?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
bridge: mdb: reduce the indentation level in br_mdb_fill_info
[linux-2.6/btrfs-unstable.git] / drivers / md / md.c
blobe55e6cf9ec17569705ac2dcea8c6f823186a4695
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 if (mddev->suspended++)
309 return;
310 synchronize_rcu();
311 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
312 mddev->pers->quiesce(mddev, 1);
313
314 del_timer_sync(&mddev->safemode_timer);
315 }
316 EXPORT_SYMBOL_GPL(mddev_suspend);
317
318 void mddev_resume(struct mddev *mddev)
319 {
320 if (--mddev->suspended)
321 return;
322 wake_up(&mddev->sb_wait);
323 mddev->pers->quiesce(mddev, 0);
324
325 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
326 md_wakeup_thread(mddev->thread);
327 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
328 }
329 EXPORT_SYMBOL_GPL(mddev_resume);
330
331 int mddev_congested(struct mddev *mddev, int bits)
332 {
333 struct md_personality *pers = mddev->pers;
334 int ret = 0;
335
336 rcu_read_lock();
337 if (mddev->suspended)
338 ret = 1;
339 else if (pers && pers->congested)
340 ret = pers->congested(mddev, bits);
341 rcu_read_unlock();
342 return ret;
343 }
344 EXPORT_SYMBOL_GPL(mddev_congested);
345 static int md_congested(void *data, int bits)
346 {
347 struct mddev *mddev = data;
348 return mddev_congested(mddev, bits);
349 }
350
351 /*
352 * Generic flush handling for md
353 */
354
355 static void md_end_flush(struct bio *bio)
356 {
357 struct md_rdev *rdev = bio->bi_private;
358 struct mddev *mddev = rdev->mddev;
359
360 rdev_dec_pending(rdev, mddev);
361
362 if (atomic_dec_and_test(&mddev->flush_pending)) {
363 /* The pre-request flush has finished */
364 queue_work(md_wq, &mddev->flush_work);
365 }
366 bio_put(bio);
367 }
368
369 static void md_submit_flush_data(struct work_struct *ws);
370
371 static void submit_flushes(struct work_struct *ws)
372 {
373 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
374 struct md_rdev *rdev;
375
376 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
377 atomic_set(&mddev->flush_pending, 1);
378 rcu_read_lock();
379 rdev_for_each_rcu(rdev, mddev)
380 if (rdev->raid_disk >= 0 &&
381 !test_bit(Faulty, &rdev->flags)) {
382 /* Take two references, one is dropped
383 * when request finishes, one after
384 * we reclaim rcu_read_lock
385 */
386 struct bio *bi;
387 atomic_inc(&rdev->nr_pending);
388 atomic_inc(&rdev->nr_pending);
389 rcu_read_unlock();
390 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
391 bi->bi_end_io = md_end_flush;
392 bi->bi_private = rdev;
393 bi->bi_bdev = rdev->bdev;
394 atomic_inc(&mddev->flush_pending);
395 submit_bio(WRITE_FLUSH, bi);
396 rcu_read_lock();
397 rdev_dec_pending(rdev, mddev);
398 }
399 rcu_read_unlock();
400 if (atomic_dec_and_test(&mddev->flush_pending))
401 queue_work(md_wq, &mddev->flush_work);
402 }
403
404 static void md_submit_flush_data(struct work_struct *ws)
405 {
406 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
407 struct bio *bio = mddev->flush_bio;
408
409 if (bio->bi_iter.bi_size == 0)
410 /* an empty barrier - all done */
411 bio_endio(bio);
412 else {
413 bio->bi_rw &= ~REQ_FLUSH;
414 mddev->pers->make_request(mddev, bio);
415 }
416
417 mddev->flush_bio = NULL;
418 wake_up(&mddev->sb_wait);
419 }
420
421 void md_flush_request(struct mddev *mddev, struct bio *bio)
422 {
423 spin_lock_irq(&mddev->lock);
424 wait_event_lock_irq(mddev->sb_wait,
425 !mddev->flush_bio,
426 mddev->lock);
427 mddev->flush_bio = bio;
428 spin_unlock_irq(&mddev->lock);
429
430 INIT_WORK(&mddev->flush_work, submit_flushes);
431 queue_work(md_wq, &mddev->flush_work);
432 }
433 EXPORT_SYMBOL(md_flush_request);
434
435 void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
436 {
437 struct mddev *mddev = cb->data;
438 md_wakeup_thread(mddev->thread);
439 kfree(cb);
440 }
441 EXPORT_SYMBOL(md_unplug);
442
443 static inline struct mddev *mddev_get(struct mddev *mddev)
444 {
445 atomic_inc(&mddev->active);
446 return mddev;
447 }
448
449 static void mddev_delayed_delete(struct work_struct *ws);
450
451 static void mddev_put(struct mddev *mddev)
452 {
453 struct bio_set *bs = NULL;
454
455 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
456 return;
457 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
458 mddev->ctime == 0 && !mddev->hold_active) {
459 /* Array is not configured at all, and not held active,
460 * so destroy it */
461 list_del_init(&mddev->all_mddevs);
462 bs = mddev->bio_set;
463 mddev->bio_set = NULL;
464 if (mddev->gendisk) {
465 /* We did a probe so need to clean up. Call
466 * queue_work inside the spinlock so that
467 * flush_workqueue() after mddev_find will
468 * succeed in waiting for the work to be done.
469 */
470 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
471 queue_work(md_misc_wq, &mddev->del_work);
472 } else
473 kfree(mddev);
474 }
475 spin_unlock(&all_mddevs_lock);
476 if (bs)
477 bioset_free(bs);
478 }
479
480 static void md_safemode_timeout(unsigned long data);
481
482 void mddev_init(struct mddev *mddev)
483 {
484 mutex_init(&mddev->open_mutex);
485 mutex_init(&mddev->reconfig_mutex);
486 mutex_init(&mddev->bitmap_info.mutex);
487 INIT_LIST_HEAD(&mddev->disks);
488 INIT_LIST_HEAD(&mddev->all_mddevs);
489 setup_timer(&mddev->safemode_timer, md_safemode_timeout,
490 (unsigned long) mddev);
491 atomic_set(&mddev->active, 1);
492 atomic_set(&mddev->openers, 0);
493 atomic_set(&mddev->active_io, 0);
494 spin_lock_init(&mddev->lock);
495 atomic_set(&mddev->flush_pending, 0);
496 init_waitqueue_head(&mddev->sb_wait);
497 init_waitqueue_head(&mddev->recovery_wait);
498 mddev->reshape_position = MaxSector;
499 mddev->reshape_backwards = 0;
500 mddev->last_sync_action = "none";
501 mddev->resync_min = 0;
502 mddev->resync_max = MaxSector;
503 mddev->level = LEVEL_NONE;
504 }
505 EXPORT_SYMBOL_GPL(mddev_init);
506
507 static struct mddev *mddev_find(dev_t unit)
508 {
509 struct mddev *mddev, *new = NULL;
510
511 if (unit && MAJOR(unit) != MD_MAJOR)
512 unit &= ~((1<<MdpMinorShift)-1);
513
514 retry:
515 spin_lock(&all_mddevs_lock);
516
517 if (unit) {
518 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
519 if (mddev->unit == unit) {
520 mddev_get(mddev);
521 spin_unlock(&all_mddevs_lock);
522 kfree(new);
523 return mddev;
524 }
525
526 if (new) {
527 list_add(&new->all_mddevs, &all_mddevs);
528 spin_unlock(&all_mddevs_lock);
529 new->hold_active = UNTIL_IOCTL;
530 return new;
531 }
532 } else if (new) {
533 /* find an unused unit number */
534 static int next_minor = 512;
535 int start = next_minor;
536 int is_free = 0;
537 int dev = 0;
538 while (!is_free) {
539 dev = MKDEV(MD_MAJOR, next_minor);
540 next_minor++;
541 if (next_minor > MINORMASK)
542 next_minor = 0;
543 if (next_minor == start) {
544 /* Oh dear, all in use. */
545 spin_unlock(&all_mddevs_lock);
546 kfree(new);
547 return NULL;
548 }
549
550 is_free = 1;
551 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
552 if (mddev->unit == dev) {
553 is_free = 0;
554 break;
555 }
556 }
557 new->unit = dev;
558 new->md_minor = MINOR(dev);
559 new->hold_active = UNTIL_STOP;
560 list_add(&new->all_mddevs, &all_mddevs);
561 spin_unlock(&all_mddevs_lock);
562 return new;
563 }
564 spin_unlock(&all_mddevs_lock);
565
566 new = kzalloc(sizeof(*new), GFP_KERNEL);
567 if (!new)
568 return NULL;
569
570 new->unit = unit;
571 if (MAJOR(unit) == MD_MAJOR)
572 new->md_minor = MINOR(unit);
573 else
574 new->md_minor = MINOR(unit) >> MdpMinorShift;
575
576 mddev_init(new);
577
578 goto retry;
579 }
580
581 static struct attribute_group md_redundancy_group;
582
583 void mddev_unlock(struct mddev *mddev)
584 {
585 if (mddev->to_remove) {
586 /* These cannot be removed under reconfig_mutex as
587 * an access to the files will try to take reconfig_mutex
588 * while holding the file unremovable, which leads to
589 * a deadlock.
590 * So hold set sysfs_active while the remove in happeing,
591 * and anything else which might set ->to_remove or my
592 * otherwise change the sysfs namespace will fail with
593 * -EBUSY if sysfs_active is still set.
594 * We set sysfs_active under reconfig_mutex and elsewhere
595 * test it under the same mutex to ensure its correct value
596 * is seen.
597 */
598 struct attribute_group *to_remove = mddev->to_remove;
599 mddev->to_remove = NULL;
600 mddev->sysfs_active = 1;
601 mutex_unlock(&mddev->reconfig_mutex);
602
603 if (mddev->kobj.sd) {
604 if (to_remove != &md_redundancy_group)
605 sysfs_remove_group(&mddev->kobj, to_remove);
606 if (mddev->pers == NULL ||
607 mddev->pers->sync_request == NULL) {
608 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
609 if (mddev->sysfs_action)
610 sysfs_put(mddev->sysfs_action);
611 mddev->sysfs_action = NULL;
612 }
613 }
614 mddev->sysfs_active = 0;
615 } else
616 mutex_unlock(&mddev->reconfig_mutex);
617
618 /* As we've dropped the mutex we need a spinlock to
619 * make sure the thread doesn't disappear
620 */
621 spin_lock(&pers_lock);
622 md_wakeup_thread(mddev->thread);
623 spin_unlock(&pers_lock);
624 }
625 EXPORT_SYMBOL_GPL(mddev_unlock);
626
627 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
628 {
629 struct md_rdev *rdev;
630
631 rdev_for_each_rcu(rdev, mddev)
632 if (rdev->desc_nr == nr)
633 return rdev;
634
635 return NULL;
636 }
637 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
638
639 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
640 {
641 struct md_rdev *rdev;
642
643 rdev_for_each(rdev, mddev)
644 if (rdev->bdev->bd_dev == dev)
645 return rdev;
646
647 return NULL;
648 }
649
650 static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
651 {
652 struct md_rdev *rdev;
653
654 rdev_for_each_rcu(rdev, mddev)
655 if (rdev->bdev->bd_dev == dev)
656 return rdev;
657
658 return NULL;
659 }
660
661 static struct md_personality *find_pers(int level, char *clevel)
662 {
663 struct md_personality *pers;
664 list_for_each_entry(pers, &pers_list, list) {
665 if (level != LEVEL_NONE && pers->level == level)
666 return pers;
667 if (strcmp(pers->name, clevel)==0)
668 return pers;
669 }
670 return NULL;
671 }
672
673 /* return the offset of the super block in 512byte sectors */
674 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
675 {
676 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
677 return MD_NEW_SIZE_SECTORS(num_sectors);
678 }
679
680 static int alloc_disk_sb(struct md_rdev *rdev)
681 {
682 rdev->sb_page = alloc_page(GFP_KERNEL);
683 if (!rdev->sb_page) {
684 printk(KERN_ALERT "md: out of memory.\n");
685 return -ENOMEM;
686 }
687
688 return 0;
689 }
690
691 void md_rdev_clear(struct md_rdev *rdev)
692 {
693 if (rdev->sb_page) {
694 put_page(rdev->sb_page);
695 rdev->sb_loaded = 0;
696 rdev->sb_page = NULL;
697 rdev->sb_start = 0;
698 rdev->sectors = 0;
699 }
700 if (rdev->bb_page) {
701 put_page(rdev->bb_page);
702 rdev->bb_page = NULL;
703 }
704 badblocks_exit(&rdev->badblocks);
705 }
706 EXPORT_SYMBOL_GPL(md_rdev_clear);
707
708 static void super_written(struct bio *bio)
709 {
710 struct md_rdev *rdev = bio->bi_private;
711 struct mddev *mddev = rdev->mddev;
712
713 if (bio->bi_error) {
714 printk("md: super_written gets error=%d\n", bio->bi_error);
715 md_error(mddev, rdev);
716 }
717
718 if (atomic_dec_and_test(&mddev->pending_writes))
719 wake_up(&mddev->sb_wait);
720 bio_put(bio);
721 }
722
723 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
724 sector_t sector, int size, struct page *page)
725 {
726 /* write first size bytes of page to sector of rdev
727 * Increment mddev->pending_writes before returning
728 * and decrement it on completion, waking up sb_wait
729 * if zero is reached.
730 * If an error occurred, call md_error
731 */
732 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
733
734 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
735 bio->bi_iter.bi_sector = sector;
736 bio_add_page(bio, page, size, 0);
737 bio->bi_private = rdev;
738 bio->bi_end_io = super_written;
739
740 atomic_inc(&mddev->pending_writes);
741 submit_bio(WRITE_FLUSH_FUA, bio);
742 }
743
744 void md_super_wait(struct mddev *mddev)
745 {
746 /* wait for all superblock writes that were scheduled to complete */
747 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
748 }
749
750 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
751 struct page *page, int rw, bool metadata_op)
752 {
753 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
754 int ret;
755
756 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
757 rdev->meta_bdev : rdev->bdev;
758 if (metadata_op)
759 bio->bi_iter.bi_sector = sector + rdev->sb_start;
760 else if (rdev->mddev->reshape_position != MaxSector &&
761 (rdev->mddev->reshape_backwards ==
762 (sector >= rdev->mddev->reshape_position)))
763 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
764 else
765 bio->bi_iter.bi_sector = sector + rdev->data_offset;
766 bio_add_page(bio, page, size, 0);
767 submit_bio_wait(rw, bio);
768
769 ret = !bio->bi_error;
770 bio_put(bio);
771 return ret;
772 }
773 EXPORT_SYMBOL_GPL(sync_page_io);
774
775 static int read_disk_sb(struct md_rdev *rdev, int size)
776 {
777 char b[BDEVNAME_SIZE];
778
779 if (rdev->sb_loaded)
780 return 0;
781
782 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
783 goto fail;
784 rdev->sb_loaded = 1;
785 return 0;
786
787 fail:
788 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
789 bdevname(rdev->bdev,b));
790 return -EINVAL;
791 }
792
793 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
794 {
795 return sb1->set_uuid0 == sb2->set_uuid0 &&
796 sb1->set_uuid1 == sb2->set_uuid1 &&
797 sb1->set_uuid2 == sb2->set_uuid2 &&
798 sb1->set_uuid3 == sb2->set_uuid3;
799 }
800
801 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
802 {
803 int ret;
804 mdp_super_t *tmp1, *tmp2;
805
806 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
807 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
808
809 if (!tmp1 || !tmp2) {
810 ret = 0;
811 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
812 goto abort;
813 }
814
815 *tmp1 = *sb1;
816 *tmp2 = *sb2;
817
818 /*
819 * nr_disks is not constant
820 */
821 tmp1->nr_disks = 0;
822 tmp2->nr_disks = 0;
823
824 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
825 abort:
826 kfree(tmp1);
827 kfree(tmp2);
828 return ret;
829 }
830
831 static u32 md_csum_fold(u32 csum)
832 {
833 csum = (csum & 0xffff) + (csum >> 16);
834 return (csum & 0xffff) + (csum >> 16);
835 }
836
837 static unsigned int calc_sb_csum(mdp_super_t *sb)
838 {
839 u64 newcsum = 0;
840 u32 *sb32 = (u32*)sb;
841 int i;
842 unsigned int disk_csum, csum;
843
844 disk_csum = sb->sb_csum;
845 sb->sb_csum = 0;
846
847 for (i = 0; i < MD_SB_BYTES/4 ; i++)
848 newcsum += sb32[i];
849 csum = (newcsum & 0xffffffff) + (newcsum>>32);
850
851 #ifdef CONFIG_ALPHA
852 /* This used to use csum_partial, which was wrong for several
853 * reasons including that different results are returned on
854 * different architectures. It isn't critical that we get exactly
855 * the same return value as before (we always csum_fold before
856 * testing, and that removes any differences). However as we
857 * know that csum_partial always returned a 16bit value on
858 * alphas, do a fold to maximise conformity to previous behaviour.
859 */
860 sb->sb_csum = md_csum_fold(disk_csum);
861 #else
862 sb->sb_csum = disk_csum;
863 #endif
864 return csum;
865 }
866
867 /*
868 * Handle superblock details.
869 * We want to be able to handle multiple superblock formats
870 * so we have a common interface to them all, and an array of
871 * different handlers.
872 * We rely on user-space to write the initial superblock, and support
873 * reading and updating of superblocks.
874 * Interface methods are:
875 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
876 * loads and validates a superblock on dev.
877 * if refdev != NULL, compare superblocks on both devices
878 * Return:
879 * 0 - dev has a superblock that is compatible with refdev
880 * 1 - dev has a superblock that is compatible and newer than refdev
881 * so dev should be used as the refdev in future
882 * -EINVAL superblock incompatible or invalid
883 * -othererror e.g. -EIO
884 *
885 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
886 * Verify that dev is acceptable into mddev.
887 * The first time, mddev->raid_disks will be 0, and data from
888 * dev should be merged in. Subsequent calls check that dev
889 * is new enough. Return 0 or -EINVAL
890 *
891 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
892 * Update the superblock for rdev with data in mddev
893 * This does not write to disc.
894 *
895 */
896
897 struct super_type {
898 char *name;
899 struct module *owner;
900 int (*load_super)(struct md_rdev *rdev,
901 struct md_rdev *refdev,
902 int minor_version);
903 int (*validate_super)(struct mddev *mddev,
904 struct md_rdev *rdev);
905 void (*sync_super)(struct mddev *mddev,
906 struct md_rdev *rdev);
907 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
908 sector_t num_sectors);
909 int (*allow_new_offset)(struct md_rdev *rdev,
910 unsigned long long new_offset);
911 };
912
913 /*
914 * Check that the given mddev has no bitmap.
915 *
916 * This function is called from the run method of all personalities that do not
917 * support bitmaps. It prints an error message and returns non-zero if mddev
918 * has a bitmap. Otherwise, it returns 0.
919 *
920 */
921 int md_check_no_bitmap(struct mddev *mddev)
922 {
923 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
924 return 0;
925 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
926 mdname(mddev), mddev->pers->name);
927 return 1;
928 }
929 EXPORT_SYMBOL(md_check_no_bitmap);
930
931 /*
932 * load_super for 0.90.0
933 */
934 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
935 {
936 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
937 mdp_super_t *sb;
938 int ret;
939
940 /*
941 * Calculate the position of the superblock (512byte sectors),
942 * it's at the end of the disk.
943 *
944 * It also happens to be a multiple of 4Kb.
945 */
946 rdev->sb_start = calc_dev_sboffset(rdev);
947
948 ret = read_disk_sb(rdev, MD_SB_BYTES);
949 if (ret) return ret;
950
951 ret = -EINVAL;
952
953 bdevname(rdev->bdev, b);
954 sb = page_address(rdev->sb_page);
955
956 if (sb->md_magic != MD_SB_MAGIC) {
957 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
958 b);
959 goto abort;
960 }
961
962 if (sb->major_version != 0 ||
963 sb->minor_version < 90 ||
964 sb->minor_version > 91) {
965 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
966 sb->major_version, sb->minor_version,
967 b);
968 goto abort;
969 }
970
971 if (sb->raid_disks <= 0)
972 goto abort;
973
974 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
975 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
976 b);
977 goto abort;
978 }
979
980 rdev->preferred_minor = sb->md_minor;
981 rdev->data_offset = 0;
982 rdev->new_data_offset = 0;
983 rdev->sb_size = MD_SB_BYTES;
984 rdev->badblocks.shift = -1;
985
986 if (sb->level == LEVEL_MULTIPATH)
987 rdev->desc_nr = -1;
988 else
989 rdev->desc_nr = sb->this_disk.number;
990
991 if (!refdev) {
992 ret = 1;
993 } else {
994 __u64 ev1, ev2;
995 mdp_super_t *refsb = page_address(refdev->sb_page);
996 if (!uuid_equal(refsb, sb)) {
997 printk(KERN_WARNING "md: %s has different UUID to %s\n",
998 b, bdevname(refdev->bdev,b2));
999 goto abort;
1000 }
1001 if (!sb_equal(refsb, sb)) {
1002 printk(KERN_WARNING "md: %s has same UUID"
1003 " but different superblock to %s\n",
1004 b, bdevname(refdev->bdev, b2));
1005 goto abort;
1006 }
1007 ev1 = md_event(sb);
1008 ev2 = md_event(refsb);
1009 if (ev1 > ev2)
1010 ret = 1;
1011 else
1012 ret = 0;
1013 }
1014 rdev->sectors = rdev->sb_start;
1015 /* Limit to 4TB as metadata cannot record more than that.
1016 * (not needed for Linear and RAID0 as metadata doesn't
1017 * record this size)
1018 */
1019 if (IS_ENABLED(CONFIG_LBDAF) && (u64)rdev->sectors >= (2ULL << 32) &&
1020 sb->level >= 1)
1021 rdev->sectors = (sector_t)(2ULL << 32) - 2;
1022
1023 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1024 /* "this cannot possibly happen" ... */
1025 ret = -EINVAL;
1026
1027 abort:
1028 return ret;
1029 }
1030
1031 /*
1032 * validate_super for 0.90.0
1033 */
1034 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1035 {
1036 mdp_disk_t *desc;
1037 mdp_super_t *sb = page_address(rdev->sb_page);
1038 __u64 ev1 = md_event(sb);
1039
1040 rdev->raid_disk = -1;
1041 clear_bit(Faulty, &rdev->flags);
1042 clear_bit(In_sync, &rdev->flags);
1043 clear_bit(Bitmap_sync, &rdev->flags);
1044 clear_bit(WriteMostly, &rdev->flags);
1045
1046 if (mddev->raid_disks == 0) {
1047 mddev->major_version = 0;
1048 mddev->minor_version = sb->minor_version;
1049 mddev->patch_version = sb->patch_version;
1050 mddev->external = 0;
1051 mddev->chunk_sectors = sb->chunk_size >> 9;
1052 mddev->ctime = sb->ctime;
1053 mddev->utime = sb->utime;
1054 mddev->level = sb->level;
1055 mddev->clevel[0] = 0;
1056 mddev->layout = sb->layout;
1057 mddev->raid_disks = sb->raid_disks;
1058 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1059 mddev->events = ev1;
1060 mddev->bitmap_info.offset = 0;
1061 mddev->bitmap_info.space = 0;
1062 /* bitmap can use 60 K after the 4K superblocks */
1063 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1064 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1065 mddev->reshape_backwards = 0;
1066
1067 if (mddev->minor_version >= 91) {
1068 mddev->reshape_position = sb->reshape_position;
1069 mddev->delta_disks = sb->delta_disks;
1070 mddev->new_level = sb->new_level;
1071 mddev->new_layout = sb->new_layout;
1072 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1073 if (mddev->delta_disks < 0)
1074 mddev->reshape_backwards = 1;
1075 } else {
1076 mddev->reshape_position = MaxSector;
1077 mddev->delta_disks = 0;
1078 mddev->new_level = mddev->level;
1079 mddev->new_layout = mddev->layout;
1080 mddev->new_chunk_sectors = mddev->chunk_sectors;
1081 }
1082
1083 if (sb->state & (1<<MD_SB_CLEAN))
1084 mddev->recovery_cp = MaxSector;
1085 else {
1086 if (sb->events_hi == sb->cp_events_hi &&
1087 sb->events_lo == sb->cp_events_lo) {
1088 mddev->recovery_cp = sb->recovery_cp;
1089 } else
1090 mddev->recovery_cp = 0;
1091 }
1092
1093 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1094 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1095 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1096 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1097
1098 mddev->max_disks = MD_SB_DISKS;
1099
1100 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1101 mddev->bitmap_info.file == NULL) {
1102 mddev->bitmap_info.offset =
1103 mddev->bitmap_info.default_offset;
1104 mddev->bitmap_info.space =
1105 mddev->bitmap_info.default_space;
1106 }
1107
1108 } else if (mddev->pers == NULL) {
1109 /* Insist on good event counter while assembling, except
1110 * for spares (which don't need an event count) */
1111 ++ev1;
1112 if (sb->disks[rdev->desc_nr].state & (
1113 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1114 if (ev1 < mddev->events)
1115 return -EINVAL;
1116 } else if (mddev->bitmap) {
1117 /* if adding to array with a bitmap, then we can accept an
1118 * older device ... but not too old.
1119 */
1120 if (ev1 < mddev->bitmap->events_cleared)
1121 return 0;
1122 if (ev1 < mddev->events)
1123 set_bit(Bitmap_sync, &rdev->flags);
1124 } else {
1125 if (ev1 < mddev->events)
1126 /* just a hot-add of a new device, leave raid_disk at -1 */
1127 return 0;
1128 }
1129
1130 if (mddev->level != LEVEL_MULTIPATH) {
1131 desc = sb->disks + rdev->desc_nr;
1132
1133 if (desc->state & (1<<MD_DISK_FAULTY))
1134 set_bit(Faulty, &rdev->flags);
1135 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1136 desc->raid_disk < mddev->raid_disks */) {
1137 set_bit(In_sync, &rdev->flags);
1138 rdev->raid_disk = desc->raid_disk;
1139 rdev->saved_raid_disk = desc->raid_disk;
1140 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1141 /* active but not in sync implies recovery up to
1142 * reshape position. We don't know exactly where
1143 * that is, so set to zero for now */
1144 if (mddev->minor_version >= 91) {
1145 rdev->recovery_offset = 0;
1146 rdev->raid_disk = desc->raid_disk;
1147 }
1148 }
1149 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1150 set_bit(WriteMostly, &rdev->flags);
1151 } else /* MULTIPATH are always insync */
1152 set_bit(In_sync, &rdev->flags);
1153 return 0;
1154 }
1155
1156 /*
1157 * sync_super for 0.90.0
1158 */
1159 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1160 {
1161 mdp_super_t *sb;
1162 struct md_rdev *rdev2;
1163 int next_spare = mddev->raid_disks;
1164
1165 /* make rdev->sb match mddev data..
1166 *
1167 * 1/ zero out disks
1168 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1169 * 3/ any empty disks < next_spare become removed
1170 *
1171 * disks[0] gets initialised to REMOVED because
1172 * we cannot be sure from other fields if it has
1173 * been initialised or not.
1174 */
1175 int i;
1176 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1177
1178 rdev->sb_size = MD_SB_BYTES;
1179
1180 sb = page_address(rdev->sb_page);
1181
1182 memset(sb, 0, sizeof(*sb));
1183
1184 sb->md_magic = MD_SB_MAGIC;
1185 sb->major_version = mddev->major_version;
1186 sb->patch_version = mddev->patch_version;
1187 sb->gvalid_words = 0; /* ignored */
1188 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1189 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1190 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1191 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1192
1193 sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1194 sb->level = mddev->level;
1195 sb->size = mddev->dev_sectors / 2;
1196 sb->raid_disks = mddev->raid_disks;
1197 sb->md_minor = mddev->md_minor;
1198 sb->not_persistent = 0;
1199 sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1200 sb->state = 0;
1201 sb->events_hi = (mddev->events>>32);
1202 sb->events_lo = (u32)mddev->events;
1203
1204 if (mddev->reshape_position == MaxSector)
1205 sb->minor_version = 90;
1206 else {
1207 sb->minor_version = 91;
1208 sb->reshape_position = mddev->reshape_position;
1209 sb->new_level = mddev->new_level;
1210 sb->delta_disks = mddev->delta_disks;
1211 sb->new_layout = mddev->new_layout;
1212 sb->new_chunk = mddev->new_chunk_sectors << 9;
1213 }
1214 mddev->minor_version = sb->minor_version;
1215 if (mddev->in_sync)
1216 {
1217 sb->recovery_cp = mddev->recovery_cp;
1218 sb->cp_events_hi = (mddev->events>>32);
1219 sb->cp_events_lo = (u32)mddev->events;
1220 if (mddev->recovery_cp == MaxSector)
1221 sb->state = (1<< MD_SB_CLEAN);
1222 } else
1223 sb->recovery_cp = 0;
1224
1225 sb->layout = mddev->layout;
1226 sb->chunk_size = mddev->chunk_sectors << 9;
1227
1228 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1229 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1230
1231 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1232 rdev_for_each(rdev2, mddev) {
1233 mdp_disk_t *d;
1234 int desc_nr;
1235 int is_active = test_bit(In_sync, &rdev2->flags);
1236
1237 if (rdev2->raid_disk >= 0 &&
1238 sb->minor_version >= 91)
1239 /* we have nowhere to store the recovery_offset,
1240 * but if it is not below the reshape_position,
1241 * we can piggy-back on that.
1242 */
1243 is_active = 1;
1244 if (rdev2->raid_disk < 0 ||
1245 test_bit(Faulty, &rdev2->flags))
1246 is_active = 0;
1247 if (is_active)
1248 desc_nr = rdev2->raid_disk;
1249 else
1250 desc_nr = next_spare++;
1251 rdev2->desc_nr = desc_nr;
1252 d = &sb->disks[rdev2->desc_nr];
1253 nr_disks++;
1254 d->number = rdev2->desc_nr;
1255 d->major = MAJOR(rdev2->bdev->bd_dev);
1256 d->minor = MINOR(rdev2->bdev->bd_dev);
1257 if (is_active)
1258 d->raid_disk = rdev2->raid_disk;
1259 else
1260 d->raid_disk = rdev2->desc_nr; /* compatibility */
1261 if (test_bit(Faulty, &rdev2->flags))
1262 d->state = (1<<MD_DISK_FAULTY);
1263 else if (is_active) {
1264 d->state = (1<<MD_DISK_ACTIVE);
1265 if (test_bit(In_sync, &rdev2->flags))
1266 d->state |= (1<<MD_DISK_SYNC);
1267 active++;
1268 working++;
1269 } else {
1270 d->state = 0;
1271 spare++;
1272 working++;
1273 }
1274 if (test_bit(WriteMostly, &rdev2->flags))
1275 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1276 }
1277 /* now set the "removed" and "faulty" bits on any missing devices */
1278 for (i=0 ; i < mddev->raid_disks ; i++) {
1279 mdp_disk_t *d = &sb->disks[i];
1280 if (d->state == 0 && d->number == 0) {
1281 d->number = i;
1282 d->raid_disk = i;
1283 d->state = (1<<MD_DISK_REMOVED);
1284 d->state |= (1<<MD_DISK_FAULTY);
1285 failed++;
1286 }
1287 }
1288 sb->nr_disks = nr_disks;
1289 sb->active_disks = active;
1290 sb->working_disks = working;
1291 sb->failed_disks = failed;
1292 sb->spare_disks = spare;
1293
1294 sb->this_disk = sb->disks[rdev->desc_nr];
1295 sb->sb_csum = calc_sb_csum(sb);
1296 }
1297
1298 /*
1299 * rdev_size_change for 0.90.0
1300 */
1301 static unsigned long long
1302 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1303 {
1304 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1305 return 0; /* component must fit device */
1306 if (rdev->mddev->bitmap_info.offset)
1307 return 0; /* can't move bitmap */
1308 rdev->sb_start = calc_dev_sboffset(rdev);
1309 if (!num_sectors || num_sectors > rdev->sb_start)
1310 num_sectors = rdev->sb_start;
1311 /* Limit to 4TB as metadata cannot record more than that.
1312 * 4TB == 2^32 KB, or 2*2^32 sectors.
1313 */
1314 if (IS_ENABLED(CONFIG_LBDAF) && (u64)num_sectors >= (2ULL << 32) &&
1315 rdev->mddev->level >= 1)
1316 num_sectors = (sector_t)(2ULL << 32) - 2;
1317 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1318 rdev->sb_page);
1319 md_super_wait(rdev->mddev);
1320 return num_sectors;
1321 }
1322
1323 static int
1324 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1325 {
1326 /* non-zero offset changes not possible with v0.90 */
1327 return new_offset == 0;
1328 }
1329
1330 /*
1331 * version 1 superblock
1332 */
1333
1334 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1335 {
1336 __le32 disk_csum;
1337 u32 csum;
1338 unsigned long long newcsum;
1339 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1340 __le32 *isuper = (__le32*)sb;
1341
1342 disk_csum = sb->sb_csum;
1343 sb->sb_csum = 0;
1344 newcsum = 0;
1345 for (; size >= 4; size -= 4)
1346 newcsum += le32_to_cpu(*isuper++);
1347
1348 if (size == 2)
1349 newcsum += le16_to_cpu(*(__le16*) isuper);
1350
1351 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1352 sb->sb_csum = disk_csum;
1353 return cpu_to_le32(csum);
1354 }
1355
1356 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1357 {
1358 struct mdp_superblock_1 *sb;
1359 int ret;
1360 sector_t sb_start;
1361 sector_t sectors;
1362 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1363 int bmask;
1364
1365 /*
1366 * Calculate the position of the superblock in 512byte sectors.
1367 * It is always aligned to a 4K boundary and
1368 * depeding on minor_version, it can be:
1369 * 0: At least 8K, but less than 12K, from end of device
1370 * 1: At start of device
1371 * 2: 4K from start of device.
