?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
serial core: fix new kernel-doc warnings
[linux/fpc-iii.git] / drivers / md / multipath.c
blobd2d3fd54cc681184cab146bf90ddc000f3b706de
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/raid/md_u.h>
24 #include <linux/seq_file.h>
25 #include "md.h"
26 #include "multipath.h"
27
28 #define MAX_WORK_PER_DISK 128
29
30 #define NR_RESERVED_BUFS 32
31
32
33 static int multipath_map (multipath_conf_t *conf)
34 {
35 int i, disks = conf->raid_disks;
36
37 /*
38 * Later we do read balancing on the read side
39 * now we use the first available disk.
40 */
41
42 rcu_read_lock();
43 for (i = 0; i < disks; i++) {
44 mdk_rdev_t *rdev = rcu_dereference(conf->multipaths[i].rdev);
45 if (rdev && test_bit(In_sync, &rdev->flags)) {
46 atomic_inc(&rdev->nr_pending);
47 rcu_read_unlock();
48 return i;
49 }
50 }
51 rcu_read_unlock();
52
53 printk(KERN_ERR "multipath_map(): no more operational IO paths?\n");
54 return (-1);
55 }
56
57 static void multipath_reschedule_retry (struct multipath_bh *mp_bh)
58 {
59 unsigned long flags;
60 mddev_t *mddev = mp_bh->mddev;
61 multipath_conf_t *conf = mddev->private;
62
63 spin_lock_irqsave(&conf->device_lock, flags);
64 list_add(&mp_bh->retry_list, &conf->retry_list);
65 spin_unlock_irqrestore(&conf->device_lock, flags);
66 md_wakeup_thread(mddev->thread);
67 }
68
69
70 /*
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 */
75 static void multipath_end_bh_io (struct multipath_bh *mp_bh, int err)
76 {
77 struct bio *bio = mp_bh->master_bio;
78 multipath_conf_t *conf = mp_bh->mddev->private;
79
80 bio_endio(bio, err);
81 mempool_free(mp_bh, conf->pool);
82 }
83
84 static void multipath_end_request(struct bio *bio, int error)
85 {
86 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
87 struct multipath_bh * mp_bh = (struct multipath_bh *)(bio->bi_private);
88 multipath_conf_t *conf = mp_bh->mddev->private;
89 mdk_rdev_t *rdev = conf->multipaths[mp_bh->path].rdev;
90
91 if (uptodate)
92 multipath_end_bh_io(mp_bh, 0);
93 else if (!bio_rw_flagged(bio, BIO_RW_AHEAD)) {
94 /*
95 * oops, IO error:
96 */
97 char b[BDEVNAME_SIZE];
98 md_error (mp_bh->mddev, rdev);
99 printk(KERN_ERR "multipath: %s: rescheduling sector %llu\n",
100 bdevname(rdev->bdev,b),
101 (unsigned long long)bio->bi_sector);
102 multipath_reschedule_retry(mp_bh);
103 } else
104 multipath_end_bh_io(mp_bh, error);
105 rdev_dec_pending(rdev, conf->mddev);
106 }
107
108 static void unplug_slaves(mddev_t *mddev)
109 {
110 multipath_conf_t *conf = mddev->private;
111 int i;
112
113 rcu_read_lock();
114 for (i=0; i<mddev->raid_disks; i++) {
115 mdk_rdev_t *rdev = rcu_dereference(conf->multipaths[i].rdev);
116 if (rdev && !test_bit(Faulty, &rdev->flags)
117 && atomic_read(&rdev->nr_pending)) {
118 struct request_queue *r_queue = bdev_get_queue(rdev->bdev);
119
120 atomic_inc(&rdev->nr_pending);
121 rcu_read_unlock();
122
123 blk_unplug(r_queue);
124
