[ARM] Support register switch in nommu mode
[linux-2.6/verdex.git] / drivers / md / multipath.c
blob96f7af4ae400ad4eaa2dd7d47d9e1a1ce1ca3efa
1 /*
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)
15 * any later version.
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
30 #define MD_DRIVER
31 #define MD_PERSONALITY
33 #define MAX_WORK_PER_DISK 128
35 #define NR_RESERVED_BUFS 32
38 static void *mp_pool_alloc(gfp_t gfp_flags, void *data)
40 struct multipath_bh *mpb;
41 mpb = kzalloc(sizeof(*mpb), gfp_flags);
42 return mpb;
45 static void mp_pool_free(void *mpb, void *data)
47 kfree(mpb);
50 static int multipath_map (multipath_conf_t *conf)
52 int i, disks = conf->raid_disks;
55 * Later we do read balancing on the read side
56 * now we use the first available disk.
59 rcu_read_lock();
60 for (i = 0; i < disks; i++) {
61 mdk_rdev_t *rdev = rcu_dereference(conf->multipaths[i].rdev);
62 if (rdev && test_bit(In_sync, &rdev->flags)) {
63 atomic_inc(&rdev->nr_pending);
64 rcu_read_unlock();
65 return i;
68 rcu_read_unlock();
70 printk(KERN_ERR "multipath_map(): no more operational IO paths?\n");
71 return (-1);
74 static void multipath_reschedule_retry (struct multipath_bh *mp_bh)
76 unsigned long flags;
77 mddev_t *mddev = mp_bh->mddev;
78 multipath_conf_t *conf = mddev_to_conf(mddev);
80 spin_lock_irqsave(&conf->device_lock, flags);
81 list_add(&mp_bh->retry_list, &conf->retry_list);
82 spin_unlock_irqrestore(&conf->device_lock, flags);
83 md_wakeup_thread(mddev->thread);
88 * multipath_end_bh_io() is called when we have finished servicing a multipathed
89 * operation and are ready to return a success/failure code to the buffer
90 * cache layer.
92 static void multipath_end_bh_io (struct multipath_bh *mp_bh, int err)
94 struct bio *bio = mp_bh->master_bio;
95 multipath_conf_t *conf = mddev_to_conf(mp_bh->mddev);
97 bio_endio(bio, bio->bi_size, err);
98 mempool_free(mp_bh, conf->pool);
101 static int multipath_end_request(struct bio *bio, unsigned int bytes_done,
102 int error)
104 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
105 struct multipath_bh * mp_bh = (struct multipath_bh *)(bio->bi_private);
106 multipath_conf_t *conf = mddev_to_conf(mp_bh->mddev);
107 mdk_rdev_t *rdev = conf->multipaths[mp_bh->path].rdev;
109 if (bio->bi_size)
110 return 1;
112 if (uptodate)
113 multipath_end_bh_io(mp_bh, 0);
114 else if (!bio_rw_ahead(bio)) {
116 * oops, IO error:
118 char b[BDEVNAME_SIZE];
119 md_error (mp_bh->mddev, rdev);
120 printk(KERN_ERR "multipath: %s: rescheduling sector %llu\n",
121 bdevname(rdev->bdev,b),
122 (unsigned long long)bio->bi_sector);
123 multipath_reschedule_retry(mp_bh);
124 } else
125 multipath_end_bh_io(mp_bh, error);
126 rdev_dec_pending(rdev, conf->mddev);
127 return 0;
130 static void unplug_slaves(mddev_t *mddev)
132 multipath_conf_t *conf = mddev_to_conf(mddev);
133 int i;
135 rcu_read_lock();
136 for (i=0; i<mddev->raid_disks; i++) {
137 mdk_rdev_t *rdev = rcu_dereference(conf->multipaths[i].rdev);
138 if (rdev && !test_bit(Faulty, &rdev->flags)
139 && atomic_read(&rdev->nr_pending)) {
140 request_queue_t *r_queue = bdev_get_queue(rdev->bdev);
142 atomic_inc(&rdev->nr_pending);
143 rcu_read_unlock();
145 if (r_queue->unplug_fn)
146 r_queue->unplug_fn(r_queue);
148 rdev_dec_pending(rdev, mddev);
149 rcu_read_lock();
152 rcu_read_unlock();
155 static void multipath_unplug(request_queue_t *q)
157 unplug_slaves(q->queuedata);
161 static int multipath_make_request (request_queue_t *q, struct bio * bio)
163 mddev_t *mddev = q->queuedata;
164 multipath_conf_t *conf = mddev_to_conf(mddev);
165 struct multipath_bh * mp_bh;
166 struct multipath_info *multipath;
167 const int rw = bio_data_dir(bio);
169 if (unlikely(bio_barrier(bio))) {
170 bio_endio(bio, bio->bi_size, -EOPNOTSUPP);
171 return 0;
174 mp_bh = mempool_alloc(conf->pool, GFP_NOIO);
176 mp_bh->master_bio = bio;
