projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge tag 'modules-next-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
[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
57 #ifndef MODULE
58 static void autostart_arrays(int part);
59 #endif
60
61 /* pers_list is a list of registered personalities protected
62 * by pers_lock.
63 * pers_lock does extra service to protect accesses to
64 * mddev->thread when the mutex cannot be held.
65 */
66 static LIST_HEAD(pers_list);
67 static DEFINE_SPINLOCK(pers_lock);
68
69 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
70 static struct workqueue_struct *md_wq;
71 static struct workqueue_struct *md_misc_wq;
72
73 static int remove_and_add_spares(struct mddev *mddev,
74 struct md_rdev *this);
75 static void mddev_detach(struct mddev *mddev);
76
77 /*
78 * Default number of read corrections we'll attempt on an rdev
79 * before ejecting it from the array. We divide the read error
80 * count by 2 for every hour elapsed between read errors.
81 */
82 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
83 /*
84 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
85 * is 1000 KB/sec, so the extra system load does not show up that much.
86 * Increase it if you want to have more _guaranteed_ speed. Note that
87 * the RAID driver will use the maximum available bandwidth if the IO
88 * subsystem is idle. There is also an 'absolute maximum' reconstruction
89 * speed limit - in case reconstruction slows down your system despite
90 * idle IO detection.
91 *
92 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
93 * or /sys/block/mdX/md/sync_speed_{min,max}
94 */
95
96 static int sysctl_speed_limit_min = 1000;
97 static int sysctl_speed_limit_max = 200000;
98 static inline int speed_min(struct mddev *mddev)
99 {
100 return mddev->sync_speed_min ?
101 mddev->sync_speed_min : sysctl_speed_limit_min;
102 }
103
104 static inline int speed_max(struct mddev *mddev)
105 {
106 return mddev->sync_speed_max ?
107 mddev->sync_speed_max : sysctl_speed_limit_max;
108 }
109
110 static struct ctl_table_header *raid_table_header;
111
112 static struct ctl_table raid_table[] = {
113 {
114 .procname = "speed_limit_min",
115 .data = &sysctl_speed_limit_min,
116 .maxlen = sizeof(int),
117 .mode = S_IRUGO|S_IWUSR,
118 .proc_handler = proc_dointvec,
119 },
120 {
121 .procname = "speed_limit_max",
122 .data = &sysctl_speed_limit_max,
123 .maxlen = sizeof(int),
124 .mode = S_IRUGO|S_IWUSR,
125 .proc_handler = proc_dointvec,
126 },
127 { }
128 };
129
130 static struct ctl_table raid_dir_table[] = {
131 {
132 .procname = "raid",
133 .maxlen = 0,
134 .mode = S_IRUGO|S_IXUGO,
135 .child = raid_table,
136 },
137 { }
138 };
139
140 static struct ctl_table raid_root_table[] = {
141 {
142 .procname = "dev",
143 .maxlen = 0,
144 .mode = 0555,
145 .child = raid_dir_table,
146 },
147 { }
148 };
149
150 static const struct block_device_operations md_fops;
151
152 static int start_readonly;
153
154 /* bio_clone_mddev
155 * like bio_clone, but with a local bio set
156 */
157
158 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
159 struct mddev *mddev)
160 {
161 struct bio *b;
162
163 if (!mddev || !mddev->bio_set)
164 return bio_alloc(gfp_mask, nr_iovecs);
165
166 b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set);
167 if (!b)
168 return NULL;
169 return b;
170 }
171 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
172
173 struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
174 struct mddev *mddev)
175 {
176 if (!mddev || !mddev->bio_set)
177 return bio_clone(bio, gfp_mask);
178
179 return bio_clone_bioset(bio, gfp_mask, mddev->bio_set);
180 }
181 EXPORT_SYMBOL_GPL(bio_clone_mddev);
182
183 /*
184 * We have a system wide 'event count' that is incremented
185 * on any 'interesting' event, and readers of /proc/mdstat
186 * can use 'poll' or 'select' to find out when the event
187 * count increases.
188 *
189 * Events are:
190 * start array, stop array, error, add device, remove device,
191 * start build, activate spare
192 */
193 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
194 static atomic_t md_event_count;
195 void md_new_event(struct mddev *mddev)
196 {
197 atomic_inc(&md_event_count);
198 wake_up(&md_event_waiters);
199 }
200 EXPORT_SYMBOL_GPL(md_new_event);
201
202 /* Alternate version that can be called from interrupts
203 * when calling sysfs_notify isn't needed.
204 */
205 static void md_new_event_inintr(struct mddev *mddev)
206 {
207 atomic_inc(&md_event_count);
208 wake_up(&md_event_waiters);
209 }
210
211 /*
212 * Enables to iterate over all existing md arrays
213 * all_mddevs_lock protects this list.
214 */
215 static LIST_HEAD(all_mddevs);
216 static DEFINE_SPINLOCK(all_mddevs_lock);
217
218 /*
219 * iterates through all used mddevs in the system.
220 * We take care to grab the all_mddevs_lock whenever navigating
221 * the list, and to always hold a refcount when unlocked.
222 * Any code which breaks out of this loop while own
223 * a reference to the current mddev and must mddev_put it.
224 */
225 #define for_each_mddev(_mddev,_tmp) \
226 \
227 for (({ spin_lock(&all_mddevs_lock); \
228 _tmp = all_mddevs.next; \
229 _mddev = NULL;}); \
230 ({ if (_tmp != &all_mddevs) \
231 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
232 spin_unlock(&all_mddevs_lock); \
233 if (_mddev) mddev_put(_mddev); \
234 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
235 _tmp != &all_mddevs;}); \
236 ({ spin_lock(&all_mddevs_lock); \
237 _tmp = _tmp->next;}) \
238 )
239
240 /* Rather than calling directly into the personality make_request function,
241 * IO requests come here first so that we can check if the device is
242 * being suspended pending a reconfiguration.
243 * We hold a refcount over the call to ->make_request. By the time that
244 * call has finished, the bio has been linked into some internal structure
245 * and so is visible to ->quiesce(), so we don't need the refcount any more.
246 */
247 static void md_make_request(struct request_queue *q, struct bio *bio)
248 {
249 const int rw = bio_data_dir(bio);
250 struct mddev *mddev = q->queuedata;
251 unsigned int sectors;
252
253 if (mddev == NULL || mddev->pers == NULL
254 || !mddev->ready) {
255 bio_io_error(bio);
256 return;
257 }
258 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
259 bio_endio(bio, bio_sectors(bio) == 0 ? 0 : -EROFS);
260 return;
261 }
262 smp_rmb(); /* Ensure implications of 'active' are visible */
263 rcu_read_lock();
264 if (mddev->suspended) {
265 DEFINE_WAIT(__wait);
266 for (;;) {
267 prepare_to_wait(&mddev->sb_wait, &__wait,
268 TASK_UNINTERRUPTIBLE);
269 if (!mddev->suspended)
270 break;
271 rcu_read_unlock();
272 schedule();
273 rcu_read_lock();
274 }
275 finish_wait(&mddev->sb_wait, &__wait);
276 }
277 atomic_inc(&mddev->active_io);
278 rcu_read_unlock();
279
280 /*
281 * save the sectors now since our bio can
282 * go away inside make_request
283 */
284 sectors = bio_sectors(bio);
285 mddev->pers->make_request(mddev, bio);
286
287 generic_start_io_acct(rw, sectors, &mddev->gendisk->part0);
288
289 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
290 wake_up(&mddev->sb_wait);
291 }
292
293 /* mddev_suspend makes sure no new requests are submitted
294 * to the device, and that any requests that have been submitted
295 * are completely handled.
296 * Once mddev_detach() is called and completes, the module will be
297 * completely unused.
298 */
299 void mddev_suspend(struct mddev *mddev)
300 {
301 BUG_ON(mddev->suspended);
302 mddev->suspended = 1;
303 synchronize_rcu();
304 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
305 mddev->pers->quiesce(mddev, 1);
306
307 del_timer_sync(&mddev->safemode_timer);
308 }
309 EXPORT_SYMBOL_GPL(mddev_suspend);
310
311 void mddev_resume(struct mddev *mddev)
312 {
313 mddev->suspended = 0;
314 wake_up(&mddev->sb_wait);
315 mddev->pers->quiesce(mddev, 0);
316
317 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
318 md_wakeup_thread(mddev->thread);
319 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
320 }
321 EXPORT_SYMBOL_GPL(mddev_resume);
322
323 int mddev_congested(struct mddev *mddev, int bits)
324 {
325 struct md_personality *pers = mddev->pers;
326 int ret = 0;
327
328 rcu_read_lock();
329 if (mddev->suspended)
330 ret = 1;
331 else if (pers && pers->congested)
332 ret = pers->congested(mddev, bits);
333 rcu_read_unlock();
334 return ret;
335 }
336 EXPORT_SYMBOL_GPL(mddev_congested);
337 static int md_congested(void *data, int bits)
338 {
339 struct mddev *mddev = data;
340 return mddev_congested(mddev, bits);
341 }
342
343 static int md_mergeable_bvec(struct request_queue *q,
344 struct bvec_merge_data *bvm,
345 struct bio_vec *biovec)
346 {
347 struct mddev *mddev = q->queuedata;
348 int ret;
349 rcu_read_lock();
350 if (mddev->suspended) {
351 /* Must always allow one vec */
352 if (bvm->bi_size == 0)
353 ret = biovec->bv_len;
354 else
355 ret = 0;
356 } else {
357 struct md_personality *pers = mddev->pers;
358 if (pers && pers->mergeable_bvec)
359 ret = pers->mergeable_bvec(mddev, bvm, biovec);
360 else
361 ret = biovec->bv_len;
362 }
363 rcu_read_unlock();
364 return ret;
365 }
366 /*
367 * Generic flush handling for md
368 */
369
370 static void md_end_flush(struct bio *bio, int err)
371 {
372 struct md_rdev *rdev = bio->bi_private;
373 struct mddev *mddev = rdev->mddev;
374
375 rdev_dec_pending(rdev, mddev);
376
377 if (atomic_dec_and_test(&mddev->flush_pending)) {
378 /* The pre-request flush has finished */
379 queue_work(md_wq, &mddev->flush_work);
380 }
381 bio_put(bio);
382 }
383
384 static void md_submit_flush_data(struct work_struct *ws);
385
386 static void submit_flushes(struct work_struct *ws)
387 {
388 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
389 struct md_rdev *rdev;
390
391 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
392 atomic_set(&mddev->flush_pending, 1);
393 rcu_read_lock();
394 rdev_for_each_rcu(rdev, mddev)
395 if (rdev->raid_disk >= 0 &&
396 !test_bit(Faulty, &rdev->flags)) {
397 /* Take two references, one is dropped
398 * when request finishes, one after
399 * we reclaim rcu_read_lock
400 */
401 struct bio *bi;
402 atomic_inc(&rdev->nr_pending);
403 atomic_inc(&rdev->nr_pending);
404 rcu_read_unlock();
405 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
406 bi->bi_end_io = md_end_flush;
407 bi->bi_private = rdev;
408 bi->bi_bdev = rdev->bdev;
409 atomic_inc(&mddev->flush_pending);
410 submit_bio(WRITE_FLUSH, bi);
411 rcu_read_lock();
412 rdev_dec_pending(rdev, mddev);
413 }
414 rcu_read_unlock();
415 if (atomic_dec_and_test(&mddev->flush_pending))
416 queue_work(md_wq, &mddev->flush_work);
417 }
418
419 static void md_submit_flush_data(struct work_struct *ws)
420 {
421 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
422 struct bio *bio = mddev->flush_bio;
423
424 if (bio->bi_iter.bi_size == 0)
425 /* an empty barrier - all done */
426 bio_endio(bio, 0);
427 else {
428 bio->bi_rw &= ~REQ_FLUSH;
429 mddev->pers->make_request(mddev, bio);
430 }
431
432 mddev->flush_bio = NULL;
433 wake_up(&mddev->sb_wait);
434 }
435
436 void md_flush_request(struct mddev *mddev, struct bio *bio)
437 {
438 spin_lock_irq(&mddev->lock);
439 wait_event_lock_irq(mddev->sb_wait,
440 !mddev->flush_bio,
441 mddev->lock);
442 mddev->flush_bio = bio;
443 spin_unlock_irq(&mddev->lock);
444
445 INIT_WORK(&mddev->flush_work, submit_flushes);
446 queue_work(md_wq, &mddev->flush_work);
447 }
448 EXPORT_SYMBOL(md_flush_request);
449
450 void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
451 {
452 struct mddev *mddev = cb->data;
453 md_wakeup_thread(mddev->thread);
454 kfree(cb);
455 }
456 EXPORT_SYMBOL(md_unplug);
457
458 static inline struct mddev *mddev_get(struct mddev *mddev)
459 {
460 atomic_inc(&mddev->active);
461 return mddev;
462 }
463
464 static void mddev_delayed_delete(struct work_struct *ws);
465
466 static void mddev_put(struct mddev *mddev)
467 {
468 struct bio_set *bs = NULL;
469
470 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
471 return;
472 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
473 mddev->ctime == 0 && !mddev->hold_active) {
474 /* Array is not configured at all, and not held active,
475 * so destroy it */
476 list_del_init(&mddev->all_mddevs);
477 bs = mddev->bio_set;
478 mddev->bio_set = NULL;
479 if (mddev->gendisk) {
480 /* We did a probe so need to clean up. Call
481 * queue_work inside the spinlock so that
482 * flush_workqueue() after mddev_find will
483 * succeed in waiting for the work to be done.
484 */
485 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
486 queue_work(md_misc_wq, &mddev->del_work);
487 } else
488 kfree(mddev);
489 }
490 spin_unlock(&all_mddevs_lock);
491 if (bs)
492 bioset_free(bs);
493 }
494
495 void mddev_init(struct mddev *mddev)
496 {
497 mutex_init(&mddev->open_mutex);
498 mutex_init(&mddev->reconfig_mutex);
499 mutex_init(&mddev->bitmap_info.mutex);
500 INIT_LIST_HEAD(&mddev->disks);
501 INIT_LIST_HEAD(&mddev->all_mddevs);
502 init_timer(&mddev->safemode_timer);
503 atomic_set(&mddev->active, 1);
504 atomic_set(&mddev->openers, 0);
505 atomic_set(&mddev->active_io, 0);
506 spin_lock_init(&mddev->lock);
507 atomic_set(&mddev->flush_pending, 0);
508 init_waitqueue_head(&mddev->sb_wait);
509 init_waitqueue_head(&mddev->recovery_wait);
510 mddev->reshape_position = MaxSector;
511 mddev->reshape_backwards = 0;
512 mddev->last_sync_action = "none";
513 mddev->resync_min = 0;
514 mddev->resync_max = MaxSector;
515 mddev->level = LEVEL_NONE;
516 }
517 EXPORT_SYMBOL_GPL(mddev_init);
518
519 static struct mddev *mddev_find(dev_t unit)
520 {
521 struct mddev *mddev, *new = NULL;
522
523 if (unit && MAJOR(unit) != MD_MAJOR)
524 unit &= ~((1<<MdpMinorShift)-1);
525
526 retry:
527 spin_lock(&all_mddevs_lock);
528
529 if (unit) {
530 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
531 if (mddev->unit == unit) {
532 mddev_get(mddev);
533 spin_unlock(&all_mddevs_lock);
534 kfree(new);
535 return mddev;
536 }
537
538 if (new) {
539 list_add(&new->all_mddevs, &all_mddevs);
540 spin_unlock(&all_mddevs_lock);
541 new->hold_active = UNTIL_IOCTL;
542 return new;
543 }
544 } else if (new) {
545 /* find an unused unit number */
546 static int next_minor = 512;
547 int start = next_minor;
548 int is_free = 0;
549 int dev = 0;
550 while (!is_free) {
551 dev = MKDEV(MD_MAJOR, next_minor);
552 next_minor++;
553 if (next_minor > MINORMASK)
554 next_minor = 0;
555 if (next_minor == start) {
556 /* Oh dear, all in use. */
557 spin_unlock(&all_mddevs_lock);
558 kfree(new);
559 return NULL;
560 }
561
562 is_free = 1;
563 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
564 if (mddev->unit == dev) {
565 is_free = 0;
566 break;
567 }
568 }
569 new->unit = dev;
570 new->md_minor = MINOR(dev);
571 new->hold_active = UNTIL_STOP;
572 list_add(&new->all_mddevs, &all_mddevs);
573 spin_unlock(&all_mddevs_lock);
574 return new;
575 }
576 spin_unlock(&all_mddevs_lock);
577
578 new = kzalloc(sizeof(*new), GFP_KERNEL);
579 if (!new)
580 return NULL;
581
582 new->unit = unit;
583 if (MAJOR(unit) == MD_MAJOR)
584 new->md_minor = MINOR(unit);
585 else
586 new->md_minor = MINOR(unit) >> MdpMinorShift;
587
588 mddev_init(new);
589
590 goto retry;
591 }
592
593 static struct attribute_group md_redundancy_group;
594
595 void mddev_unlock(struct mddev *mddev)
596 {
597 if (mddev->to_remove) {
598 /* These cannot be removed under reconfig_mutex as
599 * an access to the files will try to take reconfig_mutex
600 * while holding the file unremovable, which leads to
601 * a deadlock.
602 * So hold set sysfs_active while the remove in happeing,
603 * and anything else which might set ->to_remove or my
604 * otherwise change the sysfs namespace will fail with
605 * -EBUSY if sysfs_active is still set.
606 * We set sysfs_active under reconfig_mutex and elsewhere
607 * test it under the same mutex to ensure its correct value
608 * is seen.
609 */
610 struct attribute_group *to_remove = mddev->to_remove;
611 mddev->to_remove = NULL;
612 mddev->sysfs_active = 1;
613 mutex_unlock(&mddev->reconfig_mutex);
614
615 if (mddev->kobj.sd) {
616 if (to_remove != &md_redundancy_group)
617 sysfs_remove_group(&mddev->kobj, to_remove);
618 if (mddev->pers == NULL ||
619 mddev->pers->sync_request == NULL) {
620 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
621 if (mddev->sysfs_action)
622 sysfs_put(mddev->sysfs_action);
623 mddev->sysfs_action = NULL;
624 }
625 }
626 mddev->sysfs_active = 0;
627 } else
628 mutex_unlock(&mddev->reconfig_mutex);
629
630 /* As we've dropped the mutex we need a spinlock to
631 * make sure the thread doesn't disappear
632 */
633 spin_lock(&pers_lock);
634 md_wakeup_thread(mddev->thread);
635 spin_unlock(&pers_lock);
636 }
637 EXPORT_SYMBOL_GPL(mddev_unlock);
638
639 static struct md_rdev *find_rdev_nr_rcu(struct mddev *mddev, int nr)
640 {
641 struct md_rdev *rdev;
642
643 rdev_for_each_rcu(rdev, mddev)
644 if (rdev->desc_nr == nr)
645 return rdev;
646
647 return NULL;
648 }
649
650 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
651 {
652 struct md_rdev *rdev;
653
654 rdev_for_each(rdev, mddev)
655 if (rdev->bdev->bd_dev == dev)
656 return rdev;
657
658 return NULL;
659 }
660
661 static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
662 {
663 struct md_rdev *rdev;
664
665 rdev_for_each_rcu(rdev, mddev)
666 if (rdev->bdev->bd_dev == dev)
667 return rdev;
668
669 return NULL;
670 }
671
672 static struct md_personality *find_pers(int level, char *clevel)
673 {
674 struct md_personality *pers;
675 list_for_each_entry(pers, &pers_list, list) {
676 if (level != LEVEL_NONE && pers->level == level)
677 return pers;
678 if (strcmp(pers->name, clevel)==0)
679 return pers;
680 }
681 return NULL;
682 }
683
684 /* return the offset of the super block in 512byte sectors */
685 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
686 {
687 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
688 return MD_NEW_SIZE_SECTORS(num_sectors);
689 }
690
691 static int alloc_disk_sb(struct md_rdev *rdev)
692 {
693 rdev->sb_page = alloc_page(GFP_KERNEL);
694 if (!rdev->sb_page) {
695 printk(KERN_ALERT "md: out of memory.\n");
696 return -ENOMEM;
697 }
698
699 return 0;
700 }
701
702 void md_rdev_clear(struct md_rdev *rdev)
703 {
704 if (rdev->sb_page) {
705 put_page(rdev->sb_page);
706 rdev->sb_loaded = 0;
707 rdev->sb_page = NULL;
708 rdev->sb_start = 0;
709 rdev->sectors = 0;
710 }
711 if (rdev->bb_page) {
712 put_page(rdev->bb_page);
713 rdev->bb_page = NULL;
714 }
715 kfree(rdev->badblocks.page);
716 rdev->badblocks.page = NULL;
717 }
718 EXPORT_SYMBOL_GPL(md_rdev_clear);
719
720 static void super_written(struct bio *bio, int error)
721 {
722 struct md_rdev *rdev = bio->bi_private;
723 struct mddev *mddev = rdev->mddev;
724
725 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
726 printk("md: super_written gets error=%d, uptodate=%d\n",
727 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
728 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
729 md_error(mddev, rdev);
730 }
731
732 if (atomic_dec_and_test(&mddev->pending_writes))
733 wake_up(&mddev->sb_wait);
734 bio_put(bio);
735 }
736
737 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
738 sector_t sector, int size, struct page *page)
739 {
740 /* write first size bytes of page to sector of rdev
741 * Increment mddev->pending_writes before returning
742 * and decrement it on completion, waking up sb_wait
743 * if zero is reached.
744 * If an error occurred, call md_error
745 */
746 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
747
748 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
749 bio->bi_iter.bi_sector = sector;
750 bio_add_page(bio, page, size, 0);
751 bio->bi_private = rdev;
752 bio->bi_end_io = super_written;
753
754 atomic_inc(&mddev->pending_writes);
755 submit_bio(WRITE_FLUSH_FUA, bio);
756 }
757
758 void md_super_wait(struct mddev *mddev)
759 {
760 /* wait for all superblock writes that were scheduled to complete */
761 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
762 }
763
764 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
765 struct page *page, int rw, bool metadata_op)
766 {
767 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
768 int ret;
769
770 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
771 rdev->meta_bdev : rdev->bdev;
772 if (metadata_op)
773 bio->bi_iter.bi_sector = sector + rdev->sb_start;
774 else if (rdev->mddev->reshape_position != MaxSector &&
775 (rdev->mddev->reshape_backwards ==
776 (sector >= rdev->mddev->reshape_position)))
777 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
778 else
779 bio->bi_iter.bi_sector = sector + rdev->data_offset;
780 bio_add_page(bio, page, size, 0);
781 submit_bio_wait(rw, bio);
782
783 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
784 bio_put(bio);
785 return ret;
786 }
787 EXPORT_SYMBOL_GPL(sync_page_io);
788
789 static int read_disk_sb(struct md_rdev *rdev, int size)
790 {
791 char b[BDEVNAME_SIZE];
792
793 if (rdev->sb_loaded)
794 return 0;
795
796 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
797 goto fail;
798 rdev->sb_loaded = 1;
799 return 0;
800
801 fail:
802 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
803 bdevname(rdev->bdev,b));
804 return -EINVAL;
805 }
806
807 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
808 {
809 return sb1->set_uuid0 == sb2->set_uuid0 &&
810 sb1->set_uuid1 == sb2->set_uuid1 &&
811 sb1->set_uuid2 == sb2->set_uuid2 &&
812 sb1->set_uuid3 == sb2->set_uuid3;
813 }
814
815 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
816 {
817 int ret;
818 mdp_super_t *tmp1, *tmp2;
819
820 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
821 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
822
823 if (!tmp1 || !tmp2) {
824 ret = 0;
825 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
826 goto abort;
827 }
828
829 *tmp1 = *sb1;
830 *tmp2 = *sb2;
831
832 /*
833 * nr_disks is not constant
834 */
835 tmp1->nr_disks = 0;
836 tmp2->nr_disks = 0;
837
838 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
839 abort:
840 kfree(tmp1);
841 kfree(tmp2);
842 return ret;
843 }
844
845 static u32 md_csum_fold(u32 csum)
846 {
847 csum = (csum & 0xffff) + (csum >> 16);
848 return (csum & 0xffff) + (csum >> 16);
849 }
850
851 static unsigned int calc_sb_csum(mdp_super_t *sb)
852 {
853 u64 newcsum = 0;
854 u32 *sb32 = (u32*)sb;
855 int i;
856 unsigned int disk_csum, csum;
857
858 disk_csum = sb->sb_csum;
859 sb->sb_csum = 0;
860
861 for (i = 0; i < MD_SB_BYTES/4 ; i++)
862 newcsum += sb32[i];
863 csum = (newcsum & 0xffffffff) + (newcsum>>32);
864
865 #ifdef CONFIG_ALPHA
866 /* This used to use csum_partial, which was wrong for several
867 * reasons including that different results are returned on
868 * different architectures. It isn't critical that we get exactly
869 * the same return value as before (we always csum_fold before
870 * testing, and that removes any differences). However as we
871 * know that csum_partial always returned a 16bit value on
872 * alphas, do a fold to maximise conformity to previous behaviour.
873 */
874 sb->sb_csum = md_csum_fold(disk_csum);
875 #else
876 sb->sb_csum = disk_csum;
877 #endif
878 return csum;
879 }
880
881 /*
882 * Handle superblock details.
883 * We want to be able to handle multiple superblock formats
884 * so we have a common interface to them all, and an array of
885 * different handlers.
886 * We rely on user-space to write the initial superblock, and support
887 * reading and updating of superblocks.
888 * Interface methods are:
889 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
890 * loads and validates a superblock on dev.
891 * if refdev != NULL, compare superblocks on both devices
892 * Return:
893 * 0 - dev has a superblock that is compatible with refdev
894 * 1 - dev has a superblock that is compatible and newer than refdev
895 * so dev should be used as the refdev in future
896 * -EINVAL superblock incompatible or invalid
897 * -othererror e.g. -EIO
898 *
899 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
900 * Verify that dev is acceptable into mddev.
901 * The first time, mddev->raid_disks will be 0, and data from
902 * dev should be merged in. Subsequent calls check that dev
903 * is new enough. Return 0 or -EINVAL
904 *
905 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
906 * Update the superblock for rdev with data in mddev
907 * This does not write to disc.
908 *
909 */
910
911 struct super_type {
912 char *name;
913 struct module *owner;
914 int (*load_super)(struct md_rdev *rdev,
915 struct md_rdev *refdev,
916 int minor_version);
917 int (*validate_super)(struct mddev *mddev,
918 struct md_rdev *rdev);
919 void (*sync_super)(struct mddev *mddev,
920 struct md_rdev *rdev);
921 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
922 sector_t num_sectors);
923 int (*allow_new_offset)(struct md_rdev *rdev,
924 unsigned long long new_offset);
925 };
926
927 /*
928 * Check that the given mddev has no bitmap.
929 *
930 * This function is called from the run method of all personalities that do not
931 * support bitmaps. It prints an error message and returns non-zero if mddev
932 * has a bitmap. Otherwise, it returns 0.
933 *
934 */
935 int md_check_no_bitmap(struct mddev *mddev)
936 {
937 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
938 return 0;
939 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
940 mdname(mddev), mddev->pers->name);
941 return 1;
942 }
943 EXPORT_SYMBOL(md_check_no_bitmap);
944
945 /*
946 * load_super for 0.90.0
947 */
948 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
949 {
950 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
951 mdp_super_t *sb;
952 int ret;
953
954 /*
955 * Calculate the position of the superblock (512byte sectors),
956 * it's at the end of the disk.
957 *
958 * It also happens to be a multiple of 4Kb.
959 */
960 rdev->sb_start = calc_dev_sboffset(rdev);
961
962 ret = read_disk_sb(rdev, MD_SB_BYTES);
963 if (ret) return ret;
964
965 ret = -EINVAL;
966
967 bdevname(rdev->bdev, b);
968 sb = page_address(rdev->sb_page);
969
970 if (sb->md_magic != MD_SB_MAGIC) {
971 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
972 b);
973 goto abort;
974 }
975
976 if (sb->major_version != 0 ||
977 sb->minor_version < 90 ||
978 sb->minor_version > 91) {
979 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
980 sb->major_version, sb->minor_version,
981 b);
982 goto abort;
983 }
984
985 if (sb->raid_disks <= 0)
986 goto abort;
987
988 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
989 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
990 b);
991 goto abort;
992 }
993
994 rdev->preferred_minor = sb->md_minor;
995 rdev->data_offset = 0;
996 rdev->new_data_offset = 0;
997 rdev->sb_size = MD_SB_BYTES;
998 rdev->badblocks.shift = -1;
999
1000 if (sb->level == LEVEL_MULTIPATH)
1001 rdev->desc_nr = -1;
1002 else
1003 rdev->desc_nr = sb->this_disk.number;
1004
1005 if (!refdev) {
1006 ret = 1;
1007 } else {
1008 __u64 ev1, ev2;
1009 mdp_super_t *refsb = page_address(refdev->sb_page);
1010 if (!uuid_equal(refsb, sb)) {
1011 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1012 b, bdevname(refdev->bdev,b2));
1013 goto abort;
1014 }
1015 if (!sb_equal(refsb, sb)) {
1016 printk(KERN_WARNING "md: %s has same UUID"
1017 " but different superblock to %s\n",
1018 b, bdevname(refdev->bdev, b2));
1019 goto abort;
1020 }
1021 ev1 = md_event(sb);
1022 ev2 = md_event(refsb);
1023 if (ev1 > ev2)
1024 ret = 1;
1025 else
1026 ret = 0;
1027 }
1028 rdev->sectors = rdev->sb_start;
1029 /* Limit to 4TB as metadata cannot record more than that.
1030 * (not needed for Linear and RAID0 as metadata doesn't
1031 * record this size)
1032 */
1033 if (rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1034 rdev->sectors = (2ULL << 32) - 2;
1035
1036 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1037 /* "this cannot possibly happen" ... */
1038 ret = -EINVAL;
1039
1040 abort:
1041 return ret;
1042 }
1043
1044 /*
1045 * validate_super for 0.90.0
1046 */
1047 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1048 {
1049 mdp_disk_t *desc;
1050 mdp_super_t *sb = page_address(rdev->sb_page);
1051 __u64 ev1 = md_event(sb);
1052
1053 rdev->raid_disk = -1;
1054 clear_bit(Faulty, &rdev->flags);
1055 clear_bit(In_sync, &rdev->flags);
1056 clear_bit(Bitmap_sync, &rdev->flags);
1057 clear_bit(WriteMostly, &rdev->flags);
1058
1059 if (mddev->raid_disks == 0) {
1060 mddev->major_version = 0;
1061 mddev->minor_version = sb->minor_version;
1062 mddev->patch_version = sb->patch_version;
1063 mddev->external = 0;
1064 mddev->chunk_sectors = sb->chunk_size >> 9;
1065 mddev->ctime = sb->ctime;
1066 mddev->utime = sb->utime;
1067 mddev->level = sb->level;
1068 mddev->clevel[0] = 0;
1069 mddev->layout = sb->layout;
1070 mddev->raid_disks = sb->raid_disks;
1071 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1072 mddev->events = ev1;
1073 mddev->bitmap_info.offset = 0;
1074 mddev->bitmap_info.space = 0;
1075 /* bitmap can use 60 K after the 4K superblocks */
1076 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1077 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1078 mddev->reshape_backwards = 0;
1079
1080 if (mddev->minor_version >= 91) {
1081 mddev->reshape_position = sb->reshape_position;
1082 mddev->delta_disks = sb->delta_disks;
1083 mddev->new_level = sb->new_level;
1084 mddev->new_layout = sb->new_layout;
1085 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1086 if (mddev->delta_disks < 0)
1087 mddev->reshape_backwards = 1;
1088 } else {
1089 mddev->reshape_position = MaxSector;
1090 mddev->delta_disks = 0;
1091 mddev->new_level = mddev->level;
1092 mddev->new_layout = mddev->layout;
1093 mddev->new_chunk_sectors = mddev->chunk_sectors;
1094 }
1095
1096 if (sb->state & (1<<MD_SB_CLEAN))
1097 mddev->recovery_cp = MaxSector;
1098 else {
1099 if (sb->events_hi == sb->cp_events_hi &&
1100 sb->events_lo == sb->cp_events_lo) {
1101 mddev->recovery_cp = sb->recovery_cp;
1102 } else
1103 mddev->recovery_cp = 0;
1104 }
1105
1106 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1107 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1108 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1109 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1110
1111 mddev->max_disks = MD_SB_DISKS;
1112
1113 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1114 mddev->bitmap_info.file == NULL) {
1115 mddev->bitmap_info.offset =
1116 mddev->bitmap_info.default_offset;
1117 mddev->bitmap_info.space =
1118 mddev->bitmap_info.default_space;
1119 }
1120
1121 } else if (mddev->pers == NULL) {
1122 /* Insist on good event counter while assembling, except
1123 * for spares (which don't need an event count) */
1124 ++ev1;
1125 if (sb->disks[rdev->desc_nr].state & (
1126 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1127 if (ev1 < mddev->events)
1128 return -EINVAL;
1129 } else if (mddev->bitmap) {
1130 /* if adding to array with a bitmap, then we can accept an
1131 * older device ... but not too old.
1132 */
1133 if (ev1 < mddev->bitmap->events_cleared)
1134 return 0;
1135 if (ev1 < mddev->events)
1136 set_bit(Bitmap_sync, &rdev->flags);
1137 } else {
1138 if (ev1 < mddev->events)
1139 /* just a hot-add of a new device, leave raid_disk at -1 */
1140 return 0;
1141 }
1142
1143 if (mddev->level != LEVEL_MULTIPATH) {
1144 desc = sb->disks + rdev->desc_nr;
1145
1146 if (desc->state & (1<<MD_DISK_FAULTY))
1147 set_bit(Faulty, &rdev->flags);
1148 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1149 desc->raid_disk < mddev->raid_disks */) {
1150 set_bit(In_sync, &rdev->flags);
1151 rdev->raid_disk = desc->raid_disk;
1152 rdev->saved_raid_disk = desc->raid_disk;
1153 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1154 /* active but not in sync implies recovery up to
1155 * reshape position. We don't know exactly where
1156 * that is, so set to zero for now */
1157 if (mddev->minor_version >= 91) {
1158 rdev->recovery_offset = 0;
1159 rdev->raid_disk = desc->raid_disk;
1160 }
1161 }
1162 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1163 set_bit(WriteMostly, &rdev->flags);
1164 } else /* MULTIPATH are always insync */
1165 set_bit(In_sync, &rdev->flags);
1166 return 0;
1167 }
1168
1169 /*
1170 * sync_super for 0.90.0
1171 */
1172 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1173 {
1174 mdp_super_t *sb;
1175 struct md_rdev *rdev2;
1176 int next_spare = mddev->raid_disks;
1177
1178 /* make rdev->sb match mddev data..
