?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'merge' of git://git.kernel.org/pub/scm/linux/kernel/git/paulus/powerpc
[wrt350n-kernel.git] / drivers / md / multipath.c
blob3f299d835a2b1956e210c0a383e65f2d9c70874f
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/module.h>
23 #include <linux/slab.h>
24 #include <linux/spinlock.h>
25 #include <linux/raid/multipath.h>
26 #include <linux/buffer_head.h>
27 #include <asm/atomic.h>
28
29 #define MAJOR_NR MD_MAJOR
30 #define MD_DRIVER
31 #define MD_PERSONALITY
32
33 #define MAX_WORK_PER_DISK 128
34
35 #define NR_RESERVED_BUFS 32
36
37
38 static int multipath_map (multipath_conf_t *conf)
39 {
40 int i, disks = conf->raid_disks;
41
42 /*
43 * Later we do read balancing on the read side
44 * now we use the first available disk.
45 */
46
47 rcu_read_lock();
48 for (i = 0; i < disks; i++) {
49 mdk_rdev_t *rdev = rcu_dereference(conf->multipaths[i].rdev);
50 if (rdev && test_bit(In_sync, &rdev->flags)) {
51 atomic_inc(&rdev->nr_pending);
52 rcu_read_unlock();
53 return i;
54 }
55 }
56 rcu_read_unlock();
57
58 printk(KERN_ERR "multipath_map(): no more operational IO paths?\n");
59 return (-1);
60 }
61
62 static void multipath_reschedule_retry (struct multipath_bh *mp_bh)
63 {
64 unsigned long flags;
65 mddev_t *mddev = mp_bh->mddev;
66 multipath_conf_t *conf = mddev_to_conf(mddev);
67
68 spin_lock_irqsave(&conf->device_lock, flags);
69 list_add(&mp_bh->retry_list, &conf->retry_list);
70 spin_unlock_irqrestore(&conf->device_lock, flags);
71 md_wakeup_thread(mddev->thread);
72 }
73
74
75 /*
76 * multipath_end_bh_io() is called when we have finished servicing a multipathed
77 * operation and are ready to return a success/failure code to the buffer
78 * cache layer.
79 */
80 static void multipath_end_bh_io (struct multipath_bh *mp_bh, int err)
81 {
82 struct bio *bio = mp_bh->master_bio;
83 multipath_conf_t *conf = mddev_to_conf(mp_bh->mddev);
84
85 bio_endio(bio, err);
86 mempool_free(mp_bh, conf->pool);
87 }
88
89 static void multipath_end_request(struct bio *bio, int error)
90 {
91 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
92 struct multipath_bh * mp_bh = (struct multipath_bh *)(bio->bi_private);
93 multipath_conf_t *conf = mddev_to_conf(mp_bh->mddev);
94 mdk_rdev_t *rdev = conf->multipaths[mp_bh->path].rdev;
95
96 if (uptodate)
97 multipath_end_bh_io(mp_bh, 0);
98 else if (!bio_rw_ahead(bio)) {
99 /*
100 * oops, IO error:
101 */
102 char b[BDEVNAME_SIZE];
103 md_error (mp_bh->mddev, rdev);
104 printk(KERN_ERR "multipath: %s: rescheduling sector %llu\n",
105 bdevname(rdev->bdev,b),
106 (unsigned long long)bio->bi_sector);
107 multipath_reschedule_retry(mp_bh);
108 } else
109 multipath_end_bh_io(mp_bh, error);
110 rdev_dec_pending(rdev, conf->mddev);
111 }
112
113 static void unplug_slaves(mddev_t *mddev)
114 {
115 multipath_conf_t *conf = mddev_to_conf(mddev);
116 int i;
117
118 rcu_read_lock();
119 for (i=0; i<mddev->raid_disks; i++) {
120 mdk_rdev_t *rdev = rcu_dereference(conf->multipaths[i].rdev);
121 if (rdev && !test_bit(Faulty, &rdev->flags)
122 && atomic_read(&rdev->nr_pending)) {
123 struct request_queue *r_queue = bdev_get_queue(rdev->bdev);
124
125 atomic_inc(&rdev->nr_pending);
126 rcu_read_unlock();
127
128 blk_unplug(r_queue);
129
130 rdev_dec_pending(rdev, mddev);
131 rcu_read_lock();
132 }
133 }
134 rcu_read_unlock();
135 }
136
137 static void multipath_unplug(struct request_queue *q)
138 {
139 unplug_slaves(q->queuedata);
140 }
141
142
143 static int multipath_make_request (struct request_queue *q, struct bio * bio)
