projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Gather on-going resync information of other nodes
[deliverable/linux.git] / drivers / md / md.c
1 /*
2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
4
5 completely rewritten, based on the MD driver code from Marc Zyngier
6
7 Changes:
8
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
19
20 Neil Brown <neilb@cse.unsw.edu.au>.
21
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
28 any later version.
29
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33 */
34
35 #include <linux/kthread.h>
36 #include <linux/blkdev.h>
37 #include <linux/sysctl.h>
38 #include <linux/seq_file.h>
39 #include <linux/fs.h>
40 #include <linux/poll.h>
41 #include <linux/ctype.h>
42 #include <linux/string.h>
43 #include <linux/hdreg.h>
44 #include <linux/proc_fs.h>
45 #include <linux/random.h>
46 #include <linux/module.h>
47 #include <linux/reboot.h>
48 #include <linux/file.h>
49 #include <linux/compat.h>
50 #include <linux/delay.h>
51 #include <linux/raid/md_p.h>
52 #include <linux/raid/md_u.h>
53 #include <linux/slab.h>
54 #include "md.h"
55 #include "bitmap.h"
56 #include "md-cluster.h"
57
58 #ifndef MODULE
59 static void autostart_arrays(int part);
60 #endif
61
62 /* pers_list is a list of registered personalities protected
63 * by pers_lock.
64 * pers_lock does extra service to protect accesses to
65 * mddev->thread when the mutex cannot be held.
66 */
67 static LIST_HEAD(pers_list);
68 static DEFINE_SPINLOCK(pers_lock);
69
70 struct md_cluster_operations *md_cluster_ops;
71 struct module *md_cluster_mod;
72 EXPORT_SYMBOL(md_cluster_mod);
73
74 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
75 static struct workqueue_struct *md_wq;
76 static struct workqueue_struct *md_misc_wq;
77
78 static int remove_and_add_spares(struct mddev *mddev,
79 struct md_rdev *this);
80 static void mddev_detach(struct mddev *mddev);
81
82 /*
83 * Default number of read corrections we'll attempt on an rdev
84 * before ejecting it from the array. We divide the read error
85 * count by 2 for every hour elapsed between read errors.
86 */
87 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
88 /*
89 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
90 * is 1000 KB/sec, so the extra system load does not show up that much.
91 * Increase it if you want to have more _guaranteed_ speed. Note that
92 * the RAID driver will use the maximum available bandwidth if the IO
93 * subsystem is idle. There is also an 'absolute maximum' reconstruction
94 * speed limit - in case reconstruction slows down your system despite
95 * idle IO detection.
96 *
97 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
98 * or /sys/block/mdX/md/sync_speed_{min,max}
99 */
100
101 static int sysctl_speed_limit_min = 1000;
102 static int sysctl_speed_limit_max = 200000;
103 static inline int speed_min(struct mddev *mddev)
104 {
105 return mddev->sync_speed_min ?
106 mddev->sync_speed_min : sysctl_speed_limit_min;
107 }
108
109 static inline int speed_max(struct mddev *mddev)
110 {
111 return mddev->sync_speed_max ?
112 mddev->sync_speed_max : sysctl_speed_limit_max;
113 }
114
115 static struct ctl_table_header *raid_table_header;
116
117 static struct ctl_table raid_table[] = {
118 {
119 .procname = "speed_limit_min",
120 .data = &sysctl_speed_limit_min,
121 .maxlen = sizeof(int),
122 .mode = S_IRUGO|S_IWUSR,
123 .proc_handler = proc_dointvec,
124 },
125 {
126 .procname = "speed_limit_max",
127 .data = &sysctl_speed_limit_max,
128 .maxlen = sizeof(int),
129 .mode = S_IRUGO|S_IWUSR,
130 .proc_handler = proc_dointvec,
131 },
132 { }
133 };
134
135 static struct ctl_table raid_dir_table[] = {
136 {
137 .procname = "raid",
138 .maxlen = 0,
139 .mode = S_IRUGO|S_IXUGO,
140 .child = raid_table,
141 },
142 { }
143 };
144
145 static struct ctl_table raid_root_table[] = {
146 {
147 .procname = "dev",
148 .maxlen = 0,
149 .mode = 0555,
150 .child = raid_dir_table,
151 },
152 { }
153 };
154
155 static const struct block_device_operations md_fops;
156
157 static int start_readonly;
158
159 /* bio_clone_mddev
160 * like bio_clone, but with a local bio set
161 */
162
163 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
164 struct mddev *mddev)
165 {
166 struct bio *b;
167
168 if (!mddev || !mddev->bio_set)
169 return bio_alloc(gfp_mask, nr_iovecs);
170
171 b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set);
172 if (!b)
173 return NULL;
174 return b;
175 }
176 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
177
178 struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
179 struct mddev *mddev)
180 {
181 if (!mddev || !mddev->bio_set)
182 return bio_clone(bio, gfp_mask);
183
184 return bio_clone_bioset(bio, gfp_mask, mddev->bio_set);
185 }
186 EXPORT_SYMBOL_GPL(bio_clone_mddev);
187
188 /*
189 * We have a system wide 'event count' that is incremented
190 * on any 'interesting' event, and readers of /proc/mdstat
191 * can use 'poll' or 'select' to find out when the event
192 * count increases.
193 *
194 * Events are:
195 * start array, stop array, error, add device, remove device,
196 * start build, activate spare
197 */
198 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
199 static atomic_t md_event_count;
200 void md_new_event(struct mddev *mddev)
201 {
202 atomic_inc(&md_event_count);
203 wake_up(&md_event_waiters);
204 }
205 EXPORT_SYMBOL_GPL(md_new_event);
206
207 /* Alternate version that can be called from interrupts
208 * when calling sysfs_notify isn't needed.
209 */
210 static void md_new_event_inintr(struct mddev *mddev)
211 {
212 atomic_inc(&md_event_count);
213 wake_up(&md_event_waiters);
214 }
215
216 /*
217 * Enables to iterate over all existing md arrays
218 * all_mddevs_lock protects this list.
219 */
220 static LIST_HEAD(all_mddevs);
221 static DEFINE_SPINLOCK(all_mddevs_lock);
222
223 /*
224 * iterates through all used mddevs in the system.
225 * We take care to grab the all_mddevs_lock whenever navigating
226 * the list, and to always hold a refcount when unlocked.
227 * Any code which breaks out of this loop while own
228 * a reference to the current mddev and must mddev_put it.
229 */
230 #define for_each_mddev(_mddev,_tmp) \
231 \
232 for (({ spin_lock(&all_mddevs_lock); \
233 _tmp = all_mddevs.next; \
234 _mddev = NULL;}); \
235 ({ if (_tmp != &all_mddevs) \
236 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
237 spin_unlock(&all_mddevs_lock); \
238 if (_mddev) mddev_put(_mddev); \
239 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
240 _tmp != &all_mddevs;}); \
241 ({ spin_lock(&all_mddevs_lock); \
242 _tmp = _tmp->next;}) \
243 )
244
245 /* Rather than calling directly into the personality make_request function,
246 * IO requests come here first so that we can check if the device is
247 * being suspended pending a reconfiguration.
248 * We hold a refcount over the call to ->make_request. By the time that
249 * call has finished, the bio has been linked into some internal structure
250 * and so is visible to ->quiesce(), so we don't need the refcount any more.
251 */
252 static void md_make_request(struct request_queue *q, struct bio *bio)
253 {
254 const int rw = bio_data_dir(bio);
255 struct mddev *mddev = q->queuedata;
256 unsigned int sectors;
257
258 if (mddev == NULL || mddev->pers == NULL
259 || !mddev->ready) {
260 bio_io_error(bio);
261 return;
262 }
263 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
264 bio_endio(bio, bio_sectors(bio) == 0 ? 0 : -EROFS);
265 return;
266 }
267 smp_rmb(); /* Ensure implications of 'active' are visible */
268 rcu_read_lock();
269 if (mddev->suspended) {
270 DEFINE_WAIT(__wait);
271 for (;;) {
272 prepare_to_wait(&mddev->sb_wait, &__wait,
273 TASK_UNINTERRUPTIBLE);
274 if (!mddev->suspended)
275 break;
276 rcu_read_unlock();
277 schedule();
278 rcu_read_lock();
279 }
280 finish_wait(&mddev->sb_wait, &__wait);
281 }
282 atomic_inc(&mddev->active_io);
283 rcu_read_unlock();
284
285 /*
286 * save the sectors now since our bio can
287 * go away inside make_request
288 */
289 sectors = bio_sectors(bio);
290 mddev->pers->make_request(mddev, bio);
291
292 generic_start_io_acct(rw, sectors, &mddev->gendisk->part0);
293
294 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
295 wake_up(&mddev->sb_wait);
296 }
297
298 /* mddev_suspend makes sure no new requests are submitted
299 * to the device, and that any requests that have been submitted
300 * are completely handled.
301 * Once mddev_detach() is called and completes, the module will be
302 * completely unused.
303 */
304 void mddev_suspend(struct mddev *mddev)
305 {
306 BUG_ON(mddev->suspended);
307 mddev->suspended = 1;
308 synchronize_rcu();
309 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
310 mddev->pers->quiesce(mddev, 1);
311
312 del_timer_sync(&mddev->safemode_timer);
313 }
314 EXPORT_SYMBOL_GPL(mddev_suspend);
315
316 void mddev_resume(struct mddev *mddev)
317 {
318 mddev->suspended = 0;
319 wake_up(&mddev->sb_wait);
320 mddev->pers->quiesce(mddev, 0);
321
322 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
323 md_wakeup_thread(mddev->thread);
324 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
325 }
326 EXPORT_SYMBOL_GPL(mddev_resume);
327
328 int mddev_congested(struct mddev *mddev, int bits)
329 {
330 struct md_personality *pers = mddev->pers;
331 int ret = 0;
332
333 rcu_read_lock();
334 if (mddev->suspended)
335 ret = 1;
336 else if (pers && pers->congested)
337 ret = pers->congested(mddev, bits);
338 rcu_read_unlock();
339 return ret;
340 }
341 EXPORT_SYMBOL_GPL(mddev_congested);
342 static int md_congested(void *data, int bits)
343 {
344 struct mddev *mddev = data;
345 return mddev_congested(mddev, bits);
346 }
347
348 static int md_mergeable_bvec(struct request_queue *q,
349 struct bvec_merge_data *bvm,
350 struct bio_vec *biovec)
351 {
352 struct mddev *mddev = q->queuedata;
353 int ret;
354 rcu_read_lock();
355 if (mddev->suspended) {
356 /* Must always allow one vec */
357 if (bvm->bi_size == 0)
358 ret = biovec->bv_len;
359 else
360 ret = 0;
361 } else {
362 struct md_personality *pers = mddev->pers;
363 if (pers && pers->mergeable_bvec)
364 ret = pers->mergeable_bvec(mddev, bvm, biovec);
365 else
366 ret = biovec->bv_len;
367 }
368 rcu_read_unlock();
369 return ret;
370 }
371 /*
372 * Generic flush handling for md
373 */
374
375 static void md_end_flush(struct bio *bio, int err)
376 {
377 struct md_rdev *rdev = bio->bi_private;
378 struct mddev *mddev = rdev->mddev;
379
380 rdev_dec_pending(rdev, mddev);
381
382 if (atomic_dec_and_test(&mddev->flush_pending)) {
383 /* The pre-request flush has finished */
384 queue_work(md_wq, &mddev->flush_work);
385 }
386 bio_put(bio);
387 }
388
389 static void md_submit_flush_data(struct work_struct *ws);
390
391 static void submit_flushes(struct work_struct *ws)
392 {
393 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
394 struct md_rdev *rdev;
395
396 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
397 atomic_set(&mddev->flush_pending, 1);
398 rcu_read_lock();
399 rdev_for_each_rcu(rdev, mddev)
400 if (rdev->raid_disk >= 0 &&
401 !test_bit(Faulty, &rdev->flags)) {
402 /* Take two references, one is dropped
403 * when request finishes, one after
404 * we reclaim rcu_read_lock
405 */
406 struct bio *bi;
407 atomic_inc(&rdev->nr_pending);
408 atomic_inc(&rdev->nr_pending);
409 rcu_read_unlock();
410 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
411 bi->bi_end_io = md_end_flush;
412 bi->bi_private = rdev;
413 bi->bi_bdev = rdev->bdev;
414 atomic_inc(&mddev->flush_pending);
415 submit_bio(WRITE_FLUSH, bi);
416 rcu_read_lock();
417 rdev_dec_pending(rdev, mddev);
418 }
419 rcu_read_unlock();
420 if (atomic_dec_and_test(&mddev->flush_pending))
421 queue_work(md_wq, &mddev->flush_work);
422 }
423
424 static void md_submit_flush_data(struct work_struct *ws)
425 {
426 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
427 struct bio *bio = mddev->flush_bio;
428
429 if (bio->bi_iter.bi_size == 0)
430 /* an empty barrier - all done */
431 bio_endio(bio, 0);
432 else {
433 bio->bi_rw &= ~REQ_FLUSH;
434 mddev->pers->make_request(mddev, bio);
435 }
436
437 mddev->flush_bio = NULL;
438 wake_up(&mddev->sb_wait);
439 }
440
441 void md_flush_request(struct mddev *mddev, struct bio *bio)
442 {
443 spin_lock_irq(&mddev->lock);
444 wait_event_lock_irq(mddev->sb_wait,
445 !mddev->flush_bio,
446 mddev->lock);
447 mddev->flush_bio = bio;
448 spin_unlock_irq(&mddev->lock);
449
450 INIT_WORK(&mddev->flush_work, submit_flushes);
451 queue_work(md_wq, &mddev->flush_work);
452 }
453 EXPORT_SYMBOL(md_flush_request);
454
455 void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
456 {
457 struct mddev *mddev = cb->data;
458 md_wakeup_thread(mddev->thread);
459 kfree(cb);
460 }
461 EXPORT_SYMBOL(md_unplug);
462
463 static inline struct mddev *mddev_get(struct mddev *mddev)
464 {
465 atomic_inc(&mddev->active);
466 return mddev;
467 }
468
469 static void mddev_delayed_delete(struct work_struct *ws);
470
471 static void mddev_put(struct mddev *mddev)
472 {
473 struct bio_set *bs = NULL;
474
475 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
476 return;
477 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
478 mddev->ctime == 0 && !mddev->hold_active) {
479 /* Array is not configured at all, and not held active,
480 * so destroy it */
481 list_del_init(&mddev->all_mddevs);
482 bs = mddev->bio_set;
483 mddev->bio_set = NULL;
484 if (mddev->gendisk) {
485 /* We did a probe so need to clean up. Call
486 * queue_work inside the spinlock so that
487 * flush_workqueue() after mddev_find will
488 * succeed in waiting for the work to be done.
489 */
490 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
491 queue_work(md_misc_wq, &mddev->del_work);
492 } else
493 kfree(mddev);
494 }
495 spin_unlock(&all_mddevs_lock);
496 if (bs)
497 bioset_free(bs);
498 }
499
500 void mddev_init(struct mddev *mddev)
501 {
502 mutex_init(&mddev->open_mutex);
503 mutex_init(&mddev->reconfig_mutex);
504 mutex_init(&mddev->bitmap_info.mutex);
505 INIT_LIST_HEAD(&mddev->disks);
506 INIT_LIST_HEAD(&mddev->all_mddevs);
507 init_timer(&mddev->safemode_timer);
508 atomic_set(&mddev->active, 1);
509 atomic_set(&mddev->openers, 0);
510 atomic_set(&mddev->active_io, 0);
511 spin_lock_init(&mddev->lock);
512 atomic_set(&mddev->flush_pending, 0);
513 init_waitqueue_head(&mddev->sb_wait);
514 init_waitqueue_head(&mddev->recovery_wait);
515 mddev->reshape_position = MaxSector;
516 mddev->reshape_backwards = 0;
517 mddev->last_sync_action = "none";
518 mddev->resync_min = 0;
519 mddev->resync_max = MaxSector;
520 mddev->level = LEVEL_NONE;
521 }
522 EXPORT_SYMBOL_GPL(mddev_init);
523
524 static struct mddev *mddev_find(dev_t unit)
525 {
526 struct mddev *mddev, *new = NULL;
527
528 if (unit && MAJOR(unit) != MD_MAJOR)
529 unit &= ~((1<<MdpMinorShift)-1);
530
531 retry:
532 spin_lock(&all_mddevs_lock);
533
534 if (unit) {
535 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
536 if (mddev->unit == unit) {
537 mddev_get(mddev);
538 spin_unlock(&all_mddevs_lock);
539 kfree(new);
540 return mddev;
541 }
542
543 if (new) {
544 list_add(&new->all_mddevs, &all_mddevs);
545 spin_unlock(&all_mddevs_lock);
546 new->hold_active = UNTIL_IOCTL;
547 return new;
548 }
549 } else if (new) {
550 /* find an unused unit number */
551 static int next_minor = 512;
552 int start = next_minor;
553 int is_free = 0;
554 int dev = 0;
555 while (!is_free) {
556 dev = MKDEV(MD_MAJOR, next_minor);
557 next_minor++;
558 if (next_minor > MINORMASK)
559 next_minor = 0;
560 if (next_minor == start) {
561 /* Oh dear, all in use. */
562 spin_unlock(&all_mddevs_lock);
563 kfree(new);
564 return NULL;
565 }
566
567 is_free = 1;
568 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
569 if (mddev->unit == dev) {
570 is_free = 0;
571 break;
572 }
573 }
574 new->unit = dev;
575 new->md_minor = MINOR(dev);
576 new->hold_active = UNTIL_STOP;
577 list_add(&new->all_mddevs, &all_mddevs);
578 spin_unlock(&all_mddevs_lock);
579 return new;
580 }
581 spin_unlock(&all_mddevs_lock);
582
583 new = kzalloc(sizeof(*new), GFP_KERNEL);
584 if (!new)
585 return NULL;
586
587 new->unit = unit;
588 if (MAJOR(unit) == MD_MAJOR)
589 new->md_minor = MINOR(unit);
590 else
591 new->md_minor = MINOR(unit) >> MdpMinorShift;
592
593 mddev_init(new);
594
595 goto retry;
596 }
597
598 static struct attribute_group md_redundancy_group;
599
600 void mddev_unlock(struct mddev *mddev)
601 {
602 if (mddev->to_remove) {
603 /* These cannot be removed under reconfig_mutex as
604 * an access to the files will try to take reconfig_mutex
605 * while holding the file unremovable, which leads to
606 * a deadlock.
607 * So hold set sysfs_active while the remove in happeing,
608 * and anything else which might set ->to_remove or my
609 * otherwise change the sysfs namespace will fail with
610 * -EBUSY if sysfs_active is still set.
611 * We set sysfs_active under reconfig_mutex and elsewhere
612 * test it under the same mutex to ensure its correct value
613 * is seen.
614 */
615 struct attribute_group *to_remove = mddev->to_remove;
616 mddev->to_remove = NULL;
617 mddev->sysfs_active = 1;
618 mutex_unlock(&mddev->reconfig_mutex);
619
620 if (mddev->kobj.sd) {
621 if (to_remove != &md_redundancy_group)
622 sysfs_remove_group(&mddev->kobj, to_remove);
623 if (mddev->pers == NULL ||
624 mddev->pers->sync_request == NULL) {
625 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
626 if (mddev->sysfs_action)
627 sysfs_put(mddev->sysfs_action);
628 mddev->sysfs_action = NULL;
629 }
630 }
631 mddev->sysfs_active = 0;
632 } else
633 mutex_unlock(&mddev->reconfig_mutex);
634
635 /* As we've dropped the mutex we need a spinlock to
636 * make sure the thread doesn't disappear
637 */
638 spin_lock(&pers_lock);
639 md_wakeup_thread(mddev->thread);
640 spin_unlock(&pers_lock);
641 }
642 EXPORT_SYMBOL_GPL(mddev_unlock);
643
644 static struct md_rdev *find_rdev_nr_rcu(struct mddev *mddev, int nr)
645 {
646 struct md_rdev *rdev;
647
648 rdev_for_each_rcu(rdev, mddev)
649 if (rdev->desc_nr == nr)
650 return rdev;
651
652 return NULL;
653 }
654
655 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
656 {
657 struct md_rdev *rdev;
658
659 rdev_for_each(rdev, mddev)
660 if (rdev->bdev->bd_dev == dev)
661 return rdev;
662
663 return NULL;
664 }
665
666 static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
667 {
668 struct md_rdev *rdev;
669
670 rdev_for_each_rcu(rdev, mddev)
671 if (rdev->bdev->bd_dev == dev)
672 return rdev;
673
674 return NULL;
675 }
676
677 static struct md_personality *find_pers(int level, char *clevel)
678 {
679 struct md_personality *pers;
680 list_for_each_entry(pers, &pers_list, list) {
681 if (level != LEVEL_NONE && pers->level == level)
682 return pers;
683 if (strcmp(pers->name, clevel)==0)
684 return pers;
685 }
686 return NULL;
687 }
688
689 /* return the offset of the super block in 512byte sectors */
690 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
691 {
692 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
693 return MD_NEW_SIZE_SECTORS(num_sectors);
694 }
695
696 static int alloc_disk_sb(struct md_rdev *rdev)
697 {
698 rdev->sb_page = alloc_page(GFP_KERNEL);
699 if (!rdev->sb_page) {
700 printk(KERN_ALERT "md: out of memory.\n");
701 return -ENOMEM;
702 }
703
704 return 0;
705 }
706
707 void md_rdev_clear(struct md_rdev *rdev)
708 {
709 if (rdev->sb_page) {
710 put_page(rdev->sb_page);
711 rdev->sb_loaded = 0;
712 rdev->sb_page = NULL;
713 rdev->sb_start = 0;
714 rdev->sectors = 0;
715 }
716 if (rdev->bb_page) {
717 put_page(rdev->bb_page);
718 rdev->bb_page = NULL;
719 }
720 kfree(rdev->badblocks.page);
721 rdev->badblocks.page = NULL;
722 }
723 EXPORT_SYMBOL_GPL(md_rdev_clear);
724
725 static void super_written(struct bio *bio, int error)
726 {
727 struct md_rdev *rdev = bio->bi_private;
728 struct mddev *mddev = rdev->mddev;
729
730 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
731 printk("md: super_written gets error=%d, uptodate=%d\n",
732 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
733 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
734 md_error(mddev, rdev);
735 }
736
737 if (atomic_dec_and_test(&mddev->pending_writes))
738 wake_up(&mddev->sb_wait);
739 bio_put(bio);
740 }
741
742 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
743 sector_t sector, int size, struct page *page)
744 {
745 /* write first size bytes of page to sector of rdev
746 * Increment mddev->pending_writes before returning
747 * and decrement it on completion, waking up sb_wait
748 * if zero is reached.
749 * If an error occurred, call md_error
750 */
751 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
752
753 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
754 bio->bi_iter.bi_sector = sector;
755 bio_add_page(bio, page, size, 0);
756 bio->bi_private = rdev;
757 bio->bi_end_io = super_written;
758
759 atomic_inc(&mddev->pending_writes);
760 submit_bio(WRITE_FLUSH_FUA, bio);
761 }
762
763 void md_super_wait(struct mddev *mddev)
764 {
765 /* wait for all superblock writes that were scheduled to complete */
766 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
767 }
768
769 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
770 struct page *page, int rw, bool metadata_op)
771 {
772 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
773 int ret;
774
775 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
776 rdev->meta_bdev : rdev->bdev;
777 if (metadata_op)
778 bio->bi_iter.bi_sector = sector + rdev->sb_start;
779 else if (rdev->mddev->reshape_position != MaxSector &&
780 (rdev->mddev->reshape_backwards ==
781 (sector >= rdev->mddev->reshape_position)))
782 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
783 else
784 bio->bi_iter.bi_sector = sector + rdev->data_offset;
785 bio_add_page(bio, page, size, 0);
786 submit_bio_wait(rw, bio);
787
788 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
789 bio_put(bio);
790 return ret;
791 }
792 EXPORT_SYMBOL_GPL(sync_page_io);
793
794 static int read_disk_sb(struct md_rdev *rdev, int size)
795 {
796 char b[BDEVNAME_SIZE];
797
798 if (rdev->sb_loaded)
799 return 0;
800
801 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
802 goto fail;
803 rdev->sb_loaded = 1;
804 return 0;
805
806 fail:
807 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
808 bdevname(rdev->bdev,b));
809 return -EINVAL;
810 }
811
812 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
813 {
814 return sb1->set_uuid0 == sb2->set_uuid0 &&
815 sb1->set_uuid1 == sb2->set_uuid1 &&
816 sb1->set_uuid2 == sb2->set_uuid2 &&
817 sb1->set_uuid3 == sb2->set_uuid3;
818 }
819
820 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
821 {
822 int ret;
823 mdp_super_t *tmp1, *tmp2;
824
825 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
826 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
827
828 if (!tmp1 || !tmp2) {
829 ret = 0;
830 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
831 goto abort;
832 }
833
834 *tmp1 = *sb1;
835 *tmp2 = *sb2;
836
837 /*
838 * nr_disks is not constant
839 */
840 tmp1->nr_disks = 0;
841 tmp2->nr_disks = 0;
842
843 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
844 abort:
845 kfree(tmp1);
846 kfree(tmp2);
847 return ret;
848 }
849
850 static u32 md_csum_fold(u32 csum)
851 {
852 csum = (csum & 0xffff) + (csum >> 16);
853 return (csum & 0xffff) + (csum >> 16);
854 }
855
856 static unsigned int calc_sb_csum(mdp_super_t *sb)
857 {
858 u64 newcsum = 0;
859 u32 *sb32 = (u32*)sb;
860 int i;
861 unsigned int disk_csum, csum;
862
863 disk_csum = sb->sb_csum;
864 sb->sb_csum = 0;
865
866 for (i = 0; i < MD_SB_BYTES/4 ; i++)
867 newcsum += sb32[i];
868 csum = (newcsum & 0xffffffff) + (newcsum>>32);
869
870 #ifdef CONFIG_ALPHA
871 /* This used to use csum_partial, which was wrong for several
872 * reasons including that different results are returned on
873 * different architectures. It isn't critical that we get exactly
874 * the same return value as before (we always csum_fold before
875 * testing, and that removes any differences). However as we
876 * know that csum_partial always returned a 16bit value on
877 * alphas, do a fold to maximise conformity to previous behaviour.
878 */
879 sb->sb_csum = md_csum_fold(disk_csum);
880 #else
881 sb->sb_csum = disk_csum;
882 #endif
883 return csum;
884 }
885
886 /*
887 * Handle superblock details.
888 * We want to be able to handle multiple superblock formats
889 * so we have a common interface to them all, and an array of
890 * different handlers.
891 * We rely on user-space to write the initial superblock, and support
892 * reading and updating of superblocks.
893 * Interface methods are:
894 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
895 * loads and validates a superblock on dev.
896 * if refdev != NULL, compare superblocks on both devices
897 * Return:
898 * 0 - dev has a superblock that is compatible with refdev
899 * 1 - dev has a superblock that is compatible and newer than refdev
900 * so dev should be used as the refdev in future
901 * -EINVAL superblock incompatible or invalid
902 * -othererror e.g. -EIO
903 *
904 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
905 * Verify that dev is acceptable into mddev.
906 * The first time, mddev->raid_disks will be 0, and data from
907 * dev should be merged in. Subsequent calls check that dev
908 * is new enough. Return 0 or -EINVAL
909 *
910 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
911 * Update the superblock for rdev with data in mddev
912 * This does not write to disc.
913 *
914 */
915
916 struct super_type {
917 char *name;
918 struct module *owner;
919 int (*load_super)(struct md_rdev *rdev,
920 struct md_rdev *refdev,
921 int minor_version);
922 int (*validate_super)(struct mddev *mddev,
923 struct md_rdev *rdev);
924 void (*sync_super)(struct mddev *mddev,
925 struct md_rdev *rdev);
926 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
927 sector_t num_sectors);
928 int (*allow_new_offset)(struct md_rdev *rdev,
929 unsigned long long new_offset);
930 };
931
932 /*
933 * Check that the given mddev has no bitmap.
934 *
935 * This function is called from the run method of all personalities that do not
936 * support bitmaps. It prints an error message and returns non-zero if mddev
937 * has a bitmap. Otherwise, it returns 0.
938 *
939 */
940 int md_check_no_bitmap(struct mddev *mddev)
941 {
942 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
943 return 0;
944 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
945 mdname(mddev), mddev->pers->name);
946 return 1;
947 }
948 EXPORT_SYMBOL(md_check_no_bitmap);
949
950 /*
951 * load_super for 0.90.0
952 */
953 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
954 {
955 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
956 mdp_super_t *sb;
957 int ret;
958
959 /*
960 * Calculate the position of the superblock (512byte sectors),
961 * it's at the end of the disk.
962 *
963 * It also happens to be a multiple of 4Kb.
964 */
965 rdev->sb_start = calc_dev_sboffset(rdev);
966
967 ret = read_disk_sb(rdev, MD_SB_BYTES);
968 if (ret) return ret;
969
970 ret = -EINVAL;
971
972 bdevname(rdev->bdev, b);
973 sb = page_address(rdev->sb_page);
974
975 if (sb->md_magic != MD_SB_MAGIC) {
976 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
977 b);
978 goto abort;
979 }
980
981 if (sb->major_version != 0 ||
982 sb->minor_version < 90 ||
983 sb->minor_version > 91) {
984 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
985 sb->major_version, sb->minor_version,
986 b);
987 goto abort;
988 }
989
990 if (sb->raid_disks <= 0)
991 goto abort;
992
993 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
994 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
995 b);
996 goto abort;
997 }
998
999 rdev->preferred_minor = sb->md_minor;
1000 rdev->data_offset = 0;
1001 rdev->new_data_offset = 0;
1002 rdev->sb_size = MD_SB_BYTES;
1003 rdev->badblocks.shift = -1;
1004
1005 if (sb->level == LEVEL_MULTIPATH)
1006 rdev->desc_nr = -1;
1007 else
1008 rdev->desc_nr = sb->this_disk.number;
1009
1010 if (!refdev) {
1011 ret = 1;
1012 } else {
1013 __u64 ev1, ev2;
1014 mdp_super_t *refsb = page_address(refdev->sb_page);
1015 if (!uuid_equal(refsb, sb)) {
1016 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1017 b, bdevname(refdev->bdev,b2));
1018 goto abort;
1019 }
1020 if (!sb_equal(refsb, sb)) {
1021 printk(KERN_WARNING "md: %s has same UUID"
1022 " but different superblock to %s\n",
1023 b, bdevname(refdev->bdev, b2));
1024 goto abort;
1025 }
1026 ev1 = md_event(sb);
1027 ev2 = md_event(refsb);
1028 if (ev1 > ev2)
1029 ret = 1;
1030 else
1031 ret = 0;
1032 }
1033 rdev->sectors = rdev->sb_start;
1034 /* Limit to 4TB as metadata cannot record more than that.
1035 * (not needed for Linear and RAID0 as metadata doesn't
1036 * record this size)
1037 */
1038 if (rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1039 rdev->sectors = (2ULL << 32) - 2;
1040
1041 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1042 /* "this cannot possibly happen" ... */
1043 ret = -EINVAL;
1044
1045 abort:
1046 return ret;
1047 }
1048
1049 /*
1050 * validate_super for 0.90.0
1051 */
1052 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1053 {
1054 mdp_disk_t *desc;
1055 mdp_super_t *sb = page_address(rdev->sb_page);
1056 __u64 ev1 = md_event(sb);
1057
1058 rdev->raid_disk = -1;
1059 clear_bit(Faulty, &rdev->flags);
1060 clear_bit(In_sync, &rdev->flags);
1061 clear_bit(Bitmap_sync, &rdev->flags);
1062 clear_bit(WriteMostly, &rdev->flags);
1063
1064 if (mddev->raid_disks == 0) {
1065 mddev->major_version = 0;
1066 mddev->minor_version = sb->minor_version;
1067 mddev->patch_version = sb->patch_version;
1068 mddev->external = 0;
1069 mddev->chunk_sectors = sb->chunk_size >> 9;
1070 mddev->ctime = sb->ctime;
1071 mddev->utime = sb->utime;
1072 mddev->level = sb->level;
1073 mddev->clevel[0] = 0;
1074 mddev->layout = sb->layout;
1075 mddev->raid_disks = sb->raid_disks;
1076 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1077 mddev->events = ev1;
1078 mddev->bitmap_info.offset = 0;
1079 mddev->bitmap_info.space = 0;
1080 /* bitmap can use 60 K after the 4K superblocks */
1081 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1082 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1083 mddev->reshape_backwards = 0;
1084
1085 if (mddev->minor_version >= 91) {
1086 mddev->reshape_position = sb->reshape_position;
1087 mddev->delta_disks = sb->delta_disks;
1088 mddev->new_level = sb->new_level;
1089 mddev->new_layout = sb->new_layout;
1090 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1091 if (mddev->delta_disks < 0)
1092 mddev->reshape_backwards = 1;
1093 } else {
1094 mddev->reshape_position = MaxSector;
1095 mddev->delta_disks = 0;
1096 mddev->new_level = mddev->level;
1097 mddev->new_layout = mddev->layout;
1098 mddev->new_chunk_sectors = mddev->chunk_sectors;
1099 }
1100
1101 if (sb->state & (1<<MD_SB_CLEAN))
1102 mddev->recovery_cp = MaxSector;
1103 else {
1104 if (sb->events_hi == sb->cp_events_hi &&
1105 sb->events_lo == sb->cp_events_lo) {
1106 mddev->recovery_cp = sb->recovery_cp;
1107 } else
1108 mddev->recovery_cp = 0;
1109 }
1110
1111 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1112 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1113 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1114 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1115
1116 mddev->max_disks = MD_SB_DISKS;
1117
1118 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1119 mddev->bitmap_info.file == NULL) {
1120 mddev->bitmap_info.offset =
1121 mddev->bitmap_info.default_offset;
1122 mddev->bitmap_info.space =
1123 mddev->bitmap_info.default_space;
1124 }
1125
1126 } else if (mddev->pers == NULL) {
1127 /* Insist on good event counter while assembling, except
1128 * for spares (which don't need an event count) */
1129 ++ev1;
1130 if (sb->disks[rdev->desc_nr].state & (
1131 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1132 if (ev1 < mddev->events)
1133 return -EINVAL;
1134 } else if (mddev->bitmap) {
1135 /* if adding to array with a bitmap, then we can accept an
1136 * older device ... but not too old.
