blob: 7331a80d89f1987a42d22a2ae7510210519e6bae [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * multipath.c : Multiple Devices driver for Linux
3 *
4 * Copyright (C) 1999, 2000, 2001 Ingo Molnar, Red Hat
5 *
6 * Copyright (C) 1996, 1997, 1998 Ingo Molnar, Miguel de Icaza, Gadi Oxman
7 *
8 * MULTIPATH management functions.
9 *
10 * derived from raid1.c.
11 *
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.
16 *
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.
20 */
21
NeilBrownbff61972009-03-31 14:33:13 +110022#include <linux/blkdev.h>
Paul Gortmaker056075c2011-07-03 13:58:33 -040023#include <linux/module.h>
NeilBrownbff61972009-03-31 14:33:13 +110024#include <linux/raid/md_u.h>
NeilBrownbff61972009-03-31 14:33:13 +110025#include <linux/seq_file.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090026#include <linux/slab.h>
NeilBrown43b2e5d2009-03-31 14:33:13 +110027#include "md.h"
Christoph Hellwigef740c32009-03-31 14:27:03 +110028#include "multipath.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029
30#define MAX_WORK_PER_DISK 128
31
32#define NR_RESERVED_BUFS 32
33
NeilBrown69724e22011-10-11 16:48:57 +110034static int multipath_map (struct mpconf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -070035{
36 int i, disks = conf->raid_disks;
37
38 /*
NeilBrownf72ffdd2014-09-30 14:23:59 +100039 * Later we do read balancing on the read side
Linus Torvalds1da177e2005-04-16 15:20:36 -070040 * now we use the first available disk.
41 */
42
43 rcu_read_lock();
44 for (i = 0; i < disks; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +110045 struct md_rdev *rdev = rcu_dereference(conf->multipaths[i].rdev);
NeilBrownb2d444d2005-11-08 21:39:31 -080046 if (rdev && test_bit(In_sync, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070047 atomic_inc(&rdev->nr_pending);
48 rcu_read_unlock();
49 return i;
50 }
51 }
52 rcu_read_unlock();
53
54 printk(KERN_ERR "multipath_map(): no more operational IO paths?\n");
55 return (-1);
56}
57
58static void multipath_reschedule_retry (struct multipath_bh *mp_bh)
59{
60 unsigned long flags;
NeilBrownfd01b882011-10-11 16:47:53 +110061 struct mddev *mddev = mp_bh->mddev;
NeilBrown69724e22011-10-11 16:48:57 +110062 struct mpconf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -070063
64 spin_lock_irqsave(&conf->device_lock, flags);
65 list_add(&mp_bh->retry_list, &conf->retry_list);
66 spin_unlock_irqrestore(&conf->device_lock, flags);
67 md_wakeup_thread(mddev->thread);
68}
69
Linus Torvalds1da177e2005-04-16 15:20:36 -070070/*
71 * multipath_end_bh_io() is called when we have finished servicing a multipathed
72 * operation and are ready to return a success/failure code to the buffer
73 * cache layer.
74 */
75static void multipath_end_bh_io (struct multipath_bh *mp_bh, int err)
76{
77 struct bio *bio = mp_bh->master_bio;
NeilBrown69724e22011-10-11 16:48:57 +110078 struct mpconf *conf = mp_bh->mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -070079
Christoph Hellwig4246a0b2015-07-20 15:29:37 +020080 bio->bi_error = err;
81 bio_endio(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 mempool_free(mp_bh, conf->pool);
83}
84
Christoph Hellwig4246a0b2015-07-20 15:29:37 +020085static void multipath_end_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -070086{
H Hartley Sweeten7b928132010-03-08 16:02:40 +110087 struct multipath_bh *mp_bh = bio->bi_private;
NeilBrown69724e22011-10-11 16:48:57 +110088 struct mpconf *conf = mp_bh->mddev->private;
NeilBrown3cb03002011-10-11 16:45:26 +110089 struct md_rdev *rdev = conf->multipaths[mp_bh->path].rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
Christoph Hellwig4246a0b2015-07-20 15:29:37 +020091 if (!bio->bi_error)
Linus Torvalds1da177e2005-04-16 15:20:36 -070092 multipath_end_bh_io(mp_bh, 0);
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +020093 else if (!(bio->bi_rw & REQ_RAHEAD)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070094 /*
95 * oops, IO error:
96 */
97 char b[BDEVNAME_SIZE];
98 md_error (mp_bh->mddev, rdev);
NeilBrownf72ffdd2014-09-30 14:23:59 +100099 printk(KERN_ERR "multipath: %s: rescheduling sector %llu\n",
100 bdevname(rdev->bdev,b),
