?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/evalenti/linux...
[linux/fpc-iii.git] / drivers / md / md.c
blob194580fba7fd8b7a965ea87f9856cada61e15be6
1 /*
2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
4
5 completely rewritten, based on the MD driver code from Marc Zyngier
6
7 Changes:
8
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
19
20 Neil Brown <neilb@cse.unsw.edu.au>.
21
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
28 any later version.
29
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33 */
34
35 #include <linux/kthread.h>
36 #include <linux/blkdev.h>
37 #include <linux/badblocks.h>
38 #include <linux/sysctl.h>
39 #include <linux/seq_file.h>
40 #include <linux/fs.h>
41 #include <linux/poll.h>
42 #include <linux/ctype.h>
43 #include <linux/string.h>
44 #include <linux/hdreg.h>
45 #include <linux/proc_fs.h>
46 #include <linux/random.h>
47 #include <linux/module.h>
48 #include <linux/reboot.h>
49 #include <linux/file.h>
50 #include <linux/compat.h>
51 #include <linux/delay.h>
52 #include <linux/raid/md_p.h>
53 #include <linux/raid/md_u.h>
54 #include <linux/slab.h>
55 #include "md.h"
56 #include "bitmap.h"
57 #include "md-cluster.h"
58
59 #ifndef MODULE
60 static void autostart_arrays(int part);
61 #endif
62
63 /* pers_list is a list of registered personalities protected
64 * by pers_lock.
65 * pers_lock does extra service to protect accesses to
66 * mddev->thread when the mutex cannot be held.
67 */
68 static LIST_HEAD(pers_list);
69 static DEFINE_SPINLOCK(pers_lock);
70
71 struct md_cluster_operations *md_cluster_ops;
72 EXPORT_SYMBOL(md_cluster_ops);
73 struct module *md_cluster_mod;
74 EXPORT_SYMBOL(md_cluster_mod);
75
76 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
77 static struct workqueue_struct *md_wq;
78 static struct workqueue_struct *md_misc_wq;
79
80 static int remove_and_add_spares(struct mddev *mddev,
81 struct md_rdev *this);
82 static void mddev_detach(struct mddev *mddev);
83
84 /*
85 * Default number of read corrections we'll attempt on an rdev
86 * before ejecting it from the array. We divide the read error
87 * count by 2 for every hour elapsed between read errors.
88 */
89 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
90 /*
91 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
92 * is 1000 KB/sec, so the extra system load does not show up that much.
93 * Increase it if you want to have more _guaranteed_ speed. Note that
94 * the RAID driver will use the maximum available bandwidth if the IO
95 * subsystem is idle. There is also an 'absolute maximum' reconstruction
96 * speed limit - in case reconstruction slows down your system despite
97 * idle IO detection.
98 *
99 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
100 * or /sys/block/mdX/md/sync_speed_{min,max}
101 */
102
103 static int sysctl_speed_limit_min = 1000;
104 static int sysctl_speed_limit_max = 200000;
105 static inline int speed_min(struct mddev *mddev)
106 {
107 return mddev->sync_speed_min ?
108 mddev->sync_speed_min : sysctl_speed_limit_min;
109 }
110
111 static inline int speed_max(struct mddev *mddev)
112 {
113 return mddev->sync_speed_max ?
114 mddev->sync_speed_max : sysctl_speed_limit_max;
115 }
116
117 static struct ctl_table_header *raid_table_header;
118
119 static struct ctl_table raid_table[] = {
120 {
121 .procname = "speed_limit_min",
122 .data = &sysctl_speed_limit_min,
123 .maxlen = sizeof(int),
124 .mode = S_IRUGO|S_IWUSR,
125 .proc_handler = proc_dointvec,
126 },
127 {
128 .procname = "speed_limit_max",
129 .data = &sysctl_speed_limit_max,
130 .maxlen = sizeof(int),
131 .mode = S_IRUGO|S_IWUSR,
132 .proc_handler = proc_dointvec,
133 },
134 { }
135 };
136
137 static struct ctl_table raid_dir_table[] = {
138 {
139 .procname = "raid",
140 .maxlen = 0,
141 .mode = S_IRUGO|S_IXUGO,
142 .child = raid_table,
143 },
144 { }
145 };
146
147 static struct ctl_table raid_root_table[] = {
148 {
149 .procname = "dev",
150 .maxlen = 0,
151 .mode = 0555,
152 .child = raid_dir_table,
153 },
154 { }
155 };
156
157 static const struct block_device_operations md_fops;
158
159 static int start_readonly;
160
161 /* bio_clone_mddev
162 * like bio_clone, but with a local bio set
163 */
164
165 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
166 struct mddev *mddev)
167 {
168 struct bio *b;
169
170 if (!mddev || !mddev->bio_set)
171 return bio_alloc(gfp_mask, nr_iovecs);
172
173 b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set);
174 if (!b)
175 return NULL;
176 return b;
177 }
178 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
179
180 struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
181 struct mddev *mddev)
182 {
183 if (!mddev || !mddev->bio_set)
184 return bio_clone(bio, gfp_mask);
185
186 return bio_clone_bioset(bio, gfp_mask, mddev->bio_set);
187 }
188 EXPORT_SYMBOL_GPL(bio_clone_mddev);
189
190 /*
191 * We have a system wide 'event count' that is incremented
192 * on any 'interesting' event, and readers of /proc/mdstat
193 * can use 'poll' or 'select' to find out when the event
194 * count increases.
195 *
196 * Events are:
197 * start array, stop array, error, add device, remove device,
198 * start build, activate spare
199 */
200 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
201 static atomic_t md_event_count;
202 void md_new_event(struct mddev *mddev)
203 {
204 atomic_inc(&md_event_count);
205 wake_up(&md_event_waiters);
206 }
207 EXPORT_SYMBOL_GPL(md_new_event);
208
209 /*
210 * Enables to iterate over all existing md arrays
211 * all_mddevs_lock protects this list.
212 */
213 static LIST_HEAD(all_mddevs);
214 static DEFINE_SPINLOCK(all_mddevs_lock);
215
216 /*
217 * iterates through all used mddevs in the system.
218 * We take care to grab the all_mddevs_lock whenever navigating
219 * the list, and to always hold a refcount when unlocked.
220 * Any code which breaks out of this loop while own
221 * a reference to the current mddev and must mddev_put it.
222 */
223 #define for_each_mddev(_mddev,_tmp) \
224 \
225 for (({ spin_lock(&all_mddevs_lock); \
226 _tmp = all_mddevs.next; \
227 _mddev = NULL;}); \
228 ({ if (_tmp != &all_mddevs) \
229 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
230 spin_unlock(&all_mddevs_lock); \
231 if (_mddev) mddev_put(_mddev); \
232 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
233 _tmp != &all_mddevs;}); \
234 ({ spin_lock(&all_mddevs_lock); \
235 _tmp = _tmp->next;}) \
236 )
237
238 /* Rather than calling directly into the personality make_request function,
239 * IO requests come here first so that we can check if the device is
240 * being suspended pending a reconfiguration.
241 * We hold a refcount over the call to ->make_request. By the time that
242 * call has finished, the bio has been linked into some internal structure
243 * and so is visible to ->quiesce(), so we don't need the refcount any more.
244 */
245 static blk_qc_t md_make_request(struct request_queue *q, struct bio *bio)
246 {
247 const int rw = bio_data_dir(bio);
248 struct mddev *mddev = q->queuedata;
249 unsigned int sectors;
250 int cpu;
251
252 blk_queue_split(q, &bio, q->bio_split);
253
254 if (mddev == NULL || mddev->pers == NULL) {
255 bio_io_error(bio);
256 return BLK_QC_T_NONE;
257 }
258 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
259 if (bio_sectors(bio) != 0)
260 bio->bi_error = -EROFS;
261 bio_endio(bio);
262 return BLK_QC_T_NONE;
263 }
264 smp_rmb(); /* Ensure implications of 'active' are visible */
265 rcu_read_lock();
266 if (mddev->suspended) {
267 DEFINE_WAIT(__wait);
268 for (;;) {
269 prepare_to_wait(&mddev->sb_wait, &__wait,
270 TASK_UNINTERRUPTIBLE);
271 if (!mddev->suspended)
272 break;
273 rcu_read_unlock();
274 schedule();
275 rcu_read_lock();
276 }
277 finish_wait(&mddev->sb_wait, &__wait);
278 }
279 atomic_inc(&mddev->active_io);
280 rcu_read_unlock();
281
282 /*
283 * save the sectors now since our bio can
284 * go away inside make_request
285 */
286 sectors = bio_sectors(bio);
287 mddev->pers->make_request(mddev, bio);
288
289 cpu = part_stat_lock();
290 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
291 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
292 part_stat_unlock();
293
294 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
295 wake_up(&mddev->sb_wait);
296
297 return BLK_QC_T_NONE;
298 }
299
300 /* mddev_suspend makes sure no new requests are submitted
301 * to the device, and that any requests that have been submitted
302 * are completely handled.
303 * Once mddev_detach() is called and completes, the module will be
304 * completely unused.
305 */
306 void mddev_suspend(struct mddev *mddev)
307 {
308 WARN_ON_ONCE(current == mddev->thread->tsk);
309 if (mddev->suspended++)
310 return;
311 synchronize_rcu();
312 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
313 mddev->pers->quiesce(mddev, 1);
314
315 del_timer_sync(&mddev->safemode_timer);
316 }
317 EXPORT_SYMBOL_GPL(mddev_suspend);
318
319 void mddev_resume(struct mddev *mddev)
320 {
321 if (--mddev->suspended)
322 return;
323 wake_up(&mddev->sb_wait);
324 mddev->pers->quiesce(mddev, 0);
325
326 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
327 md_wakeup_thread(mddev->thread);
328 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
329 }
330 EXPORT_SYMBOL_GPL(mddev_resume);
331
332 int mddev_congested(struct mddev *mddev, int bits)
333 {
334 struct md_personality *pers = mddev->pers;
335 int ret = 0;
336
337 rcu_read_lock();
338 if (mddev->suspended)
339 ret = 1;
340 else if (pers && pers->congested)
341 ret = pers->congested(mddev, bits);
342 rcu_read_unlock();
343 return ret;
344 }
345 EXPORT_SYMBOL_GPL(mddev_congested);
346 static int md_congested(void *data, int bits)
347 {
348 struct mddev *mddev = data;
349 return mddev_congested(mddev, bits);
350 }
351
352 /*
353 * Generic flush handling for md
354 */
355
356 static void md_end_flush(struct bio *bio)
357 {
358 struct md_rdev *rdev = bio->bi_private;
359 struct mddev *mddev = rdev->mddev;
360
361 rdev_dec_pending(rdev, mddev);
362
363 if (atomic_dec_and_test(&mddev->flush_pending)) {
364 /* The pre-request flush has finished */
365 queue_work(md_wq, &mddev->flush_work);
366 }
367 bio_put(bio);
368 }
369
370 static void md_submit_flush_data(struct work_struct *ws);
371
372 static void submit_flushes(struct work_struct *ws)
373 {
374 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
375 struct md_rdev *rdev;
376
377 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
378 atomic_set(&mddev->flush_pending, 1);
379 rcu_read_lock();
380 rdev_for_each_rcu(rdev, mddev)
381 if (rdev->raid_disk >= 0 &&
382 !test_bit(Faulty, &rdev->flags)) {
383 /* Take two references, one is dropped
384 * when request finishes, one after
385 * we reclaim rcu_read_lock
386 */
387 struct bio *bi;
388 atomic_inc(&rdev->nr_pending);
389 atomic_inc(&rdev->nr_pending);
390 rcu_read_unlock();
391 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
392 bi->bi_end_io = md_end_flush;
393 bi->bi_private = rdev;
394 bi->bi_bdev = rdev->bdev;
395 atomic_inc(&mddev->flush_pending);
396 submit_bio(WRITE_FLUSH, bi);
397 rcu_read_lock();
398 rdev_dec_pending(rdev, mddev);
399 }
400 rcu_read_unlock();
401 if (atomic_dec_and_test(&mddev->flush_pending))
402 queue_work(md_wq, &mddev->flush_work);
403 }
404
405 static void md_submit_flush_data(struct work_struct *ws)
406 {
407 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
408 struct bio *bio = mddev->flush_bio;
409
410 if (bio->bi_iter.bi_size == 0)
411 /* an empty barrier - all done */
412 bio_endio(bio);
413 else {
414 bio->bi_rw &= ~REQ_FLUSH;
415 mddev->pers->make_request(mddev, bio);
416 }
417
418 mddev->flush_bio = NULL;
419 wake_up(&mddev->sb_wait);
420 }
421
422 void md_flush_request(struct mddev *mddev, struct bio *bio)
423 {
424 spin_lock_irq(&mddev->lock);
425 wait_event_lock_irq(mddev->sb_wait,
426 !mddev->flush_bio,
427 mddev->lock);
428 mddev->flush_bio = bio;
429 spin_unlock_irq(&mddev->lock);
430
431 INIT_WORK(&mddev->flush_work, submit_flushes);
432 queue_work(md_wq, &mddev->flush_work);
433 }
434 EXPORT_SYMBOL(md_flush_request);
435
436 void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
437 {
438 struct mddev *mddev = cb->data;
439 md_wakeup_thread(mddev->thread);
440 kfree(cb);
441 }
442 EXPORT_SYMBOL(md_unplug);
443
444 static inline struct mddev *mddev_get(struct mddev *mddev)
445 {
446 atomic_inc(&mddev->active);
447 return mddev;
448 }
449
450 static void mddev_delayed_delete(struct work_struct *ws);
451
452 static void mddev_put(struct mddev *mddev)
453 {
454 struct bio_set *bs = NULL;
455
456 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
457 return;
458 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
459 mddev->ctime == 0 && !mddev->hold_active) {
460 /* Array is not configured at all, and not held active,
461 * so destroy it */
462 list_del_init(&mddev->all_mddevs);
463 bs = mddev->bio_set;
464 mddev->bio_set = NULL;
465 if (mddev->gendisk) {
466 /* We did a probe so need to clean up. Call
467 * queue_work inside the spinlock so that
468 * flush_workqueue() after mddev_find will
469 * succeed in waiting for the work to be done.
470 */
471 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
472 queue_work(md_misc_wq, &mddev->del_work);
473 } else
474 kfree(mddev);
475 }
476 spin_unlock(&all_mddevs_lock);
477 if (bs)
478 bioset_free(bs);
479 }
480
481 static void md_safemode_timeout(unsigned long data);
482
483 void mddev_init(struct mddev *mddev)
484 {
485 mutex_init(&mddev->open_mutex);
486 mutex_init(&mddev->reconfig_mutex);
487 mutex_init(&mddev->bitmap_info.mutex);
488 INIT_LIST_HEAD(&mddev->disks);
489 INIT_LIST_HEAD(&mddev->all_mddevs);
490 setup_timer(&mddev->safemode_timer, md_safemode_timeout,
491 (unsigned long) mddev);
492 atomic_set(&mddev->active, 1);
493 atomic_set(&mddev->openers, 0);
494 atomic_set(&mddev->active_io, 0);
495 spin_lock_init(&mddev->lock);
496 atomic_set(&mddev->flush_pending, 0);
497 init_waitqueue_head(&mddev->sb_wait);
498 init_waitqueue_head(&mddev->recovery_wait);
499 mddev->reshape_position = MaxSector;
500 mddev->reshape_backwards = 0;
501 mddev->last_sync_action = "none";
502 mddev->resync_min = 0;
503 mddev->resync_max = MaxSector;
504 mddev->level = LEVEL_NONE;
505 }
506 EXPORT_SYMBOL_GPL(mddev_init);
507
508 static struct mddev *mddev_find(dev_t unit)
509 {
510 struct mddev *mddev, *new = NULL;
511
512 if (unit && MAJOR(unit) != MD_MAJOR)
513 unit &= ~((1<<MdpMinorShift)-1);
514
515 retry:
516 spin_lock(&all_mddevs_lock);
517
518 if (unit) {
519 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
520 if (mddev->unit == unit) {
521 mddev_get(mddev);
522 spin_unlock(&all_mddevs_lock);
523 kfree(new);
524 return mddev;
525 }
526
527 if (new) {
528 list_add(&new->all_mddevs, &all_mddevs);
529 spin_unlock(&all_mddevs_lock);
530 new->hold_active = UNTIL_IOCTL;
531 return new;
532 }
533 } else if (new) {
534 /* find an unused unit number */
535 static int next_minor = 512;
536 int start = next_minor;
537 int is_free = 0;
538 int dev = 0;
539 while (!is_free) {
540 dev = MKDEV(MD_MAJOR, next_minor);
541 next_minor++;
542 if (next_minor > MINORMASK)
543 next_minor = 0;
544 if (next_minor == start) {
545 /* Oh dear, all in use. */
546 spin_unlock(&all_mddevs_lock);
547 kfree(new);
548 return NULL;
549 }
550
551 is_free = 1;
552 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
553 if (mddev->unit == dev) {
554 is_free = 0;
555 break;
556 }
557 }
558 new->unit = dev;
559 new->md_minor = MINOR(dev);
560 new->hold_active = UNTIL_STOP;
561 list_add(&new->all_mddevs, &all_mddevs);
562 spin_unlock(&all_mddevs_lock);
563 return new;
564 }
565 spin_unlock(&all_mddevs_lock);
566
567 new = kzalloc(sizeof(*new), GFP_KERNEL);
568 if (!new)
569 return NULL;
570
571 new->unit = unit;
572 if (MAJOR(unit) == MD_MAJOR)
573 new->md_minor = MINOR(unit);
574 else
575 new->md_minor = MINOR(unit) >> MdpMinorShift;
576
577 mddev_init(new);
578
579 goto retry;
580 }
581
582 static struct attribute_group md_redundancy_group;
583
584 void mddev_unlock(struct mddev *mddev)
585 {
586 if (mddev->to_remove) {
587 /* These cannot be removed under reconfig_mutex as
588 * an access to the files will try to take reconfig_mutex
589 * while holding the file unremovable, which leads to
590 * a deadlock.
591 * So hold set sysfs_active while the remove in happeing,
592 * and anything else which might set ->to_remove or my
593 * otherwise change the sysfs namespace will fail with
594 * -EBUSY if sysfs_active is still set.
595 * We set sysfs_active under reconfig_mutex and elsewhere
596 * test it under the same mutex to ensure its correct value
597 * is seen.
598 */
599 struct attribute_group *to_remove = mddev->to_remove;
600 mddev->to_remove = NULL;
601 mddev->sysfs_active = 1;
602 mutex_unlock(&mddev->reconfig_mutex);
603
604 if (mddev->kobj.sd) {
605 if (to_remove != &md_redundancy_group)
606 sysfs_remove_group(&mddev->kobj, to_remove);
607 if (mddev->pers == NULL ||
608 mddev->pers->sync_request == NULL) {
609 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
610 if (mddev->sysfs_action)
611 sysfs_put(mddev->sysfs_action);
612 mddev->sysfs_action = NULL;
613 }
614 }
615 mddev->sysfs_active = 0;
616 } else
617 mutex_unlock(&mddev->reconfig_mutex);
618
619 /* As we've dropped the mutex we need a spinlock to
620 * make sure the thread doesn't disappear
621 */
622 spin_lock(&pers_lock);
623 md_wakeup_thread(mddev->thread);
624 spin_unlock(&pers_lock);
625 }
626 EXPORT_SYMBOL_GPL(mddev_unlock);
627
628 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
629 {
630 struct md_rdev *rdev;
631
632 rdev_for_each_rcu(rdev, mddev)
633 if (rdev->desc_nr == nr)
634 return rdev;
635
636 return NULL;
637 }
638 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
639
640 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
641 {
642 struct md_rdev *rdev;
643
644 rdev_for_each(rdev, mddev)
645 if (rdev->bdev->bd_dev == dev)
646 return rdev;
647
648 return NULL;
649 }
650
651 static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
652 {
653 struct md_rdev *rdev;
654
655 rdev_for_each_rcu(rdev, mddev)
656 if (rdev->bdev->bd_dev == dev)
657 return rdev;
658
659 return NULL;
660 }
661
662 static struct md_personality *find_pers(int level, char *clevel)
663 {
664 struct md_personality *pers;
665 list_for_each_entry(pers, &pers_list, list) {
666 if (level != LEVEL_NONE && pers->level == level)
667 return pers;
668 if (strcmp(pers->name, clevel)==0)
669 return pers;
670 }
671 return NULL;
672 }
673
674 /* return the offset of the super block in 512byte sectors */
675 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
676 {
677 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
678 return MD_NEW_SIZE_SECTORS(num_sectors);
679 }
680
681 static int alloc_disk_sb(struct md_rdev *rdev)
682 {
683 rdev->sb_page = alloc_page(GFP_KERNEL);
684 if (!rdev->sb_page) {
685 printk(KERN_ALERT "md: out of memory.\n");
686 return -ENOMEM;
687 }
688
689 return 0;
690 }
691
692 void md_rdev_clear(struct md_rdev *rdev)
693 {
694 if (rdev->sb_page) {
695 put_page(rdev->sb_page);
696 rdev->sb_loaded = 0;
697 rdev->sb_page = NULL;
698 rdev->sb_start = 0;
699 rdev->sectors = 0;
700 }
701 if (rdev->bb_page) {
702 put_page(rdev->bb_page);
703 rdev->bb_page = NULL;
704 }
705 badblocks_exit(&rdev->badblocks);
706 }
707 EXPORT_SYMBOL_GPL(md_rdev_clear);
708
709 static void super_written(struct bio *bio)
710 {
711 struct md_rdev *rdev = bio->bi_private;
712 struct mddev *mddev = rdev->mddev;
713
714 if (bio->bi_error) {
715 printk("md: super_written gets error=%d\n", bio->bi_error);
716 md_error(mddev, rdev);
717 }
718
719 if (atomic_dec_and_test(&mddev->pending_writes))
720 wake_up(&mddev->sb_wait);
721 rdev_dec_pending(rdev, mddev);
722 bio_put(bio);
723 }
724
725 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
726 sector_t sector, int size, struct page *page)
727 {
728 /* write first size bytes of page to sector of rdev
729 * Increment mddev->pending_writes before returning
730 * and decrement it on completion, waking up sb_wait
731 * if zero is reached.
732 * If an error occurred, call md_error
733 */
734 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
735
736 atomic_inc(&rdev->nr_pending);
737
738 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
739 bio->bi_iter.bi_sector = sector;
740 bio_add_page(bio, page, size, 0);
741 bio->bi_private = rdev;
742 bio->bi_end_io = super_written;
743
744 atomic_inc(&mddev->pending_writes);
745 submit_bio(WRITE_FLUSH_FUA, bio);
746 }
747
748 void md_super_wait(struct mddev *mddev)
749 {
750 /* wait for all superblock writes that were scheduled to complete */
751 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
752 }
753
754 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
755 struct page *page, int rw, bool metadata_op)
756 {
757 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
758 int ret;
759
760 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
761 rdev->meta_bdev : rdev->bdev;
762 if (metadata_op)
763 bio->bi_iter.bi_sector = sector + rdev->sb_start;
764 else if (rdev->mddev->reshape_position != MaxSector &&
765 (rdev->mddev->reshape_backwards ==
766 (sector >= rdev->mddev->reshape_position)))
767 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
768 else
769 bio->bi_iter.bi_sector = sector + rdev->data_offset;
770 bio_add_page(bio, page, size, 0);
771 submit_bio_wait(rw, bio);
772
773 ret = !bio->bi_error;
774 bio_put(bio);
775 return ret;
776 }
777 EXPORT_SYMBOL_GPL(sync_page_io);
778
779 static int read_disk_sb(struct md_rdev *rdev, int size)
780 {
781 char b[BDEVNAME_SIZE];
782
783 if (rdev->sb_loaded)
784 return 0;
785
786 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
787 goto fail;
788 rdev->sb_loaded = 1;
789 return 0;
790
791 fail:
792 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
793 bdevname(rdev->bdev,b));
794 return -EINVAL;
795 }
796
797 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
798 {
799 return sb1->set_uuid0 == sb2->set_uuid0 &&
800 sb1->set_uuid1 == sb2->set_uuid1 &&
801 sb1->set_uuid2 == sb2->set_uuid2 &&
802 sb1->set_uuid3 == sb2->set_uuid3;
803 }
804
805 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
806 {
807 int ret;
808 mdp_super_t *tmp1, *tmp2;
809
810 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
811 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
812
813 if (!tmp1 || !tmp2) {
814 ret = 0;
815 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
816 goto abort;
817 }
818
819 *tmp1 = *sb1;
820 *tmp2 = *sb2;
821
822 /*
823 * nr_disks is not constant
824 */
825 tmp1->nr_disks = 0;
826 tmp2->nr_disks = 0;
827
828 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
829 abort:
830 kfree(tmp1);
831 kfree(tmp2);
832 return ret;
833 }
834
835 static u32 md_csum_fold(u32 csum)
836 {
837 csum = (csum & 0xffff) + (csum >> 16);
838 return (csum & 0xffff) + (csum >> 16);
839 }
840
841 static unsigned int calc_sb_csum(mdp_super_t *sb)
842 {
843 u64 newcsum = 0;
844 u32 *sb32 = (u32*)sb;
845 int i;
846 unsigned int disk_csum, csum;
847
848 disk_csum = sb->sb_csum;
849 sb->sb_csum = 0;
850
851 for (i = 0; i < MD_SB_BYTES/4 ; i++)
852 newcsum += sb32[i];
853 csum = (newcsum & 0xffffffff) + (newcsum>>32);
854
855 #ifdef CONFIG_ALPHA
856 /* This used to use csum_partial, which was wrong for several
857 * reasons including that different results are returned on
858 * different architectures. It isn't critical that we get exactly
859 * the same return value as before (we always csum_fold before
860 * testing, and that removes any differences). However as we
861 * know that csum_partial always returned a 16bit value on
862 * alphas, do a fold to maximise conformity to previous behaviour.
863 */
864 sb->sb_csum = md_csum_fold(disk_csum);
865 #else
866 sb->sb_csum = disk_csum;
867 #endif
868 return csum;
869 }
870
871 /*
872 * Handle superblock details.
873 * We want to be able to handle multiple superblock formats
874 * so we have a common interface to them all, and an array of
875 * different handlers.
876 * We rely on user-space to write the initial superblock, and support
877 * reading and updating of superblocks.
878 * Interface methods are:
879 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
880 * loads and validates a superblock on dev.
881 * if refdev != NULL, compare superblocks on both devices
882 * Return:
883 * 0 - dev has a superblock that is compatible with refdev
884 * 1 - dev has a superblock that is compatible and newer than refdev
885 * so dev should be used as the refdev in future
886 * -EINVAL superblock incompatible or invalid
887 * -othererror e.g. -EIO
888 *
889 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
890 * Verify that dev is acceptable into mddev.
891 * The first time, mddev->raid_disks will be 0, and data from
892 * dev should be merged in. Subsequent calls check that dev
893 * is new enough. Return 0 or -EINVAL
894 *
895 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
896 * Update the superblock for rdev with data in mddev
897 * This does not write to disc.
898 *
899 */
900
901 struct super_type {
902 char *name;
903 struct module *owner;
904 int (*load_super)(struct md_rdev *rdev,
905 struct md_rdev *refdev,
906 int minor_version);
907 int (*validate_super)(struct mddev *mddev,
908 struct md_rdev *rdev);
909 void (*sync_super)(struct mddev *mddev,
910 struct md_rdev *rdev);
911 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
912 sector_t num_sectors);
913 int (*allow_new_offset)(struct md_rdev *rdev,
914 unsigned long long new_offset);
915 };
916
917 /*
918 * Check that the given mddev has no bitmap.
919 *
920 * This function is called from the run method of all personalities that do not
921 * support bitmaps. It prints an error message and returns non-zero if mddev
922 * has a bitmap. Otherwise, it returns 0.
923 *
924 */
925 int md_check_no_bitmap(struct mddev *mddev)
926 {
927 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
928 return 0;
929 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
930 mdname(mddev), mddev->pers->name);
931 return 1;
932 }
933 EXPORT_SYMBOL(md_check_no_bitmap);
934
935 /*
936 * load_super for 0.90.0
937 */
938 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
939 {
940 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
941 mdp_super_t *sb;
942 int ret;
943
944 /*
945 * Calculate the position of the superblock (512byte sectors),
946 * it's at the end of the disk.
947 *
948 * It also happens to be a multiple of 4Kb.
949 */
950 rdev->sb_start = calc_dev_sboffset(rdev);
951
952 ret = read_disk_sb(rdev, MD_SB_BYTES);
953 if (ret) return ret;
954
955 ret = -EINVAL;
956
957 bdevname(rdev->bdev, b);
958 sb = page_address(rdev->sb_page);
959
960 if (sb->md_magic != MD_SB_MAGIC) {
961 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
962 b);
963 goto abort;
964 }
965
966 if (sb->major_version != 0 ||
967 sb->minor_version < 90 ||
968 sb->minor_version > 91) {
969 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
970 sb->major_version, sb->minor_version,
971 b);
972 goto abort;
973 }
974
975 if (sb->raid_disks <= 0)
976 goto abort;
977
978 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
979 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
980 b);
981 goto abort;
982 }
983
984 rdev->preferred_minor = sb->md_minor;
985 rdev->data_offset = 0;
986 rdev->new_data_offset = 0;
987 rdev->sb_size = MD_SB_BYTES;
988 rdev->badblocks.shift = -1;
989
990 if (sb->level == LEVEL_MULTIPATH)
991 rdev->desc_nr = -1;
992 else
993 rdev->desc_nr = sb->this_disk.number;
994
995 if (!refdev) {
996 ret = 1;
997 } else {
998 __u64 ev1, ev2;
999 mdp_super_t *refsb = page_address(refdev->sb_page);
1000 if (!uuid_equal(refsb, sb)) {
1001 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1002 b, bdevname(refdev->bdev,b2));
1003 goto abort;
1004 }
1005 if (!sb_equal(refsb, sb)) {
1006 printk(KERN_WARNING "md: %s has same UUID"
1007 " but different superblock to %s\n",
1008 b, bdevname(refdev->bdev, b2));
1009 goto abort;
1010 }
1011 ev1 = md_event(sb);
1012 ev2 = md_event(refsb);
1013 if (ev1 > ev2)
1014 ret = 1;
1015 else
1016 ret = 0;
1017 }
1018 rdev->sectors = rdev->sb_start;
1019 /* Limit to 4TB as metadata cannot record more than that.
1020 * (not needed for Linear and RAID0 as metadata doesn't
1021 * record this size)
1022 */
1023 if (IS_ENABLED(CONFIG_LBDAF) && (u64)rdev->sectors >= (2ULL << 32) &&
1024 sb->level >= 1)
1025 rdev->sectors = (sector_t)(2ULL << 32) - 2;
1026
1027 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1028 /* "this cannot possibly happen" ... */
1029 ret = -EINVAL;
1030
1031 abort:
1032 return ret;
1033 }
1034
1035 /*
1036 * validate_super for 0.90.0
1037 */
1038 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1039 {
1040 mdp_disk_t *desc;
1041 mdp_super_t *sb = page_address(rdev->sb_page);
1042 __u64 ev1 = md_event(sb);
1043
1044 rdev->raid_disk = -1;
1045 clear_bit(Faulty, &rdev->flags);
1046 clear_bit(In_sync, &rdev->flags);
1047 clear_bit(Bitmap_sync, &rdev->flags);
1048 clear_bit(WriteMostly, &rdev->flags);
1049
1050 if (mddev->raid_disks == 0) {
1051 mddev->major_version = 0;
1052 mddev->minor_version = sb->minor_version;
1053 mddev->patch_version = sb->patch_version;
1054 mddev->external = 0;
1055 mddev->chunk_sectors = sb->chunk_size >> 9;
1056 mddev->ctime = sb->ctime;
1057 mddev->utime = sb->utime;
1058 mddev->level = sb->level;
1059 mddev->clevel[0] = 0;
1060 mddev->layout = sb->layout;
1061 mddev->raid_disks = sb->raid_disks;
1062 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1063 mddev->events = ev1;
1064 mddev->bitmap_info.offset = 0;
1065 mddev->bitmap_info.space = 0;
1066 /* bitmap can use 60 K after the 4K superblocks */
1067 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1068 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1069 mddev->reshape_backwards = 0;
1070
1071 if (mddev->minor_version >= 91) {
1072 mddev->reshape_position = sb->reshape_position;
1073 mddev->delta_disks = sb->delta_disks;
1074 mddev->new_level = sb->new_level;
1075 mddev->new_layout = sb->new_layout;
1076 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1077 if (mddev->delta_disks < 0)
1078 mddev->reshape_backwards = 1;
1079 } else {
1080 mddev->reshape_position = MaxSector;
1081 mddev->delta_disks = 0;
1082 mddev->new_level = mddev->level;
1083 mddev->new_layout = mddev->layout;
1084 mddev->new_chunk_sectors = mddev->chunk_sectors;
1085 }
1086
1087 if (sb->state & (1<<MD_SB_CLEAN))
1088 mddev->recovery_cp = MaxSector;
1089 else {
1090 if (sb->events_hi == sb->cp_events_hi &&
1091 sb->events_lo == sb->cp_events_lo) {
1092 mddev->recovery_cp = sb->recovery_cp;
1093 } else
1094 mddev->recovery_cp = 0;
1095 }
1096
1097 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1098 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1099 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1100 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1101
1102 mddev->max_disks = MD_SB_DISKS;
1103
1104 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1105 mddev->bitmap_info.file == NULL) {
1106 mddev->bitmap_info.offset =
1107 mddev->bitmap_info.default_offset;
1108 mddev->bitmap_info.space =
1109 mddev->bitmap_info.default_space;
1110 }
1111
1112 } else if (mddev->pers == NULL) {
1113 /* Insist on good event counter while assembling, except
1114 * for spares (which don't need an event count) */
1115 ++ev1;
1116 if (sb->disks[rdev->desc_nr].state & (
1117 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1118 if (ev1 < mddev->events)
1119 return -EINVAL;
1120 } else if (mddev->bitmap) {
1121 /* if adding to array with a bitmap, then we can accept an
1122 * older device ... but not too old.
1123 */
1124 if (ev1 < mddev->bitmap->events_cleared)
1125 return 0;
1126 if (ev1 < mddev->events)
1127 set_bit(Bitmap_sync, &rdev->flags);
1128 } else {
1129 if (ev1 < mddev->events)
1130 /* just a hot-add of a new device, leave raid_disk at -1 */
1131 return 0;
1132 }
1133
1134 if (mddev->level != LEVEL_MULTIPATH) {
1135 desc = sb->disks + rdev->desc_nr;
1136
1137 if (desc->state & (1<<MD_DISK_FAULTY))
1138 set_bit(Faulty, &rdev->flags);
1139 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1140 desc->raid_disk < mddev->raid_disks */) {
1141 set_bit(In_sync, &rdev->flags);
1142 rdev->raid_disk = desc->raid_disk;
1143 rdev->saved_raid_disk = desc->raid_disk;
1144 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1145 /* active but not in sync implies recovery up to
1146 * reshape position. We don't know exactly where
1147 * that is, so set to zero for now */
1148 if (mddev->minor_version >= 91) {
1149 rdev->recovery_offset = 0;
1150 rdev->raid_disk = desc->raid_disk;
1151 }
1152 }
1153 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1154 set_bit(WriteMostly, &rdev->flags);
1155 } else /* MULTIPATH are always insync */
1156 set_bit(In_sync, &rdev->flags);
1157 return 0;
1158 }
1159
1160 /*
1161 * sync_super for 0.90.0
1162 */
1163 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1164 {
1165 mdp_super_t *sb;
1166 struct md_rdev *rdev2;
1167 int next_spare = mddev->raid_disks;
1168
1169 /* make rdev->sb match mddev data..
