1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * multipath.c : Multiple Devices driver for Linux
5 * Copyright (C) 1999, 2000, 2001 Ingo Molnar, Red Hat
7 * Copyright (C) 1996, 1997, 1998 Ingo Molnar, Miguel de Icaza, Gadi Oxman
9 * MULTIPATH management functions.
11 * derived from raid1.c.
14 #include <linux/blkdev.h>
15 #include <linux/module.h>
16 #include <linux/raid/md_u.h>
17 #include <linux/seq_file.h>
18 #include <linux/slab.h>
20 #include "md-multipath.h"
22 #define MAX_WORK_PER_DISK 128
24 #define NR_RESERVED_BUFS 32
26 static int multipath_map (struct mpconf *conf)
28 int i, disks = conf->raid_disks;
31 * Later we do read balancing on the read side
32 * now we use the first available disk.
36 for (i = 0; i < disks; i++) {
37 struct md_rdev *rdev = rcu_dereference(conf->multipaths[i].rdev);
38 if (rdev && test_bit(In_sync, &rdev->flags) &&
39 !test_bit(Faulty, &rdev->flags)) {
40 atomic_inc(&rdev->nr_pending);
47 pr_crit_ratelimited("multipath_map(): no more operational IO paths?\n");
51 static void multipath_reschedule_retry (struct multipath_bh *mp_bh)
54 struct mddev *mddev = mp_bh->mddev;
55 struct mpconf *conf = mddev->private;
57 spin_lock_irqsave(&conf->device_lock, flags);
58 list_add(&mp_bh->retry_list, &conf->retry_list);
59 spin_unlock_irqrestore(&conf->device_lock, flags);
60 md_wakeup_thread(mddev->thread);
64 * multipath_end_bh_io() is called when we have finished servicing a multipathed
65 * operation and are ready to return a success/failure code to the buffer
68 static void multipath_end_bh_io(struct multipath_bh *mp_bh, blk_status_t status)
70 struct bio *bio = mp_bh->master_bio;
71 struct mpconf *conf = mp_bh->mddev->private;
73 bio->bi_status = status;
75 mempool_free(mp_bh, &conf->pool);
78 static void multipath_end_request(struct bio *bio)
80 struct multipath_bh *mp_bh = bio->bi_private;
81 struct mpconf *conf = mp_bh->mddev->private;
82 struct md_rdev *rdev = conf->multipaths[mp_bh->path].rdev;
85 multipath_end_bh_io(mp_bh, 0);
86 else if (!(bio->bi_opf & REQ_RAHEAD)) {
90 char b[BDEVNAME_SIZE];
91 md_error (mp_bh->mddev, rdev);
92 pr_info("multipath: %s: rescheduling sector %llu\n",
93 bdevname(rdev->bdev,b),
94 (unsigned long long)bio->bi_iter.bi_sector);
95 multipath_reschedule_retry(mp_bh);
97 multipath_end_bh_io(mp_bh, bio->bi_status);
98 rdev_dec_pending(rdev, conf->mddev);
101 static bool multipath_make_request(struct mddev *mddev, struct bio * bio)
103 struct mpconf *conf = mddev->private;
104 struct multipath_bh * mp_bh;
105 struct multipath_info *multipath;
107 if (unlikely(bio->bi_opf & REQ_PREFLUSH)
108 && md_flush_request(mddev, bio))
111 mp_bh = mempool_alloc(&conf->pool, GFP_NOIO);
113 mp_bh->master_bio = bio;
114 mp_bh->mddev = mddev;
116 mp_bh->path = multipath_map(conf);
117 if (mp_bh->path < 0) {
119 mempool_free(mp_bh, &conf->pool);
122 multipath = conf->multipaths + mp_bh->path;
124 bio_init_clone(multipath->rdev->bdev, &mp_bh->bio, bio, GFP_NOIO);
126 mp_bh->bio.bi_iter.bi_sector += multipath->rdev->data_offset;
127 mp_bh->bio.bi_opf |= REQ_FAILFAST_TRANSPORT;
128 mp_bh->bio.bi_end_io = multipath_end_request;
129 mp_bh->bio.bi_private = mp_bh;
130 mddev_check_write_zeroes(mddev, &mp_bh->bio);
131 submit_bio_noacct(&mp_bh->bio);
135 static void multipath_status(struct seq_file *seq, struct mddev *mddev)
137 struct mpconf *conf = mddev->private;
140 seq_printf (seq, " [%d/%d] [", conf->raid_disks,
141 conf->raid_disks - mddev->degraded);
143 for (i = 0; i < conf->raid_disks; i++) {
144 struct md_rdev *rdev = rcu_dereference(conf->multipaths[i].rdev);
145 seq_printf (seq, "%s", rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_");
152 * Careful, this can execute in IRQ contexts as well!
