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 md_error (mp_bh->mddev, rdev);
91 pr_info("multipath: %pg: rescheduling sector %llu\n",
93 (unsigned long long)bio->bi_iter.bi_sector);
94 multipath_reschedule_retry(mp_bh);
96 multipath_end_bh_io(mp_bh, bio->bi_status);
97 rdev_dec_pending(rdev, conf->mddev);
100 static bool multipath_make_request(struct mddev *mddev, struct bio * bio)
102 struct mpconf *conf = mddev->private;
103 struct multipath_bh * mp_bh;
104 struct multipath_info *multipath;
106 if (unlikely(bio->bi_opf & REQ_PREFLUSH)
107 && md_flush_request(mddev, bio))
110 md_account_bio(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;
158 if (conf->raid_disks - mddev->degraded <= 1) {
160 * Uh oh, we can do nothing if this is our last path, but
161 * first check if this is a queued request for a device
162 * which has just failed.
164 pr_warn("multipath: only one IO path left and IO error.\n");
165 /* leave it active... it's all we have */
169 * Mark disk as unusable
171 if (test_and_clear_bit(In_sync, &rdev->flags)) {
173 spin_lock_irqsave(&conf->device_lock, flags);
175 spin_unlock_irqrestore(&conf->device_lock, flags);
177 set_bit(Faulty, &rdev->flags);
178 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
179 pr_err("multipath: IO failure on %pg, disabling IO path.\n"
180 "multipath: Operation continuing on %d IO paths.\n",
182 conf->raid_disks - mddev->degraded);
185 static void print_multipath_conf (struct mpconf *conf)
188 struct multipath_info *tmp;
190 pr_debug("MULTIPATH conf printout:\n");
192 pr_debug("(conf==NULL)\n");
195 pr_debug(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded,
198 for (i = 0; i < conf->raid_disks; i++) {
199 tmp = conf->multipaths + i;
201 pr_debug(" disk%d, o:%d, dev:%pg\n",
202 i,!test_bit(Faulty, &tmp->rdev->flags),
207 static int multipath_add_disk(struct mddev *mddev, struct md_rdev *rdev)
209 struct mpconf *conf = mddev->private;
212 struct multipath_info *p;
214 int last = mddev->raid_disks - 1;
216 if (rdev->raid_disk >= 0)
217 first = last = rdev->raid_disk;
219 print_multipath_conf(conf);
221 for (path = first; path <= last; path++)
222 if ((p=conf->multipaths+path)->rdev == NULL) {
223 disk_stack_limits(mddev->gendisk, rdev->bdev,
224 rdev->data_offset << 9);
226 err = md_integrity_add_rdev(rdev, mddev);
229 spin_lock_irq(&conf->device_lock);
231 rdev->raid_disk = path;
232 set_bit(In_sync, &rdev->flags);
233 spin_unlock_irq(&conf->device_lock);
234 rcu_assign_pointer(p->rdev, rdev);
239 print_multipath_conf(conf);
244 static int multipath_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
246 struct mpconf *conf = mddev->private;
248 int number = rdev->raid_disk;
249 struct multipath_info *p = conf->multipaths + number;
251 print_multipath_conf(conf);
253 if (rdev == p->rdev) {
254 if (test_bit(In_sync, &rdev->flags) ||
255 atomic_read(&rdev->nr_pending)) {
256 pr_warn("hot-remove-disk, slot %d is identified but is still operational!\n", number);
261 if (!test_bit(RemoveSynchronized, &rdev->flags)) {
263 if (atomic_read(&rdev->nr_pending)) {
264 /* lost the race, try later */
270 err = md_integrity_register(mddev);
274 print_multipath_conf(conf);
279 * This is a kernel thread which:
281 * 1. Retries failed read operations on working multipaths.
282 * 2. Updates the raid superblock when problems encounter.
283 * 3. Performs writes following reads for array syncronising.