1137 */
1138 if (ev1 < mddev->bitmap->events_cleared)
1139 return 0;
1140 if (ev1 < mddev->events)
1141 set_bit(Bitmap_sync, &rdev->flags);
1142 } else {
1143 if (ev1 < mddev->events)
1144 /* just a hot-add of a new device, leave raid_disk at -1 */
1145 return 0;
1146 }
1147
1148 if (mddev->level != LEVEL_MULTIPATH) {
1149 desc = sb->disks + rdev->desc_nr;
1150
1151 if (desc->state & (1<<MD_DISK_FAULTY))
1152 set_bit(Faulty, &rdev->flags);
1153 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1154 desc->raid_disk < mddev->raid_disks */) {
1155 set_bit(In_sync, &rdev->flags);
1156 rdev->raid_disk = desc->raid_disk;
1157 rdev->saved_raid_disk = desc->raid_disk;
1158 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1159 /* active but not in sync implies recovery up to
1160 * reshape position. We don't know exactly where
1161 * that is, so set to zero for now */
1162 if (mddev->minor_version >= 91) {
1163 rdev->recovery_offset = 0;
1164 rdev->raid_disk = desc->raid_disk;
1165 }
1166 }
1167 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1168 set_bit(WriteMostly, &rdev->flags);
1169 } else /* MULTIPATH are always insync */
1170 set_bit(In_sync, &rdev->flags);
1171 return 0;
1172 }
1173
1174 /*
1175 * sync_super for 0.90.0
1176 */
1177 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1178 {
1179 mdp_super_t *sb;
1180 struct md_rdev *rdev2;
1181 int next_spare = mddev->raid_disks;
1182
1183 /* make rdev->sb match mddev data..
1184 *
1185 * 1/ zero out disks
1186 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1187 * 3/ any empty disks < next_spare become removed
1188 *
1189 * disks[0] gets initialised to REMOVED because
1190 * we cannot be sure from other fields if it has
1191 * been initialised or not.
1192 */
1193 int i;
1194 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1195
1196 rdev->sb_size = MD_SB_BYTES;
1197
1198 sb = page_address(rdev->sb_page);
1199
1200 memset(sb, 0, sizeof(*sb));
1201
1202 sb->md_magic = MD_SB_MAGIC;
1203 sb->major_version = mddev->major_version;
1204 sb->patch_version = mddev->patch_version;
1205 sb->gvalid_words = 0; /* ignored */
1206 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1207 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1208 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1209 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1210
1211 sb->ctime = mddev->ctime;
1212 sb->level = mddev->level;
1213 sb->size = mddev->dev_sectors / 2;
1214 sb->raid_disks = mddev->raid_disks;
1215 sb->md_minor = mddev->md_minor;
1216 sb->not_persistent = 0;
1217 sb->utime = mddev->utime;
1218 sb->state = 0;
1219 sb->events_hi = (mddev->events>>32);
1220 sb->events_lo = (u32)mddev->events;
1221
1222 if (mddev->reshape_position == MaxSector)
1223 sb->minor_version = 90;
1224 else {
1225 sb->minor_version = 91;
1226 sb->reshape_position = mddev->reshape_position;
1227 sb->new_level = mddev->new_level;
1228 sb->delta_disks = mddev->delta_disks;
1229 sb->new_layout = mddev->new_layout;
1230 sb->new_chunk = mddev->new_chunk_sectors << 9;
1231 }
1232 mddev->minor_version = sb->minor_version;
1233 if (mddev->in_sync)
1234 {
1235 sb->recovery_cp = mddev->recovery_cp;
1236 sb->cp_events_hi = (mddev->events>>32);
1237 sb->cp_events_lo = (u32)mddev->events;
1238 if (mddev->recovery_cp == MaxSector)
1239 sb->state = (1<< MD_SB_CLEAN);
1240 } else
1241 sb->recovery_cp = 0;
1242
1243 sb->layout = mddev->layout;
1244 sb->chunk_size = mddev->chunk_sectors << 9;
1245
1246 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1247 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1248
1249 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1250 rdev_for_each(rdev2, mddev) {
1251 mdp_disk_t *d;
1252 int desc_nr;
1253 int is_active = test_bit(In_sync, &rdev2->flags);
1254
1255 if (rdev2->raid_disk >= 0 &&
1256 sb->minor_version >= 91)
1257 /* we have nowhere to store the recovery_offset,
1258 * but if it is not below the reshape_position,
1259 * we can piggy-back on that.
1260 */
1261 is_active = 1;
1262 if (rdev2->raid_disk < 0 ||
1263 test_bit(Faulty, &rdev2->flags))
1264 is_active = 0;
1265 if (is_active)
1266 desc_nr = rdev2->raid_disk;
1267 else
1268 desc_nr = next_spare++;
1269 rdev2->desc_nr = desc_nr;
1270 d = &sb->disks[rdev2->desc_nr];
1271 nr_disks++;
1272 d->number = rdev2->desc_nr;
1273 d->major = MAJOR(rdev2->bdev->bd_dev);
1274 d->minor = MINOR(rdev2->bdev->bd_dev);
1275 if (is_active)
1276 d->raid_disk = rdev2->raid_disk;
1277 else
1278 d->raid_disk = rdev2->desc_nr; /* compatibility */
1279 if (test_bit(Faulty, &rdev2->flags))
1280 d->state = (1<<MD_DISK_FAULTY);
1281 else if (is_active) {
1282 d->state = (1<<MD_DISK_ACTIVE);
1283 if (test_bit(In_sync, &rdev2->flags))
1284 d->state |= (1<<MD_DISK_SYNC);
1285 active++;
1286 working++;
1287 } else {
1288 d->state = 0;
1289 spare++;
1290 working++;
1291 }
1292 if (test_bit(WriteMostly, &rdev2->flags))
1293 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1294 }
1295 /* now set the "removed" and "faulty" bits on any missing devices */
1296 for (i=0 ; i < mddev->raid_disks ; i++) {
1297 mdp_disk_t *d = &sb->disks[i];
1298 if (d->state == 0 && d->number == 0) {
1299 d->number = i;
1300 d->raid_disk = i;
1301 d->state = (1<<MD_DISK_REMOVED);
1302 d->state |= (1<<MD_DISK_FAULTY);
1303 failed++;
1304 }
1305 }
1306 sb->nr_disks = nr_disks;
1307 sb->active_disks = active;
1308 sb->working_disks = working;
1309 sb->failed_disks = failed;
1310 sb->spare_disks = spare;
1311
1312 sb->this_disk = sb->disks[rdev->desc_nr];
1313 sb->sb_csum = calc_sb_csum(sb);
1314 }
1315
1316 /*
1317 * rdev_size_change for 0.90.0
1318 */
1319 static unsigned long long
1320 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1321 {
1322 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1323 return 0; /* component must fit device */
1324 if (rdev->mddev->bitmap_info.offset)
1325 return 0; /* can't move bitmap */
1326 rdev->sb_start = calc_dev_sboffset(rdev);
1327 if (!num_sectors || num_sectors > rdev->sb_start)
1328 num_sectors = rdev->sb_start;
1329 /* Limit to 4TB as metadata cannot record more than that.
1330 * 4TB == 2^32 KB, or 2*2^32 sectors.
1331 */
1332 if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1333 num_sectors = (2ULL << 32) - 2;
1334 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1335 rdev->sb_page);
1336 md_super_wait(rdev->mddev);
1337 return num_sectors;
1338 }
1339
1340 static int
1341 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1342 {
1343 /* non-zero offset changes not possible with v0.90 */
1344 return new_offset == 0;
1345 }
1346
1347 /*
1348 * version 1 superblock
1349 */
1350
1351 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1352 {
1353 __le32 disk_csum;
1354 u32 csum;
1355 unsigned long long newcsum;
1356 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1357 __le32 *isuper = (__le32*)sb;
1358
1359 disk_csum = sb->sb_csum;
1360 sb->sb_csum = 0;
1361 newcsum = 0;
1362 for (; size >= 4; size -= 4)
1363 newcsum += le32_to_cpu(*isuper++);
1364
1365 if (size == 2)
1366 newcsum += le16_to_cpu(*(__le16*) isuper);
1367
1368 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1369 sb->sb_csum = disk_csum;
1370 return cpu_to_le32(csum);
1371 }
1372
1373 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1374 int acknowledged);
1375 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1376 {
1377 struct mdp_superblock_1 *sb;
1378 int ret;
1379 sector_t sb_start;
1380 sector_t sectors;
1381 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1382 int bmask;
1383
1384 /*
1385 * Calculate the position of the superblock in 512byte sectors.
1386 * It is always aligned to a 4K boundary and
1387 * depeding on minor_version, it can be:
1388 * 0: At least 8K, but less than 12K, from end of device
1389 * 1: At start of device
1390 * 2: 4K from start of device.
1391 */
1392 switch(minor_version) {
1393 case 0:
1394 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1395 sb_start -= 8*2;
1396 sb_start &= ~(sector_t)(4*2-1);
1397 break;
1398 case 1:
1399 sb_start = 0;
1400 break;
1401 case 2:
1402 sb_start = 8;
1403 break;
1404 default:
1405 return -EINVAL;
1406 }
1407 rdev->sb_start = sb_start;
1408
1409 /* superblock is rarely larger than 1K, but it can be larger,
1410 * and it is safe to read 4k, so we do that
1411 */
1412 ret = read_disk_sb(rdev, 4096);
1413 if (ret) return ret;
1414
1415 sb = page_address(rdev->sb_page);
1416
1417 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1418 sb->major_version != cpu_to_le32(1) ||
1419 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1420 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1421 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1422 return -EINVAL;
1423
1424 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1425 printk("md: invalid superblock checksum on %s\n",
1426 bdevname(rdev->bdev,b));
1427 return -EINVAL;
1428 }
1429 if (le64_to_cpu(sb->data_size) < 10) {
1430 printk("md: data_size too small on %s\n",
1431 bdevname(rdev->bdev,b));
1432 return -EINVAL;
1433 }
1434 if (sb->pad0 ||
1435 sb->pad3[0] ||
1436 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1437 /* Some padding is non-zero, might be a new feature */
1438 return -EINVAL;
1439
1440 rdev->preferred_minor = 0xffff;
1441 rdev->data_offset = le64_to_cpu(sb->data_offset);
1442 rdev->new_data_offset = rdev->data_offset;
1443 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1444 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1445 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1446 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1447
1448 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1449 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1450 if (rdev->sb_size & bmask)
1451 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1452
1453 if (minor_version
1454 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1455 return -EINVAL;
1456 if (minor_version
1457 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1458 return -EINVAL;
1459
1460 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1461 rdev->desc_nr = -1;
1462 else
1463 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1464
1465 if (!rdev->bb_page) {
1466 rdev->bb_page = alloc_page(GFP_KERNEL);
1467 if (!rdev->bb_page)
1468 return -ENOMEM;
1469 }
1470 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1471 rdev->badblocks.count == 0) {
1472 /* need to load the bad block list.
1473 * Currently we limit it to one page.
1474 */
1475 s32 offset;
1476 sector_t bb_sector;
1477 u64 *bbp;
1478 int i;
1479 int sectors = le16_to_cpu(sb->bblog_size);
1480 if (sectors > (PAGE_SIZE / 512))
1481 return -EINVAL;
1482 offset = le32_to_cpu(sb->bblog_offset);
1483 if (offset == 0)
1484 return -EINVAL;
1485 bb_sector = (long long)offset;
1486 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1487 rdev->bb_page, READ, true))
1488 return -EIO;
1489 bbp = (u64 *)page_address(rdev->bb_page);
1490 rdev->badblocks.shift = sb->bblog_shift;
1491 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1492 u64 bb = le64_to_cpu(*bbp);
1493 int count = bb & (0x3ff);
1494 u64 sector = bb >> 10;
1495 sector <<= sb->bblog_shift;
1496 count <<= sb->bblog_shift;
1497 if (bb + 1 == 0)
1498 break;
1499 if (md_set_badblocks(&rdev->badblocks,
1500 sector, count, 1) == 0)
1501 return -EINVAL;
1502 }
1503 } else if (sb->bblog_offset != 0)
1504 rdev->badblocks.shift = 0;
1505
1506 if (!refdev) {
1507 ret = 1;
1508 } else {
1509 __u64 ev1, ev2;
1510 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1511
1512 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1513 sb->level != refsb->level ||
1514 sb->layout != refsb->layout ||
1515 sb->chunksize != refsb->chunksize) {
1516 printk(KERN_WARNING "md: %s has strangely different"
1517 " superblock to %s\n",
1518 bdevname(rdev->bdev,b),
1519 bdevname(refdev->bdev,b2));
1520 return -EINVAL;
1521 }
1522 ev1 = le64_to_cpu(sb->events);
1523 ev2 = le64_to_cpu(refsb->events);
1524
1525 if (ev1 > ev2)
1526 ret = 1;
1527 else
1528 ret = 0;
1529 }
1530 if (minor_version) {
1531 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1532 sectors -= rdev->data_offset;
1533 } else
1534 sectors = rdev->sb_start;
1535 if (sectors < le64_to_cpu(sb->data_size))
1536 return -EINVAL;
1537 rdev->sectors = le64_to_cpu(sb->data_size);
1538 return ret;
1539 }
1540
1541 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1542 {
1543 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1544 __u64 ev1 = le64_to_cpu(sb->events);
1545
1546 rdev->raid_disk = -1;
1547 clear_bit(Faulty, &rdev->flags);
1548 clear_bit(In_sync, &rdev->flags);
1549 clear_bit(Bitmap_sync, &rdev->flags);
1550 clear_bit(WriteMostly, &rdev->flags);
1551
1552 if (mddev->raid_disks == 0) {
1553 mddev->major_version = 1;
1554 mddev->patch_version = 0;
1555 mddev->external = 0;
1556 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1557 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1558 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1559 mddev->level = le32_to_cpu(sb->level);
1560 mddev->clevel[0] = 0;
1561 mddev->layout = le32_to_cpu(sb->layout);
1562 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1563 mddev->dev_sectors = le64_to_cpu(sb->size);
1564 mddev->events = ev1;
1565 mddev->bitmap_info.offset = 0;
1566 mddev->bitmap_info.space = 0;
1567 /* Default location for bitmap is 1K after superblock
1568 * using 3K - total of 4K
1569 */
1570 mddev->bitmap_info.default_offset = 1024 >> 9;
1571 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1572 mddev->reshape_backwards = 0;
1573
1574 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1575 memcpy(mddev->uuid, sb->set_uuid, 16);
1576
1577 mddev->max_disks = (4096-256)/2;
1578
1579 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1580 mddev->bitmap_info.file == NULL) {
1581 mddev->bitmap_info.offset =
1582 (__s32)le32_to_cpu(sb->bitmap_offset);
1583 /* Metadata doesn't record how much space is available.
1584 * For 1.0, we assume we can use up to the superblock
1585 * if before, else to 4K beyond superblock.
1586 * For others, assume no change is possible.
1587 */
1588 if (mddev->minor_version > 0)
1589 mddev->bitmap_info.space = 0;
1590 else if (mddev->bitmap_info.offset > 0)
1591 mddev->bitmap_info.space =
1592 8 - mddev->bitmap_info.offset;
1593 else
1594 mddev->bitmap_info.space =
1595 -mddev->bitmap_info.offset;
1596 }
1597
1598 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1599 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1600 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1601 mddev->new_level = le32_to_cpu(sb->new_level);
1602 mddev->new_layout = le32_to_cpu(sb->new_layout);
1603 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1604 if (mddev->delta_disks < 0 ||
1605 (mddev->delta_disks == 0 &&
1606 (le32_to_cpu(sb->feature_map)
1607 & MD_FEATURE_RESHAPE_BACKWARDS)))
1608 mddev->reshape_backwards = 1;
1609 } else {
1610 mddev->reshape_position = MaxSector;
1611 mddev->delta_disks = 0;
1612 mddev->new_level = mddev->level;
1613 mddev->new_layout = mddev->layout;
1614 mddev->new_chunk_sectors = mddev->chunk_sectors;
1615 }
1616
1617 } else if (mddev->pers == NULL) {
1618 /* Insist of good event counter while assembling, except for
1619 * spares (which don't need an event count) */
1620 ++ev1;
1621 if (rdev->desc_nr >= 0 &&
1622 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1623 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1624 if (ev1 < mddev->events)
1625 return -EINVAL;
1626 } else if (mddev->bitmap) {
1627 /* If adding to array with a bitmap, then we can accept an
1628 * older device, but not too old.
1629 */
1630 if (ev1 < mddev->bitmap->events_cleared)
1631 return 0;
1632 if (ev1 < mddev->events)
1633 set_bit(Bitmap_sync, &rdev->flags);
1634 } else {
1635 if (ev1 < mddev->events)
1636 /* just a hot-add of a new device, leave raid_disk at -1 */
1637 return 0;
1638 }
1639 if (mddev->level != LEVEL_MULTIPATH) {
1640 int role;
1641 if (rdev->desc_nr < 0 ||
1642 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1643 role = 0xffff;
1644 rdev->desc_nr = -1;
1645 } else
1646 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1647 switch(role) {
1648 case 0xffff: /* spare */
1649 break;
1650 case 0xfffe: /* faulty */
1651 set_bit(Faulty, &rdev->flags);
1652 break;
1653 default:
1654 rdev->saved_raid_disk = role;
1655 if ((le32_to_cpu(sb->feature_map) &
1656 MD_FEATURE_RECOVERY_OFFSET)) {
1657 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1658 if (!(le32_to_cpu(sb->feature_map) &
1659 MD_FEATURE_RECOVERY_BITMAP))
1660 rdev->saved_raid_disk = -1;
1661 } else
1662 set_bit(In_sync, &rdev->flags);
1663 rdev->raid_disk = role;
1664 break;
1665 }
1666 if (sb->devflags & WriteMostly1)
1667 set_bit(WriteMostly, &rdev->flags);
1668 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1669 set_bit(Replacement, &rdev->flags);
1670 } else /* MULTIPATH are always insync */
1671 set_bit(In_sync, &rdev->flags);
1672
1673 return 0;
1674 }
1675
1676 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1677 {
1678 struct mdp_superblock_1 *sb;
1679 struct md_rdev *rdev2;
1680 int max_dev, i;
1681 /* make rdev->sb match mddev and rdev data. */
1682
1683 sb = page_address(rdev->sb_page);
1684
1685 sb->feature_map = 0;
1686 sb->pad0 = 0;
1687 sb->recovery_offset = cpu_to_le64(0);
1688 memset(sb->pad3, 0, sizeof(sb->pad3));
1689
1690 sb->utime = cpu_to_le64((__u64)mddev->utime);
1691 sb->events = cpu_to_le64(mddev->events);
1692 if (mddev->in_sync)
1693 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1694 else
1695 sb->resync_offset = cpu_to_le64(0);
1696
1697 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1698
1699 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1700 sb->size = cpu_to_le64(mddev->dev_sectors);
1701 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1702 sb->level = cpu_to_le32(mddev->level);
1703 sb->layout = cpu_to_le32(mddev->layout);
1704
1705 if (test_bit(WriteMostly, &rdev->flags))
1706 sb->devflags |= WriteMostly1;
1707 else
1708 sb->devflags &= ~WriteMostly1;
1709 sb->data_offset = cpu_to_le64(rdev->data_offset);
1710 sb->data_size = cpu_to_le64(rdev->sectors);
1711
1712 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1713 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1714 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1715 }
1716
1717 if (rdev->raid_disk >= 0 &&
1718 !test_bit(In_sync, &rdev->flags)) {
1719 sb->feature_map |=
1720 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1721 sb->recovery_offset =
1722 cpu_to_le64(rdev->recovery_offset);
1723 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1724 sb->feature_map |=
1725 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
1726 }
1727 if (test_bit(Replacement, &rdev->flags))
1728 sb->feature_map |=
1729 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1730
1731 if (mddev->reshape_position != MaxSector) {
1732 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1733 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1734 sb->new_layout = cpu_to_le32(mddev->new_layout);
1735 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1736 sb->new_level = cpu_to_le32(mddev->new_level);
1737 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1738 if (mddev->delta_disks == 0 &&
1739 mddev->reshape_backwards)
1740 sb->feature_map
1741 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1742 if (rdev->new_data_offset != rdev->data_offset) {
1743 sb->feature_map
1744 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1745 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1746 - rdev->data_offset));
1747 }
1748 }
1749
1750 if (rdev->badblocks.count == 0)
1751 /* Nothing to do for bad blocks*/ ;
1752 else if (sb->bblog_offset == 0)
1753 /* Cannot record bad blocks on this device */
1754 md_error(mddev, rdev);
1755 else {
1756 struct badblocks *bb = &rdev->badblocks;
1757 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1758 u64 *p = bb->page;
1759 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1760 if (bb->changed) {
1761 unsigned seq;
1762
1763 retry:
1764 seq = read_seqbegin(&bb->lock);
1765
1766 memset(bbp, 0xff, PAGE_SIZE);
1767
1768 for (i = 0 ; i < bb->count ; i++) {
1769 u64 internal_bb = p[i];
1770 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1771 | BB_LEN(internal_bb));
1772 bbp[i] = cpu_to_le64(store_bb);
1773 }
1774 bb->changed = 0;
1775 if (read_seqretry(&bb->lock, seq))
1776 goto retry;
1777
1778 bb->sector = (rdev->sb_start +
1779 (int)le32_to_cpu(sb->bblog_offset));
1780 bb->size = le16_to_cpu(sb->bblog_size);
1781 }
1782 }
1783
1784 max_dev = 0;
1785 rdev_for_each(rdev2, mddev)
1786 if (rdev2->desc_nr+1 > max_dev)
1787 max_dev = rdev2->desc_nr+1;
1788
1789 if (max_dev > le32_to_cpu(sb->max_dev)) {
1790 int bmask;
1791 sb->max_dev = cpu_to_le32(max_dev);
1792 rdev->sb_size = max_dev * 2 + 256;
1793 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1794 if (rdev->sb_size & bmask)
1795 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1796 } else
1797 max_dev = le32_to_cpu(sb->max_dev);
1798
1799 for (i=0; i<max_dev;i++)
1800 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1801
1802 rdev_for_each(rdev2, mddev) {
1803 i = rdev2->desc_nr;
1804 if (test_bit(Faulty, &rdev2->flags))
1805 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1806 else if (test_bit(In_sync, &rdev2->flags))
1807 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1808 else if (rdev2->raid_disk >= 0)
1809 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1810 else
1811 sb->dev_roles[i] = cpu_to_le16(0xffff);
1812 }
1813
1814 sb->sb_csum = calc_sb_1_csum(sb);
1815 }
1816
1817 static unsigned long long
1818 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1819 {
1820 struct mdp_superblock_1 *sb;
1821 sector_t max_sectors;
1822 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1823 return 0; /* component must fit device */
1824 if (rdev->data_offset != rdev->new_data_offset)
1825 return 0; /* too confusing */
1826 if (rdev->sb_start < rdev->data_offset) {
1827 /* minor versions 1 and 2; superblock before data */
1828 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1829 max_sectors -= rdev->data_offset;
1830 if (!num_sectors || num_sectors > max_sectors)
1831 num_sectors = max_sectors;
1832 } else if (rdev->mddev->bitmap_info.offset) {
1833 /* minor version 0 with bitmap we can't move */
1834 return 0;
1835 } else {
1836 /* minor version 0; superblock after data */
1837 sector_t sb_start;
1838 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1839 sb_start &= ~(sector_t)(4*2 - 1);
1840 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1841 if (!num_sectors || num_sectors > max_sectors)
1842 num_sectors = max_sectors;
1843 rdev->sb_start = sb_start;
1844 }
1845 sb = page_address(rdev->sb_page);
1846 sb->data_size = cpu_to_le64(num_sectors);
1847 sb->super_offset = rdev->sb_start;
1848 sb->sb_csum = calc_sb_1_csum(sb);
1849 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1850 rdev->sb_page);
1851 md_super_wait(rdev->mddev);
1852 return num_sectors;
1853
1854 }
1855
1856 static int
1857 super_1_allow_new_offset(struct md_rdev *rdev,
1858 unsigned long long new_offset)
1859 {
1860 /* All necessary checks on new >= old have been done */
1861 struct bitmap *bitmap;
1862 if (new_offset >= rdev->data_offset)
1863 return 1;
1864
1865 /* with 1.0 metadata, there is no metadata to tread on
1866 * so we can always move back */
1867 if (rdev->mddev->minor_version == 0)
1868 return 1;
1869
1870 /* otherwise we must be sure not to step on
1871 * any metadata, so stay:
1872 * 36K beyond start of superblock
1873 * beyond end of badblocks
1874 * beyond write-intent bitmap
1875 */
1876 if (rdev->sb_start + (32+4)*2 > new_offset)
1877 return 0;
1878 bitmap = rdev->mddev->bitmap;
1879 if (bitmap && !rdev->mddev->bitmap_info.file &&
1880 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1881 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
1882 return 0;
1883 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1884 return 0;
1885
1886 return 1;
1887 }
1888
1889 static struct super_type super_types[] = {
1890 [0] = {
1891 .name = "0.90.0",
1892 .owner = THIS_MODULE,
1893 .load_super = super_90_load,
1894 .validate_super = super_90_validate,
1895 .sync_super = super_90_sync,
1896 .rdev_size_change = super_90_rdev_size_change,
1897 .allow_new_offset = super_90_allow_new_offset,
1898 },
1899 [1] = {
1900 .name = "md-1",
1901 .owner = THIS_MODULE,
1902 .load_super = super_1_load,
1903 .validate_super = super_1_validate,
1904 .sync_super = super_1_sync,
1905 .rdev_size_change = super_1_rdev_size_change,
1906 .allow_new_offset = super_1_allow_new_offset,
1907 },
1908 };
1909
1910 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1911 {
1912 if (mddev->sync_super) {
1913 mddev->sync_super(mddev, rdev);
1914 return;
1915 }
1916
1917 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1918
1919 super_types[mddev->major_version].sync_super(mddev, rdev);
1920 }
1921
1922 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1923 {
1924 struct md_rdev *rdev, *rdev2;
1925
1926 rcu_read_lock();
1927 rdev_for_each_rcu(rdev, mddev1)
1928 rdev_for_each_rcu(rdev2, mddev2)
1929 if (rdev->bdev->bd_contains ==
1930 rdev2->bdev->bd_contains) {
1931 rcu_read_unlock();
1932 return 1;
1933 }
1934 rcu_read_unlock();
1935 return 0;
1936 }
1937
1938 static LIST_HEAD(pending_raid_disks);
1939
1940 /*
1941 * Try to register data integrity profile for an mddev
1942 *
1943 * This is called when an array is started and after a disk has been kicked
1944 * from the array. It only succeeds if all working and active component devices
1945 * are integrity capable with matching profiles.
1946 */
1947 int md_integrity_register(struct mddev *mddev)
1948 {
1949 struct md_rdev *rdev, *reference = NULL;
1950
1951 if (list_empty(&mddev->disks))
1952 return 0; /* nothing to do */
1953 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1954 return 0; /* shouldn't register, or already is */
1955 rdev_for_each(rdev, mddev) {
1956 /* skip spares and non-functional disks */
1957 if (test_bit(Faulty, &rdev->flags))
1958 continue;
1959 if (rdev->raid_disk < 0)
1960 continue;
1961 if (!reference) {
1962 /* Use the first rdev as the reference */
1963 reference = rdev;
1964 continue;
1965 }
1966 /* does this rdev's profile match the reference profile? */
1967 if (blk_integrity_compare(reference->bdev->bd_disk,
1968 rdev->bdev->bd_disk) < 0)
1969 return -EINVAL;
1970 }
1971 if (!reference || !bdev_get_integrity(reference->bdev))
1972 return 0;
1973 /*
1974 * All component devices are integrity capable and have matching
1975 * profiles, register the common profile for the md device.
1976 */
1977 if (blk_integrity_register(mddev->gendisk,
1978 bdev_get_integrity(reference->bdev)) != 0) {
1979 printk(KERN_ERR "md: failed to register integrity for %s\n",
1980 mdname(mddev));
1981 return -EINVAL;
1982 }
1983 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
1984 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
1985 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
1986 mdname(mddev));
1987 return -EINVAL;
1988 }
1989 return 0;
1990 }
1991 EXPORT_SYMBOL(md_integrity_register);
1992
1993 /* Disable data integrity if non-capable/non-matching disk is being added */
1994 void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
1995 {
1996 struct blk_integrity *bi_rdev;
1997 struct blk_integrity *bi_mddev;
1998
1999 if (!mddev->gendisk)
2000 return;
2001
2002 bi_rdev = bdev_get_integrity(rdev->bdev);
2003 bi_mddev = blk_get_integrity(mddev->gendisk);
2004
2005 if (!bi_mddev) /* nothing to do */
2006 return;
2007 if (rdev->raid_disk < 0) /* skip spares */
2008 return;
2009 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2010 rdev->bdev->bd_disk) >= 0)
2011 return;
2012 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2013 blk_integrity_unregister(mddev->gendisk);
2014 }
2015 EXPORT_SYMBOL(md_integrity_add_rdev);
2016
2017 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2018 {
2019 char b[BDEVNAME_SIZE];
2020 struct kobject *ko;
2021 char *s;
2022 int err;
2023
2024 /* prevent duplicates */
2025 if (find_rdev(mddev, rdev->bdev->bd_dev))
2026 return -EEXIST;
2027
2028 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2029 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2030 rdev->sectors < mddev->dev_sectors)) {
2031 if (mddev->pers) {
2032 /* Cannot change size, so fail
2033 * If mddev->level <= 0, then we don't care
2034 * about aligning sizes (e.g. linear)
2035 */
2036 if (mddev->level > 0)
2037 return -ENOSPC;
2038 } else
2039 mddev->dev_sectors = rdev->sectors;
2040 }
2041
2042 /* Verify rdev->desc_nr is unique.
2043 * If it is -1, assign a free number, else
2044 * check number is not in use
2045 */
2046 rcu_read_lock();
2047 if (rdev->desc_nr < 0) {
2048 int choice = 0;
2049 if (mddev->pers)
2050 choice = mddev->raid_disks;
2051 while (find_rdev_nr_rcu(mddev, choice))
2052 choice++;
2053 rdev->desc_nr = choice;
2054 } else {
2055 if (find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2056 rcu_read_unlock();
2057 return -EBUSY;
2058 }
2059 }
2060 rcu_read_unlock();
2061 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2062 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2063 mdname(mddev), mddev->max_disks);
2064 return -EBUSY;
2065 }
2066 bdevname(rdev->bdev,b);
2067 while ( (s=strchr(b, '/')) != NULL)
2068 *s = '!';
2069
2070 rdev->mddev = mddev;
2071 printk(KERN_INFO "md: bind<%s>\n", b);
2072
2073 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2074 goto fail;
2075
2076 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2077 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2078 /* failure here is OK */;
2079 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2080
2081 list_add_rcu(&rdev->same_set, &mddev->disks);
2082 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2083
2084 /* May as well allow recovery to be retried once */
2085 mddev->recovery_disabled++;
2086
2087 return 0;
2088
2089 fail:
2090 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2091 b, mdname(mddev));
2092 return err;
2093 }
2094
2095 static void md_delayed_delete(struct work_struct *ws)
2096 {
2097 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2098 kobject_del(&rdev->kobj);
2099 kobject_put(&rdev->kobj);
2100 }
2101
2102 static void unbind_rdev_from_array(struct md_rdev *rdev)
2103 {
2104 char b[BDEVNAME_SIZE];
2105
2106 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2107 list_del_rcu(&rdev->same_set);
2108 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2109 rdev->mddev = NULL;
2110 sysfs_remove_link(&rdev->kobj, "block");
2111 sysfs_put(rdev->sysfs_state);
2112 rdev->sysfs_state = NULL;
2113 rdev->badblocks.count = 0;
2114 /* We need to delay this, otherwise we can deadlock when
2115 * writing to 'remove' to "dev/state". We also need
2116 * to delay it due to rcu usage.
2117 */
2118 synchronize_rcu();
2119 INIT_WORK(&rdev->del_work, md_delayed_delete);
2120 kobject_get(&rdev->kobj);
2121 queue_work(md_misc_wq, &rdev->del_work);
2122 }
2123
2124 /*
2125 * prevent the device from being mounted, repartitioned or
2126 * otherwise reused by a RAID array (or any other kernel
2127 * subsystem), by bd_claiming the device.
2128 */
2129 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2130 {
2131 int err = 0;
2132 struct block_device *bdev;
2133 char b[BDEVNAME_SIZE];
2134
2135 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2136 shared ? (struct md_rdev *)lock_rdev : rdev);
2137 if (IS_ERR(bdev)) {
2138 printk(KERN_ERR "md: could not open %s.\n",
2139 __bdevname(dev, b));
2140 return PTR_ERR(bdev);
2141 }
2142 rdev->bdev = bdev;
2143 return err;
2144 }
2145
2146 static void unlock_rdev(struct md_rdev *rdev)
2147 {
2148 struct block_device *bdev = rdev->bdev;
2149 rdev->bdev = NULL;
2150 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2151 }
2152
2153 void md_autodetect_dev(dev_t dev);
2154
2155 static void export_rdev(struct md_rdev *rdev)
2156 {
2157 char b[BDEVNAME_SIZE];
2158
2159 printk(KERN_INFO "md: export_rdev(%s)\n",
2160 bdevname(rdev->bdev,b));
2161 md_rdev_clear(rdev);
2162 #ifndef MODULE
2163 if (test_bit(AutoDetected, &rdev->flags))
2164 md_autodetect_dev(rdev->bdev->bd_dev);
2165 #endif
2166 unlock_rdev(rdev);
2167 kobject_put(&rdev->kobj);
2168 }
2169
2170 static void kick_rdev_from_array(struct md_rdev *rdev)
2171 {
2172 unbind_rdev_from_array(rdev);
2173 export_rdev(rdev);
2174 }
2175
2176 static void export_array(struct mddev *mddev)
2177 {
2178 struct md_rdev *rdev;
2179
2180 while (!list_empty(&mddev->disks)) {
2181 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2182 same_set);
2183 kick_rdev_from_array(rdev);
2184 }
2185 mddev->raid_disks = 0;
2186 mddev->major_version = 0;
2187 }
2188
2189 static void sync_sbs(struct mddev *mddev, int nospares)
2190 {
2191 /* Update each superblock (in-memory image), but
2192 * if we are allowed to, skip spares which already
2193 * have the right event counter, or have one earlier
2194 * (which would mean they aren't being marked as dirty
2195 * with the rest of the array)
2196 */
2197 struct md_rdev *rdev;
2198 rdev_for_each(rdev, mddev) {
2199 if (rdev->sb_events == mddev->events ||
2200 (nospares &&
2201 rdev->raid_disk < 0 &&
2202 rdev->sb_events+1 == mddev->events)) {
2203 /* Don't update this superblock */
2204 rdev->sb_loaded = 2;
2205 } else {
2206 sync_super(mddev, rdev);
2207 rdev->sb_loaded = 1;
2208 }
2209 }
2210 }
2211
2212 static void md_update_sb(struct mddev *mddev, int force_change)
2213 {
2214 struct md_rdev *rdev;
2215 int sync_req;
2216 int nospares = 0;
2217 int any_badblocks_changed = 0;
2218
2219 if (mddev->ro) {
2220 if (force_change)
2221 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2222 return;
2223 }
2224 repeat:
2225 /* First make sure individual recovery_offsets are correct */
2226 rdev_for_each(rdev, mddev) {
2227 if (rdev->raid_disk >= 0 &&
2228 mddev->delta_disks >= 0 &&
2229 !test_bit(In_sync, &rdev->flags) &&
2230 mddev->curr_resync_completed > rdev->recovery_offset)
2231 rdev->recovery_offset = mddev->curr_resync_completed;
2232
2233 }
2234 if (!mddev->persistent) {
2235 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2236 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2237 if (!mddev->external) {
2238 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2239 rdev_for_each(rdev, mddev) {
2240 if (rdev->badblocks.changed) {
2241 rdev->badblocks.changed = 0;
2242 md_ack_all_badblocks(&rdev->badblocks);
2243 md_error(mddev, rdev);
2244 }
2245 clear_bit(Blocked, &rdev->flags);
2246 clear_bit(BlockedBadBlocks, &rdev->flags);
2247 wake_up(&rdev->blocked_wait);
2248 }
2249 }
2250 wake_up(&mddev->sb_wait);
2251 return;
2252 }
2253
2254 spin_lock(&mddev->lock);
2255
2256 mddev->utime = get_seconds();
2257
2258 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2259 force_change = 1;
2260 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2261 /* just a clean<-> dirty transition, possibly leave spares alone,
2262 * though if events isn't the right even/odd, we will have to do
2263 * spares after all
2264 */
2265 nospares = 1;
2266 if (force_change)
2267 nospares = 0;
2268 if (mddev->degraded)
2269 /* If the array is degraded, then skipping spares is both
2270 * dangerous and fairly pointless.