1372 */
1373 switch(minor_version) {
1374 case 0:
1375 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1376 sb_start -= 8*2;
1377 sb_start &= ~(sector_t)(4*2-1);
1378 break;
1379 case 1:
1380 sb_start = 0;
1381 break;
1382 case 2:
1383 sb_start = 8;
1384 break;
1385 default:
1386 return -EINVAL;
1387 }
1388 rdev->sb_start = sb_start;
1389
1390 /* superblock is rarely larger than 1K, but it can be larger,
1391 * and it is safe to read 4k, so we do that
1392 */
1393 ret = read_disk_sb(rdev, 4096);
1394 if (ret) return ret;
1395
1396 sb = page_address(rdev->sb_page);
1397
1398 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1399 sb->major_version != cpu_to_le32(1) ||
1400 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1401 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1402 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1403 return -EINVAL;
1404
1405 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1406 printk("md: invalid superblock checksum on %s\n",
1407 bdevname(rdev->bdev,b));
1408 return -EINVAL;
1409 }
1410 if (le64_to_cpu(sb->data_size) < 10) {
1411 printk("md: data_size too small on %s\n",
1412 bdevname(rdev->bdev,b));
1413 return -EINVAL;
1414 }
1415 if (sb->pad0 ||
1416 sb->pad3[0] ||
1417 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1418 /* Some padding is non-zero, might be a new feature */
1419 return -EINVAL;
1420
1421 rdev->preferred_minor = 0xffff;
1422 rdev->data_offset = le64_to_cpu(sb->data_offset);
1423 rdev->new_data_offset = rdev->data_offset;
1424 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1425 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1426 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1427 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1428
1429 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1430 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1431 if (rdev->sb_size & bmask)
1432 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1433
1434 if (minor_version
1435 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1436 return -EINVAL;
1437 if (minor_version
1438 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1439 return -EINVAL;
1440
1441 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1442 rdev->desc_nr = -1;
1443 else
1444 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1445
1446 if (!rdev->bb_page) {
1447 rdev->bb_page = alloc_page(GFP_KERNEL);
1448 if (!rdev->bb_page)
1449 return -ENOMEM;
1450 }
1451 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1452 rdev->badblocks.count == 0) {
1453 /* need to load the bad block list.
1454 * Currently we limit it to one page.
1455 */
1456 s32 offset;
1457 sector_t bb_sector;
1458 u64 *bbp;
1459 int i;
1460 int sectors = le16_to_cpu(sb->bblog_size);
1461 if (sectors > (PAGE_SIZE / 512))
1462 return -EINVAL;
1463 offset = le32_to_cpu(sb->bblog_offset);
1464 if (offset == 0)
1465 return -EINVAL;
1466 bb_sector = (long long)offset;
1467 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1468 rdev->bb_page, READ, true))
1469 return -EIO;
1470 bbp = (u64 *)page_address(rdev->bb_page);
1471 rdev->badblocks.shift = sb->bblog_shift;
1472 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1473 u64 bb = le64_to_cpu(*bbp);
1474 int count = bb & (0x3ff);
1475 u64 sector = bb >> 10;
1476 sector <<= sb->bblog_shift;
1477 count <<= sb->bblog_shift;
1478 if (bb + 1 == 0)
1479 break;
1480 if (badblocks_set(&rdev->badblocks, sector, count, 1))
1481 return -EINVAL;
1482 }
1483 } else if (sb->bblog_offset != 0)
1484 rdev->badblocks.shift = 0;
1485
1486 if (!refdev) {
1487 ret = 1;
1488 } else {
1489 __u64 ev1, ev2;
1490 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1491
1492 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1493 sb->level != refsb->level ||
1494 sb->layout != refsb->layout ||
1495 sb->chunksize != refsb->chunksize) {
1496 printk(KERN_WARNING "md: %s has strangely different"
1497 " superblock to %s\n",
1498 bdevname(rdev->bdev,b),
1499 bdevname(refdev->bdev,b2));
1500 return -EINVAL;
1501 }
1502 ev1 = le64_to_cpu(sb->events);
1503 ev2 = le64_to_cpu(refsb->events);
1504
1505 if (ev1 > ev2)
1506 ret = 1;
1507 else
1508 ret = 0;
1509 }
1510 if (minor_version) {
1511 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1512 sectors -= rdev->data_offset;
1513 } else
1514 sectors = rdev->sb_start;
1515 if (sectors < le64_to_cpu(sb->data_size))
1516 return -EINVAL;
1517 rdev->sectors = le64_to_cpu(sb->data_size);
1518 return ret;
1519 }
1520
1521 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1522 {
1523 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1524 __u64 ev1 = le64_to_cpu(sb->events);
1525
1526 rdev->raid_disk = -1;
1527 clear_bit(Faulty, &rdev->flags);
1528 clear_bit(In_sync, &rdev->flags);
1529 clear_bit(Bitmap_sync, &rdev->flags);
1530 clear_bit(WriteMostly, &rdev->flags);
1531
1532 if (mddev->raid_disks == 0) {
1533 mddev->major_version = 1;
1534 mddev->patch_version = 0;
1535 mddev->external = 0;
1536 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1537 mddev->ctime = le64_to_cpu(sb->ctime);
1538 mddev->utime = le64_to_cpu(sb->utime);
1539 mddev->level = le32_to_cpu(sb->level);
1540 mddev->clevel[0] = 0;
1541 mddev->layout = le32_to_cpu(sb->layout);
1542 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1543 mddev->dev_sectors = le64_to_cpu(sb->size);
1544 mddev->events = ev1;
1545 mddev->bitmap_info.offset = 0;
1546 mddev->bitmap_info.space = 0;
1547 /* Default location for bitmap is 1K after superblock
1548 * using 3K - total of 4K
1549 */
1550 mddev->bitmap_info.default_offset = 1024 >> 9;
1551 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1552 mddev->reshape_backwards = 0;
1553
1554 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1555 memcpy(mddev->uuid, sb->set_uuid, 16);
1556
1557 mddev->max_disks = (4096-256)/2;
1558
1559 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1560 mddev->bitmap_info.file == NULL) {
1561 mddev->bitmap_info.offset =
1562 (__s32)le32_to_cpu(sb->bitmap_offset);
1563 /* Metadata doesn't record how much space is available.
1564 * For 1.0, we assume we can use up to the superblock
1565 * if before, else to 4K beyond superblock.
1566 * For others, assume no change is possible.
1567 */
1568 if (mddev->minor_version > 0)
1569 mddev->bitmap_info.space = 0;
1570 else if (mddev->bitmap_info.offset > 0)
1571 mddev->bitmap_info.space =
1572 8 - mddev->bitmap_info.offset;
1573 else
1574 mddev->bitmap_info.space =
1575 -mddev->bitmap_info.offset;
1576 }
1577
1578 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1579 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1580 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1581 mddev->new_level = le32_to_cpu(sb->new_level);
1582 mddev->new_layout = le32_to_cpu(sb->new_layout);
1583 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1584 if (mddev->delta_disks < 0 ||
1585 (mddev->delta_disks == 0 &&
1586 (le32_to_cpu(sb->feature_map)
1587 & MD_FEATURE_RESHAPE_BACKWARDS)))
1588 mddev->reshape_backwards = 1;
1589 } else {
1590 mddev->reshape_position = MaxSector;
1591 mddev->delta_disks = 0;
1592 mddev->new_level = mddev->level;
1593 mddev->new_layout = mddev->layout;
1594 mddev->new_chunk_sectors = mddev->chunk_sectors;
1595 }
1596
1597 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL) {
1598 set_bit(MD_HAS_JOURNAL, &mddev->flags);
1599 if (mddev->recovery_cp == MaxSector)
1600 set_bit(MD_JOURNAL_CLEAN, &mddev->flags);
1601 }
1602 } else if (mddev->pers == NULL) {
1603 /* Insist of good event counter while assembling, except for
1604 * spares (which don't need an event count) */
1605 ++ev1;
1606 if (rdev->desc_nr >= 0 &&
1607 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1608 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1609 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1610 if (ev1 < mddev->events)
1611 return -EINVAL;
1612 } else if (mddev->bitmap) {
1613 /* If adding to array with a bitmap, then we can accept an
1614 * older device, but not too old.
1615 */
1616 if (ev1 < mddev->bitmap->events_cleared)
1617 return 0;
1618 if (ev1 < mddev->events)
1619 set_bit(Bitmap_sync, &rdev->flags);
1620 } else {
1621 if (ev1 < mddev->events)
1622 /* just a hot-add of a new device, leave raid_disk at -1 */
1623 return 0;
1624 }
1625 if (mddev->level != LEVEL_MULTIPATH) {
1626 int role;
1627 if (rdev->desc_nr < 0 ||
1628 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1629 role = MD_DISK_ROLE_SPARE;
1630 rdev->desc_nr = -1;
1631 } else
1632 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1633 switch(role) {
1634 case MD_DISK_ROLE_SPARE: /* spare */
1635 break;
1636 case MD_DISK_ROLE_FAULTY: /* faulty */
1637 set_bit(Faulty, &rdev->flags);
1638 break;
1639 case MD_DISK_ROLE_JOURNAL: /* journal device */
1640 if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1641 /* journal device without journal feature */
1642 printk(KERN_WARNING
1643 "md: journal device provided without journal feature, ignoring the device\n");
1644 return -EINVAL;
1645 }
1646 set_bit(Journal, &rdev->flags);
1647 rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1648 rdev->raid_disk = 0;
1649 break;
1650 default:
1651 rdev->saved_raid_disk = role;
1652 if ((le32_to_cpu(sb->feature_map) &
1653 MD_FEATURE_RECOVERY_OFFSET)) {
1654 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1655 if (!(le32_to_cpu(sb->feature_map) &
1656 MD_FEATURE_RECOVERY_BITMAP))
1657 rdev->saved_raid_disk = -1;
1658 } else
1659 set_bit(In_sync, &rdev->flags);
1660 rdev->raid_disk = role;
1661 break;
1662 }
1663 if (sb->devflags & WriteMostly1)
1664 set_bit(WriteMostly, &rdev->flags);
1665 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1666 set_bit(Replacement, &rdev->flags);
1667 } else /* MULTIPATH are always insync */
1668 set_bit(In_sync, &rdev->flags);
1669
1670 return 0;
1671 }
1672
1673 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1674 {
1675 struct mdp_superblock_1 *sb;
1676 struct md_rdev *rdev2;
1677 int max_dev, i;
1678 /* make rdev->sb match mddev and rdev data. */
1679
1680 sb = page_address(rdev->sb_page);
1681
1682 sb->feature_map = 0;
1683 sb->pad0 = 0;
1684 sb->recovery_offset = cpu_to_le64(0);
1685 memset(sb->pad3, 0, sizeof(sb->pad3));
1686
1687 sb->utime = cpu_to_le64((__u64)mddev->utime);
1688 sb->events = cpu_to_le64(mddev->events);
1689 if (mddev->in_sync)
1690 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1691 else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
1692 sb->resync_offset = cpu_to_le64(MaxSector);
1693 else
1694 sb->resync_offset = cpu_to_le64(0);
1695
1696 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1697
1698 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1699 sb->size = cpu_to_le64(mddev->dev_sectors);
1700 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1701 sb->level = cpu_to_le32(mddev->level);
1702 sb->layout = cpu_to_le32(mddev->layout);
1703
1704 if (test_bit(WriteMostly, &rdev->flags))
1705 sb->devflags |= WriteMostly1;
1706 else
1707 sb->devflags &= ~WriteMostly1;
1708 sb->data_offset = cpu_to_le64(rdev->data_offset);
1709 sb->data_size = cpu_to_le64(rdev->sectors);
1710
1711 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1712 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1713 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1714 }
1715
1716 if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
1717 !test_bit(In_sync, &rdev->flags)) {
1718 sb->feature_map |=
1719 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1720 sb->recovery_offset =
1721 cpu_to_le64(rdev->recovery_offset);
1722 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1723 sb->feature_map |=
1724 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
1725 }
1726 /* Note: recovery_offset and journal_tail share space */
1727 if (test_bit(Journal, &rdev->flags))
1728 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
1729 if (test_bit(Replacement, &rdev->flags))
1730 sb->feature_map |=
1731 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1732
1733 if (mddev->reshape_position != MaxSector) {
1734 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1735 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1736 sb->new_layout = cpu_to_le32(mddev->new_layout);
1737 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1738 sb->new_level = cpu_to_le32(mddev->new_level);
1739 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1740 if (mddev->delta_disks == 0 &&
1741 mddev->reshape_backwards)
1742 sb->feature_map
1743 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1744 if (rdev->new_data_offset != rdev->data_offset) {
1745 sb->feature_map
1746 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1747 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1748 - rdev->data_offset));
1749 }
1750 }
1751
1752 if (mddev_is_clustered(mddev))
1753 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
1754
1755 if (rdev->badblocks.count == 0)
1756 /* Nothing to do for bad blocks*/ ;
1757 else if (sb->bblog_offset == 0)
1758 /* Cannot record bad blocks on this device */
1759 md_error(mddev, rdev);
1760 else {
1761 struct badblocks *bb = &rdev->badblocks;
1762 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1763 u64 *p = bb->page;
1764 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1765 if (bb->changed) {
1766 unsigned seq;
1767
1768 retry:
1769 seq = read_seqbegin(&bb->lock);
1770
1771 memset(bbp, 0xff, PAGE_SIZE);
1772
1773 for (i = 0 ; i < bb->count ; i++) {
1774 u64 internal_bb = p[i];
1775 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1776 | BB_LEN(internal_bb));
1777 bbp[i] = cpu_to_le64(store_bb);
1778 }
1779 bb->changed = 0;
1780 if (read_seqretry(&bb->lock, seq))
1781 goto retry;
1782
1783 bb->sector = (rdev->sb_start +
1784 (int)le32_to_cpu(sb->bblog_offset));
1785 bb->size = le16_to_cpu(sb->bblog_size);
1786 }
1787 }
1788
1789 max_dev = 0;
1790 rdev_for_each(rdev2, mddev)
1791 if (rdev2->desc_nr+1 > max_dev)
1792 max_dev = rdev2->desc_nr+1;
1793
1794 if (max_dev > le32_to_cpu(sb->max_dev)) {
1795 int bmask;
1796 sb->max_dev = cpu_to_le32(max_dev);
1797 rdev->sb_size = max_dev * 2 + 256;
1798 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1799 if (rdev->sb_size & bmask)
1800 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1801 } else
1802 max_dev = le32_to_cpu(sb->max_dev);
1803
1804 for (i=0; i<max_dev;i++)
1805 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
1806
1807 if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
1808 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
1809
1810 rdev_for_each(rdev2, mddev) {
1811 i = rdev2->desc_nr;
1812 if (test_bit(Faulty, &rdev2->flags))
1813 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
1814 else if (test_bit(In_sync, &rdev2->flags))
1815 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1816 else if (test_bit(Journal, &rdev2->flags))
1817 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
1818 else if (rdev2->raid_disk >= 0)
1819 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1820 else
1821 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
1822 }
1823
1824 sb->sb_csum = calc_sb_1_csum(sb);
1825 }
1826
1827 static unsigned long long
1828 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1829 {
1830 struct mdp_superblock_1 *sb;
1831 sector_t max_sectors;
1832 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1833 return 0; /* component must fit device */
1834 if (rdev->data_offset != rdev->new_data_offset)
1835 return 0; /* too confusing */
1836 if (rdev->sb_start < rdev->data_offset) {
1837 /* minor versions 1 and 2; superblock before data */
1838 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1839 max_sectors -= rdev->data_offset;
1840 if (!num_sectors || num_sectors > max_sectors)
1841 num_sectors = max_sectors;
1842 } else if (rdev->mddev->bitmap_info.offset) {
1843 /* minor version 0 with bitmap we can't move */
1844 return 0;
1845 } else {
1846 /* minor version 0; superblock after data */
1847 sector_t sb_start;
1848 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1849 sb_start &= ~(sector_t)(4*2 - 1);
1850 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1851 if (!num_sectors || num_sectors > max_sectors)
1852 num_sectors = max_sectors;
1853 rdev->sb_start = sb_start;
1854 }
1855 sb = page_address(rdev->sb_page);
1856 sb->data_size = cpu_to_le64(num_sectors);
1857 sb->super_offset = rdev->sb_start;
1858 sb->sb_csum = calc_sb_1_csum(sb);
1859 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1860 rdev->sb_page);
1861 md_super_wait(rdev->mddev);
1862 return num_sectors;
1863
1864 }
1865
1866 static int
1867 super_1_allow_new_offset(struct md_rdev *rdev,
1868 unsigned long long new_offset)
1869 {
1870 /* All necessary checks on new >= old have been done */
1871 struct bitmap *bitmap;
1872 if (new_offset >= rdev->data_offset)
1873 return 1;
1874
1875 /* with 1.0 metadata, there is no metadata to tread on
1876 * so we can always move back */
1877 if (rdev->mddev->minor_version == 0)
1878 return 1;
1879
1880 /* otherwise we must be sure not to step on
1881 * any metadata, so stay:
1882 * 36K beyond start of superblock
1883 * beyond end of badblocks
1884 * beyond write-intent bitmap
1885 */
1886 if (rdev->sb_start + (32+4)*2 > new_offset)
1887 return 0;
1888 bitmap = rdev->mddev->bitmap;
1889 if (bitmap && !rdev->mddev->bitmap_info.file &&
1890 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1891 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
1892 return 0;
1893 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1894 return 0;
1895
1896 return 1;
1897 }
1898
1899 static struct super_type super_types[] = {
1900 [0] = {
1901 .name = "0.90.0",
1902 .owner = THIS_MODULE,
1903 .load_super = super_90_load,
1904 .validate_super = super_90_validate,
1905 .sync_super = super_90_sync,
1906 .rdev_size_change = super_90_rdev_size_change,
1907 .allow_new_offset = super_90_allow_new_offset,
1908 },
1909 [1] = {
1910 .name = "md-1",
1911 .owner = THIS_MODULE,
1912 .load_super = super_1_load,
1913 .validate_super = super_1_validate,
1914 .sync_super = super_1_sync,
1915 .rdev_size_change = super_1_rdev_size_change,
1916 .allow_new_offset = super_1_allow_new_offset,
1917 },
1918 };
1919
1920 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1921 {
1922 if (mddev->sync_super) {
1923 mddev->sync_super(mddev, rdev);
1924 return;
1925 }
1926
1927 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1928
1929 super_types[mddev->major_version].sync_super(mddev, rdev);
1930 }
1931
1932 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1933 {
1934 struct md_rdev *rdev, *rdev2;
1935
1936 rcu_read_lock();
1937 rdev_for_each_rcu(rdev, mddev1) {
1938 if (test_bit(Faulty, &rdev->flags) ||
1939 test_bit(Journal, &rdev->flags) ||
1940 rdev->raid_disk == -1)
1941 continue;
1942 rdev_for_each_rcu(rdev2, mddev2) {
1943 if (test_bit(Faulty, &rdev2->flags) ||
1944 test_bit(Journal, &rdev2->flags) ||
1945 rdev2->raid_disk == -1)
1946 continue;
1947 if (rdev->bdev->bd_contains ==
1948 rdev2->bdev->bd_contains) {
1949 rcu_read_unlock();
1950 return 1;
1951 }
1952 }
1953 }
1954 rcu_read_unlock();
1955 return 0;
1956 }
1957
1958 static LIST_HEAD(pending_raid_disks);
1959
1960 /*
1961 * Try to register data integrity profile for an mddev
1962 *
1963 * This is called when an array is started and after a disk has been kicked
1964 * from the array. It only succeeds if all working and active component devices
1965 * are integrity capable with matching profiles.
1966 */
1967 int md_integrity_register(struct mddev *mddev)
1968 {
1969 struct md_rdev *rdev, *reference = NULL;
1970
1971 if (list_empty(&mddev->disks))
1972 return 0; /* nothing to do */
1973 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1974 return 0; /* shouldn't register, or already is */
1975 rdev_for_each(rdev, mddev) {
1976 /* skip spares and non-functional disks */
1977 if (test_bit(Faulty, &rdev->flags))
1978 continue;
1979 if (rdev->raid_disk < 0)
1980 continue;
1981 if (!reference) {
1982 /* Use the first rdev as the reference */
1983 reference = rdev;
1984 continue;
1985 }
1986 /* does this rdev's profile match the reference profile? */
1987 if (blk_integrity_compare(reference->bdev->bd_disk,
1988 rdev->bdev->bd_disk) < 0)
1989 return -EINVAL;
1990 }
1991 if (!reference || !bdev_get_integrity(reference->bdev))
1992 return 0;
1993 /*
1994 * All component devices are integrity capable and have matching
1995 * profiles, register the common profile for the md device.
1996 */
1997 blk_integrity_register(mddev->gendisk,
1998 bdev_get_integrity(reference->bdev));
1999
2000 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
2001 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
2002 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
2003 mdname(mddev));
2004 return -EINVAL;
2005 }
2006 return 0;
2007 }
2008 EXPORT_SYMBOL(md_integrity_register);
2009
2010 /*
2011 * Attempt to add an rdev, but only if it is consistent with the current
2012 * integrity profile
2013 */
2014 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2015 {
2016 struct blk_integrity *bi_rdev;
2017 struct blk_integrity *bi_mddev;
2018 char name[BDEVNAME_SIZE];
2019
2020 if (!mddev->gendisk)
2021 return 0;
2022
2023 bi_rdev = bdev_get_integrity(rdev->bdev);
2024 bi_mddev = blk_get_integrity(mddev->gendisk);
2025
2026 if (!bi_mddev) /* nothing to do */
2027 return 0;
2028
2029 if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2030 printk(KERN_NOTICE "%s: incompatible integrity profile for %s\n",
2031 mdname(mddev), bdevname(rdev->bdev, name));
2032 return -ENXIO;
2033 }
2034
2035 return 0;
2036 }
2037 EXPORT_SYMBOL(md_integrity_add_rdev);
2038
2039 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2040 {
2041 char b[BDEVNAME_SIZE];
2042 struct kobject *ko;
2043 int err;
2044
2045 /* prevent duplicates */
2046 if (find_rdev(mddev, rdev->bdev->bd_dev))
2047 return -EEXIST;
2048
2049 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2050 if (!test_bit(Journal, &rdev->flags) &&
2051 rdev->sectors &&
2052 (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2053 if (mddev->pers) {
2054 /* Cannot change size, so fail
2055 * If mddev->level <= 0, then we don't care
2056 * about aligning sizes (e.g. linear)
2057 */
2058 if (mddev->level > 0)
2059 return -ENOSPC;
2060 } else
2061 mddev->dev_sectors = rdev->sectors;
2062 }
2063
2064 /* Verify rdev->desc_nr is unique.
2065 * If it is -1, assign a free number, else
2066 * check number is not in use
2067 */
2068 rcu_read_lock();
2069 if (rdev->desc_nr < 0) {
2070 int choice = 0;
2071 if (mddev->pers)
2072 choice = mddev->raid_disks;
2073 while (md_find_rdev_nr_rcu(mddev, choice))
2074 choice++;
2075 rdev->desc_nr = choice;
2076 } else {
2077 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2078 rcu_read_unlock();
2079 return -EBUSY;
2080 }
2081 }
2082 rcu_read_unlock();
2083 if (!test_bit(Journal, &rdev->flags) &&
2084 mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2085 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2086 mdname(mddev), mddev->max_disks);
2087 return -EBUSY;
2088 }
2089 bdevname(rdev->bdev,b);
2090 strreplace(b, '/', '!');
2091
2092 rdev->mddev = mddev;
2093 printk(KERN_INFO "md: bind<%s>\n", b);
2094
2095 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2096 goto fail;
2097
2098 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2099 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2100 /* failure here is OK */;
2101 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2102
2103 list_add_rcu(&rdev->same_set, &mddev->disks);
2104 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2105
2106 /* May as well allow recovery to be retried once */
2107 mddev->recovery_disabled++;
2108
2109 return 0;
2110
2111 fail:
2112 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2113 b, mdname(mddev));
2114 return err;
2115 }
2116
2117 static void md_delayed_delete(struct work_struct *ws)
2118 {
2119 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2120 kobject_del(&rdev->kobj);
2121 kobject_put(&rdev->kobj);
2122 }
2123
2124 static void unbind_rdev_from_array(struct md_rdev *rdev)
2125 {
2126 char b[BDEVNAME_SIZE];
2127
2128 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2129 list_del_rcu(&rdev->same_set);
2130 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2131 rdev->mddev = NULL;
2132 sysfs_remove_link(&rdev->kobj, "block");
2133 sysfs_put(rdev->sysfs_state);
2134 rdev->sysfs_state = NULL;
2135 rdev->badblocks.count = 0;
2136 /* We need to delay this, otherwise we can deadlock when
2137 * writing to 'remove' to "dev/state". We also need
2138 * to delay it due to rcu usage.
2139 */
2140 synchronize_rcu();
2141 INIT_WORK(&rdev->del_work, md_delayed_delete);
2142 kobject_get(&rdev->kobj);
2143 queue_work(md_misc_wq, &rdev->del_work);
2144 }
2145
2146 /*
2147 * prevent the device from being mounted, repartitioned or
2148 * otherwise reused by a RAID array (or any other kernel
2149 * subsystem), by bd_claiming the device.
2150 */
2151 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2152 {
2153 int err = 0;
2154 struct block_device *bdev;
2155 char b[BDEVNAME_SIZE];
2156
2157 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2158 shared ? (struct md_rdev *)lock_rdev : rdev);
2159 if (IS_ERR(bdev)) {
2160 printk(KERN_ERR "md: could not open %s.\n",
2161 __bdevname(dev, b));
2162 return PTR_ERR(bdev);
2163 }
2164 rdev->bdev = bdev;
2165 return err;
2166 }
2167
2168 static void unlock_rdev(struct md_rdev *rdev)
2169 {
2170 struct block_device *bdev = rdev->bdev;
2171 rdev->bdev = NULL;
2172 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2173 }
2174
2175 void md_autodetect_dev(dev_t dev);
2176
2177 static void export_rdev(struct md_rdev *rdev)
2178 {
2179 char b[BDEVNAME_SIZE];
2180
2181 printk(KERN_INFO "md: export_rdev(%s)\n",
2182 bdevname(rdev->bdev,b));
2183 md_rdev_clear(rdev);
2184 #ifndef MODULE
2185 if (test_bit(AutoDetected, &rdev->flags))
2186 md_autodetect_dev(rdev->bdev->bd_dev);
2187 #endif
2188 unlock_rdev(rdev);
2189 kobject_put(&rdev->kobj);
2190 }
2191
2192 void md_kick_rdev_from_array(struct md_rdev *rdev)
2193 {
2194 unbind_rdev_from_array(rdev);
2195 export_rdev(rdev);
2196 }
2197 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2198
2199 static void export_array(struct mddev *mddev)
2200 {
2201 struct md_rdev *rdev;
2202
2203 while (!list_empty(&mddev->disks)) {
2204 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2205 same_set);
2206 md_kick_rdev_from_array(rdev);
2207 }
2208 mddev->raid_disks = 0;
2209 mddev->major_version = 0;
2210 }
2211
2212 static void sync_sbs(struct mddev *mddev, int nospares)
2213 {
2214 /* Update each superblock (in-memory image), but
2215 * if we are allowed to, skip spares which already
2216 * have the right event counter, or have one earlier
2217 * (which would mean they aren't being marked as dirty
2218 * with the rest of the array)
2219 */
2220 struct md_rdev *rdev;
2221 rdev_for_each(rdev, mddev) {
2222 if (rdev->sb_events == mddev->events ||
2223 (nospares &&
2224 rdev->raid_disk < 0 &&
2225 rdev->sb_events+1 == mddev->events)) {
2226 /* Don't update this superblock */
2227 rdev->sb_loaded = 2;
2228 } else {
2229 sync_super(mddev, rdev);
2230 rdev->sb_loaded = 1;
2231 }
2232 }
2233 }
2234
2235 static bool does_sb_need_changing(struct mddev *mddev)
2236 {
2237 struct md_rdev *rdev;
2238 struct mdp_superblock_1 *sb;
2239 int role;
2240
2241 /* Find a good rdev */
2242 rdev_for_each(rdev, mddev)
2243 if ((rdev->raid_disk >= 0) && !test_bit(Faulty, &rdev->flags))
2244 break;
2245
2246 /* No good device found. */
2247 if (!rdev)
2248 return false;
2249
2250 sb = page_address(rdev->sb_page);
2251 /* Check if a device has become faulty or a spare become active */
2252 rdev_for_each(rdev, mddev) {
2253 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2254 /* Device activated? */
2255 if (role == 0xffff && rdev->raid_disk >=0 &&
2256 !test_bit(Faulty, &rdev->flags))
2257 return true;
2258 /* Device turned faulty? */
2259 if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd))
2260 return true;
2261 }
2262
2263 /* Check if any mddev parameters have changed */
2264 if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2265 (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2266 (mddev->layout != le64_to_cpu(sb->layout)) ||
2267 (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2268 (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2269 return true;
2270
2271 return false;
2272 }
2273
2274 void md_update_sb(struct mddev *mddev, int force_change)
2275 {
2276 struct md_rdev *rdev;
2277 int sync_req;
2278 int nospares = 0;
2279 int any_badblocks_changed = 0;
2280 int ret = -1;
2281
2282 if (mddev->ro) {
2283 if (force_change)
2284 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2285 return;
2286 }
2287
2288 if (mddev_is_clustered(mddev)) {
2289 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2290 force_change = 1;
2291 ret = md_cluster_ops->metadata_update_start(mddev);
2292 /* Has someone else has updated the sb */
2293 if (!does_sb_need_changing(mddev)) {
2294 if (ret == 0)
2295 md_cluster_ops->metadata_update_cancel(mddev);
2296 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2297 return;
2298 }
2299 }
2300 repeat:
2301 /* First make sure individual recovery_offsets are correct */
2302 rdev_for_each(rdev, mddev) {
2303 if (rdev->raid_disk >= 0 &&
2304 mddev->delta_disks >= 0 &&
2305 !test_bit(Journal, &rdev->flags) &&
2306 !test_bit(In_sync, &rdev->flags) &&
2307 mddev->curr_resync_completed > rdev->recovery_offset)
2308 rdev->recovery_offset = mddev->curr_resync_completed;
2309
2310 }
2311 if (!mddev->persistent) {
2312 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2313 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2314 if (!mddev->external) {
2315 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2316 rdev_for_each(rdev, mddev) {
2317 if (rdev->badblocks.changed) {
2318 rdev->badblocks.changed = 0;
2319 ack_all_badblocks(&rdev->badblocks);
2320 md_error(mddev, rdev);
2321 }
2322 clear_bit(Blocked, &rdev->flags);
2323 clear_bit(BlockedBadBlocks, &rdev->flags);
2324 wake_up(&rdev->blocked_wait);
2325 }
2326 }
2327 wake_up(&mddev->sb_wait);
2328 return;
2329 }
2330
2331 spin_lock(&mddev->lock);
2332
2333 mddev->utime = ktime_get_real_seconds();
2334
2335 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2336 force_change = 1;
2337 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2338 /* just a clean<-> dirty transition, possibly leave spares alone,
2339 * though if events isn't the right even/odd, we will have to do
2340 * spares after all
2341 */
2342 nospares = 1;
2343 if (force_change)
2344 nospares = 0;
2345 if (mddev->degraded)
2346 /* If the array is degraded, then skipping spares is both
2347 * dangerous and fairly pointless.
2348 * Dangerous because a device that was removed from the array
2349 * might have a event_count that still looks up-to-date,
2350 * so it can be re-added without a resync.
2351 * Pointless because if there are any spares to skip,
2352 * then a recovery will happen and soon that array won't
2353 * be degraded any more and the spare can go back to sleep then.
2354 */
2355 nospares = 0;
2356
2357 sync_req = mddev->in_sync;
2358
2359 /* If this is just a dirty<->clean transition, and the array is clean
2360 * and 'events' is odd, we can roll back to the previous clean state */
2361 if (nospares
2362 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2363 && mddev->can_decrease_events
2364 && mddev->events != 1) {
2365 mddev->events--;
2366 mddev->can_decrease_events = 0;
2367 } else {
2368 /* otherwise we have to go forward and ... */
2369 mddev->events ++;
2370 mddev->can_decrease_events = nospares;
2371 }
2372
2373 /*
2374 * This 64-bit counter should never wrap.
2375 * Either we are in around ~1 trillion A.C., assuming
2376 * 1 reboot per second, or we have a bug...
2377 */
2378 WARN_ON(mddev->events == 0);
2379
2380 rdev_for_each(rdev, mddev) {
2381 if (rdev->badblocks.changed)
2382 any_badblocks_changed++;
2383 if (test_bit(Faulty, &rdev->flags))
2384 set_bit(FaultRecorded, &rdev->flags);
2385 }
2386
2387 sync_sbs(mddev, nospares);
2388 spin_unlock(&mddev->lock);
2389
2390 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2391 mdname(mddev), mddev->in_sync);
2392
2393 bitmap_update_sb(mddev->bitmap);
2394 rdev_for_each(rdev, mddev) {
2395 char b[BDEVNAME_SIZE];
2396
2397 if (rdev->sb_loaded != 1)
2398 continue; /* no noise on spare devices */
2399
2400 if (!test_bit(Faulty, &rdev->flags)) {
2401 md_super_write(mddev,rdev,
2402 rdev->sb_start, rdev->sb_size,
2403 rdev->sb_page);
2404 pr_debug("md: (write) %s's sb offset: %llu\n",
2405 bdevname(rdev->bdev, b),
2406 (unsigned long long)rdev->sb_start);
2407 rdev->sb_events = mddev->events;
2408 if (rdev->badblocks.size) {
2409 md_super_write(mddev, rdev,
2410 rdev->badblocks.sector,
2411 rdev->badblocks.size << 9,
2412 rdev->bb_page);
2413 rdev->badblocks.size = 0;
2414 }
2415
2416 } else
2417 pr_debug("md: %s (skipping faulty)\n",
2418 bdevname(rdev->bdev, b));
2419
2420 if (mddev->level == LEVEL_MULTIPATH)
2421 /* only need to write one superblock... */
2422 break;
2423 }
2424 md_super_wait(mddev);
2425 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2426
2427 spin_lock(&mddev->lock);
2428 if (mddev->in_sync != sync_req ||
2429 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2430 /* have to write it out again */
2431 spin_unlock(&mddev->lock);
2432 goto repeat;
2433 }
2434 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2435 spin_unlock(&mddev->lock);
2436 wake_up(&mddev->sb_wait);
2437 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2438 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2439
2440 rdev_for_each(rdev, mddev) {
2441 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2442 clear_bit(Blocked, &rdev->flags);
2443
2444 if (any_badblocks_changed)
2445 ack_all_badblocks(&rdev->badblocks);
2446 clear_bit(BlockedBadBlocks, &rdev->flags);
2447 wake_up(&rdev->blocked_wait);
2448 }
2449
2450 if (mddev_is_clustered(mddev) && ret == 0)
2451 md_cluster_ops->metadata_update_finish(mddev);
2452 }
2453 EXPORT_SYMBOL(md_update_sb);
2454
2455 static int add_bound_rdev(struct md_rdev *rdev)
2456 {
2457 struct mddev *mddev = rdev->mddev;
2458 int err = 0;
2459 bool add_journal = test_bit(Journal, &rdev->flags);
2460
2461 if (!mddev->pers->hot_remove_disk || add_journal) {
2462 /* If there is hot_add_disk but no hot_remove_disk
2463 * then added disks for geometry changes,
2464 * and should be added immediately.
