1/*
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
22#include <linux/blkdev.h>
23#include <linux/module.h>
24#include <linux/raid/md_u.h>
25#include <linux/seq_file.h>
26#include <linux/slab.h>
27#include "md.h"
28#include "md-multipath.h"
29
30#define MAX_WORK_PER_DISK 128
31
32#define NR_RESERVED_BUFS 32
33
34static int multipath_map (struct mpconf *conf)
35{
36 int i, disks = conf->raid_disks;
37
38 /*
39 * Later we do read balancing on the read side
40 * now we use the first available disk.
41 */
42
43 rcu_read_lock();
44 for (i = 0; i < disks; i++) {
45 struct md_rdev *rdev = rcu_dereference(conf->multipaths[i].rdev);
46 if (rdev && test_bit(In_sync, &rdev->flags) &&
47 !test_bit(Faulty, &rdev->flags)) {
48 atomic_inc(&rdev->nr_pending);
49 rcu_read_unlock();
50 return i;
51 }
52 }
53 rcu_read_unlock();
54
55 pr_crit_ratelimited("multipath_map(): no more operational IO paths?\n");
56 return (-1);
57}
58
59static void multipath_reschedule_retry (struct multipath_bh *mp_bh)
60{
61 unsigned long flags;
62 struct mddev *mddev = mp_bh->mddev;
63 struct mpconf *conf = mddev->private;
64
65 spin_lock_irqsave(&conf->device_lock, flags);
66 list_add(&mp_bh->retry_list, &conf->retry_list);
67 spin_unlock_irqrestore(&conf->device_lock, flags);
68 md_wakeup_thread(mddev->thread);
69}
70
71/*
72 * multipath_end_bh_io() is called when we have finished servicing a multipathed
73 * operation and are ready to return a success/failure code to the buffer
74 * cache layer.
75 */
76static void multipath_end_bh_io(struct multipath_bh *mp_bh, blk_status_t status)
77{
78 struct bio *bio = mp_bh->master_bio;
79 struct mpconf *conf = mp_bh->mddev->private;
80
81 bio->bi_status = status;
82 bio_endio(bio);
83 mempool_free(mp_bh, &conf->pool);
84}
85
86static void multipath_end_request(struct bio *bio)
87{
88 struct multipath_bh *mp_bh = bio->bi_private;
89 struct mpconf *conf = mp_bh->mddev->private;
90 struct md_rdev *rdev = conf->multipaths[mp_bh->path].rdev;
91
92 if (!bio->bi_status)
93 multipath_end_bh_io(mp_bh, 0);
94 else if (!(bio->bi_opf & REQ_RAHEAD)) {
95 /*
96 * oops, IO error:
97 */
98 char b[BDEVNAME_SIZE];
99 md_error (mp_bh->mddev, rdev);
100 pr_info("multipath: %s: rescheduling sector %llu\n",
101 bdevname(rdev->bdev,b),
102 (unsigned long long)bio->bi_iter.bi_sector);
103 multipath_reschedule_retry(mp_bh);
104 } else
105 multipath_end_bh_io(mp_bh, bio->bi_status);
106 rdev_dec_pending(rdev, conf->mddev);
107}
108
109static bool multipath_make_request(struct mddev *mddev, struct bio * bio)
110{
111 struct mpconf *conf = mddev->private;
112 struct multipath_bh * mp_bh;
113 struct multipath_info *multipath;
114
115 if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
116 md_flush_request(mddev, bio);
117 return true;
118 }
119
120 mp_bh = mempool_alloc(&conf->pool, GFP_NOIO);
121
122 mp_bh->master_bio = bio;
123 mp_bh->mddev = mddev;
124
125 mp_bh->path = multipath_map(conf);
126 if (mp_bh->path < 0) {
127 bio_io_error(bio);
128 mempool_free(mp_bh, &conf->pool);
129 return true;
130 }
131 multipath = conf->multipaths + mp_bh->path;
132
133 bio_init(&mp_bh->bio, NULL, 0);
134 __bio_clone_fast(&mp_bh->bio, bio);
135
136 mp_bh->bio.bi_iter.bi_sector += multipath->rdev->data_offset;
137 bio_set_dev(&mp_bh->bio, multipath->rdev->bdev);
138 mp_bh->bio.bi_opf |= REQ_FAILFAST_TRANSPORT;
139 mp_bh->bio.bi_end_io = multipath_end_request;
140 mp_bh->bio.bi_private = mp_bh;
