]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - drivers/md/md.c
Merge branch 'stable-4.10' of git://git.infradead.org/users/pcmoore/selinux into...
[mirror_ubuntu-artful-kernel.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 Errors, Warnings, etc.
35 Please use:
36 pr_crit() for error conditions that risk data loss
37 pr_err() for error conditions that are unexpected, like an IO error
38 or internal inconsistency
39 pr_warn() for error conditions that could have been predicated, like
40 adding a device to an array when it has incompatible metadata
41 pr_info() for every interesting, very rare events, like an array starting
42 or stopping, or resync starting or stopping
43 pr_debug() for everything else.
44
45 */
46
47 #include <linux/kthread.h>
48 #include <linux/blkdev.h>
49 #include <linux/badblocks.h>
50 #include <linux/sysctl.h>
51 #include <linux/seq_file.h>
52 #include <linux/fs.h>
53 #include <linux/poll.h>
54 #include <linux/ctype.h>
55 #include <linux/string.h>
56 #include <linux/hdreg.h>
57 #include <linux/proc_fs.h>
58 #include <linux/random.h>
59 #include <linux/module.h>
60 #include <linux/reboot.h>
61 #include <linux/file.h>
62 #include <linux/compat.h>
63 #include <linux/delay.h>
64 #include <linux/raid/md_p.h>
65 #include <linux/raid/md_u.h>
66 #include <linux/slab.h>
67 #include <trace/events/block.h>
68 #include "md.h"
69 #include "bitmap.h"
70 #include "md-cluster.h"
71
72 #ifndef MODULE
73 static void autostart_arrays(int part);
74 #endif
75
76 /* pers_list is a list of registered personalities protected
77 * by pers_lock.
78 * pers_lock does extra service to protect accesses to
79 * mddev->thread when the mutex cannot be held.
80 */
81 static LIST_HEAD(pers_list);
82 static DEFINE_SPINLOCK(pers_lock);
83
84 struct md_cluster_operations *md_cluster_ops;
85 EXPORT_SYMBOL(md_cluster_ops);
86 struct module *md_cluster_mod;
87 EXPORT_SYMBOL(md_cluster_mod);
88
89 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
90 static struct workqueue_struct *md_wq;
91 static struct workqueue_struct *md_misc_wq;
92
93 static int remove_and_add_spares(struct mddev *mddev,
94 struct md_rdev *this);
95 static void mddev_detach(struct mddev *mddev);
96
97 /*
98 * Default number of read corrections we'll attempt on an rdev
99 * before ejecting it from the array. We divide the read error
100 * count by 2 for every hour elapsed between read errors.
101 */
102 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
103 /*
104 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
105 * is 1000 KB/sec, so the extra system load does not show up that much.
106 * Increase it if you want to have more _guaranteed_ speed. Note that
107 * the RAID driver will use the maximum available bandwidth if the IO
108 * subsystem is idle. There is also an 'absolute maximum' reconstruction
109 * speed limit - in case reconstruction slows down your system despite
110 * idle IO detection.
111 *
112 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
113 * or /sys/block/mdX/md/sync_speed_{min,max}
114 */
115
116 static int sysctl_speed_limit_min = 1000;
117 static int sysctl_speed_limit_max = 200000;
118 static inline int speed_min(struct mddev *mddev)
119 {
120 return mddev->sync_speed_min ?
121 mddev->sync_speed_min : sysctl_speed_limit_min;
122 }
123
124 static inline int speed_max(struct mddev *mddev)
125 {
126 return mddev->sync_speed_max ?
127 mddev->sync_speed_max : sysctl_speed_limit_max;
128 }
129
130 static struct ctl_table_header *raid_table_header;
131
132 static struct ctl_table raid_table[] = {
133 {
134 .procname = "speed_limit_min",
135 .data = &sysctl_speed_limit_min,
136 .maxlen = sizeof(int),
137 .mode = S_IRUGO|S_IWUSR,
138 .proc_handler = proc_dointvec,
139 },
140 {
141 .procname = "speed_limit_max",
142 .data = &sysctl_speed_limit_max,
143 .maxlen = sizeof(int),
144 .mode = S_IRUGO|S_IWUSR,
145 .proc_handler = proc_dointvec,
146 },
147 { }
148 };
149
150 static struct ctl_table raid_dir_table[] = {
151 {
152 .procname = "raid",
153 .maxlen = 0,
154 .mode = S_IRUGO|S_IXUGO,
155 .child = raid_table,
156 },
157 { }
158 };
159
160 static struct ctl_table raid_root_table[] = {
161 {
162 .procname = "dev",
163 .maxlen = 0,
164 .mode = 0555,
165 .child = raid_dir_table,
166 },
167 { }
168 };
169
170 static const struct block_device_operations md_fops;
171
172 static int start_readonly;
173
174 /* bio_clone_mddev
175 * like bio_clone, but with a local bio set
176 */
177
178 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
179 struct mddev *mddev)
180 {
181 struct bio *b;
182
183 if (!mddev || !mddev->bio_set)
184 return bio_alloc(gfp_mask, nr_iovecs);
185
186 b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set);
187 if (!b)
188 return NULL;
189 return b;
190 }
191 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
192
193 struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
194 struct mddev *mddev)
195 {
196 if (!mddev || !mddev->bio_set)
197 return bio_clone(bio, gfp_mask);
198
199 return bio_clone_bioset(bio, gfp_mask, mddev->bio_set);
200 }
201 EXPORT_SYMBOL_GPL(bio_clone_mddev);
202
203 /*
204 * We have a system wide 'event count' that is incremented
205 * on any 'interesting' event, and readers of /proc/mdstat
206 * can use 'poll' or 'select' to find out when the event
207 * count increases.
208 *
209 * Events are:
210 * start array, stop array, error, add device, remove device,
211 * start build, activate spare
212 */
213 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
214 static atomic_t md_event_count;
215 void md_new_event(struct mddev *mddev)
216 {
217 atomic_inc(&md_event_count);
218 wake_up(&md_event_waiters);
219 }
220 EXPORT_SYMBOL_GPL(md_new_event);
221
222 /*
223 * Enables to iterate over all existing md arrays
224 * all_mddevs_lock protects this list.
225 */
226 static LIST_HEAD(all_mddevs);
227 static DEFINE_SPINLOCK(all_mddevs_lock);
228
229 /*
230 * iterates through all used mddevs in the system.
231 * We take care to grab the all_mddevs_lock whenever navigating
232 * the list, and to always hold a refcount when unlocked.
233 * Any code which breaks out of this loop while own
234 * a reference to the current mddev and must mddev_put it.
235 */
236 #define for_each_mddev(_mddev,_tmp) \
237 \
238 for (({ spin_lock(&all_mddevs_lock); \
239 _tmp = all_mddevs.next; \
240 _mddev = NULL;}); \
241 ({ if (_tmp != &all_mddevs) \
242 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
243 spin_unlock(&all_mddevs_lock); \
244 if (_mddev) mddev_put(_mddev); \
245 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
246 _tmp != &all_mddevs;}); \
247 ({ spin_lock(&all_mddevs_lock); \
248 _tmp = _tmp->next;}) \
249 )
250
251 /* Rather than calling directly into the personality make_request function,
252 * IO requests come here first so that we can check if the device is
253 * being suspended pending a reconfiguration.
254 * We hold a refcount over the call to ->make_request. By the time that
255 * call has finished, the bio has been linked into some internal structure
256 * and so is visible to ->quiesce(), so we don't need the refcount any more.
257 */
258 static blk_qc_t md_make_request(struct request_queue *q, struct bio *bio)
259 {
260 const int rw = bio_data_dir(bio);
261 struct mddev *mddev = q->queuedata;
262 unsigned int sectors;
263 int cpu;
264
265 blk_queue_split(q, &bio, q->bio_split);
266
267 if (mddev == NULL || mddev->pers == NULL) {
268 bio_io_error(bio);
269 return BLK_QC_T_NONE;
270 }
271 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
272 if (bio_sectors(bio) != 0)
273 bio->bi_error = -EROFS;
274 bio_endio(bio);
275 return BLK_QC_T_NONE;
276 }
277 smp_rmb(); /* Ensure implications of 'active' are visible */
278 rcu_read_lock();
279 if (mddev->suspended) {
280 DEFINE_WAIT(__wait);
281 for (;;) {
282 prepare_to_wait(&mddev->sb_wait, &__wait,
283 TASK_UNINTERRUPTIBLE);
284 if (!mddev->suspended)
285 break;
286 rcu_read_unlock();
287 schedule();
288 rcu_read_lock();
289 }
290 finish_wait(&mddev->sb_wait, &__wait);
291 }
292 atomic_inc(&mddev->active_io);
293 rcu_read_unlock();
294
295 /*
296 * save the sectors now since our bio can
297 * go away inside make_request
298 */
299 sectors = bio_sectors(bio);
300 /* bio could be mergeable after passing to underlayer */
301 bio->bi_opf &= ~REQ_NOMERGE;
302 mddev->pers->make_request(mddev, bio);
303
304 cpu = part_stat_lock();
305 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
306 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
307 part_stat_unlock();
308
309 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
310 wake_up(&mddev->sb_wait);
311
312 return BLK_QC_T_NONE;
313 }
314
315 /* mddev_suspend makes sure no new requests are submitted
316 * to the device, and that any requests that have been submitted
317 * are completely handled.
318 * Once mddev_detach() is called and completes, the module will be
319 * completely unused.
320 */
321 void mddev_suspend(struct mddev *mddev)
322 {
323 WARN_ON_ONCE(mddev->thread && current == mddev->thread->tsk);
324 if (mddev->suspended++)
325 return;
326 synchronize_rcu();
327 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
328 mddev->pers->quiesce(mddev, 1);
329
330 del_timer_sync(&mddev->safemode_timer);
331 }
332 EXPORT_SYMBOL_GPL(mddev_suspend);
333
334 void mddev_resume(struct mddev *mddev)
335 {
336 if (--mddev->suspended)
337 return;
338 wake_up(&mddev->sb_wait);
339 mddev->pers->quiesce(mddev, 0);
340
341 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
342 md_wakeup_thread(mddev->thread);
343 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
344 }
345 EXPORT_SYMBOL_GPL(mddev_resume);
346
347 int mddev_congested(struct mddev *mddev, int bits)
348 {
349 struct md_personality *pers = mddev->pers;
350 int ret = 0;
351
352 rcu_read_lock();
353 if (mddev->suspended)
354 ret = 1;
355 else if (pers && pers->congested)
356 ret = pers->congested(mddev, bits);
357 rcu_read_unlock();
358 return ret;
359 }
360 EXPORT_SYMBOL_GPL(mddev_congested);
361 static int md_congested(void *data, int bits)
362 {
363 struct mddev *mddev = data;
364 return mddev_congested(mddev, bits);
365 }
366
367 /*
368 * Generic flush handling for md
369 */
370
371 static void md_end_flush(struct bio *bio)
372 {
373 struct md_rdev *rdev = bio->bi_private;
374 struct mddev *mddev = rdev->mddev;
375
376 rdev_dec_pending(rdev, mddev);
377
378 if (atomic_dec_and_test(&mddev->flush_pending)) {
379 /* The pre-request flush has finished */
380 queue_work(md_wq, &mddev->flush_work);
381 }
382 bio_put(bio);
383 }
384
385 static void md_submit_flush_data(struct work_struct *ws);
386
387 static void submit_flushes(struct work_struct *ws)
388 {
389 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
390 struct md_rdev *rdev;
391
392 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
393 atomic_set(&mddev->flush_pending, 1);
394 rcu_read_lock();
395 rdev_for_each_rcu(rdev, mddev)
396 if (rdev->raid_disk >= 0 &&
397 !test_bit(Faulty, &rdev->flags)) {
398 /* Take two references, one is dropped
399 * when request finishes, one after
400 * we reclaim rcu_read_lock
401 */
402 struct bio *bi;
403 atomic_inc(&rdev->nr_pending);
404 atomic_inc(&rdev->nr_pending);
405 rcu_read_unlock();
406 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
407 bi->bi_end_io = md_end_flush;
408 bi->bi_private = rdev;
409 bi->bi_bdev = rdev->bdev;
410 bi->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
411 atomic_inc(&mddev->flush_pending);
412 submit_bio(bi);
413 rcu_read_lock();
414 rdev_dec_pending(rdev, mddev);
415 }
416 rcu_read_unlock();
417 if (atomic_dec_and_test(&mddev->flush_pending))
418 queue_work(md_wq, &mddev->flush_work);
419 }
420
421 static void md_submit_flush_data(struct work_struct *ws)
422 {
423 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
424 struct bio *bio = mddev->flush_bio;
425
426 if (bio->bi_iter.bi_size == 0)
427 /* an empty barrier - all done */
428 bio_endio(bio);
429 else {
430 bio->bi_opf &= ~REQ_PREFLUSH;
431 mddev->pers->make_request(mddev, bio);
432 }
433
434 mddev->flush_bio = NULL;
435 wake_up(&mddev->sb_wait);
436 }
437
438 void md_flush_request(struct mddev *mddev, struct bio *bio)
439 {
440 spin_lock_irq(&mddev->lock);
441 wait_event_lock_irq(mddev->sb_wait,
442 !mddev->flush_bio,
443 mddev->lock);
444 mddev->flush_bio = bio;
445 spin_unlock_irq(&mddev->lock);
446
447 INIT_WORK(&mddev->flush_work, submit_flushes);
448 queue_work(md_wq, &mddev->flush_work);
449 }
450 EXPORT_SYMBOL(md_flush_request);
451
452 void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
453 {
454 struct mddev *mddev = cb->data;
455 md_wakeup_thread(mddev->thread);
456 kfree(cb);
457 }
458 EXPORT_SYMBOL(md_unplug);
459
460 static inline struct mddev *mddev_get(struct mddev *mddev)
461 {
462 atomic_inc(&mddev->active);
463 return mddev;
464 }
465
466 static void mddev_delayed_delete(struct work_struct *ws);
467
468 static void mddev_put(struct mddev *mddev)
469 {
470 struct bio_set *bs = NULL;
471
472 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
473 return;
474 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
475 mddev->ctime == 0 && !mddev->hold_active) {
476 /* Array is not configured at all, and not held active,
477 * so destroy it */
478 list_del_init(&mddev->all_mddevs);
479 bs = mddev->bio_set;
480 mddev->bio_set = NULL;
481 if (mddev->gendisk) {
482 /* We did a probe so need to clean up. Call
483 * queue_work inside the spinlock so that
484 * flush_workqueue() after mddev_find will
485 * succeed in waiting for the work to be done.
486 */
487 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
488 queue_work(md_misc_wq, &mddev->del_work);
489 } else
490 kfree(mddev);
491 }
492 spin_unlock(&all_mddevs_lock);
493 if (bs)
494 bioset_free(bs);
495 }
496
497 static void md_safemode_timeout(unsigned long data);
498
499 void mddev_init(struct mddev *mddev)
500 {
501 mutex_init(&mddev->open_mutex);
502 mutex_init(&mddev->reconfig_mutex);
503 mutex_init(&mddev->bitmap_info.mutex);
504 INIT_LIST_HEAD(&mddev->disks);
505 INIT_LIST_HEAD(&mddev->all_mddevs);
506 setup_timer(&mddev->safemode_timer, md_safemode_timeout,
507 (unsigned long) mddev);
508 atomic_set(&mddev->active, 1);
509 atomic_set(&mddev->openers, 0);
510 atomic_set(&mddev->active_io, 0);
511 spin_lock_init(&mddev->lock);
512 atomic_set(&mddev->flush_pending, 0);
513 init_waitqueue_head(&mddev->sb_wait);
514 init_waitqueue_head(&mddev->recovery_wait);
515 mddev->reshape_position = MaxSector;
516 mddev->reshape_backwards = 0;
517 mddev->last_sync_action = "none";
518 mddev->resync_min = 0;
519 mddev->resync_max = MaxSector;
520 mddev->level = LEVEL_NONE;
521 }
522 EXPORT_SYMBOL_GPL(mddev_init);
523
524 static struct mddev *mddev_find(dev_t unit)
525 {
526 struct mddev *mddev, *new = NULL;
527
528 if (unit && MAJOR(unit) != MD_MAJOR)
529 unit &= ~((1<<MdpMinorShift)-1);
530
531 retry:
532 spin_lock(&all_mddevs_lock);
533
534 if (unit) {
535 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
536 if (mddev->unit == unit) {
537 mddev_get(mddev);
538 spin_unlock(&all_mddevs_lock);
539 kfree(new);
540 return mddev;
541 }
542
543 if (new) {
544 list_add(&new->all_mddevs, &all_mddevs);
545 spin_unlock(&all_mddevs_lock);
546 new->hold_active = UNTIL_IOCTL;
547 return new;
548 }
549 } else if (new) {
550 /* find an unused unit number */
551 static int next_minor = 512;
552 int start = next_minor;
553 int is_free = 0;
554 int dev = 0;
555 while (!is_free) {
556 dev = MKDEV(MD_MAJOR, next_minor);
557 next_minor++;
558 if (next_minor > MINORMASK)
559 next_minor = 0;
560 if (next_minor == start) {
561 /* Oh dear, all in use. */
562 spin_unlock(&all_mddevs_lock);
563 kfree(new);
564 return NULL;
565 }
566
567 is_free = 1;
568 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
569 if (mddev->unit == dev) {
570 is_free = 0;
571 break;
572 }
573 }
574 new->unit = dev;
575 new->md_minor = MINOR(dev);
576 new->hold_active = UNTIL_STOP;
577 list_add(&new->all_mddevs, &all_mddevs);
578 spin_unlock(&all_mddevs_lock);
579 return new;
580 }
581 spin_unlock(&all_mddevs_lock);
582
583 new = kzalloc(sizeof(*new), GFP_KERNEL);
584 if (!new)
585 return NULL;
586
587 new->unit = unit;
588 if (MAJOR(unit) == MD_MAJOR)
589 new->md_minor = MINOR(unit);
590 else
591 new->md_minor = MINOR(unit) >> MdpMinorShift;
592
593 mddev_init(new);
594
595 goto retry;
596 }
597
598 static struct attribute_group md_redundancy_group;
599
600 void mddev_unlock(struct mddev *mddev)
601 {
602 if (mddev->to_remove) {
603 /* These cannot be removed under reconfig_mutex as
604 * an access to the files will try to take reconfig_mutex
605 * while holding the file unremovable, which leads to
606 * a deadlock.
607 * So hold set sysfs_active while the remove in happeing,
608 * and anything else which might set ->to_remove or my
609 * otherwise change the sysfs namespace will fail with
610 * -EBUSY if sysfs_active is still set.
611 * We set sysfs_active under reconfig_mutex and elsewhere
612 * test it under the same mutex to ensure its correct value
613 * is seen.
614 */
615 struct attribute_group *to_remove = mddev->to_remove;
616 mddev->to_remove = NULL;
617 mddev->sysfs_active = 1;
618 mutex_unlock(&mddev->reconfig_mutex);
619
620 if (mddev->kobj.sd) {
621 if (to_remove != &md_redundancy_group)
622 sysfs_remove_group(&mddev->kobj, to_remove);
623 if (mddev->pers == NULL ||
624 mddev->pers->sync_request == NULL) {
625 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
626 if (mddev->sysfs_action)
627 sysfs_put(mddev->sysfs_action);
628 mddev->sysfs_action = NULL;
629 }
630 }
631 mddev->sysfs_active = 0;
632 } else
633 mutex_unlock(&mddev->reconfig_mutex);
634
635 /* As we've dropped the mutex we need a spinlock to
636 * make sure the thread doesn't disappear
637 */
638 spin_lock(&pers_lock);
639 md_wakeup_thread(mddev->thread);
640 spin_unlock(&pers_lock);
641 }
642 EXPORT_SYMBOL_GPL(mddev_unlock);
643
644 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
645 {
646 struct md_rdev *rdev;
647
648 rdev_for_each_rcu(rdev, mddev)
649 if (rdev->desc_nr == nr)
650 return rdev;
651
652 return NULL;
653 }
654 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
655
656 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
657 {
658 struct md_rdev *rdev;
659
660 rdev_for_each(rdev, mddev)
661 if (rdev->bdev->bd_dev == dev)
662 return rdev;
663
664 return NULL;
665 }
666
667 static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
668 {
669 struct md_rdev *rdev;
670
671 rdev_for_each_rcu(rdev, mddev)
672 if (rdev->bdev->bd_dev == dev)
673 return rdev;
674
675 return NULL;
676 }
677
678 static struct md_personality *find_pers(int level, char *clevel)
679 {
680 struct md_personality *pers;
681 list_for_each_entry(pers, &pers_list, list) {
682 if (level != LEVEL_NONE && pers->level == level)
683 return pers;
684 if (strcmp(pers->name, clevel)==0)
685 return pers;
686 }
687 return NULL;
688 }
689
690 /* return the offset of the super block in 512byte sectors */
691 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
692 {
693 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
694 return MD_NEW_SIZE_SECTORS(num_sectors);
695 }
696
697 static int alloc_disk_sb(struct md_rdev *rdev)
698 {
699 rdev->sb_page = alloc_page(GFP_KERNEL);
700 if (!rdev->sb_page)
701 return -ENOMEM;
702 return 0;
703 }
704
705 void md_rdev_clear(struct md_rdev *rdev)
706 {
707 if (rdev->sb_page) {
708 put_page(rdev->sb_page);
709 rdev->sb_loaded = 0;
710 rdev->sb_page = NULL;
711 rdev->sb_start = 0;
712 rdev->sectors = 0;
713 }
714 if (rdev->bb_page) {
715 put_page(rdev->bb_page);
716 rdev->bb_page = NULL;
717 }
718 badblocks_exit(&rdev->badblocks);
719 }
720 EXPORT_SYMBOL_GPL(md_rdev_clear);
721
722 static void super_written(struct bio *bio)
723 {
724 struct md_rdev *rdev = bio->bi_private;
725 struct mddev *mddev = rdev->mddev;
726
727 if (bio->bi_error) {
728 pr_err("md: super_written gets error=%d\n", bio->bi_error);
729 md_error(mddev, rdev);
730 if (!test_bit(Faulty, &rdev->flags)
731 && (bio->bi_opf & MD_FAILFAST)) {
732 set_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags);
733 set_bit(LastDev, &rdev->flags);
734 }
735 } else
736 clear_bit(LastDev, &rdev->flags);
737
738 if (atomic_dec_and_test(&mddev->pending_writes))
739 wake_up(&mddev->sb_wait);
740 rdev_dec_pending(rdev, mddev);
741 bio_put(bio);
742 }
743
744 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
745 sector_t sector, int size, struct page *page)
746 {
747 /* write first size bytes of page to sector of rdev
748 * Increment mddev->pending_writes before returning
749 * and decrement it on completion, waking up sb_wait
750 * if zero is reached.
751 * If an error occurred, call md_error
752 */
753 struct bio *bio;
754 int ff = 0;
755
756 if (test_bit(Faulty, &rdev->flags))
757 return;
758
759 bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
760
761 atomic_inc(&rdev->nr_pending);
762
763 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
764 bio->bi_iter.bi_sector = sector;
765 bio_add_page(bio, page, size, 0);
766 bio->bi_private = rdev;
767 bio->bi_end_io = super_written;
768
769 if (test_bit(MD_FAILFAST_SUPPORTED, &mddev->flags) &&
770 test_bit(FailFast, &rdev->flags) &&
771 !test_bit(LastDev, &rdev->flags))
772 ff = MD_FAILFAST;
773 bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_FUA | ff;
774
775 atomic_inc(&mddev->pending_writes);
776 submit_bio(bio);
777 }
778
779 int md_super_wait(struct mddev *mddev)
780 {
781 /* wait for all superblock writes that were scheduled to complete */
782 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
783 if (test_and_clear_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags))
784 return -EAGAIN;
785 return 0;
786 }
787
788 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
789 struct page *page, int op, int op_flags, bool metadata_op)
790 {
791 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
792 int ret;
793
794 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
795 rdev->meta_bdev : rdev->bdev;
796 bio_set_op_attrs(bio, op, op_flags);
797 if (metadata_op)
798 bio->bi_iter.bi_sector = sector + rdev->sb_start;
799 else if (rdev->mddev->reshape_position != MaxSector &&
800 (rdev->mddev->reshape_backwards ==
801 (sector >= rdev->mddev->reshape_position)))
802 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
803 else
804 bio->bi_iter.bi_sector = sector + rdev->data_offset;
805 bio_add_page(bio, page, size, 0);
806
807 submit_bio_wait(bio);
808
809 ret = !bio->bi_error;
810 bio_put(bio);
811 return ret;
812 }
813 EXPORT_SYMBOL_GPL(sync_page_io);
814
815 static int read_disk_sb(struct md_rdev *rdev, int size)
816 {
817 char b[BDEVNAME_SIZE];
818
819 if (rdev->sb_loaded)
820 return 0;
821
822 if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true))
823 goto fail;
824 rdev->sb_loaded = 1;
825 return 0;
826
827 fail:
828 pr_err("md: disabled device %s, could not read superblock.\n",
829 bdevname(rdev->bdev,b));
830 return -EINVAL;
831 }
832
833 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
834 {
835 return sb1->set_uuid0 == sb2->set_uuid0 &&
836 sb1->set_uuid1 == sb2->set_uuid1 &&
837 sb1->set_uuid2 == sb2->set_uuid2 &&
838 sb1->set_uuid3 == sb2->set_uuid3;
839 }
840
841 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
842 {
843 int ret;
844 mdp_super_t *tmp1, *tmp2;
845
846 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
847 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
848
849 if (!tmp1 || !tmp2) {
850 ret = 0;
851 goto abort;
852 }
853
854 *tmp1 = *sb1;
855 *tmp2 = *sb2;
856
857 /*
858 * nr_disks is not constant
859 */
860 tmp1->nr_disks = 0;
861 tmp2->nr_disks = 0;
862
863 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
864 abort:
865 kfree(tmp1);
866 kfree(tmp2);
867 return ret;
868 }
869
870 static u32 md_csum_fold(u32 csum)
871 {
872 csum = (csum & 0xffff) + (csum >> 16);
873 return (csum & 0xffff) + (csum >> 16);
874 }
875
876 static unsigned int calc_sb_csum(mdp_super_t *sb)
877 {
878 u64 newcsum = 0;
879 u32 *sb32 = (u32*)sb;
880 int i;
881 unsigned int disk_csum, csum;
882
883 disk_csum = sb->sb_csum;
884 sb->sb_csum = 0;
885
886 for (i = 0; i < MD_SB_BYTES/4 ; i++)
887 newcsum += sb32[i];
888 csum = (newcsum & 0xffffffff) + (newcsum>>32);
889
890 #ifdef CONFIG_ALPHA
891 /* This used to use csum_partial, which was wrong for several
892 * reasons including that different results are returned on
893 * different architectures. It isn't critical that we get exactly
894 * the same return value as before (we always csum_fold before
895 * testing, and that removes any differences). However as we
896 * know that csum_partial always returned a 16bit value on
897 * alphas, do a fold to maximise conformity to previous behaviour.
898 */
899 sb->sb_csum = md_csum_fold(disk_csum);
900 #else
901 sb->sb_csum = disk_csum;
902 #endif
903 return csum;
904 }
905
906 /*
907 * Handle superblock details.
908 * We want to be able to handle multiple superblock formats
909 * so we have a common interface to them all, and an array of
910 * different handlers.
911 * We rely on user-space to write the initial superblock, and support
912 * reading and updating of superblocks.
913 * Interface methods are:
914 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
915 * loads and validates a superblock on dev.
916 * if refdev != NULL, compare superblocks on both devices
917 * Return:
918 * 0 - dev has a superblock that is compatible with refdev
919 * 1 - dev has a superblock that is compatible and newer than refdev
920 * so dev should be used as the refdev in future
921 * -EINVAL superblock incompatible or invalid
922 * -othererror e.g. -EIO
923 *
924 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
925 * Verify that dev is acceptable into mddev.
926 * The first time, mddev->raid_disks will be 0, and data from
927 * dev should be merged in. Subsequent calls check that dev
928 * is new enough. Return 0 or -EINVAL
929 *
930 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
931 * Update the superblock for rdev with data in mddev
932 * This does not write to disc.
933 *
934 */
935
936 struct super_type {
937 char *name;
938 struct module *owner;
939 int (*load_super)(struct md_rdev *rdev,
940 struct md_rdev *refdev,
941 int minor_version);
942 int (*validate_super)(struct mddev *mddev,
943 struct md_rdev *rdev);
944 void (*sync_super)(struct mddev *mddev,
945 struct md_rdev *rdev);
946 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
947 sector_t num_sectors);
948 int (*allow_new_offset)(struct md_rdev *rdev,
949 unsigned long long new_offset);
950 };
951
952 /*
953 * Check that the given mddev has no bitmap.
954 *
955 * This function is called from the run method of all personalities that do not
956 * support bitmaps. It prints an error message and returns non-zero if mddev
957 * has a bitmap. Otherwise, it returns 0.
958 *
959 */
960 int md_check_no_bitmap(struct mddev *mddev)
961 {
962 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
963 return 0;
964 pr_warn("%s: bitmaps are not supported for %s\n",
965 mdname(mddev), mddev->pers->name);
966 return 1;
967 }
968 EXPORT_SYMBOL(md_check_no_bitmap);
969
970 /*
971 * load_super for 0.90.0
972 */
973 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
974 {
975 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
976 mdp_super_t *sb;
977 int ret;
978
979 /*
980 * Calculate the position of the superblock (512byte sectors),
981 * it's at the end of the disk.
982 *
983 * It also happens to be a multiple of 4Kb.
984 */
985 rdev->sb_start = calc_dev_sboffset(rdev);
986
987 ret = read_disk_sb(rdev, MD_SB_BYTES);
988 if (ret)
989 return ret;
990
991 ret = -EINVAL;
992
993 bdevname(rdev->bdev, b);
994 sb = page_address(rdev->sb_page);
995
996 if (sb->md_magic != MD_SB_MAGIC) {
997 pr_warn("md: invalid raid superblock magic on %s\n", b);
998 goto abort;
999 }
1000
1001 if (sb->major_version != 0 ||
1002 sb->minor_version < 90 ||
1003 sb->minor_version > 91) {
1004 pr_warn("Bad version number %d.%d on %s\n",
1005 sb->major_version, sb->minor_version, b);
1006 goto abort;
1007 }
1008
1009 if (sb->raid_disks <= 0)
1010 goto abort;
1011
1012 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1013 pr_warn("md: invalid superblock checksum on %s\n", b);
1014 goto abort;
1015 }
1016
1017 rdev->preferred_minor = sb->md_minor;
1018 rdev->data_offset = 0;
1019 rdev->new_data_offset = 0;
1020 rdev->sb_size = MD_SB_BYTES;
1021 rdev->badblocks.shift = -1;
1022
1023 if (sb->level == LEVEL_MULTIPATH)
1024 rdev->desc_nr = -1;
1025 else
1026 rdev->desc_nr = sb->this_disk.number;
1027
1028 if (!refdev) {
1029 ret = 1;
1030 } else {
1031 __u64 ev1, ev2;
1032 mdp_super_t *refsb = page_address(refdev->sb_page);
1033 if (!uuid_equal(refsb, sb)) {
1034 pr_warn("md: %s has different UUID to %s\n",
1035 b, bdevname(refdev->bdev,b2));
1036 goto abort;
1037 }
1038 if (!sb_equal(refsb, sb)) {
1039 pr_warn("md: %s has same UUID but different superblock to %s\n",
1040 b, bdevname(refdev->bdev, b2));
1041 goto abort;
1042 }
1043 ev1 = md_event(sb);
1044 ev2 = md_event(refsb);
1045 if (ev1 > ev2)
1046 ret = 1;
1047 else
1048 ret = 0;
1049 }
1050 rdev->sectors = rdev->sb_start;
1051 /* Limit to 4TB as metadata cannot record more than that.
1052 * (not needed for Linear and RAID0 as metadata doesn't
1053 * record this size)
1054 */
1055 if (IS_ENABLED(CONFIG_LBDAF) && (u64)rdev->sectors >= (2ULL << 32) &&
1056 sb->level >= 1)
1057 rdev->sectors = (sector_t)(2ULL << 32) - 2;
1058
1059 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1060 /* "this cannot possibly happen" ... */
1061 ret = -EINVAL;
1062
1063 abort:
1064 return ret;
1065 }
1066
1067 /*
1068 * validate_super for 0.90.0
1069 */
1070 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1071 {
1072 mdp_disk_t *desc;
1073 mdp_super_t *sb = page_address(rdev->sb_page);
1074 __u64 ev1 = md_event(sb);
1075
1076 rdev->raid_disk = -1;
1077 clear_bit(Faulty, &rdev->flags);
1078 clear_bit(In_sync, &rdev->flags);
1079 clear_bit(Bitmap_sync, &rdev->flags);
1080 clear_bit(WriteMostly, &rdev->flags);
1081
1082 if (mddev->raid_disks == 0) {
1083 mddev->major_version = 0;
1084 mddev->minor_version = sb->minor_version;
1085 mddev->patch_version = sb->patch_version;
1086 mddev->external = 0;
1087 mddev->chunk_sectors = sb->chunk_size >> 9;
1088 mddev->ctime = sb->ctime;
1089 mddev->utime = sb->utime;
1090 mddev->level = sb->level;
1091 mddev->clevel[0] = 0;
1092 mddev->layout = sb->layout;
1093 mddev->raid_disks = sb->raid_disks;
1094 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1095 mddev->events = ev1;
1096 mddev->bitmap_info.offset = 0;
1097 mddev->bitmap_info.space = 0;
1098 /* bitmap can use 60 K after the 4K superblocks */
1099 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1100 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1101 mddev->reshape_backwards = 0;
1102
1103 if (mddev->minor_version >= 91) {
1104 mddev->reshape_position = sb->reshape_position;
1105 mddev->delta_disks = sb->delta_disks;
1106 mddev->new_level = sb->new_level;
1107 mddev->new_layout = sb->new_layout;
1108 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1109 if (mddev->delta_disks < 0)
1110 mddev->reshape_backwards = 1;
1111 } else {
1112 mddev->reshape_position = MaxSector;
1113 mddev->delta_disks = 0;
1114 mddev->new_level = mddev->level;
1115 mddev->new_layout = mddev->layout;
1116 mddev->new_chunk_sectors = mddev->chunk_sectors;
1117 }
1118
1119 if (sb->state & (1<<MD_SB_CLEAN))
1120 mddev->recovery_cp = MaxSector;
1121 else {
1122 if (sb->events_hi == sb->cp_events_hi &&
1123 sb->events_lo == sb->cp_events_lo) {
1124 mddev->recovery_cp = sb->recovery_cp;
1125 } else
1126 mddev->recovery_cp = 0;
1127 }
1128
1129 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1130 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1131 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1132 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1133
1134 mddev->max_disks = MD_SB_DISKS;
1135
1136 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1137 mddev->bitmap_info.file == NULL) {
1138 mddev->bitmap_info.offset =
1139 mddev->bitmap_info.default_offset;
1140 mddev->bitmap_info.space =
1141 mddev->bitmap_info.default_space;
1142 }
1143
1144 } else if (mddev->pers == NULL) {
1145 /* Insist on good event counter while assembling, except
1146 * for spares (which don't need an event count) */
1147 ++ev1;
1148 if (sb->disks[rdev->desc_nr].state & (
1149 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1150 if (ev1 < mddev->events)
1151 return -EINVAL;
1152 } else if (mddev->bitmap) {
1153 /* if adding to array with a bitmap, then we can accept an
1154 * older device ... but not too old.