Kent Overstreet4f024f32013-10-11 15:44:27 -0700101 (unsigned long long)bio->bi_iter.bi_sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102 multipath_reschedule_retry(mp_bh);
103 } else
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200104 multipath_end_bh_io(mp_bh, bio->bi_error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 rdev_dec_pending(rdev, conf->mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106}
107
Linus Torvaldsb4fdcb02011-11-04 17:06:58 -0700108static void multipath_make_request(struct mddev *mddev, struct bio * bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109{
NeilBrown69724e22011-10-11 16:48:57 +1100110 struct mpconf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 struct multipath_bh * mp_bh;
112 struct multipath_info *multipath;
113
Tejun Heoe9c74692010-09-03 11:56:18 +0200114 if (unlikely(bio->bi_rw & REQ_FLUSH)) {
115 md_flush_request(mddev, bio);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200116 return;
NeilBrowne5dcdd82005-09-09 16:23:41 -0700117 }
118
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 mp_bh = mempool_alloc(conf->pool, GFP_NOIO);
120
121 mp_bh->master_bio = bio;
122 mp_bh->mddev = mddev;
123
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124 mp_bh->path = multipath_map(conf);
125 if (mp_bh->path < 0) {
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200126 bio_io_error(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 mempool_free(mp_bh, conf->pool);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200128 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 }
130 multipath = conf->multipaths + mp_bh->path;
131
132 mp_bh->bio = *bio;
Kent Overstreet4f024f32013-10-11 15:44:27 -0700133 mp_bh->bio.bi_iter.bi_sector += multipath->rdev->data_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 mp_bh->bio.bi_bdev = multipath->rdev->bdev;
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +0200135 mp_bh->bio.bi_rw |= REQ_FAILFAST_TRANSPORT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 mp_bh->bio.bi_end_io = multipath_end_request;
137 mp_bh->bio.bi_private = mp_bh;
138 generic_make_request(&mp_bh->bio);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200139 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140}
141
NeilBrownfd01b882011-10-11 16:47:53 +1100142static void multipath_status (struct seq_file *seq, struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143{
NeilBrown69724e22011-10-11 16:48:57 +1100144 struct mpconf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 int i;
NeilBrownf72ffdd2014-09-30 14:23:59 +1000146
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 seq_printf (seq, " [%d/%d] [", conf->raid_disks,
NeilBrown92f861a2011-05-11 14:38:02 +1000148 conf->raid_disks - mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 for (i = 0; i < conf->raid_disks; i++)
150 seq_printf (seq, "%s",
NeilBrownf72ffdd2014-09-30 14:23:59 +1000151 conf->multipaths[i].rdev &&
NeilBrownb2d444d2005-11-08 21:39:31 -0800152 test_bit(In_sync, &conf->multipaths[i].rdev->flags) ? "U" : "_");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 seq_printf (seq, "]");
154}
155
NeilBrown5c675f82014-12-15 12:56:56 +1100156static int multipath_congested(struct mddev *mddev, int bits)
NeilBrown0d129222006-10-03 01:15:54 -0700157{
NeilBrown69724e22011-10-11 16:48:57 +1100158 struct mpconf *conf = mddev->private;
NeilBrown0d129222006-10-03 01:15:54 -0700159 int i, ret = 0;
160
161 rcu_read_lock();
162 for (i = 0; i < mddev->raid_disks ; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +1100163 struct md_rdev *rdev = rcu_dereference(conf->multipaths[i].rdev);
NeilBrown0d129222006-10-03 01:15:54 -0700164 if (rdev && !test_bit(Faulty, &rdev->flags)) {
Jens Axboe165125e2007-07-24 09:28:11 +0200165 struct request_queue *q = bdev_get_queue(rdev->bdev);
NeilBrown0d129222006-10-03 01:15:54 -0700166
167 ret |= bdi_congested(&q->backing_dev_info, bits);
168 /* Just like multipath_map, we just check the
169 * first available device
170 */
171 break;
172 }
173 }
174 rcu_read_unlock();
175 return ret;
176}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177
178/*
179 * Careful, this can execute in IRQ contexts as well!