141 mddev_check_writesame(mddev, &mp_bh->bio);
142 mddev_check_write_zeroes(mddev, &mp_bh->bio);
143 generic_make_request(&mp_bh->bio);
144 return true;
145}
146
147static void multipath_status(struct seq_file *seq, struct mddev *mddev)
148{
149 struct mpconf *conf = mddev->private;
150 int i;
151
152 seq_printf (seq, " [%d/%d] [", conf->raid_disks,
153 conf->raid_disks - mddev->degraded);
154 rcu_read_lock();
155 for (i = 0; i < conf->raid_disks; i++) {
156 struct md_rdev *rdev = rcu_dereference(conf->multipaths[i].rdev);
157 seq_printf (seq, "%s", rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_");
158 }
159 rcu_read_unlock();
160 seq_putc(seq, ']');
161}
162
163static int multipath_congested(struct mddev *mddev, int bits)
164{
165 struct mpconf *conf = mddev->private;
166 int i, ret = 0;
167
168 rcu_read_lock();
169 for (i = 0; i < mddev->raid_disks ; i++) {
170 struct md_rdev *rdev = rcu_dereference(conf->multipaths[i].rdev);
171 if (rdev && !test_bit(Faulty, &rdev->flags)) {
172 struct request_queue *q = bdev_get_queue(rdev->bdev);
173
174 ret |= bdi_congested(q->backing_dev_info, bits);
175 /* Just like multipath_map, we just check the
176 * first available device
177 */
178 break;
179 }
180 }
181 rcu_read_unlock();
182 return ret;
183}
184
185/*
186 * Careful, this can execute in IRQ contexts as well!
187 */
188static void multipath_error (struct mddev *mddev, struct md_rdev *rdev)
189{
190 struct mpconf *conf = mddev->private;
191 char b[BDEVNAME_SIZE];
192
193 if (conf->raid_disks - mddev->degraded <= 1) {
194 /*
195 * Uh oh, we can do nothing if this is our last path, but
196 * first check if this is a queued request for a device
197 * which has just failed.
198 */
199 pr_warn("multipath: only one IO path left and IO error.\n");
200 /* leave it active... it's all we have */
201 return;
202 }
203 /*
204 * Mark disk as unusable
205 */
206 if (test_and_clear_bit(In_sync, &rdev->flags)) {
207 unsigned long flags;
208 spin_lock_irqsave(&conf->device_lock, flags);
209 mddev->degraded++;
210 spin_unlock_irqrestore(&conf->device_lock, flags);
211 }
212 set_bit(Faulty, &rdev->flags);
213 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
214 pr_err("multipath: IO failure on %s, disabling IO path.\n"
215 "multipath: Operation continuing on %d IO paths.\n",
216 bdevname(rdev->bdev, b),
217 conf->raid_disks - mddev->degraded);
218}
219
220static void print_multipath_conf (struct mpconf *conf)
221{
222 int i;
223 struct multipath_info *tmp;
224
225 pr_debug("MULTIPATH conf printout:\n");
226 if (!conf) {
227 pr_debug("(conf==NULL)\n");
228 return;
229 }
230 pr_debug(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded,
231 conf->raid_disks);
232
233 for (i = 0; i < conf->raid_disks; i++) {
234 char b[BDEVNAME_SIZE];
235 tmp = conf->multipaths + i;
236 if (tmp->rdev)
237 pr_debug(" disk%d, o:%d, dev:%s\n",
238 i,!test_bit(Faulty, &tmp->rdev->flags),
239 bdevname(tmp->rdev->bdev,b));
240 }
241}
242
243static int multipath_add_disk(struct mddev *mddev, struct md_rdev *rdev)
244{
245 struct mpconf *conf = mddev->private;
246 int err = -EEXIST;
247 int path;
248 struct multipath_info *p;
249 int first = 0;
250 int last = mddev->raid_disks - 1;
251
252 if (rdev->raid_disk >= 0)
253 first = last = rdev->raid_disk;
254
255 print_multipath_conf(conf);
256
257 for (path = first; path <= last; path++)
258 if ((p=conf->multipaths+path)->rdev == NULL) {