1179 *
1180 * 1/ zero out disks
1181 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1182 * 3/ any empty disks < next_spare become removed
1183 *
1184 * disks[0] gets initialised to REMOVED because
1185 * we cannot be sure from other fields if it has
1186 * been initialised or not.
1187 */
1188 int i;
1189 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1190
1191 rdev->sb_size = MD_SB_BYTES;
1192
1193 sb = page_address(rdev->sb_page);
1194
1195 memset(sb, 0, sizeof(*sb));
1196
1197 sb->md_magic = MD_SB_MAGIC;
1198 sb->major_version = mddev->major_version;
1199 sb->patch_version = mddev->patch_version;
1200 sb->gvalid_words = 0; /* ignored */
1201 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1202 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1203 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1204 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1205
1206 sb->ctime = mddev->ctime;
1207 sb->level = mddev->level;
1208 sb->size = mddev->dev_sectors / 2;
1209 sb->raid_disks = mddev->raid_disks;
1210 sb->md_minor = mddev->md_minor;
1211 sb->not_persistent = 0;
1212 sb->utime = mddev->utime;
1213 sb->state = 0;
1214 sb->events_hi = (mddev->events>>32);
1215 sb->events_lo = (u32)mddev->events;
1216
1217 if (mddev->reshape_position == MaxSector)
1218 sb->minor_version = 90;
1219 else {
1220 sb->minor_version = 91;
1221 sb->reshape_position = mddev->reshape_position;
1222 sb->new_level = mddev->new_level;
1223 sb->delta_disks = mddev->delta_disks;
1224 sb->new_layout = mddev->new_layout;
1225 sb->new_chunk = mddev->new_chunk_sectors << 9;
1226 }
1227 mddev->minor_version = sb->minor_version;
1228 if (mddev->in_sync)
1229 {
1230 sb->recovery_cp = mddev->recovery_cp;
1231 sb->cp_events_hi = (mddev->events>>32);
1232 sb->cp_events_lo = (u32)mddev->events;
1233 if (mddev->recovery_cp == MaxSector)
1234 sb->state = (1<< MD_SB_CLEAN);
1235 } else
1236 sb->recovery_cp = 0;
1237
1238 sb->layout = mddev->layout;
1239 sb->chunk_size = mddev->chunk_sectors << 9;
1240
1241 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1242 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1243
1244 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1245 rdev_for_each(rdev2, mddev) {
1246 mdp_disk_t *d;
1247 int desc_nr;
1248 int is_active = test_bit(In_sync, &rdev2->flags);
1249
1250 if (rdev2->raid_disk >= 0 &&
1251 sb->minor_version >= 91)
1252 /* we have nowhere to store the recovery_offset,
1253 * but if it is not below the reshape_position,
1254 * we can piggy-back on that.
1255 */
1256 is_active = 1;
1257 if (rdev2->raid_disk < 0 ||
1258 test_bit(Faulty, &rdev2->flags))
1259 is_active = 0;
1260 if (is_active)
1261 desc_nr = rdev2->raid_disk;
1262 else
1263 desc_nr = next_spare++;
1264 rdev2->desc_nr = desc_nr;
1265 d = &sb->disks[rdev2->desc_nr];
1266 nr_disks++;
1267 d->number = rdev2->desc_nr;
1268 d->major = MAJOR(rdev2->bdev->bd_dev);
1269 d->minor = MINOR(rdev2->bdev->bd_dev);
1270 if (is_active)
1271 d->raid_disk = rdev2->raid_disk;
1272 else
1273 d->raid_disk = rdev2->desc_nr; /* compatibility */
1274 if (test_bit(Faulty, &rdev2->flags))
1275 d->state = (1<<MD_DISK_FAULTY);
1276 else if (is_active) {
1277 d->state = (1<<MD_DISK_ACTIVE);
1278 if (test_bit(In_sync, &rdev2->flags))
1279 d->state |= (1<<MD_DISK_SYNC);
1280 active++;
1281 working++;
1282 } else {
1283 d->state = 0;
1284 spare++;
1285 working++;
1286 }
1287 if (test_bit(WriteMostly, &rdev2->flags))
1288 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1289 }
1290 /* now set the "removed" and "faulty" bits on any missing devices */
1291 for (i=0 ; i < mddev->raid_disks ; i++) {
1292 mdp_disk_t *d = &sb->disks[i];
1293 if (d->state == 0 && d->number == 0) {
1294 d->number = i;
1295 d->raid_disk = i;
1296 d->state = (1<<MD_DISK_REMOVED);
1297 d->state |= (1<<MD_DISK_FAULTY);
1298 failed++;
1299 }
1300 }
1301 sb->nr_disks = nr_disks;
1302 sb->active_disks = active;
1303 sb->working_disks = working;
1304 sb->failed_disks = failed;
1305 sb->spare_disks = spare;
1306
1307 sb->this_disk = sb->disks[rdev->desc_nr];
1308 sb->sb_csum = calc_sb_csum(sb);
1309 }
1310
1311 /*
1312 * rdev_size_change for 0.90.0
1313 */
1314 static unsigned long long
1315 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1316 {
1317 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1318 return 0; /* component must fit device */
1319 if (rdev->mddev->bitmap_info.offset)
1320 return 0; /* can't move bitmap */
1321 rdev->sb_start = calc_dev_sboffset(rdev);
1322 if (!num_sectors || num_sectors > rdev->sb_start)
1323 num_sectors = rdev->sb_start;
1324 /* Limit to 4TB as metadata cannot record more than that.
1325 * 4TB == 2^32 KB, or 2*2^32 sectors.
1326 */
1327 if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1328 num_sectors = (2ULL << 32) - 2;
1329 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1330 rdev->sb_page);
1331 md_super_wait(rdev->mddev);
1332 return num_sectors;
1333 }
1334
1335 static int
1336 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1337 {
1338 /* non-zero offset changes not possible with v0.90 */
1339 return new_offset == 0;
1340 }
1341
1342 /*
1343 * version 1 superblock
1344 */
1345
1346 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1347 {
1348 __le32 disk_csum;
1349 u32 csum;
1350 unsigned long long newcsum;
1351 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1352 __le32 *isuper = (__le32*)sb;
1353
1354 disk_csum = sb->sb_csum;
1355 sb->sb_csum = 0;
1356 newcsum = 0;
1357 for (; size >= 4; size -= 4)
1358 newcsum += le32_to_cpu(*isuper++);
1359
1360 if (size == 2)
1361 newcsum += le16_to_cpu(*(__le16*) isuper);
1362
1363 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1364 sb->sb_csum = disk_csum;
1365 return cpu_to_le32(csum);
1366 }
1367
1368 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1369 int acknowledged);
1370 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1371 {
1372 struct mdp_superblock_1 *sb;
1373 int ret;
1374 sector_t sb_start;
1375 sector_t sectors;
1376 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1377 int bmask;
1378
1379 /*
1380 * Calculate the position of the superblock in 512byte sectors.
1381 * It is always aligned to a 4K boundary and
1382 * depeding on minor_version, it can be:
1383 * 0: At least 8K, but less than 12K, from end of device
1384 * 1: At start of device
1385 * 2: 4K from start of device.
1386 */
1387 switch(minor_version) {
1388 case 0:
1389 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1390 sb_start -= 8*2;
1391 sb_start &= ~(sector_t)(4*2-1);
1392 break;
1393 case 1:
1394 sb_start = 0;
1395 break;
1396 case 2:
1397 sb_start = 8;
1398 break;
1399 default:
1400 return -EINVAL;
1401 }
1402 rdev->sb_start = sb_start;
1403
1404 /* superblock is rarely larger than 1K, but it can be larger,
1405 * and it is safe to read 4k, so we do that
1406 */
1407 ret = read_disk_sb(rdev, 4096);
1408 if (ret) return ret;
1409
1410 sb = page_address(rdev->sb_page);
1411
1412 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1413 sb->major_version != cpu_to_le32(1) ||
1414 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1415 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1416 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1417 return -EINVAL;
1418
1419 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1420 printk("md: invalid superblock checksum on %s\n",
1421 bdevname(rdev->bdev,b));
1422 return -EINVAL;
1423 }
1424 if (le64_to_cpu(sb->data_size) < 10) {
1425 printk("md: data_size too small on %s\n",
1426 bdevname(rdev->bdev,b));
1427 return -EINVAL;
1428 }
1429 if (sb->pad0 ||
1430 sb->pad3[0] ||
1431 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1432 /* Some padding is non-zero, might be a new feature */
1433 return -EINVAL;
1434
1435 rdev->preferred_minor = 0xffff;
1436 rdev->data_offset = le64_to_cpu(sb->data_offset);
1437 rdev->new_data_offset = rdev->data_offset;
1438 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1439 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1440 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1441 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1442
1443 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1444 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1445 if (rdev->sb_size & bmask)
1446 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1447
1448 if (minor_version
1449 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1450 return -EINVAL;
1451 if (minor_version
1452 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1453 return -EINVAL;
1454
1455 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1456 rdev->desc_nr = -1;
1457 else
1458 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1459
1460 if (!rdev->bb_page) {
1461 rdev->bb_page = alloc_page(GFP_KERNEL);
1462 if (!rdev->bb_page)
1463 return -ENOMEM;
1464 }
1465 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1466 rdev->badblocks.count == 0) {
1467 /* need to load the bad block list.
1468 * Currently we limit it to one page.
1469 */
1470 s32 offset;
1471 sector_t bb_sector;
1472 u64 *bbp;
1473 int i;
1474 int sectors = le16_to_cpu(sb->bblog_size);
1475 if (sectors > (PAGE_SIZE / 512))
1476 return -EINVAL;
1477 offset = le32_to_cpu(sb->bblog_offset);
1478 if (offset == 0)
1479 return -EINVAL;
1480 bb_sector = (long long)offset;
1481 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1482 rdev->bb_page, READ, true))
1483 return -EIO;
1484 bbp = (u64 *)page_address(rdev->bb_page);
1485 rdev->badblocks.shift = sb->bblog_shift;
1486 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1487 u64 bb = le64_to_cpu(*bbp);
1488 int count = bb & (0x3ff);
1489 u64 sector = bb >> 10;
1490 sector <<= sb->bblog_shift;
1491 count <<= sb->bblog_shift;
1492 if (bb + 1 == 0)
1493 break;
1494 if (md_set_badblocks(&rdev->badblocks,
1495 sector, count, 1) == 0)
1496 return -EINVAL;
1497 }
1498 } else if (sb->bblog_offset != 0)
1499 rdev->badblocks.shift = 0;
1500
1501 if (!refdev) {
1502 ret = 1;
1503 } else {
1504 __u64 ev1, ev2;
1505 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1506
1507 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1508 sb->level != refsb->level ||
1509 sb->layout != refsb->layout ||
1510 sb->chunksize != refsb->chunksize) {
1511 printk(KERN_WARNING "md: %s has strangely different"
1512 " superblock to %s\n",
1513 bdevname(rdev->bdev,b),
1514 bdevname(refdev->bdev,b2));
1515 return -EINVAL;
1516 }
1517 ev1 = le64_to_cpu(sb->events);
1518 ev2 = le64_to_cpu(refsb->events);
1519
1520 if (ev1 > ev2)
1521 ret = 1;
1522 else
1523 ret = 0;
1524 }
1525 if (minor_version) {
1526 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1527 sectors -= rdev->data_offset;
1528 } else
1529 sectors = rdev->sb_start;
1530 if (sectors < le64_to_cpu(sb->data_size))
1531 return -EINVAL;
1532 rdev->sectors = le64_to_cpu(sb->data_size);
1533 return ret;
1534 }
1535
1536 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1537 {
1538 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1539 __u64 ev1 = le64_to_cpu(sb->events);
1540
1541 rdev->raid_disk = -1;
1542 clear_bit(Faulty, &rdev->flags);
1543 clear_bit(In_sync, &rdev->flags);
1544 clear_bit(Bitmap_sync, &rdev->flags);
1545 clear_bit(WriteMostly, &rdev->flags);
1546
1547 if (mddev->raid_disks == 0) {
1548 mddev->major_version = 1;
1549 mddev->patch_version = 0;
1550 mddev->external = 0;
1551 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1552 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1553 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1554 mddev->level = le32_to_cpu(sb->level);
1555 mddev->clevel[0] = 0;
1556 mddev->layout = le32_to_cpu(sb->layout);
1557 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1558 mddev->dev_sectors = le64_to_cpu(sb->size);
1559 mddev->events = ev1;
1560 mddev->bitmap_info.offset = 0;
1561 mddev->bitmap_info.space = 0;
1562 /* Default location for bitmap is 1K after superblock
1563 * using 3K - total of 4K
1564 */
1565 mddev->bitmap_info.default_offset = 1024 >> 9;
1566 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1567 mddev->reshape_backwards = 0;
1568
1569 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1570 memcpy(mddev->uuid, sb->set_uuid, 16);
1571
1572 mddev->max_disks = (4096-256)/2;
1573
1574 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1575 mddev->bitmap_info.file == NULL) {
1576 mddev->bitmap_info.offset =
1577 (__s32)le32_to_cpu(sb->bitmap_offset);
1578 /* Metadata doesn't record how much space is available.
1579 * For 1.0, we assume we can use up to the superblock
1580 * if before, else to 4K beyond superblock.
1581 * For others, assume no change is possible.
1582 */
1583 if (mddev->minor_version > 0)
1584 mddev->bitmap_info.space = 0;
1585 else if (mddev->bitmap_info.offset > 0)
1586 mddev->bitmap_info.space =
1587 8 - mddev->bitmap_info.offset;
1588 else
1589 mddev->bitmap_info.space =
1590 -mddev->bitmap_info.offset;
1591 }
1592
1593 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1594 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1595 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1596 mddev->new_level = le32_to_cpu(sb->new_level);
1597 mddev->new_layout = le32_to_cpu(sb->new_layout);
1598 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1599 if (mddev->delta_disks < 0 ||
1600 (mddev->delta_disks == 0 &&
1601 (le32_to_cpu(sb->feature_map)
1602 & MD_FEATURE_RESHAPE_BACKWARDS)))
1603 mddev->reshape_backwards = 1;
1604 } else {
1605 mddev->reshape_position = MaxSector;
1606 mddev->delta_disks = 0;
1607 mddev->new_level = mddev->level;
1608 mddev->new_layout = mddev->layout;
1609 mddev->new_chunk_sectors = mddev->chunk_sectors;
1610 }
1611
1612 } else if (mddev->pers == NULL) {
1613 /* Insist of good event counter while assembling, except for
1614 * spares (which don't need an event count) */
1615 ++ev1;
1616 if (rdev->desc_nr >= 0 &&
1617 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1618 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1619 if (ev1 < mddev->events)
1620 return -EINVAL;
1621 } else if (mddev->bitmap) {
1622 /* If adding to array with a bitmap, then we can accept an
1623 * older device, but not too old.
1624 */
1625 if (ev1 < mddev->bitmap->events_cleared)
1626 return 0;
1627 if (ev1 < mddev->events)
1628 set_bit(Bitmap_sync, &rdev->flags);
1629 } else {
1630 if (ev1 < mddev->events)
1631 /* just a hot-add of a new device, leave raid_disk at -1 */
1632 return 0;
1633 }
1634 if (mddev->level != LEVEL_MULTIPATH) {
1635 int role;
1636 if (rdev->desc_nr < 0 ||
1637 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1638 role = 0xffff;
1639 rdev->desc_nr = -1;
1640 } else
1641 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1642 switch(role) {
1643 case 0xffff: /* spare */
1644 break;
1645 case 0xfffe: /* faulty */
1646 set_bit(Faulty, &rdev->flags);
1647 break;
1648 default:
1649 rdev->saved_raid_disk = role;
1650 if ((le32_to_cpu(sb->feature_map) &
1651 MD_FEATURE_RECOVERY_OFFSET)) {
1652 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1653 if (!(le32_to_cpu(sb->feature_map) &
1654 MD_FEATURE_RECOVERY_BITMAP))
1655 rdev->saved_raid_disk = -1;
1656 } else
1657 set_bit(In_sync, &rdev->flags);
1658 rdev->raid_disk = role;
1659 break;
1660 }
1661 if (sb->devflags & WriteMostly1)
1662 set_bit(WriteMostly, &rdev->flags);
1663 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1664 set_bit(Replacement, &rdev->flags);
1665 } else /* MULTIPATH are always insync */
1666 set_bit(In_sync, &rdev->flags);
1667
1668 return 0;
1669 }
1670
1671 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1672 {
1673 struct mdp_superblock_1 *sb;
1674 struct md_rdev *rdev2;
1675 int max_dev, i;
1676 /* make rdev->sb match mddev and rdev data. */
1677
1678 sb = page_address(rdev->sb_page);
1679
1680 sb->feature_map = 0;
1681 sb->pad0 = 0;
1682 sb->recovery_offset = cpu_to_le64(0);
1683 memset(sb->pad3, 0, sizeof(sb->pad3));
1684
1685 sb->utime = cpu_to_le64((__u64)mddev->utime);
1686 sb->events = cpu_to_le64(mddev->events);
1687 if (mddev->in_sync)
1688 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1689 else
1690 sb->resync_offset = cpu_to_le64(0);
1691
1692 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1693
1694 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1695 sb->size = cpu_to_le64(mddev->dev_sectors);
1696 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1697 sb->level = cpu_to_le32(mddev->level);
1698 sb->layout = cpu_to_le32(mddev->layout);
1699
1700 if (test_bit(WriteMostly, &rdev->flags))
1701 sb->devflags |= WriteMostly1;
1702 else
1703 sb->devflags &= ~WriteMostly1;
1704 sb->data_offset = cpu_to_le64(rdev->data_offset);
1705 sb->data_size = cpu_to_le64(rdev->sectors);
1706
1707 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1708 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1709 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1710 }
1711
1712 if (rdev->raid_disk >= 0 &&
1713 !test_bit(In_sync, &rdev->flags)) {
1714 sb->feature_map |=
1715 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1716 sb->recovery_offset =
1717 cpu_to_le64(rdev->recovery_offset);
1718 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1719 sb->feature_map |=
1720 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
1721 }
1722 if (test_bit(Replacement, &rdev->flags))
1723 sb->feature_map |=
1724 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1725
1726 if (mddev->reshape_position != MaxSector) {
1727 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1728 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1729 sb->new_layout = cpu_to_le32(mddev->new_layout);
1730 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1731 sb->new_level = cpu_to_le32(mddev->new_level);
1732 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1733 if (mddev->delta_disks == 0 &&
1734 mddev->reshape_backwards)
1735 sb->feature_map
1736 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1737 if (rdev->new_data_offset != rdev->data_offset) {
1738 sb->feature_map
1739 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1740 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1741 - rdev->data_offset));
1742 }
1743 }
1744
1745 if (rdev->badblocks.count == 0)
1746 /* Nothing to do for bad blocks*/ ;
1747 else if (sb->bblog_offset == 0)
1748 /* Cannot record bad blocks on this device */
1749 md_error(mddev, rdev);
1750 else {
1751 struct badblocks *bb = &rdev->badblocks;
1752 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1753 u64 *p = bb->page;
1754 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1755 if (bb->changed) {
1756 unsigned seq;
1757
1758 retry:
1759 seq = read_seqbegin(&bb->lock);
1760
1761 memset(bbp, 0xff, PAGE_SIZE);
1762
1763 for (i = 0 ; i < bb->count ; i++) {
1764 u64 internal_bb = p[i];
1765 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1766 | BB_LEN(internal_bb));
1767 bbp[i] = cpu_to_le64(store_bb);
1768 }
1769 bb->changed = 0;
1770 if (read_seqretry(&bb->lock, seq))
1771 goto retry;
1772
1773 bb->sector = (rdev->sb_start +
1774 (int)le32_to_cpu(sb->bblog_offset));
1775 bb->size = le16_to_cpu(sb->bblog_size);
1776 }
1777 }
1778
1779 max_dev = 0;
1780 rdev_for_each(rdev2, mddev)
1781 if (rdev2->desc_nr+1 > max_dev)
1782 max_dev = rdev2->desc_nr+1;
1783
1784 if (max_dev > le32_to_cpu(sb->max_dev)) {
1785 int bmask;
1786 sb->max_dev = cpu_to_le32(max_dev);
1787 rdev->sb_size = max_dev * 2 + 256;
1788 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1789 if (rdev->sb_size & bmask)
1790 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1791 } else
1792 max_dev = le32_to_cpu(sb->max_dev);
1793
1794 for (i=0; i<max_dev;i++)
1795 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1796
1797 rdev_for_each(rdev2, mddev) {
1798 i = rdev2->desc_nr;
1799 if (test_bit(Faulty, &rdev2->flags))
1800 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1801 else if (test_bit(In_sync, &rdev2->flags))
1802 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1803 else if (rdev2->raid_disk >= 0)
1804 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1805 else
1806 sb->dev_roles[i] = cpu_to_le16(0xffff);
1807 }
1808
1809 sb->sb_csum = calc_sb_1_csum(sb);
1810 }
1811
1812 static unsigned long long
1813 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1814 {
1815 struct mdp_superblock_1 *sb;
1816 sector_t max_sectors;
1817 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1818 return 0; /* component must fit device */
1819 if (rdev->data_offset != rdev->new_data_offset)
1820 return 0; /* too confusing */
1821 if (rdev->sb_start < rdev->data_offset) {
1822 /* minor versions 1 and 2; superblock before data */
1823 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1824 max_sectors -= rdev->data_offset;
1825 if (!num_sectors || num_sectors > max_sectors)
1826 num_sectors = max_sectors;
1827 } else if (rdev->mddev->bitmap_info.offset) {
1828 /* minor version 0 with bitmap we can't move */
1829 return 0;
1830 } else {
1831 /* minor version 0; superblock after data */
1832 sector_t sb_start;
1833 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1834 sb_start &= ~(sector_t)(4*2 - 1);
1835 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1836 if (!num_sectors || num_sectors > max_sectors)
1837 num_sectors = max_sectors;
1838 rdev->sb_start = sb_start;
1839 }
1840 sb = page_address(rdev->sb_page);
1841 sb->data_size = cpu_to_le64(num_sectors);
1842 sb->super_offset = rdev->sb_start;
1843 sb->sb_csum = calc_sb_1_csum(sb);
1844 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1845 rdev->sb_page);
1846 md_super_wait(rdev->mddev);
1847 return num_sectors;
1848
1849 }
1850
1851 static int
1852 super_1_allow_new_offset(struct md_rdev *rdev,
1853 unsigned long long new_offset)
1854 {
1855 /* All necessary checks on new >= old have been done */
1856 struct bitmap *bitmap;
1857 if (new_offset >= rdev->data_offset)
1858 return 1;
1859
1860 /* with 1.0 metadata, there is no metadata to tread on
1861 * so we can always move back */
1862 if (rdev->mddev->minor_version == 0)
1863 return 1;
1864
1865 /* otherwise we must be sure not to step on
1866 * any metadata, so stay:
1867 * 36K beyond start of superblock
1868 * beyond end of badblocks
1869 * beyond write-intent bitmap
1870 */
1871 if (rdev->sb_start + (32+4)*2 > new_offset)
1872 return 0;
1873 bitmap = rdev->mddev->bitmap;
1874 if (bitmap && !rdev->mddev->bitmap_info.file &&
1875 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1876 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
1877 return 0;
1878 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1879 return 0;
1880
1881 return 1;
1882 }
1883
1884 static struct super_type super_types[] = {
1885 [0] = {
1886 .name = "0.90.0",
1887 .owner = THIS_MODULE,
1888 .load_super = super_90_load,
1889 .validate_super = super_90_validate,
1890 .sync_super = super_90_sync,
1891 .rdev_size_change = super_90_rdev_size_change,
1892 .allow_new_offset = super_90_allow_new_offset,
1893 },
1894 [1] = {
1895 .name = "md-1",
1896 .owner = THIS_MODULE,
1897 .load_super = super_1_load,
1898 .validate_super = super_1_validate,
1899 .sync_super = super_1_sync,
1900 .rdev_size_change = super_1_rdev_size_change,
1901 .allow_new_offset = super_1_allow_new_offset,
1902 },
1903 };
1904
1905 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1906 {
1907 if (mddev->sync_super) {
1908 mddev->sync_super(mddev, rdev);
1909 return;
1910 }
1911
1912 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1913
1914 super_types[mddev->major_version].sync_super(mddev, rdev);
1915 }
1916
1917 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1918 {
1919 struct md_rdev *rdev, *rdev2;
1920
1921 rcu_read_lock();
1922 rdev_for_each_rcu(rdev, mddev1)
1923 rdev_for_each_rcu(rdev2, mddev2)
1924 if (rdev->bdev->bd_contains ==
1925 rdev2->bdev->bd_contains) {
1926 rcu_read_unlock();
1927 return 1;
1928 }
1929 rcu_read_unlock();
1930 return 0;
1931 }
1932
1933 static LIST_HEAD(pending_raid_disks);
1934
1935 /*
1936 * Try to register data integrity profile for an mddev
1937 *
1938 * This is called when an array is started and after a disk has been kicked
1939 * from the array. It only succeeds if all working and active component devices
1940 * are integrity capable with matching profiles.
1941 */
1942 int md_integrity_register(struct mddev *mddev)
1943 {
1944 struct md_rdev *rdev, *reference = NULL;
1945
1946 if (list_empty(&mddev->disks))
1947 return 0; /* nothing to do */
1948 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1949 return 0; /* shouldn't register, or already is */
1950 rdev_for_each(rdev, mddev) {
1951 /* skip spares and non-functional disks */
1952 if (test_bit(Faulty, &rdev->flags))
1953 continue;
1954 if (rdev->raid_disk < 0)
1955 continue;
1956 if (!reference) {
1957 /* Use the first rdev as the reference */
1958 reference = rdev;
1959 continue;
1960 }
1961 /* does this rdev's profile match the reference profile? */
1962 if (blk_integrity_compare(reference->bdev->bd_disk,
1963 rdev->bdev->bd_disk) < 0)
1964 return -EINVAL;
1965 }
1966 if (!reference || !bdev_get_integrity(reference->bdev))
1967 return 0;
1968 /*
1969 * All component devices are integrity capable and have matching
1970 * profiles, register the common profile for the md device.
1971 */
1972 if (blk_integrity_register(mddev->gendisk,
1973 bdev_get_integrity(reference->bdev)) != 0) {
1974 printk(KERN_ERR "md: failed to register integrity for %s\n",
1975 mdname(mddev));
1976 return -EINVAL;
1977 }
1978 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
1979 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
1980 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
1981 mdname(mddev));
1982 return -EINVAL;
1983 }
1984 return 0;
1985 }
1986 EXPORT_SYMBOL(md_integrity_register);
1987
1988 /* Disable data integrity if non-capable/non-matching disk is being added */
1989 void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
1990 {
1991 struct blk_integrity *bi_rdev;
1992 struct blk_integrity *bi_mddev;
1993
1994 if (!mddev->gendisk)
1995 return;
1996
1997 bi_rdev = bdev_get_integrity(rdev->bdev);
1998 bi_mddev = blk_get_integrity(mddev->gendisk);
1999
2000 if (!bi_mddev) /* nothing to do */
2001 return;
2002 if (rdev->raid_disk < 0) /* skip spares */
2003 return;
2004 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2005 rdev->bdev->bd_disk) >= 0)
2006 return;
2007 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2008 blk_integrity_unregister(mddev->gendisk);
2009 }
2010 EXPORT_SYMBOL(md_integrity_add_rdev);
2011
2012 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2013 {
2014 char b[BDEVNAME_SIZE];
2015 struct kobject *ko;
2016 char *s;
2017 int err;
2018
2019 /* prevent duplicates */
2020 if (find_rdev(mddev, rdev->bdev->bd_dev))
2021 return -EEXIST;
2022
2023 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2024 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2025 rdev->sectors < mddev->dev_sectors)) {
2026 if (mddev->pers) {
2027 /* Cannot change size, so fail
2028 * If mddev->level <= 0, then we don't care
2029 * about aligning sizes (e.g. linear)
2030 */
2031 if (mddev->level > 0)
2032 return -ENOSPC;
2033 } else
2034 mddev->dev_sectors = rdev->sectors;
2035 }
2036
2037 /* Verify rdev->desc_nr is unique.
2038 * If it is -1, assign a free number, else
2039 * check number is not in use
2040 */
2041 rcu_read_lock();
2042 if (rdev->desc_nr < 0) {
2043 int choice = 0;
2044 if (mddev->pers)
2045 choice = mddev->raid_disks;
2046 while (find_rdev_nr_rcu(mddev, choice))
2047 choice++;
2048 rdev->desc_nr = choice;
2049 } else {
2050 if (find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2051 rcu_read_unlock();
2052 return -EBUSY;
2053 }
2054 }
2055 rcu_read_unlock();
2056 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2057 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2058 mdname(mddev), mddev->max_disks);
2059 return -EBUSY;
2060 }
2061 bdevname(rdev->bdev,b);
2062 while ( (s=strchr(b, '/')) != NULL)
2063 *s = '!';
2064
2065 rdev->mddev = mddev;
2066 printk(KERN_INFO "md: bind<%s>\n", b);
2067
2068 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2069 goto fail;
2070
2071 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2072 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2073 /* failure here is OK */;
2074 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2075
2076 list_add_rcu(&rdev->same_set, &mddev->disks);
2077 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2078
2079 /* May as well allow recovery to be retried once */
2080 mddev->recovery_disabled++;
2081
2082 return 0;
2083
2084 fail:
2085 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2086 b, mdname(mddev));
2087 return err;
2088 }
2089
2090 static void md_delayed_delete(struct work_struct *ws)
2091 {
2092 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2093 kobject_del(&rdev->kobj);
2094 kobject_put(&rdev->kobj);
2095 }
2096
2097 static void unbind_rdev_from_array(struct md_rdev *rdev)
2098 {
2099 char b[BDEVNAME_SIZE];
2100
2101 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2102 list_del_rcu(&rdev->same_set);
2103 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2104 rdev->mddev = NULL;
2105 sysfs_remove_link(&rdev->kobj, "block");
2106 sysfs_put(rdev->sysfs_state);
2107 rdev->sysfs_state = NULL;
2108 rdev->badblocks.count = 0;
2109 /* We need to delay this, otherwise we can deadlock when
2110 * writing to 'remove' to "dev/state". We also need
2111 * to delay it due to rcu usage.
2112 */
2113 synchronize_rcu();
2114 INIT_WORK(&rdev->del_work, md_delayed_delete);
2115 kobject_get(&rdev->kobj);
2116 queue_work(md_misc_wq, &rdev->del_work);
2117 }
2118
2119 /*
2120 * prevent the device from being mounted, repartitioned or
2121 * otherwise reused by a RAID array (or any other kernel
2122 * subsystem), by bd_claiming the device.
2123 */
2124 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2125 {
2126 int err = 0;
2127 struct block_device *bdev;
2128 char b[BDEVNAME_SIZE];
2129
2130 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2131 shared ? (struct md_rdev *)lock_rdev : rdev);
2132 if (IS_ERR(bdev)) {
2133 printk(KERN_ERR "md: could not open %s.\n",
2134 __bdevname(dev, b));
2135 return PTR_ERR(bdev);
2136 }
2137 rdev->bdev = bdev;
2138 return err;
2139 }
2140
2141 static void unlock_rdev(struct md_rdev *rdev)
2142 {
2143 struct block_device *bdev = rdev->bdev;
2144 rdev->bdev = NULL;
2145 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2146 }
2147
2148 void md_autodetect_dev(dev_t dev);
2149
2150 static void export_rdev(struct md_rdev *rdev)
2151 {
2152 char b[BDEVNAME_SIZE];
2153
2154 printk(KERN_INFO "md: export_rdev(%s)\n",
2155 bdevname(rdev->bdev,b));
2156 md_rdev_clear(rdev);
2157 #ifndef MODULE
2158 if (test_bit(AutoDetected, &rdev->flags))
2159 md_autodetect_dev(rdev->bdev->bd_dev);
2160 #endif
2161 unlock_rdev(rdev);
2162 kobject_put(&rdev->kobj);
2163 }
2164
2165 static void kick_rdev_from_array(struct md_rdev *rdev)
2166 {
2167 unbind_rdev_from_array(rdev);
2168 export_rdev(rdev);
2169 }
2170
2171 static void export_array(struct mddev *mddev)
2172 {
2173 struct md_rdev *rdev;
2174
2175 while (!list_empty(&mddev->disks)) {
2176 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2177 same_set);
2178 kick_rdev_from_array(rdev);
2179 }
2180 mddev->raid_disks = 0;
2181 mddev->major_version = 0;
2182 }
2183
2184 static void sync_sbs(struct mddev *mddev, int nospares)
2185 {
2186 /* Update each superblock (in-memory image), but
2187 * if we are allowed to, skip spares which already
2188 * have the right event counter, or have one earlier
2189 * (which would mean they aren't being marked as dirty
2190 * with the rest of the array)
2191 */
2192 struct md_rdev *rdev;
2193 rdev_for_each(rdev, mddev) {
2194 if (rdev->sb_events == mddev->events ||
2195 (nospares &&
2196 rdev->raid_disk < 0 &&
2197 rdev->sb_events+1 == mddev->events)) {
2198 /* Don't update this superblock */
2199 rdev->sb_loaded = 2;
2200 } else {
2201 sync_super(mddev, rdev);
2202 rdev->sb_loaded = 1;
2203 }
2204 }
2205 }
2206
2207 static void md_update_sb(struct mddev *mddev, int force_change)
2208 {
2209 struct md_rdev *rdev;
2210 int sync_req;
2211 int nospares = 0;
2212 int any_badblocks_changed = 0;
2213
2214 if (mddev->ro) {
2215 if (force_change)
2216 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2217 return;
2218 }
2219 repeat:
2220 /* First make sure individual recovery_offsets are correct */
2221 rdev_for_each(rdev, mddev) {
2222 if (rdev->raid_disk >= 0 &&
2223 mddev->delta_disks >= 0 &&
2224 !test_bit(In_sync, &rdev->flags) &&
2225 mddev->curr_resync_completed > rdev->recovery_offset)
2226 rdev->recovery_offset = mddev->curr_resync_completed;
2227
2228 }
2229 if (!mddev->persistent) {
2230 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2231 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2232 if (!mddev->external) {
2233 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2234 rdev_for_each(rdev, mddev) {
2235 if (rdev->badblocks.changed) {
2236 rdev->badblocks.changed = 0;
2237 md_ack_all_badblocks(&rdev->badblocks);
2238 md_error(mddev, rdev);
2239 }
2240 clear_bit(Blocked, &rdev->flags);
2241 clear_bit(BlockedBadBlocks, &rdev->flags);
2242 wake_up(&rdev->blocked_wait);
2243 }
2244 }
2245 wake_up(&mddev->sb_wait);
2246 return;
2247 }
2248
2249 spin_lock(&mddev->lock);
2250
2251 mddev->utime = get_seconds();
2252
2253 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2254 force_change = 1;
2255 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2256 /* just a clean<-> dirty transition, possibly leave spares alone,
2257 * though if events isn't the right even/odd, we will have to do
2258 * spares after all
2259 */
2260 nospares = 1;
2261 if (force_change)
2262 nospares = 0;
2263 if (mddev->degraded)
2264 /* If the array is degraded, then skipping spares is both
2265 * dangerous and fairly pointless.