125 rdev_dec_pending(rdev, mddev);
126 rcu_read_lock();
127 }
128 }
129 rcu_read_unlock();
130 }
131
132 static void multipath_unplug(struct request_queue *q)
133 {
134 unplug_slaves(q->queuedata);
135 }
136
137
138 static int multipath_make_request (struct request_queue *q, struct bio * bio)
139 {
140 mddev_t *mddev = q->queuedata;
141 multipath_conf_t *conf = mddev->private;
142 struct multipath_bh * mp_bh;
143 struct multipath_info *multipath;
144 const int rw = bio_data_dir(bio);
145 int cpu;
146
147 if (unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER))) {
148 bio_endio(bio, -EOPNOTSUPP);
149 return 0;
150 }
151
152 mp_bh = mempool_alloc(conf->pool, GFP_NOIO);
153 memset(mp_bh, 0, sizeof(*mp_bh));
154
155 mp_bh->master_bio = bio;
156 mp_bh->mddev = mddev;
157
158 cpu = part_stat_lock();
159 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
160 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
161 bio_sectors(bio));
162 part_stat_unlock();
163
164 mp_bh->path = multipath_map(conf);
165 if (mp_bh->path < 0) {
166 bio_endio(bio, -EIO);
167 mempool_free(mp_bh, conf->pool);
168 return 0;
169 }
170 multipath = conf->multipaths + mp_bh->path;
171
172 mp_bh->bio = *bio;
173 mp_bh->bio.bi_sector += multipath->rdev->data_offset;
174 mp_bh->bio.bi_bdev = multipath->rdev->bdev;
175 mp_bh->bio.bi_rw |= (1 << BIO_RW_FAILFAST_TRANSPORT);
176 mp_bh->bio.bi_end_io = multipath_end_request;
177 mp_bh->bio.bi_private = mp_bh;
178 generic_make_request(&mp_bh->bio);
179 return 0;
180 }
181
182 static void multipath_status (struct seq_file *seq, mddev_t *mddev)
183 {
184 multipath_conf_t *conf = mddev->private;
185 int i;
186
187 seq_printf (seq, " [%d/%d] [", conf->raid_disks,
188 conf->working_disks);
189 for (i = 0; i < conf->raid_disks; i++)
190 seq_printf (seq, "%s",
191 conf->multipaths[i].rdev &&
192 test_bit(In_sync, &conf->multipaths[i].rdev->flags) ? "U" : "_");
193 seq_printf (seq, "]");
194 }
195
196 static int multipath_congested(void *data, int bits)
197 {
198 mddev_t *mddev = data;
199 multipath_conf_t *conf = mddev->private;
200 int i, ret = 0;
201
202 rcu_read_lock();
203 for (i = 0; i < mddev->raid_disks ; i++) {
204 mdk_rdev_t *rdev = rcu_dereference(conf->multipaths[i].rdev);
205 if (rdev && !test_bit(Faulty, &rdev->flags)) {
206 struct request_queue *q = bdev_get_queue(rdev->bdev);
207
208 ret |= bdi_congested(&q->backing_dev_info, bits);
209 /* Just like multipath_map, we just check the
210 * first available device
211 */
212 break;
213 }
214 }
215 rcu_read_unlock();
216 return ret;
217 }
218
219 /*
220 * Careful, this can execute in IRQ contexts as well!
221 */
222 static void multipath_error (mddev_t *mddev, mdk_rdev_t *rdev)
223 {
224 multipath_conf_t *conf = mddev->private;
225
226 if (conf->working_disks <= 1) {
227 /*
228 * Uh oh, we can do nothing if this is our last path, but
229 * first check if this is a queued request for a device
230 * which has just failed.