1170 *
1171 * 1/ zero out disks
1172 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1173 * 3/ any empty disks < next_spare become removed
1174 *
1175 * disks[0] gets initialised to REMOVED because
1176 * we cannot be sure from other fields if it has
1177 * been initialised or not.
1178 */
1179 int i;
1180 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1181
1182 rdev->sb_size = MD_SB_BYTES;
1183
1184 sb = page_address(rdev->sb_page);
1185
1186 memset(sb, 0, sizeof(*sb));
1187
1188 sb->md_magic = MD_SB_MAGIC;
1189 sb->major_version = mddev->major_version;
1190 sb->patch_version = mddev->patch_version;
1191 sb->gvalid_words = 0; /* ignored */
1192 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1193 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1194 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1195 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1196
1197 sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1198 sb->level = mddev->level;
1199 sb->size = mddev->dev_sectors / 2;
1200 sb->raid_disks = mddev->raid_disks;
1201 sb->md_minor = mddev->md_minor;
1202 sb->not_persistent = 0;
1203 sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1204 sb->state = 0;
1205 sb->events_hi = (mddev->events>>32);
1206 sb->events_lo = (u32)mddev->events;
1207
1208 if (mddev->reshape_position == MaxSector)
1209 sb->minor_version = 90;
1210 else {
1211 sb->minor_version = 91;
1212 sb->reshape_position = mddev->reshape_position;
1213 sb->new_level = mddev->new_level;
1214 sb->delta_disks = mddev->delta_disks;
1215 sb->new_layout = mddev->new_layout;
1216 sb->new_chunk = mddev->new_chunk_sectors << 9;
1217 }
1218 mddev->minor_version = sb->minor_version;
1219 if (mddev->in_sync)
1220 {
1221 sb->recovery_cp = mddev->recovery_cp;
1222 sb->cp_events_hi = (mddev->events>>32);
1223 sb->cp_events_lo = (u32)mddev->events;
1224 if (mddev->recovery_cp == MaxSector)
1225 sb->state = (1<< MD_SB_CLEAN);
1226 } else
1227 sb->recovery_cp = 0;
1228
1229 sb->layout = mddev->layout;
1230 sb->chunk_size = mddev->chunk_sectors << 9;
1231
1232 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1233 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1234
1235 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1236 rdev_for_each(rdev2, mddev) {
1237 mdp_disk_t *d;
1238 int desc_nr;
1239 int is_active = test_bit(In_sync, &rdev2->flags);
1240
1241 if (rdev2->raid_disk >= 0 &&
1242 sb->minor_version >= 91)
1243 /* we have nowhere to store the recovery_offset,
1244 * but if it is not below the reshape_position,
1245 * we can piggy-back on that.
1246 */
1247 is_active = 1;
1248 if (rdev2->raid_disk < 0 ||
1249 test_bit(Faulty, &rdev2->flags))
1250 is_active = 0;
1251 if (is_active)
1252 desc_nr = rdev2->raid_disk;
1253 else
1254 desc_nr = next_spare++;
1255 rdev2->desc_nr = desc_nr;
1256 d = &sb->disks[rdev2->desc_nr];
1257 nr_disks++;
1258 d->number = rdev2->desc_nr;
1259 d->major = MAJOR(rdev2->bdev->bd_dev);
1260 d->minor = MINOR(rdev2->bdev->bd_dev);
1261 if (is_active)
1262 d->raid_disk = rdev2->raid_disk;
1263 else
1264 d->raid_disk = rdev2->desc_nr; /* compatibility */
1265 if (test_bit(Faulty, &rdev2->flags))
1266 d->state = (1<<MD_DISK_FAULTY);
1267 else if (is_active) {
1268 d->state = (1<<MD_DISK_ACTIVE);
1269 if (test_bit(In_sync, &rdev2->flags))
1270 d->state |= (1<<MD_DISK_SYNC);
1271 active++;
1272 working++;
1273 } else {
1274 d->state = 0;
1275 spare++;
1276 working++;
1277 }
1278 if (test_bit(WriteMostly, &rdev2->flags))
1279 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1280 }
1281 /* now set the "removed" and "faulty" bits on any missing devices */
1282 for (i=0 ; i < mddev->raid_disks ; i++) {
1283 mdp_disk_t *d = &sb->disks[i];
1284 if (d->state == 0 && d->number == 0) {
1285 d->number = i;
1286 d->raid_disk = i;
1287 d->state = (1<<MD_DISK_REMOVED);
1288 d->state |= (1<<MD_DISK_FAULTY);
1289 failed++;
1290 }
1291 }
1292 sb->nr_disks = nr_disks;
1293 sb->active_disks = active;
1294 sb->working_disks = working;
1295 sb->failed_disks = failed;
1296 sb->spare_disks = spare;
1297
1298 sb->this_disk = sb->disks[rdev->desc_nr];
1299 sb->sb_csum = calc_sb_csum(sb);
1300 }
1301
1302 /*
1303 * rdev_size_change for 0.90.0
1304 */
1305 static unsigned long long
1306 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1307 {
1308 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1309 return 0; /* component must fit device */
1310 if (rdev->mddev->bitmap_info.offset)
1311 return 0; /* can't move bitmap */
1312 rdev->sb_start = calc_dev_sboffset(rdev);
1313 if (!num_sectors || num_sectors > rdev->sb_start)
1314 num_sectors = rdev->sb_start;
1315 /* Limit to 4TB as metadata cannot record more than that.
1316 * 4TB == 2^32 KB, or 2*2^32 sectors.
1317 */
1318 if (IS_ENABLED(CONFIG_LBDAF) && (u64)num_sectors >= (2ULL << 32) &&
1319 rdev->mddev->level >= 1)
1320 num_sectors = (sector_t)(2ULL << 32) - 2;
1321 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1322 rdev->sb_page);
1323 md_super_wait(rdev->mddev);
1324 return num_sectors;
1325 }
1326
1327 static int
1328 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1329 {
1330 /* non-zero offset changes not possible with v0.90 */
1331 return new_offset == 0;
1332 }
1333
1334 /*
1335 * version 1 superblock
1336 */
1337
1338 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1339 {
1340 __le32 disk_csum;
1341 u32 csum;
1342 unsigned long long newcsum;
1343 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1344 __le32 *isuper = (__le32*)sb;
1345
1346 disk_csum = sb->sb_csum;
1347 sb->sb_csum = 0;
1348 newcsum = 0;
1349 for (; size >= 4; size -= 4)
1350 newcsum += le32_to_cpu(*isuper++);
1351
1352 if (size == 2)
1353 newcsum += le16_to_cpu(*(__le16*) isuper);
1354
1355 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1356 sb->sb_csum = disk_csum;
1357 return cpu_to_le32(csum);
1358 }
1359
1360 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1361 {
1362 struct mdp_superblock_1 *sb;
1363 int ret;
1364 sector_t sb_start;
1365 sector_t sectors;
1366 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1367 int bmask;
1368
1369 /*
1370 * Calculate the position of the superblock in 512byte sectors.
1371 * It is always aligned to a 4K boundary and
1372 * depeding on minor_version, it can be:
1373 * 0: At least 8K, but less than 12K, from end of device
1374 * 1: At start of device
1375 * 2: 4K from start of device.
1376 */
1377 switch(minor_version) {
1378 case 0:
1379 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1380 sb_start -= 8*2;
1381 sb_start &= ~(sector_t)(4*2-1);
1382 break;
1383 case 1:
1384 sb_start = 0;
1385 break;
1386 case 2:
1387 sb_start = 8;
1388 break;
1389 default:
1390 return -EINVAL;
1391 }
1392 rdev->sb_start = sb_start;
1393
1394 /* superblock is rarely larger than 1K, but it can be larger,
1395 * and it is safe to read 4k, so we do that
1396 */
1397 ret = read_disk_sb(rdev, 4096);
1398 if (ret) return ret;
1399
1400 sb = page_address(rdev->sb_page);
1401
1402 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1403 sb->major_version != cpu_to_le32(1) ||
1404 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1405 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1406 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1407 return -EINVAL;
1408
1409 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1410 printk("md: invalid superblock checksum on %s\n",
1411 bdevname(rdev->bdev,b));
1412 return -EINVAL;
1413 }
1414 if (le64_to_cpu(sb->data_size) < 10) {
1415 printk("md: data_size too small on %s\n",
1416 bdevname(rdev->bdev,b));
1417 return -EINVAL;
1418 }
1419 if (sb->pad0 ||
1420 sb->pad3[0] ||
1421 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1422 /* Some padding is non-zero, might be a new feature */
1423 return -EINVAL;
1424
1425 rdev->preferred_minor = 0xffff;
1426 rdev->data_offset = le64_to_cpu(sb->data_offset);
1427 rdev->new_data_offset = rdev->data_offset;
1428 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1429 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1430 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1431 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1432
1433 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1434 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1435 if (rdev->sb_size & bmask)
1436 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1437
1438 if (minor_version
1439 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1440 return -EINVAL;
1441 if (minor_version
1442 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1443 return -EINVAL;
1444
1445 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1446 rdev->desc_nr = -1;
1447 else
1448 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1449
1450 if (!rdev->bb_page) {
1451 rdev->bb_page = alloc_page(GFP_KERNEL);
1452 if (!rdev->bb_page)
1453 return -ENOMEM;
1454 }
1455 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1456 rdev->badblocks.count == 0) {
1457 /* need to load the bad block list.
1458 * Currently we limit it to one page.
1459 */
1460 s32 offset;
1461 sector_t bb_sector;
1462 u64 *bbp;
1463 int i;
1464 int sectors = le16_to_cpu(sb->bblog_size);
1465 if (sectors > (PAGE_SIZE / 512))
1466 return -EINVAL;
1467 offset = le32_to_cpu(sb->bblog_offset);
1468 if (offset == 0)
1469 return -EINVAL;
1470 bb_sector = (long long)offset;
1471 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1472 rdev->bb_page, READ, true))
1473 return -EIO;
1474 bbp = (u64 *)page_address(rdev->bb_page);
1475 rdev->badblocks.shift = sb->bblog_shift;
1476 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1477 u64 bb = le64_to_cpu(*bbp);
1478 int count = bb & (0x3ff);
1479 u64 sector = bb >> 10;
1480 sector <<= sb->bblog_shift;
1481 count <<= sb->bblog_shift;
1482 if (bb + 1 == 0)
1483 break;
1484 if (badblocks_set(&rdev->badblocks, sector, count, 1))
1485 return -EINVAL;
1486 }
1487 } else if (sb->bblog_offset != 0)
1488 rdev->badblocks.shift = 0;
1489
1490 if (!refdev) {
1491 ret = 1;
1492 } else {
1493 __u64 ev1, ev2;
1494 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1495
1496 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1497 sb->level != refsb->level ||
1498 sb->layout != refsb->layout ||
1499 sb->chunksize != refsb->chunksize) {
1500 printk(KERN_WARNING "md: %s has strangely different"
1501 " superblock to %s\n",
1502 bdevname(rdev->bdev,b),
1503 bdevname(refdev->bdev,b2));
1504 return -EINVAL;
1505 }
1506 ev1 = le64_to_cpu(sb->events);
1507 ev2 = le64_to_cpu(refsb->events);
1508
1509 if (ev1 > ev2)
1510 ret = 1;
1511 else
1512 ret = 0;
1513 }
1514 if (minor_version) {
1515 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1516 sectors -= rdev->data_offset;
1517 } else
1518 sectors = rdev->sb_start;
1519 if (sectors < le64_to_cpu(sb->data_size))
1520 return -EINVAL;
1521 rdev->sectors = le64_to_cpu(sb->data_size);
1522 return ret;
1523 }
1524
1525 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1526 {
1527 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1528 __u64 ev1 = le64_to_cpu(sb->events);
1529
1530 rdev->raid_disk = -1;
1531 clear_bit(Faulty, &rdev->flags);
1532 clear_bit(In_sync, &rdev->flags);
1533 clear_bit(Bitmap_sync, &rdev->flags);
1534 clear_bit(WriteMostly, &rdev->flags);
1535
1536 if (mddev->raid_disks == 0) {
1537 mddev->major_version = 1;
1538 mddev->patch_version = 0;
1539 mddev->external = 0;
1540 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1541 mddev->ctime = le64_to_cpu(sb->ctime);
1542 mddev->utime = le64_to_cpu(sb->utime);
1543 mddev->level = le32_to_cpu(sb->level);
1544 mddev->clevel[0] = 0;
1545 mddev->layout = le32_to_cpu(sb->layout);
1546 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1547 mddev->dev_sectors = le64_to_cpu(sb->size);
1548 mddev->events = ev1;
1549 mddev->bitmap_info.offset = 0;
1550 mddev->bitmap_info.space = 0;
1551 /* Default location for bitmap is 1K after superblock
1552 * using 3K - total of 4K
1553 */
1554 mddev->bitmap_info.default_offset = 1024 >> 9;
1555 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1556 mddev->reshape_backwards = 0;
1557
1558 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1559 memcpy(mddev->uuid, sb->set_uuid, 16);
1560
1561 mddev->max_disks = (4096-256)/2;
1562
1563 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1564 mddev->bitmap_info.file == NULL) {
1565 mddev->bitmap_info.offset =
1566 (__s32)le32_to_cpu(sb->bitmap_offset);
1567 /* Metadata doesn't record how much space is available.
1568 * For 1.0, we assume we can use up to the superblock
1569 * if before, else to 4K beyond superblock.
1570 * For others, assume no change is possible.
1571 */
1572 if (mddev->minor_version > 0)
1573 mddev->bitmap_info.space = 0;
1574 else if (mddev->bitmap_info.offset > 0)
1575 mddev->bitmap_info.space =
1576 8 - mddev->bitmap_info.offset;
1577 else
1578 mddev->bitmap_info.space =
1579 -mddev->bitmap_info.offset;
1580 }
1581
1582 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1583 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1584 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1585 mddev->new_level = le32_to_cpu(sb->new_level);
1586 mddev->new_layout = le32_to_cpu(sb->new_layout);
1587 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1588 if (mddev->delta_disks < 0 ||
1589 (mddev->delta_disks == 0 &&
1590 (le32_to_cpu(sb->feature_map)
1591 & MD_FEATURE_RESHAPE_BACKWARDS)))
1592 mddev->reshape_backwards = 1;
1593 } else {
1594 mddev->reshape_position = MaxSector;
1595 mddev->delta_disks = 0;
1596 mddev->new_level = mddev->level;
1597 mddev->new_layout = mddev->layout;
1598 mddev->new_chunk_sectors = mddev->chunk_sectors;
1599 }
1600
1601 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL) {
1602 set_bit(MD_HAS_JOURNAL, &mddev->flags);
1603 if (mddev->recovery_cp == MaxSector)
1604 set_bit(MD_JOURNAL_CLEAN, &mddev->flags);
1605 }
1606 } else if (mddev->pers == NULL) {
1607 /* Insist of good event counter while assembling, except for
1608 * spares (which don't need an event count) */
1609 ++ev1;
1610 if (rdev->desc_nr >= 0 &&
1611 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1612 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1613 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1614 if (ev1 < mddev->events)
1615 return -EINVAL;
1616 } else if (mddev->bitmap) {
1617 /* If adding to array with a bitmap, then we can accept an
1618 * older device, but not too old.
1619 */
1620 if (ev1 < mddev->bitmap->events_cleared)
1621 return 0;
1622 if (ev1 < mddev->events)
1623 set_bit(Bitmap_sync, &rdev->flags);
1624 } else {
1625 if (ev1 < mddev->events)
1626 /* just a hot-add of a new device, leave raid_disk at -1 */
1627 return 0;
1628 }
1629 if (mddev->level != LEVEL_MULTIPATH) {
1630 int role;
1631 if (rdev->desc_nr < 0 ||
1632 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1633 role = MD_DISK_ROLE_SPARE;
1634 rdev->desc_nr = -1;
1635 } else
1636 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1637 switch(role) {
1638 case MD_DISK_ROLE_SPARE: /* spare */
1639 break;
1640 case MD_DISK_ROLE_FAULTY: /* faulty */
1641 set_bit(Faulty, &rdev->flags);
1642 break;
1643 case MD_DISK_ROLE_JOURNAL: /* journal device */
1644 if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1645 /* journal device without journal feature */
1646 printk(KERN_WARNING
1647 "md: journal device provided without journal feature, ignoring the device\n");
1648 return -EINVAL;
1649 }
1650 set_bit(Journal, &rdev->flags);
1651 rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1652 rdev->raid_disk = 0;
1653 break;
1654 default:
1655 rdev->saved_raid_disk = role;
1656 if ((le32_to_cpu(sb->feature_map) &
1657 MD_FEATURE_RECOVERY_OFFSET)) {
1658 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1659 if (!(le32_to_cpu(sb->feature_map) &
1660 MD_FEATURE_RECOVERY_BITMAP))
1661 rdev->saved_raid_disk = -1;
1662 } else
1663 set_bit(In_sync, &rdev->flags);
1664 rdev->raid_disk = role;
1665 break;
1666 }
1667 if (sb->devflags & WriteMostly1)
1668 set_bit(WriteMostly, &rdev->flags);
1669 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1670 set_bit(Replacement, &rdev->flags);
1671 } else /* MULTIPATH are always insync */
1672 set_bit(In_sync, &rdev->flags);
1673
1674 return 0;
1675 }
1676
1677 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1678 {
1679 struct mdp_superblock_1 *sb;
1680 struct md_rdev *rdev2;
1681 int max_dev, i;
1682 /* make rdev->sb match mddev and rdev data. */
1683
1684 sb = page_address(rdev->sb_page);
1685
1686 sb->feature_map = 0;
1687 sb->pad0 = 0;
1688 sb->recovery_offset = cpu_to_le64(0);
1689 memset(sb->pad3, 0, sizeof(sb->pad3));
1690
1691 sb->utime = cpu_to_le64((__u64)mddev->utime);
1692 sb->events = cpu_to_le64(mddev->events);
1693 if (mddev->in_sync)
1694 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1695 else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
1696 sb->resync_offset = cpu_to_le64(MaxSector);
1697 else
1698 sb->resync_offset = cpu_to_le64(0);
1699
1700 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1701
1702 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1703 sb->size = cpu_to_le64(mddev->dev_sectors);
1704 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1705 sb->level = cpu_to_le32(mddev->level);
1706 sb->layout = cpu_to_le32(mddev->layout);
1707
1708 if (test_bit(WriteMostly, &rdev->flags))
1709 sb->devflags |= WriteMostly1;
1710 else
1711 sb->devflags &= ~WriteMostly1;
1712 sb->data_offset = cpu_to_le64(rdev->data_offset);
1713 sb->data_size = cpu_to_le64(rdev->sectors);
1714
1715 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1716 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1717 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1718 }
1719
1720 if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
1721 !test_bit(In_sync, &rdev->flags)) {
1722 sb->feature_map |=
1723 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1724 sb->recovery_offset =
1725 cpu_to_le64(rdev->recovery_offset);
1726 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1727 sb->feature_map |=
1728 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
1729 }
1730 /* Note: recovery_offset and journal_tail share space */
1731 if (test_bit(Journal, &rdev->flags))
1732 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
1733 if (test_bit(Replacement, &rdev->flags))
1734 sb->feature_map |=
1735 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1736
1737 if (mddev->reshape_position != MaxSector) {
1738 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1739 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1740 sb->new_layout = cpu_to_le32(mddev->new_layout);
1741 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1742 sb->new_level = cpu_to_le32(mddev->new_level);
1743 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1744 if (mddev->delta_disks == 0 &&
1745 mddev->reshape_backwards)
1746 sb->feature_map
1747 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1748 if (rdev->new_data_offset != rdev->data_offset) {
1749 sb->feature_map
1750 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1751 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1752 - rdev->data_offset));
1753 }
1754 }
1755
1756 if (mddev_is_clustered(mddev))
1757 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
1758
1759 if (rdev->badblocks.count == 0)
1760 /* Nothing to do for bad blocks*/ ;
1761 else if (sb->bblog_offset == 0)
1762 /* Cannot record bad blocks on this device */
1763 md_error(mddev, rdev);
1764 else {
1765 struct badblocks *bb = &rdev->badblocks;
1766 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1767 u64 *p = bb->page;
1768 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1769 if (bb->changed) {
1770 unsigned seq;
1771
1772 retry:
1773 seq = read_seqbegin(&bb->lock);
1774
1775 memset(bbp, 0xff, PAGE_SIZE);
1776
1777 for (i = 0 ; i < bb->count ; i++) {
1778 u64 internal_bb = p[i];
1779 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1780 | BB_LEN(internal_bb));
1781 bbp[i] = cpu_to_le64(store_bb);
1782 }
1783 bb->changed = 0;
1784 if (read_seqretry(&bb->lock, seq))
1785 goto retry;
1786
1787 bb->sector = (rdev->sb_start +
1788 (int)le32_to_cpu(sb->bblog_offset));
1789 bb->size = le16_to_cpu(sb->bblog_size);
1790 }
1791 }
1792
1793 max_dev = 0;
1794 rdev_for_each(rdev2, mddev)
1795 if (rdev2->desc_nr+1 > max_dev)
1796 max_dev = rdev2->desc_nr+1;
1797
1798 if (max_dev > le32_to_cpu(sb->max_dev)) {
1799 int bmask;
1800 sb->max_dev = cpu_to_le32(max_dev);
1801 rdev->sb_size = max_dev * 2 + 256;
1802 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1803 if (rdev->sb_size & bmask)
1804 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1805 } else
1806 max_dev = le32_to_cpu(sb->max_dev);
1807
1808 for (i=0; i<max_dev;i++)
1809 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
1810
1811 if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
1812 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
1813
1814 rdev_for_each(rdev2, mddev) {
1815 i = rdev2->desc_nr;
1816 if (test_bit(Faulty, &rdev2->flags))
1817 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
1818 else if (test_bit(In_sync, &rdev2->flags))
1819 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1820 else if (test_bit(Journal, &rdev2->flags))
1821 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
1822 else if (rdev2->raid_disk >= 0)
1823 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1824 else
1825 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
1826 }
1827
1828 sb->sb_csum = calc_sb_1_csum(sb);
1829 }
1830
1831 static unsigned long long
1832 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1833 {
1834 struct mdp_superblock_1 *sb;
1835 sector_t max_sectors;
1836 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1837 return 0; /* component must fit device */
1838 if (rdev->data_offset != rdev->new_data_offset)
1839 return 0; /* too confusing */
1840 if (rdev->sb_start < rdev->data_offset) {
1841 /* minor versions 1 and 2; superblock before data */
1842 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1843 max_sectors -= rdev->data_offset;
1844 if (!num_sectors || num_sectors > max_sectors)
1845 num_sectors = max_sectors;
1846 } else if (rdev->mddev->bitmap_info.offset) {
1847 /* minor version 0 with bitmap we can't move */
1848 return 0;
1849 } else {
1850 /* minor version 0; superblock after data */
1851 sector_t sb_start;
1852 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1853 sb_start &= ~(sector_t)(4*2 - 1);
1854 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1855 if (!num_sectors || num_sectors > max_sectors)
1856 num_sectors = max_sectors;
1857 rdev->sb_start = sb_start;
1858 }
1859 sb = page_address(rdev->sb_page);
1860 sb->data_size = cpu_to_le64(num_sectors);
1861 sb->super_offset = rdev->sb_start;
1862 sb->sb_csum = calc_sb_1_csum(sb);
1863 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1864 rdev->sb_page);
1865 md_super_wait(rdev->mddev);
1866 return num_sectors;
1867
1868 }
1869
1870 static int
1871 super_1_allow_new_offset(struct md_rdev *rdev,
1872 unsigned long long new_offset)
1873 {
1874 /* All necessary checks on new >= old have been done */
1875 struct bitmap *bitmap;
1876 if (new_offset >= rdev->data_offset)
1877 return 1;
1878
1879 /* with 1.0 metadata, there is no metadata to tread on
1880 * so we can always move back */
1881 if (rdev->mddev->minor_version == 0)
1882 return 1;
1883
1884 /* otherwise we must be sure not to step on
1885 * any metadata, so stay:
1886 * 36K beyond start of superblock
1887 * beyond end of badblocks
1888 * beyond write-intent bitmap
1889 */
1890 if (rdev->sb_start + (32+4)*2 > new_offset)
1891 return 0;
1892 bitmap = rdev->mddev->bitmap;
1893 if (bitmap && !rdev->mddev->bitmap_info.file &&
1894 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1895 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
1896 return 0;
1897 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1898 return 0;
1899
1900 return 1;
1901 }
1902
1903 static struct super_type super_types[] = {
1904 [0] = {
1905 .name = "0.90.0",
1906 .owner = THIS_MODULE,
1907 .load_super = super_90_load,
1908 .validate_super = super_90_validate,
1909 .sync_super = super_90_sync,
1910 .rdev_size_change = super_90_rdev_size_change,
1911 .allow_new_offset = super_90_allow_new_offset,
1912 },
1913 [1] = {
1914 .name = "md-1",
1915 .owner = THIS_MODULE,
1916 .load_super = super_1_load,
1917 .validate_super = super_1_validate,
1918 .sync_super = super_1_sync,
1919 .rdev_size_change = super_1_rdev_size_change,
1920 .allow_new_offset = super_1_allow_new_offset,
1921 },
1922 };
1923
1924 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1925 {
1926 if (mddev->sync_super) {
1927 mddev->sync_super(mddev, rdev);
1928 return;
1929 }
1930
1931 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1932
1933 super_types[mddev->major_version].sync_super(mddev, rdev);
1934 }
1935
1936 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1937 {
1938 struct md_rdev *rdev, *rdev2;
1939
1940 rcu_read_lock();
1941 rdev_for_each_rcu(rdev, mddev1) {
1942 if (test_bit(Faulty, &rdev->flags) ||
1943 test_bit(Journal, &rdev->flags) ||
1944 rdev->raid_disk == -1)
1945 continue;
1946 rdev_for_each_rcu(rdev2, mddev2) {
1947 if (test_bit(Faulty, &rdev2->flags) ||
1948 test_bit(Journal, &rdev2->flags) ||
1949 rdev2->raid_disk == -1)
1950 continue;
1951 if (rdev->bdev->bd_contains ==
1952 rdev2->bdev->bd_contains) {
1953 rcu_read_unlock();
1954 return 1;
1955 }
1956 }
1957 }
1958 rcu_read_unlock();
1959 return 0;
1960 }
1961
1962 static LIST_HEAD(pending_raid_disks);
1963
1964 /*
1965 * Try to register data integrity profile for an mddev
1966 *
1967 * This is called when an array is started and after a disk has been kicked
1968 * from the array. It only succeeds if all working and active component devices
1969 * are integrity capable with matching profiles.
1970 */
1971 int md_integrity_register(struct mddev *mddev)
1972 {
1973 struct md_rdev *rdev, *reference = NULL;
1974
1975 if (list_empty(&mddev->disks))
1976 return 0; /* nothing to do */
1977 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1978 return 0; /* shouldn't register, or already is */
1979 rdev_for_each(rdev, mddev) {
1980 /* skip spares and non-functional disks */
1981 if (test_bit(Faulty, &rdev->flags))
1982 continue;
1983 if (rdev->raid_disk < 0)
1984 continue;
1985 if (!reference) {
1986 /* Use the first rdev as the reference */
1987 reference = rdev;
1988 continue;
1989 }
1990 /* does this rdev's profile match the reference profile? */
1991 if (blk_integrity_compare(reference->bdev->bd_disk,
1992 rdev->bdev->bd_disk) < 0)
1993 return -EINVAL;
1994 }
1995 if (!reference || !bdev_get_integrity(reference->bdev))
1996 return 0;
1997 /*
1998 * All component devices are integrity capable and have matching
1999 * profiles, register the common profile for the md device.
2000 */
2001 blk_integrity_register(mddev->gendisk,
2002 bdev_get_integrity(reference->bdev));
2003
2004 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
2005 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
2006 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
2007 mdname(mddev));
2008 return -EINVAL;
2009 }
2010 return 0;
2011 }
2012 EXPORT_SYMBOL(md_integrity_register);
2013
2014 /*
2015 * Attempt to add an rdev, but only if it is consistent with the current
2016 * integrity profile
2017 */
2018 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2019 {
2020 struct blk_integrity *bi_rdev;
2021 struct blk_integrity *bi_mddev;
2022 char name[BDEVNAME_SIZE];
2023
2024 if (!mddev->gendisk)
2025 return 0;
2026
2027 bi_rdev = bdev_get_integrity(rdev->bdev);
2028 bi_mddev = blk_get_integrity(mddev->gendisk);
2029
2030 if (!bi_mddev) /* nothing to do */
2031 return 0;
2032
2033 if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2034 printk(KERN_NOTICE "%s: incompatible integrity profile for %s\n",
2035 mdname(mddev), bdevname(rdev->bdev, name));
2036 return -ENXIO;
2037 }
2038
2039 return 0;
2040 }
2041 EXPORT_SYMBOL(md_integrity_add_rdev);
2042
2043 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2044 {
2045 char b[BDEVNAME_SIZE];
2046 struct kobject *ko;
2047 int err;
2048
2049 /* prevent duplicates */
2050 if (find_rdev(mddev, rdev->bdev->bd_dev))
2051 return -EEXIST;
2052
2053 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2054 if (!test_bit(Journal, &rdev->flags) &&
2055 rdev->sectors &&
2056 (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2057 if (mddev->pers) {
2058 /* Cannot change size, so fail
2059 * If mddev->level <= 0, then we don't care
2060 * about aligning sizes (e.g. linear)
2061 */
2062 if (mddev->level > 0)
2063 return -ENOSPC;
2064 } else
2065 mddev->dev_sectors = rdev->sectors;
2066 }
2067
2068 /* Verify rdev->desc_nr is unique.
2069 * If it is -1, assign a free number, else
2070 * check number is not in use
2071 */
2072 rcu_read_lock();
2073 if (rdev->desc_nr < 0) {
2074 int choice = 0;
2075 if (mddev->pers)
2076 choice = mddev->raid_disks;
2077 while (md_find_rdev_nr_rcu(mddev, choice))
2078 choice++;
2079 rdev->desc_nr = choice;
2080 } else {
2081 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2082 rcu_read_unlock();
2083 return -EBUSY;
2084 }
2085 }
2086 rcu_read_unlock();
2087 if (!test_bit(Journal, &rdev->flags) &&
2088 mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2089 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2090 mdname(mddev), mddev->max_disks);
2091 return -EBUSY;
2092 }
2093 bdevname(rdev->bdev,b);
2094 strreplace(b, '/', '!');
2095
2096 rdev->mddev = mddev;
2097 printk(KERN_INFO "md: bind<%s>\n", b);
2098
2099 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2100 goto fail;
2101
2102 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2103 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2104 /* failure here is OK */;
2105 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2106
2107 list_add_rcu(&rdev->same_set, &mddev->disks);
2108 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2109
2110 /* May as well allow recovery to be retried once */
2111 mddev->recovery_disabled++;
2112
2113 return 0;
2114
2115 fail:
2116 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2117 b, mdname(mddev));
2118 return err;
2119 }
2120
2121 static void md_delayed_delete(struct work_struct *ws)
2122 {
2123 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2124 kobject_del(&rdev->kobj);
2125 kobject_put(&rdev->kobj);
2126 }
2127
2128 static void unbind_rdev_from_array(struct md_rdev *rdev)
2129 {
2130 char b[BDEVNAME_SIZE];
2131
2132 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2133 list_del_rcu(&rdev->same_set);
2134 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2135 rdev->mddev = NULL;
2136 sysfs_remove_link(&rdev->kobj, "block");
2137 sysfs_put(rdev->sysfs_state);
2138 rdev->sysfs_state = NULL;
2139 rdev->badblocks.count = 0;
2140 /* We need to delay this, otherwise we can deadlock when
2141 * writing to 'remove' to "dev/state". We also need
2142 * to delay it due to rcu usage.
2143 */
2144 synchronize_rcu();
2145 INIT_WORK(&rdev->del_work, md_delayed_delete);
2146 kobject_get(&rdev->kobj);
2147 queue_work(md_misc_wq, &rdev->del_work);
2148 }
2149
2150 /*
2151 * prevent the device from being mounted, repartitioned or
2152 * otherwise reused by a RAID array (or any other kernel
2153 * subsystem), by bd_claiming the device.