2465 */
2466 super_types[mddev->major_version].
2467 validate_super(mddev, rdev);
2468 if (add_journal)
2469 mddev_suspend(mddev);
2470 err = mddev->pers->hot_add_disk(mddev, rdev);
2471 if (add_journal)
2472 mddev_resume(mddev);
2473 if (err) {
2474 unbind_rdev_from_array(rdev);
2475 export_rdev(rdev);
2476 return err;
2477 }
2478 }
2479 sysfs_notify_dirent_safe(rdev->sysfs_state);
2480
2481 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2482 if (mddev->degraded)
2483 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2484 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2485 md_new_event(mddev);
2486 md_wakeup_thread(mddev->thread);
2487 return 0;
2488 }
2489
2490 /* words written to sysfs files may, or may not, be \n terminated.
2491 * We want to accept with case. For this we use cmd_match.
2492 */
2493 static int cmd_match(const char *cmd, const char *str)
2494 {
2495 /* See if cmd, written into a sysfs file, matches
2496 * str. They must either be the same, or cmd can
2497 * have a trailing newline
2498 */
2499 while (*cmd && *str && *cmd == *str) {
2500 cmd++;
2501 str++;
2502 }
2503 if (*cmd == '\n')
2504 cmd++;
2505 if (*str || *cmd)
2506 return 0;
2507 return 1;
2508 }
2509
2510 struct rdev_sysfs_entry {
2511 struct attribute attr;
2512 ssize_t (*show)(struct md_rdev *, char *);
2513 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2514 };
2515
2516 static ssize_t
2517 state_show(struct md_rdev *rdev, char *page)
2518 {
2519 char *sep = "";
2520 size_t len = 0;
2521 unsigned long flags = ACCESS_ONCE(rdev->flags);
2522
2523 if (test_bit(Faulty, &flags) ||
2524 rdev->badblocks.unacked_exist) {
2525 len+= sprintf(page+len, "%sfaulty",sep);
2526 sep = ",";
2527 }
2528 if (test_bit(In_sync, &flags)) {
2529 len += sprintf(page+len, "%sin_sync",sep);
2530 sep = ",";
2531 }
2532 if (test_bit(Journal, &flags)) {
2533 len += sprintf(page+len, "%sjournal",sep);
2534 sep = ",";
2535 }
2536 if (test_bit(WriteMostly, &flags)) {
2537 len += sprintf(page+len, "%swrite_mostly",sep);
2538 sep = ",";
2539 }
2540 if (test_bit(Blocked, &flags) ||
2541 (rdev->badblocks.unacked_exist
2542 && !test_bit(Faulty, &flags))) {
2543 len += sprintf(page+len, "%sblocked", sep);
2544 sep = ",";
2545 }
2546 if (!test_bit(Faulty, &flags) &&
2547 !test_bit(Journal, &flags) &&
2548 !test_bit(In_sync, &flags)) {
2549 len += sprintf(page+len, "%sspare", sep);
2550 sep = ",";
2551 }
2552 if (test_bit(WriteErrorSeen, &flags)) {
2553 len += sprintf(page+len, "%swrite_error", sep);
2554 sep = ",";
2555 }
2556 if (test_bit(WantReplacement, &flags)) {
2557 len += sprintf(page+len, "%swant_replacement", sep);
2558 sep = ",";
2559 }
2560 if (test_bit(Replacement, &flags)) {
2561 len += sprintf(page+len, "%sreplacement", sep);
2562 sep = ",";
2563 }
2564
2565 return len+sprintf(page+len, "\n");
2566 }
2567
2568 static ssize_t
2569 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2570 {
2571 /* can write
2572 * faulty - simulates an error
2573 * remove - disconnects the device
2574 * writemostly - sets write_mostly
2575 * -writemostly - clears write_mostly
2576 * blocked - sets the Blocked flags
2577 * -blocked - clears the Blocked and possibly simulates an error
2578 * insync - sets Insync providing device isn't active
2579 * -insync - clear Insync for a device with a slot assigned,
2580 * so that it gets rebuilt based on bitmap
2581 * write_error - sets WriteErrorSeen
2582 * -write_error - clears WriteErrorSeen
2583 */
2584 int err = -EINVAL;
2585 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2586 md_error(rdev->mddev, rdev);
2587 if (test_bit(Faulty, &rdev->flags))
2588 err = 0;
2589 else
2590 err = -EBUSY;
2591 } else if (cmd_match(buf, "remove")) {
2592 if (rdev->raid_disk >= 0)
2593 err = -EBUSY;
2594 else {
2595 struct mddev *mddev = rdev->mddev;
2596 err = 0;
2597 if (mddev_is_clustered(mddev))
2598 err = md_cluster_ops->remove_disk(mddev, rdev);
2599
2600 if (err == 0) {
2601 md_kick_rdev_from_array(rdev);
2602 if (mddev->pers)
2603 md_update_sb(mddev, 1);
2604 md_new_event(mddev);
2605 }
2606 }
2607 } else if (cmd_match(buf, "writemostly")) {
2608 set_bit(WriteMostly, &rdev->flags);
2609 err = 0;
2610 } else if (cmd_match(buf, "-writemostly")) {
2611 clear_bit(WriteMostly, &rdev->flags);
2612 err = 0;
2613 } else if (cmd_match(buf, "blocked")) {
2614 set_bit(Blocked, &rdev->flags);
2615 err = 0;
2616 } else if (cmd_match(buf, "-blocked")) {
2617 if (!test_bit(Faulty, &rdev->flags) &&
2618 rdev->badblocks.unacked_exist) {
2619 /* metadata handler doesn't understand badblocks,
2620 * so we need to fail the device
2621 */
2622 md_error(rdev->mddev, rdev);
2623 }
2624 clear_bit(Blocked, &rdev->flags);
2625 clear_bit(BlockedBadBlocks, &rdev->flags);
2626 wake_up(&rdev->blocked_wait);
2627 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2628 md_wakeup_thread(rdev->mddev->thread);
2629
2630 err = 0;
2631 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2632 set_bit(In_sync, &rdev->flags);
2633 err = 0;
2634 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
2635 !test_bit(Journal, &rdev->flags)) {
2636 if (rdev->mddev->pers == NULL) {
2637 clear_bit(In_sync, &rdev->flags);
2638 rdev->saved_raid_disk = rdev->raid_disk;
2639 rdev->raid_disk = -1;
2640 err = 0;
2641 }
2642 } else if (cmd_match(buf, "write_error")) {
2643 set_bit(WriteErrorSeen, &rdev->flags);
2644 err = 0;
2645 } else if (cmd_match(buf, "-write_error")) {
2646 clear_bit(WriteErrorSeen, &rdev->flags);
2647 err = 0;
2648 } else if (cmd_match(buf, "want_replacement")) {
2649 /* Any non-spare device that is not a replacement can
2650 * become want_replacement at any time, but we then need to
2651 * check if recovery is needed.
2652 */
2653 if (rdev->raid_disk >= 0 &&
2654 !test_bit(Journal, &rdev->flags) &&
2655 !test_bit(Replacement, &rdev->flags))
2656 set_bit(WantReplacement, &rdev->flags);
2657 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2658 md_wakeup_thread(rdev->mddev->thread);
2659 err = 0;
2660 } else if (cmd_match(buf, "-want_replacement")) {
2661 /* Clearing 'want_replacement' is always allowed.
2662 * Once replacements starts it is too late though.
2663 */
2664 err = 0;
2665 clear_bit(WantReplacement, &rdev->flags);
2666 } else if (cmd_match(buf, "replacement")) {
2667 /* Can only set a device as a replacement when array has not
2668 * yet been started. Once running, replacement is automatic
2669 * from spares, or by assigning 'slot'.
2670 */
2671 if (rdev->mddev->pers)
2672 err = -EBUSY;
2673 else {
2674 set_bit(Replacement, &rdev->flags);
2675 err = 0;
2676 }
2677 } else if (cmd_match(buf, "-replacement")) {
2678 /* Similarly, can only clear Replacement before start */
2679 if (rdev->mddev->pers)
2680 err = -EBUSY;
2681 else {
2682 clear_bit(Replacement, &rdev->flags);
2683 err = 0;
2684 }
2685 } else if (cmd_match(buf, "re-add")) {
2686 if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1)) {
2687 /* clear_bit is performed _after_ all the devices
2688 * have their local Faulty bit cleared. If any writes
2689 * happen in the meantime in the local node, they
2690 * will land in the local bitmap, which will be synced
2691 * by this node eventually
2692 */
2693 if (!mddev_is_clustered(rdev->mddev) ||
2694 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
2695 clear_bit(Faulty, &rdev->flags);
2696 err = add_bound_rdev(rdev);
2697 }
2698 } else
2699 err = -EBUSY;
2700 }
2701 if (!err)
2702 sysfs_notify_dirent_safe(rdev->sysfs_state);
2703 return err ? err : len;
2704 }
2705 static struct rdev_sysfs_entry rdev_state =
2706 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
2707
2708 static ssize_t
2709 errors_show(struct md_rdev *rdev, char *page)
2710 {
2711 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2712 }
2713
2714 static ssize_t
2715 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2716 {
2717 unsigned int n;
2718 int rv;
2719
2720 rv = kstrtouint(buf, 10, &n);
2721 if (rv < 0)
2722 return rv;
2723 atomic_set(&rdev->corrected_errors, n);
2724 return len;
2725 }
2726 static struct rdev_sysfs_entry rdev_errors =
2727 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2728
2729 static ssize_t
2730 slot_show(struct md_rdev *rdev, char *page)
2731 {
2732 if (test_bit(Journal, &rdev->flags))
2733 return sprintf(page, "journal\n");
2734 else if (rdev->raid_disk < 0)
2735 return sprintf(page, "none\n");
2736 else
2737 return sprintf(page, "%d\n", rdev->raid_disk);
2738 }
2739
2740 static ssize_t
2741 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2742 {
2743 int slot;
2744 int err;
2745
2746 if (test_bit(Journal, &rdev->flags))
2747 return -EBUSY;
2748 if (strncmp(buf, "none", 4)==0)
2749 slot = -1;
2750 else {
2751 err = kstrtouint(buf, 10, (unsigned int *)&slot);
2752 if (err < 0)
2753 return err;
2754 }
2755 if (rdev->mddev->pers && slot == -1) {
2756 /* Setting 'slot' on an active array requires also
2757 * updating the 'rd%d' link, and communicating
2758 * with the personality with ->hot_*_disk.
2759 * For now we only support removing
2760 * failed/spare devices. This normally happens automatically,
2761 * but not when the metadata is externally managed.
2762 */
2763 if (rdev->raid_disk == -1)
2764 return -EEXIST;
2765 /* personality does all needed checks */
2766 if (rdev->mddev->pers->hot_remove_disk == NULL)
2767 return -EINVAL;
2768 clear_bit(Blocked, &rdev->flags);
2769 remove_and_add_spares(rdev->mddev, rdev);
2770 if (rdev->raid_disk >= 0)
2771 return -EBUSY;
2772 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2773 md_wakeup_thread(rdev->mddev->thread);
2774 } else if (rdev->mddev->pers) {
2775 /* Activating a spare .. or possibly reactivating
2776 * if we ever get bitmaps working here.
2777 */
2778 int err;
2779
2780 if (rdev->raid_disk != -1)
2781 return -EBUSY;
2782
2783 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2784 return -EBUSY;
2785
2786 if (rdev->mddev->pers->hot_add_disk == NULL)
2787 return -EINVAL;
2788
2789 if (slot >= rdev->mddev->raid_disks &&
2790 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2791 return -ENOSPC;
2792
2793 rdev->raid_disk = slot;
2794 if (test_bit(In_sync, &rdev->flags))
2795 rdev->saved_raid_disk = slot;
2796 else
2797 rdev->saved_raid_disk = -1;
2798 clear_bit(In_sync, &rdev->flags);
2799 clear_bit(Bitmap_sync, &rdev->flags);
2800 err = rdev->mddev->pers->
2801 hot_add_disk(rdev->mddev, rdev);
2802 if (err) {
2803 rdev->raid_disk = -1;
2804 return err;
2805 } else
2806 sysfs_notify_dirent_safe(rdev->sysfs_state);
2807 if (sysfs_link_rdev(rdev->mddev, rdev))
2808 /* failure here is OK */;
2809 /* don't wakeup anyone, leave that to userspace. */
2810 } else {
2811 if (slot >= rdev->mddev->raid_disks &&
2812 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2813 return -ENOSPC;
2814 rdev->raid_disk = slot;
2815 /* assume it is working */
2816 clear_bit(Faulty, &rdev->flags);
2817 clear_bit(WriteMostly, &rdev->flags);
2818 set_bit(In_sync, &rdev->flags);
2819 sysfs_notify_dirent_safe(rdev->sysfs_state);
2820 }
2821 return len;
2822 }
2823
2824 static struct rdev_sysfs_entry rdev_slot =
2825 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2826
2827 static ssize_t
2828 offset_show(struct md_rdev *rdev, char *page)
2829 {
2830 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2831 }
2832
2833 static ssize_t
2834 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2835 {
2836 unsigned long long offset;
2837 if (kstrtoull(buf, 10, &offset) < 0)
2838 return -EINVAL;
2839 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2840 return -EBUSY;
2841 if (rdev->sectors && rdev->mddev->external)
2842 /* Must set offset before size, so overlap checks
2843 * can be sane */
2844 return -EBUSY;
2845 rdev->data_offset = offset;
2846 rdev->new_data_offset = offset;
2847 return len;
2848 }
2849
2850 static struct rdev_sysfs_entry rdev_offset =
2851 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2852
2853 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2854 {
2855 return sprintf(page, "%llu\n",
2856 (unsigned long long)rdev->new_data_offset);
2857 }
2858
2859 static ssize_t new_offset_store(struct md_rdev *rdev,
2860 const char *buf, size_t len)
2861 {
2862 unsigned long long new_offset;
2863 struct mddev *mddev = rdev->mddev;
2864
2865 if (kstrtoull(buf, 10, &new_offset) < 0)
2866 return -EINVAL;
2867
2868 if (mddev->sync_thread ||
2869 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
2870 return -EBUSY;
2871 if (new_offset == rdev->data_offset)
2872 /* reset is always permitted */
2873 ;
2874 else if (new_offset > rdev->data_offset) {
2875 /* must not push array size beyond rdev_sectors */
2876 if (new_offset - rdev->data_offset
2877 + mddev->dev_sectors > rdev->sectors)
2878 return -E2BIG;
2879 }
2880 /* Metadata worries about other space details. */
2881
2882 /* decreasing the offset is inconsistent with a backwards
2883 * reshape.
2884 */
2885 if (new_offset < rdev->data_offset &&
2886 mddev->reshape_backwards)
2887 return -EINVAL;
2888 /* Increasing offset is inconsistent with forwards
2889 * reshape. reshape_direction should be set to
2890 * 'backwards' first.
2891 */
2892 if (new_offset > rdev->data_offset &&
2893 !mddev->reshape_backwards)
2894 return -EINVAL;
2895
2896 if (mddev->pers && mddev->persistent &&
2897 !super_types[mddev->major_version]
2898 .allow_new_offset(rdev, new_offset))
2899 return -E2BIG;
2900 rdev->new_data_offset = new_offset;
2901 if (new_offset > rdev->data_offset)
2902 mddev->reshape_backwards = 1;
2903 else if (new_offset < rdev->data_offset)
2904 mddev->reshape_backwards = 0;
2905
2906 return len;
2907 }
2908 static struct rdev_sysfs_entry rdev_new_offset =
2909 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2910
2911 static ssize_t
2912 rdev_size_show(struct md_rdev *rdev, char *page)
2913 {
2914 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2915 }
2916
2917 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2918 {
2919 /* check if two start/length pairs overlap */
2920 if (s1+l1 <= s2)
2921 return 0;
2922 if (s2+l2 <= s1)
2923 return 0;
2924 return 1;
2925 }
2926
2927 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2928 {
2929 unsigned long long blocks;
2930 sector_t new;
2931
2932 if (kstrtoull(buf, 10, &blocks) < 0)
2933 return -EINVAL;
2934
2935 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2936 return -EINVAL; /* sector conversion overflow */
2937
2938 new = blocks * 2;
2939 if (new != blocks * 2)
2940 return -EINVAL; /* unsigned long long to sector_t overflow */
2941
2942 *sectors = new;
2943 return 0;
2944 }
2945
2946 static ssize_t
2947 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
2948 {
2949 struct mddev *my_mddev = rdev->mddev;
2950 sector_t oldsectors = rdev->sectors;
2951 sector_t sectors;
2952
2953 if (test_bit(Journal, &rdev->flags))
2954 return -EBUSY;
2955 if (strict_blocks_to_sectors(buf, &sectors) < 0)
2956 return -EINVAL;
2957 if (rdev->data_offset != rdev->new_data_offset)
2958 return -EINVAL; /* too confusing */
2959 if (my_mddev->pers && rdev->raid_disk >= 0) {
2960 if (my_mddev->persistent) {
2961 sectors = super_types[my_mddev->major_version].
2962 rdev_size_change(rdev, sectors);
2963 if (!sectors)
2964 return -EBUSY;
2965 } else if (!sectors)
2966 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2967 rdev->data_offset;
2968 if (!my_mddev->pers->resize)
2969 /* Cannot change size for RAID0 or Linear etc */
2970 return -EINVAL;
2971 }
2972 if (sectors < my_mddev->dev_sectors)
2973 return -EINVAL; /* component must fit device */
2974
2975 rdev->sectors = sectors;
2976 if (sectors > oldsectors && my_mddev->external) {
2977 /* Need to check that all other rdevs with the same
2978 * ->bdev do not overlap. 'rcu' is sufficient to walk
2979 * the rdev lists safely.
2980 * This check does not provide a hard guarantee, it
2981 * just helps avoid dangerous mistakes.
2982 */
2983 struct mddev *mddev;
2984 int overlap = 0;
2985 struct list_head *tmp;
2986
2987 rcu_read_lock();
2988 for_each_mddev(mddev, tmp) {
2989 struct md_rdev *rdev2;
2990
2991 rdev_for_each(rdev2, mddev)
2992 if (rdev->bdev == rdev2->bdev &&
2993 rdev != rdev2 &&
2994 overlaps(rdev->data_offset, rdev->sectors,
2995 rdev2->data_offset,
2996 rdev2->sectors)) {
2997 overlap = 1;
2998 break;
2999 }
3000 if (overlap) {
3001 mddev_put(mddev);
3002 break;
3003 }
3004 }
3005 rcu_read_unlock();
3006 if (overlap) {
3007 /* Someone else could have slipped in a size
3008 * change here, but doing so is just silly.
3009 * We put oldsectors back because we *know* it is
3010 * safe, and trust userspace not to race with
3011 * itself
3012 */
3013 rdev->sectors = oldsectors;
3014 return -EBUSY;
3015 }
3016 }
3017 return len;
3018 }
3019
3020 static struct rdev_sysfs_entry rdev_size =
3021 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3022
3023 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3024 {
3025 unsigned long long recovery_start = rdev->recovery_offset;
3026
3027 if (test_bit(In_sync, &rdev->flags) ||
3028 recovery_start == MaxSector)
3029 return sprintf(page, "none\n");
3030
3031 return sprintf(page, "%llu\n", recovery_start);
3032 }
3033
3034 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3035 {
3036 unsigned long long recovery_start;
3037
3038 if (cmd_match(buf, "none"))
3039 recovery_start = MaxSector;
3040 else if (kstrtoull(buf, 10, &recovery_start))
3041 return -EINVAL;
3042
3043 if (rdev->mddev->pers &&
3044 rdev->raid_disk >= 0)
3045 return -EBUSY;
3046
3047 rdev->recovery_offset = recovery_start;
3048 if (recovery_start == MaxSector)
3049 set_bit(In_sync, &rdev->flags);
3050 else
3051 clear_bit(In_sync, &rdev->flags);
3052 return len;
3053 }
3054
3055 static struct rdev_sysfs_entry rdev_recovery_start =
3056 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3057
3058 /* sysfs access to bad-blocks list.
3059 * We present two files.
3060 * 'bad-blocks' lists sector numbers and lengths of ranges that
3061 * are recorded as bad. The list is truncated to fit within
3062 * the one-page limit of sysfs.
3063 * Writing "sector length" to this file adds an acknowledged
3064 * bad block list.
3065 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3066 * been acknowledged. Writing to this file adds bad blocks
3067 * without acknowledging them. This is largely for testing.
3068 */
3069 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3070 {
3071 return badblocks_show(&rdev->badblocks, page, 0);
3072 }
3073 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3074 {
3075 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3076 /* Maybe that ack was all we needed */
3077 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3078 wake_up(&rdev->blocked_wait);
3079 return rv;
3080 }
3081 static struct rdev_sysfs_entry rdev_bad_blocks =
3082 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3083
3084 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3085 {
3086 return badblocks_show(&rdev->badblocks, page, 1);
3087 }
3088 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3089 {
3090 return badblocks_store(&rdev->badblocks, page, len, 1);
3091 }
3092 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3093 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3094
3095 static struct attribute *rdev_default_attrs[] = {
3096 &rdev_state.attr,
3097 &rdev_errors.attr,
3098 &rdev_slot.attr,
3099 &rdev_offset.attr,
3100 &rdev_new_offset.attr,
3101 &rdev_size.attr,
3102 &rdev_recovery_start.attr,
3103 &rdev_bad_blocks.attr,
3104 &rdev_unack_bad_blocks.attr,
3105 NULL,
3106 };
3107 static ssize_t
3108 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3109 {
3110 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3111 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3112
3113 if (!entry->show)
3114 return -EIO;
3115 if (!rdev->mddev)
3116 return -EBUSY;
3117 return entry->show(rdev, page);
3118 }
3119
3120 static ssize_t
3121 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3122 const char *page, size_t length)
3123 {
3124 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3125 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3126 ssize_t rv;
3127 struct mddev *mddev = rdev->mddev;
3128
3129 if (!entry->store)
3130 return -EIO;
3131 if (!capable(CAP_SYS_ADMIN))
3132 return -EACCES;
3133 rv = mddev ? mddev_lock(mddev): -EBUSY;
3134 if (!rv) {
3135 if (rdev->mddev == NULL)
3136 rv = -EBUSY;
3137 else
3138 rv = entry->store(rdev, page, length);
3139 mddev_unlock(mddev);
3140 }
3141 return rv;
3142 }
3143
3144 static void rdev_free(struct kobject *ko)
3145 {
3146 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3147 kfree(rdev);
3148 }
3149 static const struct sysfs_ops rdev_sysfs_ops = {
3150 .show = rdev_attr_show,
3151 .store = rdev_attr_store,
3152 };
3153 static struct kobj_type rdev_ktype = {
3154 .release = rdev_free,
3155 .sysfs_ops = &rdev_sysfs_ops,
3156 .default_attrs = rdev_default_attrs,
3157 };
3158
3159 int md_rdev_init(struct md_rdev *rdev)
3160 {
3161 rdev->desc_nr = -1;
3162 rdev->saved_raid_disk = -1;
3163 rdev->raid_disk = -1;
3164 rdev->flags = 0;
3165 rdev->data_offset = 0;
3166 rdev->new_data_offset = 0;
3167 rdev->sb_events = 0;
3168 rdev->last_read_error.tv_sec = 0;
3169 rdev->last_read_error.tv_nsec = 0;
3170 rdev->sb_loaded = 0;
3171 rdev->bb_page = NULL;
3172 atomic_set(&rdev->nr_pending, 0);
3173 atomic_set(&rdev->read_errors, 0);
3174 atomic_set(&rdev->corrected_errors, 0);
3175
3176 INIT_LIST_HEAD(&rdev->same_set);
3177 init_waitqueue_head(&rdev->blocked_wait);
3178
3179 /* Add space to store bad block list.
3180 * This reserves the space even on arrays where it cannot
3181 * be used - I wonder if that matters
3182 */
3183 return badblocks_init(&rdev->badblocks, 0);
3184 }
3185 EXPORT_SYMBOL_GPL(md_rdev_init);
3186 /*
3187 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3188 *
3189 * mark the device faulty if:
3190 *
3191 * - the device is nonexistent (zero size)
3192 * - the device has no valid superblock
3193 *
3194 * a faulty rdev _never_ has rdev->sb set.
3195 */
3196 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3197 {
3198 char b[BDEVNAME_SIZE];
3199 int err;
3200 struct md_rdev *rdev;
3201 sector_t size;
3202
3203 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3204 if (!rdev) {
3205 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3206 return ERR_PTR(-ENOMEM);
3207 }
3208
3209 err = md_rdev_init(rdev);
3210 if (err)
3211 goto abort_free;
3212 err = alloc_disk_sb(rdev);
3213 if (err)
3214 goto abort_free;
3215
3216 err = lock_rdev(rdev, newdev, super_format == -2);
3217 if (err)
3218 goto abort_free;
3219
3220 kobject_init(&rdev->kobj, &rdev_ktype);
3221
3222 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3223 if (!size) {
3224 printk(KERN_WARNING
3225 "md: %s has zero or unknown size, marking faulty!\n",
3226 bdevname(rdev->bdev,b));
3227 err = -EINVAL;
3228 goto abort_free;
3229 }
3230
3231 if (super_format >= 0) {
3232 err = super_types[super_format].
3233 load_super(rdev, NULL, super_minor);
3234 if (err == -EINVAL) {
3235 printk(KERN_WARNING
3236 "md: %s does not have a valid v%d.%d "
3237 "superblock, not importing!\n",
3238 bdevname(rdev->bdev,b),
3239 super_format, super_minor);
3240 goto abort_free;
3241 }
3242 if (err < 0) {
3243 printk(KERN_WARNING
3244 "md: could not read %s's sb, not importing!\n",
3245 bdevname(rdev->bdev,b));
3246 goto abort_free;
3247 }
3248 }
3249
3250 return rdev;
3251
3252 abort_free:
3253 if (rdev->bdev)
3254 unlock_rdev(rdev);
3255 md_rdev_clear(rdev);
3256 kfree(rdev);
3257 return ERR_PTR(err);
3258 }
3259
3260 /*
3261 * Check a full RAID array for plausibility
3262 */
3263
3264 static void analyze_sbs(struct mddev *mddev)
3265 {
3266 int i;
3267 struct md_rdev *rdev, *freshest, *tmp;
3268 char b[BDEVNAME_SIZE];
3269
3270 freshest = NULL;
3271 rdev_for_each_safe(rdev, tmp, mddev)
3272 switch (super_types[mddev->major_version].
3273 load_super(rdev, freshest, mddev->minor_version)) {
3274 case 1:
3275 freshest = rdev;
3276 break;
3277 case 0:
3278 break;
3279 default:
3280 printk( KERN_ERR \
3281 "md: fatal superblock inconsistency in %s"
3282 " -- removing from array\n",
3283 bdevname(rdev->bdev,b));
3284 md_kick_rdev_from_array(rdev);
3285 }
3286
3287 super_types[mddev->major_version].
3288 validate_super(mddev, freshest);
3289
3290 i = 0;
3291 rdev_for_each_safe(rdev, tmp, mddev) {
3292 if (mddev->max_disks &&
3293 (rdev->desc_nr >= mddev->max_disks ||
3294 i > mddev->max_disks)) {
3295 printk(KERN_WARNING
3296 "md: %s: %s: only %d devices permitted\n",
3297 mdname(mddev), bdevname(rdev->bdev, b),
3298 mddev->max_disks);
3299 md_kick_rdev_from_array(rdev);
3300 continue;
3301 }
3302 if (rdev != freshest) {
3303 if (super_types[mddev->major_version].
3304 validate_super(mddev, rdev)) {
3305 printk(KERN_WARNING "md: kicking non-fresh %s"
3306 " from array!\n",
3307 bdevname(rdev->bdev,b));
3308 md_kick_rdev_from_array(rdev);
3309 continue;
3310 }
3311 }
3312 if (mddev->level == LEVEL_MULTIPATH) {
3313 rdev->desc_nr = i++;
3314 rdev->raid_disk = rdev->desc_nr;
3315 set_bit(In_sync, &rdev->flags);
3316 } else if (rdev->raid_disk >=
3317 (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3318 !test_bit(Journal, &rdev->flags)) {
3319 rdev->raid_disk = -1;
3320 clear_bit(In_sync, &rdev->flags);
3321 }
3322 }
3323 }
3324
3325 /* Read a fixed-point number.
3326 * Numbers in sysfs attributes should be in "standard" units where
3327 * possible, so time should be in seconds.
3328 * However we internally use a a much smaller unit such as
3329 * milliseconds or jiffies.
3330 * This function takes a decimal number with a possible fractional
3331 * component, and produces an integer which is the result of
3332 * multiplying that number by 10^'scale'.
3333 * all without any floating-point arithmetic.
3334 */
3335 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3336 {
3337 unsigned long result = 0;
3338 long decimals = -1;
3339 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3340 if (*cp == '.')
3341 decimals = 0;
3342 else if (decimals < scale) {
3343 unsigned int value;
3344 value = *cp - '0';
3345 result = result * 10 + value;
3346 if (decimals >= 0)
3347 decimals++;
3348 }
3349 cp++;
3350 }
3351 if (*cp == '\n')
3352 cp++;
3353 if (*cp)
3354 return -EINVAL;
3355 if (decimals < 0)
3356 decimals = 0;
3357 while (decimals < scale) {
3358 result *= 10;
3359 decimals ++;
3360 }
3361 *res = result;
3362 return 0;
3363 }
3364
3365 static ssize_t
3366 safe_delay_show(struct mddev *mddev, char *page)
3367 {
3368 int msec = (mddev->safemode_delay*1000)/HZ;
3369 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3370 }
3371 static ssize_t
3372 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3373 {
3374 unsigned long msec;
3375
3376 if (mddev_is_clustered(mddev)) {
3377 pr_info("md: Safemode is disabled for clustered mode\n");
3378 return -EINVAL;
3379 }
3380
3381 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3382 return -EINVAL;
3383 if (msec == 0)
3384 mddev->safemode_delay = 0;
3385 else {
3386 unsigned long old_delay = mddev->safemode_delay;
3387 unsigned long new_delay = (msec*HZ)/1000;
3388
3389 if (new_delay == 0)
3390 new_delay = 1;
3391 mddev->safemode_delay = new_delay;
3392 if (new_delay < old_delay || old_delay == 0)
3393 mod_timer(&mddev->safemode_timer, jiffies+1);
3394 }
3395 return len;
3396 }
3397 static struct md_sysfs_entry md_safe_delay =
3398 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3399
3400 static ssize_t
3401 level_show(struct mddev *mddev, char *page)
3402 {
3403 struct md_personality *p;
3404 int ret;
3405 spin_lock(&mddev->lock);
3406 p = mddev->pers;
3407 if (p)
3408 ret = sprintf(page, "%s\n", p->name);
3409 else if (mddev->clevel[0])
3410 ret = sprintf(page, "%s\n", mddev->clevel);
3411 else if (mddev->level != LEVEL_NONE)
3412 ret = sprintf(page, "%d\n", mddev->level);
3413 else
3414 ret = 0;
3415 spin_unlock(&mddev->lock);
3416 return ret;
3417 }
3418
3419 static ssize_t
3420 level_store(struct mddev *mddev, const char *buf, size_t len)
3421 {
3422 char clevel[16];
3423 ssize_t rv;
3424 size_t slen = len;
3425 struct md_personality *pers, *oldpers;
3426 long level;
3427 void *priv, *oldpriv;
3428 struct md_rdev *rdev;
3429
3430 if (slen == 0 || slen >= sizeof(clevel))
3431 return -EINVAL;
3432
3433 rv = mddev_lock(mddev);
3434 if (rv)
3435 return rv;
3436
3437 if (mddev->pers == NULL) {
3438 strncpy(mddev->clevel, buf, slen);
3439 if (mddev->clevel[slen-1] == '\n')
3440 slen--;
3441 mddev->clevel[slen] = 0;
3442 mddev->level = LEVEL_NONE;
3443 rv = len;
3444 goto out_unlock;
3445 }
3446 rv = -EROFS;
3447 if (mddev->ro)
3448 goto out_unlock;
3449
3450 /* request to change the personality. Need to ensure:
3451 * - array is not engaged in resync/recovery/reshape
3452 * - old personality can be suspended
3453 * - new personality will access other array.