2271 * Dangerous because a device that was removed from the array
2272 * might have a event_count that still looks up-to-date,
2273 * so it can be re-added without a resync.
2274 * Pointless because if there are any spares to skip,
2275 * then a recovery will happen and soon that array won't
2276 * be degraded any more and the spare can go back to sleep then.
2277 */
2278 nospares = 0;
2279
2280 sync_req = mddev->in_sync;
2281
2282 /* If this is just a dirty<->clean transition, and the array is clean
2283 * and 'events' is odd, we can roll back to the previous clean state */
2284 if (nospares
2285 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2286 && mddev->can_decrease_events
2287 && mddev->events != 1) {
2288 mddev->events--;
2289 mddev->can_decrease_events = 0;
2290 } else {
2291 /* otherwise we have to go forward and ... */
2292 mddev->events ++;
2293 mddev->can_decrease_events = nospares;
2294 }
2295
2296 /*
2297 * This 64-bit counter should never wrap.
2298 * Either we are in around ~1 trillion A.C., assuming
2299 * 1 reboot per second, or we have a bug...
2300 */
2301 WARN_ON(mddev->events == 0);
2302
2303 rdev_for_each(rdev, mddev) {
2304 if (rdev->badblocks.changed)
2305 any_badblocks_changed++;
2306 if (test_bit(Faulty, &rdev->flags))
2307 set_bit(FaultRecorded, &rdev->flags);
2308 }
2309
2310 sync_sbs(mddev, nospares);
2311 spin_unlock(&mddev->lock);
2312
2313 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2314 mdname(mddev), mddev->in_sync);
2315
2316 bitmap_update_sb(mddev->bitmap);
2317 rdev_for_each(rdev, mddev) {
2318 char b[BDEVNAME_SIZE];
2319
2320 if (rdev->sb_loaded != 1)
2321 continue; /* no noise on spare devices */
2322
2323 if (!test_bit(Faulty, &rdev->flags)) {
2324 md_super_write(mddev,rdev,
2325 rdev->sb_start, rdev->sb_size,
2326 rdev->sb_page);
2327 pr_debug("md: (write) %s's sb offset: %llu\n",
2328 bdevname(rdev->bdev, b),
2329 (unsigned long long)rdev->sb_start);
2330 rdev->sb_events = mddev->events;
2331 if (rdev->badblocks.size) {
2332 md_super_write(mddev, rdev,
2333 rdev->badblocks.sector,
2334 rdev->badblocks.size << 9,
2335 rdev->bb_page);
2336 rdev->badblocks.size = 0;
2337 }
2338
2339 } else
2340 pr_debug("md: %s (skipping faulty)\n",
2341 bdevname(rdev->bdev, b));
2342
2343 if (mddev->level == LEVEL_MULTIPATH)
2344 /* only need to write one superblock... */
2345 break;
2346 }
2347 md_super_wait(mddev);
2348 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2349
2350 spin_lock(&mddev->lock);
2351 if (mddev->in_sync != sync_req ||
2352 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2353 /* have to write it out again */
2354 spin_unlock(&mddev->lock);
2355 goto repeat;
2356 }
2357 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2358 spin_unlock(&mddev->lock);
2359 wake_up(&mddev->sb_wait);
2360 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2361 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2362
2363 rdev_for_each(rdev, mddev) {
2364 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2365 clear_bit(Blocked, &rdev->flags);
2366
2367 if (any_badblocks_changed)
2368 md_ack_all_badblocks(&rdev->badblocks);
2369 clear_bit(BlockedBadBlocks, &rdev->flags);
2370 wake_up(&rdev->blocked_wait);
2371 }
2372 }
2373
2374 /* words written to sysfs files may, or may not, be \n terminated.
2375 * We want to accept with case. For this we use cmd_match.
2376 */
2377 static int cmd_match(const char *cmd, const char *str)
2378 {
2379 /* See if cmd, written into a sysfs file, matches
2380 * str. They must either be the same, or cmd can
2381 * have a trailing newline
2382 */
2383 while (*cmd && *str && *cmd == *str) {
2384 cmd++;
2385 str++;
2386 }
2387 if (*cmd == '\n')
2388 cmd++;
2389 if (*str || *cmd)
2390 return 0;
2391 return 1;
2392 }
2393
2394 struct rdev_sysfs_entry {
2395 struct attribute attr;
2396 ssize_t (*show)(struct md_rdev *, char *);
2397 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2398 };
2399
2400 static ssize_t
2401 state_show(struct md_rdev *rdev, char *page)
2402 {
2403 char *sep = "";
2404 size_t len = 0;
2405 unsigned long flags = ACCESS_ONCE(rdev->flags);
2406
2407 if (test_bit(Faulty, &flags) ||
2408 rdev->badblocks.unacked_exist) {
2409 len+= sprintf(page+len, "%sfaulty",sep);
2410 sep = ",";
2411 }
2412 if (test_bit(In_sync, &flags)) {
2413 len += sprintf(page+len, "%sin_sync",sep);
2414 sep = ",";
2415 }
2416 if (test_bit(WriteMostly, &flags)) {
2417 len += sprintf(page+len, "%swrite_mostly",sep);
2418 sep = ",";
2419 }
2420 if (test_bit(Blocked, &flags) ||
2421 (rdev->badblocks.unacked_exist
2422 && !test_bit(Faulty, &flags))) {
2423 len += sprintf(page+len, "%sblocked", sep);
2424 sep = ",";
2425 }
2426 if (!test_bit(Faulty, &flags) &&
2427 !test_bit(In_sync, &flags)) {
2428 len += sprintf(page+len, "%sspare", sep);
2429 sep = ",";
2430 }
2431 if (test_bit(WriteErrorSeen, &flags)) {
2432 len += sprintf(page+len, "%swrite_error", sep);
2433 sep = ",";
2434 }
2435 if (test_bit(WantReplacement, &flags)) {
2436 len += sprintf(page+len, "%swant_replacement", sep);
2437 sep = ",";
2438 }
2439 if (test_bit(Replacement, &flags)) {
2440 len += sprintf(page+len, "%sreplacement", sep);
2441 sep = ",";
2442 }
2443
2444 return len+sprintf(page+len, "\n");
2445 }
2446
2447 static ssize_t
2448 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2449 {
2450 /* can write
2451 * faulty - simulates an error
2452 * remove - disconnects the device
2453 * writemostly - sets write_mostly
2454 * -writemostly - clears write_mostly
2455 * blocked - sets the Blocked flags
2456 * -blocked - clears the Blocked and possibly simulates an error
2457 * insync - sets Insync providing device isn't active
2458 * -insync - clear Insync for a device with a slot assigned,
2459 * so that it gets rebuilt based on bitmap
2460 * write_error - sets WriteErrorSeen
2461 * -write_error - clears WriteErrorSeen
2462 */
2463 int err = -EINVAL;
2464 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2465 md_error(rdev->mddev, rdev);
2466 if (test_bit(Faulty, &rdev->flags))
2467 err = 0;
2468 else
2469 err = -EBUSY;
2470 } else if (cmd_match(buf, "remove")) {
2471 if (rdev->raid_disk >= 0)
2472 err = -EBUSY;
2473 else {
2474 struct mddev *mddev = rdev->mddev;
2475 kick_rdev_from_array(rdev);
2476 if (mddev->pers)
2477 md_update_sb(mddev, 1);
2478 md_new_event(mddev);
2479 err = 0;
2480 }
2481 } else if (cmd_match(buf, "writemostly")) {
2482 set_bit(WriteMostly, &rdev->flags);
2483 err = 0;
2484 } else if (cmd_match(buf, "-writemostly")) {
2485 clear_bit(WriteMostly, &rdev->flags);
2486 err = 0;
2487 } else if (cmd_match(buf, "blocked")) {
2488 set_bit(Blocked, &rdev->flags);
2489 err = 0;
2490 } else if (cmd_match(buf, "-blocked")) {
2491 if (!test_bit(Faulty, &rdev->flags) &&
2492 rdev->badblocks.unacked_exist) {
2493 /* metadata handler doesn't understand badblocks,
2494 * so we need to fail the device
2495 */
2496 md_error(rdev->mddev, rdev);
2497 }
2498 clear_bit(Blocked, &rdev->flags);
2499 clear_bit(BlockedBadBlocks, &rdev->flags);
2500 wake_up(&rdev->blocked_wait);
2501 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2502 md_wakeup_thread(rdev->mddev->thread);
2503
2504 err = 0;
2505 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2506 set_bit(In_sync, &rdev->flags);
2507 err = 0;
2508 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0) {
2509 if (rdev->mddev->pers == NULL) {
2510 clear_bit(In_sync, &rdev->flags);
2511 rdev->saved_raid_disk = rdev->raid_disk;
2512 rdev->raid_disk = -1;
2513 err = 0;
2514 }
2515 } else if (cmd_match(buf, "write_error")) {
2516 set_bit(WriteErrorSeen, &rdev->flags);
2517 err = 0;
2518 } else if (cmd_match(buf, "-write_error")) {
2519 clear_bit(WriteErrorSeen, &rdev->flags);
2520 err = 0;
2521 } else if (cmd_match(buf, "want_replacement")) {
2522 /* Any non-spare device that is not a replacement can
2523 * become want_replacement at any time, but we then need to
2524 * check if recovery is needed.
2525 */
2526 if (rdev->raid_disk >= 0 &&
2527 !test_bit(Replacement, &rdev->flags))
2528 set_bit(WantReplacement, &rdev->flags);
2529 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2530 md_wakeup_thread(rdev->mddev->thread);
2531 err = 0;
2532 } else if (cmd_match(buf, "-want_replacement")) {
2533 /* Clearing 'want_replacement' is always allowed.
2534 * Once replacements starts it is too late though.
2535 */
2536 err = 0;
2537 clear_bit(WantReplacement, &rdev->flags);
2538 } else if (cmd_match(buf, "replacement")) {
2539 /* Can only set a device as a replacement when array has not
2540 * yet been started. Once running, replacement is automatic
2541 * from spares, or by assigning 'slot'.
2542 */
2543 if (rdev->mddev->pers)
2544 err = -EBUSY;
2545 else {
2546 set_bit(Replacement, &rdev->flags);
2547 err = 0;
2548 }
2549 } else if (cmd_match(buf, "-replacement")) {
2550 /* Similarly, can only clear Replacement before start */
2551 if (rdev->mddev->pers)
2552 err = -EBUSY;
2553 else {
2554 clear_bit(Replacement, &rdev->flags);
2555 err = 0;
2556 }
2557 }
2558 if (!err)
2559 sysfs_notify_dirent_safe(rdev->sysfs_state);
2560 return err ? err : len;
2561 }
2562 static struct rdev_sysfs_entry rdev_state =
2563 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2564
2565 static ssize_t
2566 errors_show(struct md_rdev *rdev, char *page)
2567 {
2568 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2569 }
2570
2571 static ssize_t
2572 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2573 {
2574 char *e;
2575 unsigned long n = simple_strtoul(buf, &e, 10);
2576 if (*buf && (*e == 0 || *e == '\n')) {
2577 atomic_set(&rdev->corrected_errors, n);
2578 return len;
2579 }
2580 return -EINVAL;
2581 }
2582 static struct rdev_sysfs_entry rdev_errors =
2583 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2584
2585 static ssize_t
2586 slot_show(struct md_rdev *rdev, char *page)
2587 {
2588 if (rdev->raid_disk < 0)
2589 return sprintf(page, "none\n");
2590 else
2591 return sprintf(page, "%d\n", rdev->raid_disk);
2592 }
2593
2594 static ssize_t
2595 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2596 {
2597 char *e;
2598 int err;
2599 int slot = simple_strtoul(buf, &e, 10);
2600 if (strncmp(buf, "none", 4)==0)
2601 slot = -1;
2602 else if (e==buf || (*e && *e!= '\n'))
2603 return -EINVAL;
2604 if (rdev->mddev->pers && slot == -1) {
2605 /* Setting 'slot' on an active array requires also
2606 * updating the 'rd%d' link, and communicating
2607 * with the personality with ->hot_*_disk.
2608 * For now we only support removing
2609 * failed/spare devices. This normally happens automatically,
2610 * but not when the metadata is externally managed.
2611 */
2612 if (rdev->raid_disk == -1)
2613 return -EEXIST;
2614 /* personality does all needed checks */
2615 if (rdev->mddev->pers->hot_remove_disk == NULL)
2616 return -EINVAL;
2617 clear_bit(Blocked, &rdev->flags);
2618 remove_and_add_spares(rdev->mddev, rdev);
2619 if (rdev->raid_disk >= 0)
2620 return -EBUSY;
2621 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2622 md_wakeup_thread(rdev->mddev->thread);
2623 } else if (rdev->mddev->pers) {
2624 /* Activating a spare .. or possibly reactivating
2625 * if we ever get bitmaps working here.
2626 */
2627
2628 if (rdev->raid_disk != -1)
2629 return -EBUSY;
2630
2631 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2632 return -EBUSY;
2633
2634 if (rdev->mddev->pers->hot_add_disk == NULL)
2635 return -EINVAL;
2636
2637 if (slot >= rdev->mddev->raid_disks &&
2638 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2639 return -ENOSPC;
2640
2641 rdev->raid_disk = slot;
2642 if (test_bit(In_sync, &rdev->flags))
2643 rdev->saved_raid_disk = slot;
2644 else
2645 rdev->saved_raid_disk = -1;
2646 clear_bit(In_sync, &rdev->flags);
2647 clear_bit(Bitmap_sync, &rdev->flags);
2648 err = rdev->mddev->pers->
2649 hot_add_disk(rdev->mddev, rdev);
2650 if (err) {
2651 rdev->raid_disk = -1;
2652 return err;
2653 } else
2654 sysfs_notify_dirent_safe(rdev->sysfs_state);
2655 if (sysfs_link_rdev(rdev->mddev, rdev))
2656 /* failure here is OK */;
2657 /* don't wakeup anyone, leave that to userspace. */
2658 } else {
2659 if (slot >= rdev->mddev->raid_disks &&
2660 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2661 return -ENOSPC;
2662 rdev->raid_disk = slot;
2663 /* assume it is working */
2664 clear_bit(Faulty, &rdev->flags);
2665 clear_bit(WriteMostly, &rdev->flags);
2666 set_bit(In_sync, &rdev->flags);
2667 sysfs_notify_dirent_safe(rdev->sysfs_state);
2668 }
2669 return len;
2670 }
2671
2672 static struct rdev_sysfs_entry rdev_slot =
2673 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2674
2675 static ssize_t
2676 offset_show(struct md_rdev *rdev, char *page)
2677 {
2678 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2679 }
2680
2681 static ssize_t
2682 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2683 {
2684 unsigned long long offset;
2685 if (kstrtoull(buf, 10, &offset) < 0)
2686 return -EINVAL;
2687 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2688 return -EBUSY;
2689 if (rdev->sectors && rdev->mddev->external)
2690 /* Must set offset before size, so overlap checks
2691 * can be sane */
2692 return -EBUSY;
2693 rdev->data_offset = offset;
2694 rdev->new_data_offset = offset;
2695 return len;
2696 }
2697
2698 static struct rdev_sysfs_entry rdev_offset =
2699 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2700
2701 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2702 {
2703 return sprintf(page, "%llu\n",
2704 (unsigned long long)rdev->new_data_offset);
2705 }
2706
2707 static ssize_t new_offset_store(struct md_rdev *rdev,
2708 const char *buf, size_t len)
2709 {
2710 unsigned long long new_offset;
2711 struct mddev *mddev = rdev->mddev;
2712
2713 if (kstrtoull(buf, 10, &new_offset) < 0)
2714 return -EINVAL;
2715
2716 if (mddev->sync_thread ||
2717 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
2718 return -EBUSY;
2719 if (new_offset == rdev->data_offset)
2720 /* reset is always permitted */
2721 ;
2722 else if (new_offset > rdev->data_offset) {
2723 /* must not push array size beyond rdev_sectors */
2724 if (new_offset - rdev->data_offset
2725 + mddev->dev_sectors > rdev->sectors)
2726 return -E2BIG;
2727 }
2728 /* Metadata worries about other space details. */
2729
2730 /* decreasing the offset is inconsistent with a backwards
2731 * reshape.
2732 */
2733 if (new_offset < rdev->data_offset &&
2734 mddev->reshape_backwards)
2735 return -EINVAL;
2736 /* Increasing offset is inconsistent with forwards
2737 * reshape. reshape_direction should be set to
2738 * 'backwards' first.
2739 */
2740 if (new_offset > rdev->data_offset &&
2741 !mddev->reshape_backwards)
2742 return -EINVAL;
2743
2744 if (mddev->pers && mddev->persistent &&
2745 !super_types[mddev->major_version]
2746 .allow_new_offset(rdev, new_offset))
2747 return -E2BIG;
2748 rdev->new_data_offset = new_offset;
2749 if (new_offset > rdev->data_offset)
2750 mddev->reshape_backwards = 1;
2751 else if (new_offset < rdev->data_offset)
2752 mddev->reshape_backwards = 0;
2753
2754 return len;
2755 }
2756 static struct rdev_sysfs_entry rdev_new_offset =
2757 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2758
2759 static ssize_t
2760 rdev_size_show(struct md_rdev *rdev, char *page)
2761 {
2762 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2763 }
2764
2765 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2766 {
2767 /* check if two start/length pairs overlap */
2768 if (s1+l1 <= s2)
2769 return 0;
2770 if (s2+l2 <= s1)
2771 return 0;
2772 return 1;
2773 }
2774
2775 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2776 {
2777 unsigned long long blocks;
2778 sector_t new;
2779
2780 if (kstrtoull(buf, 10, &blocks) < 0)
2781 return -EINVAL;
2782
2783 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2784 return -EINVAL; /* sector conversion overflow */
2785
2786 new = blocks * 2;
2787 if (new != blocks * 2)
2788 return -EINVAL; /* unsigned long long to sector_t overflow */
2789
2790 *sectors = new;
2791 return 0;
2792 }
2793
2794 static ssize_t
2795 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
2796 {
2797 struct mddev *my_mddev = rdev->mddev;
2798 sector_t oldsectors = rdev->sectors;
2799 sector_t sectors;
2800
2801 if (strict_blocks_to_sectors(buf, &sectors) < 0)
2802 return -EINVAL;
2803 if (rdev->data_offset != rdev->new_data_offset)
2804 return -EINVAL; /* too confusing */
2805 if (my_mddev->pers && rdev->raid_disk >= 0) {
2806 if (my_mddev->persistent) {
2807 sectors = super_types[my_mddev->major_version].
2808 rdev_size_change(rdev, sectors);
2809 if (!sectors)
2810 return -EBUSY;
2811 } else if (!sectors)
2812 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2813 rdev->data_offset;
2814 if (!my_mddev->pers->resize)
2815 /* Cannot change size for RAID0 or Linear etc */
2816 return -EINVAL;
2817 }
2818 if (sectors < my_mddev->dev_sectors)
2819 return -EINVAL; /* component must fit device */
2820
2821 rdev->sectors = sectors;
2822 if (sectors > oldsectors && my_mddev->external) {
2823 /* Need to check that all other rdevs with the same
2824 * ->bdev do not overlap. 'rcu' is sufficient to walk
2825 * the rdev lists safely.
2826 * This check does not provide a hard guarantee, it
2827 * just helps avoid dangerous mistakes.
2828 */
2829 struct mddev *mddev;
2830 int overlap = 0;
2831 struct list_head *tmp;
2832
2833 rcu_read_lock();
2834 for_each_mddev(mddev, tmp) {
2835 struct md_rdev *rdev2;
2836
2837 rdev_for_each(rdev2, mddev)
2838 if (rdev->bdev == rdev2->bdev &&
2839 rdev != rdev2 &&
2840 overlaps(rdev->data_offset, rdev->sectors,
2841 rdev2->data_offset,
2842 rdev2->sectors)) {
2843 overlap = 1;
2844 break;
2845 }
2846 if (overlap) {
2847 mddev_put(mddev);
2848 break;
2849 }
2850 }
2851 rcu_read_unlock();
2852 if (overlap) {
2853 /* Someone else could have slipped in a size
2854 * change here, but doing so is just silly.
2855 * We put oldsectors back because we *know* it is
2856 * safe, and trust userspace not to race with
2857 * itself
2858 */
2859 rdev->sectors = oldsectors;
2860 return -EBUSY;
2861 }
2862 }
2863 return len;
2864 }
2865
2866 static struct rdev_sysfs_entry rdev_size =
2867 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2868
2869 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
2870 {
2871 unsigned long long recovery_start = rdev->recovery_offset;
2872
2873 if (test_bit(In_sync, &rdev->flags) ||
2874 recovery_start == MaxSector)
2875 return sprintf(page, "none\n");
2876
2877 return sprintf(page, "%llu\n", recovery_start);
2878 }
2879
2880 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
2881 {
2882 unsigned long long recovery_start;
2883
2884 if (cmd_match(buf, "none"))
2885 recovery_start = MaxSector;
2886 else if (kstrtoull(buf, 10, &recovery_start))
2887 return -EINVAL;
2888
2889 if (rdev->mddev->pers &&
2890 rdev->raid_disk >= 0)
2891 return -EBUSY;
2892
2893 rdev->recovery_offset = recovery_start;
2894 if (recovery_start == MaxSector)
2895 set_bit(In_sync, &rdev->flags);
2896 else
2897 clear_bit(In_sync, &rdev->flags);
2898 return len;
2899 }
2900
2901 static struct rdev_sysfs_entry rdev_recovery_start =
2902 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2903
2904 static ssize_t
2905 badblocks_show(struct badblocks *bb, char *page, int unack);
2906 static ssize_t
2907 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
2908
2909 static ssize_t bb_show(struct md_rdev *rdev, char *page)
2910 {
2911 return badblocks_show(&rdev->badblocks, page, 0);
2912 }
2913 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
2914 {
2915 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
2916 /* Maybe that ack was all we needed */
2917 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
2918 wake_up(&rdev->blocked_wait);
2919 return rv;
2920 }
2921 static struct rdev_sysfs_entry rdev_bad_blocks =
2922 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
2923
2924 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
2925 {
2926 return badblocks_show(&rdev->badblocks, page, 1);
2927 }
2928 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
2929 {
2930 return badblocks_store(&rdev->badblocks, page, len, 1);
2931 }
2932 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
2933 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
2934
2935 static struct attribute *rdev_default_attrs[] = {
2936 &rdev_state.attr,
2937 &rdev_errors.attr,
2938 &rdev_slot.attr,
2939 &rdev_offset.attr,
2940 &rdev_new_offset.attr,
2941 &rdev_size.attr,
2942 &rdev_recovery_start.attr,
2943 &rdev_bad_blocks.attr,
2944 &rdev_unack_bad_blocks.attr,
2945 NULL,
2946 };
2947 static ssize_t
2948 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2949 {
2950 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2951 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
2952
2953 if (!entry->show)
2954 return -EIO;
2955 if (!rdev->mddev)
2956 return -EBUSY;
2957 return entry->show(rdev, page);
2958 }
2959
2960 static ssize_t
2961 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2962 const char *page, size_t length)
2963 {
2964 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2965 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
2966 ssize_t rv;
2967 struct mddev *mddev = rdev->mddev;
2968
2969 if (!entry->store)
2970 return -EIO;
2971 if (!capable(CAP_SYS_ADMIN))
2972 return -EACCES;
2973 rv = mddev ? mddev_lock(mddev): -EBUSY;
2974 if (!rv) {
2975 if (rdev->mddev == NULL)
2976 rv = -EBUSY;
2977 else
2978 rv = entry->store(rdev, page, length);
2979 mddev_unlock(mddev);
2980 }
2981 return rv;
2982 }
2983
2984 static void rdev_free(struct kobject *ko)
2985 {
2986 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
2987 kfree(rdev);
2988 }
2989 static const struct sysfs_ops rdev_sysfs_ops = {
2990 .show = rdev_attr_show,
2991 .store = rdev_attr_store,
2992 };
2993 static struct kobj_type rdev_ktype = {
2994 .release = rdev_free,
2995 .sysfs_ops = &rdev_sysfs_ops,
2996 .default_attrs = rdev_default_attrs,
2997 };
2998
2999 int md_rdev_init(struct md_rdev *rdev)
3000 {
3001 rdev->desc_nr = -1;
3002 rdev->saved_raid_disk = -1;
3003 rdev->raid_disk = -1;
3004 rdev->flags = 0;
3005 rdev->data_offset = 0;
3006 rdev->new_data_offset = 0;
3007 rdev->sb_events = 0;
3008 rdev->last_read_error.tv_sec = 0;
3009 rdev->last_read_error.tv_nsec = 0;
3010 rdev->sb_loaded = 0;
3011 rdev->bb_page = NULL;
3012 atomic_set(&rdev->nr_pending, 0);
3013 atomic_set(&rdev->read_errors, 0);
3014 atomic_set(&rdev->corrected_errors, 0);
3015
3016 INIT_LIST_HEAD(&rdev->same_set);
3017 init_waitqueue_head(&rdev->blocked_wait);
3018
3019 /* Add space to store bad block list.
3020 * This reserves the space even on arrays where it cannot
3021 * be used - I wonder if that matters
3022 */
3023 rdev->badblocks.count = 0;
3024 rdev->badblocks.shift = -1; /* disabled until explicitly enabled */
3025 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3026 seqlock_init(&rdev->badblocks.lock);
3027 if (rdev->badblocks.page == NULL)
3028 return -ENOMEM;
3029
3030 return 0;
3031 }
3032 EXPORT_SYMBOL_GPL(md_rdev_init);
3033 /*
3034 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3035 *
3036 * mark the device faulty if:
3037 *
3038 * - the device is nonexistent (zero size)
3039 * - the device has no valid superblock
3040 *
3041 * a faulty rdev _never_ has rdev->sb set.
3042 */
3043 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3044 {
3045 char b[BDEVNAME_SIZE];
3046 int err;
3047 struct md_rdev *rdev;
3048 sector_t size;
3049
3050 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3051 if (!rdev) {
3052 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3053 return ERR_PTR(-ENOMEM);
3054 }
3055
3056 err = md_rdev_init(rdev);
3057 if (err)
3058 goto abort_free;
3059 err = alloc_disk_sb(rdev);
3060 if (err)
3061 goto abort_free;
3062
3063 err = lock_rdev(rdev, newdev, super_format == -2);
3064 if (err)
3065 goto abort_free;
3066
3067 kobject_init(&rdev->kobj, &rdev_ktype);
3068
3069 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3070 if (!size) {
3071 printk(KERN_WARNING
3072 "md: %s has zero or unknown size, marking faulty!\n",
3073 bdevname(rdev->bdev,b));
3074 err = -EINVAL;
3075 goto abort_free;
3076 }
3077
3078 if (super_format >= 0) {
3079 err = super_types[super_format].
3080 load_super(rdev, NULL, super_minor);
3081 if (err == -EINVAL) {
3082 printk(KERN_WARNING
3083 "md: %s does not have a valid v%d.%d "
3084 "superblock, not importing!\n",
3085 bdevname(rdev->bdev,b),
3086 super_format, super_minor);
3087 goto abort_free;
3088 }
3089 if (err < 0) {
3090 printk(KERN_WARNING
3091 "md: could not read %s's sb, not importing!\n",
3092 bdevname(rdev->bdev,b));
3093 goto abort_free;
3094 }
3095 }
3096
3097 return rdev;
3098
3099 abort_free:
3100 if (rdev->bdev)
3101 unlock_rdev(rdev);
3102 md_rdev_clear(rdev);
3103 kfree(rdev);
3104 return ERR_PTR(err);
3105 }
3106
3107 /*
3108 * Check a full RAID array for plausibility
3109 */
3110
3111 static void analyze_sbs(struct mddev *mddev)
3112 {
3113 int i;
3114 struct md_rdev *rdev, *freshest, *tmp;
3115 char b[BDEVNAME_SIZE];
3116
3117 freshest = NULL;
3118 rdev_for_each_safe(rdev, tmp, mddev)
3119 switch (super_types[mddev->major_version].
3120 load_super(rdev, freshest, mddev->minor_version)) {
3121 case 1:
3122 freshest = rdev;
3123 break;
3124 case 0:
3125 break;
3126 default:
3127 printk( KERN_ERR \
3128 "md: fatal superblock inconsistency in %s"
3129 " -- removing from array\n",
3130 bdevname(rdev->bdev,b));
3131 kick_rdev_from_array(rdev);
3132 }
3133
3134 super_types[mddev->major_version].
3135 validate_super(mddev, freshest);
3136
3137 i = 0;
3138 rdev_for_each_safe(rdev, tmp, mddev) {
3139 if (mddev->max_disks &&
3140 (rdev->desc_nr >= mddev->max_disks ||
3141 i > mddev->max_disks)) {
3142 printk(KERN_WARNING
3143 "md: %s: %s: only %d devices permitted\n",
3144 mdname(mddev), bdevname(rdev->bdev, b),
3145 mddev->max_disks);
3146 kick_rdev_from_array(rdev);
3147 continue;
3148 }
3149 if (rdev != freshest)
3150 if (super_types[mddev->major_version].
3151 validate_super(mddev, rdev)) {
3152 printk(KERN_WARNING "md: kicking non-fresh %s"
3153 " from array!\n",
3154 bdevname(rdev->bdev,b));
3155 kick_rdev_from_array(rdev);
3156 continue;
3157 }
3158 if (mddev->level == LEVEL_MULTIPATH) {
3159 rdev->desc_nr = i++;
3160 rdev->raid_disk = rdev->desc_nr;
3161 set_bit(In_sync, &rdev->flags);
3162 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
3163 rdev->raid_disk = -1;
3164 clear_bit(In_sync, &rdev->flags);
3165 }
3166 }
3167 }
3168
3169 /* Read a fixed-point number.
3170 * Numbers in sysfs attributes should be in "standard" units where
3171 * possible, so time should be in seconds.
3172 * However we internally use a a much smaller unit such as
3173 * milliseconds or jiffies.
3174 * This function takes a decimal number with a possible fractional
3175 * component, and produces an integer which is the result of
3176 * multiplying that number by 10^'scale'.
3177 * all without any floating-point arithmetic.
3178 */
3179 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3180 {
3181 unsigned long result = 0;
3182 long decimals = -1;
3183 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3184 if (*cp == '.')
3185 decimals = 0;
3186 else if (decimals < scale) {
3187 unsigned int value;
3188 value = *cp - '0';
3189 result = result * 10 + value;
3190 if (decimals >= 0)
3191 decimals++;
3192 }
3193 cp++;
3194 }
3195 if (*cp == '\n')
3196 cp++;
3197 if (*cp)
3198 return -EINVAL;
3199 if (decimals < 0)
3200 decimals = 0;
3201 while (decimals < scale) {
3202 result *= 10;
3203 decimals ++;
3204 }
3205 *res = result;
3206 return 0;
3207 }
3208
3209 static void md_safemode_timeout(unsigned long data);
3210
3211 static ssize_t
3212 safe_delay_show(struct mddev *mddev, char *page)
3213 {
3214 int msec = (mddev->safemode_delay*1000)/HZ;
3215 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3216 }
3217 static ssize_t
3218 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3219 {
3220 unsigned long msec;
3221
3222 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3223 return -EINVAL;
3224 if (msec == 0)
3225 mddev->safemode_delay = 0;
3226 else {
3227 unsigned long old_delay = mddev->safemode_delay;
3228 unsigned long new_delay = (msec*HZ)/1000;
3229
3230 if (new_delay == 0)
3231 new_delay = 1;
3232 mddev->safemode_delay = new_delay;
3233 if (new_delay < old_delay || old_delay == 0)
3234 mod_timer(&mddev->safemode_timer, jiffies+1);
3235 }
3236 return len;
3237 }
3238 static struct md_sysfs_entry md_safe_delay =
3239 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3240
3241 static ssize_t
3242 level_show(struct mddev *mddev, char *page)
3243 {
3244 struct md_personality *p;
3245 int ret;
3246 spin_lock(&mddev->lock);
3247 p = mddev->pers;
3248 if (p)
3249 ret = sprintf(page, "%s\n", p->name);
3250 else if (mddev->clevel[0])
3251 ret = sprintf(page, "%s\n", mddev->clevel);
3252 else if (mddev->level != LEVEL_NONE)
3253 ret = sprintf(page, "%d\n", mddev->level);
3254 else
3255 ret = 0;
3256 spin_unlock(&mddev->lock);
3257 return ret;
3258 }
3259
3260 static ssize_t
3261 level_store(struct mddev *mddev, const char *buf, size_t len)
3262 {
3263 char clevel[16];
3264 ssize_t rv;
3265 size_t slen = len;
3266 struct md_personality *pers, *oldpers;
3267 long level;
3268 void *priv, *oldpriv;
3269 struct md_rdev *rdev;
3270
3271 if (slen == 0 || slen >= sizeof(clevel))
3272 return -EINVAL;
3273
3274 rv = mddev_lock(mddev);
3275 if (rv)
3276 return rv;
3277
3278 if (mddev->pers == NULL) {
3279 strncpy(mddev->clevel, buf, slen);
3280 if (mddev->clevel[slen-1] == '\n')
3281 slen--;
3282 mddev->clevel[slen] = 0;
3283 mddev->level = LEVEL_NONE;
3284 rv = len;
3285 goto out_unlock;
3286 }
3287 rv = -EROFS;
3288 if (mddev->ro)
3289 goto out_unlock;
3290
3291 /* request to change the personality. Need to ensure:
3292 * - array is not engaged in resync/recovery/reshape
3293 * - old personality can be suspended
3294 * - new personality will access other array.