1155 */
1156 if (ev1 < mddev->bitmap->events_cleared)
1157 return 0;
1158 if (ev1 < mddev->events)
1159 set_bit(Bitmap_sync, &rdev->flags);
1160 } else {
1161 if (ev1 < mddev->events)
1162 /* just a hot-add of a new device, leave raid_disk at -1 */
1163 return 0;
1164 }
1165
1166 if (mddev->level != LEVEL_MULTIPATH) {
1167 desc = sb->disks + rdev->desc_nr;
1168
1169 if (desc->state & (1<<MD_DISK_FAULTY))
1170 set_bit(Faulty, &rdev->flags);
1171 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1172 desc->raid_disk < mddev->raid_disks */) {
1173 set_bit(In_sync, &rdev->flags);
1174 rdev->raid_disk = desc->raid_disk;
1175 rdev->saved_raid_disk = desc->raid_disk;
1176 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1177 /* active but not in sync implies recovery up to
1178 * reshape position. We don't know exactly where
1179 * that is, so set to zero for now */
1180 if (mddev->minor_version >= 91) {
1181 rdev->recovery_offset = 0;
1182 rdev->raid_disk = desc->raid_disk;
1183 }
1184 }
1185 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1186 set_bit(WriteMostly, &rdev->flags);
1187 if (desc->state & (1<<MD_DISK_FAILFAST))
1188 set_bit(FailFast, &rdev->flags);
1189 } else /* MULTIPATH are always insync */
1190 set_bit(In_sync, &rdev->flags);
1191 return 0;
1192 }
1193
1194 /*
1195 * sync_super for 0.90.0
1196 */
1197 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1198 {
1199 mdp_super_t *sb;
1200 struct md_rdev *rdev2;
1201 int next_spare = mddev->raid_disks;
1202
1203 /* make rdev->sb match mddev data..
1204 *
1205 * 1/ zero out disks
1206 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1207 * 3/ any empty disks < next_spare become removed
1208 *
1209 * disks[0] gets initialised to REMOVED because
1210 * we cannot be sure from other fields if it has
1211 * been initialised or not.
1212 */
1213 int i;
1214 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1215
1216 rdev->sb_size = MD_SB_BYTES;
1217
1218 sb = page_address(rdev->sb_page);
1219
1220 memset(sb, 0, sizeof(*sb));
1221
1222 sb->md_magic = MD_SB_MAGIC;
1223 sb->major_version = mddev->major_version;
1224 sb->patch_version = mddev->patch_version;
1225 sb->gvalid_words = 0; /* ignored */
1226 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1227 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1228 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1229 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1230
1231 sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1232 sb->level = mddev->level;
1233 sb->size = mddev->dev_sectors / 2;
1234 sb->raid_disks = mddev->raid_disks;
1235 sb->md_minor = mddev->md_minor;
1236 sb->not_persistent = 0;
1237 sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1238 sb->state = 0;
1239 sb->events_hi = (mddev->events>>32);
1240 sb->events_lo = (u32)mddev->events;
1241
1242 if (mddev->reshape_position == MaxSector)
1243 sb->minor_version = 90;
1244 else {
1245 sb->minor_version = 91;
1246 sb->reshape_position = mddev->reshape_position;
1247 sb->new_level = mddev->new_level;
1248 sb->delta_disks = mddev->delta_disks;
1249 sb->new_layout = mddev->new_layout;
1250 sb->new_chunk = mddev->new_chunk_sectors << 9;
1251 }
1252 mddev->minor_version = sb->minor_version;
1253 if (mddev->in_sync)
1254 {
1255 sb->recovery_cp = mddev->recovery_cp;
1256 sb->cp_events_hi = (mddev->events>>32);
1257 sb->cp_events_lo = (u32)mddev->events;
1258 if (mddev->recovery_cp == MaxSector)
1259 sb->state = (1<< MD_SB_CLEAN);
1260 } else
1261 sb->recovery_cp = 0;
1262
1263 sb->layout = mddev->layout;
1264 sb->chunk_size = mddev->chunk_sectors << 9;
1265
1266 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1267 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1268
1269 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1270 rdev_for_each(rdev2, mddev) {
1271 mdp_disk_t *d;
1272 int desc_nr;
1273 int is_active = test_bit(In_sync, &rdev2->flags);
1274
1275 if (rdev2->raid_disk >= 0 &&
1276 sb->minor_version >= 91)
1277 /* we have nowhere to store the recovery_offset,
1278 * but if it is not below the reshape_position,
1279 * we can piggy-back on that.
1280 */
1281 is_active = 1;
1282 if (rdev2->raid_disk < 0 ||
1283 test_bit(Faulty, &rdev2->flags))
1284 is_active = 0;
1285 if (is_active)
1286 desc_nr = rdev2->raid_disk;
1287 else
1288 desc_nr = next_spare++;
1289 rdev2->desc_nr = desc_nr;
1290 d = &sb->disks[rdev2->desc_nr];
1291 nr_disks++;
1292 d->number = rdev2->desc_nr;
1293 d->major = MAJOR(rdev2->bdev->bd_dev);
1294 d->minor = MINOR(rdev2->bdev->bd_dev);
1295 if (is_active)
1296 d->raid_disk = rdev2->raid_disk;
1297 else
1298 d->raid_disk = rdev2->desc_nr; /* compatibility */
1299 if (test_bit(Faulty, &rdev2->flags))
1300 d->state = (1<<MD_DISK_FAULTY);
1301 else if (is_active) {
1302 d->state = (1<<MD_DISK_ACTIVE);
1303 if (test_bit(In_sync, &rdev2->flags))
1304 d->state |= (1<<MD_DISK_SYNC);
1305 active++;
1306 working++;
1307 } else {
1308 d->state = 0;
1309 spare++;
1310 working++;
1311 }
1312 if (test_bit(WriteMostly, &rdev2->flags))
1313 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1314 if (test_bit(FailFast, &rdev2->flags))
1315 d->state |= (1<<MD_DISK_FAILFAST);
1316 }
1317 /* now set the "removed" and "faulty" bits on any missing devices */
1318 for (i=0 ; i < mddev->raid_disks ; i++) {
1319 mdp_disk_t *d = &sb->disks[i];
1320 if (d->state == 0 && d->number == 0) {
1321 d->number = i;
1322 d->raid_disk = i;
1323 d->state = (1<<MD_DISK_REMOVED);
1324 d->state |= (1<<MD_DISK_FAULTY);
1325 failed++;
1326 }
1327 }
1328 sb->nr_disks = nr_disks;
1329 sb->active_disks = active;
1330 sb->working_disks = working;
1331 sb->failed_disks = failed;
1332 sb->spare_disks = spare;
1333
1334 sb->this_disk = sb->disks[rdev->desc_nr];
1335 sb->sb_csum = calc_sb_csum(sb);
1336 }
1337
1338 /*
1339 * rdev_size_change for 0.90.0
1340 */
1341 static unsigned long long
1342 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1343 {
1344 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1345 return 0; /* component must fit device */
1346 if (rdev->mddev->bitmap_info.offset)
1347 return 0; /* can't move bitmap */
1348 rdev->sb_start = calc_dev_sboffset(rdev);
1349 if (!num_sectors || num_sectors > rdev->sb_start)
1350 num_sectors = rdev->sb_start;
1351 /* Limit to 4TB as metadata cannot record more than that.
1352 * 4TB == 2^32 KB, or 2*2^32 sectors.
1353 */
1354 if (IS_ENABLED(CONFIG_LBDAF) && (u64)num_sectors >= (2ULL << 32) &&
1355 rdev->mddev->level >= 1)
1356 num_sectors = (sector_t)(2ULL << 32) - 2;
1357 do {
1358 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1359 rdev->sb_page);
1360 } while (md_super_wait(rdev->mddev) < 0);
1361 return num_sectors;
1362 }
1363
1364 static int
1365 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1366 {
1367 /* non-zero offset changes not possible with v0.90 */
1368 return new_offset == 0;
1369 }
1370
1371 /*
1372 * version 1 superblock
1373 */
1374
1375 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1376 {
1377 __le32 disk_csum;
1378 u32 csum;
1379 unsigned long long newcsum;
1380 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1381 __le32 *isuper = (__le32*)sb;
1382
1383 disk_csum = sb->sb_csum;
1384 sb->sb_csum = 0;
1385 newcsum = 0;
1386 for (; size >= 4; size -= 4)
1387 newcsum += le32_to_cpu(*isuper++);
1388
1389 if (size == 2)
1390 newcsum += le16_to_cpu(*(__le16*) isuper);
1391
1392 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1393 sb->sb_csum = disk_csum;
1394 return cpu_to_le32(csum);
1395 }
1396
1397 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1398 {
1399 struct mdp_superblock_1 *sb;
1400 int ret;
1401 sector_t sb_start;
1402 sector_t sectors;
1403 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1404 int bmask;
1405
1406 /*
1407 * Calculate the position of the superblock in 512byte sectors.
1408 * It is always aligned to a 4K boundary and
1409 * depeding on minor_version, it can be:
1410 * 0: At least 8K, but less than 12K, from end of device
1411 * 1: At start of device
1412 * 2: 4K from start of device.
1413 */
1414 switch(minor_version) {
1415 case 0:
1416 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1417 sb_start -= 8*2;
1418 sb_start &= ~(sector_t)(4*2-1);
1419 break;
1420 case 1:
1421 sb_start = 0;
1422 break;
1423 case 2:
1424 sb_start = 8;
1425 break;
1426 default:
1427 return -EINVAL;
1428 }
1429 rdev->sb_start = sb_start;
1430
1431 /* superblock is rarely larger than 1K, but it can be larger,
1432 * and it is safe to read 4k, so we do that
1433 */
1434 ret = read_disk_sb(rdev, 4096);
1435 if (ret) return ret;
1436
1437 sb = page_address(rdev->sb_page);
1438
1439 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1440 sb->major_version != cpu_to_le32(1) ||
1441 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1442 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1443 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1444 return -EINVAL;
1445
1446 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1447 pr_warn("md: invalid superblock checksum on %s\n",
1448 bdevname(rdev->bdev,b));
1449 return -EINVAL;
1450 }
1451 if (le64_to_cpu(sb->data_size) < 10) {
1452 pr_warn("md: data_size too small on %s\n",
1453 bdevname(rdev->bdev,b));
1454 return -EINVAL;
1455 }
1456 if (sb->pad0 ||
1457 sb->pad3[0] ||
1458 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1459 /* Some padding is non-zero, might be a new feature */
1460 return -EINVAL;
1461
1462 rdev->preferred_minor = 0xffff;
1463 rdev->data_offset = le64_to_cpu(sb->data_offset);
1464 rdev->new_data_offset = rdev->data_offset;
1465 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1466 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1467 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1468 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1469
1470 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1471 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1472 if (rdev->sb_size & bmask)
1473 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1474
1475 if (minor_version
1476 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1477 return -EINVAL;
1478 if (minor_version
1479 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1480 return -EINVAL;
1481
1482 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1483 rdev->desc_nr = -1;
1484 else
1485 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1486
1487 if (!rdev->bb_page) {
1488 rdev->bb_page = alloc_page(GFP_KERNEL);
1489 if (!rdev->bb_page)
1490 return -ENOMEM;
1491 }
1492 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1493 rdev->badblocks.count == 0) {
1494 /* need to load the bad block list.
1495 * Currently we limit it to one page.
1496 */
1497 s32 offset;
1498 sector_t bb_sector;
1499 u64 *bbp;
1500 int i;
1501 int sectors = le16_to_cpu(sb->bblog_size);
1502 if (sectors > (PAGE_SIZE / 512))
1503 return -EINVAL;
1504 offset = le32_to_cpu(sb->bblog_offset);
1505 if (offset == 0)
1506 return -EINVAL;
1507 bb_sector = (long long)offset;
1508 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1509 rdev->bb_page, REQ_OP_READ, 0, true))
1510 return -EIO;
1511 bbp = (u64 *)page_address(rdev->bb_page);
1512 rdev->badblocks.shift = sb->bblog_shift;
1513 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1514 u64 bb = le64_to_cpu(*bbp);
1515 int count = bb & (0x3ff);
1516 u64 sector = bb >> 10;
1517 sector <<= sb->bblog_shift;
1518 count <<= sb->bblog_shift;
1519 if (bb + 1 == 0)
1520 break;
1521 if (badblocks_set(&rdev->badblocks, sector, count, 1))
1522 return -EINVAL;
1523 }
1524 } else if (sb->bblog_offset != 0)
1525 rdev->badblocks.shift = 0;
1526
1527 if (!refdev) {
1528 ret = 1;
1529 } else {
1530 __u64 ev1, ev2;
1531 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1532
1533 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1534 sb->level != refsb->level ||
1535 sb->layout != refsb->layout ||
1536 sb->chunksize != refsb->chunksize) {
1537 pr_warn("md: %s has strangely different superblock to %s\n",
1538 bdevname(rdev->bdev,b),
1539 bdevname(refdev->bdev,b2));
1540 return -EINVAL;
1541 }
1542 ev1 = le64_to_cpu(sb->events);
1543 ev2 = le64_to_cpu(refsb->events);
1544
1545 if (ev1 > ev2)
1546 ret = 1;
1547 else
1548 ret = 0;
1549 }
1550 if (minor_version) {
1551 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1552 sectors -= rdev->data_offset;
1553 } else
1554 sectors = rdev->sb_start;
1555 if (sectors < le64_to_cpu(sb->data_size))
1556 return -EINVAL;
1557 rdev->sectors = le64_to_cpu(sb->data_size);
1558 return ret;
1559 }
1560
1561 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1562 {
1563 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1564 __u64 ev1 = le64_to_cpu(sb->events);
1565
1566 rdev->raid_disk = -1;
1567 clear_bit(Faulty, &rdev->flags);
1568 clear_bit(In_sync, &rdev->flags);
1569 clear_bit(Bitmap_sync, &rdev->flags);
1570 clear_bit(WriteMostly, &rdev->flags);
1571
1572 if (mddev->raid_disks == 0) {
1573 mddev->major_version = 1;
1574 mddev->patch_version = 0;
1575 mddev->external = 0;
1576 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1577 mddev->ctime = le64_to_cpu(sb->ctime);
1578 mddev->utime = le64_to_cpu(sb->utime);
1579 mddev->level = le32_to_cpu(sb->level);
1580 mddev->clevel[0] = 0;
1581 mddev->layout = le32_to_cpu(sb->layout);
1582 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1583 mddev->dev_sectors = le64_to_cpu(sb->size);
1584 mddev->events = ev1;
1585 mddev->bitmap_info.offset = 0;
1586 mddev->bitmap_info.space = 0;
1587 /* Default location for bitmap is 1K after superblock
1588 * using 3K - total of 4K
1589 */
1590 mddev->bitmap_info.default_offset = 1024 >> 9;
1591 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1592 mddev->reshape_backwards = 0;
1593
1594 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1595 memcpy(mddev->uuid, sb->set_uuid, 16);
1596
1597 mddev->max_disks = (4096-256)/2;
1598
1599 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1600 mddev->bitmap_info.file == NULL) {
1601 mddev->bitmap_info.offset =
1602 (__s32)le32_to_cpu(sb->bitmap_offset);
1603 /* Metadata doesn't record how much space is available.
1604 * For 1.0, we assume we can use up to the superblock
1605 * if before, else to 4K beyond superblock.
1606 * For others, assume no change is possible.
1607 */
1608 if (mddev->minor_version > 0)
1609 mddev->bitmap_info.space = 0;
1610 else if (mddev->bitmap_info.offset > 0)
1611 mddev->bitmap_info.space =
1612 8 - mddev->bitmap_info.offset;
1613 else
1614 mddev->bitmap_info.space =
1615 -mddev->bitmap_info.offset;
1616 }
1617
1618 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1619 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1620 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1621 mddev->new_level = le32_to_cpu(sb->new_level);
1622 mddev->new_layout = le32_to_cpu(sb->new_layout);
1623 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1624 if (mddev->delta_disks < 0 ||
1625 (mddev->delta_disks == 0 &&
1626 (le32_to_cpu(sb->feature_map)
1627 & MD_FEATURE_RESHAPE_BACKWARDS)))
1628 mddev->reshape_backwards = 1;
1629 } else {
1630 mddev->reshape_position = MaxSector;
1631 mddev->delta_disks = 0;
1632 mddev->new_level = mddev->level;
1633 mddev->new_layout = mddev->layout;
1634 mddev->new_chunk_sectors = mddev->chunk_sectors;
1635 }
1636
1637 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1638 set_bit(MD_HAS_JOURNAL, &mddev->flags);
1639 } else if (mddev->pers == NULL) {
1640 /* Insist of good event counter while assembling, except for
1641 * spares (which don't need an event count) */
1642 ++ev1;
1643 if (rdev->desc_nr >= 0 &&
1644 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1645 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1646 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1647 if (ev1 < mddev->events)
1648 return -EINVAL;
1649 } else if (mddev->bitmap) {
1650 /* If adding to array with a bitmap, then we can accept an
1651 * older device, but not too old.
1652 */
1653 if (ev1 < mddev->bitmap->events_cleared)
1654 return 0;
1655 if (ev1 < mddev->events)
1656 set_bit(Bitmap_sync, &rdev->flags);
1657 } else {
1658 if (ev1 < mddev->events)
1659 /* just a hot-add of a new device, leave raid_disk at -1 */
1660 return 0;
1661 }
1662 if (mddev->level != LEVEL_MULTIPATH) {
1663 int role;
1664 if (rdev->desc_nr < 0 ||
1665 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1666 role = MD_DISK_ROLE_SPARE;
1667 rdev->desc_nr = -1;
1668 } else
1669 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1670 switch(role) {
1671 case MD_DISK_ROLE_SPARE: /* spare */
1672 break;
1673 case MD_DISK_ROLE_FAULTY: /* faulty */
1674 set_bit(Faulty, &rdev->flags);
1675 break;
1676 case MD_DISK_ROLE_JOURNAL: /* journal device */
1677 if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1678 /* journal device without journal feature */
1679 pr_warn("md: journal device provided without journal feature, ignoring the device\n");
1680 return -EINVAL;
1681 }
1682 set_bit(Journal, &rdev->flags);
1683 rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1684 rdev->raid_disk = 0;
1685 break;
1686 default:
1687 rdev->saved_raid_disk = role;
1688 if ((le32_to_cpu(sb->feature_map) &
1689 MD_FEATURE_RECOVERY_OFFSET)) {
1690 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1691 if (!(le32_to_cpu(sb->feature_map) &
1692 MD_FEATURE_RECOVERY_BITMAP))
1693 rdev->saved_raid_disk = -1;
1694 } else
1695 set_bit(In_sync, &rdev->flags);
1696 rdev->raid_disk = role;
1697 break;
1698 }
1699 if (sb->devflags & WriteMostly1)
1700 set_bit(WriteMostly, &rdev->flags);
1701 if (sb->devflags & FailFast1)
1702 set_bit(FailFast, &rdev->flags);
1703 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1704 set_bit(Replacement, &rdev->flags);
1705 } else /* MULTIPATH are always insync */
1706 set_bit(In_sync, &rdev->flags);
1707
1708 return 0;
1709 }
1710
1711 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1712 {
1713 struct mdp_superblock_1 *sb;
1714 struct md_rdev *rdev2;
1715 int max_dev, i;
1716 /* make rdev->sb match mddev and rdev data. */
1717
1718 sb = page_address(rdev->sb_page);
1719
1720 sb->feature_map = 0;
1721 sb->pad0 = 0;
1722 sb->recovery_offset = cpu_to_le64(0);
1723 memset(sb->pad3, 0, sizeof(sb->pad3));
1724
1725 sb->utime = cpu_to_le64((__u64)mddev->utime);
1726 sb->events = cpu_to_le64(mddev->events);
1727 if (mddev->in_sync)
1728 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1729 else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
1730 sb->resync_offset = cpu_to_le64(MaxSector);
1731 else
1732 sb->resync_offset = cpu_to_le64(0);
1733
1734 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1735
1736 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1737 sb->size = cpu_to_le64(mddev->dev_sectors);
1738 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1739 sb->level = cpu_to_le32(mddev->level);
1740 sb->layout = cpu_to_le32(mddev->layout);
1741 if (test_bit(FailFast, &rdev->flags))
1742 sb->devflags |= FailFast1;
1743 else
1744 sb->devflags &= ~FailFast1;
1745
1746 if (test_bit(WriteMostly, &rdev->flags))
1747 sb->devflags |= WriteMostly1;
1748 else
1749 sb->devflags &= ~WriteMostly1;
1750 sb->data_offset = cpu_to_le64(rdev->data_offset);
1751 sb->data_size = cpu_to_le64(rdev->sectors);
1752
1753 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1754 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1755 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1756 }
1757
1758 if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
1759 !test_bit(In_sync, &rdev->flags)) {
1760 sb->feature_map |=
1761 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1762 sb->recovery_offset =
1763 cpu_to_le64(rdev->recovery_offset);
1764 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1765 sb->feature_map |=
1766 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
1767 }
1768 /* Note: recovery_offset and journal_tail share space */
1769 if (test_bit(Journal, &rdev->flags))
1770 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
1771 if (test_bit(Replacement, &rdev->flags))
1772 sb->feature_map |=
1773 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1774
1775 if (mddev->reshape_position != MaxSector) {
1776 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1777 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1778 sb->new_layout = cpu_to_le32(mddev->new_layout);
1779 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1780 sb->new_level = cpu_to_le32(mddev->new_level);
1781 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1782 if (mddev->delta_disks == 0 &&
1783 mddev->reshape_backwards)
1784 sb->feature_map
1785 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1786 if (rdev->new_data_offset != rdev->data_offset) {
1787 sb->feature_map
1788 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1789 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1790 - rdev->data_offset));
1791 }
1792 }
1793
1794 if (mddev_is_clustered(mddev))
1795 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
1796
1797 if (rdev->badblocks.count == 0)
1798 /* Nothing to do for bad blocks*/ ;
1799 else if (sb->bblog_offset == 0)
1800 /* Cannot record bad blocks on this device */
1801 md_error(mddev, rdev);
1802 else {
1803 struct badblocks *bb = &rdev->badblocks;
1804 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1805 u64 *p = bb->page;
1806 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1807 if (bb->changed) {
1808 unsigned seq;
1809
1810 retry:
1811 seq = read_seqbegin(&bb->lock);
1812
1813 memset(bbp, 0xff, PAGE_SIZE);
1814
1815 for (i = 0 ; i < bb->count ; i++) {
1816 u64 internal_bb = p[i];
1817 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1818 | BB_LEN(internal_bb));
1819 bbp[i] = cpu_to_le64(store_bb);
1820 }
1821 bb->changed = 0;
1822 if (read_seqretry(&bb->lock, seq))
1823 goto retry;
1824
1825 bb->sector = (rdev->sb_start +
1826 (int)le32_to_cpu(sb->bblog_offset));
1827 bb->size = le16_to_cpu(sb->bblog_size);
1828 }
1829 }
1830
1831 max_dev = 0;
1832 rdev_for_each(rdev2, mddev)
1833 if (rdev2->desc_nr+1 > max_dev)
1834 max_dev = rdev2->desc_nr+1;
1835
1836 if (max_dev > le32_to_cpu(sb->max_dev)) {
1837 int bmask;
1838 sb->max_dev = cpu_to_le32(max_dev);
1839 rdev->sb_size = max_dev * 2 + 256;
1840 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1841 if (rdev->sb_size & bmask)
1842 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1843 } else
1844 max_dev = le32_to_cpu(sb->max_dev);
1845
1846 for (i=0; i<max_dev;i++)
1847 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
1848
1849 if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
1850 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
1851
1852 rdev_for_each(rdev2, mddev) {
1853 i = rdev2->desc_nr;
1854 if (test_bit(Faulty, &rdev2->flags))
1855 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
1856 else if (test_bit(In_sync, &rdev2->flags))
1857 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1858 else if (test_bit(Journal, &rdev2->flags))
1859 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
1860 else if (rdev2->raid_disk >= 0)
1861 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1862 else
1863 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
1864 }
1865
1866 sb->sb_csum = calc_sb_1_csum(sb);
1867 }
1868
1869 static unsigned long long
1870 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1871 {
1872 struct mdp_superblock_1 *sb;
1873 sector_t max_sectors;
1874 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1875 return 0; /* component must fit device */
1876 if (rdev->data_offset != rdev->new_data_offset)
1877 return 0; /* too confusing */
1878 if (rdev->sb_start < rdev->data_offset) {
1879 /* minor versions 1 and 2; superblock before data */
1880 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1881 max_sectors -= rdev->data_offset;
1882 if (!num_sectors || num_sectors > max_sectors)
1883 num_sectors = max_sectors;
1884 } else if (rdev->mddev->bitmap_info.offset) {
1885 /* minor version 0 with bitmap we can't move */
1886 return 0;
1887 } else {
1888 /* minor version 0; superblock after data */
1889 sector_t sb_start;
1890 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1891 sb_start &= ~(sector_t)(4*2 - 1);
1892 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1893 if (!num_sectors || num_sectors > max_sectors)
1894 num_sectors = max_sectors;
1895 rdev->sb_start = sb_start;
1896 }
1897 sb = page_address(rdev->sb_page);
1898 sb->data_size = cpu_to_le64(num_sectors);
1899 sb->super_offset = rdev->sb_start;
1900 sb->sb_csum = calc_sb_1_csum(sb);
1901 do {
1902 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1903 rdev->sb_page);
1904 } while (md_super_wait(rdev->mddev) < 0);
1905 return num_sectors;
1906
1907 }
1908
1909 static int
1910 super_1_allow_new_offset(struct md_rdev *rdev,
1911 unsigned long long new_offset)
1912 {
1913 /* All necessary checks on new >= old have been done */
1914 struct bitmap *bitmap;
1915 if (new_offset >= rdev->data_offset)
1916 return 1;
1917
1918 /* with 1.0 metadata, there is no metadata to tread on
1919 * so we can always move back */
1920 if (rdev->mddev->minor_version == 0)
1921 return 1;
1922
1923 /* otherwise we must be sure not to step on
1924 * any metadata, so stay:
1925 * 36K beyond start of superblock
1926 * beyond end of badblocks
1927 * beyond write-intent bitmap
1928 */
1929 if (rdev->sb_start + (32+4)*2 > new_offset)
1930 return 0;
1931 bitmap = rdev->mddev->bitmap;
1932 if (bitmap && !rdev->mddev->bitmap_info.file &&
1933 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1934 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
1935 return 0;
1936 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1937 return 0;
1938
1939 return 1;
1940 }
1941
1942 static struct super_type super_types[] = {
1943 [0] = {
1944 .name = "0.90.0",
1945 .owner = THIS_MODULE,
1946 .load_super = super_90_load,
1947 .validate_super = super_90_validate,
1948 .sync_super = super_90_sync,
1949 .rdev_size_change = super_90_rdev_size_change,
1950 .allow_new_offset = super_90_allow_new_offset,
1951 },
1952 [1] = {
1953 .name = "md-1",
1954 .owner = THIS_MODULE,
1955 .load_super = super_1_load,
1956 .validate_super = super_1_validate,
1957 .sync_super = super_1_sync,
1958 .rdev_size_change = super_1_rdev_size_change,
1959 .allow_new_offset = super_1_allow_new_offset,
1960 },
1961 };
1962
1963 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1964 {
1965 if (mddev->sync_super) {
1966 mddev->sync_super(mddev, rdev);
1967 return;
1968 }
1969
1970 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1971
1972 super_types[mddev->major_version].sync_super(mddev, rdev);
1973 }
1974
1975 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1976 {
1977 struct md_rdev *rdev, *rdev2;
1978
1979 rcu_read_lock();
1980 rdev_for_each_rcu(rdev, mddev1) {
1981 if (test_bit(Faulty, &rdev->flags) ||
1982 test_bit(Journal, &rdev->flags) ||
1983 rdev->raid_disk == -1)
1984 continue;
1985 rdev_for_each_rcu(rdev2, mddev2) {
1986 if (test_bit(Faulty, &rdev2->flags) ||
1987 test_bit(Journal, &rdev2->flags) ||
1988 rdev2->raid_disk == -1)
1989 continue;
1990 if (rdev->bdev->bd_contains ==
1991 rdev2->bdev->bd_contains) {
1992 rcu_read_unlock();
1993 return 1;
1994 }
1995 }
1996 }
1997 rcu_read_unlock();
1998 return 0;
1999 }
2000
2001 static LIST_HEAD(pending_raid_disks);
2002
2003 /*
2004 * Try to register data integrity profile for an mddev
2005 *
2006 * This is called when an array is started and after a disk has been kicked
2007 * from the array. It only succeeds if all working and active component devices
2008 * are integrity capable with matching profiles.
2009 */
2010 int md_integrity_register(struct mddev *mddev)
2011 {
2012 struct md_rdev *rdev, *reference = NULL;
2013
2014 if (list_empty(&mddev->disks))
2015 return 0; /* nothing to do */
2016 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2017 return 0; /* shouldn't register, or already is */
2018 rdev_for_each(rdev, mddev) {
2019 /* skip spares and non-functional disks */
2020 if (test_bit(Faulty, &rdev->flags))
2021 continue;
2022 if (rdev->raid_disk < 0)
2023 continue;
2024 if (!reference) {
2025 /* Use the first rdev as the reference */
2026 reference = rdev;
2027 continue;
2028 }
2029 /* does this rdev's profile match the reference profile? */
2030 if (blk_integrity_compare(reference->bdev->bd_disk,
2031 rdev->bdev->bd_disk) < 0)
2032 return -EINVAL;
2033 }
2034 if (!reference || !bdev_get_integrity(reference->bdev))
2035 return 0;
2036 /*
2037 * All component devices are integrity capable and have matching
2038 * profiles, register the common profile for the md device.
2039 */
2040 blk_integrity_register(mddev->gendisk,
2041 bdev_get_integrity(reference->bdev));
2042
2043 pr_debug("md: data integrity enabled on %s\n", mdname(mddev));
2044 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
2045 pr_err("md: failed to create integrity pool for %s\n",
2046 mdname(mddev));
2047 return -EINVAL;
2048 }
2049 return 0;
2050 }
2051 EXPORT_SYMBOL(md_integrity_register);
2052
2053 /*
2054 * Attempt to add an rdev, but only if it is consistent with the current
2055 * integrity profile
2056 */
2057 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2058 {
2059 struct blk_integrity *bi_rdev;
2060 struct blk_integrity *bi_mddev;
2061 char name[BDEVNAME_SIZE];
2062
2063 if (!mddev->gendisk)
2064 return 0;
2065
2066 bi_rdev = bdev_get_integrity(rdev->bdev);
2067 bi_mddev = blk_get_integrity(mddev->gendisk);
2068
2069 if (!bi_mddev) /* nothing to do */
2070 return 0;
2071
2072 if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2073 pr_err("%s: incompatible integrity profile for %s\n",
2074 mdname(mddev), bdevname(rdev->bdev, name));
2075 return -ENXIO;
2076 }
2077
2078 return 0;
2079 }
2080 EXPORT_SYMBOL(md_integrity_add_rdev);
2081
2082 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2083 {
2084 char b[BDEVNAME_SIZE];
2085 struct kobject *ko;
2086 int err;
2087
2088 /* prevent duplicates */
2089 if (find_rdev(mddev, rdev->bdev->bd_dev))
2090 return -EEXIST;
2091
2092 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2093 if (!test_bit(Journal, &rdev->flags) &&
2094 rdev->sectors &&
2095 (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2096 if (mddev->pers) {
2097 /* Cannot change size, so fail
2098 * If mddev->level <= 0, then we don't care
2099 * about aligning sizes (e.g. linear)
2100 */
2101 if (mddev->level > 0)
2102 return -ENOSPC;
2103 } else
2104 mddev->dev_sectors = rdev->sectors;
2105 }
2106
2107 /* Verify rdev->desc_nr is unique.
2108 * If it is -1, assign a free number, else
2109 * check number is not in use
2110 */
2111 rcu_read_lock();
2112 if (rdev->desc_nr < 0) {
2113 int choice = 0;
2114 if (mddev->pers)
2115 choice = mddev->raid_disks;
2116 while (md_find_rdev_nr_rcu(mddev, choice))
2117 choice++;
2118 rdev->desc_nr = choice;
2119 } else {
2120 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2121 rcu_read_unlock();
2122 return -EBUSY;
2123 }
2124 }
2125 rcu_read_unlock();
2126 if (!test_bit(Journal, &rdev->flags) &&
2127 mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2128 pr_warn("md: %s: array is limited to %d devices\n",
2129 mdname(mddev), mddev->max_disks);
2130 return -EBUSY;
2131 }
2132 bdevname(rdev->bdev,b);
2133 strreplace(b, '/', '!');
2134
2135 rdev->mddev = mddev;
2136 pr_debug("md: bind<%s>\n", b);
2137
2138 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2139 goto fail;
2140
2141 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2142 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2143 /* failure here is OK */;
2144 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2145
2146 list_add_rcu(&rdev->same_set, &mddev->disks);
2147 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2148
2149 /* May as well allow recovery to be retried once */
2150 mddev->recovery_disabled++;
2151
2152 return 0;
2153
2154 fail:
2155 pr_warn("md: failed to register dev-%s for %s\n",
2156 b, mdname(mddev));
2157 return err;
2158 }
2159
2160 static void md_delayed_delete(struct work_struct *ws)
2161 {
2162 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2163 kobject_del(&rdev->kobj);
2164 kobject_put(&rdev->kobj);
2165 }
2166
2167 static void unbind_rdev_from_array(struct md_rdev *rdev)
2168 {
2169 char b[BDEVNAME_SIZE];
2170
2171 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2172 list_del_rcu(&rdev->same_set);
2173 pr_debug("md: unbind<%s>\n", bdevname(rdev->bdev,b));
2174 rdev->mddev = NULL;
2175 sysfs_remove_link(&rdev->kobj, "block");
2176 sysfs_put(rdev->sysfs_state);
2177 rdev->sysfs_state = NULL;
2178 rdev->badblocks.count = 0;
2179 /* We need to delay this, otherwise we can deadlock when
2180 * writing to 'remove' to "dev/state". We also need
2181 * to delay it due to rcu usage.
2182 */
2183 synchronize_rcu();
2184 INIT_WORK(&rdev->del_work, md_delayed_delete);
2185 kobject_get(&rdev->kobj);
2186 queue_work(md_misc_wq, &rdev->del_work);
2187 }
2188
2189 /*
2190 * prevent the device from being mounted, repartitioned or
2191 * otherwise reused by a RAID array (or any other kernel
2192 * subsystem), by bd_claiming the device.