180 */
NeilBrownfd01b882011-10-11 16:47:53 +1100181static void multipath_error (struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182{
NeilBrown69724e22011-10-11 16:48:57 +1100183 struct mpconf *conf = mddev->private;
NeilBrown6f8d0c72011-05-11 14:38:44 +1000184 char b[BDEVNAME_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185
NeilBrown92f861a2011-05-11 14:38:02 +1000186 if (conf->raid_disks - mddev->degraded <= 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 /*
188 * Uh oh, we can do nothing if this is our last path, but
189 * first check if this is a queued request for a device
190 * which has just failed.
191 */
NeilBrownf72ffdd2014-09-30 14:23:59 +1000192 printk(KERN_ALERT
NeilBrown6f8d0c72011-05-11 14:38:44 +1000193 "multipath: only one IO path left and IO error.\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 /* leave it active... it's all we have */
NeilBrown6f8d0c72011-05-11 14:38:44 +1000195 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 }
NeilBrown6f8d0c72011-05-11 14:38:44 +1000197 /*
198 * Mark disk as unusable
199 */
200 if (test_and_clear_bit(In_sync, &rdev->flags)) {
201 unsigned long flags;
202 spin_lock_irqsave(&conf->device_lock, flags);
203 mddev->degraded++;
204 spin_unlock_irqrestore(&conf->device_lock, flags);
205 }
206 set_bit(Faulty, &rdev->flags);
207 set_bit(MD_CHANGE_DEVS, &mddev->flags);
208 printk(KERN_ALERT "multipath: IO failure on %s,"
209 " disabling IO path.\n"
210 "multipath: Operation continuing"
211 " on %d IO paths.\n",
212 bdevname(rdev->bdev, b),
213 conf->raid_disks - mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214}
215
NeilBrown69724e22011-10-11 16:48:57 +1100216static void print_multipath_conf (struct mpconf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217{
218 int i;
219 struct multipath_info *tmp;
220
221 printk("MULTIPATH conf printout:\n");
222 if (!conf) {
223 printk("(conf==NULL)\n");
224 return;
225 }
NeilBrown92f861a2011-05-11 14:38:02 +1000226 printk(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 conf->raid_disks);
228
229 for (i = 0; i < conf->raid_disks; i++) {
230 char b[BDEVNAME_SIZE];
231 tmp = conf->multipaths + i;
232 if (tmp->rdev)
233 printk(" disk%d, o:%d, dev:%s\n",
NeilBrownb2d444d2005-11-08 21:39:31 -0800234 i,!test_bit(Faulty, &tmp->rdev->flags),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 bdevname(tmp->rdev->bdev,b));
236 }
237}
238
NeilBrownfd01b882011-10-11 16:47:53 +1100239static int multipath_add_disk(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240{
NeilBrown69724e22011-10-11 16:48:57 +1100241 struct mpconf *conf = mddev->private;
Jesper Juhl352d7682006-01-09 20:54:44 -0800242 struct request_queue *q;
Neil Brown199050e2008-06-28 08:31:33 +1000243 int err = -EEXIST;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 int path;
245 struct multipath_info *p;
Neil Brown6c2fce22008-06-28 08:31:31 +1000246 int first = 0;
247 int last = mddev->raid_disks - 1;
248
249 if (rdev->raid_disk >= 0)
250 first = last = rdev->raid_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251
252 print_multipath_conf(conf);
253
Neil Brown6c2fce22008-06-28 08:31:31 +1000254 for (path = first; path <= last; path++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 if ((p=conf->multipaths+path)->rdev == NULL) {
Jesper Juhl352d7682006-01-09 20:54:44 -0800256 q = rdev->bdev->bd_disk->queue;
Martin K. Petersen8f6c2e42009-07-01 11:13:45 +1000257 disk_stack_limits(mddev->gendisk, rdev->bdev,
258 rdev->data_offset << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259
NeilBrown6f8d0c72011-05-11 14:38:44 +1000260 spin_lock_irq(&conf->device_lock);
NeilBrown750a8f32006-10-28 10:38:31 -0700261 mddev->degraded--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 rdev->raid_disk = path;
NeilBrownb2d444d2005-11-08 21:39:31 -0800263 set_bit(In_sync, &rdev->flags);
NeilBrown6f8d0c72011-05-11 14:38:44 +1000264 spin_unlock_irq(&conf->device_lock);
Suzanne Woodd6065f72005-11-08 21:39:27 -0800265 rcu_assign_pointer(p->rdev, rdev);
Neil Brown199050e2008-06-28 08:31:33 +1000266 err = 0;