3295 */
3296
3297 rv = -EBUSY;
3298 if (mddev->sync_thread ||
3299 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3300 mddev->reshape_position != MaxSector ||
3301 mddev->sysfs_active)
3302 goto out_unlock;
3303
3304 rv = -EINVAL;
3305 if (!mddev->pers->quiesce) {
3306 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3307 mdname(mddev), mddev->pers->name);
3308 goto out_unlock;
3309 }
3310
3311 /* Now find the new personality */
3312 strncpy(clevel, buf, slen);
3313 if (clevel[slen-1] == '\n')
3314 slen--;
3315 clevel[slen] = 0;
3316 if (kstrtol(clevel, 10, &level))
3317 level = LEVEL_NONE;
3318
3319 if (request_module("md-%s", clevel) != 0)
3320 request_module("md-level-%s", clevel);
3321 spin_lock(&pers_lock);
3322 pers = find_pers(level, clevel);
3323 if (!pers || !try_module_get(pers->owner)) {
3324 spin_unlock(&pers_lock);
3325 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3326 rv = -EINVAL;
3327 goto out_unlock;
3328 }
3329 spin_unlock(&pers_lock);
3330
3331 if (pers == mddev->pers) {
3332 /* Nothing to do! */
3333 module_put(pers->owner);
3334 rv = len;
3335 goto out_unlock;
3336 }
3337 if (!pers->takeover) {
3338 module_put(pers->owner);
3339 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3340 mdname(mddev), clevel);
3341 rv = -EINVAL;
3342 goto out_unlock;
3343 }
3344
3345 rdev_for_each(rdev, mddev)
3346 rdev->new_raid_disk = rdev->raid_disk;
3347
3348 /* ->takeover must set new_* and/or delta_disks
3349 * if it succeeds, and may set them when it fails.
3350 */
3351 priv = pers->takeover(mddev);
3352 if (IS_ERR(priv)) {
3353 mddev->new_level = mddev->level;
3354 mddev->new_layout = mddev->layout;
3355 mddev->new_chunk_sectors = mddev->chunk_sectors;
3356 mddev->raid_disks -= mddev->delta_disks;
3357 mddev->delta_disks = 0;
3358 mddev->reshape_backwards = 0;
3359 module_put(pers->owner);
3360 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3361 mdname(mddev), clevel);
3362 rv = PTR_ERR(priv);
3363 goto out_unlock;
3364 }
3365
3366 /* Looks like we have a winner */
3367 mddev_suspend(mddev);
3368 mddev_detach(mddev);
3369
3370 spin_lock(&mddev->lock);
3371 oldpers = mddev->pers;
3372 oldpriv = mddev->private;
3373 mddev->pers = pers;
3374 mddev->private = priv;
3375 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3376 mddev->level = mddev->new_level;
3377 mddev->layout = mddev->new_layout;
3378 mddev->chunk_sectors = mddev->new_chunk_sectors;
3379 mddev->delta_disks = 0;
3380 mddev->reshape_backwards = 0;
3381 mddev->degraded = 0;
3382 spin_unlock(&mddev->lock);
3383
3384 if (oldpers->sync_request == NULL &&
3385 mddev->external) {
3386 /* We are converting from a no-redundancy array
3387 * to a redundancy array and metadata is managed
3388 * externally so we need to be sure that writes
3389 * won't block due to a need to transition
3390 * clean->dirty
3391 * until external management is started.
3392 */
3393 mddev->in_sync = 0;
3394 mddev->safemode_delay = 0;
3395 mddev->safemode = 0;
3396 }
3397
3398 oldpers->free(mddev, oldpriv);
3399
3400 if (oldpers->sync_request == NULL &&
3401 pers->sync_request != NULL) {
3402 /* need to add the md_redundancy_group */
3403 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3404 printk(KERN_WARNING
3405 "md: cannot register extra attributes for %s\n",
3406 mdname(mddev));
3407 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3408 }
3409 if (oldpers->sync_request != NULL &&
3410 pers->sync_request == NULL) {
3411 /* need to remove the md_redundancy_group */
3412 if (mddev->to_remove == NULL)
3413 mddev->to_remove = &md_redundancy_group;
3414 }
3415
3416 rdev_for_each(rdev, mddev) {
3417 if (rdev->raid_disk < 0)
3418 continue;
3419 if (rdev->new_raid_disk >= mddev->raid_disks)
3420 rdev->new_raid_disk = -1;
3421 if (rdev->new_raid_disk == rdev->raid_disk)
3422 continue;
3423 sysfs_unlink_rdev(mddev, rdev);
3424 }
3425 rdev_for_each(rdev, mddev) {
3426 if (rdev->raid_disk < 0)
3427 continue;
3428 if (rdev->new_raid_disk == rdev->raid_disk)
3429 continue;
3430 rdev->raid_disk = rdev->new_raid_disk;
3431 if (rdev->raid_disk < 0)
3432 clear_bit(In_sync, &rdev->flags);
3433 else {
3434 if (sysfs_link_rdev(mddev, rdev))
3435 printk(KERN_WARNING "md: cannot register rd%d"
3436 " for %s after level change\n",
3437 rdev->raid_disk, mdname(mddev));
3438 }
3439 }
3440
3441 if (pers->sync_request == NULL) {
3442 /* this is now an array without redundancy, so
3443 * it must always be in_sync
3444 */
3445 mddev->in_sync = 1;
3446 del_timer_sync(&mddev->safemode_timer);
3447 }
3448 blk_set_stacking_limits(&mddev->queue->limits);
3449 pers->run(mddev);
3450 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3451 mddev_resume(mddev);
3452 if (!mddev->thread)
3453 md_update_sb(mddev, 1);
3454 sysfs_notify(&mddev->kobj, NULL, "level");
3455 md_new_event(mddev);
3456 rv = len;
3457 out_unlock:
3458 mddev_unlock(mddev);
3459 return rv;
3460 }
3461
3462 static struct md_sysfs_entry md_level =
3463 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3464
3465 static ssize_t
3466 layout_show(struct mddev *mddev, char *page)
3467 {
3468 /* just a number, not meaningful for all levels */
3469 if (mddev->reshape_position != MaxSector &&
3470 mddev->layout != mddev->new_layout)
3471 return sprintf(page, "%d (%d)\n",
3472 mddev->new_layout, mddev->layout);
3473 return sprintf(page, "%d\n", mddev->layout);
3474 }
3475
3476 static ssize_t
3477 layout_store(struct mddev *mddev, const char *buf, size_t len)
3478 {
3479 char *e;
3480 unsigned long n = simple_strtoul(buf, &e, 10);
3481 int err;
3482
3483 if (!*buf || (*e && *e != '\n'))
3484 return -EINVAL;
3485 err = mddev_lock(mddev);
3486 if (err)
3487 return err;
3488
3489 if (mddev->pers) {
3490 if (mddev->pers->check_reshape == NULL)
3491 err = -EBUSY;
3492 else if (mddev->ro)
3493 err = -EROFS;
3494 else {
3495 mddev->new_layout = n;
3496 err = mddev->pers->check_reshape(mddev);
3497 if (err)
3498 mddev->new_layout = mddev->layout;
3499 }
3500 } else {
3501 mddev->new_layout = n;
3502 if (mddev->reshape_position == MaxSector)
3503 mddev->layout = n;
3504 }
3505 mddev_unlock(mddev);
3506 return err ?: len;
3507 }
3508 static struct md_sysfs_entry md_layout =
3509 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3510
3511 static ssize_t
3512 raid_disks_show(struct mddev *mddev, char *page)
3513 {
3514 if (mddev->raid_disks == 0)
3515 return 0;
3516 if (mddev->reshape_position != MaxSector &&
3517 mddev->delta_disks != 0)
3518 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3519 mddev->raid_disks - mddev->delta_disks);
3520 return sprintf(page, "%d\n", mddev->raid_disks);
3521 }
3522
3523 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3524
3525 static ssize_t
3526 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3527 {
3528 char *e;
3529 int err;
3530 unsigned long n = simple_strtoul(buf, &e, 10);
3531
3532 if (!*buf || (*e && *e != '\n'))
3533 return -EINVAL;
3534
3535 err = mddev_lock(mddev);
3536 if (err)
3537 return err;
3538 if (mddev->pers)
3539 err = update_raid_disks(mddev, n);
3540 else if (mddev->reshape_position != MaxSector) {
3541 struct md_rdev *rdev;
3542 int olddisks = mddev->raid_disks - mddev->delta_disks;
3543
3544 err = -EINVAL;
3545 rdev_for_each(rdev, mddev) {
3546 if (olddisks < n &&
3547 rdev->data_offset < rdev->new_data_offset)
3548 goto out_unlock;
3549 if (olddisks > n &&
3550 rdev->data_offset > rdev->new_data_offset)
3551 goto out_unlock;
3552 }
3553 err = 0;
3554 mddev->delta_disks = n - olddisks;
3555 mddev->raid_disks = n;
3556 mddev->reshape_backwards = (mddev->delta_disks < 0);
3557 } else
3558 mddev->raid_disks = n;
3559 out_unlock:
3560 mddev_unlock(mddev);
3561 return err ? err : len;
3562 }
3563 static struct md_sysfs_entry md_raid_disks =
3564 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3565
3566 static ssize_t
3567 chunk_size_show(struct mddev *mddev, char *page)
3568 {
3569 if (mddev->reshape_position != MaxSector &&
3570 mddev->chunk_sectors != mddev->new_chunk_sectors)
3571 return sprintf(page, "%d (%d)\n",
3572 mddev->new_chunk_sectors << 9,
3573 mddev->chunk_sectors << 9);
3574 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3575 }
3576
3577 static ssize_t
3578 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3579 {
3580 int err;
3581 char *e;
3582 unsigned long n = simple_strtoul(buf, &e, 10);
3583
3584 if (!*buf || (*e && *e != '\n'))
3585 return -EINVAL;
3586
3587 err = mddev_lock(mddev);
3588 if (err)
3589 return err;
3590 if (mddev->pers) {
3591 if (mddev->pers->check_reshape == NULL)
3592 err = -EBUSY;
3593 else if (mddev->ro)
3594 err = -EROFS;
3595 else {
3596 mddev->new_chunk_sectors = n >> 9;
3597 err = mddev->pers->check_reshape(mddev);
3598 if (err)
3599 mddev->new_chunk_sectors = mddev->chunk_sectors;
3600 }
3601 } else {
3602 mddev->new_chunk_sectors = n >> 9;
3603 if (mddev->reshape_position == MaxSector)
3604 mddev->chunk_sectors = n >> 9;
3605 }
3606 mddev_unlock(mddev);
3607 return err ?: len;
3608 }
3609 static struct md_sysfs_entry md_chunk_size =
3610 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3611
3612 static ssize_t
3613 resync_start_show(struct mddev *mddev, char *page)
3614 {
3615 if (mddev->recovery_cp == MaxSector)
3616 return sprintf(page, "none\n");
3617 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3618 }
3619
3620 static ssize_t
3621 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3622 {
3623 int err;
3624 char *e;
3625 unsigned long long n = simple_strtoull(buf, &e, 10);
3626
3627 err = mddev_lock(mddev);
3628 if (err)
3629 return err;
3630 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3631 err = -EBUSY;
3632 else if (cmd_match(buf, "none"))
3633 n = MaxSector;
3634 else if (!*buf || (*e && *e != '\n'))
3635 err = -EINVAL;
3636
3637 if (!err) {
3638 mddev->recovery_cp = n;
3639 if (mddev->pers)
3640 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3641 }
3642 mddev_unlock(mddev);
3643 return err ?: len;
3644 }
3645 static struct md_sysfs_entry md_resync_start =
3646 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3647
3648 /*
3649 * The array state can be:
3650 *
3651 * clear
3652 * No devices, no size, no level
3653 * Equivalent to STOP_ARRAY ioctl
3654 * inactive
3655 * May have some settings, but array is not active
3656 * all IO results in error
3657 * When written, doesn't tear down array, but just stops it
3658 * suspended (not supported yet)
3659 * All IO requests will block. The array can be reconfigured.
3660 * Writing this, if accepted, will block until array is quiescent
3661 * readonly
3662 * no resync can happen. no superblocks get written.
3663 * write requests fail
3664 * read-auto
3665 * like readonly, but behaves like 'clean' on a write request.
3666 *
3667 * clean - no pending writes, but otherwise active.
3668 * When written to inactive array, starts without resync
3669 * If a write request arrives then
3670 * if metadata is known, mark 'dirty' and switch to 'active'.
3671 * if not known, block and switch to write-pending
3672 * If written to an active array that has pending writes, then fails.
3673 * active
3674 * fully active: IO and resync can be happening.
3675 * When written to inactive array, starts with resync
3676 *
3677 * write-pending
3678 * clean, but writes are blocked waiting for 'active' to be written.
3679 *
3680 * active-idle
3681 * like active, but no writes have been seen for a while (100msec).
3682 *
3683 */
3684 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3685 write_pending, active_idle, bad_word};
3686 static char *array_states[] = {
3687 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3688 "write-pending", "active-idle", NULL };
3689
3690 static int match_word(const char *word, char **list)
3691 {
3692 int n;
3693 for (n=0; list[n]; n++)
3694 if (cmd_match(word, list[n]))
3695 break;
3696 return n;
3697 }
3698
3699 static ssize_t
3700 array_state_show(struct mddev *mddev, char *page)
3701 {
3702 enum array_state st = inactive;
3703
3704 if (mddev->pers)
3705 switch(mddev->ro) {
3706 case 1:
3707 st = readonly;
3708 break;
3709 case 2:
3710 st = read_auto;
3711 break;
3712 case 0:
3713 if (mddev->in_sync)
3714 st = clean;
3715 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3716 st = write_pending;
3717 else if (mddev->safemode)
3718 st = active_idle;
3719 else
3720 st = active;
3721 }
3722 else {
3723 if (list_empty(&mddev->disks) &&
3724 mddev->raid_disks == 0 &&
3725 mddev->dev_sectors == 0)
3726 st = clear;
3727 else
3728 st = inactive;
3729 }
3730 return sprintf(page, "%s\n", array_states[st]);
3731 }
3732
3733 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
3734 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
3735 static int do_md_run(struct mddev *mddev);
3736 static int restart_array(struct mddev *mddev);
3737
3738 static ssize_t
3739 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3740 {
3741 int err;
3742 enum array_state st = match_word(buf, array_states);
3743
3744 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
3745 /* don't take reconfig_mutex when toggling between
3746 * clean and active
3747 */
3748 spin_lock(&mddev->lock);
3749 if (st == active) {
3750 restart_array(mddev);
3751 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3752 wake_up(&mddev->sb_wait);
3753 err = 0;
3754 } else /* st == clean */ {
3755 restart_array(mddev);
3756 if (atomic_read(&mddev->writes_pending) == 0) {
3757 if (mddev->in_sync == 0) {
3758 mddev->in_sync = 1;
3759 if (mddev->safemode == 1)
3760 mddev->safemode = 0;
3761 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3762 }
3763 err = 0;
3764 } else
3765 err = -EBUSY;
3766 }
3767 spin_unlock(&mddev->lock);
3768 return err;
3769 }
3770 err = mddev_lock(mddev);
3771 if (err)
3772 return err;
3773 err = -EINVAL;
3774 switch(st) {
3775 case bad_word:
3776 break;
3777 case clear:
3778 /* stopping an active array */
3779 err = do_md_stop(mddev, 0, NULL);
3780 break;
3781 case inactive:
3782 /* stopping an active array */
3783 if (mddev->pers)
3784 err = do_md_stop(mddev, 2, NULL);
3785 else
3786 err = 0; /* already inactive */
3787 break;
3788 case suspended:
3789 break; /* not supported yet */
3790 case readonly:
3791 if (mddev->pers)
3792 err = md_set_readonly(mddev, NULL);
3793 else {
3794 mddev->ro = 1;
3795 set_disk_ro(mddev->gendisk, 1);
3796 err = do_md_run(mddev);
3797 }
3798 break;
3799 case read_auto:
3800 if (mddev->pers) {
3801 if (mddev->ro == 0)
3802 err = md_set_readonly(mddev, NULL);
3803 else if (mddev->ro == 1)
3804 err = restart_array(mddev);
3805 if (err == 0) {
3806 mddev->ro = 2;
3807 set_disk_ro(mddev->gendisk, 0);
3808 }
3809 } else {
3810 mddev->ro = 2;
3811 err = do_md_run(mddev);
3812 }
3813 break;
3814 case clean:
3815 if (mddev->pers) {
3816 restart_array(mddev);
3817 spin_lock(&mddev->lock);
3818 if (atomic_read(&mddev->writes_pending) == 0) {
3819 if (mddev->in_sync == 0) {
3820 mddev->in_sync = 1;
3821 if (mddev->safemode == 1)
3822 mddev->safemode = 0;
3823 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3824 }
3825 err = 0;
3826 } else
3827 err = -EBUSY;
3828 spin_unlock(&mddev->lock);
3829 } else
3830 err = -EINVAL;
3831 break;
3832 case active:
3833 if (mddev->pers) {
3834 restart_array(mddev);
3835 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3836 wake_up(&mddev->sb_wait);
3837 err = 0;
3838 } else {
3839 mddev->ro = 0;
3840 set_disk_ro(mddev->gendisk, 0);
3841 err = do_md_run(mddev);
3842 }
3843 break;
3844 case write_pending:
3845 case active_idle:
3846 /* these cannot be set */
3847 break;
3848 }
3849
3850 if (!err) {
3851 if (mddev->hold_active == UNTIL_IOCTL)
3852 mddev->hold_active = 0;
3853 sysfs_notify_dirent_safe(mddev->sysfs_state);
3854 }
3855 mddev_unlock(mddev);
3856 return err ?: len;
3857 }
3858 static struct md_sysfs_entry md_array_state =
3859 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3860
3861 static ssize_t
3862 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
3863 return sprintf(page, "%d\n",
3864 atomic_read(&mddev->max_corr_read_errors));
3865 }
3866
3867 static ssize_t
3868 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
3869 {
3870 char *e;
3871 unsigned long n = simple_strtoul(buf, &e, 10);
3872
3873 if (*buf && (*e == 0 || *e == '\n')) {
3874 atomic_set(&mddev->max_corr_read_errors, n);
3875 return len;
3876 }
3877 return -EINVAL;
3878 }
3879
3880 static struct md_sysfs_entry max_corr_read_errors =
3881 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3882 max_corrected_read_errors_store);
3883
3884 static ssize_t
3885 null_show(struct mddev *mddev, char *page)
3886 {
3887 return -EINVAL;
3888 }
3889
3890 static ssize_t
3891 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
3892 {
3893 /* buf must be %d:%d\n? giving major and minor numbers */
3894 /* The new device is added to the array.
3895 * If the array has a persistent superblock, we read the
3896 * superblock to initialise info and check validity.
3897 * Otherwise, only checking done is that in bind_rdev_to_array,
3898 * which mainly checks size.
3899 */
3900 char *e;
3901 int major = simple_strtoul(buf, &e, 10);
3902 int minor;
3903 dev_t dev;
3904 struct md_rdev *rdev;
3905 int err;
3906
3907 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3908 return -EINVAL;
3909 minor = simple_strtoul(e+1, &e, 10);
3910 if (*e && *e != '\n')
3911 return -EINVAL;
3912 dev = MKDEV(major, minor);
3913 if (major != MAJOR(dev) ||
3914 minor != MINOR(dev))
3915 return -EOVERFLOW;
3916
3917 flush_workqueue(md_misc_wq);
3918
3919 err = mddev_lock(mddev);
3920 if (err)
3921 return err;
3922 if (mddev->persistent) {
3923 rdev = md_import_device(dev, mddev->major_version,
3924 mddev->minor_version);
3925 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3926 struct md_rdev *rdev0
3927 = list_entry(mddev->disks.next,
3928 struct md_rdev, same_set);
3929 err = super_types[mddev->major_version]
3930 .load_super(rdev, rdev0, mddev->minor_version);
3931 if (err < 0)
3932 goto out;
3933 }
3934 } else if (mddev->external)
3935 rdev = md_import_device(dev, -2, -1);
3936 else
3937 rdev = md_import_device(dev, -1, -1);
3938
3939 if (IS_ERR(rdev))
3940 return PTR_ERR(rdev);
3941 err = bind_rdev_to_array(rdev, mddev);
3942 out:
3943 if (err)
3944 export_rdev(rdev);
3945 mddev_unlock(mddev);
3946 return err ? err : len;
3947 }
3948
3949 static struct md_sysfs_entry md_new_device =
3950 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3951
3952 static ssize_t
3953 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
3954 {
3955 char *end;
3956 unsigned long chunk, end_chunk;
3957 int err;
3958
3959 err = mddev_lock(mddev);
3960 if (err)
3961 return err;
3962 if (!mddev->bitmap)
3963 goto out;
3964 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3965 while (*buf) {
3966 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3967 if (buf == end) break;
3968 if (*end == '-') { /* range */
3969 buf = end + 1;
3970 end_chunk = simple_strtoul(buf, &end, 0);
3971 if (buf == end) break;
3972 }
3973 if (*end && !isspace(*end)) break;
3974 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3975 buf = skip_spaces(end);
3976 }
3977 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3978 out:
3979 mddev_unlock(mddev);
3980 return len;
3981 }
3982
3983 static struct md_sysfs_entry md_bitmap =
3984 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3985
3986 static ssize_t
3987 size_show(struct mddev *mddev, char *page)
3988 {
3989 return sprintf(page, "%llu\n",
3990 (unsigned long long)mddev->dev_sectors / 2);
3991 }
3992
3993 static int update_size(struct mddev *mddev, sector_t num_sectors);
3994
3995 static ssize_t
3996 size_store(struct mddev *mddev, const char *buf, size_t len)
3997 {
3998 /* If array is inactive, we can reduce the component size, but
3999 * not increase it (except from 0).
4000 * If array is active, we can try an on-line resize
4001 */
4002 sector_t sectors;
4003 int err = strict_blocks_to_sectors(buf, &sectors);
4004
4005 if (err < 0)
4006 return err;
4007 err = mddev_lock(mddev);
4008 if (err)
4009 return err;
4010 if (mddev->pers) {
4011 err = update_size(mddev, sectors);
4012 md_update_sb(mddev, 1);
4013 } else {
4014 if (mddev->dev_sectors == 0 ||
4015 mddev->dev_sectors > sectors)
4016 mddev->dev_sectors = sectors;
4017 else
4018 err = -ENOSPC;
4019 }
4020 mddev_unlock(mddev);
4021 return err ? err : len;
4022 }
4023
4024 static struct md_sysfs_entry md_size =
4025 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4026
4027 /* Metadata version.
4028 * This is one of
4029 * 'none' for arrays with no metadata (good luck...)
4030 * 'external' for arrays with externally managed metadata,
4031 * or N.M for internally known formats
4032 */
4033 static ssize_t
4034 metadata_show(struct mddev *mddev, char *page)
4035 {
4036 if (mddev->persistent)
4037 return sprintf(page, "%d.%d\n",
4038 mddev->major_version, mddev->minor_version);
4039 else if (mddev->external)
4040 return sprintf(page, "external:%s\n", mddev->metadata_type);
4041 else
4042 return sprintf(page, "none\n");
4043 }
4044
4045 static ssize_t
4046 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4047 {
4048 int major, minor;
4049 char *e;
4050 int err;
4051 /* Changing the details of 'external' metadata is
4052 * always permitted. Otherwise there must be
4053 * no devices attached to the array.
4054 */
4055
4056 err = mddev_lock(mddev);
4057 if (err)
4058 return err;
4059 err = -EBUSY;
4060 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4061 ;
4062 else if (!list_empty(&mddev->disks))
4063 goto out_unlock;
4064
4065 err = 0;
4066 if (cmd_match(buf, "none")) {
4067 mddev->persistent = 0;
4068 mddev->external = 0;
4069 mddev->major_version = 0;
4070 mddev->minor_version = 90;
4071 goto out_unlock;
4072 }
4073 if (strncmp(buf, "external:", 9) == 0) {
4074 size_t namelen = len-9;
4075 if (namelen >= sizeof(mddev->metadata_type))
4076 namelen = sizeof(mddev->metadata_type)-1;
4077 strncpy(mddev->metadata_type, buf+9, namelen);
4078 mddev->metadata_type[namelen] = 0;
4079 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4080 mddev->metadata_type[--namelen] = 0;
4081 mddev->persistent = 0;
4082 mddev->external = 1;
4083 mddev->major_version = 0;
4084 mddev->minor_version = 90;
4085 goto out_unlock;
4086 }
4087 major = simple_strtoul(buf, &e, 10);
4088 err = -EINVAL;
4089 if (e==buf || *e != '.')