2154 */
2155 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2156 {
2157 int err = 0;
2158 struct block_device *bdev;
2159 char b[BDEVNAME_SIZE];
2160
2161 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2162 shared ? (struct md_rdev *)lock_rdev : rdev);
2163 if (IS_ERR(bdev)) {
2164 printk(KERN_ERR "md: could not open %s.\n",
2165 __bdevname(dev, b));
2166 return PTR_ERR(bdev);
2167 }
2168 rdev->bdev = bdev;
2169 return err;
2170 }
2171
2172 static void unlock_rdev(struct md_rdev *rdev)
2173 {
2174 struct block_device *bdev = rdev->bdev;
2175 rdev->bdev = NULL;
2176 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2177 }
2178
2179 void md_autodetect_dev(dev_t dev);
2180
2181 static void export_rdev(struct md_rdev *rdev)
2182 {
2183 char b[BDEVNAME_SIZE];
2184
2185 printk(KERN_INFO "md: export_rdev(%s)\n",
2186 bdevname(rdev->bdev,b));
2187 md_rdev_clear(rdev);
2188 #ifndef MODULE
2189 if (test_bit(AutoDetected, &rdev->flags))
2190 md_autodetect_dev(rdev->bdev->bd_dev);
2191 #endif
2192 unlock_rdev(rdev);
2193 kobject_put(&rdev->kobj);
2194 }
2195
2196 void md_kick_rdev_from_array(struct md_rdev *rdev)
2197 {
2198 unbind_rdev_from_array(rdev);
2199 export_rdev(rdev);
2200 }
2201 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2202
2203 static void export_array(struct mddev *mddev)
2204 {
2205 struct md_rdev *rdev;
2206
2207 while (!list_empty(&mddev->disks)) {
2208 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2209 same_set);
2210 md_kick_rdev_from_array(rdev);
2211 }
2212 mddev->raid_disks = 0;
2213 mddev->major_version = 0;
2214 }
2215
2216 static void sync_sbs(struct mddev *mddev, int nospares)
2217 {
2218 /* Update each superblock (in-memory image), but
2219 * if we are allowed to, skip spares which already
2220 * have the right event counter, or have one earlier
2221 * (which would mean they aren't being marked as dirty
2222 * with the rest of the array)
2223 */
2224 struct md_rdev *rdev;
2225 rdev_for_each(rdev, mddev) {
2226 if (rdev->sb_events == mddev->events ||
2227 (nospares &&
2228 rdev->raid_disk < 0 &&
2229 rdev->sb_events+1 == mddev->events)) {
2230 /* Don't update this superblock */
2231 rdev->sb_loaded = 2;
2232 } else {
2233 sync_super(mddev, rdev);
2234 rdev->sb_loaded = 1;
2235 }
2236 }
2237 }
2238
2239 static bool does_sb_need_changing(struct mddev *mddev)
2240 {
2241 struct md_rdev *rdev;
2242 struct mdp_superblock_1 *sb;
2243 int role;
2244
2245 /* Find a good rdev */
2246 rdev_for_each(rdev, mddev)
2247 if ((rdev->raid_disk >= 0) && !test_bit(Faulty, &rdev->flags))
2248 break;
2249
2250 /* No good device found. */
2251 if (!rdev)
2252 return false;
2253
2254 sb = page_address(rdev->sb_page);
2255 /* Check if a device has become faulty or a spare become active */
2256 rdev_for_each(rdev, mddev) {
2257 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2258 /* Device activated? */
2259 if (role == 0xffff && rdev->raid_disk >=0 &&
2260 !test_bit(Faulty, &rdev->flags))
2261 return true;
2262 /* Device turned faulty? */
2263 if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd))
2264 return true;
2265 }
2266
2267 /* Check if any mddev parameters have changed */
2268 if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2269 (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2270 (mddev->layout != le64_to_cpu(sb->layout)) ||
2271 (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2272 (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2273 return true;
2274
2275 return false;
2276 }
2277
2278 void md_update_sb(struct mddev *mddev, int force_change)
2279 {
2280 struct md_rdev *rdev;
2281 int sync_req;
2282 int nospares = 0;
2283 int any_badblocks_changed = 0;
2284 int ret = -1;
2285
2286 if (mddev->ro) {
2287 if (force_change)
2288 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2289 return;
2290 }
2291
2292 if (mddev_is_clustered(mddev)) {
2293 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2294 force_change = 1;
2295 ret = md_cluster_ops->metadata_update_start(mddev);
2296 /* Has someone else has updated the sb */
2297 if (!does_sb_need_changing(mddev)) {
2298 if (ret == 0)
2299 md_cluster_ops->metadata_update_cancel(mddev);
2300 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2301 return;
2302 }
2303 }
2304 repeat:
2305 /* First make sure individual recovery_offsets are correct */
2306 rdev_for_each(rdev, mddev) {
2307 if (rdev->raid_disk >= 0 &&
2308 mddev->delta_disks >= 0 &&
2309 !test_bit(Journal, &rdev->flags) &&
2310 !test_bit(In_sync, &rdev->flags) &&
2311 mddev->curr_resync_completed > rdev->recovery_offset)
2312 rdev->recovery_offset = mddev->curr_resync_completed;
2313
2314 }
2315 if (!mddev->persistent) {
2316 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2317 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2318 if (!mddev->external) {
2319 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2320 rdev_for_each(rdev, mddev) {
2321 if (rdev->badblocks.changed) {
2322 rdev->badblocks.changed = 0;
2323 ack_all_badblocks(&rdev->badblocks);
2324 md_error(mddev, rdev);
2325 }
2326 clear_bit(Blocked, &rdev->flags);
2327 clear_bit(BlockedBadBlocks, &rdev->flags);
2328 wake_up(&rdev->blocked_wait);
2329 }
2330 }
2331 wake_up(&mddev->sb_wait);
2332 return;
2333 }
2334
2335 spin_lock(&mddev->lock);
2336
2337 mddev->utime = ktime_get_real_seconds();
2338
2339 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2340 force_change = 1;
2341 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2342 /* just a clean<-> dirty transition, possibly leave spares alone,
2343 * though if events isn't the right even/odd, we will have to do
2344 * spares after all
2345 */
2346 nospares = 1;
2347 if (force_change)
2348 nospares = 0;
2349 if (mddev->degraded)
2350 /* If the array is degraded, then skipping spares is both
2351 * dangerous and fairly pointless.
2352 * Dangerous because a device that was removed from the array
2353 * might have a event_count that still looks up-to-date,
2354 * so it can be re-added without a resync.
2355 * Pointless because if there are any spares to skip,
2356 * then a recovery will happen and soon that array won't
2357 * be degraded any more and the spare can go back to sleep then.
2358 */
2359 nospares = 0;
2360
2361 sync_req = mddev->in_sync;
2362
2363 /* If this is just a dirty<->clean transition, and the array is clean
2364 * and 'events' is odd, we can roll back to the previous clean state */
2365 if (nospares
2366 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2367 && mddev->can_decrease_events
2368 && mddev->events != 1) {
2369 mddev->events--;
2370 mddev->can_decrease_events = 0;
2371 } else {
2372 /* otherwise we have to go forward and ... */
2373 mddev->events ++;
2374 mddev->can_decrease_events = nospares;
2375 }
2376
2377 /*
2378 * This 64-bit counter should never wrap.
2379 * Either we are in around ~1 trillion A.C., assuming
2380 * 1 reboot per second, or we have a bug...
2381 */
2382 WARN_ON(mddev->events == 0);
2383
2384 rdev_for_each(rdev, mddev) {
2385 if (rdev->badblocks.changed)
2386 any_badblocks_changed++;
2387 if (test_bit(Faulty, &rdev->flags))
2388 set_bit(FaultRecorded, &rdev->flags);
2389 }
2390
2391 sync_sbs(mddev, nospares);
2392 spin_unlock(&mddev->lock);
2393
2394 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2395 mdname(mddev), mddev->in_sync);
2396
2397 bitmap_update_sb(mddev->bitmap);
2398 rdev_for_each(rdev, mddev) {
2399 char b[BDEVNAME_SIZE];
2400
2401 if (rdev->sb_loaded != 1)
2402 continue; /* no noise on spare devices */
2403
2404 if (!test_bit(Faulty, &rdev->flags)) {
2405 md_super_write(mddev,rdev,
2406 rdev->sb_start, rdev->sb_size,
2407 rdev->sb_page);
2408 pr_debug("md: (write) %s's sb offset: %llu\n",
2409 bdevname(rdev->bdev, b),
2410 (unsigned long long)rdev->sb_start);
2411 rdev->sb_events = mddev->events;
2412 if (rdev->badblocks.size) {
2413 md_super_write(mddev, rdev,
2414 rdev->badblocks.sector,
2415 rdev->badblocks.size << 9,
2416 rdev->bb_page);
2417 rdev->badblocks.size = 0;
2418 }
2419
2420 } else
2421 pr_debug("md: %s (skipping faulty)\n",
2422 bdevname(rdev->bdev, b));
2423
2424 if (mddev->level == LEVEL_MULTIPATH)
2425 /* only need to write one superblock... */
2426 break;
2427 }
2428 md_super_wait(mddev);
2429 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2430
2431 spin_lock(&mddev->lock);
2432 if (mddev->in_sync != sync_req ||
2433 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2434 /* have to write it out again */
2435 spin_unlock(&mddev->lock);
2436 goto repeat;
2437 }
2438 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2439 spin_unlock(&mddev->lock);
2440 wake_up(&mddev->sb_wait);
2441 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2442 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2443
2444 rdev_for_each(rdev, mddev) {
2445 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2446 clear_bit(Blocked, &rdev->flags);
2447
2448 if (any_badblocks_changed)
2449 ack_all_badblocks(&rdev->badblocks);
2450 clear_bit(BlockedBadBlocks, &rdev->flags);
2451 wake_up(&rdev->blocked_wait);
2452 }
2453
2454 if (mddev_is_clustered(mddev) && ret == 0)
2455 md_cluster_ops->metadata_update_finish(mddev);
2456 }
2457 EXPORT_SYMBOL(md_update_sb);
2458
2459 static int add_bound_rdev(struct md_rdev *rdev)
2460 {
2461 struct mddev *mddev = rdev->mddev;
2462 int err = 0;
2463 bool add_journal = test_bit(Journal, &rdev->flags);
2464
2465 if (!mddev->pers->hot_remove_disk || add_journal) {
2466 /* If there is hot_add_disk but no hot_remove_disk
2467 * then added disks for geometry changes,
2468 * and should be added immediately.
2469 */
2470 super_types[mddev->major_version].
2471 validate_super(mddev, rdev);
2472 if (add_journal)
2473 mddev_suspend(mddev);
2474 err = mddev->pers->hot_add_disk(mddev, rdev);
2475 if (add_journal)
2476 mddev_resume(mddev);
2477 if (err) {
2478 unbind_rdev_from_array(rdev);
2479 export_rdev(rdev);
2480 return err;
2481 }
2482 }
2483 sysfs_notify_dirent_safe(rdev->sysfs_state);
2484
2485 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2486 if (mddev->degraded)
2487 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2488 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2489 md_new_event(mddev);
2490 md_wakeup_thread(mddev->thread);
2491 return 0;
2492 }
2493
2494 /* words written to sysfs files may, or may not, be \n terminated.
2495 * We want to accept with case. For this we use cmd_match.
2496 */
2497 static int cmd_match(const char *cmd, const char *str)
2498 {
2499 /* See if cmd, written into a sysfs file, matches
2500 * str. They must either be the same, or cmd can
2501 * have a trailing newline
2502 */
2503 while (*cmd && *str && *cmd == *str) {
2504 cmd++;
2505 str++;
2506 }
2507 if (*cmd == '\n')
2508 cmd++;
2509 if (*str || *cmd)
2510 return 0;
2511 return 1;
2512 }
2513
2514 struct rdev_sysfs_entry {
2515 struct attribute attr;
2516 ssize_t (*show)(struct md_rdev *, char *);
2517 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2518 };
2519
2520 static ssize_t
2521 state_show(struct md_rdev *rdev, char *page)
2522 {
2523 char *sep = "";
2524 size_t len = 0;
2525 unsigned long flags = ACCESS_ONCE(rdev->flags);
2526
2527 if (test_bit(Faulty, &flags) ||
2528 rdev->badblocks.unacked_exist) {
2529 len+= sprintf(page+len, "%sfaulty",sep);
2530 sep = ",";
2531 }
2532 if (test_bit(In_sync, &flags)) {
2533 len += sprintf(page+len, "%sin_sync",sep);
2534 sep = ",";
2535 }
2536 if (test_bit(Journal, &flags)) {
2537 len += sprintf(page+len, "%sjournal",sep);
2538 sep = ",";
2539 }
2540 if (test_bit(WriteMostly, &flags)) {
2541 len += sprintf(page+len, "%swrite_mostly",sep);
2542 sep = ",";
2543 }
2544 if (test_bit(Blocked, &flags) ||
2545 (rdev->badblocks.unacked_exist
2546 && !test_bit(Faulty, &flags))) {
2547 len += sprintf(page+len, "%sblocked", sep);
2548 sep = ",";
2549 }
2550 if (!test_bit(Faulty, &flags) &&
2551 !test_bit(Journal, &flags) &&
2552 !test_bit(In_sync, &flags)) {
2553 len += sprintf(page+len, "%sspare", sep);
2554 sep = ",";
2555 }
2556 if (test_bit(WriteErrorSeen, &flags)) {
2557 len += sprintf(page+len, "%swrite_error", sep);
2558 sep = ",";
2559 }
2560 if (test_bit(WantReplacement, &flags)) {
2561 len += sprintf(page+len, "%swant_replacement", sep);
2562 sep = ",";
2563 }
2564 if (test_bit(Replacement, &flags)) {
2565 len += sprintf(page+len, "%sreplacement", sep);
2566 sep = ",";
2567 }
2568
2569 return len+sprintf(page+len, "\n");
2570 }
2571
2572 static ssize_t
2573 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2574 {
2575 /* can write
2576 * faulty - simulates an error
2577 * remove - disconnects the device
2578 * writemostly - sets write_mostly
2579 * -writemostly - clears write_mostly
2580 * blocked - sets the Blocked flags
2581 * -blocked - clears the Blocked and possibly simulates an error
2582 * insync - sets Insync providing device isn't active
2583 * -insync - clear Insync for a device with a slot assigned,
2584 * so that it gets rebuilt based on bitmap
2585 * write_error - sets WriteErrorSeen
2586 * -write_error - clears WriteErrorSeen
2587 */
2588 int err = -EINVAL;
2589 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2590 md_error(rdev->mddev, rdev);
2591 if (test_bit(Faulty, &rdev->flags))
2592 err = 0;
2593 else
2594 err = -EBUSY;
2595 } else if (cmd_match(buf, "remove")) {
2596 if (rdev->raid_disk >= 0)
2597 err = -EBUSY;
2598 else {
2599 struct mddev *mddev = rdev->mddev;
2600 err = 0;
2601 if (mddev_is_clustered(mddev))
2602 err = md_cluster_ops->remove_disk(mddev, rdev);
2603
2604 if (err == 0) {
2605 md_kick_rdev_from_array(rdev);
2606 if (mddev->pers)
2607 md_update_sb(mddev, 1);
2608 md_new_event(mddev);
2609 }
2610 }
2611 } else if (cmd_match(buf, "writemostly")) {
2612 set_bit(WriteMostly, &rdev->flags);
2613 err = 0;
2614 } else if (cmd_match(buf, "-writemostly")) {
2615 clear_bit(WriteMostly, &rdev->flags);
2616 err = 0;
2617 } else if (cmd_match(buf, "blocked")) {
2618 set_bit(Blocked, &rdev->flags);
2619 err = 0;
2620 } else if (cmd_match(buf, "-blocked")) {
2621 if (!test_bit(Faulty, &rdev->flags) &&
2622 rdev->badblocks.unacked_exist) {
2623 /* metadata handler doesn't understand badblocks,
2624 * so we need to fail the device
2625 */
2626 md_error(rdev->mddev, rdev);
2627 }
2628 clear_bit(Blocked, &rdev->flags);
2629 clear_bit(BlockedBadBlocks, &rdev->flags);
2630 wake_up(&rdev->blocked_wait);
2631 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2632 md_wakeup_thread(rdev->mddev->thread);
2633
2634 err = 0;
2635 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2636 set_bit(In_sync, &rdev->flags);
2637 err = 0;
2638 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
2639 !test_bit(Journal, &rdev->flags)) {
2640 if (rdev->mddev->pers == NULL) {
2641 clear_bit(In_sync, &rdev->flags);
2642 rdev->saved_raid_disk = rdev->raid_disk;
2643 rdev->raid_disk = -1;
2644 err = 0;
2645 }
2646 } else if (cmd_match(buf, "write_error")) {
2647 set_bit(WriteErrorSeen, &rdev->flags);
2648 err = 0;
2649 } else if (cmd_match(buf, "-write_error")) {
2650 clear_bit(WriteErrorSeen, &rdev->flags);
2651 err = 0;
2652 } else if (cmd_match(buf, "want_replacement")) {
2653 /* Any non-spare device that is not a replacement can
2654 * become want_replacement at any time, but we then need to
2655 * check if recovery is needed.
2656 */
2657 if (rdev->raid_disk >= 0 &&
2658 !test_bit(Journal, &rdev->flags) &&
2659 !test_bit(Replacement, &rdev->flags))
2660 set_bit(WantReplacement, &rdev->flags);
2661 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2662 md_wakeup_thread(rdev->mddev->thread);
2663 err = 0;
2664 } else if (cmd_match(buf, "-want_replacement")) {
2665 /* Clearing 'want_replacement' is always allowed.
2666 * Once replacements starts it is too late though.
2667 */
2668 err = 0;
2669 clear_bit(WantReplacement, &rdev->flags);
2670 } else if (cmd_match(buf, "replacement")) {
2671 /* Can only set a device as a replacement when array has not
2672 * yet been started. Once running, replacement is automatic
2673 * from spares, or by assigning 'slot'.
2674 */
2675 if (rdev->mddev->pers)
2676 err = -EBUSY;
2677 else {
2678 set_bit(Replacement, &rdev->flags);
2679 err = 0;
2680 }
2681 } else if (cmd_match(buf, "-replacement")) {
2682 /* Similarly, can only clear Replacement before start */
2683 if (rdev->mddev->pers)
2684 err = -EBUSY;
2685 else {
2686 clear_bit(Replacement, &rdev->flags);
2687 err = 0;
2688 }
2689 } else if (cmd_match(buf, "re-add")) {
2690 if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1)) {
2691 /* clear_bit is performed _after_ all the devices
2692 * have their local Faulty bit cleared. If any writes
2693 * happen in the meantime in the local node, they
2694 * will land in the local bitmap, which will be synced
2695 * by this node eventually
2696 */
2697 if (!mddev_is_clustered(rdev->mddev) ||
2698 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
2699 clear_bit(Faulty, &rdev->flags);
2700 err = add_bound_rdev(rdev);
2701 }
2702 } else
2703 err = -EBUSY;
2704 }
2705 if (!err)
2706 sysfs_notify_dirent_safe(rdev->sysfs_state);
2707 return err ? err : len;
2708 }
2709 static struct rdev_sysfs_entry rdev_state =
2710 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
2711
2712 static ssize_t
2713 errors_show(struct md_rdev *rdev, char *page)
2714 {
2715 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2716 }
2717
2718 static ssize_t
2719 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2720 {
2721 unsigned int n;
2722 int rv;
2723
2724 rv = kstrtouint(buf, 10, &n);
2725 if (rv < 0)
2726 return rv;
2727 atomic_set(&rdev->corrected_errors, n);
2728 return len;
2729 }
2730 static struct rdev_sysfs_entry rdev_errors =
2731 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2732
2733 static ssize_t
2734 slot_show(struct md_rdev *rdev, char *page)
2735 {
2736 if (test_bit(Journal, &rdev->flags))
2737 return sprintf(page, "journal\n");
2738 else if (rdev->raid_disk < 0)
2739 return sprintf(page, "none\n");
2740 else
2741 return sprintf(page, "%d\n", rdev->raid_disk);
2742 }
2743
2744 static ssize_t
2745 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2746 {
2747 int slot;
2748 int err;
2749
2750 if (test_bit(Journal, &rdev->flags))
2751 return -EBUSY;
2752 if (strncmp(buf, "none", 4)==0)
2753 slot = -1;
2754 else {
2755 err = kstrtouint(buf, 10, (unsigned int *)&slot);
2756 if (err < 0)
2757 return err;
2758 }
2759 if (rdev->mddev->pers && slot == -1) {
2760 /* Setting 'slot' on an active array requires also
2761 * updating the 'rd%d' link, and communicating
2762 * with the personality with ->hot_*_disk.
2763 * For now we only support removing
2764 * failed/spare devices. This normally happens automatically,
2765 * but not when the metadata is externally managed.
2766 */
2767 if (rdev->raid_disk == -1)
2768 return -EEXIST;
2769 /* personality does all needed checks */
2770 if (rdev->mddev->pers->hot_remove_disk == NULL)
2771 return -EINVAL;
2772 clear_bit(Blocked, &rdev->flags);
2773 remove_and_add_spares(rdev->mddev, rdev);
2774 if (rdev->raid_disk >= 0)
2775 return -EBUSY;
2776 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2777 md_wakeup_thread(rdev->mddev->thread);
2778 } else if (rdev->mddev->pers) {
2779 /* Activating a spare .. or possibly reactivating
2780 * if we ever get bitmaps working here.
2781 */
2782 int err;
2783
2784 if (rdev->raid_disk != -1)
2785 return -EBUSY;
2786
2787 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2788 return -EBUSY;
2789
2790 if (rdev->mddev->pers->hot_add_disk == NULL)
2791 return -EINVAL;
2792
2793 if (slot >= rdev->mddev->raid_disks &&
2794 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2795 return -ENOSPC;
2796
2797 rdev->raid_disk = slot;
2798 if (test_bit(In_sync, &rdev->flags))
2799 rdev->saved_raid_disk = slot;
2800 else
2801 rdev->saved_raid_disk = -1;
2802 clear_bit(In_sync, &rdev->flags);
2803 clear_bit(Bitmap_sync, &rdev->flags);
2804 err = rdev->mddev->pers->
2805 hot_add_disk(rdev->mddev, rdev);
2806 if (err) {
2807 rdev->raid_disk = -1;
2808 return err;
2809 } else
2810 sysfs_notify_dirent_safe(rdev->sysfs_state);
2811 if (sysfs_link_rdev(rdev->mddev, rdev))
2812 /* failure here is OK */;
2813 /* don't wakeup anyone, leave that to userspace. */
2814 } else {
2815 if (slot >= rdev->mddev->raid_disks &&
2816 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2817 return -ENOSPC;
2818 rdev->raid_disk = slot;
2819 /* assume it is working */
2820 clear_bit(Faulty, &rdev->flags);
2821 clear_bit(WriteMostly, &rdev->flags);
2822 set_bit(In_sync, &rdev->flags);
2823 sysfs_notify_dirent_safe(rdev->sysfs_state);
2824 }
2825 return len;
2826 }
2827
2828 static struct rdev_sysfs_entry rdev_slot =
2829 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2830
2831 static ssize_t
2832 offset_show(struct md_rdev *rdev, char *page)
2833 {
2834 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2835 }
2836
2837 static ssize_t
2838 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2839 {
2840 unsigned long long offset;
2841 if (kstrtoull(buf, 10, &offset) < 0)
2842 return -EINVAL;
2843 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2844 return -EBUSY;
2845 if (rdev->sectors && rdev->mddev->external)
2846 /* Must set offset before size, so overlap checks
2847 * can be sane */
2848 return -EBUSY;
2849 rdev->data_offset = offset;
2850 rdev->new_data_offset = offset;
2851 return len;
2852 }
2853
2854 static struct rdev_sysfs_entry rdev_offset =
2855 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2856
2857 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2858 {
2859 return sprintf(page, "%llu\n",
2860 (unsigned long long)rdev->new_data_offset);
2861 }
2862
2863 static ssize_t new_offset_store(struct md_rdev *rdev,
2864 const char *buf, size_t len)
2865 {
2866 unsigned long long new_offset;
2867 struct mddev *mddev = rdev->mddev;
2868
2869 if (kstrtoull(buf, 10, &new_offset) < 0)
2870 return -EINVAL;
2871
2872 if (mddev->sync_thread ||
2873 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
2874 return -EBUSY;
2875 if (new_offset == rdev->data_offset)
2876 /* reset is always permitted */
2877 ;
2878 else if (new_offset > rdev->data_offset) {
2879 /* must not push array size beyond rdev_sectors */
2880 if (new_offset - rdev->data_offset
2881 + mddev->dev_sectors > rdev->sectors)
2882 return -E2BIG;
2883 }
2884 /* Metadata worries about other space details. */
2885
2886 /* decreasing the offset is inconsistent with a backwards
2887 * reshape.
2888 */
2889 if (new_offset < rdev->data_offset &&
2890 mddev->reshape_backwards)
2891 return -EINVAL;
2892 /* Increasing offset is inconsistent with forwards
2893 * reshape. reshape_direction should be set to
2894 * 'backwards' first.
2895 */
2896 if (new_offset > rdev->data_offset &&
2897 !mddev->reshape_backwards)
2898 return -EINVAL;
2899
2900 if (mddev->pers && mddev->persistent &&
2901 !super_types[mddev->major_version]
2902 .allow_new_offset(rdev, new_offset))
2903 return -E2BIG;
2904 rdev->new_data_offset = new_offset;
2905 if (new_offset > rdev->data_offset)
2906 mddev->reshape_backwards = 1;
2907 else if (new_offset < rdev->data_offset)
2908 mddev->reshape_backwards = 0;
2909
2910 return len;
2911 }
2912 static struct rdev_sysfs_entry rdev_new_offset =
2913 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2914
2915 static ssize_t
2916 rdev_size_show(struct md_rdev *rdev, char *page)
2917 {
2918 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2919 }
2920
2921 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2922 {
2923 /* check if two start/length pairs overlap */
2924 if (s1+l1 <= s2)
2925 return 0;
2926 if (s2+l2 <= s1)
2927 return 0;
2928 return 1;
2929 }
2930
2931 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2932 {
2933 unsigned long long blocks;
2934 sector_t new;
2935
2936 if (kstrtoull(buf, 10, &blocks) < 0)
2937 return -EINVAL;
2938
2939 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2940 return -EINVAL; /* sector conversion overflow */
2941
2942 new = blocks * 2;
2943 if (new != blocks * 2)
2944 return -EINVAL; /* unsigned long long to sector_t overflow */
2945
2946 *sectors = new;
2947 return 0;
2948 }
2949
2950 static ssize_t
2951 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
2952 {
2953 struct mddev *my_mddev = rdev->mddev;
2954 sector_t oldsectors = rdev->sectors;
2955 sector_t sectors;
2956
2957 if (test_bit(Journal, &rdev->flags))
2958 return -EBUSY;
2959 if (strict_blocks_to_sectors(buf, &sectors) < 0)
2960 return -EINVAL;
2961 if (rdev->data_offset != rdev->new_data_offset)
2962 return -EINVAL; /* too confusing */
2963 if (my_mddev->pers && rdev->raid_disk >= 0) {
2964 if (my_mddev->persistent) {
2965 sectors = super_types[my_mddev->major_version].
2966 rdev_size_change(rdev, sectors);
2967 if (!sectors)
2968 return -EBUSY;
2969 } else if (!sectors)
2970 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2971 rdev->data_offset;
2972 if (!my_mddev->pers->resize)
2973 /* Cannot change size for RAID0 or Linear etc */
2974 return -EINVAL;
2975 }
2976 if (sectors < my_mddev->dev_sectors)
2977 return -EINVAL; /* component must fit device */
2978
2979 rdev->sectors = sectors;
2980 if (sectors > oldsectors && my_mddev->external) {
2981 /* Need to check that all other rdevs with the same
2982 * ->bdev do not overlap. 'rcu' is sufficient to walk
2983 * the rdev lists safely.
2984 * This check does not provide a hard guarantee, it
2985 * just helps avoid dangerous mistakes.
2986 */
2987 struct mddev *mddev;
2988 int overlap = 0;
2989 struct list_head *tmp;
2990
2991 rcu_read_lock();
2992 for_each_mddev(mddev, tmp) {
2993 struct md_rdev *rdev2;
2994
2995 rdev_for_each(rdev2, mddev)
2996 if (rdev->bdev == rdev2->bdev &&
2997 rdev != rdev2 &&
2998 overlaps(rdev->data_offset, rdev->sectors,
2999 rdev2->data_offset,
3000 rdev2->sectors)) {
3001 overlap = 1;
3002 break;
3003 }
3004 if (overlap) {
3005 mddev_put(mddev);
3006 break;
3007 }
3008 }
3009 rcu_read_unlock();
3010 if (overlap) {
3011 /* Someone else could have slipped in a size
3012 * change here, but doing so is just silly.
3013 * We put oldsectors back because we *know* it is
3014 * safe, and trust userspace not to race with
3015 * itself
3016 */
3017 rdev->sectors = oldsectors;
3018 return -EBUSY;
3019 }
3020 }
3021 return len;
3022 }
3023
3024 static struct rdev_sysfs_entry rdev_size =
3025 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3026
3027 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3028 {
3029 unsigned long long recovery_start = rdev->recovery_offset;
3030
3031 if (test_bit(In_sync, &rdev->flags) ||
3032 recovery_start == MaxSector)
3033 return sprintf(page, "none\n");
3034
3035 return sprintf(page, "%llu\n", recovery_start);
3036 }
3037
3038 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3039 {
3040 unsigned long long recovery_start;
3041
3042 if (cmd_match(buf, "none"))
3043 recovery_start = MaxSector;
3044 else if (kstrtoull(buf, 10, &recovery_start))
3045 return -EINVAL;
3046
3047 if (rdev->mddev->pers &&
3048 rdev->raid_disk >= 0)
3049 return -EBUSY;
3050
3051 rdev->recovery_offset = recovery_start;
3052 if (recovery_start == MaxSector)
3053 set_bit(In_sync, &rdev->flags);
3054 else
3055 clear_bit(In_sync, &rdev->flags);
3056 return len;
3057 }
3058
3059 static struct rdev_sysfs_entry rdev_recovery_start =
3060 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3061
3062 /* sysfs access to bad-blocks list.
3063 * We present two files.
3064 * 'bad-blocks' lists sector numbers and lengths of ranges that
3065 * are recorded as bad. The list is truncated to fit within
3066 * the one-page limit of sysfs.
3067 * Writing "sector length" to this file adds an acknowledged
3068 * bad block list.
3069 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3070 * been acknowledged. Writing to this file adds bad blocks
3071 * without acknowledging them. This is largely for testing.
3072 */
3073 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3074 {
3075 return badblocks_show(&rdev->badblocks, page, 0);
3076 }
3077 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3078 {
3079 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3080 /* Maybe that ack was all we needed */
3081 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3082 wake_up(&rdev->blocked_wait);
3083 return rv;
3084 }
3085 static struct rdev_sysfs_entry rdev_bad_blocks =
3086 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3087
3088 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3089 {
3090 return badblocks_show(&rdev->badblocks, page, 1);
3091 }
3092 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3093 {
3094 return badblocks_store(&rdev->badblocks, page, len, 1);
3095 }
3096 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3097 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3098
3099 static struct attribute *rdev_default_attrs[] = {
3100 &rdev_state.attr,
3101 &rdev_errors.attr,
3102 &rdev_slot.attr,
3103 &rdev_offset.attr,
3104 &rdev_new_offset.attr,
3105 &rdev_size.attr,
3106 &rdev_recovery_start.attr,
3107 &rdev_bad_blocks.attr,
3108 &rdev_unack_bad_blocks.attr,
3109 NULL,
3110 };
3111 static ssize_t
3112 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3113 {
3114 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3115 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3116
3117 if (!entry->show)
3118 return -EIO;
3119 if (!rdev->mddev)
3120 return -EBUSY;
3121 return entry->show(rdev, page);
3122 }
3123
3124 static ssize_t
3125 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3126 const char *page, size_t length)
3127 {
3128 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3129 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3130 ssize_t rv;
3131 struct mddev *mddev = rdev->mddev;
3132
3133 if (!entry->store)
3134 return -EIO;
3135 if (!capable(CAP_SYS_ADMIN))
3136 return -EACCES;
3137 rv = mddev ? mddev_lock(mddev): -EBUSY;
3138 if (!rv) {
3139 if (rdev->mddev == NULL)
3140 rv = -EBUSY;
3141 else
3142 rv = entry->store(rdev, page, length);
3143 mddev_unlock(mddev);
3144 }
3145 return rv;
3146 }
3147
3148 static void rdev_free(struct kobject *ko)
3149 {
3150 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3151 kfree(rdev);
3152 }
3153 static const struct sysfs_ops rdev_sysfs_ops = {
3154 .show = rdev_attr_show,
3155 .store = rdev_attr_store,
3156 };
3157 static struct kobj_type rdev_ktype = {
3158 .release = rdev_free,
3159 .sysfs_ops = &rdev_sysfs_ops,
3160 .default_attrs = rdev_default_attrs,
3161 };
3162
3163 int md_rdev_init(struct md_rdev *rdev)
3164 {
3165 rdev->desc_nr = -1;
3166 rdev->saved_raid_disk = -1;
3167 rdev->raid_disk = -1;
3168 rdev->flags = 0;
3169 rdev->data_offset = 0;
3170 rdev->new_data_offset = 0;
3171 rdev->sb_events = 0;
3172 rdev->last_read_error.tv_sec = 0;
3173 rdev->last_read_error.tv_nsec = 0;
3174 rdev->sb_loaded = 0;
3175 rdev->bb_page = NULL;
3176 atomic_set(&rdev->nr_pending, 0);
3177 atomic_set(&rdev->read_errors, 0);
3178 atomic_set(&rdev->corrected_errors, 0);
3179
3180 INIT_LIST_HEAD(&rdev->same_set);
3181 init_waitqueue_head(&rdev->blocked_wait);
3182
3183 /* Add space to store bad block list.
3184 * This reserves the space even on arrays where it cannot
3185 * be used - I wonder if that matters
3186 */
3187 return badblocks_init(&rdev->badblocks, 0);
3188 }
3189 EXPORT_SYMBOL_GPL(md_rdev_init);
3190 /*
3191 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3192 *
3193 * mark the device faulty if:
3194 *
3195 * - the device is nonexistent (zero size)
3196 * - the device has no valid superblock
3197 *
3198 * a faulty rdev _never_ has rdev->sb set.
3199 */
3200 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3201 {
3202 char b[BDEVNAME_SIZE];
3203 int err;
3204 struct md_rdev *rdev;
3205 sector_t size;
3206
3207 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3208 if (!rdev) {
3209 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3210 return ERR_PTR(-ENOMEM);
3211 }
3212
3213 err = md_rdev_init(rdev);
3214 if (err)
3215 goto abort_free;
3216 err = alloc_disk_sb(rdev);
3217 if (err)
3218 goto abort_free;
3219
3220 err = lock_rdev(rdev, newdev, super_format == -2);
3221 if (err)
3222 goto abort_free;
3223
3224 kobject_init(&rdev->kobj, &rdev_ktype);
3225
3226 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3227 if (!size) {
3228 printk(KERN_WARNING
3229 "md: %s has zero or unknown size, marking faulty!\n",
3230 bdevname(rdev->bdev,b));
3231 err = -EINVAL;
3232 goto abort_free;
3233 }
3234
3235 if (super_format >= 0) {
3236 err = super_types[super_format].
3237 load_super(rdev, NULL, super_minor);
3238 if (err == -EINVAL) {
3239 printk(KERN_WARNING
3240 "md: %s does not have a valid v%d.%d "
3241 "superblock, not importing!\n",
3242 bdevname(rdev->bdev,b),
3243 super_format, super_minor);
3244 goto abort_free;
3245 }
3246 if (err < 0) {
3247 printk(KERN_WARNING
3248 "md: could not read %s's sb, not importing!\n",
3249 bdevname(rdev->bdev,b));
3250 goto abort_free;
3251 }
3252 }
3253
3254 return rdev;
3255
3256 abort_free:
3257 if (rdev->bdev)
3258 unlock_rdev(rdev);
3259 md_rdev_clear(rdev);
3260 kfree(rdev);
3261 return ERR_PTR(err);
3262 }
3263
3264 /*
3265 * Check a full RAID array for plausibility
3266 */
3267
3268 static void analyze_sbs(struct mddev *mddev)
3269 {
3270 int i;
3271 struct md_rdev *rdev, *freshest, *tmp;
3272 char b[BDEVNAME_SIZE];
3273
3274 freshest = NULL;
3275 rdev_for_each_safe(rdev, tmp, mddev)
3276 switch (super_types[mddev->major_version].
3277 load_super(rdev, freshest, mddev->minor_version)) {
3278 case 1:
3279 freshest = rdev;
3280 break;
3281 case 0:
3282 break;
3283 default:
3284 printk( KERN_ERR \
3285 "md: fatal superblock inconsistency in %s"
3286 " -- removing from array\n",
3287 bdevname(rdev->bdev,b));
3288 md_kick_rdev_from_array(rdev);
3289 }
3290
3291 super_types[mddev->major_version].