Dan Williamsc7bfced2015-10-21 13:20:02 -0400267 mddev_suspend(mddev);
Andre Nollac5e7112009-08-03 10:59:47 +1000268 md_integrity_add_rdev(rdev, mddev);
Dan Williamsc7bfced2015-10-21 13:20:02 -0400269 mddev_resume(mddev);
Neil Brown199050e2008-06-28 08:31:33 +1000270 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 }
272
273 print_multipath_conf(conf);
Neil Brown199050e2008-06-28 08:31:33 +1000274
275 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276}
277
NeilBrownb8321b62011-12-23 10:17:51 +1100278static int multipath_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279{
NeilBrown69724e22011-10-11 16:48:57 +1100280 struct mpconf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 int err = 0;
NeilBrownb8321b62011-12-23 10:17:51 +1100282 int number = rdev->raid_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 struct multipath_info *p = conf->multipaths + number;
284
285 print_multipath_conf(conf);
286
NeilBrownb8321b62011-12-23 10:17:51 +1100287 if (rdev == p->rdev) {
NeilBrownb2d444d2005-11-08 21:39:31 -0800288 if (test_bit(In_sync, &rdev->flags) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 atomic_read(&rdev->nr_pending)) {
NeilBrowndfc70642008-05-23 13:04:39 -0700290 printk(KERN_ERR "hot-remove-disk, slot %d is identified"
291 " but is still operational!\n", number);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 err = -EBUSY;
293 goto abort;
294 }
295 p->rdev = NULL;
Paul E. McKenneyfbd568a3e2005-05-01 08:59:04 -0700296 synchronize_rcu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 if (atomic_read(&rdev->nr_pending)) {
298 /* lost the race, try later */
299 err = -EBUSY;
300 p->rdev = rdev;
Andre Nollac5e7112009-08-03 10:59:47 +1000301 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 }
Martin K. Petersena91a2782011-03-17 11:11:05 +0100303 err = md_integrity_register(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304 }
305abort:
306
307 print_multipath_conf(conf);
308 return err;
309}
310
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311/*
312 * This is a kernel thread which:
313 *
314 * 1. Retries failed read operations on working multipaths.
315 * 2. Updates the raid superblock when problems encounter.
316 * 3. Performs writes following reads for array syncronising.
317 */
318
Shaohua Li4ed87312012-10-11 13:34:00 +1100319static void multipathd(struct md_thread *thread)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320{
Shaohua Li4ed87312012-10-11 13:34:00 +1100321 struct mddev *mddev = thread->mddev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 struct multipath_bh *mp_bh;
323 struct bio *bio;
324 unsigned long flags;
NeilBrown69724e22011-10-11 16:48:57 +1100325 struct mpconf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 struct list_head *head = &conf->retry_list;
327
328 md_check_recovery(mddev);
329 for (;;) {
330 char b[BDEVNAME_SIZE];
331 spin_lock_irqsave(&conf->device_lock, flags);
332 if (list_empty(head))
333 break;
334 mp_bh = list_entry(head->prev, struct multipath_bh, retry_list);
335 list_del(head->prev);
336 spin_unlock_irqrestore(&conf->device_lock, flags);
337
338 bio = &mp_bh->bio;
Kent Overstreet4f024f32013-10-11 15:44:27 -0700339 bio->bi_iter.bi_sector = mp_bh->master_bio->bi_iter.bi_sector;
NeilBrownf72ffdd2014-09-30 14:23:59 +1000340
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 if ((mp_bh->path = multipath_map (conf))<0) {
342 printk(KERN_ALERT "multipath: %s: unrecoverable IO read"
343 " error for block %llu\n",
344 bdevname(bio->bi_bdev,b),
Kent Overstreet4f024f32013-10-11 15:44:27 -0700345 (unsigned long long)bio->bi_iter.bi_sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 multipath_end_bh_io(mp_bh, -EIO);
347 } else {
348 printk(KERN_ERR "multipath: %s: redirecting sector %llu"
349 " to another IO path\n",
350 bdevname(bio->bi_bdev,b),
Kent Overstreet4f024f32013-10-11 15:44:27 -0700351 (unsigned long long)bio->bi_iter.bi_sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 *bio = *(mp_bh->master_bio);