3454 */
3455
3456 rv = -EBUSY;
3457 if (mddev->sync_thread ||
3458 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3459 mddev->reshape_position != MaxSector ||
3460 mddev->sysfs_active)
3461 goto out_unlock;
3462
3463 rv = -EINVAL;
3464 if (!mddev->pers->quiesce) {
3465 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3466 mdname(mddev), mddev->pers->name);
3467 goto out_unlock;
3468 }
3469
3470 /* Now find the new personality */
3471 strncpy(clevel, buf, slen);
3472 if (clevel[slen-1] == '\n')
3473 slen--;
3474 clevel[slen] = 0;
3475 if (kstrtol(clevel, 10, &level))
3476 level = LEVEL_NONE;
3477
3478 if (request_module("md-%s", clevel) != 0)
3479 request_module("md-level-%s", clevel);
3480 spin_lock(&pers_lock);
3481 pers = find_pers(level, clevel);
3482 if (!pers || !try_module_get(pers->owner)) {
3483 spin_unlock(&pers_lock);
3484 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3485 rv = -EINVAL;
3486 goto out_unlock;
3487 }
3488 spin_unlock(&pers_lock);
3489
3490 if (pers == mddev->pers) {
3491 /* Nothing to do! */
3492 module_put(pers->owner);
3493 rv = len;
3494 goto out_unlock;
3495 }
3496 if (!pers->takeover) {
3497 module_put(pers->owner);
3498 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3499 mdname(mddev), clevel);
3500 rv = -EINVAL;
3501 goto out_unlock;
3502 }
3503
3504 rdev_for_each(rdev, mddev)
3505 rdev->new_raid_disk = rdev->raid_disk;
3506
3507 /* ->takeover must set new_* and/or delta_disks
3508 * if it succeeds, and may set them when it fails.
3509 */
3510 priv = pers->takeover(mddev);
3511 if (IS_ERR(priv)) {
3512 mddev->new_level = mddev->level;
3513 mddev->new_layout = mddev->layout;
3514 mddev->new_chunk_sectors = mddev->chunk_sectors;
3515 mddev->raid_disks -= mddev->delta_disks;
3516 mddev->delta_disks = 0;
3517 mddev->reshape_backwards = 0;
3518 module_put(pers->owner);
3519 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3520 mdname(mddev), clevel);
3521 rv = PTR_ERR(priv);
3522 goto out_unlock;
3523 }
3524
3525 /* Looks like we have a winner */
3526 mddev_suspend(mddev);
3527 mddev_detach(mddev);
3528
3529 spin_lock(&mddev->lock);
3530 oldpers = mddev->pers;
3531 oldpriv = mddev->private;
3532 mddev->pers = pers;
3533 mddev->private = priv;
3534 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3535 mddev->level = mddev->new_level;
3536 mddev->layout = mddev->new_layout;
3537 mddev->chunk_sectors = mddev->new_chunk_sectors;
3538 mddev->delta_disks = 0;
3539 mddev->reshape_backwards = 0;
3540 mddev->degraded = 0;
3541 spin_unlock(&mddev->lock);
3542
3543 if (oldpers->sync_request == NULL &&
3544 mddev->external) {
3545 /* We are converting from a no-redundancy array
3546 * to a redundancy array and metadata is managed
3547 * externally so we need to be sure that writes
3548 * won't block due to a need to transition
3549 * clean->dirty
3550 * until external management is started.
3551 */
3552 mddev->in_sync = 0;
3553 mddev->safemode_delay = 0;
3554 mddev->safemode = 0;
3555 }
3556
3557 oldpers->free(mddev, oldpriv);
3558
3559 if (oldpers->sync_request == NULL &&
3560 pers->sync_request != NULL) {
3561 /* need to add the md_redundancy_group */
3562 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3563 printk(KERN_WARNING
3564 "md: cannot register extra attributes for %s\n",
3565 mdname(mddev));
3566 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3567 }
3568 if (oldpers->sync_request != NULL &&
3569 pers->sync_request == NULL) {
3570 /* need to remove the md_redundancy_group */
3571 if (mddev->to_remove == NULL)
3572 mddev->to_remove = &md_redundancy_group;
3573 }
3574
3575 rdev_for_each(rdev, mddev) {
3576 if (rdev->raid_disk < 0)
3577 continue;
3578 if (rdev->new_raid_disk >= mddev->raid_disks)
3579 rdev->new_raid_disk = -1;
3580 if (rdev->new_raid_disk == rdev->raid_disk)
3581 continue;
3582 sysfs_unlink_rdev(mddev, rdev);
3583 }
3584 rdev_for_each(rdev, mddev) {
3585 if (rdev->raid_disk < 0)
3586 continue;
3587 if (rdev->new_raid_disk == rdev->raid_disk)
3588 continue;
3589 rdev->raid_disk = rdev->new_raid_disk;
3590 if (rdev->raid_disk < 0)
3591 clear_bit(In_sync, &rdev->flags);
3592 else {
3593 if (sysfs_link_rdev(mddev, rdev))
3594 printk(KERN_WARNING "md: cannot register rd%d"
3595 " for %s after level change\n",
3596 rdev->raid_disk, mdname(mddev));
3597 }
3598 }
3599
3600 if (pers->sync_request == NULL) {
3601 /* this is now an array without redundancy, so
3602 * it must always be in_sync
3603 */
3604 mddev->in_sync = 1;
3605 del_timer_sync(&mddev->safemode_timer);
3606 }
3607 blk_set_stacking_limits(&mddev->queue->limits);
3608 pers->run(mddev);
3609 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3610 mddev_resume(mddev);
3611 if (!mddev->thread)
3612 md_update_sb(mddev, 1);
3613 sysfs_notify(&mddev->kobj, NULL, "level");
3614 md_new_event(mddev);
3615 rv = len;
3616 out_unlock:
3617 mddev_unlock(mddev);
3618 return rv;
3619 }
3620
3621 static struct md_sysfs_entry md_level =
3622 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3623
3624 static ssize_t
3625 layout_show(struct mddev *mddev, char *page)
3626 {
3627 /* just a number, not meaningful for all levels */
3628 if (mddev->reshape_position != MaxSector &&
3629 mddev->layout != mddev->new_layout)
3630 return sprintf(page, "%d (%d)\n",
3631 mddev->new_layout, mddev->layout);
3632 return sprintf(page, "%d\n", mddev->layout);
3633 }
3634
3635 static ssize_t
3636 layout_store(struct mddev *mddev, const char *buf, size_t len)
3637 {
3638 unsigned int n;
3639 int err;
3640
3641 err = kstrtouint(buf, 10, &n);
3642 if (err < 0)
3643 return err;
3644 err = mddev_lock(mddev);
3645 if (err)
3646 return err;
3647
3648 if (mddev->pers) {
3649 if (mddev->pers->check_reshape == NULL)
3650 err = -EBUSY;
3651 else if (mddev->ro)
3652 err = -EROFS;
3653 else {
3654 mddev->new_layout = n;
3655 err = mddev->pers->check_reshape(mddev);
3656 if (err)
3657 mddev->new_layout = mddev->layout;
3658 }
3659 } else {
3660 mddev->new_layout = n;
3661 if (mddev->reshape_position == MaxSector)
3662 mddev->layout = n;
3663 }
3664 mddev_unlock(mddev);
3665 return err ?: len;
3666 }
3667 static struct md_sysfs_entry md_layout =
3668 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3669
3670 static ssize_t
3671 raid_disks_show(struct mddev *mddev, char *page)
3672 {
3673 if (mddev->raid_disks == 0)
3674 return 0;
3675 if (mddev->reshape_position != MaxSector &&
3676 mddev->delta_disks != 0)
3677 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3678 mddev->raid_disks - mddev->delta_disks);
3679 return sprintf(page, "%d\n", mddev->raid_disks);
3680 }
3681
3682 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3683
3684 static ssize_t
3685 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3686 {
3687 unsigned int n;
3688 int err;
3689
3690 err = kstrtouint(buf, 10, &n);
3691 if (err < 0)
3692 return err;
3693
3694 err = mddev_lock(mddev);
3695 if (err)
3696 return err;
3697 if (mddev->pers)
3698 err = update_raid_disks(mddev, n);
3699 else if (mddev->reshape_position != MaxSector) {
3700 struct md_rdev *rdev;
3701 int olddisks = mddev->raid_disks - mddev->delta_disks;
3702
3703 err = -EINVAL;
3704 rdev_for_each(rdev, mddev) {
3705 if (olddisks < n &&
3706 rdev->data_offset < rdev->new_data_offset)
3707 goto out_unlock;
3708 if (olddisks > n &&
3709 rdev->data_offset > rdev->new_data_offset)
3710 goto out_unlock;
3711 }
3712 err = 0;
3713 mddev->delta_disks = n - olddisks;
3714 mddev->raid_disks = n;
3715 mddev->reshape_backwards = (mddev->delta_disks < 0);
3716 } else
3717 mddev->raid_disks = n;
3718 out_unlock:
3719 mddev_unlock(mddev);
3720 return err ? err : len;
3721 }
3722 static struct md_sysfs_entry md_raid_disks =
3723 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3724
3725 static ssize_t
3726 chunk_size_show(struct mddev *mddev, char *page)
3727 {
3728 if (mddev->reshape_position != MaxSector &&
3729 mddev->chunk_sectors != mddev->new_chunk_sectors)
3730 return sprintf(page, "%d (%d)\n",
3731 mddev->new_chunk_sectors << 9,
3732 mddev->chunk_sectors << 9);
3733 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3734 }
3735
3736 static ssize_t
3737 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3738 {
3739 unsigned long n;
3740 int err;
3741
3742 err = kstrtoul(buf, 10, &n);
3743 if (err < 0)
3744 return err;
3745
3746 err = mddev_lock(mddev);
3747 if (err)
3748 return err;
3749 if (mddev->pers) {
3750 if (mddev->pers->check_reshape == NULL)
3751 err = -EBUSY;
3752 else if (mddev->ro)
3753 err = -EROFS;
3754 else {
3755 mddev->new_chunk_sectors = n >> 9;
3756 err = mddev->pers->check_reshape(mddev);
3757 if (err)
3758 mddev->new_chunk_sectors = mddev->chunk_sectors;
3759 }
3760 } else {
3761 mddev->new_chunk_sectors = n >> 9;
3762 if (mddev->reshape_position == MaxSector)
3763 mddev->chunk_sectors = n >> 9;
3764 }
3765 mddev_unlock(mddev);
3766 return err ?: len;
3767 }
3768 static struct md_sysfs_entry md_chunk_size =
3769 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3770
3771 static ssize_t
3772 resync_start_show(struct mddev *mddev, char *page)
3773 {
3774 if (mddev->recovery_cp == MaxSector)
3775 return sprintf(page, "none\n");
3776 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3777 }
3778
3779 static ssize_t
3780 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3781 {
3782 unsigned long long n;
3783 int err;
3784
3785 if (cmd_match(buf, "none"))
3786 n = MaxSector;
3787 else {
3788 err = kstrtoull(buf, 10, &n);
3789 if (err < 0)
3790 return err;
3791 if (n != (sector_t)n)
3792 return -EINVAL;
3793 }
3794
3795 err = mddev_lock(mddev);
3796 if (err)
3797 return err;
3798 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3799 err = -EBUSY;
3800
3801 if (!err) {
3802 mddev->recovery_cp = n;
3803 if (mddev->pers)
3804 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3805 }
3806 mddev_unlock(mddev);
3807 return err ?: len;
3808 }
3809 static struct md_sysfs_entry md_resync_start =
3810 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
3811 resync_start_show, resync_start_store);
3812
3813 /*
3814 * The array state can be:
3815 *
3816 * clear
3817 * No devices, no size, no level
3818 * Equivalent to STOP_ARRAY ioctl
3819 * inactive
3820 * May have some settings, but array is not active
3821 * all IO results in error
3822 * When written, doesn't tear down array, but just stops it
3823 * suspended (not supported yet)
3824 * All IO requests will block. The array can be reconfigured.
3825 * Writing this, if accepted, will block until array is quiescent
3826 * readonly
3827 * no resync can happen. no superblocks get written.
3828 * write requests fail
3829 * read-auto
3830 * like readonly, but behaves like 'clean' on a write request.
3831 *
3832 * clean - no pending writes, but otherwise active.
3833 * When written to inactive array, starts without resync
3834 * If a write request arrives then
3835 * if metadata is known, mark 'dirty' and switch to 'active'.
3836 * if not known, block and switch to write-pending
3837 * If written to an active array that has pending writes, then fails.
3838 * active
3839 * fully active: IO and resync can be happening.
3840 * When written to inactive array, starts with resync
3841 *
3842 * write-pending
3843 * clean, but writes are blocked waiting for 'active' to be written.
3844 *
3845 * active-idle
3846 * like active, but no writes have been seen for a while (100msec).
3847 *
3848 */
3849 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3850 write_pending, active_idle, bad_word};
3851 static char *array_states[] = {
3852 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3853 "write-pending", "active-idle", NULL };
3854
3855 static int match_word(const char *word, char **list)
3856 {
3857 int n;
3858 for (n=0; list[n]; n++)
3859 if (cmd_match(word, list[n]))
3860 break;
3861 return n;
3862 }
3863
3864 static ssize_t
3865 array_state_show(struct mddev *mddev, char *page)
3866 {
3867 enum array_state st = inactive;
3868
3869 if (mddev->pers)
3870 switch(mddev->ro) {
3871 case 1:
3872 st = readonly;
3873 break;
3874 case 2:
3875 st = read_auto;
3876 break;
3877 case 0:
3878 if (mddev->in_sync)
3879 st = clean;
3880 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3881 st = write_pending;
3882 else if (mddev->safemode)
3883 st = active_idle;
3884 else
3885 st = active;
3886 }
3887 else {
3888 if (list_empty(&mddev->disks) &&
3889 mddev->raid_disks == 0 &&
3890 mddev->dev_sectors == 0)
3891 st = clear;
3892 else
3893 st = inactive;
3894 }
3895 return sprintf(page, "%s\n", array_states[st]);
3896 }
3897
3898 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
3899 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
3900 static int do_md_run(struct mddev *mddev);
3901 static int restart_array(struct mddev *mddev);
3902
3903 static ssize_t
3904 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3905 {
3906 int err;
3907 enum array_state st = match_word(buf, array_states);
3908
3909 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
3910 /* don't take reconfig_mutex when toggling between
3911 * clean and active
3912 */
3913 spin_lock(&mddev->lock);
3914 if (st == active) {
3915 restart_array(mddev);
3916 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3917 wake_up(&mddev->sb_wait);
3918 err = 0;
3919 } else /* st == clean */ {
3920 restart_array(mddev);
3921 if (atomic_read(&mddev->writes_pending) == 0) {
3922 if (mddev->in_sync == 0) {
3923 mddev->in_sync = 1;
3924 if (mddev->safemode == 1)
3925 mddev->safemode = 0;
3926 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3927 }
3928 err = 0;
3929 } else
3930 err = -EBUSY;
3931 }
3932 spin_unlock(&mddev->lock);
3933 return err ?: len;
3934 }
3935 err = mddev_lock(mddev);
3936 if (err)
3937 return err;
3938 err = -EINVAL;
3939 switch(st) {
3940 case bad_word:
3941 break;
3942 case clear:
3943 /* stopping an active array */
3944 err = do_md_stop(mddev, 0, NULL);
3945 break;
3946 case inactive:
3947 /* stopping an active array */
3948 if (mddev->pers)
3949 err = do_md_stop(mddev, 2, NULL);
3950 else
3951 err = 0; /* already inactive */
3952 break;
3953 case suspended:
3954 break; /* not supported yet */
3955 case readonly:
3956 if (mddev->pers)
3957 err = md_set_readonly(mddev, NULL);
3958 else {
3959 mddev->ro = 1;
3960 set_disk_ro(mddev->gendisk, 1);
3961 err = do_md_run(mddev);
3962 }
3963 break;
3964 case read_auto:
3965 if (mddev->pers) {
3966 if (mddev->ro == 0)
3967 err = md_set_readonly(mddev, NULL);
3968 else if (mddev->ro == 1)
3969 err = restart_array(mddev);
3970 if (err == 0) {
3971 mddev->ro = 2;
3972 set_disk_ro(mddev->gendisk, 0);
3973 }
3974 } else {
3975 mddev->ro = 2;
3976 err = do_md_run(mddev);
3977 }
3978 break;
3979 case clean:
3980 if (mddev->pers) {
3981 err = restart_array(mddev);
3982 if (err)
3983 break;
3984 spin_lock(&mddev->lock);
3985 if (atomic_read(&mddev->writes_pending) == 0) {
3986 if (mddev->in_sync == 0) {
3987 mddev->in_sync = 1;
3988 if (mddev->safemode == 1)
3989 mddev->safemode = 0;
3990 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3991 }
3992 err = 0;
3993 } else
3994 err = -EBUSY;
3995 spin_unlock(&mddev->lock);
3996 } else
3997 err = -EINVAL;
3998 break;
3999 case active:
4000 if (mddev->pers) {
4001 err = restart_array(mddev);
4002 if (err)
4003 break;
4004 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
4005 wake_up(&mddev->sb_wait);
4006 err = 0;
4007 } else {
4008 mddev->ro = 0;
4009 set_disk_ro(mddev->gendisk, 0);
4010 err = do_md_run(mddev);
4011 }
4012 break;
4013 case write_pending:
4014 case active_idle:
4015 /* these cannot be set */
4016 break;
4017 }
4018
4019 if (!err) {
4020 if (mddev->hold_active == UNTIL_IOCTL)
4021 mddev->hold_active = 0;
4022 sysfs_notify_dirent_safe(mddev->sysfs_state);
4023 }
4024 mddev_unlock(mddev);
4025 return err ?: len;
4026 }
4027 static struct md_sysfs_entry md_array_state =
4028 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4029
4030 static ssize_t
4031 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4032 return sprintf(page, "%d\n",
4033 atomic_read(&mddev->max_corr_read_errors));
4034 }
4035
4036 static ssize_t
4037 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4038 {
4039 unsigned int n;
4040 int rv;
4041
4042 rv = kstrtouint(buf, 10, &n);
4043 if (rv < 0)
4044 return rv;
4045 atomic_set(&mddev->max_corr_read_errors, n);
4046 return len;
4047 }
4048
4049 static struct md_sysfs_entry max_corr_read_errors =
4050 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4051 max_corrected_read_errors_store);
4052
4053 static ssize_t
4054 null_show(struct mddev *mddev, char *page)
4055 {
4056 return -EINVAL;
4057 }
4058
4059 static ssize_t
4060 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4061 {
4062 /* buf must be %d:%d\n? giving major and minor numbers */
4063 /* The new device is added to the array.
4064 * If the array has a persistent superblock, we read the
4065 * superblock to initialise info and check validity.
4066 * Otherwise, only checking done is that in bind_rdev_to_array,
4067 * which mainly checks size.
4068 */
4069 char *e;
4070 int major = simple_strtoul(buf, &e, 10);
4071 int minor;
4072 dev_t dev;
4073 struct md_rdev *rdev;
4074 int err;
4075
4076 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4077 return -EINVAL;
4078 minor = simple_strtoul(e+1, &e, 10);
4079 if (*e && *e != '\n')
4080 return -EINVAL;
4081 dev = MKDEV(major, minor);
4082 if (major != MAJOR(dev) ||
4083 minor != MINOR(dev))
4084 return -EOVERFLOW;
4085
4086 flush_workqueue(md_misc_wq);
4087
4088 err = mddev_lock(mddev);
4089 if (err)
4090 return err;
4091 if (mddev->persistent) {
4092 rdev = md_import_device(dev, mddev->major_version,
4093 mddev->minor_version);
4094 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4095 struct md_rdev *rdev0
4096 = list_entry(mddev->disks.next,
4097 struct md_rdev, same_set);
4098 err = super_types[mddev->major_version]
4099 .load_super(rdev, rdev0, mddev->minor_version);
4100 if (err < 0)
4101 goto out;
4102 }
4103 } else if (mddev->external)
4104 rdev = md_import_device(dev, -2, -1);
4105 else
4106 rdev = md_import_device(dev, -1, -1);
4107
4108 if (IS_ERR(rdev)) {
4109 mddev_unlock(mddev);
4110 return PTR_ERR(rdev);
4111 }
4112 err = bind_rdev_to_array(rdev, mddev);
4113 out:
4114 if (err)
4115 export_rdev(rdev);
4116 mddev_unlock(mddev);
4117 return err ? err : len;
4118 }
4119
4120 static struct md_sysfs_entry md_new_device =
4121 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4122
4123 static ssize_t
4124 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4125 {
4126 char *end;
4127 unsigned long chunk, end_chunk;
4128 int err;
4129
4130 err = mddev_lock(mddev);
4131 if (err)
4132 return err;
4133 if (!mddev->bitmap)
4134 goto out;
4135 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4136 while (*buf) {
4137 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4138 if (buf == end) break;
4139 if (*end == '-') { /* range */
4140 buf = end + 1;
4141 end_chunk = simple_strtoul(buf, &end, 0);
4142 if (buf == end) break;
4143 }
4144 if (*end && !isspace(*end)) break;
4145 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4146 buf = skip_spaces(end);
4147 }
4148 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4149 out:
4150 mddev_unlock(mddev);
4151 return len;
4152 }
4153
4154 static struct md_sysfs_entry md_bitmap =
4155 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4156
4157 static ssize_t
4158 size_show(struct mddev *mddev, char *page)
4159 {
4160 return sprintf(page, "%llu\n",
4161 (unsigned long long)mddev->dev_sectors / 2);
4162 }
4163
4164 static int update_size(struct mddev *mddev, sector_t num_sectors);
4165
4166 static ssize_t
4167 size_store(struct mddev *mddev, const char *buf, size_t len)
4168 {
4169 /* If array is inactive, we can reduce the component size, but
4170 * not increase it (except from 0).
4171 * If array is active, we can try an on-line resize
4172 */
4173 sector_t sectors;
4174 int err = strict_blocks_to_sectors(buf, &sectors);
4175
4176 if (err < 0)
4177 return err;
4178 err = mddev_lock(mddev);
4179 if (err)
4180 return err;
4181 if (mddev->pers) {
4182 err = update_size(mddev, sectors);
4183 md_update_sb(mddev, 1);
4184 } else {
4185 if (mddev->dev_sectors == 0 ||
4186 mddev->dev_sectors > sectors)
4187 mddev->dev_sectors = sectors;
4188 else
4189 err = -ENOSPC;
4190 }
4191 mddev_unlock(mddev);
4192 return err ? err : len;
4193 }
4194
4195 static struct md_sysfs_entry md_size =
4196 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4197
4198 /* Metadata version.
4199 * This is one of
4200 * 'none' for arrays with no metadata (good luck...)
4201 * 'external' for arrays with externally managed metadata,
4202 * or N.M for internally known formats
4203 */
4204 static ssize_t
4205 metadata_show(struct mddev *mddev, char *page)
4206 {
4207 if (mddev->persistent)
4208 return sprintf(page, "%d.%d\n",
4209 mddev->major_version, mddev->minor_version);
4210 else if (mddev->external)
4211 return sprintf(page, "external:%s\n", mddev->metadata_type);
4212 else
4213 return sprintf(page, "none\n");
4214 }
4215
4216 static ssize_t
4217 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4218 {
4219 int major, minor;
4220 char *e;
4221 int err;
4222 /* Changing the details of 'external' metadata is
4223 * always permitted. Otherwise there must be
4224 * no devices attached to the array.
4225 */
4226
4227 err = mddev_lock(mddev);
4228 if (err)
4229 return err;
4230 err = -EBUSY;
4231 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4232 ;
4233 else if (!list_empty(&mddev->disks))
4234 goto out_unlock;
4235
4236 err = 0;
4237 if (cmd_match(buf, "none")) {
4238 mddev->persistent = 0;
4239 mddev->external = 0;
4240 mddev->major_version = 0;
4241 mddev->minor_version = 90;
4242 goto out_unlock;
4243 }
4244 if (strncmp(buf, "external:", 9) == 0) {
4245 size_t namelen = len-9;
4246 if (namelen >= sizeof(mddev->metadata_type))
4247 namelen = sizeof(mddev->metadata_type)-1;
4248 strncpy(mddev->metadata_type, buf+9, namelen);
4249 mddev->metadata_type[namelen] = 0;
4250 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4251 mddev->metadata_type[--namelen] = 0;
4252 mddev->persistent = 0;
4253 mddev->external = 1;
4254 mddev->major_version = 0;
4255 mddev->minor_version = 90;
4256 goto out_unlock;
4257 }
4258 major = simple_strtoul(buf, &e, 10);
4259 err = -EINVAL;
4260 if (e==buf || *e != '.')
