?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
dm thin metadata: fix __udivdi3 undefined on 32-bit
[linux/fpc-iii.git] / drivers / md / md.c
blob07f307402351b79ed121b0d07e13b9762e3d5c69
1 /*
2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
4
5 completely rewritten, based on the MD driver code from Marc Zyngier
6
7 Changes:
8
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
19
20 Neil Brown <neilb@cse.unsw.edu.au>.
21
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
28 any later version.
29
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33 */
34
35 #include <linux/kthread.h>
36 #include <linux/blkdev.h>
37 #include <linux/sysctl.h>
38 #include <linux/seq_file.h>
39 #include <linux/fs.h>
40 #include <linux/poll.h>
41 #include <linux/ctype.h>
42 #include <linux/string.h>
43 #include <linux/hdreg.h>
44 #include <linux/proc_fs.h>
45 #include <linux/random.h>
46 #include <linux/module.h>
47 #include <linux/reboot.h>
48 #include <linux/file.h>
49 #include <linux/compat.h>
50 #include <linux/delay.h>
51 #include <linux/raid/md_p.h>
52 #include <linux/raid/md_u.h>
53 #include <linux/slab.h>
54 #include "md.h"
55 #include "bitmap.h"
56 #include "md-cluster.h"
57
58 #ifndef MODULE
59 static void autostart_arrays(int part);
60 #endif
61
62 /* pers_list is a list of registered personalities protected
63 * by pers_lock.
64 * pers_lock does extra service to protect accesses to
65 * mddev->thread when the mutex cannot be held.
66 */
67 static LIST_HEAD(pers_list);
68 static DEFINE_SPINLOCK(pers_lock);
69
70 struct md_cluster_operations *md_cluster_ops;
71 EXPORT_SYMBOL(md_cluster_ops);
72 struct module *md_cluster_mod;
73 EXPORT_SYMBOL(md_cluster_mod);
74
75 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
76 static struct workqueue_struct *md_wq;
77 static struct workqueue_struct *md_misc_wq;
78
79 static int remove_and_add_spares(struct mddev *mddev,
80 struct md_rdev *this);
81 static void mddev_detach(struct mddev *mddev);
82
83 /*
84 * Default number of read corrections we'll attempt on an rdev
85 * before ejecting it from the array. We divide the read error
86 * count by 2 for every hour elapsed between read errors.
87 */
88 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
89 /*
90 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
91 * is 1000 KB/sec, so the extra system load does not show up that much.
92 * Increase it if you want to have more _guaranteed_ speed. Note that
93 * the RAID driver will use the maximum available bandwidth if the IO
94 * subsystem is idle. There is also an 'absolute maximum' reconstruction
95 * speed limit - in case reconstruction slows down your system despite
96 * idle IO detection.
97 *
98 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
99 * or /sys/block/mdX/md/sync_speed_{min,max}
100 */
101
102 static int sysctl_speed_limit_min = 1000;
103 static int sysctl_speed_limit_max = 200000;
104 static inline int speed_min(struct mddev *mddev)
105 {
106 return mddev->sync_speed_min ?
107 mddev->sync_speed_min : sysctl_speed_limit_min;
108 }
109
110 static inline int speed_max(struct mddev *mddev)
111 {
112 return mddev->sync_speed_max ?