2266 * Dangerous because a device that was removed from the array
2267 * might have a event_count that still looks up-to-date,
2268 * so it can be re-added without a resync.
2269 * Pointless because if there are any spares to skip,
2270 * then a recovery will happen and soon that array won't
2271 * be degraded any more and the spare can go back to sleep then.
2272 */
2273 nospares = 0;
2274
2275 sync_req = mddev->in_sync;
2276
2277 /* If this is just a dirty<->clean transition, and the array is clean
2278 * and 'events' is odd, we can roll back to the previous clean state */
2279 if (nospares
2280 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2281 && mddev->can_decrease_events
2282 && mddev->events != 1) {
2283 mddev->events--;
2284 mddev->can_decrease_events = 0;
2285 } else {
2286 /* otherwise we have to go forward and ... */
2287 mddev->events ++;
2288 mddev->can_decrease_events = nospares;
2289 }
2290
2291 /*
2292 * This 64-bit counter should never wrap.
2293 * Either we are in around ~1 trillion A.C., assuming
2294 * 1 reboot per second, or we have a bug...
2295 */
2296 WARN_ON(mddev->events == 0);
2297
2298 rdev_for_each(rdev, mddev) {
2299 if (rdev->badblocks.changed)
2300 any_badblocks_changed++;
2301 if (test_bit(Faulty, &rdev->flags))
2302 set_bit(FaultRecorded, &rdev->flags);
2303 }
2304
2305 sync_sbs(mddev, nospares);
2306 spin_unlock(&mddev->lock);
2307
2308 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2309 mdname(mddev), mddev->in_sync);
2310
2311 bitmap_update_sb(mddev->bitmap);
2312 rdev_for_each(rdev, mddev) {
2313 char b[BDEVNAME_SIZE];
2314
2315 if (rdev->sb_loaded != 1)
2316 continue; /* no noise on spare devices */
2317
2318 if (!test_bit(Faulty, &rdev->flags)) {
2319 md_super_write(mddev,rdev,
2320 rdev->sb_start, rdev->sb_size,
2321 rdev->sb_page);
2322 pr_debug("md: (write) %s's sb offset: %llu\n",
2323 bdevname(rdev->bdev, b),
2324 (unsigned long long)rdev->sb_start);
2325 rdev->sb_events = mddev->events;
2326 if (rdev->badblocks.size) {
2327 md_super_write(mddev, rdev,
2328 rdev->badblocks.sector,
2329 rdev->badblocks.size << 9,
2330 rdev->bb_page);
2331 rdev->badblocks.size = 0;
2332 }
2333
2334 } else
2335 pr_debug("md: %s (skipping faulty)\n",
2336 bdevname(rdev->bdev, b));
2337
2338 if (mddev->level == LEVEL_MULTIPATH)
2339 /* only need to write one superblock... */
2340 break;
2341 }
2342 md_super_wait(mddev);
2343 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2344
2345 spin_lock(&mddev->lock);
2346 if (mddev->in_sync != sync_req ||
2347 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2348 /* have to write it out again */
2349 spin_unlock(&mddev->lock);
2350 goto repeat;
2351 }
2352 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2353 spin_unlock(&mddev->lock);
2354 wake_up(&mddev->sb_wait);
2355 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2356 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2357
2358 rdev_for_each(rdev, mddev) {
2359 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2360 clear_bit(Blocked, &rdev->flags);
2361
2362 if (any_badblocks_changed)
2363 md_ack_all_badblocks(&rdev->badblocks);
2364 clear_bit(BlockedBadBlocks, &rdev->flags);
2365 wake_up(&rdev->blocked_wait);
2366 }
2367 }
2368
2369 /* words written to sysfs files may, or may not, be \n terminated.
2370 * We want to accept with case. For this we use cmd_match.
2371 */
2372 static int cmd_match(const char *cmd, const char *str)
2373 {
2374 /* See if cmd, written into a sysfs file, matches
2375 * str. They must either be the same, or cmd can
2376 * have a trailing newline
2377 */
2378 while (*cmd && *str && *cmd == *str) {
2379 cmd++;
2380 str++;
2381 }
2382 if (*cmd == '\n')
2383 cmd++;
2384 if (*str || *cmd)
2385 return 0;
2386 return 1;
2387 }
2388
2389 struct rdev_sysfs_entry {
2390 struct attribute attr;
2391 ssize_t (*show)(struct md_rdev *, char *);
2392 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2393 };
2394
2395 static ssize_t
2396 state_show(struct md_rdev *rdev, char *page)
2397 {
2398 char *sep = "";
2399 size_t len = 0;
2400 unsigned long flags = ACCESS_ONCE(rdev->flags);
2401
2402 if (test_bit(Faulty, &flags) ||
2403 rdev->badblocks.unacked_exist) {
2404 len+= sprintf(page+len, "%sfaulty",sep);
2405 sep = ",";
2406 }
2407 if (test_bit(In_sync, &flags)) {
2408 len += sprintf(page+len, "%sin_sync",sep);
2409 sep = ",";
2410 }
2411 if (test_bit(WriteMostly, &flags)) {
2412 len += sprintf(page+len, "%swrite_mostly",sep);
2413 sep = ",";
2414 }
2415 if (test_bit(Blocked, &flags) ||
2416 (rdev->badblocks.unacked_exist
2417 && !test_bit(Faulty, &flags))) {
2418 len += sprintf(page+len, "%sblocked", sep);
2419 sep = ",";
2420 }
2421 if (!test_bit(Faulty, &flags) &&
2422 !test_bit(In_sync, &flags)) {
2423 len += sprintf(page+len, "%sspare", sep);
2424 sep = ",";
2425 }
2426 if (test_bit(WriteErrorSeen, &flags)) {
2427 len += sprintf(page+len, "%swrite_error", sep);
2428 sep = ",";
2429 }
2430 if (test_bit(WantReplacement, &flags)) {
2431 len += sprintf(page+len, "%swant_replacement", sep);
2432 sep = ",";
2433 }
2434 if (test_bit(Replacement, &flags)) {
2435 len += sprintf(page+len, "%sreplacement", sep);
2436 sep = ",";
2437 }
2438
2439 return len+sprintf(page+len, "\n");
2440 }
2441
2442 static ssize_t
2443 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2444 {
2445 /* can write
2446 * faulty - simulates an error
2447 * remove - disconnects the device
2448 * writemostly - sets write_mostly
2449 * -writemostly - clears write_mostly
2450 * blocked - sets the Blocked flags
2451 * -blocked - clears the Blocked and possibly simulates an error
2452 * insync - sets Insync providing device isn't active
2453 * -insync - clear Insync for a device with a slot assigned,
2454 * so that it gets rebuilt based on bitmap
2455 * write_error - sets WriteErrorSeen
2456 * -write_error - clears WriteErrorSeen
2457 */
2458 int err = -EINVAL;
2459 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2460 md_error(rdev->mddev, rdev);
2461 if (test_bit(Faulty, &rdev->flags))
2462 err = 0;
2463 else
2464 err = -EBUSY;
2465 } else if (cmd_match(buf, "remove")) {
2466 if (rdev->raid_disk >= 0)
2467 err = -EBUSY;
2468 else {
2469 struct mddev *mddev = rdev->mddev;
2470 kick_rdev_from_array(rdev);
2471 if (mddev->pers)
2472 md_update_sb(mddev, 1);
2473 md_new_event(mddev);
2474 err = 0;
2475 }
2476 } else if (cmd_match(buf, "writemostly")) {
2477 set_bit(WriteMostly, &rdev->flags);
2478 err = 0;
2479 } else if (cmd_match(buf, "-writemostly")) {
2480 clear_bit(WriteMostly, &rdev->flags);
2481 err = 0;
2482 } else if (cmd_match(buf, "blocked")) {
2483 set_bit(Blocked, &rdev->flags);
2484 err = 0;
2485 } else if (cmd_match(buf, "-blocked")) {
2486 if (!test_bit(Faulty, &rdev->flags) &&
2487 rdev->badblocks.unacked_exist) {
2488 /* metadata handler doesn't understand badblocks,
2489 * so we need to fail the device
2490 */
2491 md_error(rdev->mddev, rdev);
2492 }
2493 clear_bit(Blocked, &rdev->flags);
2494 clear_bit(BlockedBadBlocks, &rdev->flags);
2495 wake_up(&rdev->blocked_wait);
2496 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2497 md_wakeup_thread(rdev->mddev->thread);
2498
2499 err = 0;
2500 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2501 set_bit(In_sync, &rdev->flags);
2502 err = 0;
2503 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0) {
2504 if (rdev->mddev->pers == NULL) {
2505 clear_bit(In_sync, &rdev->flags);
2506 rdev->saved_raid_disk = rdev->raid_disk;
2507 rdev->raid_disk = -1;
2508 err = 0;
2509 }
2510 } else if (cmd_match(buf, "write_error")) {
2511 set_bit(WriteErrorSeen, &rdev->flags);
2512 err = 0;
2513 } else if (cmd_match(buf, "-write_error")) {
2514 clear_bit(WriteErrorSeen, &rdev->flags);
2515 err = 0;
2516 } else if (cmd_match(buf, "want_replacement")) {
2517 /* Any non-spare device that is not a replacement can
2518 * become want_replacement at any time, but we then need to
2519 * check if recovery is needed.
2520 */
2521 if (rdev->raid_disk >= 0 &&
2522 !test_bit(Replacement, &rdev->flags))
2523 set_bit(WantReplacement, &rdev->flags);
2524 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2525 md_wakeup_thread(rdev->mddev->thread);
2526 err = 0;
2527 } else if (cmd_match(buf, "-want_replacement")) {
2528 /* Clearing 'want_replacement' is always allowed.
2529 * Once replacements starts it is too late though.
2530 */
2531 err = 0;
2532 clear_bit(WantReplacement, &rdev->flags);
2533 } else if (cmd_match(buf, "replacement")) {
2534 /* Can only set a device as a replacement when array has not
2535 * yet been started. Once running, replacement is automatic
2536 * from spares, or by assigning 'slot'.
2537 */
2538 if (rdev->mddev->pers)
2539 err = -EBUSY;
2540 else {
2541 set_bit(Replacement, &rdev->flags);
2542 err = 0;
2543 }
2544 } else if (cmd_match(buf, "-replacement")) {
2545 /* Similarly, can only clear Replacement before start */
2546 if (rdev->mddev->pers)
2547 err = -EBUSY;
2548 else {
2549 clear_bit(Replacement, &rdev->flags);
2550 err = 0;
2551 }
2552 }
2553 if (!err)
2554 sysfs_notify_dirent_safe(rdev->sysfs_state);
2555 return err ? err : len;
2556 }
2557 static struct rdev_sysfs_entry rdev_state =
2558 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2559
2560 static ssize_t
2561 errors_show(struct md_rdev *rdev, char *page)
2562 {
2563 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2564 }
2565
2566 static ssize_t
2567 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2568 {
2569 char *e;
2570 unsigned long n = simple_strtoul(buf, &e, 10);
2571 if (*buf && (*e == 0 || *e == '\n')) {
2572 atomic_set(&rdev->corrected_errors, n);
2573 return len;
2574 }
2575 return -EINVAL;
2576 }
2577 static struct rdev_sysfs_entry rdev_errors =
2578 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2579
2580 static ssize_t
2581 slot_show(struct md_rdev *rdev, char *page)
2582 {
2583 if (rdev->raid_disk < 0)
2584 return sprintf(page, "none\n");
2585 else
2586 return sprintf(page, "%d\n", rdev->raid_disk);
2587 }
2588
2589 static ssize_t
2590 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2591 {
2592 char *e;
2593 int err;
2594 int slot = simple_strtoul(buf, &e, 10);
2595 if (strncmp(buf, "none", 4)==0)
2596 slot = -1;
2597 else if (e==buf || (*e && *e!= '\n'))
2598 return -EINVAL;
2599 if (rdev->mddev->pers && slot == -1) {
2600 /* Setting 'slot' on an active array requires also
2601 * updating the 'rd%d' link, and communicating
2602 * with the personality with ->hot_*_disk.
2603 * For now we only support removing
2604 * failed/spare devices. This normally happens automatically,
2605 * but not when the metadata is externally managed.
2606 */
2607 if (rdev->raid_disk == -1)
2608 return -EEXIST;
2609 /* personality does all needed checks */
2610 if (rdev->mddev->pers->hot_remove_disk == NULL)
2611 return -EINVAL;
2612 clear_bit(Blocked, &rdev->flags);
2613 remove_and_add_spares(rdev->mddev, rdev);
2614 if (rdev->raid_disk >= 0)
2615 return -EBUSY;
2616 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2617 md_wakeup_thread(rdev->mddev->thread);
2618 } else if (rdev->mddev->pers) {
2619 /* Activating a spare .. or possibly reactivating
2620 * if we ever get bitmaps working here.
2621 */
2622
2623 if (rdev->raid_disk != -1)
2624 return -EBUSY;
2625
2626 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2627 return -EBUSY;
2628
2629 if (rdev->mddev->pers->hot_add_disk == NULL)
2630 return -EINVAL;
2631
2632 if (slot >= rdev->mddev->raid_disks &&
2633 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2634 return -ENOSPC;
2635
2636 rdev->raid_disk = slot;
2637 if (test_bit(In_sync, &rdev->flags))
2638 rdev->saved_raid_disk = slot;
2639 else
2640 rdev->saved_raid_disk = -1;
2641 clear_bit(In_sync, &rdev->flags);
2642 clear_bit(Bitmap_sync, &rdev->flags);
2643 err = rdev->mddev->pers->
2644 hot_add_disk(rdev->mddev, rdev);
2645 if (err) {
2646 rdev->raid_disk = -1;
2647 return err;
2648 } else
2649 sysfs_notify_dirent_safe(rdev->sysfs_state);
2650 if (sysfs_link_rdev(rdev->mddev, rdev))
2651 /* failure here is OK */;
2652 /* don't wakeup anyone, leave that to userspace. */
2653 } else {
2654 if (slot >= rdev->mddev->raid_disks &&
2655 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2656 return -ENOSPC;
2657 rdev->raid_disk = slot;
2658 /* assume it is working */
2659 clear_bit(Faulty, &rdev->flags);
2660 clear_bit(WriteMostly, &rdev->flags);
2661 set_bit(In_sync, &rdev->flags);
2662 sysfs_notify_dirent_safe(rdev->sysfs_state);
2663 }
2664 return len;
2665 }
2666
2667 static struct rdev_sysfs_entry rdev_slot =
2668 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2669
2670 static ssize_t
2671 offset_show(struct md_rdev *rdev, char *page)
2672 {
2673 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2674 }
2675
2676 static ssize_t
2677 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2678 {
2679 unsigned long long offset;
2680 if (kstrtoull(buf, 10, &offset) < 0)
2681 return -EINVAL;
2682 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2683 return -EBUSY;
2684 if (rdev->sectors && rdev->mddev->external)
2685 /* Must set offset before size, so overlap checks
2686 * can be sane */
2687 return -EBUSY;
2688 rdev->data_offset = offset;
2689 rdev->new_data_offset = offset;
2690 return len;
2691 }
2692
2693 static struct rdev_sysfs_entry rdev_offset =
2694 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2695
2696 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2697 {
2698 return sprintf(page, "%llu\n",
2699 (unsigned long long)rdev->new_data_offset);
2700 }
2701
2702 static ssize_t new_offset_store(struct md_rdev *rdev,
2703 const char *buf, size_t len)
2704 {
2705 unsigned long long new_offset;
2706 struct mddev *mddev = rdev->mddev;
2707
2708 if (kstrtoull(buf, 10, &new_offset) < 0)
2709 return -EINVAL;
2710
2711 if (mddev->sync_thread ||
2712 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
2713 return -EBUSY;
2714 if (new_offset == rdev->data_offset)
2715 /* reset is always permitted */
2716 ;
2717 else if (new_offset > rdev->data_offset) {
2718 /* must not push array size beyond rdev_sectors */
2719 if (new_offset - rdev->data_offset
2720 + mddev->dev_sectors > rdev->sectors)
2721 return -E2BIG;
2722 }
2723 /* Metadata worries about other space details. */
2724
2725 /* decreasing the offset is inconsistent with a backwards
2726 * reshape.
2727 */
2728 if (new_offset < rdev->data_offset &&
2729 mddev->reshape_backwards)
2730 return -EINVAL;
2731 /* Increasing offset is inconsistent with forwards
2732 * reshape. reshape_direction should be set to
2733 * 'backwards' first.
2734 */
2735 if (new_offset > rdev->data_offset &&
2736 !mddev->reshape_backwards)
2737 return -EINVAL;
2738
2739 if (mddev->pers && mddev->persistent &&
2740 !super_types[mddev->major_version]
2741 .allow_new_offset(rdev, new_offset))
2742 return -E2BIG;
2743 rdev->new_data_offset = new_offset;
2744 if (new_offset > rdev->data_offset)
2745 mddev->reshape_backwards = 1;
2746 else if (new_offset < rdev->data_offset)
2747 mddev->reshape_backwards = 0;
2748
2749 return len;
2750 }
2751 static struct rdev_sysfs_entry rdev_new_offset =
2752 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2753
2754 static ssize_t
2755 rdev_size_show(struct md_rdev *rdev, char *page)
2756 {
2757 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2758 }
2759
2760 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2761 {
2762 /* check if two start/length pairs overlap */
2763 if (s1+l1 <= s2)
2764 return 0;
2765 if (s2+l2 <= s1)
2766 return 0;
2767 return 1;
2768 }
2769
2770 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2771 {
2772 unsigned long long blocks;
2773 sector_t new;
2774
2775 if (kstrtoull(buf, 10, &blocks) < 0)
2776 return -EINVAL;
2777
2778 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2779 return -EINVAL; /* sector conversion overflow */
2780
2781 new = blocks * 2;
2782 if (new != blocks * 2)
2783 return -EINVAL; /* unsigned long long to sector_t overflow */
2784
2785 *sectors = new;
2786 return 0;
2787 }
2788
2789 static ssize_t
2790 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
2791 {
2792 struct mddev *my_mddev = rdev->mddev;
2793 sector_t oldsectors = rdev->sectors;
2794 sector_t sectors;
2795
2796 if (strict_blocks_to_sectors(buf, &sectors) < 0)
2797 return -EINVAL;
2798 if (rdev->data_offset != rdev->new_data_offset)
2799 return -EINVAL; /* too confusing */
2800 if (my_mddev->pers && rdev->raid_disk >= 0) {
2801 if (my_mddev->persistent) {
2802 sectors = super_types[my_mddev->major_version].
2803 rdev_size_change(rdev, sectors);
2804 if (!sectors)
2805 return -EBUSY;
2806 } else if (!sectors)
2807 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2808 rdev->data_offset;
2809 if (!my_mddev->pers->resize)
2810 /* Cannot change size for RAID0 or Linear etc */
2811 return -EINVAL;
2812 }
2813 if (sectors < my_mddev->dev_sectors)
2814 return -EINVAL; /* component must fit device */
2815
2816 rdev->sectors = sectors;
2817 if (sectors > oldsectors && my_mddev->external) {
2818 /* Need to check that all other rdevs with the same
2819 * ->bdev do not overlap. 'rcu' is sufficient to walk
2820 * the rdev lists safely.
2821 * This check does not provide a hard guarantee, it
2822 * just helps avoid dangerous mistakes.
2823 */
2824 struct mddev *mddev;
2825 int overlap = 0;
2826 struct list_head *tmp;
2827
2828 rcu_read_lock();
2829 for_each_mddev(mddev, tmp) {
2830 struct md_rdev *rdev2;
2831
2832 rdev_for_each(rdev2, mddev)
2833 if (rdev->bdev == rdev2->bdev &&
2834 rdev != rdev2 &&
2835 overlaps(rdev->data_offset, rdev->sectors,
2836 rdev2->data_offset,
2837 rdev2->sectors)) {
2838 overlap = 1;
2839 break;
2840 }
2841 if (overlap) {
2842 mddev_put(mddev);
2843 break;
2844 }
2845 }
2846 rcu_read_unlock();
2847 if (overlap) {
2848 /* Someone else could have slipped in a size
2849 * change here, but doing so is just silly.
2850 * We put oldsectors back because we *know* it is
2851 * safe, and trust userspace not to race with
2852 * itself
2853 */
2854 rdev->sectors = oldsectors;
2855 return -EBUSY;
2856 }
2857 }
2858 return len;
2859 }
2860
2861 static struct rdev_sysfs_entry rdev_size =
2862 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2863
2864 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
2865 {
2866 unsigned long long recovery_start = rdev->recovery_offset;
2867
2868 if (test_bit(In_sync, &rdev->flags) ||
2869 recovery_start == MaxSector)
2870 return sprintf(page, "none\n");
2871
2872 return sprintf(page, "%llu\n", recovery_start);
2873 }
2874
2875 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
2876 {
2877 unsigned long long recovery_start;
2878
2879 if (cmd_match(buf, "none"))
2880 recovery_start = MaxSector;
2881 else if (kstrtoull(buf, 10, &recovery_start))
2882 return -EINVAL;
2883
2884 if (rdev->mddev->pers &&
2885 rdev->raid_disk >= 0)
2886 return -EBUSY;
2887
2888 rdev->recovery_offset = recovery_start;
2889 if (recovery_start == MaxSector)
2890 set_bit(In_sync, &rdev->flags);
2891 else
2892 clear_bit(In_sync, &rdev->flags);
2893 return len;
2894 }
2895
2896 static struct rdev_sysfs_entry rdev_recovery_start =
2897 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2898
2899 static ssize_t
2900 badblocks_show(struct badblocks *bb, char *page, int unack);
2901 static ssize_t
2902 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
2903
2904 static ssize_t bb_show(struct md_rdev *rdev, char *page)
2905 {
2906 return badblocks_show(&rdev->badblocks, page, 0);
2907 }
2908 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
2909 {
2910 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
2911 /* Maybe that ack was all we needed */
2912 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
2913 wake_up(&rdev->blocked_wait);
2914 return rv;
2915 }
2916 static struct rdev_sysfs_entry rdev_bad_blocks =
2917 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
2918
2919 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
2920 {
2921 return badblocks_show(&rdev->badblocks, page, 1);
2922 }
2923 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
2924 {
2925 return badblocks_store(&rdev->badblocks, page, len, 1);
2926 }
2927 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
2928 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
2929
2930 static struct attribute *rdev_default_attrs[] = {
2931 &rdev_state.attr,
2932 &rdev_errors.attr,
2933 &rdev_slot.attr,
2934 &rdev_offset.attr,
2935 &rdev_new_offset.attr,
2936 &rdev_size.attr,
2937 &rdev_recovery_start.attr,
2938 &rdev_bad_blocks.attr,
2939 &rdev_unack_bad_blocks.attr,
2940 NULL,
2941 };
2942 static ssize_t
2943 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2944 {
2945 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2946 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
2947
2948 if (!entry->show)
2949 return -EIO;
2950 if (!rdev->mddev)
2951 return -EBUSY;
2952 return entry->show(rdev, page);
2953 }
2954
2955 static ssize_t
2956 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2957 const char *page, size_t length)
2958 {
2959 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2960 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
2961 ssize_t rv;
2962 struct mddev *mddev = rdev->mddev;
2963
2964 if (!entry->store)
2965 return -EIO;
2966 if (!capable(CAP_SYS_ADMIN))
2967 return -EACCES;
2968 rv = mddev ? mddev_lock(mddev): -EBUSY;
2969 if (!rv) {
2970 if (rdev->mddev == NULL)
2971 rv = -EBUSY;
2972 else
2973 rv = entry->store(rdev, page, length);
2974 mddev_unlock(mddev);
2975 }
2976 return rv;
2977 }
2978
2979 static void rdev_free(struct kobject *ko)
2980 {
2981 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
2982 kfree(rdev);
2983 }
2984 static const struct sysfs_ops rdev_sysfs_ops = {
2985 .show = rdev_attr_show,
2986 .store = rdev_attr_store,
2987 };
2988 static struct kobj_type rdev_ktype = {
2989 .release = rdev_free,
2990 .sysfs_ops = &rdev_sysfs_ops,
2991 .default_attrs = rdev_default_attrs,
2992 };
2993
2994 int md_rdev_init(struct md_rdev *rdev)
2995 {
2996 rdev->desc_nr = -1;
2997 rdev->saved_raid_disk = -1;
2998 rdev->raid_disk = -1;
2999 rdev->flags = 0;
3000 rdev->data_offset = 0;
3001 rdev->new_data_offset = 0;
3002 rdev->sb_events = 0;
3003 rdev->last_read_error.tv_sec = 0;
3004 rdev->last_read_error.tv_nsec = 0;
3005 rdev->sb_loaded = 0;
3006 rdev->bb_page = NULL;
3007 atomic_set(&rdev->nr_pending, 0);
3008 atomic_set(&rdev->read_errors, 0);
3009 atomic_set(&rdev->corrected_errors, 0);
3010
3011 INIT_LIST_HEAD(&rdev->same_set);
3012 init_waitqueue_head(&rdev->blocked_wait);
3013
3014 /* Add space to store bad block list.
3015 * This reserves the space even on arrays where it cannot
3016 * be used - I wonder if that matters
3017 */
3018 rdev->badblocks.count = 0;
3019 rdev->badblocks.shift = -1; /* disabled until explicitly enabled */
3020 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3021 seqlock_init(&rdev->badblocks.lock);
3022 if (rdev->badblocks.page == NULL)
3023 return -ENOMEM;
3024
3025 return 0;
3026 }
3027 EXPORT_SYMBOL_GPL(md_rdev_init);
3028 /*
3029 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3030 *
3031 * mark the device faulty if:
3032 *
3033 * - the device is nonexistent (zero size)
3034 * - the device has no valid superblock
3035 *
3036 * a faulty rdev _never_ has rdev->sb set.
3037 */
3038 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3039 {
3040 char b[BDEVNAME_SIZE];
3041 int err;
3042 struct md_rdev *rdev;
3043 sector_t size;
3044
3045 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3046 if (!rdev) {
3047 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3048 return ERR_PTR(-ENOMEM);
3049 }
3050
3051 err = md_rdev_init(rdev);
3052 if (err)
3053 goto abort_free;
3054 err = alloc_disk_sb(rdev);
3055 if (err)
3056 goto abort_free;
3057
3058 err = lock_rdev(rdev, newdev, super_format == -2);
3059 if (err)
3060 goto abort_free;
3061
3062 kobject_init(&rdev->kobj, &rdev_ktype);
3063
3064 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3065 if (!size) {
3066 printk(KERN_WARNING
3067 "md: %s has zero or unknown size, marking faulty!\n",
3068 bdevname(rdev->bdev,b));
3069 err = -EINVAL;
3070 goto abort_free;
3071 }
3072
3073 if (super_format >= 0) {
3074 err = super_types[super_format].
3075 load_super(rdev, NULL, super_minor);
3076 if (err == -EINVAL) {
3077 printk(KERN_WARNING
3078 "md: %s does not have a valid v%d.%d "
3079 "superblock, not importing!\n",
3080 bdevname(rdev->bdev,b),
3081 super_format, super_minor);
3082 goto abort_free;
3083 }
3084 if (err < 0) {
3085 printk(KERN_WARNING
3086 "md: could not read %s's sb, not importing!\n",
3087 bdevname(rdev->bdev,b));
3088 goto abort_free;
3089 }
3090 }
3091
3092 return rdev;
3093
3094 abort_free:
3095 if (rdev->bdev)
3096 unlock_rdev(rdev);
3097 md_rdev_clear(rdev);
3098 kfree(rdev);
3099 return ERR_PTR(err);
3100 }
3101
3102 /*
3103 * Check a full RAID array for plausibility
3104 */
3105
3106 static void analyze_sbs(struct mddev *mddev)
3107 {
3108 int i;
3109 struct md_rdev *rdev, *freshest, *tmp;
3110 char b[BDEVNAME_SIZE];
3111
3112 freshest = NULL;
3113 rdev_for_each_safe(rdev, tmp, mddev)
3114 switch (super_types[mddev->major_version].
3115 load_super(rdev, freshest, mddev->minor_version)) {
3116 case 1:
3117 freshest = rdev;
3118 break;
3119 case 0:
3120 break;
3121 default:
3122 printk( KERN_ERR \
3123 "md: fatal superblock inconsistency in %s"
3124 " -- removing from array\n",
3125 bdevname(rdev->bdev,b));
3126 kick_rdev_from_array(rdev);
3127 }
3128
3129 super_types[mddev->major_version].
3130 validate_super(mddev, freshest);
3131
3132 i = 0;
3133 rdev_for_each_safe(rdev, tmp, mddev) {
3134 if (mddev->max_disks &&
3135 (rdev->desc_nr >= mddev->max_disks ||
3136 i > mddev->max_disks)) {
3137 printk(KERN_WARNING
3138 "md: %s: %s: only %d devices permitted\n",
3139 mdname(mddev), bdevname(rdev->bdev, b),
3140 mddev->max_disks);
3141 kick_rdev_from_array(rdev);
3142 continue;
3143 }
3144 if (rdev != freshest)
3145 if (super_types[mddev->major_version].
3146 validate_super(mddev, rdev)) {
3147 printk(KERN_WARNING "md: kicking non-fresh %s"
3148 " from array!\n",
3149 bdevname(rdev->bdev,b));
3150 kick_rdev_from_array(rdev);
3151 continue;
3152 }
3153 if (mddev->level == LEVEL_MULTIPATH) {
3154 rdev->desc_nr = i++;
3155 rdev->raid_disk = rdev->desc_nr;
3156 set_bit(In_sync, &rdev->flags);
3157 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
3158 rdev->raid_disk = -1;
3159 clear_bit(In_sync, &rdev->flags);
3160 }
3161 }
3162 }
3163
3164 /* Read a fixed-point number.
3165 * Numbers in sysfs attributes should be in "standard" units where
3166 * possible, so time should be in seconds.
3167 * However we internally use a a much smaller unit such as
3168 * milliseconds or jiffies.
3169 * This function takes a decimal number with a possible fractional
3170 * component, and produces an integer which is the result of
3171 * multiplying that number by 10^'scale'.
3172 * all without any floating-point arithmetic.
3173 */
3174 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3175 {
3176 unsigned long result = 0;
3177 long decimals = -1;
3178 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3179 if (*cp == '.')
3180 decimals = 0;
3181 else if (decimals < scale) {
3182 unsigned int value;
3183 value = *cp - '0';
3184 result = result * 10 + value;
3185 if (decimals >= 0)
3186 decimals++;
3187 }
3188 cp++;
3189 }
3190 if (*cp == '\n')
3191 cp++;
3192 if (*cp)
3193 return -EINVAL;
3194 if (decimals < 0)
3195 decimals = 0;
3196 while (decimals < scale) {
3197 result *= 10;
3198 decimals ++;
3199 }
3200 *res = result;
3201 return 0;
3202 }
3203
3204 static void md_safemode_timeout(unsigned long data);
3205
3206 static ssize_t
3207 safe_delay_show(struct mddev *mddev, char *page)
3208 {
3209 int msec = (mddev->safemode_delay*1000)/HZ;
3210 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3211 }
3212 static ssize_t
3213 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3214 {
3215 unsigned long msec;
3216
3217 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3218 return -EINVAL;
3219 if (msec == 0)
3220 mddev->safemode_delay = 0;
3221 else {
3222 unsigned long old_delay = mddev->safemode_delay;
3223 unsigned long new_delay = (msec*HZ)/1000;
3224
3225 if (new_delay == 0)
3226 new_delay = 1;
3227 mddev->safemode_delay = new_delay;
3228 if (new_delay < old_delay || old_delay == 0)
3229 mod_timer(&mddev->safemode_timer, jiffies+1);
3230 }
3231 return len;
3232 }
3233 static struct md_sysfs_entry md_safe_delay =
3234 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3235
3236 static ssize_t
3237 level_show(struct mddev *mddev, char *page)
3238 {
3239 struct md_personality *p;
3240 int ret;
3241 spin_lock(&mddev->lock);
3242 p = mddev->pers;
3243 if (p)
3244 ret = sprintf(page, "%s\n", p->name);
3245 else if (mddev->clevel[0])
3246 ret = sprintf(page, "%s\n", mddev->clevel);
3247 else if (mddev->level != LEVEL_NONE)
3248 ret = sprintf(page, "%d\n", mddev->level);
3249 else
3250 ret = 0;
3251 spin_unlock(&mddev->lock);
3252 return ret;
3253 }
3254
3255 static ssize_t
3256 level_store(struct mddev *mddev, const char *buf, size_t len)
3257 {
3258 char clevel[16];
3259 ssize_t rv;
3260 size_t slen = len;
3261 struct md_personality *pers, *oldpers;
3262 long level;
3263 void *priv, *oldpriv;
3264 struct md_rdev *rdev;
3265
3266 if (slen == 0 || slen >= sizeof(clevel))
3267 return -EINVAL;
3268
3269 rv = mddev_lock(mddev);
3270 if (rv)
3271 return rv;
3272
3273 if (mddev->pers == NULL) {
3274 strncpy(mddev->clevel, buf, slen);
3275 if (mddev->clevel[slen-1] == '\n')
3276 slen--;
3277 mddev->clevel[slen] = 0;
3278 mddev->level = LEVEL_NONE;
3279 rv = len;
3280 goto out_unlock;
3281 }
3282 rv = -EROFS;
3283 if (mddev->ro)
3284 goto out_unlock;
3285
3286 /* request to change the personality. Need to ensure:
3287 * - array is not engaged in resync/recovery/reshape
3288 * - old personality can be suspended
3289 * - new personality will access other array.