3292 validate_super(mddev, freshest);
3293
3294 i = 0;
3295 rdev_for_each_safe(rdev, tmp, mddev) {
3296 if (mddev->max_disks &&
3297 (rdev->desc_nr >= mddev->max_disks ||
3298 i > mddev->max_disks)) {
3299 printk(KERN_WARNING
3300 "md: %s: %s: only %d devices permitted\n",
3301 mdname(mddev), bdevname(rdev->bdev, b),
3302 mddev->max_disks);
3303 md_kick_rdev_from_array(rdev);
3304 continue;
3305 }
3306 if (rdev != freshest) {
3307 if (super_types[mddev->major_version].
3308 validate_super(mddev, rdev)) {
3309 printk(KERN_WARNING "md: kicking non-fresh %s"
3310 " from array!\n",
3311 bdevname(rdev->bdev,b));
3312 md_kick_rdev_from_array(rdev);
3313 continue;
3314 }
3315 }
3316 if (mddev->level == LEVEL_MULTIPATH) {
3317 rdev->desc_nr = i++;
3318 rdev->raid_disk = rdev->desc_nr;
3319 set_bit(In_sync, &rdev->flags);
3320 } else if (rdev->raid_disk >=
3321 (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3322 !test_bit(Journal, &rdev->flags)) {
3323 rdev->raid_disk = -1;
3324 clear_bit(In_sync, &rdev->flags);
3325 }
3326 }
3327 }
3328
3329 /* Read a fixed-point number.
3330 * Numbers in sysfs attributes should be in "standard" units where
3331 * possible, so time should be in seconds.
3332 * However we internally use a a much smaller unit such as
3333 * milliseconds or jiffies.
3334 * This function takes a decimal number with a possible fractional
3335 * component, and produces an integer which is the result of
3336 * multiplying that number by 10^'scale'.
3337 * all without any floating-point arithmetic.
3338 */
3339 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3340 {
3341 unsigned long result = 0;
3342 long decimals = -1;
3343 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3344 if (*cp == '.')
3345 decimals = 0;
3346 else if (decimals < scale) {
3347 unsigned int value;
3348 value = *cp - '0';
3349 result = result * 10 + value;
3350 if (decimals >= 0)
3351 decimals++;
3352 }
3353 cp++;
3354 }
3355 if (*cp == '\n')
3356 cp++;
3357 if (*cp)
3358 return -EINVAL;
3359 if (decimals < 0)
3360 decimals = 0;
3361 while (decimals < scale) {
3362 result *= 10;
3363 decimals ++;
3364 }
3365 *res = result;
3366 return 0;
3367 }
3368
3369 static ssize_t
3370 safe_delay_show(struct mddev *mddev, char *page)
3371 {
3372 int msec = (mddev->safemode_delay*1000)/HZ;
3373 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3374 }
3375 static ssize_t
3376 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3377 {
3378 unsigned long msec;
3379
3380 if (mddev_is_clustered(mddev)) {
3381 pr_info("md: Safemode is disabled for clustered mode\n");
3382 return -EINVAL;
3383 }
3384
3385 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3386 return -EINVAL;
3387 if (msec == 0)
3388 mddev->safemode_delay = 0;
3389 else {
3390 unsigned long old_delay = mddev->safemode_delay;
3391 unsigned long new_delay = (msec*HZ)/1000;
3392
3393 if (new_delay == 0)
3394 new_delay = 1;
3395 mddev->safemode_delay = new_delay;
3396 if (new_delay < old_delay || old_delay == 0)
3397 mod_timer(&mddev->safemode_timer, jiffies+1);
3398 }
3399 return len;
3400 }
3401 static struct md_sysfs_entry md_safe_delay =
3402 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3403
3404 static ssize_t
3405 level_show(struct mddev *mddev, char *page)
3406 {
3407 struct md_personality *p;
3408 int ret;
3409 spin_lock(&mddev->lock);
3410 p = mddev->pers;
3411 if (p)
3412 ret = sprintf(page, "%s\n", p->name);
3413 else if (mddev->clevel[0])
3414 ret = sprintf(page, "%s\n", mddev->clevel);
3415 else if (mddev->level != LEVEL_NONE)
3416 ret = sprintf(page, "%d\n", mddev->level);
3417 else
3418 ret = 0;
3419 spin_unlock(&mddev->lock);
3420 return ret;
3421 }
3422
3423 static ssize_t
3424 level_store(struct mddev *mddev, const char *buf, size_t len)
3425 {
3426 char clevel[16];
3427 ssize_t rv;
3428 size_t slen = len;
3429 struct md_personality *pers, *oldpers;
3430 long level;
3431 void *priv, *oldpriv;
3432 struct md_rdev *rdev;
3433
3434 if (slen == 0 || slen >= sizeof(clevel))
3435 return -EINVAL;
3436
3437 rv = mddev_lock(mddev);
3438 if (rv)
3439 return rv;
3440
3441 if (mddev->pers == NULL) {
3442 strncpy(mddev->clevel, buf, slen);
3443 if (mddev->clevel[slen-1] == '\n')
3444 slen--;
3445 mddev->clevel[slen] = 0;
3446 mddev->level = LEVEL_NONE;
3447 rv = len;
3448 goto out_unlock;
3449 }
3450 rv = -EROFS;
3451 if (mddev->ro)
3452 goto out_unlock;
3453
3454 /* request to change the personality. Need to ensure:
3455 * - array is not engaged in resync/recovery/reshape
3456 * - old personality can be suspended
3457 * - new personality will access other array.
3458 */
3459
3460 rv = -EBUSY;
3461 if (mddev->sync_thread ||
3462 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3463 mddev->reshape_position != MaxSector ||
3464 mddev->sysfs_active)
3465 goto out_unlock;
3466
3467 rv = -EINVAL;
3468 if (!mddev->pers->quiesce) {
3469 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3470 mdname(mddev), mddev->pers->name);
3471 goto out_unlock;
3472 }
3473
3474 /* Now find the new personality */
3475 strncpy(clevel, buf, slen);
3476 if (clevel[slen-1] == '\n')
3477 slen--;
3478 clevel[slen] = 0;
3479 if (kstrtol(clevel, 10, &level))
3480 level = LEVEL_NONE;
3481
3482 if (request_module("md-%s", clevel) != 0)
3483 request_module("md-level-%s", clevel);
3484 spin_lock(&pers_lock);
3485 pers = find_pers(level, clevel);
3486 if (!pers || !try_module_get(pers->owner)) {
3487 spin_unlock(&pers_lock);
3488 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3489 rv = -EINVAL;
3490 goto out_unlock;
3491 }
3492 spin_unlock(&pers_lock);
3493
3494 if (pers == mddev->pers) {
3495 /* Nothing to do! */
3496 module_put(pers->owner);
3497 rv = len;
3498 goto out_unlock;
3499 }
3500 if (!pers->takeover) {
3501 module_put(pers->owner);
3502 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3503 mdname(mddev), clevel);
3504 rv = -EINVAL;
3505 goto out_unlock;
3506 }
3507
3508 rdev_for_each(rdev, mddev)
3509 rdev->new_raid_disk = rdev->raid_disk;
3510
3511 /* ->takeover must set new_* and/or delta_disks
3512 * if it succeeds, and may set them when it fails.
3513 */
3514 priv = pers->takeover(mddev);
3515 if (IS_ERR(priv)) {
3516 mddev->new_level = mddev->level;
3517 mddev->new_layout = mddev->layout;
3518 mddev->new_chunk_sectors = mddev->chunk_sectors;
3519 mddev->raid_disks -= mddev->delta_disks;
3520 mddev->delta_disks = 0;
3521 mddev->reshape_backwards = 0;
3522 module_put(pers->owner);
3523 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3524 mdname(mddev), clevel);
3525 rv = PTR_ERR(priv);
3526 goto out_unlock;
3527 }
3528
3529 /* Looks like we have a winner */
3530 mddev_suspend(mddev);
3531 mddev_detach(mddev);
3532
3533 spin_lock(&mddev->lock);
3534 oldpers = mddev->pers;
3535 oldpriv = mddev->private;
3536 mddev->pers = pers;
3537 mddev->private = priv;
3538 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3539 mddev->level = mddev->new_level;
3540 mddev->layout = mddev->new_layout;
3541 mddev->chunk_sectors = mddev->new_chunk_sectors;
3542 mddev->delta_disks = 0;
3543 mddev->reshape_backwards = 0;
3544 mddev->degraded = 0;
3545 spin_unlock(&mddev->lock);
3546
3547 if (oldpers->sync_request == NULL &&
3548 mddev->external) {
3549 /* We are converting from a no-redundancy array
3550 * to a redundancy array and metadata is managed
3551 * externally so we need to be sure that writes
3552 * won't block due to a need to transition
3553 * clean->dirty
3554 * until external management is started.
3555 */
3556 mddev->in_sync = 0;
3557 mddev->safemode_delay = 0;
3558 mddev->safemode = 0;
3559 }
3560
3561 oldpers->free(mddev, oldpriv);
3562
3563 if (oldpers->sync_request == NULL &&
3564 pers->sync_request != NULL) {
3565 /* need to add the md_redundancy_group */
3566 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3567 printk(KERN_WARNING
3568 "md: cannot register extra attributes for %s\n",
3569 mdname(mddev));
3570 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3571 }
3572 if (oldpers->sync_request != NULL &&
3573 pers->sync_request == NULL) {
3574 /* need to remove the md_redundancy_group */
3575 if (mddev->to_remove == NULL)
3576 mddev->to_remove = &md_redundancy_group;
3577 }
3578
3579 rdev_for_each(rdev, mddev) {
3580 if (rdev->raid_disk < 0)
3581 continue;
3582 if (rdev->new_raid_disk >= mddev->raid_disks)
3583 rdev->new_raid_disk = -1;
3584 if (rdev->new_raid_disk == rdev->raid_disk)
3585 continue;
3586 sysfs_unlink_rdev(mddev, rdev);
3587 }
3588 rdev_for_each(rdev, mddev) {
3589 if (rdev->raid_disk < 0)
3590 continue;
3591 if (rdev->new_raid_disk == rdev->raid_disk)
3592 continue;
3593 rdev->raid_disk = rdev->new_raid_disk;
3594 if (rdev->raid_disk < 0)
3595 clear_bit(In_sync, &rdev->flags);
3596 else {
3597 if (sysfs_link_rdev(mddev, rdev))
3598 printk(KERN_WARNING "md: cannot register rd%d"
3599 " for %s after level change\n",
3600 rdev->raid_disk, mdname(mddev));
3601 }
3602 }
3603
3604 if (pers->sync_request == NULL) {
3605 /* this is now an array without redundancy, so
3606 * it must always be in_sync
3607 */
3608 mddev->in_sync = 1;
3609 del_timer_sync(&mddev->safemode_timer);
3610 }
3611 blk_set_stacking_limits(&mddev->queue->limits);
3612 pers->run(mddev);
3613 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3614 mddev_resume(mddev);
3615 if (!mddev->thread)
3616 md_update_sb(mddev, 1);
3617 sysfs_notify(&mddev->kobj, NULL, "level");
3618 md_new_event(mddev);
3619 rv = len;
3620 out_unlock:
3621 mddev_unlock(mddev);
3622 return rv;
3623 }
3624
3625 static struct md_sysfs_entry md_level =
3626 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3627
3628 static ssize_t
3629 layout_show(struct mddev *mddev, char *page)
3630 {
3631 /* just a number, not meaningful for all levels */
3632 if (mddev->reshape_position != MaxSector &&
3633 mddev->layout != mddev->new_layout)
3634 return sprintf(page, "%d (%d)\n",
3635 mddev->new_layout, mddev->layout);
3636 return sprintf(page, "%d\n", mddev->layout);
3637 }
3638
3639 static ssize_t
3640 layout_store(struct mddev *mddev, const char *buf, size_t len)
3641 {
3642 unsigned int n;
3643 int err;
3644
3645 err = kstrtouint(buf, 10, &n);
3646 if (err < 0)
3647 return err;
3648 err = mddev_lock(mddev);
3649 if (err)
3650 return err;
3651
3652 if (mddev->pers) {
3653 if (mddev->pers->check_reshape == NULL)
3654 err = -EBUSY;
3655 else if (mddev->ro)
3656 err = -EROFS;
3657 else {
3658 mddev->new_layout = n;
3659 err = mddev->pers->check_reshape(mddev);
3660 if (err)
3661 mddev->new_layout = mddev->layout;
3662 }
3663 } else {
3664 mddev->new_layout = n;
3665 if (mddev->reshape_position == MaxSector)
3666 mddev->layout = n;
3667 }
3668 mddev_unlock(mddev);
3669 return err ?: len;
3670 }
3671 static struct md_sysfs_entry md_layout =
3672 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3673
3674 static ssize_t
3675 raid_disks_show(struct mddev *mddev, char *page)
3676 {
3677 if (mddev->raid_disks == 0)
3678 return 0;
3679 if (mddev->reshape_position != MaxSector &&
3680 mddev->delta_disks != 0)
3681 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3682 mddev->raid_disks - mddev->delta_disks);
3683 return sprintf(page, "%d\n", mddev->raid_disks);
3684 }
3685
3686 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3687
3688 static ssize_t
3689 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3690 {
3691 unsigned int n;
3692 int err;
3693
3694 err = kstrtouint(buf, 10, &n);
3695 if (err < 0)
3696 return err;
3697
3698 err = mddev_lock(mddev);
3699 if (err)
3700 return err;
3701 if (mddev->pers)
3702 err = update_raid_disks(mddev, n);
3703 else if (mddev->reshape_position != MaxSector) {
3704 struct md_rdev *rdev;
3705 int olddisks = mddev->raid_disks - mddev->delta_disks;
3706
3707 err = -EINVAL;
3708 rdev_for_each(rdev, mddev) {
3709 if (olddisks < n &&
3710 rdev->data_offset < rdev->new_data_offset)
3711 goto out_unlock;
3712 if (olddisks > n &&
3713 rdev->data_offset > rdev->new_data_offset)
3714 goto out_unlock;
3715 }
3716 err = 0;
3717 mddev->delta_disks = n - olddisks;
3718 mddev->raid_disks = n;
3719 mddev->reshape_backwards = (mddev->delta_disks < 0);
3720 } else
3721 mddev->raid_disks = n;
3722 out_unlock:
3723 mddev_unlock(mddev);
3724 return err ? err : len;
3725 }
3726 static struct md_sysfs_entry md_raid_disks =
3727 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3728
3729 static ssize_t
3730 chunk_size_show(struct mddev *mddev, char *page)
3731 {
3732 if (mddev->reshape_position != MaxSector &&
3733 mddev->chunk_sectors != mddev->new_chunk_sectors)
3734 return sprintf(page, "%d (%d)\n",
3735 mddev->new_chunk_sectors << 9,
3736 mddev->chunk_sectors << 9);
3737 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3738 }
3739
3740 static ssize_t
3741 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3742 {
3743 unsigned long n;
3744 int err;
3745
3746 err = kstrtoul(buf, 10, &n);
3747 if (err < 0)
3748 return err;
3749
3750 err = mddev_lock(mddev);
3751 if (err)
3752 return err;
3753 if (mddev->pers) {
3754 if (mddev->pers->check_reshape == NULL)
3755 err = -EBUSY;
3756 else if (mddev->ro)
3757 err = -EROFS;
3758 else {
3759 mddev->new_chunk_sectors = n >> 9;
3760 err = mddev->pers->check_reshape(mddev);
3761 if (err)
3762 mddev->new_chunk_sectors = mddev->chunk_sectors;
3763 }
3764 } else {
3765 mddev->new_chunk_sectors = n >> 9;
3766 if (mddev->reshape_position == MaxSector)
3767 mddev->chunk_sectors = n >> 9;
3768 }
3769 mddev_unlock(mddev);
3770 return err ?: len;
3771 }
3772 static struct md_sysfs_entry md_chunk_size =
3773 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3774
3775 static ssize_t
3776 resync_start_show(struct mddev *mddev, char *page)
3777 {
3778 if (mddev->recovery_cp == MaxSector)
3779 return sprintf(page, "none\n");
3780 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3781 }
3782
3783 static ssize_t
3784 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3785 {
3786 unsigned long long n;
3787 int err;
3788
3789 if (cmd_match(buf, "none"))
3790 n = MaxSector;
3791 else {
3792 err = kstrtoull(buf, 10, &n);
3793 if (err < 0)
3794 return err;
3795 if (n != (sector_t)n)
3796 return -EINVAL;
3797 }
3798
3799 err = mddev_lock(mddev);
3800 if (err)
3801 return err;
3802 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3803 err = -EBUSY;
3804
3805 if (!err) {
3806 mddev->recovery_cp = n;
3807 if (mddev->pers)
3808 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3809 }
3810 mddev_unlock(mddev);
3811 return err ?: len;
3812 }
3813 static struct md_sysfs_entry md_resync_start =
3814 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
3815 resync_start_show, resync_start_store);
3816
3817 /*
3818 * The array state can be:
3819 *
3820 * clear
3821 * No devices, no size, no level
3822 * Equivalent to STOP_ARRAY ioctl
3823 * inactive
3824 * May have some settings, but array is not active
3825 * all IO results in error
3826 * When written, doesn't tear down array, but just stops it
3827 * suspended (not supported yet)
3828 * All IO requests will block. The array can be reconfigured.
3829 * Writing this, if accepted, will block until array is quiescent
3830 * readonly
3831 * no resync can happen. no superblocks get written.
3832 * write requests fail
3833 * read-auto
3834 * like readonly, but behaves like 'clean' on a write request.
3835 *
3836 * clean - no pending writes, but otherwise active.
3837 * When written to inactive array, starts without resync
3838 * If a write request arrives then
3839 * if metadata is known, mark 'dirty' and switch to 'active'.
3840 * if not known, block and switch to write-pending
3841 * If written to an active array that has pending writes, then fails.
3842 * active
3843 * fully active: IO and resync can be happening.
3844 * When written to inactive array, starts with resync
3845 *
3846 * write-pending
3847 * clean, but writes are blocked waiting for 'active' to be written.
3848 *
3849 * active-idle
3850 * like active, but no writes have been seen for a while (100msec).
3851 *
3852 */
3853 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3854 write_pending, active_idle, bad_word};
3855 static char *array_states[] = {
3856 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3857 "write-pending", "active-idle", NULL };
3858
3859 static int match_word(const char *word, char **list)
3860 {
3861 int n;
3862 for (n=0; list[n]; n++)
3863 if (cmd_match(word, list[n]))
3864 break;
3865 return n;
3866 }
3867
3868 static ssize_t
3869 array_state_show(struct mddev *mddev, char *page)
3870 {
3871 enum array_state st = inactive;
3872
3873 if (mddev->pers)
3874 switch(mddev->ro) {
3875 case 1:
3876 st = readonly;
3877 break;
3878 case 2:
3879 st = read_auto;
3880 break;
3881 case 0:
3882 if (mddev->in_sync)
3883 st = clean;
3884 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3885 st = write_pending;
3886 else if (mddev->safemode)
3887 st = active_idle;
3888 else
3889 st = active;
3890 }
3891 else {
3892 if (list_empty(&mddev->disks) &&
3893 mddev->raid_disks == 0 &&
3894 mddev->dev_sectors == 0)
3895 st = clear;
3896 else
3897 st = inactive;
3898 }
3899 return sprintf(page, "%s\n", array_states[st]);
3900 }
3901
3902 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
3903 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
3904 static int do_md_run(struct mddev *mddev);
3905 static int restart_array(struct mddev *mddev);
3906
3907 static ssize_t
3908 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3909 {
3910 int err;
3911 enum array_state st = match_word(buf, array_states);
3912
3913 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
3914 /* don't take reconfig_mutex when toggling between
3915 * clean and active
3916 */
3917 spin_lock(&mddev->lock);
3918 if (st == active) {
3919 restart_array(mddev);
3920 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3921 wake_up(&mddev->sb_wait);
3922 err = 0;
3923 } else /* st == clean */ {
3924 restart_array(mddev);
3925 if (atomic_read(&mddev->writes_pending) == 0) {
3926 if (mddev->in_sync == 0) {
3927 mddev->in_sync = 1;
3928 if (mddev->safemode == 1)
3929 mddev->safemode = 0;
3930 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3931 }
3932 err = 0;
3933 } else
3934 err = -EBUSY;
3935 }
3936 spin_unlock(&mddev->lock);
3937 return err ?: len;
3938 }
3939 err = mddev_lock(mddev);
3940 if (err)
3941 return err;
3942 err = -EINVAL;
3943 switch(st) {
3944 case bad_word:
3945 break;
3946 case clear:
3947 /* stopping an active array */
3948 err = do_md_stop(mddev, 0, NULL);
3949 break;
3950 case inactive:
3951 /* stopping an active array */
3952 if (mddev->pers)
3953 err = do_md_stop(mddev, 2, NULL);
3954 else
3955 err = 0; /* already inactive */
3956 break;
3957 case suspended:
3958 break; /* not supported yet */
3959 case readonly:
3960 if (mddev->pers)
3961 err = md_set_readonly(mddev, NULL);
3962 else {
3963 mddev->ro = 1;
3964 set_disk_ro(mddev->gendisk, 1);
3965 err = do_md_run(mddev);
3966 }
3967 break;
3968 case read_auto:
3969 if (mddev->pers) {
3970 if (mddev->ro == 0)
3971 err = md_set_readonly(mddev, NULL);
3972 else if (mddev->ro == 1)
3973 err = restart_array(mddev);
3974 if (err == 0) {
3975 mddev->ro = 2;
3976 set_disk_ro(mddev->gendisk, 0);
3977 }
3978 } else {
3979 mddev->ro = 2;
3980 err = do_md_run(mddev);
3981 }
3982 break;
3983 case clean:
3984 if (mddev->pers) {
3985 err = restart_array(mddev);
3986 if (err)
3987 break;
3988 spin_lock(&mddev->lock);
3989 if (atomic_read(&mddev->writes_pending) == 0) {
3990 if (mddev->in_sync == 0) {
3991 mddev->in_sync = 1;
3992 if (mddev->safemode == 1)
3993 mddev->safemode = 0;
3994 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3995 }
3996 err = 0;
3997 } else
3998 err = -EBUSY;
3999 spin_unlock(&mddev->lock);
4000 } else
4001 err = -EINVAL;
4002 break;
4003 case active:
4004 if (mddev->pers) {
4005 err = restart_array(mddev);
4006 if (err)
4007 break;
4008 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
4009 wake_up(&mddev->sb_wait);
4010 err = 0;
4011 } else {
4012 mddev->ro = 0;
4013 set_disk_ro(mddev->gendisk, 0);
4014 err = do_md_run(mddev);
4015 }
4016 break;
4017 case write_pending:
4018 case active_idle:
4019 /* these cannot be set */
4020 break;
4021 }
4022
4023 if (!err) {
4024 if (mddev->hold_active == UNTIL_IOCTL)
4025 mddev->hold_active = 0;
4026 sysfs_notify_dirent_safe(mddev->sysfs_state);
4027 }
4028 mddev_unlock(mddev);
4029 return err ?: len;
4030 }
4031 static struct md_sysfs_entry md_array_state =
4032 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4033
4034 static ssize_t
4035 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4036 return sprintf(page, "%d\n",
4037 atomic_read(&mddev->max_corr_read_errors));
4038 }
4039
4040 static ssize_t
4041 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4042 {
4043 unsigned int n;
4044 int rv;
4045
4046 rv = kstrtouint(buf, 10, &n);
4047 if (rv < 0)
4048 return rv;
4049 atomic_set(&mddev->max_corr_read_errors, n);
4050 return len;
4051 }
4052
4053 static struct md_sysfs_entry max_corr_read_errors =
4054 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4055 max_corrected_read_errors_store);
4056
4057 static ssize_t
4058 null_show(struct mddev *mddev, char *page)
4059 {
4060 return -EINVAL;
4061 }
4062
4063 static ssize_t
4064 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4065 {
4066 /* buf must be %d:%d\n? giving major and minor numbers */
4067 /* The new device is added to the array.
4068 * If the array has a persistent superblock, we read the
4069 * superblock to initialise info and check validity.
4070 * Otherwise, only checking done is that in bind_rdev_to_array,
4071 * which mainly checks size.
4072 */
4073 char *e;
4074 int major = simple_strtoul(buf, &e, 10);
4075 int minor;
4076 dev_t dev;
4077 struct md_rdev *rdev;
4078 int err;
4079
4080 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4081 return -EINVAL;
4082 minor = simple_strtoul(e+1, &e, 10);
4083 if (*e && *e != '\n')
4084 return -EINVAL;
4085 dev = MKDEV(major, minor);
4086 if (major != MAJOR(dev) ||
4087 minor != MINOR(dev))
4088 return -EOVERFLOW;
4089
4090 flush_workqueue(md_misc_wq);
4091
4092 err = mddev_lock(mddev);
4093 if (err)
4094 return err;
4095 if (mddev->persistent) {
4096 rdev = md_import_device(dev, mddev->major_version,
4097 mddev->minor_version);
4098 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4099 struct md_rdev *rdev0
4100 = list_entry(mddev->disks.next,
4101 struct md_rdev, same_set);
4102 err = super_types[mddev->major_version]
4103 .load_super(rdev, rdev0, mddev->minor_version);
4104 if (err < 0)
4105 goto out;
4106 }
4107 } else if (mddev->external)
4108 rdev = md_import_device(dev, -2, -1);
4109 else
4110 rdev = md_import_device(dev, -1, -1);
4111
4112 if (IS_ERR(rdev)) {
4113 mddev_unlock(mddev);
4114 return PTR_ERR(rdev);
4115 }
4116 err = bind_rdev_to_array(rdev, mddev);
4117 out:
4118 if (err)
4119 export_rdev(rdev);
4120 mddev_unlock(mddev);
4121 return err ? err : len;
4122 }
4123
4124 static struct md_sysfs_entry md_new_device =
4125 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4126
4127 static ssize_t
4128 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4129 {
4130 char *end;
4131 unsigned long chunk, end_chunk;
4132 int err;
4133
4134 err = mddev_lock(mddev);
4135 if (err)
4136 return err;
4137 if (!mddev->bitmap)
4138 goto out;
4139 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4140 while (*buf) {
4141 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4142 if (buf == end) break;
4143 if (*end == '-') { /* range */
4144 buf = end + 1;
4145 end_chunk = simple_strtoul(buf, &end, 0);
4146 if (buf == end) break;
4147 }
4148 if (*end && !isspace(*end)) break;
4149 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4150 buf = skip_spaces(end);
4151 }
4152 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4153 out:
4154 mddev_unlock(mddev);
4155 return len;
4156 }
4157
4158 static struct md_sysfs_entry md_bitmap =
4159 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4160
4161 static ssize_t
4162 size_show(struct mddev *mddev, char *page)
4163 {
4164 return sprintf(page, "%llu\n",
4165 (unsigned long long)mddev->dev_sectors / 2);
4166 }
4167
4168 static int update_size(struct mddev *mddev, sector_t num_sectors);
4169
4170 static ssize_t
4171 size_store(struct mddev *mddev, const char *buf, size_t len)
4172 {
4173 /* If array is inactive, we can reduce the component size, but
4174 * not increase it (except from 0).
4175 * If array is active, we can try an on-line resize
4176 */
4177 sector_t sectors;
4178 int err = strict_blocks_to_sectors(buf, &sectors);
4179
4180 if (err < 0)
4181 return err;
4182 err = mddev_lock(mddev);
4183 if (err)
4184 return err;
4185 if (mddev->pers) {
4186 err = update_size(mddev, sectors);
4187 md_update_sb(mddev, 1);
4188 } else {
4189 if (mddev->dev_sectors == 0 ||
4190 mddev->dev_sectors > sectors)
4191 mddev->dev_sectors = sectors;
4192 else
4193 err = -ENOSPC;
4194 }
4195 mddev_unlock(mddev);
4196 return err ? err : len;
4197 }
4198
4199 static struct md_sysfs_entry md_size =
4200 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4201
4202 /* Metadata version.
4203 * This is one of
4204 * 'none' for arrays with no metadata (good luck...)
4205 * 'external' for arrays with externally managed metadata,
4206 * or N.M for internally known formats
4207 */
4208 static ssize_t
4209 metadata_show(struct mddev *mddev, char *page)
4210 {
4211 if (mddev->persistent)
4212 return sprintf(page, "%d.%d\n",
4213 mddev->major_version, mddev->minor_version);
4214 else if (mddev->external)
4215 return sprintf(page, "external:%s\n", mddev->metadata_type);
4216 else
4217 return sprintf(page, "none\n");
4218 }
4219
4220 static ssize_t
4221 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4222 {
4223 int major, minor;
4224 char *e;
4225 int err;
4226 /* Changing the details of 'external' metadata is
4227 * always permitted. Otherwise there must be
4228 * no devices attached to the array.
4229 */
4230
4231 err = mddev_lock(mddev);
4232 if (err)
4233 return err;
4234 err = -EBUSY;
4235 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4236 ;
4237 else if (!list_empty(&mddev->disks))
4238 goto out_unlock;
4239
4240 err = 0;
4241 if (cmd_match(buf, "none")) {
4242 mddev->persistent = 0;
4243 mddev->external = 0;
4244 mddev->major_version = 0;
4245 mddev->minor_version = 90;
4246 goto out_unlock;
4247 }
4248 if (strncmp(buf, "external:", 9) == 0) {
4249 size_t namelen = len-9;
4250 if (namelen >= sizeof(mddev->metadata_type))
4251 namelen = sizeof(mddev->metadata_type)-1;
4252 strncpy(mddev->metadata_type, buf+9, namelen);
4253 mddev->metadata_type[namelen] = 0;
4254 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4255 mddev->metadata_type[--namelen] = 0;
4256 mddev->persistent = 0;
4257 mddev->external = 1;
4258 mddev->major_version = 0;
4259 mddev->minor_version = 90;
4260 goto out_unlock;
4261 }
4262 major = simple_strtoul(buf, &e, 10);
4263 err = -EINVAL;
4264 if (e==buf || *e != '.')