231 */
232 printk(KERN_ALERT
233 "multipath: only one IO path left and IO error.\n");
234 /* leave it active... it's all we have */
235 } else {
236 /*
237 * Mark disk as unusable
238 */
239 if (!test_bit(Faulty, &rdev->flags)) {
240 char b[BDEVNAME_SIZE];
241 clear_bit(In_sync, &rdev->flags);
242 set_bit(Faulty, &rdev->flags);
243 set_bit(MD_CHANGE_DEVS, &mddev->flags);
244 conf->working_disks--;
245 mddev->degraded++;
246 printk(KERN_ALERT "multipath: IO failure on %s,"
247 " disabling IO path.\n"
248 "multipath: Operation continuing"
249 " on %d IO paths.\n",
250 bdevname (rdev->bdev,b),
251 conf->working_disks);
252 }
253 }
254 }
255
256 static void print_multipath_conf (multipath_conf_t *conf)
257 {
258 int i;
259 struct multipath_info *tmp;
260
261 printk("MULTIPATH conf printout:\n");
262 if (!conf) {
263 printk("(conf==NULL)\n");
264 return;
265 }
266 printk(" --- wd:%d rd:%d\n", conf->working_disks,
267 conf->raid_disks);
268
269 for (i = 0; i < conf->raid_disks; i++) {
270 char b[BDEVNAME_SIZE];
271 tmp = conf->multipaths + i;
272 if (tmp->rdev)
273 printk(" disk%d, o:%d, dev:%s\n",
274 i,!test_bit(Faulty, &tmp->rdev->flags),
275 bdevname(tmp->rdev->bdev,b));
276 }
277 }
278
279
280 static int multipath_add_disk(mddev_t *mddev, mdk_rdev_t *rdev)
281 {
282 multipath_conf_t *conf = mddev->private;
283 struct request_queue *q;
284 int err = -EEXIST;
285 int path;
286 struct multipath_info *p;
287 int first = 0;
288 int last = mddev->raid_disks - 1;
289
290 if (rdev->raid_disk >= 0)
291 first = last = rdev->raid_disk;
292
293 print_multipath_conf(conf);
294
295 for (path = first; path <= last; path++)
296 if ((p=conf->multipaths+path)->rdev == NULL) {
297 q = rdev->bdev->bd_disk->queue;
298 disk_stack_limits(mddev->gendisk, rdev->bdev,
299 rdev->data_offset << 9);
300
301 /* as we don't honour merge_bvec_fn, we must never risk
302 * violating it, so limit ->max_sector to one PAGE, as
303 * a one page request is never in violation.
304 * (Note: it is very unlikely that a device with
305 * merge_bvec_fn will be involved in multipath.)
306 */
307 if (q->merge_bvec_fn &&
308 queue_max_sectors(q) > (PAGE_SIZE>>9))
309 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
310
311 conf->working_disks++;
312 mddev->degraded--;
313 rdev->raid_disk = path;
314 set_bit(In_sync, &rdev->flags);
315 rcu_assign_pointer(p->rdev, rdev);
316 err = 0;
317 md_integrity_add_rdev(rdev, mddev);
318 break;
319 }
320
321 print_multipath_conf(conf);
322
323 return err;
324 }
325
326 static int multipath_remove_disk(mddev_t *mddev, int number)
327 {
328 multipath_conf_t *conf = mddev->private;
329 int err = 0;
330 mdk_rdev_t *rdev;
331 struct multipath_info *p = conf->multipaths + number;
332
333 print_multipath_conf(conf);
334
335 rdev = p->rdev;
336 if (rdev) {
337 if (test_bit(In_sync, &rdev->flags) ||
338 atomic_read(&rdev->nr_pending)) {
339 printk(KERN_ERR "hot-remove-disk, slot %d is identified"
340 " but is still operational!\n", number);
341 err = -EBUSY;
342 goto abort;
343 }
344 p->rdev = NULL;
345 synchronize_rcu();
346 if (atomic_read(&rdev->nr_pending)) {
347 /* lost the race, try later */
348 err = -EBUSY;
349 p->rdev = rdev;
350 goto abort;
351 }
352 md_integrity_register(mddev);
353 }
354 abort:
355
356 print_multipath_conf(conf);
357 return err;
358 }
359
360
361
362 /*
363 * This is a kernel thread which:
364 *
365 * 1. Retries failed read operations on working multipaths.
366 * 2. Updates the raid superblock when problems encounter.
