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 mdk_personality_t multipath_personality;
39 static spinlock_t retry_list_lock = SPIN_LOCK_UNLOCKED;
40 struct multipath_bh *multipath_retry_list = NULL, **multipath_retry_tail;
43 static void *mp_pool_alloc(int gfp_flags, void *data)
45 struct multipath_bh *mpb;
46 mpb = kmalloc(sizeof(*mpb), gfp_flags);
48 memset(mpb, 0, sizeof(*mpb));
52 static void mp_pool_free(void *mpb, void *data)
57 static int multipath_map (multipath_conf_t *conf)
59 int i, disks = conf->raid_disks;
62 * Later we do read balancing on the read side
63 * now we use the first available disk.
66 spin_lock_irq(&conf->device_lock);
67 for (i = 0; i < disks; i++) {
68 mdk_rdev_t *rdev = conf->multipaths[i].rdev;
69 if (rdev && rdev->in_sync) {
70 atomic_inc(&rdev->nr_pending);
71 spin_unlock_irq(&conf->device_lock);
75 spin_unlock_irq(&conf->device_lock);
77 printk(KERN_ERR "multipath_map(): no more operational IO paths?\n");
81 static void multipath_reschedule_retry (struct multipath_bh *mp_bh)
84 mddev_t *mddev = mp_bh->mddev;
86 spin_lock_irqsave(&retry_list_lock, flags);
87 if (multipath_retry_list == NULL)
88 multipath_retry_tail = &multipath_retry_list;
89 *multipath_retry_tail = mp_bh;
90 multipath_retry_tail = &mp_bh->next_mp;
91 mp_bh->next_mp = NULL;
92 spin_unlock_irqrestore(&retry_list_lock, flags);
93 md_wakeup_thread(mddev->thread);
98 * multipath_end_bh_io() is called when we have finished servicing a multipathed
99 * operation and are ready to return a success/failure code to the buffer
102 static void multipath_end_bh_io (struct multipath_bh *mp_bh, int uptodate)
104 struct bio *bio = mp_bh->master_bio;
105 multipath_conf_t *conf = mddev_to_conf(mp_bh->mddev);
107 bio_endio(bio, bio->bi_size, uptodate ? 0 : -EIO);
108 mempool_free(mp_bh, conf->pool);
111 int multipath_end_request(struct bio *bio, unsigned int bytes_done, int error)
113 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
114 struct multipath_bh * mp_bh = (struct multipath_bh *)(bio->bi_private);
115 multipath_conf_t *conf = mddev_to_conf(mp_bh->mddev);
116 mdk_rdev_t *rdev = conf->multipaths[mp_bh->path].rdev;
122 multipath_end_bh_io(mp_bh, uptodate);
123 else if ((bio->bi_rw & (1 << BIO_RW_AHEAD)) == 0) {
127 char b[BDEVNAME_SIZE];
128 md_error (mp_bh->mddev, rdev);
129 printk(KERN_ERR "multipath: %s: rescheduling sector %llu\n",
130 bdevname(rdev->bdev,b),
131 (unsigned long long)bio->bi_sector);
132 multipath_reschedule_retry(mp_bh);
134 multipath_end_bh_io(mp_bh, 0);
135 rdev_dec_pending(rdev, conf->mddev);
139 static void unplug_slaves(mddev_t *mddev)
141 multipath_conf_t *conf = mddev_to_conf(mddev);
145 spin_lock_irqsave(&conf->device_lock, flags);
146 for (i=0; i<mddev->raid_disks; i++) {
147 mdk_rdev_t *rdev = conf->multipaths[i].rdev;
148 if (rdev && !rdev->faulty) {
149 request_queue_t *r_queue = bdev_get_queue(rdev->bdev);
151 atomic_inc(&rdev->nr_pending);
152 spin_unlock_irqrestore(&conf->device_lock, flags);
154 if (r_queue->unplug_fn)
155 r_queue->unplug_fn(r_queue);
157 spin_lock_irqsave(&conf->device_lock, flags);
158 atomic_dec(&rdev->nr_pending);
161 spin_unlock_irqrestore(&conf->device_lock, flags);
163 static void multipath_unplug(request_queue_t *q)
165 unplug_slaves(q->queuedata);
169 static int multipath_make_request (request_queue_t *q, struct bio * bio)
171 mddev_t *mddev = q->queuedata;
172 multipath_conf_t *conf = mddev_to_conf(mddev);
173 struct multipath_bh * mp_bh;
174 struct multipath_info *multipath;
176 mp_bh = mempool_alloc(conf->pool, GFP_NOIO);
178 mp_bh->master_bio = bio;
179 mp_bh->mddev = mddev;
181 if (bio_data_dir(bio)==WRITE) {
182 disk_stat_inc(mddev->gendisk, writes);
183 disk_stat_add(mddev->gendisk, write_sectors, bio_sectors(bio));
185 disk_stat_inc(mddev->gendisk, reads);
186 disk_stat_add(mddev->gendisk, read_sectors, bio_sectors(bio));
189 mp_bh->path = multipath_map(conf);
190 if (mp_bh->path < 0) {
191 bio_endio(bio, bio->bi_size, -EIO);
192 mempool_free(mp_bh, conf->pool);
195 multipath = conf->multipaths + mp_bh->path;
198 mp_bh->bio.bi_bdev = multipath->rdev->bdev;
199 mp_bh->bio.bi_rw |= (1 << BIO_RW_FAILFAST);
200 mp_bh->bio.bi_end_io = multipath_end_request;
201 mp_bh->bio.bi_private = mp_bh;
202 generic_make_request(&mp_bh->bio);
206 static void multipath_status (struct seq_file *seq, mddev_t *mddev)
208 multipath_conf_t *conf = mddev_to_conf(mddev);
211 seq_printf (seq, " [%d/%d] [", conf->raid_disks,
212 conf->working_disks);
213 for (i = 0; i < conf->raid_disks; i++)
214 seq_printf (seq, "%s",
215 conf->multipaths[i].rdev &&
216 conf->multipaths[i].rdev->in_sync ? "U" : "_");
217 seq_printf (seq, "]");
222 * Careful, this can execute in IRQ contexts as well!