Kent Overstreet4f024f32013-10-11 15:44:27 -0700353 bio->bi_iter.bi_sector +=
354 conf->multipaths[mp_bh->path].rdev->data_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 bio->bi_bdev = conf->multipaths[mp_bh->path].rdev->bdev;
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +0200356 bio->bi_rw |= REQ_FAILFAST_TRANSPORT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 bio->bi_end_io = multipath_end_request;
358 bio->bi_private = mp_bh;
359 generic_make_request(bio);
360 }
361 }
362 spin_unlock_irqrestore(&conf->device_lock, flags);
363}
364
NeilBrownfd01b882011-10-11 16:47:53 +1100365static sector_t multipath_size(struct mddev *mddev, sector_t sectors, int raid_disks)
Dan Williams80c3a6c2009-03-17 18:10:40 -0700366{
367 WARN_ONCE(sectors || raid_disks,
368 "%s does not support generic reshape\n", __func__);
369
370 return mddev->dev_sectors;
371}
372
NeilBrownfd01b882011-10-11 16:47:53 +1100373static int multipath_run (struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374{
NeilBrown69724e22011-10-11 16:48:57 +1100375 struct mpconf *conf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 int disk_idx;
377 struct multipath_info *disk;
NeilBrown3cb03002011-10-11 16:45:26 +1100378 struct md_rdev *rdev;
NeilBrown92f861a2011-05-11 14:38:02 +1000379 int working_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380
Andre Noll0894cc32009-06-18 08:49:23 +1000381 if (md_check_no_bitmap(mddev))
382 return -EINVAL;
383
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 if (mddev->level != LEVEL_MULTIPATH) {
385 printk("multipath: %s: raid level not set to multipath IO (%d)\n",
386 mdname(mddev), mddev->level);
387 goto out;
388 }
389 /*
390 * copy the already verified devices into our private MULTIPATH
391 * bookkeeping area. [whatever we allocate in multipath_run(),
NeilBrownafa0f552014-12-15 12:56:58 +1100392 * should be freed in multipath_free()]
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 */
394
NeilBrown69724e22011-10-11 16:48:57 +1100395 conf = kzalloc(sizeof(struct mpconf), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 mddev->private = conf;
397 if (!conf) {
NeilBrownf72ffdd2014-09-30 14:23:59 +1000398 printk(KERN_ERR
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 "multipath: couldn't allocate memory for %s\n",
400 mdname(mddev));
401 goto out;
402 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403
NeilBrown9ffae0c2006-01-06 00:20:32 -0800404 conf->multipaths = kzalloc(sizeof(struct multipath_info)*mddev->raid_disks,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 GFP_KERNEL);
406 if (!conf->multipaths) {
NeilBrownf72ffdd2014-09-30 14:23:59 +1000407 printk(KERN_ERR
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 "multipath: couldn't allocate memory for %s\n",
409 mdname(mddev));
410 goto out_free_conf;
411 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412
NeilBrown92f861a2011-05-11 14:38:02 +1000413 working_disks = 0;
NeilBrowndafb20f2012-03-19 12:46:39 +1100414 rdev_for_each(rdev, mddev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 disk_idx = rdev->raid_disk;
416 if (disk_idx < 0 ||
417 disk_idx >= mddev->raid_disks)
418 continue;
419
420 disk = conf->multipaths + disk_idx;
421 disk->rdev = rdev;
Martin K. Petersen8f6c2e42009-07-01 11:13:45 +1000422 disk_stack_limits(mddev->gendisk, rdev->bdev,
423 rdev->data_offset << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424
NeilBrownb2d444d2005-11-08 21:39:31 -0800425 if (!test_bit(Faulty, &rdev->flags))
NeilBrown92f861a2011-05-11 14:38:02 +1000426 working_disks++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 }
428
429 conf->raid_disks = mddev->raid_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 conf->mddev = mddev;
431 spin_lock_init(&conf->device_lock);
432 INIT_LIST_HEAD(&conf->retry_list);
433
NeilBrown92f861a2011-05-11 14:38:02 +1000434 if (!working_disks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 printk(KERN_ERR "multipath: no operational IO paths for %s\n",
436 mdname(mddev));
437 goto out_free_conf;
438 }