3290 */
3291
3292 rv = -EBUSY;
3293 if (mddev->sync_thread ||
3294 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3295 mddev->reshape_position != MaxSector ||
3296 mddev->sysfs_active)
3297 goto out_unlock;
3298
3299 rv = -EINVAL;
3300 if (!mddev->pers->quiesce) {
3301 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3302 mdname(mddev), mddev->pers->name);
3303 goto out_unlock;
3304 }
3305
3306 /* Now find the new personality */
3307 strncpy(clevel, buf, slen);
3308 if (clevel[slen-1] == '\n')
3309 slen--;
3310 clevel[slen] = 0;
3311 if (kstrtol(clevel, 10, &level))
3312 level = LEVEL_NONE;
3313
3314 if (request_module("md-%s", clevel) != 0)
3315 request_module("md-level-%s", clevel);
3316 spin_lock(&pers_lock);
3317 pers = find_pers(level, clevel);
3318 if (!pers || !try_module_get(pers->owner)) {
3319 spin_unlock(&pers_lock);
3320 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3321 rv = -EINVAL;
3322 goto out_unlock;
3323 }
3324 spin_unlock(&pers_lock);
3325
3326 if (pers == mddev->pers) {
3327 /* Nothing to do! */
3328 module_put(pers->owner);
3329 rv = len;
3330 goto out_unlock;
3331 }
3332 if (!pers->takeover) {
3333 module_put(pers->owner);
3334 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3335 mdname(mddev), clevel);
3336 rv = -EINVAL;
3337 goto out_unlock;
3338 }
3339
3340 rdev_for_each(rdev, mddev)
3341 rdev->new_raid_disk = rdev->raid_disk;
3342
3343 /* ->takeover must set new_* and/or delta_disks
3344 * if it succeeds, and may set them when it fails.
3345 */
3346 priv = pers->takeover(mddev);
3347 if (IS_ERR(priv)) {
3348 mddev->new_level = mddev->level;
3349 mddev->new_layout = mddev->layout;
3350 mddev->new_chunk_sectors = mddev->chunk_sectors;
3351 mddev->raid_disks -= mddev->delta_disks;
3352 mddev->delta_disks = 0;
3353 mddev->reshape_backwards = 0;
3354 module_put(pers->owner);
3355 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3356 mdname(mddev), clevel);
3357 rv = PTR_ERR(priv);
3358 goto out_unlock;
3359 }
3360
3361 /* Looks like we have a winner */
3362 mddev_suspend(mddev);
3363 mddev_detach(mddev);
3364
3365 spin_lock(&mddev->lock);
3366 oldpers = mddev->pers;
3367 oldpriv = mddev->private;
3368 mddev->pers = pers;
3369 mddev->private = priv;
3370 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3371 mddev->level = mddev->new_level;
3372 mddev->layout = mddev->new_layout;
3373 mddev->chunk_sectors = mddev->new_chunk_sectors;
3374 mddev->delta_disks = 0;
3375 mddev->reshape_backwards = 0;
3376 mddev->degraded = 0;
3377 spin_unlock(&mddev->lock);
3378
3379 if (oldpers->sync_request == NULL &&
3380 mddev->external) {
3381 /* We are converting from a no-redundancy array
3382 * to a redundancy array and metadata is managed
3383 * externally so we need to be sure that writes
3384 * won't block due to a need to transition
3385 * clean->dirty
3386 * until external management is started.
3387 */
3388 mddev->in_sync = 0;
3389 mddev->safemode_delay = 0;
3390 mddev->safemode = 0;
3391 }
3392
3393 oldpers->free(mddev, oldpriv);
3394
3395 if (oldpers->sync_request == NULL &&
3396 pers->sync_request != NULL) {
3397 /* need to add the md_redundancy_group */
3398 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3399 printk(KERN_WARNING
3400 "md: cannot register extra attributes for %s\n",
3401 mdname(mddev));
3402 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3403 }
3404 if (oldpers->sync_request != NULL &&
3405 pers->sync_request == NULL) {
3406 /* need to remove the md_redundancy_group */
3407 if (mddev->to_remove == NULL)
3408 mddev->to_remove = &md_redundancy_group;
3409 }
3410
3411 rdev_for_each(rdev, mddev) {
3412 if (rdev->raid_disk < 0)
3413 continue;
3414 if (rdev->new_raid_disk >= mddev->raid_disks)
3415 rdev->new_raid_disk = -1;
3416 if (rdev->new_raid_disk == rdev->raid_disk)
3417 continue;
3418 sysfs_unlink_rdev(mddev, rdev);
3419 }
3420 rdev_for_each(rdev, mddev) {
3421 if (rdev->raid_disk < 0)
3422 continue;
3423 if (rdev->new_raid_disk == rdev->raid_disk)
3424 continue;
3425 rdev->raid_disk = rdev->new_raid_disk;
3426 if (rdev->raid_disk < 0)
3427 clear_bit(In_sync, &rdev->flags);
3428 else {
3429 if (sysfs_link_rdev(mddev, rdev))
3430 printk(KERN_WARNING "md: cannot register rd%d"
3431 " for %s after level change\n",
3432 rdev->raid_disk, mdname(mddev));
3433 }
3434 }
3435
3436 if (pers->sync_request == NULL) {
3437 /* this is now an array without redundancy, so
3438 * it must always be in_sync
3439 */
3440 mddev->in_sync = 1;
3441 del_timer_sync(&mddev->safemode_timer);
3442 }
3443 blk_set_stacking_limits(&mddev->queue->limits);
3444 pers->run(mddev);
3445 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3446 mddev_resume(mddev);
3447 if (!mddev->thread)
3448 md_update_sb(mddev, 1);
3449 sysfs_notify(&mddev->kobj, NULL, "level");
3450 md_new_event(mddev);
3451 rv = len;
3452 out_unlock:
3453 mddev_unlock(mddev);
3454 return rv;
3455 }
3456
3457 static struct md_sysfs_entry md_level =
3458 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3459
3460 static ssize_t
3461 layout_show(struct mddev *mddev, char *page)
3462 {
3463 /* just a number, not meaningful for all levels */
3464 if (mddev->reshape_position != MaxSector &&
3465 mddev->layout != mddev->new_layout)
3466 return sprintf(page, "%d (%d)\n",
3467 mddev->new_layout, mddev->layout);
3468 return sprintf(page, "%d\n", mddev->layout);
3469 }
3470
3471 static ssize_t
3472 layout_store(struct mddev *mddev, const char *buf, size_t len)
3473 {
3474 char *e;
3475 unsigned long n = simple_strtoul(buf, &e, 10);
3476 int err;
3477
3478 if (!*buf || (*e && *e != '\n'))
3479 return -EINVAL;
3480 err = mddev_lock(mddev);
3481 if (err)
3482 return err;
3483
3484 if (mddev->pers) {
3485 if (mddev->pers->check_reshape == NULL)
3486 err = -EBUSY;
3487 else if (mddev->ro)
3488 err = -EROFS;
3489 else {
3490 mddev->new_layout = n;
3491 err = mddev->pers->check_reshape(mddev);
3492 if (err)
3493 mddev->new_layout = mddev->layout;
3494 }
3495 } else {
3496 mddev->new_layout = n;
3497 if (mddev->reshape_position == MaxSector)
3498 mddev->layout = n;
3499 }
3500 mddev_unlock(mddev);
3501 return err ?: len;
3502 }
3503 static struct md_sysfs_entry md_layout =
3504 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3505
3506 static ssize_t
3507 raid_disks_show(struct mddev *mddev, char *page)
3508 {
3509 if (mddev->raid_disks == 0)
3510 return 0;
3511 if (mddev->reshape_position != MaxSector &&
3512 mddev->delta_disks != 0)
3513 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3514 mddev->raid_disks - mddev->delta_disks);
3515 return sprintf(page, "%d\n", mddev->raid_disks);
3516 }
3517
3518 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3519
3520 static ssize_t
3521 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3522 {
3523 char *e;
3524 int err;
3525 unsigned long n = simple_strtoul(buf, &e, 10);
3526
3527 if (!*buf || (*e && *e != '\n'))
3528 return -EINVAL;
3529
3530 err = mddev_lock(mddev);
3531 if (err)
3532 return err;
3533 if (mddev->pers)
3534 err = update_raid_disks(mddev, n);
3535 else if (mddev->reshape_position != MaxSector) {
3536 struct md_rdev *rdev;
3537 int olddisks = mddev->raid_disks - mddev->delta_disks;
3538
3539 err = -EINVAL;
3540 rdev_for_each(rdev, mddev) {
3541 if (olddisks < n &&
3542 rdev->data_offset < rdev->new_data_offset)
3543 goto out_unlock;
3544 if (olddisks > n &&
3545 rdev->data_offset > rdev->new_data_offset)
3546 goto out_unlock;
3547 }
3548 err = 0;
3549 mddev->delta_disks = n - olddisks;
3550 mddev->raid_disks = n;
3551 mddev->reshape_backwards = (mddev->delta_disks < 0);
3552 } else
3553 mddev->raid_disks = n;
3554 out_unlock:
3555 mddev_unlock(mddev);
3556 return err ? err : len;
3557 }
3558 static struct md_sysfs_entry md_raid_disks =
3559 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3560
3561 static ssize_t
3562 chunk_size_show(struct mddev *mddev, char *page)
3563 {
3564 if (mddev->reshape_position != MaxSector &&
3565 mddev->chunk_sectors != mddev->new_chunk_sectors)
3566 return sprintf(page, "%d (%d)\n",
3567 mddev->new_chunk_sectors << 9,
3568 mddev->chunk_sectors << 9);
3569 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3570 }
3571
3572 static ssize_t
3573 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3574 {
3575 int err;
3576 char *e;
3577 unsigned long n = simple_strtoul(buf, &e, 10);
3578
3579 if (!*buf || (*e && *e != '\n'))
3580 return -EINVAL;
3581
3582 err = mddev_lock(mddev);
3583 if (err)
3584 return err;
3585 if (mddev->pers) {
3586 if (mddev->pers->check_reshape == NULL)
3587 err = -EBUSY;
3588 else if (mddev->ro)
3589 err = -EROFS;
3590 else {
3591 mddev->new_chunk_sectors = n >> 9;
3592 err = mddev->pers->check_reshape(mddev);
3593 if (err)
3594 mddev->new_chunk_sectors = mddev->chunk_sectors;
3595 }
3596 } else {
3597 mddev->new_chunk_sectors = n >> 9;
3598 if (mddev->reshape_position == MaxSector)
3599 mddev->chunk_sectors = n >> 9;
3600 }
3601 mddev_unlock(mddev);
3602 return err ?: len;
3603 }
3604 static struct md_sysfs_entry md_chunk_size =
3605 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3606
3607 static ssize_t
3608 resync_start_show(struct mddev *mddev, char *page)
3609 {
3610 if (mddev->recovery_cp == MaxSector)
3611 return sprintf(page, "none\n");
3612 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3613 }
3614
3615 static ssize_t
3616 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3617 {
3618 int err;
3619 char *e;
3620 unsigned long long n = simple_strtoull(buf, &e, 10);
3621
3622 err = mddev_lock(mddev);
3623 if (err)
3624 return err;
3625 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3626 err = -EBUSY;
3627 else if (cmd_match(buf, "none"))
3628 n = MaxSector;
3629 else if (!*buf || (*e && *e != '\n'))
3630 err = -EINVAL;
3631
3632 if (!err) {
3633 mddev->recovery_cp = n;
3634 if (mddev->pers)
3635 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3636 }
3637 mddev_unlock(mddev);
3638 return err ?: len;
3639 }
3640 static struct md_sysfs_entry md_resync_start =
3641 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3642
3643 /*
3644 * The array state can be:
3645 *
3646 * clear
3647 * No devices, no size, no level
3648 * Equivalent to STOP_ARRAY ioctl
3649 * inactive
3650 * May have some settings, but array is not active
3651 * all IO results in error
3652 * When written, doesn't tear down array, but just stops it
3653 * suspended (not supported yet)
3654 * All IO requests will block. The array can be reconfigured.
3655 * Writing this, if accepted, will block until array is quiescent
3656 * readonly
3657 * no resync can happen. no superblocks get written.
3658 * write requests fail
3659 * read-auto
3660 * like readonly, but behaves like 'clean' on a write request.
3661 *
3662 * clean - no pending writes, but otherwise active.
3663 * When written to inactive array, starts without resync
3664 * If a write request arrives then
3665 * if metadata is known, mark 'dirty' and switch to 'active'.
3666 * if not known, block and switch to write-pending
3667 * If written to an active array that has pending writes, then fails.
3668 * active
3669 * fully active: IO and resync can be happening.
3670 * When written to inactive array, starts with resync
3671 *
3672 * write-pending
3673 * clean, but writes are blocked waiting for 'active' to be written.
3674 *
3675 * active-idle
3676 * like active, but no writes have been seen for a while (100msec).
3677 *
3678 */
3679 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3680 write_pending, active_idle, bad_word};
3681 static char *array_states[] = {
3682 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3683 "write-pending", "active-idle", NULL };
3684
3685 static int match_word(const char *word, char **list)
3686 {
3687 int n;
3688 for (n=0; list[n]; n++)
3689 if (cmd_match(word, list[n]))
3690 break;
3691 return n;
3692 }
3693
3694 static ssize_t
3695 array_state_show(struct mddev *mddev, char *page)
3696 {
3697 enum array_state st = inactive;
3698
3699 if (mddev->pers)
3700 switch(mddev->ro) {
3701 case 1:
3702 st = readonly;
3703 break;
3704 case 2:
3705 st = read_auto;
3706 break;
3707 case 0:
3708 if (mddev->in_sync)
3709 st = clean;
3710 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3711 st = write_pending;
3712 else if (mddev->safemode)
3713 st = active_idle;
3714 else
3715 st = active;
3716 }
3717 else {
3718 if (list_empty(&mddev->disks) &&
3719 mddev->raid_disks == 0 &&
3720 mddev->dev_sectors == 0)
3721 st = clear;
3722 else
3723 st = inactive;
3724 }
3725 return sprintf(page, "%s\n", array_states[st]);
3726 }
3727
3728 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
3729 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
3730 static int do_md_run(struct mddev *mddev);
3731 static int restart_array(struct mddev *mddev);
3732
3733 static ssize_t
3734 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3735 {
3736 int err;
3737 enum array_state st = match_word(buf, array_states);
3738
3739 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
3740 /* don't take reconfig_mutex when toggling between
3741 * clean and active
3742 */
3743 spin_lock(&mddev->lock);
3744 if (st == active) {
3745 restart_array(mddev);
3746 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3747 wake_up(&mddev->sb_wait);
3748 err = 0;
3749 } else /* st == clean */ {
3750 restart_array(mddev);
3751 if (atomic_read(&mddev->writes_pending) == 0) {
3752 if (mddev->in_sync == 0) {
3753 mddev->in_sync = 1;
3754 if (mddev->safemode == 1)
3755 mddev->safemode = 0;
3756 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3757 }
3758 err = 0;
3759 } else
3760 err = -EBUSY;
3761 }
3762 spin_unlock(&mddev->lock);
3763 return err;
3764 }
3765 err = mddev_lock(mddev);
3766 if (err)
3767 return err;
3768 err = -EINVAL;
3769 switch(st) {
3770 case bad_word:
3771 break;
3772 case clear:
3773 /* stopping an active array */
3774 err = do_md_stop(mddev, 0, NULL);
3775 break;
3776 case inactive:
3777 /* stopping an active array */
3778 if (mddev->pers)
3779 err = do_md_stop(mddev, 2, NULL);
3780 else
3781 err = 0; /* already inactive */
3782 break;
3783 case suspended:
3784 break; /* not supported yet */
3785 case readonly:
3786 if (mddev->pers)
3787 err = md_set_readonly(mddev, NULL);
3788 else {
3789 mddev->ro = 1;
3790 set_disk_ro(mddev->gendisk, 1);
3791 err = do_md_run(mddev);
3792 }
3793 break;
3794 case read_auto:
3795 if (mddev->pers) {
3796 if (mddev->ro == 0)
3797 err = md_set_readonly(mddev, NULL);
3798 else if (mddev->ro == 1)
3799 err = restart_array(mddev);
3800 if (err == 0) {
3801 mddev->ro = 2;
3802 set_disk_ro(mddev->gendisk, 0);
3803 }
3804 } else {
3805 mddev->ro = 2;
3806 err = do_md_run(mddev);
3807 }
3808 break;
3809 case clean:
3810 if (mddev->pers) {
3811 restart_array(mddev);
3812 spin_lock(&mddev->lock);
3813 if (atomic_read(&mddev->writes_pending) == 0) {
3814 if (mddev->in_sync == 0) {
3815 mddev->in_sync = 1;
3816 if (mddev->safemode == 1)
3817 mddev->safemode = 0;
3818 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3819 }
3820 err = 0;
3821 } else
3822 err = -EBUSY;
3823 spin_unlock(&mddev->lock);
3824 } else
3825 err = -EINVAL;
3826 break;
3827 case active:
3828 if (mddev->pers) {
3829 restart_array(mddev);
3830 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3831 wake_up(&mddev->sb_wait);
3832 err = 0;
3833 } else {
3834 mddev->ro = 0;
3835 set_disk_ro(mddev->gendisk, 0);
3836 err = do_md_run(mddev);
3837 }
3838 break;
3839 case write_pending:
3840 case active_idle:
3841 /* these cannot be set */
3842 break;
3843 }
3844
3845 if (!err) {
3846 if (mddev->hold_active == UNTIL_IOCTL)
3847 mddev->hold_active = 0;
3848 sysfs_notify_dirent_safe(mddev->sysfs_state);
3849 }
3850 mddev_unlock(mddev);
3851 return err ?: len;
3852 }
3853 static struct md_sysfs_entry md_array_state =
3854 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3855
3856 static ssize_t
3857 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
3858 return sprintf(page, "%d\n",
3859 atomic_read(&mddev->max_corr_read_errors));
3860 }
3861
3862 static ssize_t
3863 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
3864 {
3865 char *e;
3866 unsigned long n = simple_strtoul(buf, &e, 10);
3867
3868 if (*buf && (*e == 0 || *e == '\n')) {
3869 atomic_set(&mddev->max_corr_read_errors, n);
3870 return len;
3871 }
3872 return -EINVAL;
3873 }
3874
3875 static struct md_sysfs_entry max_corr_read_errors =
3876 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3877 max_corrected_read_errors_store);
3878
3879 static ssize_t
3880 null_show(struct mddev *mddev, char *page)
3881 {
3882 return -EINVAL;
3883 }
3884
3885 static ssize_t
3886 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
3887 {
3888 /* buf must be %d:%d\n? giving major and minor numbers */
3889 /* The new device is added to the array.
3890 * If the array has a persistent superblock, we read the
3891 * superblock to initialise info and check validity.
3892 * Otherwise, only checking done is that in bind_rdev_to_array,
3893 * which mainly checks size.
3894 */
3895 char *e;
3896 int major = simple_strtoul(buf, &e, 10);
3897 int minor;
3898 dev_t dev;
3899 struct md_rdev *rdev;
3900 int err;
3901
3902 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3903 return -EINVAL;
3904 minor = simple_strtoul(e+1, &e, 10);
3905 if (*e && *e != '\n')
3906 return -EINVAL;
3907 dev = MKDEV(major, minor);
3908 if (major != MAJOR(dev) ||
3909 minor != MINOR(dev))
3910 return -EOVERFLOW;
3911
3912 flush_workqueue(md_misc_wq);
3913
3914 err = mddev_lock(mddev);
3915 if (err)
3916 return err;
3917 if (mddev->persistent) {
3918 rdev = md_import_device(dev, mddev->major_version,
3919 mddev->minor_version);
3920 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3921 struct md_rdev *rdev0
3922 = list_entry(mddev->disks.next,
3923 struct md_rdev, same_set);
3924 err = super_types[mddev->major_version]
3925 .load_super(rdev, rdev0, mddev->minor_version);
3926 if (err < 0)
3927 goto out;
3928 }
3929 } else if (mddev->external)
3930 rdev = md_import_device(dev, -2, -1);
3931 else
3932 rdev = md_import_device(dev, -1, -1);
3933
3934 if (IS_ERR(rdev))
3935 return PTR_ERR(rdev);
3936 err = bind_rdev_to_array(rdev, mddev);
3937 out:
3938 if (err)
3939 export_rdev(rdev);
3940 mddev_unlock(mddev);
3941 return err ? err : len;
3942 }
3943
3944 static struct md_sysfs_entry md_new_device =
3945 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3946
3947 static ssize_t
3948 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
3949 {
3950 char *end;
3951 unsigned long chunk, end_chunk;
3952 int err;
3953
3954 err = mddev_lock(mddev);
3955 if (err)
3956 return err;
3957 if (!mddev->bitmap)
3958 goto out;
3959 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3960 while (*buf) {
3961 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3962 if (buf == end) break;
3963 if (*end == '-') { /* range */
3964 buf = end + 1;
3965 end_chunk = simple_strtoul(buf, &end, 0);
3966 if (buf == end) break;
3967 }
3968 if (*end && !isspace(*end)) break;
3969 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3970 buf = skip_spaces(end);
3971 }
3972 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3973 out:
3974 mddev_unlock(mddev);
3975 return len;
3976 }
3977
3978 static struct md_sysfs_entry md_bitmap =
3979 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3980
3981 static ssize_t
3982 size_show(struct mddev *mddev, char *page)
3983 {
3984 return sprintf(page, "%llu\n",
3985 (unsigned long long)mddev->dev_sectors / 2);
3986 }
3987
3988 static int update_size(struct mddev *mddev, sector_t num_sectors);
3989
3990 static ssize_t
3991 size_store(struct mddev *mddev, const char *buf, size_t len)
3992 {
3993 /* If array is inactive, we can reduce the component size, but
3994 * not increase it (except from 0).
3995 * If array is active, we can try an on-line resize
3996 */
3997 sector_t sectors;
3998 int err = strict_blocks_to_sectors(buf, &sectors);
3999
4000 if (err < 0)
4001 return err;
4002 err = mddev_lock(mddev);
4003 if (err)
4004 return err;
4005 if (mddev->pers) {
4006 err = update_size(mddev, sectors);
4007 md_update_sb(mddev, 1);
4008 } else {
4009 if (mddev->dev_sectors == 0 ||
4010 mddev->dev_sectors > sectors)
4011 mddev->dev_sectors = sectors;
4012 else
4013 err = -ENOSPC;
4014 }
4015 mddev_unlock(mddev);
4016 return err ? err : len;
4017 }
4018
4019 static struct md_sysfs_entry md_size =
4020 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4021
4022 /* Metadata version.
4023 * This is one of
4024 * 'none' for arrays with no metadata (good luck...)
4025 * 'external' for arrays with externally managed metadata,
4026 * or N.M for internally known formats
4027 */
4028 static ssize_t
4029 metadata_show(struct mddev *mddev, char *page)
4030 {
4031 if (mddev->persistent)
4032 return sprintf(page, "%d.%d\n",
4033 mddev->major_version, mddev->minor_version);
4034 else if (mddev->external)
4035 return sprintf(page, "external:%s\n", mddev->metadata_type);
4036 else
4037 return sprintf(page, "none\n");
4038 }
4039
4040 static ssize_t
4041 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4042 {
4043 int major, minor;
4044 char *e;
4045 int err;
4046 /* Changing the details of 'external' metadata is
4047 * always permitted. Otherwise there must be
4048 * no devices attached to the array.
4049 */
4050
4051 err = mddev_lock(mddev);
4052 if (err)
4053 return err;
4054 err = -EBUSY;
4055 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4056 ;
4057 else if (!list_empty(&mddev->disks))
4058 goto out_unlock;
4059
4060 err = 0;
4061 if (cmd_match(buf, "none")) {
4062 mddev->persistent = 0;
4063 mddev->external = 0;
4064 mddev->major_version = 0;
4065 mddev->minor_version = 90;
4066 goto out_unlock;
4067 }
4068 if (strncmp(buf, "external:", 9) == 0) {
4069 size_t namelen = len-9;
4070 if (namelen >= sizeof(mddev->metadata_type))
4071 namelen = sizeof(mddev->metadata_type)-1;
4072 strncpy(mddev->metadata_type, buf+9, namelen);
4073 mddev->metadata_type[namelen] = 0;
4074 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4075 mddev->metadata_type[--namelen] = 0;
4076 mddev->persistent = 0;
4077 mddev->external = 1;
4078 mddev->major_version = 0;
4079 mddev->minor_version = 90;
4080 goto out_unlock;
4081 }
4082 major = simple_strtoul(buf, &e, 10);
4083 err = -EINVAL;
4084 if (e==buf || *e != '.')