224 static void multipath_error (mddev_t *mddev, mdk_rdev_t *rdev)
226 multipath_conf_t *conf = mddev_to_conf(mddev);
228 if (conf->working_disks <= 1) {
230 * Uh oh, we can do nothing if this is our last path, but
231 * first check if this is a queued request for a device
232 * which has just failed.
235 "multipath: only one IO path left and IO error.\n");
236 /* leave it active... it's all we have */
239 * Mark disk as unusable
242 char b[BDEVNAME_SIZE];
246 conf->working_disks--;
247 printk(KERN_ALERT "multipath: IO failure on %s,"
248 " disabling IO path. \n Operation continuing"
249 " on %d IO paths.\n",
250 bdevname (rdev->bdev,b),
251 conf->working_disks);
256 static void print_multipath_conf (multipath_conf_t *conf)
259 struct multipath_info *tmp;
261 printk("MULTIPATH conf printout:\n");
263 printk("(conf==NULL)\n");
266 printk(" --- wd:%d rd:%d\n", conf->working_disks,
269 for (i = 0; i < conf->raid_disks; i++) {
270 char b[BDEVNAME_SIZE];
271 tmp = conf->multipaths + i;
273 printk(" disk%d, o:%d, dev:%s\n",
274 i,!tmp->rdev->faulty,
275 bdevname(tmp->rdev->bdev,b));
280 static int multipath_add_disk(mddev_t *mddev, mdk_rdev_t *rdev)
282 multipath_conf_t *conf = mddev->private;
285 struct multipath_info *p;
287 print_multipath_conf(conf);
288 spin_lock_irq(&conf->device_lock);
289 for (path=0; path<mddev->raid_disks; path++)
290 if ((p=conf->multipaths+path)->rdev == NULL) {
292 blk_queue_stack_limits(mddev->queue,
293 rdev->bdev->bd_disk->queue);
295 /* as we don't honour merge_bvec_fn, we must never risk
296 * violating it, so limit ->max_sector to one PAGE, as
297 * a one page request is never in violation.
298 * (Note: it is very unlikely that a device with
299 * merge_bvec_fn will be involved in multipath.)
301 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
302 mddev->queue->max_sectors > (PAGE_SIZE>>9))
303 mddev->queue->max_sectors = (PAGE_SIZE>>9);
305 conf->working_disks++;
306 rdev->raid_disk = path;
310 spin_unlock_irq(&conf->device_lock);
312 print_multipath_conf(conf);
316 static int multipath_remove_disk(mddev_t *mddev, int number)
318 multipath_conf_t *conf = mddev->private;
320 struct multipath_info *p = conf->multipaths + number;
322 print_multipath_conf(conf);
323 spin_lock_irq(&conf->device_lock);
326 if (p->rdev->in_sync ||
327 atomic_read(&p->rdev->nr_pending)) {
328 printk(KERN_ERR "hot-remove-disk, slot %d is identified" " but is still operational!\n", number);
338 spin_unlock_irq(&conf->device_lock);
340 print_multipath_conf(conf);
347 * This is a kernel thread which:
349 * 1. Retries failed read operations on working multipaths.
350 * 2. Updates the raid superblock when problems encounter.
351 * 3. Performs writes following reads for array syncronising.