NeilBrown92f861a2011-05-11 14:38:02 +1000439 mddev->degraded = conf->raid_disks - working_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440
Sage Weilbbba8092009-09-21 17:02:55 -0700441 conf->pool = mempool_create_kmalloc_pool(NR_RESERVED_BUFS,
Matthew Dobson26b6e052006-03-26 01:37:48 -0800442 sizeof(struct multipath_bh));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 if (conf->pool == NULL) {
NeilBrownf72ffdd2014-09-30 14:23:59 +1000444 printk(KERN_ERR
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 "multipath: couldn't allocate memory for %s\n",
446 mdname(mddev));
447 goto out_free_conf;
448 }
449
450 {
NeilBrown02326052012-07-03 15:56:52 +1000451 mddev->thread = md_register_thread(multipathd, mddev,
452 "multipath");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 if (!mddev->thread) {
454 printk(KERN_ERR "multipath: couldn't allocate thread"
455 " for %s\n", mdname(mddev));
456 goto out_free_conf;
457 }
458 }
459
NeilBrownf72ffdd2014-09-30 14:23:59 +1000460 printk(KERN_INFO
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 "multipath: array %s active with %d out of %d IO paths\n",
NeilBrown92f861a2011-05-11 14:38:02 +1000462 mdname(mddev), conf->raid_disks - mddev->degraded,
463 mddev->raid_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 /*
465 * Ok, everything is just fine now
466 */
Dan Williams1f403622009-03-31 14:59:03 +1100467 md_set_array_sectors(mddev, multipath_size(mddev, 0, 0));
NeilBrown7a5febe2005-05-16 21:53:16 -0700468
Martin K. Petersena91a2782011-03-17 11:11:05 +0100469 if (md_integrity_register(mddev))
470 goto out_free_conf;
471
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 return 0;
473
474out_free_conf:
Julia Lawall644df1a2015-09-13 14:15:10 +0200475 mempool_destroy(conf->pool);
Jesper Juhl990a8ba2005-06-21 17:17:30 -0700476 kfree(conf->multipaths);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 kfree(conf);
478 mddev->private = NULL;
479out:
480 return -EIO;
481}
482
NeilBrownafa0f552014-12-15 12:56:58 +1100483static void multipath_free(struct mddev *mddev, void *priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484{
NeilBrownafa0f552014-12-15 12:56:58 +1100485 struct mpconf *conf = priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 mempool_destroy(conf->pool);
488 kfree(conf->multipaths);
489 kfree(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490}
491
NeilBrown84fc4b52011-10-11 16:49:58 +1100492static struct md_personality multipath_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493{
494 .name = "multipath",
NeilBrown2604b702006-01-06 00:20:36 -0800495 .level = LEVEL_MULTIPATH,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 .owner = THIS_MODULE,
497 .make_request = multipath_make_request,
498 .run = multipath_run,
NeilBrownafa0f552014-12-15 12:56:58 +1100499 .free = multipath_free,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 .status = multipath_status,
501 .error_handler = multipath_error,
502 .hot_add_disk = multipath_add_disk,
503 .hot_remove_disk= multipath_remove_disk,
Dan Williams80c3a6c2009-03-17 18:10:40 -0700504 .size = multipath_size,
NeilBrown5c675f82014-12-15 12:56:56 +1100505 .congested = multipath_congested,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506};
507
508static int __init multipath_init (void)
509{
NeilBrown2604b702006-01-06 00:20:36 -0800510 return register_md_personality (&multipath_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511}
512
513static void __exit multipath_exit (void)
514{
NeilBrown2604b702006-01-06 00:20:36 -0800515 unregister_md_personality (&multipath_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516}
517
518module_init(multipath_init);
519module_exit(multipath_exit);
520MODULE_LICENSE("GPL");
NeilBrown0efb9e62009-12-14 12:49:58 +1100521MODULE_DESCRIPTION("simple multi-path personality for MD");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522MODULE_ALIAS("md-personality-7"); /* MULTIPATH */
NeilBrownd9d166c2006-01-06 00:20:51 -0800523MODULE_ALIAS("md-multipath");
NeilBrown2604b702006-01-06 00:20:36 -0800524MODULE_ALIAS("md-level--4");