259 disk_stack_limits(mddev->gendisk, rdev->bdev,
260 rdev->data_offset << 9);
261
262 err = md_integrity_add_rdev(rdev, mddev);
263 if (err)
264 break;
265 spin_lock_irq(&conf->device_lock);
266 mddev->degraded--;
267 rdev->raid_disk = path;
268 set_bit(In_sync, &rdev->flags);
269 spin_unlock_irq(&conf->device_lock);
270 rcu_assign_pointer(p->rdev, rdev);
271 err = 0;
272 break;
273 }
274
275 print_multipath_conf(conf);
276
277 return err;
278}
279
280static int multipath_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
281{
282 struct mpconf *conf = mddev->private;
283 int err = 0;
284 int number = rdev->raid_disk;
285 struct multipath_info *p = conf->multipaths + number;
286
287 print_multipath_conf(conf);
288
289 if (rdev == p->rdev) {
290 if (test_bit(In_sync, &rdev->flags) ||
291 atomic_read(&rdev->nr_pending)) {
292 pr_warn("hot-remove-disk, slot %d is identified but is still operational!\n", number);
293 err = -EBUSY;
294 goto abort;
295 }
296 p->rdev = NULL;
297 if (!test_bit(RemoveSynchronized, &rdev->flags)) {
298 synchronize_rcu();
299 if (atomic_read(&rdev->nr_pending)) {
300 /* lost the race, try later */
301 err = -EBUSY;
302 p->rdev = rdev;
303 goto abort;
304 }
305 }
306 err = md_integrity_register(mddev);
307 }
308abort:
309
310 print_multipath_conf(conf);
311 return err;
312}
313
314/*
315 * This is a kernel thread which:
316 *
317 * 1. Retries failed read operations on working multipaths.
318 * 2. Updates the raid superblock when problems encounter.
319 * 3. Performs writes following reads for array syncronising.
320 */
321
322static void multipathd(struct md_thread *thread)
323{
324 struct mddev *mddev = thread->mddev;
325 struct multipath_bh *mp_bh;
326 struct bio *bio;
327 unsigned long flags;
328 struct mpconf *conf = mddev->private;
329 struct list_head *head = &conf->retry_list;
330
331 md_check_recovery(mddev);
332 for (;;) {
333 char b[BDEVNAME_SIZE];
334 spin_lock_irqsave(&conf->device_lock, flags);
335 if (list_empty(head))
336 break;
337 mp_bh = list_entry(head->prev, struct multipath_bh, retry_list);
338 list_del(head->prev);
339 spin_unlock_irqrestore(&conf->device_lock, flags);
340
341 bio = &mp_bh->bio;
342 bio->bi_iter.bi_sector = mp_bh->master_bio->bi_iter.bi_sector;
343
344 if ((mp_bh->path = multipath_map (conf))<0) {
345 pr_err("multipath: %s: unrecoverable IO read error for block %llu\n",
346 bio_devname(bio, b),
347 (unsigned long long)bio->bi_iter.bi_sector);
348 multipath_end_bh_io(mp_bh, BLK_STS_IOERR);
349 } else {
350 pr_err("multipath: %s: redirecting sector %llu to another IO path\n",
351 bio_devname(bio, b),
352 (unsigned long long)bio->bi_iter.bi_sector);
353 *bio = *(mp_bh->master_bio);
354 bio->bi_iter.bi_sector +=
355 conf->multipaths[mp_bh->path].rdev->data_offset;
356 bio_set_dev(bio, conf->multipaths[mp_bh->path].rdev->bdev);
357 bio->bi_opf |= REQ_FAILFAST_TRANSPORT;
358 bio->bi_end_io = multipath_end_request;
359 bio->bi_private = mp_bh;
360 generic_make_request(bio);
361 }
362 }
363 spin_unlock_irqrestore(&conf->device_lock, flags);
364}
365
366static sector_t multipath_size(struct mddev *mddev, sector_t sectors, int raid_disks)
367{
368 WARN_ONCE(sectors || raid_disks,
369 "%s does not support generic reshape\n", __func__);
370
371 return mddev->dev_sectors;
372}
373
374static int multipath_run (struct mddev *mddev)
375{
376 struct mpconf *conf;
377 int disk_idx;
378 struct multipath_info *disk;