2193 */
2194 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2195 {
2196 int err = 0;
2197 struct block_device *bdev;
2198 char b[BDEVNAME_SIZE];
2199
2200 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2201 shared ? (struct md_rdev *)lock_rdev : rdev);
2202 if (IS_ERR(bdev)) {
2203 pr_warn("md: could not open %s.\n", __bdevname(dev, b));
2204 return PTR_ERR(bdev);
2205 }
2206 rdev->bdev = bdev;
2207 return err;
2208 }
2209
2210 static void unlock_rdev(struct md_rdev *rdev)
2211 {
2212 struct block_device *bdev = rdev->bdev;
2213 rdev->bdev = NULL;
2214 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2215 }
2216
2217 void md_autodetect_dev(dev_t dev);
2218
2219 static void export_rdev(struct md_rdev *rdev)
2220 {
2221 char b[BDEVNAME_SIZE];
2222
2223 pr_debug("md: export_rdev(%s)\n", bdevname(rdev->bdev,b));
2224 md_rdev_clear(rdev);
2225 #ifndef MODULE
2226 if (test_bit(AutoDetected, &rdev->flags))
2227 md_autodetect_dev(rdev->bdev->bd_dev);
2228 #endif
2229 unlock_rdev(rdev);
2230 kobject_put(&rdev->kobj);
2231 }
2232
2233 void md_kick_rdev_from_array(struct md_rdev *rdev)
2234 {
2235 unbind_rdev_from_array(rdev);
2236 export_rdev(rdev);
2237 }
2238 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2239
2240 static void export_array(struct mddev *mddev)
2241 {
2242 struct md_rdev *rdev;
2243
2244 while (!list_empty(&mddev->disks)) {
2245 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2246 same_set);
2247 md_kick_rdev_from_array(rdev);
2248 }
2249 mddev->raid_disks = 0;
2250 mddev->major_version = 0;
2251 }
2252
2253 static void sync_sbs(struct mddev *mddev, int nospares)
2254 {
2255 /* Update each superblock (in-memory image), but
2256 * if we are allowed to, skip spares which already
2257 * have the right event counter, or have one earlier
2258 * (which would mean they aren't being marked as dirty
2259 * with the rest of the array)
2260 */
2261 struct md_rdev *rdev;
2262 rdev_for_each(rdev, mddev) {
2263 if (rdev->sb_events == mddev->events ||
2264 (nospares &&
2265 rdev->raid_disk < 0 &&
2266 rdev->sb_events+1 == mddev->events)) {
2267 /* Don't update this superblock */
2268 rdev->sb_loaded = 2;
2269 } else {
2270 sync_super(mddev, rdev);
2271 rdev->sb_loaded = 1;
2272 }
2273 }
2274 }
2275
2276 static bool does_sb_need_changing(struct mddev *mddev)
2277 {
2278 struct md_rdev *rdev;
2279 struct mdp_superblock_1 *sb;
2280 int role;
2281
2282 /* Find a good rdev */
2283 rdev_for_each(rdev, mddev)
2284 if ((rdev->raid_disk >= 0) && !test_bit(Faulty, &rdev->flags))
2285 break;
2286
2287 /* No good device found. */
2288 if (!rdev)
2289 return false;
2290
2291 sb = page_address(rdev->sb_page);
2292 /* Check if a device has become faulty or a spare become active */
2293 rdev_for_each(rdev, mddev) {
2294 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2295 /* Device activated? */
2296 if (role == 0xffff && rdev->raid_disk >=0 &&
2297 !test_bit(Faulty, &rdev->flags))
2298 return true;
2299 /* Device turned faulty? */
2300 if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd))
2301 return true;
2302 }
2303
2304 /* Check if any mddev parameters have changed */
2305 if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2306 (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2307 (mddev->layout != le64_to_cpu(sb->layout)) ||
2308 (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2309 (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2310 return true;
2311
2312 return false;
2313 }
2314
2315 void md_update_sb(struct mddev *mddev, int force_change)
2316 {
2317 struct md_rdev *rdev;
2318 int sync_req;
2319 int nospares = 0;
2320 int any_badblocks_changed = 0;
2321 int ret = -1;
2322
2323 if (mddev->ro) {
2324 if (force_change)
2325 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2326 return;
2327 }
2328
2329 repeat:
2330 if (mddev_is_clustered(mddev)) {
2331 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2332 force_change = 1;
2333 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2334 nospares = 1;
2335 ret = md_cluster_ops->metadata_update_start(mddev);
2336 /* Has someone else has updated the sb */
2337 if (!does_sb_need_changing(mddev)) {
2338 if (ret == 0)
2339 md_cluster_ops->metadata_update_cancel(mddev);
2340 bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2341 BIT(MD_SB_CHANGE_DEVS) |
2342 BIT(MD_SB_CHANGE_CLEAN));
2343 return;
2344 }
2345 }
2346
2347 /* First make sure individual recovery_offsets are correct */
2348 rdev_for_each(rdev, mddev) {
2349 if (rdev->raid_disk >= 0 &&
2350 mddev->delta_disks >= 0 &&
2351 !test_bit(Journal, &rdev->flags) &&
2352 !test_bit(In_sync, &rdev->flags) &&
2353 mddev->curr_resync_completed > rdev->recovery_offset)
2354 rdev->recovery_offset = mddev->curr_resync_completed;
2355
2356 }
2357 if (!mddev->persistent) {
2358 clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2359 clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2360 if (!mddev->external) {
2361 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
2362 rdev_for_each(rdev, mddev) {
2363 if (rdev->badblocks.changed) {
2364 rdev->badblocks.changed = 0;
2365 ack_all_badblocks(&rdev->badblocks);
2366 md_error(mddev, rdev);
2367 }
2368 clear_bit(Blocked, &rdev->flags);
2369 clear_bit(BlockedBadBlocks, &rdev->flags);
2370 wake_up(&rdev->blocked_wait);
2371 }
2372 }
2373 wake_up(&mddev->sb_wait);
2374 return;
2375 }
2376
2377 spin_lock(&mddev->lock);
2378
2379 mddev->utime = ktime_get_real_seconds();
2380
2381 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2382 force_change = 1;
2383 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2384 /* just a clean<-> dirty transition, possibly leave spares alone,
2385 * though if events isn't the right even/odd, we will have to do
2386 * spares after all
2387 */
2388 nospares = 1;
2389 if (force_change)
2390 nospares = 0;
2391 if (mddev->degraded)
2392 /* If the array is degraded, then skipping spares is both
2393 * dangerous and fairly pointless.
2394 * Dangerous because a device that was removed from the array
2395 * might have a event_count that still looks up-to-date,
2396 * so it can be re-added without a resync.
2397 * Pointless because if there are any spares to skip,
2398 * then a recovery will happen and soon that array won't
2399 * be degraded any more and the spare can go back to sleep then.
2400 */
2401 nospares = 0;
2402
2403 sync_req = mddev->in_sync;
2404
2405 /* If this is just a dirty<->clean transition, and the array is clean
2406 * and 'events' is odd, we can roll back to the previous clean state */
2407 if (nospares
2408 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2409 && mddev->can_decrease_events
2410 && mddev->events != 1) {
2411 mddev->events--;
2412 mddev->can_decrease_events = 0;
2413 } else {
2414 /* otherwise we have to go forward and ... */
2415 mddev->events ++;
2416 mddev->can_decrease_events = nospares;
2417 }
2418
2419 /*
2420 * This 64-bit counter should never wrap.
2421 * Either we are in around ~1 trillion A.C., assuming
2422 * 1 reboot per second, or we have a bug...
2423 */
2424 WARN_ON(mddev->events == 0);
2425
2426 rdev_for_each(rdev, mddev) {
2427 if (rdev->badblocks.changed)
2428 any_badblocks_changed++;
2429 if (test_bit(Faulty, &rdev->flags))
2430 set_bit(FaultRecorded, &rdev->flags);
2431 }
2432
2433 sync_sbs(mddev, nospares);
2434 spin_unlock(&mddev->lock);
2435
2436 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2437 mdname(mddev), mddev->in_sync);
2438
2439 if (mddev->queue)
2440 blk_add_trace_msg(mddev->queue, "md md_update_sb");
2441 rewrite:
2442 bitmap_update_sb(mddev->bitmap);
2443 rdev_for_each(rdev, mddev) {
2444 char b[BDEVNAME_SIZE];
2445
2446 if (rdev->sb_loaded != 1)
2447 continue; /* no noise on spare devices */
2448
2449 if (!test_bit(Faulty, &rdev->flags)) {
2450 md_super_write(mddev,rdev,
2451 rdev->sb_start, rdev->sb_size,
2452 rdev->sb_page);
2453 pr_debug("md: (write) %s's sb offset: %llu\n",
2454 bdevname(rdev->bdev, b),
2455 (unsigned long long)rdev->sb_start);
2456 rdev->sb_events = mddev->events;
2457 if (rdev->badblocks.size) {
2458 md_super_write(mddev, rdev,
2459 rdev->badblocks.sector,
2460 rdev->badblocks.size << 9,
2461 rdev->bb_page);
2462 rdev->badblocks.size = 0;
2463 }
2464
2465 } else
2466 pr_debug("md: %s (skipping faulty)\n",
2467 bdevname(rdev->bdev, b));
2468
2469 if (mddev->level == LEVEL_MULTIPATH)
2470 /* only need to write one superblock... */
2471 break;
2472 }
2473 if (md_super_wait(mddev) < 0)
2474 goto rewrite;
2475 /* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */
2476
2477 if (mddev_is_clustered(mddev) && ret == 0)
2478 md_cluster_ops->metadata_update_finish(mddev);
2479
2480 if (mddev->in_sync != sync_req ||
2481 !bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2482 BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_CLEAN)))
2483 /* have to write it out again */
2484 goto repeat;
2485 wake_up(&mddev->sb_wait);
2486 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2487 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2488
2489 rdev_for_each(rdev, mddev) {
2490 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2491 clear_bit(Blocked, &rdev->flags);
2492
2493 if (any_badblocks_changed)
2494 ack_all_badblocks(&rdev->badblocks);
2495 clear_bit(BlockedBadBlocks, &rdev->flags);
2496 wake_up(&rdev->blocked_wait);
2497 }
2498 }
2499 EXPORT_SYMBOL(md_update_sb);
2500
2501 static int add_bound_rdev(struct md_rdev *rdev)
2502 {
2503 struct mddev *mddev = rdev->mddev;
2504 int err = 0;
2505 bool add_journal = test_bit(Journal, &rdev->flags);
2506
2507 if (!mddev->pers->hot_remove_disk || add_journal) {
2508 /* If there is hot_add_disk but no hot_remove_disk
2509 * then added disks for geometry changes,
2510 * and should be added immediately.
2511 */
2512 super_types[mddev->major_version].
2513 validate_super(mddev, rdev);
2514 if (add_journal)
2515 mddev_suspend(mddev);
2516 err = mddev->pers->hot_add_disk(mddev, rdev);
2517 if (add_journal)
2518 mddev_resume(mddev);
2519 if (err) {
2520 md_kick_rdev_from_array(rdev);
2521 return err;
2522 }
2523 }
2524 sysfs_notify_dirent_safe(rdev->sysfs_state);
2525
2526 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2527 if (mddev->degraded)
2528 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2529 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2530 md_new_event(mddev);
2531 md_wakeup_thread(mddev->thread);
2532 return 0;
2533 }
2534
2535 /* words written to sysfs files may, or may not, be \n terminated.
2536 * We want to accept with case. For this we use cmd_match.
2537 */
2538 static int cmd_match(const char *cmd, const char *str)
2539 {
2540 /* See if cmd, written into a sysfs file, matches
2541 * str. They must either be the same, or cmd can
2542 * have a trailing newline
2543 */
2544 while (*cmd && *str && *cmd == *str) {
2545 cmd++;
2546 str++;
2547 }
2548 if (*cmd == '\n')
2549 cmd++;
2550 if (*str || *cmd)
2551 return 0;
2552 return 1;
2553 }
2554
2555 struct rdev_sysfs_entry {
2556 struct attribute attr;
2557 ssize_t (*show)(struct md_rdev *, char *);
2558 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2559 };
2560
2561 static ssize_t
2562 state_show(struct md_rdev *rdev, char *page)
2563 {
2564 char *sep = ",";
2565 size_t len = 0;
2566 unsigned long flags = ACCESS_ONCE(rdev->flags);
2567
2568 if (test_bit(Faulty, &flags) ||
2569 (!test_bit(ExternalBbl, &flags) &&
2570 rdev->badblocks.unacked_exist))
2571 len += sprintf(page+len, "faulty%s", sep);
2572 if (test_bit(In_sync, &flags))
2573 len += sprintf(page+len, "in_sync%s", sep);
2574 if (test_bit(Journal, &flags))
2575 len += sprintf(page+len, "journal%s", sep);
2576 if (test_bit(WriteMostly, &flags))
2577 len += sprintf(page+len, "write_mostly%s", sep);
2578 if (test_bit(Blocked, &flags) ||
2579 (rdev->badblocks.unacked_exist
2580 && !test_bit(Faulty, &flags)))
2581 len += sprintf(page+len, "blocked%s", sep);
2582 if (!test_bit(Faulty, &flags) &&
2583 !test_bit(Journal, &flags) &&
2584 !test_bit(In_sync, &flags))
2585 len += sprintf(page+len, "spare%s", sep);
2586 if (test_bit(WriteErrorSeen, &flags))
2587 len += sprintf(page+len, "write_error%s", sep);
2588 if (test_bit(WantReplacement, &flags))
2589 len += sprintf(page+len, "want_replacement%s", sep);
2590 if (test_bit(Replacement, &flags))
2591 len += sprintf(page+len, "replacement%s", sep);
2592 if (test_bit(ExternalBbl, &flags))
2593 len += sprintf(page+len, "external_bbl%s", sep);
2594 if (test_bit(FailFast, &flags))
2595 len += sprintf(page+len, "failfast%s", sep);
2596
2597 if (len)
2598 len -= strlen(sep);
2599
2600 return len+sprintf(page+len, "\n");
2601 }
2602
2603 static ssize_t
2604 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2605 {
2606 /* can write
2607 * faulty - simulates an error
2608 * remove - disconnects the device
2609 * writemostly - sets write_mostly
2610 * -writemostly - clears write_mostly
2611 * blocked - sets the Blocked flags
2612 * -blocked - clears the Blocked and possibly simulates an error
2613 * insync - sets Insync providing device isn't active
2614 * -insync - clear Insync for a device with a slot assigned,
2615 * so that it gets rebuilt based on bitmap
2616 * write_error - sets WriteErrorSeen
2617 * -write_error - clears WriteErrorSeen
2618 * {,-}failfast - set/clear FailFast
2619 */
2620 int err = -EINVAL;
2621 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2622 md_error(rdev->mddev, rdev);
2623 if (test_bit(Faulty, &rdev->flags))
2624 err = 0;
2625 else
2626 err = -EBUSY;
2627 } else if (cmd_match(buf, "remove")) {
2628 if (rdev->mddev->pers) {
2629 clear_bit(Blocked, &rdev->flags);
2630 remove_and_add_spares(rdev->mddev, rdev);
2631 }
2632 if (rdev->raid_disk >= 0)
2633 err = -EBUSY;
2634 else {
2635 struct mddev *mddev = rdev->mddev;
2636 err = 0;
2637 if (mddev_is_clustered(mddev))
2638 err = md_cluster_ops->remove_disk(mddev, rdev);
2639
2640 if (err == 0) {
2641 md_kick_rdev_from_array(rdev);
2642 if (mddev->pers) {
2643 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2644 md_wakeup_thread(mddev->thread);
2645 }
2646 md_new_event(mddev);
2647 }
2648 }
2649 } else if (cmd_match(buf, "writemostly")) {
2650 set_bit(WriteMostly, &rdev->flags);
2651 err = 0;
2652 } else if (cmd_match(buf, "-writemostly")) {
2653 clear_bit(WriteMostly, &rdev->flags);
2654 err = 0;
2655 } else if (cmd_match(buf, "blocked")) {
2656 set_bit(Blocked, &rdev->flags);
2657 err = 0;
2658 } else if (cmd_match(buf, "-blocked")) {
2659 if (!test_bit(Faulty, &rdev->flags) &&
2660 !test_bit(ExternalBbl, &rdev->flags) &&
2661 rdev->badblocks.unacked_exist) {
2662 /* metadata handler doesn't understand badblocks,
2663 * so we need to fail the device
2664 */
2665 md_error(rdev->mddev, rdev);
2666 }
2667 clear_bit(Blocked, &rdev->flags);
2668 clear_bit(BlockedBadBlocks, &rdev->flags);
2669 wake_up(&rdev->blocked_wait);
2670 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2671 md_wakeup_thread(rdev->mddev->thread);
2672
2673 err = 0;
2674 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2675 set_bit(In_sync, &rdev->flags);
2676 err = 0;
2677 } else if (cmd_match(buf, "failfast")) {
2678 set_bit(FailFast, &rdev->flags);
2679 err = 0;
2680 } else if (cmd_match(buf, "-failfast")) {
2681 clear_bit(FailFast, &rdev->flags);
2682 err = 0;
2683 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
2684 !test_bit(Journal, &rdev->flags)) {
2685 if (rdev->mddev->pers == NULL) {
2686 clear_bit(In_sync, &rdev->flags);
2687 rdev->saved_raid_disk = rdev->raid_disk;
2688 rdev->raid_disk = -1;
2689 err = 0;
2690 }
2691 } else if (cmd_match(buf, "write_error")) {
2692 set_bit(WriteErrorSeen, &rdev->flags);
2693 err = 0;
2694 } else if (cmd_match(buf, "-write_error")) {
2695 clear_bit(WriteErrorSeen, &rdev->flags);
2696 err = 0;
2697 } else if (cmd_match(buf, "want_replacement")) {
2698 /* Any non-spare device that is not a replacement can
2699 * become want_replacement at any time, but we then need to
2700 * check if recovery is needed.
2701 */
2702 if (rdev->raid_disk >= 0 &&
2703 !test_bit(Journal, &rdev->flags) &&
2704 !test_bit(Replacement, &rdev->flags))
2705 set_bit(WantReplacement, &rdev->flags);
2706 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2707 md_wakeup_thread(rdev->mddev->thread);
2708 err = 0;
2709 } else if (cmd_match(buf, "-want_replacement")) {
2710 /* Clearing 'want_replacement' is always allowed.
2711 * Once replacements starts it is too late though.
2712 */
2713 err = 0;
2714 clear_bit(WantReplacement, &rdev->flags);
2715 } else if (cmd_match(buf, "replacement")) {
2716 /* Can only set a device as a replacement when array has not
2717 * yet been started. Once running, replacement is automatic
2718 * from spares, or by assigning 'slot'.
2719 */
2720 if (rdev->mddev->pers)
2721 err = -EBUSY;
2722 else {
2723 set_bit(Replacement, &rdev->flags);
2724 err = 0;
2725 }
2726 } else if (cmd_match(buf, "-replacement")) {
2727 /* Similarly, can only clear Replacement before start */
2728 if (rdev->mddev->pers)
2729 err = -EBUSY;
2730 else {
2731 clear_bit(Replacement, &rdev->flags);
2732 err = 0;
2733 }
2734 } else if (cmd_match(buf, "re-add")) {
2735 if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1)) {
2736 /* clear_bit is performed _after_ all the devices
2737 * have their local Faulty bit cleared. If any writes
2738 * happen in the meantime in the local node, they
2739 * will land in the local bitmap, which will be synced
2740 * by this node eventually
2741 */
2742 if (!mddev_is_clustered(rdev->mddev) ||
2743 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
2744 clear_bit(Faulty, &rdev->flags);
2745 err = add_bound_rdev(rdev);
2746 }
2747 } else
2748 err = -EBUSY;
2749 } else if (cmd_match(buf, "external_bbl") && (rdev->mddev->external)) {
2750 set_bit(ExternalBbl, &rdev->flags);
2751 rdev->badblocks.shift = 0;
2752 err = 0;
2753 } else if (cmd_match(buf, "-external_bbl") && (rdev->mddev->external)) {
2754 clear_bit(ExternalBbl, &rdev->flags);
2755 err = 0;
2756 }
2757 if (!err)
2758 sysfs_notify_dirent_safe(rdev->sysfs_state);
2759 return err ? err : len;
2760 }
2761 static struct rdev_sysfs_entry rdev_state =
2762 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
2763
2764 static ssize_t
2765 errors_show(struct md_rdev *rdev, char *page)
2766 {
2767 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2768 }
2769
2770 static ssize_t
2771 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2772 {
2773 unsigned int n;
2774 int rv;
2775
2776 rv = kstrtouint(buf, 10, &n);
2777 if (rv < 0)
2778 return rv;
2779 atomic_set(&rdev->corrected_errors, n);
2780 return len;
2781 }
2782 static struct rdev_sysfs_entry rdev_errors =
2783 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2784
2785 static ssize_t
2786 slot_show(struct md_rdev *rdev, char *page)
2787 {
2788 if (test_bit(Journal, &rdev->flags))
2789 return sprintf(page, "journal\n");
2790 else if (rdev->raid_disk < 0)
2791 return sprintf(page, "none\n");
2792 else
2793 return sprintf(page, "%d\n", rdev->raid_disk);
2794 }
2795
2796 static ssize_t
2797 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2798 {
2799 int slot;
2800 int err;
2801
2802 if (test_bit(Journal, &rdev->flags))
2803 return -EBUSY;
2804 if (strncmp(buf, "none", 4)==0)
2805 slot = -1;
2806 else {
2807 err = kstrtouint(buf, 10, (unsigned int *)&slot);
2808 if (err < 0)
2809 return err;
2810 }
2811 if (rdev->mddev->pers && slot == -1) {
2812 /* Setting 'slot' on an active array requires also
2813 * updating the 'rd%d' link, and communicating
2814 * with the personality with ->hot_*_disk.
2815 * For now we only support removing
2816 * failed/spare devices. This normally happens automatically,
2817 * but not when the metadata is externally managed.
2818 */
2819 if (rdev->raid_disk == -1)
2820 return -EEXIST;
2821 /* personality does all needed checks */
2822 if (rdev->mddev->pers->hot_remove_disk == NULL)
2823 return -EINVAL;
2824 clear_bit(Blocked, &rdev->flags);
2825 remove_and_add_spares(rdev->mddev, rdev);
2826 if (rdev->raid_disk >= 0)
2827 return -EBUSY;
2828 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2829 md_wakeup_thread(rdev->mddev->thread);
2830 } else if (rdev->mddev->pers) {
2831 /* Activating a spare .. or possibly reactivating
2832 * if we ever get bitmaps working here.
2833 */
2834 int err;
2835
2836 if (rdev->raid_disk != -1)
2837 return -EBUSY;
2838
2839 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2840 return -EBUSY;
2841
2842 if (rdev->mddev->pers->hot_add_disk == NULL)
2843 return -EINVAL;
2844
2845 if (slot >= rdev->mddev->raid_disks &&
2846 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2847 return -ENOSPC;
2848
2849 rdev->raid_disk = slot;
2850 if (test_bit(In_sync, &rdev->flags))
2851 rdev->saved_raid_disk = slot;
2852 else
2853 rdev->saved_raid_disk = -1;
2854 clear_bit(In_sync, &rdev->flags);
2855 clear_bit(Bitmap_sync, &rdev->flags);
2856 err = rdev->mddev->pers->
2857 hot_add_disk(rdev->mddev, rdev);
2858 if (err) {
2859 rdev->raid_disk = -1;
2860 return err;
2861 } else
2862 sysfs_notify_dirent_safe(rdev->sysfs_state);
2863 if (sysfs_link_rdev(rdev->mddev, rdev))
2864 /* failure here is OK */;
2865 /* don't wakeup anyone, leave that to userspace. */
2866 } else {
2867 if (slot >= rdev->mddev->raid_disks &&
2868 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2869 return -ENOSPC;
2870 rdev->raid_disk = slot;
2871 /* assume it is working */
2872 clear_bit(Faulty, &rdev->flags);
2873 clear_bit(WriteMostly, &rdev->flags);
2874 set_bit(In_sync, &rdev->flags);
2875 sysfs_notify_dirent_safe(rdev->sysfs_state);
2876 }
2877 return len;
2878 }
2879
2880 static struct rdev_sysfs_entry rdev_slot =
2881 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2882
2883 static ssize_t
2884 offset_show(struct md_rdev *rdev, char *page)
2885 {
2886 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2887 }
2888
2889 static ssize_t
2890 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2891 {
2892 unsigned long long offset;
2893 if (kstrtoull(buf, 10, &offset) < 0)
2894 return -EINVAL;
2895 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2896 return -EBUSY;
2897 if (rdev->sectors && rdev->mddev->external)
2898 /* Must set offset before size, so overlap checks
2899 * can be sane */
2900 return -EBUSY;
2901 rdev->data_offset = offset;
2902 rdev->new_data_offset = offset;
2903 return len;
2904 }
2905
2906 static struct rdev_sysfs_entry rdev_offset =
2907 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2908
2909 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2910 {
2911 return sprintf(page, "%llu\n",
2912 (unsigned long long)rdev->new_data_offset);
2913 }
2914
2915 static ssize_t new_offset_store(struct md_rdev *rdev,
2916 const char *buf, size_t len)
2917 {
2918 unsigned long long new_offset;
2919 struct mddev *mddev = rdev->mddev;
2920
2921 if (kstrtoull(buf, 10, &new_offset) < 0)
2922 return -EINVAL;
2923
2924 if (mddev->sync_thread ||
2925 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
2926 return -EBUSY;
2927 if (new_offset == rdev->data_offset)
2928 /* reset is always permitted */
2929 ;
2930 else if (new_offset > rdev->data_offset) {
2931 /* must not push array size beyond rdev_sectors */
2932 if (new_offset - rdev->data_offset
2933 + mddev->dev_sectors > rdev->sectors)
2934 return -E2BIG;
2935 }
2936 /* Metadata worries about other space details. */
2937
2938 /* decreasing the offset is inconsistent with a backwards
2939 * reshape.
2940 */
2941 if (new_offset < rdev->data_offset &&
2942 mddev->reshape_backwards)
2943 return -EINVAL;
2944 /* Increasing offset is inconsistent with forwards
2945 * reshape. reshape_direction should be set to
2946 * 'backwards' first.
2947 */
2948 if (new_offset > rdev->data_offset &&
2949 !mddev->reshape_backwards)
2950 return -EINVAL;
2951
2952 if (mddev->pers && mddev->persistent &&
2953 !super_types[mddev->major_version]
2954 .allow_new_offset(rdev, new_offset))
2955 return -E2BIG;
2956 rdev->new_data_offset = new_offset;
2957 if (new_offset > rdev->data_offset)
2958 mddev->reshape_backwards = 1;
2959 else if (new_offset < rdev->data_offset)
2960 mddev->reshape_backwards = 0;
2961
2962 return len;
2963 }
2964 static struct rdev_sysfs_entry rdev_new_offset =
2965 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2966
2967 static ssize_t
2968 rdev_size_show(struct md_rdev *rdev, char *page)
2969 {
2970 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2971 }
2972
2973 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2974 {
2975 /* check if two start/length pairs overlap */
2976 if (s1+l1 <= s2)
2977 return 0;
2978 if (s2+l2 <= s1)
2979 return 0;
2980 return 1;
2981 }
2982
2983 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2984 {
2985 unsigned long long blocks;
2986 sector_t new;
2987
2988 if (kstrtoull(buf, 10, &blocks) < 0)
2989 return -EINVAL;
2990
2991 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2992 return -EINVAL; /* sector conversion overflow */
2993
2994 new = blocks * 2;
2995 if (new != blocks * 2)
2996 return -EINVAL; /* unsigned long long to sector_t overflow */
2997
2998 *sectors = new;
2999 return 0;
3000 }
3001
3002 static ssize_t
3003 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3004 {
3005 struct mddev *my_mddev = rdev->mddev;
3006 sector_t oldsectors = rdev->sectors;
3007 sector_t sectors;
3008
3009 if (test_bit(Journal, &rdev->flags))
3010 return -EBUSY;
3011 if (strict_blocks_to_sectors(buf, &sectors) < 0)
3012 return -EINVAL;
3013 if (rdev->data_offset != rdev->new_data_offset)
3014 return -EINVAL; /* too confusing */
3015 if (my_mddev->pers && rdev->raid_disk >= 0) {
3016 if (my_mddev->persistent) {
3017 sectors = super_types[my_mddev->major_version].
3018 rdev_size_change(rdev, sectors);
3019 if (!sectors)
3020 return -EBUSY;
3021 } else if (!sectors)
3022 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
3023 rdev->data_offset;
3024 if (!my_mddev->pers->resize)
3025 /* Cannot change size for RAID0 or Linear etc */
3026 return -EINVAL;
3027 }
3028 if (sectors < my_mddev->dev_sectors)
3029 return -EINVAL; /* component must fit device */
3030
3031 rdev->sectors = sectors;
3032 if (sectors > oldsectors && my_mddev->external) {
3033 /* Need to check that all other rdevs with the same
3034 * ->bdev do not overlap. 'rcu' is sufficient to walk
3035 * the rdev lists safely.
3036 * This check does not provide a hard guarantee, it
3037 * just helps avoid dangerous mistakes.
3038 */
3039 struct mddev *mddev;
3040 int overlap = 0;
3041 struct list_head *tmp;
3042
3043 rcu_read_lock();
3044 for_each_mddev(mddev, tmp) {
3045 struct md_rdev *rdev2;
3046
3047 rdev_for_each(rdev2, mddev)
3048 if (rdev->bdev == rdev2->bdev &&
3049 rdev != rdev2 &&
3050 overlaps(rdev->data_offset, rdev->sectors,
3051 rdev2->data_offset,
3052 rdev2->sectors)) {
3053 overlap = 1;
3054 break;
3055 }
3056 if (overlap) {
3057 mddev_put(mddev);
3058 break;
3059 }
3060 }
3061 rcu_read_unlock();
3062 if (overlap) {
3063 /* Someone else could have slipped in a size
3064 * change here, but doing so is just silly.
3065 * We put oldsectors back because we *know* it is
3066 * safe, and trust userspace not to race with
3067 * itself
3068 */
3069 rdev->sectors = oldsectors;
3070 return -EBUSY;
3071 }
3072 }
3073 return len;
3074 }
3075
3076 static struct rdev_sysfs_entry rdev_size =
3077 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3078
3079 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3080 {
3081 unsigned long long recovery_start = rdev->recovery_offset;
3082
3083 if (test_bit(In_sync, &rdev->flags) ||
3084 recovery_start == MaxSector)
3085 return sprintf(page, "none\n");
3086
3087 return sprintf(page, "%llu\n", recovery_start);
3088 }
3089
3090 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3091 {
3092 unsigned long long recovery_start;
3093
3094 if (cmd_match(buf, "none"))
3095 recovery_start = MaxSector;
3096 else if (kstrtoull(buf, 10, &recovery_start))
3097 return -EINVAL;
3098
3099 if (rdev->mddev->pers &&
3100 rdev->raid_disk >= 0)
3101 return -EBUSY;
3102
3103 rdev->recovery_offset = recovery_start;
3104 if (recovery_start == MaxSector)
3105 set_bit(In_sync, &rdev->flags);
3106 else
3107 clear_bit(In_sync, &rdev->flags);
3108 return len;
3109 }
3110
3111 static struct rdev_sysfs_entry rdev_recovery_start =
3112 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3113
3114 /* sysfs access to bad-blocks list.
3115 * We present two files.
3116 * 'bad-blocks' lists sector numbers and lengths of ranges that
3117 * are recorded as bad. The list is truncated to fit within
3118 * the one-page limit of sysfs.
3119 * Writing "sector length" to this file adds an acknowledged
3120 * bad block list.
3121 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3122 * been acknowledged. Writing to this file adds bad blocks
3123 * without acknowledging them. This is largely for testing.
3124 */
3125 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3126 {
3127 return badblocks_show(&rdev->badblocks, page, 0);
3128 }
3129 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3130 {
3131 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3132 /* Maybe that ack was all we needed */
3133 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3134 wake_up(&rdev->blocked_wait);
3135 return rv;
3136 }
3137 static struct rdev_sysfs_entry rdev_bad_blocks =
3138 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3139
3140 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3141 {
3142 return badblocks_show(&rdev->badblocks, page, 1);
3143 }
3144 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3145 {
3146 return badblocks_store(&rdev->badblocks, page, len, 1);
3147 }
3148 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3149 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3150
3151 static struct attribute *rdev_default_attrs[] = {
3152 &rdev_state.attr,
3153 &rdev_errors.attr,
3154 &rdev_slot.attr,
3155 &rdev_offset.attr,
3156 &rdev_new_offset.attr,
3157 &rdev_size.attr,
3158 &rdev_recovery_start.attr,
3159 &rdev_bad_blocks.attr,
3160 &rdev_unack_bad_blocks.attr,
3161 NULL,
3162 };
3163 static ssize_t
3164 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3165 {
3166 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3167 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3168
3169 if (!entry->show)
3170 return -EIO;
3171 if (!rdev->mddev)
3172 return -EBUSY;
3173 return entry->show(rdev, page);
3174 }
3175
3176 static ssize_t
3177 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3178 const char *page, size_t length)
3179 {
3180 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3181 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3182 ssize_t rv;
3183 struct mddev *mddev = rdev->mddev;
3184
3185 if (!entry->store)
3186 return -EIO;
3187 if (!capable(CAP_SYS_ADMIN))
3188 return -EACCES;
3189 rv = mddev ? mddev_lock(mddev): -EBUSY;
3190 if (!rv) {
3191 if (rdev->mddev == NULL)
3192 rv = -EBUSY;
3193 else
3194 rv = entry->store(rdev, page, length);
3195 mddev_unlock(mddev);
3196 }
3197 return rv;
3198 }
3199
3200 static void rdev_free(struct kobject *ko)
3201 {
3202 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3203 kfree(rdev);
3204 }
3205 static const struct sysfs_ops rdev_sysfs_ops = {
3206 .show = rdev_attr_show,
3207 .store = rdev_attr_store,
3208 };
3209 static struct kobj_type rdev_ktype = {
3210 .release = rdev_free,
3211 .sysfs_ops = &rdev_sysfs_ops,
3212 .default_attrs = rdev_default_attrs,
3213 };
3214
3215 int md_rdev_init(struct md_rdev *rdev)
3216 {
3217 rdev->desc_nr = -1;
3218 rdev->saved_raid_disk = -1;
3219 rdev->raid_disk = -1;
3220 rdev->flags = 0;
3221 rdev->data_offset = 0;
3222 rdev->new_data_offset = 0;
3223 rdev->sb_events = 0;
3224 rdev->last_read_error = 0;
3225 rdev->sb_loaded = 0;
3226 rdev->bb_page = NULL;
3227 atomic_set(&rdev->nr_pending, 0);
3228 atomic_set(&rdev->read_errors, 0);
3229 atomic_set(&rdev->corrected_errors, 0);
3230
3231 INIT_LIST_HEAD(&rdev->same_set);
3232 init_waitqueue_head(&rdev->blocked_wait);
3233
3234 /* Add space to store bad block list.
3235 * This reserves the space even on arrays where it cannot
3236 * be used - I wonder if that matters
3237 */
3238 return badblocks_init(&rdev->badblocks, 0);
3239 }
3240 EXPORT_SYMBOL_GPL(md_rdev_init);
3241 /*
3242 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3243 *
3244 * mark the device faulty if:
3245 *
3246 * - the device is nonexistent (zero size)
3247 * - the device has no valid superblock
3248 *
3249 * a faulty rdev _never_ has rdev->sb set.
3250 */
3251 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3252 {
3253 char b[BDEVNAME_SIZE];
3254 int err;
3255 struct md_rdev *rdev;
3256 sector_t size;
3257
3258 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3259 if (!rdev)
3260 return ERR_PTR(-ENOMEM);
3261
3262 err = md_rdev_init(rdev);
3263 if (err)
3264 goto abort_free;
3265 err = alloc_disk_sb(rdev);
3266 if (err)
3267 goto abort_free;
3268
3269 err = lock_rdev(rdev, newdev, super_format == -2);
3270 if (err)
3271 goto abort_free;
3272
3273 kobject_init(&rdev->kobj, &rdev_ktype);
3274
3275 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3276 if (!size) {
3277 pr_warn("md: %s has zero or unknown size, marking faulty!\n",
3278 bdevname(rdev->bdev,b));
3279 err = -EINVAL;
3280 goto abort_free;
3281 }
3282
3283 if (super_format >= 0) {
3284 err = super_types[super_format].
3285 load_super(rdev, NULL, super_minor);
3286 if (err == -EINVAL) {
3287 pr_warn("md: %s does not have a valid v%d.%d superblock, not importing!\n",
3288 bdevname(rdev->bdev,b),
3289 super_format, super_minor);
3290 goto abort_free;
3291 }
3292 if (err < 0) {
3293 pr_warn("md: could not read %s's sb, not importing!\n",
3294 bdevname(rdev->bdev,b));
3295 goto abort_free;
3296 }
3297 }
3298
3299 return rdev;
3300
3301 abort_free:
3302 if (rdev->bdev)
3303 unlock_rdev(rdev);
3304 md_rdev_clear(rdev);
3305 kfree(rdev);
3306 return ERR_PTR(err);
3307 }
3308
3309 /*
3310 * Check a full RAID array for plausibility
3311 */
3312
3313 static void analyze_sbs(struct mddev *mddev)
3314 {
3315 int i;
3316 struct md_rdev *rdev, *freshest, *tmp;
3317 char b[BDEVNAME_SIZE];
3318
3319 freshest = NULL;
3320 rdev_for_each_safe(rdev, tmp, mddev)
3321 switch (super_types[mddev->major_version].
3322 load_super(rdev, freshest, mddev->minor_version)) {
3323 case 1:
3324 freshest = rdev;
3325 break;
3326 case 0:
3327 break;
3328 default:
3329 pr_warn("md: fatal superblock inconsistency in %s -- removing from array\n",
3330 bdevname(rdev->bdev,b));
3331 md_kick_rdev_from_array(rdev);
3332 }
3333
3334 super_types[mddev->major_version].