154 static void multipath_error (struct mddev *mddev, struct md_rdev *rdev)
156 struct mpconf *conf = mddev->private;
157 char b[BDEVNAME_SIZE];
159 if (conf->raid_disks - mddev->degraded <= 1) {
161 * Uh oh, we can do nothing if this is our last path, but
162 * first check if this is a queued request for a device
163 * which has just failed.
165 pr_warn("multipath: only one IO path left and IO error.\n");
166 /* leave it active... it's all we have */
170 * Mark disk as unusable
172 if (test_and_clear_bit(In_sync, &rdev->flags)) {
174 spin_lock_irqsave(&conf->device_lock, flags);
176 spin_unlock_irqrestore(&conf->device_lock, flags);
178 set_bit(Faulty, &rdev->flags);
179 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
180 pr_err("multipath: IO failure on %s, disabling IO path.\n"
181 "multipath: Operation continuing on %d IO paths.\n",
182 bdevname(rdev->bdev, b),
183 conf->raid_disks - mddev->degraded);
186 static void print_multipath_conf (struct mpconf *conf)
189 struct multipath_info *tmp;
191 pr_debug("MULTIPATH conf printout:\n");
193 pr_debug("(conf==NULL)\n");
196 pr_debug(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded,
199 for (i = 0; i < conf->raid_disks; i++) {
200 char b[BDEVNAME_SIZE];
201 tmp = conf->multipaths + i;
203 pr_debug(" disk%d, o:%d, dev:%s\n",
204 i,!test_bit(Faulty, &tmp->rdev->flags),
205 bdevname(tmp->rdev->bdev,b));
209 static int multipath_add_disk(struct mddev *mddev, struct md_rdev *rdev)
211 struct mpconf *conf = mddev->private;
214 struct multipath_info *p;
216 int last = mddev->raid_disks - 1;
218 if (rdev->raid_disk >= 0)
219 first = last = rdev->raid_disk;
221 print_multipath_conf(conf);
223 for (path = first; path <= last; path++)
224 if ((p=conf->multipaths+path)->rdev == NULL) {
225 disk_stack_limits(mddev->gendisk, rdev->bdev,
226 rdev->data_offset << 9);
228 err = md_integrity_add_rdev(rdev, mddev);
231 spin_lock_irq(&conf->device_lock);
233 rdev->raid_disk = path;
234 set_bit(In_sync, &rdev->flags);
235 spin_unlock_irq(&conf->device_lock);
236 rcu_assign_pointer(p->rdev, rdev);
241 print_multipath_conf(conf);
246 static int multipath_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
248 struct mpconf *conf = mddev->private;
250 int number = rdev->raid_disk;
251 struct multipath_info *p = conf->multipaths + number;
253 print_multipath_conf(conf);
255 if (rdev == p->rdev) {
256 if (test_bit(In_sync, &rdev->flags) ||
257 atomic_read(&rdev->nr_pending)) {
258 pr_warn("hot-remove-disk, slot %d is identified but is still operational!\n", number);
263 if (!test_bit(RemoveSynchronized, &rdev->flags)) {
265 if (atomic_read(&rdev->nr_pending)) {
266 /* lost the race, try later */
272 err = md_integrity_register(mddev);
276 print_multipath_conf(conf);
281 * This is a kernel thread which:
283 * 1. Retries failed read operations on working multipaths.
284 * 2. Updates the raid superblock when problems encounter.
285 * 3. Performs writes following reads for array syncronising.