113 mddev->sync_speed_max : sysctl_speed_limit_max;
114 }
115
116 static struct ctl_table_header *raid_table_header;
117
118 static struct ctl_table raid_table[] = {
119 {
120 .procname = "speed_limit_min",
121 .data = &sysctl_speed_limit_min,
122 .maxlen = sizeof(int),
123 .mode = S_IRUGO|S_IWUSR,
124 .proc_handler = proc_dointvec,
125 },
126 {
127 .procname = "speed_limit_max",
128 .data = &sysctl_speed_limit_max,
129 .maxlen = sizeof(int),
130 .mode = S_IRUGO|S_IWUSR,
131 .proc_handler = proc_dointvec,
132 },
133 { }
134 };
135
136 static struct ctl_table raid_dir_table[] = {
137 {
138 .procname = "raid",
139 .maxlen = 0,
140 .mode = S_IRUGO|S_IXUGO,
141 .child = raid_table,
142 },
143 { }
144 };
145
146 static struct ctl_table raid_root_table[] = {
147 {
148 .procname = "dev",
149 .maxlen = 0,
150 .mode = 0555,
151 .child = raid_dir_table,
152 },
153 { }
154 };
155
156 static const struct block_device_operations md_fops;
157
158 static int start_readonly;
159
160 /* bio_clone_mddev
161 * like bio_clone, but with a local bio set
162 */
163
164 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
165 struct mddev *mddev)
166 {
167 struct bio *b;
168
169 if (!mddev || !mddev->bio_set)
170 return bio_alloc(gfp_mask, nr_iovecs);
171
172 b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set);
173 if (!b)
174 return NULL;
175 return b;
176 }
177 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
178
179 struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
180 struct mddev *mddev)
181 {
182 if (!mddev || !mddev->bio_set)
183 return bio_clone(bio, gfp_mask);
184
185 return bio_clone_bioset(bio, gfp_mask, mddev->bio_set);
186 }
187 EXPORT_SYMBOL_GPL(bio_clone_mddev);
188
189 /*
190 * We have a system wide 'event count' that is incremented
191 * on any 'interesting' event, and readers of /proc/mdstat
192 * can use 'poll' or 'select' to find out when the event
193 * count increases.
194 *
195 * Events are:
196 * start array, stop array, error, add device, remove device,
197 * start build, activate spare
198 */
199 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
200 static atomic_t md_event_count;
201 void md_new_event(struct mddev *mddev)
202 {
203 atomic_inc(&md_event_count);
204 wake_up(&md_event_waiters);
205 }
206 EXPORT_SYMBOL_GPL(md_new_event);
207
208 /* Alternate version that can be called from interrupts
209 * when calling sysfs_notify isn't needed.
210 */
211 static void md_new_event_inintr(struct mddev *mddev)
212 {
213 atomic_inc(&md_event_count);
214 wake_up(&md_event_waiters);
215 }
216
217 /*
218 * Enables to iterate over all existing md arrays
219 * all_mddevs_lock protects this list.
220 */
221 static LIST_HEAD(all_mddevs);
222 static DEFINE_SPINLOCK(all_mddevs_lock);
223
224 /*
225 * iterates through all used mddevs in the system.
226 * We take care to grab the all_mddevs_lock whenever navigating
227 * the list, and to always hold a refcount when unlocked.
228 * Any code which breaks out of this loop while own
229 * a reference to the current mddev and must mddev_put it.
230 */
231 #define for_each_mddev(_mddev,_tmp) \
232 \
233 for (({ spin_lock(&all_mddevs_lock); \
234 _tmp = all_mddevs.next; \
235 _mddev = NULL;}); \
236 ({ if (_tmp != &all_mddevs) \
237 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
238 spin_unlock(&all_mddevs_lock); \
239 if (_mddev) mddev_put(_mddev); \
240 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
241 _tmp != &all_mddevs;}); \
242 ({ spin_lock(&all_mddevs_lock); \
243 _tmp = _tmp->next;}) \
244 )
245
246 /* Rather than calling directly into the personality make_request function,
247 * IO requests come here first so that we can check if the device is
248 * being suspended pending a reconfiguration.
249 * We hold a refcount over the call to ->make_request. By the time that
250 * call has finished, the bio has been linked into some internal structure
251 * and so is visible to ->quiesce(), so we don't need the refcount any more.