177 mp_bh->mddev = mddev;
179 disk_stat_inc(mddev->gendisk, ios[rw]);
180 disk_stat_add(mddev->gendisk, sectors[rw], bio_sectors(bio));
182 mp_bh->path = multipath_map(conf);
183 if (mp_bh->path < 0) {
184 bio_endio(bio, bio->bi_size, -EIO);
185 mempool_free(mp_bh, conf->pool);
186 return 0;
188 multipath = conf->multipaths + mp_bh->path;
190 mp_bh->bio = *bio;
191 mp_bh->bio.bi_sector += multipath->rdev->data_offset;
192 mp_bh->bio.bi_bdev = multipath->rdev->bdev;
193 mp_bh->bio.bi_rw |= (1 << BIO_RW_FAILFAST);
194 mp_bh->bio.bi_end_io = multipath_end_request;
195 mp_bh->bio.bi_private = mp_bh;
196 generic_make_request(&mp_bh->bio);
197 return 0;
200 static void multipath_status (struct seq_file *seq, mddev_t *mddev)
202 multipath_conf_t *conf = mddev_to_conf(mddev);
203 int i;
205 seq_printf (seq, " [%d/%d] [", conf->raid_disks,
206 conf->working_disks);
207 for (i = 0; i < conf->raid_disks; i++)
208 seq_printf (seq, "%s",
209 conf->multipaths[i].rdev &&
210 test_bit(In_sync, &conf->multipaths[i].rdev->flags) ? "U" : "_");
211 seq_printf (seq, "]");
214 static int multipath_issue_flush(request_queue_t *q, struct gendisk *disk,
215 sector_t *error_sector)
217 mddev_t *mddev = q->queuedata;
218 multipath_conf_t *conf = mddev_to_conf(mddev);
219 int i, ret = 0;
221 rcu_read_lock();
222 for (i=0; i<mddev->raid_disks && ret == 0; i++) {
223 mdk_rdev_t *rdev = rcu_dereference(conf->multipaths[i].rdev);
224 if (rdev && !test_bit(Faulty, &rdev->flags)) {
225 struct block_device *bdev = rdev->bdev;
226 request_queue_t *r_queue = bdev_get_queue(bdev);
228 if (!r_queue->issue_flush_fn)
229 ret = -EOPNOTSUPP;
230 else {
231 atomic_inc(&rdev->nr_pending);
232 rcu_read_unlock();
233 ret = r_queue->issue_flush_fn(r_queue, bdev->bd_disk,
234 error_sector);
235 rdev_dec_pending(rdev, mddev);
236 rcu_read_lock();
240 rcu_read_unlock();
241 return ret;
245 * Careful, this can execute in IRQ contexts as well!
247 static void multipath_error (mddev_t *mddev, mdk_rdev_t *rdev)
249 multipath_conf_t *conf = mddev_to_conf(mddev);
251 if (conf->working_disks <= 1) {
253 * Uh oh, we can do nothing if this is our last path, but
254 * first check if this is a queued request for a device
255 * which has just failed.
257 printk(KERN_ALERT
258 "multipath: only one IO path left and IO error.\n");
259 /* leave it active... it's all we have */
260 } else {
262 * Mark disk as unusable
264 if (!test_bit(Faulty, &rdev->flags)) {
265 char b[BDEVNAME_SIZE];
266 clear_bit(In_sync, &rdev->flags);
267 set_bit(Faulty, &rdev->flags);
268 mddev->sb_dirty = 1;
269 conf->working_disks--;
270 printk(KERN_ALERT "multipath: IO failure on %s,"
271 " disabling IO path. \n Operation continuing"
272 " on %d IO paths.\n",
273 bdevname (rdev->bdev,b),
274 conf->working_disks);
279 static void print_multipath_conf (multipath_conf_t *conf)
281 int i;
282 struct multipath_info *tmp;
284 printk("MULTIPATH conf printout:\n");
285 if (!conf) {
286 printk("(conf==NULL)\n");
287 return;
289 printk(" --- wd:%d rd:%d\n", conf->working_disks,
290 conf->raid_disks);
292 for (i = 0; i < conf->raid_disks; i++) {
293 char b[BDEVNAME_SIZE];
294 tmp = conf->multipaths + i;
295 if (tmp->rdev)
296 printk(" disk%d, o:%d, dev:%s\n",
297 i,!test_bit(Faulty, &tmp->rdev->flags),
298 bdevname(tmp->rdev->bdev,b));
303 static int multipath_add_disk(mddev_t *mddev, mdk_rdev_t *rdev)
305 multipath_conf_t *conf = mddev->private;
306 struct request_queue *q;
307 int found = 0;
308 int path;
309 struct multipath_info *p;
311 print_multipath_conf(conf);
313 for (path=0; path<mddev->raid_disks; path++)
314 if ((p=conf->multipaths+path)->rdev == NULL) {
315 q = rdev->bdev->bd_disk->queue;
316 blk_queue_stack_limits(mddev->queue, q);
318 /* as we don't honour merge_bvec_fn, we must never risk
319 * violating it, so limit ->max_sector to one PAGE, as
320 * a one page request is never in violation.