144 {
145 mddev_t *mddev = q->queuedata;
146 multipath_conf_t *conf = mddev_to_conf(mddev);
147 struct multipath_bh * mp_bh;
148 struct multipath_info *multipath;
149 const int rw = bio_data_dir(bio);
150
151 if (unlikely(bio_barrier(bio))) {
152 bio_endio(bio, -EOPNOTSUPP);
153 return 0;
154 }
155
156 mp_bh = mempool_alloc(conf->pool, GFP_NOIO);
157
158 mp_bh->master_bio = bio;
159 mp_bh->mddev = mddev;
160
161 disk_stat_inc(mddev->gendisk, ios[rw]);
162 disk_stat_add(mddev->gendisk, sectors[rw], bio_sectors(bio));
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);
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_to_conf(mddev);
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_to_conf(mddev);
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_to_conf(mddev);
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 Operation continuing"
248 " on %d IO paths.\n",
249 bdevname (rdev->bdev,b),
250 conf->working_disks);
251 }
252 }
253 }
254
255 static void print_multipath_conf (multipath_conf_t *conf)
256 {
257 int i;
258 struct multipath_info *tmp;
259
260 printk("MULTIPATH conf printout:\n");
261 if (!conf) {
262 printk("(conf==NULL)\n");
263 return;
264 }
265 printk(" --- wd:%d rd:%d\n", conf->working_disks,
266 conf->raid_disks);
267
268 for (i = 0; i < conf->raid_disks; i++) {
269 char b[BDEVNAME_SIZE];
270 tmp = conf->multipaths + i;
271 if (tmp->rdev)
272 printk(" disk%d, o:%d, dev:%s\n",
273 i,!test_bit(Faulty, &tmp->rdev->flags),
274 bdevname(tmp->rdev->bdev,b));
275 }
276 }
277
278
279 static int multipath_add_disk(mddev_t *mddev, mdk_rdev_t *rdev)
280 {
281 multipath_conf_t *conf = mddev->private;
282 struct request_queue *q;
283 int found = 0;
284 int path;
285 struct multipath_info *p;
286
287 print_multipath_conf(conf);
288
289 for (path=0; path<mddev->raid_disks; path++)
290 if ((p=conf->multipaths+path)->rdev == NULL) {
291 q = rdev->bdev->bd_disk->queue;
292 blk_queue_stack_limits(mddev->queue, q);
293
294 /* as we don't honour merge_bvec_fn, we must never risk
295 * violating it, so limit ->max_sector to one PAGE, as
296 * a one page request is never in violation.
297 * (Note: it is very unlikely that a device with
298 * merge_bvec_fn will be involved in multipath.)
299 */
300 if (q->merge_bvec_fn &&
301 mddev->queue->max_sectors > (PAGE_SIZE>>9))
302 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
303
304 conf->working_disks++;
305 mddev->degraded--;
306 rdev->raid_disk = path;
307 set_bit(In_sync, &rdev->flags);
308 rcu_assign_pointer(p->rdev, rdev);
309 found = 1;
310 }
311
312 print_multipath_conf(conf);
313 return found;
314 }
315
316 static int multipath_remove_disk(mddev_t *mddev, int number)
317 {
318 multipath_conf_t *conf = mddev->private;
319 int err = 0;
320 mdk_rdev_t *rdev;
321 struct multipath_info *p = conf->multipaths + number;
322
323 print_multipath_conf(conf);
324
325 rdev = p->rdev;
326 if (rdev) {
327 if (test_bit(In_sync, &rdev->flags) ||
328 atomic_read(&rdev->nr_pending)) {
329 printk(KERN_ERR "hot-remove-disk, slot %d is identified" " but is still operational!\n", number);
330 err = -EBUSY;
331 goto abort;
332 }
333 p->rdev = NULL;
334 synchronize_rcu();
335 if (atomic_read(&rdev->nr_pending)) {
336 /* lost the race, try later */
337 err = -EBUSY;
338 p->rdev = rdev;
339 }
340 }
341 abort:
342
343 print_multipath_conf(conf);
344 return err;
345 }
346
347
348
349 /*
350 * This is a kernel thread which:
351 *
352 * 1. Retries failed read operations on working multipaths.