4085 goto out_unlock;
4086 buf = e+1;
4087 minor = simple_strtoul(buf, &e, 10);
4088 if (e==buf || (*e && *e != '\n') )
4089 goto out_unlock;
4090 err = -ENOENT;
4091 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4092 goto out_unlock;
4093 mddev->major_version = major;
4094 mddev->minor_version = minor;
4095 mddev->persistent = 1;
4096 mddev->external = 0;
4097 err = 0;
4098 out_unlock:
4099 mddev_unlock(mddev);
4100 return err ?: len;
4101 }
4102
4103 static struct md_sysfs_entry md_metadata =
4104 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4105
4106 static ssize_t
4107 action_show(struct mddev *mddev, char *page)
4108 {
4109 char *type = "idle";
4110 unsigned long recovery = mddev->recovery;
4111 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4112 type = "frozen";
4113 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4114 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4115 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4116 type = "reshape";
4117 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4118 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4119 type = "resync";
4120 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4121 type = "check";
4122 else
4123 type = "repair";
4124 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4125 type = "recover";
4126 }
4127 return sprintf(page, "%s\n", type);
4128 }
4129
4130 static ssize_t
4131 action_store(struct mddev *mddev, const char *page, size_t len)
4132 {
4133 if (!mddev->pers || !mddev->pers->sync_request)
4134 return -EINVAL;
4135
4136 if (cmd_match(page, "frozen"))
4137 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4138 else
4139 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4140
4141 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4142 flush_workqueue(md_misc_wq);
4143 if (mddev->sync_thread) {
4144 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4145 if (mddev_lock(mddev) == 0) {
4146 md_reap_sync_thread(mddev);
4147 mddev_unlock(mddev);
4148 }
4149 }
4150 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4151 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
4152 return -EBUSY;
4153 else if (cmd_match(page, "resync"))
4154 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4155 else if (cmd_match(page, "recover")) {
4156 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4157 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4158 } else if (cmd_match(page, "reshape")) {
4159 int err;
4160 if (mddev->pers->start_reshape == NULL)
4161 return -EINVAL;
4162 err = mddev_lock(mddev);
4163 if (!err) {
4164 err = mddev->pers->start_reshape(mddev);
4165 mddev_unlock(mddev);
4166 }
4167 if (err)
4168 return err;
4169 sysfs_notify(&mddev->kobj, NULL, "degraded");
4170 } else {
4171 if (cmd_match(page, "check"))
4172 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4173 else if (!cmd_match(page, "repair"))
4174 return -EINVAL;
4175 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4176 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4177 }
4178 if (mddev->ro == 2) {
4179 /* A write to sync_action is enough to justify
4180 * canceling read-auto mode
4181 */
4182 mddev->ro = 0;
4183 md_wakeup_thread(mddev->sync_thread);
4184 }
4185 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4186 md_wakeup_thread(mddev->thread);
4187 sysfs_notify_dirent_safe(mddev->sysfs_action);
4188 return len;
4189 }
4190
4191 static struct md_sysfs_entry md_scan_mode =
4192 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4193
4194 static ssize_t
4195 last_sync_action_show(struct mddev *mddev, char *page)
4196 {
4197 return sprintf(page, "%s\n", mddev->last_sync_action);
4198 }
4199
4200 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4201
4202 static ssize_t
4203 mismatch_cnt_show(struct mddev *mddev, char *page)
4204 {
4205 return sprintf(page, "%llu\n",
4206 (unsigned long long)
4207 atomic64_read(&mddev->resync_mismatches));
4208 }
4209
4210 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4211
4212 static ssize_t
4213 sync_min_show(struct mddev *mddev, char *page)
4214 {
4215 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4216 mddev->sync_speed_min ? "local": "system");
4217 }
4218
4219 static ssize_t
4220 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4221 {
4222 int min;
4223 char *e;
4224 if (strncmp(buf, "system", 6)==0) {
4225 mddev->sync_speed_min = 0;
4226 return len;
4227 }
4228 min = simple_strtoul(buf, &e, 10);
4229 if (buf == e || (*e && *e != '\n') || min <= 0)
4230 return -EINVAL;
4231 mddev->sync_speed_min = min;
4232 return len;
4233 }
4234
4235 static struct md_sysfs_entry md_sync_min =
4236 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4237
4238 static ssize_t
4239 sync_max_show(struct mddev *mddev, char *page)
4240 {
4241 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4242 mddev->sync_speed_max ? "local": "system");
4243 }
4244
4245 static ssize_t
4246 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4247 {
4248 int max;
4249 char *e;
4250 if (strncmp(buf, "system", 6)==0) {
4251 mddev->sync_speed_max = 0;
4252 return len;
4253 }
4254 max = simple_strtoul(buf, &e, 10);
4255 if (buf == e || (*e && *e != '\n') || max <= 0)
4256 return -EINVAL;
4257 mddev->sync_speed_max = max;
4258 return len;
4259 }
4260
4261 static struct md_sysfs_entry md_sync_max =
4262 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4263
4264 static ssize_t
4265 degraded_show(struct mddev *mddev, char *page)
4266 {
4267 return sprintf(page, "%d\n", mddev->degraded);
4268 }
4269 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4270
4271 static ssize_t
4272 sync_force_parallel_show(struct mddev *mddev, char *page)
4273 {
4274 return sprintf(page, "%d\n", mddev->parallel_resync);
4275 }
4276
4277 static ssize_t
4278 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4279 {
4280 long n;
4281
4282 if (kstrtol(buf, 10, &n))
4283 return -EINVAL;
4284
4285 if (n != 0 && n != 1)
4286 return -EINVAL;
4287
4288 mddev->parallel_resync = n;
4289
4290 if (mddev->sync_thread)
4291 wake_up(&resync_wait);
4292
4293 return len;
4294 }
4295
4296 /* force parallel resync, even with shared block devices */
4297 static struct md_sysfs_entry md_sync_force_parallel =
4298 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4299 sync_force_parallel_show, sync_force_parallel_store);
4300
4301 static ssize_t
4302 sync_speed_show(struct mddev *mddev, char *page)
4303 {
4304 unsigned long resync, dt, db;
4305 if (mddev->curr_resync == 0)
4306 return sprintf(page, "none\n");
4307 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4308 dt = (jiffies - mddev->resync_mark) / HZ;
4309 if (!dt) dt++;
4310 db = resync - mddev->resync_mark_cnt;
4311 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4312 }
4313
4314 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4315
4316 static ssize_t
4317 sync_completed_show(struct mddev *mddev, char *page)
4318 {
4319 unsigned long long max_sectors, resync;
4320
4321 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4322 return sprintf(page, "none\n");
4323
4324 if (mddev->curr_resync == 1 ||
4325 mddev->curr_resync == 2)
4326 return sprintf(page, "delayed\n");
4327
4328 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4329 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4330 max_sectors = mddev->resync_max_sectors;
4331 else
4332 max_sectors = mddev->dev_sectors;
4333
4334 resync = mddev->curr_resync_completed;
4335 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4336 }
4337
4338 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
4339
4340 static ssize_t
4341 min_sync_show(struct mddev *mddev, char *page)
4342 {
4343 return sprintf(page, "%llu\n",
4344 (unsigned long long)mddev->resync_min);
4345 }
4346 static ssize_t
4347 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4348 {
4349 unsigned long long min;
4350 int err;
4351 int chunk;
4352
4353 if (kstrtoull(buf, 10, &min))
4354 return -EINVAL;
4355
4356 spin_lock(&mddev->lock);
4357 err = -EINVAL;
4358 if (min > mddev->resync_max)
4359 goto out_unlock;
4360
4361 err = -EBUSY;
4362 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4363 goto out_unlock;
4364
4365 /* Must be a multiple of chunk_size */
4366 chunk = mddev->chunk_sectors;
4367 if (chunk) {
4368 sector_t temp = min;
4369
4370 err = -EINVAL;
4371 if (sector_div(temp, chunk))
4372 goto out_unlock;
4373 }
4374 mddev->resync_min = min;
4375 err = 0;
4376
4377 out_unlock:
4378 spin_unlock(&mddev->lock);
4379 return err ?: len;
4380 }
4381
4382 static struct md_sysfs_entry md_min_sync =
4383 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4384
4385 static ssize_t
4386 max_sync_show(struct mddev *mddev, char *page)
4387 {
4388 if (mddev->resync_max == MaxSector)
4389 return sprintf(page, "max\n");
4390 else
4391 return sprintf(page, "%llu\n",
4392 (unsigned long long)mddev->resync_max);
4393 }
4394 static ssize_t
4395 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4396 {
4397 int err;
4398 spin_lock(&mddev->lock);
4399 if (strncmp(buf, "max", 3) == 0)
4400 mddev->resync_max = MaxSector;
4401 else {
4402 unsigned long long max;
4403 int chunk;
4404
4405 err = -EINVAL;
4406 if (kstrtoull(buf, 10, &max))
4407 goto out_unlock;
4408 if (max < mddev->resync_min)
4409 goto out_unlock;
4410
4411 err = -EBUSY;
4412 if (max < mddev->resync_max &&
4413 mddev->ro == 0 &&
4414 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4415 goto out_unlock;
4416
4417 /* Must be a multiple of chunk_size */
4418 chunk = mddev->chunk_sectors;
4419 if (chunk) {
4420 sector_t temp = max;
4421
4422 err = -EINVAL;
4423 if (sector_div(temp, chunk))
4424 goto out_unlock;
4425 }
4426 mddev->resync_max = max;
4427 }
4428 wake_up(&mddev->recovery_wait);
4429 err = 0;
4430 out_unlock:
4431 spin_unlock(&mddev->lock);
4432 return err ?: len;
4433 }
4434
4435 static struct md_sysfs_entry md_max_sync =
4436 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4437
4438 static ssize_t
4439 suspend_lo_show(struct mddev *mddev, char *page)
4440 {
4441 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4442 }
4443
4444 static ssize_t
4445 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4446 {
4447 char *e;
4448 unsigned long long new = simple_strtoull(buf, &e, 10);
4449 unsigned long long old;
4450 int err;
4451
4452 if (buf == e || (*e && *e != '\n'))
4453 return -EINVAL;
4454
4455 err = mddev_lock(mddev);
4456 if (err)
4457 return err;
4458 err = -EINVAL;
4459 if (mddev->pers == NULL ||
4460 mddev->pers->quiesce == NULL)
4461 goto unlock;
4462 old = mddev->suspend_lo;
4463 mddev->suspend_lo = new;
4464 if (new >= old)
4465 /* Shrinking suspended region */
4466 mddev->pers->quiesce(mddev, 2);
4467 else {
4468 /* Expanding suspended region - need to wait */
4469 mddev->pers->quiesce(mddev, 1);
4470 mddev->pers->quiesce(mddev, 0);
4471 }
4472 err = 0;
4473 unlock:
4474 mddev_unlock(mddev);
4475 return err ?: len;
4476 }
4477 static struct md_sysfs_entry md_suspend_lo =
4478 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4479
4480 static ssize_t
4481 suspend_hi_show(struct mddev *mddev, char *page)
4482 {
4483 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4484 }
4485
4486 static ssize_t
4487 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4488 {
4489 char *e;
4490 unsigned long long new = simple_strtoull(buf, &e, 10);
4491 unsigned long long old;
4492 int err;
4493
4494 if (buf == e || (*e && *e != '\n'))
4495 return -EINVAL;
4496
4497 err = mddev_lock(mddev);
4498 if (err)
4499 return err;
4500 err = -EINVAL;
4501 if (mddev->pers == NULL ||
4502 mddev->pers->quiesce == NULL)
4503 goto unlock;
4504 old = mddev->suspend_hi;
4505 mddev->suspend_hi = new;
4506 if (new <= old)
4507 /* Shrinking suspended region */
4508 mddev->pers->quiesce(mddev, 2);
4509 else {
4510 /* Expanding suspended region - need to wait */
4511 mddev->pers->quiesce(mddev, 1);
4512 mddev->pers->quiesce(mddev, 0);
4513 }
4514 err = 0;
4515 unlock:
4516 mddev_unlock(mddev);
4517 return err ?: len;
4518 }
4519 static struct md_sysfs_entry md_suspend_hi =
4520 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4521
4522 static ssize_t
4523 reshape_position_show(struct mddev *mddev, char *page)
4524 {
4525 if (mddev->reshape_position != MaxSector)
4526 return sprintf(page, "%llu\n",
4527 (unsigned long long)mddev->reshape_position);
4528 strcpy(page, "none\n");
4529 return 5;
4530 }
4531
4532 static ssize_t
4533 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4534 {
4535 struct md_rdev *rdev;
4536 char *e;
4537 int err;
4538 unsigned long long new = simple_strtoull(buf, &e, 10);
4539
4540 if (buf == e || (*e && *e != '\n'))
4541 return -EINVAL;
4542 err = mddev_lock(mddev);
4543 if (err)
4544 return err;
4545 err = -EBUSY;
4546 if (mddev->pers)
4547 goto unlock;
4548 mddev->reshape_position = new;
4549 mddev->delta_disks = 0;
4550 mddev->reshape_backwards = 0;
4551 mddev->new_level = mddev->level;
4552 mddev->new_layout = mddev->layout;
4553 mddev->new_chunk_sectors = mddev->chunk_sectors;
4554 rdev_for_each(rdev, mddev)
4555 rdev->new_data_offset = rdev->data_offset;
4556 err = 0;
4557 unlock:
4558 mddev_unlock(mddev);
4559 return err ?: len;
4560 }
4561
4562 static struct md_sysfs_entry md_reshape_position =
4563 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4564 reshape_position_store);
4565
4566 static ssize_t
4567 reshape_direction_show(struct mddev *mddev, char *page)
4568 {
4569 return sprintf(page, "%s\n",
4570 mddev->reshape_backwards ? "backwards" : "forwards");
4571 }
4572
4573 static ssize_t
4574 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4575 {
4576 int backwards = 0;
4577 int err;
4578
4579 if (cmd_match(buf, "forwards"))
4580 backwards = 0;
4581 else if (cmd_match(buf, "backwards"))
4582 backwards = 1;
4583 else
4584 return -EINVAL;
4585 if (mddev->reshape_backwards == backwards)
4586 return len;
4587
4588 err = mddev_lock(mddev);
4589 if (err)
4590 return err;
4591 /* check if we are allowed to change */
4592 if (mddev->delta_disks)
4593 err = -EBUSY;
4594 else if (mddev->persistent &&
4595 mddev->major_version == 0)
4596 err = -EINVAL;
4597 else
4598 mddev->reshape_backwards = backwards;
4599 mddev_unlock(mddev);
4600 return err ?: len;
4601 }
4602
4603 static struct md_sysfs_entry md_reshape_direction =
4604 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4605 reshape_direction_store);
4606
4607 static ssize_t
4608 array_size_show(struct mddev *mddev, char *page)
4609 {
4610 if (mddev->external_size)
4611 return sprintf(page, "%llu\n",
4612 (unsigned long long)mddev->array_sectors/2);
4613 else
4614 return sprintf(page, "default\n");
4615 }
4616
4617 static ssize_t
4618 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4619 {
4620 sector_t sectors;
4621 int err;
4622
4623 err = mddev_lock(mddev);
4624 if (err)
4625 return err;
4626
4627 if (strncmp(buf, "default", 7) == 0) {
4628 if (mddev->pers)
4629 sectors = mddev->pers->size(mddev, 0, 0);
4630 else
4631 sectors = mddev->array_sectors;
4632
4633 mddev->external_size = 0;
4634 } else {
4635 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4636 err = -EINVAL;
4637 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4638 err = -E2BIG;
4639 else
4640 mddev->external_size = 1;
4641 }
4642
4643 if (!err) {
4644 mddev->array_sectors = sectors;
4645 if (mddev->pers) {
4646 set_capacity(mddev->gendisk, mddev->array_sectors);
4647 revalidate_disk(mddev->gendisk);
4648 }
4649 }
4650 mddev_unlock(mddev);
4651 return err ?: len;
4652 }
4653
4654 static struct md_sysfs_entry md_array_size =
4655 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4656 array_size_store);
4657
4658 static struct attribute *md_default_attrs[] = {
4659 &md_level.attr,
4660 &md_layout.attr,
4661 &md_raid_disks.attr,
4662 &md_chunk_size.attr,
4663 &md_size.attr,
4664 &md_resync_start.attr,
4665 &md_metadata.attr,
4666 &md_new_device.attr,
4667 &md_safe_delay.attr,
4668 &md_array_state.attr,
4669 &md_reshape_position.attr,
4670 &md_reshape_direction.attr,
4671 &md_array_size.attr,
4672 &max_corr_read_errors.attr,
4673 NULL,
4674 };
4675
4676 static struct attribute *md_redundancy_attrs[] = {
4677 &md_scan_mode.attr,
4678 &md_last_scan_mode.attr,
4679 &md_mismatches.attr,
4680 &md_sync_min.attr,
4681 &md_sync_max.attr,
4682 &md_sync_speed.attr,
4683 &md_sync_force_parallel.attr,
4684 &md_sync_completed.attr,
4685 &md_min_sync.attr,
4686 &md_max_sync.attr,
4687 &md_suspend_lo.attr,
4688 &md_suspend_hi.attr,
4689 &md_bitmap.attr,
4690 &md_degraded.attr,
4691 NULL,
4692 };
4693 static struct attribute_group md_redundancy_group = {
4694 .name = NULL,
4695 .attrs = md_redundancy_attrs,
4696 };
4697
4698 static ssize_t
4699 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4700 {
4701 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4702 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4703 ssize_t rv;
4704
4705 if (!entry->show)
4706 return -EIO;
4707 spin_lock(&all_mddevs_lock);
4708 if (list_empty(&mddev->all_mddevs)) {
4709 spin_unlock(&all_mddevs_lock);
4710 return -EBUSY;
4711 }
4712 mddev_get(mddev);
4713 spin_unlock(&all_mddevs_lock);
4714
4715 rv = entry->show(mddev, page);
4716 mddev_put(mddev);
4717 return rv;
4718 }
4719
4720 static ssize_t
4721 md_attr_store(struct kobject *kobj, struct attribute *attr,
4722 const char *page, size_t length)
4723 {
4724 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4725 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4726 ssize_t rv;
4727
4728 if (!entry->store)
4729 return -EIO;
4730 if (!capable(CAP_SYS_ADMIN))
4731 return -EACCES;
4732 spin_lock(&all_mddevs_lock);
4733 if (list_empty(&mddev->all_mddevs)) {
4734 spin_unlock(&all_mddevs_lock);
4735 return -EBUSY;
4736 }
4737 mddev_get(mddev);
4738 spin_unlock(&all_mddevs_lock);
4739 rv = entry->store(mddev, page, length);
4740 mddev_put(mddev);
4741 return rv;
4742 }
4743
4744 static void md_free(struct kobject *ko)
4745 {
4746 struct mddev *mddev = container_of(ko, struct mddev, kobj);
4747
4748 if (mddev->sysfs_state)
4749 sysfs_put(mddev->sysfs_state);
4750
4751 if (mddev->gendisk) {
4752 del_gendisk(mddev->gendisk);
4753 put_disk(mddev->gendisk);
4754 }
4755 if (mddev->queue)
4756 blk_cleanup_queue(mddev->queue);
4757
4758 kfree(mddev);
4759 }
4760
4761 static const struct sysfs_ops md_sysfs_ops = {
4762 .show = md_attr_show,
4763 .store = md_attr_store,
4764 };
4765 static struct kobj_type md_ktype = {
4766 .release = md_free,
4767 .sysfs_ops = &md_sysfs_ops,
4768 .default_attrs = md_default_attrs,
4769 };
4770
4771 int mdp_major = 0;
4772
4773 static void mddev_delayed_delete(struct work_struct *ws)
4774 {
4775 struct mddev *mddev = container_of(ws, struct mddev, del_work);
4776
4777 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4778 kobject_del(&mddev->kobj);
4779 kobject_put(&mddev->kobj);
4780 }
4781
4782 static int md_alloc(dev_t dev, char *name)
4783 {
4784 static DEFINE_MUTEX(disks_mutex);
4785 struct mddev *mddev = mddev_find(dev);
4786 struct gendisk *disk;
4787 int partitioned;
4788 int shift;
4789 int unit;
4790 int error;
4791
4792 if (!mddev)
4793 return -ENODEV;
4794
4795 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4796 shift = partitioned ? MdpMinorShift : 0;
4797 unit = MINOR(mddev->unit) >> shift;
4798
4799 /* wait for any previous instance of this device to be
4800 * completely removed (mddev_delayed_delete).
4801 */
4802 flush_workqueue(md_misc_wq);
4803
4804 mutex_lock(&disks_mutex);
4805 error = -EEXIST;
4806 if (mddev->gendisk)
4807 goto abort;
4808
4809 if (name) {
4810 /* Need to ensure that 'name' is not a duplicate.
4811 */
4812 struct mddev *mddev2;
4813 spin_lock(&all_mddevs_lock);
4814
4815 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4816 if (mddev2->gendisk &&
4817 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4818 spin_unlock(&all_mddevs_lock);
4819 goto abort;
4820 }
4821 spin_unlock(&all_mddevs_lock);
4822 }
4823
4824 error = -ENOMEM;
4825 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4826 if (!mddev->queue)
4827 goto abort;
4828 mddev->queue->queuedata = mddev;
4829
4830 blk_queue_make_request(mddev->queue, md_make_request);
4831 blk_set_stacking_limits(&mddev->queue->limits);
4832
4833 disk = alloc_disk(1 << shift);
4834 if (!disk) {
4835 blk_cleanup_queue(mddev->queue);
4836 mddev->queue = NULL;
4837 goto abort;
4838 }
4839 disk->major = MAJOR(mddev->unit);
4840 disk->first_minor = unit << shift;
4841 if (name)
4842 strcpy(disk->disk_name, name);
4843 else if (partitioned)
4844 sprintf(disk->disk_name, "md_d%d", unit);
4845 else
4846 sprintf(disk->disk_name, "md%d", unit);
4847 disk->fops = &md_fops;
4848 disk->private_data = mddev;
4849 disk->queue = mddev->queue;
4850 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
4851 /* Allow extended partitions. This makes the
4852 * 'mdp' device redundant, but we can't really
4853 * remove it now.
4854 */
4855 disk->flags |= GENHD_FL_EXT_DEVT;
4856 mddev->gendisk = disk;
4857 /* As soon as we call add_disk(), another thread could get
4858 * through to md_open, so make sure it doesn't get too far
4859 */
4860 mutex_lock(&mddev->open_mutex);
4861 add_disk(disk);
4862
4863 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4864 &disk_to_dev(disk)->kobj, "%s", "md");
4865 if (error) {
4866 /* This isn't possible, but as kobject_init_and_add is marked
4867 * __must_check, we must do something with the result
4868 */
4869 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4870 disk->disk_name);
4871 error = 0;
4872 }
4873 if (mddev->kobj.sd &&
4874 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4875 printk(KERN_DEBUG "pointless warning\n");
4876 mutex_unlock(&mddev->open_mutex);
4877 abort:
4878 mutex_unlock(&disks_mutex);
4879 if (!error && mddev->kobj.sd) {
4880 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4881 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4882 }
4883 mddev_put(mddev);
4884 return error;
4885 }
4886
4887 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4888 {
4889 md_alloc(dev, NULL);
4890 return NULL;
4891 }
4892
4893 static int add_named_array(const char *val, struct kernel_param *kp)
4894 {
4895 /* val must be "md_*" where * is not all digits.
4896 * We allocate an array with a large free minor number, and
4897 * set the name to val. val must not already be an active name.
4898 */
4899 int len = strlen(val);
4900 char buf[DISK_NAME_LEN];
4901
4902 while (len && val[len-1] == '\n')
4903 len--;
4904 if (len >= DISK_NAME_LEN)
4905 return -E2BIG;
4906 strlcpy(buf, val, len+1);
4907 if (strncmp(buf, "md_", 3) != 0)
4908 return -EINVAL;
4909 return md_alloc(0, buf);
4910 }
4911
4912 static void md_safemode_timeout(unsigned long data)
4913 {
4914 struct mddev *mddev = (struct mddev *) data;
4915
4916 if (!atomic_read(&mddev->writes_pending)) {
4917 mddev->safemode = 1;
4918 if (mddev->external)
4919 sysfs_notify_dirent_safe(mddev->sysfs_state);
4920 }
4921 md_wakeup_thread(mddev->thread);
4922 }
4923
4924 static int start_dirty_degraded;
4925
4926 int md_run(struct mddev *mddev)
4927 {
4928 int err;
4929 struct md_rdev *rdev;
4930 struct md_personality *pers;
4931
4932 if (list_empty(&mddev->disks))
4933 /* cannot run an array with no devices.. */
4934 return -EINVAL;
4935
4936 if (mddev->pers)
4937 return -EBUSY;
4938 /* Cannot run until previous stop completes properly */
4939 if (mddev->sysfs_active)
4940 return -EBUSY;
4941
4942 /*
4943 * Analyze all RAID superblock(s)
4944 */
4945 if (!mddev->raid_disks) {
4946 if (!mddev->persistent)
4947 return -EINVAL;
4948 analyze_sbs(mddev);
4949 }
4950
4951 if (mddev->level != LEVEL_NONE)
4952 request_module("md-level-%d", mddev->level);
4953 else if (mddev->clevel[0])
4954 request_module("md-%s", mddev->clevel);
4955
4956 /*
4957 * Drop all container device buffers, from now on
4958 * the only valid external interface is through the md
4959 * device.
4960 */
4961 rdev_for_each(rdev, mddev) {
4962 if (test_bit(Faulty, &rdev->flags))
4963 continue;
4964 sync_blockdev(rdev->bdev);
4965 invalidate_bdev(rdev->bdev);
4966
4967 /* perform some consistency tests on the device.
4968 * We don't want the data to overlap the metadata,
4969 * Internal Bitmap issues have been handled elsewhere.
4970 */
4971 if (rdev->meta_bdev) {
4972 /* Nothing to check */;
4973 } else if (rdev->data_offset < rdev->sb_start) {
4974 if (mddev->dev_sectors &&
4975 rdev->data_offset + mddev->dev_sectors
4976 > rdev->sb_start) {
4977 printk("md: %s: data overlaps metadata\n",
4978 mdname(mddev));
4979 return -EINVAL;
4980 }
4981 } else {
4982 if (rdev->sb_start + rdev->sb_size/512
4983 > rdev->data_offset) {
4984 printk("md: %s: metadata overlaps data\n",
4985 mdname(mddev));
4986 return -EINVAL;
4987 }
4988 }
4989 sysfs_notify_dirent_safe(rdev->sysfs_state);
4990 }
4991
4992 if (mddev->bio_set == NULL)
4993 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
4994
4995 spin_lock(&pers_lock);
4996 pers = find_pers(mddev->level, mddev->clevel);
4997 if (!pers || !try_module_get(pers->owner)) {
4998 spin_unlock(&pers_lock);
4999 if (mddev->level != LEVEL_NONE)
5000 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5001 mddev->level);
5002 else
5003 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5004 mddev->clevel);
5005 return -EINVAL;
5006 }
5007 spin_unlock(&pers_lock);
5008 if (mddev->level != pers->level) {
5009 mddev->level = pers->level;
5010 mddev->new_level = pers->level;
5011 }
5012 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5013
5014 if (mddev->reshape_position != MaxSector &&
5015 pers->start_reshape == NULL) {
5016 /* This personality cannot handle reshaping... */
5017 module_put(pers->owner);
5018 return -EINVAL;
5019 }
5020
5021 if (pers->sync_request) {
5022 /* Warn if this is a potentially silly
5023 * configuration.
5024 */
5025 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5026 struct md_rdev *rdev2;
5027 int warned = 0;
5028
5029 rdev_for_each(rdev, mddev)
5030 rdev_for_each(rdev2, mddev) {
5031 if (rdev < rdev2 &&
5032 rdev->bdev->bd_contains ==
5033 rdev2->bdev->bd_contains) {
5034 printk(KERN_WARNING
5035 "%s: WARNING: %s appears to be"
5036 " on the same physical disk as"
5037 " %s.\n",
5038 mdname(mddev),
5039 bdevname(rdev->bdev,b),
5040 bdevname(rdev2->bdev,b2));
5041 warned = 1;
5042 }
5043 }
5044
5045 if (warned)
5046 printk(KERN_WARNING
5047 "True protection against single-disk"
5048 " failure might be compromised.\n");
5049 }
5050
5051 mddev->recovery = 0;
5052 /* may be over-ridden by personality */
5053 mddev->resync_max_sectors = mddev->dev_sectors;
5054
5055 mddev->ok_start_degraded = start_dirty_degraded;
5056
5057 if (start_readonly && mddev->ro == 0)
5058 mddev->ro = 2; /* read-only, but switch on first write */
5059
5060 err = pers->run(mddev);
5061 if (err)
5062 printk(KERN_ERR "md: pers->run() failed ...\n");
5063 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5064 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5065 " but 'external_size' not in effect?\n", __func__);
5066 printk(KERN_ERR
5067 "md: invalid array_size %llu > default size %llu\n",
5068 (unsigned long long)mddev->array_sectors / 2,
5069 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5070 err = -EINVAL;
5071 }
5072 if (err == 0 && pers->sync_request &&
5073 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5074 err = bitmap_create(mddev);
5075 if (err)
5076 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5077 mdname(mddev), err);
5078 }
5079 if (err) {
5080 mddev_detach(mddev);
5081 pers->free(mddev, mddev->private);
5082 module_put(pers->owner);
5083 bitmap_destroy(mddev);
5084 return err;
5085 }
5086 if (mddev->queue) {
5087 mddev->queue->backing_dev_info.congested_data = mddev;
5088 mddev->queue->backing_dev_info.congested_fn = md_congested;
5089 blk_queue_merge_bvec(mddev->queue, md_mergeable_bvec);
5090 }
5091 if (pers->sync_request) {
5092 if (mddev->kobj.sd &&
5093 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5094 printk(KERN_WARNING
5095 "md: cannot register extra attributes for %s\n",
5096 mdname(mddev));
5097 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5098 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5099 mddev->ro = 0;
5100
5101 atomic_set(&mddev->writes_pending,0);
5102 atomic_set(&mddev->max_corr_read_errors,
5103 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5104 mddev->safemode = 0;
5105 mddev->safemode_timer.function = md_safemode_timeout;
5106 mddev->safemode_timer.data = (unsigned long) mddev;
5107 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5108 mddev->in_sync = 1;
5109 smp_wmb();
5110 spin_lock(&mddev->lock);
5111 mddev->pers = pers;
5112 mddev->ready = 1;
5113 spin_unlock(&mddev->lock);
5114 rdev_for_each(rdev, mddev)
5115 if (rdev->raid_disk >= 0)
5116 if (sysfs_link_rdev(mddev, rdev))
5117 /* failure here is OK */;
5118
5119 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5120
5121 if (mddev->flags & MD_UPDATE_SB_FLAGS)
5122 md_update_sb(mddev, 0);
5123
5124 md_new_event(mddev);
5125 sysfs_notify_dirent_safe(mddev->sysfs_state);
5126 sysfs_notify_dirent_safe(mddev->sysfs_action);
5127 sysfs_notify(&mddev->kobj, NULL, "degraded");
5128 return 0;
5129 }
5130 EXPORT_SYMBOL_GPL(md_run);
5131
5132 static int do_md_run(struct mddev *mddev)
5133 {
5134 int err;
5135
5136 err = md_run(mddev);
5137 if (err)
5138 goto out;
5139 err = bitmap_load(mddev);
5140 if (err) {
5141 bitmap_destroy(mddev);
5142 goto out;
5143 }
5144
5145 md_wakeup_thread(mddev->thread);
5146 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5147
5148 set_capacity(mddev->gendisk, mddev->array_sectors);
5149 revalidate_disk(mddev->gendisk);
5150 mddev->changed = 1;
5151 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5152 out:
5153 return err;
5154 }
5155
5156 static int restart_array(struct mddev *mddev)
5157 {
5158 struct gendisk *disk = mddev->gendisk;
5159
5160 /* Complain if it has no devices */
5161 if (list_empty(&mddev->disks))
5162 return -ENXIO;
5163 if (!mddev->pers)
5164 return -EINVAL;
5165 if (!mddev->ro)
5166 return -EBUSY;
5167 mddev->safemode = 0;
5168 mddev->ro = 0;
5169 set_disk_ro(disk, 0);
5170 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5171 mdname(mddev));
5172 /* Kick recovery or resync if necessary */
5173 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5174 md_wakeup_thread(mddev->thread);
5175 md_wakeup_thread(mddev->sync_thread);
5176 sysfs_notify_dirent_safe(mddev->sysfs_state);
5177 return 0;
5178 }
5179
5180 static void md_clean(struct mddev *mddev)
5181 {
5182 mddev->array_sectors = 0;
5183 mddev->external_size = 0;
5184 mddev->dev_sectors = 0;
5185 mddev->raid_disks = 0;
5186 mddev->recovery_cp = 0;
5187 mddev->resync_min = 0;
5188 mddev->resync_max = MaxSector;
5189 mddev->reshape_position = MaxSector;
5190 mddev->external = 0;
5191 mddev->persistent = 0;
5192 mddev->level = LEVEL_NONE;
5193 mddev->clevel[0] = 0;
5194 mddev->flags = 0;
5195 mddev->ro = 0;
5196 mddev->metadata_type[0] = 0;
5197 mddev->chunk_sectors = 0;
5198 mddev->ctime = mddev->utime = 0;
5199 mddev->layout = 0;
5200 mddev->max_disks = 0;
5201 mddev->events = 0;
5202 mddev->can_decrease_events = 0;
5203 mddev->delta_disks = 0;
5204 mddev->reshape_backwards = 0;
5205 mddev->new_level = LEVEL_NONE;
5206 mddev->new_layout = 0;
5207 mddev->new_chunk_sectors = 0;
5208 mddev->curr_resync = 0;
5209 atomic64_set(&mddev->resync_mismatches, 0);
5210 mddev->suspend_lo = mddev->suspend_hi = 0;
5211 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5212 mddev->recovery = 0;
5213 mddev->in_sync = 0;
5214 mddev->changed = 0;
5215 mddev->degraded = 0;
5216 mddev->safemode = 0;
5217 mddev->merge_check_needed = 0;
5218 mddev->bitmap_info.offset = 0;
5219 mddev->bitmap_info.default_offset = 0;
5220 mddev->bitmap_info.default_space = 0;
5221 mddev->bitmap_info.chunksize = 0;
5222 mddev->bitmap_info.daemon_sleep = 0;
5223 mddev->bitmap_info.max_write_behind = 0;
5224 }
5225
5226 static void __md_stop_writes(struct mddev *mddev)
5227 {
5228 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5229 flush_workqueue(md_misc_wq);
5230 if (mddev->sync_thread) {
5231 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5232 md_reap_sync_thread(mddev);
5233 }
5234
5235 del_timer_sync(&mddev->safemode_timer);
5236
5237 bitmap_flush(mddev);
5238 md_super_wait(mddev);
5239
5240 if (mddev->ro == 0 &&
5241 (!mddev->in_sync || (mddev->flags & MD_UPDATE_SB_FLAGS))) {
5242 /* mark array as shutdown cleanly */
5243 mddev->in_sync = 1;
5244 md_update_sb(mddev, 1);
5245 }
5246 }
5247
5248 void md_stop_writes(struct mddev *mddev)
5249 {
5250 mddev_lock_nointr(mddev);
5251 __md_stop_writes(mddev);
5252 mddev_unlock(mddev);
5253 }
5254 EXPORT_SYMBOL_GPL(md_stop_writes);
5255
5256 static void mddev_detach(struct mddev *mddev)
5257 {
5258 struct bitmap *bitmap = mddev->bitmap;
5259 /* wait for behind writes to complete */
5260 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
5261 printk(KERN_INFO "md:%s: behind writes in progress - waiting to stop.\n",
5262 mdname(mddev));
5263 /* need to kick something here to make sure I/O goes? */
5264 wait_event(bitmap->behind_wait,
5265 atomic_read(&bitmap->behind_writes) == 0);
5266 }
5267 if (mddev->pers && mddev->pers->quiesce) {
5268 mddev->pers->quiesce(mddev, 1);
5269 mddev->pers->quiesce(mddev, 0);
5270 }
5271 md_unregister_thread(&mddev->thread);
5272 if (mddev->queue)
5273 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5274 }
5275
5276 static void __md_stop(struct mddev *mddev)
5277 {
5278 struct md_personality *pers = mddev->pers;
5279 mddev_detach(mddev);
5280 spin_lock(&mddev->lock);
5281 mddev->ready = 0;
5282 mddev->pers = NULL;
5283 spin_unlock(&mddev->lock);
5284 pers->free(mddev, mddev->private);
5285 if (pers->sync_request && mddev->to_remove == NULL)
5286 mddev->to_remove = &md_redundancy_group;
5287 module_put(pers->owner);
5288 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5289 }
5290
5291 void md_stop(struct mddev *mddev)
5292 {
5293 /* stop the array and free an attached data structures.
5294 * This is called from dm-raid
5295 */
5296 __md_stop(mddev);
5297 bitmap_destroy(mddev);
5298 if (mddev->bio_set)
5299 bioset_free(mddev->bio_set);
5300 }
5301
5302 EXPORT_SYMBOL_GPL(md_stop);
5303
5304 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5305 {
5306 int err = 0;
5307 int did_freeze = 0;
5308
5309 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5310 did_freeze = 1;
5311 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5312 md_wakeup_thread(mddev->thread);
5313 }
5314 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5315 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5316 if (mddev->sync_thread)
5317 /* Thread might be blocked waiting for metadata update
5318 * which will now never happen */
5319 wake_up_process(mddev->sync_thread->tsk);
5320
5321 mddev_unlock(mddev);
5322 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5323 &mddev->recovery));
5324 mddev_lock_nointr(mddev);
5325
5326 mutex_lock(&mddev->open_mutex);
5327 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5328 mddev->sync_thread ||
5329 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5330 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5331 printk("md: %s still in use.\n",mdname(mddev));
5332 if (did_freeze) {
5333 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5334 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5335 md_wakeup_thread(mddev->thread);
5336 }
5337 err = -EBUSY;
5338 goto out;
5339 }
5340 if (mddev->pers) {
5341 __md_stop_writes(mddev);
5342
5343 err = -ENXIO;
5344 if (mddev->ro==1)
5345 goto out;
5346 mddev->ro = 1;
5347 set_disk_ro(mddev->gendisk, 1);
5348 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5349 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5350 md_wakeup_thread(mddev->thread);
5351 sysfs_notify_dirent_safe(mddev->sysfs_state);
5352 err = 0;
5353 }
5354 out:
5355 mutex_unlock(&mddev->open_mutex);
5356 return err;
5357 }
5358
5359 /* mode:
5360 * 0 - completely stop and dis-assemble array
5361 * 2 - stop but do not disassemble array
5362 */
5363 static int do_md_stop(struct mddev *mddev, int mode,
5364 struct block_device *bdev)
5365 {
5366 struct gendisk *disk = mddev->gendisk;
5367 struct md_rdev *rdev;
5368 int did_freeze = 0;
5369
5370 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5371 did_freeze = 1;
5372 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5373 md_wakeup_thread(mddev->thread);
5374 }
5375 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5376 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5377 if (mddev->sync_thread)
5378 /* Thread might be blocked waiting for metadata update
5379 * which will now never happen */
5380 wake_up_process(mddev->sync_thread->tsk);
5381
5382 mddev_unlock(mddev);
5383 wait_event(resync_wait, (mddev->sync_thread == NULL &&
5384 !test_bit(MD_RECOVERY_RUNNING,
5385 &mddev->recovery)));
5386 mddev_lock_nointr(mddev);
5387
5388 mutex_lock(&mddev->open_mutex);
5389 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5390 mddev->sysfs_active ||
5391 mddev->sync_thread ||
5392 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5393 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5394 printk("md: %s still in use.\n",mdname(mddev));
5395 mutex_unlock(&mddev->open_mutex);
5396 if (did_freeze) {
5397 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5398 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5399 md_wakeup_thread(mddev->thread);
5400 }
5401 return -EBUSY;
5402 }
5403 if (mddev->pers) {
5404 if (mddev->ro)
5405 set_disk_ro(disk, 0);
5406
5407 __md_stop_writes(mddev);
5408 __md_stop(mddev);
5409 mddev->queue->merge_bvec_fn = NULL;
5410 mddev->queue->backing_dev_info.congested_fn = NULL;
5411
5412 /* tell userspace to handle 'inactive' */
5413 sysfs_notify_dirent_safe(mddev->sysfs_state);
5414
5415 rdev_for_each(rdev, mddev)
5416 if (rdev->raid_disk >= 0)
5417 sysfs_unlink_rdev(mddev, rdev);
5418
5419 set_capacity(disk, 0);
5420 mutex_unlock(&mddev->open_mutex);
5421 mddev->changed = 1;
5422 revalidate_disk(disk);
5423
5424 if (mddev->ro)
5425 mddev->ro = 0;
5426 } else
5427 mutex_unlock(&mddev->open_mutex);
5428 /*
5429 * Free resources if final stop
5430 */
5431 if (mode == 0) {
5432 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5433
5434 bitmap_destroy(mddev);
5435 if (mddev->bitmap_info.file) {
5436 struct file *f = mddev->bitmap_info.file;
5437 spin_lock(&mddev->lock);
5438 mddev->bitmap_info.file = NULL;
5439 spin_unlock(&mddev->lock);
5440 fput(f);
5441 }
5442 mddev->bitmap_info.offset = 0;
5443
5444 export_array(mddev);
5445
5446 md_clean(mddev);
5447 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5448 if (mddev->hold_active == UNTIL_STOP)
5449 mddev->hold_active = 0;
5450 }
5451 blk_integrity_unregister(disk);
5452 md_new_event(mddev);
5453 sysfs_notify_dirent_safe(mddev->sysfs_state);
5454 return 0;
5455 }
5456
5457 #ifndef MODULE
5458 static void autorun_array(struct mddev *mddev)
5459 {
5460 struct md_rdev *rdev;
5461 int err;
5462
5463 if (list_empty(&mddev->disks))
5464 return;
5465
5466 printk(KERN_INFO "md: running: ");
5467
5468 rdev_for_each(rdev, mddev) {
5469 char b[BDEVNAME_SIZE];
5470 printk("<%s>", bdevname(rdev->bdev,b));
5471 }
5472 printk("\n");
5473
5474 err = do_md_run(mddev);
5475 if (err) {
5476 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
5477 do_md_stop(mddev, 0, NULL);
5478 }
5479 }
5480
5481 /*
5482 * lets try to run arrays based on all disks that have arrived
5483 * until now. (those are in pending_raid_disks)
5484 *
5485 * the method: pick the first pending disk, collect all disks with
5486 * the same UUID, remove all from the pending list and put them into
5487 * the 'same_array' list. Then order this list based on superblock
5488 * update time (freshest comes first), kick out 'old' disks and
5489 * compare superblocks. If everything's fine then run it.
5490 *
5491 * If "unit" is allocated, then bump its reference count
5492 */
5493 static void autorun_devices(int part)
5494 {
5495 struct md_rdev *rdev0, *rdev, *tmp;
5496 struct mddev *mddev;
5497 char b[BDEVNAME_SIZE];
5498
5499 printk(KERN_INFO "md: autorun ...\n");
5500 while (!list_empty(&pending_raid_disks)) {
5501 int unit;
5502 dev_t dev;
5503 LIST_HEAD(candidates);
5504 rdev0 = list_entry(pending_raid_disks.next,
5505 struct md_rdev, same_set);
5506
5507 printk(KERN_INFO "md: considering %s ...\n",
5508 bdevname(rdev0->bdev,b));
5509 INIT_LIST_HEAD(&candidates);
5510 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5511 if (super_90_load(rdev, rdev0, 0) >= 0) {
5512 printk(KERN_INFO "md: adding %s ...\n",
5513 bdevname(rdev->bdev,b));
5514 list_move(&rdev->same_set, &candidates);
5515 }
5516 /*
5517 * now we have a set of devices, with all of them having
5518 * mostly sane superblocks. It's time to allocate the
5519 * mddev.