252 */
253 static blk_qc_t md_make_request(struct request_queue *q, struct bio *bio)
254 {
255 const int rw = bio_data_dir(bio);
256 struct mddev *mddev = q->queuedata;
257 unsigned int sectors;
258 int cpu;
259
260 blk_queue_split(q, &bio, q->bio_split);
261
262 if (mddev == NULL || mddev->pers == NULL
263 || !mddev->ready) {
264 bio_io_error(bio);
265 return BLK_QC_T_NONE;
266 }
267 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
268 if (bio_sectors(bio) != 0)
269 bio->bi_error = -EROFS;
270 bio_endio(bio);
271 return BLK_QC_T_NONE;
272 }
273 smp_rmb(); /* Ensure implications of 'active' are visible */
274 rcu_read_lock();
275 if (mddev->suspended) {
276 DEFINE_WAIT(__wait);
277 for (;;) {
278 prepare_to_wait(&mddev->sb_wait, &__wait,
279 TASK_UNINTERRUPTIBLE);
280 if (!mddev->suspended)
281 break;
282 rcu_read_unlock();
283 schedule();
284 rcu_read_lock();
285 }
286 finish_wait(&mddev->sb_wait, &__wait);
287 }
288 atomic_inc(&mddev->active_io);
289 rcu_read_unlock();
290
291 /*
292 * save the sectors now since our bio can
293 * go away inside make_request
294 */
295 sectors = bio_sectors(bio);
296 /* bio could be mergeable after passing to underlayer */
297 bio->bi_rw &= ~REQ_NOMERGE;
298 mddev->pers->make_request(mddev, bio);
299
300 cpu = part_stat_lock();
301 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
302 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
303 part_stat_unlock();
304
305 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
306 wake_up(&mddev->sb_wait);
307
308 return BLK_QC_T_NONE;
309 }
310
311 /* mddev_suspend makes sure no new requests are submitted
312 * to the device, and that any requests that have been submitted
313 * are completely handled.
314 * Once mddev_detach() is called and completes, the module will be
315 * completely unused.
316 */
317 void mddev_suspend(struct mddev *mddev)
318 {
319 if (mddev->suspended++)
320 return;
321 synchronize_rcu();
322 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
323 mddev->pers->quiesce(mddev, 1);
324
325 del_timer_sync(&mddev->safemode_timer);
326 }
327 EXPORT_SYMBOL_GPL(mddev_suspend);
328
329 void mddev_resume(struct mddev *mddev)
330 {
331 if (--mddev->suspended)
332 return;
333 wake_up(&mddev->sb_wait);
334 mddev->pers->quiesce(mddev, 0);
335
336 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
337 md_wakeup_thread(mddev->thread);
338 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
339 }
340 EXPORT_SYMBOL_GPL(mddev_resume);
341
342 int mddev_congested(struct mddev *mddev, int bits)
343 {
344 struct md_personality *pers = mddev->pers;
345 int ret = 0;
346
347 rcu_read_lock();
348 if (mddev->suspended)
349 ret = 1;
350 else if (pers && pers->congested)
351 ret = pers->congested(mddev, bits);
352 rcu_read_unlock();
353 return ret;
354 }
355 EXPORT_SYMBOL_GPL(mddev_congested);
356 static int md_congested(void *data, int bits)
357 {
358 struct mddev *mddev = data;
359 return mddev_congested(mddev, bits);
360 }
361
362 /*
363 * Generic flush handling for md
364 */
365
366 static void md_end_flush(struct bio *bio)
367 {
368 struct md_rdev *rdev = bio->bi_private;
369 struct mddev *mddev = rdev->mddev;