4265 goto out_unlock;
4266 buf = e+1;
4267 minor = simple_strtoul(buf, &e, 10);
4268 if (e==buf || (*e && *e != '\n') )
4269 goto out_unlock;
4270 err = -ENOENT;
4271 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4272 goto out_unlock;
4273 mddev->major_version = major;
4274 mddev->minor_version = minor;
4275 mddev->persistent = 1;
4276 mddev->external = 0;
4277 err = 0;
4278 out_unlock:
4279 mddev_unlock(mddev);
4280 return err ?: len;
4281 }
4282
4283 static struct md_sysfs_entry md_metadata =
4284 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4285
4286 static ssize_t
4287 action_show(struct mddev *mddev, char *page)
4288 {
4289 char *type = "idle";
4290 unsigned long recovery = mddev->recovery;
4291 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4292 type = "frozen";
4293 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4294 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4295 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4296 type = "reshape";
4297 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4298 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4299 type = "resync";
4300 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4301 type = "check";
4302 else
4303 type = "repair";
4304 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4305 type = "recover";
4306 else if (mddev->reshape_position != MaxSector)
4307 type = "reshape";
4308 }
4309 return sprintf(page, "%s\n", type);
4310 }
4311
4312 static ssize_t
4313 action_store(struct mddev *mddev, const char *page, size_t len)
4314 {
4315 if (!mddev->pers || !mddev->pers->sync_request)
4316 return -EINVAL;
4317
4318
4319 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4320 if (cmd_match(page, "frozen"))
4321 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4322 else
4323 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4324 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4325 mddev_lock(mddev) == 0) {
4326 flush_workqueue(md_misc_wq);
4327 if (mddev->sync_thread) {
4328 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4329 md_reap_sync_thread(mddev);
4330 }
4331 mddev_unlock(mddev);
4332 }
4333 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4334 return -EBUSY;
4335 else if (cmd_match(page, "resync"))
4336 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4337 else if (cmd_match(page, "recover")) {
4338 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4339 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4340 } else if (cmd_match(page, "reshape")) {
4341 int err;
4342 if (mddev->pers->start_reshape == NULL)
4343 return -EINVAL;
4344 err = mddev_lock(mddev);
4345 if (!err) {
4346 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4347 err = -EBUSY;
4348 else {
4349 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4350 err = mddev->pers->start_reshape(mddev);
4351 }
4352 mddev_unlock(mddev);
4353 }
4354 if (err)
4355 return err;
4356 sysfs_notify(&mddev->kobj, NULL, "degraded");
4357 } else {
4358 if (cmd_match(page, "check"))
4359 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4360 else if (!cmd_match(page, "repair"))
4361 return -EINVAL;
4362 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4363 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4364 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4365 }
4366 if (mddev->ro == 2) {
4367 /* A write to sync_action is enough to justify
4368 * canceling read-auto mode
4369 */
4370 mddev->ro = 0;
4371 md_wakeup_thread(mddev->sync_thread);
4372 }
4373 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4374 md_wakeup_thread(mddev->thread);
4375 sysfs_notify_dirent_safe(mddev->sysfs_action);
4376 return len;
4377 }
4378
4379 static struct md_sysfs_entry md_scan_mode =
4380 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4381
4382 static ssize_t
4383 last_sync_action_show(struct mddev *mddev, char *page)
4384 {
4385 return sprintf(page, "%s\n", mddev->last_sync_action);
4386 }
4387
4388 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4389
4390 static ssize_t
4391 mismatch_cnt_show(struct mddev *mddev, char *page)
4392 {
4393 return sprintf(page, "%llu\n",
4394 (unsigned long long)
4395 atomic64_read(&mddev->resync_mismatches));
4396 }
4397
4398 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4399
4400 static ssize_t
4401 sync_min_show(struct mddev *mddev, char *page)
4402 {
4403 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4404 mddev->sync_speed_min ? "local": "system");
4405 }
4406
4407 static ssize_t
4408 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4409 {
4410 unsigned int min;
4411 int rv;
4412
4413 if (strncmp(buf, "system", 6)==0) {
4414 min = 0;
4415 } else {
4416 rv = kstrtouint(buf, 10, &min);
4417 if (rv < 0)
4418 return rv;
4419 if (min == 0)
4420 return -EINVAL;
4421 }
4422 mddev->sync_speed_min = min;
4423 return len;
4424 }
4425
4426 static struct md_sysfs_entry md_sync_min =
4427 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4428
4429 static ssize_t
4430 sync_max_show(struct mddev *mddev, char *page)
4431 {
4432 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4433 mddev->sync_speed_max ? "local": "system");
4434 }
4435
4436 static ssize_t
4437 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4438 {
4439 unsigned int max;
4440 int rv;
4441
4442 if (strncmp(buf, "system", 6)==0) {
4443 max = 0;
4444 } else {
4445 rv = kstrtouint(buf, 10, &max);
4446 if (rv < 0)
4447 return rv;
4448 if (max == 0)
4449 return -EINVAL;
4450 }
4451 mddev->sync_speed_max = max;
4452 return len;
4453 }
4454
4455 static struct md_sysfs_entry md_sync_max =
4456 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4457
4458 static ssize_t
4459 degraded_show(struct mddev *mddev, char *page)
4460 {
4461 return sprintf(page, "%d\n", mddev->degraded);
4462 }
4463 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4464
4465 static ssize_t
4466 sync_force_parallel_show(struct mddev *mddev, char *page)
4467 {
4468 return sprintf(page, "%d\n", mddev->parallel_resync);
4469 }
4470
4471 static ssize_t
4472 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4473 {
4474 long n;
4475
4476 if (kstrtol(buf, 10, &n))
4477 return -EINVAL;
4478
4479 if (n != 0 && n != 1)
4480 return -EINVAL;
4481
4482 mddev->parallel_resync = n;
4483
4484 if (mddev->sync_thread)
4485 wake_up(&resync_wait);
4486
4487 return len;
4488 }
4489
4490 /* force parallel resync, even with shared block devices */
4491 static struct md_sysfs_entry md_sync_force_parallel =
4492 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4493 sync_force_parallel_show, sync_force_parallel_store);
4494
4495 static ssize_t
4496 sync_speed_show(struct mddev *mddev, char *page)
4497 {
4498 unsigned long resync, dt, db;
4499 if (mddev->curr_resync == 0)
4500 return sprintf(page, "none\n");
4501 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4502 dt = (jiffies - mddev->resync_mark) / HZ;
4503 if (!dt) dt++;
4504 db = resync - mddev->resync_mark_cnt;
4505 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4506 }
4507
4508 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4509
4510 static ssize_t
4511 sync_completed_show(struct mddev *mddev, char *page)
4512 {
4513 unsigned long long max_sectors, resync;
4514
4515 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4516 return sprintf(page, "none\n");
4517
4518 if (mddev->curr_resync == 1 ||
4519 mddev->curr_resync == 2)
4520 return sprintf(page, "delayed\n");
4521
4522 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4523 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4524 max_sectors = mddev->resync_max_sectors;
4525 else
4526 max_sectors = mddev->dev_sectors;
4527
4528 resync = mddev->curr_resync_completed;
4529 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4530 }
4531
4532 static struct md_sysfs_entry md_sync_completed =
4533 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
4534
4535 static ssize_t
4536 min_sync_show(struct mddev *mddev, char *page)
4537 {
4538 return sprintf(page, "%llu\n",
4539 (unsigned long long)mddev->resync_min);
4540 }
4541 static ssize_t
4542 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4543 {
4544 unsigned long long min;
4545 int err;
4546
4547 if (kstrtoull(buf, 10, &min))
4548 return -EINVAL;
4549
4550 spin_lock(&mddev->lock);
4551 err = -EINVAL;
4552 if (min > mddev->resync_max)
4553 goto out_unlock;
4554
4555 err = -EBUSY;
4556 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4557 goto out_unlock;
4558
4559 /* Round down to multiple of 4K for safety */
4560 mddev->resync_min = round_down(min, 8);
4561 err = 0;
4562
4563 out_unlock:
4564 spin_unlock(&mddev->lock);
4565 return err ?: len;
4566 }
4567
4568 static struct md_sysfs_entry md_min_sync =
4569 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4570
4571 static ssize_t
4572 max_sync_show(struct mddev *mddev, char *page)
4573 {
4574 if (mddev->resync_max == MaxSector)
4575 return sprintf(page, "max\n");
4576 else
4577 return sprintf(page, "%llu\n",
4578 (unsigned long long)mddev->resync_max);
4579 }
4580 static ssize_t
4581 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4582 {
4583 int err;
4584 spin_lock(&mddev->lock);
4585 if (strncmp(buf, "max", 3) == 0)
4586 mddev->resync_max = MaxSector;
4587 else {
4588 unsigned long long max;
4589 int chunk;
4590
4591 err = -EINVAL;
4592 if (kstrtoull(buf, 10, &max))
4593 goto out_unlock;
4594 if (max < mddev->resync_min)
4595 goto out_unlock;
4596
4597 err = -EBUSY;
4598 if (max < mddev->resync_max &&
4599 mddev->ro == 0 &&
4600 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4601 goto out_unlock;
4602
4603 /* Must be a multiple of chunk_size */
4604 chunk = mddev->chunk_sectors;
4605 if (chunk) {
4606 sector_t temp = max;
4607
4608 err = -EINVAL;
4609 if (sector_div(temp, chunk))
4610 goto out_unlock;
4611 }
4612 mddev->resync_max = max;
4613 }
4614 wake_up(&mddev->recovery_wait);
4615 err = 0;
4616 out_unlock:
4617 spin_unlock(&mddev->lock);
4618 return err ?: len;
4619 }
4620
4621 static struct md_sysfs_entry md_max_sync =
4622 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4623
4624 static ssize_t
4625 suspend_lo_show(struct mddev *mddev, char *page)
4626 {
4627 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4628 }
4629
4630 static ssize_t
4631 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4632 {
4633 unsigned long long old, new;
4634 int err;
4635
4636 err = kstrtoull(buf, 10, &new);
4637 if (err < 0)
4638 return err;
4639 if (new != (sector_t)new)
4640 return -EINVAL;
4641
4642 err = mddev_lock(mddev);
4643 if (err)
4644 return err;
4645 err = -EINVAL;
4646 if (mddev->pers == NULL ||
4647 mddev->pers->quiesce == NULL)
4648 goto unlock;
4649 old = mddev->suspend_lo;
4650 mddev->suspend_lo = new;
4651 if (new >= old)
4652 /* Shrinking suspended region */
4653 mddev->pers->quiesce(mddev, 2);
4654 else {
4655 /* Expanding suspended region - need to wait */
4656 mddev->pers->quiesce(mddev, 1);
4657 mddev->pers->quiesce(mddev, 0);
4658 }
4659 err = 0;
4660 unlock:
4661 mddev_unlock(mddev);
4662 return err ?: len;
4663 }
4664 static struct md_sysfs_entry md_suspend_lo =
4665 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4666
4667 static ssize_t
4668 suspend_hi_show(struct mddev *mddev, char *page)
4669 {
4670 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4671 }
4672
4673 static ssize_t
4674 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4675 {
4676 unsigned long long old, new;
4677 int err;
4678
4679 err = kstrtoull(buf, 10, &new);
4680 if (err < 0)
4681 return err;
4682 if (new != (sector_t)new)
4683 return -EINVAL;
4684
4685 err = mddev_lock(mddev);
4686 if (err)
4687 return err;
4688 err = -EINVAL;
4689 if (mddev->pers == NULL ||
4690 mddev->pers->quiesce == NULL)
4691 goto unlock;
4692 old = mddev->suspend_hi;
4693 mddev->suspend_hi = new;
4694 if (new <= old)
4695 /* Shrinking suspended region */
4696 mddev->pers->quiesce(mddev, 2);
4697 else {
4698 /* Expanding suspended region - need to wait */
4699 mddev->pers->quiesce(mddev, 1);
4700 mddev->pers->quiesce(mddev, 0);
4701 }
4702 err = 0;
4703 unlock:
4704 mddev_unlock(mddev);
4705 return err ?: len;
4706 }
4707 static struct md_sysfs_entry md_suspend_hi =
4708 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4709
4710 static ssize_t
4711 reshape_position_show(struct mddev *mddev, char *page)
4712 {
4713 if (mddev->reshape_position != MaxSector)
4714 return sprintf(page, "%llu\n",
4715 (unsigned long long)mddev->reshape_position);
4716 strcpy(page, "none\n");
4717 return 5;
4718 }
4719
4720 static ssize_t
4721 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4722 {
4723 struct md_rdev *rdev;
4724 unsigned long long new;
4725 int err;
4726
4727 err = kstrtoull(buf, 10, &new);
4728 if (err < 0)
4729 return err;
4730 if (new != (sector_t)new)
4731 return -EINVAL;
4732 err = mddev_lock(mddev);
4733 if (err)
4734 return err;
4735 err = -EBUSY;
4736 if (mddev->pers)
4737 goto unlock;
4738 mddev->reshape_position = new;
4739 mddev->delta_disks = 0;
4740 mddev->reshape_backwards = 0;
4741 mddev->new_level = mddev->level;
4742 mddev->new_layout = mddev->layout;
4743 mddev->new_chunk_sectors = mddev->chunk_sectors;
4744 rdev_for_each(rdev, mddev)
4745 rdev->new_data_offset = rdev->data_offset;
4746 err = 0;
4747 unlock:
4748 mddev_unlock(mddev);
4749 return err ?: len;
4750 }
4751
4752 static struct md_sysfs_entry md_reshape_position =
4753 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4754 reshape_position_store);
4755
4756 static ssize_t
4757 reshape_direction_show(struct mddev *mddev, char *page)
4758 {
4759 return sprintf(page, "%s\n",
4760 mddev->reshape_backwards ? "backwards" : "forwards");
4761 }
4762
4763 static ssize_t
4764 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4765 {
4766 int backwards = 0;
4767 int err;
4768
4769 if (cmd_match(buf, "forwards"))
4770 backwards = 0;
4771 else if (cmd_match(buf, "backwards"))
4772 backwards = 1;
4773 else
4774 return -EINVAL;
4775 if (mddev->reshape_backwards == backwards)
4776 return len;
4777
4778 err = mddev_lock(mddev);
4779 if (err)
4780 return err;
4781 /* check if we are allowed to change */
4782 if (mddev->delta_disks)
4783 err = -EBUSY;
4784 else if (mddev->persistent &&
4785 mddev->major_version == 0)
4786 err = -EINVAL;
4787 else
4788 mddev->reshape_backwards = backwards;
4789 mddev_unlock(mddev);
4790 return err ?: len;
4791 }
4792
4793 static struct md_sysfs_entry md_reshape_direction =
4794 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4795 reshape_direction_store);
4796
4797 static ssize_t
4798 array_size_show(struct mddev *mddev, char *page)
4799 {
4800 if (mddev->external_size)
4801 return sprintf(page, "%llu\n",
4802 (unsigned long long)mddev->array_sectors/2);
4803 else
4804 return sprintf(page, "default\n");
4805 }
4806
4807 static ssize_t
4808 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4809 {
4810 sector_t sectors;
4811 int err;
4812
4813 err = mddev_lock(mddev);
4814 if (err)
4815 return err;
4816
4817 if (strncmp(buf, "default", 7) == 0) {
4818 if (mddev->pers)
4819 sectors = mddev->pers->size(mddev, 0, 0);
4820 else
4821 sectors = mddev->array_sectors;
4822
4823 mddev->external_size = 0;
4824 } else {
4825 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4826 err = -EINVAL;
4827 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4828 err = -E2BIG;
4829 else
4830 mddev->external_size = 1;
4831 }
4832
4833 if (!err) {
4834 mddev->array_sectors = sectors;
4835 if (mddev->pers) {
4836 set_capacity(mddev->gendisk, mddev->array_sectors);
4837 revalidate_disk(mddev->gendisk);
4838 }
4839 }
4840 mddev_unlock(mddev);
4841 return err ?: len;
4842 }
4843
4844 static struct md_sysfs_entry md_array_size =
4845 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4846 array_size_store);
4847
4848 static struct attribute *md_default_attrs[] = {
4849 &md_level.attr,
4850 &md_layout.attr,
4851 &md_raid_disks.attr,
4852 &md_chunk_size.attr,
4853 &md_size.attr,
4854 &md_resync_start.attr,
4855 &md_metadata.attr,
4856 &md_new_device.attr,
4857 &md_safe_delay.attr,
4858 &md_array_state.attr,
4859 &md_reshape_position.attr,
4860 &md_reshape_direction.attr,
4861 &md_array_size.attr,
4862 &max_corr_read_errors.attr,
4863 NULL,
4864 };
4865
4866 static struct attribute *md_redundancy_attrs[] = {
4867 &md_scan_mode.attr,
4868 &md_last_scan_mode.attr,
4869 &md_mismatches.attr,
4870 &md_sync_min.attr,
4871 &md_sync_max.attr,
4872 &md_sync_speed.attr,
4873 &md_sync_force_parallel.attr,
4874 &md_sync_completed.attr,
4875 &md_min_sync.attr,
4876 &md_max_sync.attr,
4877 &md_suspend_lo.attr,
4878 &md_suspend_hi.attr,
4879 &md_bitmap.attr,
4880 &md_degraded.attr,
4881 NULL,
4882 };
4883 static struct attribute_group md_redundancy_group = {
4884 .name = NULL,
4885 .attrs = md_redundancy_attrs,
4886 };
4887
4888 static ssize_t
4889 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4890 {
4891 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4892 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4893 ssize_t rv;
4894
4895 if (!entry->show)
4896 return -EIO;
4897 spin_lock(&all_mddevs_lock);
4898 if (list_empty(&mddev->all_mddevs)) {
4899 spin_unlock(&all_mddevs_lock);
4900 return -EBUSY;
4901 }
4902 mddev_get(mddev);
4903 spin_unlock(&all_mddevs_lock);
4904
4905 rv = entry->show(mddev, page);
4906 mddev_put(mddev);
4907 return rv;
4908 }
4909
4910 static ssize_t
4911 md_attr_store(struct kobject *kobj, struct attribute *attr,
4912 const char *page, size_t length)
4913 {
4914 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4915 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4916 ssize_t rv;
4917
4918 if (!entry->store)
4919 return -EIO;
4920 if (!capable(CAP_SYS_ADMIN))
4921 return -EACCES;
4922 spin_lock(&all_mddevs_lock);
4923 if (list_empty(&mddev->all_mddevs)) {
4924 spin_unlock(&all_mddevs_lock);
4925 return -EBUSY;
4926 }
4927 mddev_get(mddev);
4928 spin_unlock(&all_mddevs_lock);
4929 rv = entry->store(mddev, page, length);
4930 mddev_put(mddev);
4931 return rv;
4932 }
4933
4934 static void md_free(struct kobject *ko)
4935 {
4936 struct mddev *mddev = container_of(ko, struct mddev, kobj);
4937
4938 if (mddev->sysfs_state)
4939 sysfs_put(mddev->sysfs_state);
4940
4941 if (mddev->queue)
4942 blk_cleanup_queue(mddev->queue);
4943 if (mddev->gendisk) {
4944 del_gendisk(mddev->gendisk);
4945 put_disk(mddev->gendisk);
4946 }
4947
4948 kfree(mddev);
4949 }
4950
4951 static const struct sysfs_ops md_sysfs_ops = {
4952 .show = md_attr_show,
4953 .store = md_attr_store,
4954 };
4955 static struct kobj_type md_ktype = {
4956 .release = md_free,
4957 .sysfs_ops = &md_sysfs_ops,
4958 .default_attrs = md_default_attrs,
4959 };
4960
4961 int mdp_major = 0;
4962
4963 static void mddev_delayed_delete(struct work_struct *ws)
4964 {
4965 struct mddev *mddev = container_of(ws, struct mddev, del_work);
4966
4967 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4968 kobject_del(&mddev->kobj);
4969 kobject_put(&mddev->kobj);
4970 }
4971
4972 static int md_alloc(dev_t dev, char *name)
4973 {
4974 static DEFINE_MUTEX(disks_mutex);
4975 struct mddev *mddev = mddev_find(dev);
4976 struct gendisk *disk;
4977 int partitioned;
4978 int shift;
4979 int unit;
4980 int error;
4981
4982 if (!mddev)
4983 return -ENODEV;
4984
4985 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4986 shift = partitioned ? MdpMinorShift : 0;
4987 unit = MINOR(mddev->unit) >> shift;
4988
4989 /* wait for any previous instance of this device to be
4990 * completely removed (mddev_delayed_delete).
4991 */
4992 flush_workqueue(md_misc_wq);
4993
4994 mutex_lock(&disks_mutex);
4995 error = -EEXIST;
4996 if (mddev->gendisk)
4997 goto abort;
4998
4999 if (name) {
5000 /* Need to ensure that 'name' is not a duplicate.
5001 */
5002 struct mddev *mddev2;
5003 spin_lock(&all_mddevs_lock);
5004
5005 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5006 if (mddev2->gendisk &&
5007 strcmp(mddev2->gendisk->disk_name, name) == 0) {
5008 spin_unlock(&all_mddevs_lock);
5009 goto abort;
5010 }
5011 spin_unlock(&all_mddevs_lock);
5012 }
5013
5014 error = -ENOMEM;
5015 mddev->queue = blk_alloc_queue(GFP_KERNEL);
5016 if (!mddev->queue)
5017 goto abort;
5018 mddev->queue->queuedata = mddev;
5019
5020 blk_queue_make_request(mddev->queue, md_make_request);
5021 blk_set_stacking_limits(&mddev->queue->limits);
5022
5023 disk = alloc_disk(1 << shift);
5024 if (!disk) {
5025 blk_cleanup_queue(mddev->queue);
5026 mddev->queue = NULL;
5027 goto abort;
5028 }
5029 disk->major = MAJOR(mddev->unit);
5030 disk->first_minor = unit << shift;
5031 if (name)
5032 strcpy(disk->disk_name, name);
5033 else if (partitioned)
5034 sprintf(disk->disk_name, "md_d%d", unit);
5035 else
5036 sprintf(disk->disk_name, "md%d", unit);
5037 disk->fops = &md_fops;
5038 disk->private_data = mddev;
5039 disk->queue = mddev->queue;
5040 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
5041 /* Allow extended partitions. This makes the
5042 * 'mdp' device redundant, but we can't really
5043 * remove it now.
5044 */
5045 disk->flags |= GENHD_FL_EXT_DEVT;
5046 mddev->gendisk = disk;
5047 /* As soon as we call add_disk(), another thread could get
5048 * through to md_open, so make sure it doesn't get too far
5049 */
5050 mutex_lock(&mddev->open_mutex);
5051 add_disk(disk);
5052
5053 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
5054 &disk_to_dev(disk)->kobj, "%s", "md");
5055 if (error) {
5056 /* This isn't possible, but as kobject_init_and_add is marked
5057 * __must_check, we must do something with the result
5058 */
5059 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
5060 disk->disk_name);
5061 error = 0;
5062 }
5063 if (mddev->kobj.sd &&
5064 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
5065 printk(KERN_DEBUG "pointless warning\n");
5066 mutex_unlock(&mddev->open_mutex);
5067 abort:
5068 mutex_unlock(&disks_mutex);
5069 if (!error && mddev->kobj.sd) {
5070 kobject_uevent(&mddev->kobj, KOBJ_ADD);
5071 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5072 }
5073 mddev_put(mddev);
5074 return error;
5075 }
5076
5077 static struct kobject *md_probe(dev_t dev, int *part, void *data)
5078 {
5079 md_alloc(dev, NULL);
5080 return NULL;
5081 }
5082
5083 static int add_named_array(const char *val, struct kernel_param *kp)
5084 {
5085 /* val must be "md_*" where * is not all digits.
5086 * We allocate an array with a large free minor number, and
5087 * set the name to val. val must not already be an active name.
5088 */
5089 int len = strlen(val);
5090 char buf[DISK_NAME_LEN];
5091
5092 while (len && val[len-1] == '\n')
5093 len--;
5094 if (len >= DISK_NAME_LEN)
5095 return -E2BIG;
5096 strlcpy(buf, val, len+1);
5097 if (strncmp(buf, "md_", 3) != 0)
5098 return -EINVAL;
5099 return md_alloc(0, buf);
5100 }
5101
5102 static void md_safemode_timeout(unsigned long data)
5103 {
5104 struct mddev *mddev = (struct mddev *) data;
5105
5106 if (!atomic_read(&mddev->writes_pending)) {
5107 mddev->safemode = 1;
5108 if (mddev->external)
5109 sysfs_notify_dirent_safe(mddev->sysfs_state);
5110 }
5111 md_wakeup_thread(mddev->thread);
5112 }
5113
5114 static int start_dirty_degraded;
5115
5116 int md_run(struct mddev *mddev)
5117 {
5118 int err;
5119 struct md_rdev *rdev;
5120 struct md_personality *pers;
5121
5122 if (list_empty(&mddev->disks))
5123 /* cannot run an array with no devices.. */
5124 return -EINVAL;
5125
5126 if (mddev->pers)
5127 return -EBUSY;
5128 /* Cannot run until previous stop completes properly */
5129 if (mddev->sysfs_active)
5130 return -EBUSY;
5131
5132 /*
5133 * Analyze all RAID superblock(s)
5134 */
5135 if (!mddev->raid_disks) {
5136 if (!mddev->persistent)
5137 return -EINVAL;
5138 analyze_sbs(mddev);
5139 }
5140
5141 if (mddev->level != LEVEL_NONE)
5142 request_module("md-level-%d", mddev->level);
5143 else if (mddev->clevel[0])
5144 request_module("md-%s", mddev->clevel);
5145
5146 /*
5147 * Drop all container device buffers, from now on
5148 * the only valid external interface is through the md
5149 * device.
5150 */
5151 rdev_for_each(rdev, mddev) {
5152 if (test_bit(Faulty, &rdev->flags))
5153 continue;
5154 sync_blockdev(rdev->bdev);
5155 invalidate_bdev(rdev->bdev);
5156
5157 /* perform some consistency tests on the device.
5158 * We don't want the data to overlap the metadata,
5159 * Internal Bitmap issues have been handled elsewhere.
5160 */
5161 if (rdev->meta_bdev) {
5162 /* Nothing to check */;
5163 } else if (rdev->data_offset < rdev->sb_start) {
5164 if (mddev->dev_sectors &&
5165 rdev->data_offset + mddev->dev_sectors
5166 > rdev->sb_start) {
5167 printk("md: %s: data overlaps metadata\n",
5168 mdname(mddev));
5169 return -EINVAL;
5170 }
5171 } else {
5172 if (rdev->sb_start + rdev->sb_size/512
5173 > rdev->data_offset) {
5174 printk("md: %s: metadata overlaps data\n",
5175 mdname(mddev));
5176 return -EINVAL;
5177 }
5178 }
5179 sysfs_notify_dirent_safe(rdev->sysfs_state);
5180 }
5181
5182 if (mddev->bio_set == NULL)
5183 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
5184
5185 spin_lock(&pers_lock);
5186 pers = find_pers(mddev->level, mddev->clevel);
5187 if (!pers || !try_module_get(pers->owner)) {
5188 spin_unlock(&pers_lock);
5189 if (mddev->level != LEVEL_NONE)
5190 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5191 mddev->level);
5192 else
5193 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5194 mddev->clevel);
5195 return -EINVAL;
5196 }
5197 spin_unlock(&pers_lock);
5198 if (mddev->level != pers->level) {
5199 mddev->level = pers->level;
5200 mddev->new_level = pers->level;
5201 }
5202 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5203
5204 if (mddev->reshape_position != MaxSector &&
5205 pers->start_reshape == NULL) {
5206 /* This personality cannot handle reshaping... */
5207 module_put(pers->owner);
5208 return -EINVAL;
5209 }
5210
5211 if (pers->sync_request) {
5212 /* Warn if this is a potentially silly
5213 * configuration.
5214 */
5215 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5216 struct md_rdev *rdev2;
5217 int warned = 0;
5218
5219 rdev_for_each(rdev, mddev)
5220 rdev_for_each(rdev2, mddev) {
5221 if (rdev < rdev2 &&
5222 rdev->bdev->bd_contains ==
5223 rdev2->bdev->bd_contains) {
5224 printk(KERN_WARNING
5225 "%s: WARNING: %s appears to be"
5226 " on the same physical disk as"
5227 " %s.\n",
5228 mdname(mddev),
5229 bdevname(rdev->bdev,b),
5230 bdevname(rdev2->bdev,b2));
5231 warned = 1;
5232 }
5233 }
5234
5235 if (warned)
5236 printk(KERN_WARNING
5237 "True protection against single-disk"
5238 " failure might be compromised.\n");
5239 }
5240
5241 mddev->recovery = 0;
5242 /* may be over-ridden by personality */
5243 mddev->resync_max_sectors = mddev->dev_sectors;
5244
5245 mddev->ok_start_degraded = start_dirty_degraded;
5246
5247 if (start_readonly && mddev->ro == 0)
5248 mddev->ro = 2; /* read-only, but switch on first write */
5249
5250 err = pers->run(mddev);
5251 if (err)
5252 printk(KERN_ERR "md: pers->run() failed ...\n");
5253 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5254 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5255 " but 'external_size' not in effect?\n", __func__);
5256 printk(KERN_ERR
5257 "md: invalid array_size %llu > default size %llu\n",
5258 (unsigned long long)mddev->array_sectors / 2,
5259 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5260 err = -EINVAL;
5261 }
5262 if (err == 0 && pers->sync_request &&
5263 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5264 struct bitmap *bitmap;
5265
5266 bitmap = bitmap_create(mddev, -1);
5267 if (IS_ERR(bitmap)) {
5268 err = PTR_ERR(bitmap);
5269 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5270 mdname(mddev), err);
5271 } else
5272 mddev->bitmap = bitmap;
5273
5274 }
5275 if (err) {
5276 mddev_detach(mddev);
5277 if (mddev->private)
5278 pers->free(mddev, mddev->private);
5279 mddev->private = NULL;
5280 module_put(pers->owner);
5281 bitmap_destroy(mddev);
5282 return err;
5283 }
5284 if (mddev->queue) {
5285 mddev->queue->backing_dev_info.congested_data = mddev;
5286 mddev->queue->backing_dev_info.congested_fn = md_congested;
5287 }
5288 if (pers->sync_request) {
5289 if (mddev->kobj.sd &&
5290 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5291 printk(KERN_WARNING
5292 "md: cannot register extra attributes for %s\n",
5293 mdname(mddev));
5294 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5295 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5296 mddev->ro = 0;
5297
5298 atomic_set(&mddev->writes_pending,0);
5299 atomic_set(&mddev->max_corr_read_errors,
5300 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5301 mddev->safemode = 0;
5302 if (mddev_is_clustered(mddev))
5303 mddev->safemode_delay = 0;
5304 else
5305 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5306 mddev->in_sync = 1;
5307 smp_wmb();
5308 spin_lock(&mddev->lock);
5309 mddev->pers = pers;
5310 spin_unlock(&mddev->lock);
5311 rdev_for_each(rdev, mddev)
5312 if (rdev->raid_disk >= 0)
5313 if (sysfs_link_rdev(mddev, rdev))
5314 /* failure here is OK */;
5315
5316 if (mddev->degraded && !mddev->ro)
5317 /* This ensures that recovering status is reported immediately
5318 * via sysfs - until a lack of spares is confirmed.
5319 */
5320 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5321 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5322
5323 if (mddev->flags & MD_UPDATE_SB_FLAGS)
5324 md_update_sb(mddev, 0);
5325
5326 md_new_event(mddev);
5327 sysfs_notify_dirent_safe(mddev->sysfs_state);
5328 sysfs_notify_dirent_safe(mddev->sysfs_action);
5329 sysfs_notify(&mddev->kobj, NULL, "degraded");
5330 return 0;
5331 }
5332 EXPORT_SYMBOL_GPL(md_run);
5333
5334 static int do_md_run(struct mddev *mddev)
5335 {
5336 int err;
5337
5338 err = md_run(mddev);
5339 if (err)
5340 goto out;
5341 err = bitmap_load(mddev);
5342 if (err) {
5343 bitmap_destroy(mddev);
5344 goto out;
5345 }
5346
5347 if (mddev_is_clustered(mddev))
5348 md_allow_write(mddev);
5349
5350 md_wakeup_thread(mddev->thread);
5351 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5352
5353 set_capacity(mddev->gendisk, mddev->array_sectors);
5354 revalidate_disk(mddev->gendisk);
5355 mddev->changed = 1;
5356 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5357 out:
5358 return err;
5359 }
5360
5361 static int restart_array(struct mddev *mddev)
5362 {
5363 struct gendisk *disk = mddev->gendisk;
5364
5365 /* Complain if it has no devices */
5366 if (list_empty(&mddev->disks))
5367 return -ENXIO;
5368 if (!mddev->pers)
5369 return -EINVAL;
5370 if (!mddev->ro)
5371 return -EBUSY;
5372 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5373 struct md_rdev *rdev;
5374 bool has_journal = false;
5375
5376 rcu_read_lock();
5377 rdev_for_each_rcu(rdev, mddev) {
5378 if (test_bit(Journal, &rdev->flags) &&
5379 !test_bit(Faulty, &rdev->flags)) {
5380 has_journal = true;
5381 break;
5382 }
5383 }
5384 rcu_read_unlock();
5385
5386 /* Don't restart rw with journal missing/faulty */
5387 if (!has_journal)
5388 return -EINVAL;
5389 }
5390
5391 mddev->safemode = 0;
5392 mddev->ro = 0;
5393 set_disk_ro(disk, 0);
5394 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5395 mdname(mddev));
5396 /* Kick recovery or resync if necessary */
5397 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5398 md_wakeup_thread(mddev->thread);
5399 md_wakeup_thread(mddev->sync_thread);
5400 sysfs_notify_dirent_safe(mddev->sysfs_state);
5401 return 0;
5402 }
5403
5404 static void md_clean(struct mddev *mddev)
5405 {
5406 mddev->array_sectors = 0;
5407 mddev->external_size = 0;
5408 mddev->dev_sectors = 0;
5409 mddev->raid_disks = 0;
5410 mddev->recovery_cp = 0;
5411 mddev->resync_min = 0;
5412 mddev->resync_max = MaxSector;
5413 mddev->reshape_position = MaxSector;
5414 mddev->external = 0;
5415 mddev->persistent = 0;
5416 mddev->level = LEVEL_NONE;
5417 mddev->clevel[0] = 0;
5418 mddev->flags = 0;
5419 mddev->ro = 0;
5420 mddev->metadata_type[0] = 0;
5421 mddev->chunk_sectors = 0;
5422 mddev->ctime = mddev->utime = 0;
5423 mddev->layout = 0;
5424 mddev->max_disks = 0;
5425 mddev->events = 0;
5426 mddev->can_decrease_events = 0;
5427 mddev->delta_disks = 0;
5428 mddev->reshape_backwards = 0;
5429 mddev->new_level = LEVEL_NONE;
5430 mddev->new_layout = 0;
5431 mddev->new_chunk_sectors = 0;
5432 mddev->curr_resync = 0;
5433 atomic64_set(&mddev->resync_mismatches, 0);
5434 mddev->suspend_lo = mddev->suspend_hi = 0;
5435 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5436 mddev->recovery = 0;
5437 mddev->in_sync = 0;
5438 mddev->changed = 0;
5439 mddev->degraded = 0;
5440 mddev->safemode = 0;
5441 mddev->private = NULL;
5442 mddev->bitmap_info.offset = 0;
5443 mddev->bitmap_info.default_offset = 0;
5444 mddev->bitmap_info.default_space = 0;
5445 mddev->bitmap_info.chunksize = 0;
5446 mddev->bitmap_info.daemon_sleep = 0;
5447 mddev->bitmap_info.max_write_behind = 0;
5448 }
5449
5450 static void __md_stop_writes(struct mddev *mddev)
5451 {
5452 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5453 flush_workqueue(md_misc_wq);
5454 if (mddev->sync_thread) {
5455 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5456 md_reap_sync_thread(mddev);
5457 }
5458
5459 del_timer_sync(&mddev->safemode_timer);
5460
5461 bitmap_flush(mddev);
5462 md_super_wait(mddev);
5463
5464 if (mddev->ro == 0 &&
5465 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
5466 (mddev->flags & MD_UPDATE_SB_FLAGS))) {
5467 /* mark array as shutdown cleanly */
5468 if (!mddev_is_clustered(mddev))
5469 mddev->in_sync = 1;
5470 md_update_sb(mddev, 1);
5471 }
5472 }
5473
5474 void md_stop_writes(struct mddev *mddev)
5475 {
5476 mddev_lock_nointr(mddev);
5477 __md_stop_writes(mddev);
5478 mddev_unlock(mddev);
5479 }
5480 EXPORT_SYMBOL_GPL(md_stop_writes);
5481
5482 static void mddev_detach(struct mddev *mddev)
5483 {
5484 struct bitmap *bitmap = mddev->bitmap;
5485 /* wait for behind writes to complete */
5486 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
5487 printk(KERN_INFO "md:%s: behind writes in progress - waiting to stop.\n",
5488 mdname(mddev));
5489 /* need to kick something here to make sure I/O goes? */
5490 wait_event(bitmap->behind_wait,
5491 atomic_read(&bitmap->behind_writes) == 0);
5492 }
5493 if (mddev->pers && mddev->pers->quiesce) {
5494 mddev->pers->quiesce(mddev, 1);
5495 mddev->pers->quiesce(mddev, 0);
5496 }
5497 md_unregister_thread(&mddev->thread);
5498 if (mddev->queue)
5499 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5500 }
5501
5502 static void __md_stop(struct mddev *mddev)
5503 {
5504 struct md_personality *pers = mddev->pers;
5505 mddev_detach(mddev);
5506 /* Ensure ->event_work is done */
5507 flush_workqueue(md_misc_wq);
5508 spin_lock(&mddev->lock);
5509 mddev->pers = NULL;
5510 spin_unlock(&mddev->lock);
5511 pers->free(mddev, mddev->private);
5512 mddev->private = NULL;
5513 if (pers->sync_request && mddev->to_remove == NULL)
5514 mddev->to_remove = &md_redundancy_group;
5515 module_put(pers->owner);
5516 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5517 }
5518
5519 void md_stop(struct mddev *mddev)
5520 {
5521 /* stop the array and free an attached data structures.