5520 */
5521 if (part) {
5522 dev = MKDEV(mdp_major,
5523 rdev0->preferred_minor << MdpMinorShift);
5524 unit = MINOR(dev) >> MdpMinorShift;
5525 } else {
5526 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5527 unit = MINOR(dev);
5528 }
5529 if (rdev0->preferred_minor != unit) {
5530 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5531 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5532 break;
5533 }
5534
5535 md_probe(dev, NULL, NULL);
5536 mddev = mddev_find(dev);
5537 if (!mddev || !mddev->gendisk) {
5538 if (mddev)
5539 mddev_put(mddev);
5540 printk(KERN_ERR
5541 "md: cannot allocate memory for md drive.\n");
5542 break;
5543 }
5544 if (mddev_lock(mddev))
5545 printk(KERN_WARNING "md: %s locked, cannot run\n",
5546 mdname(mddev));
5547 else if (mddev->raid_disks || mddev->major_version
5548 || !list_empty(&mddev->disks)) {
5549 printk(KERN_WARNING
5550 "md: %s already running, cannot run %s\n",
5551 mdname(mddev), bdevname(rdev0->bdev,b));
5552 mddev_unlock(mddev);
5553 } else {
5554 printk(KERN_INFO "md: created %s\n", mdname(mddev));
5555 mddev->persistent = 1;
5556 rdev_for_each_list(rdev, tmp, &candidates) {
5557 list_del_init(&rdev->same_set);
5558 if (bind_rdev_to_array(rdev, mddev))
5559 export_rdev(rdev);
5560 }
5561 autorun_array(mddev);
5562 mddev_unlock(mddev);
5563 }
5564 /* on success, candidates will be empty, on error
5565 * it won't...
5566 */
5567 rdev_for_each_list(rdev, tmp, &candidates) {
5568 list_del_init(&rdev->same_set);
5569 export_rdev(rdev);
5570 }
5571 mddev_put(mddev);
5572 }
5573 printk(KERN_INFO "md: ... autorun DONE.\n");
5574 }
5575 #endif /* !MODULE */
5576
5577 static int get_version(void __user *arg)
5578 {
5579 mdu_version_t ver;
5580
5581 ver.major = MD_MAJOR_VERSION;
5582 ver.minor = MD_MINOR_VERSION;
5583 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5584
5585 if (copy_to_user(arg, &ver, sizeof(ver)))
5586 return -EFAULT;
5587
5588 return 0;
5589 }
5590
5591 static int get_array_info(struct mddev *mddev, void __user *arg)
5592 {
5593 mdu_array_info_t info;
5594 int nr,working,insync,failed,spare;
5595 struct md_rdev *rdev;
5596
5597 nr = working = insync = failed = spare = 0;
5598 rcu_read_lock();
5599 rdev_for_each_rcu(rdev, mddev) {
5600 nr++;
5601 if (test_bit(Faulty, &rdev->flags))
5602 failed++;
5603 else {
5604 working++;
5605 if (test_bit(In_sync, &rdev->flags))
5606 insync++;
5607 else
5608 spare++;
5609 }
5610 }
5611 rcu_read_unlock();
5612
5613 info.major_version = mddev->major_version;
5614 info.minor_version = mddev->minor_version;
5615 info.patch_version = MD_PATCHLEVEL_VERSION;
5616 info.ctime = mddev->ctime;
5617 info.level = mddev->level;
5618 info.size = mddev->dev_sectors / 2;
5619 if (info.size != mddev->dev_sectors / 2) /* overflow */
5620 info.size = -1;
5621 info.nr_disks = nr;
5622 info.raid_disks = mddev->raid_disks;
5623 info.md_minor = mddev->md_minor;
5624 info.not_persistent= !mddev->persistent;
5625
5626 info.utime = mddev->utime;
5627 info.state = 0;
5628 if (mddev->in_sync)
5629 info.state = (1<<MD_SB_CLEAN);
5630 if (mddev->bitmap && mddev->bitmap_info.offset)
5631 info.state |= (1<<MD_SB_BITMAP_PRESENT);
5632 info.active_disks = insync;
5633 info.working_disks = working;
5634 info.failed_disks = failed;
5635 info.spare_disks = spare;
5636
5637 info.layout = mddev->layout;
5638 info.chunk_size = mddev->chunk_sectors << 9;
5639
5640 if (copy_to_user(arg, &info, sizeof(info)))
5641 return -EFAULT;
5642
5643 return 0;
5644 }
5645
5646 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
5647 {
5648 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5649 char *ptr;
5650 int err;
5651
5652 file = kmalloc(sizeof(*file), GFP_NOIO);
5653 if (!file)
5654 return -ENOMEM;
5655
5656 err = 0;
5657 spin_lock(&mddev->lock);
5658 /* bitmap disabled, zero the first byte and copy out */
5659 if (!mddev->bitmap_info.file)
5660 file->pathname[0] = '\0';
5661 else if ((ptr = d_path(&mddev->bitmap_info.file->f_path,
5662 file->pathname, sizeof(file->pathname))),
5663 IS_ERR(ptr))
5664 err = PTR_ERR(ptr);
5665 else
5666 memmove(file->pathname, ptr,
5667 sizeof(file->pathname)-(ptr-file->pathname));
5668 spin_unlock(&mddev->lock);
5669
5670 if (err == 0 &&
5671 copy_to_user(arg, file, sizeof(*file)))
5672 err = -EFAULT;
5673
5674 kfree(file);
5675 return err;
5676 }
5677
5678 static int get_disk_info(struct mddev *mddev, void __user * arg)
5679 {
5680 mdu_disk_info_t info;
5681 struct md_rdev *rdev;
5682
5683 if (copy_from_user(&info, arg, sizeof(info)))
5684 return -EFAULT;
5685
5686 rcu_read_lock();
5687 rdev = find_rdev_nr_rcu(mddev, info.number);
5688 if (rdev) {
5689 info.major = MAJOR(rdev->bdev->bd_dev);
5690 info.minor = MINOR(rdev->bdev->bd_dev);
5691 info.raid_disk = rdev->raid_disk;
5692 info.state = 0;
5693 if (test_bit(Faulty, &rdev->flags))
5694 info.state |= (1<<MD_DISK_FAULTY);
5695 else if (test_bit(In_sync, &rdev->flags)) {
5696 info.state |= (1<<MD_DISK_ACTIVE);
5697 info.state |= (1<<MD_DISK_SYNC);
5698 }
5699 if (test_bit(WriteMostly, &rdev->flags))
5700 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5701 } else {
5702 info.major = info.minor = 0;
5703 info.raid_disk = -1;
5704 info.state = (1<<MD_DISK_REMOVED);
5705 }
5706 rcu_read_unlock();
5707
5708 if (copy_to_user(arg, &info, sizeof(info)))
5709 return -EFAULT;
5710
5711 return 0;
5712 }
5713
5714 static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
5715 {
5716 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5717 struct md_rdev *rdev;
5718 dev_t dev = MKDEV(info->major,info->minor);
5719
5720 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5721 return -EOVERFLOW;
5722
5723 if (!mddev->raid_disks) {
5724 int err;
5725 /* expecting a device which has a superblock */
5726 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5727 if (IS_ERR(rdev)) {
5728 printk(KERN_WARNING
5729 "md: md_import_device returned %ld\n",
5730 PTR_ERR(rdev));
5731 return PTR_ERR(rdev);
5732 }
5733 if (!list_empty(&mddev->disks)) {
5734 struct md_rdev *rdev0
5735 = list_entry(mddev->disks.next,
5736 struct md_rdev, same_set);
5737 err = super_types[mddev->major_version]
5738 .load_super(rdev, rdev0, mddev->minor_version);
5739 if (err < 0) {
5740 printk(KERN_WARNING
5741 "md: %s has different UUID to %s\n",
5742 bdevname(rdev->bdev,b),
5743 bdevname(rdev0->bdev,b2));
5744 export_rdev(rdev);
5745 return -EINVAL;
5746 }
5747 }
5748 err = bind_rdev_to_array(rdev, mddev);
5749 if (err)
5750 export_rdev(rdev);
5751 return err;
5752 }
5753
5754 /*
5755 * add_new_disk can be used once the array is assembled
5756 * to add "hot spares". They must already have a superblock
5757 * written
5758 */
5759 if (mddev->pers) {
5760 int err;
5761 if (!mddev->pers->hot_add_disk) {
5762 printk(KERN_WARNING
5763 "%s: personality does not support diskops!\n",
5764 mdname(mddev));
5765 return -EINVAL;
5766 }
5767 if (mddev->persistent)
5768 rdev = md_import_device(dev, mddev->major_version,
5769 mddev->minor_version);
5770 else
5771 rdev = md_import_device(dev, -1, -1);
5772 if (IS_ERR(rdev)) {
5773 printk(KERN_WARNING
5774 "md: md_import_device returned %ld\n",
5775 PTR_ERR(rdev));
5776 return PTR_ERR(rdev);
5777 }
5778 /* set saved_raid_disk if appropriate */
5779 if (!mddev->persistent) {
5780 if (info->state & (1<<MD_DISK_SYNC) &&
5781 info->raid_disk < mddev->raid_disks) {
5782 rdev->raid_disk = info->raid_disk;
5783 set_bit(In_sync, &rdev->flags);
5784 clear_bit(Bitmap_sync, &rdev->flags);
5785 } else
5786 rdev->raid_disk = -1;
5787 rdev->saved_raid_disk = rdev->raid_disk;
5788 } else
5789 super_types[mddev->major_version].
5790 validate_super(mddev, rdev);
5791 if ((info->state & (1<<MD_DISK_SYNC)) &&
5792 rdev->raid_disk != info->raid_disk) {
5793 /* This was a hot-add request, but events doesn't
5794 * match, so reject it.
5795 */
5796 export_rdev(rdev);
5797 return -EINVAL;
5798 }
5799
5800 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5801 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5802 set_bit(WriteMostly, &rdev->flags);
5803 else
5804 clear_bit(WriteMostly, &rdev->flags);
5805
5806 rdev->raid_disk = -1;
5807 err = bind_rdev_to_array(rdev, mddev);
5808 if (!err && !mddev->pers->hot_remove_disk) {
5809 /* If there is hot_add_disk but no hot_remove_disk
5810 * then added disks for geometry changes,
5811 * and should be added immediately.
5812 */
5813 super_types[mddev->major_version].
5814 validate_super(mddev, rdev);
5815 err = mddev->pers->hot_add_disk(mddev, rdev);
5816 if (err)
5817 unbind_rdev_from_array(rdev);
5818 }
5819 if (err)
5820 export_rdev(rdev);
5821 else
5822 sysfs_notify_dirent_safe(rdev->sysfs_state);
5823
5824 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5825 if (mddev->degraded)
5826 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5827 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5828 if (!err)
5829 md_new_event(mddev);
5830 md_wakeup_thread(mddev->thread);
5831 return err;
5832 }
5833
5834 /* otherwise, add_new_disk is only allowed
5835 * for major_version==0 superblocks
5836 */
5837 if (mddev->major_version != 0) {
5838 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5839 mdname(mddev));
5840 return -EINVAL;
5841 }
5842
5843 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5844 int err;
5845 rdev = md_import_device(dev, -1, 0);
5846 if (IS_ERR(rdev)) {
5847 printk(KERN_WARNING
5848 "md: error, md_import_device() returned %ld\n",
5849 PTR_ERR(rdev));
5850 return PTR_ERR(rdev);
5851 }
5852 rdev->desc_nr = info->number;
5853 if (info->raid_disk < mddev->raid_disks)
5854 rdev->raid_disk = info->raid_disk;
5855 else
5856 rdev->raid_disk = -1;
5857
5858 if (rdev->raid_disk < mddev->raid_disks)
5859 if (info->state & (1<<MD_DISK_SYNC))
5860 set_bit(In_sync, &rdev->flags);
5861
5862 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5863 set_bit(WriteMostly, &rdev->flags);
5864
5865 if (!mddev->persistent) {
5866 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5867 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5868 } else
5869 rdev->sb_start = calc_dev_sboffset(rdev);
5870 rdev->sectors = rdev->sb_start;
5871
5872 err = bind_rdev_to_array(rdev, mddev);
5873 if (err) {
5874 export_rdev(rdev);
5875 return err;
5876 }
5877 }
5878
5879 return 0;
5880 }
5881
5882 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
5883 {
5884 char b[BDEVNAME_SIZE];
5885 struct md_rdev *rdev;
5886
5887 rdev = find_rdev(mddev, dev);
5888 if (!rdev)
5889 return -ENXIO;
5890
5891 clear_bit(Blocked, &rdev->flags);
5892 remove_and_add_spares(mddev, rdev);
5893
5894 if (rdev->raid_disk >= 0)
5895 goto busy;
5896
5897 kick_rdev_from_array(rdev);
5898 md_update_sb(mddev, 1);
5899 md_new_event(mddev);
5900
5901 return 0;
5902 busy:
5903 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5904 bdevname(rdev->bdev,b), mdname(mddev));
5905 return -EBUSY;
5906 }
5907
5908 static int hot_add_disk(struct mddev *mddev, dev_t dev)
5909 {
5910 char b[BDEVNAME_SIZE];
5911 int err;
5912 struct md_rdev *rdev;
5913
5914 if (!mddev->pers)
5915 return -ENODEV;
5916
5917 if (mddev->major_version != 0) {
5918 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5919 " version-0 superblocks.\n",
5920 mdname(mddev));
5921 return -EINVAL;
5922 }
5923 if (!mddev->pers->hot_add_disk) {
5924 printk(KERN_WARNING
5925 "%s: personality does not support diskops!\n",
5926 mdname(mddev));
5927 return -EINVAL;
5928 }
5929
5930 rdev = md_import_device(dev, -1, 0);
5931 if (IS_ERR(rdev)) {
5932 printk(KERN_WARNING
5933 "md: error, md_import_device() returned %ld\n",
5934 PTR_ERR(rdev));
5935 return -EINVAL;
5936 }
5937
5938 if (mddev->persistent)
5939 rdev->sb_start = calc_dev_sboffset(rdev);
5940 else
5941 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5942
5943 rdev->sectors = rdev->sb_start;
5944
5945 if (test_bit(Faulty, &rdev->flags)) {
5946 printk(KERN_WARNING
5947 "md: can not hot-add faulty %s disk to %s!\n",
5948 bdevname(rdev->bdev,b), mdname(mddev));
5949 err = -EINVAL;
5950 goto abort_export;
5951 }
5952 clear_bit(In_sync, &rdev->flags);
5953 rdev->desc_nr = -1;
5954 rdev->saved_raid_disk = -1;
5955 err = bind_rdev_to_array(rdev, mddev);
5956 if (err)
5957 goto abort_export;
5958
5959 /*
5960 * The rest should better be atomic, we can have disk failures
5961 * noticed in interrupt contexts ...
5962 */
5963
5964 rdev->raid_disk = -1;
5965
5966 md_update_sb(mddev, 1);
5967
5968 /*
5969 * Kick recovery, maybe this spare has to be added to the
5970 * array immediately.
5971 */
5972 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5973 md_wakeup_thread(mddev->thread);
5974 md_new_event(mddev);
5975 return 0;
5976
5977 abort_export:
5978 export_rdev(rdev);
5979 return err;
5980 }
5981
5982 static int set_bitmap_file(struct mddev *mddev, int fd)
5983 {
5984 int err = 0;
5985
5986 if (mddev->pers) {
5987 if (!mddev->pers->quiesce || !mddev->thread)
5988 return -EBUSY;
5989 if (mddev->recovery || mddev->sync_thread)
5990 return -EBUSY;
5991 /* we should be able to change the bitmap.. */
5992 }
5993
5994 if (fd >= 0) {
5995 struct inode *inode;
5996 struct file *f;
5997
5998 if (mddev->bitmap || mddev->bitmap_info.file)
5999 return -EEXIST; /* cannot add when bitmap is present */
6000 f = fget(fd);
6001
6002 if (f == NULL) {
6003 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
6004 mdname(mddev));
6005 return -EBADF;
6006 }
6007
6008 inode = f->f_mapping->host;
6009 if (!S_ISREG(inode->i_mode)) {
6010 printk(KERN_ERR "%s: error: bitmap file must be a regular file\n",
6011 mdname(mddev));
6012 err = -EBADF;
6013 } else if (!(f->f_mode & FMODE_WRITE)) {
6014 printk(KERN_ERR "%s: error: bitmap file must open for write\n",
6015 mdname(mddev));
6016 err = -EBADF;
6017 } else if (atomic_read(&inode->i_writecount) != 1) {
6018 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
6019 mdname(mddev));
6020 err = -EBUSY;
6021 }
6022 if (err) {
6023 fput(f);
6024 return err;
6025 }
6026 mddev->bitmap_info.file = f;
6027 mddev->bitmap_info.offset = 0; /* file overrides offset */
6028 } else if (mddev->bitmap == NULL)
6029 return -ENOENT; /* cannot remove what isn't there */
6030 err = 0;
6031 if (mddev->pers) {
6032 mddev->pers->quiesce(mddev, 1);
6033 if (fd >= 0) {
6034 err = bitmap_create(mddev);
6035 if (!err)
6036 err = bitmap_load(mddev);
6037 }
6038 if (fd < 0 || err) {
6039 bitmap_destroy(mddev);
6040 fd = -1; /* make sure to put the file */
6041 }
6042 mddev->pers->quiesce(mddev, 0);
6043 }
6044 if (fd < 0) {
6045 struct file *f = mddev->bitmap_info.file;
6046 if (f) {
6047 spin_lock(&mddev->lock);
6048 mddev->bitmap_info.file = NULL;
6049 spin_unlock(&mddev->lock);
6050 fput(f);
6051 }
6052 }
6053
6054 return err;
6055 }
6056
6057 /*
6058 * set_array_info is used two different ways
6059 * The original usage is when creating a new array.
6060 * In this usage, raid_disks is > 0 and it together with
6061 * level, size, not_persistent,layout,chunksize determine the
6062 * shape of the array.
6063 * This will always create an array with a type-0.90.0 superblock.
6064 * The newer usage is when assembling an array.
6065 * In this case raid_disks will be 0, and the major_version field is
6066 * use to determine which style super-blocks are to be found on the devices.
6067 * The minor and patch _version numbers are also kept incase the
6068 * super_block handler wishes to interpret them.
6069 */
6070 static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
6071 {
6072
6073 if (info->raid_disks == 0) {
6074 /* just setting version number for superblock loading */
6075 if (info->major_version < 0 ||
6076 info->major_version >= ARRAY_SIZE(super_types) ||
6077 super_types[info->major_version].name == NULL) {
6078 /* maybe try to auto-load a module? */
6079 printk(KERN_INFO
6080 "md: superblock version %d not known\n",
6081 info->major_version);
6082 return -EINVAL;
6083 }
6084 mddev->major_version = info->major_version;
6085 mddev->minor_version = info->minor_version;
6086 mddev->patch_version = info->patch_version;
6087 mddev->persistent = !info->not_persistent;
6088 /* ensure mddev_put doesn't delete this now that there
6089 * is some minimal configuration.
6090 */
6091 mddev->ctime = get_seconds();
6092 return 0;
6093 }
6094 mddev->major_version = MD_MAJOR_VERSION;
6095 mddev->minor_version = MD_MINOR_VERSION;
6096 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6097 mddev->ctime = get_seconds();
6098
6099 mddev->level = info->level;
6100 mddev->clevel[0] = 0;
6101 mddev->dev_sectors = 2 * (sector_t)info->size;
6102 mddev->raid_disks = info->raid_disks;
6103 /* don't set md_minor, it is determined by which /dev/md* was
6104 * openned
6105 */
6106 if (info->state & (1<<MD_SB_CLEAN))
6107 mddev->recovery_cp = MaxSector;
6108 else
6109 mddev->recovery_cp = 0;
6110 mddev->persistent = ! info->not_persistent;
6111 mddev->external = 0;
6112
6113 mddev->layout = info->layout;
6114 mddev->chunk_sectors = info->chunk_size >> 9;
6115
6116 mddev->max_disks = MD_SB_DISKS;
6117
6118 if (mddev->persistent)
6119 mddev->flags = 0;
6120 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6121
6122 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6123 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6124 mddev->bitmap_info.offset = 0;
6125
6126 mddev->reshape_position = MaxSector;
6127
6128 /*
6129 * Generate a 128 bit UUID
6130 */
6131 get_random_bytes(mddev->uuid, 16);
6132
6133 mddev->new_level = mddev->level;
6134 mddev->new_chunk_sectors = mddev->chunk_sectors;
6135 mddev->new_layout = mddev->layout;
6136 mddev->delta_disks = 0;
6137 mddev->reshape_backwards = 0;
6138
6139 return 0;
6140 }
6141
6142 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6143 {
6144 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6145
6146 if (mddev->external_size)
6147 return;
6148
6149 mddev->array_sectors = array_sectors;
6150 }
6151 EXPORT_SYMBOL(md_set_array_sectors);
6152
6153 static int update_size(struct mddev *mddev, sector_t num_sectors)
6154 {
6155 struct md_rdev *rdev;
6156 int rv;
6157 int fit = (num_sectors == 0);
6158
6159 if (mddev->pers->resize == NULL)
6160 return -EINVAL;
6161 /* The "num_sectors" is the number of sectors of each device that
6162 * is used. This can only make sense for arrays with redundancy.
6163 * linear and raid0 always use whatever space is available. We can only
6164 * consider changing this number if no resync or reconstruction is
6165 * happening, and if the new size is acceptable. It must fit before the
6166 * sb_start or, if that is <data_offset, it must fit before the size
6167 * of each device. If num_sectors is zero, we find the largest size
6168 * that fits.
6169 */
6170 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6171 mddev->sync_thread)
6172 return -EBUSY;
6173 if (mddev->ro)
6174 return -EROFS;
6175
6176 rdev_for_each(rdev, mddev) {
6177 sector_t avail = rdev->sectors;
6178
6179 if (fit && (num_sectors == 0 || num_sectors > avail))
6180 num_sectors = avail;
6181 if (avail < num_sectors)
6182 return -ENOSPC;
6183 }
6184 rv = mddev->pers->resize(mddev, num_sectors);
6185 if (!rv)
6186 revalidate_disk(mddev->gendisk);
6187 return rv;
6188 }
6189
6190 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6191 {
6192 int rv;
6193 struct md_rdev *rdev;
6194 /* change the number of raid disks */
6195 if (mddev->pers->check_reshape == NULL)
6196 return -EINVAL;
6197 if (mddev->ro)
6198 return -EROFS;
6199 if (raid_disks <= 0 ||
6200 (mddev->max_disks && raid_disks >= mddev->max_disks))
6201 return -EINVAL;
6202 if (mddev->sync_thread ||
6203 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6204 mddev->reshape_position != MaxSector)
6205 return -EBUSY;
6206
6207 rdev_for_each(rdev, mddev) {
6208 if (mddev->raid_disks < raid_disks &&
6209 rdev->data_offset < rdev->new_data_offset)
6210 return -EINVAL;
6211 if (mddev->raid_disks > raid_disks &&
6212 rdev->data_offset > rdev->new_data_offset)
6213 return -EINVAL;
6214 }
6215
6216 mddev->delta_disks = raid_disks - mddev->raid_disks;
6217 if (mddev->delta_disks < 0)
6218 mddev->reshape_backwards = 1;
6219 else if (mddev->delta_disks > 0)
6220 mddev->reshape_backwards = 0;
6221
6222 rv = mddev->pers->check_reshape(mddev);
6223 if (rv < 0) {
6224 mddev->delta_disks = 0;
6225 mddev->reshape_backwards = 0;
6226 }
6227 return rv;
6228 }
6229
6230 /*
6231 * update_array_info is used to change the configuration of an
6232 * on-line array.
6233 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6234 * fields in the info are checked against the array.
6235 * Any differences that cannot be handled will cause an error.
6236 * Normally, only one change can be managed at a time.
6237 */
6238 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6239 {
6240 int rv = 0;
6241 int cnt = 0;
6242 int state = 0;
6243
6244 /* calculate expected state,ignoring low bits */
6245 if (mddev->bitmap && mddev->bitmap_info.offset)
6246 state |= (1 << MD_SB_BITMAP_PRESENT);
6247
6248 if (mddev->major_version != info->major_version ||
6249 mddev->minor_version != info->minor_version ||
6250 /* mddev->patch_version != info->patch_version || */
6251 mddev->ctime != info->ctime ||
6252 mddev->level != info->level ||
6253 /* mddev->layout != info->layout || */
6254 !mddev->persistent != info->not_persistent||
6255 mddev->chunk_sectors != info->chunk_size >> 9 ||
6256 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6257 ((state^info->state) & 0xfffffe00)
6258 )
6259 return -EINVAL;
6260 /* Check there is only one change */
6261 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6262 cnt++;
6263 if (mddev->raid_disks != info->raid_disks)
6264 cnt++;
6265 if (mddev->layout != info->layout)
6266 cnt++;
6267 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6268 cnt++;
6269 if (cnt == 0)
6270 return 0;
6271 if (cnt > 1)
6272 return -EINVAL;
6273
6274 if (mddev->layout != info->layout) {
6275 /* Change layout
6276 * we don't need to do anything at the md level, the
6277 * personality will take care of it all.
6278 */
6279 if (mddev->pers->check_reshape == NULL)
6280 return -EINVAL;
6281 else {
6282 mddev->new_layout = info->layout;
6283 rv = mddev->pers->check_reshape(mddev);
6284 if (rv)
6285 mddev->new_layout = mddev->layout;
6286 return rv;
6287 }
6288 }
6289 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6290 rv = update_size(mddev, (sector_t)info->size * 2);
6291
6292 if (mddev->raid_disks != info->raid_disks)
6293 rv = update_raid_disks(mddev, info->raid_disks);
6294
6295 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6296 if (mddev->pers->quiesce == NULL || mddev->thread == NULL)
6297 return -EINVAL;
6298 if (mddev->recovery || mddev->sync_thread)
6299 return -EBUSY;
6300 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6301 /* add the bitmap */
6302 if (mddev->bitmap)
6303 return -EEXIST;
6304 if (mddev->bitmap_info.default_offset == 0)
6305 return -EINVAL;
6306 mddev->bitmap_info.offset =
6307 mddev->bitmap_info.default_offset;
6308 mddev->bitmap_info.space =
6309 mddev->bitmap_info.default_space;
6310 mddev->pers->quiesce(mddev, 1);
6311 rv = bitmap_create(mddev);
6312 if (!rv)
6313 rv = bitmap_load(mddev);
6314 if (rv)
6315 bitmap_destroy(mddev);
6316 mddev->pers->quiesce(mddev, 0);
6317 } else {
6318 /* remove the bitmap */
6319 if (!mddev->bitmap)
6320 return -ENOENT;
6321 if (mddev->bitmap->storage.file)
6322 return -EINVAL;
6323 mddev->pers->quiesce(mddev, 1);
6324 bitmap_destroy(mddev);
6325 mddev->pers->quiesce(mddev, 0);
6326 mddev->bitmap_info.offset = 0;
6327 }
6328 }
6329 md_update_sb(mddev, 1);
6330 return rv;
6331 }
6332
6333 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6334 {
6335 struct md_rdev *rdev;
6336 int err = 0;
6337
6338 if (mddev->pers == NULL)
6339 return -ENODEV;
6340
6341 rcu_read_lock();
6342 rdev = find_rdev_rcu(mddev, dev);
6343 if (!rdev)
6344 err = -ENODEV;
6345 else {
6346 md_error(mddev, rdev);
6347 if (!test_bit(Faulty, &rdev->flags))
6348 err = -EBUSY;
6349 }
6350 rcu_read_unlock();
6351 return err;
6352 }
6353
6354 /*
6355 * We have a problem here : there is no easy way to give a CHS
6356 * virtual geometry. We currently pretend that we have a 2 heads
6357 * 4 sectors (with a BIG number of cylinders...). This drives
6358 * dosfs just mad... ;-)
6359 */
6360 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6361 {
6362 struct mddev *mddev = bdev->bd_disk->private_data;
6363
6364 geo->heads = 2;
6365 geo->sectors = 4;
6366 geo->cylinders = mddev->array_sectors / 8;
6367 return 0;
6368 }
6369
6370 static inline bool md_ioctl_valid(unsigned int cmd)
6371 {
6372 switch (cmd) {
6373 case ADD_NEW_DISK:
6374 case BLKROSET:
6375 case GET_ARRAY_INFO:
6376 case GET_BITMAP_FILE:
6377 case GET_DISK_INFO:
6378 case HOT_ADD_DISK:
6379 case HOT_REMOVE_DISK:
6380 case RAID_AUTORUN:
6381 case RAID_VERSION:
6382 case RESTART_ARRAY_RW:
6383 case RUN_ARRAY:
6384 case SET_ARRAY_INFO:
6385 case SET_BITMAP_FILE:
6386 case SET_DISK_FAULTY:
6387 case STOP_ARRAY:
6388 case STOP_ARRAY_RO:
6389 return true;
6390 default:
6391 return false;
6392 }
6393 }
6394
6395 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6396 unsigned int cmd, unsigned long arg)
6397 {
6398 int err = 0;
6399 void __user *argp = (void __user *)arg;
6400 struct mddev *mddev = NULL;
6401 int ro;
6402
6403 if (!md_ioctl_valid(cmd))
6404 return -ENOTTY;
6405
6406 switch (cmd) {
6407 case RAID_VERSION:
6408 case GET_ARRAY_INFO:
6409 case GET_DISK_INFO:
6410 break;
6411 default:
6412 if (!capable(CAP_SYS_ADMIN))
6413 return -EACCES;
6414 }
6415
6416 /*
6417 * Commands dealing with the RAID driver but not any
6418 * particular array:
6419 */
6420 switch (cmd) {
6421 case RAID_VERSION:
6422 err = get_version(argp);
6423 goto out;
6424
6425 #ifndef MODULE
6426 case RAID_AUTORUN:
6427 err = 0;
6428 autostart_arrays(arg);
6429 goto out;
6430 #endif
6431 default:;
6432 }
6433
6434 /*
6435 * Commands creating/starting a new array:
6436 */
6437
6438 mddev = bdev->bd_disk->private_data;
6439
6440 if (!mddev) {
6441 BUG();
6442 goto out;
6443 }
6444
6445 /* Some actions do not requires the mutex */
6446 switch (cmd) {
6447 case GET_ARRAY_INFO:
6448 if (!mddev->raid_disks && !mddev->external)
6449 err = -ENODEV;
6450 else
6451 err = get_array_info(mddev, argp);
6452 goto out;
6453
6454 case GET_DISK_INFO:
6455 if (!mddev->raid_disks && !mddev->external)
6456 err = -ENODEV;
6457 else
6458 err = get_disk_info(mddev, argp);
6459 goto out;
6460
6461 case SET_DISK_FAULTY:
6462 err = set_disk_faulty(mddev, new_decode_dev(arg));
6463 goto out;
6464
6465 case GET_BITMAP_FILE:
6466 err = get_bitmap_file(mddev, argp);
6467 goto out;
6468
6469 }
6470
6471 if (cmd == ADD_NEW_DISK)
6472 /* need to ensure md_delayed_delete() has completed */
6473 flush_workqueue(md_misc_wq);
6474
6475 if (cmd == HOT_REMOVE_DISK)
6476 /* need to ensure recovery thread has run */
6477 wait_event_interruptible_timeout(mddev->sb_wait,
6478 !test_bit(MD_RECOVERY_NEEDED,
6479 &mddev->flags),
6480 msecs_to_jiffies(5000));
6481 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6482 /* Need to flush page cache, and ensure no-one else opens
6483 * and writes
6484 */
6485 mutex_lock(&mddev->open_mutex);
6486 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
6487 mutex_unlock(&mddev->open_mutex);
6488 err = -EBUSY;
6489 goto out;
6490 }
6491 set_bit(MD_STILL_CLOSED, &mddev->flags);
6492 mutex_unlock(&mddev->open_mutex);
6493 sync_blockdev(bdev);
6494 }
6495 err = mddev_lock(mddev);
6496 if (err) {
6497 printk(KERN_INFO
6498 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6499 err, cmd);
6500 goto out;
6501 }
6502
6503 if (cmd == SET_ARRAY_INFO) {
6504 mdu_array_info_t info;
6505 if (!arg)
6506 memset(&info, 0, sizeof(info));
6507 else if (copy_from_user(&info, argp, sizeof(info))) {
6508 err = -EFAULT;
6509 goto unlock;
6510 }
6511 if (mddev->pers) {
6512 err = update_array_info(mddev, &info);
6513 if (err) {
6514 printk(KERN_WARNING "md: couldn't update"
6515 " array info. %d\n", err);
6516 goto unlock;
6517 }
6518 goto unlock;
6519 }
6520 if (!list_empty(&mddev->disks)) {
6521 printk(KERN_WARNING
6522 "md: array %s already has disks!\n",
6523 mdname(mddev));
6524 err = -EBUSY;
6525 goto unlock;
6526 }
6527 if (mddev->raid_disks) {
6528 printk(KERN_WARNING
6529 "md: array %s already initialised!\n",
6530 mdname(mddev));
6531 err = -EBUSY;
6532 goto unlock;
6533 }
6534 err = set_array_info(mddev, &info);
6535 if (err) {
6536 printk(KERN_WARNING "md: couldn't set"
6537 " array info. %d\n", err);
6538 goto unlock;
6539 }
6540 goto unlock;
6541 }
6542
6543 /*
6544 * Commands querying/configuring an existing array:
6545 */
6546 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6547 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6548 if ((!mddev->raid_disks && !mddev->external)
6549 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6550 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6551 && cmd != GET_BITMAP_FILE) {
6552 err = -ENODEV;
6553 goto unlock;
6554 }
6555
6556 /*
6557 * Commands even a read-only array can execute:
6558 */
6559 switch (cmd) {
6560 case RESTART_ARRAY_RW:
6561 err = restart_array(mddev);
6562 goto unlock;
6563
6564 case STOP_ARRAY:
6565 err = do_md_stop(mddev, 0, bdev);
6566 goto unlock;
6567
6568 case STOP_ARRAY_RO:
6569 err = md_set_readonly(mddev, bdev);
6570 goto unlock;
6571
6572 case HOT_REMOVE_DISK:
6573 err = hot_remove_disk(mddev, new_decode_dev(arg));
6574 goto unlock;
6575
6576 case ADD_NEW_DISK:
6577 /* We can support ADD_NEW_DISK on read-only arrays
6578 * on if we are re-adding a preexisting device.