286 static void multipathd(struct md_thread *thread)
288 struct mddev *mddev = thread->mddev;
289 struct multipath_bh *mp_bh;
292 struct mpconf *conf = mddev->private;
293 struct list_head *head = &conf->retry_list;
295 md_check_recovery(mddev);
297 spin_lock_irqsave(&conf->device_lock, flags);
298 if (list_empty(head))
300 mp_bh = list_entry(head->prev, struct multipath_bh, retry_list);
301 list_del(head->prev);
302 spin_unlock_irqrestore(&conf->device_lock, flags);
305 bio->bi_iter.bi_sector = mp_bh->master_bio->bi_iter.bi_sector;
307 if ((mp_bh->path = multipath_map (conf))<0) {
308 pr_err("multipath: %pg: unrecoverable IO read error for block %llu\n",
310 (unsigned long long)bio->bi_iter.bi_sector);
311 multipath_end_bh_io(mp_bh, BLK_STS_IOERR);
313 pr_err("multipath: %pg: redirecting sector %llu to another IO path\n",
315 (unsigned long long)bio->bi_iter.bi_sector);
316 *bio = *(mp_bh->master_bio);
317 bio->bi_iter.bi_sector +=
318 conf->multipaths[mp_bh->path].rdev->data_offset;
319 bio_set_dev(bio, conf->multipaths[mp_bh->path].rdev->bdev);
320 bio->bi_opf |= REQ_FAILFAST_TRANSPORT;
321 bio->bi_end_io = multipath_end_request;
322 bio->bi_private = mp_bh;
323 submit_bio_noacct(bio);
326 spin_unlock_irqrestore(&conf->device_lock, flags);
329 static sector_t multipath_size(struct mddev *mddev, sector_t sectors, int raid_disks)
331 WARN_ONCE(sectors || raid_disks,
332 "%s does not support generic reshape\n", __func__);
334 return mddev->dev_sectors;
337 static int multipath_run (struct mddev *mddev)
341 struct multipath_info *disk;
342 struct md_rdev *rdev;
346 if (md_check_no_bitmap(mddev))
349 if (mddev->level != LEVEL_MULTIPATH) {
350 pr_warn("multipath: %s: raid level not set to multipath IO (%d)\n",
351 mdname(mddev), mddev->level);
355 * copy the already verified devices into our private MULTIPATH
356 * bookkeeping area. [whatever we allocate in multipath_run(),
357 * should be freed in multipath_free()]
360 conf = kzalloc(sizeof(struct mpconf), GFP_KERNEL);
361 mddev->private = conf;
365 conf->multipaths = kcalloc(mddev->raid_disks,
366 sizeof(struct multipath_info),
368 if (!conf->multipaths)
372 rdev_for_each(rdev, mddev) {
373 disk_idx = rdev->raid_disk;
375 disk_idx >= mddev->raid_disks)
378 disk = conf->multipaths + disk_idx;
380 disk_stack_limits(mddev->gendisk, rdev->bdev,
381 rdev->data_offset << 9);
383 if (!test_bit(Faulty, &rdev->flags))
387 conf->raid_disks = mddev->raid_disks;
389 spin_lock_init(&conf->device_lock);
390 INIT_LIST_HEAD(&conf->retry_list);
392 if (!working_disks) {
393 pr_warn("multipath: no operational IO paths for %s\n",
397 mddev->degraded = conf->raid_disks - working_disks;
399 ret = mempool_init_kmalloc_pool(&conf->pool, NR_RESERVED_BUFS,
400 sizeof(struct multipath_bh));
404 rcu_assign_pointer(mddev->thread,
405 md_register_thread(multipathd, mddev, "multipath"));
409 pr_info("multipath: array %s active with %d out of %d IO paths\n",
410 mdname(mddev), conf->raid_disks - mddev->degraded,
413 * Ok, everything is just fine now
415 md_set_array_sectors(mddev, multipath_size(mddev, 0, 0));
417 if (md_integrity_register(mddev))
423 mempool_exit(&conf->pool);
424 kfree(conf->multipaths);
426 mddev->private = NULL;
431 static void multipath_free(struct mddev *mddev, void *priv)
433 struct mpconf *conf = priv;
435 mempool_exit(&conf->pool);
436 kfree(conf->multipaths);
440 static struct md_personality multipath_personality =
443 .level = LEVEL_MULTIPATH,
444 .owner = THIS_MODULE,
445 .make_request = multipath_make_request,
446 .run = multipath_run,
447 .free = multipath_free,
448 .status = multipath_status,
449 .error_handler = multipath_error,
450 .hot_add_disk = multipath_add_disk,
451 .hot_remove_disk= multipath_remove_disk,
452 .size = multipath_size,
455 static int __init multipath_init (void)
457 return register_md_personality (&multipath_personality);
460 static void __exit multipath_exit (void)
462 unregister_md_personality (&multipath_personality);
465 module_init(multipath_init);
466 module_exit(multipath_exit);
467 MODULE_LICENSE("GPL");
468 MODULE_DESCRIPTION("simple multi-path personality for MD (deprecated)");
469 MODULE_ALIAS("md-personality-7"); /* MULTIPATH */
470 MODULE_ALIAS("md-multipath");
471 MODULE_ALIAS("md-level--4");