370
371 rdev_dec_pending(rdev, mddev);
372
373 if (atomic_dec_and_test(&mddev->flush_pending)) {
374 /* The pre-request flush has finished */
375 queue_work(md_wq, &mddev->flush_work);
376 }
377 bio_put(bio);
378 }
379
380 static void md_submit_flush_data(struct work_struct *ws);
381
382 static void submit_flushes(struct work_struct *ws)
383 {
384 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
385 struct md_rdev *rdev;
386
387 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
388 atomic_set(&mddev->flush_pending, 1);
389 rcu_read_lock();
390 rdev_for_each_rcu(rdev, mddev)
391 if (rdev->raid_disk >= 0 &&
392 !test_bit(Faulty, &rdev->flags)) {
393 /* Take two references, one is dropped
394 * when request finishes, one after
395 * we reclaim rcu_read_lock
396 */
397 struct bio *bi;
398 atomic_inc(&rdev->nr_pending);
399 atomic_inc(&rdev->nr_pending);
400 rcu_read_unlock();
401 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
402 bi->bi_end_io = md_end_flush;
403 bi->bi_private = rdev;
404 bi->bi_bdev = rdev->bdev;
405 atomic_inc(&mddev->flush_pending);
406 submit_bio(WRITE_FLUSH, bi);
407 rcu_read_lock();
408 rdev_dec_pending(rdev, mddev);
409 }
410 rcu_read_unlock();
411 if (atomic_dec_and_test(&mddev->flush_pending))
412 queue_work(md_wq, &mddev->flush_work);
413 }
414
415 static void md_submit_flush_data(struct work_struct *ws)
416 {
417 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
418 struct bio *bio = mddev->flush_bio;
419
420 if (bio->bi_iter.bi_size == 0)
421 /* an empty barrier - all done */
422 bio_endio(bio);
423 else {
424 bio->bi_rw &= ~REQ_FLUSH;
425 mddev->pers->make_request(mddev, bio);
426 }
427
428 mddev->flush_bio = NULL;
429 wake_up(&mddev->sb_wait);
430 }
431
432 void md_flush_request(struct mddev *mddev, struct bio *bio)
433 {
434 spin_lock_irq(&mddev->lock);
435 wait_event_lock_irq(mddev->sb_wait,
436 !mddev->flush_bio,
437 mddev->lock);
438 mddev->flush_bio = bio;
439 spin_unlock_irq(&mddev->lock);
440
441 INIT_WORK(&mddev->flush_work, submit_flushes);
442 queue_work(md_wq, &mddev->flush_work);
443 }
444 EXPORT_SYMBOL(md_flush_request);
445
446 void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
447 {
448 struct mddev *mddev = cb->data;
449 md_wakeup_thread(mddev->thread);
450 kfree(cb);
451 }
452 EXPORT_SYMBOL(md_unplug);
453
454 static inline struct mddev *mddev_get(struct mddev *mddev)
455 {
456 atomic_inc(&mddev->active);
457 return mddev;
458 }
459
460 static void mddev_delayed_delete(struct work_struct *ws);
461
462 static void mddev_put(struct mddev *mddev)
463 {
464 struct bio_set *bs = NULL;
465
466 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
467 return;
468 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
469 mddev->ctime == 0 && !mddev->hold_active) {
470 /* Array is not configured at all, and not held active,
471 * so destroy it */
472 list_del_init(&mddev->all_mddevs);
473 bs = mddev->bio_set;
474 mddev->bio_set = NULL;
475 if (mddev->gendisk) {
476 /* We did a probe so need to clean up. Call
477 * queue_work inside the spinlock so that
478 * flush_workqueue() after mddev_find will
479 * succeed in waiting for the work to be done.