4090 goto out_unlock;
4091 buf = e+1;
4092 minor = simple_strtoul(buf, &e, 10);
4093 if (e==buf || (*e && *e != '\n') )
4094 goto out_unlock;
4095 err = -ENOENT;
4096 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4097 goto out_unlock;
4098 mddev->major_version = major;
4099 mddev->minor_version = minor;
4100 mddev->persistent = 1;
4101 mddev->external = 0;
4102 err = 0;
4103 out_unlock:
4104 mddev_unlock(mddev);
4105 return err ?: len;
4106 }
4107
4108 static struct md_sysfs_entry md_metadata =
4109 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4110
4111 static ssize_t
4112 action_show(struct mddev *mddev, char *page)
4113 {
4114 char *type = "idle";
4115 unsigned long recovery = mddev->recovery;
4116 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4117 type = "frozen";
4118 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4119 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4120 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4121 type = "reshape";
4122 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4123 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4124 type = "resync";
4125 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4126 type = "check";
4127 else
4128 type = "repair";
4129 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4130 type = "recover";
4131 }
4132 return sprintf(page, "%s\n", type);
4133 }
4134
4135 static ssize_t
4136 action_store(struct mddev *mddev, const char *page, size_t len)
4137 {
4138 if (!mddev->pers || !mddev->pers->sync_request)
4139 return -EINVAL;
4140
4141 if (cmd_match(page, "frozen"))
4142 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4143 else
4144 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4145
4146 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4147 flush_workqueue(md_misc_wq);
4148 if (mddev->sync_thread) {
4149 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4150 if (mddev_lock(mddev) == 0) {
4151 md_reap_sync_thread(mddev);
4152 mddev_unlock(mddev);
4153 }
4154 }
4155 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4156 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
4157 return -EBUSY;
4158 else if (cmd_match(page, "resync"))
4159 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4160 else if (cmd_match(page, "recover")) {
4161 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4162 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4163 } else if (cmd_match(page, "reshape")) {
4164 int err;
4165 if (mddev->pers->start_reshape == NULL)
4166 return -EINVAL;
4167 err = mddev_lock(mddev);
4168 if (!err) {
4169 err = mddev->pers->start_reshape(mddev);
4170 mddev_unlock(mddev);
4171 }
4172 if (err)
4173 return err;
4174 sysfs_notify(&mddev->kobj, NULL, "degraded");
4175 } else {
4176 if (cmd_match(page, "check"))
4177 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4178 else if (!cmd_match(page, "repair"))
4179 return -EINVAL;
4180 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4181 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4182 }
4183 if (mddev->ro == 2) {
4184 /* A write to sync_action is enough to justify
4185 * canceling read-auto mode
4186 */
4187 mddev->ro = 0;
4188 md_wakeup_thread(mddev->sync_thread);
4189 }
4190 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4191 md_wakeup_thread(mddev->thread);
4192 sysfs_notify_dirent_safe(mddev->sysfs_action);
4193 return len;
4194 }
4195
4196 static struct md_sysfs_entry md_scan_mode =
4197 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4198
4199 static ssize_t
4200 last_sync_action_show(struct mddev *mddev, char *page)
4201 {
4202 return sprintf(page, "%s\n", mddev->last_sync_action);
4203 }
4204
4205 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4206
4207 static ssize_t
4208 mismatch_cnt_show(struct mddev *mddev, char *page)
4209 {
4210 return sprintf(page, "%llu\n",
4211 (unsigned long long)
4212 atomic64_read(&mddev->resync_mismatches));
4213 }
4214
4215 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4216
4217 static ssize_t
4218 sync_min_show(struct mddev *mddev, char *page)
4219 {
4220 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4221 mddev->sync_speed_min ? "local": "system");
4222 }
4223
4224 static ssize_t
4225 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4226 {
4227 int min;
4228 char *e;
4229 if (strncmp(buf, "system", 6)==0) {
4230 mddev->sync_speed_min = 0;
4231 return len;
4232 }
4233 min = simple_strtoul(buf, &e, 10);
4234 if (buf == e || (*e && *e != '\n') || min <= 0)
4235 return -EINVAL;
4236 mddev->sync_speed_min = min;
4237 return len;
4238 }
4239
4240 static struct md_sysfs_entry md_sync_min =
4241 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4242
4243 static ssize_t
4244 sync_max_show(struct mddev *mddev, char *page)
4245 {
4246 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4247 mddev->sync_speed_max ? "local": "system");
4248 }
4249
4250 static ssize_t
4251 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4252 {
4253 int max;
4254 char *e;
4255 if (strncmp(buf, "system", 6)==0) {
4256 mddev->sync_speed_max = 0;
4257 return len;
4258 }
4259 max = simple_strtoul(buf, &e, 10);
4260 if (buf == e || (*e && *e != '\n') || max <= 0)
4261 return -EINVAL;
4262 mddev->sync_speed_max = max;
4263 return len;
4264 }
4265
4266 static struct md_sysfs_entry md_sync_max =
4267 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4268
4269 static ssize_t
4270 degraded_show(struct mddev *mddev, char *page)
4271 {
4272 return sprintf(page, "%d\n", mddev->degraded);
4273 }
4274 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4275
4276 static ssize_t
4277 sync_force_parallel_show(struct mddev *mddev, char *page)
4278 {
4279 return sprintf(page, "%d\n", mddev->parallel_resync);
4280 }
4281
4282 static ssize_t
4283 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4284 {
4285 long n;
4286
4287 if (kstrtol(buf, 10, &n))
4288 return -EINVAL;
4289
4290 if (n != 0 && n != 1)
4291 return -EINVAL;
4292
4293 mddev->parallel_resync = n;
4294
4295 if (mddev->sync_thread)
4296 wake_up(&resync_wait);
4297
4298 return len;
4299 }
4300
4301 /* force parallel resync, even with shared block devices */
4302 static struct md_sysfs_entry md_sync_force_parallel =
4303 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4304 sync_force_parallel_show, sync_force_parallel_store);
4305
4306 static ssize_t
4307 sync_speed_show(struct mddev *mddev, char *page)
4308 {
4309 unsigned long resync, dt, db;
4310 if (mddev->curr_resync == 0)
4311 return sprintf(page, "none\n");
4312 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4313 dt = (jiffies - mddev->resync_mark) / HZ;
4314 if (!dt) dt++;
4315 db = resync - mddev->resync_mark_cnt;
4316 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4317 }
4318
4319 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4320
4321 static ssize_t
4322 sync_completed_show(struct mddev *mddev, char *page)
4323 {
4324 unsigned long long max_sectors, resync;
4325
4326 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4327 return sprintf(page, "none\n");
4328
4329 if (mddev->curr_resync == 1 ||
4330 mddev->curr_resync == 2)
4331 return sprintf(page, "delayed\n");
4332
4333 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4334 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4335 max_sectors = mddev->resync_max_sectors;
4336 else
4337 max_sectors = mddev->dev_sectors;
4338
4339 resync = mddev->curr_resync_completed;
4340 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4341 }
4342
4343 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
4344
4345 static ssize_t
4346 min_sync_show(struct mddev *mddev, char *page)
4347 {
4348 return sprintf(page, "%llu\n",
4349 (unsigned long long)mddev->resync_min);
4350 }
4351 static ssize_t
4352 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4353 {
4354 unsigned long long min;
4355 int err;
4356 int chunk;
4357
4358 if (kstrtoull(buf, 10, &min))
4359 return -EINVAL;
4360
4361 spin_lock(&mddev->lock);
4362 err = -EINVAL;
4363 if (min > mddev->resync_max)
4364 goto out_unlock;
4365
4366 err = -EBUSY;
4367 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4368 goto out_unlock;
4369
4370 /* Must be a multiple of chunk_size */
4371 chunk = mddev->chunk_sectors;
4372 if (chunk) {
4373 sector_t temp = min;
4374
4375 err = -EINVAL;
4376 if (sector_div(temp, chunk))
4377 goto out_unlock;
4378 }
4379 mddev->resync_min = min;
4380 err = 0;
4381
4382 out_unlock:
4383 spin_unlock(&mddev->lock);
4384 return err ?: len;
4385 }
4386
4387 static struct md_sysfs_entry md_min_sync =
4388 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4389
4390 static ssize_t
4391 max_sync_show(struct mddev *mddev, char *page)
4392 {
4393 if (mddev->resync_max == MaxSector)
4394 return sprintf(page, "max\n");
4395 else
4396 return sprintf(page, "%llu\n",
4397 (unsigned long long)mddev->resync_max);
4398 }
4399 static ssize_t
4400 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4401 {
4402 int err;
4403 spin_lock(&mddev->lock);
4404 if (strncmp(buf, "max", 3) == 0)
4405 mddev->resync_max = MaxSector;
4406 else {
4407 unsigned long long max;
4408 int chunk;
4409
4410 err = -EINVAL;
4411 if (kstrtoull(buf, 10, &max))
4412 goto out_unlock;
4413 if (max < mddev->resync_min)
4414 goto out_unlock;
4415
4416 err = -EBUSY;
4417 if (max < mddev->resync_max &&
4418 mddev->ro == 0 &&
4419 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4420 goto out_unlock;
4421
4422 /* Must be a multiple of chunk_size */
4423 chunk = mddev->chunk_sectors;
4424 if (chunk) {
4425 sector_t temp = max;
4426
4427 err = -EINVAL;
4428 if (sector_div(temp, chunk))
4429 goto out_unlock;
4430 }
4431 mddev->resync_max = max;
4432 }
4433 wake_up(&mddev->recovery_wait);
4434 err = 0;
4435 out_unlock:
4436 spin_unlock(&mddev->lock);
4437 return err ?: len;
4438 }
4439
4440 static struct md_sysfs_entry md_max_sync =
4441 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4442
4443 static ssize_t
4444 suspend_lo_show(struct mddev *mddev, char *page)
4445 {
4446 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4447 }
4448
4449 static ssize_t
4450 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4451 {
4452 char *e;
4453 unsigned long long new = simple_strtoull(buf, &e, 10);
4454 unsigned long long old;
4455 int err;
4456
4457 if (buf == e || (*e && *e != '\n'))
4458 return -EINVAL;
4459
4460 err = mddev_lock(mddev);
4461 if (err)
4462 return err;
4463 err = -EINVAL;
4464 if (mddev->pers == NULL ||
4465 mddev->pers->quiesce == NULL)
4466 goto unlock;
4467 old = mddev->suspend_lo;
4468 mddev->suspend_lo = new;
4469 if (new >= old)
4470 /* Shrinking suspended region */
4471 mddev->pers->quiesce(mddev, 2);
4472 else {
4473 /* Expanding suspended region - need to wait */
4474 mddev->pers->quiesce(mddev, 1);
4475 mddev->pers->quiesce(mddev, 0);
4476 }
4477 err = 0;
4478 unlock:
4479 mddev_unlock(mddev);
4480 return err ?: len;
4481 }
4482 static struct md_sysfs_entry md_suspend_lo =
4483 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4484
4485 static ssize_t
4486 suspend_hi_show(struct mddev *mddev, char *page)
4487 {
4488 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4489 }
4490
4491 static ssize_t
4492 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4493 {
4494 char *e;
4495 unsigned long long new = simple_strtoull(buf, &e, 10);
4496 unsigned long long old;
4497 int err;
4498
4499 if (buf == e || (*e && *e != '\n'))
4500 return -EINVAL;
4501
4502 err = mddev_lock(mddev);
4503 if (err)
4504 return err;
4505 err = -EINVAL;
4506 if (mddev->pers == NULL ||
4507 mddev->pers->quiesce == NULL)
4508 goto unlock;
4509 old = mddev->suspend_hi;
4510 mddev->suspend_hi = new;
4511 if (new <= old)
4512 /* Shrinking suspended region */
4513 mddev->pers->quiesce(mddev, 2);
4514 else {
4515 /* Expanding suspended region - need to wait */
4516 mddev->pers->quiesce(mddev, 1);
4517 mddev->pers->quiesce(mddev, 0);
4518 }
4519 err = 0;
4520 unlock:
4521 mddev_unlock(mddev);
4522 return err ?: len;
4523 }
4524 static struct md_sysfs_entry md_suspend_hi =
4525 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4526
4527 static ssize_t
4528 reshape_position_show(struct mddev *mddev, char *page)
4529 {
4530 if (mddev->reshape_position != MaxSector)
4531 return sprintf(page, "%llu\n",
4532 (unsigned long long)mddev->reshape_position);
4533 strcpy(page, "none\n");
4534 return 5;
4535 }
4536
4537 static ssize_t
4538 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4539 {
4540 struct md_rdev *rdev;
4541 char *e;
4542 int err;
4543 unsigned long long new = simple_strtoull(buf, &e, 10);
4544
4545 if (buf == e || (*e && *e != '\n'))
4546 return -EINVAL;
4547 err = mddev_lock(mddev);
4548 if (err)
4549 return err;
4550 err = -EBUSY;
4551 if (mddev->pers)
4552 goto unlock;
4553 mddev->reshape_position = new;
4554 mddev->delta_disks = 0;
4555 mddev->reshape_backwards = 0;
4556 mddev->new_level = mddev->level;
4557 mddev->new_layout = mddev->layout;
4558 mddev->new_chunk_sectors = mddev->chunk_sectors;
4559 rdev_for_each(rdev, mddev)
4560 rdev->new_data_offset = rdev->data_offset;
4561 err = 0;
4562 unlock:
4563 mddev_unlock(mddev);
4564 return err ?: len;
4565 }
4566
4567 static struct md_sysfs_entry md_reshape_position =
4568 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4569 reshape_position_store);
4570
4571 static ssize_t
4572 reshape_direction_show(struct mddev *mddev, char *page)
4573 {
4574 return sprintf(page, "%s\n",
4575 mddev->reshape_backwards ? "backwards" : "forwards");
4576 }
4577
4578 static ssize_t
4579 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4580 {
4581 int backwards = 0;
4582 int err;
4583
4584 if (cmd_match(buf, "forwards"))
4585 backwards = 0;
4586 else if (cmd_match(buf, "backwards"))
4587 backwards = 1;
4588 else
4589 return -EINVAL;
4590 if (mddev->reshape_backwards == backwards)
4591 return len;
4592
4593 err = mddev_lock(mddev);
4594 if (err)
4595 return err;
4596 /* check if we are allowed to change */
4597 if (mddev->delta_disks)
4598 err = -EBUSY;
4599 else if (mddev->persistent &&
4600 mddev->major_version == 0)
4601 err = -EINVAL;
4602 else
4603 mddev->reshape_backwards = backwards;
4604 mddev_unlock(mddev);
4605 return err ?: len;
4606 }
4607
4608 static struct md_sysfs_entry md_reshape_direction =
4609 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4610 reshape_direction_store);
4611
4612 static ssize_t
4613 array_size_show(struct mddev *mddev, char *page)
4614 {
4615 if (mddev->external_size)
4616 return sprintf(page, "%llu\n",
4617 (unsigned long long)mddev->array_sectors/2);
4618 else
4619 return sprintf(page, "default\n");
4620 }
4621
4622 static ssize_t
4623 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4624 {
4625 sector_t sectors;
4626 int err;
4627
4628 err = mddev_lock(mddev);
4629 if (err)
4630 return err;
4631
4632 if (strncmp(buf, "default", 7) == 0) {
4633 if (mddev->pers)
4634 sectors = mddev->pers->size(mddev, 0, 0);
4635 else
4636 sectors = mddev->array_sectors;
4637
4638 mddev->external_size = 0;
4639 } else {
4640 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4641 err = -EINVAL;
4642 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4643 err = -E2BIG;
4644 else
4645 mddev->external_size = 1;
4646 }
4647
4648 if (!err) {
4649 mddev->array_sectors = sectors;
4650 if (mddev->pers) {
4651 set_capacity(mddev->gendisk, mddev->array_sectors);
4652 revalidate_disk(mddev->gendisk);
4653 }
4654 }
4655 mddev_unlock(mddev);
4656 return err ?: len;
4657 }
4658
4659 static struct md_sysfs_entry md_array_size =
4660 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4661 array_size_store);
4662
4663 static struct attribute *md_default_attrs[] = {
4664 &md_level.attr,
4665 &md_layout.attr,
4666 &md_raid_disks.attr,
4667 &md_chunk_size.attr,
4668 &md_size.attr,
4669 &md_resync_start.attr,
4670 &md_metadata.attr,
4671 &md_new_device.attr,
4672 &md_safe_delay.attr,
4673 &md_array_state.attr,
4674 &md_reshape_position.attr,
4675 &md_reshape_direction.attr,
4676 &md_array_size.attr,
4677 &max_corr_read_errors.attr,
4678 NULL,
4679 };
4680
4681 static struct attribute *md_redundancy_attrs[] = {
4682 &md_scan_mode.attr,
4683 &md_last_scan_mode.attr,
4684 &md_mismatches.attr,
4685 &md_sync_min.attr,
4686 &md_sync_max.attr,
4687 &md_sync_speed.attr,
4688 &md_sync_force_parallel.attr,
4689 &md_sync_completed.attr,
4690 &md_min_sync.attr,
4691 &md_max_sync.attr,
4692 &md_suspend_lo.attr,
4693 &md_suspend_hi.attr,
4694 &md_bitmap.attr,
4695 &md_degraded.attr,
4696 NULL,
4697 };
4698 static struct attribute_group md_redundancy_group = {
4699 .name = NULL,
4700 .attrs = md_redundancy_attrs,
4701 };
4702
4703 static ssize_t
4704 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4705 {
4706 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4707 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4708 ssize_t rv;
4709
4710 if (!entry->show)
4711 return -EIO;
4712 spin_lock(&all_mddevs_lock);
4713 if (list_empty(&mddev->all_mddevs)) {
4714 spin_unlock(&all_mddevs_lock);
4715 return -EBUSY;
4716 }
4717 mddev_get(mddev);
4718 spin_unlock(&all_mddevs_lock);
4719
4720 rv = entry->show(mddev, page);
4721 mddev_put(mddev);
4722 return rv;
4723 }
4724
4725 static ssize_t
4726 md_attr_store(struct kobject *kobj, struct attribute *attr,
4727 const char *page, size_t length)
4728 {
4729 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4730 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4731 ssize_t rv;
4732
4733 if (!entry->store)
4734 return -EIO;
4735 if (!capable(CAP_SYS_ADMIN))
4736 return -EACCES;
4737 spin_lock(&all_mddevs_lock);
4738 if (list_empty(&mddev->all_mddevs)) {
4739 spin_unlock(&all_mddevs_lock);
4740 return -EBUSY;
4741 }
4742 mddev_get(mddev);
4743 spin_unlock(&all_mddevs_lock);
4744 rv = entry->store(mddev, page, length);
4745 mddev_put(mddev);
4746 return rv;
4747 }
4748
4749 static void md_free(struct kobject *ko)
4750 {
4751 struct mddev *mddev = container_of(ko, struct mddev, kobj);
4752
4753 if (mddev->sysfs_state)
4754 sysfs_put(mddev->sysfs_state);
4755
4756 if (mddev->gendisk) {
4757 del_gendisk(mddev->gendisk);
4758 put_disk(mddev->gendisk);
4759 }
4760 if (mddev->queue)
4761 blk_cleanup_queue(mddev->queue);
4762
4763 kfree(mddev);
4764 }
4765
4766 static const struct sysfs_ops md_sysfs_ops = {
4767 .show = md_attr_show,
4768 .store = md_attr_store,
4769 };
4770 static struct kobj_type md_ktype = {
4771 .release = md_free,
4772 .sysfs_ops = &md_sysfs_ops,
4773 .default_attrs = md_default_attrs,
4774 };
4775
4776 int mdp_major = 0;
4777
4778 static void mddev_delayed_delete(struct work_struct *ws)
4779 {
4780 struct mddev *mddev = container_of(ws, struct mddev, del_work);
4781
4782 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4783 kobject_del(&mddev->kobj);
4784 kobject_put(&mddev->kobj);
4785 }
4786
4787 static int md_alloc(dev_t dev, char *name)
4788 {
4789 static DEFINE_MUTEX(disks_mutex);
4790 struct mddev *mddev = mddev_find(dev);
4791 struct gendisk *disk;
4792 int partitioned;
4793 int shift;
4794 int unit;
4795 int error;
4796
4797 if (!mddev)
4798 return -ENODEV;
4799
4800 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4801 shift = partitioned ? MdpMinorShift : 0;
4802 unit = MINOR(mddev->unit) >> shift;
4803
4804 /* wait for any previous instance of this device to be
4805 * completely removed (mddev_delayed_delete).
4806 */
4807 flush_workqueue(md_misc_wq);
4808
4809 mutex_lock(&disks_mutex);
4810 error = -EEXIST;
4811 if (mddev->gendisk)
4812 goto abort;
4813
4814 if (name) {
4815 /* Need to ensure that 'name' is not a duplicate.
4816 */
4817 struct mddev *mddev2;
4818 spin_lock(&all_mddevs_lock);
4819
4820 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4821 if (mddev2->gendisk &&
4822 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4823 spin_unlock(&all_mddevs_lock);
4824 goto abort;
4825 }
4826 spin_unlock(&all_mddevs_lock);
4827 }
4828
4829 error = -ENOMEM;
4830 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4831 if (!mddev->queue)
4832 goto abort;
4833 mddev->queue->queuedata = mddev;
4834
4835 blk_queue_make_request(mddev->queue, md_make_request);
4836 blk_set_stacking_limits(&mddev->queue->limits);
4837
4838 disk = alloc_disk(1 << shift);
4839 if (!disk) {
4840 blk_cleanup_queue(mddev->queue);
4841 mddev->queue = NULL;
4842 goto abort;
4843 }
4844 disk->major = MAJOR(mddev->unit);
4845 disk->first_minor = unit << shift;
4846 if (name)
4847 strcpy(disk->disk_name, name);
4848 else if (partitioned)
4849 sprintf(disk->disk_name, "md_d%d", unit);
4850 else
4851 sprintf(disk->disk_name, "md%d", unit);
4852 disk->fops = &md_fops;
4853 disk->private_data = mddev;
4854 disk->queue = mddev->queue;
4855 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
4856 /* Allow extended partitions. This makes the
4857 * 'mdp' device redundant, but we can't really
4858 * remove it now.
4859 */
4860 disk->flags |= GENHD_FL_EXT_DEVT;
4861 mddev->gendisk = disk;
4862 /* As soon as we call add_disk(), another thread could get
4863 * through to md_open, so make sure it doesn't get too far
4864 */
4865 mutex_lock(&mddev->open_mutex);
4866 add_disk(disk);
4867
4868 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4869 &disk_to_dev(disk)->kobj, "%s", "md");
4870 if (error) {
4871 /* This isn't possible, but as kobject_init_and_add is marked
4872 * __must_check, we must do something with the result
4873 */
4874 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4875 disk->disk_name);
4876 error = 0;
4877 }
4878 if (mddev->kobj.sd &&
4879 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4880 printk(KERN_DEBUG "pointless warning\n");
4881 mutex_unlock(&mddev->open_mutex);
4882 abort:
4883 mutex_unlock(&disks_mutex);
4884 if (!error && mddev->kobj.sd) {
4885 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4886 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4887 }
4888 mddev_put(mddev);
4889 return error;
4890 }
4891
4892 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4893 {
4894 md_alloc(dev, NULL);
4895 return NULL;
4896 }
4897
4898 static int add_named_array(const char *val, struct kernel_param *kp)
4899 {
4900 /* val must be "md_*" where * is not all digits.
4901 * We allocate an array with a large free minor number, and
4902 * set the name to val. val must not already be an active name.
4903 */
4904 int len = strlen(val);
4905 char buf[DISK_NAME_LEN];
4906
4907 while (len && val[len-1] == '\n')
4908 len--;
4909 if (len >= DISK_NAME_LEN)
4910 return -E2BIG;
4911 strlcpy(buf, val, len+1);
4912 if (strncmp(buf, "md_", 3) != 0)
4913 return -EINVAL;
4914 return md_alloc(0, buf);
4915 }
4916
4917 static void md_safemode_timeout(unsigned long data)
4918 {
4919 struct mddev *mddev = (struct mddev *) data;
4920
4921 if (!atomic_read(&mddev->writes_pending)) {
4922 mddev->safemode = 1;
4923 if (mddev->external)
4924 sysfs_notify_dirent_safe(mddev->sysfs_state);
4925 }
4926 md_wakeup_thread(mddev->thread);
4927 }
4928
4929 static int start_dirty_degraded;
4930
4931 int md_run(struct mddev *mddev)
4932 {
4933 int err;
4934 struct md_rdev *rdev;
4935 struct md_personality *pers;
4936
4937 if (list_empty(&mddev->disks))
4938 /* cannot run an array with no devices.. */
4939 return -EINVAL;
4940
4941 if (mddev->pers)
4942 return -EBUSY;
4943 /* Cannot run until previous stop completes properly */
4944 if (mddev->sysfs_active)
4945 return -EBUSY;
4946
4947 /*
4948 * Analyze all RAID superblock(s)
4949 */
4950 if (!mddev->raid_disks) {
4951 if (!mddev->persistent)
4952 return -EINVAL;
4953 analyze_sbs(mddev);
4954 }
4955
4956 if (mddev->level != LEVEL_NONE)
4957 request_module("md-level-%d", mddev->level);
4958 else if (mddev->clevel[0])
4959 request_module("md-%s", mddev->clevel);
4960
4961 /*
4962 * Drop all container device buffers, from now on
4963 * the only valid external interface is through the md
4964 * device.
4965 */
4966 rdev_for_each(rdev, mddev) {
4967 if (test_bit(Faulty, &rdev->flags))
4968 continue;
4969 sync_blockdev(rdev->bdev);
4970 invalidate_bdev(rdev->bdev);
4971
4972 /* perform some consistency tests on the device.
4973 * We don't want the data to overlap the metadata,
4974 * Internal Bitmap issues have been handled elsewhere.
4975 */
4976 if (rdev->meta_bdev) {
4977 /* Nothing to check */;
4978 } else if (rdev->data_offset < rdev->sb_start) {
4979 if (mddev->dev_sectors &&
4980 rdev->data_offset + mddev->dev_sectors
4981 > rdev->sb_start) {
4982 printk("md: %s: data overlaps metadata\n",
4983 mdname(mddev));
4984 return -EINVAL;
4985 }
4986 } else {
4987 if (rdev->sb_start + rdev->sb_size/512
4988 > rdev->data_offset) {
4989 printk("md: %s: metadata overlaps data\n",
4990 mdname(mddev));
4991 return -EINVAL;
4992 }
4993 }
4994 sysfs_notify_dirent_safe(rdev->sysfs_state);
4995 }
4996
4997 if (mddev->bio_set == NULL)
4998 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
4999
5000 spin_lock(&pers_lock);
5001 pers = find_pers(mddev->level, mddev->clevel);
5002 if (!pers || !try_module_get(pers->owner)) {
5003 spin_unlock(&pers_lock);
5004 if (mddev->level != LEVEL_NONE)
5005 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5006 mddev->level);
5007 else
5008 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5009 mddev->clevel);
5010 return -EINVAL;
5011 }
5012 spin_unlock(&pers_lock);
5013 if (mddev->level != pers->level) {
5014 mddev->level = pers->level;
5015 mddev->new_level = pers->level;
5016 }
5017 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5018
5019 if (mddev->reshape_position != MaxSector &&
5020 pers->start_reshape == NULL) {
5021 /* This personality cannot handle reshaping... */
5022 module_put(pers->owner);
5023 return -EINVAL;
5024 }
5025
5026 if (pers->sync_request) {
5027 /* Warn if this is a potentially silly
5028 * configuration.
5029 */
5030 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5031 struct md_rdev *rdev2;
5032 int warned = 0;
5033
5034 rdev_for_each(rdev, mddev)
5035 rdev_for_each(rdev2, mddev) {
5036 if (rdev < rdev2 &&
5037 rdev->bdev->bd_contains ==
5038 rdev2->bdev->bd_contains) {
5039 printk(KERN_WARNING
5040 "%s: WARNING: %s appears to be"
5041 " on the same physical disk as"
5042 " %s.\n",
5043 mdname(mddev),
5044 bdevname(rdev->bdev,b),
5045 bdevname(rdev2->bdev,b2));
5046 warned = 1;
5047 }
5048 }
5049
5050 if (warned)
5051 printk(KERN_WARNING
5052 "True protection against single-disk"
5053 " failure might be compromised.\n");
5054 }
5055
5056 mddev->recovery = 0;
5057 /* may be over-ridden by personality */
5058 mddev->resync_max_sectors = mddev->dev_sectors;
5059
5060 mddev->ok_start_degraded = start_dirty_degraded;
5061
5062 if (start_readonly && mddev->ro == 0)
5063 mddev->ro = 2; /* read-only, but switch on first write */
5064
5065 err = pers->run(mddev);
5066 if (err)
5067 printk(KERN_ERR "md: pers->run() failed ...\n");
5068 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5069 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5070 " but 'external_size' not in effect?\n", __func__);
5071 printk(KERN_ERR
5072 "md: invalid array_size %llu > default size %llu\n",
5073 (unsigned long long)mddev->array_sectors / 2,
5074 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5075 err = -EINVAL;
5076 }
5077 if (err == 0 && pers->sync_request &&
5078 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5079 err = bitmap_create(mddev);
5080 if (err)
5081 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5082 mdname(mddev), err);
5083 }
5084 if (err) {
5085 mddev_detach(mddev);
5086 pers->free(mddev, mddev->private);
5087 module_put(pers->owner);
5088 bitmap_destroy(mddev);
5089 return err;
5090 }
5091 if (mddev->queue) {
5092 mddev->queue->backing_dev_info.congested_data = mddev;
5093 mddev->queue->backing_dev_info.congested_fn = md_congested;
5094 blk_queue_merge_bvec(mddev->queue, md_mergeable_bvec);
5095 }
5096 if (pers->sync_request) {
5097 if (mddev->kobj.sd &&
5098 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5099 printk(KERN_WARNING
5100 "md: cannot register extra attributes for %s\n",
5101 mdname(mddev));
5102 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5103 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5104 mddev->ro = 0;
5105
5106 atomic_set(&mddev->writes_pending,0);
5107 atomic_set(&mddev->max_corr_read_errors,
5108 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5109 mddev->safemode = 0;
5110 mddev->safemode_timer.function = md_safemode_timeout;
5111 mddev->safemode_timer.data = (unsigned long) mddev;
5112 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5113 mddev->in_sync = 1;
5114 smp_wmb();
5115 spin_lock(&mddev->lock);
5116 mddev->pers = pers;
5117 mddev->ready = 1;
5118 spin_unlock(&mddev->lock);
5119 rdev_for_each(rdev, mddev)
5120 if (rdev->raid_disk >= 0)
5121 if (sysfs_link_rdev(mddev, rdev))
5122 /* failure here is OK */;
5123
5124 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5125
5126 if (mddev->flags & MD_UPDATE_SB_FLAGS)
5127 md_update_sb(mddev, 0);
5128
5129 md_new_event(mddev);
5130 sysfs_notify_dirent_safe(mddev->sysfs_state);
5131 sysfs_notify_dirent_safe(mddev->sysfs_action);
5132 sysfs_notify(&mddev->kobj, NULL, "degraded");
5133 return 0;
5134 }
5135 EXPORT_SYMBOL_GPL(md_run);
5136
5137 static int do_md_run(struct mddev *mddev)
5138 {
5139 int err;
5140
5141 err = md_run(mddev);
5142 if (err)
5143 goto out;
5144 err = bitmap_load(mddev);
5145 if (err) {
5146 bitmap_destroy(mddev);
5147 goto out;
5148 }
5149
5150 md_wakeup_thread(mddev->thread);
5151 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5152
5153 set_capacity(mddev->gendisk, mddev->array_sectors);
5154 revalidate_disk(mddev->gendisk);
5155 mddev->changed = 1;
5156 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5157 out:
5158 return err;
5159 }
5160
5161 static int restart_array(struct mddev *mddev)
5162 {
5163 struct gendisk *disk = mddev->gendisk;
5164
5165 /* Complain if it has no devices */
5166 if (list_empty(&mddev->disks))
5167 return -ENXIO;
5168 if (!mddev->pers)
5169 return -EINVAL;
5170 if (!mddev->ro)
5171 return -EBUSY;
5172 mddev->safemode = 0;
5173 mddev->ro = 0;
5174 set_disk_ro(disk, 0);
5175 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5176 mdname(mddev));
5177 /* Kick recovery or resync if necessary */
5178 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5179 md_wakeup_thread(mddev->thread);
5180 md_wakeup_thread(mddev->sync_thread);
5181 sysfs_notify_dirent_safe(mddev->sysfs_state);
5182 return 0;
5183 }
5184
5185 static void md_clean(struct mddev *mddev)
5186 {
5187 mddev->array_sectors = 0;
5188 mddev->external_size = 0;
5189 mddev->dev_sectors = 0;
5190 mddev->raid_disks = 0;
5191 mddev->recovery_cp = 0;
5192 mddev->resync_min = 0;
5193 mddev->resync_max = MaxSector;
5194 mddev->reshape_position = MaxSector;
5195 mddev->external = 0;
5196 mddev->persistent = 0;
5197 mddev->level = LEVEL_NONE;
5198 mddev->clevel[0] = 0;
5199 mddev->flags = 0;
5200 mddev->ro = 0;
5201 mddev->metadata_type[0] = 0;
5202 mddev->chunk_sectors = 0;
5203 mddev->ctime = mddev->utime = 0;
5204 mddev->layout = 0;
5205 mddev->max_disks = 0;
5206 mddev->events = 0;
5207 mddev->can_decrease_events = 0;
5208 mddev->delta_disks = 0;
5209 mddev->reshape_backwards = 0;
5210 mddev->new_level = LEVEL_NONE;
5211 mddev->new_layout = 0;
5212 mddev->new_chunk_sectors = 0;
5213 mddev->curr_resync = 0;
5214 atomic64_set(&mddev->resync_mismatches, 0);
5215 mddev->suspend_lo = mddev->suspend_hi = 0;
5216 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5217 mddev->recovery = 0;
5218 mddev->in_sync = 0;
5219 mddev->changed = 0;
5220 mddev->degraded = 0;
5221 mddev->safemode = 0;
5222 mddev->merge_check_needed = 0;
5223 mddev->bitmap_info.offset = 0;
5224 mddev->bitmap_info.default_offset = 0;
5225 mddev->bitmap_info.default_space = 0;
5226 mddev->bitmap_info.chunksize = 0;
5227 mddev->bitmap_info.daemon_sleep = 0;
5228 mddev->bitmap_info.max_write_behind = 0;
5229 }
5230
5231 static void __md_stop_writes(struct mddev *mddev)
5232 {
5233 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5234 flush_workqueue(md_misc_wq);
5235 if (mddev->sync_thread) {
5236 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5237 md_reap_sync_thread(mddev);
5238 }
5239
5240 del_timer_sync(&mddev->safemode_timer);
5241
5242 bitmap_flush(mddev);
5243 md_super_wait(mddev);
5244
5245 if (mddev->ro == 0 &&
5246 (!mddev->in_sync || (mddev->flags & MD_UPDATE_SB_FLAGS))) {
5247 /* mark array as shutdown cleanly */
5248 mddev->in_sync = 1;
5249 md_update_sb(mddev, 1);
5250 }
5251 }
5252
5253 void md_stop_writes(struct mddev *mddev)
5254 {
5255 mddev_lock_nointr(mddev);
5256 __md_stop_writes(mddev);
5257 mddev_unlock(mddev);
5258 }
5259 EXPORT_SYMBOL_GPL(md_stop_writes);
5260
5261 static void mddev_detach(struct mddev *mddev)
5262 {
5263 struct bitmap *bitmap = mddev->bitmap;
5264 /* wait for behind writes to complete */
5265 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
5266 printk(KERN_INFO "md:%s: behind writes in progress - waiting to stop.\n",
5267 mdname(mddev));
5268 /* need to kick something here to make sure I/O goes? */
5269 wait_event(bitmap->behind_wait,
5270 atomic_read(&bitmap->behind_writes) == 0);
5271 }
5272 if (mddev->pers && mddev->pers->quiesce) {
5273 mddev->pers->quiesce(mddev, 1);
5274 mddev->pers->quiesce(mddev, 0);
5275 }
5276 md_unregister_thread(&mddev->thread);
5277 if (mddev->queue)
5278 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5279 }
5280
5281 static void __md_stop(struct mddev *mddev)
5282 {
5283 struct md_personality *pers = mddev->pers;
5284 mddev_detach(mddev);
5285 spin_lock(&mddev->lock);
5286 mddev->ready = 0;
5287 mddev->pers = NULL;
5288 spin_unlock(&mddev->lock);
5289 pers->free(mddev, mddev->private);
5290 if (pers->sync_request && mddev->to_remove == NULL)
5291 mddev->to_remove = &md_redundancy_group;
5292 module_put(pers->owner);
5293 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5294 }
5295
5296 void md_stop(struct mddev *mddev)
5297 {
5298 /* stop the array and free an attached data structures.
5299 * This is called from dm-raid
5300 */
5301 __md_stop(mddev);
5302 bitmap_destroy(mddev);
5303 if (mddev->bio_set)
5304 bioset_free(mddev->bio_set);
5305 }
5306
5307 EXPORT_SYMBOL_GPL(md_stop);
5308
5309 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5310 {
5311 int err = 0;
5312 int did_freeze = 0;
5313
5314 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5315 did_freeze = 1;
5316 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5317 md_wakeup_thread(mddev->thread);
5318 }
5319 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5320 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5321 if (mddev->sync_thread)
5322 /* Thread might be blocked waiting for metadata update
5323 * which will now never happen */
5324 wake_up_process(mddev->sync_thread->tsk);
5325
5326 mddev_unlock(mddev);
5327 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5328 &mddev->recovery));
5329 mddev_lock_nointr(mddev);
5330
5331 mutex_lock(&mddev->open_mutex);
5332 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5333 mddev->sync_thread ||
5334 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5335 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5336 printk("md: %s still in use.\n",mdname(mddev));
5337 if (did_freeze) {
5338 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5339 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5340 md_wakeup_thread(mddev->thread);
5341 }
5342 err = -EBUSY;
5343 goto out;
5344 }
5345 if (mddev->pers) {
5346 __md_stop_writes(mddev);
5347
5348 err = -ENXIO;
5349 if (mddev->ro==1)
5350 goto out;
5351 mddev->ro = 1;
5352 set_disk_ro(mddev->gendisk, 1);
5353 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5354 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5355 md_wakeup_thread(mddev->thread);
5356 sysfs_notify_dirent_safe(mddev->sysfs_state);
5357 err = 0;
5358 }
5359 out:
5360 mutex_unlock(&mddev->open_mutex);
5361 return err;
5362 }
5363
5364 /* mode:
5365 * 0 - completely stop and dis-assemble array
5366 * 2 - stop but do not disassemble array
5367 */
5368 static int do_md_stop(struct mddev *mddev, int mode,
5369 struct block_device *bdev)
5370 {
5371 struct gendisk *disk = mddev->gendisk;
5372 struct md_rdev *rdev;
5373 int did_freeze = 0;
5374
5375 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5376 did_freeze = 1;
5377 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5378 md_wakeup_thread(mddev->thread);
5379 }
5380 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5381 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5382 if (mddev->sync_thread)
5383 /* Thread might be blocked waiting for metadata update
5384 * which will now never happen */
5385 wake_up_process(mddev->sync_thread->tsk);
5386
5387 mddev_unlock(mddev);
5388 wait_event(resync_wait, (mddev->sync_thread == NULL &&
5389 !test_bit(MD_RECOVERY_RUNNING,
5390 &mddev->recovery)));
5391 mddev_lock_nointr(mddev);
5392
5393 mutex_lock(&mddev->open_mutex);
5394 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5395 mddev->sysfs_active ||
5396 mddev->sync_thread ||
5397 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5398 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5399 printk("md: %s still in use.\n",mdname(mddev));
5400 mutex_unlock(&mddev->open_mutex);
5401 if (did_freeze) {
5402 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5403 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5404 md_wakeup_thread(mddev->thread);
5405 }
5406 return -EBUSY;
5407 }
5408 if (mddev->pers) {
5409 if (mddev->ro)
5410 set_disk_ro(disk, 0);
5411
5412 __md_stop_writes(mddev);
5413 __md_stop(mddev);
5414 mddev->queue->merge_bvec_fn = NULL;
5415 mddev->queue->backing_dev_info.congested_fn = NULL;
5416
5417 /* tell userspace to handle 'inactive' */
5418 sysfs_notify_dirent_safe(mddev->sysfs_state);
5419
5420 rdev_for_each(rdev, mddev)
5421 if (rdev->raid_disk >= 0)
5422 sysfs_unlink_rdev(mddev, rdev);
5423
5424 set_capacity(disk, 0);
5425 mutex_unlock(&mddev->open_mutex);
5426 mddev->changed = 1;
5427 revalidate_disk(disk);
5428
5429 if (mddev->ro)
5430 mddev->ro = 0;
5431 } else
5432 mutex_unlock(&mddev->open_mutex);
5433 /*
5434 * Free resources if final stop
5435 */
5436 if (mode == 0) {
5437 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5438
5439 bitmap_destroy(mddev);
5440 if (mddev->bitmap_info.file) {
5441 struct file *f = mddev->bitmap_info.file;
5442 spin_lock(&mddev->lock);
5443 mddev->bitmap_info.file = NULL;
5444 spin_unlock(&mddev->lock);
5445 fput(f);
5446 }
5447 mddev->bitmap_info.offset = 0;
5448
5449 export_array(mddev);
5450
5451 md_clean(mddev);
5452 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5453 if (mddev->hold_active == UNTIL_STOP)
5454 mddev->hold_active = 0;
5455 }
5456 blk_integrity_unregister(disk);
5457 md_new_event(mddev);
5458 sysfs_notify_dirent_safe(mddev->sysfs_state);
5459 return 0;
5460 }
5461
5462 #ifndef MODULE
5463 static void autorun_array(struct mddev *mddev)
5464 {
5465 struct md_rdev *rdev;
5466 int err;
5467
5468 if (list_empty(&mddev->disks))
5469 return;
5470
5471 printk(KERN_INFO "md: running: ");
5472
5473 rdev_for_each(rdev, mddev) {
5474 char b[BDEVNAME_SIZE];
5475 printk("<%s>", bdevname(rdev->bdev,b));
5476 }
5477 printk("\n");
5478
5479 err = do_md_run(mddev);
5480 if (err) {
5481 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
5482 do_md_stop(mddev, 0, NULL);
5483 }
5484 }
5485
5486 /*
5487 * lets try to run arrays based on all disks that have arrived
5488 * until now. (those are in pending_raid_disks)
5489 *
5490 * the method: pick the first pending disk, collect all disks with
5491 * the same UUID, remove all from the pending list and put them into
5492 * the 'same_array' list. Then order this list based on superblock
5493 * update time (freshest comes first), kick out 'old' disks and
5494 * compare superblocks. If everything's fine then run it.
5495 *
5496 * If "unit" is allocated, then bump its reference count
5497 */
5498 static void autorun_devices(int part)
5499 {
5500 struct md_rdev *rdev0, *rdev, *tmp;
5501 struct mddev *mddev;
5502 char b[BDEVNAME_SIZE];
5503
5504 printk(KERN_INFO "md: autorun ...\n");
5505 while (!list_empty(&pending_raid_disks)) {
5506 int unit;
5507 dev_t dev;
5508 LIST_HEAD(candidates);
5509 rdev0 = list_entry(pending_raid_disks.next,
5510 struct md_rdev, same_set);
5511
5512 printk(KERN_INFO "md: considering %s ...\n",
5513 bdevname(rdev0->bdev,b));
5514 INIT_LIST_HEAD(&candidates);
5515 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5516 if (super_90_load(rdev, rdev0, 0) >= 0) {
5517 printk(KERN_INFO "md: adding %s ...\n",
5518 bdevname(rdev->bdev,b));
5519 list_move(&rdev->same_set, &candidates);
5520 }
5521 /*
5522 * now we have a set of devices, with all of them having
5523 * mostly sane superblocks. It's time to allocate the
5524 * mddev.