321 * (Note: it is very unlikely that a device with
322 * merge_bvec_fn will be involved in multipath.)
324 if (q->merge_bvec_fn &&
325 mddev->queue->max_sectors > (PAGE_SIZE>>9))
326 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
328 conf->working_disks++;
329 rdev->raid_disk = path;
330 set_bit(In_sync, &rdev->flags);
331 rcu_assign_pointer(p->rdev, rdev);
332 found = 1;
335 print_multipath_conf(conf);
336 return found;
339 static int multipath_remove_disk(mddev_t *mddev, int number)
341 multipath_conf_t *conf = mddev->private;
342 int err = 0;
343 mdk_rdev_t *rdev;
344 struct multipath_info *p = conf->multipaths + number;
346 print_multipath_conf(conf);
348 rdev = p->rdev;
349 if (rdev) {
350 if (test_bit(In_sync, &rdev->flags) ||
351 atomic_read(&rdev->nr_pending)) {
352 printk(KERN_ERR "hot-remove-disk, slot %d is identified" " but is still operational!\n", number);
353 err = -EBUSY;
354 goto abort;
356 p->rdev = NULL;
357 synchronize_rcu();
358 if (atomic_read(&rdev->nr_pending)) {
359 /* lost the race, try later */
360 err = -EBUSY;
361 p->rdev = rdev;
364 abort:
366 print_multipath_conf(conf);
367 return err;
373 * This is a kernel thread which:
375 * 1. Retries failed read operations on working multipaths.
376 * 2. Updates the raid superblock when problems encounter.
377 * 3. Performs writes following reads for array syncronising.
380 static void multipathd (mddev_t *mddev)
382 struct multipath_bh *mp_bh;
383 struct bio *bio;
384 unsigned long flags;
385 multipath_conf_t *conf = mddev_to_conf(mddev);
386 struct list_head *head = &conf->retry_list;
388 md_check_recovery(mddev);
389 for (;;) {
390 char b[BDEVNAME_SIZE];
391 spin_lock_irqsave(&conf->device_lock, flags);
392 if (list_empty(head))
393 break;
394 mp_bh = list_entry(head->prev, struct multipath_bh, retry_list);
395 list_del(head->prev);
396 spin_unlock_irqrestore(&conf->device_lock, flags);
398 bio = &mp_bh->bio;
399 bio->bi_sector = mp_bh->master_bio->bi_sector;
401 if ((mp_bh->path = multipath_map (conf))<0) {
402 printk(KERN_ALERT "multipath: %s: unrecoverable IO read"
403 " error for block %llu\n",
404 bdevname(bio->bi_bdev,b),
405 (unsigned long long)bio->bi_sector);
406 multipath_end_bh_io(mp_bh, -EIO);
407 } else {
408 printk(KERN_ERR "multipath: %s: redirecting sector %llu"
409 " to another IO path\n",
410 bdevname(bio->bi_bdev,b),
411 (unsigned long long)bio->bi_sector);
412 *bio = *(mp_bh->master_bio);
413 bio->bi_sector += conf->multipaths[mp_bh->path].rdev->data_offset;
414 bio->bi_bdev = conf->multipaths[mp_bh->path].rdev->bdev;
415 bio->bi_rw |= (1 << BIO_RW_FAILFAST);
416 bio->bi_end_io = multipath_end_request;
417 bio->bi_private = mp_bh;
418 generic_make_request(bio);
421 spin_unlock_irqrestore(&conf->device_lock, flags);
424 static int multipath_run (mddev_t *mddev)
426 multipath_conf_t *conf;
427 int disk_idx;
428 struct multipath_info *disk;
429 mdk_rdev_t *rdev;
430 struct list_head *tmp;
432 if (mddev->level != LEVEL_MULTIPATH) {
433 printk("multipath: %s: raid level not set to multipath IO (%d)\n",
434 mdname(mddev), mddev->level);
435 goto out;
438 * copy the already verified devices into our private MULTIPATH
439 * bookkeeping area. [whatever we allocate in multipath_run(),
440 * should be freed in multipath_stop()]
443 conf = kzalloc(sizeof(multipath_conf_t), GFP_KERNEL);
444 mddev->private = conf;
445 if (!conf) {
446 printk(KERN_ERR
447 "multipath: couldn't allocate memory for %s\n",
448 mdname(mddev));