5522 * This is called from dm-raid
5523 */
5524 __md_stop(mddev);
5525 bitmap_destroy(mddev);
5526 if (mddev->bio_set)
5527 bioset_free(mddev->bio_set);
5528 }
5529
5530 EXPORT_SYMBOL_GPL(md_stop);
5531
5532 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5533 {
5534 int err = 0;
5535 int did_freeze = 0;
5536
5537 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5538 did_freeze = 1;
5539 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5540 md_wakeup_thread(mddev->thread);
5541 }
5542 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5543 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5544 if (mddev->sync_thread)
5545 /* Thread might be blocked waiting for metadata update
5546 * which will now never happen */
5547 wake_up_process(mddev->sync_thread->tsk);
5548
5549 if (mddev->external && test_bit(MD_CHANGE_PENDING, &mddev->flags))
5550 return -EBUSY;
5551 mddev_unlock(mddev);
5552 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5553 &mddev->recovery));
5554 wait_event(mddev->sb_wait,
5555 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5556 mddev_lock_nointr(mddev);
5557
5558 mutex_lock(&mddev->open_mutex);
5559 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5560 mddev->sync_thread ||
5561 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5562 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5563 printk("md: %s still in use.\n",mdname(mddev));
5564 if (did_freeze) {
5565 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5566 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5567 md_wakeup_thread(mddev->thread);
5568 }
5569 err = -EBUSY;
5570 goto out;
5571 }
5572 if (mddev->pers) {
5573 __md_stop_writes(mddev);
5574
5575 err = -ENXIO;
5576 if (mddev->ro==1)
5577 goto out;
5578 mddev->ro = 1;
5579 set_disk_ro(mddev->gendisk, 1);
5580 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5581 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5582 md_wakeup_thread(mddev->thread);
5583 sysfs_notify_dirent_safe(mddev->sysfs_state);
5584 err = 0;
5585 }
5586 out:
5587 mutex_unlock(&mddev->open_mutex);
5588 return err;
5589 }
5590
5591 /* mode:
5592 * 0 - completely stop and dis-assemble array
5593 * 2 - stop but do not disassemble array
5594 */
5595 static int do_md_stop(struct mddev *mddev, int mode,
5596 struct block_device *bdev)
5597 {
5598 struct gendisk *disk = mddev->gendisk;
5599 struct md_rdev *rdev;
5600 int did_freeze = 0;
5601
5602 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5603 did_freeze = 1;
5604 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5605 md_wakeup_thread(mddev->thread);
5606 }
5607 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5608 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5609 if (mddev->sync_thread)
5610 /* Thread might be blocked waiting for metadata update
5611 * which will now never happen */
5612 wake_up_process(mddev->sync_thread->tsk);
5613
5614 mddev_unlock(mddev);
5615 wait_event(resync_wait, (mddev->sync_thread == NULL &&
5616 !test_bit(MD_RECOVERY_RUNNING,
5617 &mddev->recovery)));
5618 mddev_lock_nointr(mddev);
5619
5620 mutex_lock(&mddev->open_mutex);
5621 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5622 mddev->sysfs_active ||
5623 mddev->sync_thread ||
5624 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5625 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5626 printk("md: %s still in use.\n",mdname(mddev));
5627 mutex_unlock(&mddev->open_mutex);
5628 if (did_freeze) {
5629 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5630 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5631 md_wakeup_thread(mddev->thread);
5632 }
5633 return -EBUSY;
5634 }
5635 if (mddev->pers) {
5636 if (mddev->ro)
5637 set_disk_ro(disk, 0);
5638
5639 __md_stop_writes(mddev);
5640 __md_stop(mddev);
5641 mddev->queue->backing_dev_info.congested_fn = NULL;
5642
5643 /* tell userspace to handle 'inactive' */
5644 sysfs_notify_dirent_safe(mddev->sysfs_state);
5645
5646 rdev_for_each(rdev, mddev)
5647 if (rdev->raid_disk >= 0)
5648 sysfs_unlink_rdev(mddev, rdev);
5649
5650 set_capacity(disk, 0);
5651 mutex_unlock(&mddev->open_mutex);
5652 mddev->changed = 1;
5653 revalidate_disk(disk);
5654
5655 if (mddev->ro)
5656 mddev->ro = 0;
5657 } else
5658 mutex_unlock(&mddev->open_mutex);
5659 /*
5660 * Free resources if final stop
5661 */
5662 if (mode == 0) {
5663 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5664
5665 bitmap_destroy(mddev);
5666 if (mddev->bitmap_info.file) {
5667 struct file *f = mddev->bitmap_info.file;
5668 spin_lock(&mddev->lock);
5669 mddev->bitmap_info.file = NULL;
5670 spin_unlock(&mddev->lock);
5671 fput(f);
5672 }
5673 mddev->bitmap_info.offset = 0;
5674
5675 export_array(mddev);
5676
5677 md_clean(mddev);
5678 if (mddev->hold_active == UNTIL_STOP)
5679 mddev->hold_active = 0;
5680 }
5681 md_new_event(mddev);
5682 sysfs_notify_dirent_safe(mddev->sysfs_state);
5683 return 0;
5684 }
5685
5686 #ifndef MODULE
5687 static void autorun_array(struct mddev *mddev)
5688 {
5689 struct md_rdev *rdev;
5690 int err;
5691
5692 if (list_empty(&mddev->disks))
5693 return;
5694
5695 printk(KERN_INFO "md: running: ");
5696
5697 rdev_for_each(rdev, mddev) {
5698 char b[BDEVNAME_SIZE];
5699 printk("<%s>", bdevname(rdev->bdev,b));
5700 }
5701 printk("\n");
5702
5703 err = do_md_run(mddev);
5704 if (err) {
5705 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
5706 do_md_stop(mddev, 0, NULL);
5707 }
5708 }
5709
5710 /*
5711 * lets try to run arrays based on all disks that have arrived
5712 * until now. (those are in pending_raid_disks)
5713 *
5714 * the method: pick the first pending disk, collect all disks with
5715 * the same UUID, remove all from the pending list and put them into
5716 * the 'same_array' list. Then order this list based on superblock
5717 * update time (freshest comes first), kick out 'old' disks and
5718 * compare superblocks. If everything's fine then run it.
5719 *
5720 * If "unit" is allocated, then bump its reference count
5721 */
5722 static void autorun_devices(int part)
5723 {
5724 struct md_rdev *rdev0, *rdev, *tmp;
5725 struct mddev *mddev;
5726 char b[BDEVNAME_SIZE];
5727
5728 printk(KERN_INFO "md: autorun ...\n");
5729 while (!list_empty(&pending_raid_disks)) {
5730 int unit;
5731 dev_t dev;
5732 LIST_HEAD(candidates);
5733 rdev0 = list_entry(pending_raid_disks.next,
5734 struct md_rdev, same_set);
5735
5736 printk(KERN_INFO "md: considering %s ...\n",
5737 bdevname(rdev0->bdev,b));
5738 INIT_LIST_HEAD(&candidates);
5739 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5740 if (super_90_load(rdev, rdev0, 0) >= 0) {
5741 printk(KERN_INFO "md: adding %s ...\n",
5742 bdevname(rdev->bdev,b));
5743 list_move(&rdev->same_set, &candidates);
5744 }
5745 /*
5746 * now we have a set of devices, with all of them having
5747 * mostly sane superblocks. It's time to allocate the
5748 * mddev.
5749 */
5750 if (part) {
5751 dev = MKDEV(mdp_major,
5752 rdev0->preferred_minor << MdpMinorShift);
5753 unit = MINOR(dev) >> MdpMinorShift;
5754 } else {
5755 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5756 unit = MINOR(dev);
5757 }
5758 if (rdev0->preferred_minor != unit) {
5759 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5760 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5761 break;
5762 }
5763
5764 md_probe(dev, NULL, NULL);
5765 mddev = mddev_find(dev);
5766 if (!mddev || !mddev->gendisk) {
5767 if (mddev)
5768 mddev_put(mddev);
5769 printk(KERN_ERR
5770 "md: cannot allocate memory for md drive.\n");
5771 break;
5772 }
5773 if (mddev_lock(mddev))
5774 printk(KERN_WARNING "md: %s locked, cannot run\n",
5775 mdname(mddev));
5776 else if (mddev->raid_disks || mddev->major_version
5777 || !list_empty(&mddev->disks)) {
5778 printk(KERN_WARNING
5779 "md: %s already running, cannot run %s\n",
5780 mdname(mddev), bdevname(rdev0->bdev,b));
5781 mddev_unlock(mddev);
5782 } else {
5783 printk(KERN_INFO "md: created %s\n", mdname(mddev));
5784 mddev->persistent = 1;
5785 rdev_for_each_list(rdev, tmp, &candidates) {
5786 list_del_init(&rdev->same_set);
5787 if (bind_rdev_to_array(rdev, mddev))
5788 export_rdev(rdev);
5789 }
5790 autorun_array(mddev);
5791 mddev_unlock(mddev);
5792 }
5793 /* on success, candidates will be empty, on error
5794 * it won't...
5795 */
5796 rdev_for_each_list(rdev, tmp, &candidates) {
5797 list_del_init(&rdev->same_set);
5798 export_rdev(rdev);
5799 }
5800 mddev_put(mddev);
5801 }
5802 printk(KERN_INFO "md: ... autorun DONE.\n");
5803 }
5804 #endif /* !MODULE */
5805
5806 static int get_version(void __user *arg)
5807 {
5808 mdu_version_t ver;
5809
5810 ver.major = MD_MAJOR_VERSION;
5811 ver.minor = MD_MINOR_VERSION;
5812 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5813
5814 if (copy_to_user(arg, &ver, sizeof(ver)))
5815 return -EFAULT;
5816
5817 return 0;
5818 }
5819
5820 static int get_array_info(struct mddev *mddev, void __user *arg)
5821 {
5822 mdu_array_info_t info;
5823 int nr,working,insync,failed,spare;
5824 struct md_rdev *rdev;
5825
5826 nr = working = insync = failed = spare = 0;
5827 rcu_read_lock();
5828 rdev_for_each_rcu(rdev, mddev) {
5829 nr++;
5830 if (test_bit(Faulty, &rdev->flags))
5831 failed++;
5832 else {
5833 working++;
5834 if (test_bit(In_sync, &rdev->flags))
5835 insync++;
5836 else
5837 spare++;
5838 }
5839 }
5840 rcu_read_unlock();
5841
5842 info.major_version = mddev->major_version;
5843 info.minor_version = mddev->minor_version;
5844 info.patch_version = MD_PATCHLEVEL_VERSION;
5845 info.ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
5846 info.level = mddev->level;
5847 info.size = mddev->dev_sectors / 2;
5848 if (info.size != mddev->dev_sectors / 2) /* overflow */
5849 info.size = -1;
5850 info.nr_disks = nr;
5851 info.raid_disks = mddev->raid_disks;
5852 info.md_minor = mddev->md_minor;
5853 info.not_persistent= !mddev->persistent;
5854
5855 info.utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
5856 info.state = 0;
5857 if (mddev->in_sync)
5858 info.state = (1<<MD_SB_CLEAN);
5859 if (mddev->bitmap && mddev->bitmap_info.offset)
5860 info.state |= (1<<MD_SB_BITMAP_PRESENT);
5861 if (mddev_is_clustered(mddev))
5862 info.state |= (1<<MD_SB_CLUSTERED);
5863 info.active_disks = insync;
5864 info.working_disks = working;
5865 info.failed_disks = failed;
5866 info.spare_disks = spare;
5867
5868 info.layout = mddev->layout;
5869 info.chunk_size = mddev->chunk_sectors << 9;
5870
5871 if (copy_to_user(arg, &info, sizeof(info)))
5872 return -EFAULT;
5873
5874 return 0;
5875 }
5876
5877 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
5878 {
5879 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5880 char *ptr;
5881 int err;
5882
5883 file = kzalloc(sizeof(*file), GFP_NOIO);
5884 if (!file)
5885 return -ENOMEM;
5886
5887 err = 0;
5888 spin_lock(&mddev->lock);
5889 /* bitmap enabled */
5890 if (mddev->bitmap_info.file) {
5891 ptr = file_path(mddev->bitmap_info.file, file->pathname,
5892 sizeof(file->pathname));
5893 if (IS_ERR(ptr))
5894 err = PTR_ERR(ptr);
5895 else
5896 memmove(file->pathname, ptr,
5897 sizeof(file->pathname)-(ptr-file->pathname));
5898 }
5899 spin_unlock(&mddev->lock);
5900
5901 if (err == 0 &&
5902 copy_to_user(arg, file, sizeof(*file)))
5903 err = -EFAULT;
5904
5905 kfree(file);
5906 return err;
5907 }
5908
5909 static int get_disk_info(struct mddev *mddev, void __user * arg)
5910 {
5911 mdu_disk_info_t info;
5912 struct md_rdev *rdev;
5913
5914 if (copy_from_user(&info, arg, sizeof(info)))
5915 return -EFAULT;
5916
5917 rcu_read_lock();
5918 rdev = md_find_rdev_nr_rcu(mddev, info.number);
5919 if (rdev) {
5920 info.major = MAJOR(rdev->bdev->bd_dev);
5921 info.minor = MINOR(rdev->bdev->bd_dev);
5922 info.raid_disk = rdev->raid_disk;
5923 info.state = 0;
5924 if (test_bit(Faulty, &rdev->flags))
5925 info.state |= (1<<MD_DISK_FAULTY);
5926 else if (test_bit(In_sync, &rdev->flags)) {
5927 info.state |= (1<<MD_DISK_ACTIVE);
5928 info.state |= (1<<MD_DISK_SYNC);
5929 }
5930 if (test_bit(Journal, &rdev->flags))
5931 info.state |= (1<<MD_DISK_JOURNAL);
5932 if (test_bit(WriteMostly, &rdev->flags))
5933 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5934 } else {
5935 info.major = info.minor = 0;
5936 info.raid_disk = -1;
5937 info.state = (1<<MD_DISK_REMOVED);
5938 }
5939 rcu_read_unlock();
5940
5941 if (copy_to_user(arg, &info, sizeof(info)))
5942 return -EFAULT;
5943
5944 return 0;
5945 }
5946
5947 static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
5948 {
5949 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5950 struct md_rdev *rdev;
5951 dev_t dev = MKDEV(info->major,info->minor);
5952
5953 if (mddev_is_clustered(mddev) &&
5954 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
5955 pr_err("%s: Cannot add to clustered mddev.\n",
5956 mdname(mddev));
5957 return -EINVAL;
5958 }
5959
5960 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5961 return -EOVERFLOW;
5962
5963 if (!mddev->raid_disks) {
5964 int err;
5965 /* expecting a device which has a superblock */
5966 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5967 if (IS_ERR(rdev)) {
5968 printk(KERN_WARNING
5969 "md: md_import_device returned %ld\n",
5970 PTR_ERR(rdev));
5971 return PTR_ERR(rdev);
5972 }
5973 if (!list_empty(&mddev->disks)) {
5974 struct md_rdev *rdev0
5975 = list_entry(mddev->disks.next,
5976 struct md_rdev, same_set);
5977 err = super_types[mddev->major_version]
5978 .load_super(rdev, rdev0, mddev->minor_version);
5979 if (err < 0) {
5980 printk(KERN_WARNING
5981 "md: %s has different UUID to %s\n",
5982 bdevname(rdev->bdev,b),
5983 bdevname(rdev0->bdev,b2));
5984 export_rdev(rdev);
5985 return -EINVAL;
5986 }
5987 }
5988 err = bind_rdev_to_array(rdev, mddev);
5989 if (err)
5990 export_rdev(rdev);
5991 return err;
5992 }
5993
5994 /*
5995 * add_new_disk can be used once the array is assembled
5996 * to add "hot spares". They must already have a superblock
5997 * written
5998 */
5999 if (mddev->pers) {
6000 int err;
6001 if (!mddev->pers->hot_add_disk) {
6002 printk(KERN_WARNING
6003 "%s: personality does not support diskops!\n",
6004 mdname(mddev));
6005 return -EINVAL;
6006 }
6007 if (mddev->persistent)
6008 rdev = md_import_device(dev, mddev->major_version,
6009 mddev->minor_version);
6010 else
6011 rdev = md_import_device(dev, -1, -1);
6012 if (IS_ERR(rdev)) {
6013 printk(KERN_WARNING
6014 "md: md_import_device returned %ld\n",
6015 PTR_ERR(rdev));
6016 return PTR_ERR(rdev);
6017 }
6018 /* set saved_raid_disk if appropriate */
6019 if (!mddev->persistent) {
6020 if (info->state & (1<<MD_DISK_SYNC) &&
6021 info->raid_disk < mddev->raid_disks) {
6022 rdev->raid_disk = info->raid_disk;
6023 set_bit(In_sync, &rdev->flags);
6024 clear_bit(Bitmap_sync, &rdev->flags);
6025 } else
6026 rdev->raid_disk = -1;
6027 rdev->saved_raid_disk = rdev->raid_disk;
6028 } else
6029 super_types[mddev->major_version].
6030 validate_super(mddev, rdev);
6031 if ((info->state & (1<<MD_DISK_SYNC)) &&
6032 rdev->raid_disk != info->raid_disk) {
6033 /* This was a hot-add request, but events doesn't
6034 * match, so reject it.
6035 */
6036 export_rdev(rdev);
6037 return -EINVAL;
6038 }
6039
6040 clear_bit(In_sync, &rdev->flags); /* just to be sure */
6041 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6042 set_bit(WriteMostly, &rdev->flags);
6043 else
6044 clear_bit(WriteMostly, &rdev->flags);
6045
6046 if (info->state & (1<<MD_DISK_JOURNAL)) {
6047 struct md_rdev *rdev2;
6048 bool has_journal = false;
6049
6050 /* make sure no existing journal disk */
6051 rdev_for_each(rdev2, mddev) {
6052 if (test_bit(Journal, &rdev2->flags)) {
6053 has_journal = true;
6054 break;
6055 }
6056 }
6057 if (has_journal) {
6058 export_rdev(rdev);
6059 return -EBUSY;
6060 }
6061 set_bit(Journal, &rdev->flags);
6062 }
6063 /*
6064 * check whether the device shows up in other nodes
6065 */
6066 if (mddev_is_clustered(mddev)) {
6067 if (info->state & (1 << MD_DISK_CANDIDATE))
6068 set_bit(Candidate, &rdev->flags);
6069 else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6070 /* --add initiated by this node */
6071 err = md_cluster_ops->add_new_disk(mddev, rdev);
6072 if (err) {
6073 export_rdev(rdev);
6074 return err;
6075 }
6076 }
6077 }
6078
6079 rdev->raid_disk = -1;
6080 err = bind_rdev_to_array(rdev, mddev);
6081
6082 if (err)
6083 export_rdev(rdev);
6084
6085 if (mddev_is_clustered(mddev)) {
6086 if (info->state & (1 << MD_DISK_CANDIDATE))
6087 md_cluster_ops->new_disk_ack(mddev, (err == 0));
6088 else {
6089 if (err)
6090 md_cluster_ops->add_new_disk_cancel(mddev);
6091 else
6092 err = add_bound_rdev(rdev);
6093 }
6094
6095 } else if (!err)
6096 err = add_bound_rdev(rdev);
6097
6098 return err;
6099 }
6100
6101 /* otherwise, add_new_disk is only allowed
6102 * for major_version==0 superblocks
6103 */
6104 if (mddev->major_version != 0) {
6105 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
6106 mdname(mddev));
6107 return -EINVAL;
6108 }
6109
6110 if (!(info->state & (1<<MD_DISK_FAULTY))) {
6111 int err;
6112 rdev = md_import_device(dev, -1, 0);
6113 if (IS_ERR(rdev)) {
6114 printk(KERN_WARNING
6115 "md: error, md_import_device() returned %ld\n",
6116 PTR_ERR(rdev));
6117 return PTR_ERR(rdev);
6118 }
6119 rdev->desc_nr = info->number;
6120 if (info->raid_disk < mddev->raid_disks)
6121 rdev->raid_disk = info->raid_disk;
6122 else
6123 rdev->raid_disk = -1;
6124
6125 if (rdev->raid_disk < mddev->raid_disks)
6126 if (info->state & (1<<MD_DISK_SYNC))
6127 set_bit(In_sync, &rdev->flags);
6128
6129 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6130 set_bit(WriteMostly, &rdev->flags);
6131
6132 if (!mddev->persistent) {
6133 printk(KERN_INFO "md: nonpersistent superblock ...\n");
6134 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6135 } else
6136 rdev->sb_start = calc_dev_sboffset(rdev);
6137 rdev->sectors = rdev->sb_start;
6138
6139 err = bind_rdev_to_array(rdev, mddev);
6140 if (err) {
6141 export_rdev(rdev);
6142 return err;
6143 }
6144 }
6145
6146 return 0;
6147 }
6148
6149 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6150 {
6151 char b[BDEVNAME_SIZE];
6152 struct md_rdev *rdev;
6153
6154 rdev = find_rdev(mddev, dev);
6155 if (!rdev)
6156 return -ENXIO;
6157
6158 if (rdev->raid_disk < 0)
6159 goto kick_rdev;
6160
6161 clear_bit(Blocked, &rdev->flags);
6162 remove_and_add_spares(mddev, rdev);
6163
6164 if (rdev->raid_disk >= 0)
6165 goto busy;
6166
6167 kick_rdev:
6168 if (mddev_is_clustered(mddev))
6169 md_cluster_ops->remove_disk(mddev, rdev);
6170
6171 md_kick_rdev_from_array(rdev);
6172 md_update_sb(mddev, 1);
6173 md_new_event(mddev);
6174
6175 return 0;
6176 busy:
6177 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
6178 bdevname(rdev->bdev,b), mdname(mddev));
6179 return -EBUSY;
6180 }
6181
6182 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6183 {
6184 char b[BDEVNAME_SIZE];
6185 int err;
6186 struct md_rdev *rdev;
6187
6188 if (!mddev->pers)
6189 return -ENODEV;
6190
6191 if (mddev->major_version != 0) {
6192 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
6193 " version-0 superblocks.\n",
6194 mdname(mddev));
6195 return -EINVAL;
6196 }
6197 if (!mddev->pers->hot_add_disk) {
6198 printk(KERN_WARNING
6199 "%s: personality does not support diskops!\n",
6200 mdname(mddev));
6201 return -EINVAL;
6202 }
6203
6204 rdev = md_import_device(dev, -1, 0);
6205 if (IS_ERR(rdev)) {
6206 printk(KERN_WARNING
6207 "md: error, md_import_device() returned %ld\n",
6208 PTR_ERR(rdev));
6209 return -EINVAL;
6210 }
6211
6212 if (mddev->persistent)
6213 rdev->sb_start = calc_dev_sboffset(rdev);
6214 else
6215 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6216
6217 rdev->sectors = rdev->sb_start;
6218
6219 if (test_bit(Faulty, &rdev->flags)) {
6220 printk(KERN_WARNING
6221 "md: can not hot-add faulty %s disk to %s!\n",
6222 bdevname(rdev->bdev,b), mdname(mddev));
6223 err = -EINVAL;
6224 goto abort_export;
6225 }
6226
6227 clear_bit(In_sync, &rdev->flags);
6228 rdev->desc_nr = -1;
6229 rdev->saved_raid_disk = -1;
6230 err = bind_rdev_to_array(rdev, mddev);
6231 if (err)
6232 goto abort_export;
6233
6234 /*
6235 * The rest should better be atomic, we can have disk failures
6236 * noticed in interrupt contexts ...
6237 */
6238
6239 rdev->raid_disk = -1;
6240
6241 md_update_sb(mddev, 1);
6242 /*
6243 * Kick recovery, maybe this spare has to be added to the
6244 * array immediately.
6245 */
6246 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6247 md_wakeup_thread(mddev->thread);
6248 md_new_event(mddev);
6249 return 0;
6250
6251 abort_export:
6252 export_rdev(rdev);
6253 return err;
6254 }
6255
6256 static int set_bitmap_file(struct mddev *mddev, int fd)
6257 {
6258 int err = 0;
6259
6260 if (mddev->pers) {
6261 if (!mddev->pers->quiesce || !mddev->thread)
6262 return -EBUSY;
6263 if (mddev->recovery || mddev->sync_thread)
6264 return -EBUSY;
6265 /* we should be able to change the bitmap.. */
6266 }
6267
6268 if (fd >= 0) {
6269 struct inode *inode;
6270 struct file *f;
6271
6272 if (mddev->bitmap || mddev->bitmap_info.file)
6273 return -EEXIST; /* cannot add when bitmap is present */
6274 f = fget(fd);
6275
6276 if (f == NULL) {
6277 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
6278 mdname(mddev));
6279 return -EBADF;
6280 }
6281
6282 inode = f->f_mapping->host;
6283 if (!S_ISREG(inode->i_mode)) {
6284 printk(KERN_ERR "%s: error: bitmap file must be a regular file\n",
6285 mdname(mddev));
6286 err = -EBADF;
6287 } else if (!(f->f_mode & FMODE_WRITE)) {
6288 printk(KERN_ERR "%s: error: bitmap file must open for write\n",
6289 mdname(mddev));
6290 err = -EBADF;
6291 } else if (atomic_read(&inode->i_writecount) != 1) {
6292 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
6293 mdname(mddev));
6294 err = -EBUSY;
6295 }
6296 if (err) {
6297 fput(f);
6298 return err;
6299 }
6300 mddev->bitmap_info.file = f;
6301 mddev->bitmap_info.offset = 0; /* file overrides offset */
6302 } else if (mddev->bitmap == NULL)
6303 return -ENOENT; /* cannot remove what isn't there */
6304 err = 0;
6305 if (mddev->pers) {
6306 mddev->pers->quiesce(mddev, 1);
6307 if (fd >= 0) {
6308 struct bitmap *bitmap;
6309
6310 bitmap = bitmap_create(mddev, -1);
6311 if (!IS_ERR(bitmap)) {
6312 mddev->bitmap = bitmap;
6313 err = bitmap_load(mddev);
6314 } else
6315 err = PTR_ERR(bitmap);
6316 }
6317 if (fd < 0 || err) {
6318 bitmap_destroy(mddev);
6319 fd = -1; /* make sure to put the file */
6320 }
6321 mddev->pers->quiesce(mddev, 0);
6322 }
6323 if (fd < 0) {
6324 struct file *f = mddev->bitmap_info.file;
6325 if (f) {
6326 spin_lock(&mddev->lock);
6327 mddev->bitmap_info.file = NULL;
6328 spin_unlock(&mddev->lock);
6329 fput(f);
6330 }
6331 }
6332
6333 return err;
6334 }
6335
6336 /*
6337 * set_array_info is used two different ways
6338 * The original usage is when creating a new array.
6339 * In this usage, raid_disks is > 0 and it together with
6340 * level, size, not_persistent,layout,chunksize determine the
6341 * shape of the array.
6342 * This will always create an array with a type-0.90.0 superblock.
6343 * The newer usage is when assembling an array.
6344 * In this case raid_disks will be 0, and the major_version field is
6345 * use to determine which style super-blocks are to be found on the devices.
6346 * The minor and patch _version numbers are also kept incase the
6347 * super_block handler wishes to interpret them.
6348 */
6349 static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
6350 {
6351
6352 if (info->raid_disks == 0) {
6353 /* just setting version number for superblock loading */
6354 if (info->major_version < 0 ||
6355 info->major_version >= ARRAY_SIZE(super_types) ||
6356 super_types[info->major_version].name == NULL) {
6357 /* maybe try to auto-load a module? */
6358 printk(KERN_INFO
6359 "md: superblock version %d not known\n",
6360 info->major_version);
6361 return -EINVAL;
6362 }
6363 mddev->major_version = info->major_version;
6364 mddev->minor_version = info->minor_version;
6365 mddev->patch_version = info->patch_version;
6366 mddev->persistent = !info->not_persistent;
6367 /* ensure mddev_put doesn't delete this now that there
6368 * is some minimal configuration.
6369 */
6370 mddev->ctime = ktime_get_real_seconds();
6371 return 0;
6372 }
6373 mddev->major_version = MD_MAJOR_VERSION;
6374 mddev->minor_version = MD_MINOR_VERSION;
6375 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6376 mddev->ctime = ktime_get_real_seconds();
6377
6378 mddev->level = info->level;
6379 mddev->clevel[0] = 0;
6380 mddev->dev_sectors = 2 * (sector_t)info->size;
6381 mddev->raid_disks = info->raid_disks;
6382 /* don't set md_minor, it is determined by which /dev/md* was
6383 * openned
6384 */
6385 if (info->state & (1<<MD_SB_CLEAN))
6386 mddev->recovery_cp = MaxSector;
6387 else
6388 mddev->recovery_cp = 0;
6389 mddev->persistent = ! info->not_persistent;
6390 mddev->external = 0;
6391
6392 mddev->layout = info->layout;
6393 mddev->chunk_sectors = info->chunk_size >> 9;
6394
6395 mddev->max_disks = MD_SB_DISKS;
6396
6397 if (mddev->persistent)
6398 mddev->flags = 0;
6399 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6400
6401 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6402 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6403 mddev->bitmap_info.offset = 0;
6404
6405 mddev->reshape_position = MaxSector;
6406
6407 /*
6408 * Generate a 128 bit UUID
6409 */
6410 get_random_bytes(mddev->uuid, 16);
6411
6412 mddev->new_level = mddev->level;
6413 mddev->new_chunk_sectors = mddev->chunk_sectors;
6414 mddev->new_layout = mddev->layout;
6415 mddev->delta_disks = 0;
6416 mddev->reshape_backwards = 0;
6417
6418 return 0;
6419 }
6420
6421 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6422 {
6423 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6424
6425 if (mddev->external_size)
6426 return;
6427
6428 mddev->array_sectors = array_sectors;
6429 }
6430 EXPORT_SYMBOL(md_set_array_sectors);
6431
6432 static int update_size(struct mddev *mddev, sector_t num_sectors)
6433 {
6434 struct md_rdev *rdev;
6435 int rv;
6436 int fit = (num_sectors == 0);
6437
6438 if (mddev->pers->resize == NULL)
6439 return -EINVAL;
6440 /* The "num_sectors" is the number of sectors of each device that
6441 * is used. This can only make sense for arrays with redundancy.
6442 * linear and raid0 always use whatever space is available. We can only
6443 * consider changing this number if no resync or reconstruction is
6444 * happening, and if the new size is acceptable. It must fit before the
6445 * sb_start or, if that is <data_offset, it must fit before the size
6446 * of each device. If num_sectors is zero, we find the largest size
6447 * that fits.
6448 */
6449 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6450 mddev->sync_thread)
6451 return -EBUSY;
6452 if (mddev->ro)
6453 return -EROFS;
6454
6455 rdev_for_each(rdev, mddev) {
6456 sector_t avail = rdev->sectors;
6457
6458 if (fit && (num_sectors == 0 || num_sectors > avail))
6459 num_sectors = avail;
6460 if (avail < num_sectors)
6461 return -ENOSPC;
6462 }
6463 rv = mddev->pers->resize(mddev, num_sectors);
6464 if (!rv)
6465 revalidate_disk(mddev->gendisk);
6466 return rv;
6467 }
6468
6469 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6470 {
6471 int rv;
6472 struct md_rdev *rdev;
6473 /* change the number of raid disks */
6474 if (mddev->pers->check_reshape == NULL)
6475 return -EINVAL;
6476 if (mddev->ro)
6477 return -EROFS;
6478 if (raid_disks <= 0 ||
6479 (mddev->max_disks && raid_disks >= mddev->max_disks))
6480 return -EINVAL;
6481 if (mddev->sync_thread ||
6482 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6483 mddev->reshape_position != MaxSector)
6484 return -EBUSY;
6485
6486 rdev_for_each(rdev, mddev) {
6487 if (mddev->raid_disks < raid_disks &&
6488 rdev->data_offset < rdev->new_data_offset)
6489 return -EINVAL;
6490 if (mddev->raid_disks > raid_disks &&
6491 rdev->data_offset > rdev->new_data_offset)
6492 return -EINVAL;
6493 }
6494
6495 mddev->delta_disks = raid_disks - mddev->raid_disks;
6496 if (mddev->delta_disks < 0)
6497 mddev->reshape_backwards = 1;
6498 else if (mddev->delta_disks > 0)
6499 mddev->reshape_backwards = 0;
6500
6501 rv = mddev->pers->check_reshape(mddev);
6502 if (rv < 0) {
6503 mddev->delta_disks = 0;
6504 mddev->reshape_backwards = 0;
6505 }
6506 return rv;
6507 }
6508
6509 /*
6510 * update_array_info is used to change the configuration of an
6511 * on-line array.
6512 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6513 * fields in the info are checked against the array.
6514 * Any differences that cannot be handled will cause an error.
6515 * Normally, only one change can be managed at a time.
6516 */
6517 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6518 {
6519 int rv = 0;
6520 int cnt = 0;
6521 int state = 0;
6522
6523 /* calculate expected state,ignoring low bits */
6524 if (mddev->bitmap && mddev->bitmap_info.offset)
6525 state |= (1 << MD_SB_BITMAP_PRESENT);
6526
6527 if (mddev->major_version != info->major_version ||
6528 mddev->minor_version != info->minor_version ||
6529 /* mddev->patch_version != info->patch_version || */
6530 mddev->ctime != info->ctime ||
6531 mddev->level != info->level ||
6532 /* mddev->layout != info->layout || */
6533 mddev->persistent != !info->not_persistent ||
6534 mddev->chunk_sectors != info->chunk_size >> 9 ||
6535 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6536 ((state^info->state) & 0xfffffe00)
6537 )
6538 return -EINVAL;
6539 /* Check there is only one change */
6540 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6541 cnt++;
6542 if (mddev->raid_disks != info->raid_disks)
6543 cnt++;
6544 if (mddev->layout != info->layout)
6545 cnt++;
6546 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6547 cnt++;
6548 if (cnt == 0)
6549 return 0;
6550 if (cnt > 1)
6551 return -EINVAL;
6552
6553 if (mddev->layout != info->layout) {
6554 /* Change layout
6555 * we don't need to do anything at the md level, the
6556 * personality will take care of it all.