480 */
481 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
482 queue_work(md_misc_wq, &mddev->del_work);
483 } else
484 kfree(mddev);
485 }
486 spin_unlock(&all_mddevs_lock);
487 if (bs)
488 bioset_free(bs);
489 }
490
491 static void md_safemode_timeout(unsigned long data);
492
493 void mddev_init(struct mddev *mddev)
494 {
495 mutex_init(&mddev->open_mutex);
496 mutex_init(&mddev->reconfig_mutex);
497 mutex_init(&mddev->bitmap_info.mutex);
498 INIT_LIST_HEAD(&mddev->disks);
499 INIT_LIST_HEAD(&mddev->all_mddevs);
500 setup_timer(&mddev->safemode_timer, md_safemode_timeout,
501 (unsigned long) mddev);
502 atomic_set(&mddev->active, 1);
503 atomic_set(&mddev->openers, 0);
504 atomic_set(&mddev->active_io, 0);
505 spin_lock_init(&mddev->lock);
506 atomic_set(&mddev->flush_pending, 0);
507 init_waitqueue_head(&mddev->sb_wait);
508 init_waitqueue_head(&mddev->recovery_wait);
509 mddev->reshape_position = MaxSector;
510 mddev->reshape_backwards = 0;
511 mddev->last_sync_action = "none";
512 mddev->resync_min = 0;
513 mddev->resync_max = MaxSector;
514 mddev->level = LEVEL_NONE;
515 }
516 EXPORT_SYMBOL_GPL(mddev_init);
517
518 static struct mddev *mddev_find(dev_t unit)
519 {
520 struct mddev *mddev, *new = NULL;
521
522 if (unit && MAJOR(unit) != MD_MAJOR)
523 unit &= ~((1<<MdpMinorShift)-1);
524
525 retry:
526 spin_lock(&all_mddevs_lock);
527
528 if (unit) {
529 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
530 if (mddev->unit == unit) {
531 mddev_get(mddev);
532 spin_unlock(&all_mddevs_lock);
533 kfree(new);
534 return mddev;
535 }
536
537 if (new) {
538 list_add(&new->all_mddevs, &all_mddevs);
539 spin_unlock(&all_mddevs_lock);
540 new->hold_active = UNTIL_IOCTL;
541 return new;
542 }
543 } else if (new) {
544 /* find an unused unit number */
545 static int next_minor = 512;
546 int start = next_minor;
547 int is_free = 0;
548 int dev = 0;
549 while (!is_free) {
550 dev = MKDEV(MD_MAJOR, next_minor);
551 next_minor++;
552 if (next_minor > MINORMASK)
553 next_minor = 0;
554 if (next_minor == start) {
555 /* Oh dear, all in use. */
556 spin_unlock(&all_mddevs_lock);
557 kfree(new);
558 return NULL;
559 }
560
561 is_free = 1;
562 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
563 if (mddev->unit == dev) {
564 is_free = 0;
565 break;
566 }
567 }
568 new->unit = dev;
569 new->md_minor = MINOR(dev);
570 new->hold_active = UNTIL_STOP;
571 list_add(&new->all_mddevs, &all_mddevs);
572 spin_unlock(&all_mddevs_lock);
573 return new;
574 }
575 spin_unlock(&all_mddevs_lock);
576
577 new = kzalloc(sizeof(*new), GFP_KERNEL);
578 if (!new)
579 return NULL;
580
581 new->unit = unit;
582 if (MAJOR(unit) == MD_MAJOR)
583 new->md_minor = MINOR(unit);
584 else
585 new->md_minor = MINOR(unit) >> MdpMinorShift;
586
587 mddev_init(new);
588
589 goto retry;
590 }
591
592 static struct attribute_group md_redundancy_group;
593
594 void mddev_unlock(struct mddev *mddev)
595 {
596 if (mddev->to_remove) {
597 /* These cannot be removed under reconfig_mutex as
598 * an access to the files will try to take reconfig_mutex
599 * while holding the file unremovable, which leads to
600 * a deadlock.
601 * So hold set sysfs_active while the remove in happeing,
602 * and anything else which might set ->to_remove or my
603 * otherwise change the sysfs namespace will fail with
604 * -EBUSY if sysfs_active is still set.
605 * We set sysfs_active under reconfig_mutex and elsewhere
606 * test it under the same mutex to ensure its correct value
607 * is seen.
