473 disk_idx
= rdev
->raid_disk
;
475 disk_idx
>= mddev
->raid_disks
)
478 disk
= conf
->multipaths
+ disk_idx
;
481 blk_queue_stack_limits(mddev
->queue
,
482 rdev
->bdev
->bd_disk
->queue
);
483 /* as we don't honour merge_bvec_fn, we must never risk
484 * violating it, not that we ever expect a device with
485 * a merge_bvec_fn to be involved in multipath */
486 if (rdev
->bdev
->bd_disk
->queue
->merge_bvec_fn
&&
487 mddev
->queue
->max_sectors
> (PAGE_SIZE
>>9))
488 blk_queue_max_sectors(mddev
->queue
, PAGE_SIZE
>>9);
490 if (!test_bit(Faulty
, &rdev
->flags
))
491 conf
->working_disks
++;
494 conf
->raid_disks
= mddev
->raid_disks
;
496 spin_lock_init(&conf
->device_lock
);
497 INIT_LIST_HEAD(&conf
->retry_list
);
499 if (!conf
->working_disks
) {
500 printk(KERN_ERR
"multipath: no operational IO paths for %s\n",
504 mddev
->degraded
= conf
->raid_disks
= conf
->working_disks
;
506 conf
->pool
= mempool_create_kzalloc_pool(NR_RESERVED_BUFS
,
507 sizeof(struct multipath_bh
));
508 if (conf
->pool
== NULL
) {
510 "multipath: couldn't allocate memory for %s\n",
516 mddev
->thread
= md_register_thread(multipathd
, mddev
, "%s_multipath");
517 if (!mddev
->thread
) {
518 printk(KERN_ERR
"multipath: couldn't allocate thread"
519 " for %s\n", mdname(mddev
));
525 "multipath: array %s active with %d out of %d IO paths\n",
526 mdname(mddev
), conf
->working_disks
, mddev
->raid_disks
);
528 * Ok, everything is just fine now
530 mddev
->array_size
= mddev
->size
;
532 mddev
->queue
->unplug_fn
= multipath_unplug
;
533 mddev
->queue
->issue_flush_fn
= multipath_issue_flush
;
534 mddev
->queue
->backing_dev_info
.congested_fn
= multipath_congested
;
535 mddev
->queue
->backing_dev_info
.congested_data
= mddev
;
541 mempool_destroy(conf
->pool
);
542 kfree(conf
->multipaths
);
544 mddev
->private = NULL
;
550 static int multipath_stop (mddev_t
*mddev
)
552 multipath_conf_t
*conf
= mddev_to_conf(mddev
);
554 md_unregister_thread(mddev
->thread
);
555 mddev
->thread
= NULL
;
556 blk_sync_queue(mddev
->queue
); /* the unplug fn references 'conf'*/
557 mempool_destroy(conf
->pool
);
558 kfree(conf
->multipaths
);
560 mddev
->private = NULL
;
564 static struct mdk_personality multipath_personality
=
567 .level
= LEVEL_MULTIPATH
,
568 .owner
= THIS_MODULE
,
569 .make_request
= multipath_make_request
,
570 .run
= multipath_run
,
571 .stop
= multipath_stop
,
572 .status
= multipath_status
,
573 .error_handler
= multipath_error
,
574 .hot_add_disk
= multipath_add_disk
,
575 .hot_remove_disk
= multipath_remove_disk
,
578 static int __init
multipath_init (void)
580 return register_md_personality (&multipath_personality
);
583 static void __exit
multipath_exit (void)
585 unregister_md_personality (&multipath_personality
);
588 module_init(multipath_init
);
589 module_exit(multipath_exit
);
590 MODULE_LICENSE("GPL");
591 MODULE_ALIAS("md-personality-7"); /* MULTIPATH */
592 MODULE_ALIAS("md-multipath");
593 MODULE_ALIAS("md-level--4");