353 * 2. Updates the raid superblock when problems encounter.
354 * 3. Performs writes following reads for array syncronising.
355 */
356
357 static void multipathd (mddev_t *mddev)
358 {
359 struct multipath_bh *mp_bh;
360 struct bio *bio;
361 unsigned long flags;
362 multipath_conf_t *conf = mddev_to_conf(mddev);
363 struct list_head *head = &conf->retry_list;
364
365 md_check_recovery(mddev);
366 for (;;) {
367 char b[BDEVNAME_SIZE];
368 spin_lock_irqsave(&conf->device_lock, flags);
369 if (list_empty(head))
370 break;
371 mp_bh = list_entry(head->prev, struct multipath_bh, retry_list);
372 list_del(head->prev);
373 spin_unlock_irqrestore(&conf->device_lock, flags);
374
375 bio = &mp_bh->bio;
376 bio->bi_sector = mp_bh->master_bio->bi_sector;
377
378 if ((mp_bh->path = multipath_map (conf))<0) {
379 printk(KERN_ALERT "multipath: %s: unrecoverable IO read"
380 " error for block %llu\n",
381 bdevname(bio->bi_bdev,b),
382 (unsigned long long)bio->bi_sector);
383 multipath_end_bh_io(mp_bh, -EIO);
384 } else {
385 printk(KERN_ERR "multipath: %s: redirecting sector %llu"
386 " to another IO path\n",
387 bdevname(bio->bi_bdev,b),
388 (unsigned long long)bio->bi_sector);
389 *bio = *(mp_bh->master_bio);
390 bio->bi_sector += conf->multipaths[mp_bh->path].rdev->data_offset;
391 bio->bi_bdev = conf->multipaths[mp_bh->path].rdev->bdev;
392 bio->bi_rw |= (1 << BIO_RW_FAILFAST);
393 bio->bi_end_io = multipath_end_request;
394 bio->bi_private = mp_bh;
395 generic_make_request(bio);
396 }
397 }
398 spin_unlock_irqrestore(&conf->device_lock, flags);
399 }
400
401 static int multipath_run (mddev_t *mddev)
402 {
403 multipath_conf_t *conf;
404 int disk_idx;
405 struct multipath_info *disk;
406 mdk_rdev_t *rdev;
407 struct list_head *tmp;
408
409 if (mddev->level != LEVEL_MULTIPATH) {
410 printk("multipath: %s: raid level not set to multipath IO (%d)\n",
411 mdname(mddev), mddev->level);
412 goto out;
413 }
414 /*
415 * copy the already verified devices into our private MULTIPATH
416 * bookkeeping area. [whatever we allocate in multipath_run(),
417 * should be freed in multipath_stop()]
418 */
419
420 conf = kzalloc(sizeof(multipath_conf_t), GFP_KERNEL);
421 mddev->private = conf;
422 if (!conf) {
423 printk(KERN_ERR
424 "multipath: couldn't allocate memory for %s\n",
425 mdname(mddev));
426 goto out;
427 }
428
429 conf->multipaths = kzalloc(sizeof(struct multipath_info)*mddev->raid_disks,
430 GFP_KERNEL);
431 if (!conf->multipaths) {
432 printk(KERN_ERR
433 "multipath: couldn't allocate memory for %s\n",
434 mdname(mddev));
435 goto out_free_conf;
436 }
437
438 conf->working_disks = 0;
439 rdev_for_each(rdev, tmp, mddev) {
440 disk_idx = rdev->raid_disk;
441 if (disk_idx < 0 ||
442 disk_idx >= mddev->raid_disks)
443 continue;
444
445 disk = conf->multipaths + disk_idx;
446 disk->rdev = rdev;
447
448 blk_queue_stack_limits(mddev->queue,
449 rdev->bdev->bd_disk->queue);
450 /* as we don't honour merge_bvec_fn, we must never risk
451 * violating it, not that we ever expect a device with