5525 */
5526 if (part) {
5527 dev = MKDEV(mdp_major,
5528 rdev0->preferred_minor << MdpMinorShift);
5529 unit = MINOR(dev) >> MdpMinorShift;
5530 } else {
5531 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5532 unit = MINOR(dev);
5533 }
5534 if (rdev0->preferred_minor != unit) {
5535 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5536 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5537 break;
5538 }
5539
5540 md_probe(dev, NULL, NULL);
5541 mddev = mddev_find(dev);
5542 if (!mddev || !mddev->gendisk) {
5543 if (mddev)
5544 mddev_put(mddev);
5545 printk(KERN_ERR
5546 "md: cannot allocate memory for md drive.\n");
5547 break;
5548 }
5549 if (mddev_lock(mddev))
5550 printk(KERN_WARNING "md: %s locked, cannot run\n",
5551 mdname(mddev));
5552 else if (mddev->raid_disks || mddev->major_version
5553 || !list_empty(&mddev->disks)) {
5554 printk(KERN_WARNING
5555 "md: %s already running, cannot run %s\n",
5556 mdname(mddev), bdevname(rdev0->bdev,b));
5557 mddev_unlock(mddev);
5558 } else {
5559 printk(KERN_INFO "md: created %s\n", mdname(mddev));
5560 mddev->persistent = 1;
5561 rdev_for_each_list(rdev, tmp, &candidates) {
5562 list_del_init(&rdev->same_set);
5563 if (bind_rdev_to_array(rdev, mddev))
5564 export_rdev(rdev);
5565 }
5566 autorun_array(mddev);
5567 mddev_unlock(mddev);
5568 }
5569 /* on success, candidates will be empty, on error
5570 * it won't...
5571 */
5572 rdev_for_each_list(rdev, tmp, &candidates) {
5573 list_del_init(&rdev->same_set);
5574 export_rdev(rdev);
5575 }
5576 mddev_put(mddev);
5577 }
5578 printk(KERN_INFO "md: ... autorun DONE.\n");
5579 }
5580 #endif /* !MODULE */
5581
5582 static int get_version(void __user *arg)
5583 {
5584 mdu_version_t ver;
5585
5586 ver.major = MD_MAJOR_VERSION;
5587 ver.minor = MD_MINOR_VERSION;
5588 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5589
5590 if (copy_to_user(arg, &ver, sizeof(ver)))
5591 return -EFAULT;
5592
5593 return 0;
5594 }
5595
5596 static int get_array_info(struct mddev *mddev, void __user *arg)
5597 {
5598 mdu_array_info_t info;
5599 int nr,working,insync,failed,spare;
5600 struct md_rdev *rdev;
5601
5602 nr = working = insync = failed = spare = 0;
5603 rcu_read_lock();
5604 rdev_for_each_rcu(rdev, mddev) {
5605 nr++;
5606 if (test_bit(Faulty, &rdev->flags))
5607 failed++;
5608 else {
5609 working++;
5610 if (test_bit(In_sync, &rdev->flags))
5611 insync++;
5612 else
5613 spare++;
5614 }
5615 }
5616 rcu_read_unlock();
5617
5618 info.major_version = mddev->major_version;
5619 info.minor_version = mddev->minor_version;
5620 info.patch_version = MD_PATCHLEVEL_VERSION;
5621 info.ctime = mddev->ctime;
5622 info.level = mddev->level;
5623 info.size = mddev->dev_sectors / 2;
5624 if (info.size != mddev->dev_sectors / 2) /* overflow */
5625 info.size = -1;
5626 info.nr_disks = nr;
5627 info.raid_disks = mddev->raid_disks;
5628 info.md_minor = mddev->md_minor;
5629 info.not_persistent= !mddev->persistent;
5630
5631 info.utime = mddev->utime;
5632 info.state = 0;
5633 if (mddev->in_sync)
5634 info.state = (1<<MD_SB_CLEAN);
5635 if (mddev->bitmap && mddev->bitmap_info.offset)
5636 info.state |= (1<<MD_SB_BITMAP_PRESENT);
5637 if (mddev_is_clustered(mddev))
5638 info.state |= (1<<MD_SB_CLUSTERED);
5639
5640 info.active_disks = insync;
5641 info.working_disks = working;
5642 info.failed_disks = failed;
5643 info.spare_disks = spare;
5644
5645 info.layout = mddev->layout;
5646 info.chunk_size = mddev->chunk_sectors << 9;
5647
5648 if (copy_to_user(arg, &info, sizeof(info)))
5649 return -EFAULT;
5650
5651 return 0;
5652 }
5653
5654 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
5655 {
5656 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5657 char *ptr;
5658 int err;
5659
5660 file = kmalloc(sizeof(*file), GFP_NOIO);
5661 if (!file)
5662 return -ENOMEM;
5663
5664 err = 0;
5665 spin_lock(&mddev->lock);
5666 /* bitmap disabled, zero the first byte and copy out */
5667 if (!mddev->bitmap_info.file)
5668 file->pathname[0] = '\0';
5669 else if ((ptr = d_path(&mddev->bitmap_info.file->f_path,
5670 file->pathname, sizeof(file->pathname))),
5671 IS_ERR(ptr))
5672 err = PTR_ERR(ptr);
5673 else
5674 memmove(file->pathname, ptr,
5675 sizeof(file->pathname)-(ptr-file->pathname));
5676 spin_unlock(&mddev->lock);
5677
5678 if (err == 0 &&
5679 copy_to_user(arg, file, sizeof(*file)))
5680 err = -EFAULT;
5681
5682 kfree(file);
5683 return err;
5684 }
5685
5686 static int get_disk_info(struct mddev *mddev, void __user * arg)
5687 {
5688 mdu_disk_info_t info;
5689 struct md_rdev *rdev;
5690
5691 if (copy_from_user(&info, arg, sizeof(info)))
5692 return -EFAULT;
5693
5694 rcu_read_lock();
5695 rdev = find_rdev_nr_rcu(mddev, info.number);
5696 if (rdev) {
5697 info.major = MAJOR(rdev->bdev->bd_dev);
5698 info.minor = MINOR(rdev->bdev->bd_dev);
5699 info.raid_disk = rdev->raid_disk;
5700 info.state = 0;
5701 if (test_bit(Faulty, &rdev->flags))
5702 info.state |= (1<<MD_DISK_FAULTY);
5703 else if (test_bit(In_sync, &rdev->flags)) {
5704 info.state |= (1<<MD_DISK_ACTIVE);
5705 info.state |= (1<<MD_DISK_SYNC);
5706 }
5707 if (test_bit(WriteMostly, &rdev->flags))
5708 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5709 } else {
5710 info.major = info.minor = 0;
5711 info.raid_disk = -1;
5712 info.state = (1<<MD_DISK_REMOVED);
5713 }
5714 rcu_read_unlock();
5715
5716 if (copy_to_user(arg, &info, sizeof(info)))
5717 return -EFAULT;
5718
5719 return 0;
5720 }
5721
5722 static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
5723 {
5724 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5725 struct md_rdev *rdev;
5726 dev_t dev = MKDEV(info->major,info->minor);
5727
5728 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5729 return -EOVERFLOW;
5730
5731 if (!mddev->raid_disks) {
5732 int err;
5733 /* expecting a device which has a superblock */
5734 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5735 if (IS_ERR(rdev)) {
5736 printk(KERN_WARNING
5737 "md: md_import_device returned %ld\n",
5738 PTR_ERR(rdev));
5739 return PTR_ERR(rdev);
5740 }
5741 if (!list_empty(&mddev->disks)) {
5742 struct md_rdev *rdev0
5743 = list_entry(mddev->disks.next,
5744 struct md_rdev, same_set);
5745 err = super_types[mddev->major_version]
5746 .load_super(rdev, rdev0, mddev->minor_version);
5747 if (err < 0) {
5748 printk(KERN_WARNING
5749 "md: %s has different UUID to %s\n",
5750 bdevname(rdev->bdev,b),
5751 bdevname(rdev0->bdev,b2));
5752 export_rdev(rdev);
5753 return -EINVAL;
5754 }
5755 }
5756 err = bind_rdev_to_array(rdev, mddev);
5757 if (err)
5758 export_rdev(rdev);
5759 return err;
5760 }
5761
5762 /*
5763 * add_new_disk can be used once the array is assembled
5764 * to add "hot spares". They must already have a superblock
5765 * written
5766 */
5767 if (mddev->pers) {
5768 int err;
5769 if (!mddev->pers->hot_add_disk) {
5770 printk(KERN_WARNING
5771 "%s: personality does not support diskops!\n",
5772 mdname(mddev));
5773 return -EINVAL;
5774 }
5775 if (mddev->persistent)
5776 rdev = md_import_device(dev, mddev->major_version,
5777 mddev->minor_version);
5778 else
5779 rdev = md_import_device(dev, -1, -1);
5780 if (IS_ERR(rdev)) {
5781 printk(KERN_WARNING
5782 "md: md_import_device returned %ld\n",
5783 PTR_ERR(rdev));
5784 return PTR_ERR(rdev);
5785 }
5786 /* set saved_raid_disk if appropriate */
5787 if (!mddev->persistent) {
5788 if (info->state & (1<<MD_DISK_SYNC) &&
5789 info->raid_disk < mddev->raid_disks) {
5790 rdev->raid_disk = info->raid_disk;
5791 set_bit(In_sync, &rdev->flags);
5792 clear_bit(Bitmap_sync, &rdev->flags);
5793 } else
5794 rdev->raid_disk = -1;
5795 rdev->saved_raid_disk = rdev->raid_disk;
5796 } else
5797 super_types[mddev->major_version].
5798 validate_super(mddev, rdev);
5799 if ((info->state & (1<<MD_DISK_SYNC)) &&
5800 rdev->raid_disk != info->raid_disk) {
5801 /* This was a hot-add request, but events doesn't
5802 * match, so reject it.
5803 */
5804 export_rdev(rdev);
5805 return -EINVAL;
5806 }
5807
5808 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5809 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5810 set_bit(WriteMostly, &rdev->flags);
5811 else
5812 clear_bit(WriteMostly, &rdev->flags);
5813
5814 rdev->raid_disk = -1;
5815 err = bind_rdev_to_array(rdev, mddev);
5816 if (!err && !mddev->pers->hot_remove_disk) {
5817 /* If there is hot_add_disk but no hot_remove_disk
5818 * then added disks for geometry changes,
5819 * and should be added immediately.
5820 */
5821 super_types[mddev->major_version].
5822 validate_super(mddev, rdev);
5823 err = mddev->pers->hot_add_disk(mddev, rdev);
5824 if (err)
5825 unbind_rdev_from_array(rdev);
5826 }
5827 if (err)
5828 export_rdev(rdev);
5829 else
5830 sysfs_notify_dirent_safe(rdev->sysfs_state);
5831
5832 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5833 if (mddev->degraded)
5834 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5835 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5836 if (!err)
5837 md_new_event(mddev);
5838 md_wakeup_thread(mddev->thread);
5839 return err;
5840 }
5841
5842 /* otherwise, add_new_disk is only allowed
5843 * for major_version==0 superblocks
5844 */
5845 if (mddev->major_version != 0) {
5846 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5847 mdname(mddev));
5848 return -EINVAL;
5849 }
5850
5851 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5852 int err;
5853 rdev = md_import_device(dev, -1, 0);
5854 if (IS_ERR(rdev)) {
5855 printk(KERN_WARNING
5856 "md: error, md_import_device() returned %ld\n",
5857 PTR_ERR(rdev));
5858 return PTR_ERR(rdev);
5859 }
5860 rdev->desc_nr = info->number;
5861 if (info->raid_disk < mddev->raid_disks)
5862 rdev->raid_disk = info->raid_disk;
5863 else
5864 rdev->raid_disk = -1;
5865
5866 if (rdev->raid_disk < mddev->raid_disks)
5867 if (info->state & (1<<MD_DISK_SYNC))
5868 set_bit(In_sync, &rdev->flags);
5869
5870 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5871 set_bit(WriteMostly, &rdev->flags);
5872
5873 if (!mddev->persistent) {
5874 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5875 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5876 } else
5877 rdev->sb_start = calc_dev_sboffset(rdev);
5878 rdev->sectors = rdev->sb_start;
5879
5880 err = bind_rdev_to_array(rdev, mddev);
5881 if (err) {
5882 export_rdev(rdev);
5883 return err;
5884 }
5885 }
5886
5887 return 0;
5888 }
5889
5890 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
5891 {
5892 char b[BDEVNAME_SIZE];
5893 struct md_rdev *rdev;
5894
5895 rdev = find_rdev(mddev, dev);
5896 if (!rdev)
5897 return -ENXIO;
5898
5899 clear_bit(Blocked, &rdev->flags);
5900 remove_and_add_spares(mddev, rdev);
5901
5902 if (rdev->raid_disk >= 0)
5903 goto busy;
5904
5905 kick_rdev_from_array(rdev);
5906 md_update_sb(mddev, 1);
5907 md_new_event(mddev);
5908
5909 return 0;
5910 busy:
5911 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5912 bdevname(rdev->bdev,b), mdname(mddev));
5913 return -EBUSY;
5914 }
5915
5916 static int hot_add_disk(struct mddev *mddev, dev_t dev)
5917 {
5918 char b[BDEVNAME_SIZE];
5919 int err;
5920 struct md_rdev *rdev;
5921
5922 if (!mddev->pers)
5923 return -ENODEV;
5924
5925 if (mddev->major_version != 0) {
5926 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5927 " version-0 superblocks.\n",
5928 mdname(mddev));
5929 return -EINVAL;
5930 }
5931 if (!mddev->pers->hot_add_disk) {
5932 printk(KERN_WARNING
5933 "%s: personality does not support diskops!\n",
5934 mdname(mddev));
5935 return -EINVAL;
5936 }
5937
5938 rdev = md_import_device(dev, -1, 0);
5939 if (IS_ERR(rdev)) {
5940 printk(KERN_WARNING
5941 "md: error, md_import_device() returned %ld\n",
5942 PTR_ERR(rdev));
5943 return -EINVAL;
5944 }
5945
5946 if (mddev->persistent)
5947 rdev->sb_start = calc_dev_sboffset(rdev);
5948 else
5949 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5950
5951 rdev->sectors = rdev->sb_start;
5952
5953 if (test_bit(Faulty, &rdev->flags)) {
5954 printk(KERN_WARNING
5955 "md: can not hot-add faulty %s disk to %s!\n",
5956 bdevname(rdev->bdev,b), mdname(mddev));
5957 err = -EINVAL;
5958 goto abort_export;
5959 }
5960 clear_bit(In_sync, &rdev->flags);
5961 rdev->desc_nr = -1;
5962 rdev->saved_raid_disk = -1;
5963 err = bind_rdev_to_array(rdev, mddev);
5964 if (err)
5965 goto abort_export;
5966
5967 /*
5968 * The rest should better be atomic, we can have disk failures
5969 * noticed in interrupt contexts ...
5970 */
5971
5972 rdev->raid_disk = -1;
5973
5974 md_update_sb(mddev, 1);
5975
5976 /*
5977 * Kick recovery, maybe this spare has to be added to the
5978 * array immediately.
5979 */
5980 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5981 md_wakeup_thread(mddev->thread);
5982 md_new_event(mddev);
5983 return 0;
5984
5985 abort_export:
5986 export_rdev(rdev);
5987 return err;
5988 }
5989
5990 static int set_bitmap_file(struct mddev *mddev, int fd)
5991 {
5992 int err = 0;
5993
5994 if (mddev->pers) {
5995 if (!mddev->pers->quiesce || !mddev->thread)
5996 return -EBUSY;
5997 if (mddev->recovery || mddev->sync_thread)
5998 return -EBUSY;
5999 /* we should be able to change the bitmap.. */
6000 }
6001
6002 if (fd >= 0) {
6003 struct inode *inode;
6004 struct file *f;
6005
6006 if (mddev->bitmap || mddev->bitmap_info.file)
6007 return -EEXIST; /* cannot add when bitmap is present */
6008 f = fget(fd);
6009
6010 if (f == NULL) {
6011 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
6012 mdname(mddev));
6013 return -EBADF;
6014 }
6015
6016 inode = f->f_mapping->host;
6017 if (!S_ISREG(inode->i_mode)) {
6018 printk(KERN_ERR "%s: error: bitmap file must be a regular file\n",
6019 mdname(mddev));
6020 err = -EBADF;
6021 } else if (!(f->f_mode & FMODE_WRITE)) {
6022 printk(KERN_ERR "%s: error: bitmap file must open for write\n",
6023 mdname(mddev));
6024 err = -EBADF;
6025 } else if (atomic_read(&inode->i_writecount) != 1) {
6026 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
6027 mdname(mddev));
6028 err = -EBUSY;
6029 }
6030 if (err) {
6031 fput(f);
6032 return err;
6033 }
6034 mddev->bitmap_info.file = f;
6035 mddev->bitmap_info.offset = 0; /* file overrides offset */
6036 } else if (mddev->bitmap == NULL)
6037 return -ENOENT; /* cannot remove what isn't there */
6038 err = 0;
6039 if (mddev->pers) {
6040 mddev->pers->quiesce(mddev, 1);
6041 if (fd >= 0) {
6042 err = bitmap_create(mddev);
6043 if (!err)
6044 err = bitmap_load(mddev);
6045 }
6046 if (fd < 0 || err) {
6047 bitmap_destroy(mddev);
6048 fd = -1; /* make sure to put the file */
6049 }
6050 mddev->pers->quiesce(mddev, 0);
6051 }
6052 if (fd < 0) {
6053 struct file *f = mddev->bitmap_info.file;
6054 if (f) {
6055 spin_lock(&mddev->lock);
6056 mddev->bitmap_info.file = NULL;
6057 spin_unlock(&mddev->lock);
6058 fput(f);
6059 }
6060 }
6061
6062 return err;
6063 }
6064
6065 /*
6066 * set_array_info is used two different ways
6067 * The original usage is when creating a new array.
6068 * In this usage, raid_disks is > 0 and it together with
6069 * level, size, not_persistent,layout,chunksize determine the
6070 * shape of the array.
6071 * This will always create an array with a type-0.90.0 superblock.
6072 * The newer usage is when assembling an array.
6073 * In this case raid_disks will be 0, and the major_version field is
6074 * use to determine which style super-blocks are to be found on the devices.
6075 * The minor and patch _version numbers are also kept incase the
6076 * super_block handler wishes to interpret them.
6077 */
6078 static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
6079 {
6080
6081 if (info->raid_disks == 0) {
6082 /* just setting version number for superblock loading */
6083 if (info->major_version < 0 ||
6084 info->major_version >= ARRAY_SIZE(super_types) ||
6085 super_types[info->major_version].name == NULL) {
6086 /* maybe try to auto-load a module? */
6087 printk(KERN_INFO
6088 "md: superblock version %d not known\n",
6089 info->major_version);
6090 return -EINVAL;
6091 }
6092 mddev->major_version = info->major_version;
6093 mddev->minor_version = info->minor_version;
6094 mddev->patch_version = info->patch_version;
6095 mddev->persistent = !info->not_persistent;
6096 /* ensure mddev_put doesn't delete this now that there
6097 * is some minimal configuration.
6098 */
6099 mddev->ctime = get_seconds();
6100 return 0;
6101 }
6102 mddev->major_version = MD_MAJOR_VERSION;
6103 mddev->minor_version = MD_MINOR_VERSION;
6104 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6105 mddev->ctime = get_seconds();
6106
6107 mddev->level = info->level;
6108 mddev->clevel[0] = 0;
6109 mddev->dev_sectors = 2 * (sector_t)info->size;
6110 mddev->raid_disks = info->raid_disks;
6111 /* don't set md_minor, it is determined by which /dev/md* was
6112 * openned
6113 */
6114 if (info->state & (1<<MD_SB_CLEAN))
6115 mddev->recovery_cp = MaxSector;
6116 else
6117 mddev->recovery_cp = 0;
6118 mddev->persistent = ! info->not_persistent;
6119 mddev->external = 0;
6120
6121 mddev->layout = info->layout;
6122 mddev->chunk_sectors = info->chunk_size >> 9;
6123
6124 mddev->max_disks = MD_SB_DISKS;
6125
6126 if (mddev->persistent)
6127 mddev->flags = 0;
6128 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6129
6130 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6131 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6132 mddev->bitmap_info.offset = 0;
6133
6134 mddev->reshape_position = MaxSector;
6135
6136 /*
6137 * Generate a 128 bit UUID
6138 */
6139 get_random_bytes(mddev->uuid, 16);
6140
6141 mddev->new_level = mddev->level;
6142 mddev->new_chunk_sectors = mddev->chunk_sectors;
6143 mddev->new_layout = mddev->layout;
6144 mddev->delta_disks = 0;
6145 mddev->reshape_backwards = 0;
6146
6147 return 0;
6148 }
6149
6150 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6151 {
6152 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6153
6154 if (mddev->external_size)
6155 return;
6156
6157 mddev->array_sectors = array_sectors;
6158 }
6159 EXPORT_SYMBOL(md_set_array_sectors);
6160
6161 static int update_size(struct mddev *mddev, sector_t num_sectors)
6162 {
6163 struct md_rdev *rdev;
6164 int rv;
6165 int fit = (num_sectors == 0);
6166
6167 if (mddev->pers->resize == NULL)
6168 return -EINVAL;
6169 /* The "num_sectors" is the number of sectors of each device that
6170 * is used. This can only make sense for arrays with redundancy.
6171 * linear and raid0 always use whatever space is available. We can only
6172 * consider changing this number if no resync or reconstruction is
6173 * happening, and if the new size is acceptable. It must fit before the
6174 * sb_start or, if that is <data_offset, it must fit before the size
6175 * of each device. If num_sectors is zero, we find the largest size
6176 * that fits.
6177 */
6178 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6179 mddev->sync_thread)
6180 return -EBUSY;
6181 if (mddev->ro)
6182 return -EROFS;
6183
6184 rdev_for_each(rdev, mddev) {
6185 sector_t avail = rdev->sectors;
6186
6187 if (fit && (num_sectors == 0 || num_sectors > avail))
6188 num_sectors = avail;
6189 if (avail < num_sectors)
6190 return -ENOSPC;
6191 }
6192 rv = mddev->pers->resize(mddev, num_sectors);
6193 if (!rv)
6194 revalidate_disk(mddev->gendisk);
6195 return rv;
6196 }
6197
6198 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6199 {
6200 int rv;
6201 struct md_rdev *rdev;
6202 /* change the number of raid disks */
6203 if (mddev->pers->check_reshape == NULL)
6204 return -EINVAL;
6205 if (mddev->ro)
6206 return -EROFS;
6207 if (raid_disks <= 0 ||
6208 (mddev->max_disks && raid_disks >= mddev->max_disks))
6209 return -EINVAL;
6210 if (mddev->sync_thread ||
6211 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6212 mddev->reshape_position != MaxSector)
6213 return -EBUSY;
6214
6215 rdev_for_each(rdev, mddev) {
6216 if (mddev->raid_disks < raid_disks &&
6217 rdev->data_offset < rdev->new_data_offset)
6218 return -EINVAL;
6219 if (mddev->raid_disks > raid_disks &&
6220 rdev->data_offset > rdev->new_data_offset)
6221 return -EINVAL;
6222 }
6223
6224 mddev->delta_disks = raid_disks - mddev->raid_disks;
6225 if (mddev->delta_disks < 0)
6226 mddev->reshape_backwards = 1;
6227 else if (mddev->delta_disks > 0)
6228 mddev->reshape_backwards = 0;
6229
6230 rv = mddev->pers->check_reshape(mddev);
6231 if (rv < 0) {
6232 mddev->delta_disks = 0;
6233 mddev->reshape_backwards = 0;
6234 }
6235 return rv;
6236 }
6237
6238 /*
6239 * update_array_info is used to change the configuration of an
6240 * on-line array.
6241 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6242 * fields in the info are checked against the array.
6243 * Any differences that cannot be handled will cause an error.
6244 * Normally, only one change can be managed at a time.
6245 */
6246 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6247 {
6248 int rv = 0;
6249 int cnt = 0;
6250 int state = 0;
6251
6252 /* calculate expected state,ignoring low bits */
6253 if (mddev->bitmap && mddev->bitmap_info.offset)
6254 state |= (1 << MD_SB_BITMAP_PRESENT);
6255
6256 if (mddev->major_version != info->major_version ||
6257 mddev->minor_version != info->minor_version ||
6258 /* mddev->patch_version != info->patch_version || */
6259 mddev->ctime != info->ctime ||
6260 mddev->level != info->level ||
6261 /* mddev->layout != info->layout || */
6262 !mddev->persistent != info->not_persistent||
6263 mddev->chunk_sectors != info->chunk_size >> 9 ||
6264 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6265 ((state^info->state) & 0xfffffe00)
6266 )
6267 return -EINVAL;
6268 /* Check there is only one change */
6269 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6270 cnt++;
6271 if (mddev->raid_disks != info->raid_disks)
6272 cnt++;
6273 if (mddev->layout != info->layout)
6274 cnt++;
6275 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6276 cnt++;
6277 if (cnt == 0)
6278 return 0;
6279 if (cnt > 1)
6280 return -EINVAL;
6281
6282 if (mddev->layout != info->layout) {
6283 /* Change layout
6284 * we don't need to do anything at the md level, the
6285 * personality will take care of it all.
6286 */
6287 if (mddev->pers->check_reshape == NULL)
6288 return -EINVAL;
6289 else {
6290 mddev->new_layout = info->layout;
6291 rv = mddev->pers->check_reshape(mddev);
6292 if (rv)
6293 mddev->new_layout = mddev->layout;
6294 return rv;
6295 }
6296 }
6297 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6298 rv = update_size(mddev, (sector_t)info->size * 2);
6299
6300 if (mddev->raid_disks != info->raid_disks)
6301 rv = update_raid_disks(mddev, info->raid_disks);
6302
6303 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6304 if (mddev->pers->quiesce == NULL || mddev->thread == NULL)
6305 return -EINVAL;
6306 if (mddev->recovery || mddev->sync_thread)
6307 return -EBUSY;
6308 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6309 /* add the bitmap */
6310 if (mddev->bitmap)
6311 return -EEXIST;
6312 if (mddev->bitmap_info.default_offset == 0)
6313 return -EINVAL;
6314 mddev->bitmap_info.offset =
6315 mddev->bitmap_info.default_offset;
6316 mddev->bitmap_info.space =
6317 mddev->bitmap_info.default_space;
6318 mddev->pers->quiesce(mddev, 1);
6319 rv = bitmap_create(mddev);
6320 if (!rv)
6321 rv = bitmap_load(mddev);
6322 if (rv)
6323 bitmap_destroy(mddev);
6324 mddev->pers->quiesce(mddev, 0);
6325 } else {
6326 /* remove the bitmap */
6327 if (!mddev->bitmap)
6328 return -ENOENT;
6329 if (mddev->bitmap->storage.file)
6330 return -EINVAL;
6331 mddev->pers->quiesce(mddev, 1);
6332 bitmap_destroy(mddev);
6333 mddev->pers->quiesce(mddev, 0);
6334 mddev->bitmap_info.offset = 0;
6335 }
6336 }
6337 md_update_sb(mddev, 1);
6338 return rv;
6339 }
6340
6341 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6342 {
6343 struct md_rdev *rdev;
6344 int err = 0;
6345
6346 if (mddev->pers == NULL)
6347 return -ENODEV;
6348
6349 rcu_read_lock();
6350 rdev = find_rdev_rcu(mddev, dev);
6351 if (!rdev)
6352 err = -ENODEV;
6353 else {
6354 md_error(mddev, rdev);
6355 if (!test_bit(Faulty, &rdev->flags))
6356 err = -EBUSY;
6357 }
6358 rcu_read_unlock();
6359 return err;
6360 }
6361
6362 /*
6363 * We have a problem here : there is no easy way to give a CHS
6364 * virtual geometry. We currently pretend that we have a 2 heads
6365 * 4 sectors (with a BIG number of cylinders...). This drives
6366 * dosfs just mad... ;-)
6367 */
6368 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6369 {
6370 struct mddev *mddev = bdev->bd_disk->private_data;
6371
6372 geo->heads = 2;
6373 geo->sectors = 4;
6374 geo->cylinders = mddev->array_sectors / 8;
6375 return 0;
6376 }
6377
6378 static inline bool md_ioctl_valid(unsigned int cmd)
6379 {
6380 switch (cmd) {
6381 case ADD_NEW_DISK:
6382 case BLKROSET:
6383 case GET_ARRAY_INFO:
6384 case GET_BITMAP_FILE:
6385 case GET_DISK_INFO:
6386 case HOT_ADD_DISK:
6387 case HOT_REMOVE_DISK:
6388 case RAID_AUTORUN:
6389 case RAID_VERSION:
6390 case RESTART_ARRAY_RW:
6391 case RUN_ARRAY:
6392 case SET_ARRAY_INFO:
6393 case SET_BITMAP_FILE:
6394 case SET_DISK_FAULTY:
6395 case STOP_ARRAY:
6396 case STOP_ARRAY_RO:
6397 return true;
6398 default:
6399 return false;
6400 }
6401 }
6402
6403 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6404 unsigned int cmd, unsigned long arg)
6405 {
6406 int err = 0;
6407 void __user *argp = (void __user *)arg;
6408 struct mddev *mddev = NULL;
6409 int ro;
6410
6411 if (!md_ioctl_valid(cmd))
6412 return -ENOTTY;
6413
6414 switch (cmd) {
6415 case RAID_VERSION:
6416 case GET_ARRAY_INFO:
6417 case GET_DISK_INFO:
6418 break;
6419 default:
6420 if (!capable(CAP_SYS_ADMIN))
6421 return -EACCES;
6422 }
6423
6424 /*
6425 * Commands dealing with the RAID driver but not any
6426 * particular array:
6427 */
6428 switch (cmd) {
6429 case RAID_VERSION:
6430 err = get_version(argp);
6431 goto out;
6432
6433 #ifndef MODULE
6434 case RAID_AUTORUN:
6435 err = 0;
6436 autostart_arrays(arg);
6437 goto out;
6438 #endif
6439 default:;
6440 }
6441
6442 /*
6443 * Commands creating/starting a new array:
6444 */
6445
6446 mddev = bdev->bd_disk->private_data;
6447
6448 if (!mddev) {
6449 BUG();
6450 goto out;
6451 }
6452
6453 /* Some actions do not requires the mutex */
6454 switch (cmd) {
6455 case GET_ARRAY_INFO:
6456 if (!mddev->raid_disks && !mddev->external)
6457 err = -ENODEV;
6458 else
6459 err = get_array_info(mddev, argp);
6460 goto out;
6461
6462 case GET_DISK_INFO:
6463 if (!mddev->raid_disks && !mddev->external)
6464 err = -ENODEV;
6465 else
6466 err = get_disk_info(mddev, argp);
6467 goto out;
6468
6469 case SET_DISK_FAULTY:
6470 err = set_disk_faulty(mddev, new_decode_dev(arg));
6471 goto out;
6472
6473 case GET_BITMAP_FILE:
6474 err = get_bitmap_file(mddev, argp);
6475 goto out;
6476
6477 }
6478
6479 if (cmd == ADD_NEW_DISK)
6480 /* need to ensure md_delayed_delete() has completed */
6481 flush_workqueue(md_misc_wq);
6482
6483 if (cmd == HOT_REMOVE_DISK)
6484 /* need to ensure recovery thread has run */
6485 wait_event_interruptible_timeout(mddev->sb_wait,
6486 !test_bit(MD_RECOVERY_NEEDED,
6487 &mddev->flags),
6488 msecs_to_jiffies(5000));
6489 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6490 /* Need to flush page cache, and ensure no-one else opens
6491 * and writes
6492 */
6493 mutex_lock(&mddev->open_mutex);
6494 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
6495 mutex_unlock(&mddev->open_mutex);
6496 err = -EBUSY;
6497 goto out;
6498 }
6499 set_bit(MD_STILL_CLOSED, &mddev->flags);
6500 mutex_unlock(&mddev->open_mutex);
6501 sync_blockdev(bdev);
6502 }
6503 err = mddev_lock(mddev);
6504 if (err) {
6505 printk(KERN_INFO
6506 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6507 err, cmd);
6508 goto out;
6509 }
6510
6511 if (cmd == SET_ARRAY_INFO) {
6512 mdu_array_info_t info;
6513 if (!arg)
6514 memset(&info, 0, sizeof(info));
6515 else if (copy_from_user(&info, argp, sizeof(info))) {
6516 err = -EFAULT;
6517 goto unlock;
6518 }
6519 if (mddev->pers) {
6520 err = update_array_info(mddev, &info);
6521 if (err) {
6522 printk(KERN_WARNING "md: couldn't update"
6523 " array info. %d\n", err);
6524 goto unlock;
6525 }
6526 goto unlock;
6527 }
6528 if (!list_empty(&mddev->disks)) {
6529 printk(KERN_WARNING
6530 "md: array %s already has disks!\n",
6531 mdname(mddev));
6532 err = -EBUSY;
6533 goto unlock;
6534 }
6535 if (mddev->raid_disks) {
6536 printk(KERN_WARNING
6537 "md: array %s already initialised!\n",
6538 mdname(mddev));
6539 err = -EBUSY;
6540 goto unlock;
6541 }
6542 err = set_array_info(mddev, &info);
6543 if (err) {
6544 printk(KERN_WARNING "md: couldn't set"
6545 " array info. %d\n", err);
6546 goto unlock;
6547 }
6548 goto unlock;
6549 }
6550
6551 /*
6552 * Commands querying/configuring an existing array:
6553 */
6554 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6555 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6556 if ((!mddev->raid_disks && !mddev->external)
6557 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6558 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6559 && cmd != GET_BITMAP_FILE) {
6560 err = -ENODEV;
6561 goto unlock;
6562 }
6563
6564 /*
6565 * Commands even a read-only array can execute:
6566 */
6567 switch (cmd) {
6568 case RESTART_ARRAY_RW:
6569 err = restart_array(mddev);
6570 goto unlock;
6571
6572 case STOP_ARRAY:
6573 err = do_md_stop(mddev, 0, bdev);
6574 goto unlock;
6575
6576 case STOP_ARRAY_RO:
6577 err = md_set_readonly(mddev, bdev);
6578 goto unlock;
6579
6580 case HOT_REMOVE_DISK:
6581 err = hot_remove_disk(mddev, new_decode_dev(arg));
6582 goto unlock;
6583
6584 case ADD_NEW_DISK:
6585 /* We can support ADD_NEW_DISK on read-only arrays
6586 * on if we are re-adding a preexisting device.
6587 * So require mddev->pers and MD_DISK_SYNC.
6588 */
6589 if (mddev->pers) {
6590 mdu_disk_info_t info;
6591 if (copy_from_user(&info, argp, sizeof(info)))
6592 err = -EFAULT;
6593 else if (!(info.state & (1<<MD_DISK_SYNC)))
6594 /* Need to clear read-only for this */
6595 break;
6596 else
6597 err = add_new_disk(mddev, &info);
6598 goto unlock;
6599 }
6600 break;
6601
6602 case BLKROSET:
6603 if (get_user(ro, (int __user *)(arg))) {
6604 err = -EFAULT;
6605 goto unlock;
6606 }
6607 err = -EINVAL;
6608
6609 /* if the bdev is going readonly the value of mddev->ro
6610 * does not matter, no writes are coming
6611 */
6612 if (ro)
6613 goto unlock;
6614
6615 /* are we are already prepared for writes? */
6616 if (mddev->ro != 1)
6617 goto unlock;
6618
6619 /* transitioning to readauto need only happen for
6620 * arrays that call md_write_start
6621 */
6622 if (mddev->pers) {
6623 err = restart_array(mddev);
6624 if (err == 0) {
6625 mddev->ro = 2;
6626 set_disk_ro(mddev->gendisk, 0);
6627 }
6628 }
6629 goto unlock;
6630 }
6631
6632 /*
6633 * The remaining ioctls are changing the state of the
6634 * superblock, so we do not allow them on read-only arrays.