367 * 3. Performs writes following reads for array syncronising.
368 */
369
370 static void multipathd (mddev_t *mddev)
371 {
372 struct multipath_bh *mp_bh;
373 struct bio *bio;
374 unsigned long flags;
375 multipath_conf_t *conf = mddev->private;
376 struct list_head *head = &conf->retry_list;
377
378 md_check_recovery(mddev);
379 for (;;) {
380 char b[BDEVNAME_SIZE];
381 spin_lock_irqsave(&conf->device_lock, flags);
382 if (list_empty(head))
383 break;
384 mp_bh = list_entry(head->prev, struct multipath_bh, retry_list);
385 list_del(head->prev);
386 spin_unlock_irqrestore(&conf->device_lock, flags);
387
388 bio = &mp_bh->bio;
389 bio->bi_sector = mp_bh->master_bio->bi_sector;
390
391 if ((mp_bh->path = multipath_map (conf))<0) {
392 printk(KERN_ALERT "multipath: %s: unrecoverable IO read"
393 " error for block %llu\n",
394 bdevname(bio->bi_bdev,b),
395 (unsigned long long)bio->bi_sector);
396 multipath_end_bh_io(mp_bh, -EIO);
397 } else {
398 printk(KERN_ERR "multipath: %s: redirecting sector %llu"
399 " to another IO path\n",
400 bdevname(bio->bi_bdev,b),
401 (unsigned long long)bio->bi_sector);
402 *bio = *(mp_bh->master_bio);
403 bio->bi_sector += conf->multipaths[mp_bh->path].rdev->data_offset;
404 bio->bi_bdev = conf->multipaths[mp_bh->path].rdev->bdev;
405 bio->bi_rw |= (1 << BIO_RW_FAILFAST_TRANSPORT);
406 bio->bi_end_io = multipath_end_request;
407 bio->bi_private = mp_bh;
408 generic_make_request(bio);
409 }
410 }
411 spin_unlock_irqrestore(&conf->device_lock, flags);
412 }
413
414 static sector_t multipath_size(mddev_t *mddev, sector_t sectors, int raid_disks)
415 {
416 WARN_ONCE(sectors || raid_disks,
417 "%s does not support generic reshape\n", __func__);
418
419 return mddev->dev_sectors;
420 }
421
422 static int multipath_run (mddev_t *mddev)
423 {
424 multipath_conf_t *conf;
425 int disk_idx;
426 struct multipath_info *disk;
427 mdk_rdev_t *rdev;
428
429 if (md_check_no_bitmap(mddev))
430 return -EINVAL;
431
432 if (mddev->level != LEVEL_MULTIPATH) {
433 printk("multipath: %s: raid level not set to multipath IO (%d)\n",
434 mdname(mddev), mddev->level);
435 goto out;
436 }
437 /*
438 * copy the already verified devices into our private MULTIPATH
439 * bookkeeping area. [whatever we allocate in multipath_run(),
440 * should be freed in multipath_stop()]
441 */
442 mddev->queue->queue_lock = &mddev->queue->__queue_lock;
443
444 conf = kzalloc(sizeof(multipath_conf_t), GFP_KERNEL);
445 mddev->private = conf;
446 if (!conf) {
447 printk(KERN_ERR
448 "multipath: couldn't allocate memory for %s\n",
449 mdname(mddev));
450 goto out;
451 }
452
453 conf->multipaths = kzalloc(sizeof(struct multipath_info)*mddev->raid_disks,
454 GFP_KERNEL);
455 if (!conf->multipaths) {
456 printk(KERN_ERR
457 "multipath: couldn't allocate memory for %s\n",
458 mdname(mddev));
459 goto out_free_conf;
460 }
461
462 conf->working_disks = 0;
463 list_for_each_entry(rdev, &mddev->disks, same_set) {
464 disk_idx = rdev->raid_disk;
465 if (disk_idx < 0 ||
466 disk_idx >= mddev->raid_disks)
467 continue;
468
469 disk = conf->multipaths + disk_idx;
470 disk->rdev = rdev;
471 disk_stack_limits(mddev->gendisk, rdev->bdev,
472 rdev->data_offset << 9);
473