608 */
609 struct attribute_group *to_remove = mddev->to_remove;
610 mddev->to_remove = NULL;
611 mddev->sysfs_active = 1;
612 mutex_unlock(&mddev->reconfig_mutex);
613
614 if (mddev->kobj.sd) {
615 if (to_remove != &md_redundancy_group)
616 sysfs_remove_group(&mddev->kobj, to_remove);
617 if (mddev->pers == NULL ||
618 mddev->pers->sync_request == NULL) {
619 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
620 if (mddev->sysfs_action)
621 sysfs_put(mddev->sysfs_action);
622 mddev->sysfs_action = NULL;
623 }
624 }
625 mddev->sysfs_active = 0;
626 } else
627 mutex_unlock(&mddev->reconfig_mutex);
628
629 /* As we've dropped the mutex we need a spinlock to
630 * make sure the thread doesn't disappear
631 */
632 spin_lock(&pers_lock);
633 md_wakeup_thread(mddev->thread);
634 spin_unlock(&pers_lock);
635 }
636 EXPORT_SYMBOL_GPL(mddev_unlock);
637
638 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
639 {
640 struct md_rdev *rdev;
641
642 rdev_for_each_rcu(rdev, mddev)
643 if (rdev->desc_nr == nr)
644 return rdev;
645
646 return NULL;
647 }
648 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
649
650 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
651 {
652 struct md_rdev *rdev;
653
654 rdev_for_each(rdev, mddev)
655 if (rdev->bdev->bd_dev == dev)
656 return rdev;
657
658 return NULL;
659 }
660
661 static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
662 {
663 struct md_rdev *rdev;
664
665 rdev_for_each_rcu(rdev, mddev)
666 if (rdev->bdev->bd_dev == dev)
667 return rdev;
668
669 return NULL;
670 }
671
672 static struct md_personality *find_pers(int level, char *clevel)
673 {
674 struct md_personality *pers;
675 list_for_each_entry(pers, &pers_list, list) {
676 if (level != LEVEL_NONE && pers->level == level)
677 return pers;
678 if (strcmp(pers->name, clevel)==0)
679 return pers;
680 }
681 return NULL;
682 }
683
684 /* return the offset of the super block in 512byte sectors */
685 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
686 {
687 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
688 return MD_NEW_SIZE_SECTORS(num_sectors);
689 }
690
691 static int alloc_disk_sb(struct md_rdev *rdev)
692 {
693 rdev->sb_page = alloc_page(GFP_KERNEL);
694 if (!rdev->sb_page) {
695 printk(KERN_ALERT "md: out of memory.\n");
696 return -ENOMEM;
697 }
698
699 return 0;
700 }
701
702 void md_rdev_clear(struct md_rdev *rdev)
703 {
704 if (rdev->sb_page) {
705 put_page(rdev->sb_page);
706 rdev->sb_loaded = 0;
707 rdev->sb_page = NULL;
708 rdev->sb_start = 0;
709 rdev->sectors = 0;
710 }
711 if (rdev->bb_page) {
712 put_page(rdev->bb_page);
713 rdev->bb_page = NULL;
714 }
715 kfree(rdev->badblocks.page);
716 rdev->badblocks.page = NULL;
717 }
718 EXPORT_SYMBOL_GPL(md_rdev_clear);
719
720 static void super_written(struct bio *bio)
721 {
722 struct md_rdev *rdev = bio->bi_private;
723 struct mddev *mddev = rdev->mddev;
724
725 if (bio->bi_error) {
726 printk("md: super_written gets error=%d\n", bio->bi_error);
727 md_error(mddev, rdev);
728 }
729
730 if (atomic_dec_and_test(&mddev->pending_writes))
731 wake_up(&mddev->sb_wait);
732 bio_put(bio);
733 }
734
735 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
736 sector_t sector, int size, struct page *page)
737 {
738 /* write first size bytes of page to sector of rdev
739 * Increment mddev->pending_writes before returning
740 * and decrement it on completion, waking up sb_wait
741 * if zero is reached.