6557 */
6558 if (mddev->pers->check_reshape == NULL)
6559 return -EINVAL;
6560 else {
6561 mddev->new_layout = info->layout;
6562 rv = mddev->pers->check_reshape(mddev);
6563 if (rv)
6564 mddev->new_layout = mddev->layout;
6565 return rv;
6566 }
6567 }
6568 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6569 rv = update_size(mddev, (sector_t)info->size * 2);
6570
6571 if (mddev->raid_disks != info->raid_disks)
6572 rv = update_raid_disks(mddev, info->raid_disks);
6573
6574 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6575 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
6576 rv = -EINVAL;
6577 goto err;
6578 }
6579 if (mddev->recovery || mddev->sync_thread) {
6580 rv = -EBUSY;
6581 goto err;
6582 }
6583 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6584 struct bitmap *bitmap;
6585 /* add the bitmap */
6586 if (mddev->bitmap) {
6587 rv = -EEXIST;
6588 goto err;
6589 }
6590 if (mddev->bitmap_info.default_offset == 0) {
6591 rv = -EINVAL;
6592 goto err;
6593 }
6594 mddev->bitmap_info.offset =
6595 mddev->bitmap_info.default_offset;
6596 mddev->bitmap_info.space =
6597 mddev->bitmap_info.default_space;
6598 mddev->pers->quiesce(mddev, 1);
6599 bitmap = bitmap_create(mddev, -1);
6600 if (!IS_ERR(bitmap)) {
6601 mddev->bitmap = bitmap;
6602 rv = bitmap_load(mddev);
6603 } else
6604 rv = PTR_ERR(bitmap);
6605 if (rv)
6606 bitmap_destroy(mddev);
6607 mddev->pers->quiesce(mddev, 0);
6608 } else {
6609 /* remove the bitmap */
6610 if (!mddev->bitmap) {
6611 rv = -ENOENT;
6612 goto err;
6613 }
6614 if (mddev->bitmap->storage.file) {
6615 rv = -EINVAL;
6616 goto err;
6617 }
6618 if (mddev->bitmap_info.nodes) {
6619 /* hold PW on all the bitmap lock */
6620 if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) {
6621 printk("md: can't change bitmap to none since the"
6622 " array is in use by more than one node\n");
6623 rv = -EPERM;
6624 md_cluster_ops->unlock_all_bitmaps(mddev);
6625 goto err;
6626 }
6627
6628 mddev->bitmap_info.nodes = 0;
6629 md_cluster_ops->leave(mddev);
6630 }
6631 mddev->pers->quiesce(mddev, 1);
6632 bitmap_destroy(mddev);
6633 mddev->pers->quiesce(mddev, 0);
6634 mddev->bitmap_info.offset = 0;
6635 }
6636 }
6637 md_update_sb(mddev, 1);
6638 return rv;
6639 err:
6640 return rv;
6641 }
6642
6643 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6644 {
6645 struct md_rdev *rdev;
6646 int err = 0;
6647
6648 if (mddev->pers == NULL)
6649 return -ENODEV;
6650
6651 rcu_read_lock();
6652 rdev = find_rdev_rcu(mddev, dev);
6653 if (!rdev)
6654 err = -ENODEV;
6655 else {
6656 md_error(mddev, rdev);
6657 if (!test_bit(Faulty, &rdev->flags))
6658 err = -EBUSY;
6659 }
6660 rcu_read_unlock();
6661 return err;
6662 }
6663
6664 /*
6665 * We have a problem here : there is no easy way to give a CHS
6666 * virtual geometry. We currently pretend that we have a 2 heads
6667 * 4 sectors (with a BIG number of cylinders...). This drives
6668 * dosfs just mad... ;-)
6669 */
6670 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6671 {
6672 struct mddev *mddev = bdev->bd_disk->private_data;
6673
6674 geo->heads = 2;
6675 geo->sectors = 4;
6676 geo->cylinders = mddev->array_sectors / 8;
6677 return 0;
6678 }
6679
6680 static inline bool md_ioctl_valid(unsigned int cmd)
6681 {
6682 switch (cmd) {
6683 case ADD_NEW_DISK:
6684 case BLKROSET:
6685 case GET_ARRAY_INFO:
6686 case GET_BITMAP_FILE:
6687 case GET_DISK_INFO:
6688 case HOT_ADD_DISK:
6689 case HOT_REMOVE_DISK:
6690 case RAID_AUTORUN:
6691 case RAID_VERSION:
6692 case RESTART_ARRAY_RW:
6693 case RUN_ARRAY:
6694 case SET_ARRAY_INFO:
6695 case SET_BITMAP_FILE:
6696 case SET_DISK_FAULTY:
6697 case STOP_ARRAY:
6698 case STOP_ARRAY_RO:
6699 case CLUSTERED_DISK_NACK:
6700 return true;
6701 default:
6702 return false;
6703 }
6704 }
6705
6706 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6707 unsigned int cmd, unsigned long arg)
6708 {
6709 int err = 0;
6710 void __user *argp = (void __user *)arg;
6711 struct mddev *mddev = NULL;
6712 int ro;
6713
6714 if (!md_ioctl_valid(cmd))
6715 return -ENOTTY;
6716
6717 switch (cmd) {
6718 case RAID_VERSION:
6719 case GET_ARRAY_INFO:
6720 case GET_DISK_INFO:
6721 break;
6722 default:
6723 if (!capable(CAP_SYS_ADMIN))
6724 return -EACCES;
6725 }
6726
6727 /*
6728 * Commands dealing with the RAID driver but not any
6729 * particular array:
6730 */
6731 switch (cmd) {
6732 case RAID_VERSION:
6733 err = get_version(argp);
6734 goto out;
6735
6736 #ifndef MODULE
6737 case RAID_AUTORUN:
6738 err = 0;
6739 autostart_arrays(arg);
6740 goto out;
6741 #endif
6742 default:;
6743 }
6744
6745 /*
6746 * Commands creating/starting a new array:
6747 */
6748
6749 mddev = bdev->bd_disk->private_data;
6750
6751 if (!mddev) {
6752 BUG();
6753 goto out;
6754 }
6755
6756 /* Some actions do not requires the mutex */
6757 switch (cmd) {
6758 case GET_ARRAY_INFO:
6759 if (!mddev->raid_disks && !mddev->external)
6760 err = -ENODEV;
6761 else
6762 err = get_array_info(mddev, argp);
6763 goto out;
6764
6765 case GET_DISK_INFO:
6766 if (!mddev->raid_disks && !mddev->external)
6767 err = -ENODEV;
6768 else
6769 err = get_disk_info(mddev, argp);
6770 goto out;
6771
6772 case SET_DISK_FAULTY:
6773 err = set_disk_faulty(mddev, new_decode_dev(arg));
6774 goto out;
6775
6776 case GET_BITMAP_FILE:
6777 err = get_bitmap_file(mddev, argp);
6778 goto out;
6779
6780 }
6781
6782 if (cmd == ADD_NEW_DISK)
6783 /* need to ensure md_delayed_delete() has completed */
6784 flush_workqueue(md_misc_wq);
6785
6786 if (cmd == HOT_REMOVE_DISK)
6787 /* need to ensure recovery thread has run */
6788 wait_event_interruptible_timeout(mddev->sb_wait,
6789 !test_bit(MD_RECOVERY_NEEDED,
6790 &mddev->flags),
6791 msecs_to_jiffies(5000));
6792 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6793 /* Need to flush page cache, and ensure no-one else opens
6794 * and writes
6795 */
6796 mutex_lock(&mddev->open_mutex);
6797 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
6798 mutex_unlock(&mddev->open_mutex);
6799 err = -EBUSY;
6800 goto out;
6801 }
6802 set_bit(MD_STILL_CLOSED, &mddev->flags);
6803 mutex_unlock(&mddev->open_mutex);
6804 sync_blockdev(bdev);
6805 }
6806 err = mddev_lock(mddev);
6807 if (err) {
6808 printk(KERN_INFO
6809 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6810 err, cmd);
6811 goto out;
6812 }
6813
6814 if (cmd == SET_ARRAY_INFO) {
6815 mdu_array_info_t info;
6816 if (!arg)
6817 memset(&info, 0, sizeof(info));
6818 else if (copy_from_user(&info, argp, sizeof(info))) {
6819 err = -EFAULT;
6820 goto unlock;
6821 }
6822 if (mddev->pers) {
6823 err = update_array_info(mddev, &info);
6824 if (err) {
6825 printk(KERN_WARNING "md: couldn't update"
6826 " array info. %d\n", err);
6827 goto unlock;
6828 }
6829 goto unlock;
6830 }
6831 if (!list_empty(&mddev->disks)) {
6832 printk(KERN_WARNING
6833 "md: array %s already has disks!\n",
6834 mdname(mddev));
6835 err = -EBUSY;
6836 goto unlock;
6837 }
6838 if (mddev->raid_disks) {
6839 printk(KERN_WARNING
6840 "md: array %s already initialised!\n",
6841 mdname(mddev));
6842 err = -EBUSY;
6843 goto unlock;
6844 }
6845 err = set_array_info(mddev, &info);
6846 if (err) {
6847 printk(KERN_WARNING "md: couldn't set"
6848 " array info. %d\n", err);
6849 goto unlock;
6850 }
6851 goto unlock;
6852 }
6853
6854 /*
6855 * Commands querying/configuring an existing array:
6856 */
6857 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6858 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6859 if ((!mddev->raid_disks && !mddev->external)
6860 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6861 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6862 && cmd != GET_BITMAP_FILE) {
6863 err = -ENODEV;
6864 goto unlock;
6865 }
6866
6867 /*
6868 * Commands even a read-only array can execute:
6869 */
6870 switch (cmd) {
6871 case RESTART_ARRAY_RW:
6872 err = restart_array(mddev);
6873 goto unlock;
6874
6875 case STOP_ARRAY:
6876 err = do_md_stop(mddev, 0, bdev);
6877 goto unlock;
6878
6879 case STOP_ARRAY_RO:
6880 err = md_set_readonly(mddev, bdev);
6881 goto unlock;
6882
6883 case HOT_REMOVE_DISK:
6884 err = hot_remove_disk(mddev, new_decode_dev(arg));
6885 goto unlock;
6886
6887 case ADD_NEW_DISK:
6888 /* We can support ADD_NEW_DISK on read-only arrays
6889 * only if we are re-adding a preexisting device.
6890 * So require mddev->pers and MD_DISK_SYNC.
6891 */
6892 if (mddev->pers) {
6893 mdu_disk_info_t info;
6894 if (copy_from_user(&info, argp, sizeof(info)))
6895 err = -EFAULT;
6896 else if (!(info.state & (1<<MD_DISK_SYNC)))
6897 /* Need to clear read-only for this */
6898 break;
6899 else
6900 err = add_new_disk(mddev, &info);
6901 goto unlock;
6902 }
6903 break;
6904
6905 case BLKROSET:
6906 if (get_user(ro, (int __user *)(arg))) {
6907 err = -EFAULT;
6908 goto unlock;
6909 }
6910 err = -EINVAL;
6911
6912 /* if the bdev is going readonly the value of mddev->ro
6913 * does not matter, no writes are coming
6914 */
6915 if (ro)
6916 goto unlock;
6917
6918 /* are we are already prepared for writes? */
6919 if (mddev->ro != 1)
6920 goto unlock;
6921
6922 /* transitioning to readauto need only happen for
6923 * arrays that call md_write_start
6924 */
6925 if (mddev->pers) {
6926 err = restart_array(mddev);
6927 if (err == 0) {
6928 mddev->ro = 2;
6929 set_disk_ro(mddev->gendisk, 0);
6930 }
6931 }
6932 goto unlock;
6933 }
6934
6935 /*
6936 * The remaining ioctls are changing the state of the
6937 * superblock, so we do not allow them on read-only arrays.
6938 */
6939 if (mddev->ro && mddev->pers) {
6940 if (mddev->ro == 2) {
6941 mddev->ro = 0;
6942 sysfs_notify_dirent_safe(mddev->sysfs_state);
6943 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6944 /* mddev_unlock will wake thread */
6945 /* If a device failed while we were read-only, we
6946 * need to make sure the metadata is updated now.
6947 */
6948 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6949 mddev_unlock(mddev);
6950 wait_event(mddev->sb_wait,
6951 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6952 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6953 mddev_lock_nointr(mddev);
6954 }
6955 } else {
6956 err = -EROFS;
6957 goto unlock;
6958 }
6959 }
6960
6961 switch (cmd) {
6962 case ADD_NEW_DISK:
6963 {
6964 mdu_disk_info_t info;
6965 if (copy_from_user(&info, argp, sizeof(info)))
6966 err = -EFAULT;
6967 else
6968 err = add_new_disk(mddev, &info);
6969 goto unlock;
6970 }
6971
6972 case CLUSTERED_DISK_NACK:
6973 if (mddev_is_clustered(mddev))
6974 md_cluster_ops->new_disk_ack(mddev, false);
6975 else
6976 err = -EINVAL;
6977 goto unlock;
6978
6979 case HOT_ADD_DISK:
6980 err = hot_add_disk(mddev, new_decode_dev(arg));
6981 goto unlock;
6982
6983 case RUN_ARRAY:
6984 err = do_md_run(mddev);
6985 goto unlock;
6986
6987 case SET_BITMAP_FILE:
6988 err = set_bitmap_file(mddev, (int)arg);
6989 goto unlock;
6990
6991 default:
6992 err = -EINVAL;
6993 goto unlock;
6994 }
6995
6996 unlock:
6997 if (mddev->hold_active == UNTIL_IOCTL &&
6998 err != -EINVAL)
6999 mddev->hold_active = 0;
7000 mddev_unlock(mddev);
7001 out:
7002 return err;
7003 }
7004 #ifdef CONFIG_COMPAT
7005 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
7006 unsigned int cmd, unsigned long arg)
7007 {
7008 switch (cmd) {
7009 case HOT_REMOVE_DISK:
7010 case HOT_ADD_DISK:
7011 case SET_DISK_FAULTY:
7012 case SET_BITMAP_FILE:
7013 /* These take in integer arg, do not convert */
7014 break;
7015 default:
7016 arg = (unsigned long)compat_ptr(arg);
7017 break;
7018 }
7019
7020 return md_ioctl(bdev, mode, cmd, arg);
7021 }
7022 #endif /* CONFIG_COMPAT */
7023
7024 static int md_open(struct block_device *bdev, fmode_t mode)
7025 {
7026 /*
7027 * Succeed if we can lock the mddev, which confirms that
7028 * it isn't being stopped right now.
7029 */
7030 struct mddev *mddev = mddev_find(bdev->bd_dev);
7031 int err;
7032
7033 if (!mddev)
7034 return -ENODEV;
7035
7036 if (mddev->gendisk != bdev->bd_disk) {
7037 /* we are racing with mddev_put which is discarding this
7038 * bd_disk.
7039 */
7040 mddev_put(mddev);
7041 /* Wait until bdev->bd_disk is definitely gone */
7042 flush_workqueue(md_misc_wq);
7043 /* Then retry the open from the top */
7044 return -ERESTARTSYS;
7045 }
7046 BUG_ON(mddev != bdev->bd_disk->private_data);
7047
7048 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
7049 goto out;
7050
7051 err = 0;
7052 atomic_inc(&mddev->openers);
7053 clear_bit(MD_STILL_CLOSED, &mddev->flags);
7054 mutex_unlock(&mddev->open_mutex);
7055
7056 check_disk_change(bdev);
7057 out:
7058 return err;
7059 }
7060
7061 static void md_release(struct gendisk *disk, fmode_t mode)
7062 {
7063 struct mddev *mddev = disk->private_data;
7064
7065 BUG_ON(!mddev);
7066 atomic_dec(&mddev->openers);
7067 mddev_put(mddev);
7068 }
7069
7070 static int md_media_changed(struct gendisk *disk)
7071 {
7072 struct mddev *mddev = disk->private_data;
7073
7074 return mddev->changed;
7075 }
7076
7077 static int md_revalidate(struct gendisk *disk)
7078 {
7079 struct mddev *mddev = disk->private_data;
7080
7081 mddev->changed = 0;
7082 return 0;
7083 }
7084 static const struct block_device_operations md_fops =
7085 {
7086 .owner = THIS_MODULE,
7087 .open = md_open,
7088 .release = md_release,
7089 .ioctl = md_ioctl,
7090 #ifdef CONFIG_COMPAT
7091 .compat_ioctl = md_compat_ioctl,
7092 #endif
7093 .getgeo = md_getgeo,
7094 .media_changed = md_media_changed,
7095 .revalidate_disk= md_revalidate,
7096 };
7097
7098 static int md_thread(void *arg)
7099 {
7100 struct md_thread *thread = arg;
7101
7102 /*
7103 * md_thread is a 'system-thread', it's priority should be very
7104 * high. We avoid resource deadlocks individually in each
7105 * raid personality. (RAID5 does preallocation) We also use RR and
7106 * the very same RT priority as kswapd, thus we will never get
7107 * into a priority inversion deadlock.
7108 *
7109 * we definitely have to have equal or higher priority than
7110 * bdflush, otherwise bdflush will deadlock if there are too
7111 * many dirty RAID5 blocks.
7112 */
7113
7114 allow_signal(SIGKILL);
7115 while (!kthread_should_stop()) {
7116
7117 /* We need to wait INTERRUPTIBLE so that
7118 * we don't add to the load-average.
7119 * That means we need to be sure no signals are
7120 * pending
7121 */
7122 if (signal_pending(current))
7123 flush_signals(current);
7124
7125 wait_event_interruptible_timeout
7126 (thread->wqueue,
7127 test_bit(THREAD_WAKEUP, &thread->flags)
7128 || kthread_should_stop(),
7129 thread->timeout);
7130
7131 clear_bit(THREAD_WAKEUP, &thread->flags);
7132 if (!kthread_should_stop())
7133 thread->run(thread);
7134 }
7135
7136 return 0;
7137 }
7138
7139 void md_wakeup_thread(struct md_thread *thread)
7140 {
7141 if (thread) {
7142 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
7143 set_bit(THREAD_WAKEUP, &thread->flags);
7144 wake_up(&thread->wqueue);
7145 }
7146 }
7147 EXPORT_SYMBOL(md_wakeup_thread);
7148
7149 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7150 struct mddev *mddev, const char *name)
7151 {
7152 struct md_thread *thread;
7153
7154 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7155 if (!thread)
7156 return NULL;
7157
7158 init_waitqueue_head(&thread->wqueue);
7159
7160 thread->run = run;
7161 thread->mddev = mddev;
7162 thread->timeout = MAX_SCHEDULE_TIMEOUT;
7163 thread->tsk = kthread_run(md_thread, thread,
7164 "%s_%s",
7165 mdname(thread->mddev),
7166 name);
7167 if (IS_ERR(thread->tsk)) {
7168 kfree(thread);
7169 return NULL;
7170 }
7171 return thread;
7172 }
7173 EXPORT_SYMBOL(md_register_thread);
7174
7175 void md_unregister_thread(struct md_thread **threadp)
7176 {
7177 struct md_thread *thread = *threadp;
7178 if (!thread)
7179 return;
7180 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7181 /* Locking ensures that mddev_unlock does not wake_up a
7182 * non-existent thread
7183 */
7184 spin_lock(&pers_lock);
7185 *threadp = NULL;
7186 spin_unlock(&pers_lock);
7187
7188 kthread_stop(thread->tsk);
7189 kfree(thread);
7190 }
7191 EXPORT_SYMBOL(md_unregister_thread);
7192
7193 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7194 {
7195 if (!rdev || test_bit(Faulty, &rdev->flags))
7196 return;
7197
7198 if (!mddev->pers || !mddev->pers->error_handler)
7199 return;
7200 mddev->pers->error_handler(mddev,rdev);
7201 if (mddev->degraded)
7202 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7203 sysfs_notify_dirent_safe(rdev->sysfs_state);
7204 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7205 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7206 md_wakeup_thread(mddev->thread);
7207 if (mddev->event_work.func)
7208 queue_work(md_misc_wq, &mddev->event_work);
7209 md_new_event(mddev);
7210 }
7211 EXPORT_SYMBOL(md_error);
7212
7213 /* seq_file implementation /proc/mdstat */
7214
7215 static void status_unused(struct seq_file *seq)
7216 {
7217 int i = 0;
7218 struct md_rdev *rdev;
7219
7220 seq_printf(seq, "unused devices: ");
7221
7222 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
7223 char b[BDEVNAME_SIZE];
7224 i++;
7225 seq_printf(seq, "%s ",
7226 bdevname(rdev->bdev,b));
7227 }
7228 if (!i)
7229 seq_printf(seq, "<none>");
7230
7231 seq_printf(seq, "\n");
7232 }
7233
7234 static int status_resync(struct seq_file *seq, struct mddev *mddev)
7235 {
7236 sector_t max_sectors, resync, res;
7237 unsigned long dt, db;
7238 sector_t rt;
7239 int scale;
7240 unsigned int per_milli;
7241
7242 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7243 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7244 max_sectors = mddev->resync_max_sectors;
7245 else
7246 max_sectors = mddev->dev_sectors;
7247
7248 resync = mddev->curr_resync;
7249 if (resync <= 3) {
7250 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7251 /* Still cleaning up */
7252 resync = max_sectors;
7253 } else
7254 resync -= atomic_read(&mddev->recovery_active);
7255
7256 if (resync == 0) {
7257 if (mddev->recovery_cp < MaxSector) {
7258 seq_printf(seq, "\tresync=PENDING");
7259 return 1;
7260 }
7261 return 0;
7262 }
7263 if (resync < 3) {
7264 seq_printf(seq, "\tresync=DELAYED");
7265 return 1;
7266 }
7267
7268 WARN_ON(max_sectors == 0);
7269 /* Pick 'scale' such that (resync>>scale)*1000 will fit
7270 * in a sector_t, and (max_sectors>>scale) will fit in a
7271 * u32, as those are the requirements for sector_div.
7272 * Thus 'scale' must be at least 10
7273 */
7274 scale = 10;
7275 if (sizeof(sector_t) > sizeof(unsigned long)) {
7276 while ( max_sectors/2 > (1ULL<<(scale+32)))
7277 scale++;
7278 }
7279 res = (resync>>scale)*1000;
7280 sector_div(res, (u32)((max_sectors>>scale)+1));
7281
7282 per_milli = res;
7283 {
7284 int i, x = per_milli/50, y = 20-x;
7285 seq_printf(seq, "[");
7286 for (i = 0; i < x; i++)
7287 seq_printf(seq, "=");
7288 seq_printf(seq, ">");
7289 for (i = 0; i < y; i++)
7290 seq_printf(seq, ".");
7291 seq_printf(seq, "] ");
7292 }
7293 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
7294 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
7295 "reshape" :
7296 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
7297 "check" :
7298 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
7299 "resync" : "recovery"))),
7300 per_milli/10, per_milli % 10,
7301 (unsigned long long) resync/2,
7302 (unsigned long long) max_sectors/2);
7303
7304 /*
7305 * dt: time from mark until now
7306 * db: blocks written from mark until now
7307 * rt: remaining time
7308 *
7309 * rt is a sector_t, so could be 32bit or 64bit.
7310 * So we divide before multiply in case it is 32bit and close
7311 * to the limit.
7312 * We scale the divisor (db) by 32 to avoid losing precision
7313 * near the end of resync when the number of remaining sectors
7314 * is close to 'db'.
7315 * We then divide rt by 32 after multiplying by db to compensate.
7316 * The '+1' avoids division by zero if db is very small.
7317 */
7318 dt = ((jiffies - mddev->resync_mark) / HZ);
7319 if (!dt) dt++;
7320 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
7321 - mddev->resync_mark_cnt;
7322
7323 rt = max_sectors - resync; /* number of remaining sectors */
7324 sector_div(rt, db/32+1);
7325 rt *= dt;
7326 rt >>= 5;
7327
7328 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7329 ((unsigned long)rt % 60)/6);
7330
7331 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
7332 return 1;
7333 }
7334
7335 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7336 {
7337 struct list_head *tmp;
7338 loff_t l = *pos;
7339 struct mddev *mddev;
7340
7341 if (l >= 0x10000)
7342 return NULL;
7343 if (!l--)
7344 /* header */
7345 return (void*)1;
7346
7347 spin_lock(&all_mddevs_lock);
7348 list_for_each(tmp,&all_mddevs)
7349 if (!l--) {
7350 mddev = list_entry(tmp, struct mddev, all_mddevs);
7351 mddev_get(mddev);
7352 spin_unlock(&all_mddevs_lock);
7353 return mddev;
7354 }
7355 spin_unlock(&all_mddevs_lock);
7356 if (!l--)
7357 return (void*)2;/* tail */
7358 return NULL;
7359 }
7360
7361 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7362 {
7363 struct list_head *tmp;
7364 struct mddev *next_mddev, *mddev = v;
7365
7366 ++*pos;
7367 if (v == (void*)2)
7368 return NULL;
7369
7370 spin_lock(&all_mddevs_lock);
7371 if (v == (void*)1)
7372 tmp = all_mddevs.next;
7373 else
7374 tmp = mddev->all_mddevs.next;
7375 if (tmp != &all_mddevs)
7376 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
7377 else {
7378 next_mddev = (void*)2;
7379 *pos = 0x10000;
7380 }
7381 spin_unlock(&all_mddevs_lock);
7382
7383 if (v != (void*)1)
7384 mddev_put(mddev);
7385 return next_mddev;
7386
7387 }
7388
7389 static void md_seq_stop(struct seq_file *seq, void *v)
7390 {
7391 struct mddev *mddev = v;
7392
7393 if (mddev && v != (void*)1 && v != (void*)2)
7394 mddev_put(mddev);
7395 }
7396
7397 static int md_seq_show(struct seq_file *seq, void *v)
7398 {
7399 struct mddev *mddev = v;
7400 sector_t sectors;
7401 struct md_rdev *rdev;
7402
7403 if (v == (void*)1) {
7404 struct md_personality *pers;
7405 seq_printf(seq, "Personalities : ");
7406 spin_lock(&pers_lock);
7407 list_for_each_entry(pers, &pers_list, list)
7408 seq_printf(seq, "[%s] ", pers->name);
7409
7410 spin_unlock(&pers_lock);
7411 seq_printf(seq, "\n");
7412 seq->poll_event = atomic_read(&md_event_count);
7413 return 0;
7414 }
7415 if (v == (void*)2) {
7416 status_unused(seq);
7417 return 0;
7418 }
7419
7420 spin_lock(&mddev->lock);
7421 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7422 seq_printf(seq, "%s : %sactive", mdname(mddev),
7423 mddev->pers ? "" : "in");
7424 if (mddev->pers) {
7425 if (mddev->ro==1)
7426 seq_printf(seq, " (read-only)");
7427 if (mddev->ro==2)
7428 seq_printf(seq, " (auto-read-only)");
7429 seq_printf(seq, " %s", mddev->pers->name);
7430 }
7431
7432 sectors = 0;
7433 rcu_read_lock();
7434 rdev_for_each_rcu(rdev, mddev) {
7435 char b[BDEVNAME_SIZE];
7436 seq_printf(seq, " %s[%d]",
7437 bdevname(rdev->bdev,b), rdev->desc_nr);
7438 if (test_bit(WriteMostly, &rdev->flags))
7439 seq_printf(seq, "(W)");
7440 if (test_bit(Journal, &rdev->flags))
7441 seq_printf(seq, "(J)");
7442 if (test_bit(Faulty, &rdev->flags)) {
7443 seq_printf(seq, "(F)");
7444 continue;
7445 }
7446 if (rdev->raid_disk < 0)
7447 seq_printf(seq, "(S)"); /* spare */
7448 if (test_bit(Replacement, &rdev->flags))
7449 seq_printf(seq, "(R)");
7450 sectors += rdev->sectors;
7451 }
7452 rcu_read_unlock();
7453
7454 if (!list_empty(&mddev->disks)) {
7455 if (mddev->pers)
7456 seq_printf(seq, "\n %llu blocks",
7457 (unsigned long long)
7458 mddev->array_sectors / 2);
7459 else
7460 seq_printf(seq, "\n %llu blocks",
7461 (unsigned long long)sectors / 2);
7462 }
7463 if (mddev->persistent) {
7464 if (mddev->major_version != 0 ||
7465 mddev->minor_version != 90) {
7466 seq_printf(seq," super %d.%d",
7467 mddev->major_version,
7468 mddev->minor_version);
7469 }
7470 } else if (mddev->external)
7471 seq_printf(seq, " super external:%s",
7472 mddev->metadata_type);
7473 else
7474 seq_printf(seq, " super non-persistent");
7475
7476 if (mddev->pers) {
7477 mddev->pers->status(seq, mddev);
7478 seq_printf(seq, "\n ");
7479 if (mddev->pers->sync_request) {
7480 if (status_resync(seq, mddev))
7481 seq_printf(seq, "\n ");
7482 }
7483 } else
7484 seq_printf(seq, "\n ");
7485
7486 bitmap_status(seq, mddev->bitmap);
7487
7488 seq_printf(seq, "\n");
7489 }
7490 spin_unlock(&mddev->lock);
7491
7492 return 0;
7493 }
7494
7495 static const struct seq_operations md_seq_ops = {
7496 .start = md_seq_start,
7497 .next = md_seq_next,
7498 .stop = md_seq_stop,
7499 .show = md_seq_show,
7500 };
7501
7502 static int md_seq_open(struct inode *inode, struct file *file)
7503 {
7504 struct seq_file *seq;
7505 int error;
7506
7507 error = seq_open(file, &md_seq_ops);
7508 if (error)
7509 return error;
7510
7511 seq = file->private_data;
7512 seq->poll_event = atomic_read(&md_event_count);
7513 return error;
7514 }
7515
7516 static int md_unloading;
7517 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7518 {
7519 struct seq_file *seq = filp->private_data;
7520 int mask;
7521
7522 if (md_unloading)
7523 return POLLIN|POLLRDNORM|POLLERR|POLLPRI;
7524 poll_wait(filp, &md_event_waiters, wait);
7525
7526 /* always allow read */
7527 mask = POLLIN | POLLRDNORM;
7528
7529 if (seq->poll_event != atomic_read(&md_event_count))
7530 mask |= POLLERR | POLLPRI;
7531 return mask;
7532 }
7533
7534 static const struct file_operations md_seq_fops = {
7535 .owner = THIS_MODULE,
7536 .open = md_seq_open,
7537 .read = seq_read,
7538 .llseek = seq_lseek,
7539 .release = seq_release_private,
7540 .poll = mdstat_poll,
7541 };
7542
7543 int register_md_personality(struct md_personality *p)
7544 {
7545 printk(KERN_INFO "md: %s personality registered for level %d\n",
7546 p->name, p->level);
7547 spin_lock(&pers_lock);
7548 list_add_tail(&p->list, &pers_list);
7549 spin_unlock(&pers_lock);
7550 return 0;
7551 }
7552 EXPORT_SYMBOL(register_md_personality);
7553
7554 int unregister_md_personality(struct md_personality *p)
7555 {
7556 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
7557 spin_lock(&pers_lock);
7558 list_del_init(&p->list);
7559 spin_unlock(&pers_lock);
7560 return 0;
7561 }
7562 EXPORT_SYMBOL(unregister_md_personality);
7563
7564 int register_md_cluster_operations(struct md_cluster_operations *ops,
7565 struct module *module)
7566 {
7567 int ret = 0;
7568 spin_lock(&pers_lock);
7569 if (md_cluster_ops != NULL)
7570 ret = -EALREADY;
7571 else {
7572 md_cluster_ops = ops;
7573 md_cluster_mod = module;
7574 }
7575 spin_unlock(&pers_lock);
7576 return ret;
7577 }
7578 EXPORT_SYMBOL(register_md_cluster_operations);
7579
7580 int unregister_md_cluster_operations(void)
7581 {
7582 spin_lock(&pers_lock);
7583 md_cluster_ops = NULL;
7584 spin_unlock(&pers_lock);
7585 return 0;
7586 }
7587 EXPORT_SYMBOL(unregister_md_cluster_operations);
7588
7589 int md_setup_cluster(struct mddev *mddev, int nodes)
7590 {
7591 int err;
7592
7593 err = request_module("md-cluster");
7594 if (err) {
7595 pr_err("md-cluster module not found.\n");
7596 return -ENOENT;
7597 }
7598
7599 spin_lock(&pers_lock);
7600 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
7601 spin_unlock(&pers_lock);
7602 return -ENOENT;
7603 }
7604 spin_unlock(&pers_lock);
7605
7606 return md_cluster_ops->join(mddev, nodes);
7607 }
7608
7609 void md_cluster_stop(struct mddev *mddev)
7610 {
7611 if (!md_cluster_ops)
7612 return;
7613 md_cluster_ops->leave(mddev);
7614 module_put(md_cluster_mod);
7615 }
7616
7617 static int is_mddev_idle(struct mddev *mddev, int init)
7618 {
7619 struct md_rdev *rdev;
7620 int idle;
7621 int curr_events;
7622
7623 idle = 1;
7624 rcu_read_lock();
7625 rdev_for_each_rcu(rdev, mddev) {
7626 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7627 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7628 (int)part_stat_read(&disk->part0, sectors[1]) -
7629 atomic_read(&disk->sync_io);
7630 /* sync IO will cause sync_io to increase before the disk_stats
7631 * as sync_io is counted when a request starts, and
7632 * disk_stats is counted when it completes.
7633 * So resync activity will cause curr_events to be smaller than
7634 * when there was no such activity.
7635 * non-sync IO will cause disk_stat to increase without
7636 * increasing sync_io so curr_events will (eventually)
7637 * be larger than it was before. Once it becomes
7638 * substantially larger, the test below will cause
7639 * the array to appear non-idle, and resync will slow
7640 * down.
7641 * If there is a lot of outstanding resync activity when
7642 * we set last_event to curr_events, then all that activity
7643 * completing might cause the array to appear non-idle
7644 * and resync will be slowed down even though there might
7645 * not have been non-resync activity. This will only
7646 * happen once though. 'last_events' will soon reflect
7647 * the state where there is little or no outstanding
7648 * resync requests, and further resync activity will
7649 * always make curr_events less than last_events.
7650 *
7651 */
7652 if (init || curr_events - rdev->last_events > 64) {
7653 rdev->last_events = curr_events;
7654 idle = 0;
7655 }
7656 }
7657 rcu_read_unlock();
7658 return idle;
7659 }
7660
7661 void md_done_sync(struct mddev *mddev, int blocks, int ok)
7662 {
7663 /* another "blocks" (512byte) blocks have been synced */
7664 atomic_sub(blocks, &mddev->recovery_active);
7665 wake_up(&mddev->recovery_wait);
7666 if (!ok) {
7667 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7668 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
7669 md_wakeup_thread(mddev->thread);
7670 // stop recovery, signal do_sync ....
7671 }
7672 }
7673 EXPORT_SYMBOL(md_done_sync);
7674
7675 /* md_write_start(mddev, bi)
7676 * If we need to update some array metadata (e.g. 'active' flag
7677 * in superblock) before writing, schedule a superblock update
7678 * and wait for it to complete.
7679 */
7680 void md_write_start(struct mddev *mddev, struct bio *bi)
7681 {
7682 int did_change = 0;
7683 if (bio_data_dir(bi) != WRITE)
7684 return;
7685
7686 BUG_ON(mddev->ro == 1);
7687 if (mddev->ro == 2) {
7688 /* need to switch to read/write */
7689 mddev->ro = 0;
7690 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7691 md_wakeup_thread(mddev->thread);
7692 md_wakeup_thread(mddev->sync_thread);
7693 did_change = 1;
7694 }
7695 atomic_inc(&mddev->writes_pending);
7696 if (mddev->safemode == 1)
7697 mddev->safemode = 0;
7698 if (mddev->in_sync) {
7699 spin_lock(&mddev->lock);
7700 if (mddev->in_sync) {
7701 mddev->in_sync = 0;
7702 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7703 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7704 md_wakeup_thread(mddev->thread);
7705 did_change = 1;
7706 }
7707 spin_unlock(&mddev->lock);
7708 }
7709 if (did_change)
7710 sysfs_notify_dirent_safe(mddev->sysfs_state);
7711 wait_event(mddev->sb_wait,
7712 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
7713 }
7714 EXPORT_SYMBOL(md_write_start);
7715
7716 void md_write_end(struct mddev *mddev)
7717 {
7718 if (atomic_dec_and_test(&mddev->writes_pending)) {
7719 if (mddev->safemode == 2)
7720 md_wakeup_thread(mddev->thread);
7721 else if (mddev->safemode_delay)
7722 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7723 }
7724 }
7725 EXPORT_SYMBOL(md_write_end);
7726
7727 /* md_allow_write(mddev)
7728 * Calling this ensures that the array is marked 'active' so that writes
7729 * may proceed without blocking. It is important to call this before
7730 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7731 * Must be called with mddev_lock held.
7732 *
7733 * In the ->external case MD_CHANGE_PENDING can not be cleared until mddev->lock
7734 * is dropped, so return -EAGAIN after notifying userspace.