3335 validate_super(mddev, freshest);
3336
3337 i = 0;
3338 rdev_for_each_safe(rdev, tmp, mddev) {
3339 if (mddev->max_disks &&
3340 (rdev->desc_nr >= mddev->max_disks ||
3341 i > mddev->max_disks)) {
3342 pr_warn("md: %s: %s: only %d devices permitted\n",
3343 mdname(mddev), bdevname(rdev->bdev, b),
3344 mddev->max_disks);
3345 md_kick_rdev_from_array(rdev);
3346 continue;
3347 }
3348 if (rdev != freshest) {
3349 if (super_types[mddev->major_version].
3350 validate_super(mddev, rdev)) {
3351 pr_warn("md: kicking non-fresh %s from array!\n",
3352 bdevname(rdev->bdev,b));
3353 md_kick_rdev_from_array(rdev);
3354 continue;
3355 }
3356 }
3357 if (mddev->level == LEVEL_MULTIPATH) {
3358 rdev->desc_nr = i++;
3359 rdev->raid_disk = rdev->desc_nr;
3360 set_bit(In_sync, &rdev->flags);
3361 } else if (rdev->raid_disk >=
3362 (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3363 !test_bit(Journal, &rdev->flags)) {
3364 rdev->raid_disk = -1;
3365 clear_bit(In_sync, &rdev->flags);
3366 }
3367 }
3368 }
3369
3370 /* Read a fixed-point number.
3371 * Numbers in sysfs attributes should be in "standard" units where
3372 * possible, so time should be in seconds.
3373 * However we internally use a a much smaller unit such as
3374 * milliseconds or jiffies.
3375 * This function takes a decimal number with a possible fractional
3376 * component, and produces an integer which is the result of
3377 * multiplying that number by 10^'scale'.
3378 * all without any floating-point arithmetic.
3379 */
3380 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3381 {
3382 unsigned long result = 0;
3383 long decimals = -1;
3384 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3385 if (*cp == '.')
3386 decimals = 0;
3387 else if (decimals < scale) {
3388 unsigned int value;
3389 value = *cp - '0';
3390 result = result * 10 + value;
3391 if (decimals >= 0)
3392 decimals++;
3393 }
3394 cp++;
3395 }
3396 if (*cp == '\n')
3397 cp++;
3398 if (*cp)
3399 return -EINVAL;
3400 if (decimals < 0)
3401 decimals = 0;
3402 while (decimals < scale) {
3403 result *= 10;
3404 decimals ++;
3405 }
3406 *res = result;
3407 return 0;
3408 }
3409
3410 static ssize_t
3411 safe_delay_show(struct mddev *mddev, char *page)
3412 {
3413 int msec = (mddev->safemode_delay*1000)/HZ;
3414 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3415 }
3416 static ssize_t
3417 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3418 {
3419 unsigned long msec;
3420
3421 if (mddev_is_clustered(mddev)) {
3422 pr_warn("md: Safemode is disabled for clustered mode\n");
3423 return -EINVAL;
3424 }
3425
3426 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3427 return -EINVAL;
3428 if (msec == 0)
3429 mddev->safemode_delay = 0;
3430 else {
3431 unsigned long old_delay = mddev->safemode_delay;
3432 unsigned long new_delay = (msec*HZ)/1000;
3433
3434 if (new_delay == 0)
3435 new_delay = 1;
3436 mddev->safemode_delay = new_delay;
3437 if (new_delay < old_delay || old_delay == 0)
3438 mod_timer(&mddev->safemode_timer, jiffies+1);
3439 }
3440 return len;
3441 }
3442 static struct md_sysfs_entry md_safe_delay =
3443 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3444
3445 static ssize_t
3446 level_show(struct mddev *mddev, char *page)
3447 {
3448 struct md_personality *p;
3449 int ret;
3450 spin_lock(&mddev->lock);
3451 p = mddev->pers;
3452 if (p)
3453 ret = sprintf(page, "%s\n", p->name);
3454 else if (mddev->clevel[0])
3455 ret = sprintf(page, "%s\n", mddev->clevel);
3456 else if (mddev->level != LEVEL_NONE)
3457 ret = sprintf(page, "%d\n", mddev->level);
3458 else
3459 ret = 0;
3460 spin_unlock(&mddev->lock);
3461 return ret;
3462 }
3463
3464 static ssize_t
3465 level_store(struct mddev *mddev, const char *buf, size_t len)
3466 {
3467 char clevel[16];
3468 ssize_t rv;
3469 size_t slen = len;
3470 struct md_personality *pers, *oldpers;
3471 long level;
3472 void *priv, *oldpriv;
3473 struct md_rdev *rdev;
3474
3475 if (slen == 0 || slen >= sizeof(clevel))
3476 return -EINVAL;
3477
3478 rv = mddev_lock(mddev);
3479 if (rv)
3480 return rv;
3481
3482 if (mddev->pers == NULL) {
3483 strncpy(mddev->clevel, buf, slen);
3484 if (mddev->clevel[slen-1] == '\n')
3485 slen--;
3486 mddev->clevel[slen] = 0;
3487 mddev->level = LEVEL_NONE;
3488 rv = len;
3489 goto out_unlock;
3490 }
3491 rv = -EROFS;
3492 if (mddev->ro)
3493 goto out_unlock;
3494
3495 /* request to change the personality. Need to ensure:
3496 * - array is not engaged in resync/recovery/reshape
3497 * - old personality can be suspended
3498 * - new personality will access other array.
3499 */
3500
3501 rv = -EBUSY;
3502 if (mddev->sync_thread ||
3503 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3504 mddev->reshape_position != MaxSector ||
3505 mddev->sysfs_active)
3506 goto out_unlock;
3507
3508 rv = -EINVAL;
3509 if (!mddev->pers->quiesce) {
3510 pr_warn("md: %s: %s does not support online personality change\n",
3511 mdname(mddev), mddev->pers->name);
3512 goto out_unlock;
3513 }
3514
3515 /* Now find the new personality */
3516 strncpy(clevel, buf, slen);
3517 if (clevel[slen-1] == '\n')
3518 slen--;
3519 clevel[slen] = 0;
3520 if (kstrtol(clevel, 10, &level))
3521 level = LEVEL_NONE;
3522
3523 if (request_module("md-%s", clevel) != 0)
3524 request_module("md-level-%s", clevel);
3525 spin_lock(&pers_lock);
3526 pers = find_pers(level, clevel);
3527 if (!pers || !try_module_get(pers->owner)) {
3528 spin_unlock(&pers_lock);
3529 pr_warn("md: personality %s not loaded\n", clevel);
3530 rv = -EINVAL;
3531 goto out_unlock;
3532 }
3533 spin_unlock(&pers_lock);
3534
3535 if (pers == mddev->pers) {
3536 /* Nothing to do! */
3537 module_put(pers->owner);
3538 rv = len;
3539 goto out_unlock;
3540 }
3541 if (!pers->takeover) {
3542 module_put(pers->owner);
3543 pr_warn("md: %s: %s does not support personality takeover\n",
3544 mdname(mddev), clevel);
3545 rv = -EINVAL;
3546 goto out_unlock;
3547 }
3548
3549 rdev_for_each(rdev, mddev)
3550 rdev->new_raid_disk = rdev->raid_disk;
3551
3552 /* ->takeover must set new_* and/or delta_disks
3553 * if it succeeds, and may set them when it fails.
3554 */
3555 priv = pers->takeover(mddev);
3556 if (IS_ERR(priv)) {
3557 mddev->new_level = mddev->level;
3558 mddev->new_layout = mddev->layout;
3559 mddev->new_chunk_sectors = mddev->chunk_sectors;
3560 mddev->raid_disks -= mddev->delta_disks;
3561 mddev->delta_disks = 0;
3562 mddev->reshape_backwards = 0;
3563 module_put(pers->owner);
3564 pr_warn("md: %s: %s would not accept array\n",
3565 mdname(mddev), clevel);
3566 rv = PTR_ERR(priv);
3567 goto out_unlock;
3568 }
3569
3570 /* Looks like we have a winner */
3571 mddev_suspend(mddev);
3572 mddev_detach(mddev);
3573
3574 spin_lock(&mddev->lock);
3575 oldpers = mddev->pers;
3576 oldpriv = mddev->private;
3577 mddev->pers = pers;
3578 mddev->private = priv;
3579 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3580 mddev->level = mddev->new_level;
3581 mddev->layout = mddev->new_layout;
3582 mddev->chunk_sectors = mddev->new_chunk_sectors;
3583 mddev->delta_disks = 0;
3584 mddev->reshape_backwards = 0;
3585 mddev->degraded = 0;
3586 spin_unlock(&mddev->lock);
3587
3588 if (oldpers->sync_request == NULL &&
3589 mddev->external) {
3590 /* We are converting from a no-redundancy array
3591 * to a redundancy array and metadata is managed
3592 * externally so we need to be sure that writes
3593 * won't block due to a need to transition
3594 * clean->dirty
3595 * until external management is started.
3596 */
3597 mddev->in_sync = 0;
3598 mddev->safemode_delay = 0;
3599 mddev->safemode = 0;
3600 }
3601
3602 oldpers->free(mddev, oldpriv);
3603
3604 if (oldpers->sync_request == NULL &&
3605 pers->sync_request != NULL) {
3606 /* need to add the md_redundancy_group */
3607 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3608 pr_warn("md: cannot register extra attributes for %s\n",
3609 mdname(mddev));
3610 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3611 }
3612 if (oldpers->sync_request != NULL &&
3613 pers->sync_request == NULL) {
3614 /* need to remove the md_redundancy_group */
3615 if (mddev->to_remove == NULL)
3616 mddev->to_remove = &md_redundancy_group;
3617 }
3618
3619 module_put(oldpers->owner);
3620
3621 rdev_for_each(rdev, mddev) {
3622 if (rdev->raid_disk < 0)
3623 continue;
3624 if (rdev->new_raid_disk >= mddev->raid_disks)
3625 rdev->new_raid_disk = -1;
3626 if (rdev->new_raid_disk == rdev->raid_disk)
3627 continue;
3628 sysfs_unlink_rdev(mddev, rdev);
3629 }
3630 rdev_for_each(rdev, mddev) {
3631 if (rdev->raid_disk < 0)
3632 continue;
3633 if (rdev->new_raid_disk == rdev->raid_disk)
3634 continue;
3635 rdev->raid_disk = rdev->new_raid_disk;
3636 if (rdev->raid_disk < 0)
3637 clear_bit(In_sync, &rdev->flags);
3638 else {
3639 if (sysfs_link_rdev(mddev, rdev))
3640 pr_warn("md: cannot register rd%d for %s after level change\n",
3641 rdev->raid_disk, mdname(mddev));
3642 }
3643 }
3644
3645 if (pers->sync_request == NULL) {
3646 /* this is now an array without redundancy, so
3647 * it must always be in_sync
3648 */
3649 mddev->in_sync = 1;
3650 del_timer_sync(&mddev->safemode_timer);
3651 }
3652 blk_set_stacking_limits(&mddev->queue->limits);
3653 pers->run(mddev);
3654 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
3655 mddev_resume(mddev);
3656 if (!mddev->thread)
3657 md_update_sb(mddev, 1);
3658 sysfs_notify(&mddev->kobj, NULL, "level");
3659 md_new_event(mddev);
3660 rv = len;
3661 out_unlock:
3662 mddev_unlock(mddev);
3663 return rv;
3664 }
3665
3666 static struct md_sysfs_entry md_level =
3667 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3668
3669 static ssize_t
3670 layout_show(struct mddev *mddev, char *page)
3671 {
3672 /* just a number, not meaningful for all levels */
3673 if (mddev->reshape_position != MaxSector &&
3674 mddev->layout != mddev->new_layout)
3675 return sprintf(page, "%d (%d)\n",
3676 mddev->new_layout, mddev->layout);
3677 return sprintf(page, "%d\n", mddev->layout);
3678 }
3679
3680 static ssize_t
3681 layout_store(struct mddev *mddev, const char *buf, size_t len)
3682 {
3683 unsigned int n;
3684 int err;
3685
3686 err = kstrtouint(buf, 10, &n);
3687 if (err < 0)
3688 return err;
3689 err = mddev_lock(mddev);
3690 if (err)
3691 return err;
3692
3693 if (mddev->pers) {
3694 if (mddev->pers->check_reshape == NULL)
3695 err = -EBUSY;
3696 else if (mddev->ro)
3697 err = -EROFS;
3698 else {
3699 mddev->new_layout = n;
3700 err = mddev->pers->check_reshape(mddev);
3701 if (err)
3702 mddev->new_layout = mddev->layout;
3703 }
3704 } else {
3705 mddev->new_layout = n;
3706 if (mddev->reshape_position == MaxSector)
3707 mddev->layout = n;
3708 }
3709 mddev_unlock(mddev);
3710 return err ?: len;
3711 }
3712 static struct md_sysfs_entry md_layout =
3713 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3714
3715 static ssize_t
3716 raid_disks_show(struct mddev *mddev, char *page)
3717 {
3718 if (mddev->raid_disks == 0)
3719 return 0;
3720 if (mddev->reshape_position != MaxSector &&
3721 mddev->delta_disks != 0)
3722 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3723 mddev->raid_disks - mddev->delta_disks);
3724 return sprintf(page, "%d\n", mddev->raid_disks);
3725 }
3726
3727 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3728
3729 static ssize_t
3730 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3731 {
3732 unsigned int n;
3733 int err;
3734
3735 err = kstrtouint(buf, 10, &n);
3736 if (err < 0)
3737 return err;
3738
3739 err = mddev_lock(mddev);
3740 if (err)
3741 return err;
3742 if (mddev->pers)
3743 err = update_raid_disks(mddev, n);
3744 else if (mddev->reshape_position != MaxSector) {
3745 struct md_rdev *rdev;
3746 int olddisks = mddev->raid_disks - mddev->delta_disks;
3747
3748 err = -EINVAL;
3749 rdev_for_each(rdev, mddev) {
3750 if (olddisks < n &&
3751 rdev->data_offset < rdev->new_data_offset)
3752 goto out_unlock;
3753 if (olddisks > n &&
3754 rdev->data_offset > rdev->new_data_offset)
3755 goto out_unlock;
3756 }
3757 err = 0;
3758 mddev->delta_disks = n - olddisks;
3759 mddev->raid_disks = n;
3760 mddev->reshape_backwards = (mddev->delta_disks < 0);
3761 } else
3762 mddev->raid_disks = n;
3763 out_unlock:
3764 mddev_unlock(mddev);
3765 return err ? err : len;
3766 }
3767 static struct md_sysfs_entry md_raid_disks =
3768 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3769
3770 static ssize_t
3771 chunk_size_show(struct mddev *mddev, char *page)
3772 {
3773 if (mddev->reshape_position != MaxSector &&
3774 mddev->chunk_sectors != mddev->new_chunk_sectors)
3775 return sprintf(page, "%d (%d)\n",
3776 mddev->new_chunk_sectors << 9,
3777 mddev->chunk_sectors << 9);
3778 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3779 }
3780
3781 static ssize_t
3782 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3783 {
3784 unsigned long n;
3785 int err;
3786
3787 err = kstrtoul(buf, 10, &n);
3788 if (err < 0)
3789 return err;
3790
3791 err = mddev_lock(mddev);
3792 if (err)
3793 return err;
3794 if (mddev->pers) {
3795 if (mddev->pers->check_reshape == NULL)
3796 err = -EBUSY;
3797 else if (mddev->ro)
3798 err = -EROFS;
3799 else {
3800 mddev->new_chunk_sectors = n >> 9;
3801 err = mddev->pers->check_reshape(mddev);
3802 if (err)
3803 mddev->new_chunk_sectors = mddev->chunk_sectors;
3804 }
3805 } else {
3806 mddev->new_chunk_sectors = n >> 9;
3807 if (mddev->reshape_position == MaxSector)
3808 mddev->chunk_sectors = n >> 9;
3809 }
3810 mddev_unlock(mddev);
3811 return err ?: len;
3812 }
3813 static struct md_sysfs_entry md_chunk_size =
3814 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3815
3816 static ssize_t
3817 resync_start_show(struct mddev *mddev, char *page)
3818 {
3819 if (mddev->recovery_cp == MaxSector)
3820 return sprintf(page, "none\n");
3821 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3822 }
3823
3824 static ssize_t
3825 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3826 {
3827 unsigned long long n;
3828 int err;
3829
3830 if (cmd_match(buf, "none"))
3831 n = MaxSector;
3832 else {
3833 err = kstrtoull(buf, 10, &n);
3834 if (err < 0)
3835 return err;
3836 if (n != (sector_t)n)
3837 return -EINVAL;
3838 }
3839
3840 err = mddev_lock(mddev);
3841 if (err)
3842 return err;
3843 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3844 err = -EBUSY;
3845
3846 if (!err) {
3847 mddev->recovery_cp = n;
3848 if (mddev->pers)
3849 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
3850 }
3851 mddev_unlock(mddev);
3852 return err ?: len;
3853 }
3854 static struct md_sysfs_entry md_resync_start =
3855 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
3856 resync_start_show, resync_start_store);
3857
3858 /*
3859 * The array state can be:
3860 *
3861 * clear
3862 * No devices, no size, no level
3863 * Equivalent to STOP_ARRAY ioctl
3864 * inactive
3865 * May have some settings, but array is not active
3866 * all IO results in error
3867 * When written, doesn't tear down array, but just stops it
3868 * suspended (not supported yet)
3869 * All IO requests will block. The array can be reconfigured.
3870 * Writing this, if accepted, will block until array is quiescent
3871 * readonly
3872 * no resync can happen. no superblocks get written.
3873 * write requests fail
3874 * read-auto
3875 * like readonly, but behaves like 'clean' on a write request.
3876 *
3877 * clean - no pending writes, but otherwise active.
3878 * When written to inactive array, starts without resync
3879 * If a write request arrives then
3880 * if metadata is known, mark 'dirty' and switch to 'active'.
3881 * if not known, block and switch to write-pending
3882 * If written to an active array that has pending writes, then fails.
3883 * active
3884 * fully active: IO and resync can be happening.
3885 * When written to inactive array, starts with resync
3886 *
3887 * write-pending
3888 * clean, but writes are blocked waiting for 'active' to be written.
3889 *
3890 * active-idle
3891 * like active, but no writes have been seen for a while (100msec).
3892 *
3893 */
3894 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3895 write_pending, active_idle, bad_word};
3896 static char *array_states[] = {
3897 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3898 "write-pending", "active-idle", NULL };
3899
3900 static int match_word(const char *word, char **list)
3901 {
3902 int n;
3903 for (n=0; list[n]; n++)
3904 if (cmd_match(word, list[n]))
3905 break;
3906 return n;
3907 }
3908
3909 static ssize_t
3910 array_state_show(struct mddev *mddev, char *page)
3911 {
3912 enum array_state st = inactive;
3913
3914 if (mddev->pers)
3915 switch(mddev->ro) {
3916 case 1:
3917 st = readonly;
3918 break;
3919 case 2:
3920 st = read_auto;
3921 break;
3922 case 0:
3923 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
3924 st = write_pending;
3925 else if (mddev->in_sync)
3926 st = clean;
3927 else if (mddev->safemode)
3928 st = active_idle;
3929 else
3930 st = active;
3931 }
3932 else {
3933 if (list_empty(&mddev->disks) &&
3934 mddev->raid_disks == 0 &&
3935 mddev->dev_sectors == 0)
3936 st = clear;
3937 else
3938 st = inactive;
3939 }
3940 return sprintf(page, "%s\n", array_states[st]);
3941 }
3942
3943 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
3944 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
3945 static int do_md_run(struct mddev *mddev);
3946 static int restart_array(struct mddev *mddev);
3947
3948 static ssize_t
3949 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3950 {
3951 int err;
3952 enum array_state st = match_word(buf, array_states);
3953
3954 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
3955 /* don't take reconfig_mutex when toggling between
3956 * clean and active
3957 */
3958 spin_lock(&mddev->lock);
3959 if (st == active) {
3960 restart_array(mddev);
3961 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
3962 md_wakeup_thread(mddev->thread);
3963 wake_up(&mddev->sb_wait);
3964 err = 0;
3965 } else /* st == clean */ {
3966 restart_array(mddev);
3967 if (atomic_read(&mddev->writes_pending) == 0) {
3968 if (mddev->in_sync == 0) {
3969 mddev->in_sync = 1;
3970 if (mddev->safemode == 1)
3971 mddev->safemode = 0;
3972 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
3973 }
3974 err = 0;
3975 } else
3976 err = -EBUSY;
3977 }
3978 if (!err)
3979 sysfs_notify_dirent_safe(mddev->sysfs_state);
3980 spin_unlock(&mddev->lock);
3981 return err ?: len;
3982 }
3983 err = mddev_lock(mddev);
3984 if (err)
3985 return err;
3986 err = -EINVAL;
3987 switch(st) {
3988 case bad_word:
3989 break;
3990 case clear:
3991 /* stopping an active array */
3992 err = do_md_stop(mddev, 0, NULL);
3993 break;
3994 case inactive:
3995 /* stopping an active array */
3996 if (mddev->pers)
3997 err = do_md_stop(mddev, 2, NULL);
3998 else
3999 err = 0; /* already inactive */
4000 break;
4001 case suspended:
4002 break; /* not supported yet */
4003 case readonly:
4004 if (mddev->pers)
4005 err = md_set_readonly(mddev, NULL);
4006 else {
4007 mddev->ro = 1;
4008 set_disk_ro(mddev->gendisk, 1);
4009 err = do_md_run(mddev);
4010 }
4011 break;
4012 case read_auto:
4013 if (mddev->pers) {
4014 if (mddev->ro == 0)
4015 err = md_set_readonly(mddev, NULL);
4016 else if (mddev->ro == 1)
4017 err = restart_array(mddev);
4018 if (err == 0) {
4019 mddev->ro = 2;
4020 set_disk_ro(mddev->gendisk, 0);
4021 }
4022 } else {
4023 mddev->ro = 2;
4024 err = do_md_run(mddev);
4025 }
4026 break;
4027 case clean:
4028 if (mddev->pers) {
4029 err = restart_array(mddev);
4030 if (err)
4031 break;
4032 spin_lock(&mddev->lock);
4033 if (atomic_read(&mddev->writes_pending) == 0) {
4034 if (mddev->in_sync == 0) {
4035 mddev->in_sync = 1;
4036 if (mddev->safemode == 1)
4037 mddev->safemode = 0;
4038 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
4039 }
4040 err = 0;
4041 } else
4042 err = -EBUSY;
4043 spin_unlock(&mddev->lock);
4044 } else
4045 err = -EINVAL;
4046 break;
4047 case active:
4048 if (mddev->pers) {
4049 err = restart_array(mddev);
4050 if (err)
4051 break;
4052 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4053 wake_up(&mddev->sb_wait);
4054 err = 0;
4055 } else {
4056 mddev->ro = 0;
4057 set_disk_ro(mddev->gendisk, 0);
4058 err = do_md_run(mddev);
4059 }
4060 break;
4061 case write_pending:
4062 case active_idle:
4063 /* these cannot be set */
4064 break;
4065 }
4066
4067 if (!err) {
4068 if (mddev->hold_active == UNTIL_IOCTL)
4069 mddev->hold_active = 0;
4070 sysfs_notify_dirent_safe(mddev->sysfs_state);
4071 }
4072 mddev_unlock(mddev);
4073 return err ?: len;
4074 }
4075 static struct md_sysfs_entry md_array_state =
4076 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4077
4078 static ssize_t
4079 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4080 return sprintf(page, "%d\n",
4081 atomic_read(&mddev->max_corr_read_errors));
4082 }
4083
4084 static ssize_t
4085 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4086 {
4087 unsigned int n;
4088 int rv;
4089
4090 rv = kstrtouint(buf, 10, &n);
4091 if (rv < 0)
4092 return rv;
4093 atomic_set(&mddev->max_corr_read_errors, n);
4094 return len;
4095 }
4096
4097 static struct md_sysfs_entry max_corr_read_errors =
4098 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4099 max_corrected_read_errors_store);
4100
4101 static ssize_t
4102 null_show(struct mddev *mddev, char *page)
4103 {
4104 return -EINVAL;
4105 }
4106
4107 static ssize_t
4108 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4109 {
4110 /* buf must be %d:%d\n? giving major and minor numbers */
4111 /* The new device is added to the array.
4112 * If the array has a persistent superblock, we read the
4113 * superblock to initialise info and check validity.
4114 * Otherwise, only checking done is that in bind_rdev_to_array,
4115 * which mainly checks size.
4116 */
4117 char *e;
4118 int major = simple_strtoul(buf, &e, 10);
4119 int minor;
4120 dev_t dev;
4121 struct md_rdev *rdev;
4122 int err;
4123
4124 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4125 return -EINVAL;
4126 minor = simple_strtoul(e+1, &e, 10);
4127 if (*e && *e != '\n')
4128 return -EINVAL;
4129 dev = MKDEV(major, minor);
4130 if (major != MAJOR(dev) ||
4131 minor != MINOR(dev))
4132 return -EOVERFLOW;
4133
4134 flush_workqueue(md_misc_wq);
4135
4136 err = mddev_lock(mddev);
4137 if (err)
4138 return err;
4139 if (mddev->persistent) {
4140 rdev = md_import_device(dev, mddev->major_version,
4141 mddev->minor_version);
4142 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4143 struct md_rdev *rdev0
4144 = list_entry(mddev->disks.next,
4145 struct md_rdev, same_set);
4146 err = super_types[mddev->major_version]
4147 .load_super(rdev, rdev0, mddev->minor_version);
4148 if (err < 0)
4149 goto out;
4150 }
4151 } else if (mddev->external)
4152 rdev = md_import_device(dev, -2, -1);
4153 else
4154 rdev = md_import_device(dev, -1, -1);
4155
4156 if (IS_ERR(rdev)) {
4157 mddev_unlock(mddev);
4158 return PTR_ERR(rdev);
4159 }
4160 err = bind_rdev_to_array(rdev, mddev);
4161 out:
4162 if (err)
4163 export_rdev(rdev);
4164 mddev_unlock(mddev);
4165 return err ? err : len;
4166 }
4167
4168 static struct md_sysfs_entry md_new_device =
4169 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4170
4171 static ssize_t
4172 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4173 {
4174 char *end;
4175 unsigned long chunk, end_chunk;
4176 int err;
4177
4178 err = mddev_lock(mddev);
4179 if (err)
4180 return err;
4181 if (!mddev->bitmap)
4182 goto out;
4183 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4184 while (*buf) {
4185 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4186 if (buf == end) break;
4187 if (*end == '-') { /* range */
4188 buf = end + 1;
4189 end_chunk = simple_strtoul(buf, &end, 0);
4190 if (buf == end) break;
4191 }
4192 if (*end && !isspace(*end)) break;
4193 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4194 buf = skip_spaces(end);
4195 }
4196 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4197 out:
4198 mddev_unlock(mddev);
4199 return len;
4200 }
4201
4202 static struct md_sysfs_entry md_bitmap =
4203 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4204
4205 static ssize_t
4206 size_show(struct mddev *mddev, char *page)
4207 {
4208 return sprintf(page, "%llu\n",
4209 (unsigned long long)mddev->dev_sectors / 2);
4210 }
4211
4212 static int update_size(struct mddev *mddev, sector_t num_sectors);
4213
4214 static ssize_t
4215 size_store(struct mddev *mddev, const char *buf, size_t len)
4216 {
4217 /* If array is inactive, we can reduce the component size, but
4218 * not increase it (except from 0).
4219 * If array is active, we can try an on-line resize
4220 */
4221 sector_t sectors;
4222 int err = strict_blocks_to_sectors(buf, &sectors);
4223
4224 if (err < 0)
4225 return err;
4226 err = mddev_lock(mddev);
4227 if (err)
4228 return err;
4229 if (mddev->pers) {
4230 err = update_size(mddev, sectors);
4231 if (err == 0)
4232 md_update_sb(mddev, 1);
4233 } else {
4234 if (mddev->dev_sectors == 0 ||
4235 mddev->dev_sectors > sectors)
4236 mddev->dev_sectors = sectors;
4237 else
4238 err = -ENOSPC;
4239 }
4240 mddev_unlock(mddev);
4241 return err ? err : len;
4242 }
4243
4244 static struct md_sysfs_entry md_size =
4245 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4246
4247 /* Metadata version.
4248 * This is one of
4249 * 'none' for arrays with no metadata (good luck...)
4250 * 'external' for arrays with externally managed metadata,
4251 * or N.M for internally known formats
4252 */
4253 static ssize_t
4254 metadata_show(struct mddev *mddev, char *page)
4255 {
4256 if (mddev->persistent)
4257 return sprintf(page, "%d.%d\n",
4258 mddev->major_version, mddev->minor_version);
4259 else if (mddev->external)
4260 return sprintf(page, "external:%s\n", mddev->metadata_type);
4261 else
4262 return sprintf(page, "none\n");
4263 }
4264
4265 static ssize_t
4266 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4267 {
4268 int major, minor;
4269 char *e;
4270 int err;
4271 /* Changing the details of 'external' metadata is
4272 * always permitted. Otherwise there must be
4273 * no devices attached to the array.
4274 */
4275
4276 err = mddev_lock(mddev);
4277 if (err)
4278 return err;
4279 err = -EBUSY;
4280 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4281 ;
4282 else if (!list_empty(&mddev->disks))
4283 goto out_unlock;
4284
4285 err = 0;
4286 if (cmd_match(buf, "none")) {
4287 mddev->persistent = 0;
4288 mddev->external = 0;
4289 mddev->major_version = 0;
4290 mddev->minor_version = 90;
4291 goto out_unlock;
4292 }
4293 if (strncmp(buf, "external:", 9) == 0) {
4294 size_t namelen = len-9;
4295 if (namelen >= sizeof(mddev->metadata_type))
4296 namelen = sizeof(mddev->metadata_type)-1;
4297 strncpy(mddev->metadata_type, buf+9, namelen);
4298 mddev->metadata_type[namelen] = 0;
4299 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4300 mddev->metadata_type[--namelen] = 0;
4301 mddev->persistent = 0;
4302 mddev->external = 1;
4303 mddev->major_version = 0;
4304 mddev->minor_version = 90;
4305 goto out_unlock;
4306 }
4307 major = simple_strtoul(buf, &e, 10);
4308 err = -EINVAL;
4309 if (e==buf || *e != '.')