452 * a merge_bvec_fn to be involved in multipath */
453 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
454 mddev->queue->max_sectors > (PAGE_SIZE>>9))
455 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
456
457 if (!test_bit(Faulty, &rdev->flags))
458 conf->working_disks++;
459 }
460
461 conf->raid_disks = mddev->raid_disks;
462 conf->mddev = mddev;
463 spin_lock_init(&conf->device_lock);
464 INIT_LIST_HEAD(&conf->retry_list);
465
466 if (!conf->working_disks) {
467 printk(KERN_ERR "multipath: no operational IO paths for %s\n",
468 mdname(mddev));
469 goto out_free_conf;
470 }
471 mddev->degraded = conf->raid_disks - conf->working_disks;
472
473 conf->pool = mempool_create_kzalloc_pool(NR_RESERVED_BUFS,
474 sizeof(struct multipath_bh));
475 if (conf->pool == NULL) {
476 printk(KERN_ERR
477 "multipath: couldn't allocate memory for %s\n",
478 mdname(mddev));
479 goto out_free_conf;
480 }
481
482 {
483 mddev->thread = md_register_thread(multipathd, mddev, "%s_multipath");
484 if (!mddev->thread) {
485 printk(KERN_ERR "multipath: couldn't allocate thread"
486 " for %s\n", mdname(mddev));
487 goto out_free_conf;
488 }
489 }
490
491 printk(KERN_INFO
492 "multipath: array %s active with %d out of %d IO paths\n",
493 mdname(mddev), conf->working_disks, mddev->raid_disks);
494 /*
495 * Ok, everything is just fine now
496 */
497 mddev->array_size = mddev->size;
498
499 mddev->queue->unplug_fn = multipath_unplug;
500 mddev->queue->backing_dev_info.congested_fn = multipath_congested;
501 mddev->queue->backing_dev_info.congested_data = mddev;
502
503 return 0;
504
505 out_free_conf:
506 if (conf->pool)
507 mempool_destroy(conf->pool);
508 kfree(conf->multipaths);
509 kfree(conf);
510 mddev->private = NULL;
511 out:
512 return -EIO;
513 }
514
515
516 static int multipath_stop (mddev_t *mddev)
517 {
518 multipath_conf_t *conf = mddev_to_conf(mddev);
519
520 md_unregister_thread(mddev->thread);
521 mddev->thread = NULL;
522 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
523 mempool_destroy(conf->pool);
524 kfree(conf->multipaths);
525 kfree(conf);
526 mddev->private = NULL;
527 return 0;
528 }
529
530 static struct mdk_personality multipath_personality =
531 {
532 .name = "multipath",
533 .level = LEVEL_MULTIPATH,
534 .owner = THIS_MODULE,
535 .make_request = multipath_make_request,
536 .run = multipath_run,
537 .stop = multipath_stop,
538 .status = multipath_status,
539 .error_handler = multipath_error,
540 .hot_add_disk = multipath_add_disk,
541 .hot_remove_disk= multipath_remove_disk,
542 };
543
544 static int __init multipath_init (void)
545 {
546 return register_md_personality (&multipath_personality);
547 }
548
549 static void __exit multipath_exit (void)
550 {
551 unregister_md_personality (&multipath_personality);
552 }
553
554 module_init(multipath_init);
555 module_exit(multipath_exit);
556 MODULE_LICENSE("GPL");
557 MODULE_ALIAS("md-personality-7"); /* MULTIPATH */
558 MODULE_ALIAS("md-multipath");
559 MODULE_ALIAS("md-level--4");
WRT350N Kernel Patches