4261 goto out_unlock;
4262 buf = e+1;
4263 minor = simple_strtoul(buf, &e, 10);
4264 if (e==buf || (*e && *e != '\n') )
4265 goto out_unlock;
4266 err = -ENOENT;
4267 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4268 goto out_unlock;
4269 mddev->major_version = major;
4270 mddev->minor_version = minor;
4271 mddev->persistent = 1;
4272 mddev->external = 0;
4273 err = 0;
4274 out_unlock:
4275 mddev_unlock(mddev);
4276 return err ?: len;
4277 }
4278
4279 static struct md_sysfs_entry md_metadata =
4280 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4281
4282 static ssize_t
4283 action_show(struct mddev *mddev, char *page)
4284 {
4285 char *type = "idle";
4286 unsigned long recovery = mddev->recovery;
4287 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4288 type = "frozen";
4289 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4290 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4291 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4292 type = "reshape";
4293 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4294 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4295 type = "resync";
4296 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4297 type = "check";
4298 else
4299 type = "repair";
4300 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4301 type = "recover";
4302 else if (mddev->reshape_position != MaxSector)
4303 type = "reshape";
4304 }
4305 return sprintf(page, "%s\n", type);
4306 }
4307
4308 static ssize_t
4309 action_store(struct mddev *mddev, const char *page, size_t len)
4310 {
4311 if (!mddev->pers || !mddev->pers->sync_request)
4312 return -EINVAL;
4313
4314
4315 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4316 if (cmd_match(page, "frozen"))
4317 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4318 else
4319 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4320 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4321 mddev_lock(mddev) == 0) {
4322 flush_workqueue(md_misc_wq);
4323 if (mddev->sync_thread) {
4324 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4325 md_reap_sync_thread(mddev);
4326 }
4327 mddev_unlock(mddev);
4328 }
4329 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4330 return -EBUSY;
4331 else if (cmd_match(page, "resync"))
4332 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4333 else if (cmd_match(page, "recover")) {
4334 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4335 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4336 } else if (cmd_match(page, "reshape")) {
4337 int err;
4338 if (mddev->pers->start_reshape == NULL)
4339 return -EINVAL;
4340 err = mddev_lock(mddev);
4341 if (!err) {
4342 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4343 err = -EBUSY;
4344 else {
4345 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4346 err = mddev->pers->start_reshape(mddev);
4347 }
4348 mddev_unlock(mddev);
4349 }
4350 if (err)
4351 return err;
4352 sysfs_notify(&mddev->kobj, NULL, "degraded");
4353 } else {
4354 if (cmd_match(page, "check"))
4355 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4356 else if (!cmd_match(page, "repair"))
4357 return -EINVAL;
4358 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4359 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4360 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4361 }
4362 if (mddev->ro == 2) {
4363 /* A write to sync_action is enough to justify
4364 * canceling read-auto mode
4365 */
4366 mddev->ro = 0;
4367 md_wakeup_thread(mddev->sync_thread);
4368 }
4369 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4370 md_wakeup_thread(mddev->thread);
4371 sysfs_notify_dirent_safe(mddev->sysfs_action);
4372 return len;
4373 }
4374
4375 static struct md_sysfs_entry md_scan_mode =
4376 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4377
4378 static ssize_t
4379 last_sync_action_show(struct mddev *mddev, char *page)
4380 {
4381 return sprintf(page, "%s\n", mddev->last_sync_action);
4382 }
4383
4384 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4385
4386 static ssize_t
4387 mismatch_cnt_show(struct mddev *mddev, char *page)
4388 {
4389 return sprintf(page, "%llu\n",
4390 (unsigned long long)
4391 atomic64_read(&mddev->resync_mismatches));
4392 }
4393
4394 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4395
4396 static ssize_t
4397 sync_min_show(struct mddev *mddev, char *page)
4398 {
4399 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4400 mddev->sync_speed_min ? "local": "system");
4401 }
4402
4403 static ssize_t
4404 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4405 {
4406 unsigned int min;
4407 int rv;
4408
4409 if (strncmp(buf, "system", 6)==0) {
4410 min = 0;
4411 } else {
4412 rv = kstrtouint(buf, 10, &min);
4413 if (rv < 0)
4414 return rv;
4415 if (min == 0)
4416 return -EINVAL;
4417 }
4418 mddev->sync_speed_min = min;
4419 return len;
4420 }
4421
4422 static struct md_sysfs_entry md_sync_min =
4423 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4424
4425 static ssize_t
4426 sync_max_show(struct mddev *mddev, char *page)
4427 {
4428 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4429 mddev->sync_speed_max ? "local": "system");
4430 }
4431
4432 static ssize_t
4433 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4434 {
4435 unsigned int max;
4436 int rv;
4437
4438 if (strncmp(buf, "system", 6)==0) {
4439 max = 0;
4440 } else {
4441 rv = kstrtouint(buf, 10, &max);
4442 if (rv < 0)
4443 return rv;
4444 if (max == 0)
4445 return -EINVAL;
4446 }
4447 mddev->sync_speed_max = max;
4448 return len;
4449 }
4450
4451 static struct md_sysfs_entry md_sync_max =
4452 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4453
4454 static ssize_t
4455 degraded_show(struct mddev *mddev, char *page)
4456 {
4457 return sprintf(page, "%d\n", mddev->degraded);
4458 }
4459 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4460
4461 static ssize_t
4462 sync_force_parallel_show(struct mddev *mddev, char *page)
4463 {
4464 return sprintf(page, "%d\n", mddev->parallel_resync);
4465 }
4466
4467 static ssize_t
4468 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4469 {
4470 long n;
4471
4472 if (kstrtol(buf, 10, &n))
4473 return -EINVAL;
4474
4475 if (n != 0 && n != 1)
4476 return -EINVAL;
4477
4478 mddev->parallel_resync = n;
4479
4480 if (mddev->sync_thread)
4481 wake_up(&resync_wait);
4482
4483 return len;
4484 }
4485
4486 /* force parallel resync, even with shared block devices */
4487 static struct md_sysfs_entry md_sync_force_parallel =
4488 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4489 sync_force_parallel_show, sync_force_parallel_store);
4490
4491 static ssize_t
4492 sync_speed_show(struct mddev *mddev, char *page)
4493 {
4494 unsigned long resync, dt, db;
4495 if (mddev->curr_resync == 0)
4496 return sprintf(page, "none\n");
4497 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4498 dt = (jiffies - mddev->resync_mark) / HZ;
4499 if (!dt) dt++;
4500 db = resync - mddev->resync_mark_cnt;
4501 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4502 }
4503
4504 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4505
4506 static ssize_t
4507 sync_completed_show(struct mddev *mddev, char *page)
4508 {
4509 unsigned long long max_sectors, resync;
4510
4511 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4512 return sprintf(page, "none\n");
4513
4514 if (mddev->curr_resync == 1 ||
4515 mddev->curr_resync == 2)
4516 return sprintf(page, "delayed\n");
4517
4518 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4519 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4520 max_sectors = mddev->resync_max_sectors;
4521 else
4522 max_sectors = mddev->dev_sectors;
4523
4524 resync = mddev->curr_resync_completed;
4525 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4526 }
4527
4528 static struct md_sysfs_entry md_sync_completed =
4529 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
4530
4531 static ssize_t
4532 min_sync_show(struct mddev *mddev, char *page)
4533 {
4534 return sprintf(page, "%llu\n",
4535 (unsigned long long)mddev->resync_min);
4536 }
4537 static ssize_t
4538 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4539 {
4540 unsigned long long min;
4541 int err;
4542
4543 if (kstrtoull(buf, 10, &min))
4544 return -EINVAL;
4545
4546 spin_lock(&mddev->lock);
4547 err = -EINVAL;
4548 if (min > mddev->resync_max)
4549 goto out_unlock;
4550
4551 err = -EBUSY;
4552 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4553 goto out_unlock;
4554
4555 /* Round down to multiple of 4K for safety */
4556 mddev->resync_min = round_down(min, 8);
4557 err = 0;
4558
4559 out_unlock:
4560 spin_unlock(&mddev->lock);
4561 return err ?: len;
4562 }
4563
4564 static struct md_sysfs_entry md_min_sync =
4565 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4566
4567 static ssize_t
4568 max_sync_show(struct mddev *mddev, char *page)
4569 {
4570 if (mddev->resync_max == MaxSector)
4571 return sprintf(page, "max\n");
4572 else
4573 return sprintf(page, "%llu\n",
4574 (unsigned long long)mddev->resync_max);
4575 }
4576 static ssize_t
4577 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4578 {
4579 int err;
4580 spin_lock(&mddev->lock);
4581 if (strncmp(buf, "max", 3) == 0)
4582 mddev->resync_max = MaxSector;
4583 else {
4584 unsigned long long max;
4585 int chunk;
4586
4587 err = -EINVAL;
4588 if (kstrtoull(buf, 10, &max))
4589 goto out_unlock;
4590 if (max < mddev->resync_min)
4591 goto out_unlock;
4592
4593 err = -EBUSY;
4594 if (max < mddev->resync_max &&
4595 mddev->ro == 0 &&
4596 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4597 goto out_unlock;
4598
4599 /* Must be a multiple of chunk_size */
4600 chunk = mddev->chunk_sectors;
4601 if (chunk) {
4602 sector_t temp = max;
4603
4604 err = -EINVAL;
4605 if (sector_div(temp, chunk))
4606 goto out_unlock;
4607 }
4608 mddev->resync_max = max;
4609 }
4610 wake_up(&mddev->recovery_wait);
4611 err = 0;
4612 out_unlock:
4613 spin_unlock(&mddev->lock);
4614 return err ?: len;
4615 }
4616
4617 static struct md_sysfs_entry md_max_sync =
4618 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4619
4620 static ssize_t
4621 suspend_lo_show(struct mddev *mddev, char *page)
4622 {
4623 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4624 }
4625
4626 static ssize_t
4627 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4628 {
4629 unsigned long long old, new;
4630 int err;
4631
4632 err = kstrtoull(buf, 10, &new);
4633 if (err < 0)
4634 return err;
4635 if (new != (sector_t)new)
4636 return -EINVAL;
4637
4638 err = mddev_lock(mddev);
4639 if (err)
4640 return err;
4641 err = -EINVAL;
4642 if (mddev->pers == NULL ||
4643 mddev->pers->quiesce == NULL)
4644 goto unlock;
4645 old = mddev->suspend_lo;
4646 mddev->suspend_lo = new;
4647 if (new >= old)
4648 /* Shrinking suspended region */
4649 mddev->pers->quiesce(mddev, 2);
4650 else {
4651 /* Expanding suspended region - need to wait */
4652 mddev->pers->quiesce(mddev, 1);
4653 mddev->pers->quiesce(mddev, 0);
4654 }
4655 err = 0;
4656 unlock:
4657 mddev_unlock(mddev);
4658 return err ?: len;
4659 }
4660 static struct md_sysfs_entry md_suspend_lo =
4661 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4662
4663 static ssize_t
4664 suspend_hi_show(struct mddev *mddev, char *page)
4665 {
4666 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4667 }
4668
4669 static ssize_t
4670 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4671 {
4672 unsigned long long old, new;
4673 int err;
4674
4675 err = kstrtoull(buf, 10, &new);
4676 if (err < 0)
4677 return err;
4678 if (new != (sector_t)new)
4679 return -EINVAL;
4680
4681 err = mddev_lock(mddev);
4682 if (err)
4683 return err;
4684 err = -EINVAL;
4685 if (mddev->pers == NULL ||
4686 mddev->pers->quiesce == NULL)
4687 goto unlock;
4688 old = mddev->suspend_hi;
4689 mddev->suspend_hi = new;
4690 if (new <= old)
4691 /* Shrinking suspended region */
4692 mddev->pers->quiesce(mddev, 2);
4693 else {
4694 /* Expanding suspended region - need to wait */
4695 mddev->pers->quiesce(mddev, 1);
4696 mddev->pers->quiesce(mddev, 0);
4697 }
4698 err = 0;
4699 unlock:
4700 mddev_unlock(mddev);
4701 return err ?: len;
4702 }
4703 static struct md_sysfs_entry md_suspend_hi =
4704 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4705
4706 static ssize_t
4707 reshape_position_show(struct mddev *mddev, char *page)
4708 {
4709 if (mddev->reshape_position != MaxSector)
4710 return sprintf(page, "%llu\n",
4711 (unsigned long long)mddev->reshape_position);
4712 strcpy(page, "none\n");
4713 return 5;
4714 }
4715
4716 static ssize_t
4717 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4718 {
4719 struct md_rdev *rdev;
4720 unsigned long long new;
4721 int err;
4722
4723 err = kstrtoull(buf, 10, &new);
4724 if (err < 0)
4725 return err;
4726 if (new != (sector_t)new)
4727 return -EINVAL;
4728 err = mddev_lock(mddev);
4729 if (err)
4730 return err;
4731 err = -EBUSY;
4732 if (mddev->pers)
4733 goto unlock;
4734 mddev->reshape_position = new;
4735 mddev->delta_disks = 0;
4736 mddev->reshape_backwards = 0;
4737 mddev->new_level = mddev->level;
4738 mddev->new_layout = mddev->layout;
4739 mddev->new_chunk_sectors = mddev->chunk_sectors;
4740 rdev_for_each(rdev, mddev)
4741 rdev->new_data_offset = rdev->data_offset;
4742 err = 0;
4743 unlock:
4744 mddev_unlock(mddev);
4745 return err ?: len;
4746 }
4747
4748 static struct md_sysfs_entry md_reshape_position =
4749 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4750 reshape_position_store);
4751
4752 static ssize_t
4753 reshape_direction_show(struct mddev *mddev, char *page)
4754 {
4755 return sprintf(page, "%s\n",
4756 mddev->reshape_backwards ? "backwards" : "forwards");
4757 }
4758
4759 static ssize_t
4760 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4761 {
4762 int backwards = 0;
4763 int err;
4764
4765 if (cmd_match(buf, "forwards"))
4766 backwards = 0;
4767 else if (cmd_match(buf, "backwards"))
4768 backwards = 1;
4769 else
4770 return -EINVAL;
4771 if (mddev->reshape_backwards == backwards)
4772 return len;
4773
4774 err = mddev_lock(mddev);
4775 if (err)
4776 return err;
4777 /* check if we are allowed to change */
4778 if (mddev->delta_disks)
4779 err = -EBUSY;
4780 else if (mddev->persistent &&
4781 mddev->major_version == 0)
4782 err = -EINVAL;
4783 else
4784 mddev->reshape_backwards = backwards;
4785 mddev_unlock(mddev);
4786 return err ?: len;
4787 }
4788
4789 static struct md_sysfs_entry md_reshape_direction =
4790 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4791 reshape_direction_store);
4792
4793 static ssize_t
4794 array_size_show(struct mddev *mddev, char *page)
4795 {
4796 if (mddev->external_size)
4797 return sprintf(page, "%llu\n",
4798 (unsigned long long)mddev->array_sectors/2);
4799 else
4800 return sprintf(page, "default\n");
4801 }
4802
4803 static ssize_t
4804 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4805 {
4806 sector_t sectors;
4807 int err;
4808
4809 err = mddev_lock(mddev);
4810 if (err)
4811 return err;
4812
4813 if (strncmp(buf, "default", 7) == 0) {
4814 if (mddev->pers)
4815 sectors = mddev->pers->size(mddev, 0, 0);
4816 else
4817 sectors = mddev->array_sectors;
4818
4819 mddev->external_size = 0;
4820 } else {
4821 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4822 err = -EINVAL;
4823 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4824 err = -E2BIG;
4825 else
4826 mddev->external_size = 1;
4827 }
4828
4829 if (!err) {
4830 mddev->array_sectors = sectors;
4831 if (mddev->pers) {
4832 set_capacity(mddev->gendisk, mddev->array_sectors);
4833 revalidate_disk(mddev->gendisk);
4834 }
4835 }
4836 mddev_unlock(mddev);
4837 return err ?: len;
4838 }
4839
4840 static struct md_sysfs_entry md_array_size =
4841 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4842 array_size_store);
4843
4844 static struct attribute *md_default_attrs[] = {
4845 &md_level.attr,
4846 &md_layout.attr,
4847 &md_raid_disks.attr,
4848 &md_chunk_size.attr,
4849 &md_size.attr,
4850 &md_resync_start.attr,
4851 &md_metadata.attr,
4852 &md_new_device.attr,
4853 &md_safe_delay.attr,
4854 &md_array_state.attr,
4855 &md_reshape_position.attr,
4856 &md_reshape_direction.attr,
4857 &md_array_size.attr,
4858 &max_corr_read_errors.attr,
4859 NULL,
4860 };
4861
4862 static struct attribute *md_redundancy_attrs[] = {
4863 &md_scan_mode.attr,
4864 &md_last_scan_mode.attr,
4865 &md_mismatches.attr,
4866 &md_sync_min.attr,
4867 &md_sync_max.attr,
4868 &md_sync_speed.attr,
4869 &md_sync_force_parallel.attr,
4870 &md_sync_completed.attr,
4871 &md_min_sync.attr,
4872 &md_max_sync.attr,
4873 &md_suspend_lo.attr,
4874 &md_suspend_hi.attr,
4875 &md_bitmap.attr,
4876 &md_degraded.attr,
4877 NULL,
4878 };
4879 static struct attribute_group md_redundancy_group = {
4880 .name = NULL,
4881 .attrs = md_redundancy_attrs,
4882 };
4883
4884 static ssize_t
4885 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4886 {
4887 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4888 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4889 ssize_t rv;
4890
4891 if (!entry->show)
4892 return -EIO;
4893 spin_lock(&all_mddevs_lock);
4894 if (list_empty(&mddev->all_mddevs)) {
4895 spin_unlock(&all_mddevs_lock);
4896 return -EBUSY;
4897 }
4898 mddev_get(mddev);
4899 spin_unlock(&all_mddevs_lock);
4900
4901 rv = entry->show(mddev, page);
4902 mddev_put(mddev);
4903 return rv;
4904 }
4905
4906 static ssize_t
4907 md_attr_store(struct kobject *kobj, struct attribute *attr,
4908 const char *page, size_t length)
4909 {
4910 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4911 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4912 ssize_t rv;
4913
4914 if (!entry->store)
4915 return -EIO;
4916 if (!capable(CAP_SYS_ADMIN))
4917 return -EACCES;
4918 spin_lock(&all_mddevs_lock);
4919 if (list_empty(&mddev->all_mddevs)) {
4920 spin_unlock(&all_mddevs_lock);
4921 return -EBUSY;
4922 }
4923 mddev_get(mddev);
4924 spin_unlock(&all_mddevs_lock);
4925 rv = entry->store(mddev, page, length);
4926 mddev_put(mddev);
4927 return rv;
4928 }
4929
4930 static void md_free(struct kobject *ko)
4931 {
4932 struct mddev *mddev = container_of(ko, struct mddev, kobj);
4933
4934 if (mddev->sysfs_state)
4935 sysfs_put(mddev->sysfs_state);
4936
4937 if (mddev->queue)
4938 blk_cleanup_queue(mddev->queue);
4939 if (mddev->gendisk) {
4940 del_gendisk(mddev->gendisk);
4941 put_disk(mddev->gendisk);
4942 }
4943
4944 kfree(mddev);
4945 }
4946
4947 static const struct sysfs_ops md_sysfs_ops = {
4948 .show = md_attr_show,
4949 .store = md_attr_store,
4950 };
4951 static struct kobj_type md_ktype = {
4952 .release = md_free,
4953 .sysfs_ops = &md_sysfs_ops,
4954 .default_attrs = md_default_attrs,
4955 };
4956
4957 int mdp_major = 0;
4958
4959 static void mddev_delayed_delete(struct work_struct *ws)
4960 {
4961 struct mddev *mddev = container_of(ws, struct mddev, del_work);
4962
4963 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4964 kobject_del(&mddev->kobj);
4965 kobject_put(&mddev->kobj);
4966 }
4967
4968 static int md_alloc(dev_t dev, char *name)
4969 {
4970 static DEFINE_MUTEX(disks_mutex);
4971 struct mddev *mddev = mddev_find(dev);
4972 struct gendisk *disk;
4973 int partitioned;
4974 int shift;
4975 int unit;
4976 int error;
4977
4978 if (!mddev)
4979 return -ENODEV;
4980
4981 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4982 shift = partitioned ? MdpMinorShift : 0;
4983 unit = MINOR(mddev->unit) >> shift;
4984
4985 /* wait for any previous instance of this device to be
4986 * completely removed (mddev_delayed_delete).
4987 */
4988 flush_workqueue(md_misc_wq);
4989
4990 mutex_lock(&disks_mutex);
4991 error = -EEXIST;
4992 if (mddev->gendisk)
4993 goto abort;
4994
4995 if (name) {
4996 /* Need to ensure that 'name' is not a duplicate.
4997 */
4998 struct mddev *mddev2;
4999 spin_lock(&all_mddevs_lock);
5000
5001 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5002 if (mddev2->gendisk &&
5003 strcmp(mddev2->gendisk->disk_name, name) == 0) {
5004 spin_unlock(&all_mddevs_lock);
5005 goto abort;
5006 }
5007 spin_unlock(&all_mddevs_lock);
5008 }
5009
5010 error = -ENOMEM;
5011 mddev->queue = blk_alloc_queue(GFP_KERNEL);
5012 if (!mddev->queue)
5013 goto abort;
5014 mddev->queue->queuedata = mddev;
5015
5016 blk_queue_make_request(mddev->queue, md_make_request);
5017 blk_set_stacking_limits(&mddev->queue->limits);
5018
5019 disk = alloc_disk(1 << shift);
5020 if (!disk) {
5021 blk_cleanup_queue(mddev->queue);
5022 mddev->queue = NULL;
5023 goto abort;
5024 }
5025 disk->major = MAJOR(mddev->unit);
5026 disk->first_minor = unit << shift;
5027 if (name)
5028 strcpy(disk->disk_name, name);
5029 else if (partitioned)
5030 sprintf(disk->disk_name, "md_d%d", unit);
5031 else
5032 sprintf(disk->disk_name, "md%d", unit);
5033 disk->fops = &md_fops;
5034 disk->private_data = mddev;
5035 disk->queue = mddev->queue;
5036 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
5037 /* Allow extended partitions. This makes the
5038 * 'mdp' device redundant, but we can't really
5039 * remove it now.
5040 */
5041 disk->flags |= GENHD_FL_EXT_DEVT;
5042 mddev->gendisk = disk;
5043 /* As soon as we call add_disk(), another thread could get
5044 * through to md_open, so make sure it doesn't get too far
5045 */
5046 mutex_lock(&mddev->open_mutex);
5047 add_disk(disk);
5048
5049 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
5050 &disk_to_dev(disk)->kobj, "%s", "md");
5051 if (error) {
5052 /* This isn't possible, but as kobject_init_and_add is marked
5053 * __must_check, we must do something with the result
5054 */
5055 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
5056 disk->disk_name);
5057 error = 0;
5058 }
5059 if (mddev->kobj.sd &&
5060 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
5061 printk(KERN_DEBUG "pointless warning\n");
5062 mutex_unlock(&mddev->open_mutex);
5063 abort:
5064 mutex_unlock(&disks_mutex);
5065 if (!error && mddev->kobj.sd) {
5066 kobject_uevent(&mddev->kobj, KOBJ_ADD);
5067 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5068 }
5069 mddev_put(mddev);
5070 return error;
5071 }
5072
5073 static struct kobject *md_probe(dev_t dev, int *part, void *data)
5074 {
5075 md_alloc(dev, NULL);
5076 return NULL;
5077 }
5078
5079 static int add_named_array(const char *val, struct kernel_param *kp)
5080 {
5081 /* val must be "md_*" where * is not all digits.
5082 * We allocate an array with a large free minor number, and
5083 * set the name to val. val must not already be an active name.
5084 */
5085 int len = strlen(val);
5086 char buf[DISK_NAME_LEN];
5087
5088 while (len && val[len-1] == '\n')
5089 len--;
5090 if (len >= DISK_NAME_LEN)
5091 return -E2BIG;
5092 strlcpy(buf, val, len+1);
5093 if (strncmp(buf, "md_", 3) != 0)
5094 return -EINVAL;
5095 return md_alloc(0, buf);
5096 }
5097
5098 static void md_safemode_timeout(unsigned long data)
5099 {
5100 struct mddev *mddev = (struct mddev *) data;
5101
5102 if (!atomic_read(&mddev->writes_pending)) {
5103 mddev->safemode = 1;
5104 if (mddev->external)
5105 sysfs_notify_dirent_safe(mddev->sysfs_state);
5106 }
5107 md_wakeup_thread(mddev->thread);
5108 }
5109
5110 static int start_dirty_degraded;
5111
5112 int md_run(struct mddev *mddev)
5113 {
5114 int err;
5115 struct md_rdev *rdev;
5116 struct md_personality *pers;
5117
5118 if (list_empty(&mddev->disks))
5119 /* cannot run an array with no devices.. */
5120 return -EINVAL;
5121
5122 if (mddev->pers)
5123 return -EBUSY;
5124 /* Cannot run until previous stop completes properly */
5125 if (mddev->sysfs_active)
5126 return -EBUSY;
5127
5128 /*
5129 * Analyze all RAID superblock(s)
5130 */
5131 if (!mddev->raid_disks) {
5132 if (!mddev->persistent)
5133 return -EINVAL;
5134 analyze_sbs(mddev);
5135 }
5136
5137 if (mddev->level != LEVEL_NONE)
5138 request_module("md-level-%d", mddev->level);
5139 else if (mddev->clevel[0])
5140 request_module("md-%s", mddev->clevel);
5141
5142 /*
5143 * Drop all container device buffers, from now on
5144 * the only valid external interface is through the md
5145 * device.
5146 */
5147 rdev_for_each(rdev, mddev) {
5148 if (test_bit(Faulty, &rdev->flags))
5149 continue;
5150 sync_blockdev(rdev->bdev);
5151 invalidate_bdev(rdev->bdev);
5152
5153 /* perform some consistency tests on the device.
5154 * We don't want the data to overlap the metadata,
5155 * Internal Bitmap issues have been handled elsewhere.
5156 */
5157 if (rdev->meta_bdev) {
5158 /* Nothing to check */;
5159 } else if (rdev->data_offset < rdev->sb_start) {
5160 if (mddev->dev_sectors &&
5161 rdev->data_offset + mddev->dev_sectors
5162 > rdev->sb_start) {
5163 printk("md: %s: data overlaps metadata\n",
5164 mdname(mddev));
5165 return -EINVAL;
5166 }
5167 } else {
5168 if (rdev->sb_start + rdev->sb_size/512
5169 > rdev->data_offset) {
5170 printk("md: %s: metadata overlaps data\n",
5171 mdname(mddev));
5172 return -EINVAL;
5173 }
5174 }
5175 sysfs_notify_dirent_safe(rdev->sysfs_state);
5176 }
5177
5178 if (mddev->bio_set == NULL)
5179 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
5180
5181 spin_lock(&pers_lock);
5182 pers = find_pers(mddev->level, mddev->clevel);
5183 if (!pers || !try_module_get(pers->owner)) {
5184 spin_unlock(&pers_lock);
5185 if (mddev->level != LEVEL_NONE)
5186 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5187 mddev->level);
5188 else
5189 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5190 mddev->clevel);
5191 return -EINVAL;
5192 }
5193 spin_unlock(&pers_lock);
5194 if (mddev->level != pers->level) {
5195 mddev->level = pers->level;
5196 mddev->new_level = pers->level;
5197 }
5198 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5199
5200 if (mddev->reshape_position != MaxSector &&
5201 pers->start_reshape == NULL) {
5202 /* This personality cannot handle reshaping... */
5203 module_put(pers->owner);
5204 return -EINVAL;
5205 }
5206
5207 if (pers->sync_request) {
5208 /* Warn if this is a potentially silly
5209 * configuration.
5210 */
5211 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5212 struct md_rdev *rdev2;
5213 int warned = 0;
5214
5215 rdev_for_each(rdev, mddev)
5216 rdev_for_each(rdev2, mddev) {
5217 if (rdev < rdev2 &&
5218 rdev->bdev->bd_contains ==
5219 rdev2->bdev->bd_contains) {
5220 printk(KERN_WARNING
5221 "%s: WARNING: %s appears to be"
5222 " on the same physical disk as"
5223 " %s.\n",
5224 mdname(mddev),
5225 bdevname(rdev->bdev,b),
5226 bdevname(rdev2->bdev,b2));
5227 warned = 1;
5228 }
5229 }
5230
5231 if (warned)
5232 printk(KERN_WARNING
5233 "True protection against single-disk"
5234 " failure might be compromised.\n");
5235 }
5236
5237 mddev->recovery = 0;
5238 /* may be over-ridden by personality */
5239 mddev->resync_max_sectors = mddev->dev_sectors;
5240
5241 mddev->ok_start_degraded = start_dirty_degraded;
5242
5243 if (start_readonly && mddev->ro == 0)
5244 mddev->ro = 2; /* read-only, but switch on first write */
5245
5246 err = pers->run(mddev);
5247 if (err)
5248 printk(KERN_ERR "md: pers->run() failed ...\n");
5249 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5250 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5251 " but 'external_size' not in effect?\n", __func__);
5252 printk(KERN_ERR
5253 "md: invalid array_size %llu > default size %llu\n",
5254 (unsigned long long)mddev->array_sectors / 2,
5255 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5256 err = -EINVAL;
5257 }
5258 if (err == 0 && pers->sync_request &&
5259 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5260 struct bitmap *bitmap;
5261
5262 bitmap = bitmap_create(mddev, -1);
5263 if (IS_ERR(bitmap)) {
5264 err = PTR_ERR(bitmap);
5265 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5266 mdname(mddev), err);
5267 } else
5268 mddev->bitmap = bitmap;
5269
5270 }
5271 if (err) {
5272 mddev_detach(mddev);
5273 if (mddev->private)
5274 pers->free(mddev, mddev->private);
5275 mddev->private = NULL;
5276 module_put(pers->owner);
5277 bitmap_destroy(mddev);
5278 return err;
5279 }
5280 if (mddev->queue) {
5281 mddev->queue->backing_dev_info.congested_data = mddev;
5282 mddev->queue->backing_dev_info.congested_fn = md_congested;
5283 }
5284 if (pers->sync_request) {
5285 if (mddev->kobj.sd &&
5286 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5287 printk(KERN_WARNING
5288 "md: cannot register extra attributes for %s\n",
5289 mdname(mddev));
5290 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5291 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5292 mddev->ro = 0;
5293
5294 atomic_set(&mddev->writes_pending,0);
5295 atomic_set(&mddev->max_corr_read_errors,
5296 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5297 mddev->safemode = 0;
5298 if (mddev_is_clustered(mddev))
5299 mddev->safemode_delay = 0;
5300 else
5301 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5302 mddev->in_sync = 1;
5303 smp_wmb();
5304 spin_lock(&mddev->lock);
5305 mddev->pers = pers;
5306 spin_unlock(&mddev->lock);
5307 rdev_for_each(rdev, mddev)
5308 if (rdev->raid_disk >= 0)
5309 if (sysfs_link_rdev(mddev, rdev))
5310 /* failure here is OK */;
5311
5312 if (mddev->degraded && !mddev->ro)
5313 /* This ensures that recovering status is reported immediately
5314 * via sysfs - until a lack of spares is confirmed.
5315 */
5316 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5317 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5318
5319 if (mddev->flags & MD_UPDATE_SB_FLAGS)
5320 md_update_sb(mddev, 0);
5321
5322 md_new_event(mddev);
5323 sysfs_notify_dirent_safe(mddev->sysfs_state);
5324 sysfs_notify_dirent_safe(mddev->sysfs_action);
5325 sysfs_notify(&mddev->kobj, NULL, "degraded");
5326 return 0;
5327 }
5328 EXPORT_SYMBOL_GPL(md_run);
5329
5330 static int do_md_run(struct mddev *mddev)
5331 {
5332 int err;
5333
5334 err = md_run(mddev);
5335 if (err)
5336 goto out;
5337 err = bitmap_load(mddev);
5338 if (err) {
5339 bitmap_destroy(mddev);
5340 goto out;
5341 }
5342
5343 if (mddev_is_clustered(mddev))
5344 md_allow_write(mddev);
5345
5346 md_wakeup_thread(mddev->thread);
5347 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5348
5349 set_capacity(mddev->gendisk, mddev->array_sectors);
5350 revalidate_disk(mddev->gendisk);
5351 mddev->changed = 1;
5352 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5353 out:
5354 return err;
5355 }
5356
5357 static int restart_array(struct mddev *mddev)
5358 {
5359 struct gendisk *disk = mddev->gendisk;
5360
5361 /* Complain if it has no devices */
5362 if (list_empty(&mddev->disks))
5363 return -ENXIO;
5364 if (!mddev->pers)
5365 return -EINVAL;
5366 if (!mddev->ro)
5367 return -EBUSY;
5368 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5369 struct md_rdev *rdev;
5370 bool has_journal = false;
5371
5372 rcu_read_lock();
5373 rdev_for_each_rcu(rdev, mddev) {
5374 if (test_bit(Journal, &rdev->flags) &&
5375 !test_bit(Faulty, &rdev->flags)) {
5376 has_journal = true;
5377 break;
5378 }
5379 }
5380 rcu_read_unlock();
5381
5382 /* Don't restart rw with journal missing/faulty */
5383 if (!has_journal)
5384 return -EINVAL;
5385 }
5386
5387 mddev->safemode = 0;
5388 mddev->ro = 0;
5389 set_disk_ro(disk, 0);
5390 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5391 mdname(mddev));
5392 /* Kick recovery or resync if necessary */
5393 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5394 md_wakeup_thread(mddev->thread);
5395 md_wakeup_thread(mddev->sync_thread);
5396 sysfs_notify_dirent_safe(mddev->sysfs_state);
5397 return 0;
5398 }
5399
5400 static void md_clean(struct mddev *mddev)
5401 {
5402 mddev->array_sectors = 0;
5403 mddev->external_size = 0;
5404 mddev->dev_sectors = 0;
5405 mddev->raid_disks = 0;
5406 mddev->recovery_cp = 0;
5407 mddev->resync_min = 0;
5408 mddev->resync_max = MaxSector;
5409 mddev->reshape_position = MaxSector;
5410 mddev->external = 0;
5411 mddev->persistent = 0;
5412 mddev->level = LEVEL_NONE;
5413 mddev->clevel[0] = 0;
5414 mddev->flags = 0;
5415 mddev->ro = 0;
5416 mddev->metadata_type[0] = 0;
5417 mddev->chunk_sectors = 0;
5418 mddev->ctime = mddev->utime = 0;
5419 mddev->layout = 0;
5420 mddev->max_disks = 0;
5421 mddev->events = 0;
5422 mddev->can_decrease_events = 0;
5423 mddev->delta_disks = 0;
5424 mddev->reshape_backwards = 0;
5425 mddev->new_level = LEVEL_NONE;
5426 mddev->new_layout = 0;
5427 mddev->new_chunk_sectors = 0;
5428 mddev->curr_resync = 0;
5429 atomic64_set(&mddev->resync_mismatches, 0);
5430 mddev->suspend_lo = mddev->suspend_hi = 0;
5431 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5432 mddev->recovery = 0;
5433 mddev->in_sync = 0;
5434 mddev->changed = 0;
5435 mddev->degraded = 0;
5436 mddev->safemode = 0;
5437 mddev->private = NULL;
5438 mddev->bitmap_info.offset = 0;
5439 mddev->bitmap_info.default_offset = 0;
5440 mddev->bitmap_info.default_space = 0;
5441 mddev->bitmap_info.chunksize = 0;
5442 mddev->bitmap_info.daemon_sleep = 0;
5443 mddev->bitmap_info.max_write_behind = 0;
5444 }
5445
5446 static void __md_stop_writes(struct mddev *mddev)
5447 {
5448 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5449 flush_workqueue(md_misc_wq);
5450 if (mddev->sync_thread) {
5451 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5452 md_reap_sync_thread(mddev);
5453 }
5454
5455 del_timer_sync(&mddev->safemode_timer);
5456
5457 bitmap_flush(mddev);
5458 md_super_wait(mddev);
5459
5460 if (mddev->ro == 0 &&
5461 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
5462 (mddev->flags & MD_UPDATE_SB_FLAGS))) {
5463 /* mark array as shutdown cleanly */
5464 if (!mddev_is_clustered(mddev))
5465 mddev->in_sync = 1;
5466 md_update_sb(mddev, 1);
5467 }
5468 }
5469
5470 void md_stop_writes(struct mddev *mddev)
5471 {
5472 mddev_lock_nointr(mddev);
5473 __md_stop_writes(mddev);
5474 mddev_unlock(mddev);
5475 }
5476 EXPORT_SYMBOL_GPL(md_stop_writes);
5477
5478 static void mddev_detach(struct mddev *mddev)
5479 {
5480 struct bitmap *bitmap = mddev->bitmap;
5481 /* wait for behind writes to complete */
5482 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
5483 printk(KERN_INFO "md:%s: behind writes in progress - waiting to stop.\n",
5484 mdname(mddev));
5485 /* need to kick something here to make sure I/O goes? */
5486 wait_event(bitmap->behind_wait,
5487 atomic_read(&bitmap->behind_writes) == 0);
5488 }
5489 if (mddev->pers && mddev->pers->quiesce) {
5490 mddev->pers->quiesce(mddev, 1);
5491 mddev->pers->quiesce(mddev, 0);
5492 }
5493 md_unregister_thread(&mddev->thread);
5494 if (mddev->queue)
5495 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5496 }
5497
5498 static void __md_stop(struct mddev *mddev)
5499 {
5500 struct md_personality *pers = mddev->pers;
5501 mddev_detach(mddev);
5502 /* Ensure ->event_work is done */
5503 flush_workqueue(md_misc_wq);
5504 spin_lock(&mddev->lock);
5505 mddev->pers = NULL;
5506 spin_unlock(&mddev->lock);
5507 pers->free(mddev, mddev->private);
5508 mddev->private = NULL;
5509 if (pers->sync_request && mddev->to_remove == NULL)
5510 mddev->to_remove = &md_redundancy_group;
5511 module_put(pers->owner);
5512 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5513 }
5514
5515 void md_stop(struct mddev *mddev)
5516 {
5517 /* stop the array and free an attached data structures.
5518 * This is called from dm-raid
5519 */
5520 __md_stop(mddev);
5521 bitmap_destroy(mddev);
5522 if (mddev->bio_set)
5523 bioset_free(mddev->bio_set);
5524 }
5525
5526 EXPORT_SYMBOL_GPL(md_stop);
5527
5528 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5529 {
5530 int err = 0;
5531 int did_freeze = 0;
5532
5533 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5534 did_freeze = 1;
5535 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5536 md_wakeup_thread(mddev->thread);
5537 }
5538 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5539 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5540 if (mddev->sync_thread)
5541 /* Thread might be blocked waiting for metadata update
5542 * which will now never happen */
5543 wake_up_process(mddev->sync_thread->tsk);
5544
5545 if (mddev->external && test_bit(MD_CHANGE_PENDING, &mddev->flags))
5546 return -EBUSY;
5547 mddev_unlock(mddev);
5548 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5549 &mddev->recovery));
5550 wait_event(mddev->sb_wait,
5551 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5552 mddev_lock_nointr(mddev);
5553
5554 mutex_lock(&mddev->open_mutex);
5555 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5556 mddev->sync_thread ||
5557 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5558 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5559 printk("md: %s still in use.\n",mdname(mddev));
5560 if (did_freeze) {
5561 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5562 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5563 md_wakeup_thread(mddev->thread);
5564 }
5565 err = -EBUSY;
5566 goto out;
5567 }
5568 if (mddev->pers) {
5569 __md_stop_writes(mddev);
5570
5571 err = -ENXIO;
5572 if (mddev->ro==1)
5573 goto out;
5574 mddev->ro = 1;
5575 set_disk_ro(mddev->gendisk, 1);
5576 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5577 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5578 md_wakeup_thread(mddev->thread);
5579 sysfs_notify_dirent_safe(mddev->sysfs_state);
5580 err = 0;
5581 }
5582 out:
5583 mutex_unlock(&mddev->open_mutex);
5584 return err;
5585 }
5586
5587 /* mode:
5588 * 0 - completely stop and dis-assemble array
5589 * 2 - stop but do not disassemble array
5590 */
5591 static int do_md_stop(struct mddev *mddev, int mode,
5592 struct block_device *bdev)
5593 {
5594 struct gendisk *disk = mddev->gendisk;
5595 struct md_rdev *rdev;
5596 int did_freeze = 0;
5597
5598 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5599 did_freeze = 1;
5600 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5601 md_wakeup_thread(mddev->thread);
5602 }
5603 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5604 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5605 if (mddev->sync_thread)
5606 /* Thread might be blocked waiting for metadata update
5607 * which will now never happen */
5608 wake_up_process(mddev->sync_thread->tsk);
5609
5610 mddev_unlock(mddev);
5611 wait_event(resync_wait, (mddev->sync_thread == NULL &&
5612 !test_bit(MD_RECOVERY_RUNNING,
5613 &mddev->recovery)));
5614 mddev_lock_nointr(mddev);
5615
5616 mutex_lock(&mddev->open_mutex);
5617 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5618 mddev->sysfs_active ||
5619 mddev->sync_thread ||
5620 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5621 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5622 printk("md: %s still in use.\n",mdname(mddev));
5623 mutex_unlock(&mddev->open_mutex);
5624 if (did_freeze) {
5625 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5626 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5627 md_wakeup_thread(mddev->thread);
5628 }
5629 return -EBUSY;
5630 }
5631 if (mddev->pers) {
5632 if (mddev->ro)
5633 set_disk_ro(disk, 0);
5634
5635 __md_stop_writes(mddev);
5636 __md_stop(mddev);
5637 mddev->queue->backing_dev_info.congested_fn = NULL;
5638
5639 /* tell userspace to handle 'inactive' */
5640 sysfs_notify_dirent_safe(mddev->sysfs_state);
5641
5642 rdev_for_each(rdev, mddev)
5643 if (rdev->raid_disk >= 0)
5644 sysfs_unlink_rdev(mddev, rdev);
5645
5646 set_capacity(disk, 0);
5647 mutex_unlock(&mddev->open_mutex);
5648 mddev->changed = 1;
5649 revalidate_disk(disk);
5650
5651 if (mddev->ro)
5652 mddev->ro = 0;
5653 } else
5654 mutex_unlock(&mddev->open_mutex);
5655 /*
5656 * Free resources if final stop
5657 */
5658 if (mode == 0) {
5659 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5660
5661 bitmap_destroy(mddev);
5662 if (mddev->bitmap_info.file) {
5663 struct file *f = mddev->bitmap_info.file;
5664 spin_lock(&mddev->lock);
5665 mddev->bitmap_info.file = NULL;
5666 spin_unlock(&mddev->lock);
5667 fput(f);
5668 }
5669 mddev->bitmap_info.offset = 0;
5670
5671 export_array(mddev);
5672
5673 md_clean(mddev);
5674 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5675 if (mddev->hold_active == UNTIL_STOP)
5676 mddev->hold_active = 0;
5677 }
5678 md_new_event(mddev);
5679 sysfs_notify_dirent_safe(mddev->sysfs_state);
5680 return 0;
5681 }
5682
5683 #ifndef MODULE
5684 static void autorun_array(struct mddev *mddev)
5685 {
5686 struct md_rdev *rdev;
5687 int err;
5688
5689 if (list_empty(&mddev->disks))
5690 return;
5691
5692 printk(KERN_INFO "md: running: ");
5693
5694 rdev_for_each(rdev, mddev) {
5695 char b[BDEVNAME_SIZE];
5696 printk("<%s>", bdevname(rdev->bdev,b));
5697 }
5698 printk("\n");
5699
5700 err = do_md_run(mddev);
5701 if (err) {
5702 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
5703 do_md_stop(mddev, 0, NULL);
5704 }
5705 }
5706
5707 /*
5708 * lets try to run arrays based on all disks that have arrived
5709 * until now. (those are in pending_raid_disks)
5710 *
5711 * the method: pick the first pending disk, collect all disks with
5712 * the same UUID, remove all from the pending list and put them into
5713 * the 'same_array' list. Then order this list based on superblock
5714 * update time (freshest comes first), kick out 'old' disks and
5715 * compare superblocks. If everything's fine then run it.