379 struct md_rdev *rdev;
380 int working_disks;
381 int ret;
382
383 if (md_check_no_bitmap(mddev))
384 return -EINVAL;
385
386 if (mddev->level != LEVEL_MULTIPATH) {
387 pr_warn("multipath: %s: raid level not set to multipath IO (%d)\n",
388 mdname(mddev), mddev->level);
389 goto out;
390 }
391 /*
392 * copy the already verified devices into our private MULTIPATH
393 * bookkeeping area. [whatever we allocate in multipath_run(),
394 * should be freed in multipath_free()]
395 */
396
397 conf = kzalloc(sizeof(struct mpconf), GFP_KERNEL);
398 mddev->private = conf;
399 if (!conf)
400 goto out;
401
402 conf->multipaths = kcalloc(mddev->raid_disks,
403 sizeof(struct multipath_info),
404 GFP_KERNEL);
405 if (!conf->multipaths)
406 goto out_free_conf;
407
408 working_disks = 0;
409 rdev_for_each(rdev, mddev) {
410 disk_idx = rdev->raid_disk;
411 if (disk_idx < 0 ||
412 disk_idx >= mddev->raid_disks)
413 continue;
414
415 disk = conf->multipaths + disk_idx;
416 disk->rdev = rdev;
417 disk_stack_limits(mddev->gendisk, rdev->bdev,
418 rdev->data_offset << 9);
419
420 if (!test_bit(Faulty, &rdev->flags))
421 working_disks++;
422 }
423
424 conf->raid_disks = mddev->raid_disks;
425 conf->mddev = mddev;
426 spin_lock_init(&conf->device_lock);
427 INIT_LIST_HEAD(&conf->retry_list);
428
429 if (!working_disks) {
430 pr_warn("multipath: no operational IO paths for %s\n",
431 mdname(mddev));
432 goto out_free_conf;
433 }
434 mddev->degraded = conf->raid_disks - working_disks;
435
436 ret = mempool_init_kmalloc_pool(&conf->pool, NR_RESERVED_BUFS,
437 sizeof(struct multipath_bh));
438 if (ret)
439 goto out_free_conf;
440
441 mddev->thread = md_register_thread(multipathd, mddev,
442 "multipath");
443 if (!mddev->thread)
444 goto out_free_conf;
445
446 pr_info("multipath: array %s active with %d out of %d IO paths\n",
447 mdname(mddev), conf->raid_disks - mddev->degraded,
448 mddev->raid_disks);
449 /*
450 * Ok, everything is just fine now
451 */
452 md_set_array_sectors(mddev, multipath_size(mddev, 0, 0));
453
454 if (md_integrity_register(mddev))
455 goto out_free_conf;
456
457 return 0;
458
459out_free_conf:
460 mempool_exit(&conf->pool);
461 kfree(conf->multipaths);
462 kfree(conf);
463 mddev->private = NULL;
464out:
465 return -EIO;
466}
467
468static void multipath_free(struct mddev *mddev, void *priv)
469{
470 struct mpconf *conf = priv;
471
472 mempool_exit(&conf->pool);
473 kfree(conf->multipaths);
474 kfree(conf);
475}
476
477static struct md_personality multipath_personality =
478{
479 .name = "multipath",
480 .level = LEVEL_MULTIPATH,
481 .owner = THIS_MODULE,
482 .make_request = multipath_make_request,
483 .run = multipath_run,
484 .free = multipath_free,
485 .status = multipath_status,
486 .error_handler = multipath_error,
487 .hot_add_disk = multipath_add_disk,
488 .hot_remove_disk= multipath_remove_disk,
489 .size = multipath_size,
490 .congested = multipath_congested,
491};
492
493static int __init multipath_init (void)
494{
495 return register_md_personality (&multipath_personality);
496}
497
498static void __exit multipath_exit (void)
499{
500 unregister_md_personality (&multipath_personality);
501}
502
503module_init(multipath_init);
504module_exit(multipath_exit);
505MODULE_LICENSE("GPL");
506MODULE_DESCRIPTION("simple multi-path personality for MD");
507MODULE_ALIAS("md-personality-7"); /* MULTIPATH */
508MODULE_ALIAS("md-multipath");
509MODULE_ALIAS("md-level--4");
510