474 /* as we don't honour merge_bvec_fn, we must never risk
475 * violating it, not that we ever expect a device with
476 * a merge_bvec_fn to be involved in multipath */
477 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
478 queue_max_sectors(mddev->queue) > (PAGE_SIZE>>9))
479 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
480
481 if (!test_bit(Faulty, &rdev->flags))
482 conf->working_disks++;
483 }
484
485 conf->raid_disks = mddev->raid_disks;
486 conf->mddev = mddev;
487 spin_lock_init(&conf->device_lock);
488 INIT_LIST_HEAD(&conf->retry_list);
489
490 if (!conf->working_disks) {
491 printk(KERN_ERR "multipath: no operational IO paths for %s\n",
492 mdname(mddev));
493 goto out_free_conf;
494 }
495 mddev->degraded = conf->raid_disks - conf->working_disks;
496
497 conf->pool = mempool_create_kmalloc_pool(NR_RESERVED_BUFS,
498 sizeof(struct multipath_bh));
499 if (conf->pool == NULL) {
500 printk(KERN_ERR
501 "multipath: couldn't allocate memory for %s\n",
502 mdname(mddev));
503 goto out_free_conf;
504 }
505
506 {
507 mddev->thread = md_register_thread(multipathd, mddev, "%s_multipath");
508 if (!mddev->thread) {
509 printk(KERN_ERR "multipath: couldn't allocate thread"
510 " for %s\n", mdname(mddev));
511 goto out_free_conf;
512 }
513 }
514
515 printk(KERN_INFO
516 "multipath: array %s active with %d out of %d IO paths\n",
517 mdname(mddev), conf->working_disks, mddev->raid_disks);
518 /*
519 * Ok, everything is just fine now
520 */
521 md_set_array_sectors(mddev, multipath_size(mddev, 0, 0));
522
523 mddev->queue->unplug_fn = multipath_unplug;
524 mddev->queue->backing_dev_info.congested_fn = multipath_congested;
525 mddev->queue->backing_dev_info.congested_data = mddev;
526 md_integrity_register(mddev);
527 return 0;
528
529 out_free_conf:
530 if (conf->pool)
531 mempool_destroy(conf->pool);
532 kfree(conf->multipaths);
533 kfree(conf);
534 mddev->private = NULL;
535 out:
536 return -EIO;
537 }
538
539
540 static int multipath_stop (mddev_t *mddev)
541 {
542 multipath_conf_t *conf = mddev->private;
543
544 md_unregister_thread(mddev->thread);
545 mddev->thread = NULL;
546 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
547 mempool_destroy(conf->pool);
548 kfree(conf->multipaths);
549 kfree(conf);
550 mddev->private = NULL;
551 return 0;
552 }
553
554 static struct mdk_personality multipath_personality =
555 {
556 .name = "multipath",
557 .level = LEVEL_MULTIPATH,
558 .owner = THIS_MODULE,
559 .make_request = multipath_make_request,
560 .run = multipath_run,
561 .stop = multipath_stop,
562 .status = multipath_status,
563 .error_handler = multipath_error,
564 .hot_add_disk = multipath_add_disk,
565 .hot_remove_disk= multipath_remove_disk,
566 .size = multipath_size,
567 };
568
569 static int __init multipath_init (void)
570 {
571 return register_md_personality (&multipath_personality);
572 }
573
574 static void __exit multipath_exit (void)
575 {
576 unregister_md_personality (&multipath_personality);
577 }
578
579 module_init(multipath_init);
580 module_exit(multipath_exit);
581 MODULE_LICENSE("GPL");
582 MODULE_ALIAS("md-personality-7"); /* MULTIPATH */
583 MODULE_ALIAS("md-multipath");
584 MODULE_ALIAS("md-level--4");
Linux with added FPC-III support