742 * If an error occurred, call md_error
743 */
744 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
745
746 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
747 bio->bi_iter.bi_sector = sector;
748 bio_add_page(bio, page, size, 0);
749 bio->bi_private = rdev;
750 bio->bi_end_io = super_written;
751
752 atomic_inc(&mddev->pending_writes);
753 submit_bio(WRITE_FLUSH_FUA, bio);
754 }
755
756 void md_super_wait(struct mddev *mddev)
757 {
758 /* wait for all superblock writes that were scheduled to complete */
759 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
760 }
761
762 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
763 struct page *page, int rw, bool metadata_op)
764 {
765 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
766 int ret;
767
768 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
769 rdev->meta_bdev : rdev->bdev;
770 if (metadata_op)
771 bio->bi_iter.bi_sector = sector + rdev->sb_start;
772 else if (rdev->mddev->reshape_position != MaxSector &&
773 (rdev->mddev->reshape_backwards ==
774 (sector >= rdev->mddev->reshape_position)))
775 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
776 else
777 bio->bi_iter.bi_sector = sector + rdev->data_offset;
778 bio_add_page(bio, page, size, 0);
779 submit_bio_wait(rw, bio);
780
781 ret = !bio->bi_error;
782 bio_put(bio);
783 return ret;
784 }
785 EXPORT_SYMBOL_GPL(sync_page_io);
786
787 static int read_disk_sb(struct md_rdev *rdev, int size)
788 {
789 char b[BDEVNAME_SIZE];
790
791 if (rdev->sb_loaded)
792 return 0;
793
794 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
795 goto fail;
796 rdev->sb_loaded = 1;
797 return 0;
798
799 fail:
800 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
801 bdevname(rdev->bdev,b));
802 return -EINVAL;
803 }
804
805 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
806 {
807 return sb1->set_uuid0 == sb2->set_uuid0 &&
808 sb1->set_uuid1 == sb2->set_uuid1 &&
809 sb1->set_uuid2 == sb2->set_uuid2 &&
810 sb1->set_uuid3 == sb2->set_uuid3;
811 }
812
813 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
814 {
815 int ret;
816 mdp_super_t *tmp1, *tmp2;
817
818 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
819 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
820
821 if (!tmp1 || !tmp2) {
822 ret = 0;
823 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
824 goto abort;
825 }
826
827 *tmp1 = *sb1;
828 *tmp2 = *sb2;
829
830 /*
831 * nr_disks is not constant
832 */
833 tmp1->nr_disks = 0;
834 tmp2->nr_disks = 0;
835
836 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
837 abort:
838 kfree(tmp1);
839 kfree(tmp2);
840 return ret;
841 }
842
843 static u32 md_csum_fold(u32 csum)
844 {
845 csum = (csum & 0xffff) + (csum >> 16);
846 return (csum & 0xffff) + (csum >> 16);
847 }
848
849 static unsigned int calc_sb_csum(mdp_super_t *sb)
850 {
851 u64 newcsum = 0;
852 u32 *sb32 = (u32*)sb;
853 int i;
854 unsigned int disk_csum, csum;
855
856 disk_csum = sb->sb_csum;
857 sb->sb_csum = 0;
858
859 for (i = 0; i < MD_SB_BYTES/4 ; i++)
860 newcsum += sb32[i];
861 csum = (newcsum & 0xffffffff) + (newcsum>>32);
862
863 #ifdef CONFIG_ALPHA
864 /* This used to use csum_partial, which was wrong for several
865 * reasons including that different results are returned on
866 * different architectures. It isn't critical that we get exactly
867 * the same return value as before (we always csum_fold before
868 * testing, and that removes any differences). However as we
869 * know that csum_partial always returned a 16bit value on
870 * alphas, do a fold to maximise conformity to previous behaviour.
871 */
872 sb->sb_csum = md_csum_fold(disk_csum);
873 #else
874 sb->sb_csum = disk_csum;
875 #endif
876 return csum;
877 }
878
879 /*
880 * Handle superblock details.
881 * We want to be able to handle multiple superblock formats
882 * so we have a common interface to them all, and an array of
883 * different handlers.