288 static void multipathd(struct md_thread *thread)
290 struct mddev *mddev = thread->mddev;
291 struct multipath_bh *mp_bh;
294 struct mpconf *conf = mddev->private;
295 struct list_head *head = &conf->retry_list;
297 md_check_recovery(mddev);
299 spin_lock_irqsave(&conf->device_lock, flags);
300 if (list_empty(head))
302 mp_bh = list_entry(head->prev, struct multipath_bh, retry_list);
303 list_del(head->prev);
304 spin_unlock_irqrestore(&conf->device_lock, flags);
307 bio->bi_iter.bi_sector = mp_bh->master_bio->bi_iter.bi_sector;
309 if ((mp_bh->path = multipath_map (conf))<0) {
310 pr_err("multipath: %pg: unrecoverable IO read error for block %llu\n",
312 (unsigned long long)bio->bi_iter.bi_sector);
313 multipath_end_bh_io(mp_bh, BLK_STS_IOERR);
315 pr_err("multipath: %pg: redirecting sector %llu to another IO path\n",
317 (unsigned long long)bio->bi_iter.bi_sector);
318 *bio = *(mp_bh->master_bio);
319 bio->bi_iter.bi_sector +=
320 conf->multipaths[mp_bh->path].rdev->data_offset;
321 bio_set_dev(bio, conf->multipaths[mp_bh->path].rdev->bdev);
322 bio->bi_opf |= REQ_FAILFAST_TRANSPORT;
323 bio->bi_end_io = multipath_end_request;
324 bio->bi_private = mp_bh;
325 submit_bio_noacct(bio);
328 spin_unlock_irqrestore(&conf->device_lock, flags);
331 static sector_t multipath_size(struct mddev *mddev, sector_t sectors, int raid_disks)
333 WARN_ONCE(sectors || raid_disks,
334 "%s does not support generic reshape\n", __func__);
336 return mddev->dev_sectors;
339 static int multipath_run (struct mddev *mddev)
343 struct multipath_info *disk;
344 struct md_rdev *rdev;
348 if (md_check_no_bitmap(mddev))
351 if (mddev->level != LEVEL_MULTIPATH) {
352 pr_warn("multipath: %s: raid level not set to multipath IO (%d)\n",
353 mdname(mddev), mddev->level);
357 * copy the already verified devices into our private MULTIPATH
358 * bookkeeping area. [whatever we allocate in multipath_run(),
359 * should be freed in multipath_free()]
362 conf = kzalloc(sizeof(struct mpconf), GFP_KERNEL);
363 mddev->private = conf;
367 conf->multipaths = kcalloc(mddev->raid_disks,
368 sizeof(struct multipath_info),
370 if (!conf->multipaths)
374 rdev_for_each(rdev, mddev) {
375 disk_idx = rdev->raid_disk;
377 disk_idx >= mddev->raid_disks)
380 disk = conf->multipaths + disk_idx;
382 disk_stack_limits(mddev->gendisk, rdev->bdev,
383 rdev->data_offset << 9);
385 if (!test_bit(Faulty, &rdev->flags))
389 conf->raid_disks = mddev->raid_disks;
391 spin_lock_init(&conf->device_lock);
392 INIT_LIST_HEAD(&conf->retry_list);
394 if (!working_disks) {
395 pr_warn("multipath: no operational IO paths for %s\n",
399 mddev->degraded = conf->raid_disks - working_disks;
401 ret = mempool_init_kmalloc_pool(&conf->pool, NR_RESERVED_BUFS,
402 sizeof(struct multipath_bh));
406 mddev->thread = md_register_thread(multipathd, mddev,
411 pr_info("multipath: array %s active with %d out of %d IO paths\n",
412 mdname(mddev), conf->raid_disks - mddev->degraded,
415 * Ok, everything is just fine now
417 md_set_array_sectors(mddev, multipath_size(mddev, 0, 0));
419 if (md_integrity_register(mddev))
425 mempool_exit(&conf->pool);
426 kfree(conf->multipaths);
428 mddev->private = NULL;
433 static void multipath_free(struct mddev *mddev, void *priv)
435 struct mpconf *conf = priv;
437 mempool_exit(&conf->pool);
438 kfree(conf->multipaths);
442 static struct md_personality multipath_personality =
445 .level = LEVEL_MULTIPATH,
446 .owner = THIS_MODULE,
447 .make_request = multipath_make_request,
448 .run = multipath_run,
449 .free = multipath_free,
450 .status = multipath_status,
451 .error_handler = multipath_error,
452 .hot_add_disk = multipath_add_disk,
453 .hot_remove_disk= multipath_remove_disk,
454 .size = multipath_size,
457 static int __init multipath_init (void)
459 return register_md_personality (&multipath_personality);
462 static void __exit multipath_exit (void)
464 unregister_md_personality (&multipath_personality);
467 module_init(multipath_init);
468 module_exit(multipath_exit);
469 MODULE_LICENSE("GPL");
470 MODULE_DESCRIPTION("simple multi-path personality for MD (deprecated)");
471 MODULE_ALIAS("md-personality-7"); /* MULTIPATH */
472 MODULE_ALIAS("md-multipath");
473 MODULE_ALIAS("md-level--4");