4310 goto out_unlock;
4311 buf = e+1;
4312 minor = simple_strtoul(buf, &e, 10);
4313 if (e==buf || (*e && *e != '\n') )
4314 goto out_unlock;
4315 err = -ENOENT;
4316 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4317 goto out_unlock;
4318 mddev->major_version = major;
4319 mddev->minor_version = minor;
4320 mddev->persistent = 1;
4321 mddev->external = 0;
4322 err = 0;
4323 out_unlock:
4324 mddev_unlock(mddev);
4325 return err ?: len;
4326 }
4327
4328 static struct md_sysfs_entry md_metadata =
4329 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4330
4331 static ssize_t
4332 action_show(struct mddev *mddev, char *page)
4333 {
4334 char *type = "idle";
4335 unsigned long recovery = mddev->recovery;
4336 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4337 type = "frozen";
4338 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4339 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4340 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4341 type = "reshape";
4342 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4343 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4344 type = "resync";
4345 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4346 type = "check";
4347 else
4348 type = "repair";
4349 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4350 type = "recover";
4351 else if (mddev->reshape_position != MaxSector)
4352 type = "reshape";
4353 }
4354 return sprintf(page, "%s\n", type);
4355 }
4356
4357 static ssize_t
4358 action_store(struct mddev *mddev, const char *page, size_t len)
4359 {
4360 if (!mddev->pers || !mddev->pers->sync_request)
4361 return -EINVAL;
4362
4363
4364 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4365 if (cmd_match(page, "frozen"))
4366 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4367 else
4368 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4369 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4370 mddev_lock(mddev) == 0) {
4371 flush_workqueue(md_misc_wq);
4372 if (mddev->sync_thread) {
4373 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4374 md_reap_sync_thread(mddev);
4375 }
4376 mddev_unlock(mddev);
4377 }
4378 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4379 return -EBUSY;
4380 else if (cmd_match(page, "resync"))
4381 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4382 else if (cmd_match(page, "recover")) {
4383 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4384 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4385 } else if (cmd_match(page, "reshape")) {
4386 int err;
4387 if (mddev->pers->start_reshape == NULL)
4388 return -EINVAL;
4389 err = mddev_lock(mddev);
4390 if (!err) {
4391 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4392 err = -EBUSY;
4393 else {
4394 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4395 err = mddev->pers->start_reshape(mddev);
4396 }
4397 mddev_unlock(mddev);
4398 }
4399 if (err)
4400 return err;
4401 sysfs_notify(&mddev->kobj, NULL, "degraded");
4402 } else {
4403 if (cmd_match(page, "check"))
4404 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4405 else if (!cmd_match(page, "repair"))
4406 return -EINVAL;
4407 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4408 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4409 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4410 }
4411 if (mddev->ro == 2) {
4412 /* A write to sync_action is enough to justify
4413 * canceling read-auto mode
4414 */
4415 mddev->ro = 0;
4416 md_wakeup_thread(mddev->sync_thread);
4417 }
4418 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4419 md_wakeup_thread(mddev->thread);
4420 sysfs_notify_dirent_safe(mddev->sysfs_action);
4421 return len;
4422 }
4423
4424 static struct md_sysfs_entry md_scan_mode =
4425 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4426
4427 static ssize_t
4428 last_sync_action_show(struct mddev *mddev, char *page)
4429 {
4430 return sprintf(page, "%s\n", mddev->last_sync_action);
4431 }
4432
4433 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4434
4435 static ssize_t
4436 mismatch_cnt_show(struct mddev *mddev, char *page)
4437 {
4438 return sprintf(page, "%llu\n",
4439 (unsigned long long)
4440 atomic64_read(&mddev->resync_mismatches));
4441 }
4442
4443 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4444
4445 static ssize_t
4446 sync_min_show(struct mddev *mddev, char *page)
4447 {
4448 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4449 mddev->sync_speed_min ? "local": "system");
4450 }
4451
4452 static ssize_t
4453 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4454 {
4455 unsigned int min;
4456 int rv;
4457
4458 if (strncmp(buf, "system", 6)==0) {
4459 min = 0;
4460 } else {
4461 rv = kstrtouint(buf, 10, &min);
4462 if (rv < 0)
4463 return rv;
4464 if (min == 0)
4465 return -EINVAL;
4466 }
4467 mddev->sync_speed_min = min;
4468 return len;
4469 }
4470
4471 static struct md_sysfs_entry md_sync_min =
4472 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4473
4474 static ssize_t
4475 sync_max_show(struct mddev *mddev, char *page)
4476 {
4477 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4478 mddev->sync_speed_max ? "local": "system");
4479 }
4480
4481 static ssize_t
4482 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4483 {
4484 unsigned int max;
4485 int rv;
4486
4487 if (strncmp(buf, "system", 6)==0) {
4488 max = 0;
4489 } else {
4490 rv = kstrtouint(buf, 10, &max);
4491 if (rv < 0)
4492 return rv;
4493 if (max == 0)
4494 return -EINVAL;
4495 }
4496 mddev->sync_speed_max = max;
4497 return len;
4498 }
4499
4500 static struct md_sysfs_entry md_sync_max =
4501 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4502
4503 static ssize_t
4504 degraded_show(struct mddev *mddev, char *page)
4505 {
4506 return sprintf(page, "%d\n", mddev->degraded);
4507 }
4508 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4509
4510 static ssize_t
4511 sync_force_parallel_show(struct mddev *mddev, char *page)
4512 {
4513 return sprintf(page, "%d\n", mddev->parallel_resync);
4514 }
4515
4516 static ssize_t
4517 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4518 {
4519 long n;
4520
4521 if (kstrtol(buf, 10, &n))
4522 return -EINVAL;
4523
4524 if (n != 0 && n != 1)
4525 return -EINVAL;
4526
4527 mddev->parallel_resync = n;
4528
4529 if (mddev->sync_thread)
4530 wake_up(&resync_wait);
4531
4532 return len;
4533 }
4534
4535 /* force parallel resync, even with shared block devices */
4536 static struct md_sysfs_entry md_sync_force_parallel =
4537 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4538 sync_force_parallel_show, sync_force_parallel_store);
4539
4540 static ssize_t
4541 sync_speed_show(struct mddev *mddev, char *page)
4542 {
4543 unsigned long resync, dt, db;
4544 if (mddev->curr_resync == 0)
4545 return sprintf(page, "none\n");
4546 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4547 dt = (jiffies - mddev->resync_mark) / HZ;
4548 if (!dt) dt++;
4549 db = resync - mddev->resync_mark_cnt;
4550 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4551 }
4552
4553 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4554
4555 static ssize_t
4556 sync_completed_show(struct mddev *mddev, char *page)
4557 {
4558 unsigned long long max_sectors, resync;
4559
4560 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4561 return sprintf(page, "none\n");
4562
4563 if (mddev->curr_resync == 1 ||
4564 mddev->curr_resync == 2)
4565 return sprintf(page, "delayed\n");
4566
4567 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4568 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4569 max_sectors = mddev->resync_max_sectors;
4570 else
4571 max_sectors = mddev->dev_sectors;
4572
4573 resync = mddev->curr_resync_completed;
4574 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4575 }
4576
4577 static struct md_sysfs_entry md_sync_completed =
4578 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
4579
4580 static ssize_t
4581 min_sync_show(struct mddev *mddev, char *page)
4582 {
4583 return sprintf(page, "%llu\n",
4584 (unsigned long long)mddev->resync_min);
4585 }
4586 static ssize_t
4587 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4588 {
4589 unsigned long long min;
4590 int err;
4591
4592 if (kstrtoull(buf, 10, &min))
4593 return -EINVAL;
4594
4595 spin_lock(&mddev->lock);
4596 err = -EINVAL;
4597 if (min > mddev->resync_max)
4598 goto out_unlock;
4599
4600 err = -EBUSY;
4601 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4602 goto out_unlock;
4603
4604 /* Round down to multiple of 4K for safety */
4605 mddev->resync_min = round_down(min, 8);
4606 err = 0;
4607
4608 out_unlock:
4609 spin_unlock(&mddev->lock);
4610 return err ?: len;
4611 }
4612
4613 static struct md_sysfs_entry md_min_sync =
4614 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4615
4616 static ssize_t
4617 max_sync_show(struct mddev *mddev, char *page)
4618 {
4619 if (mddev->resync_max == MaxSector)
4620 return sprintf(page, "max\n");
4621 else
4622 return sprintf(page, "%llu\n",
4623 (unsigned long long)mddev->resync_max);
4624 }
4625 static ssize_t
4626 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4627 {
4628 int err;
4629 spin_lock(&mddev->lock);
4630 if (strncmp(buf, "max", 3) == 0)
4631 mddev->resync_max = MaxSector;
4632 else {
4633 unsigned long long max;
4634 int chunk;
4635
4636 err = -EINVAL;
4637 if (kstrtoull(buf, 10, &max))
4638 goto out_unlock;
4639 if (max < mddev->resync_min)
4640 goto out_unlock;
4641
4642 err = -EBUSY;
4643 if (max < mddev->resync_max &&
4644 mddev->ro == 0 &&
4645 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4646 goto out_unlock;
4647
4648 /* Must be a multiple of chunk_size */
4649 chunk = mddev->chunk_sectors;
4650 if (chunk) {
4651 sector_t temp = max;
4652
4653 err = -EINVAL;
4654 if (sector_div(temp, chunk))
4655 goto out_unlock;
4656 }
4657 mddev->resync_max = max;
4658 }
4659 wake_up(&mddev->recovery_wait);
4660 err = 0;
4661 out_unlock:
4662 spin_unlock(&mddev->lock);
4663 return err ?: len;
4664 }
4665
4666 static struct md_sysfs_entry md_max_sync =
4667 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4668
4669 static ssize_t
4670 suspend_lo_show(struct mddev *mddev, char *page)
4671 {
4672 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4673 }
4674
4675 static ssize_t
4676 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4677 {
4678 unsigned long long old, new;
4679 int err;
4680
4681 err = kstrtoull(buf, 10, &new);
4682 if (err < 0)
4683 return err;
4684 if (new != (sector_t)new)
4685 return -EINVAL;
4686
4687 err = mddev_lock(mddev);
4688 if (err)
4689 return err;
4690 err = -EINVAL;
4691 if (mddev->pers == NULL ||
4692 mddev->pers->quiesce == NULL)
4693 goto unlock;
4694 old = mddev->suspend_lo;
4695 mddev->suspend_lo = new;
4696 if (new >= old)
4697 /* Shrinking suspended region */
4698 mddev->pers->quiesce(mddev, 2);
4699 else {
4700 /* Expanding suspended region - need to wait */
4701 mddev->pers->quiesce(mddev, 1);
4702 mddev->pers->quiesce(mddev, 0);
4703 }
4704 err = 0;
4705 unlock:
4706 mddev_unlock(mddev);
4707 return err ?: len;
4708 }
4709 static struct md_sysfs_entry md_suspend_lo =
4710 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4711
4712 static ssize_t
4713 suspend_hi_show(struct mddev *mddev, char *page)
4714 {
4715 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4716 }
4717
4718 static ssize_t
4719 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4720 {
4721 unsigned long long old, new;
4722 int err;
4723
4724 err = kstrtoull(buf, 10, &new);
4725 if (err < 0)
4726 return err;
4727 if (new != (sector_t)new)
4728 return -EINVAL;
4729
4730 err = mddev_lock(mddev);
4731 if (err)
4732 return err;
4733 err = -EINVAL;
4734 if (mddev->pers == NULL ||
4735 mddev->pers->quiesce == NULL)
4736 goto unlock;
4737 old = mddev->suspend_hi;
4738 mddev->suspend_hi = new;
4739 if (new <= old)
4740 /* Shrinking suspended region */
4741 mddev->pers->quiesce(mddev, 2);
4742 else {
4743 /* Expanding suspended region - need to wait */
4744 mddev->pers->quiesce(mddev, 1);
4745 mddev->pers->quiesce(mddev, 0);
4746 }
4747 err = 0;
4748 unlock:
4749 mddev_unlock(mddev);
4750 return err ?: len;
4751 }
4752 static struct md_sysfs_entry md_suspend_hi =
4753 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4754
4755 static ssize_t
4756 reshape_position_show(struct mddev *mddev, char *page)
4757 {
4758 if (mddev->reshape_position != MaxSector)
4759 return sprintf(page, "%llu\n",
4760 (unsigned long long)mddev->reshape_position);
4761 strcpy(page, "none\n");
4762 return 5;
4763 }
4764
4765 static ssize_t
4766 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4767 {
4768 struct md_rdev *rdev;
4769 unsigned long long new;
4770 int err;
4771
4772 err = kstrtoull(buf, 10, &new);
4773 if (err < 0)
4774 return err;
4775 if (new != (sector_t)new)
4776 return -EINVAL;
4777 err = mddev_lock(mddev);
4778 if (err)
4779 return err;
4780 err = -EBUSY;
4781 if (mddev->pers)
4782 goto unlock;
4783 mddev->reshape_position = new;
4784 mddev->delta_disks = 0;
4785 mddev->reshape_backwards = 0;
4786 mddev->new_level = mddev->level;
4787 mddev->new_layout = mddev->layout;
4788 mddev->new_chunk_sectors = mddev->chunk_sectors;
4789 rdev_for_each(rdev, mddev)
4790 rdev->new_data_offset = rdev->data_offset;
4791 err = 0;
4792 unlock:
4793 mddev_unlock(mddev);
4794 return err ?: len;
4795 }
4796
4797 static struct md_sysfs_entry md_reshape_position =
4798 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4799 reshape_position_store);
4800
4801 static ssize_t
4802 reshape_direction_show(struct mddev *mddev, char *page)
4803 {
4804 return sprintf(page, "%s\n",
4805 mddev->reshape_backwards ? "backwards" : "forwards");
4806 }
4807
4808 static ssize_t
4809 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4810 {
4811 int backwards = 0;
4812 int err;
4813
4814 if (cmd_match(buf, "forwards"))
4815 backwards = 0;
4816 else if (cmd_match(buf, "backwards"))
4817 backwards = 1;
4818 else
4819 return -EINVAL;
4820 if (mddev->reshape_backwards == backwards)
4821 return len;
4822
4823 err = mddev_lock(mddev);
4824 if (err)
4825 return err;
4826 /* check if we are allowed to change */
4827 if (mddev->delta_disks)
4828 err = -EBUSY;
4829 else if (mddev->persistent &&
4830 mddev->major_version == 0)
4831 err = -EINVAL;
4832 else
4833 mddev->reshape_backwards = backwards;
4834 mddev_unlock(mddev);
4835 return err ?: len;
4836 }
4837
4838 static struct md_sysfs_entry md_reshape_direction =
4839 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4840 reshape_direction_store);
4841
4842 static ssize_t
4843 array_size_show(struct mddev *mddev, char *page)
4844 {
4845 if (mddev->external_size)
4846 return sprintf(page, "%llu\n",
4847 (unsigned long long)mddev->array_sectors/2);
4848 else
4849 return sprintf(page, "default\n");
4850 }
4851
4852 static ssize_t
4853 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4854 {
4855 sector_t sectors;
4856 int err;
4857
4858 err = mddev_lock(mddev);
4859 if (err)
4860 return err;
4861
4862 /* cluster raid doesn't support change array_sectors */
4863 if (mddev_is_clustered(mddev))
4864 return -EINVAL;
4865
4866 if (strncmp(buf, "default", 7) == 0) {
4867 if (mddev->pers)
4868 sectors = mddev->pers->size(mddev, 0, 0);
4869 else
4870 sectors = mddev->array_sectors;
4871
4872 mddev->external_size = 0;
4873 } else {
4874 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4875 err = -EINVAL;
4876 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4877 err = -E2BIG;
4878 else
4879 mddev->external_size = 1;
4880 }
4881
4882 if (!err) {
4883 mddev->array_sectors = sectors;
4884 if (mddev->pers) {
4885 set_capacity(mddev->gendisk, mddev->array_sectors);
4886 revalidate_disk(mddev->gendisk);
4887 }
4888 }
4889 mddev_unlock(mddev);
4890 return err ?: len;
4891 }
4892
4893 static struct md_sysfs_entry md_array_size =
4894 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4895 array_size_store);
4896
4897 static struct attribute *md_default_attrs[] = {
4898 &md_level.attr,
4899 &md_layout.attr,
4900 &md_raid_disks.attr,
4901 &md_chunk_size.attr,
4902 &md_size.attr,
4903 &md_resync_start.attr,
4904 &md_metadata.attr,
4905 &md_new_device.attr,
4906 &md_safe_delay.attr,
4907 &md_array_state.attr,
4908 &md_reshape_position.attr,
4909 &md_reshape_direction.attr,
4910 &md_array_size.attr,
4911 &max_corr_read_errors.attr,
4912 NULL,
4913 };
4914
4915 static struct attribute *md_redundancy_attrs[] = {
4916 &md_scan_mode.attr,
4917 &md_last_scan_mode.attr,
4918 &md_mismatches.attr,
4919 &md_sync_min.attr,
4920 &md_sync_max.attr,
4921 &md_sync_speed.attr,
4922 &md_sync_force_parallel.attr,
4923 &md_sync_completed.attr,
4924 &md_min_sync.attr,
4925 &md_max_sync.attr,
4926 &md_suspend_lo.attr,
4927 &md_suspend_hi.attr,
4928 &md_bitmap.attr,
4929 &md_degraded.attr,
4930 NULL,
4931 };
4932 static struct attribute_group md_redundancy_group = {
4933 .name = NULL,
4934 .attrs = md_redundancy_attrs,
4935 };
4936
4937 static ssize_t
4938 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4939 {
4940 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4941 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4942 ssize_t rv;
4943
4944 if (!entry->show)
4945 return -EIO;
4946 spin_lock(&all_mddevs_lock);
4947 if (list_empty(&mddev->all_mddevs)) {
4948 spin_unlock(&all_mddevs_lock);
4949 return -EBUSY;
4950 }
4951 mddev_get(mddev);
4952 spin_unlock(&all_mddevs_lock);
4953
4954 rv = entry->show(mddev, page);
4955 mddev_put(mddev);
4956 return rv;
4957 }
4958
4959 static ssize_t
4960 md_attr_store(struct kobject *kobj, struct attribute *attr,
4961 const char *page, size_t length)
4962 {
4963 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4964 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4965 ssize_t rv;
4966
4967 if (!entry->store)
4968 return -EIO;
4969 if (!capable(CAP_SYS_ADMIN))
4970 return -EACCES;
4971 spin_lock(&all_mddevs_lock);
4972 if (list_empty(&mddev->all_mddevs)) {
4973 spin_unlock(&all_mddevs_lock);
4974 return -EBUSY;
4975 }
4976 mddev_get(mddev);
4977 spin_unlock(&all_mddevs_lock);
4978 rv = entry->store(mddev, page, length);
4979 mddev_put(mddev);
4980 return rv;
4981 }
4982
4983 static void md_free(struct kobject *ko)
4984 {
4985 struct mddev *mddev = container_of(ko, struct mddev, kobj);
4986
4987 if (mddev->sysfs_state)
4988 sysfs_put(mddev->sysfs_state);
4989
4990 if (mddev->queue)
4991 blk_cleanup_queue(mddev->queue);
4992 if (mddev->gendisk) {
4993 del_gendisk(mddev->gendisk);
4994 put_disk(mddev->gendisk);
4995 }
4996
4997 kfree(mddev);
4998 }
4999
5000 static const struct sysfs_ops md_sysfs_ops = {
5001 .show = md_attr_show,
5002 .store = md_attr_store,
5003 };
5004 static struct kobj_type md_ktype = {
5005 .release = md_free,
5006 .sysfs_ops = &md_sysfs_ops,
5007 .default_attrs = md_default_attrs,
5008 };
5009
5010 int mdp_major = 0;
5011
5012 static void mddev_delayed_delete(struct work_struct *ws)
5013 {
5014 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5015
5016 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5017 kobject_del(&mddev->kobj);
5018 kobject_put(&mddev->kobj);
5019 }
5020
5021 static int md_alloc(dev_t dev, char *name)
5022 {
5023 static DEFINE_MUTEX(disks_mutex);
5024 struct mddev *mddev = mddev_find(dev);
5025 struct gendisk *disk;
5026 int partitioned;
5027 int shift;
5028 int unit;
5029 int error;
5030
5031 if (!mddev)
5032 return -ENODEV;
5033
5034 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
5035 shift = partitioned ? MdpMinorShift : 0;
5036 unit = MINOR(mddev->unit) >> shift;
5037
5038 /* wait for any previous instance of this device to be
5039 * completely removed (mddev_delayed_delete).
5040 */
5041 flush_workqueue(md_misc_wq);
5042
5043 mutex_lock(&disks_mutex);
5044 error = -EEXIST;
5045 if (mddev->gendisk)
5046 goto abort;
5047
5048 if (name) {
5049 /* Need to ensure that 'name' is not a duplicate.
5050 */
5051 struct mddev *mddev2;
5052 spin_lock(&all_mddevs_lock);
5053
5054 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5055 if (mddev2->gendisk &&
5056 strcmp(mddev2->gendisk->disk_name, name) == 0) {
5057 spin_unlock(&all_mddevs_lock);
5058 goto abort;
5059 }
5060 spin_unlock(&all_mddevs_lock);
5061 }
5062
5063 error = -ENOMEM;
5064 mddev->queue = blk_alloc_queue(GFP_KERNEL);
5065 if (!mddev->queue)
5066 goto abort;
5067 mddev->queue->queuedata = mddev;
5068
5069 blk_queue_make_request(mddev->queue, md_make_request);
5070 blk_set_stacking_limits(&mddev->queue->limits);
5071
5072 disk = alloc_disk(1 << shift);
5073 if (!disk) {
5074 blk_cleanup_queue(mddev->queue);
5075 mddev->queue = NULL;
5076 goto abort;
5077 }
5078 disk->major = MAJOR(mddev->unit);
5079 disk->first_minor = unit << shift;
5080 if (name)
5081 strcpy(disk->disk_name, name);
5082 else if (partitioned)
5083 sprintf(disk->disk_name, "md_d%d", unit);
5084 else
5085 sprintf(disk->disk_name, "md%d", unit);
5086 disk->fops = &md_fops;
5087 disk->private_data = mddev;
5088 disk->queue = mddev->queue;
5089 blk_queue_write_cache(mddev->queue, true, true);
5090 /* Allow extended partitions. This makes the
5091 * 'mdp' device redundant, but we can't really
5092 * remove it now.
5093 */
5094 disk->flags |= GENHD_FL_EXT_DEVT;
5095 mddev->gendisk = disk;
5096 /* As soon as we call add_disk(), another thread could get
5097 * through to md_open, so make sure it doesn't get too far
5098 */
5099 mutex_lock(&mddev->open_mutex);
5100 add_disk(disk);
5101
5102 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
5103 &disk_to_dev(disk)->kobj, "%s", "md");
5104 if (error) {
5105 /* This isn't possible, but as kobject_init_and_add is marked
5106 * __must_check, we must do something with the result
5107 */
5108 pr_debug("md: cannot register %s/md - name in use\n",
5109 disk->disk_name);
5110 error = 0;
5111 }
5112 if (mddev->kobj.sd &&
5113 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
5114 pr_debug("pointless warning\n");
5115 mutex_unlock(&mddev->open_mutex);
5116 abort:
5117 mutex_unlock(&disks_mutex);
5118 if (!error && mddev->kobj.sd) {
5119 kobject_uevent(&mddev->kobj, KOBJ_ADD);
5120 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5121 }
5122 mddev_put(mddev);
5123 return error;
5124 }
5125
5126 static struct kobject *md_probe(dev_t dev, int *part, void *data)
5127 {
5128 md_alloc(dev, NULL);
5129 return NULL;
5130 }
5131
5132 static int add_named_array(const char *val, struct kernel_param *kp)
5133 {
5134 /* val must be "md_*" where * is not all digits.
5135 * We allocate an array with a large free minor number, and
5136 * set the name to val. val must not already be an active name.
5137 */
5138 int len = strlen(val);
5139 char buf[DISK_NAME_LEN];
5140
5141 while (len && val[len-1] == '\n')
5142 len--;
5143 if (len >= DISK_NAME_LEN)
5144 return -E2BIG;
5145 strlcpy(buf, val, len+1);
5146 if (strncmp(buf, "md_", 3) != 0)
5147 return -EINVAL;
5148 return md_alloc(0, buf);
5149 }
5150
5151 static void md_safemode_timeout(unsigned long data)
5152 {
5153 struct mddev *mddev = (struct mddev *) data;
5154
5155 if (!atomic_read(&mddev->writes_pending)) {
5156 mddev->safemode = 1;
5157 if (mddev->external)
5158 sysfs_notify_dirent_safe(mddev->sysfs_state);
5159 }
5160 md_wakeup_thread(mddev->thread);
5161 }
5162
5163 static int start_dirty_degraded;
5164
5165 int md_run(struct mddev *mddev)
5166 {
5167 int err;
5168 struct md_rdev *rdev;
5169 struct md_personality *pers;
5170
5171 if (list_empty(&mddev->disks))
5172 /* cannot run an array with no devices.. */
5173 return -EINVAL;
5174
5175 if (mddev->pers)
5176 return -EBUSY;
5177 /* Cannot run until previous stop completes properly */
5178 if (mddev->sysfs_active)
5179 return -EBUSY;
5180
5181 /*
5182 * Analyze all RAID superblock(s)
5183 */
5184 if (!mddev->raid_disks) {
5185 if (!mddev->persistent)
5186 return -EINVAL;
5187 analyze_sbs(mddev);
5188 }
5189
5190 if (mddev->level != LEVEL_NONE)
5191 request_module("md-level-%d", mddev->level);
5192 else if (mddev->clevel[0])
5193 request_module("md-%s", mddev->clevel);
5194
5195 /*
5196 * Drop all container device buffers, from now on
5197 * the only valid external interface is through the md
5198 * device.
5199 */
5200 rdev_for_each(rdev, mddev) {
5201 if (test_bit(Faulty, &rdev->flags))
5202 continue;
5203 sync_blockdev(rdev->bdev);
5204 invalidate_bdev(rdev->bdev);
5205
5206 /* perform some consistency tests on the device.
5207 * We don't want the data to overlap the metadata,
5208 * Internal Bitmap issues have been handled elsewhere.
5209 */
5210 if (rdev->meta_bdev) {
5211 /* Nothing to check */;
5212 } else if (rdev->data_offset < rdev->sb_start) {
5213 if (mddev->dev_sectors &&
5214 rdev->data_offset + mddev->dev_sectors
5215 > rdev->sb_start) {
5216 pr_warn("md: %s: data overlaps metadata\n",
5217 mdname(mddev));
5218 return -EINVAL;
5219 }
5220 } else {
5221 if (rdev->sb_start + rdev->sb_size/512
5222 > rdev->data_offset) {
5223 pr_warn("md: %s: metadata overlaps data\n",
5224 mdname(mddev));
5225 return -EINVAL;
5226 }
5227 }
5228 sysfs_notify_dirent_safe(rdev->sysfs_state);
5229 }
5230
5231 if (mddev->bio_set == NULL)
5232 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
5233
5234 spin_lock(&pers_lock);
5235 pers = find_pers(mddev->level, mddev->clevel);
5236 if (!pers || !try_module_get(pers->owner)) {
5237 spin_unlock(&pers_lock);
5238 if (mddev->level != LEVEL_NONE)
5239 pr_warn("md: personality for level %d is not loaded!\n",
5240 mddev->level);
5241 else
5242 pr_warn("md: personality for level %s is not loaded!\n",
5243 mddev->clevel);
5244 return -EINVAL;
5245 }
5246 spin_unlock(&pers_lock);
5247 if (mddev->level != pers->level) {
5248 mddev->level = pers->level;
5249 mddev->new_level = pers->level;
5250 }
5251 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5252
5253 if (mddev->reshape_position != MaxSector &&
5254 pers->start_reshape == NULL) {
5255 /* This personality cannot handle reshaping... */
5256 module_put(pers->owner);
5257 return -EINVAL;
5258 }
5259
5260 if (pers->sync_request) {
5261 /* Warn if this is a potentially silly
5262 * configuration.
5263 */
5264 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5265 struct md_rdev *rdev2;
5266 int warned = 0;
5267
5268 rdev_for_each(rdev, mddev)
5269 rdev_for_each(rdev2, mddev) {
5270 if (rdev < rdev2 &&
5271 rdev->bdev->bd_contains ==
5272 rdev2->bdev->bd_contains) {
5273 pr_warn("%s: WARNING: %s appears to be on the same physical disk as %s.\n",
5274 mdname(mddev),
5275 bdevname(rdev->bdev,b),
5276 bdevname(rdev2->bdev,b2));
5277 warned = 1;
5278 }
5279 }
5280
5281 if (warned)
5282 pr_warn("True protection against single-disk failure might be compromised.\n");
5283 }
5284
5285 mddev->recovery = 0;
5286 /* may be over-ridden by personality */
5287 mddev->resync_max_sectors = mddev->dev_sectors;
5288
5289 mddev->ok_start_degraded = start_dirty_degraded;
5290
5291 if (start_readonly && mddev->ro == 0)
5292 mddev->ro = 2; /* read-only, but switch on first write */
5293
5294 err = pers->run(mddev);
5295 if (err)
5296 pr_warn("md: pers->run() failed ...\n");
5297 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5298 WARN_ONCE(!mddev->external_size,
5299 "%s: default size too small, but 'external_size' not in effect?\n",
5300 __func__);
5301 pr_warn("md: invalid array_size %llu > default size %llu\n",
5302 (unsigned long long)mddev->array_sectors / 2,
5303 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5304 err = -EINVAL;
5305 }
5306 if (err == 0 && pers->sync_request &&
5307 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5308 struct bitmap *bitmap;
5309
5310 bitmap = bitmap_create(mddev, -1);
5311 if (IS_ERR(bitmap)) {
5312 err = PTR_ERR(bitmap);
5313 pr_warn("%s: failed to create bitmap (%d)\n",
5314 mdname(mddev), err);
5315 } else
5316 mddev->bitmap = bitmap;
5317
5318 }
5319 if (err) {
5320 mddev_detach(mddev);
5321 if (mddev->private)
5322 pers->free(mddev, mddev->private);
5323 mddev->private = NULL;
5324 module_put(pers->owner);
5325 bitmap_destroy(mddev);
5326 return err;
5327 }
5328 if (mddev->queue) {
5329 bool nonrot = true;
5330
5331 rdev_for_each(rdev, mddev) {
5332 if (rdev->raid_disk >= 0 &&
5333 !blk_queue_nonrot(bdev_get_queue(rdev->bdev))) {
5334 nonrot = false;
5335 break;
5336 }
5337 }
5338 if (mddev->degraded)
5339 nonrot = false;
5340 if (nonrot)
5341 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, mddev->queue);
5342 else
5343 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, mddev->queue);
5344 mddev->queue->backing_dev_info.congested_data = mddev;
5345 mddev->queue->backing_dev_info.congested_fn = md_congested;
5346 }
5347 if (pers->sync_request) {
5348 if (mddev->kobj.sd &&
5349 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5350 pr_warn("md: cannot register extra attributes for %s\n",
5351 mdname(mddev));
5352 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5353 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5354 mddev->ro = 0;
5355
5356 atomic_set(&mddev->writes_pending,0);
5357 atomic_set(&mddev->max_corr_read_errors,
5358 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5359 mddev->safemode = 0;
5360 if (mddev_is_clustered(mddev))
5361 mddev->safemode_delay = 0;
5362 else
5363 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5364 mddev->in_sync = 1;
5365 smp_wmb();
5366 spin_lock(&mddev->lock);
5367 mddev->pers = pers;
5368 spin_unlock(&mddev->lock);
5369 rdev_for_each(rdev, mddev)
5370 if (rdev->raid_disk >= 0)
5371 if (sysfs_link_rdev(mddev, rdev))
5372 /* failure here is OK */;
5373
5374 if (mddev->degraded && !mddev->ro)
5375 /* This ensures that recovering status is reported immediately
5376 * via sysfs - until a lack of spares is confirmed.
5377 */
5378 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5379 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5380
5381 if (mddev->sb_flags)
5382 md_update_sb(mddev, 0);
5383
5384 md_new_event(mddev);
5385 sysfs_notify_dirent_safe(mddev->sysfs_state);
5386 sysfs_notify_dirent_safe(mddev->sysfs_action);
5387 sysfs_notify(&mddev->kobj, NULL, "degraded");
5388 return 0;
5389 }
5390 EXPORT_SYMBOL_GPL(md_run);
5391
5392 static int do_md_run(struct mddev *mddev)
5393 {
5394 int err;
5395
5396 err = md_run(mddev);
5397 if (err)
5398 goto out;
5399 err = bitmap_load(mddev);
5400 if (err) {
5401 bitmap_destroy(mddev);
5402 goto out;
5403 }
5404
5405 if (mddev_is_clustered(mddev))
5406 md_allow_write(mddev);
5407
5408 md_wakeup_thread(mddev->thread);
5409 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5410
5411 set_capacity(mddev->gendisk, mddev->array_sectors);
5412 revalidate_disk(mddev->gendisk);
5413 mddev->changed = 1;
5414 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5415 out:
5416 return err;
5417 }
5418
5419 static int restart_array(struct mddev *mddev)
5420 {
5421 struct gendisk *disk = mddev->gendisk;
5422
5423 /* Complain if it has no devices */
5424 if (list_empty(&mddev->disks))
5425 return -ENXIO;
5426 if (!mddev->pers)
5427 return -EINVAL;
5428 if (!mddev->ro)
5429 return -EBUSY;
5430 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5431 struct md_rdev *rdev;
5432 bool has_journal = false;
5433
5434 rcu_read_lock();
5435 rdev_for_each_rcu(rdev, mddev) {
5436 if (test_bit(Journal, &rdev->flags) &&
5437 !test_bit(Faulty, &rdev->flags)) {
5438 has_journal = true;
5439 break;
5440 }
5441 }
5442 rcu_read_unlock();
5443
5444 /* Don't restart rw with journal missing/faulty */
5445 if (!has_journal)
5446 return -EINVAL;
5447 }
5448
5449 mddev->safemode = 0;
5450 mddev->ro = 0;
5451 set_disk_ro(disk, 0);
5452 pr_debug("md: %s switched to read-write mode.\n", mdname(mddev));
5453 /* Kick recovery or resync if necessary */
5454 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5455 md_wakeup_thread(mddev->thread);
5456 md_wakeup_thread(mddev->sync_thread);
5457 sysfs_notify_dirent_safe(mddev->sysfs_state);
5458 return 0;
5459 }
5460
5461 static void md_clean(struct mddev *mddev)
5462 {
5463 mddev->array_sectors = 0;
5464 mddev->external_size = 0;
5465 mddev->dev_sectors = 0;
5466 mddev->raid_disks = 0;
5467 mddev->recovery_cp = 0;
5468 mddev->resync_min = 0;
5469 mddev->resync_max = MaxSector;
5470 mddev->reshape_position = MaxSector;
5471 mddev->external = 0;
5472 mddev->persistent = 0;
5473 mddev->level = LEVEL_NONE;
5474 mddev->clevel[0] = 0;
5475 mddev->flags = 0;
5476 mddev->sb_flags = 0;
5477 mddev->ro = 0;
5478 mddev->metadata_type[0] = 0;
5479 mddev->chunk_sectors = 0;
5480 mddev->ctime = mddev->utime = 0;
5481 mddev->layout = 0;
5482 mddev->max_disks = 0;
5483 mddev->events = 0;
5484 mddev->can_decrease_events = 0;
5485 mddev->delta_disks = 0;
5486 mddev->reshape_backwards = 0;
5487 mddev->new_level = LEVEL_NONE;
5488 mddev->new_layout = 0;
5489 mddev->new_chunk_sectors = 0;
5490 mddev->curr_resync = 0;
5491 atomic64_set(&mddev->resync_mismatches, 0);
5492 mddev->suspend_lo = mddev->suspend_hi = 0;
5493 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5494 mddev->recovery = 0;
5495 mddev->in_sync = 0;
5496 mddev->changed = 0;
5497 mddev->degraded = 0;
5498 mddev->safemode = 0;
5499 mddev->private = NULL;
5500 mddev->cluster_info = NULL;
5501 mddev->bitmap_info.offset = 0;
5502 mddev->bitmap_info.default_offset = 0;
5503 mddev->bitmap_info.default_space = 0;
5504 mddev->bitmap_info.chunksize = 0;
5505 mddev->bitmap_info.daemon_sleep = 0;
5506 mddev->bitmap_info.max_write_behind = 0;
5507 mddev->bitmap_info.nodes = 0;
5508 }
5509
5510 static void __md_stop_writes(struct mddev *mddev)
5511 {
5512 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5513 flush_workqueue(md_misc_wq);
5514 if (mddev->sync_thread) {
5515 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5516 md_reap_sync_thread(mddev);
5517 }
5518
5519 del_timer_sync(&mddev->safemode_timer);
5520
5521 if (mddev->pers && mddev->pers->quiesce) {
5522 mddev->pers->quiesce(mddev, 1);
5523 mddev->pers->quiesce(mddev, 0);
5524 }
5525 bitmap_flush(mddev);
5526
5527 if (mddev->ro == 0 &&
5528 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
5529 mddev->sb_flags)) {
5530 /* mark array as shutdown cleanly */
5531 if (!mddev_is_clustered(mddev))
5532 mddev->in_sync = 1;
5533 md_update_sb(mddev, 1);
5534 }
5535 }
5536
5537 void md_stop_writes(struct mddev *mddev)
5538 {
5539 mddev_lock_nointr(mddev);
5540 __md_stop_writes(mddev);
5541 mddev_unlock(mddev);
5542 }
5543 EXPORT_SYMBOL_GPL(md_stop_writes);
5544
5545 static void mddev_detach(struct mddev *mddev)
5546 {
5547 struct bitmap *bitmap = mddev->bitmap;
5548 /* wait for behind writes to complete */
5549 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
5550 pr_debug("md:%s: behind writes in progress - waiting to stop.\n",
5551 mdname(mddev));
5552 /* need to kick something here to make sure I/O goes? */
5553 wait_event(bitmap->behind_wait,
5554 atomic_read(&bitmap->behind_writes) == 0);
5555 }
5556 if (mddev->pers && mddev->pers->quiesce) {
5557 mddev->pers->quiesce(mddev, 1);
5558 mddev->pers->quiesce(mddev, 0);
5559 }
5560 md_unregister_thread(&mddev->thread);
5561 if (mddev->queue)
5562 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5563 }
5564
5565 static void __md_stop(struct mddev *mddev)
5566 {
5567 struct md_personality *pers = mddev->pers;
5568 mddev_detach(mddev);
5569 /* Ensure ->event_work is done */
5570 flush_workqueue(md_misc_wq);
5571 spin_lock(&mddev->lock);
5572 mddev->pers = NULL;
5573 spin_unlock(&mddev->lock);
5574 pers->free(mddev, mddev->private);
5575 mddev->private = NULL;
5576 if (pers->sync_request && mddev->to_remove == NULL)
5577 mddev->to_remove = &md_redundancy_group;
5578 module_put(pers->owner);
5579 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5580 }
5581
5582 void md_stop(struct mddev *mddev)
5583 {
5584 /* stop the array and free an attached data structures.