6635 */
6636 if (mddev->ro && mddev->pers) {
6637 if (mddev->ro == 2) {
6638 mddev->ro = 0;
6639 sysfs_notify_dirent_safe(mddev->sysfs_state);
6640 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6641 /* mddev_unlock will wake thread */
6642 /* If a device failed while we were read-only, we
6643 * need to make sure the metadata is updated now.
6644 */
6645 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6646 mddev_unlock(mddev);
6647 wait_event(mddev->sb_wait,
6648 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6649 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6650 mddev_lock_nointr(mddev);
6651 }
6652 } else {
6653 err = -EROFS;
6654 goto unlock;
6655 }
6656 }
6657
6658 switch (cmd) {
6659 case ADD_NEW_DISK:
6660 {
6661 mdu_disk_info_t info;
6662 if (copy_from_user(&info, argp, sizeof(info)))
6663 err = -EFAULT;
6664 else
6665 err = add_new_disk(mddev, &info);
6666 goto unlock;
6667 }
6668
6669 case HOT_ADD_DISK:
6670 err = hot_add_disk(mddev, new_decode_dev(arg));
6671 goto unlock;
6672
6673 case RUN_ARRAY:
6674 err = do_md_run(mddev);
6675 goto unlock;
6676
6677 case SET_BITMAP_FILE:
6678 err = set_bitmap_file(mddev, (int)arg);
6679 goto unlock;
6680
6681 default:
6682 err = -EINVAL;
6683 goto unlock;
6684 }
6685
6686 unlock:
6687 if (mddev->hold_active == UNTIL_IOCTL &&
6688 err != -EINVAL)
6689 mddev->hold_active = 0;
6690 mddev_unlock(mddev);
6691 out:
6692 return err;
6693 }
6694 #ifdef CONFIG_COMPAT
6695 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6696 unsigned int cmd, unsigned long arg)
6697 {
6698 switch (cmd) {
6699 case HOT_REMOVE_DISK:
6700 case HOT_ADD_DISK:
6701 case SET_DISK_FAULTY:
6702 case SET_BITMAP_FILE:
6703 /* These take in integer arg, do not convert */
6704 break;
6705 default:
6706 arg = (unsigned long)compat_ptr(arg);
6707 break;
6708 }
6709
6710 return md_ioctl(bdev, mode, cmd, arg);
6711 }
6712 #endif /* CONFIG_COMPAT */
6713
6714 static int md_open(struct block_device *bdev, fmode_t mode)
6715 {
6716 /*
6717 * Succeed if we can lock the mddev, which confirms that
6718 * it isn't being stopped right now.
6719 */
6720 struct mddev *mddev = mddev_find(bdev->bd_dev);
6721 int err;
6722
6723 if (!mddev)
6724 return -ENODEV;
6725
6726 if (mddev->gendisk != bdev->bd_disk) {
6727 /* we are racing with mddev_put which is discarding this
6728 * bd_disk.
6729 */
6730 mddev_put(mddev);
6731 /* Wait until bdev->bd_disk is definitely gone */
6732 flush_workqueue(md_misc_wq);
6733 /* Then retry the open from the top */
6734 return -ERESTARTSYS;
6735 }
6736 BUG_ON(mddev != bdev->bd_disk->private_data);
6737
6738 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
6739 goto out;
6740
6741 err = 0;
6742 atomic_inc(&mddev->openers);
6743 clear_bit(MD_STILL_CLOSED, &mddev->flags);
6744 mutex_unlock(&mddev->open_mutex);
6745
6746 check_disk_change(bdev);
6747 out:
6748 return err;
6749 }
6750
6751 static void md_release(struct gendisk *disk, fmode_t mode)
6752 {
6753 struct mddev *mddev = disk->private_data;
6754
6755 BUG_ON(!mddev);
6756 atomic_dec(&mddev->openers);
6757 mddev_put(mddev);
6758 }
6759
6760 static int md_media_changed(struct gendisk *disk)
6761 {
6762 struct mddev *mddev = disk->private_data;
6763
6764 return mddev->changed;
6765 }
6766
6767 static int md_revalidate(struct gendisk *disk)
6768 {
6769 struct mddev *mddev = disk->private_data;
6770
6771 mddev->changed = 0;
6772 return 0;
6773 }
6774 static const struct block_device_operations md_fops =
6775 {
6776 .owner = THIS_MODULE,
6777 .open = md_open,
6778 .release = md_release,
6779 .ioctl = md_ioctl,
6780 #ifdef CONFIG_COMPAT
6781 .compat_ioctl = md_compat_ioctl,
6782 #endif
6783 .getgeo = md_getgeo,
6784 .media_changed = md_media_changed,
6785 .revalidate_disk= md_revalidate,
6786 };
6787
6788 static int md_thread(void *arg)
6789 {
6790 struct md_thread *thread = arg;
6791
6792 /*
6793 * md_thread is a 'system-thread', it's priority should be very
6794 * high. We avoid resource deadlocks individually in each
6795 * raid personality. (RAID5 does preallocation) We also use RR and
6796 * the very same RT priority as kswapd, thus we will never get
6797 * into a priority inversion deadlock.
6798 *
6799 * we definitely have to have equal or higher priority than
6800 * bdflush, otherwise bdflush will deadlock if there are too
6801 * many dirty RAID5 blocks.
6802 */
6803
6804 allow_signal(SIGKILL);
6805 while (!kthread_should_stop()) {
6806
6807 /* We need to wait INTERRUPTIBLE so that
6808 * we don't add to the load-average.
6809 * That means we need to be sure no signals are
6810 * pending
6811 */
6812 if (signal_pending(current))
6813 flush_signals(current);
6814
6815 wait_event_interruptible_timeout
6816 (thread->wqueue,
6817 test_bit(THREAD_WAKEUP, &thread->flags)
6818 || kthread_should_stop(),
6819 thread->timeout);
6820
6821 clear_bit(THREAD_WAKEUP, &thread->flags);
6822 if (!kthread_should_stop())
6823 thread->run(thread);
6824 }
6825
6826 return 0;
6827 }
6828
6829 void md_wakeup_thread(struct md_thread *thread)
6830 {
6831 if (thread) {
6832 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
6833 set_bit(THREAD_WAKEUP, &thread->flags);
6834 wake_up(&thread->wqueue);
6835 }
6836 }
6837 EXPORT_SYMBOL(md_wakeup_thread);
6838
6839 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
6840 struct mddev *mddev, const char *name)
6841 {
6842 struct md_thread *thread;
6843
6844 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
6845 if (!thread)
6846 return NULL;
6847
6848 init_waitqueue_head(&thread->wqueue);
6849
6850 thread->run = run;
6851 thread->mddev = mddev;
6852 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6853 thread->tsk = kthread_run(md_thread, thread,
6854 "%s_%s",
6855 mdname(thread->mddev),
6856 name);
6857 if (IS_ERR(thread->tsk)) {
6858 kfree(thread);
6859 return NULL;
6860 }
6861 return thread;
6862 }
6863 EXPORT_SYMBOL(md_register_thread);
6864
6865 void md_unregister_thread(struct md_thread **threadp)
6866 {
6867 struct md_thread *thread = *threadp;
6868 if (!thread)
6869 return;
6870 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6871 /* Locking ensures that mddev_unlock does not wake_up a
6872 * non-existent thread
6873 */
6874 spin_lock(&pers_lock);
6875 *threadp = NULL;
6876 spin_unlock(&pers_lock);
6877
6878 kthread_stop(thread->tsk);
6879 kfree(thread);
6880 }
6881 EXPORT_SYMBOL(md_unregister_thread);
6882
6883 void md_error(struct mddev *mddev, struct md_rdev *rdev)
6884 {
6885 if (!rdev || test_bit(Faulty, &rdev->flags))
6886 return;
6887
6888 if (!mddev->pers || !mddev->pers->error_handler)
6889 return;
6890 mddev->pers->error_handler(mddev,rdev);
6891 if (mddev->degraded)
6892 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6893 sysfs_notify_dirent_safe(rdev->sysfs_state);
6894 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6895 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6896 md_wakeup_thread(mddev->thread);
6897 if (mddev->event_work.func)
6898 queue_work(md_misc_wq, &mddev->event_work);
6899 md_new_event_inintr(mddev);
6900 }
6901 EXPORT_SYMBOL(md_error);
6902
6903 /* seq_file implementation /proc/mdstat */
6904
6905 static void status_unused(struct seq_file *seq)
6906 {
6907 int i = 0;
6908 struct md_rdev *rdev;
6909
6910 seq_printf(seq, "unused devices: ");
6911
6912 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6913 char b[BDEVNAME_SIZE];
6914 i++;
6915 seq_printf(seq, "%s ",
6916 bdevname(rdev->bdev,b));
6917 }
6918 if (!i)
6919 seq_printf(seq, "<none>");
6920
6921 seq_printf(seq, "\n");
6922 }
6923
6924 static void status_resync(struct seq_file *seq, struct mddev *mddev)
6925 {
6926 sector_t max_sectors, resync, res;
6927 unsigned long dt, db;
6928 sector_t rt;
6929 int scale;
6930 unsigned int per_milli;
6931
6932 if (mddev->curr_resync <= 3)
6933 resync = 0;
6934 else
6935 resync = mddev->curr_resync
6936 - atomic_read(&mddev->recovery_active);
6937
6938 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
6939 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6940 max_sectors = mddev->resync_max_sectors;
6941 else
6942 max_sectors = mddev->dev_sectors;
6943
6944 WARN_ON(max_sectors == 0);
6945 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6946 * in a sector_t, and (max_sectors>>scale) will fit in a
6947 * u32, as those are the requirements for sector_div.
6948 * Thus 'scale' must be at least 10
6949 */
6950 scale = 10;
6951 if (sizeof(sector_t) > sizeof(unsigned long)) {
6952 while ( max_sectors/2 > (1ULL<<(scale+32)))
6953 scale++;
6954 }
6955 res = (resync>>scale)*1000;
6956 sector_div(res, (u32)((max_sectors>>scale)+1));
6957
6958 per_milli = res;
6959 {
6960 int i, x = per_milli/50, y = 20-x;
6961 seq_printf(seq, "[");
6962 for (i = 0; i < x; i++)
6963 seq_printf(seq, "=");
6964 seq_printf(seq, ">");
6965 for (i = 0; i < y; i++)
6966 seq_printf(seq, ".");
6967 seq_printf(seq, "] ");
6968 }
6969 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6970 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6971 "reshape" :
6972 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6973 "check" :
6974 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6975 "resync" : "recovery"))),
6976 per_milli/10, per_milli % 10,
6977 (unsigned long long) resync/2,
6978 (unsigned long long) max_sectors/2);
6979
6980 /*
6981 * dt: time from mark until now
6982 * db: blocks written from mark until now
6983 * rt: remaining time
6984 *
6985 * rt is a sector_t, so could be 32bit or 64bit.
6986 * So we divide before multiply in case it is 32bit and close
6987 * to the limit.
6988 * We scale the divisor (db) by 32 to avoid losing precision
6989 * near the end of resync when the number of remaining sectors
6990 * is close to 'db'.
6991 * We then divide rt by 32 after multiplying by db to compensate.
6992 * The '+1' avoids division by zero if db is very small.
6993 */
6994 dt = ((jiffies - mddev->resync_mark) / HZ);
6995 if (!dt) dt++;
6996 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6997 - mddev->resync_mark_cnt;
6998
6999 rt = max_sectors - resync; /* number of remaining sectors */
7000 sector_div(rt, db/32+1);
7001 rt *= dt;
7002 rt >>= 5;
7003
7004 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7005 ((unsigned long)rt % 60)/6);
7006
7007 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
7008 }
7009
7010 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7011 {
7012 struct list_head *tmp;
7013 loff_t l = *pos;
7014 struct mddev *mddev;
7015
7016 if (l >= 0x10000)
7017 return NULL;
7018 if (!l--)
7019 /* header */
7020 return (void*)1;
7021
7022 spin_lock(&all_mddevs_lock);
7023 list_for_each(tmp,&all_mddevs)
7024 if (!l--) {
7025 mddev = list_entry(tmp, struct mddev, all_mddevs);
7026 mddev_get(mddev);
7027 spin_unlock(&all_mddevs_lock);
7028 return mddev;
7029 }
7030 spin_unlock(&all_mddevs_lock);
7031 if (!l--)
7032 return (void*)2;/* tail */
7033 return NULL;
7034 }
7035
7036 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7037 {
7038 struct list_head *tmp;
7039 struct mddev *next_mddev, *mddev = v;
7040
7041 ++*pos;
7042 if (v == (void*)2)
7043 return NULL;
7044
7045 spin_lock(&all_mddevs_lock);
7046 if (v == (void*)1)
7047 tmp = all_mddevs.next;
7048 else
7049 tmp = mddev->all_mddevs.next;
7050 if (tmp != &all_mddevs)
7051 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
7052 else {
7053 next_mddev = (void*)2;
7054 *pos = 0x10000;
7055 }
7056 spin_unlock(&all_mddevs_lock);
7057
7058 if (v != (void*)1)
7059 mddev_put(mddev);
7060 return next_mddev;
7061
7062 }
7063
7064 static void md_seq_stop(struct seq_file *seq, void *v)
7065 {
7066 struct mddev *mddev = v;
7067
7068 if (mddev && v != (void*)1 && v != (void*)2)
7069 mddev_put(mddev);
7070 }
7071
7072 static int md_seq_show(struct seq_file *seq, void *v)
7073 {
7074 struct mddev *mddev = v;
7075 sector_t sectors;
7076 struct md_rdev *rdev;
7077
7078 if (v == (void*)1) {
7079 struct md_personality *pers;
7080 seq_printf(seq, "Personalities : ");
7081 spin_lock(&pers_lock);
7082 list_for_each_entry(pers, &pers_list, list)
7083 seq_printf(seq, "[%s] ", pers->name);
7084
7085 spin_unlock(&pers_lock);
7086 seq_printf(seq, "\n");
7087 seq->poll_event = atomic_read(&md_event_count);
7088 return 0;
7089 }
7090 if (v == (void*)2) {
7091 status_unused(seq);
7092 return 0;
7093 }
7094
7095 spin_lock(&mddev->lock);
7096 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7097 seq_printf(seq, "%s : %sactive", mdname(mddev),
7098 mddev->pers ? "" : "in");
7099 if (mddev->pers) {
7100 if (mddev->ro==1)
7101 seq_printf(seq, " (read-only)");
7102 if (mddev->ro==2)
7103 seq_printf(seq, " (auto-read-only)");
7104 seq_printf(seq, " %s", mddev->pers->name);
7105 }
7106
7107 sectors = 0;
7108 rcu_read_lock();
7109 rdev_for_each_rcu(rdev, mddev) {
7110 char b[BDEVNAME_SIZE];
7111 seq_printf(seq, " %s[%d]",
7112 bdevname(rdev->bdev,b), rdev->desc_nr);
7113 if (test_bit(WriteMostly, &rdev->flags))
7114 seq_printf(seq, "(W)");
7115 if (test_bit(Faulty, &rdev->flags)) {
7116 seq_printf(seq, "(F)");
7117 continue;
7118 }
7119 if (rdev->raid_disk < 0)
7120 seq_printf(seq, "(S)"); /* spare */
7121 if (test_bit(Replacement, &rdev->flags))
7122 seq_printf(seq, "(R)");
7123 sectors += rdev->sectors;
7124 }
7125 rcu_read_unlock();
7126
7127 if (!list_empty(&mddev->disks)) {
7128 if (mddev->pers)
7129 seq_printf(seq, "\n %llu blocks",
7130 (unsigned long long)
7131 mddev->array_sectors / 2);
7132 else
7133 seq_printf(seq, "\n %llu blocks",
7134 (unsigned long long)sectors / 2);
7135 }
7136 if (mddev->persistent) {
7137 if (mddev->major_version != 0 ||
7138 mddev->minor_version != 90) {
7139 seq_printf(seq," super %d.%d",
7140 mddev->major_version,
7141 mddev->minor_version);
7142 }
7143 } else if (mddev->external)
7144 seq_printf(seq, " super external:%s",
7145 mddev->metadata_type);
7146 else
7147 seq_printf(seq, " super non-persistent");
7148
7149 if (mddev->pers) {
7150 mddev->pers->status(seq, mddev);
7151 seq_printf(seq, "\n ");
7152 if (mddev->pers->sync_request) {
7153 if (mddev->curr_resync > 2) {
7154 status_resync(seq, mddev);
7155 seq_printf(seq, "\n ");
7156 } else if (mddev->curr_resync >= 1)
7157 seq_printf(seq, "\tresync=DELAYED\n ");
7158 else if (mddev->recovery_cp < MaxSector)
7159 seq_printf(seq, "\tresync=PENDING\n ");
7160 }
7161 } else
7162 seq_printf(seq, "\n ");
7163
7164 bitmap_status(seq, mddev->bitmap);
7165
7166 seq_printf(seq, "\n");
7167 }
7168 spin_unlock(&mddev->lock);
7169
7170 return 0;
7171 }
7172
7173 static const struct seq_operations md_seq_ops = {
7174 .start = md_seq_start,
7175 .next = md_seq_next,
7176 .stop = md_seq_stop,
7177 .show = md_seq_show,
7178 };
7179
7180 static int md_seq_open(struct inode *inode, struct file *file)
7181 {
7182 struct seq_file *seq;
7183 int error;
7184
7185 error = seq_open(file, &md_seq_ops);
7186 if (error)
7187 return error;
7188
7189 seq = file->private_data;
7190 seq->poll_event = atomic_read(&md_event_count);
7191 return error;
7192 }
7193
7194 static int md_unloading;
7195 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7196 {
7197 struct seq_file *seq = filp->private_data;
7198 int mask;
7199
7200 if (md_unloading)
7201 return POLLIN|POLLRDNORM|POLLERR|POLLPRI;
7202 poll_wait(filp, &md_event_waiters, wait);
7203
7204 /* always allow read */
7205 mask = POLLIN | POLLRDNORM;
7206
7207 if (seq->poll_event != atomic_read(&md_event_count))
7208 mask |= POLLERR | POLLPRI;
7209 return mask;
7210 }
7211
7212 static const struct file_operations md_seq_fops = {
7213 .owner = THIS_MODULE,
7214 .open = md_seq_open,
7215 .read = seq_read,
7216 .llseek = seq_lseek,
7217 .release = seq_release_private,
7218 .poll = mdstat_poll,
7219 };
7220
7221 int register_md_personality(struct md_personality *p)
7222 {
7223 printk(KERN_INFO "md: %s personality registered for level %d\n",
7224 p->name, p->level);
7225 spin_lock(&pers_lock);
7226 list_add_tail(&p->list, &pers_list);
7227 spin_unlock(&pers_lock);
7228 return 0;
7229 }
7230 EXPORT_SYMBOL(register_md_personality);
7231
7232 int unregister_md_personality(struct md_personality *p)
7233 {
7234 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
7235 spin_lock(&pers_lock);
7236 list_del_init(&p->list);
7237 spin_unlock(&pers_lock);
7238 return 0;
7239 }
7240 EXPORT_SYMBOL(unregister_md_personality);
7241
7242 int register_md_cluster_operations(struct md_cluster_operations *ops, struct module *module)
7243 {
7244 if (md_cluster_ops != NULL)
7245 return -EALREADY;
7246 spin_lock(&pers_lock);
7247 md_cluster_ops = ops;
7248 md_cluster_mod = module;
7249 spin_unlock(&pers_lock);
7250 return 0;
7251 }
7252 EXPORT_SYMBOL(register_md_cluster_operations);
7253
7254 int unregister_md_cluster_operations(void)
7255 {
7256 spin_lock(&pers_lock);
7257 md_cluster_ops = NULL;
7258 spin_unlock(&pers_lock);
7259 return 0;
7260 }
7261 EXPORT_SYMBOL(unregister_md_cluster_operations);
7262
7263 int md_setup_cluster(struct mddev *mddev, int nodes)
7264 {
7265 int err;
7266
7267 err = request_module("md-cluster");
7268 if (err) {
7269 pr_err("md-cluster module not found.\n");
7270 return err;
7271 }
7272
7273 spin_lock(&pers_lock);
7274 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
7275 spin_unlock(&pers_lock);
7276 return -ENOENT;
7277 }
7278 spin_unlock(&pers_lock);
7279
7280 return md_cluster_ops->join(mddev, nodes);
7281 }
7282
7283 void md_cluster_stop(struct mddev *mddev)
7284 {
7285 if (!md_cluster_ops)
7286 return;
7287 md_cluster_ops->leave(mddev);
7288 module_put(md_cluster_mod);
7289 }
7290
7291 static int is_mddev_idle(struct mddev *mddev, int init)
7292 {
7293 struct md_rdev *rdev;
7294 int idle;
7295 int curr_events;
7296
7297 idle = 1;
7298 rcu_read_lock();
7299 rdev_for_each_rcu(rdev, mddev) {
7300 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7301 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7302 (int)part_stat_read(&disk->part0, sectors[1]) -
7303 atomic_read(&disk->sync_io);
7304 /* sync IO will cause sync_io to increase before the disk_stats
7305 * as sync_io is counted when a request starts, and
7306 * disk_stats is counted when it completes.
7307 * So resync activity will cause curr_events to be smaller than
7308 * when there was no such activity.
7309 * non-sync IO will cause disk_stat to increase without
7310 * increasing sync_io so curr_events will (eventually)
7311 * be larger than it was before. Once it becomes
7312 * substantially larger, the test below will cause
7313 * the array to appear non-idle, and resync will slow
7314 * down.
7315 * If there is a lot of outstanding resync activity when
7316 * we set last_event to curr_events, then all that activity
7317 * completing might cause the array to appear non-idle
7318 * and resync will be slowed down even though there might
7319 * not have been non-resync activity. This will only
7320 * happen once though. 'last_events' will soon reflect
7321 * the state where there is little or no outstanding
7322 * resync requests, and further resync activity will
7323 * always make curr_events less than last_events.
7324 *
7325 */
7326 if (init || curr_events - rdev->last_events > 64) {
7327 rdev->last_events = curr_events;
7328 idle = 0;
7329 }
7330 }
7331 rcu_read_unlock();
7332 return idle;
7333 }
7334
7335 void md_done_sync(struct mddev *mddev, int blocks, int ok)
7336 {
7337 /* another "blocks" (512byte) blocks have been synced */
7338 atomic_sub(blocks, &mddev->recovery_active);
7339 wake_up(&mddev->recovery_wait);
7340 if (!ok) {
7341 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7342 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
7343 md_wakeup_thread(mddev->thread);
7344 // stop recovery, signal do_sync ....
7345 }
7346 }
7347 EXPORT_SYMBOL(md_done_sync);
7348
7349 /* md_write_start(mddev, bi)
7350 * If we need to update some array metadata (e.g. 'active' flag
7351 * in superblock) before writing, schedule a superblock update
7352 * and wait for it to complete.
7353 */
7354 void md_write_start(struct mddev *mddev, struct bio *bi)
7355 {
7356 int did_change = 0;
7357 if (bio_data_dir(bi) != WRITE)
7358 return;
7359
7360 BUG_ON(mddev->ro == 1);
7361 if (mddev->ro == 2) {
7362 /* need to switch to read/write */
7363 mddev->ro = 0;
7364 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7365 md_wakeup_thread(mddev->thread);
7366 md_wakeup_thread(mddev->sync_thread);
7367 did_change = 1;
7368 }
7369 atomic_inc(&mddev->writes_pending);
7370 if (mddev->safemode == 1)
7371 mddev->safemode = 0;
7372 if (mddev->in_sync) {
7373 spin_lock(&mddev->lock);
7374 if (mddev->in_sync) {
7375 mddev->in_sync = 0;
7376 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7377 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7378 md_wakeup_thread(mddev->thread);
7379 did_change = 1;
7380 }
7381 spin_unlock(&mddev->lock);
7382 }
7383 if (did_change)
7384 sysfs_notify_dirent_safe(mddev->sysfs_state);
7385 wait_event(mddev->sb_wait,
7386 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
7387 }
7388 EXPORT_SYMBOL(md_write_start);
7389
7390 void md_write_end(struct mddev *mddev)
7391 {
7392 if (atomic_dec_and_test(&mddev->writes_pending)) {
7393 if (mddev->safemode == 2)
7394 md_wakeup_thread(mddev->thread);
7395 else if (mddev->safemode_delay)
7396 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7397 }
7398 }
7399 EXPORT_SYMBOL(md_write_end);
7400
7401 /* md_allow_write(mddev)
7402 * Calling this ensures that the array is marked 'active' so that writes
7403 * may proceed without blocking. It is important to call this before
7404 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7405 * Must be called with mddev_lock held.
7406 *
7407 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7408 * is dropped, so return -EAGAIN after notifying userspace.
7409 */
7410 int md_allow_write(struct mddev *mddev)
7411 {
7412 if (!mddev->pers)
7413 return 0;
7414 if (mddev->ro)
7415 return 0;
7416 if (!mddev->pers->sync_request)
7417 return 0;
7418
7419 spin_lock(&mddev->lock);
7420 if (mddev->in_sync) {
7421 mddev->in_sync = 0;
7422 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7423 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7424 if (mddev->safemode_delay &&
7425 mddev->safemode == 0)
7426 mddev->safemode = 1;
7427 spin_unlock(&mddev->lock);
7428 md_update_sb(mddev, 0);
7429 sysfs_notify_dirent_safe(mddev->sysfs_state);
7430 } else
7431 spin_unlock(&mddev->lock);
7432
7433 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
7434 return -EAGAIN;
7435 else
7436 return 0;
7437 }
7438 EXPORT_SYMBOL_GPL(md_allow_write);
7439
7440 #define SYNC_MARKS 10
7441 #define SYNC_MARK_STEP (3*HZ)
7442 #define UPDATE_FREQUENCY (5*60*HZ)
7443 void md_do_sync(struct md_thread *thread)
7444 {
7445 struct mddev *mddev = thread->mddev;
7446 struct mddev *mddev2;
7447 unsigned int currspeed = 0,
7448 window;
7449 sector_t max_sectors,j, io_sectors, recovery_done;
7450 unsigned long mark[SYNC_MARKS];
7451 unsigned long update_time;
7452 sector_t mark_cnt[SYNC_MARKS];
7453 int last_mark,m;
7454 struct list_head *tmp;
7455 sector_t last_check;
7456 int skipped = 0;
7457 struct md_rdev *rdev;
7458 char *desc, *action = NULL;
7459 struct blk_plug plug;
7460
7461 /* just incase thread restarts... */
7462 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7463 return;
7464 if (mddev->ro) {/* never try to sync a read-only array */
7465 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7466 return;
7467 }
7468
7469 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7470 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
7471 desc = "data-check";
7472 action = "check";
7473 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7474 desc = "requested-resync";
7475 action = "repair";
7476 } else
7477 desc = "resync";
7478 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7479 desc = "reshape";
7480 else
7481 desc = "recovery";
7482
7483 mddev->last_sync_action = action ?: desc;
7484
7485 /* we overload curr_resync somewhat here.
7486 * 0 == not engaged in resync at all
7487 * 2 == checking that there is no conflict with another sync
7488 * 1 == like 2, but have yielded to allow conflicting resync to
7489 * commense
7490 * other == active in resync - this many blocks
7491 *
7492 * Before starting a resync we must have set curr_resync to
7493 * 2, and then checked that every "conflicting" array has curr_resync
7494 * less than ours. When we find one that is the same or higher
7495 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7496 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7497 * This will mean we have to start checking from the beginning again.
7498 *
7499 */
7500
7501 do {
7502 mddev->curr_resync = 2;
7503
7504 try_again:
7505 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7506 goto skip;
7507 for_each_mddev(mddev2, tmp) {
7508 if (mddev2 == mddev)
7509 continue;
7510 if (!mddev->parallel_resync
7511 && mddev2->curr_resync
7512 && match_mddev_units(mddev, mddev2)) {
7513 DEFINE_WAIT(wq);
7514 if (mddev < mddev2 && mddev->curr_resync == 2) {
7515 /* arbitrarily yield */
7516 mddev->curr_resync = 1;
7517 wake_up(&resync_wait);
7518 }
7519 if (mddev > mddev2 && mddev->curr_resync == 1)
7520 /* no need to wait here, we can wait the next
7521 * time 'round when curr_resync == 2
7522 */
7523 continue;
7524 /* We need to wait 'interruptible' so as not to
7525 * contribute to the load average, and not to
7526 * be caught by 'softlockup'
7527 */
7528 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7529 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7530 mddev2->curr_resync >= mddev->curr_resync) {
7531 printk(KERN_INFO "md: delaying %s of %s"
7532 " until %s has finished (they"
7533 " share one or more physical units)\n",
7534 desc, mdname(mddev), mdname(mddev2));
7535 mddev_put(mddev2);
7536 if (signal_pending(current))
7537 flush_signals(current);
7538 schedule();
7539 finish_wait(&resync_wait, &wq);
7540 goto try_again;
7541 }
7542 finish_wait(&resync_wait, &wq);
7543 }
7544 }
7545 } while (mddev->curr_resync < 2);
7546
7547 j = 0;
7548 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7549 /* resync follows the size requested by the personality,
7550 * which defaults to physical size, but can be virtual size
7551 */
7552 max_sectors = mddev->resync_max_sectors;
7553 atomic64_set(&mddev->resync_mismatches, 0);
7554 /* we don't use the checkpoint if there's a bitmap */
7555 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7556 j = mddev->resync_min;
7557 else if (!mddev->bitmap)
7558 j = mddev->recovery_cp;
7559
7560 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7561 max_sectors = mddev->resync_max_sectors;
7562 else {
7563 /* recovery follows the physical size of devices */
7564 max_sectors = mddev->dev_sectors;
7565 j = MaxSector;
7566 rcu_read_lock();
7567 rdev_for_each_rcu(rdev, mddev)
7568 if (rdev->raid_disk >= 0 &&
7569 !test_bit(Faulty, &rdev->flags) &&
7570 !test_bit(In_sync, &rdev->flags) &&
7571 rdev->recovery_offset < j)
7572 j = rdev->recovery_offset;
7573 rcu_read_unlock();
7574
7575 /* If there is a bitmap, we need to make sure all
7576 * writes that started before we added a spare
7577 * complete before we start doing a recovery.
7578 * Otherwise the write might complete and (via
7579 * bitmap_endwrite) set a bit in the bitmap after the
7580 * recovery has checked that bit and skipped that
7581 * region.
7582 */
7583 if (mddev->bitmap) {
7584 mddev->pers->quiesce(mddev, 1);
7585 mddev->pers->quiesce(mddev, 0);
7586 }
7587 }
7588
7589 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7590 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7591 " %d KB/sec/disk.\n", speed_min(mddev));
7592 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
7593 "(but not more than %d KB/sec) for %s.\n",
7594 speed_max(mddev), desc);
7595
7596 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
7597
7598 io_sectors = 0;
7599 for (m = 0; m < SYNC_MARKS; m++) {
7600 mark[m] = jiffies;
7601 mark_cnt[m] = io_sectors;
7602 }
7603 last_mark = 0;
7604 mddev->resync_mark = mark[last_mark];
7605 mddev->resync_mark_cnt = mark_cnt[last_mark];
7606
7607 /*
7608 * Tune reconstruction:
7609 */
7610 window = 32*(PAGE_SIZE/512);
7611 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7612 window/2, (unsigned long long)max_sectors/2);
7613
7614 atomic_set(&mddev->recovery_active, 0);
7615 last_check = 0;
7616
7617 if (j>2) {
7618 printk(KERN_INFO
7619 "md: resuming %s of %s from checkpoint.\n",
7620 desc, mdname(mddev));
7621 mddev->curr_resync = j;
7622 } else
7623 mddev->curr_resync = 3; /* no longer delayed */
7624 mddev->curr_resync_completed = j;
7625 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7626 md_new_event(mddev);
7627 update_time = jiffies;
7628
7629 if (mddev_is_clustered(mddev))
7630 md_cluster_ops->resync_info_update(mddev, j, max_sectors);
7631
7632 blk_start_plug(&plug);
7633 while (j < max_sectors) {
7634 sector_t sectors;
7635
7636 skipped = 0;
7637
7638 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7639 ((mddev->curr_resync > mddev->curr_resync_completed &&
7640 (mddev->curr_resync - mddev->curr_resync_completed)
7641 > (max_sectors >> 4)) ||
7642 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7643 (j - mddev->curr_resync_completed)*2
7644 >= mddev->resync_max - mddev->curr_resync_completed
7645 )) {
7646 /* time to update curr_resync_completed */
7647 wait_event(mddev->recovery_wait,
7648 atomic_read(&mddev->recovery_active) == 0);
7649 mddev->curr_resync_completed = j;
7650 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7651 j > mddev->recovery_cp)
7652 mddev->recovery_cp = j;
7653 update_time = jiffies;
7654 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7655 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7656 }
7657
7658 while (j >= mddev->resync_max &&
7659 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7660 /* As this condition is controlled by user-space,
7661 * we can block indefinitely, so use '_interruptible'
7662 * to avoid triggering warnings.
7663 */
7664 flush_signals(current); /* just in case */
7665 wait_event_interruptible(mddev->recovery_wait,
7666 mddev->resync_max > j
7667 || test_bit(MD_RECOVERY_INTR,
7668 &mddev->recovery));
7669 }
7670
7671 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7672 break;
7673
7674 sectors = mddev->pers->sync_request(mddev, j, &skipped,
7675 currspeed < speed_min(mddev));
7676 if (sectors == 0) {
7677 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7678 break;
7679 }
7680
7681 if (!skipped) { /* actual IO requested */
7682 io_sectors += sectors;
7683 atomic_add(sectors, &mddev->recovery_active);
7684 }
7685
7686 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7687 break;
7688
7689 j += sectors;
7690 if (j > 2)
7691 mddev->curr_resync = j;
7692 if (mddev_is_clustered(mddev))
7693 md_cluster_ops->resync_info_update(mddev, j, max_sectors);
7694 mddev->curr_mark_cnt = io_sectors;
7695 if (last_check == 0)
7696 /* this is the earliest that rebuild will be
7697 * visible in /proc/mdstat
7698 */
7699 md_new_event(mddev);
7700
7701 if (last_check + window > io_sectors || j == max_sectors)
7702 continue;
7703
7704 last_check = io_sectors;
7705 repeat:
7706 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7707 /* step marks */
7708 int next = (last_mark+1) % SYNC_MARKS;
7709
7710 mddev->resync_mark = mark[next];
7711 mddev->resync_mark_cnt = mark_cnt[next];
7712 mark[next] = jiffies;
7713 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
7714 last_mark = next;
7715 }
7716
7717 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7718 break;
7719
7720 /*
7721 * this loop exits only if either when we are slower than
7722 * the 'hard' speed limit, or the system was IO-idle for
7723 * a jiffy.