354 static void multipathd (mddev_t *mddev)
356 struct multipath_bh *mp_bh;
359 multipath_conf_t *conf = mddev_to_conf(mddev);
361 md_check_recovery(mddev);
363 char b[BDEVNAME_SIZE];
364 spin_lock_irqsave(&retry_list_lock, flags);
365 mp_bh = multipath_retry_list;
368 multipath_retry_list = mp_bh->next_mp;
369 spin_unlock_irqrestore(&retry_list_lock, flags);
371 mddev = mp_bh->mddev;
373 bio->bi_sector = mp_bh->master_bio->bi_sector;
375 if ((mp_bh->path = multipath_map (conf))<0) {
376 printk(KERN_ALERT "multipath: %s: unrecoverable IO read"
377 " error for block %llu\n",
378 bdevname(bio->bi_bdev,b),
379 (unsigned long long)bio->bi_sector);
380 multipath_end_bh_io(mp_bh, 0);
382 printk(KERN_ERR "multipath: %s: redirecting sector %llu"
383 " to another IO path\n",
384 bdevname(bio->bi_bdev,b),
385 (unsigned long long)bio->bi_sector);
386 *bio = *(mp_bh->master_bio);
387 bio->bi_bdev = conf->multipaths[mp_bh->path].rdev->bdev;
388 bio->bi_rw |= (1 << BIO_RW_FAILFAST);
389 bio->bi_end_io = multipath_end_request;
390 bio->bi_private = mp_bh;
391 generic_make_request(bio);
394 spin_unlock_irqrestore(&retry_list_lock, flags);
397 static int multipath_run (mddev_t *mddev)
399 multipath_conf_t *conf;
401 struct multipath_info *disk;
403 struct list_head *tmp;
405 if (mddev->level != LEVEL_MULTIPATH) {
406 printk("multipath: %s: raid level not set to multipath IO (%d)\n",
407 mdname(mddev), mddev->level);
411 * copy the already verified devices into our private MULTIPATH
412 * bookkeeping area. [whatever we allocate in multipath_run(),
413 * should be freed in multipath_stop()]
416 conf = kmalloc(sizeof(multipath_conf_t), GFP_KERNEL);
417 mddev->private = conf;
420 "multipath: couldn't allocate memory for %s\n",
424 memset(conf, 0, sizeof(*conf));
426 conf->multipaths = kmalloc(sizeof(struct multipath_info)*mddev->raid_disks,
428 if (!conf->multipaths) {
430 "multipath: couldn't allocate memory for %s\n",
434 memset(conf->multipaths, 0, sizeof(struct multipath_info)*mddev->raid_disks);
436 mddev->queue->unplug_fn = multipath_unplug;
438 conf->working_disks = 0;
439 ITERATE_RDEV(mddev,rdev,tmp) {
440 disk_idx = rdev->raid_disk;
442 disk_idx >= mddev->raid_disks)
445 disk = conf->multipaths + disk_idx;
448 blk_queue_stack_limits(mddev->queue,
449 rdev->bdev->bd_disk->queue);
450 /* as we don't honour merge_bvec_fn, we must never risk
451 * violating it, not that we ever expect a device with
452 * a merge_bvec_fn to be involved in multipath */
453 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
454 mddev->queue->max_sectors > (PAGE_SIZE>>9))
455 mddev->queue->max_sectors = (PAGE_SIZE>>9);
458 conf->working_disks++;
461 conf->raid_disks = mddev->raid_disks;
464 conf->device_lock = SPIN_LOCK_UNLOCKED;
466 if (!conf->working_disks) {
467 printk(KERN_ERR "multipath: no operational IO paths for %s\n",
471 mddev->degraded = conf->raid_disks = conf->working_disks;
473 conf->pool = mempool_create(NR_RESERVED_BUFS,
474 mp_pool_alloc, mp_pool_free,
476 if (conf->pool == NULL) {
478 "multipath: couldn't allocate memory for %s\n",
484 mddev->thread = md_register_thread(multipathd, mddev, "%s_multipath");
485 if (!mddev->thread) {
486 printk(KERN_ERR "multipath: couldn't allocate thread"
487 " for %s\n", mdname(mddev));
493 "multipath: array %s active with %d out of %d IO paths\n",
494 mdname(mddev), conf->working_disks, mddev->raid_disks);
496 * Ok, everything is just fine now
498 mddev->array_size = mddev->size;
503 mempool_destroy(conf->pool);
504 if (conf->multipaths)
505 kfree(conf->multipaths);
507 mddev->private = NULL;
513 static int multipath_stop (mddev_t *mddev)
515 multipath_conf_t *conf = mddev_to_conf(mddev);
517 md_unregister_thread(mddev->thread);
518 mddev->thread = NULL;
519 mempool_destroy(conf->pool);
520 kfree(conf->multipaths);
522 mddev->private = NULL;
526 static mdk_personality_t multipath_personality=
529 .owner = THIS_MODULE,
530 .make_request = multipath_make_request,
531 .run = multipath_run,
532 .stop = multipath_stop,
533 .status = multipath_status,
534 .error_handler = multipath_error,
535 .hot_add_disk = multipath_add_disk,
536 .hot_remove_disk= multipath_remove_disk,
539 static int __init multipath_init (void)
541 return register_md_personality (MULTIPATH, &multipath_personality);
544 static void __exit multipath_exit (void)
546 unregister_md_personality (MULTIPATH);
549 module_init(multipath_init);
550 module_exit(multipath_exit);
551 MODULE_LICENSE("GPL");
552 MODULE_ALIAS("md-personality-7"); /* MULTIPATH */