7735 */
7736 int md_allow_write(struct mddev *mddev)
7737 {
7738 if (!mddev->pers)
7739 return 0;
7740 if (mddev->ro)
7741 return 0;
7742 if (!mddev->pers->sync_request)
7743 return 0;
7744
7745 spin_lock(&mddev->lock);
7746 if (mddev->in_sync) {
7747 mddev->in_sync = 0;
7748 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7749 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7750 if (mddev->safemode_delay &&
7751 mddev->safemode == 0)
7752 mddev->safemode = 1;
7753 spin_unlock(&mddev->lock);
7754 md_update_sb(mddev, 0);
7755 sysfs_notify_dirent_safe(mddev->sysfs_state);
7756 } else
7757 spin_unlock(&mddev->lock);
7758
7759 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
7760 return -EAGAIN;
7761 else
7762 return 0;
7763 }
7764 EXPORT_SYMBOL_GPL(md_allow_write);
7765
7766 #define SYNC_MARKS 10
7767 #define SYNC_MARK_STEP (3*HZ)
7768 #define UPDATE_FREQUENCY (5*60*HZ)
7769 void md_do_sync(struct md_thread *thread)
7770 {
7771 struct mddev *mddev = thread->mddev;
7772 struct mddev *mddev2;
7773 unsigned int currspeed = 0,
7774 window;
7775 sector_t max_sectors,j, io_sectors, recovery_done;
7776 unsigned long mark[SYNC_MARKS];
7777 unsigned long update_time;
7778 sector_t mark_cnt[SYNC_MARKS];
7779 int last_mark,m;
7780 struct list_head *tmp;
7781 sector_t last_check;
7782 int skipped = 0;
7783 struct md_rdev *rdev;
7784 char *desc, *action = NULL;
7785 struct blk_plug plug;
7786 bool cluster_resync_finished = false;
7787
7788 /* just incase thread restarts... */
7789 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7790 return;
7791 if (mddev->ro) {/* never try to sync a read-only array */
7792 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7793 return;
7794 }
7795
7796 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7797 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
7798 desc = "data-check";
7799 action = "check";
7800 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7801 desc = "requested-resync";
7802 action = "repair";
7803 } else
7804 desc = "resync";
7805 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7806 desc = "reshape";
7807 else
7808 desc = "recovery";
7809
7810 mddev->last_sync_action = action ?: desc;
7811
7812 /* we overload curr_resync somewhat here.
7813 * 0 == not engaged in resync at all
7814 * 2 == checking that there is no conflict with another sync
7815 * 1 == like 2, but have yielded to allow conflicting resync to
7816 * commense
7817 * other == active in resync - this many blocks
7818 *
7819 * Before starting a resync we must have set curr_resync to
7820 * 2, and then checked that every "conflicting" array has curr_resync
7821 * less than ours. When we find one that is the same or higher
7822 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7823 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7824 * This will mean we have to start checking from the beginning again.
7825 *
7826 */
7827
7828 do {
7829 mddev->curr_resync = 2;
7830
7831 try_again:
7832 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7833 goto skip;
7834 for_each_mddev(mddev2, tmp) {
7835 if (mddev2 == mddev)
7836 continue;
7837 if (!mddev->parallel_resync
7838 && mddev2->curr_resync
7839 && match_mddev_units(mddev, mddev2)) {
7840 DEFINE_WAIT(wq);
7841 if (mddev < mddev2 && mddev->curr_resync == 2) {
7842 /* arbitrarily yield */
7843 mddev->curr_resync = 1;
7844 wake_up(&resync_wait);
7845 }
7846 if (mddev > mddev2 && mddev->curr_resync == 1)
7847 /* no need to wait here, we can wait the next
7848 * time 'round when curr_resync == 2
7849 */
7850 continue;
7851 /* We need to wait 'interruptible' so as not to
7852 * contribute to the load average, and not to
7853 * be caught by 'softlockup'
7854 */
7855 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7856 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7857 mddev2->curr_resync >= mddev->curr_resync) {
7858 printk(KERN_INFO "md: delaying %s of %s"
7859 " until %s has finished (they"
7860 " share one or more physical units)\n",
7861 desc, mdname(mddev), mdname(mddev2));
7862 mddev_put(mddev2);
7863 if (signal_pending(current))
7864 flush_signals(current);
7865 schedule();
7866 finish_wait(&resync_wait, &wq);
7867 goto try_again;
7868 }
7869 finish_wait(&resync_wait, &wq);
7870 }
7871 }
7872 } while (mddev->curr_resync < 2);
7873
7874 j = 0;
7875 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7876 /* resync follows the size requested by the personality,
7877 * which defaults to physical size, but can be virtual size
7878 */
7879 max_sectors = mddev->resync_max_sectors;
7880 atomic64_set(&mddev->resync_mismatches, 0);
7881 /* we don't use the checkpoint if there's a bitmap */
7882 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7883 j = mddev->resync_min;
7884 else if (!mddev->bitmap)
7885 j = mddev->recovery_cp;
7886
7887 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7888 max_sectors = mddev->resync_max_sectors;
7889 else {
7890 /* recovery follows the physical size of devices */
7891 max_sectors = mddev->dev_sectors;
7892 j = MaxSector;
7893 rcu_read_lock();
7894 rdev_for_each_rcu(rdev, mddev)
7895 if (rdev->raid_disk >= 0 &&
7896 !test_bit(Journal, &rdev->flags) &&
7897 !test_bit(Faulty, &rdev->flags) &&
7898 !test_bit(In_sync, &rdev->flags) &&
7899 rdev->recovery_offset < j)
7900 j = rdev->recovery_offset;
7901 rcu_read_unlock();
7902
7903 /* If there is a bitmap, we need to make sure all
7904 * writes that started before we added a spare
7905 * complete before we start doing a recovery.
7906 * Otherwise the write might complete and (via
7907 * bitmap_endwrite) set a bit in the bitmap after the
7908 * recovery has checked that bit and skipped that
7909 * region.
7910 */
7911 if (mddev->bitmap) {
7912 mddev->pers->quiesce(mddev, 1);
7913 mddev->pers->quiesce(mddev, 0);
7914 }
7915 }
7916
7917 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7918 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7919 " %d KB/sec/disk.\n", speed_min(mddev));
7920 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
7921 "(but not more than %d KB/sec) for %s.\n",
7922 speed_max(mddev), desc);
7923
7924 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
7925
7926 io_sectors = 0;
7927 for (m = 0; m < SYNC_MARKS; m++) {
7928 mark[m] = jiffies;
7929 mark_cnt[m] = io_sectors;
7930 }
7931 last_mark = 0;
7932 mddev->resync_mark = mark[last_mark];
7933 mddev->resync_mark_cnt = mark_cnt[last_mark];
7934
7935 /*
7936 * Tune reconstruction:
7937 */
7938 window = 32*(PAGE_SIZE/512);
7939 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7940 window/2, (unsigned long long)max_sectors/2);
7941
7942 atomic_set(&mddev->recovery_active, 0);
7943 last_check = 0;
7944
7945 if (j>2) {
7946 printk(KERN_INFO
7947 "md: resuming %s of %s from checkpoint.\n",
7948 desc, mdname(mddev));
7949 mddev->curr_resync = j;
7950 } else
7951 mddev->curr_resync = 3; /* no longer delayed */
7952 mddev->curr_resync_completed = j;
7953 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7954 md_new_event(mddev);
7955 update_time = jiffies;
7956
7957 blk_start_plug(&plug);
7958 while (j < max_sectors) {
7959 sector_t sectors;
7960
7961 skipped = 0;
7962
7963 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7964 ((mddev->curr_resync > mddev->curr_resync_completed &&
7965 (mddev->curr_resync - mddev->curr_resync_completed)
7966 > (max_sectors >> 4)) ||
7967 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7968 (j - mddev->curr_resync_completed)*2
7969 >= mddev->resync_max - mddev->curr_resync_completed ||
7970 mddev->curr_resync_completed > mddev->resync_max
7971 )) {
7972 /* time to update curr_resync_completed */
7973 wait_event(mddev->recovery_wait,
7974 atomic_read(&mddev->recovery_active) == 0);
7975 mddev->curr_resync_completed = j;
7976 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7977 j > mddev->recovery_cp)
7978 mddev->recovery_cp = j;
7979 update_time = jiffies;
7980 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7981 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7982 }
7983
7984 while (j >= mddev->resync_max &&
7985 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7986 /* As this condition is controlled by user-space,
7987 * we can block indefinitely, so use '_interruptible'
7988 * to avoid triggering warnings.
7989 */
7990 flush_signals(current); /* just in case */
7991 wait_event_interruptible(mddev->recovery_wait,
7992 mddev->resync_max > j
7993 || test_bit(MD_RECOVERY_INTR,
7994 &mddev->recovery));
7995 }
7996
7997 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7998 break;
7999
8000 sectors = mddev->pers->sync_request(mddev, j, &skipped);
8001 if (sectors == 0) {
8002 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8003 break;
8004 }
8005
8006 if (!skipped) { /* actual IO requested */
8007 io_sectors += sectors;
8008 atomic_add(sectors, &mddev->recovery_active);
8009 }
8010
8011 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8012 break;
8013
8014 j += sectors;
8015 if (j > max_sectors)
8016 /* when skipping, extra large numbers can be returned. */
8017 j = max_sectors;
8018 if (j > 2)
8019 mddev->curr_resync = j;
8020 mddev->curr_mark_cnt = io_sectors;
8021 if (last_check == 0)
8022 /* this is the earliest that rebuild will be
8023 * visible in /proc/mdstat
8024 */
8025 md_new_event(mddev);
8026
8027 if (last_check + window > io_sectors || j == max_sectors)
8028 continue;
8029
8030 last_check = io_sectors;
8031 repeat:
8032 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8033 /* step marks */
8034 int next = (last_mark+1) % SYNC_MARKS;
8035
8036 mddev->resync_mark = mark[next];
8037 mddev->resync_mark_cnt = mark_cnt[next];
8038 mark[next] = jiffies;
8039 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
8040 last_mark = next;
8041 }
8042
8043 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8044 break;
8045
8046 /*
8047 * this loop exits only if either when we are slower than
8048 * the 'hard' speed limit, or the system was IO-idle for
8049 * a jiffy.
8050 * the system might be non-idle CPU-wise, but we only care
8051 * about not overloading the IO subsystem. (things like an
8052 * e2fsck being done on the RAID array should execute fast)
8053 */
8054 cond_resched();
8055
8056 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8057 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
8058 /((jiffies-mddev->resync_mark)/HZ +1) +1;
8059
8060 if (currspeed > speed_min(mddev)) {
8061 if (currspeed > speed_max(mddev)) {
8062 msleep(500);
8063 goto repeat;
8064 }
8065 if (!is_mddev_idle(mddev, 0)) {
8066 /*
8067 * Give other IO more of a chance.
8068 * The faster the devices, the less we wait.
8069 */
8070 wait_event(mddev->recovery_wait,
8071 !atomic_read(&mddev->recovery_active));
8072 }
8073 }
8074 }
8075 printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc,
8076 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
8077 ? "interrupted" : "done");
8078 /*
8079 * this also signals 'finished resyncing' to md_stop
8080 */
8081 blk_finish_plug(&plug);
8082 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
8083
8084 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8085 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8086 mddev->curr_resync > 2) {
8087 mddev->curr_resync_completed = mddev->curr_resync;
8088 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8089 }
8090 /* tell personality and other nodes that we are finished */
8091 if (mddev_is_clustered(mddev)) {
8092 md_cluster_ops->resync_finish(mddev);
8093 cluster_resync_finished = true;
8094 }
8095 mddev->pers->sync_request(mddev, max_sectors, &skipped);
8096
8097 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
8098 mddev->curr_resync > 2) {
8099 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8100 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8101 if (mddev->curr_resync >= mddev->recovery_cp) {
8102 printk(KERN_INFO
8103 "md: checkpointing %s of %s.\n",
8104 desc, mdname(mddev));
8105 if (test_bit(MD_RECOVERY_ERROR,
8106 &mddev->recovery))
8107 mddev->recovery_cp =
8108 mddev->curr_resync_completed;
8109 else
8110 mddev->recovery_cp =
8111 mddev->curr_resync;
8112 }
8113 } else
8114 mddev->recovery_cp = MaxSector;
8115 } else {
8116 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8117 mddev->curr_resync = MaxSector;
8118 rcu_read_lock();
8119 rdev_for_each_rcu(rdev, mddev)
8120 if (rdev->raid_disk >= 0 &&
8121 mddev->delta_disks >= 0 &&
8122 !test_bit(Journal, &rdev->flags) &&
8123 !test_bit(Faulty, &rdev->flags) &&
8124 !test_bit(In_sync, &rdev->flags) &&
8125 rdev->recovery_offset < mddev->curr_resync)
8126 rdev->recovery_offset = mddev->curr_resync;
8127 rcu_read_unlock();
8128 }
8129 }
8130 skip:
8131 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8132
8133 if (mddev_is_clustered(mddev) &&
8134 test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8135 !cluster_resync_finished)
8136 md_cluster_ops->resync_finish(mddev);
8137
8138 spin_lock(&mddev->lock);
8139 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8140 /* We completed so min/max setting can be forgotten if used. */
8141 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8142 mddev->resync_min = 0;
8143 mddev->resync_max = MaxSector;
8144 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8145 mddev->resync_min = mddev->curr_resync_completed;
8146 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
8147 mddev->curr_resync = 0;
8148 spin_unlock(&mddev->lock);
8149
8150 wake_up(&resync_wait);
8151 md_wakeup_thread(mddev->thread);
8152 return;
8153 }
8154 EXPORT_SYMBOL_GPL(md_do_sync);
8155
8156 static int remove_and_add_spares(struct mddev *mddev,
8157 struct md_rdev *this)
8158 {
8159 struct md_rdev *rdev;
8160 int spares = 0;
8161 int removed = 0;
8162
8163 rdev_for_each(rdev, mddev)
8164 if ((this == NULL || rdev == this) &&
8165 rdev->raid_disk >= 0 &&
8166 !test_bit(Blocked, &rdev->flags) &&
8167 (test_bit(Faulty, &rdev->flags) ||
8168 (!test_bit(In_sync, &rdev->flags) &&
8169 !test_bit(Journal, &rdev->flags))) &&
8170 atomic_read(&rdev->nr_pending)==0) {
8171 if (mddev->pers->hot_remove_disk(
8172 mddev, rdev) == 0) {
8173 sysfs_unlink_rdev(mddev, rdev);
8174 rdev->raid_disk = -1;
8175 removed++;
8176 }
8177 }
8178 if (removed && mddev->kobj.sd)
8179 sysfs_notify(&mddev->kobj, NULL, "degraded");
8180
8181 if (this && removed)
8182 goto no_add;
8183
8184 rdev_for_each(rdev, mddev) {
8185 if (this && this != rdev)
8186 continue;
8187 if (test_bit(Candidate, &rdev->flags))
8188 continue;
8189 if (rdev->raid_disk >= 0 &&
8190 !test_bit(In_sync, &rdev->flags) &&
8191 !test_bit(Journal, &rdev->flags) &&
8192 !test_bit(Faulty, &rdev->flags))
8193 spares++;
8194 if (rdev->raid_disk >= 0)
8195 continue;
8196 if (test_bit(Faulty, &rdev->flags))
8197 continue;
8198 if (!test_bit(Journal, &rdev->flags)) {
8199 if (mddev->ro &&
8200 ! (rdev->saved_raid_disk >= 0 &&
8201 !test_bit(Bitmap_sync, &rdev->flags)))
8202 continue;
8203
8204 rdev->recovery_offset = 0;
8205 }
8206 if (mddev->pers->
8207 hot_add_disk(mddev, rdev) == 0) {
8208 if (sysfs_link_rdev(mddev, rdev))
8209 /* failure here is OK */;
8210 if (!test_bit(Journal, &rdev->flags))
8211 spares++;
8212 md_new_event(mddev);
8213 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8214 }
8215 }
8216 no_add:
8217 if (removed)
8218 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8219 return spares;
8220 }
8221
8222 static void md_start_sync(struct work_struct *ws)
8223 {
8224 struct mddev *mddev = container_of(ws, struct mddev, del_work);
8225 int ret = 0;
8226
8227 if (mddev_is_clustered(mddev)) {
8228 ret = md_cluster_ops->resync_start(mddev);
8229 if (ret) {
8230 mddev->sync_thread = NULL;
8231 goto out;
8232 }
8233 }
8234
8235 mddev->sync_thread = md_register_thread(md_do_sync,
8236 mddev,
8237 "resync");
8238 out:
8239 if (!mddev->sync_thread) {
8240 if (!(mddev_is_clustered(mddev) && ret == -EAGAIN))
8241 printk(KERN_ERR "%s: could not start resync"
8242 " thread...\n",
8243 mdname(mddev));
8244 /* leave the spares where they are, it shouldn't hurt */
8245 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8246 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8247 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8248 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8249 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8250 wake_up(&resync_wait);
8251 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8252 &mddev->recovery))
8253 if (mddev->sysfs_action)
8254 sysfs_notify_dirent_safe(mddev->sysfs_action);
8255 } else
8256 md_wakeup_thread(mddev->sync_thread);
8257 sysfs_notify_dirent_safe(mddev->sysfs_action);
8258 md_new_event(mddev);
8259 }
8260
8261 /*
8262 * This routine is regularly called by all per-raid-array threads to
8263 * deal with generic issues like resync and super-block update.
8264 * Raid personalities that don't have a thread (linear/raid0) do not
8265 * need this as they never do any recovery or update the superblock.
8266 *
8267 * It does not do any resync itself, but rather "forks" off other threads
8268 * to do that as needed.
8269 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8270 * "->recovery" and create a thread at ->sync_thread.
8271 * When the thread finishes it sets MD_RECOVERY_DONE
8272 * and wakeups up this thread which will reap the thread and finish up.
8273 * This thread also removes any faulty devices (with nr_pending == 0).
8274 *
8275 * The overall approach is:
8276 * 1/ if the superblock needs updating, update it.
8277 * 2/ If a recovery thread is running, don't do anything else.
8278 * 3/ If recovery has finished, clean up, possibly marking spares active.
8279 * 4/ If there are any faulty devices, remove them.
8280 * 5/ If array is degraded, try to add spares devices
8281 * 6/ If array has spares or is not in-sync, start a resync thread.
8282 */
8283 void md_check_recovery(struct mddev *mddev)
8284 {
8285 if (mddev->suspended)
8286 return;
8287
8288 if (mddev->bitmap)
8289 bitmap_daemon_work(mddev);
8290
8291 if (signal_pending(current)) {
8292 if (mddev->pers->sync_request && !mddev->external) {
8293 printk(KERN_INFO "md: %s in immediate safe mode\n",
8294 mdname(mddev));
8295 mddev->safemode = 2;
8296 }
8297 flush_signals(current);
8298 }
8299
8300 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
8301 return;
8302 if ( ! (
8303 (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) ||
8304 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8305 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8306 test_bit(MD_RELOAD_SB, &mddev->flags) ||
8307 (mddev->external == 0 && mddev->safemode == 1) ||
8308 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
8309 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
8310 ))
8311 return;
8312
8313 if (mddev_trylock(mddev)) {
8314 int spares = 0;
8315
8316 if (mddev->ro) {
8317 struct md_rdev *rdev;
8318 if (!mddev->external && mddev->in_sync)
8319 /* 'Blocked' flag not needed as failed devices
8320 * will be recorded if array switched to read/write.
8321 * Leaving it set will prevent the device
8322 * from being removed.
8323 */
8324 rdev_for_each(rdev, mddev)
8325 clear_bit(Blocked, &rdev->flags);
8326 /* On a read-only array we can:
8327 * - remove failed devices
8328 * - add already-in_sync devices if the array itself
8329 * is in-sync.
8330 * As we only add devices that are already in-sync,
8331 * we can activate the spares immediately.
8332 */
8333 remove_and_add_spares(mddev, NULL);
8334 /* There is no thread, but we need to call
8335 * ->spare_active and clear saved_raid_disk
8336 */
8337 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8338 md_reap_sync_thread(mddev);
8339 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8340 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8341 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
8342 goto unlock;
8343 }
8344
8345 if (mddev_is_clustered(mddev)) {
8346 struct md_rdev *rdev;
8347 /* kick the device if another node issued a
8348 * remove disk.
8349 */
8350 rdev_for_each(rdev, mddev) {
8351 if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
8352 rdev->raid_disk < 0)
8353 md_kick_rdev_from_array(rdev);
8354 }
8355
8356 if (test_and_clear_bit(MD_RELOAD_SB, &mddev->flags))
8357 md_reload_sb(mddev, mddev->good_device_nr);
8358 }
8359
8360 if (!mddev->external) {
8361 int did_change = 0;
8362 spin_lock(&mddev->lock);
8363 if (mddev->safemode &&
8364 !atomic_read(&mddev->writes_pending) &&
8365 !mddev->in_sync &&
8366 mddev->recovery_cp == MaxSector) {
8367 mddev->in_sync = 1;
8368 did_change = 1;
8369 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
8370 }
8371 if (mddev->safemode == 1)
8372 mddev->safemode = 0;
8373 spin_unlock(&mddev->lock);
8374 if (did_change)
8375 sysfs_notify_dirent_safe(mddev->sysfs_state);
8376 }
8377
8378 if (mddev->flags & MD_UPDATE_SB_FLAGS)
8379 md_update_sb(mddev, 0);
8380
8381 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
8382 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
8383 /* resync/recovery still happening */
8384 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8385 goto unlock;
8386 }
8387 if (mddev->sync_thread) {
8388 md_reap_sync_thread(mddev);
8389 goto unlock;
8390 }
8391 /* Set RUNNING before clearing NEEDED to avoid
8392 * any transients in the value of "sync_action".
8393 */
8394 mddev->curr_resync_completed = 0;
8395 spin_lock(&mddev->lock);
8396 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8397 spin_unlock(&mddev->lock);
8398 /* Clear some bits that don't mean anything, but
8399 * might be left set
8400 */
8401 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
8402 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8403
8404 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8405 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
8406 goto not_running;
8407 /* no recovery is running.
8408 * remove any failed drives, then
8409 * add spares if possible.
8410 * Spares are also removed and re-added, to allow
8411 * the personality to fail the re-add.
8412 */
8413
8414 if (mddev->reshape_position != MaxSector) {
8415 if (mddev->pers->check_reshape == NULL ||
8416 mddev->pers->check_reshape(mddev) != 0)
8417 /* Cannot proceed */
8418 goto not_running;
8419 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8420 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8421 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
8422 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8423 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8424 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8425 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8426 } else if (mddev->recovery_cp < MaxSector) {
8427 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8428 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8429 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
8430 /* nothing to be done ... */
8431 goto not_running;
8432
8433 if (mddev->pers->sync_request) {
8434 if (spares) {
8435 /* We are adding a device or devices to an array
8436 * which has the bitmap stored on all devices.
8437 * So make sure all bitmap pages get written
8438 */
8439 bitmap_write_all(mddev->bitmap);
8440 }
8441 INIT_WORK(&mddev->del_work, md_start_sync);
8442 queue_work(md_misc_wq, &mddev->del_work);
8443 goto unlock;
8444 }
8445 not_running:
8446 if (!mddev->sync_thread) {
8447 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8448 wake_up(&resync_wait);
8449 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8450 &mddev->recovery))
8451 if (mddev->sysfs_action)
8452 sysfs_notify_dirent_safe(mddev->sysfs_action);
8453 }
8454 unlock:
8455 wake_up(&mddev->sb_wait);
8456 mddev_unlock(mddev);
8457 }
8458 }
8459 EXPORT_SYMBOL(md_check_recovery);
8460
8461 void md_reap_sync_thread(struct mddev *mddev)
8462 {
8463 struct md_rdev *rdev;
8464
8465 /* resync has finished, collect result */
8466 md_unregister_thread(&mddev->sync_thread);
8467 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8468 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8469 /* success...*/
8470 /* activate any spares */
8471 if (mddev->pers->spare_active(mddev)) {
8472 sysfs_notify(&mddev->kobj, NULL,
8473 "degraded");
8474 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8475 }
8476 }
8477 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8478 mddev->pers->finish_reshape)
8479 mddev->pers->finish_reshape(mddev);
8480
8481 /* If array is no-longer degraded, then any saved_raid_disk
8482 * information must be scrapped.
8483 */
8484 if (!mddev->degraded)
8485 rdev_for_each(rdev, mddev)
8486 rdev->saved_raid_disk = -1;
8487
8488 md_update_sb(mddev, 1);
8489 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8490 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8491 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8492 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8493 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8494 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8495 wake_up(&resync_wait);
8496 /* flag recovery needed just to double check */
8497 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8498 sysfs_notify_dirent_safe(mddev->sysfs_action);
8499 md_new_event(mddev);
8500 if (mddev->event_work.func)
8501 queue_work(md_misc_wq, &mddev->event_work);
8502 }
8503 EXPORT_SYMBOL(md_reap_sync_thread);
8504
8505 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
8506 {
8507 sysfs_notify_dirent_safe(rdev->sysfs_state);
8508 wait_event_timeout(rdev->blocked_wait,
8509 !test_bit(Blocked, &rdev->flags) &&
8510 !test_bit(BlockedBadBlocks, &rdev->flags),
8511 msecs_to_jiffies(5000));
8512 rdev_dec_pending(rdev, mddev);
8513 }
8514 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8515
8516 void md_finish_reshape(struct mddev *mddev)
8517 {
8518 /* called be personality module when reshape completes. */
8519 struct md_rdev *rdev;
8520
8521 rdev_for_each(rdev, mddev) {
8522 if (rdev->data_offset > rdev->new_data_offset)
8523 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8524 else
8525 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8526 rdev->data_offset = rdev->new_data_offset;
8527 }
8528 }
8529 EXPORT_SYMBOL(md_finish_reshape);
8530
8531 /* Bad block management */
8532
8533 /* Returns 1 on success, 0 on failure */
8534 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8535 int is_new)
8536 {
8537 int rv;
8538 if (is_new)
8539 s += rdev->new_data_offset;
8540 else
8541 s += rdev->data_offset;
8542 rv = badblocks_set(&rdev->badblocks, s, sectors, 0);
8543 if (rv == 0) {
8544 /* Make sure they get written out promptly */
8545 sysfs_notify_dirent_safe(rdev->sysfs_state);
8546 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8547 set_bit(MD_CHANGE_PENDING, &rdev->mddev->flags);
8548 md_wakeup_thread(rdev->mddev->thread);
8549 return 1;
8550 } else
8551 return 0;
8552 }
8553 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8554
8555 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8556 int is_new)
8557 {
8558 if (is_new)
8559 s += rdev->new_data_offset;
8560 else
8561 s += rdev->data_offset;
8562 return badblocks_clear(&rdev->badblocks,
8563 s, sectors);
8564 }
8565 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8566
8567 static int md_notify_reboot(struct notifier_block *this,
8568 unsigned long code, void *x)
8569 {
8570 struct list_head *tmp;
8571 struct mddev *mddev;
8572 int need_delay = 0;
8573
8574 for_each_mddev(mddev, tmp) {
8575 if (mddev_trylock(mddev)) {
8576 if (mddev->pers)
8577 __md_stop_writes(mddev);
8578 if (mddev->persistent)
8579 mddev->safemode = 2;
8580 mddev_unlock(mddev);
8581 }
8582 need_delay = 1;
8583 }
8584 /*
8585 * certain more exotic SCSI devices are known to be
8586 * volatile wrt too early system reboots. While the
8587 * right place to handle this issue is the given
8588 * driver, we do want to have a safe RAID driver ...
8589 */
8590 if (need_delay)
8591 mdelay(1000*1);
8592
8593 return NOTIFY_DONE;
8594 }
8595
8596 static struct notifier_block md_notifier = {
8597 .notifier_call = md_notify_reboot,
8598 .next = NULL,
8599 .priority = INT_MAX, /* before any real devices */
8600 };
8601
8602 static void md_geninit(void)
8603 {
8604 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
8605
8606 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
8607 }
8608
8609 static int __init md_init(void)
8610 {
8611 int ret = -ENOMEM;
8612
8613 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
8614 if (!md_wq)
8615 goto err_wq;
8616
8617 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8618 if (!md_misc_wq)
8619 goto err_misc_wq;
8620
8621 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8622 goto err_md;
8623
8624 if ((ret = register_blkdev(0, "mdp")) < 0)
8625 goto err_mdp;
8626 mdp_major = ret;
8627
8628 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
8629 md_probe, NULL, NULL);
8630 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
8631 md_probe, NULL, NULL);
8632
8633 register_reboot_notifier(&md_notifier);
8634 raid_table_header = register_sysctl_table(raid_root_table);
8635
8636 md_geninit();
8637 return 0;
8638
8639 err_mdp:
8640 unregister_blkdev(MD_MAJOR, "md");
8641 err_md:
8642 destroy_workqueue(md_misc_wq);
8643 err_misc_wq:
8644 destroy_workqueue(md_wq);
8645 err_wq:
8646 return ret;
8647 }
8648
8649 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
8650 {
8651 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
8652 struct md_rdev *rdev2;
8653 int role, ret;
8654 char b[BDEVNAME_SIZE];
8655
8656 /* Check for change of roles in the active devices */
8657 rdev_for_each(rdev2, mddev) {
8658 if (test_bit(Faulty, &rdev2->flags))
8659 continue;
8660
8661 /* Check if the roles changed */
8662 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
8663
8664 if (test_bit(Candidate, &rdev2->flags)) {
8665 if (role == 0xfffe) {
8666 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b));
8667 md_kick_rdev_from_array(rdev2);
8668 continue;
8669 }
8670 else
8671 clear_bit(Candidate, &rdev2->flags);
8672 }
8673
8674 if (role != rdev2->raid_disk) {
8675 /* got activated */
8676 if (rdev2->raid_disk == -1 && role != 0xffff) {
8677 rdev2->saved_raid_disk = role;
8678 ret = remove_and_add_spares(mddev, rdev2);
8679 pr_info("Activated spare: %s\n",
8680 bdevname(rdev2->bdev,b));
8681 }
8682 /* device faulty
8683 * We just want to do the minimum to mark the disk
8684 * as faulty. The recovery is performed by the
8685 * one who initiated the error.
8686 */
8687 if ((role == 0xfffe) || (role == 0xfffd)) {
8688 md_error(mddev, rdev2);
8689 clear_bit(Blocked, &rdev2->flags);
8690 }
8691 }
8692 }
8693
8694 if (mddev->raid_disks != le32_to_cpu(sb->raid_disks))
8695 update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
8696
8697 /* Finally set the event to be up to date */
8698 mddev->events = le64_to_cpu(sb->events);
8699 }
8700
8701 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
8702 {
8703 int err;
8704 struct page *swapout = rdev->sb_page;
8705 struct mdp_superblock_1 *sb;
8706
8707 /* Store the sb page of the rdev in the swapout temporary
8708 * variable in case we err in the future
8709 */
8710 rdev->sb_page = NULL;
8711 alloc_disk_sb(rdev);
8712 ClearPageUptodate(rdev->sb_page);
8713 rdev->sb_loaded = 0;
8714 err = super_types[mddev->major_version].load_super(rdev, NULL, mddev->minor_version);
8715
8716 if (err < 0) {
8717 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
8718 __func__, __LINE__, rdev->desc_nr, err);
8719 put_page(rdev->sb_page);
8720 rdev->sb_page = swapout;
8721 rdev->sb_loaded = 1;
8722 return err;
8723 }
8724
8725 sb = page_address(rdev->sb_page);
8726 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
8727 * is not set
8728 */
8729
8730 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
8731 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
8732
8733 /* The other node finished recovery, call spare_active to set
8734 * device In_sync and mddev->degraded
8735 */
8736 if (rdev->recovery_offset == MaxSector &&
8737 !test_bit(In_sync, &rdev->flags) &&
8738 mddev->pers->spare_active(mddev))
8739 sysfs_notify(&mddev->kobj, NULL, "degraded");
8740
8741 put_page(swapout);
8742 return 0;
8743 }
8744
8745 void md_reload_sb(struct mddev *mddev, int nr)
8746 {
8747 struct md_rdev *rdev;
8748 int err;
8749
8750 /* Find the rdev */
8751 rdev_for_each_rcu(rdev, mddev) {
8752 if (rdev->desc_nr == nr)
8753 break;
8754 }
8755
8756 if (!rdev || rdev->desc_nr != nr) {
8757 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
8758 return;
8759 }
8760
8761 err = read_rdev(mddev, rdev);
8762 if (err < 0)
8763 return;
8764
8765 check_sb_changes(mddev, rdev);
8766
8767 /* Read all rdev's to update recovery_offset */
8768 rdev_for_each_rcu(rdev, mddev)
8769 read_rdev(mddev, rdev);
8770 }
8771 EXPORT_SYMBOL(md_reload_sb);
8772
8773 #ifndef MODULE
8774
8775 /*
8776 * Searches all registered partitions for autorun RAID arrays
8777 * at boot time.
8778 */
8779
8780 static LIST_HEAD(all_detected_devices);
8781 struct detected_devices_node {
8782 struct list_head list;
8783 dev_t dev;
8784 };
8785
8786 void md_autodetect_dev(dev_t dev)
8787 {
8788 struct detected_devices_node *node_detected_dev;
8789
8790 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8791 if (node_detected_dev) {
8792 node_detected_dev->dev = dev;
8793 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8794 } else {
8795 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8796 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8797 }
8798 }
8799
8800 static void autostart_arrays(int part)
8801 {
8802 struct md_rdev *rdev;
8803 struct detected_devices_node *node_detected_dev;
8804 dev_t dev;
8805 int i_scanned, i_passed;
8806
8807 i_scanned = 0;
8808 i_passed = 0;
8809
8810 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
8811
8812 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8813 i_scanned++;
8814 node_detected_dev = list_entry(all_detected_devices.next,
8815 struct detected_devices_node, list);
8816 list_del(&node_detected_dev->list);
8817 dev = node_detected_dev->dev;
8818 kfree(node_detected_dev);
8819 rdev = md_import_device(dev,0, 90);
8820 if (IS_ERR(rdev))
8821 continue;
8822
8823 if (test_bit(Faulty, &rdev->flags))
8824 continue;
8825
8826 set_bit(AutoDetected, &rdev->flags);
8827 list_add(&rdev->same_set, &pending_raid_disks);
8828 i_passed++;
8829 }
8830
8831 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8832 i_scanned, i_passed);
8833
8834 autorun_devices(part);
8835 }
8836
8837 #endif /* !MODULE */
8838
8839 static __exit void md_exit(void)
8840 {
8841 struct mddev *mddev;
8842 struct list_head *tmp;
8843 int delay = 1;
8844
8845 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
8846 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
8847
8848 unregister_blkdev(MD_MAJOR,"md");
8849 unregister_blkdev(mdp_major, "mdp");
8850 unregister_reboot_notifier(&md_notifier);
8851 unregister_sysctl_table(raid_table_header);
8852
8853 /* We cannot unload the modules while some process is
8854 * waiting for us in select() or poll() - wake them up
8855 */
8856 md_unloading = 1;
8857 while (waitqueue_active(&md_event_waiters)) {
8858 /* not safe to leave yet */
8859 wake_up(&md_event_waiters);
8860 msleep(delay);
8861 delay += delay;
8862 }
8863 remove_proc_entry("mdstat", NULL);
8864
8865 for_each_mddev(mddev, tmp) {
8866 export_array(mddev);
8867 mddev->hold_active = 0;
8868 }
8869 destroy_workqueue(md_misc_wq);
8870 destroy_workqueue(md_wq);
8871 }
8872
8873 subsys_initcall(md_init);
8874 module_exit(md_exit)
8875
8876 static int get_ro(char *buffer, struct kernel_param *kp)
8877 {
8878 return sprintf(buffer, "%d", start_readonly);
8879 }
8880 static int set_ro(const char *val, struct kernel_param *kp)
8881 {
8882 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
8883 }
8884
8885 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8886 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
8887 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
8888
8889 MODULE_LICENSE("GPL");
8890 MODULE_DESCRIPTION("MD RAID framework");
8891 MODULE_ALIAS("md");
8892 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Linux with added FPC-III support