5585 * This is called from dm-raid
5586 */
5587 __md_stop(mddev);
5588 bitmap_destroy(mddev);
5589 if (mddev->bio_set)
5590 bioset_free(mddev->bio_set);
5591 }
5592
5593 EXPORT_SYMBOL_GPL(md_stop);
5594
5595 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5596 {
5597 int err = 0;
5598 int did_freeze = 0;
5599
5600 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5601 did_freeze = 1;
5602 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5603 md_wakeup_thread(mddev->thread);
5604 }
5605 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5606 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5607 if (mddev->sync_thread)
5608 /* Thread might be blocked waiting for metadata update
5609 * which will now never happen */
5610 wake_up_process(mddev->sync_thread->tsk);
5611
5612 if (mddev->external && test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
5613 return -EBUSY;
5614 mddev_unlock(mddev);
5615 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5616 &mddev->recovery));
5617 wait_event(mddev->sb_wait,
5618 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
5619 mddev_lock_nointr(mddev);
5620
5621 mutex_lock(&mddev->open_mutex);
5622 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5623 mddev->sync_thread ||
5624 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
5625 pr_warn("md: %s still in use.\n",mdname(mddev));
5626 if (did_freeze) {
5627 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5628 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5629 md_wakeup_thread(mddev->thread);
5630 }
5631 err = -EBUSY;
5632 goto out;
5633 }
5634 if (mddev->pers) {
5635 __md_stop_writes(mddev);
5636
5637 err = -ENXIO;
5638 if (mddev->ro==1)
5639 goto out;
5640 mddev->ro = 1;
5641 set_disk_ro(mddev->gendisk, 1);
5642 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5643 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5644 md_wakeup_thread(mddev->thread);
5645 sysfs_notify_dirent_safe(mddev->sysfs_state);
5646 err = 0;
5647 }
5648 out:
5649 mutex_unlock(&mddev->open_mutex);
5650 return err;
5651 }
5652
5653 /* mode:
5654 * 0 - completely stop and dis-assemble array
5655 * 2 - stop but do not disassemble array
5656 */
5657 static int do_md_stop(struct mddev *mddev, int mode,
5658 struct block_device *bdev)
5659 {
5660 struct gendisk *disk = mddev->gendisk;
5661 struct md_rdev *rdev;
5662 int did_freeze = 0;
5663
5664 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5665 did_freeze = 1;
5666 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5667 md_wakeup_thread(mddev->thread);
5668 }
5669 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5670 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5671 if (mddev->sync_thread)
5672 /* Thread might be blocked waiting for metadata update
5673 * which will now never happen */
5674 wake_up_process(mddev->sync_thread->tsk);
5675
5676 mddev_unlock(mddev);
5677 wait_event(resync_wait, (mddev->sync_thread == NULL &&
5678 !test_bit(MD_RECOVERY_RUNNING,
5679 &mddev->recovery)));
5680 mddev_lock_nointr(mddev);
5681
5682 mutex_lock(&mddev->open_mutex);
5683 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5684 mddev->sysfs_active ||
5685 mddev->sync_thread ||
5686 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
5687 pr_warn("md: %s still in use.\n",mdname(mddev));
5688 mutex_unlock(&mddev->open_mutex);
5689 if (did_freeze) {
5690 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5691 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5692 md_wakeup_thread(mddev->thread);
5693 }
5694 return -EBUSY;
5695 }
5696 if (mddev->pers) {
5697 if (mddev->ro)
5698 set_disk_ro(disk, 0);
5699
5700 __md_stop_writes(mddev);
5701 __md_stop(mddev);
5702 mddev->queue->backing_dev_info.congested_fn = NULL;
5703
5704 /* tell userspace to handle 'inactive' */
5705 sysfs_notify_dirent_safe(mddev->sysfs_state);
5706
5707 rdev_for_each(rdev, mddev)
5708 if (rdev->raid_disk >= 0)
5709 sysfs_unlink_rdev(mddev, rdev);
5710
5711 set_capacity(disk, 0);
5712 mutex_unlock(&mddev->open_mutex);
5713 mddev->changed = 1;
5714 revalidate_disk(disk);
5715
5716 if (mddev->ro)
5717 mddev->ro = 0;
5718 } else
5719 mutex_unlock(&mddev->open_mutex);
5720 /*
5721 * Free resources if final stop
5722 */
5723 if (mode == 0) {
5724 pr_info("md: %s stopped.\n", mdname(mddev));
5725
5726 bitmap_destroy(mddev);
5727 if (mddev->bitmap_info.file) {
5728 struct file *f = mddev->bitmap_info.file;
5729 spin_lock(&mddev->lock);
5730 mddev->bitmap_info.file = NULL;
5731 spin_unlock(&mddev->lock);
5732 fput(f);
5733 }
5734 mddev->bitmap_info.offset = 0;
5735
5736 export_array(mddev);
5737
5738 md_clean(mddev);
5739 if (mddev->hold_active == UNTIL_STOP)
5740 mddev->hold_active = 0;
5741 }
5742 md_new_event(mddev);
5743 sysfs_notify_dirent_safe(mddev->sysfs_state);
5744 return 0;
5745 }
5746
5747 #ifndef MODULE
5748 static void autorun_array(struct mddev *mddev)
5749 {
5750 struct md_rdev *rdev;
5751 int err;
5752
5753 if (list_empty(&mddev->disks))
5754 return;
5755
5756 pr_info("md: running: ");
5757
5758 rdev_for_each(rdev, mddev) {
5759 char b[BDEVNAME_SIZE];
5760 pr_cont("<%s>", bdevname(rdev->bdev,b));
5761 }
5762 pr_cont("\n");
5763
5764 err = do_md_run(mddev);
5765 if (err) {
5766 pr_warn("md: do_md_run() returned %d\n", err);
5767 do_md_stop(mddev, 0, NULL);
5768 }
5769 }
5770
5771 /*
5772 * lets try to run arrays based on all disks that have arrived
5773 * until now. (those are in pending_raid_disks)
5774 *
5775 * the method: pick the first pending disk, collect all disks with
5776 * the same UUID, remove all from the pending list and put them into
5777 * the 'same_array' list. Then order this list based on superblock
5778 * update time (freshest comes first), kick out 'old' disks and
5779 * compare superblocks. If everything's fine then run it.
5780 *
5781 * If "unit" is allocated, then bump its reference count
5782 */
5783 static void autorun_devices(int part)
5784 {
5785 struct md_rdev *rdev0, *rdev, *tmp;
5786 struct mddev *mddev;
5787 char b[BDEVNAME_SIZE];
5788
5789 pr_info("md: autorun ...\n");
5790 while (!list_empty(&pending_raid_disks)) {
5791 int unit;
5792 dev_t dev;
5793 LIST_HEAD(candidates);
5794 rdev0 = list_entry(pending_raid_disks.next,
5795 struct md_rdev, same_set);
5796
5797 pr_debug("md: considering %s ...\n", bdevname(rdev0->bdev,b));
5798 INIT_LIST_HEAD(&candidates);
5799 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5800 if (super_90_load(rdev, rdev0, 0) >= 0) {
5801 pr_debug("md: adding %s ...\n",
5802 bdevname(rdev->bdev,b));
5803 list_move(&rdev->same_set, &candidates);
5804 }
5805 /*
5806 * now we have a set of devices, with all of them having
5807 * mostly sane superblocks. It's time to allocate the
5808 * mddev.
5809 */
5810 if (part) {
5811 dev = MKDEV(mdp_major,
5812 rdev0->preferred_minor << MdpMinorShift);
5813 unit = MINOR(dev) >> MdpMinorShift;
5814 } else {
5815 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5816 unit = MINOR(dev);
5817 }
5818 if (rdev0->preferred_minor != unit) {
5819 pr_warn("md: unit number in %s is bad: %d\n",
5820 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5821 break;
5822 }
5823
5824 md_probe(dev, NULL, NULL);
5825 mddev = mddev_find(dev);
5826 if (!mddev || !mddev->gendisk) {
5827 if (mddev)
5828 mddev_put(mddev);
5829 break;
5830 }
5831 if (mddev_lock(mddev))
5832 pr_warn("md: %s locked, cannot run\n", mdname(mddev));
5833 else if (mddev->raid_disks || mddev->major_version
5834 || !list_empty(&mddev->disks)) {
5835 pr_warn("md: %s already running, cannot run %s\n",
5836 mdname(mddev), bdevname(rdev0->bdev,b));
5837 mddev_unlock(mddev);
5838 } else {
5839 pr_debug("md: created %s\n", mdname(mddev));
5840 mddev->persistent = 1;
5841 rdev_for_each_list(rdev, tmp, &candidates) {
5842 list_del_init(&rdev->same_set);
5843 if (bind_rdev_to_array(rdev, mddev))
5844 export_rdev(rdev);
5845 }
5846 autorun_array(mddev);
5847 mddev_unlock(mddev);
5848 }
5849 /* on success, candidates will be empty, on error
5850 * it won't...
5851 */
5852 rdev_for_each_list(rdev, tmp, &candidates) {
5853 list_del_init(&rdev->same_set);
5854 export_rdev(rdev);
5855 }
5856 mddev_put(mddev);
5857 }
5858 pr_info("md: ... autorun DONE.\n");
5859 }
5860 #endif /* !MODULE */
5861
5862 static int get_version(void __user *arg)
5863 {
5864 mdu_version_t ver;
5865
5866 ver.major = MD_MAJOR_VERSION;
5867 ver.minor = MD_MINOR_VERSION;
5868 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5869
5870 if (copy_to_user(arg, &ver, sizeof(ver)))
5871 return -EFAULT;
5872
5873 return 0;
5874 }
5875
5876 static int get_array_info(struct mddev *mddev, void __user *arg)
5877 {
5878 mdu_array_info_t info;
5879 int nr,working,insync,failed,spare;
5880 struct md_rdev *rdev;
5881
5882 nr = working = insync = failed = spare = 0;
5883 rcu_read_lock();
5884 rdev_for_each_rcu(rdev, mddev) {
5885 nr++;
5886 if (test_bit(Faulty, &rdev->flags))
5887 failed++;
5888 else {
5889 working++;
5890 if (test_bit(In_sync, &rdev->flags))
5891 insync++;
5892 else if (test_bit(Journal, &rdev->flags))
5893 /* TODO: add journal count to md_u.h */
5894 ;
5895 else
5896 spare++;
5897 }
5898 }
5899 rcu_read_unlock();
5900
5901 info.major_version = mddev->major_version;
5902 info.minor_version = mddev->minor_version;
5903 info.patch_version = MD_PATCHLEVEL_VERSION;
5904 info.ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
5905 info.level = mddev->level;
5906 info.size = mddev->dev_sectors / 2;
5907 if (info.size != mddev->dev_sectors / 2) /* overflow */
5908 info.size = -1;
5909 info.nr_disks = nr;
5910 info.raid_disks = mddev->raid_disks;
5911 info.md_minor = mddev->md_minor;
5912 info.not_persistent= !mddev->persistent;
5913
5914 info.utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
5915 info.state = 0;
5916 if (mddev->in_sync)
5917 info.state = (1<<MD_SB_CLEAN);
5918 if (mddev->bitmap && mddev->bitmap_info.offset)
5919 info.state |= (1<<MD_SB_BITMAP_PRESENT);
5920 if (mddev_is_clustered(mddev))
5921 info.state |= (1<<MD_SB_CLUSTERED);
5922 info.active_disks = insync;
5923 info.working_disks = working;
5924 info.failed_disks = failed;
5925 info.spare_disks = spare;
5926
5927 info.layout = mddev->layout;
5928 info.chunk_size = mddev->chunk_sectors << 9;
5929
5930 if (copy_to_user(arg, &info, sizeof(info)))
5931 return -EFAULT;
5932
5933 return 0;
5934 }
5935
5936 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
5937 {
5938 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5939 char *ptr;
5940 int err;
5941
5942 file = kzalloc(sizeof(*file), GFP_NOIO);
5943 if (!file)
5944 return -ENOMEM;
5945
5946 err = 0;
5947 spin_lock(&mddev->lock);
5948 /* bitmap enabled */
5949 if (mddev->bitmap_info.file) {
5950 ptr = file_path(mddev->bitmap_info.file, file->pathname,
5951 sizeof(file->pathname));
5952 if (IS_ERR(ptr))
5953 err = PTR_ERR(ptr);
5954 else
5955 memmove(file->pathname, ptr,
5956 sizeof(file->pathname)-(ptr-file->pathname));
5957 }
5958 spin_unlock(&mddev->lock);
5959
5960 if (err == 0 &&
5961 copy_to_user(arg, file, sizeof(*file)))
5962 err = -EFAULT;
5963
5964 kfree(file);
5965 return err;
5966 }
5967
5968 static int get_disk_info(struct mddev *mddev, void __user * arg)
5969 {
5970 mdu_disk_info_t info;
5971 struct md_rdev *rdev;
5972
5973 if (copy_from_user(&info, arg, sizeof(info)))
5974 return -EFAULT;
5975
5976 rcu_read_lock();
5977 rdev = md_find_rdev_nr_rcu(mddev, info.number);
5978 if (rdev) {
5979 info.major = MAJOR(rdev->bdev->bd_dev);
5980 info.minor = MINOR(rdev->bdev->bd_dev);
5981 info.raid_disk = rdev->raid_disk;
5982 info.state = 0;
5983 if (test_bit(Faulty, &rdev->flags))
5984 info.state |= (1<<MD_DISK_FAULTY);
5985 else if (test_bit(In_sync, &rdev->flags)) {
5986 info.state |= (1<<MD_DISK_ACTIVE);
5987 info.state |= (1<<MD_DISK_SYNC);
5988 }
5989 if (test_bit(Journal, &rdev->flags))
5990 info.state |= (1<<MD_DISK_JOURNAL);
5991 if (test_bit(WriteMostly, &rdev->flags))
5992 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5993 if (test_bit(FailFast, &rdev->flags))
5994 info.state |= (1<<MD_DISK_FAILFAST);
5995 } else {
5996 info.major = info.minor = 0;
5997 info.raid_disk = -1;
5998 info.state = (1<<MD_DISK_REMOVED);
5999 }
6000 rcu_read_unlock();
6001
6002 if (copy_to_user(arg, &info, sizeof(info)))
6003 return -EFAULT;
6004
6005 return 0;
6006 }
6007
6008 static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
6009 {
6010 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
6011 struct md_rdev *rdev;
6012 dev_t dev = MKDEV(info->major,info->minor);
6013
6014 if (mddev_is_clustered(mddev) &&
6015 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
6016 pr_warn("%s: Cannot add to clustered mddev.\n",
6017 mdname(mddev));
6018 return -EINVAL;
6019 }
6020
6021 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
6022 return -EOVERFLOW;
6023
6024 if (!mddev->raid_disks) {
6025 int err;
6026 /* expecting a device which has a superblock */
6027 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
6028 if (IS_ERR(rdev)) {
6029 pr_warn("md: md_import_device returned %ld\n",
6030 PTR_ERR(rdev));
6031 return PTR_ERR(rdev);
6032 }
6033 if (!list_empty(&mddev->disks)) {
6034 struct md_rdev *rdev0
6035 = list_entry(mddev->disks.next,
6036 struct md_rdev, same_set);
6037 err = super_types[mddev->major_version]
6038 .load_super(rdev, rdev0, mddev->minor_version);
6039 if (err < 0) {
6040 pr_warn("md: %s has different UUID to %s\n",
6041 bdevname(rdev->bdev,b),
6042 bdevname(rdev0->bdev,b2));
6043 export_rdev(rdev);
6044 return -EINVAL;
6045 }
6046 }
6047 err = bind_rdev_to_array(rdev, mddev);
6048 if (err)
6049 export_rdev(rdev);
6050 return err;
6051 }
6052
6053 /*
6054 * add_new_disk can be used once the array is assembled
6055 * to add "hot spares". They must already have a superblock
6056 * written
6057 */
6058 if (mddev->pers) {
6059 int err;
6060 if (!mddev->pers->hot_add_disk) {
6061 pr_warn("%s: personality does not support diskops!\n",
6062 mdname(mddev));
6063 return -EINVAL;
6064 }
6065 if (mddev->persistent)
6066 rdev = md_import_device(dev, mddev->major_version,
6067 mddev->minor_version);
6068 else
6069 rdev = md_import_device(dev, -1, -1);
6070 if (IS_ERR(rdev)) {
6071 pr_warn("md: md_import_device returned %ld\n",
6072 PTR_ERR(rdev));
6073 return PTR_ERR(rdev);
6074 }
6075 /* set saved_raid_disk if appropriate */
6076 if (!mddev->persistent) {
6077 if (info->state & (1<<MD_DISK_SYNC) &&
6078 info->raid_disk < mddev->raid_disks) {
6079 rdev->raid_disk = info->raid_disk;
6080 set_bit(In_sync, &rdev->flags);
6081 clear_bit(Bitmap_sync, &rdev->flags);
6082 } else
6083 rdev->raid_disk = -1;
6084 rdev->saved_raid_disk = rdev->raid_disk;
6085 } else
6086 super_types[mddev->major_version].
6087 validate_super(mddev, rdev);
6088 if ((info->state & (1<<MD_DISK_SYNC)) &&
6089 rdev->raid_disk != info->raid_disk) {
6090 /* This was a hot-add request, but events doesn't
6091 * match, so reject it.
6092 */
6093 export_rdev(rdev);
6094 return -EINVAL;
6095 }
6096
6097 clear_bit(In_sync, &rdev->flags); /* just to be sure */
6098 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6099 set_bit(WriteMostly, &rdev->flags);
6100 else
6101 clear_bit(WriteMostly, &rdev->flags);
6102 if (info->state & (1<<MD_DISK_FAILFAST))
6103 set_bit(FailFast, &rdev->flags);
6104 else
6105 clear_bit(FailFast, &rdev->flags);
6106
6107 if (info->state & (1<<MD_DISK_JOURNAL)) {
6108 struct md_rdev *rdev2;
6109 bool has_journal = false;
6110
6111 /* make sure no existing journal disk */
6112 rdev_for_each(rdev2, mddev) {
6113 if (test_bit(Journal, &rdev2->flags)) {
6114 has_journal = true;
6115 break;
6116 }
6117 }
6118 if (has_journal) {
6119 export_rdev(rdev);
6120 return -EBUSY;
6121 }
6122 set_bit(Journal, &rdev->flags);
6123 }
6124 /*
6125 * check whether the device shows up in other nodes
6126 */
6127 if (mddev_is_clustered(mddev)) {
6128 if (info->state & (1 << MD_DISK_CANDIDATE))
6129 set_bit(Candidate, &rdev->flags);
6130 else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6131 /* --add initiated by this node */
6132 err = md_cluster_ops->add_new_disk(mddev, rdev);
6133 if (err) {
6134 export_rdev(rdev);
6135 return err;
6136 }
6137 }
6138 }
6139
6140 rdev->raid_disk = -1;
6141 err = bind_rdev_to_array(rdev, mddev);
6142
6143 if (err)
6144 export_rdev(rdev);
6145
6146 if (mddev_is_clustered(mddev)) {
6147 if (info->state & (1 << MD_DISK_CANDIDATE)) {
6148 if (!err) {
6149 err = md_cluster_ops->new_disk_ack(mddev,
6150 err == 0);
6151 if (err)
6152 md_kick_rdev_from_array(rdev);
6153 }
6154 } else {
6155 if (err)
6156 md_cluster_ops->add_new_disk_cancel(mddev);
6157 else
6158 err = add_bound_rdev(rdev);
6159 }
6160
6161 } else if (!err)
6162 err = add_bound_rdev(rdev);
6163
6164 return err;
6165 }
6166
6167 /* otherwise, add_new_disk is only allowed
6168 * for major_version==0 superblocks
6169 */
6170 if (mddev->major_version != 0) {
6171 pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev));
6172 return -EINVAL;
6173 }
6174
6175 if (!(info->state & (1<<MD_DISK_FAULTY))) {
6176 int err;
6177 rdev = md_import_device(dev, -1, 0);
6178 if (IS_ERR(rdev)) {
6179 pr_warn("md: error, md_import_device() returned %ld\n",
6180 PTR_ERR(rdev));
6181 return PTR_ERR(rdev);
6182 }
6183 rdev->desc_nr = info->number;
6184 if (info->raid_disk < mddev->raid_disks)
6185 rdev->raid_disk = info->raid_disk;
6186 else
6187 rdev->raid_disk = -1;
6188
6189 if (rdev->raid_disk < mddev->raid_disks)
6190 if (info->state & (1<<MD_DISK_SYNC))
6191 set_bit(In_sync, &rdev->flags);
6192
6193 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6194 set_bit(WriteMostly, &rdev->flags);
6195 if (info->state & (1<<MD_DISK_FAILFAST))
6196 set_bit(FailFast, &rdev->flags);
6197
6198 if (!mddev->persistent) {
6199 pr_debug("md: nonpersistent superblock ...\n");
6200 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6201 } else
6202 rdev->sb_start = calc_dev_sboffset(rdev);
6203 rdev->sectors = rdev->sb_start;
6204
6205 err = bind_rdev_to_array(rdev, mddev);
6206 if (err) {
6207 export_rdev(rdev);
6208 return err;
6209 }
6210 }
6211
6212 return 0;
6213 }
6214
6215 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6216 {
6217 char b[BDEVNAME_SIZE];
6218 struct md_rdev *rdev;
6219
6220 rdev = find_rdev(mddev, dev);
6221 if (!rdev)
6222 return -ENXIO;
6223
6224 if (rdev->raid_disk < 0)
6225 goto kick_rdev;
6226
6227 clear_bit(Blocked, &rdev->flags);
6228 remove_and_add_spares(mddev, rdev);
6229
6230 if (rdev->raid_disk >= 0)
6231 goto busy;
6232
6233 kick_rdev:
6234 if (mddev_is_clustered(mddev))
6235 md_cluster_ops->remove_disk(mddev, rdev);
6236
6237 md_kick_rdev_from_array(rdev);
6238 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6239 if (mddev->thread)
6240 md_wakeup_thread(mddev->thread);
6241 else
6242 md_update_sb(mddev, 1);
6243 md_new_event(mddev);
6244
6245 return 0;
6246 busy:
6247 pr_debug("md: cannot remove active disk %s from %s ...\n",
6248 bdevname(rdev->bdev,b), mdname(mddev));
6249 return -EBUSY;
6250 }
6251
6252 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6253 {
6254 char b[BDEVNAME_SIZE];
6255 int err;
6256 struct md_rdev *rdev;
6257
6258 if (!mddev->pers)
6259 return -ENODEV;
6260
6261 if (mddev->major_version != 0) {
6262 pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n",
6263 mdname(mddev));
6264 return -EINVAL;
6265 }
6266 if (!mddev->pers->hot_add_disk) {
6267 pr_warn("%s: personality does not support diskops!\n",
6268 mdname(mddev));
6269 return -EINVAL;
6270 }
6271
6272 rdev = md_import_device(dev, -1, 0);
6273 if (IS_ERR(rdev)) {
6274 pr_warn("md: error, md_import_device() returned %ld\n",
6275 PTR_ERR(rdev));
6276 return -EINVAL;
6277 }
6278
6279 if (mddev->persistent)
6280 rdev->sb_start = calc_dev_sboffset(rdev);
6281 else
6282 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6283
6284 rdev->sectors = rdev->sb_start;
6285
6286 if (test_bit(Faulty, &rdev->flags)) {
6287 pr_warn("md: can not hot-add faulty %s disk to %s!\n",
6288 bdevname(rdev->bdev,b), mdname(mddev));
6289 err = -EINVAL;
6290 goto abort_export;
6291 }
6292
6293 clear_bit(In_sync, &rdev->flags);
6294 rdev->desc_nr = -1;
6295 rdev->saved_raid_disk = -1;
6296 err = bind_rdev_to_array(rdev, mddev);
6297 if (err)
6298 goto abort_export;
6299
6300 /*
6301 * The rest should better be atomic, we can have disk failures
6302 * noticed in interrupt contexts ...
6303 */
6304
6305 rdev->raid_disk = -1;
6306
6307 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6308 if (!mddev->thread)
6309 md_update_sb(mddev, 1);
6310 /*
6311 * Kick recovery, maybe this spare has to be added to the
6312 * array immediately.
6313 */
6314 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6315 md_wakeup_thread(mddev->thread);
6316 md_new_event(mddev);
6317 return 0;
6318
6319 abort_export:
6320 export_rdev(rdev);
6321 return err;
6322 }
6323
6324 static int set_bitmap_file(struct mddev *mddev, int fd)
6325 {
6326 int err = 0;
6327
6328 if (mddev->pers) {
6329 if (!mddev->pers->quiesce || !mddev->thread)
6330 return -EBUSY;
6331 if (mddev->recovery || mddev->sync_thread)
6332 return -EBUSY;
6333 /* we should be able to change the bitmap.. */
6334 }
6335
6336 if (fd >= 0) {
6337 struct inode *inode;
6338 struct file *f;
6339
6340 if (mddev->bitmap || mddev->bitmap_info.file)
6341 return -EEXIST; /* cannot add when bitmap is present */
6342 f = fget(fd);
6343
6344 if (f == NULL) {
6345 pr_warn("%s: error: failed to get bitmap file\n",
6346 mdname(mddev));
6347 return -EBADF;
6348 }
6349
6350 inode = f->f_mapping->host;
6351 if (!S_ISREG(inode->i_mode)) {
6352 pr_warn("%s: error: bitmap file must be a regular file\n",
6353 mdname(mddev));
6354 err = -EBADF;
6355 } else if (!(f->f_mode & FMODE_WRITE)) {
6356 pr_warn("%s: error: bitmap file must open for write\n",
6357 mdname(mddev));
6358 err = -EBADF;
6359 } else if (atomic_read(&inode->i_writecount) != 1) {
6360 pr_warn("%s: error: bitmap file is already in use\n",
6361 mdname(mddev));
6362 err = -EBUSY;
6363 }
6364 if (err) {
6365 fput(f);
6366 return err;
6367 }
6368 mddev->bitmap_info.file = f;
6369 mddev->bitmap_info.offset = 0; /* file overrides offset */
6370 } else if (mddev->bitmap == NULL)
6371 return -ENOENT; /* cannot remove what isn't there */
6372 err = 0;
6373 if (mddev->pers) {
6374 mddev->pers->quiesce(mddev, 1);
6375 if (fd >= 0) {
6376 struct bitmap *bitmap;
6377
6378 bitmap = bitmap_create(mddev, -1);
6379 if (!IS_ERR(bitmap)) {
6380 mddev->bitmap = bitmap;
6381 err = bitmap_load(mddev);
6382 } else
6383 err = PTR_ERR(bitmap);
6384 }
6385 if (fd < 0 || err) {
6386 bitmap_destroy(mddev);
6387 fd = -1; /* make sure to put the file */
6388 }
6389 mddev->pers->quiesce(mddev, 0);
6390 }
6391 if (fd < 0) {
6392 struct file *f = mddev->bitmap_info.file;
6393 if (f) {
6394 spin_lock(&mddev->lock);
6395 mddev->bitmap_info.file = NULL;
6396 spin_unlock(&mddev->lock);
6397 fput(f);
6398 }
6399 }
6400
6401 return err;
6402 }
6403
6404 /*
6405 * set_array_info is used two different ways
6406 * The original usage is when creating a new array.
6407 * In this usage, raid_disks is > 0 and it together with
6408 * level, size, not_persistent,layout,chunksize determine the
6409 * shape of the array.
6410 * This will always create an array with a type-0.90.0 superblock.
6411 * The newer usage is when assembling an array.
6412 * In this case raid_disks will be 0, and the major_version field is
6413 * use to determine which style super-blocks are to be found on the devices.
6414 * The minor and patch _version numbers are also kept incase the
6415 * super_block handler wishes to interpret them.
6416 */
6417 static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
6418 {
6419
6420 if (info->raid_disks == 0) {
6421 /* just setting version number for superblock loading */
6422 if (info->major_version < 0 ||
6423 info->major_version >= ARRAY_SIZE(super_types) ||
6424 super_types[info->major_version].name == NULL) {
6425 /* maybe try to auto-load a module? */
6426 pr_warn("md: superblock version %d not known\n",
6427 info->major_version);
6428 return -EINVAL;
6429 }
6430 mddev->major_version = info->major_version;
6431 mddev->minor_version = info->minor_version;
6432 mddev->patch_version = info->patch_version;
6433 mddev->persistent = !info->not_persistent;
6434 /* ensure mddev_put doesn't delete this now that there
6435 * is some minimal configuration.
6436 */
6437 mddev->ctime = ktime_get_real_seconds();
6438 return 0;
6439 }
6440 mddev->major_version = MD_MAJOR_VERSION;
6441 mddev->minor_version = MD_MINOR_VERSION;
6442 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6443 mddev->ctime = ktime_get_real_seconds();
6444
6445 mddev->level = info->level;
6446 mddev->clevel[0] = 0;
6447 mddev->dev_sectors = 2 * (sector_t)info->size;
6448 mddev->raid_disks = info->raid_disks;
6449 /* don't set md_minor, it is determined by which /dev/md* was
6450 * openned
6451 */
6452 if (info->state & (1<<MD_SB_CLEAN))
6453 mddev->recovery_cp = MaxSector;
6454 else
6455 mddev->recovery_cp = 0;
6456 mddev->persistent = ! info->not_persistent;
6457 mddev->external = 0;
6458
6459 mddev->layout = info->layout;
6460 mddev->chunk_sectors = info->chunk_size >> 9;
6461
6462 mddev->max_disks = MD_SB_DISKS;
6463
6464 if (mddev->persistent) {
6465 mddev->flags = 0;
6466 mddev->sb_flags = 0;
6467 }
6468 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6469
6470 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6471 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6472 mddev->bitmap_info.offset = 0;
6473
6474 mddev->reshape_position = MaxSector;
6475
6476 /*
6477 * Generate a 128 bit UUID
6478 */
6479 get_random_bytes(mddev->uuid, 16);
6480
6481 mddev->new_level = mddev->level;
6482 mddev->new_chunk_sectors = mddev->chunk_sectors;
6483 mddev->new_layout = mddev->layout;
6484 mddev->delta_disks = 0;
6485 mddev->reshape_backwards = 0;
6486
6487 return 0;
6488 }
6489
6490 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6491 {
6492 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6493
6494 if (mddev->external_size)
6495 return;
6496
6497 mddev->array_sectors = array_sectors;
6498 }
6499 EXPORT_SYMBOL(md_set_array_sectors);
6500
6501 static int update_size(struct mddev *mddev, sector_t num_sectors)
6502 {
6503 struct md_rdev *rdev;
6504 int rv;
6505 int fit = (num_sectors == 0);
6506
6507 /* cluster raid doesn't support update size */
6508 if (mddev_is_clustered(mddev))
6509 return -EINVAL;
6510
6511 if (mddev->pers->resize == NULL)
6512 return -EINVAL;
6513 /* The "num_sectors" is the number of sectors of each device that
6514 * is used. This can only make sense for arrays with redundancy.
6515 * linear and raid0 always use whatever space is available. We can only
6516 * consider changing this number if no resync or reconstruction is
6517 * happening, and if the new size is acceptable. It must fit before the
6518 * sb_start or, if that is <data_offset, it must fit before the size
6519 * of each device. If num_sectors is zero, we find the largest size
6520 * that fits.
6521 */
6522 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6523 mddev->sync_thread)
6524 return -EBUSY;
6525 if (mddev->ro)
6526 return -EROFS;
6527
6528 rdev_for_each(rdev, mddev) {
6529 sector_t avail = rdev->sectors;
6530
6531 if (fit && (num_sectors == 0 || num_sectors > avail))
6532 num_sectors = avail;
6533 if (avail < num_sectors)
6534 return -ENOSPC;
6535 }
6536 rv = mddev->pers->resize(mddev, num_sectors);
6537 if (!rv)
6538 revalidate_disk(mddev->gendisk);
6539 return rv;
6540 }
6541
6542 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6543 {
6544 int rv;
6545 struct md_rdev *rdev;
6546 /* change the number of raid disks */
6547 if (mddev->pers->check_reshape == NULL)
6548 return -EINVAL;
6549 if (mddev->ro)
6550 return -EROFS;
6551 if (raid_disks <= 0 ||
6552 (mddev->max_disks && raid_disks >= mddev->max_disks))
6553 return -EINVAL;
6554 if (mddev->sync_thread ||
6555 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6556 mddev->reshape_position != MaxSector)
6557 return -EBUSY;
6558
6559 rdev_for_each(rdev, mddev) {
6560 if (mddev->raid_disks < raid_disks &&
6561 rdev->data_offset < rdev->new_data_offset)
6562 return -EINVAL;
6563 if (mddev->raid_disks > raid_disks &&
6564 rdev->data_offset > rdev->new_data_offset)
6565 return -EINVAL;
6566 }
6567
6568 mddev->delta_disks = raid_disks - mddev->raid_disks;
6569 if (mddev->delta_disks < 0)
6570 mddev->reshape_backwards = 1;
6571 else if (mddev->delta_disks > 0)
6572 mddev->reshape_backwards = 0;
6573
6574 rv = mddev->pers->check_reshape(mddev);
6575 if (rv < 0) {
6576 mddev->delta_disks = 0;
6577 mddev->reshape_backwards = 0;
6578 }
6579 return rv;
6580 }
6581
6582 /*
6583 * update_array_info is used to change the configuration of an
6584 * on-line array.
6585 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6586 * fields in the info are checked against the array.
6587 * Any differences that cannot be handled will cause an error.
6588 * Normally, only one change can be managed at a time.
6589 */
6590 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6591 {
6592 int rv = 0;
6593 int cnt = 0;
6594 int state = 0;
6595
6596 /* calculate expected state,ignoring low bits */
6597 if (mddev->bitmap && mddev->bitmap_info.offset)
6598 state |= (1 << MD_SB_BITMAP_PRESENT);
6599
6600 if (mddev->major_version != info->major_version ||
6601 mddev->minor_version != info->minor_version ||
6602 /* mddev->patch_version != info->patch_version || */
6603 mddev->ctime != info->ctime ||
6604 mddev->level != info->level ||
6605 /* mddev->layout != info->layout || */
6606 mddev->persistent != !info->not_persistent ||
6607 mddev->chunk_sectors != info->chunk_size >> 9 ||
6608 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6609 ((state^info->state) & 0xfffffe00)
6610 )
6611 return -EINVAL;
6612 /* Check there is only one change */
6613 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6614 cnt++;
6615 if (mddev->raid_disks != info->raid_disks)
6616 cnt++;
6617 if (mddev->layout != info->layout)
6618 cnt++;
6619 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6620 cnt++;
6621 if (cnt == 0)
6622 return 0;
6623 if (cnt > 1)
6624 return -EINVAL;
6625
6626 if (mddev->layout != info->layout) {
6627 /* Change layout
6628 * we don't need to do anything at the md level, the
6629 * personality will take care of it all.