5716 *
5717 * If "unit" is allocated, then bump its reference count
5718 */
5719 static void autorun_devices(int part)
5720 {
5721 struct md_rdev *rdev0, *rdev, *tmp;
5722 struct mddev *mddev;
5723 char b[BDEVNAME_SIZE];
5724
5725 printk(KERN_INFO "md: autorun ...\n");
5726 while (!list_empty(&pending_raid_disks)) {
5727 int unit;
5728 dev_t dev;
5729 LIST_HEAD(candidates);
5730 rdev0 = list_entry(pending_raid_disks.next,
5731 struct md_rdev, same_set);
5732
5733 printk(KERN_INFO "md: considering %s ...\n",
5734 bdevname(rdev0->bdev,b));
5735 INIT_LIST_HEAD(&candidates);
5736 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5737 if (super_90_load(rdev, rdev0, 0) >= 0) {
5738 printk(KERN_INFO "md: adding %s ...\n",
5739 bdevname(rdev->bdev,b));
5740 list_move(&rdev->same_set, &candidates);
5741 }
5742 /*
5743 * now we have a set of devices, with all of them having
5744 * mostly sane superblocks. It's time to allocate the
5745 * mddev.
5746 */
5747 if (part) {
5748 dev = MKDEV(mdp_major,
5749 rdev0->preferred_minor << MdpMinorShift);
5750 unit = MINOR(dev) >> MdpMinorShift;
5751 } else {
5752 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5753 unit = MINOR(dev);
5754 }
5755 if (rdev0->preferred_minor != unit) {
5756 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5757 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5758 break;
5759 }
5760
5761 md_probe(dev, NULL, NULL);
5762 mddev = mddev_find(dev);
5763 if (!mddev || !mddev->gendisk) {
5764 if (mddev)
5765 mddev_put(mddev);
5766 printk(KERN_ERR
5767 "md: cannot allocate memory for md drive.\n");
5768 break;
5769 }
5770 if (mddev_lock(mddev))
5771 printk(KERN_WARNING "md: %s locked, cannot run\n",
5772 mdname(mddev));
5773 else if (mddev->raid_disks || mddev->major_version
5774 || !list_empty(&mddev->disks)) {
5775 printk(KERN_WARNING
5776 "md: %s already running, cannot run %s\n",
5777 mdname(mddev), bdevname(rdev0->bdev,b));
5778 mddev_unlock(mddev);
5779 } else {
5780 printk(KERN_INFO "md: created %s\n", mdname(mddev));
5781 mddev->persistent = 1;
5782 rdev_for_each_list(rdev, tmp, &candidates) {
5783 list_del_init(&rdev->same_set);
5784 if (bind_rdev_to_array(rdev, mddev))
5785 export_rdev(rdev);
5786 }
5787 autorun_array(mddev);
5788 mddev_unlock(mddev);
5789 }
5790 /* on success, candidates will be empty, on error
5791 * it won't...
5792 */
5793 rdev_for_each_list(rdev, tmp, &candidates) {
5794 list_del_init(&rdev->same_set);
5795 export_rdev(rdev);
5796 }
5797 mddev_put(mddev);
5798 }
5799 printk(KERN_INFO "md: ... autorun DONE.\n");
5800 }
5801 #endif /* !MODULE */
5802
5803 static int get_version(void __user *arg)
5804 {
5805 mdu_version_t ver;
5806
5807 ver.major = MD_MAJOR_VERSION;
5808 ver.minor = MD_MINOR_VERSION;
5809 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5810
5811 if (copy_to_user(arg, &ver, sizeof(ver)))
5812 return -EFAULT;
5813
5814 return 0;
5815 }
5816
5817 static int get_array_info(struct mddev *mddev, void __user *arg)
5818 {
5819 mdu_array_info_t info;
5820 int nr,working,insync,failed,spare;
5821 struct md_rdev *rdev;
5822
5823 nr = working = insync = failed = spare = 0;
5824 rcu_read_lock();
5825 rdev_for_each_rcu(rdev, mddev) {
5826 nr++;
5827 if (test_bit(Faulty, &rdev->flags))
5828 failed++;
5829 else {
5830 working++;
5831 if (test_bit(In_sync, &rdev->flags))
5832 insync++;
5833 else
5834 spare++;
5835 }
5836 }
5837 rcu_read_unlock();
5838
5839 info.major_version = mddev->major_version;
5840 info.minor_version = mddev->minor_version;
5841 info.patch_version = MD_PATCHLEVEL_VERSION;
5842 info.ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
5843 info.level = mddev->level;
5844 info.size = mddev->dev_sectors / 2;
5845 if (info.size != mddev->dev_sectors / 2) /* overflow */
5846 info.size = -1;
5847 info.nr_disks = nr;
5848 info.raid_disks = mddev->raid_disks;
5849 info.md_minor = mddev->md_minor;
5850 info.not_persistent= !mddev->persistent;
5851
5852 info.utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
5853 info.state = 0;
5854 if (mddev->in_sync)
5855 info.state = (1<<MD_SB_CLEAN);
5856 if (mddev->bitmap && mddev->bitmap_info.offset)
5857 info.state |= (1<<MD_SB_BITMAP_PRESENT);
5858 if (mddev_is_clustered(mddev))
5859 info.state |= (1<<MD_SB_CLUSTERED);
5860 info.active_disks = insync;
5861 info.working_disks = working;
5862 info.failed_disks = failed;
5863 info.spare_disks = spare;
5864
5865 info.layout = mddev->layout;
5866 info.chunk_size = mddev->chunk_sectors << 9;
5867
5868 if (copy_to_user(arg, &info, sizeof(info)))
5869 return -EFAULT;
5870
5871 return 0;
5872 }
5873
5874 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
5875 {
5876 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5877 char *ptr;
5878 int err;
5879
5880 file = kzalloc(sizeof(*file), GFP_NOIO);
5881 if (!file)
5882 return -ENOMEM;
5883
5884 err = 0;
5885 spin_lock(&mddev->lock);
5886 /* bitmap enabled */
5887 if (mddev->bitmap_info.file) {
5888 ptr = file_path(mddev->bitmap_info.file, file->pathname,
5889 sizeof(file->pathname));
5890 if (IS_ERR(ptr))
5891 err = PTR_ERR(ptr);
5892 else
5893 memmove(file->pathname, ptr,
5894 sizeof(file->pathname)-(ptr-file->pathname));
5895 }
5896 spin_unlock(&mddev->lock);
5897
5898 if (err == 0 &&
5899 copy_to_user(arg, file, sizeof(*file)))
5900 err = -EFAULT;
5901
5902 kfree(file);
5903 return err;
5904 }
5905
5906 static int get_disk_info(struct mddev *mddev, void __user * arg)
5907 {
5908 mdu_disk_info_t info;
5909 struct md_rdev *rdev;
5910
5911 if (copy_from_user(&info, arg, sizeof(info)))
5912 return -EFAULT;
5913
5914 rcu_read_lock();
5915 rdev = md_find_rdev_nr_rcu(mddev, info.number);
5916 if (rdev) {
5917 info.major = MAJOR(rdev->bdev->bd_dev);
5918 info.minor = MINOR(rdev->bdev->bd_dev);
5919 info.raid_disk = rdev->raid_disk;
5920 info.state = 0;
5921 if (test_bit(Faulty, &rdev->flags))
5922 info.state |= (1<<MD_DISK_FAULTY);
5923 else if (test_bit(In_sync, &rdev->flags)) {
5924 info.state |= (1<<MD_DISK_ACTIVE);
5925 info.state |= (1<<MD_DISK_SYNC);
5926 }
5927 if (test_bit(Journal, &rdev->flags))
5928 info.state |= (1<<MD_DISK_JOURNAL);
5929 if (test_bit(WriteMostly, &rdev->flags))
5930 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5931 } else {
5932 info.major = info.minor = 0;
5933 info.raid_disk = -1;
5934 info.state = (1<<MD_DISK_REMOVED);
5935 }
5936 rcu_read_unlock();
5937
5938 if (copy_to_user(arg, &info, sizeof(info)))
5939 return -EFAULT;
5940
5941 return 0;
5942 }
5943
5944 static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
5945 {
5946 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5947 struct md_rdev *rdev;
5948 dev_t dev = MKDEV(info->major,info->minor);
5949
5950 if (mddev_is_clustered(mddev) &&
5951 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
5952 pr_err("%s: Cannot add to clustered mddev.\n",
5953 mdname(mddev));
5954 return -EINVAL;
5955 }
5956
5957 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5958 return -EOVERFLOW;
5959
5960 if (!mddev->raid_disks) {
5961 int err;
5962 /* expecting a device which has a superblock */
5963 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5964 if (IS_ERR(rdev)) {
5965 printk(KERN_WARNING
5966 "md: md_import_device returned %ld\n",
5967 PTR_ERR(rdev));
5968 return PTR_ERR(rdev);
5969 }
5970 if (!list_empty(&mddev->disks)) {
5971 struct md_rdev *rdev0
5972 = list_entry(mddev->disks.next,
5973 struct md_rdev, same_set);
5974 err = super_types[mddev->major_version]
5975 .load_super(rdev, rdev0, mddev->minor_version);
5976 if (err < 0) {
5977 printk(KERN_WARNING
5978 "md: %s has different UUID to %s\n",
5979 bdevname(rdev->bdev,b),
5980 bdevname(rdev0->bdev,b2));
5981 export_rdev(rdev);
5982 return -EINVAL;
5983 }
5984 }
5985 err = bind_rdev_to_array(rdev, mddev);
5986 if (err)
5987 export_rdev(rdev);
5988 return err;
5989 }
5990
5991 /*
5992 * add_new_disk can be used once the array is assembled
5993 * to add "hot spares". They must already have a superblock
5994 * written
5995 */
5996 if (mddev->pers) {
5997 int err;
5998 if (!mddev->pers->hot_add_disk) {
5999 printk(KERN_WARNING
6000 "%s: personality does not support diskops!\n",
6001 mdname(mddev));
6002 return -EINVAL;
6003 }
6004 if (mddev->persistent)
6005 rdev = md_import_device(dev, mddev->major_version,
6006 mddev->minor_version);
6007 else
6008 rdev = md_import_device(dev, -1, -1);
6009 if (IS_ERR(rdev)) {
6010 printk(KERN_WARNING
6011 "md: md_import_device returned %ld\n",
6012 PTR_ERR(rdev));
6013 return PTR_ERR(rdev);
6014 }
6015 /* set saved_raid_disk if appropriate */
6016 if (!mddev->persistent) {
6017 if (info->state & (1<<MD_DISK_SYNC) &&
6018 info->raid_disk < mddev->raid_disks) {
6019 rdev->raid_disk = info->raid_disk;
6020 set_bit(In_sync, &rdev->flags);
6021 clear_bit(Bitmap_sync, &rdev->flags);
6022 } else
6023 rdev->raid_disk = -1;
6024 rdev->saved_raid_disk = rdev->raid_disk;
6025 } else
6026 super_types[mddev->major_version].
6027 validate_super(mddev, rdev);
6028 if ((info->state & (1<<MD_DISK_SYNC)) &&
6029 rdev->raid_disk != info->raid_disk) {
6030 /* This was a hot-add request, but events doesn't
6031 * match, so reject it.
6032 */
6033 export_rdev(rdev);
6034 return -EINVAL;
6035 }
6036
6037 clear_bit(In_sync, &rdev->flags); /* just to be sure */
6038 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6039 set_bit(WriteMostly, &rdev->flags);
6040 else
6041 clear_bit(WriteMostly, &rdev->flags);
6042
6043 if (info->state & (1<<MD_DISK_JOURNAL)) {
6044 struct md_rdev *rdev2;
6045 bool has_journal = false;
6046
6047 /* make sure no existing journal disk */
6048 rdev_for_each(rdev2, mddev) {
6049 if (test_bit(Journal, &rdev2->flags)) {
6050 has_journal = true;
6051 break;
6052 }
6053 }
6054 if (has_journal) {
6055 export_rdev(rdev);
6056 return -EBUSY;
6057 }
6058 set_bit(Journal, &rdev->flags);
6059 }
6060 /*
6061 * check whether the device shows up in other nodes
6062 */
6063 if (mddev_is_clustered(mddev)) {
6064 if (info->state & (1 << MD_DISK_CANDIDATE))
6065 set_bit(Candidate, &rdev->flags);
6066 else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6067 /* --add initiated by this node */
6068 err = md_cluster_ops->add_new_disk(mddev, rdev);
6069 if (err) {
6070 export_rdev(rdev);
6071 return err;
6072 }
6073 }
6074 }
6075
6076 rdev->raid_disk = -1;
6077 err = bind_rdev_to_array(rdev, mddev);
6078
6079 if (err)
6080 export_rdev(rdev);
6081
6082 if (mddev_is_clustered(mddev)) {
6083 if (info->state & (1 << MD_DISK_CANDIDATE))
6084 md_cluster_ops->new_disk_ack(mddev, (err == 0));
6085 else {
6086 if (err)
6087 md_cluster_ops->add_new_disk_cancel(mddev);
6088 else
6089 err = add_bound_rdev(rdev);
6090 }
6091
6092 } else if (!err)
6093 err = add_bound_rdev(rdev);
6094
6095 return err;
6096 }
6097
6098 /* otherwise, add_new_disk is only allowed
6099 * for major_version==0 superblocks
6100 */
6101 if (mddev->major_version != 0) {
6102 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
6103 mdname(mddev));
6104 return -EINVAL;
6105 }
6106
6107 if (!(info->state & (1<<MD_DISK_FAULTY))) {
6108 int err;
6109 rdev = md_import_device(dev, -1, 0);
6110 if (IS_ERR(rdev)) {
6111 printk(KERN_WARNING
6112 "md: error, md_import_device() returned %ld\n",
6113 PTR_ERR(rdev));
6114 return PTR_ERR(rdev);
6115 }
6116 rdev->desc_nr = info->number;
6117 if (info->raid_disk < mddev->raid_disks)
6118 rdev->raid_disk = info->raid_disk;
6119 else
6120 rdev->raid_disk = -1;
6121
6122 if (rdev->raid_disk < mddev->raid_disks)
6123 if (info->state & (1<<MD_DISK_SYNC))
6124 set_bit(In_sync, &rdev->flags);
6125
6126 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6127 set_bit(WriteMostly, &rdev->flags);
6128
6129 if (!mddev->persistent) {
6130 printk(KERN_INFO "md: nonpersistent superblock ...\n");
6131 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6132 } else
6133 rdev->sb_start = calc_dev_sboffset(rdev);
6134 rdev->sectors = rdev->sb_start;
6135
6136 err = bind_rdev_to_array(rdev, mddev);
6137 if (err) {
6138 export_rdev(rdev);
6139 return err;
6140 }
6141 }
6142
6143 return 0;
6144 }
6145
6146 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6147 {
6148 char b[BDEVNAME_SIZE];
6149 struct md_rdev *rdev;
6150
6151 rdev = find_rdev(mddev, dev);
6152 if (!rdev)
6153 return -ENXIO;
6154
6155 if (rdev->raid_disk < 0)
6156 goto kick_rdev;
6157
6158 clear_bit(Blocked, &rdev->flags);
6159 remove_and_add_spares(mddev, rdev);
6160
6161 if (rdev->raid_disk >= 0)
6162 goto busy;
6163
6164 kick_rdev:
6165 if (mddev_is_clustered(mddev))
6166 md_cluster_ops->remove_disk(mddev, rdev);
6167
6168 md_kick_rdev_from_array(rdev);
6169 md_update_sb(mddev, 1);
6170 md_new_event(mddev);
6171
6172 return 0;
6173 busy:
6174 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
6175 bdevname(rdev->bdev,b), mdname(mddev));
6176 return -EBUSY;
6177 }
6178
6179 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6180 {
6181 char b[BDEVNAME_SIZE];
6182 int err;
6183 struct md_rdev *rdev;
6184
6185 if (!mddev->pers)
6186 return -ENODEV;
6187
6188 if (mddev->major_version != 0) {
6189 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
6190 " version-0 superblocks.\n",
6191 mdname(mddev));
6192 return -EINVAL;
6193 }
6194 if (!mddev->pers->hot_add_disk) {
6195 printk(KERN_WARNING
6196 "%s: personality does not support diskops!\n",
6197 mdname(mddev));
6198 return -EINVAL;
6199 }
6200
6201 rdev = md_import_device(dev, -1, 0);
6202 if (IS_ERR(rdev)) {
6203 printk(KERN_WARNING
6204 "md: error, md_import_device() returned %ld\n",
6205 PTR_ERR(rdev));
6206 return -EINVAL;
6207 }
6208
6209 if (mddev->persistent)
6210 rdev->sb_start = calc_dev_sboffset(rdev);
6211 else
6212 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6213
6214 rdev->sectors = rdev->sb_start;
6215
6216 if (test_bit(Faulty, &rdev->flags)) {
6217 printk(KERN_WARNING
6218 "md: can not hot-add faulty %s disk to %s!\n",
6219 bdevname(rdev->bdev,b), mdname(mddev));
6220 err = -EINVAL;
6221 goto abort_export;
6222 }
6223
6224 clear_bit(In_sync, &rdev->flags);
6225 rdev->desc_nr = -1;
6226 rdev->saved_raid_disk = -1;
6227 err = bind_rdev_to_array(rdev, mddev);
6228 if (err)
6229 goto abort_export;
6230
6231 /*
6232 * The rest should better be atomic, we can have disk failures
6233 * noticed in interrupt contexts ...
6234 */
6235
6236 rdev->raid_disk = -1;
6237
6238 md_update_sb(mddev, 1);
6239 /*
6240 * Kick recovery, maybe this spare has to be added to the
6241 * array immediately.
6242 */
6243 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6244 md_wakeup_thread(mddev->thread);
6245 md_new_event(mddev);
6246 return 0;
6247
6248 abort_export:
6249 export_rdev(rdev);
6250 return err;
6251 }
6252
6253 static int set_bitmap_file(struct mddev *mddev, int fd)
6254 {
6255 int err = 0;
6256
6257 if (mddev->pers) {
6258 if (!mddev->pers->quiesce || !mddev->thread)
6259 return -EBUSY;
6260 if (mddev->recovery || mddev->sync_thread)
6261 return -EBUSY;
6262 /* we should be able to change the bitmap.. */
6263 }
6264
6265 if (fd >= 0) {
6266 struct inode *inode;
6267 struct file *f;
6268
6269 if (mddev->bitmap || mddev->bitmap_info.file)
6270 return -EEXIST; /* cannot add when bitmap is present */
6271 f = fget(fd);
6272
6273 if (f == NULL) {
6274 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
6275 mdname(mddev));
6276 return -EBADF;
6277 }
6278
6279 inode = f->f_mapping->host;
6280 if (!S_ISREG(inode->i_mode)) {
6281 printk(KERN_ERR "%s: error: bitmap file must be a regular file\n",
6282 mdname(mddev));
6283 err = -EBADF;
6284 } else if (!(f->f_mode & FMODE_WRITE)) {
6285 printk(KERN_ERR "%s: error: bitmap file must open for write\n",
6286 mdname(mddev));
6287 err = -EBADF;
6288 } else if (atomic_read(&inode->i_writecount) != 1) {
6289 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
6290 mdname(mddev));
6291 err = -EBUSY;
6292 }
6293 if (err) {
6294 fput(f);
6295 return err;
6296 }
6297 mddev->bitmap_info.file = f;
6298 mddev->bitmap_info.offset = 0; /* file overrides offset */
6299 } else if (mddev->bitmap == NULL)
6300 return -ENOENT; /* cannot remove what isn't there */
6301 err = 0;
6302 if (mddev->pers) {
6303 mddev->pers->quiesce(mddev, 1);
6304 if (fd >= 0) {
6305 struct bitmap *bitmap;
6306
6307 bitmap = bitmap_create(mddev, -1);
6308 if (!IS_ERR(bitmap)) {
6309 mddev->bitmap = bitmap;
6310 err = bitmap_load(mddev);
6311 } else
6312 err = PTR_ERR(bitmap);
6313 }
6314 if (fd < 0 || err) {
6315 bitmap_destroy(mddev);
6316 fd = -1; /* make sure to put the file */
6317 }
6318 mddev->pers->quiesce(mddev, 0);
6319 }
6320 if (fd < 0) {
6321 struct file *f = mddev->bitmap_info.file;
6322 if (f) {
6323 spin_lock(&mddev->lock);
6324 mddev->bitmap_info.file = NULL;
6325 spin_unlock(&mddev->lock);
6326 fput(f);
6327 }
6328 }
6329
6330 return err;
6331 }
6332
6333 /*
6334 * set_array_info is used two different ways
6335 * The original usage is when creating a new array.
6336 * In this usage, raid_disks is > 0 and it together with
6337 * level, size, not_persistent,layout,chunksize determine the
6338 * shape of the array.
6339 * This will always create an array with a type-0.90.0 superblock.
6340 * The newer usage is when assembling an array.
6341 * In this case raid_disks will be 0, and the major_version field is
6342 * use to determine which style super-blocks are to be found on the devices.
6343 * The minor and patch _version numbers are also kept incase the
6344 * super_block handler wishes to interpret them.
6345 */
6346 static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
6347 {
6348
6349 if (info->raid_disks == 0) {
6350 /* just setting version number for superblock loading */
6351 if (info->major_version < 0 ||
6352 info->major_version >= ARRAY_SIZE(super_types) ||
6353 super_types[info->major_version].name == NULL) {
6354 /* maybe try to auto-load a module? */
6355 printk(KERN_INFO
6356 "md: superblock version %d not known\n",
6357 info->major_version);
6358 return -EINVAL;
6359 }
6360 mddev->major_version = info->major_version;
6361 mddev->minor_version = info->minor_version;
6362 mddev->patch_version = info->patch_version;
6363 mddev->persistent = !info->not_persistent;
6364 /* ensure mddev_put doesn't delete this now that there
6365 * is some minimal configuration.
6366 */
6367 mddev->ctime = ktime_get_real_seconds();
6368 return 0;
6369 }
6370 mddev->major_version = MD_MAJOR_VERSION;
6371 mddev->minor_version = MD_MINOR_VERSION;
6372 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6373 mddev->ctime = ktime_get_real_seconds();
6374
6375 mddev->level = info->level;
6376 mddev->clevel[0] = 0;
6377 mddev->dev_sectors = 2 * (sector_t)info->size;
6378 mddev->raid_disks = info->raid_disks;
6379 /* don't set md_minor, it is determined by which /dev/md* was
6380 * openned
6381 */
6382 if (info->state & (1<<MD_SB_CLEAN))
6383 mddev->recovery_cp = MaxSector;
6384 else
6385 mddev->recovery_cp = 0;
6386 mddev->persistent = ! info->not_persistent;
6387 mddev->external = 0;
6388
6389 mddev->layout = info->layout;
6390 mddev->chunk_sectors = info->chunk_size >> 9;
6391
6392 mddev->max_disks = MD_SB_DISKS;
6393
6394 if (mddev->persistent)
6395 mddev->flags = 0;
6396 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6397
6398 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6399 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6400 mddev->bitmap_info.offset = 0;
6401
6402 mddev->reshape_position = MaxSector;
6403
6404 /*
6405 * Generate a 128 bit UUID
6406 */
6407 get_random_bytes(mddev->uuid, 16);
6408
6409 mddev->new_level = mddev->level;
6410 mddev->new_chunk_sectors = mddev->chunk_sectors;
6411 mddev->new_layout = mddev->layout;
6412 mddev->delta_disks = 0;
6413 mddev->reshape_backwards = 0;
6414
6415 return 0;
6416 }
6417
6418 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6419 {
6420 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6421
6422 if (mddev->external_size)
6423 return;
6424
6425 mddev->array_sectors = array_sectors;
6426 }
6427 EXPORT_SYMBOL(md_set_array_sectors);
6428
6429 static int update_size(struct mddev *mddev, sector_t num_sectors)
6430 {
6431 struct md_rdev *rdev;
6432 int rv;
6433 int fit = (num_sectors == 0);
6434
6435 if (mddev->pers->resize == NULL)
6436 return -EINVAL;
6437 /* The "num_sectors" is the number of sectors of each device that
6438 * is used. This can only make sense for arrays with redundancy.
6439 * linear and raid0 always use whatever space is available. We can only
6440 * consider changing this number if no resync or reconstruction is
6441 * happening, and if the new size is acceptable. It must fit before the
6442 * sb_start or, if that is <data_offset, it must fit before the size
6443 * of each device. If num_sectors is zero, we find the largest size
6444 * that fits.
6445 */
6446 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6447 mddev->sync_thread)
6448 return -EBUSY;
6449 if (mddev->ro)
6450 return -EROFS;
6451
6452 rdev_for_each(rdev, mddev) {
6453 sector_t avail = rdev->sectors;
6454
6455 if (fit && (num_sectors == 0 || num_sectors > avail))
6456 num_sectors = avail;
6457 if (avail < num_sectors)
6458 return -ENOSPC;
6459 }
6460 rv = mddev->pers->resize(mddev, num_sectors);
6461 if (!rv)
6462 revalidate_disk(mddev->gendisk);
6463 return rv;
6464 }
6465
6466 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6467 {
6468 int rv;
6469 struct md_rdev *rdev;
6470 /* change the number of raid disks */
6471 if (mddev->pers->check_reshape == NULL)
6472 return -EINVAL;
6473 if (mddev->ro)
6474 return -EROFS;
6475 if (raid_disks <= 0 ||
6476 (mddev->max_disks && raid_disks >= mddev->max_disks))
6477 return -EINVAL;
6478 if (mddev->sync_thread ||
6479 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6480 mddev->reshape_position != MaxSector)
6481 return -EBUSY;
6482
6483 rdev_for_each(rdev, mddev) {
6484 if (mddev->raid_disks < raid_disks &&
6485 rdev->data_offset < rdev->new_data_offset)
6486 return -EINVAL;
6487 if (mddev->raid_disks > raid_disks &&
6488 rdev->data_offset > rdev->new_data_offset)
6489 return -EINVAL;
6490 }
6491
6492 mddev->delta_disks = raid_disks - mddev->raid_disks;
6493 if (mddev->delta_disks < 0)
6494 mddev->reshape_backwards = 1;
6495 else if (mddev->delta_disks > 0)
6496 mddev->reshape_backwards = 0;
6497
6498 rv = mddev->pers->check_reshape(mddev);
6499 if (rv < 0) {
6500 mddev->delta_disks = 0;
6501 mddev->reshape_backwards = 0;
6502 }
6503 return rv;
6504 }
6505
6506 /*
6507 * update_array_info is used to change the configuration of an
6508 * on-line array.
6509 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6510 * fields in the info are checked against the array.
6511 * Any differences that cannot be handled will cause an error.
6512 * Normally, only one change can be managed at a time.
6513 */
6514 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6515 {
6516 int rv = 0;
6517 int cnt = 0;
6518 int state = 0;
6519
6520 /* calculate expected state,ignoring low bits */
6521 if (mddev->bitmap && mddev->bitmap_info.offset)
6522 state |= (1 << MD_SB_BITMAP_PRESENT);
6523
6524 if (mddev->major_version != info->major_version ||
6525 mddev->minor_version != info->minor_version ||
6526 /* mddev->patch_version != info->patch_version || */
6527 mddev->ctime != info->ctime ||
6528 mddev->level != info->level ||
6529 /* mddev->layout != info->layout || */
6530 mddev->persistent != !info->not_persistent ||
6531 mddev->chunk_sectors != info->chunk_size >> 9 ||
6532 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6533 ((state^info->state) & 0xfffffe00)
6534 )
6535 return -EINVAL;
6536 /* Check there is only one change */
6537 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6538 cnt++;
6539 if (mddev->raid_disks != info->raid_disks)
6540 cnt++;
6541 if (mddev->layout != info->layout)
6542 cnt++;
6543 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6544 cnt++;
6545 if (cnt == 0)
6546 return 0;
6547 if (cnt > 1)
6548 return -EINVAL;
6549
6550 if (mddev->layout != info->layout) {
6551 /* Change layout
6552 * we don't need to do anything at the md level, the
6553 * personality will take care of it all.
6554 */
6555 if (mddev->pers->check_reshape == NULL)
6556 return -EINVAL;
6557 else {
6558 mddev->new_layout = info->layout;
6559 rv = mddev->pers->check_reshape(mddev);
6560 if (rv)
6561 mddev->new_layout = mddev->layout;
6562 return rv;
6563 }
6564 }
6565 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6566 rv = update_size(mddev, (sector_t)info->size * 2);
6567
6568 if (mddev->raid_disks != info->raid_disks)
6569 rv = update_raid_disks(mddev, info->raid_disks);
6570
6571 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6572 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
6573 rv = -EINVAL;
6574 goto err;
6575 }
6576 if (mddev->recovery || mddev->sync_thread) {
6577 rv = -EBUSY;
6578 goto err;
6579 }
6580 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6581 struct bitmap *bitmap;
6582 /* add the bitmap */
6583 if (mddev->bitmap) {
6584 rv = -EEXIST;
6585 goto err;
6586 }
6587 if (mddev->bitmap_info.default_offset == 0) {
6588 rv = -EINVAL;
6589 goto err;
6590 }
6591 mddev->bitmap_info.offset =
6592 mddev->bitmap_info.default_offset;
6593 mddev->bitmap_info.space =
6594 mddev->bitmap_info.default_space;
6595 mddev->pers->quiesce(mddev, 1);
6596 bitmap = bitmap_create(mddev, -1);
6597 if (!IS_ERR(bitmap)) {
6598 mddev->bitmap = bitmap;
6599 rv = bitmap_load(mddev);
6600 } else
6601 rv = PTR_ERR(bitmap);
6602 if (rv)
6603 bitmap_destroy(mddev);
6604 mddev->pers->quiesce(mddev, 0);
6605 } else {
6606 /* remove the bitmap */
6607 if (!mddev->bitmap) {
6608 rv = -ENOENT;
6609 goto err;
6610 }
6611 if (mddev->bitmap->storage.file) {
6612 rv = -EINVAL;
6613 goto err;
6614 }
6615 if (mddev->bitmap_info.nodes) {
6616 /* hold PW on all the bitmap lock */
6617 if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) {
6618 printk("md: can't change bitmap to none since the"
6619 " array is in use by more than one node\n");
6620 rv = -EPERM;
6621 md_cluster_ops->unlock_all_bitmaps(mddev);
6622 goto err;
6623 }
6624
6625 mddev->bitmap_info.nodes = 0;
6626 md_cluster_ops->leave(mddev);
6627 }
6628 mddev->pers->quiesce(mddev, 1);
6629 bitmap_destroy(mddev);
6630 mddev->pers->quiesce(mddev, 0);
6631 mddev->bitmap_info.offset = 0;
6632 }
6633 }
6634 md_update_sb(mddev, 1);
6635 return rv;
6636 err:
6637 return rv;
6638 }
6639
6640 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6641 {
6642 struct md_rdev *rdev;
6643 int err = 0;
6644
6645 if (mddev->pers == NULL)
6646 return -ENODEV;
6647
6648 rcu_read_lock();
6649 rdev = find_rdev_rcu(mddev, dev);
6650 if (!rdev)
6651 err = -ENODEV;
6652 else {
6653 md_error(mddev, rdev);
6654 if (!test_bit(Faulty, &rdev->flags))
6655 err = -EBUSY;
6656 }
6657 rcu_read_unlock();
6658 return err;
6659 }
6660
6661 /*
6662 * We have a problem here : there is no easy way to give a CHS
6663 * virtual geometry. We currently pretend that we have a 2 heads
6664 * 4 sectors (with a BIG number of cylinders...). This drives
6665 * dosfs just mad... ;-)
6666 */
6667 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6668 {
6669 struct mddev *mddev = bdev->bd_disk->private_data;
6670
6671 geo->heads = 2;
6672 geo->sectors = 4;
6673 geo->cylinders = mddev->array_sectors / 8;
6674 return 0;
6675 }
6676
6677 static inline bool md_ioctl_valid(unsigned int cmd)
6678 {
6679 switch (cmd) {
6680 case ADD_NEW_DISK:
6681 case BLKROSET:
6682 case GET_ARRAY_INFO:
6683 case GET_BITMAP_FILE:
6684 case GET_DISK_INFO:
6685 case HOT_ADD_DISK:
6686 case HOT_REMOVE_DISK:
6687 case RAID_AUTORUN:
6688 case RAID_VERSION:
6689 case RESTART_ARRAY_RW:
6690 case RUN_ARRAY:
6691 case SET_ARRAY_INFO:
6692 case SET_BITMAP_FILE:
6693 case SET_DISK_FAULTY:
6694 case STOP_ARRAY:
6695 case STOP_ARRAY_RO:
6696 case CLUSTERED_DISK_NACK:
6697 return true;
6698 default:
6699 return false;
6700 }
6701 }
6702
6703 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6704 unsigned int cmd, unsigned long arg)
6705 {
6706 int err = 0;
6707 void __user *argp = (void __user *)arg;
6708 struct mddev *mddev = NULL;
6709 int ro;
6710
6711 if (!md_ioctl_valid(cmd))
6712 return -ENOTTY;
6713
6714 switch (cmd) {
6715 case RAID_VERSION:
6716 case GET_ARRAY_INFO:
6717 case GET_DISK_INFO:
6718 break;
6719 default:
6720 if (!capable(CAP_SYS_ADMIN))
6721 return -EACCES;
6722 }
6723
6724 /*
6725 * Commands dealing with the RAID driver but not any
6726 * particular array:
6727 */
6728 switch (cmd) {
6729 case RAID_VERSION:
6730 err = get_version(argp);
6731 goto out;
6732
6733 #ifndef MODULE
6734 case RAID_AUTORUN:
6735 err = 0;
6736 autostart_arrays(arg);
6737 goto out;
6738 #endif
6739 default:;
6740 }
6741
6742 /*
6743 * Commands creating/starting a new array:
6744 */
6745
6746 mddev = bdev->bd_disk->private_data;
6747
6748 if (!mddev) {
6749 BUG();
6750 goto out;
6751 }
6752
6753 /* Some actions do not requires the mutex */
6754 switch (cmd) {
6755 case GET_ARRAY_INFO:
6756 if (!mddev->raid_disks && !mddev->external)
6757 err = -ENODEV;
6758 else
6759 err = get_array_info(mddev, argp);
6760 goto out;
6761
6762 case GET_DISK_INFO:
6763 if (!mddev->raid_disks && !mddev->external)
6764 err = -ENODEV;
6765 else
6766 err = get_disk_info(mddev, argp);
6767 goto out;
6768
6769 case SET_DISK_FAULTY:
6770 err = set_disk_faulty(mddev, new_decode_dev(arg));
6771 goto out;
6772
6773 case GET_BITMAP_FILE:
6774 err = get_bitmap_file(mddev, argp);
6775 goto out;
6776
6777 }
6778
6779 if (cmd == ADD_NEW_DISK)
6780 /* need to ensure md_delayed_delete() has completed */
6781 flush_workqueue(md_misc_wq);
6782
6783 if (cmd == HOT_REMOVE_DISK)
6784 /* need to ensure recovery thread has run */
6785 wait_event_interruptible_timeout(mddev->sb_wait,
6786 !test_bit(MD_RECOVERY_NEEDED,
6787 &mddev->flags),
6788 msecs_to_jiffies(5000));
6789 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6790 /* Need to flush page cache, and ensure no-one else opens
6791 * and writes
6792 */
6793 mutex_lock(&mddev->open_mutex);
6794 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
6795 mutex_unlock(&mddev->open_mutex);
6796 err = -EBUSY;
6797 goto out;
6798 }
6799 set_bit(MD_STILL_CLOSED, &mddev->flags);
6800 mutex_unlock(&mddev->open_mutex);
6801 sync_blockdev(bdev);
6802 }
6803 err = mddev_lock(mddev);
6804 if (err) {
6805 printk(KERN_INFO
6806 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6807 err, cmd);
6808 goto out;
6809 }
6810
6811 if (cmd == SET_ARRAY_INFO) {
6812 mdu_array_info_t info;
6813 if (!arg)
6814 memset(&info, 0, sizeof(info));
6815 else if (copy_from_user(&info, argp, sizeof(info))) {
6816 err = -EFAULT;
6817 goto unlock;
6818 }
6819 if (mddev->pers) {
6820 err = update_array_info(mddev, &info);
6821 if (err) {
6822 printk(KERN_WARNING "md: couldn't update"
6823 " array info. %d\n", err);
6824 goto unlock;
6825 }
6826 goto unlock;
6827 }
6828 if (!list_empty(&mddev->disks)) {
6829 printk(KERN_WARNING
6830 "md: array %s already has disks!\n",
6831 mdname(mddev));
6832 err = -EBUSY;
6833 goto unlock;
6834 }
6835 if (mddev->raid_disks) {
6836 printk(KERN_WARNING
6837 "md: array %s already initialised!\n",
6838 mdname(mddev));
6839 err = -EBUSY;
6840 goto unlock;
6841 }
6842 err = set_array_info(mddev, &info);
6843 if (err) {
6844 printk(KERN_WARNING "md: couldn't set"
6845 " array info. %d\n", err);
6846 goto unlock;
6847 }
6848 goto unlock;
6849 }
6850
6851 /*
6852 * Commands querying/configuring an existing array:
6853 */
6854 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6855 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6856 if ((!mddev->raid_disks && !mddev->external)
6857 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6858 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6859 && cmd != GET_BITMAP_FILE) {
6860 err = -ENODEV;
6861 goto unlock;
6862 }
6863
6864 /*
6865 * Commands even a read-only array can execute:
6866 */
6867 switch (cmd) {
6868 case RESTART_ARRAY_RW:
6869 err = restart_array(mddev);
6870 goto unlock;
6871
6872 case STOP_ARRAY:
6873 err = do_md_stop(mddev, 0, bdev);
6874 goto unlock;
6875
6876 case STOP_ARRAY_RO:
6877 err = md_set_readonly(mddev, bdev);
6878 goto unlock;
6879
6880 case HOT_REMOVE_DISK:
6881 err = hot_remove_disk(mddev, new_decode_dev(arg));
6882 goto unlock;
6883
6884 case ADD_NEW_DISK:
6885 /* We can support ADD_NEW_DISK on read-only arrays
6886 * on if we are re-adding a preexisting device.