6579 * So require mddev->pers and MD_DISK_SYNC.
6580 */
6581 if (mddev->pers) {
6582 mdu_disk_info_t info;
6583 if (copy_from_user(&info, argp, sizeof(info)))
6584 err = -EFAULT;
6585 else if (!(info.state & (1<<MD_DISK_SYNC)))
6586 /* Need to clear read-only for this */
6587 break;
6588 else
6589 err = add_new_disk(mddev, &info);
6590 goto unlock;
6591 }
6592 break;
6593
6594 case BLKROSET:
6595 if (get_user(ro, (int __user *)(arg))) {
6596 err = -EFAULT;
6597 goto unlock;
6598 }
6599 err = -EINVAL;
6600
6601 /* if the bdev is going readonly the value of mddev->ro
6602 * does not matter, no writes are coming
6603 */
6604 if (ro)
6605 goto unlock;
6606
6607 /* are we are already prepared for writes? */
6608 if (mddev->ro != 1)
6609 goto unlock;
6610
6611 /* transitioning to readauto need only happen for
6612 * arrays that call md_write_start
6613 */
6614 if (mddev->pers) {
6615 err = restart_array(mddev);
6616 if (err == 0) {
6617 mddev->ro = 2;
6618 set_disk_ro(mddev->gendisk, 0);
6619 }
6620 }
6621 goto unlock;
6622 }
6623
6624 /*
6625 * The remaining ioctls are changing the state of the
6626 * superblock, so we do not allow them on read-only arrays.
6627 */
6628 if (mddev->ro && mddev->pers) {
6629 if (mddev->ro == 2) {
6630 mddev->ro = 0;
6631 sysfs_notify_dirent_safe(mddev->sysfs_state);
6632 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6633 /* mddev_unlock will wake thread */
6634 /* If a device failed while we were read-only, we
6635 * need to make sure the metadata is updated now.
6636 */
6637 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6638 mddev_unlock(mddev);
6639 wait_event(mddev->sb_wait,
6640 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6641 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6642 mddev_lock_nointr(mddev);
6643 }
6644 } else {
6645 err = -EROFS;
6646 goto unlock;
6647 }
6648 }
6649
6650 switch (cmd) {
6651 case ADD_NEW_DISK:
6652 {
6653 mdu_disk_info_t info;
6654 if (copy_from_user(&info, argp, sizeof(info)))
6655 err = -EFAULT;
6656 else
6657 err = add_new_disk(mddev, &info);
6658 goto unlock;
6659 }
6660
6661 case HOT_ADD_DISK:
6662 err = hot_add_disk(mddev, new_decode_dev(arg));
6663 goto unlock;
6664
6665 case RUN_ARRAY:
6666 err = do_md_run(mddev);
6667 goto unlock;
6668
6669 case SET_BITMAP_FILE:
6670 err = set_bitmap_file(mddev, (int)arg);
6671 goto unlock;
6672
6673 default:
6674 err = -EINVAL;
6675 goto unlock;
6676 }
6677
6678 unlock:
6679 if (mddev->hold_active == UNTIL_IOCTL &&
6680 err != -EINVAL)
6681 mddev->hold_active = 0;
6682 mddev_unlock(mddev);
6683 out:
6684 return err;
6685 }
6686 #ifdef CONFIG_COMPAT
6687 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6688 unsigned int cmd, unsigned long arg)
6689 {
6690 switch (cmd) {
6691 case HOT_REMOVE_DISK:
6692 case HOT_ADD_DISK:
6693 case SET_DISK_FAULTY:
6694 case SET_BITMAP_FILE:
6695 /* These take in integer arg, do not convert */
6696 break;
6697 default:
6698 arg = (unsigned long)compat_ptr(arg);
6699 break;
6700 }
6701
6702 return md_ioctl(bdev, mode, cmd, arg);
6703 }
6704 #endif /* CONFIG_COMPAT */
6705
6706 static int md_open(struct block_device *bdev, fmode_t mode)
6707 {
6708 /*
6709 * Succeed if we can lock the mddev, which confirms that
6710 * it isn't being stopped right now.
6711 */
6712 struct mddev *mddev = mddev_find(bdev->bd_dev);
6713 int err;
6714
6715 if (!mddev)
6716 return -ENODEV;
6717
6718 if (mddev->gendisk != bdev->bd_disk) {
6719 /* we are racing with mddev_put which is discarding this
6720 * bd_disk.
6721 */
6722 mddev_put(mddev);
6723 /* Wait until bdev->bd_disk is definitely gone */
6724 flush_workqueue(md_misc_wq);
6725 /* Then retry the open from the top */
6726 return -ERESTARTSYS;
6727 }
6728 BUG_ON(mddev != bdev->bd_disk->private_data);
6729
6730 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
6731 goto out;
6732
6733 err = 0;
6734 atomic_inc(&mddev->openers);
6735 clear_bit(MD_STILL_CLOSED, &mddev->flags);
6736 mutex_unlock(&mddev->open_mutex);
6737
6738 check_disk_change(bdev);
6739 out:
6740 return err;
6741 }
6742
6743 static void md_release(struct gendisk *disk, fmode_t mode)
6744 {
6745 struct mddev *mddev = disk->private_data;
6746
6747 BUG_ON(!mddev);
6748 atomic_dec(&mddev->openers);
6749 mddev_put(mddev);
6750 }
6751
6752 static int md_media_changed(struct gendisk *disk)
6753 {
6754 struct mddev *mddev = disk->private_data;
6755
6756 return mddev->changed;
6757 }
6758
6759 static int md_revalidate(struct gendisk *disk)
6760 {
6761 struct mddev *mddev = disk->private_data;
6762
6763 mddev->changed = 0;
6764 return 0;
6765 }
6766 static const struct block_device_operations md_fops =
6767 {
6768 .owner = THIS_MODULE,
6769 .open = md_open,
6770 .release = md_release,
6771 .ioctl = md_ioctl,
6772 #ifdef CONFIG_COMPAT
6773 .compat_ioctl = md_compat_ioctl,
6774 #endif
6775 .getgeo = md_getgeo,
6776 .media_changed = md_media_changed,
6777 .revalidate_disk= md_revalidate,
6778 };
6779
6780 static int md_thread(void *arg)
6781 {
6782 struct md_thread *thread = arg;
6783
6784 /*
6785 * md_thread is a 'system-thread', it's priority should be very
6786 * high. We avoid resource deadlocks individually in each
6787 * raid personality. (RAID5 does preallocation) We also use RR and
6788 * the very same RT priority as kswapd, thus we will never get
6789 * into a priority inversion deadlock.
6790 *
6791 * we definitely have to have equal or higher priority than
6792 * bdflush, otherwise bdflush will deadlock if there are too
6793 * many dirty RAID5 blocks.
6794 */
6795
6796 allow_signal(SIGKILL);
6797 while (!kthread_should_stop()) {
6798
6799 /* We need to wait INTERRUPTIBLE so that
6800 * we don't add to the load-average.
6801 * That means we need to be sure no signals are
6802 * pending
6803 */
6804 if (signal_pending(current))
6805 flush_signals(current);
6806
6807 wait_event_interruptible_timeout
6808 (thread->wqueue,
6809 test_bit(THREAD_WAKEUP, &thread->flags)
6810 || kthread_should_stop(),
6811 thread->timeout);
6812
6813 clear_bit(THREAD_WAKEUP, &thread->flags);
6814 if (!kthread_should_stop())
6815 thread->run(thread);
6816 }
6817
6818 return 0;
6819 }
6820
6821 void md_wakeup_thread(struct md_thread *thread)
6822 {
6823 if (thread) {
6824 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
6825 set_bit(THREAD_WAKEUP, &thread->flags);
6826 wake_up(&thread->wqueue);
6827 }
6828 }
6829 EXPORT_SYMBOL(md_wakeup_thread);
6830
6831 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
6832 struct mddev *mddev, const char *name)
6833 {
6834 struct md_thread *thread;
6835
6836 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
6837 if (!thread)
6838 return NULL;
6839
6840 init_waitqueue_head(&thread->wqueue);
6841
6842 thread->run = run;
6843 thread->mddev = mddev;
6844 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6845 thread->tsk = kthread_run(md_thread, thread,
6846 "%s_%s",
6847 mdname(thread->mddev),
6848 name);
6849 if (IS_ERR(thread->tsk)) {
6850 kfree(thread);
6851 return NULL;
6852 }
6853 return thread;
6854 }
6855 EXPORT_SYMBOL(md_register_thread);
6856
6857 void md_unregister_thread(struct md_thread **threadp)
6858 {
6859 struct md_thread *thread = *threadp;
6860 if (!thread)
6861 return;
6862 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6863 /* Locking ensures that mddev_unlock does not wake_up a
6864 * non-existent thread
6865 */
6866 spin_lock(&pers_lock);
6867 *threadp = NULL;
6868 spin_unlock(&pers_lock);
6869
6870 kthread_stop(thread->tsk);
6871 kfree(thread);
6872 }
6873 EXPORT_SYMBOL(md_unregister_thread);
6874
6875 void md_error(struct mddev *mddev, struct md_rdev *rdev)
6876 {
6877 if (!rdev || test_bit(Faulty, &rdev->flags))
6878 return;
6879
6880 if (!mddev->pers || !mddev->pers->error_handler)
6881 return;
6882 mddev->pers->error_handler(mddev,rdev);
6883 if (mddev->degraded)
6884 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6885 sysfs_notify_dirent_safe(rdev->sysfs_state);
6886 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6887 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6888 md_wakeup_thread(mddev->thread);
6889 if (mddev->event_work.func)
6890 queue_work(md_misc_wq, &mddev->event_work);
6891 md_new_event_inintr(mddev);
6892 }
6893 EXPORT_SYMBOL(md_error);
6894
6895 /* seq_file implementation /proc/mdstat */
6896
6897 static void status_unused(struct seq_file *seq)
6898 {
6899 int i = 0;
6900 struct md_rdev *rdev;
6901
6902 seq_printf(seq, "unused devices: ");
6903
6904 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6905 char b[BDEVNAME_SIZE];
6906 i++;
6907 seq_printf(seq, "%s ",
6908 bdevname(rdev->bdev,b));
6909 }
6910 if (!i)
6911 seq_printf(seq, "<none>");
6912
6913 seq_printf(seq, "\n");
6914 }
6915
6916 static void status_resync(struct seq_file *seq, struct mddev *mddev)
6917 {
6918 sector_t max_sectors, resync, res;
6919 unsigned long dt, db;
6920 sector_t rt;
6921 int scale;
6922 unsigned int per_milli;
6923
6924 if (mddev->curr_resync <= 3)
6925 resync = 0;
6926 else
6927 resync = mddev->curr_resync
6928 - atomic_read(&mddev->recovery_active);
6929
6930 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
6931 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6932 max_sectors = mddev->resync_max_sectors;
6933 else
6934 max_sectors = mddev->dev_sectors;
6935
6936 WARN_ON(max_sectors == 0);
6937 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6938 * in a sector_t, and (max_sectors>>scale) will fit in a
6939 * u32, as those are the requirements for sector_div.
6940 * Thus 'scale' must be at least 10
6941 */
6942 scale = 10;
6943 if (sizeof(sector_t) > sizeof(unsigned long)) {
6944 while ( max_sectors/2 > (1ULL<<(scale+32)))
6945 scale++;
6946 }
6947 res = (resync>>scale)*1000;
6948 sector_div(res, (u32)((max_sectors>>scale)+1));
6949
6950 per_milli = res;
6951 {
6952 int i, x = per_milli/50, y = 20-x;
6953 seq_printf(seq, "[");
6954 for (i = 0; i < x; i++)
6955 seq_printf(seq, "=");
6956 seq_printf(seq, ">");
6957 for (i = 0; i < y; i++)
6958 seq_printf(seq, ".");
6959 seq_printf(seq, "] ");
6960 }
6961 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6962 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6963 "reshape" :
6964 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6965 "check" :
6966 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6967 "resync" : "recovery"))),
6968 per_milli/10, per_milli % 10,
6969 (unsigned long long) resync/2,
6970 (unsigned long long) max_sectors/2);
6971
6972 /*
6973 * dt: time from mark until now
6974 * db: blocks written from mark until now
6975 * rt: remaining time
6976 *
6977 * rt is a sector_t, so could be 32bit or 64bit.
6978 * So we divide before multiply in case it is 32bit and close
6979 * to the limit.
6980 * We scale the divisor (db) by 32 to avoid losing precision
6981 * near the end of resync when the number of remaining sectors
6982 * is close to 'db'.
6983 * We then divide rt by 32 after multiplying by db to compensate.
6984 * The '+1' avoids division by zero if db is very small.
6985 */
6986 dt = ((jiffies - mddev->resync_mark) / HZ);
6987 if (!dt) dt++;
6988 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6989 - mddev->resync_mark_cnt;
6990
6991 rt = max_sectors - resync; /* number of remaining sectors */
6992 sector_div(rt, db/32+1);
6993 rt *= dt;
6994 rt >>= 5;
6995
6996 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6997 ((unsigned long)rt % 60)/6);
6998
6999 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
7000 }
7001
7002 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7003 {
7004 struct list_head *tmp;
7005 loff_t l = *pos;
7006 struct mddev *mddev;
7007
7008 if (l >= 0x10000)
7009 return NULL;
7010 if (!l--)
7011 /* header */
7012 return (void*)1;
7013
7014 spin_lock(&all_mddevs_lock);
7015 list_for_each(tmp,&all_mddevs)
7016 if (!l--) {
7017 mddev = list_entry(tmp, struct mddev, all_mddevs);
7018 mddev_get(mddev);
7019 spin_unlock(&all_mddevs_lock);
7020 return mddev;
7021 }
7022 spin_unlock(&all_mddevs_lock);
7023 if (!l--)
7024 return (void*)2;/* tail */
7025 return NULL;
7026 }
7027
7028 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7029 {
7030 struct list_head *tmp;
7031 struct mddev *next_mddev, *mddev = v;
7032
7033 ++*pos;
7034 if (v == (void*)2)
7035 return NULL;
7036
7037 spin_lock(&all_mddevs_lock);
7038 if (v == (void*)1)
7039 tmp = all_mddevs.next;
7040 else
7041 tmp = mddev->all_mddevs.next;
7042 if (tmp != &all_mddevs)
7043 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
7044 else {
7045 next_mddev = (void*)2;
7046 *pos = 0x10000;
7047 }
7048 spin_unlock(&all_mddevs_lock);
7049
7050 if (v != (void*)1)
7051 mddev_put(mddev);
7052 return next_mddev;
7053
7054 }
7055
7056 static void md_seq_stop(struct seq_file *seq, void *v)
7057 {
7058 struct mddev *mddev = v;
7059
7060 if (mddev && v != (void*)1 && v != (void*)2)
7061 mddev_put(mddev);
7062 }
7063
7064 static int md_seq_show(struct seq_file *seq, void *v)
7065 {
7066 struct mddev *mddev = v;
7067 sector_t sectors;
7068 struct md_rdev *rdev;
7069
7070 if (v == (void*)1) {
7071 struct md_personality *pers;
7072 seq_printf(seq, "Personalities : ");
7073 spin_lock(&pers_lock);
7074 list_for_each_entry(pers, &pers_list, list)
7075 seq_printf(seq, "[%s] ", pers->name);
7076
7077 spin_unlock(&pers_lock);
7078 seq_printf(seq, "\n");
7079 seq->poll_event = atomic_read(&md_event_count);
7080 return 0;
7081 }
7082 if (v == (void*)2) {
7083 status_unused(seq);
7084 return 0;
7085 }
7086
7087 spin_lock(&mddev->lock);
7088 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7089 seq_printf(seq, "%s : %sactive", mdname(mddev),
7090 mddev->pers ? "" : "in");
7091 if (mddev->pers) {
7092 if (mddev->ro==1)
7093 seq_printf(seq, " (read-only)");
7094 if (mddev->ro==2)
7095 seq_printf(seq, " (auto-read-only)");
7096 seq_printf(seq, " %s", mddev->pers->name);
7097 }
7098
7099 sectors = 0;
7100 rcu_read_lock();
7101 rdev_for_each_rcu(rdev, mddev) {
7102 char b[BDEVNAME_SIZE];
7103 seq_printf(seq, " %s[%d]",
7104 bdevname(rdev->bdev,b), rdev->desc_nr);
7105 if (test_bit(WriteMostly, &rdev->flags))
7106 seq_printf(seq, "(W)");
7107 if (test_bit(Faulty, &rdev->flags)) {
7108 seq_printf(seq, "(F)");
7109 continue;
7110 }
7111 if (rdev->raid_disk < 0)
7112 seq_printf(seq, "(S)"); /* spare */
7113 if (test_bit(Replacement, &rdev->flags))
7114 seq_printf(seq, "(R)");
7115 sectors += rdev->sectors;
7116 }
7117 rcu_read_unlock();
7118
7119 if (!list_empty(&mddev->disks)) {
7120 if (mddev->pers)
7121 seq_printf(seq, "\n %llu blocks",
7122 (unsigned long long)
7123 mddev->array_sectors / 2);
7124 else
7125 seq_printf(seq, "\n %llu blocks",
7126 (unsigned long long)sectors / 2);
7127 }
7128 if (mddev->persistent) {
7129 if (mddev->major_version != 0 ||
7130 mddev->minor_version != 90) {
7131 seq_printf(seq," super %d.%d",
7132 mddev->major_version,
7133 mddev->minor_version);
7134 }
7135 } else if (mddev->external)
7136 seq_printf(seq, " super external:%s",
7137 mddev->metadata_type);
7138 else
7139 seq_printf(seq, " super non-persistent");
7140
7141 if (mddev->pers) {
7142 mddev->pers->status(seq, mddev);
7143 seq_printf(seq, "\n ");
7144 if (mddev->pers->sync_request) {
7145 if (mddev->curr_resync > 2) {
7146 status_resync(seq, mddev);
7147 seq_printf(seq, "\n ");
7148 } else if (mddev->curr_resync >= 1)
7149 seq_printf(seq, "\tresync=DELAYED\n ");
7150 else if (mddev->recovery_cp < MaxSector)
7151 seq_printf(seq, "\tresync=PENDING\n ");
7152 }
7153 } else
7154 seq_printf(seq, "\n ");
7155
7156 bitmap_status(seq, mddev->bitmap);
7157
7158 seq_printf(seq, "\n");
7159 }
7160 spin_unlock(&mddev->lock);
7161
7162 return 0;
7163 }
7164
7165 static const struct seq_operations md_seq_ops = {
7166 .start = md_seq_start,
7167 .next = md_seq_next,
7168 .stop = md_seq_stop,
7169 .show = md_seq_show,
7170 };
7171
7172 static int md_seq_open(struct inode *inode, struct file *file)
7173 {
7174 struct seq_file *seq;
7175 int error;
7176
7177 error = seq_open(file, &md_seq_ops);
7178 if (error)
7179 return error;
7180
7181 seq = file->private_data;
7182 seq->poll_event = atomic_read(&md_event_count);
7183 return error;
7184 }
7185
7186 static int md_unloading;
7187 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7188 {
7189 struct seq_file *seq = filp->private_data;
7190 int mask;
7191
7192 if (md_unloading)
7193 return POLLIN|POLLRDNORM|POLLERR|POLLPRI;
7194 poll_wait(filp, &md_event_waiters, wait);
7195
7196 /* always allow read */
7197 mask = POLLIN | POLLRDNORM;
7198
7199 if (seq->poll_event != atomic_read(&md_event_count))
7200 mask |= POLLERR | POLLPRI;
7201 return mask;
7202 }
7203
7204 static const struct file_operations md_seq_fops = {
7205 .owner = THIS_MODULE,
7206 .open = md_seq_open,
7207 .read = seq_read,
7208 .llseek = seq_lseek,
7209 .release = seq_release_private,
7210 .poll = mdstat_poll,
7211 };
7212
7213 int register_md_personality(struct md_personality *p)
7214 {
7215 printk(KERN_INFO "md: %s personality registered for level %d\n",
7216 p->name, p->level);
7217 spin_lock(&pers_lock);
7218 list_add_tail(&p->list, &pers_list);
7219 spin_unlock(&pers_lock);
7220 return 0;
7221 }
7222 EXPORT_SYMBOL(register_md_personality);
7223
7224 int unregister_md_personality(struct md_personality *p)
7225 {
7226 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
7227 spin_lock(&pers_lock);
7228 list_del_init(&p->list);
7229 spin_unlock(&pers_lock);
7230 return 0;
7231 }
7232 EXPORT_SYMBOL(unregister_md_personality);
7233
7234 static int is_mddev_idle(struct mddev *mddev, int init)
7235 {
7236 struct md_rdev *rdev;
7237 int idle;
7238 int curr_events;
7239
7240 idle = 1;
7241 rcu_read_lock();
7242 rdev_for_each_rcu(rdev, mddev) {
7243 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7244 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7245 (int)part_stat_read(&disk->part0, sectors[1]) -
7246 atomic_read(&disk->sync_io);
7247 /* sync IO will cause sync_io to increase before the disk_stats
7248 * as sync_io is counted when a request starts, and
7249 * disk_stats is counted when it completes.
7250 * So resync activity will cause curr_events to be smaller than
7251 * when there was no such activity.
7252 * non-sync IO will cause disk_stat to increase without
7253 * increasing sync_io so curr_events will (eventually)
7254 * be larger than it was before. Once it becomes
7255 * substantially larger, the test below will cause
7256 * the array to appear non-idle, and resync will slow
7257 * down.
7258 * If there is a lot of outstanding resync activity when
7259 * we set last_event to curr_events, then all that activity
7260 * completing might cause the array to appear non-idle
7261 * and resync will be slowed down even though there might
7262 * not have been non-resync activity. This will only
7263 * happen once though. 'last_events' will soon reflect
7264 * the state where there is little or no outstanding
7265 * resync requests, and further resync activity will
7266 * always make curr_events less than last_events.
7267 *
7268 */
7269 if (init || curr_events - rdev->last_events > 64) {
7270 rdev->last_events = curr_events;
7271 idle = 0;
7272 }
7273 }
7274 rcu_read_unlock();
7275 return idle;
7276 }
7277
7278 void md_done_sync(struct mddev *mddev, int blocks, int ok)
7279 {
7280 /* another "blocks" (512byte) blocks have been synced */
7281 atomic_sub(blocks, &mddev->recovery_active);
7282 wake_up(&mddev->recovery_wait);
7283 if (!ok) {
7284 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7285 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
7286 md_wakeup_thread(mddev->thread);
7287 // stop recovery, signal do_sync ....
7288 }
7289 }
7290 EXPORT_SYMBOL(md_done_sync);
7291
7292 /* md_write_start(mddev, bi)
7293 * If we need to update some array metadata (e.g. 'active' flag
7294 * in superblock) before writing, schedule a superblock update
7295 * and wait for it to complete.
7296 */
7297 void md_write_start(struct mddev *mddev, struct bio *bi)
7298 {
7299 int did_change = 0;
7300 if (bio_data_dir(bi) != WRITE)
7301 return;
7302
7303 BUG_ON(mddev->ro == 1);
7304 if (mddev->ro == 2) {
7305 /* need to switch to read/write */
7306 mddev->ro = 0;
7307 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7308 md_wakeup_thread(mddev->thread);
7309 md_wakeup_thread(mddev->sync_thread);
7310 did_change = 1;
7311 }
7312 atomic_inc(&mddev->writes_pending);
7313 if (mddev->safemode == 1)
7314 mddev->safemode = 0;
7315 if (mddev->in_sync) {
7316 spin_lock(&mddev->lock);
7317 if (mddev->in_sync) {
7318 mddev->in_sync = 0;
7319 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7320 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7321 md_wakeup_thread(mddev->thread);
7322 did_change = 1;
7323 }
7324 spin_unlock(&mddev->lock);
7325 }
7326 if (did_change)
7327 sysfs_notify_dirent_safe(mddev->sysfs_state);
7328 wait_event(mddev->sb_wait,
7329 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
7330 }
7331 EXPORT_SYMBOL(md_write_start);
7332
7333 void md_write_end(struct mddev *mddev)
7334 {
7335 if (atomic_dec_and_test(&mddev->writes_pending)) {
7336 if (mddev->safemode == 2)
7337 md_wakeup_thread(mddev->thread);
7338 else if (mddev->safemode_delay)
7339 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7340 }
7341 }
7342 EXPORT_SYMBOL(md_write_end);
7343
7344 /* md_allow_write(mddev)
7345 * Calling this ensures that the array is marked 'active' so that writes
7346 * may proceed without blocking. It is important to call this before
7347 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7348 * Must be called with mddev_lock held.
7349 *
7350 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7351 * is dropped, so return -EAGAIN after notifying userspace.
7352 */
7353 int md_allow_write(struct mddev *mddev)
7354 {
7355 if (!mddev->pers)
7356 return 0;
7357 if (mddev->ro)
7358 return 0;
7359 if (!mddev->pers->sync_request)
7360 return 0;
7361
7362 spin_lock(&mddev->lock);
7363 if (mddev->in_sync) {
7364 mddev->in_sync = 0;
7365 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7366 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7367 if (mddev->safemode_delay &&
7368 mddev->safemode == 0)
7369 mddev->safemode = 1;
7370 spin_unlock(&mddev->lock);
7371 md_update_sb(mddev, 0);
7372 sysfs_notify_dirent_safe(mddev->sysfs_state);
7373 } else
7374 spin_unlock(&mddev->lock);
7375
7376 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
7377 return -EAGAIN;
7378 else
7379 return 0;
7380 }
7381 EXPORT_SYMBOL_GPL(md_allow_write);
7382
7383 #define SYNC_MARKS 10
7384 #define SYNC_MARK_STEP (3*HZ)
7385 #define UPDATE_FREQUENCY (5*60*HZ)
7386 void md_do_sync(struct md_thread *thread)
7387 {
7388 struct mddev *mddev = thread->mddev;
7389 struct mddev *mddev2;
7390 unsigned int currspeed = 0,
7391 window;
7392 sector_t max_sectors,j, io_sectors, recovery_done;
7393 unsigned long mark[SYNC_MARKS];
7394 unsigned long update_time;
7395 sector_t mark_cnt[SYNC_MARKS];
7396 int last_mark,m;
7397 struct list_head *tmp;
7398 sector_t last_check;
7399 int skipped = 0;
7400 struct md_rdev *rdev;
7401 char *desc, *action = NULL;
7402 struct blk_plug plug;
7403
7404 /* just incase thread restarts... */
7405 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7406 return;
7407 if (mddev->ro) {/* never try to sync a read-only array */
7408 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7409 return;
7410 }
7411
7412 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7413 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
7414 desc = "data-check";
7415 action = "check";
7416 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7417 desc = "requested-resync";
7418 action = "repair";
7419 } else
7420 desc = "resync";
7421 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7422 desc = "reshape";
7423 else
7424 desc = "recovery";
7425
7426 mddev->last_sync_action = action ?: desc;
7427
7428 /* we overload curr_resync somewhat here.
7429 * 0 == not engaged in resync at all
7430 * 2 == checking that there is no conflict with another sync
7431 * 1 == like 2, but have yielded to allow conflicting resync to
7432 * commense
7433 * other == active in resync - this many blocks
7434 *
7435 * Before starting a resync we must have set curr_resync to
7436 * 2, and then checked that every "conflicting" array has curr_resync
7437 * less than ours. When we find one that is the same or higher
7438 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7439 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7440 * This will mean we have to start checking from the beginning again.
7441 *
7442 */
7443
7444 do {
7445 mddev->curr_resync = 2;
7446
7447 try_again:
7448 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7449 goto skip;
7450 for_each_mddev(mddev2, tmp) {
7451 if (mddev2 == mddev)
7452 continue;
7453 if (!mddev->parallel_resync
7454 && mddev2->curr_resync
7455 && match_mddev_units(mddev, mddev2)) {
7456 DEFINE_WAIT(wq);
7457 if (mddev < mddev2 && mddev->curr_resync == 2) {
7458 /* arbitrarily yield */
7459 mddev->curr_resync = 1;
7460 wake_up(&resync_wait);
7461 }
7462 if (mddev > mddev2 && mddev->curr_resync == 1)
7463 /* no need to wait here, we can wait the next
7464 * time 'round when curr_resync == 2
7465 */
7466 continue;
7467 /* We need to wait 'interruptible' so as not to
7468 * contribute to the load average, and not to
7469 * be caught by 'softlockup'
7470 */
7471 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7472 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7473 mddev2->curr_resync >= mddev->curr_resync) {
7474 printk(KERN_INFO "md: delaying %s of %s"
7475 " until %s has finished (they"
7476 " share one or more physical units)\n",
7477 desc, mdname(mddev), mdname(mddev2));
7478 mddev_put(mddev2);
7479 if (signal_pending(current))
7480 flush_signals(current);
7481 schedule();
7482 finish_wait(&resync_wait, &wq);
7483 goto try_again;
7484 }
7485 finish_wait(&resync_wait, &wq);
7486 }
7487 }
7488 } while (mddev->curr_resync < 2);
7489
7490 j = 0;
7491 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7492 /* resync follows the size requested by the personality,
7493 * which defaults to physical size, but can be virtual size
7494 */
7495 max_sectors = mddev->resync_max_sectors;
7496 atomic64_set(&mddev->resync_mismatches, 0);
7497 /* we don't use the checkpoint if there's a bitmap */
7498 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7499 j = mddev->resync_min;
7500 else if (!mddev->bitmap)
7501 j = mddev->recovery_cp;
7502
7503 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7504 max_sectors = mddev->resync_max_sectors;
7505 else {
7506 /* recovery follows the physical size of devices */
7507 max_sectors = mddev->dev_sectors;
7508 j = MaxSector;
7509 rcu_read_lock();
7510 rdev_for_each_rcu(rdev, mddev)
7511 if (rdev->raid_disk >= 0 &&
7512 !test_bit(Faulty, &rdev->flags) &&
7513 !test_bit(In_sync, &rdev->flags) &&
7514 rdev->recovery_offset < j)
7515 j = rdev->recovery_offset;
7516 rcu_read_unlock();
7517
7518 /* If there is a bitmap, we need to make sure all
7519 * writes that started before we added a spare
7520 * complete before we start doing a recovery.
7521 * Otherwise the write might complete and (via
7522 * bitmap_endwrite) set a bit in the bitmap after the
7523 * recovery has checked that bit and skipped that
7524 * region.
7525 */
7526 if (mddev->bitmap) {
7527 mddev->pers->quiesce(mddev, 1);
7528 mddev->pers->quiesce(mddev, 0);
7529 }
7530 }
7531
7532 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7533 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7534 " %d KB/sec/disk.\n", speed_min(mddev));
7535 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
7536 "(but not more than %d KB/sec) for %s.\n",
7537 speed_max(mddev), desc);
7538
7539 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
7540
7541 io_sectors = 0;
7542 for (m = 0; m < SYNC_MARKS; m++) {
7543 mark[m] = jiffies;
7544 mark_cnt[m] = io_sectors;
7545 }
7546 last_mark = 0;
7547 mddev->resync_mark = mark[last_mark];
7548 mddev->resync_mark_cnt = mark_cnt[last_mark];
7549
7550 /*
7551 * Tune reconstruction:
7552 */
7553 window = 32*(PAGE_SIZE/512);
7554 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7555 window/2, (unsigned long long)max_sectors/2);
7556
7557 atomic_set(&mddev->recovery_active, 0);
7558 last_check = 0;
7559
7560 if (j>2) {
7561 printk(KERN_INFO
7562 "md: resuming %s of %s from checkpoint.\n",
7563 desc, mdname(mddev));
7564 mddev->curr_resync = j;
7565 } else
7566 mddev->curr_resync = 3; /* no longer delayed */
7567 mddev->curr_resync_completed = j;
7568 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7569 md_new_event(mddev);
7570 update_time = jiffies;
7571
7572 blk_start_plug(&plug);
7573 while (j < max_sectors) {
7574 sector_t sectors;
7575
7576 skipped = 0;
7577
7578 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7579 ((mddev->curr_resync > mddev->curr_resync_completed &&
7580 (mddev->curr_resync - mddev->curr_resync_completed)
7581 > (max_sectors >> 4)) ||
7582 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7583 (j - mddev->curr_resync_completed)*2
7584 >= mddev->resync_max - mddev->curr_resync_completed
7585 )) {
7586 /* time to update curr_resync_completed */
7587 wait_event(mddev->recovery_wait,
7588 atomic_read(&mddev->recovery_active) == 0);
7589 mddev->curr_resync_completed = j;
7590 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7591 j > mddev->recovery_cp)
7592 mddev->recovery_cp = j;
7593 update_time = jiffies;
7594 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7595 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7596 }
7597
7598 while (j >= mddev->resync_max &&
7599 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7600 /* As this condition is controlled by user-space,
7601 * we can block indefinitely, so use '_interruptible'
7602 * to avoid triggering warnings.
7603 */
7604 flush_signals(current); /* just in case */
7605 wait_event_interruptible(mddev->recovery_wait,
7606 mddev->resync_max > j
7607 || test_bit(MD_RECOVERY_INTR,
7608 &mddev->recovery));
7609 }
7610
7611 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7612 break;
7613
7614 sectors = mddev->pers->sync_request(mddev, j, &skipped,
7615 currspeed < speed_min(mddev));
7616 if (sectors == 0) {
7617 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7618 break;
7619 }
7620
7621 if (!skipped) { /* actual IO requested */
7622 io_sectors += sectors;
7623 atomic_add(sectors, &mddev->recovery_active);
7624 }
7625
7626 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7627 break;
7628
7629 j += sectors;
7630 if (j > 2)
7631 mddev->curr_resync = j;
7632 mddev->curr_mark_cnt = io_sectors;
7633 if (last_check == 0)
7634 /* this is the earliest that rebuild will be
7635 * visible in /proc/mdstat
7636 */
7637 md_new_event(mddev);
7638
7639 if (last_check + window > io_sectors || j == max_sectors)
7640 continue;
7641
7642 last_check = io_sectors;
7643 repeat:
7644 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7645 /* step marks */
7646 int next = (last_mark+1) % SYNC_MARKS;
7647
7648 mddev->resync_mark = mark[next];
7649 mddev->resync_mark_cnt = mark_cnt[next];
7650 mark[next] = jiffies;
7651 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
7652 last_mark = next;
7653 }
7654
7655 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7656 break;
7657
7658 /*
7659 * this loop exits only if either when we are slower than
7660 * the 'hard' speed limit, or the system was IO-idle for
7661 * a jiffy.