449 goto out;
452 conf->multipaths = kzalloc(sizeof(struct multipath_info)*mddev->raid_disks,
453 GFP_KERNEL);
454 if (!conf->multipaths) {
455 printk(KERN_ERR
456 "multipath: couldn't allocate memory for %s\n",
457 mdname(mddev));
458 goto out_free_conf;
461 conf->working_disks = 0;
462 ITERATE_RDEV(mddev,rdev,tmp) {
463 disk_idx = rdev->raid_disk;
464 if (disk_idx < 0 ||
465 disk_idx >= mddev->raid_disks)
466 continue;
468 disk = conf->multipaths + disk_idx;
469 disk->rdev = rdev;
471 blk_queue_stack_limits(mddev->queue,
472 rdev->bdev->bd_disk->queue);
473 /* as we don't honour merge_bvec_fn, we must never risk
474 * violating it, not that we ever expect a device with
475 * a merge_bvec_fn to be involved in multipath */
476 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
477 mddev->queue->max_sectors > (PAGE_SIZE>>9))
478 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
480 if (!test_bit(Faulty, &rdev->flags))
481 conf->working_disks++;
484 conf->raid_disks = mddev->raid_disks;
485 mddev->sb_dirty = 1;
486 conf->mddev = mddev;
487 spin_lock_init(&conf->device_lock);
488 INIT_LIST_HEAD(&conf->retry_list);
490 if (!conf->working_disks) {
491 printk(KERN_ERR "multipath: no operational IO paths for %s\n",
492 mdname(mddev));
493 goto out_free_conf;
495 mddev->degraded = conf->raid_disks = conf->working_disks;
497 conf->pool = mempool_create(NR_RESERVED_BUFS,
498 mp_pool_alloc, mp_pool_free,
499 NULL);
500 if (conf->pool == NULL) {
501 printk(KERN_ERR
502 "multipath: couldn't allocate memory for %s\n",
503 mdname(mddev));
504 goto out_free_conf;
508 mddev->thread = md_register_thread(multipathd, mddev, "%s_multipath");
509 if (!mddev->thread) {
510 printk(KERN_ERR "multipath: couldn't allocate thread"
511 " for %s\n", mdname(mddev));
512 goto out_free_conf;
516 printk(KERN_INFO
517 "multipath: array %s active with %d out of %d IO paths\n",
518 mdname(mddev), conf->working_disks, mddev->raid_disks);
520 * Ok, everything is just fine now
522 mddev->array_size = mddev->size;
524 mddev->queue->unplug_fn = multipath_unplug;
525 mddev->queue->issue_flush_fn = multipath_issue_flush;
527 return 0;
529 out_free_conf:
530 if (conf->pool)
531 mempool_destroy(conf->pool);
532 kfree(conf->multipaths);
533 kfree(conf);
534 mddev->private = NULL;
535 out:
536 return -EIO;
540 static int multipath_stop (mddev_t *mddev)
542 multipath_conf_t *conf = mddev_to_conf(mddev);
544 md_unregister_thread(mddev->thread);
545 mddev->thread = NULL;
546 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
547 mempool_destroy(conf->pool);
548 kfree(conf->multipaths);
549 kfree(conf);
550 mddev->private = NULL;
551 return 0;
554 static struct mdk_personality multipath_personality =
556 .name = "multipath",
557 .level = LEVEL_MULTIPATH,
558 .owner = THIS_MODULE,
559 .make_request = multipath_make_request,
560 .run = multipath_run,
561 .stop = multipath_stop,
562 .status = multipath_status,
563 .error_handler = multipath_error,
564 .hot_add_disk = multipath_add_disk,
565 .hot_remove_disk= multipath_remove_disk,
568 static int __init multipath_init (void)
570 return register_md_personality (&multipath_personality);
573 static void __exit multipath_exit (void)
575 unregister_md_personality (&multipath_personality);
578 module_init(multipath_init);
579 module_exit(multipath_exit);
580 MODULE_LICENSE("GPL");
581 MODULE_ALIAS("md-personality-7"); /* MULTIPATH */
582 MODULE_ALIAS("md-multipath");
583 MODULE_ALIAS("md-level--4");