6887 * So require mddev->pers and MD_DISK_SYNC.
6888 */
6889 if (mddev->pers) {
6890 mdu_disk_info_t info;
6891 if (copy_from_user(&info, argp, sizeof(info)))
6892 err = -EFAULT;
6893 else if (!(info.state & (1<<MD_DISK_SYNC)))
6894 /* Need to clear read-only for this */
6895 break;
6896 else
6897 err = add_new_disk(mddev, &info);
6898 goto unlock;
6899 }
6900 break;
6901
6902 case BLKROSET:
6903 if (get_user(ro, (int __user *)(arg))) {
6904 err = -EFAULT;
6905 goto unlock;
6906 }
6907 err = -EINVAL;
6908
6909 /* if the bdev is going readonly the value of mddev->ro
6910 * does not matter, no writes are coming
6911 */
6912 if (ro)
6913 goto unlock;
6914
6915 /* are we are already prepared for writes? */
6916 if (mddev->ro != 1)
6917 goto unlock;
6918
6919 /* transitioning to readauto need only happen for
6920 * arrays that call md_write_start
6921 */
6922 if (mddev->pers) {
6923 err = restart_array(mddev);
6924 if (err == 0) {
6925 mddev->ro = 2;
6926 set_disk_ro(mddev->gendisk, 0);
6927 }
6928 }
6929 goto unlock;
6930 }
6931
6932 /*
6933 * The remaining ioctls are changing the state of the
6934 * superblock, so we do not allow them on read-only arrays.
6935 */
6936 if (mddev->ro && mddev->pers) {
6937 if (mddev->ro == 2) {
6938 mddev->ro = 0;
6939 sysfs_notify_dirent_safe(mddev->sysfs_state);
6940 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6941 /* mddev_unlock will wake thread */
6942 /* If a device failed while we were read-only, we
6943 * need to make sure the metadata is updated now.
6944 */
6945 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6946 mddev_unlock(mddev);
6947 wait_event(mddev->sb_wait,
6948 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6949 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6950 mddev_lock_nointr(mddev);
6951 }
6952 } else {
6953 err = -EROFS;
6954 goto unlock;
6955 }
6956 }
6957
6958 switch (cmd) {
6959 case ADD_NEW_DISK:
6960 {
6961 mdu_disk_info_t info;
6962 if (copy_from_user(&info, argp, sizeof(info)))
6963 err = -EFAULT;
6964 else
6965 err = add_new_disk(mddev, &info);
6966 goto unlock;
6967 }
6968
6969 case CLUSTERED_DISK_NACK:
6970 if (mddev_is_clustered(mddev))
6971 md_cluster_ops->new_disk_ack(mddev, false);
6972 else
6973 err = -EINVAL;
6974 goto unlock;
6975
6976 case HOT_ADD_DISK:
6977 err = hot_add_disk(mddev, new_decode_dev(arg));
6978 goto unlock;
6979
6980 case RUN_ARRAY:
6981 err = do_md_run(mddev);
6982 goto unlock;
6983
6984 case SET_BITMAP_FILE:
6985 err = set_bitmap_file(mddev, (int)arg);
6986 goto unlock;
6987
6988 default:
6989 err = -EINVAL;
6990 goto unlock;
6991 }
6992
6993 unlock:
6994 if (mddev->hold_active == UNTIL_IOCTL &&
6995 err != -EINVAL)
6996 mddev->hold_active = 0;
6997 mddev_unlock(mddev);
6998 out:
6999 return err;
7000 }
7001 #ifdef CONFIG_COMPAT
7002 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
7003 unsigned int cmd, unsigned long arg)
7004 {
7005 switch (cmd) {
7006 case HOT_REMOVE_DISK:
7007 case HOT_ADD_DISK:
7008 case SET_DISK_FAULTY:
7009 case SET_BITMAP_FILE:
7010 /* These take in integer arg, do not convert */
7011 break;
7012 default:
7013 arg = (unsigned long)compat_ptr(arg);
7014 break;
7015 }
7016
7017 return md_ioctl(bdev, mode, cmd, arg);
7018 }
7019 #endif /* CONFIG_COMPAT */
7020
7021 static int md_open(struct block_device *bdev, fmode_t mode)
7022 {
7023 /*
7024 * Succeed if we can lock the mddev, which confirms that
7025 * it isn't being stopped right now.
7026 */
7027 struct mddev *mddev = mddev_find(bdev->bd_dev);
7028 int err;
7029
7030 if (!mddev)
7031 return -ENODEV;
7032
7033 if (mddev->gendisk != bdev->bd_disk) {
7034 /* we are racing with mddev_put which is discarding this
7035 * bd_disk.
7036 */
7037 mddev_put(mddev);
7038 /* Wait until bdev->bd_disk is definitely gone */
7039 flush_workqueue(md_misc_wq);
7040 /* Then retry the open from the top */
7041 return -ERESTARTSYS;
7042 }
7043 BUG_ON(mddev != bdev->bd_disk->private_data);
7044
7045 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
7046 goto out;
7047
7048 err = 0;
7049 atomic_inc(&mddev->openers);
7050 clear_bit(MD_STILL_CLOSED, &mddev->flags);
7051 mutex_unlock(&mddev->open_mutex);
7052
7053 check_disk_change(bdev);
7054 out:
7055 return err;
7056 }
7057
7058 static void md_release(struct gendisk *disk, fmode_t mode)
7059 {
7060 struct mddev *mddev = disk->private_data;
7061
7062 BUG_ON(!mddev);
7063 atomic_dec(&mddev->openers);
7064 mddev_put(mddev);
7065 }
7066
7067 static int md_media_changed(struct gendisk *disk)
7068 {
7069 struct mddev *mddev = disk->private_data;
7070
7071 return mddev->changed;
7072 }
7073
7074 static int md_revalidate(struct gendisk *disk)
7075 {
7076 struct mddev *mddev = disk->private_data;
7077
7078 mddev->changed = 0;
7079 return 0;
7080 }
7081 static const struct block_device_operations md_fops =
7082 {
7083 .owner = THIS_MODULE,
7084 .open = md_open,
7085 .release = md_release,
7086 .ioctl = md_ioctl,
7087 #ifdef CONFIG_COMPAT
7088 .compat_ioctl = md_compat_ioctl,
7089 #endif
7090 .getgeo = md_getgeo,
7091 .media_changed = md_media_changed,
7092 .revalidate_disk= md_revalidate,
7093 };
7094
7095 static int md_thread(void *arg)
7096 {
7097 struct md_thread *thread = arg;
7098
7099 /*
7100 * md_thread is a 'system-thread', it's priority should be very
7101 * high. We avoid resource deadlocks individually in each
7102 * raid personality. (RAID5 does preallocation) We also use RR and
7103 * the very same RT priority as kswapd, thus we will never get
7104 * into a priority inversion deadlock.
7105 *
7106 * we definitely have to have equal or higher priority than
7107 * bdflush, otherwise bdflush will deadlock if there are too
7108 * many dirty RAID5 blocks.
7109 */
7110
7111 allow_signal(SIGKILL);
7112 while (!kthread_should_stop()) {
7113
7114 /* We need to wait INTERRUPTIBLE so that
7115 * we don't add to the load-average.
7116 * That means we need to be sure no signals are
7117 * pending
7118 */
7119 if (signal_pending(current))
7120 flush_signals(current);
7121
7122 wait_event_interruptible_timeout
7123 (thread->wqueue,
7124 test_bit(THREAD_WAKEUP, &thread->flags)
7125 || kthread_should_stop(),
7126 thread->timeout);
7127
7128 clear_bit(THREAD_WAKEUP, &thread->flags);
7129 if (!kthread_should_stop())
7130 thread->run(thread);
7131 }
7132
7133 return 0;
7134 }
7135
7136 void md_wakeup_thread(struct md_thread *thread)
7137 {
7138 if (thread) {
7139 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
7140 set_bit(THREAD_WAKEUP, &thread->flags);
7141 wake_up(&thread->wqueue);
7142 }
7143 }
7144 EXPORT_SYMBOL(md_wakeup_thread);
7145
7146 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7147 struct mddev *mddev, const char *name)
7148 {
7149 struct md_thread *thread;
7150
7151 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7152 if (!thread)
7153 return NULL;
7154
7155 init_waitqueue_head(&thread->wqueue);
7156
7157 thread->run = run;
7158 thread->mddev = mddev;
7159 thread->timeout = MAX_SCHEDULE_TIMEOUT;
7160 thread->tsk = kthread_run(md_thread, thread,
7161 "%s_%s",
7162 mdname(thread->mddev),
7163 name);
7164 if (IS_ERR(thread->tsk)) {
7165 kfree(thread);
7166 return NULL;
7167 }
7168 return thread;
7169 }
7170 EXPORT_SYMBOL(md_register_thread);
7171
7172 void md_unregister_thread(struct md_thread **threadp)
7173 {
7174 struct md_thread *thread = *threadp;
7175 if (!thread)
7176 return;
7177 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7178 /* Locking ensures that mddev_unlock does not wake_up a
7179 * non-existent thread
7180 */
7181 spin_lock(&pers_lock);
7182 *threadp = NULL;
7183 spin_unlock(&pers_lock);
7184
7185 kthread_stop(thread->tsk);
7186 kfree(thread);
7187 }
7188 EXPORT_SYMBOL(md_unregister_thread);
7189
7190 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7191 {
7192 if (!rdev || test_bit(Faulty, &rdev->flags))
7193 return;
7194
7195 if (!mddev->pers || !mddev->pers->error_handler)
7196 return;
7197 mddev->pers->error_handler(mddev,rdev);
7198 if (mddev->degraded)
7199 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7200 sysfs_notify_dirent_safe(rdev->sysfs_state);
7201 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7202 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7203 md_wakeup_thread(mddev->thread);
7204 if (mddev->event_work.func)
7205 queue_work(md_misc_wq, &mddev->event_work);
7206 md_new_event(mddev);
7207 }
7208 EXPORT_SYMBOL(md_error);
7209
7210 /* seq_file implementation /proc/mdstat */
7211
7212 static void status_unused(struct seq_file *seq)
7213 {
7214 int i = 0;
7215 struct md_rdev *rdev;
7216
7217 seq_printf(seq, "unused devices: ");
7218
7219 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
7220 char b[BDEVNAME_SIZE];
7221 i++;
7222 seq_printf(seq, "%s ",
7223 bdevname(rdev->bdev,b));
7224 }
7225 if (!i)
7226 seq_printf(seq, "<none>");
7227
7228 seq_printf(seq, "\n");
7229 }
7230
7231 static int status_resync(struct seq_file *seq, struct mddev *mddev)
7232 {
7233 sector_t max_sectors, resync, res;
7234 unsigned long dt, db;
7235 sector_t rt;
7236 int scale;
7237 unsigned int per_milli;
7238
7239 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7240 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7241 max_sectors = mddev->resync_max_sectors;
7242 else
7243 max_sectors = mddev->dev_sectors;
7244
7245 resync = mddev->curr_resync;
7246 if (resync <= 3) {
7247 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7248 /* Still cleaning up */
7249 resync = max_sectors;
7250 } else
7251 resync -= atomic_read(&mddev->recovery_active);
7252
7253 if (resync == 0) {
7254 if (mddev->recovery_cp < MaxSector) {
7255 seq_printf(seq, "\tresync=PENDING");
7256 return 1;
7257 }
7258 return 0;
7259 }
7260 if (resync < 3) {
7261 seq_printf(seq, "\tresync=DELAYED");
7262 return 1;
7263 }
7264
7265 WARN_ON(max_sectors == 0);
7266 /* Pick 'scale' such that (resync>>scale)*1000 will fit
7267 * in a sector_t, and (max_sectors>>scale) will fit in a
7268 * u32, as those are the requirements for sector_div.
7269 * Thus 'scale' must be at least 10
7270 */
7271 scale = 10;
7272 if (sizeof(sector_t) > sizeof(unsigned long)) {
7273 while ( max_sectors/2 > (1ULL<<(scale+32)))
7274 scale++;
7275 }
7276 res = (resync>>scale)*1000;
7277 sector_div(res, (u32)((max_sectors>>scale)+1));
7278
7279 per_milli = res;
7280 {
7281 int i, x = per_milli/50, y = 20-x;
7282 seq_printf(seq, "[");
7283 for (i = 0; i < x; i++)
7284 seq_printf(seq, "=");
7285 seq_printf(seq, ">");
7286 for (i = 0; i < y; i++)
7287 seq_printf(seq, ".");
7288 seq_printf(seq, "] ");
7289 }
7290 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
7291 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
7292 "reshape" :
7293 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
7294 "check" :
7295 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
7296 "resync" : "recovery"))),
7297 per_milli/10, per_milli % 10,
7298 (unsigned long long) resync/2,
7299 (unsigned long long) max_sectors/2);
7300
7301 /*
7302 * dt: time from mark until now
7303 * db: blocks written from mark until now
7304 * rt: remaining time
7305 *
7306 * rt is a sector_t, so could be 32bit or 64bit.
7307 * So we divide before multiply in case it is 32bit and close
7308 * to the limit.
7309 * We scale the divisor (db) by 32 to avoid losing precision
7310 * near the end of resync when the number of remaining sectors
7311 * is close to 'db'.
7312 * We then divide rt by 32 after multiplying by db to compensate.
7313 * The '+1' avoids division by zero if db is very small.
7314 */
7315 dt = ((jiffies - mddev->resync_mark) / HZ);
7316 if (!dt) dt++;
7317 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
7318 - mddev->resync_mark_cnt;
7319
7320 rt = max_sectors - resync; /* number of remaining sectors */
7321 sector_div(rt, db/32+1);
7322 rt *= dt;
7323 rt >>= 5;
7324
7325 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7326 ((unsigned long)rt % 60)/6);
7327
7328 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
7329 return 1;
7330 }
7331
7332 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7333 {
7334 struct list_head *tmp;
7335 loff_t l = *pos;
7336 struct mddev *mddev;
7337
7338 if (l >= 0x10000)
7339 return NULL;
7340 if (!l--)
7341 /* header */
7342 return (void*)1;
7343
7344 spin_lock(&all_mddevs_lock);
7345 list_for_each(tmp,&all_mddevs)
7346 if (!l--) {
7347 mddev = list_entry(tmp, struct mddev, all_mddevs);
7348 mddev_get(mddev);
7349 spin_unlock(&all_mddevs_lock);
7350 return mddev;
7351 }
7352 spin_unlock(&all_mddevs_lock);
7353 if (!l--)
7354 return (void*)2;/* tail */
7355 return NULL;
7356 }
7357
7358 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7359 {
7360 struct list_head *tmp;
7361 struct mddev *next_mddev, *mddev = v;
7362
7363 ++*pos;
7364 if (v == (void*)2)
7365 return NULL;
7366
7367 spin_lock(&all_mddevs_lock);
7368 if (v == (void*)1)
7369 tmp = all_mddevs.next;
7370 else
7371 tmp = mddev->all_mddevs.next;
7372 if (tmp != &all_mddevs)
7373 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
7374 else {
7375 next_mddev = (void*)2;
7376 *pos = 0x10000;
7377 }
7378 spin_unlock(&all_mddevs_lock);
7379
7380 if (v != (void*)1)
7381 mddev_put(mddev);
7382 return next_mddev;
7383
7384 }
7385
7386 static void md_seq_stop(struct seq_file *seq, void *v)
7387 {
7388 struct mddev *mddev = v;
7389
7390 if (mddev && v != (void*)1 && v != (void*)2)
7391 mddev_put(mddev);
7392 }
7393
7394 static int md_seq_show(struct seq_file *seq, void *v)
7395 {
7396 struct mddev *mddev = v;
7397 sector_t sectors;
7398 struct md_rdev *rdev;
7399
7400 if (v == (void*)1) {
7401 struct md_personality *pers;
7402 seq_printf(seq, "Personalities : ");
7403 spin_lock(&pers_lock);
7404 list_for_each_entry(pers, &pers_list, list)
7405 seq_printf(seq, "[%s] ", pers->name);
7406
7407 spin_unlock(&pers_lock);
7408 seq_printf(seq, "\n");
7409 seq->poll_event = atomic_read(&md_event_count);
7410 return 0;
7411 }
7412 if (v == (void*)2) {
7413 status_unused(seq);
7414 return 0;
7415 }
7416
7417 spin_lock(&mddev->lock);
7418 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7419 seq_printf(seq, "%s : %sactive", mdname(mddev),
7420 mddev->pers ? "" : "in");
7421 if (mddev->pers) {
7422 if (mddev->ro==1)
7423 seq_printf(seq, " (read-only)");
7424 if (mddev->ro==2)
7425 seq_printf(seq, " (auto-read-only)");
7426 seq_printf(seq, " %s", mddev->pers->name);
7427 }
7428
7429 sectors = 0;
7430 rcu_read_lock();
7431 rdev_for_each_rcu(rdev, mddev) {
7432 char b[BDEVNAME_SIZE];
7433 seq_printf(seq, " %s[%d]",
7434 bdevname(rdev->bdev,b), rdev->desc_nr);
7435 if (test_bit(WriteMostly, &rdev->flags))
7436 seq_printf(seq, "(W)");
7437 if (test_bit(Journal, &rdev->flags))
7438 seq_printf(seq, "(J)");
7439 if (test_bit(Faulty, &rdev->flags)) {
7440 seq_printf(seq, "(F)");
7441 continue;
7442 }
7443 if (rdev->raid_disk < 0)
7444 seq_printf(seq, "(S)"); /* spare */
7445 if (test_bit(Replacement, &rdev->flags))
7446 seq_printf(seq, "(R)");
7447 sectors += rdev->sectors;
7448 }
7449 rcu_read_unlock();
7450
7451 if (!list_empty(&mddev->disks)) {
7452 if (mddev->pers)
7453 seq_printf(seq, "\n %llu blocks",
7454 (unsigned long long)
7455 mddev->array_sectors / 2);
7456 else
7457 seq_printf(seq, "\n %llu blocks",
7458 (unsigned long long)sectors / 2);
7459 }
7460 if (mddev->persistent) {
7461 if (mddev->major_version != 0 ||
7462 mddev->minor_version != 90) {
7463 seq_printf(seq," super %d.%d",
7464 mddev->major_version,
7465 mddev->minor_version);
7466 }
7467 } else if (mddev->external)
7468 seq_printf(seq, " super external:%s",
7469 mddev->metadata_type);
7470 else
7471 seq_printf(seq, " super non-persistent");
7472
7473 if (mddev->pers) {
7474 mddev->pers->status(seq, mddev);
7475 seq_printf(seq, "\n ");
7476 if (mddev->pers->sync_request) {
7477 if (status_resync(seq, mddev))
7478 seq_printf(seq, "\n ");
7479 }
7480 } else
7481 seq_printf(seq, "\n ");
7482
7483 bitmap_status(seq, mddev->bitmap);
7484
7485 seq_printf(seq, "\n");
7486 }
7487 spin_unlock(&mddev->lock);
7488
7489 return 0;
7490 }
7491
7492 static const struct seq_operations md_seq_ops = {
7493 .start = md_seq_start,
7494 .next = md_seq_next,
7495 .stop = md_seq_stop,
7496 .show = md_seq_show,
7497 };
7498
7499 static int md_seq_open(struct inode *inode, struct file *file)
7500 {
7501 struct seq_file *seq;
7502 int error;
7503
7504 error = seq_open(file, &md_seq_ops);
7505 if (error)
7506 return error;
7507
7508 seq = file->private_data;
7509 seq->poll_event = atomic_read(&md_event_count);
7510 return error;
7511 }
7512
7513 static int md_unloading;
7514 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7515 {
7516 struct seq_file *seq = filp->private_data;
7517 int mask;
7518
7519 if (md_unloading)
7520 return POLLIN|POLLRDNORM|POLLERR|POLLPRI;
7521 poll_wait(filp, &md_event_waiters, wait);
7522
7523 /* always allow read */
7524 mask = POLLIN | POLLRDNORM;
7525
7526 if (seq->poll_event != atomic_read(&md_event_count))
7527 mask |= POLLERR | POLLPRI;
7528 return mask;
7529 }
7530
7531 static const struct file_operations md_seq_fops = {
7532 .owner = THIS_MODULE,
7533 .open = md_seq_open,
7534 .read = seq_read,
7535 .llseek = seq_lseek,
7536 .release = seq_release_private,
7537 .poll = mdstat_poll,
7538 };
7539
7540 int register_md_personality(struct md_personality *p)
7541 {
7542 printk(KERN_INFO "md: %s personality registered for level %d\n",
7543 p->name, p->level);
7544 spin_lock(&pers_lock);
7545 list_add_tail(&p->list, &pers_list);
7546 spin_unlock(&pers_lock);
7547 return 0;
7548 }
7549 EXPORT_SYMBOL(register_md_personality);
7550
7551 int unregister_md_personality(struct md_personality *p)
7552 {
7553 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
7554 spin_lock(&pers_lock);
7555 list_del_init(&p->list);
7556 spin_unlock(&pers_lock);
7557 return 0;
7558 }
7559 EXPORT_SYMBOL(unregister_md_personality);
7560
7561 int register_md_cluster_operations(struct md_cluster_operations *ops,
7562 struct module *module)
7563 {
7564 int ret = 0;
7565 spin_lock(&pers_lock);
7566 if (md_cluster_ops != NULL)
7567 ret = -EALREADY;
7568 else {
7569 md_cluster_ops = ops;
7570 md_cluster_mod = module;
7571 }
7572 spin_unlock(&pers_lock);
7573 return ret;
7574 }
7575 EXPORT_SYMBOL(register_md_cluster_operations);
7576
7577 int unregister_md_cluster_operations(void)
7578 {
7579 spin_lock(&pers_lock);
7580 md_cluster_ops = NULL;
7581 spin_unlock(&pers_lock);
7582 return 0;
7583 }
7584 EXPORT_SYMBOL(unregister_md_cluster_operations);
7585
7586 int md_setup_cluster(struct mddev *mddev, int nodes)
7587 {
7588 int err;
7589
7590 err = request_module("md-cluster");
7591 if (err) {
7592 pr_err("md-cluster module not found.\n");
7593 return -ENOENT;
7594 }
7595
7596 spin_lock(&pers_lock);
7597 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
7598 spin_unlock(&pers_lock);
7599 return -ENOENT;
7600 }
7601 spin_unlock(&pers_lock);
7602
7603 return md_cluster_ops->join(mddev, nodes);
7604 }
7605
7606 void md_cluster_stop(struct mddev *mddev)
7607 {
7608 if (!md_cluster_ops)
7609 return;
7610 md_cluster_ops->leave(mddev);
7611 module_put(md_cluster_mod);
7612 }
7613
7614 static int is_mddev_idle(struct mddev *mddev, int init)
7615 {
7616 struct md_rdev *rdev;
7617 int idle;
7618 int curr_events;
7619
7620 idle = 1;
7621 rcu_read_lock();
7622 rdev_for_each_rcu(rdev, mddev) {
7623 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7624 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7625 (int)part_stat_read(&disk->part0, sectors[1]) -
7626 atomic_read(&disk->sync_io);
7627 /* sync IO will cause sync_io to increase before the disk_stats
7628 * as sync_io is counted when a request starts, and
7629 * disk_stats is counted when it completes.
7630 * So resync activity will cause curr_events to be smaller than
7631 * when there was no such activity.
7632 * non-sync IO will cause disk_stat to increase without
7633 * increasing sync_io so curr_events will (eventually)
7634 * be larger than it was before. Once it becomes
7635 * substantially larger, the test below will cause
7636 * the array to appear non-idle, and resync will slow
7637 * down.
7638 * If there is a lot of outstanding resync activity when
7639 * we set last_event to curr_events, then all that activity
7640 * completing might cause the array to appear non-idle
7641 * and resync will be slowed down even though there might
7642 * not have been non-resync activity. This will only
7643 * happen once though. 'last_events' will soon reflect
7644 * the state where there is little or no outstanding
7645 * resync requests, and further resync activity will
7646 * always make curr_events less than last_events.
7647 *
7648 */
7649 if (init || curr_events - rdev->last_events > 64) {
7650 rdev->last_events = curr_events;
7651 idle = 0;
7652 }
7653 }
7654 rcu_read_unlock();
7655 return idle;
7656 }
7657
7658 void md_done_sync(struct mddev *mddev, int blocks, int ok)
7659 {
7660 /* another "blocks" (512byte) blocks have been synced */
7661 atomic_sub(blocks, &mddev->recovery_active);
7662 wake_up(&mddev->recovery_wait);
7663 if (!ok) {
7664 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7665 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
7666 md_wakeup_thread(mddev->thread);
7667 // stop recovery, signal do_sync ....
7668 }
7669 }
7670 EXPORT_SYMBOL(md_done_sync);
7671
7672 /* md_write_start(mddev, bi)
7673 * If we need to update some array metadata (e.g. 'active' flag
7674 * in superblock) before writing, schedule a superblock update
7675 * and wait for it to complete.
7676 */
7677 void md_write_start(struct mddev *mddev, struct bio *bi)
7678 {
7679 int did_change = 0;
7680 if (bio_data_dir(bi) != WRITE)
7681 return;
7682
7683 BUG_ON(mddev->ro == 1);
7684 if (mddev->ro == 2) {
7685 /* need to switch to read/write */
7686 mddev->ro = 0;
7687 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7688 md_wakeup_thread(mddev->thread);
7689 md_wakeup_thread(mddev->sync_thread);
7690 did_change = 1;
7691 }
7692 atomic_inc(&mddev->writes_pending);
7693 if (mddev->safemode == 1)
7694 mddev->safemode = 0;
7695 if (mddev->in_sync) {
7696 spin_lock(&mddev->lock);
7697 if (mddev->in_sync) {
7698 mddev->in_sync = 0;
7699 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7700 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7701 md_wakeup_thread(mddev->thread);
7702 did_change = 1;
7703 }
7704 spin_unlock(&mddev->lock);
7705 }
7706 if (did_change)
7707 sysfs_notify_dirent_safe(mddev->sysfs_state);
7708 wait_event(mddev->sb_wait,
7709 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
7710 }
7711 EXPORT_SYMBOL(md_write_start);
7712
7713 void md_write_end(struct mddev *mddev)
7714 {
7715 if (atomic_dec_and_test(&mddev->writes_pending)) {
7716 if (mddev->safemode == 2)
7717 md_wakeup_thread(mddev->thread);
7718 else if (mddev->safemode_delay)
7719 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7720 }
7721 }
7722 EXPORT_SYMBOL(md_write_end);
7723
7724 /* md_allow_write(mddev)
7725 * Calling this ensures that the array is marked 'active' so that writes
7726 * may proceed without blocking. It is important to call this before
7727 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7728 * Must be called with mddev_lock held.
7729 *
7730 * In the ->external case MD_CHANGE_PENDING can not be cleared until mddev->lock
7731 * is dropped, so return -EAGAIN after notifying userspace.
7732 */
7733 int md_allow_write(struct mddev *mddev)
7734 {
7735 if (!mddev->pers)
7736 return 0;
7737 if (mddev->ro)
7738 return 0;
7739 if (!mddev->pers->sync_request)
7740 return 0;
7741
7742 spin_lock(&mddev->lock);
7743 if (mddev->in_sync) {
7744 mddev->in_sync = 0;
7745 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7746 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7747 if (mddev->safemode_delay &&
7748 mddev->safemode == 0)
7749 mddev->safemode = 1;
7750 spin_unlock(&mddev->lock);
7751 md_update_sb(mddev, 0);
7752 sysfs_notify_dirent_safe(mddev->sysfs_state);
7753 } else
7754 spin_unlock(&mddev->lock);
7755
7756 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
7757 return -EAGAIN;
7758 else
7759 return 0;
7760 }
7761 EXPORT_SYMBOL_GPL(md_allow_write);
7762
7763 #define SYNC_MARKS 10
7764 #define SYNC_MARK_STEP (3*HZ)
7765 #define UPDATE_FREQUENCY (5*60*HZ)
7766 void md_do_sync(struct md_thread *thread)
7767 {
7768 struct mddev *mddev = thread->mddev;
7769 struct mddev *mddev2;
7770 unsigned int currspeed = 0,
7771 window;
7772 sector_t max_sectors,j, io_sectors, recovery_done;
7773 unsigned long mark[SYNC_MARKS];
7774 unsigned long update_time;
7775 sector_t mark_cnt[SYNC_MARKS];
7776 int last_mark,m;
7777 struct list_head *tmp;
7778 sector_t last_check;
7779 int skipped = 0;
7780 struct md_rdev *rdev;
7781 char *desc, *action = NULL;
7782 struct blk_plug plug;
7783 bool cluster_resync_finished = false;
7784
7785 /* just incase thread restarts... */
7786 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7787 return;
7788 if (mddev->ro) {/* never try to sync a read-only array */
7789 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7790 return;
7791 }
7792
7793 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7794 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
7795 desc = "data-check";
7796 action = "check";
7797 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7798 desc = "requested-resync";
7799 action = "repair";
7800 } else
7801 desc = "resync";
7802 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7803 desc = "reshape";
7804 else
7805 desc = "recovery";
7806
7807 mddev->last_sync_action = action ?: desc;
7808
7809 /* we overload curr_resync somewhat here.