7662 * the system might be non-idle CPU-wise, but we only care
7663 * about not overloading the IO subsystem. (things like an
7664 * e2fsck being done on the RAID array should execute fast)
7665 */
7666 cond_resched();
7667
7668 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
7669 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
7670 /((jiffies-mddev->resync_mark)/HZ +1) +1;
7671
7672 if (currspeed > speed_min(mddev)) {
7673 if ((currspeed > speed_max(mddev)) ||
7674 !is_mddev_idle(mddev, 0)) {
7675 msleep(500);
7676 goto repeat;
7677 }
7678 }
7679 }
7680 printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc,
7681 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
7682 ? "interrupted" : "done");
7683 /*
7684 * this also signals 'finished resyncing' to md_stop
7685 */
7686 blk_finish_plug(&plug);
7687 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7688
7689 /* tell personality that we are finished */
7690 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
7691
7692 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
7693 mddev->curr_resync > 2) {
7694 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7695 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7696 if (mddev->curr_resync >= mddev->recovery_cp) {
7697 printk(KERN_INFO
7698 "md: checkpointing %s of %s.\n",
7699 desc, mdname(mddev));
7700 if (test_bit(MD_RECOVERY_ERROR,
7701 &mddev->recovery))
7702 mddev->recovery_cp =
7703 mddev->curr_resync_completed;
7704 else
7705 mddev->recovery_cp =
7706 mddev->curr_resync;
7707 }
7708 } else
7709 mddev->recovery_cp = MaxSector;
7710 } else {
7711 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7712 mddev->curr_resync = MaxSector;
7713 rcu_read_lock();
7714 rdev_for_each_rcu(rdev, mddev)
7715 if (rdev->raid_disk >= 0 &&
7716 mddev->delta_disks >= 0 &&
7717 !test_bit(Faulty, &rdev->flags) &&
7718 !test_bit(In_sync, &rdev->flags) &&
7719 rdev->recovery_offset < mddev->curr_resync)
7720 rdev->recovery_offset = mddev->curr_resync;
7721 rcu_read_unlock();
7722 }
7723 }
7724 skip:
7725 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7726
7727 spin_lock(&mddev->lock);
7728 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7729 /* We completed so min/max setting can be forgotten if used. */
7730 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7731 mddev->resync_min = 0;
7732 mddev->resync_max = MaxSector;
7733 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7734 mddev->resync_min = mddev->curr_resync_completed;
7735 mddev->curr_resync = 0;
7736 spin_unlock(&mddev->lock);
7737
7738 wake_up(&resync_wait);
7739 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7740 md_wakeup_thread(mddev->thread);
7741 return;
7742 }
7743 EXPORT_SYMBOL_GPL(md_do_sync);
7744
7745 static int remove_and_add_spares(struct mddev *mddev,
7746 struct md_rdev *this)
7747 {
7748 struct md_rdev *rdev;
7749 int spares = 0;
7750 int removed = 0;
7751
7752 rdev_for_each(rdev, mddev)
7753 if ((this == NULL || rdev == this) &&
7754 rdev->raid_disk >= 0 &&
7755 !test_bit(Blocked, &rdev->flags) &&
7756 (test_bit(Faulty, &rdev->flags) ||
7757 ! test_bit(In_sync, &rdev->flags)) &&
7758 atomic_read(&rdev->nr_pending)==0) {
7759 if (mddev->pers->hot_remove_disk(
7760 mddev, rdev) == 0) {
7761 sysfs_unlink_rdev(mddev, rdev);
7762 rdev->raid_disk = -1;
7763 removed++;
7764 }
7765 }
7766 if (removed && mddev->kobj.sd)
7767 sysfs_notify(&mddev->kobj, NULL, "degraded");
7768
7769 if (this)
7770 goto no_add;
7771
7772 rdev_for_each(rdev, mddev) {
7773 if (rdev->raid_disk >= 0 &&
7774 !test_bit(In_sync, &rdev->flags) &&
7775 !test_bit(Faulty, &rdev->flags))
7776 spares++;
7777 if (rdev->raid_disk >= 0)
7778 continue;
7779 if (test_bit(Faulty, &rdev->flags))
7780 continue;
7781 if (mddev->ro &&
7782 ! (rdev->saved_raid_disk >= 0 &&
7783 !test_bit(Bitmap_sync, &rdev->flags)))
7784 continue;
7785
7786 if (rdev->saved_raid_disk < 0)
7787 rdev->recovery_offset = 0;
7788 if (mddev->pers->
7789 hot_add_disk(mddev, rdev) == 0) {
7790 if (sysfs_link_rdev(mddev, rdev))
7791 /* failure here is OK */;
7792 spares++;
7793 md_new_event(mddev);
7794 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7795 }
7796 }
7797 no_add:
7798 if (removed)
7799 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7800 return spares;
7801 }
7802
7803 static void md_start_sync(struct work_struct *ws)
7804 {
7805 struct mddev *mddev = container_of(ws, struct mddev, del_work);
7806
7807 mddev->sync_thread = md_register_thread(md_do_sync,
7808 mddev,
7809 "resync");
7810 if (!mddev->sync_thread) {
7811 printk(KERN_ERR "%s: could not start resync"
7812 " thread...\n",
7813 mdname(mddev));
7814 /* leave the spares where they are, it shouldn't hurt */
7815 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7816 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7817 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7818 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7819 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7820 wake_up(&resync_wait);
7821 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7822 &mddev->recovery))
7823 if (mddev->sysfs_action)
7824 sysfs_notify_dirent_safe(mddev->sysfs_action);
7825 } else
7826 md_wakeup_thread(mddev->sync_thread);
7827 sysfs_notify_dirent_safe(mddev->sysfs_action);
7828 md_new_event(mddev);
7829 }
7830
7831 /*
7832 * This routine is regularly called by all per-raid-array threads to
7833 * deal with generic issues like resync and super-block update.
7834 * Raid personalities that don't have a thread (linear/raid0) do not
7835 * need this as they never do any recovery or update the superblock.
7836 *
7837 * It does not do any resync itself, but rather "forks" off other threads
7838 * to do that as needed.
7839 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7840 * "->recovery" and create a thread at ->sync_thread.
7841 * When the thread finishes it sets MD_RECOVERY_DONE
7842 * and wakeups up this thread which will reap the thread and finish up.
7843 * This thread also removes any faulty devices (with nr_pending == 0).
7844 *
7845 * The overall approach is:
7846 * 1/ if the superblock needs updating, update it.
7847 * 2/ If a recovery thread is running, don't do anything else.
7848 * 3/ If recovery has finished, clean up, possibly marking spares active.
7849 * 4/ If there are any faulty devices, remove them.
7850 * 5/ If array is degraded, try to add spares devices
7851 * 6/ If array has spares or is not in-sync, start a resync thread.
7852 */
7853 void md_check_recovery(struct mddev *mddev)
7854 {
7855 if (mddev->suspended)
7856 return;
7857
7858 if (mddev->bitmap)
7859 bitmap_daemon_work(mddev);
7860
7861 if (signal_pending(current)) {
7862 if (mddev->pers->sync_request && !mddev->external) {
7863 printk(KERN_INFO "md: %s in immediate safe mode\n",
7864 mdname(mddev));
7865 mddev->safemode = 2;
7866 }
7867 flush_signals(current);
7868 }
7869
7870 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7871 return;
7872 if ( ! (
7873 (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) ||
7874 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7875 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7876 (mddev->external == 0 && mddev->safemode == 1) ||
7877 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7878 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7879 ))
7880 return;
7881
7882 if (mddev_trylock(mddev)) {
7883 int spares = 0;
7884
7885 if (mddev->ro) {
7886 /* On a read-only array we can:
7887 * - remove failed devices
7888 * - add already-in_sync devices if the array itself
7889 * is in-sync.
7890 * As we only add devices that are already in-sync,
7891 * we can activate the spares immediately.
7892 */
7893 remove_and_add_spares(mddev, NULL);
7894 /* There is no thread, but we need to call
7895 * ->spare_active and clear saved_raid_disk
7896 */
7897 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7898 md_reap_sync_thread(mddev);
7899 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7900 goto unlock;
7901 }
7902
7903 if (!mddev->external) {
7904 int did_change = 0;
7905 spin_lock(&mddev->lock);
7906 if (mddev->safemode &&
7907 !atomic_read(&mddev->writes_pending) &&
7908 !mddev->in_sync &&
7909 mddev->recovery_cp == MaxSector) {
7910 mddev->in_sync = 1;
7911 did_change = 1;
7912 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7913 }
7914 if (mddev->safemode == 1)
7915 mddev->safemode = 0;
7916 spin_unlock(&mddev->lock);
7917 if (did_change)
7918 sysfs_notify_dirent_safe(mddev->sysfs_state);
7919 }
7920
7921 if (mddev->flags & MD_UPDATE_SB_FLAGS)
7922 md_update_sb(mddev, 0);
7923
7924 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7925 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7926 /* resync/recovery still happening */
7927 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7928 goto unlock;
7929 }
7930 if (mddev->sync_thread) {
7931 md_reap_sync_thread(mddev);
7932 goto unlock;
7933 }
7934 /* Set RUNNING before clearing NEEDED to avoid
7935 * any transients in the value of "sync_action".
7936 */
7937 mddev->curr_resync_completed = 0;
7938 spin_lock(&mddev->lock);
7939 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7940 spin_unlock(&mddev->lock);
7941 /* Clear some bits that don't mean anything, but
7942 * might be left set
7943 */
7944 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7945 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7946
7947 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7948 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7949 goto not_running;
7950 /* no recovery is running.
7951 * remove any failed drives, then
7952 * add spares if possible.
7953 * Spares are also removed and re-added, to allow
7954 * the personality to fail the re-add.
7955 */
7956
7957 if (mddev->reshape_position != MaxSector) {
7958 if (mddev->pers->check_reshape == NULL ||
7959 mddev->pers->check_reshape(mddev) != 0)
7960 /* Cannot proceed */
7961 goto not_running;
7962 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7963 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7964 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
7965 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7966 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7967 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7968 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7969 } else if (mddev->recovery_cp < MaxSector) {
7970 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7971 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7972 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7973 /* nothing to be done ... */
7974 goto not_running;
7975
7976 if (mddev->pers->sync_request) {
7977 if (spares) {
7978 /* We are adding a device or devices to an array
7979 * which has the bitmap stored on all devices.
7980 * So make sure all bitmap pages get written
7981 */
7982 bitmap_write_all(mddev->bitmap);
7983 }
7984 INIT_WORK(&mddev->del_work, md_start_sync);
7985 queue_work(md_misc_wq, &mddev->del_work);
7986 goto unlock;
7987 }
7988 not_running:
7989 if (!mddev->sync_thread) {
7990 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7991 wake_up(&resync_wait);
7992 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7993 &mddev->recovery))
7994 if (mddev->sysfs_action)
7995 sysfs_notify_dirent_safe(mddev->sysfs_action);
7996 }
7997 unlock:
7998 wake_up(&mddev->sb_wait);
7999 mddev_unlock(mddev);
8000 }
8001 }
8002 EXPORT_SYMBOL(md_check_recovery);
8003
8004 void md_reap_sync_thread(struct mddev *mddev)
8005 {
8006 struct md_rdev *rdev;
8007
8008 /* resync has finished, collect result */
8009 md_unregister_thread(&mddev->sync_thread);
8010 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8011 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8012 /* success...*/
8013 /* activate any spares */
8014 if (mddev->pers->spare_active(mddev)) {
8015 sysfs_notify(&mddev->kobj, NULL,
8016 "degraded");
8017 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8018 }
8019 }
8020 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8021 mddev->pers->finish_reshape)
8022 mddev->pers->finish_reshape(mddev);
8023
8024 /* If array is no-longer degraded, then any saved_raid_disk
8025 * information must be scrapped.
8026 */
8027 if (!mddev->degraded)
8028 rdev_for_each(rdev, mddev)
8029 rdev->saved_raid_disk = -1;
8030
8031 md_update_sb(mddev, 1);
8032 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8033 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8034 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8035 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8036 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8037 wake_up(&resync_wait);
8038 /* flag recovery needed just to double check */
8039 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8040 sysfs_notify_dirent_safe(mddev->sysfs_action);
8041 md_new_event(mddev);
8042 if (mddev->event_work.func)
8043 queue_work(md_misc_wq, &mddev->event_work);
8044 }
8045 EXPORT_SYMBOL(md_reap_sync_thread);
8046
8047 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
8048 {
8049 sysfs_notify_dirent_safe(rdev->sysfs_state);
8050 wait_event_timeout(rdev->blocked_wait,
8051 !test_bit(Blocked, &rdev->flags) &&
8052 !test_bit(BlockedBadBlocks, &rdev->flags),
8053 msecs_to_jiffies(5000));
8054 rdev_dec_pending(rdev, mddev);
8055 }
8056 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8057
8058 void md_finish_reshape(struct mddev *mddev)
8059 {
8060 /* called be personality module when reshape completes. */
8061 struct md_rdev *rdev;
8062
8063 rdev_for_each(rdev, mddev) {
8064 if (rdev->data_offset > rdev->new_data_offset)
8065 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8066 else
8067 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8068 rdev->data_offset = rdev->new_data_offset;
8069 }
8070 }
8071 EXPORT_SYMBOL(md_finish_reshape);
8072
8073 /* Bad block management.
8074 * We can record which blocks on each device are 'bad' and so just
8075 * fail those blocks, or that stripe, rather than the whole device.
8076 * Entries in the bad-block table are 64bits wide. This comprises:
8077 * Length of bad-range, in sectors: 0-511 for lengths 1-512
8078 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
8079 * A 'shift' can be set so that larger blocks are tracked and
8080 * consequently larger devices can be covered.
8081 * 'Acknowledged' flag - 1 bit. - the most significant bit.
8082 *
8083 * Locking of the bad-block table uses a seqlock so md_is_badblock
8084 * might need to retry if it is very unlucky.
8085 * We will sometimes want to check for bad blocks in a bi_end_io function,
8086 * so we use the write_seqlock_irq variant.
8087 *
8088 * When looking for a bad block we specify a range and want to
8089 * know if any block in the range is bad. So we binary-search
8090 * to the last range that starts at-or-before the given endpoint,
8091 * (or "before the sector after the target range")
8092 * then see if it ends after the given start.
8093 * We return
8094 * 0 if there are no known bad blocks in the range
8095 * 1 if there are known bad block which are all acknowledged
8096 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
8097 * plus the start/length of the first bad section we overlap.
8098 */
8099 int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
8100 sector_t *first_bad, int *bad_sectors)
8101 {
8102 int hi;
8103 int lo;
8104 u64 *p = bb->page;
8105 int rv;
8106 sector_t target = s + sectors;
8107 unsigned seq;
8108
8109 if (bb->shift > 0) {
8110 /* round the start down, and the end up */
8111 s >>= bb->shift;
8112 target += (1<<bb->shift) - 1;
8113 target >>= bb->shift;
8114 sectors = target - s;
8115 }
8116 /* 'target' is now the first block after the bad range */
8117
8118 retry:
8119 seq = read_seqbegin(&bb->lock);
8120 lo = 0;
8121 rv = 0;
8122 hi = bb->count;
8123
8124 /* Binary search between lo and hi for 'target'
8125 * i.e. for the last range that starts before 'target'
8126 */
8127 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8128 * are known not to be the last range before target.
8129 * VARIANT: hi-lo is the number of possible
8130 * ranges, and decreases until it reaches 1
8131 */
8132 while (hi - lo > 1) {
8133 int mid = (lo + hi) / 2;
8134 sector_t a = BB_OFFSET(p[mid]);
8135 if (a < target)
8136 /* This could still be the one, earlier ranges
8137 * could not. */
8138 lo = mid;
8139 else
8140 /* This and later ranges are definitely out. */
8141 hi = mid;
8142 }
8143 /* 'lo' might be the last that started before target, but 'hi' isn't */
8144 if (hi > lo) {
8145 /* need to check all range that end after 's' to see if
8146 * any are unacknowledged.
8147 */
8148 while (lo >= 0 &&
8149 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8150 if (BB_OFFSET(p[lo]) < target) {
8151 /* starts before the end, and finishes after
8152 * the start, so they must overlap
8153 */
8154 if (rv != -1 && BB_ACK(p[lo]))
8155 rv = 1;
8156 else
8157 rv = -1;
8158 *first_bad = BB_OFFSET(p[lo]);
8159 *bad_sectors = BB_LEN(p[lo]);
8160 }
8161 lo--;
8162 }
8163 }
8164
8165 if (read_seqretry(&bb->lock, seq))
8166 goto retry;
8167
8168 return rv;
8169 }
8170 EXPORT_SYMBOL_GPL(md_is_badblock);
8171
8172 /*
8173 * Add a range of bad blocks to the table.
8174 * This might extend the table, or might contract it
8175 * if two adjacent ranges can be merged.
8176 * We binary-search to find the 'insertion' point, then
8177 * decide how best to handle it.
8178 */
8179 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8180 int acknowledged)
8181 {
8182 u64 *p;
8183 int lo, hi;
8184 int rv = 1;
8185 unsigned long flags;
8186
8187 if (bb->shift < 0)
8188 /* badblocks are disabled */
8189 return 0;
8190
8191 if (bb->shift) {
8192 /* round the start down, and the end up */
8193 sector_t next = s + sectors;
8194 s >>= bb->shift;
8195 next += (1<<bb->shift) - 1;
8196 next >>= bb->shift;
8197 sectors = next - s;
8198 }
8199
8200 write_seqlock_irqsave(&bb->lock, flags);
8201
8202 p = bb->page;
8203 lo = 0;
8204 hi = bb->count;
8205 /* Find the last range that starts at-or-before 's' */
8206 while (hi - lo > 1) {
8207 int mid = (lo + hi) / 2;
8208 sector_t a = BB_OFFSET(p[mid]);
8209 if (a <= s)
8210 lo = mid;
8211 else
8212 hi = mid;
8213 }
8214 if (hi > lo && BB_OFFSET(p[lo]) > s)
8215 hi = lo;
8216
8217 if (hi > lo) {
8218 /* we found a range that might merge with the start
8219 * of our new range
8220 */
8221 sector_t a = BB_OFFSET(p[lo]);
8222 sector_t e = a + BB_LEN(p[lo]);
8223 int ack = BB_ACK(p[lo]);
8224 if (e >= s) {
8225 /* Yes, we can merge with a previous range */
8226 if (s == a && s + sectors >= e)
8227 /* new range covers old */
8228 ack = acknowledged;
8229 else
8230 ack = ack && acknowledged;
8231
8232 if (e < s + sectors)
8233 e = s + sectors;
8234 if (e - a <= BB_MAX_LEN) {
8235 p[lo] = BB_MAKE(a, e-a, ack);
8236 s = e;
8237 } else {
8238 /* does not all fit in one range,
8239 * make p[lo] maximal
8240 */
8241 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8242 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8243 s = a + BB_MAX_LEN;
8244 }
8245 sectors = e - s;
8246 }
8247 }
8248 if (sectors && hi < bb->count) {
8249 /* 'hi' points to the first range that starts after 's'.
8250 * Maybe we can merge with the start of that range */
8251 sector_t a = BB_OFFSET(p[hi]);
8252 sector_t e = a + BB_LEN(p[hi]);
8253 int ack = BB_ACK(p[hi]);
8254 if (a <= s + sectors) {
8255 /* merging is possible */
8256 if (e <= s + sectors) {
8257 /* full overlap */
8258 e = s + sectors;
8259 ack = acknowledged;
8260 } else
8261 ack = ack && acknowledged;
8262
8263 a = s;
8264 if (e - a <= BB_MAX_LEN) {
8265 p[hi] = BB_MAKE(a, e-a, ack);
8266 s = e;
8267 } else {
8268 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8269 s = a + BB_MAX_LEN;
8270 }
8271 sectors = e - s;
8272 lo = hi;
8273 hi++;
8274 }
8275 }
8276 if (sectors == 0 && hi < bb->count) {
8277 /* we might be able to combine lo and hi */
8278 /* Note: 's' is at the end of 'lo' */
8279 sector_t a = BB_OFFSET(p[hi]);
8280 int lolen = BB_LEN(p[lo]);
8281 int hilen = BB_LEN(p[hi]);
8282 int newlen = lolen + hilen - (s - a);
8283 if (s >= a && newlen < BB_MAX_LEN) {
8284 /* yes, we can combine them */
8285 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8286 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8287 memmove(p + hi, p + hi + 1,
8288 (bb->count - hi - 1) * 8);
8289 bb->count--;
8290 }
8291 }
8292 while (sectors) {
8293 /* didn't merge (it all).
8294 * Need to add a range just before 'hi' */
8295 if (bb->count >= MD_MAX_BADBLOCKS) {
8296 /* No room for more */
8297 rv = 0;
8298 break;
8299 } else {
8300 int this_sectors = sectors;
8301 memmove(p + hi + 1, p + hi,
8302 (bb->count - hi) * 8);
8303 bb->count++;
8304
8305 if (this_sectors > BB_MAX_LEN)
8306 this_sectors = BB_MAX_LEN;
8307 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8308 sectors -= this_sectors;
8309 s += this_sectors;
8310 }
8311 }
8312
8313 bb->changed = 1;
8314 if (!acknowledged)
8315 bb->unacked_exist = 1;
8316 write_sequnlock_irqrestore(&bb->lock, flags);
8317
8318 return rv;
8319 }
8320
8321 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8322 int is_new)
8323 {
8324 int rv;
8325 if (is_new)
8326 s += rdev->new_data_offset;
8327 else
8328 s += rdev->data_offset;
8329 rv = md_set_badblocks(&rdev->badblocks,
8330 s, sectors, 0);
8331 if (rv) {
8332 /* Make sure they get written out promptly */
8333 sysfs_notify_dirent_safe(rdev->sysfs_state);
8334 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8335 md_wakeup_thread(rdev->mddev->thread);
8336 }
8337 return rv;
8338 }
8339 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8340
8341 /*
8342 * Remove a range of bad blocks from the table.
8343 * This may involve extending the table if we spilt a region,
8344 * but it must not fail. So if the table becomes full, we just
8345 * drop the remove request.
8346 */
8347 static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8348 {
8349 u64 *p;
8350 int lo, hi;
8351 sector_t target = s + sectors;
8352 int rv = 0;
8353
8354 if (bb->shift > 0) {
8355 /* When clearing we round the start up and the end down.
8356 * This should not matter as the shift should align with
8357 * the block size and no rounding should ever be needed.
8358 * However it is better the think a block is bad when it
8359 * isn't than to think a block is not bad when it is.
8360 */
8361 s += (1<<bb->shift) - 1;
8362 s >>= bb->shift;
8363 target >>= bb->shift;
8364 sectors = target - s;
8365 }
8366
8367 write_seqlock_irq(&bb->lock);
8368
8369 p = bb->page;
8370 lo = 0;
8371 hi = bb->count;
8372 /* Find the last range that starts before 'target' */
8373 while (hi - lo > 1) {
8374 int mid = (lo + hi) / 2;
8375 sector_t a = BB_OFFSET(p[mid]);
8376 if (a < target)
8377 lo = mid;
8378 else
8379 hi = mid;
8380 }
8381 if (hi > lo) {
8382 /* p[lo] is the last range that could overlap the
8383 * current range. Earlier ranges could also overlap,
8384 * but only this one can overlap the end of the range.
8385 */
8386 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8387 /* Partial overlap, leave the tail of this range */
8388 int ack = BB_ACK(p[lo]);
8389 sector_t a = BB_OFFSET(p[lo]);
8390 sector_t end = a + BB_LEN(p[lo]);
8391
8392 if (a < s) {
8393 /* we need to split this range */
8394 if (bb->count >= MD_MAX_BADBLOCKS) {
8395 rv = -ENOSPC;
8396 goto out;
8397 }
8398 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8399 bb->count++;
8400 p[lo] = BB_MAKE(a, s-a, ack);
8401 lo++;
8402 }
8403 p[lo] = BB_MAKE(target, end - target, ack);
8404 /* there is no longer an overlap */
8405 hi = lo;
8406 lo--;
8407 }
8408 while (lo >= 0 &&
8409 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8410 /* This range does overlap */
8411 if (BB_OFFSET(p[lo]) < s) {
8412 /* Keep the early parts of this range. */
8413 int ack = BB_ACK(p[lo]);
8414 sector_t start = BB_OFFSET(p[lo]);
8415 p[lo] = BB_MAKE(start, s - start, ack);
8416 /* now low doesn't overlap, so.. */
8417 break;
8418 }
8419 lo--;
8420 }
8421 /* 'lo' is strictly before, 'hi' is strictly after,
8422 * anything between needs to be discarded
8423 */
8424 if (hi - lo > 1) {
8425 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8426 bb->count -= (hi - lo - 1);
8427 }
8428 }
8429
8430 bb->changed = 1;
8431 out:
8432 write_sequnlock_irq(&bb->lock);
8433 return rv;
8434 }
8435
8436 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8437 int is_new)
8438 {
8439 if (is_new)
8440 s += rdev->new_data_offset;
8441 else
8442 s += rdev->data_offset;
8443 return md_clear_badblocks(&rdev->badblocks,
8444 s, sectors);
8445 }
8446 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8447
8448 /*
8449 * Acknowledge all bad blocks in a list.
8450 * This only succeeds if ->changed is clear. It is used by
8451 * in-kernel metadata updates
8452 */
8453 void md_ack_all_badblocks(struct badblocks *bb)
8454 {
8455 if (bb->page == NULL || bb->changed)
8456 /* no point even trying */
8457 return;
8458 write_seqlock_irq(&bb->lock);
8459
8460 if (bb->changed == 0 && bb->unacked_exist) {
8461 u64 *p = bb->page;
8462 int i;
8463 for (i = 0; i < bb->count ; i++) {
8464 if (!BB_ACK(p[i])) {
8465 sector_t start = BB_OFFSET(p[i]);
8466 int len = BB_LEN(p[i]);
8467 p[i] = BB_MAKE(start, len, 1);
8468 }
8469 }
8470 bb->unacked_exist = 0;
8471 }
8472 write_sequnlock_irq(&bb->lock);
8473 }
8474 EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8475
8476 /* sysfs access to bad-blocks list.
8477 * We present two files.
8478 * 'bad-blocks' lists sector numbers and lengths of ranges that
8479 * are recorded as bad. The list is truncated to fit within
8480 * the one-page limit of sysfs.
8481 * Writing "sector length" to this file adds an acknowledged
8482 * bad block list.
8483 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8484 * been acknowledged. Writing to this file adds bad blocks
8485 * without acknowledging them. This is largely for testing.
8486 */
8487
8488 static ssize_t
8489 badblocks_show(struct badblocks *bb, char *page, int unack)
8490 {
8491 size_t len;
8492 int i;
8493 u64 *p = bb->page;
8494 unsigned seq;
8495
8496 if (bb->shift < 0)
8497 return 0;
8498
8499 retry:
8500 seq = read_seqbegin(&bb->lock);
8501
8502 len = 0;
8503 i = 0;
8504
8505 while (len < PAGE_SIZE && i < bb->count) {
8506 sector_t s = BB_OFFSET(p[i]);
8507 unsigned int length = BB_LEN(p[i]);
8508 int ack = BB_ACK(p[i]);
8509 i++;
8510
8511 if (unack && ack)
8512 continue;
8513
8514 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8515 (unsigned long long)s << bb->shift,
8516 length << bb->shift);
8517 }
8518 if (unack && len == 0)
8519 bb->unacked_exist = 0;
8520
8521 if (read_seqretry(&bb->lock, seq))
8522 goto retry;
8523
8524 return len;
8525 }
8526
8527 #define DO_DEBUG 1
8528
8529 static ssize_t
8530 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8531 {
8532 unsigned long long sector;
8533 int length;
8534 char newline;
8535 #ifdef DO_DEBUG
8536 /* Allow clearing via sysfs *only* for testing/debugging.
8537 * Normally only a successful write may clear a badblock
8538 */
8539 int clear = 0;
8540 if (page[0] == '-') {
8541 clear = 1;
8542 page++;
8543 }
8544 #endif /* DO_DEBUG */
8545
8546 switch (sscanf(page, "%llu %d%c", &sector, &length, &newline)) {
8547 case 3:
8548 if (newline != '\n')
8549 return -EINVAL;
8550 case 2:
8551 if (length <= 0)
8552 return -EINVAL;
8553 break;
8554 default:
8555 return -EINVAL;
8556 }
8557
8558 #ifdef DO_DEBUG
8559 if (clear) {
8560 md_clear_badblocks(bb, sector, length);
8561 return len;
8562 }
8563 #endif /* DO_DEBUG */
8564 if (md_set_badblocks(bb, sector, length, !unack))
8565 return len;
8566 else
8567 return -ENOSPC;
8568 }
8569
8570 static int md_notify_reboot(struct notifier_block *this,
8571 unsigned long code, void *x)
8572 {
8573 struct list_head *tmp;
8574 struct mddev *mddev;
8575 int need_delay = 0;
8576
8577 for_each_mddev(mddev, tmp) {
8578 if (mddev_trylock(mddev)) {
8579 if (mddev->pers)
8580 __md_stop_writes(mddev);
8581 if (mddev->persistent)
8582 mddev->safemode = 2;
8583 mddev_unlock(mddev);
8584 }
8585 need_delay = 1;
8586 }
8587 /*
8588 * certain more exotic SCSI devices are known to be
8589 * volatile wrt too early system reboots. While the
8590 * right place to handle this issue is the given
8591 * driver, we do want to have a safe RAID driver ...
8592 */
8593 if (need_delay)
8594 mdelay(1000*1);
8595
8596 return NOTIFY_DONE;
8597 }
8598
8599 static struct notifier_block md_notifier = {
8600 .notifier_call = md_notify_reboot,
8601 .next = NULL,
8602 .priority = INT_MAX, /* before any real devices */
8603 };
8604
8605 static void md_geninit(void)
8606 {
8607 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
8608
8609 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
8610 }
8611
8612 static int __init md_init(void)
8613 {
8614 int ret = -ENOMEM;
8615
8616 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
8617 if (!md_wq)
8618 goto err_wq;
8619
8620 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8621 if (!md_misc_wq)
8622 goto err_misc_wq;
8623
8624 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8625 goto err_md;
8626
8627 if ((ret = register_blkdev(0, "mdp")) < 0)
8628 goto err_mdp;
8629 mdp_major = ret;
8630
8631 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
8632 md_probe, NULL, NULL);
8633 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
8634 md_probe, NULL, NULL);
8635
8636 register_reboot_notifier(&md_notifier);
8637 raid_table_header = register_sysctl_table(raid_root_table);
8638
8639 md_geninit();
8640 return 0;
8641
8642 err_mdp:
8643 unregister_blkdev(MD_MAJOR, "md");
8644 err_md:
8645 destroy_workqueue(md_misc_wq);
8646 err_misc_wq:
8647 destroy_workqueue(md_wq);
8648 err_wq:
8649 return ret;
8650 }
8651
8652 #ifndef MODULE
8653
8654 /*
8655 * Searches all registered partitions for autorun RAID arrays
8656 * at boot time.
8657 */
8658
8659 static LIST_HEAD(all_detected_devices);
8660 struct detected_devices_node {
8661 struct list_head list;
8662 dev_t dev;
8663 };
8664
8665 void md_autodetect_dev(dev_t dev)
8666 {
8667 struct detected_devices_node *node_detected_dev;
8668
8669 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8670 if (node_detected_dev) {
8671 node_detected_dev->dev = dev;
8672 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8673 } else {
8674 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8675 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8676 }
8677 }
8678
8679 static void autostart_arrays(int part)
8680 {
8681 struct md_rdev *rdev;
8682 struct detected_devices_node *node_detected_dev;
8683 dev_t dev;
8684 int i_scanned, i_passed;
8685
8686 i_scanned = 0;
8687 i_passed = 0;
8688
8689 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
8690
8691 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8692 i_scanned++;
8693 node_detected_dev = list_entry(all_detected_devices.next,
8694 struct detected_devices_node, list);
8695 list_del(&node_detected_dev->list);
8696 dev = node_detected_dev->dev;
8697 kfree(node_detected_dev);
8698 rdev = md_import_device(dev,0, 90);
8699 if (IS_ERR(rdev))
8700 continue;
8701
8702 if (test_bit(Faulty, &rdev->flags))
8703 continue;
8704
8705 set_bit(AutoDetected, &rdev->flags);
8706 list_add(&rdev->same_set, &pending_raid_disks);
8707 i_passed++;
8708 }
8709
8710 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8711 i_scanned, i_passed);
8712
8713 autorun_devices(part);
8714 }
8715
8716 #endif /* !MODULE */
8717
8718 static __exit void md_exit(void)
8719 {
8720 struct mddev *mddev;
8721 struct list_head *tmp;
8722 int delay = 1;
8723
8724 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
8725 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
8726
8727 unregister_blkdev(MD_MAJOR,"md");
8728 unregister_blkdev(mdp_major, "mdp");
8729 unregister_reboot_notifier(&md_notifier);
8730 unregister_sysctl_table(raid_table_header);
8731
8732 /* We cannot unload the modules while some process is
8733 * waiting for us in select() or poll() - wake them up
8734 */
8735 md_unloading = 1;
8736 while (waitqueue_active(&md_event_waiters)) {
8737 /* not safe to leave yet */
8738 wake_up(&md_event_waiters);
8739 msleep(delay);
8740 delay += delay;
8741 }
8742 remove_proc_entry("mdstat", NULL);
8743
8744 for_each_mddev(mddev, tmp) {
8745 export_array(mddev);
8746 mddev->hold_active = 0;
8747 }
8748 destroy_workqueue(md_misc_wq);
8749 destroy_workqueue(md_wq);
8750 }
8751
8752 subsys_initcall(md_init);
8753 module_exit(md_exit)
8754
8755 static int get_ro(char *buffer, struct kernel_param *kp)
8756 {
8757 return sprintf(buffer, "%d", start_readonly);
8758 }
8759 static int set_ro(const char *val, struct kernel_param *kp)
8760 {
8761 char *e;
8762 int num = simple_strtoul(val, &e, 10);
8763 if (*val && (*e == '\0' || *e == '\n')) {
8764 start_readonly = num;
8765 return 0;
8766 }
8767 return -EINVAL;
8768 }
8769
8770 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8771 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
8772 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
8773
8774 MODULE_LICENSE("GPL");
8775 MODULE_DESCRIPTION("MD RAID framework");
8776 MODULE_ALIAS("md");
8777 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Linux kernel - Deliverables
This page took 0.450065 seconds and 6 git commands to generate.