7724 * the system might be non-idle CPU-wise, but we only care
7725 * about not overloading the IO subsystem. (things like an
7726 * e2fsck being done on the RAID array should execute fast)
7727 */
7728 cond_resched();
7729
7730 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
7731 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
7732 /((jiffies-mddev->resync_mark)/HZ +1) +1;
7733
7734 if (currspeed > speed_min(mddev)) {
7735 if ((currspeed > speed_max(mddev)) ||
7736 !is_mddev_idle(mddev, 0)) {
7737 msleep(500);
7738 goto repeat;
7739 }
7740 }
7741 }
7742 printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc,
7743 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
7744 ? "interrupted" : "done");
7745 /*
7746 * this also signals 'finished resyncing' to md_stop
7747 */
7748 blk_finish_plug(&plug);
7749 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7750
7751 /* tell personality that we are finished */
7752 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
7753
7754 if (mddev_is_clustered(mddev))
7755 md_cluster_ops->resync_info_update(mddev, 0, 0);
7756
7757 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
7758 mddev->curr_resync > 2) {
7759 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7760 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7761 if (mddev->curr_resync >= mddev->recovery_cp) {
7762 printk(KERN_INFO
7763 "md: checkpointing %s of %s.\n",
7764 desc, mdname(mddev));
7765 if (test_bit(MD_RECOVERY_ERROR,
7766 &mddev->recovery))
7767 mddev->recovery_cp =
7768 mddev->curr_resync_completed;
7769 else
7770 mddev->recovery_cp =
7771 mddev->curr_resync;
7772 }
7773 } else
7774 mddev->recovery_cp = MaxSector;
7775 } else {
7776 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7777 mddev->curr_resync = MaxSector;
7778 rcu_read_lock();
7779 rdev_for_each_rcu(rdev, mddev)
7780 if (rdev->raid_disk >= 0 &&
7781 mddev->delta_disks >= 0 &&
7782 !test_bit(Faulty, &rdev->flags) &&
7783 !test_bit(In_sync, &rdev->flags) &&
7784 rdev->recovery_offset < mddev->curr_resync)
7785 rdev->recovery_offset = mddev->curr_resync;
7786 rcu_read_unlock();
7787 }
7788 }
7789 skip:
7790 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7791
7792 spin_lock(&mddev->lock);
7793 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7794 /* We completed so min/max setting can be forgotten if used. */
7795 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7796 mddev->resync_min = 0;
7797 mddev->resync_max = MaxSector;
7798 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7799 mddev->resync_min = mddev->curr_resync_completed;
7800 mddev->curr_resync = 0;
7801 spin_unlock(&mddev->lock);
7802
7803 wake_up(&resync_wait);
7804 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7805 md_wakeup_thread(mddev->thread);
7806 return;
7807 }
7808 EXPORT_SYMBOL_GPL(md_do_sync);
7809
7810 static int remove_and_add_spares(struct mddev *mddev,
7811 struct md_rdev *this)
7812 {
7813 struct md_rdev *rdev;
7814 int spares = 0;
7815 int removed = 0;
7816
7817 rdev_for_each(rdev, mddev)
7818 if ((this == NULL || rdev == this) &&
7819 rdev->raid_disk >= 0 &&
7820 !test_bit(Blocked, &rdev->flags) &&
7821 (test_bit(Faulty, &rdev->flags) ||
7822 ! test_bit(In_sync, &rdev->flags)) &&
7823 atomic_read(&rdev->nr_pending)==0) {
7824 if (mddev->pers->hot_remove_disk(
7825 mddev, rdev) == 0) {
7826 sysfs_unlink_rdev(mddev, rdev);
7827 rdev->raid_disk = -1;
7828 removed++;
7829 }
7830 }
7831 if (removed && mddev->kobj.sd)
7832 sysfs_notify(&mddev->kobj, NULL, "degraded");
7833
7834 if (this)
7835 goto no_add;
7836
7837 rdev_for_each(rdev, mddev) {
7838 if (rdev->raid_disk >= 0 &&
7839 !test_bit(In_sync, &rdev->flags) &&
7840 !test_bit(Faulty, &rdev->flags))
7841 spares++;
7842 if (rdev->raid_disk >= 0)
7843 continue;
7844 if (test_bit(Faulty, &rdev->flags))
7845 continue;
7846 if (mddev->ro &&
7847 ! (rdev->saved_raid_disk >= 0 &&
7848 !test_bit(Bitmap_sync, &rdev->flags)))
7849 continue;
7850
7851 if (rdev->saved_raid_disk < 0)
7852 rdev->recovery_offset = 0;
7853 if (mddev->pers->
7854 hot_add_disk(mddev, rdev) == 0) {
7855 if (sysfs_link_rdev(mddev, rdev))
7856 /* failure here is OK */;
7857 spares++;
7858 md_new_event(mddev);
7859 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7860 }
7861 }
7862 no_add:
7863 if (removed)
7864 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7865 return spares;
7866 }
7867
7868 static void md_start_sync(struct work_struct *ws)
7869 {
7870 struct mddev *mddev = container_of(ws, struct mddev, del_work);
7871
7872 mddev->sync_thread = md_register_thread(md_do_sync,
7873 mddev,
7874 "resync");
7875 if (!mddev->sync_thread) {
7876 printk(KERN_ERR "%s: could not start resync"
7877 " thread...\n",
7878 mdname(mddev));
7879 /* leave the spares where they are, it shouldn't hurt */
7880 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7881 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7882 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7883 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7884 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7885 wake_up(&resync_wait);
7886 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7887 &mddev->recovery))
7888 if (mddev->sysfs_action)
7889 sysfs_notify_dirent_safe(mddev->sysfs_action);
7890 } else
7891 md_wakeup_thread(mddev->sync_thread);
7892 sysfs_notify_dirent_safe(mddev->sysfs_action);
7893 md_new_event(mddev);
7894 }
7895
7896 /*
7897 * This routine is regularly called by all per-raid-array threads to
7898 * deal with generic issues like resync and super-block update.
7899 * Raid personalities that don't have a thread (linear/raid0) do not
7900 * need this as they never do any recovery or update the superblock.
7901 *
7902 * It does not do any resync itself, but rather "forks" off other threads
7903 * to do that as needed.
7904 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7905 * "->recovery" and create a thread at ->sync_thread.
7906 * When the thread finishes it sets MD_RECOVERY_DONE
7907 * and wakeups up this thread which will reap the thread and finish up.
7908 * This thread also removes any faulty devices (with nr_pending == 0).
7909 *
7910 * The overall approach is:
7911 * 1/ if the superblock needs updating, update it.
7912 * 2/ If a recovery thread is running, don't do anything else.
7913 * 3/ If recovery has finished, clean up, possibly marking spares active.
7914 * 4/ If there are any faulty devices, remove them.
7915 * 5/ If array is degraded, try to add spares devices
7916 * 6/ If array has spares or is not in-sync, start a resync thread.
7917 */
7918 void md_check_recovery(struct mddev *mddev)
7919 {
7920 if (mddev->suspended)
7921 return;
7922
7923 if (mddev->bitmap)
7924 bitmap_daemon_work(mddev);
7925
7926 if (signal_pending(current)) {
7927 if (mddev->pers->sync_request && !mddev->external) {
7928 printk(KERN_INFO "md: %s in immediate safe mode\n",
7929 mdname(mddev));
7930 mddev->safemode = 2;
7931 }
7932 flush_signals(current);
7933 }
7934
7935 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7936 return;
7937 if ( ! (
7938 (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) ||
7939 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7940 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7941 (mddev->external == 0 && mddev->safemode == 1) ||
7942 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7943 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7944 ))
7945 return;
7946
7947 if (mddev_trylock(mddev)) {
7948 int spares = 0;
7949
7950 if (mddev->ro) {
7951 /* On a read-only array we can:
7952 * - remove failed devices
7953 * - add already-in_sync devices if the array itself
7954 * is in-sync.
7955 * As we only add devices that are already in-sync,
7956 * we can activate the spares immediately.
7957 */
7958 remove_and_add_spares(mddev, NULL);
7959 /* There is no thread, but we need to call
7960 * ->spare_active and clear saved_raid_disk
7961 */
7962 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7963 md_reap_sync_thread(mddev);
7964 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7965 goto unlock;
7966 }
7967
7968 if (!mddev->external) {
7969 int did_change = 0;
7970 spin_lock(&mddev->lock);
7971 if (mddev->safemode &&
7972 !atomic_read(&mddev->writes_pending) &&
7973 !mddev->in_sync &&
7974 mddev->recovery_cp == MaxSector) {
7975 mddev->in_sync = 1;
7976 did_change = 1;
7977 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7978 }
7979 if (mddev->safemode == 1)
7980 mddev->safemode = 0;
7981 spin_unlock(&mddev->lock);
7982 if (did_change)
7983 sysfs_notify_dirent_safe(mddev->sysfs_state);
7984 }
7985
7986 if (mddev->flags & MD_UPDATE_SB_FLAGS)
7987 md_update_sb(mddev, 0);
7988
7989 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7990 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7991 /* resync/recovery still happening */
7992 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7993 goto unlock;
7994 }
7995 if (mddev->sync_thread) {
7996 md_reap_sync_thread(mddev);
7997 goto unlock;
7998 }
7999 /* Set RUNNING before clearing NEEDED to avoid
8000 * any transients in the value of "sync_action".
8001 */
8002 mddev->curr_resync_completed = 0;
8003 spin_lock(&mddev->lock);
8004 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8005 spin_unlock(&mddev->lock);
8006 /* Clear some bits that don't mean anything, but
8007 * might be left set
8008 */
8009 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
8010 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8011
8012 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8013 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
8014 goto not_running;
8015 /* no recovery is running.
8016 * remove any failed drives, then
8017 * add spares if possible.
8018 * Spares are also removed and re-added, to allow
8019 * the personality to fail the re-add.
8020 */
8021
8022 if (mddev->reshape_position != MaxSector) {
8023 if (mddev->pers->check_reshape == NULL ||
8024 mddev->pers->check_reshape(mddev) != 0)
8025 /* Cannot proceed */
8026 goto not_running;
8027 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8028 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8029 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
8030 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8031 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8032 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8033 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8034 } else if (mddev->recovery_cp < MaxSector) {
8035 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8036 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8037 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
8038 /* nothing to be done ... */
8039 goto not_running;
8040
8041 if (mddev->pers->sync_request) {
8042 if (spares) {
8043 /* We are adding a device or devices to an array
8044 * which has the bitmap stored on all devices.
8045 * So make sure all bitmap pages get written
8046 */
8047 bitmap_write_all(mddev->bitmap);
8048 }
8049 INIT_WORK(&mddev->del_work, md_start_sync);
8050 queue_work(md_misc_wq, &mddev->del_work);
8051 goto unlock;
8052 }
8053 not_running:
8054 if (!mddev->sync_thread) {
8055 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8056 wake_up(&resync_wait);
8057 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8058 &mddev->recovery))
8059 if (mddev->sysfs_action)
8060 sysfs_notify_dirent_safe(mddev->sysfs_action);
8061 }
8062 unlock:
8063 wake_up(&mddev->sb_wait);
8064 mddev_unlock(mddev);
8065 }
8066 }
8067 EXPORT_SYMBOL(md_check_recovery);
8068
8069 void md_reap_sync_thread(struct mddev *mddev)
8070 {
8071 struct md_rdev *rdev;
8072
8073 /* resync has finished, collect result */
8074 md_unregister_thread(&mddev->sync_thread);
8075 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8076 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8077 /* success...*/
8078 /* activate any spares */
8079 if (mddev->pers->spare_active(mddev)) {
8080 sysfs_notify(&mddev->kobj, NULL,
8081 "degraded");
8082 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8083 }
8084 }
8085 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8086 mddev->pers->finish_reshape)
8087 mddev->pers->finish_reshape(mddev);
8088
8089 /* If array is no-longer degraded, then any saved_raid_disk
8090 * information must be scrapped.
8091 */
8092 if (!mddev->degraded)
8093 rdev_for_each(rdev, mddev)
8094 rdev->saved_raid_disk = -1;
8095
8096 md_update_sb(mddev, 1);
8097 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8098 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8099 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8100 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8101 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8102 wake_up(&resync_wait);
8103 /* flag recovery needed just to double check */
8104 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8105 sysfs_notify_dirent_safe(mddev->sysfs_action);
8106 md_new_event(mddev);
8107 if (mddev->event_work.func)
8108 queue_work(md_misc_wq, &mddev->event_work);
8109 }
8110 EXPORT_SYMBOL(md_reap_sync_thread);
8111
8112 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
8113 {
8114 sysfs_notify_dirent_safe(rdev->sysfs_state);
8115 wait_event_timeout(rdev->blocked_wait,
8116 !test_bit(Blocked, &rdev->flags) &&
8117 !test_bit(BlockedBadBlocks, &rdev->flags),
8118 msecs_to_jiffies(5000));
8119 rdev_dec_pending(rdev, mddev);
8120 }
8121 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8122
8123 void md_finish_reshape(struct mddev *mddev)
8124 {
8125 /* called be personality module when reshape completes. */
8126 struct md_rdev *rdev;
8127
8128 rdev_for_each(rdev, mddev) {
8129 if (rdev->data_offset > rdev->new_data_offset)
8130 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8131 else
8132 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8133 rdev->data_offset = rdev->new_data_offset;
8134 }
8135 }
8136 EXPORT_SYMBOL(md_finish_reshape);
8137
8138 /* Bad block management.
8139 * We can record which blocks on each device are 'bad' and so just
8140 * fail those blocks, or that stripe, rather than the whole device.
8141 * Entries in the bad-block table are 64bits wide. This comprises:
8142 * Length of bad-range, in sectors: 0-511 for lengths 1-512
8143 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
8144 * A 'shift' can be set so that larger blocks are tracked and
8145 * consequently larger devices can be covered.
8146 * 'Acknowledged' flag - 1 bit. - the most significant bit.
8147 *
8148 * Locking of the bad-block table uses a seqlock so md_is_badblock
8149 * might need to retry if it is very unlucky.
8150 * We will sometimes want to check for bad blocks in a bi_end_io function,
8151 * so we use the write_seqlock_irq variant.
8152 *
8153 * When looking for a bad block we specify a range and want to
8154 * know if any block in the range is bad. So we binary-search
8155 * to the last range that starts at-or-before the given endpoint,
8156 * (or "before the sector after the target range")
8157 * then see if it ends after the given start.
8158 * We return
8159 * 0 if there are no known bad blocks in the range
8160 * 1 if there are known bad block which are all acknowledged
8161 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
8162 * plus the start/length of the first bad section we overlap.
8163 */
8164 int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
8165 sector_t *first_bad, int *bad_sectors)
8166 {
8167 int hi;
8168 int lo;
8169 u64 *p = bb->page;
8170 int rv;
8171 sector_t target = s + sectors;
8172 unsigned seq;
8173
8174 if (bb->shift > 0) {
8175 /* round the start down, and the end up */
8176 s >>= bb->shift;
8177 target += (1<<bb->shift) - 1;
8178 target >>= bb->shift;
8179 sectors = target - s;
8180 }
8181 /* 'target' is now the first block after the bad range */
8182
8183 retry:
8184 seq = read_seqbegin(&bb->lock);
8185 lo = 0;
8186 rv = 0;
8187 hi = bb->count;
8188
8189 /* Binary search between lo and hi for 'target'
8190 * i.e. for the last range that starts before 'target'
8191 */
8192 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8193 * are known not to be the last range before target.
8194 * VARIANT: hi-lo is the number of possible
8195 * ranges, and decreases until it reaches 1
8196 */
8197 while (hi - lo > 1) {
8198 int mid = (lo + hi) / 2;
8199 sector_t a = BB_OFFSET(p[mid]);
8200 if (a < target)
8201 /* This could still be the one, earlier ranges
8202 * could not. */
8203 lo = mid;
8204 else
8205 /* This and later ranges are definitely out. */
8206 hi = mid;
8207 }
8208 /* 'lo' might be the last that started before target, but 'hi' isn't */
8209 if (hi > lo) {
8210 /* need to check all range that end after 's' to see if
8211 * any are unacknowledged.
8212 */
8213 while (lo >= 0 &&
8214 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8215 if (BB_OFFSET(p[lo]) < target) {
8216 /* starts before the end, and finishes after
8217 * the start, so they must overlap
8218 */
8219 if (rv != -1 && BB_ACK(p[lo]))
8220 rv = 1;
8221 else
8222 rv = -1;
8223 *first_bad = BB_OFFSET(p[lo]);
8224 *bad_sectors = BB_LEN(p[lo]);
8225 }
8226 lo--;
8227 }
8228 }
8229
8230 if (read_seqretry(&bb->lock, seq))
8231 goto retry;
8232
8233 return rv;
8234 }
8235 EXPORT_SYMBOL_GPL(md_is_badblock);
8236
8237 /*
8238 * Add a range of bad blocks to the table.
8239 * This might extend the table, or might contract it
8240 * if two adjacent ranges can be merged.
8241 * We binary-search to find the 'insertion' point, then
8242 * decide how best to handle it.
8243 */
8244 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8245 int acknowledged)
8246 {
8247 u64 *p;
8248 int lo, hi;
8249 int rv = 1;
8250 unsigned long flags;
8251
8252 if (bb->shift < 0)
8253 /* badblocks are disabled */
8254 return 0;
8255
8256 if (bb->shift) {
8257 /* round the start down, and the end up */
8258 sector_t next = s + sectors;
8259 s >>= bb->shift;
8260 next += (1<<bb->shift) - 1;
8261 next >>= bb->shift;
8262 sectors = next - s;
8263 }
8264
8265 write_seqlock_irqsave(&bb->lock, flags);
8266
8267 p = bb->page;
8268 lo = 0;
8269 hi = bb->count;
8270 /* Find the last range that starts at-or-before 's' */
8271 while (hi - lo > 1) {
8272 int mid = (lo + hi) / 2;
8273 sector_t a = BB_OFFSET(p[mid]);
8274 if (a <= s)
8275 lo = mid;
8276 else
8277 hi = mid;
8278 }
8279 if (hi > lo && BB_OFFSET(p[lo]) > s)
8280 hi = lo;
8281
8282 if (hi > lo) {
8283 /* we found a range that might merge with the start
8284 * of our new range
8285 */
8286 sector_t a = BB_OFFSET(p[lo]);
8287 sector_t e = a + BB_LEN(p[lo]);
8288 int ack = BB_ACK(p[lo]);
8289 if (e >= s) {
8290 /* Yes, we can merge with a previous range */
8291 if (s == a && s + sectors >= e)
8292 /* new range covers old */
8293 ack = acknowledged;
8294 else
8295 ack = ack && acknowledged;
8296
8297 if (e < s + sectors)
8298 e = s + sectors;
8299 if (e - a <= BB_MAX_LEN) {
8300 p[lo] = BB_MAKE(a, e-a, ack);
8301 s = e;
8302 } else {
8303 /* does not all fit in one range,
8304 * make p[lo] maximal
8305 */
8306 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8307 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8308 s = a + BB_MAX_LEN;
8309 }
8310 sectors = e - s;
8311 }
8312 }
8313 if (sectors && hi < bb->count) {
8314 /* 'hi' points to the first range that starts after 's'.
8315 * Maybe we can merge with the start of that range */
8316 sector_t a = BB_OFFSET(p[hi]);
8317 sector_t e = a + BB_LEN(p[hi]);
8318 int ack = BB_ACK(p[hi]);
8319 if (a <= s + sectors) {
8320 /* merging is possible */
8321 if (e <= s + sectors) {
8322 /* full overlap */
8323 e = s + sectors;
8324 ack = acknowledged;
8325 } else
8326 ack = ack && acknowledged;
8327
8328 a = s;
8329 if (e - a <= BB_MAX_LEN) {
8330 p[hi] = BB_MAKE(a, e-a, ack);
8331 s = e;
8332 } else {
8333 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8334 s = a + BB_MAX_LEN;
8335 }
8336 sectors = e - s;
8337 lo = hi;
8338 hi++;
8339 }
8340 }
8341 if (sectors == 0 && hi < bb->count) {
8342 /* we might be able to combine lo and hi */
8343 /* Note: 's' is at the end of 'lo' */
8344 sector_t a = BB_OFFSET(p[hi]);
8345 int lolen = BB_LEN(p[lo]);
8346 int hilen = BB_LEN(p[hi]);
8347 int newlen = lolen + hilen - (s - a);
8348 if (s >= a && newlen < BB_MAX_LEN) {
8349 /* yes, we can combine them */
8350 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8351 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8352 memmove(p + hi, p + hi + 1,
8353 (bb->count - hi - 1) * 8);
8354 bb->count--;
8355 }
8356 }
8357 while (sectors) {
8358 /* didn't merge (it all).
8359 * Need to add a range just before 'hi' */
8360 if (bb->count >= MD_MAX_BADBLOCKS) {
8361 /* No room for more */
8362 rv = 0;
8363 break;
8364 } else {
8365 int this_sectors = sectors;
8366 memmove(p + hi + 1, p + hi,
8367 (bb->count - hi) * 8);
8368 bb->count++;
8369
8370 if (this_sectors > BB_MAX_LEN)
8371 this_sectors = BB_MAX_LEN;
8372 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8373 sectors -= this_sectors;
8374 s += this_sectors;
8375 }
8376 }
8377
8378 bb->changed = 1;
8379 if (!acknowledged)
8380 bb->unacked_exist = 1;
8381 write_sequnlock_irqrestore(&bb->lock, flags);
8382
8383 return rv;
8384 }
8385
8386 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8387 int is_new)
8388 {
8389 int rv;
8390 if (is_new)
8391 s += rdev->new_data_offset;
8392 else
8393 s += rdev->data_offset;
8394 rv = md_set_badblocks(&rdev->badblocks,
8395 s, sectors, 0);
8396 if (rv) {
8397 /* Make sure they get written out promptly */
8398 sysfs_notify_dirent_safe(rdev->sysfs_state);
8399 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8400 md_wakeup_thread(rdev->mddev->thread);
8401 }
8402 return rv;
8403 }
8404 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8405
8406 /*
8407 * Remove a range of bad blocks from the table.
8408 * This may involve extending the table if we spilt a region,
8409 * but it must not fail. So if the table becomes full, we just
8410 * drop the remove request.
8411 */
8412 static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8413 {
8414 u64 *p;
8415 int lo, hi;
8416 sector_t target = s + sectors;
8417 int rv = 0;
8418
8419 if (bb->shift > 0) {
8420 /* When clearing we round the start up and the end down.
8421 * This should not matter as the shift should align with
8422 * the block size and no rounding should ever be needed.
8423 * However it is better the think a block is bad when it
8424 * isn't than to think a block is not bad when it is.
8425 */
8426 s += (1<<bb->shift) - 1;
8427 s >>= bb->shift;
8428 target >>= bb->shift;
8429 sectors = target - s;
8430 }
8431
8432 write_seqlock_irq(&bb->lock);
8433
8434 p = bb->page;
8435 lo = 0;
8436 hi = bb->count;
8437 /* Find the last range that starts before 'target' */
8438 while (hi - lo > 1) {
8439 int mid = (lo + hi) / 2;
8440 sector_t a = BB_OFFSET(p[mid]);
8441 if (a < target)
8442 lo = mid;
8443 else
8444 hi = mid;
8445 }
8446 if (hi > lo) {
8447 /* p[lo] is the last range that could overlap the
8448 * current range. Earlier ranges could also overlap,
8449 * but only this one can overlap the end of the range.
8450 */
8451 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8452 /* Partial overlap, leave the tail of this range */
8453 int ack = BB_ACK(p[lo]);
8454 sector_t a = BB_OFFSET(p[lo]);
8455 sector_t end = a + BB_LEN(p[lo]);
8456
8457 if (a < s) {
8458 /* we need to split this range */
8459 if (bb->count >= MD_MAX_BADBLOCKS) {
8460 rv = -ENOSPC;
8461 goto out;
8462 }
8463 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8464 bb->count++;
8465 p[lo] = BB_MAKE(a, s-a, ack);
8466 lo++;
8467 }
8468 p[lo] = BB_MAKE(target, end - target, ack);
8469 /* there is no longer an overlap */
8470 hi = lo;
8471 lo--;
8472 }
8473 while (lo >= 0 &&
8474 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8475 /* This range does overlap */
8476 if (BB_OFFSET(p[lo]) < s) {
8477 /* Keep the early parts of this range. */
8478 int ack = BB_ACK(p[lo]);
8479 sector_t start = BB_OFFSET(p[lo]);
8480 p[lo] = BB_MAKE(start, s - start, ack);
8481 /* now low doesn't overlap, so.. */
8482 break;
8483 }
8484 lo--;
8485 }
8486 /* 'lo' is strictly before, 'hi' is strictly after,
8487 * anything between needs to be discarded
8488 */
8489 if (hi - lo > 1) {
8490 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8491 bb->count -= (hi - lo - 1);
8492 }
8493 }
8494
8495 bb->changed = 1;
8496 out:
8497 write_sequnlock_irq(&bb->lock);
8498 return rv;
8499 }
8500
8501 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8502 int is_new)
8503 {
8504 if (is_new)
8505 s += rdev->new_data_offset;
8506 else
8507 s += rdev->data_offset;
8508 return md_clear_badblocks(&rdev->badblocks,
8509 s, sectors);
8510 }
8511 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8512
8513 /*
8514 * Acknowledge all bad blocks in a list.
8515 * This only succeeds if ->changed is clear. It is used by
8516 * in-kernel metadata updates
8517 */
8518 void md_ack_all_badblocks(struct badblocks *bb)
8519 {
8520 if (bb->page == NULL || bb->changed)
8521 /* no point even trying */
8522 return;
8523 write_seqlock_irq(&bb->lock);
8524
8525 if (bb->changed == 0 && bb->unacked_exist) {
8526 u64 *p = bb->page;
8527 int i;
8528 for (i = 0; i < bb->count ; i++) {
8529 if (!BB_ACK(p[i])) {
8530 sector_t start = BB_OFFSET(p[i]);
8531 int len = BB_LEN(p[i]);
8532 p[i] = BB_MAKE(start, len, 1);
8533 }
8534 }
8535 bb->unacked_exist = 0;
8536 }
8537 write_sequnlock_irq(&bb->lock);
8538 }
8539 EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8540
8541 /* sysfs access to bad-blocks list.
8542 * We present two files.
8543 * 'bad-blocks' lists sector numbers and lengths of ranges that
8544 * are recorded as bad. The list is truncated to fit within
8545 * the one-page limit of sysfs.
8546 * Writing "sector length" to this file adds an acknowledged
8547 * bad block list.
8548 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8549 * been acknowledged. Writing to this file adds bad blocks
8550 * without acknowledging them. This is largely for testing.
8551 */
8552
8553 static ssize_t
8554 badblocks_show(struct badblocks *bb, char *page, int unack)
8555 {
8556 size_t len;
8557 int i;
8558 u64 *p = bb->page;
8559 unsigned seq;
8560
8561 if (bb->shift < 0)
8562 return 0;
8563
8564 retry:
8565 seq = read_seqbegin(&bb->lock);
8566
8567 len = 0;
8568 i = 0;
8569
8570 while (len < PAGE_SIZE && i < bb->count) {
8571 sector_t s = BB_OFFSET(p[i]);
8572 unsigned int length = BB_LEN(p[i]);
8573 int ack = BB_ACK(p[i]);
8574 i++;
8575
8576 if (unack && ack)
8577 continue;
8578
8579 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8580 (unsigned long long)s << bb->shift,
8581 length << bb->shift);
8582 }
8583 if (unack && len == 0)
8584 bb->unacked_exist = 0;
8585
8586 if (read_seqretry(&bb->lock, seq))
8587 goto retry;
8588
8589 return len;
8590 }
8591
8592 #define DO_DEBUG 1
8593
8594 static ssize_t
8595 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8596 {
8597 unsigned long long sector;
8598 int length;
8599 char newline;
8600 #ifdef DO_DEBUG
8601 /* Allow clearing via sysfs *only* for testing/debugging.
8602 * Normally only a successful write may clear a badblock
8603 */
8604 int clear = 0;
8605 if (page[0] == '-') {
8606 clear = 1;
8607 page++;
8608 }
8609 #endif /* DO_DEBUG */
8610
8611 switch (sscanf(page, "%llu %d%c", &sector, &length, &newline)) {
8612 case 3:
8613 if (newline != '\n')
8614 return -EINVAL;
8615 case 2:
8616 if (length <= 0)
8617 return -EINVAL;
8618 break;
8619 default:
8620 return -EINVAL;
8621 }
8622
8623 #ifdef DO_DEBUG
8624 if (clear) {
8625 md_clear_badblocks(bb, sector, length);
8626 return len;
8627 }
8628 #endif /* DO_DEBUG */
8629 if (md_set_badblocks(bb, sector, length, !unack))
8630 return len;
8631 else
8632 return -ENOSPC;
8633 }
8634
8635 static int md_notify_reboot(struct notifier_block *this,
8636 unsigned long code, void *x)
8637 {
8638 struct list_head *tmp;
8639 struct mddev *mddev;
8640 int need_delay = 0;
8641
8642 for_each_mddev(mddev, tmp) {
8643 if (mddev_trylock(mddev)) {
8644 if (mddev->pers)
8645 __md_stop_writes(mddev);
8646 if (mddev->persistent)
8647 mddev->safemode = 2;
8648 mddev_unlock(mddev);
8649 }
8650 need_delay = 1;
8651 }
8652 /*
8653 * certain more exotic SCSI devices are known to be
8654 * volatile wrt too early system reboots. While the
8655 * right place to handle this issue is the given
8656 * driver, we do want to have a safe RAID driver ...
8657 */
8658 if (need_delay)
8659 mdelay(1000*1);
8660
8661 return NOTIFY_DONE;
8662 }
8663
8664 static struct notifier_block md_notifier = {
8665 .notifier_call = md_notify_reboot,
8666 .next = NULL,
8667 .priority = INT_MAX, /* before any real devices */
8668 };
8669
8670 static void md_geninit(void)
8671 {
8672 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
8673
8674 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
8675 }
8676
8677 static int __init md_init(void)
8678 {
8679 int ret = -ENOMEM;
8680
8681 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
8682 if (!md_wq)
8683 goto err_wq;
8684
8685 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8686 if (!md_misc_wq)
8687 goto err_misc_wq;
8688
8689 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8690 goto err_md;
8691
8692 if ((ret = register_blkdev(0, "mdp")) < 0)
8693 goto err_mdp;
8694 mdp_major = ret;
8695
8696 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
8697 md_probe, NULL, NULL);
8698 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
8699 md_probe, NULL, NULL);
8700
8701 register_reboot_notifier(&md_notifier);
8702 raid_table_header = register_sysctl_table(raid_root_table);
8703
8704 md_geninit();
8705 return 0;
8706
8707 err_mdp:
8708 unregister_blkdev(MD_MAJOR, "md");
8709 err_md:
8710 destroy_workqueue(md_misc_wq);
8711 err_misc_wq:
8712 destroy_workqueue(md_wq);
8713 err_wq:
8714 return ret;
8715 }
8716
8717 #ifndef MODULE
8718
8719 /*
8720 * Searches all registered partitions for autorun RAID arrays
8721 * at boot time.
8722 */
8723
8724 static LIST_HEAD(all_detected_devices);
8725 struct detected_devices_node {
8726 struct list_head list;
8727 dev_t dev;
8728 };
8729
8730 void md_autodetect_dev(dev_t dev)
8731 {
8732 struct detected_devices_node *node_detected_dev;
8733
8734 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8735 if (node_detected_dev) {
8736 node_detected_dev->dev = dev;
8737 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8738 } else {
8739 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8740 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8741 }
8742 }
8743
8744 static void autostart_arrays(int part)
8745 {
8746 struct md_rdev *rdev;
8747 struct detected_devices_node *node_detected_dev;
8748 dev_t dev;
8749 int i_scanned, i_passed;
8750
8751 i_scanned = 0;
8752 i_passed = 0;
8753
8754 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
8755
8756 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8757 i_scanned++;
8758 node_detected_dev = list_entry(all_detected_devices.next,
8759 struct detected_devices_node, list);
8760 list_del(&node_detected_dev->list);
8761 dev = node_detected_dev->dev;
8762 kfree(node_detected_dev);
8763 rdev = md_import_device(dev,0, 90);
8764 if (IS_ERR(rdev))
8765 continue;
8766
8767 if (test_bit(Faulty, &rdev->flags))
8768 continue;
8769
8770 set_bit(AutoDetected, &rdev->flags);
8771 list_add(&rdev->same_set, &pending_raid_disks);
8772 i_passed++;
8773 }
8774
8775 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8776 i_scanned, i_passed);
8777
8778 autorun_devices(part);
8779 }
8780
8781 #endif /* !MODULE */
8782
8783 static __exit void md_exit(void)
8784 {
8785 struct mddev *mddev;
8786 struct list_head *tmp;
8787 int delay = 1;
8788
8789 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
8790 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
8791
8792 unregister_blkdev(MD_MAJOR,"md");
8793 unregister_blkdev(mdp_major, "mdp");
8794 unregister_reboot_notifier(&md_notifier);
8795 unregister_sysctl_table(raid_table_header);
8796
8797 /* We cannot unload the modules while some process is
8798 * waiting for us in select() or poll() - wake them up
8799 */
8800 md_unloading = 1;
8801 while (waitqueue_active(&md_event_waiters)) {
8802 /* not safe to leave yet */
8803 wake_up(&md_event_waiters);
8804 msleep(delay);
8805 delay += delay;
8806 }
8807 remove_proc_entry("mdstat", NULL);
8808
8809 for_each_mddev(mddev, tmp) {
8810 export_array(mddev);
8811 mddev->hold_active = 0;
8812 }
8813 destroy_workqueue(md_misc_wq);
8814 destroy_workqueue(md_wq);
8815 }
8816
8817 subsys_initcall(md_init);
8818 module_exit(md_exit)
8819
8820 static int get_ro(char *buffer, struct kernel_param *kp)
8821 {
8822 return sprintf(buffer, "%d", start_readonly);
8823 }
8824 static int set_ro(const char *val, struct kernel_param *kp)
8825 {
8826 char *e;
8827 int num = simple_strtoul(val, &e, 10);
8828 if (*val && (*e == '\0' || *e == '\n')) {
8829 start_readonly = num;
8830 return 0;
8831 }
8832 return -EINVAL;
8833 }
8834
8835 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8836 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
8837 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
8838
8839 MODULE_LICENSE("GPL");
8840 MODULE_DESCRIPTION("MD RAID framework");
8841 MODULE_ALIAS("md");
8842 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Linux kernel - Deliverables
This page took 0.207057 seconds and 6 git commands to generate.