7810 * 0 == not engaged in resync at all
7811 * 2 == checking that there is no conflict with another sync
7812 * 1 == like 2, but have yielded to allow conflicting resync to
7813 * commense
7814 * other == active in resync - this many blocks
7815 *
7816 * Before starting a resync we must have set curr_resync to
7817 * 2, and then checked that every "conflicting" array has curr_resync
7818 * less than ours. When we find one that is the same or higher
7819 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7820 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7821 * This will mean we have to start checking from the beginning again.
7822 *
7823 */
7824
7825 do {
7826 mddev->curr_resync = 2;
7827
7828 try_again:
7829 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7830 goto skip;
7831 for_each_mddev(mddev2, tmp) {
7832 if (mddev2 == mddev)
7833 continue;
7834 if (!mddev->parallel_resync
7835 && mddev2->curr_resync
7836 && match_mddev_units(mddev, mddev2)) {
7837 DEFINE_WAIT(wq);
7838 if (mddev < mddev2 && mddev->curr_resync == 2) {
7839 /* arbitrarily yield */
7840 mddev->curr_resync = 1;
7841 wake_up(&resync_wait);
7842 }
7843 if (mddev > mddev2 && mddev->curr_resync == 1)
7844 /* no need to wait here, we can wait the next
7845 * time 'round when curr_resync == 2
7846 */
7847 continue;
7848 /* We need to wait 'interruptible' so as not to
7849 * contribute to the load average, and not to
7850 * be caught by 'softlockup'
7851 */
7852 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7853 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7854 mddev2->curr_resync >= mddev->curr_resync) {
7855 printk(KERN_INFO "md: delaying %s of %s"
7856 " until %s has finished (they"
7857 " share one or more physical units)\n",
7858 desc, mdname(mddev), mdname(mddev2));
7859 mddev_put(mddev2);
7860 if (signal_pending(current))
7861 flush_signals(current);
7862 schedule();
7863 finish_wait(&resync_wait, &wq);
7864 goto try_again;
7865 }
7866 finish_wait(&resync_wait, &wq);
7867 }
7868 }
7869 } while (mddev->curr_resync < 2);
7870
7871 j = 0;
7872 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7873 /* resync follows the size requested by the personality,
7874 * which defaults to physical size, but can be virtual size
7875 */
7876 max_sectors = mddev->resync_max_sectors;
7877 atomic64_set(&mddev->resync_mismatches, 0);
7878 /* we don't use the checkpoint if there's a bitmap */
7879 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7880 j = mddev->resync_min;
7881 else if (!mddev->bitmap)
7882 j = mddev->recovery_cp;
7883
7884 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7885 max_sectors = mddev->resync_max_sectors;
7886 else {
7887 /* recovery follows the physical size of devices */
7888 max_sectors = mddev->dev_sectors;
7889 j = MaxSector;
7890 rcu_read_lock();
7891 rdev_for_each_rcu(rdev, mddev)
7892 if (rdev->raid_disk >= 0 &&
7893 !test_bit(Journal, &rdev->flags) &&
7894 !test_bit(Faulty, &rdev->flags) &&
7895 !test_bit(In_sync, &rdev->flags) &&
7896 rdev->recovery_offset < j)
7897 j = rdev->recovery_offset;
7898 rcu_read_unlock();
7899
7900 /* If there is a bitmap, we need to make sure all
7901 * writes that started before we added a spare
7902 * complete before we start doing a recovery.
7903 * Otherwise the write might complete and (via
7904 * bitmap_endwrite) set a bit in the bitmap after the
7905 * recovery has checked that bit and skipped that
7906 * region.
7907 */
7908 if (mddev->bitmap) {
7909 mddev->pers->quiesce(mddev, 1);
7910 mddev->pers->quiesce(mddev, 0);
7911 }
7912 }
7913
7914 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7915 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7916 " %d KB/sec/disk.\n", speed_min(mddev));
7917 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
7918 "(but not more than %d KB/sec) for %s.\n",
7919 speed_max(mddev), desc);
7920
7921 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
7922
7923 io_sectors = 0;
7924 for (m = 0; m < SYNC_MARKS; m++) {
7925 mark[m] = jiffies;
7926 mark_cnt[m] = io_sectors;
7927 }
7928 last_mark = 0;
7929 mddev->resync_mark = mark[last_mark];
7930 mddev->resync_mark_cnt = mark_cnt[last_mark];
7931
7932 /*
7933 * Tune reconstruction:
7934 */
7935 window = 32*(PAGE_SIZE/512);
7936 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7937 window/2, (unsigned long long)max_sectors/2);
7938
7939 atomic_set(&mddev->recovery_active, 0);
7940 last_check = 0;
7941
7942 if (j>2) {
7943 printk(KERN_INFO
7944 "md: resuming %s of %s from checkpoint.\n",
7945 desc, mdname(mddev));
7946 mddev->curr_resync = j;
7947 } else
7948 mddev->curr_resync = 3; /* no longer delayed */
7949 mddev->curr_resync_completed = j;
7950 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7951 md_new_event(mddev);
7952 update_time = jiffies;
7953
7954 blk_start_plug(&plug);
7955 while (j < max_sectors) {
7956 sector_t sectors;
7957
7958 skipped = 0;
7959
7960 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7961 ((mddev->curr_resync > mddev->curr_resync_completed &&
7962 (mddev->curr_resync - mddev->curr_resync_completed)
7963 > (max_sectors >> 4)) ||
7964 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7965 (j - mddev->curr_resync_completed)*2
7966 >= mddev->resync_max - mddev->curr_resync_completed ||
7967 mddev->curr_resync_completed > mddev->resync_max
7968 )) {
7969 /* time to update curr_resync_completed */
7970 wait_event(mddev->recovery_wait,
7971 atomic_read(&mddev->recovery_active) == 0);
7972 mddev->curr_resync_completed = j;
7973 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7974 j > mddev->recovery_cp)
7975 mddev->recovery_cp = j;
7976 update_time = jiffies;
7977 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7978 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7979 }
7980
7981 while (j >= mddev->resync_max &&
7982 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7983 /* As this condition is controlled by user-space,
7984 * we can block indefinitely, so use '_interruptible'
7985 * to avoid triggering warnings.
7986 */
7987 flush_signals(current); /* just in case */
7988 wait_event_interruptible(mddev->recovery_wait,
7989 mddev->resync_max > j
7990 || test_bit(MD_RECOVERY_INTR,
7991 &mddev->recovery));
7992 }
7993
7994 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7995 break;
7996
7997 sectors = mddev->pers->sync_request(mddev, j, &skipped);
7998 if (sectors == 0) {
7999 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8000 break;
8001 }
8002
8003 if (!skipped) { /* actual IO requested */
8004 io_sectors += sectors;
8005 atomic_add(sectors, &mddev->recovery_active);
8006 }
8007
8008 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8009 break;
8010
8011 j += sectors;
8012 if (j > max_sectors)
8013 /* when skipping, extra large numbers can be returned. */
8014 j = max_sectors;
8015 if (j > 2)
8016 mddev->curr_resync = j;
8017 mddev->curr_mark_cnt = io_sectors;
8018 if (last_check == 0)
8019 /* this is the earliest that rebuild will be
8020 * visible in /proc/mdstat
8021 */
8022 md_new_event(mddev);
8023
8024 if (last_check + window > io_sectors || j == max_sectors)
8025 continue;
8026
8027 last_check = io_sectors;
8028 repeat:
8029 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8030 /* step marks */
8031 int next = (last_mark+1) % SYNC_MARKS;
8032
8033 mddev->resync_mark = mark[next];
8034 mddev->resync_mark_cnt = mark_cnt[next];
8035 mark[next] = jiffies;
8036 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
8037 last_mark = next;
8038 }
8039
8040 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8041 break;
8042
8043 /*
8044 * this loop exits only if either when we are slower than
8045 * the 'hard' speed limit, or the system was IO-idle for
8046 * a jiffy.
8047 * the system might be non-idle CPU-wise, but we only care
8048 * about not overloading the IO subsystem. (things like an
8049 * e2fsck being done on the RAID array should execute fast)
8050 */
8051 cond_resched();
8052
8053 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8054 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
8055 /((jiffies-mddev->resync_mark)/HZ +1) +1;
8056
8057 if (currspeed > speed_min(mddev)) {
8058 if (currspeed > speed_max(mddev)) {
8059 msleep(500);
8060 goto repeat;
8061 }
8062 if (!is_mddev_idle(mddev, 0)) {
8063 /*
8064 * Give other IO more of a chance.
8065 * The faster the devices, the less we wait.
8066 */
8067 wait_event(mddev->recovery_wait,
8068 !atomic_read(&mddev->recovery_active));
8069 }
8070 }
8071 }
8072 printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc,
8073 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
8074 ? "interrupted" : "done");
8075 /*
8076 * this also signals 'finished resyncing' to md_stop
8077 */
8078 blk_finish_plug(&plug);
8079 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
8080
8081 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8082 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8083 mddev->curr_resync > 2) {
8084 mddev->curr_resync_completed = mddev->curr_resync;
8085 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8086 }
8087 /* tell personality and other nodes that we are finished */
8088 if (mddev_is_clustered(mddev)) {
8089 md_cluster_ops->resync_finish(mddev);
8090 cluster_resync_finished = true;
8091 }
8092 mddev->pers->sync_request(mddev, max_sectors, &skipped);
8093
8094 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
8095 mddev->curr_resync > 2) {
8096 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8097 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8098 if (mddev->curr_resync >= mddev->recovery_cp) {
8099 printk(KERN_INFO
8100 "md: checkpointing %s of %s.\n",
8101 desc, mdname(mddev));
8102 if (test_bit(MD_RECOVERY_ERROR,
8103 &mddev->recovery))
8104 mddev->recovery_cp =
8105 mddev->curr_resync_completed;
8106 else
8107 mddev->recovery_cp =
8108 mddev->curr_resync;
8109 }
8110 } else
8111 mddev->recovery_cp = MaxSector;
8112 } else {
8113 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8114 mddev->curr_resync = MaxSector;
8115 rcu_read_lock();
8116 rdev_for_each_rcu(rdev, mddev)
8117 if (rdev->raid_disk >= 0 &&
8118 mddev->delta_disks >= 0 &&
8119 !test_bit(Journal, &rdev->flags) &&
8120 !test_bit(Faulty, &rdev->flags) &&
8121 !test_bit(In_sync, &rdev->flags) &&
8122 rdev->recovery_offset < mddev->curr_resync)
8123 rdev->recovery_offset = mddev->curr_resync;
8124 rcu_read_unlock();
8125 }
8126 }
8127 skip:
8128 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8129
8130 if (mddev_is_clustered(mddev) &&
8131 test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8132 !cluster_resync_finished)
8133 md_cluster_ops->resync_finish(mddev);
8134
8135 spin_lock(&mddev->lock);
8136 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8137 /* We completed so min/max setting can be forgotten if used. */
8138 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8139 mddev->resync_min = 0;
8140 mddev->resync_max = MaxSector;
8141 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8142 mddev->resync_min = mddev->curr_resync_completed;
8143 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
8144 mddev->curr_resync = 0;
8145 spin_unlock(&mddev->lock);
8146
8147 wake_up(&resync_wait);
8148 md_wakeup_thread(mddev->thread);
8149 return;
8150 }
8151 EXPORT_SYMBOL_GPL(md_do_sync);
8152
8153 static int remove_and_add_spares(struct mddev *mddev,
8154 struct md_rdev *this)
8155 {
8156 struct md_rdev *rdev;
8157 int spares = 0;
8158 int removed = 0;
8159
8160 rdev_for_each(rdev, mddev)
8161 if ((this == NULL || rdev == this) &&
8162 rdev->raid_disk >= 0 &&
8163 !test_bit(Blocked, &rdev->flags) &&
8164 (test_bit(Faulty, &rdev->flags) ||
8165 (!test_bit(In_sync, &rdev->flags) &&
8166 !test_bit(Journal, &rdev->flags))) &&
8167 atomic_read(&rdev->nr_pending)==0) {
8168 if (mddev->pers->hot_remove_disk(
8169 mddev, rdev) == 0) {
8170 sysfs_unlink_rdev(mddev, rdev);
8171 rdev->raid_disk = -1;
8172 removed++;
8173 }
8174 }
8175 if (removed && mddev->kobj.sd)
8176 sysfs_notify(&mddev->kobj, NULL, "degraded");
8177
8178 if (this && removed)
8179 goto no_add;
8180
8181 rdev_for_each(rdev, mddev) {
8182 if (this && this != rdev)
8183 continue;
8184 if (test_bit(Candidate, &rdev->flags))
8185 continue;
8186 if (rdev->raid_disk >= 0 &&
8187 !test_bit(In_sync, &rdev->flags) &&
8188 !test_bit(Journal, &rdev->flags) &&
8189 !test_bit(Faulty, &rdev->flags))
8190 spares++;
8191 if (rdev->raid_disk >= 0)
8192 continue;
8193 if (test_bit(Faulty, &rdev->flags))
8194 continue;
8195 if (!test_bit(Journal, &rdev->flags)) {
8196 if (mddev->ro &&
8197 ! (rdev->saved_raid_disk >= 0 &&
8198 !test_bit(Bitmap_sync, &rdev->flags)))
8199 continue;
8200
8201 rdev->recovery_offset = 0;
8202 }
8203 if (mddev->pers->
8204 hot_add_disk(mddev, rdev) == 0) {
8205 if (sysfs_link_rdev(mddev, rdev))
8206 /* failure here is OK */;
8207 if (!test_bit(Journal, &rdev->flags))
8208 spares++;
8209 md_new_event(mddev);
8210 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8211 }
8212 }
8213 no_add:
8214 if (removed)
8215 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8216 return spares;
8217 }
8218
8219 static void md_start_sync(struct work_struct *ws)
8220 {
8221 struct mddev *mddev = container_of(ws, struct mddev, del_work);
8222 int ret = 0;
8223
8224 if (mddev_is_clustered(mddev)) {
8225 ret = md_cluster_ops->resync_start(mddev);
8226 if (ret) {
8227 mddev->sync_thread = NULL;
8228 goto out;
8229 }
8230 }
8231
8232 mddev->sync_thread = md_register_thread(md_do_sync,
8233 mddev,
8234 "resync");
8235 out:
8236 if (!mddev->sync_thread) {
8237 if (!(mddev_is_clustered(mddev) && ret == -EAGAIN))
8238 printk(KERN_ERR "%s: could not start resync"
8239 " thread...\n",
8240 mdname(mddev));
8241 /* leave the spares where they are, it shouldn't hurt */
8242 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8243 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8244 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8245 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8246 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8247 wake_up(&resync_wait);
8248 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8249 &mddev->recovery))
8250 if (mddev->sysfs_action)
8251 sysfs_notify_dirent_safe(mddev->sysfs_action);
8252 } else
8253 md_wakeup_thread(mddev->sync_thread);
8254 sysfs_notify_dirent_safe(mddev->sysfs_action);
8255 md_new_event(mddev);
8256 }
8257
8258 /*
8259 * This routine is regularly called by all per-raid-array threads to
8260 * deal with generic issues like resync and super-block update.
8261 * Raid personalities that don't have a thread (linear/raid0) do not
8262 * need this as they never do any recovery or update the superblock.
8263 *
8264 * It does not do any resync itself, but rather "forks" off other threads
8265 * to do that as needed.
8266 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8267 * "->recovery" and create a thread at ->sync_thread.
8268 * When the thread finishes it sets MD_RECOVERY_DONE
8269 * and wakeups up this thread which will reap the thread and finish up.
8270 * This thread also removes any faulty devices (with nr_pending == 0).
8271 *
8272 * The overall approach is:
8273 * 1/ if the superblock needs updating, update it.
8274 * 2/ If a recovery thread is running, don't do anything else.
8275 * 3/ If recovery has finished, clean up, possibly marking spares active.
8276 * 4/ If there are any faulty devices, remove them.
8277 * 5/ If array is degraded, try to add spares devices
8278 * 6/ If array has spares or is not in-sync, start a resync thread.
8279 */
8280 void md_check_recovery(struct mddev *mddev)
8281 {
8282 if (mddev->suspended)
8283 return;
8284
8285 if (mddev->bitmap)
8286 bitmap_daemon_work(mddev);
8287
8288 if (signal_pending(current)) {
8289 if (mddev->pers->sync_request && !mddev->external) {
8290 printk(KERN_INFO "md: %s in immediate safe mode\n",
8291 mdname(mddev));
8292 mddev->safemode = 2;
8293 }
8294 flush_signals(current);
8295 }
8296
8297 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
8298 return;
8299 if ( ! (
8300 (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) ||
8301 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8302 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8303 test_bit(MD_RELOAD_SB, &mddev->flags) ||
8304 (mddev->external == 0 && mddev->safemode == 1) ||
8305 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
8306 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
8307 ))
8308 return;
8309
8310 if (mddev_trylock(mddev)) {
8311 int spares = 0;
8312
8313 if (mddev->ro) {
8314 struct md_rdev *rdev;
8315 if (!mddev->external && mddev->in_sync)
8316 /* 'Blocked' flag not needed as failed devices
8317 * will be recorded if array switched to read/write.
8318 * Leaving it set will prevent the device
8319 * from being removed.
8320 */
8321 rdev_for_each(rdev, mddev)
8322 clear_bit(Blocked, &rdev->flags);
8323 /* On a read-only array we can:
8324 * - remove failed devices
8325 * - add already-in_sync devices if the array itself
8326 * is in-sync.
8327 * As we only add devices that are already in-sync,
8328 * we can activate the spares immediately.
8329 */
8330 remove_and_add_spares(mddev, NULL);
8331 /* There is no thread, but we need to call
8332 * ->spare_active and clear saved_raid_disk
8333 */
8334 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8335 md_reap_sync_thread(mddev);
8336 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8337 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8338 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
8339 goto unlock;
8340 }
8341
8342 if (mddev_is_clustered(mddev)) {
8343 struct md_rdev *rdev;
8344 /* kick the device if another node issued a
8345 * remove disk.
8346 */
8347 rdev_for_each(rdev, mddev) {
8348 if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
8349 rdev->raid_disk < 0)
8350 md_kick_rdev_from_array(rdev);
8351 }
8352
8353 if (test_and_clear_bit(MD_RELOAD_SB, &mddev->flags))
8354 md_reload_sb(mddev, mddev->good_device_nr);
8355 }
8356
8357 if (!mddev->external) {
8358 int did_change = 0;
8359 spin_lock(&mddev->lock);
8360 if (mddev->safemode &&
8361 !atomic_read(&mddev->writes_pending) &&
8362 !mddev->in_sync &&
8363 mddev->recovery_cp == MaxSector) {
8364 mddev->in_sync = 1;
8365 did_change = 1;
8366 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
8367 }
8368 if (mddev->safemode == 1)
8369 mddev->safemode = 0;
8370 spin_unlock(&mddev->lock);
8371 if (did_change)
8372 sysfs_notify_dirent_safe(mddev->sysfs_state);
8373 }
8374
8375 if (mddev->flags & MD_UPDATE_SB_FLAGS)
8376 md_update_sb(mddev, 0);
8377
8378 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
8379 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
8380 /* resync/recovery still happening */
8381 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8382 goto unlock;
8383 }
8384 if (mddev->sync_thread) {
8385 md_reap_sync_thread(mddev);
8386 goto unlock;
8387 }
8388 /* Set RUNNING before clearing NEEDED to avoid
8389 * any transients in the value of "sync_action".
8390 */
8391 mddev->curr_resync_completed = 0;
8392 spin_lock(&mddev->lock);
8393 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8394 spin_unlock(&mddev->lock);
8395 /* Clear some bits that don't mean anything, but
8396 * might be left set
8397 */
8398 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
8399 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8400
8401 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8402 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
8403 goto not_running;
8404 /* no recovery is running.
8405 * remove any failed drives, then
8406 * add spares if possible.
8407 * Spares are also removed and re-added, to allow
8408 * the personality to fail the re-add.
8409 */
8410
8411 if (mddev->reshape_position != MaxSector) {
8412 if (mddev->pers->check_reshape == NULL ||
8413 mddev->pers->check_reshape(mddev) != 0)
8414 /* Cannot proceed */
8415 goto not_running;
8416 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8417 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8418 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
8419 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8420 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8421 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8422 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8423 } else if (mddev->recovery_cp < MaxSector) {
8424 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8425 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8426 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
8427 /* nothing to be done ... */
8428 goto not_running;
8429
8430 if (mddev->pers->sync_request) {
8431 if (spares) {
8432 /* We are adding a device or devices to an array
8433 * which has the bitmap stored on all devices.
8434 * So make sure all bitmap pages get written
8435 */
8436 bitmap_write_all(mddev->bitmap);
8437 }
8438 INIT_WORK(&mddev->del_work, md_start_sync);
8439 queue_work(md_misc_wq, &mddev->del_work);
8440 goto unlock;
8441 }
8442 not_running:
8443 if (!mddev->sync_thread) {
8444 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8445 wake_up(&resync_wait);
8446 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8447 &mddev->recovery))
8448 if (mddev->sysfs_action)
8449 sysfs_notify_dirent_safe(mddev->sysfs_action);
8450 }
8451 unlock:
8452 wake_up(&mddev->sb_wait);
8453 mddev_unlock(mddev);
8454 }
8455 }
8456 EXPORT_SYMBOL(md_check_recovery);
8457
8458 void md_reap_sync_thread(struct mddev *mddev)
8459 {
8460 struct md_rdev *rdev;
8461
8462 /* resync has finished, collect result */
8463 md_unregister_thread(&mddev->sync_thread);
8464 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8465 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8466 /* success...*/
8467 /* activate any spares */
8468 if (mddev->pers->spare_active(mddev)) {
8469 sysfs_notify(&mddev->kobj, NULL,
8470 "degraded");
8471 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8472 }
8473 }
8474 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8475 mddev->pers->finish_reshape)
8476 mddev->pers->finish_reshape(mddev);
8477
8478 /* If array is no-longer degraded, then any saved_raid_disk
8479 * information must be scrapped.
8480 */
8481 if (!mddev->degraded)
8482 rdev_for_each(rdev, mddev)
8483 rdev->saved_raid_disk = -1;
8484
8485 md_update_sb(mddev, 1);
8486 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8487 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8488 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8489 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8490 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8491 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8492 wake_up(&resync_wait);
8493 /* flag recovery needed just to double check */
8494 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8495 sysfs_notify_dirent_safe(mddev->sysfs_action);
8496 md_new_event(mddev);
8497 if (mddev->event_work.func)
8498 queue_work(md_misc_wq, &mddev->event_work);
8499 }
8500 EXPORT_SYMBOL(md_reap_sync_thread);
8501
8502 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
8503 {
8504 sysfs_notify_dirent_safe(rdev->sysfs_state);
8505 wait_event_timeout(rdev->blocked_wait,
8506 !test_bit(Blocked, &rdev->flags) &&
8507 !test_bit(BlockedBadBlocks, &rdev->flags),
8508 msecs_to_jiffies(5000));
8509 rdev_dec_pending(rdev, mddev);
8510 }
8511 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8512
8513 void md_finish_reshape(struct mddev *mddev)
8514 {
8515 /* called be personality module when reshape completes. */
8516 struct md_rdev *rdev;
8517
8518 rdev_for_each(rdev, mddev) {
8519 if (rdev->data_offset > rdev->new_data_offset)
8520 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8521 else
8522 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8523 rdev->data_offset = rdev->new_data_offset;
8524 }
8525 }
8526 EXPORT_SYMBOL(md_finish_reshape);
8527
8528 /* Bad block management */
8529
8530 /* Returns 1 on success, 0 on failure */
8531 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8532 int is_new)
8533 {
8534 int rv;
8535 if (is_new)
8536 s += rdev->new_data_offset;
8537 else
8538 s += rdev->data_offset;
8539 rv = badblocks_set(&rdev->badblocks, s, sectors, 0);
8540 if (rv == 0) {
8541 /* Make sure they get written out promptly */
8542 sysfs_notify_dirent_safe(rdev->sysfs_state);
8543 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8544 set_bit(MD_CHANGE_PENDING, &rdev->mddev->flags);
8545 md_wakeup_thread(rdev->mddev->thread);
8546 return 1;
8547 } else
8548 return 0;
8549 }
8550 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8551
8552 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8553 int is_new)
8554 {
8555 if (is_new)
8556 s += rdev->new_data_offset;
8557 else
8558 s += rdev->data_offset;
8559 return badblocks_clear(&rdev->badblocks,
8560 s, sectors);
8561 }
8562 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8563
8564 static int md_notify_reboot(struct notifier_block *this,
8565 unsigned long code, void *x)
8566 {
8567 struct list_head *tmp;
8568 struct mddev *mddev;
8569 int need_delay = 0;
8570
8571 for_each_mddev(mddev, tmp) {
8572 if (mddev_trylock(mddev)) {
8573 if (mddev->pers)
8574 __md_stop_writes(mddev);
8575 if (mddev->persistent)
8576 mddev->safemode = 2;
8577 mddev_unlock(mddev);
8578 }
8579 need_delay = 1;
8580 }
8581 /*
8582 * certain more exotic SCSI devices are known to be
8583 * volatile wrt too early system reboots. While the
8584 * right place to handle this issue is the given
8585 * driver, we do want to have a safe RAID driver ...
8586 */
8587 if (need_delay)
8588 mdelay(1000*1);
8589
8590 return NOTIFY_DONE;
8591 }
8592
8593 static struct notifier_block md_notifier = {
8594 .notifier_call = md_notify_reboot,
8595 .next = NULL,
8596 .priority = INT_MAX, /* before any real devices */
8597 };
8598
8599 static void md_geninit(void)
8600 {
8601 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
8602
8603 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
8604 }
8605
8606 static int __init md_init(void)
8607 {
8608 int ret = -ENOMEM;
8609
8610 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
8611 if (!md_wq)
8612 goto err_wq;
8613
8614 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8615 if (!md_misc_wq)
8616 goto err_misc_wq;
8617
8618 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8619 goto err_md;
8620
8621 if ((ret = register_blkdev(0, "mdp")) < 0)
8622 goto err_mdp;
8623 mdp_major = ret;
8624
8625 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
8626 md_probe, NULL, NULL);
8627 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
8628 md_probe, NULL, NULL);
8629
8630 register_reboot_notifier(&md_notifier);
8631 raid_table_header = register_sysctl_table(raid_root_table);
8632
8633 md_geninit();
8634 return 0;
8635
8636 err_mdp:
8637 unregister_blkdev(MD_MAJOR, "md");
8638 err_md:
8639 destroy_workqueue(md_misc_wq);
8640 err_misc_wq:
8641 destroy_workqueue(md_wq);
8642 err_wq:
8643 return ret;
8644 }
8645
8646 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
8647 {
8648 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
8649 struct md_rdev *rdev2;
8650 int role, ret;
8651 char b[BDEVNAME_SIZE];
8652
8653 /* Check for change of roles in the active devices */
8654 rdev_for_each(rdev2, mddev) {
8655 if (test_bit(Faulty, &rdev2->flags))
8656 continue;
8657
8658 /* Check if the roles changed */
8659 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
8660
8661 if (test_bit(Candidate, &rdev2->flags)) {
8662 if (role == 0xfffe) {
8663 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b));
8664 md_kick_rdev_from_array(rdev2);
8665 continue;
8666 }
8667 else
8668 clear_bit(Candidate, &rdev2->flags);
8669 }
8670
8671 if (role != rdev2->raid_disk) {
8672 /* got activated */
8673 if (rdev2->raid_disk == -1 && role != 0xffff) {
8674 rdev2->saved_raid_disk = role;
8675 ret = remove_and_add_spares(mddev, rdev2);
8676 pr_info("Activated spare: %s\n",
8677 bdevname(rdev2->bdev,b));
8678 }
8679 /* device faulty
8680 * We just want to do the minimum to mark the disk
8681 * as faulty. The recovery is performed by the
8682 * one who initiated the error.
8683 */
8684 if ((role == 0xfffe) || (role == 0xfffd)) {
8685 md_error(mddev, rdev2);
8686 clear_bit(Blocked, &rdev2->flags);
8687 }
8688 }
8689 }
8690
8691 if (mddev->raid_disks != le32_to_cpu(sb->raid_disks))
8692 update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
8693
8694 /* Finally set the event to be up to date */
8695 mddev->events = le64_to_cpu(sb->events);
8696 }
8697
8698 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
8699 {
8700 int err;
8701 struct page *swapout = rdev->sb_page;
8702 struct mdp_superblock_1 *sb;
8703
8704 /* Store the sb page of the rdev in the swapout temporary
8705 * variable in case we err in the future
8706 */
8707 rdev->sb_page = NULL;
8708 alloc_disk_sb(rdev);
8709 ClearPageUptodate(rdev->sb_page);
8710 rdev->sb_loaded = 0;
8711 err = super_types[mddev->major_version].load_super(rdev, NULL, mddev->minor_version);
8712
8713 if (err < 0) {
8714 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
8715 __func__, __LINE__, rdev->desc_nr, err);
8716 put_page(rdev->sb_page);
8717 rdev->sb_page = swapout;
8718 rdev->sb_loaded = 1;
8719 return err;
8720 }
8721
8722 sb = page_address(rdev->sb_page);
8723 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
8724 * is not set
8725 */
8726
8727 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
8728 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
8729
8730 /* The other node finished recovery, call spare_active to set
8731 * device In_sync and mddev->degraded
8732 */
8733 if (rdev->recovery_offset == MaxSector &&
8734 !test_bit(In_sync, &rdev->flags) &&
8735 mddev->pers->spare_active(mddev))
8736 sysfs_notify(&mddev->kobj, NULL, "degraded");
8737
8738 put_page(swapout);
8739 return 0;
8740 }
8741
8742 void md_reload_sb(struct mddev *mddev, int nr)
8743 {
8744 struct md_rdev *rdev;
8745 int err;
8746
8747 /* Find the rdev */
8748 rdev_for_each_rcu(rdev, mddev) {
8749 if (rdev->desc_nr == nr)
8750 break;
8751 }
8752
8753 if (!rdev || rdev->desc_nr != nr) {
8754 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
8755 return;
8756 }
8757
8758 err = read_rdev(mddev, rdev);
8759 if (err < 0)
8760 return;
8761
8762 check_sb_changes(mddev, rdev);
8763
8764 /* Read all rdev's to update recovery_offset */
8765 rdev_for_each_rcu(rdev, mddev)
8766 read_rdev(mddev, rdev);
8767 }
8768 EXPORT_SYMBOL(md_reload_sb);
8769
8770 #ifndef MODULE
8771
8772 /*
8773 * Searches all registered partitions for autorun RAID arrays
8774 * at boot time.
8775 */
8776
8777 static LIST_HEAD(all_detected_devices);
8778 struct detected_devices_node {
8779 struct list_head list;
8780 dev_t dev;
8781 };
8782
8783 void md_autodetect_dev(dev_t dev)
8784 {
8785 struct detected_devices_node *node_detected_dev;
8786
8787 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8788 if (node_detected_dev) {
8789 node_detected_dev->dev = dev;
8790 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8791 } else {
8792 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8793 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8794 }
8795 }
8796
8797 static void autostart_arrays(int part)
8798 {
8799 struct md_rdev *rdev;
8800 struct detected_devices_node *node_detected_dev;
8801 dev_t dev;
8802 int i_scanned, i_passed;
8803
8804 i_scanned = 0;
8805 i_passed = 0;
8806
8807 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
8808
8809 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8810 i_scanned++;
8811 node_detected_dev = list_entry(all_detected_devices.next,
8812 struct detected_devices_node, list);
8813 list_del(&node_detected_dev->list);
8814 dev = node_detected_dev->dev;
8815 kfree(node_detected_dev);
8816 rdev = md_import_device(dev,0, 90);
8817 if (IS_ERR(rdev))
8818 continue;
8819
8820 if (test_bit(Faulty, &rdev->flags))
8821 continue;
8822
8823 set_bit(AutoDetected, &rdev->flags);
8824 list_add(&rdev->same_set, &pending_raid_disks);
8825 i_passed++;
8826 }
8827
8828 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8829 i_scanned, i_passed);
8830
8831 autorun_devices(part);
8832 }
8833
8834 #endif /* !MODULE */
8835
8836 static __exit void md_exit(void)
8837 {
8838 struct mddev *mddev;
8839 struct list_head *tmp;
8840 int delay = 1;
8841
8842 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
8843 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
8844
8845 unregister_blkdev(MD_MAJOR,"md");
8846 unregister_blkdev(mdp_major, "mdp");
8847 unregister_reboot_notifier(&md_notifier);
8848 unregister_sysctl_table(raid_table_header);
8849
8850 /* We cannot unload the modules while some process is
8851 * waiting for us in select() or poll() - wake them up
8852 */
8853 md_unloading = 1;
8854 while (waitqueue_active(&md_event_waiters)) {
8855 /* not safe to leave yet */
8856 wake_up(&md_event_waiters);
8857 msleep(delay);
8858 delay += delay;
8859 }
8860 remove_proc_entry("mdstat", NULL);
8861
8862 for_each_mddev(mddev, tmp) {
8863 export_array(mddev);
8864 mddev->hold_active = 0;
8865 }
8866 destroy_workqueue(md_misc_wq);
8867 destroy_workqueue(md_wq);
8868 }
8869
8870 subsys_initcall(md_init);
8871 module_exit(md_exit)
8872
8873 static int get_ro(char *buffer, struct kernel_param *kp)
8874 {
8875 return sprintf(buffer, "%d", start_readonly);
8876 }
8877 static int set_ro(const char *val, struct kernel_param *kp)
8878 {
8879 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
8880 }
8881
8882 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8883 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
8884 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
8885
8886 MODULE_LICENSE("GPL");
8887 MODULE_DESCRIPTION("MD RAID framework");
8888 MODULE_ALIAS("md");
8889 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
(deprecated) Btrfs devel tree