6630 */
6631 if (mddev->pers->check_reshape == NULL)
6632 return -EINVAL;
6633 else {
6634 mddev->new_layout = info->layout;
6635 rv = mddev->pers->check_reshape(mddev);
6636 if (rv)
6637 mddev->new_layout = mddev->layout;
6638 return rv;
6639 }
6640 }
6641 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6642 rv = update_size(mddev, (sector_t)info->size * 2);
6643
6644 if (mddev->raid_disks != info->raid_disks)
6645 rv = update_raid_disks(mddev, info->raid_disks);
6646
6647 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6648 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
6649 rv = -EINVAL;
6650 goto err;
6651 }
6652 if (mddev->recovery || mddev->sync_thread) {
6653 rv = -EBUSY;
6654 goto err;
6655 }
6656 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6657 struct bitmap *bitmap;
6658 /* add the bitmap */
6659 if (mddev->bitmap) {
6660 rv = -EEXIST;
6661 goto err;
6662 }
6663 if (mddev->bitmap_info.default_offset == 0) {
6664 rv = -EINVAL;
6665 goto err;
6666 }
6667 mddev->bitmap_info.offset =
6668 mddev->bitmap_info.default_offset;
6669 mddev->bitmap_info.space =
6670 mddev->bitmap_info.default_space;
6671 mddev->pers->quiesce(mddev, 1);
6672 bitmap = bitmap_create(mddev, -1);
6673 if (!IS_ERR(bitmap)) {
6674 mddev->bitmap = bitmap;
6675 rv = bitmap_load(mddev);
6676 } else
6677 rv = PTR_ERR(bitmap);
6678 if (rv)
6679 bitmap_destroy(mddev);
6680 mddev->pers->quiesce(mddev, 0);
6681 } else {
6682 /* remove the bitmap */
6683 if (!mddev->bitmap) {
6684 rv = -ENOENT;
6685 goto err;
6686 }
6687 if (mddev->bitmap->storage.file) {
6688 rv = -EINVAL;
6689 goto err;
6690 }
6691 if (mddev->bitmap_info.nodes) {
6692 /* hold PW on all the bitmap lock */
6693 if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) {
6694 pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n");
6695 rv = -EPERM;
6696 md_cluster_ops->unlock_all_bitmaps(mddev);
6697 goto err;
6698 }
6699
6700 mddev->bitmap_info.nodes = 0;
6701 md_cluster_ops->leave(mddev);
6702 }
6703 mddev->pers->quiesce(mddev, 1);
6704 bitmap_destroy(mddev);
6705 mddev->pers->quiesce(mddev, 0);
6706 mddev->bitmap_info.offset = 0;
6707 }
6708 }
6709 md_update_sb(mddev, 1);
6710 return rv;
6711 err:
6712 return rv;
6713 }
6714
6715 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6716 {
6717 struct md_rdev *rdev;
6718 int err = 0;
6719
6720 if (mddev->pers == NULL)
6721 return -ENODEV;
6722
6723 rcu_read_lock();
6724 rdev = find_rdev_rcu(mddev, dev);
6725 if (!rdev)
6726 err = -ENODEV;
6727 else {
6728 md_error(mddev, rdev);
6729 if (!test_bit(Faulty, &rdev->flags))
6730 err = -EBUSY;
6731 }
6732 rcu_read_unlock();
6733 return err;
6734 }
6735
6736 /*
6737 * We have a problem here : there is no easy way to give a CHS
6738 * virtual geometry. We currently pretend that we have a 2 heads
6739 * 4 sectors (with a BIG number of cylinders...). This drives
6740 * dosfs just mad... ;-)
6741 */
6742 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6743 {
6744 struct mddev *mddev = bdev->bd_disk->private_data;
6745
6746 geo->heads = 2;
6747 geo->sectors = 4;
6748 geo->cylinders = mddev->array_sectors / 8;
6749 return 0;
6750 }
6751
6752 static inline bool md_ioctl_valid(unsigned int cmd)
6753 {
6754 switch (cmd) {
6755 case ADD_NEW_DISK:
6756 case BLKROSET:
6757 case GET_ARRAY_INFO:
6758 case GET_BITMAP_FILE:
6759 case GET_DISK_INFO:
6760 case HOT_ADD_DISK:
6761 case HOT_REMOVE_DISK:
6762 case RAID_AUTORUN:
6763 case RAID_VERSION:
6764 case RESTART_ARRAY_RW:
6765 case RUN_ARRAY:
6766 case SET_ARRAY_INFO:
6767 case SET_BITMAP_FILE:
6768 case SET_DISK_FAULTY:
6769 case STOP_ARRAY:
6770 case STOP_ARRAY_RO:
6771 case CLUSTERED_DISK_NACK:
6772 return true;
6773 default:
6774 return false;
6775 }
6776 }
6777
6778 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6779 unsigned int cmd, unsigned long arg)
6780 {
6781 int err = 0;
6782 void __user *argp = (void __user *)arg;
6783 struct mddev *mddev = NULL;
6784 int ro;
6785
6786 if (!md_ioctl_valid(cmd))
6787 return -ENOTTY;
6788
6789 switch (cmd) {
6790 case RAID_VERSION:
6791 case GET_ARRAY_INFO:
6792 case GET_DISK_INFO:
6793 break;
6794 default:
6795 if (!capable(CAP_SYS_ADMIN))
6796 return -EACCES;
6797 }
6798
6799 /*
6800 * Commands dealing with the RAID driver but not any
6801 * particular array:
6802 */
6803 switch (cmd) {
6804 case RAID_VERSION:
6805 err = get_version(argp);
6806 goto out;
6807
6808 #ifndef MODULE
6809 case RAID_AUTORUN:
6810 err = 0;
6811 autostart_arrays(arg);
6812 goto out;
6813 #endif
6814 default:;
6815 }
6816
6817 /*
6818 * Commands creating/starting a new array:
6819 */
6820
6821 mddev = bdev->bd_disk->private_data;
6822
6823 if (!mddev) {
6824 BUG();
6825 goto out;
6826 }
6827
6828 /* Some actions do not requires the mutex */
6829 switch (cmd) {
6830 case GET_ARRAY_INFO:
6831 if (!mddev->raid_disks && !mddev->external)
6832 err = -ENODEV;
6833 else
6834 err = get_array_info(mddev, argp);
6835 goto out;
6836
6837 case GET_DISK_INFO:
6838 if (!mddev->raid_disks && !mddev->external)
6839 err = -ENODEV;
6840 else
6841 err = get_disk_info(mddev, argp);
6842 goto out;
6843
6844 case SET_DISK_FAULTY:
6845 err = set_disk_faulty(mddev, new_decode_dev(arg));
6846 goto out;
6847
6848 case GET_BITMAP_FILE:
6849 err = get_bitmap_file(mddev, argp);
6850 goto out;
6851
6852 }
6853
6854 if (cmd == ADD_NEW_DISK)
6855 /* need to ensure md_delayed_delete() has completed */
6856 flush_workqueue(md_misc_wq);
6857
6858 if (cmd == HOT_REMOVE_DISK)
6859 /* need to ensure recovery thread has run */
6860 wait_event_interruptible_timeout(mddev->sb_wait,
6861 !test_bit(MD_RECOVERY_NEEDED,
6862 &mddev->recovery),
6863 msecs_to_jiffies(5000));
6864 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6865 /* Need to flush page cache, and ensure no-one else opens
6866 * and writes
6867 */
6868 mutex_lock(&mddev->open_mutex);
6869 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
6870 mutex_unlock(&mddev->open_mutex);
6871 err = -EBUSY;
6872 goto out;
6873 }
6874 set_bit(MD_CLOSING, &mddev->flags);
6875 mutex_unlock(&mddev->open_mutex);
6876 sync_blockdev(bdev);
6877 }
6878 err = mddev_lock(mddev);
6879 if (err) {
6880 pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n",
6881 err, cmd);
6882 goto out;
6883 }
6884
6885 if (cmd == SET_ARRAY_INFO) {
6886 mdu_array_info_t info;
6887 if (!arg)
6888 memset(&info, 0, sizeof(info));
6889 else if (copy_from_user(&info, argp, sizeof(info))) {
6890 err = -EFAULT;
6891 goto unlock;
6892 }
6893 if (mddev->pers) {
6894 err = update_array_info(mddev, &info);
6895 if (err) {
6896 pr_warn("md: couldn't update array info. %d\n", err);
6897 goto unlock;
6898 }
6899 goto unlock;
6900 }
6901 if (!list_empty(&mddev->disks)) {
6902 pr_warn("md: array %s already has disks!\n", mdname(mddev));
6903 err = -EBUSY;
6904 goto unlock;
6905 }
6906 if (mddev->raid_disks) {
6907 pr_warn("md: array %s already initialised!\n", mdname(mddev));
6908 err = -EBUSY;
6909 goto unlock;
6910 }
6911 err = set_array_info(mddev, &info);
6912 if (err) {
6913 pr_warn("md: couldn't set array info. %d\n", err);
6914 goto unlock;
6915 }
6916 goto unlock;
6917 }
6918
6919 /*
6920 * Commands querying/configuring an existing array:
6921 */
6922 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6923 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6924 if ((!mddev->raid_disks && !mddev->external)
6925 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6926 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6927 && cmd != GET_BITMAP_FILE) {
6928 err = -ENODEV;
6929 goto unlock;
6930 }
6931
6932 /*
6933 * Commands even a read-only array can execute:
6934 */
6935 switch (cmd) {
6936 case RESTART_ARRAY_RW:
6937 err = restart_array(mddev);
6938 goto unlock;
6939
6940 case STOP_ARRAY:
6941 err = do_md_stop(mddev, 0, bdev);
6942 goto unlock;
6943
6944 case STOP_ARRAY_RO:
6945 err = md_set_readonly(mddev, bdev);
6946 goto unlock;
6947
6948 case HOT_REMOVE_DISK:
6949 err = hot_remove_disk(mddev, new_decode_dev(arg));
6950 goto unlock;
6951
6952 case ADD_NEW_DISK:
6953 /* We can support ADD_NEW_DISK on read-only arrays
6954 * only if we are re-adding a preexisting device.
6955 * So require mddev->pers and MD_DISK_SYNC.
6956 */
6957 if (mddev->pers) {
6958 mdu_disk_info_t info;
6959 if (copy_from_user(&info, argp, sizeof(info)))
6960 err = -EFAULT;
6961 else if (!(info.state & (1<<MD_DISK_SYNC)))
6962 /* Need to clear read-only for this */
6963 break;
6964 else
6965 err = add_new_disk(mddev, &info);
6966 goto unlock;
6967 }
6968 break;
6969
6970 case BLKROSET:
6971 if (get_user(ro, (int __user *)(arg))) {
6972 err = -EFAULT;
6973 goto unlock;
6974 }
6975 err = -EINVAL;
6976
6977 /* if the bdev is going readonly the value of mddev->ro
6978 * does not matter, no writes are coming
6979 */
6980 if (ro)
6981 goto unlock;
6982
6983 /* are we are already prepared for writes? */
6984 if (mddev->ro != 1)
6985 goto unlock;
6986
6987 /* transitioning to readauto need only happen for
6988 * arrays that call md_write_start
6989 */
6990 if (mddev->pers) {
6991 err = restart_array(mddev);
6992 if (err == 0) {
6993 mddev->ro = 2;
6994 set_disk_ro(mddev->gendisk, 0);
6995 }
6996 }
6997 goto unlock;
6998 }
6999
7000 /*
7001 * The remaining ioctls are changing the state of the
7002 * superblock, so we do not allow them on read-only arrays.
7003 */
7004 if (mddev->ro && mddev->pers) {
7005 if (mddev->ro == 2) {
7006 mddev->ro = 0;
7007 sysfs_notify_dirent_safe(mddev->sysfs_state);
7008 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7009 /* mddev_unlock will wake thread */
7010 /* If a device failed while we were read-only, we
7011 * need to make sure the metadata is updated now.
7012 */
7013 if (test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) {
7014 mddev_unlock(mddev);
7015 wait_event(mddev->sb_wait,
7016 !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) &&
7017 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
7018 mddev_lock_nointr(mddev);
7019 }
7020 } else {
7021 err = -EROFS;
7022 goto unlock;
7023 }
7024 }
7025
7026 switch (cmd) {
7027 case ADD_NEW_DISK:
7028 {
7029 mdu_disk_info_t info;
7030 if (copy_from_user(&info, argp, sizeof(info)))
7031 err = -EFAULT;
7032 else
7033 err = add_new_disk(mddev, &info);
7034 goto unlock;
7035 }
7036
7037 case CLUSTERED_DISK_NACK:
7038 if (mddev_is_clustered(mddev))
7039 md_cluster_ops->new_disk_ack(mddev, false);
7040 else
7041 err = -EINVAL;
7042 goto unlock;
7043
7044 case HOT_ADD_DISK:
7045 err = hot_add_disk(mddev, new_decode_dev(arg));
7046 goto unlock;
7047
7048 case RUN_ARRAY:
7049 err = do_md_run(mddev);
7050 goto unlock;
7051
7052 case SET_BITMAP_FILE:
7053 err = set_bitmap_file(mddev, (int)arg);
7054 goto unlock;
7055
7056 default:
7057 err = -EINVAL;
7058 goto unlock;
7059 }
7060
7061 unlock:
7062 if (mddev->hold_active == UNTIL_IOCTL &&
7063 err != -EINVAL)
7064 mddev->hold_active = 0;
7065 mddev_unlock(mddev);
7066 out:
7067 return err;
7068 }
7069 #ifdef CONFIG_COMPAT
7070 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
7071 unsigned int cmd, unsigned long arg)
7072 {
7073 switch (cmd) {
7074 case HOT_REMOVE_DISK:
7075 case HOT_ADD_DISK:
7076 case SET_DISK_FAULTY:
7077 case SET_BITMAP_FILE:
7078 /* These take in integer arg, do not convert */
7079 break;
7080 default:
7081 arg = (unsigned long)compat_ptr(arg);
7082 break;
7083 }
7084
7085 return md_ioctl(bdev, mode, cmd, arg);
7086 }
7087 #endif /* CONFIG_COMPAT */
7088
7089 static int md_open(struct block_device *bdev, fmode_t mode)
7090 {
7091 /*
7092 * Succeed if we can lock the mddev, which confirms that
7093 * it isn't being stopped right now.
7094 */
7095 struct mddev *mddev = mddev_find(bdev->bd_dev);
7096 int err;
7097
7098 if (!mddev)
7099 return -ENODEV;
7100
7101 if (mddev->gendisk != bdev->bd_disk) {
7102 /* we are racing with mddev_put which is discarding this
7103 * bd_disk.
7104 */
7105 mddev_put(mddev);
7106 /* Wait until bdev->bd_disk is definitely gone */
7107 flush_workqueue(md_misc_wq);
7108 /* Then retry the open from the top */
7109 return -ERESTARTSYS;
7110 }
7111 BUG_ON(mddev != bdev->bd_disk->private_data);
7112
7113 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
7114 goto out;
7115
7116 if (test_bit(MD_CLOSING, &mddev->flags)) {
7117 mutex_unlock(&mddev->open_mutex);
7118 err = -ENODEV;
7119 goto out;
7120 }
7121
7122 err = 0;
7123 atomic_inc(&mddev->openers);
7124 mutex_unlock(&mddev->open_mutex);
7125
7126 check_disk_change(bdev);
7127 out:
7128 if (err)
7129 mddev_put(mddev);
7130 return err;
7131 }
7132
7133 static void md_release(struct gendisk *disk, fmode_t mode)
7134 {
7135 struct mddev *mddev = disk->private_data;
7136
7137 BUG_ON(!mddev);
7138 atomic_dec(&mddev->openers);
7139 mddev_put(mddev);
7140 }
7141
7142 static int md_media_changed(struct gendisk *disk)
7143 {
7144 struct mddev *mddev = disk->private_data;
7145
7146 return mddev->changed;
7147 }
7148
7149 static int md_revalidate(struct gendisk *disk)
7150 {
7151 struct mddev *mddev = disk->private_data;
7152
7153 mddev->changed = 0;
7154 return 0;
7155 }
7156 static const struct block_device_operations md_fops =
7157 {
7158 .owner = THIS_MODULE,
7159 .open = md_open,
7160 .release = md_release,
7161 .ioctl = md_ioctl,
7162 #ifdef CONFIG_COMPAT
7163 .compat_ioctl = md_compat_ioctl,
7164 #endif
7165 .getgeo = md_getgeo,
7166 .media_changed = md_media_changed,
7167 .revalidate_disk= md_revalidate,
7168 };
7169
7170 static int md_thread(void *arg)
7171 {
7172 struct md_thread *thread = arg;
7173
7174 /*
7175 * md_thread is a 'system-thread', it's priority should be very
7176 * high. We avoid resource deadlocks individually in each
7177 * raid personality. (RAID5 does preallocation) We also use RR and
7178 * the very same RT priority as kswapd, thus we will never get
7179 * into a priority inversion deadlock.
7180 *
7181 * we definitely have to have equal or higher priority than
7182 * bdflush, otherwise bdflush will deadlock if there are too
7183 * many dirty RAID5 blocks.
7184 */
7185
7186 allow_signal(SIGKILL);
7187 while (!kthread_should_stop()) {
7188
7189 /* We need to wait INTERRUPTIBLE so that
7190 * we don't add to the load-average.
7191 * That means we need to be sure no signals are
7192 * pending
7193 */
7194 if (signal_pending(current))
7195 flush_signals(current);
7196
7197 wait_event_interruptible_timeout
7198 (thread->wqueue,
7199 test_bit(THREAD_WAKEUP, &thread->flags)
7200 || kthread_should_stop() || kthread_should_park(),
7201 thread->timeout);
7202
7203 clear_bit(THREAD_WAKEUP, &thread->flags);
7204 if (kthread_should_park())
7205 kthread_parkme();
7206 if (!kthread_should_stop())
7207 thread->run(thread);
7208 }
7209
7210 return 0;
7211 }
7212
7213 void md_wakeup_thread(struct md_thread *thread)
7214 {
7215 if (thread) {
7216 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
7217 set_bit(THREAD_WAKEUP, &thread->flags);
7218 wake_up(&thread->wqueue);
7219 }
7220 }
7221 EXPORT_SYMBOL(md_wakeup_thread);
7222
7223 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7224 struct mddev *mddev, const char *name)
7225 {
7226 struct md_thread *thread;
7227
7228 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7229 if (!thread)
7230 return NULL;
7231
7232 init_waitqueue_head(&thread->wqueue);
7233
7234 thread->run = run;
7235 thread->mddev = mddev;
7236 thread->timeout = MAX_SCHEDULE_TIMEOUT;
7237 thread->tsk = kthread_run(md_thread, thread,
7238 "%s_%s",
7239 mdname(thread->mddev),
7240 name);
7241 if (IS_ERR(thread->tsk)) {
7242 kfree(thread);
7243 return NULL;
7244 }
7245 return thread;
7246 }
7247 EXPORT_SYMBOL(md_register_thread);
7248
7249 void md_unregister_thread(struct md_thread **threadp)
7250 {
7251 struct md_thread *thread = *threadp;
7252 if (!thread)
7253 return;
7254 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7255 /* Locking ensures that mddev_unlock does not wake_up a
7256 * non-existent thread
7257 */
7258 spin_lock(&pers_lock);
7259 *threadp = NULL;
7260 spin_unlock(&pers_lock);
7261
7262 kthread_stop(thread->tsk);
7263 kfree(thread);
7264 }
7265 EXPORT_SYMBOL(md_unregister_thread);
7266
7267 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7268 {
7269 if (!rdev || test_bit(Faulty, &rdev->flags))
7270 return;
7271
7272 if (!mddev->pers || !mddev->pers->error_handler)
7273 return;
7274 mddev->pers->error_handler(mddev,rdev);
7275 if (mddev->degraded)
7276 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7277 sysfs_notify_dirent_safe(rdev->sysfs_state);
7278 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7279 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7280 md_wakeup_thread(mddev->thread);
7281 if (mddev->event_work.func)
7282 queue_work(md_misc_wq, &mddev->event_work);
7283 md_new_event(mddev);
7284 }
7285 EXPORT_SYMBOL(md_error);
7286
7287 /* seq_file implementation /proc/mdstat */
7288
7289 static void status_unused(struct seq_file *seq)
7290 {
7291 int i = 0;
7292 struct md_rdev *rdev;
7293
7294 seq_printf(seq, "unused devices: ");
7295
7296 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
7297 char b[BDEVNAME_SIZE];
7298 i++;
7299 seq_printf(seq, "%s ",
7300 bdevname(rdev->bdev,b));
7301 }
7302 if (!i)
7303 seq_printf(seq, "<none>");
7304
7305 seq_printf(seq, "\n");
7306 }
7307
7308 static int status_resync(struct seq_file *seq, struct mddev *mddev)
7309 {
7310 sector_t max_sectors, resync, res;
7311 unsigned long dt, db;
7312 sector_t rt;
7313 int scale;
7314 unsigned int per_milli;
7315
7316 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7317 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7318 max_sectors = mddev->resync_max_sectors;
7319 else
7320 max_sectors = mddev->dev_sectors;
7321
7322 resync = mddev->curr_resync;
7323 if (resync <= 3) {
7324 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7325 /* Still cleaning up */
7326 resync = max_sectors;
7327 } else
7328 resync -= atomic_read(&mddev->recovery_active);
7329
7330 if (resync == 0) {
7331 if (mddev->recovery_cp < MaxSector) {
7332 seq_printf(seq, "\tresync=PENDING");
7333 return 1;
7334 }
7335 return 0;
7336 }
7337 if (resync < 3) {
7338 seq_printf(seq, "\tresync=DELAYED");
7339 return 1;
7340 }
7341
7342 WARN_ON(max_sectors == 0);
7343 /* Pick 'scale' such that (resync>>scale)*1000 will fit
7344 * in a sector_t, and (max_sectors>>scale) will fit in a
7345 * u32, as those are the requirements for sector_div.
7346 * Thus 'scale' must be at least 10
7347 */
7348 scale = 10;
7349 if (sizeof(sector_t) > sizeof(unsigned long)) {
7350 while ( max_sectors/2 > (1ULL<<(scale+32)))
7351 scale++;
7352 }
7353 res = (resync>>scale)*1000;
7354 sector_div(res, (u32)((max_sectors>>scale)+1));
7355
7356 per_milli = res;
7357 {
7358 int i, x = per_milli/50, y = 20-x;
7359 seq_printf(seq, "[");
7360 for (i = 0; i < x; i++)
7361 seq_printf(seq, "=");
7362 seq_printf(seq, ">");
7363 for (i = 0; i < y; i++)
7364 seq_printf(seq, ".");
7365 seq_printf(seq, "] ");
7366 }
7367 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
7368 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
7369 "reshape" :
7370 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
7371 "check" :
7372 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
7373 "resync" : "recovery"))),
7374 per_milli/10, per_milli % 10,
7375 (unsigned long long) resync/2,
7376 (unsigned long long) max_sectors/2);
7377
7378 /*
7379 * dt: time from mark until now
7380 * db: blocks written from mark until now
7381 * rt: remaining time
7382 *
7383 * rt is a sector_t, so could be 32bit or 64bit.
7384 * So we divide before multiply in case it is 32bit and close
7385 * to the limit.
7386 * We scale the divisor (db) by 32 to avoid losing precision
7387 * near the end of resync when the number of remaining sectors
7388 * is close to 'db'.
7389 * We then divide rt by 32 after multiplying by db to compensate.
7390 * The '+1' avoids division by zero if db is very small.
7391 */
7392 dt = ((jiffies - mddev->resync_mark) / HZ);
7393 if (!dt) dt++;
7394 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
7395 - mddev->resync_mark_cnt;
7396
7397 rt = max_sectors - resync; /* number of remaining sectors */
7398 sector_div(rt, db/32+1);
7399 rt *= dt;
7400 rt >>= 5;
7401
7402 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7403 ((unsigned long)rt % 60)/6);
7404
7405 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
7406 return 1;
7407 }
7408
7409 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7410 {
7411 struct list_head *tmp;
7412 loff_t l = *pos;
7413 struct mddev *mddev;
7414
7415 if (l >= 0x10000)
7416 return NULL;
7417 if (!l--)
7418 /* header */
7419 return (void*)1;
7420
7421 spin_lock(&all_mddevs_lock);
7422 list_for_each(tmp,&all_mddevs)
7423 if (!l--) {
7424 mddev = list_entry(tmp, struct mddev, all_mddevs);
7425 mddev_get(mddev);
7426 spin_unlock(&all_mddevs_lock);
7427 return mddev;
7428 }
7429 spin_unlock(&all_mddevs_lock);
7430 if (!l--)
7431 return (void*)2;/* tail */
7432 return NULL;
7433 }
7434
7435 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7436 {
7437 struct list_head *tmp;
7438 struct mddev *next_mddev, *mddev = v;
7439
7440 ++*pos;
7441 if (v == (void*)2)
7442 return NULL;
7443
7444 spin_lock(&all_mddevs_lock);
7445 if (v == (void*)1)
7446 tmp = all_mddevs.next;
7447 else
7448 tmp = mddev->all_mddevs.next;
7449 if (tmp != &all_mddevs)
7450 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
7451 else {
7452 next_mddev = (void*)2;
7453 *pos = 0x10000;
7454 }
7455 spin_unlock(&all_mddevs_lock);
7456
7457 if (v != (void*)1)
7458 mddev_put(mddev);
7459 return next_mddev;
7460
7461 }
7462
7463 static void md_seq_stop(struct seq_file *seq, void *v)
7464 {
7465 struct mddev *mddev = v;
7466
7467 if (mddev && v != (void*)1 && v != (void*)2)
7468 mddev_put(mddev);
7469 }
7470
7471 static int md_seq_show(struct seq_file *seq, void *v)
7472 {
7473 struct mddev *mddev = v;
7474 sector_t sectors;
7475 struct md_rdev *rdev;
7476
7477 if (v == (void*)1) {
7478 struct md_personality *pers;
7479 seq_printf(seq, "Personalities : ");
7480 spin_lock(&pers_lock);
7481 list_for_each_entry(pers, &pers_list, list)
7482 seq_printf(seq, "[%s] ", pers->name);
7483
7484 spin_unlock(&pers_lock);
7485 seq_printf(seq, "\n");
7486 seq->poll_event = atomic_read(&md_event_count);
7487 return 0;
7488 }
7489 if (v == (void*)2) {
7490 status_unused(seq);
7491 return 0;
7492 }
7493
7494 spin_lock(&mddev->lock);
7495 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7496 seq_printf(seq, "%s : %sactive", mdname(mddev),
7497 mddev->pers ? "" : "in");
7498 if (mddev->pers) {
7499 if (mddev->ro==1)
7500 seq_printf(seq, " (read-only)");
7501 if (mddev->ro==2)
7502 seq_printf(seq, " (auto-read-only)");
7503 seq_printf(seq, " %s", mddev->pers->name);
7504 }
7505
7506 sectors = 0;
7507 rcu_read_lock();
7508 rdev_for_each_rcu(rdev, mddev) {
7509 char b[BDEVNAME_SIZE];
7510 seq_printf(seq, " %s[%d]",
7511 bdevname(rdev->bdev,b), rdev->desc_nr);
7512 if (test_bit(WriteMostly, &rdev->flags))
7513 seq_printf(seq, "(W)");
7514 if (test_bit(Journal, &rdev->flags))
7515 seq_printf(seq, "(J)");
7516 if (test_bit(Faulty, &rdev->flags)) {
7517 seq_printf(seq, "(F)");
7518 continue;
7519 }
7520 if (rdev->raid_disk < 0)
7521 seq_printf(seq, "(S)"); /* spare */
7522 if (test_bit(Replacement, &rdev->flags))
7523 seq_printf(seq, "(R)");
7524 sectors += rdev->sectors;
7525 }
7526 rcu_read_unlock();
7527
7528 if (!list_empty(&mddev->disks)) {
7529 if (mddev->pers)
7530 seq_printf(seq, "\n %llu blocks",
7531 (unsigned long long)
7532 mddev->array_sectors / 2);
7533 else
7534 seq_printf(seq, "\n %llu blocks",
7535 (unsigned long long)sectors / 2);
7536 }
7537 if (mddev->persistent) {
7538 if (mddev->major_version != 0 ||
7539 mddev->minor_version != 90) {
7540 seq_printf(seq," super %d.%d",
7541 mddev->major_version,
7542 mddev->minor_version);
7543 }
7544 } else if (mddev->external)
7545 seq_printf(seq, " super external:%s",
7546 mddev->metadata_type);
7547 else
7548 seq_printf(seq, " super non-persistent");
7549
7550 if (mddev->pers) {
7551 mddev->pers->status(seq, mddev);
7552 seq_printf(seq, "\n ");
7553 if (mddev->pers->sync_request) {
7554 if (status_resync(seq, mddev))
7555 seq_printf(seq, "\n ");
7556 }
7557 } else
7558 seq_printf(seq, "\n ");
7559
7560 bitmap_status(seq, mddev->bitmap);
7561
7562 seq_printf(seq, "\n");
7563 }
7564 spin_unlock(&mddev->lock);
7565
7566 return 0;
7567 }
7568
7569 static const struct seq_operations md_seq_ops = {
7570 .start = md_seq_start,
7571 .next = md_seq_next,
7572 .stop = md_seq_stop,
7573 .show = md_seq_show,
7574 };
7575
7576 static int md_seq_open(struct inode *inode, struct file *file)
7577 {
7578 struct seq_file *seq;
7579 int error;
7580
7581 error = seq_open(file, &md_seq_ops);
7582 if (error)
7583 return error;
7584
7585 seq = file->private_data;
7586 seq->poll_event = atomic_read(&md_event_count);
7587 return error;
7588 }
7589
7590 static int md_unloading;
7591 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7592 {
7593 struct seq_file *seq = filp->private_data;
7594 int mask;
7595
7596 if (md_unloading)
7597 return POLLIN|POLLRDNORM|POLLERR|POLLPRI;
7598 poll_wait(filp, &md_event_waiters, wait);
7599
7600 /* always allow read */
7601 mask = POLLIN | POLLRDNORM;
7602
7603 if (seq->poll_event != atomic_read(&md_event_count))
7604 mask |= POLLERR | POLLPRI;
7605 return mask;
7606 }
7607
7608 static const struct file_operations md_seq_fops = {
7609 .owner = THIS_MODULE,
7610 .open = md_seq_open,
7611 .read = seq_read,
7612 .llseek = seq_lseek,
7613 .release = seq_release_private,
7614 .poll = mdstat_poll,
7615 };
7616
7617 int register_md_personality(struct md_personality *p)
7618 {
7619 pr_debug("md: %s personality registered for level %d\n",
7620 p->name, p->level);
7621 spin_lock(&pers_lock);
7622 list_add_tail(&p->list, &pers_list);
7623 spin_unlock(&pers_lock);
7624 return 0;
7625 }
7626 EXPORT_SYMBOL(register_md_personality);
7627
7628 int unregister_md_personality(struct md_personality *p)
7629 {
7630 pr_debug("md: %s personality unregistered\n", p->name);
7631 spin_lock(&pers_lock);
7632 list_del_init(&p->list);
7633 spin_unlock(&pers_lock);
7634 return 0;
7635 }
7636 EXPORT_SYMBOL(unregister_md_personality);
7637
7638 int register_md_cluster_operations(struct md_cluster_operations *ops,
7639 struct module *module)
7640 {
7641 int ret = 0;
7642 spin_lock(&pers_lock);
7643 if (md_cluster_ops != NULL)
7644 ret = -EALREADY;
7645 else {
7646 md_cluster_ops = ops;
7647 md_cluster_mod = module;
7648 }
7649 spin_unlock(&pers_lock);
7650 return ret;
7651 }
7652 EXPORT_SYMBOL(register_md_cluster_operations);
7653
7654 int unregister_md_cluster_operations(void)
7655 {
7656 spin_lock(&pers_lock);
7657 md_cluster_ops = NULL;
7658 spin_unlock(&pers_lock);
7659 return 0;
7660 }
7661 EXPORT_SYMBOL(unregister_md_cluster_operations);
7662
7663 int md_setup_cluster(struct mddev *mddev, int nodes)
7664 {
7665 if (!md_cluster_ops)
7666 request_module("md-cluster");
7667 spin_lock(&pers_lock);
7668 /* ensure module won't be unloaded */
7669 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
7670 pr_warn("can't find md-cluster module or get it's reference.\n");
7671 spin_unlock(&pers_lock);
7672 return -ENOENT;
7673 }
7674 spin_unlock(&pers_lock);
7675
7676 return md_cluster_ops->join(mddev, nodes);
7677 }
7678
7679 void md_cluster_stop(struct mddev *mddev)
7680 {
7681 if (!md_cluster_ops)
7682 return;
7683 md_cluster_ops->leave(mddev);
7684 module_put(md_cluster_mod);
7685 }
7686
7687 static int is_mddev_idle(struct mddev *mddev, int init)
7688 {
7689 struct md_rdev *rdev;
7690 int idle;
7691 int curr_events;
7692
7693 idle = 1;
7694 rcu_read_lock();
7695 rdev_for_each_rcu(rdev, mddev) {
7696 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7697 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7698 (int)part_stat_read(&disk->part0, sectors[1]) -
7699 atomic_read(&disk->sync_io);
7700 /* sync IO will cause sync_io to increase before the disk_stats
7701 * as sync_io is counted when a request starts, and
7702 * disk_stats is counted when it completes.
7703 * So resync activity will cause curr_events to be smaller than
7704 * when there was no such activity.
7705 * non-sync IO will cause disk_stat to increase without
7706 * increasing sync_io so curr_events will (eventually)
7707 * be larger than it was before. Once it becomes
7708 * substantially larger, the test below will cause
7709 * the array to appear non-idle, and resync will slow
7710 * down.
7711 * If there is a lot of outstanding resync activity when
7712 * we set last_event to curr_events, then all that activity
7713 * completing might cause the array to appear non-idle
7714 * and resync will be slowed down even though there might
7715 * not have been non-resync activity. This will only
7716 * happen once though. 'last_events' will soon reflect
7717 * the state where there is little or no outstanding
7718 * resync requests, and further resync activity will
7719 * always make curr_events less than last_events.
7720 *
7721 */
7722 if (init || curr_events - rdev->last_events > 64) {
7723 rdev->last_events = curr_events;
7724 idle = 0;
7725 }
7726 }
7727 rcu_read_unlock();
7728 return idle;
7729 }
7730
7731 void md_done_sync(struct mddev *mddev, int blocks, int ok)
7732 {
7733 /* another "blocks" (512byte) blocks have been synced */
7734 atomic_sub(blocks, &mddev->recovery_active);
7735 wake_up(&mddev->recovery_wait);
7736 if (!ok) {
7737 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7738 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
7739 md_wakeup_thread(mddev->thread);
7740 // stop recovery, signal do_sync ....
7741 }
7742 }
7743 EXPORT_SYMBOL(md_done_sync);
7744
7745 /* md_write_start(mddev, bi)
7746 * If we need to update some array metadata (e.g. 'active' flag
7747 * in superblock) before writing, schedule a superblock update
7748 * and wait for it to complete.
7749 */
7750 void md_write_start(struct mddev *mddev, struct bio *bi)
7751 {
7752 int did_change = 0;
7753 if (bio_data_dir(bi) != WRITE)
7754 return;
7755
7756 BUG_ON(mddev->ro == 1);
7757 if (mddev->ro == 2) {
7758 /* need to switch to read/write */
7759 mddev->ro = 0;
7760 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7761 md_wakeup_thread(mddev->thread);
7762 md_wakeup_thread(mddev->sync_thread);
7763 did_change = 1;
7764 }
7765 atomic_inc(&mddev->writes_pending);
7766 if (mddev->safemode == 1)
7767 mddev->safemode = 0;
7768 if (mddev->in_sync) {
7769 spin_lock(&mddev->lock);
7770 if (mddev->in_sync) {
7771 mddev->in_sync = 0;
7772 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
7773 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
7774 md_wakeup_thread(mddev->thread);
7775 did_change = 1;
7776 }
7777 spin_unlock(&mddev->lock);
7778 }
7779 if (did_change)
7780 sysfs_notify_dirent_safe(mddev->sysfs_state);
7781 wait_event(mddev->sb_wait,
7782 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
7783 }
7784 EXPORT_SYMBOL(md_write_start);
7785
7786 void md_write_end(struct mddev *mddev)
7787 {
7788 if (atomic_dec_and_test(&mddev->writes_pending)) {
7789 if (mddev->safemode == 2)
7790 md_wakeup_thread(mddev->thread);
7791 else if (mddev->safemode_delay)
7792 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7793 }
7794 }
7795 EXPORT_SYMBOL(md_write_end);
7796
7797 /* md_allow_write(mddev)
7798 * Calling this ensures that the array is marked 'active' so that writes
7799 * may proceed without blocking. It is important to call this before
7800 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7801 * Must be called with mddev_lock held.
7802 *
7803 * In the ->external case MD_SB_CHANGE_PENDING can not be cleared until mddev->lock
7804 * is dropped, so return -EAGAIN after notifying userspace.
7805 */
7806 int md_allow_write(struct mddev *mddev)
7807 {
7808 if (!mddev->pers)
7809 return 0;
7810 if (mddev->ro)
7811 return 0;
7812 if (!mddev->pers->sync_request)
7813 return 0;
7814
7815 spin_lock(&mddev->lock);
7816 if (mddev->in_sync) {
7817 mddev->in_sync = 0;
7818 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
7819 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
7820 if (mddev->safemode_delay &&
7821 mddev->safemode == 0)
7822 mddev->safemode = 1;
7823 spin_unlock(&mddev->lock);
7824 md_update_sb(mddev, 0);
7825 sysfs_notify_dirent_safe(mddev->sysfs_state);
7826 } else
7827 spin_unlock(&mddev->lock);
7828
7829 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
7830 return -EAGAIN;
7831 else
7832 return 0;
7833 }
7834 EXPORT_SYMBOL_GPL(md_allow_write);
7835
7836 #define SYNC_MARKS 10
7837 #define SYNC_MARK_STEP (3*HZ)
7838 #define UPDATE_FREQUENCY (5*60*HZ)
7839 void md_do_sync(struct md_thread *thread)
7840 {
7841 struct mddev *mddev = thread->mddev;
7842 struct mddev *mddev2;
7843 unsigned int currspeed = 0,
7844 window;
7845 sector_t max_sectors,j, io_sectors, recovery_done;
7846 unsigned long mark[SYNC_MARKS];
7847 unsigned long update_time;
7848 sector_t mark_cnt[SYNC_MARKS];
7849 int last_mark,m;
7850 struct list_head *tmp;
7851 sector_t last_check;
7852 int skipped = 0;
7853 struct md_rdev *rdev;
7854 char *desc, *action = NULL;
7855 struct blk_plug plug;
7856 int ret;
7857
7858 /* just incase thread restarts... */
7859 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7860 return;
7861 if (mddev->ro) {/* never try to sync a read-only array */
7862 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7863 return;
7864 }
7865
7866 if (mddev_is_clustered(mddev)) {
7867 ret = md_cluster_ops->resync_start(mddev);
7868 if (ret)
7869 goto skip;
7870
7871 set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags);
7872 if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7873 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
7874 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
7875 && ((unsigned long long)mddev->curr_resync_completed
7876 < (unsigned long long)mddev->resync_max_sectors))
7877 goto skip;
7878 }
7879
7880 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7881 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
7882 desc = "data-check";
7883 action = "check";
7884 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7885 desc = "requested-resync";
7886 action = "repair";
7887 } else
7888 desc = "resync";
7889 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7890 desc = "reshape";
7891 else
7892 desc = "recovery";
7893
7894 mddev->last_sync_action = action ?: desc;
7895
7896 /* we overload curr_resync somewhat here.
7897 * 0 == not engaged in resync at all
7898 * 2 == checking that there is no conflict with another sync
7899 * 1 == like 2, but have yielded to allow conflicting resync to
7900 * commense
7901 * other == active in resync - this many blocks
7902 *
7903 * Before starting a resync we must have set curr_resync to
7904 * 2, and then checked that every "conflicting" array has curr_resync
7905 * less than ours. When we find one that is the same or higher
7906 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7907 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7908 * This will mean we have to start checking from the beginning again.
7909 *
7910 */
7911
7912 do {
7913 int mddev2_minor = -1;
7914 mddev->curr_resync = 2;
7915
7916 try_again:
7917 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7918 goto skip;
7919 for_each_mddev(mddev2, tmp) {
7920 if (mddev2 == mddev)
7921 continue;
7922 if (!mddev->parallel_resync
7923 && mddev2->curr_resync
7924 && match_mddev_units(mddev, mddev2)) {
7925 DEFINE_WAIT(wq);
7926 if (mddev < mddev2 && mddev->curr_resync == 2) {
7927 /* arbitrarily yield */
7928 mddev->curr_resync = 1;
7929 wake_up(&resync_wait);
7930 }
7931 if (mddev > mddev2 && mddev->curr_resync == 1)
7932 /* no need to wait here, we can wait the next
7933 * time 'round when curr_resync == 2
7934 */
7935 continue;
7936 /* We need to wait 'interruptible' so as not to
7937 * contribute to the load average, and not to
7938 * be caught by 'softlockup'
7939 */
7940 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7941 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7942 mddev2->curr_resync >= mddev->curr_resync) {
7943 if (mddev2_minor != mddev2->md_minor) {
7944 mddev2_minor = mddev2->md_minor;
7945 pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n",
7946 desc, mdname(mddev),
7947 mdname(mddev2));
7948 }
7949 mddev_put(mddev2);
7950 if (signal_pending(current))
7951 flush_signals(current);
7952 schedule();
7953 finish_wait(&resync_wait, &wq);
7954 goto try_again;
7955 }
7956 finish_wait(&resync_wait, &wq);
7957 }
7958 }
7959 } while (mddev->curr_resync < 2);
7960
7961 j = 0;
7962 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7963 /* resync follows the size requested by the personality,
7964 * which defaults to physical size, but can be virtual size
7965 */
7966 max_sectors = mddev->resync_max_sectors;
7967 atomic64_set(&mddev->resync_mismatches, 0);
7968 /* we don't use the checkpoint if there's a bitmap */
7969 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7970 j = mddev->resync_min;
7971 else if (!mddev->bitmap)
7972 j = mddev->recovery_cp;
7973
7974 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7975 max_sectors = mddev->resync_max_sectors;
7976 else {
7977 /* recovery follows the physical size of devices */
7978 max_sectors = mddev->dev_sectors;
7979 j = MaxSector;
7980 rcu_read_lock();
7981 rdev_for_each_rcu(rdev, mddev)
7982 if (rdev->raid_disk >= 0 &&
7983 !test_bit(Journal, &rdev->flags) &&
7984 !test_bit(Faulty, &rdev->flags) &&
7985 !test_bit(In_sync, &rdev->flags) &&
7986 rdev->recovery_offset < j)
7987 j = rdev->recovery_offset;
7988 rcu_read_unlock();
7989
7990 /* If there is a bitmap, we need to make sure all
7991 * writes that started before we added a spare
7992 * complete before we start doing a recovery.
7993 * Otherwise the write might complete and (via
7994 * bitmap_endwrite) set a bit in the bitmap after the
7995 * recovery has checked that bit and skipped that
7996 * region.
7997 */
7998 if (mddev->bitmap) {
7999 mddev->pers->quiesce(mddev, 1);
8000 mddev->pers->quiesce(mddev, 0);
8001 }
8002 }
8003
8004 pr_info("md: %s of RAID array %s\n", desc, mdname(mddev));
8005 pr_debug("md: minimum _guaranteed_ speed: %d KB/sec/disk.\n", speed_min(mddev));
8006 pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n",
8007 speed_max(mddev), desc);
8008
8009 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
8010
8011 io_sectors = 0;
8012 for (m = 0; m < SYNC_MARKS; m++) {
8013 mark[m] = jiffies;
8014 mark_cnt[m] = io_sectors;
8015 }
8016 last_mark = 0;
8017 mddev->resync_mark = mark[last_mark];
8018 mddev->resync_mark_cnt = mark_cnt[last_mark];
8019
8020 /*
8021 * Tune reconstruction:
8022 */
8023 window = 32*(PAGE_SIZE/512);
8024 pr_debug("md: using %dk window, over a total of %lluk.\n",
8025 window/2, (unsigned long long)max_sectors/2);
8026
8027 atomic_set(&mddev->recovery_active, 0);
8028 last_check = 0;
8029
8030 if (j>2) {
8031 pr_debug("md: resuming %s of %s from checkpoint.\n",
8032 desc, mdname(mddev));
8033 mddev->curr_resync = j;
8034 } else
8035 mddev->curr_resync = 3; /* no longer delayed */
8036 mddev->curr_resync_completed = j;
8037 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8038 md_new_event(mddev);
8039 update_time = jiffies;
8040
8041 blk_start_plug(&plug);
8042 while (j < max_sectors) {
8043 sector_t sectors;
8044
8045 skipped = 0;
8046
8047 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8048 ((mddev->curr_resync > mddev->curr_resync_completed &&
8049 (mddev->curr_resync - mddev->curr_resync_completed)
8050 > (max_sectors >> 4)) ||
8051 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
8052 (j - mddev->curr_resync_completed)*2
8053 >= mddev->resync_max - mddev->curr_resync_completed ||
8054 mddev->curr_resync_completed > mddev->resync_max
8055 )) {
8056 /* time to update curr_resync_completed */
8057 wait_event(mddev->recovery_wait,
8058 atomic_read(&mddev->recovery_active) == 0);
8059 mddev->curr_resync_completed = j;
8060 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
8061 j > mddev->recovery_cp)
8062 mddev->recovery_cp = j;
8063 update_time = jiffies;
8064 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8065 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8066 }
8067
8068 while (j >= mddev->resync_max &&
8069 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8070 /* As this condition is controlled by user-space,
8071 * we can block indefinitely, so use '_interruptible'
8072 * to avoid triggering warnings.
8073 */
8074 flush_signals(current); /* just in case */
8075 wait_event_interruptible(mddev->recovery_wait,
8076 mddev->resync_max > j
8077 || test_bit(MD_RECOVERY_INTR,
8078 &mddev->recovery));
8079 }
8080
8081 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8082 break;
8083
8084 sectors = mddev->pers->sync_request(mddev, j, &skipped);
8085 if (sectors == 0) {
8086 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8087 break;
8088 }
8089
8090 if (!skipped) { /* actual IO requested */
8091 io_sectors += sectors;
8092 atomic_add(sectors, &mddev->recovery_active);
8093 }
8094
8095 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8096 break;
8097
8098 j += sectors;
8099 if (j > max_sectors)
8100 /* when skipping, extra large numbers can be returned. */
8101 j = max_sectors;
8102 if (j > 2)
8103 mddev->curr_resync = j;
8104 mddev->curr_mark_cnt = io_sectors;
8105 if (last_check == 0)
8106 /* this is the earliest that rebuild will be
8107 * visible in /proc/mdstat
8108 */
8109 md_new_event(mddev);
8110
8111 if (last_check + window > io_sectors || j == max_sectors)
8112 continue;
8113
8114 last_check = io_sectors;
8115 repeat:
8116 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8117 /* step marks */
8118 int next = (last_mark+1) % SYNC_MARKS;
8119
8120 mddev->resync_mark = mark[next];
8121 mddev->resync_mark_cnt = mark_cnt[next];
8122 mark[next] = jiffies;
8123 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
8124 last_mark = next;
8125 }
8126
8127 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8128 break;
8129
8130 /*
8131 * this loop exits only if either when we are slower than
8132 * the 'hard' speed limit, or the system was IO-idle for
8133 * a jiffy.
8134 * the system might be non-idle CPU-wise, but we only care
8135 * about not overloading the IO subsystem. (things like an
8136 * e2fsck being done on the RAID array should execute fast)
8137 */
8138 cond_resched();
8139
8140 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8141 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
8142 /((jiffies-mddev->resync_mark)/HZ +1) +1;
8143
8144 if (currspeed > speed_min(mddev)) {
8145 if (currspeed > speed_max(mddev)) {
8146 msleep(500);
8147 goto repeat;
8148 }
8149 if (!is_mddev_idle(mddev, 0)) {
8150 /*
8151 * Give other IO more of a chance.
8152 * The faster the devices, the less we wait.
8153 */
8154 wait_event(mddev->recovery_wait,
8155 !atomic_read(&mddev->recovery_active));
8156 }
8157 }
8158 }
8159 pr_info("md: %s: %s %s.\n",mdname(mddev), desc,
8160 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
8161 ? "interrupted" : "done");
8162 /*
8163 * this also signals 'finished resyncing' to md_stop
8164 */
8165 blk_finish_plug(&plug);
8166 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
8167
8168 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8169 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8170 mddev->curr_resync > 3) {
8171 mddev->curr_resync_completed = mddev->curr_resync;
8172 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8173 }
8174 mddev->pers->sync_request(mddev, max_sectors, &skipped);
8175
8176 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
8177 mddev->curr_resync > 3) {
8178 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8179 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8180 if (mddev->curr_resync >= mddev->recovery_cp) {
8181 pr_debug("md: checkpointing %s of %s.\n",
8182 desc, mdname(mddev));
8183 if (test_bit(MD_RECOVERY_ERROR,
8184 &mddev->recovery))
8185 mddev->recovery_cp =
8186 mddev->curr_resync_completed;
8187 else
8188 mddev->recovery_cp =
8189 mddev->curr_resync;
8190 }
8191 } else
8192 mddev->recovery_cp = MaxSector;
8193 } else {
8194 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8195 mddev->curr_resync = MaxSector;
8196 rcu_read_lock();
8197 rdev_for_each_rcu(rdev, mddev)
8198 if (rdev->raid_disk >= 0 &&
8199 mddev->delta_disks >= 0 &&
8200 !test_bit(Journal, &rdev->flags) &&
8201 !test_bit(Faulty, &rdev->flags) &&
8202 !test_bit(In_sync, &rdev->flags) &&
8203 rdev->recovery_offset < mddev->curr_resync)
8204 rdev->recovery_offset = mddev->curr_resync;
8205 rcu_read_unlock();
8206 }
8207 }
8208 skip:
8209 /* set CHANGE_PENDING here since maybe another update is needed,
8210 * so other nodes are informed. It should be harmless for normal
8211 * raid */
8212 set_mask_bits(&mddev->sb_flags, 0,
8213 BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS));
8214
8215 spin_lock(&mddev->lock);
8216 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8217 /* We completed so min/max setting can be forgotten if used. */
8218 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8219 mddev->resync_min = 0;
8220 mddev->resync_max = MaxSector;
8221 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8222 mddev->resync_min = mddev->curr_resync_completed;
8223 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
8224 mddev->curr_resync = 0;
8225 spin_unlock(&mddev->lock);
8226
8227 wake_up(&resync_wait);
8228 md_wakeup_thread(mddev->thread);
8229 return;
8230 }
8231 EXPORT_SYMBOL_GPL(md_do_sync);
8232
8233 static int remove_and_add_spares(struct mddev *mddev,
8234 struct md_rdev *this)
8235 {
8236 struct md_rdev *rdev;
8237 int spares = 0;
8238 int removed = 0;
8239 bool remove_some = false;
8240
8241 rdev_for_each(rdev, mddev) {
8242 if ((this == NULL || rdev == this) &&
8243 rdev->raid_disk >= 0 &&
8244 !test_bit(Blocked, &rdev->flags) &&
8245 test_bit(Faulty, &rdev->flags) &&
8246 atomic_read(&rdev->nr_pending)==0) {
8247 /* Faulty non-Blocked devices with nr_pending == 0
8248 * never get nr_pending incremented,
8249 * never get Faulty cleared, and never get Blocked set.
8250 * So we can synchronize_rcu now rather than once per device
8251 */
8252 remove_some = true;
8253 set_bit(RemoveSynchronized, &rdev->flags);
8254 }
8255 }
8256
8257 if (remove_some)
8258 synchronize_rcu();
8259 rdev_for_each(rdev, mddev) {
8260 if ((this == NULL || rdev == this) &&
8261 rdev->raid_disk >= 0 &&
8262 !test_bit(Blocked, &rdev->flags) &&
8263 ((test_bit(RemoveSynchronized, &rdev->flags) ||
8264 (!test_bit(In_sync, &rdev->flags) &&
8265 !test_bit(Journal, &rdev->flags))) &&
8266 atomic_read(&rdev->nr_pending)==0)) {
8267 if (mddev->pers->hot_remove_disk(
8268 mddev, rdev) == 0) {
8269 sysfs_unlink_rdev(mddev, rdev);
8270 rdev->raid_disk = -1;
8271 removed++;
8272 }
8273 }
8274 if (remove_some && test_bit(RemoveSynchronized, &rdev->flags))
8275 clear_bit(RemoveSynchronized, &rdev->flags);
8276 }
8277
8278 if (removed && mddev->kobj.sd)
8279 sysfs_notify(&mddev->kobj, NULL, "degraded");
8280
8281 if (this && removed)
8282 goto no_add;
8283
8284 rdev_for_each(rdev, mddev) {
8285 if (this && this != rdev)
8286 continue;
8287 if (test_bit(Candidate, &rdev->flags))
8288 continue;
8289 if (rdev->raid_disk >= 0 &&
8290 !test_bit(In_sync, &rdev->flags) &&
8291 !test_bit(Journal, &rdev->flags) &&
8292 !test_bit(Faulty, &rdev->flags))
8293 spares++;
8294 if (rdev->raid_disk >= 0)
8295 continue;
8296 if (test_bit(Faulty, &rdev->flags))
8297 continue;
8298 if (!test_bit(Journal, &rdev->flags)) {
8299 if (mddev->ro &&
8300 ! (rdev->saved_raid_disk >= 0 &&
8301 !test_bit(Bitmap_sync, &rdev->flags)))
8302 continue;
8303
8304 rdev->recovery_offset = 0;
8305 }
8306 if (mddev->pers->
8307 hot_add_disk(mddev, rdev) == 0) {
8308 if (sysfs_link_rdev(mddev, rdev))
8309 /* failure here is OK */;
8310 if (!test_bit(Journal, &rdev->flags))
8311 spares++;
8312 md_new_event(mddev);
8313 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
8314 }
8315 }
8316 no_add:
8317 if (removed)
8318 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
8319 return spares;
8320 }
8321
8322 static void md_start_sync(struct work_struct *ws)
8323 {
8324 struct mddev *mddev = container_of(ws, struct mddev, del_work);
8325
8326 mddev->sync_thread = md_register_thread(md_do_sync,
8327 mddev,
8328 "resync");
8329 if (!mddev->sync_thread) {
8330 pr_warn("%s: could not start resync thread...\n",
8331 mdname(mddev));
8332 /* leave the spares where they are, it shouldn't hurt */
8333 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8334 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8335 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8336 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8337 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8338 wake_up(&resync_wait);
8339 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8340 &mddev->recovery))
8341 if (mddev->sysfs_action)
8342 sysfs_notify_dirent_safe(mddev->sysfs_action);
8343 } else
8344 md_wakeup_thread(mddev->sync_thread);
8345 sysfs_notify_dirent_safe(mddev->sysfs_action);
8346 md_new_event(mddev);
8347 }
8348
8349 /*
8350 * This routine is regularly called by all per-raid-array threads to
8351 * deal with generic issues like resync and super-block update.
8352 * Raid personalities that don't have a thread (linear/raid0) do not
8353 * need this as they never do any recovery or update the superblock.
8354 *
8355 * It does not do any resync itself, but rather "forks" off other threads
8356 * to do that as needed.
8357 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8358 * "->recovery" and create a thread at ->sync_thread.
8359 * When the thread finishes it sets MD_RECOVERY_DONE
8360 * and wakeups up this thread which will reap the thread and finish up.
8361 * This thread also removes any faulty devices (with nr_pending == 0).
8362 *
8363 * The overall approach is:
8364 * 1/ if the superblock needs updating, update it.
8365 * 2/ If a recovery thread is running, don't do anything else.
8366 * 3/ If recovery has finished, clean up, possibly marking spares active.
8367 * 4/ If there are any faulty devices, remove them.
8368 * 5/ If array is degraded, try to add spares devices
8369 * 6/ If array has spares or is not in-sync, start a resync thread.
8370 */
8371 void md_check_recovery(struct mddev *mddev)
8372 {
8373 if (mddev->suspended)
8374 return;
8375
8376 if (mddev->bitmap)
8377 bitmap_daemon_work(mddev);
8378
8379 if (signal_pending(current)) {
8380 if (mddev->pers->sync_request && !mddev->external) {
8381 pr_debug("md: %s in immediate safe mode\n",
8382 mdname(mddev));
8383 mddev->safemode = 2;
8384 }
8385 flush_signals(current);
8386 }
8387
8388 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
8389 return;
8390 if ( ! (
8391 (mddev->sb_flags & ~ (1<<MD_SB_CHANGE_PENDING)) ||
8392 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8393 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8394 test_bit(MD_RELOAD_SB, &mddev->flags) ||
8395 (mddev->external == 0 && mddev->safemode == 1) ||
8396 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
8397 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
8398 ))
8399 return;
8400
8401 if (mddev_trylock(mddev)) {
8402 int spares = 0;
8403
8404 if (mddev->ro) {
8405 struct md_rdev *rdev;
8406 if (!mddev->external && mddev->in_sync)
8407 /* 'Blocked' flag not needed as failed devices
8408 * will be recorded if array switched to read/write.
8409 * Leaving it set will prevent the device
8410 * from being removed.
8411 */
8412 rdev_for_each(rdev, mddev)
8413 clear_bit(Blocked, &rdev->flags);
8414 /* On a read-only array we can:
8415 * - remove failed devices
8416 * - add already-in_sync devices if the array itself
8417 * is in-sync.
8418 * As we only add devices that are already in-sync,
8419 * we can activate the spares immediately.
8420 */
8421 remove_and_add_spares(mddev, NULL);
8422 /* There is no thread, but we need to call
8423 * ->spare_active and clear saved_raid_disk
8424 */
8425 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8426 md_reap_sync_thread(mddev);
8427 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8428 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8429 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8430 goto unlock;
8431 }
8432
8433 if (mddev_is_clustered(mddev)) {
8434 struct md_rdev *rdev;
8435 /* kick the device if another node issued a
8436 * remove disk.
8437 */
8438 rdev_for_each(rdev, mddev) {
8439 if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
8440 rdev->raid_disk < 0)
8441 md_kick_rdev_from_array(rdev);
8442 }
8443
8444 if (test_and_clear_bit(MD_RELOAD_SB, &mddev->flags))
8445 md_reload_sb(mddev, mddev->good_device_nr);
8446 }
8447
8448 if (!mddev->external) {
8449 int did_change = 0;
8450 spin_lock(&mddev->lock);
8451 if (mddev->safemode &&
8452 !atomic_read(&mddev->writes_pending) &&
8453 !mddev->in_sync &&
8454 mddev->recovery_cp == MaxSector) {
8455 mddev->in_sync = 1;
8456 did_change = 1;
8457 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8458 }
8459 if (mddev->safemode == 1)
8460 mddev->safemode = 0;
8461 spin_unlock(&mddev->lock);
8462 if (did_change)
8463 sysfs_notify_dirent_safe(mddev->sysfs_state);
8464 }
8465
8466 if (mddev->sb_flags)
8467 md_update_sb(mddev, 0);
8468
8469 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
8470 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
8471 /* resync/recovery still happening */
8472 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8473 goto unlock;
8474 }
8475 if (mddev->sync_thread) {
8476 md_reap_sync_thread(mddev);
8477 goto unlock;
8478 }
8479 /* Set RUNNING before clearing NEEDED to avoid
8480 * any transients in the value of "sync_action".
8481 */
8482 mddev->curr_resync_completed = 0;
8483 spin_lock(&mddev->lock);
8484 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8485 spin_unlock(&mddev->lock);
8486 /* Clear some bits that don't mean anything, but
8487 * might be left set
8488 */
8489 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
8490 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8491
8492 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8493 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
8494 goto not_running;
8495 /* no recovery is running.
8496 * remove any failed drives, then
8497 * add spares if possible.
8498 * Spares are also removed and re-added, to allow
8499 * the personality to fail the re-add.
8500 */
8501
8502 if (mddev->reshape_position != MaxSector) {
8503 if (mddev->pers->check_reshape == NULL ||
8504 mddev->pers->check_reshape(mddev) != 0)
8505 /* Cannot proceed */
8506 goto not_running;
8507 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8508 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8509 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
8510 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8511 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8512 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8513 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8514 } else if (mddev->recovery_cp < MaxSector) {
8515 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8516 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8517 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
8518 /* nothing to be done ... */
8519 goto not_running;
8520
8521 if (mddev->pers->sync_request) {
8522 if (spares) {
8523 /* We are adding a device or devices to an array
8524 * which has the bitmap stored on all devices.
8525 * So make sure all bitmap pages get written
8526 */
8527 bitmap_write_all(mddev->bitmap);
8528 }
8529 INIT_WORK(&mddev->del_work, md_start_sync);
8530 queue_work(md_misc_wq, &mddev->del_work);
8531 goto unlock;
8532 }
8533 not_running:
8534 if (!mddev->sync_thread) {
8535 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8536 wake_up(&resync_wait);
8537 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8538 &mddev->recovery))
8539 if (mddev->sysfs_action)
8540 sysfs_notify_dirent_safe(mddev->sysfs_action);
8541 }
8542 unlock:
8543 wake_up(&mddev->sb_wait);
8544 mddev_unlock(mddev);
8545 }
8546 }
8547 EXPORT_SYMBOL(md_check_recovery);
8548
8549 void md_reap_sync_thread(struct mddev *mddev)
8550 {
8551 struct md_rdev *rdev;
8552
8553 /* resync has finished, collect result */
8554 md_unregister_thread(&mddev->sync_thread);
8555 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8556 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8557 /* success...*/
8558 /* activate any spares */
8559 if (mddev->pers->spare_active(mddev)) {
8560 sysfs_notify(&mddev->kobj, NULL,
8561 "degraded");
8562 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
8563 }
8564 }
8565 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8566 mddev->pers->finish_reshape)
8567 mddev->pers->finish_reshape(mddev);
8568
8569 /* If array is no-longer degraded, then any saved_raid_disk
8570 * information must be scrapped.
8571 */
8572 if (!mddev->degraded)
8573 rdev_for_each(rdev, mddev)
8574 rdev->saved_raid_disk = -1;
8575
8576 md_update_sb(mddev, 1);
8577 /* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can
8578 * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by
8579 * clustered raid */
8580 if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags))
8581 md_cluster_ops->resync_finish(mddev);
8582 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8583 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8584 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8585 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8586 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8587 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8588 wake_up(&resync_wait);
8589 /* flag recovery needed just to double check */
8590 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8591 sysfs_notify_dirent_safe(mddev->sysfs_action);
8592 md_new_event(mddev);
8593 if (mddev->event_work.func)
8594 queue_work(md_misc_wq, &mddev->event_work);
8595 }
8596 EXPORT_SYMBOL(md_reap_sync_thread);
8597
8598 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
8599 {
8600 sysfs_notify_dirent_safe(rdev->sysfs_state);
8601 wait_event_timeout(rdev->blocked_wait,
8602 !test_bit(Blocked, &rdev->flags) &&
8603 !test_bit(BlockedBadBlocks, &rdev->flags),
8604 msecs_to_jiffies(5000));
8605 rdev_dec_pending(rdev, mddev);
8606 }
8607 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8608
8609 void md_finish_reshape(struct mddev *mddev)
8610 {
8611 /* called be personality module when reshape completes. */
8612 struct md_rdev *rdev;
8613
8614 rdev_for_each(rdev, mddev) {
8615 if (rdev->data_offset > rdev->new_data_offset)
8616 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8617 else
8618 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8619 rdev->data_offset = rdev->new_data_offset;
8620 }
8621 }
8622 EXPORT_SYMBOL(md_finish_reshape);
8623
8624 /* Bad block management */
8625
8626 /* Returns 1 on success, 0 on failure */
8627 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8628 int is_new)
8629 {
8630 struct mddev *mddev = rdev->mddev;
8631 int rv;
8632 if (is_new)
8633 s += rdev->new_data_offset;
8634 else
8635 s += rdev->data_offset;
8636 rv = badblocks_set(&rdev->badblocks, s, sectors, 0);
8637 if (rv == 0) {
8638 /* Make sure they get written out promptly */
8639 if (test_bit(ExternalBbl, &rdev->flags))
8640 sysfs_notify(&rdev->kobj, NULL,
8641 "unacknowledged_bad_blocks");
8642 sysfs_notify_dirent_safe(rdev->sysfs_state);
8643 set_mask_bits(&mddev->sb_flags, 0,
8644 BIT(MD_SB_CHANGE_CLEAN) | BIT(MD_SB_CHANGE_PENDING));
8645 md_wakeup_thread(rdev->mddev->thread);
8646 return 1;
8647 } else
8648 return 0;
8649 }
8650 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8651
8652 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8653 int is_new)
8654 {
8655 int rv;
8656 if (is_new)
8657 s += rdev->new_data_offset;
8658 else
8659 s += rdev->data_offset;
8660 rv = badblocks_clear(&rdev->badblocks, s, sectors);
8661 if ((rv == 0) && test_bit(ExternalBbl, &rdev->flags))
8662 sysfs_notify(&rdev->kobj, NULL, "bad_blocks");
8663 return rv;
8664 }
8665 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8666
8667 static int md_notify_reboot(struct notifier_block *this,
8668 unsigned long code, void *x)
8669 {
8670 struct list_head *tmp;
8671 struct mddev *mddev;
8672 int need_delay = 0;
8673
8674 for_each_mddev(mddev, tmp) {
8675 if (mddev_trylock(mddev)) {
8676 if (mddev->pers)
8677 __md_stop_writes(mddev);
8678 if (mddev->persistent)
8679 mddev->safemode = 2;
8680 mddev_unlock(mddev);
8681 }
8682 need_delay = 1;
8683 }
8684 /*
8685 * certain more exotic SCSI devices are known to be
8686 * volatile wrt too early system reboots. While the
8687 * right place to handle this issue is the given
8688 * driver, we do want to have a safe RAID driver ...
8689 */
8690 if (need_delay)
8691 mdelay(1000*1);
8692
8693 return NOTIFY_DONE;
8694 }
8695
8696 static struct notifier_block md_notifier = {
8697 .notifier_call = md_notify_reboot,
8698 .next = NULL,
8699 .priority = INT_MAX, /* before any real devices */
8700 };
8701
8702 static void md_geninit(void)
8703 {
8704 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
8705
8706 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
8707 }
8708
8709 static int __init md_init(void)
8710 {
8711 int ret = -ENOMEM;
8712
8713 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
8714 if (!md_wq)
8715 goto err_wq;
8716
8717 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8718 if (!md_misc_wq)
8719 goto err_misc_wq;
8720
8721 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8722 goto err_md;
8723
8724 if ((ret = register_blkdev(0, "mdp")) < 0)
8725 goto err_mdp;
8726 mdp_major = ret;
8727
8728 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
8729 md_probe, NULL, NULL);
8730 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
8731 md_probe, NULL, NULL);
8732
8733 register_reboot_notifier(&md_notifier);
8734 raid_table_header = register_sysctl_table(raid_root_table);
8735
8736 md_geninit();
8737 return 0;
8738
8739 err_mdp:
8740 unregister_blkdev(MD_MAJOR, "md");
8741 err_md:
8742 destroy_workqueue(md_misc_wq);
8743 err_misc_wq:
8744 destroy_workqueue(md_wq);
8745 err_wq:
8746 return ret;
8747 }
8748
8749 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
8750 {
8751 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
8752 struct md_rdev *rdev2;
8753 int role, ret;
8754 char b[BDEVNAME_SIZE];
8755
8756 /* Check for change of roles in the active devices */
8757 rdev_for_each(rdev2, mddev) {
8758 if (test_bit(Faulty, &rdev2->flags))
8759 continue;
8760
8761 /* Check if the roles changed */
8762 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
8763
8764 if (test_bit(Candidate, &rdev2->flags)) {
8765 if (role == 0xfffe) {
8766 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b));
8767 md_kick_rdev_from_array(rdev2);
8768 continue;
8769 }
8770 else
8771 clear_bit(Candidate, &rdev2->flags);
8772 }
8773
8774 if (role != rdev2->raid_disk) {
8775 /* got activated */
8776 if (rdev2->raid_disk == -1 && role != 0xffff) {
8777 rdev2->saved_raid_disk = role;
8778 ret = remove_and_add_spares(mddev, rdev2);
8779 pr_info("Activated spare: %s\n",
8780 bdevname(rdev2->bdev,b));
8781 /* wakeup mddev->thread here, so array could
8782 * perform resync with the new activated disk */
8783 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8784 md_wakeup_thread(mddev->thread);
8785
8786 }
8787 /* device faulty
8788 * We just want to do the minimum to mark the disk
8789 * as faulty. The recovery is performed by the
8790 * one who initiated the error.
8791 */
8792 if ((role == 0xfffe) || (role == 0xfffd)) {
8793 md_error(mddev, rdev2);
8794 clear_bit(Blocked, &rdev2->flags);
8795 }
8796 }
8797 }
8798
8799 if (mddev->raid_disks != le32_to_cpu(sb->raid_disks))
8800 update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
8801
8802 /* Finally set the event to be up to date */
8803 mddev->events = le64_to_cpu(sb->events);
8804 }
8805
8806 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
8807 {
8808 int err;
8809 struct page *swapout = rdev->sb_page;
8810 struct mdp_superblock_1 *sb;
8811
8812 /* Store the sb page of the rdev in the swapout temporary
8813 * variable in case we err in the future
8814 */
8815 rdev->sb_page = NULL;
8816 err = alloc_disk_sb(rdev);
8817 if (err == 0) {
8818 ClearPageUptodate(rdev->sb_page);
8819 rdev->sb_loaded = 0;
8820 err = super_types[mddev->major_version].
8821 load_super(rdev, NULL, mddev->minor_version);
8822 }
8823 if (err < 0) {
8824 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
8825 __func__, __LINE__, rdev->desc_nr, err);
8826 if (rdev->sb_page)
8827 put_page(rdev->sb_page);
8828 rdev->sb_page = swapout;
8829 rdev->sb_loaded = 1;
8830 return err;
8831 }
8832
8833 sb = page_address(rdev->sb_page);
8834 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
8835 * is not set
8836 */
8837
8838 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
8839 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
8840
8841 /* The other node finished recovery, call spare_active to set
8842 * device In_sync and mddev->degraded
8843 */
8844 if (rdev->recovery_offset == MaxSector &&
8845 !test_bit(In_sync, &rdev->flags) &&
8846 mddev->pers->spare_active(mddev))
8847 sysfs_notify(&mddev->kobj, NULL, "degraded");
8848
8849 put_page(swapout);
8850 return 0;
8851 }
8852
8853 void md_reload_sb(struct mddev *mddev, int nr)
8854 {
8855 struct md_rdev *rdev;
8856 int err;
8857
8858 /* Find the rdev */
8859 rdev_for_each_rcu(rdev, mddev) {
8860 if (rdev->desc_nr == nr)
8861 break;
8862 }
8863
8864 if (!rdev || rdev->desc_nr != nr) {
8865 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
8866 return;
8867 }
8868
8869 err = read_rdev(mddev, rdev);
8870 if (err < 0)
8871 return;
8872
8873 check_sb_changes(mddev, rdev);
8874
8875 /* Read all rdev's to update recovery_offset */
8876 rdev_for_each_rcu(rdev, mddev)
8877 read_rdev(mddev, rdev);
8878 }
8879 EXPORT_SYMBOL(md_reload_sb);
8880
8881 #ifndef MODULE
8882
8883 /*
8884 * Searches all registered partitions for autorun RAID arrays
8885 * at boot time.
8886 */
8887
8888 static DEFINE_MUTEX(detected_devices_mutex);
8889 static LIST_HEAD(all_detected_devices);
8890 struct detected_devices_node {
8891 struct list_head list;
8892 dev_t dev;
8893 };
8894
8895 void md_autodetect_dev(dev_t dev)
8896 {
8897 struct detected_devices_node *node_detected_dev;
8898
8899 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8900 if (node_detected_dev) {
8901 node_detected_dev->dev = dev;
8902 mutex_lock(&detected_devices_mutex);
8903 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8904 mutex_unlock(&detected_devices_mutex);
8905 }
8906 }
8907
8908 static void autostart_arrays(int part)
8909 {
8910 struct md_rdev *rdev;
8911 struct detected_devices_node *node_detected_dev;
8912 dev_t dev;
8913 int i_scanned, i_passed;
8914
8915 i_scanned = 0;
8916 i_passed = 0;
8917
8918 pr_info("md: Autodetecting RAID arrays.\n");
8919
8920 mutex_lock(&detected_devices_mutex);
8921 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8922 i_scanned++;
8923 node_detected_dev = list_entry(all_detected_devices.next,
8924 struct detected_devices_node, list);
8925 list_del(&node_detected_dev->list);
8926 dev = node_detected_dev->dev;
8927 kfree(node_detected_dev);
8928 mutex_unlock(&detected_devices_mutex);
8929 rdev = md_import_device(dev,0, 90);
8930 mutex_lock(&detected_devices_mutex);
8931 if (IS_ERR(rdev))
8932 continue;
8933
8934 if (test_bit(Faulty, &rdev->flags))
8935 continue;
8936
8937 set_bit(AutoDetected, &rdev->flags);
8938 list_add(&rdev->same_set, &pending_raid_disks);
8939 i_passed++;
8940 }
8941 mutex_unlock(&detected_devices_mutex);
8942
8943 pr_debug("md: Scanned %d and added %d devices.\n", i_scanned, i_passed);
8944
8945 autorun_devices(part);
8946 }
8947
8948 #endif /* !MODULE */
8949
8950 static __exit void md_exit(void)
8951 {
8952 struct mddev *mddev;
8953 struct list_head *tmp;
8954 int delay = 1;
8955
8956 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
8957 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
8958
8959 unregister_blkdev(MD_MAJOR,"md");
8960 unregister_blkdev(mdp_major, "mdp");
8961 unregister_reboot_notifier(&md_notifier);
8962 unregister_sysctl_table(raid_table_header);
8963
8964 /* We cannot unload the modules while some process is
8965 * waiting for us in select() or poll() - wake them up
8966 */
8967 md_unloading = 1;
8968 while (waitqueue_active(&md_event_waiters)) {
8969 /* not safe to leave yet */
8970 wake_up(&md_event_waiters);
8971 msleep(delay);
8972 delay += delay;
8973 }
8974 remove_proc_entry("mdstat", NULL);
8975
8976 for_each_mddev(mddev, tmp) {
8977 export_array(mddev);
8978 mddev->hold_active = 0;
8979 }
8980 destroy_workqueue(md_misc_wq);
8981 destroy_workqueue(md_wq);
8982 }
8983
8984 subsys_initcall(md_init);
8985 module_exit(md_exit)
8986
8987 static int get_ro(char *buffer, struct kernel_param *kp)
8988 {
8989 return sprintf(buffer, "%d", start_readonly);
8990 }
8991 static int set_ro(const char *val, struct kernel_param *kp)
8992 {
8993 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
8994 }
8995
8996 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8997 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
8998 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
8999
9000 MODULE_LICENSE("GPL");
9001 MODULE_DESCRIPTION("MD RAID framework");
9002 MODULE_ALIAS("md");
9003 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);