884 * We rely on user-space to write the initial superblock, and support
885 * reading and updating of superblocks.
886 * Interface methods are:
887 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
888 * loads and validates a superblock on dev.
889 * if refdev != NULL, compare superblocks on both devices
890 * Return:
891 * 0 - dev has a superblock that is compatible with refdev
892 * 1 - dev has a superblock that is compatible and newer than refdev
893 * so dev should be used as the refdev in future
894 * -EINVAL superblock incompatible or invalid
895 * -othererror e.g. -EIO
896 *
897 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
898 * Verify that dev is acceptable into mddev.
899 * The first time, mddev->raid_disks will be 0, and data from
900 * dev should be merged in. Subsequent calls check that dev
901 * is new enough. Return 0 or -EINVAL
902 *
903 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
904 * Update the superblock for rdev with data in mddev
905 * This does not write to disc.
906 *
907 */
908
909 struct super_type {
910 char *name;
911 struct module *owner;
912 int (*load_super)(struct md_rdev *rdev,
913 struct md_rdev *refdev,
914 int minor_version);
915 int (*validate_super)(struct mddev *mddev,
916 struct md_rdev *rdev);
917 void (*sync_super)(struct mddev *mddev,
918 struct md_rdev *rdev);
919 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
920 sector_t num_sectors);
921 int (*allow_new_offset)(struct md_rdev *rdev,
922 unsigned long long new_offset);
923 };
924
925 /*
926 * Check that the given mddev has no bitmap.
927 *
928 * This function is called from the run method of all personalities that do not
929 * support bitmaps. It prints an error message and returns non-zero if mddev
930 * has a bitmap. Otherwise, it returns 0.
931 *
932 */
933 int md_check_no_bitmap(struct mddev *mddev)
934 {
935 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
936 return 0;
937 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
938 mdname(mddev), mddev->pers->name);
939 return 1;
940 }
941 EXPORT_SYMBOL(md_check_no_bitmap);
942
943 /*
944 * load_super for 0.90.0
945 */
946 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
947 {
948 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
949 mdp_super_t *sb;
950 int ret;
951
952 /*
953 * Calculate the position of the superblock (512byte sectors),
954 * it's at the end of the disk.
955 *
956 * It also happens to be a multiple of 4Kb.
957 */
958 rdev->sb_start = calc_dev_sboffset(rdev);
959
960 ret = read_disk_sb(rdev, MD_SB_BYTES);
961 if (ret) return ret;
962
963 ret = -EINVAL;
964
965 bdevname(rdev->bdev, b);
966 sb = page_address(rdev->sb_page);
967
968 if (sb->md_magic != MD_SB_MAGIC) {
969 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
970 b);
971 goto abort;
972 }
973
974 if (sb->major_version != 0 ||
975 sb->minor_version < 90 ||
976 sb->minor_version > 91) {
977 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
978 sb->major_version, sb->minor_version,
979 b);
980 goto abort;
981 }
982
983 if (sb->raid_disks <= 0)
984 goto abort;
985
986 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
987 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
988 b);
989 goto abort;
990 }
991
992 rdev->preferred_minor = sb->md_minor;
993 rdev->data_offset = 0;
994 rdev->new_data_offset = 0;
995 rdev->sb_size = MD_SB_BYTES;
996 rdev->badblocks.shift = -1;
997
998 if (sb->level == LEVEL_MULTIPATH)
999 rdev->desc_nr = -1;
1000 else
1001 rdev->desc_nr = sb->this_disk.number;
1002
1003 if (!refdev) {
1004 ret = 1;
1005 } else {
1006 __u64 ev1, ev2;
1007 mdp_super_t *refsb = page_address(refdev->sb_page);
1008 if (!uuid_equal(refsb, sb)) {
1009 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1010 b, bdevname(refdev->bdev,b2));
1011 goto abort;
1012 }
1013 if (!sb_equal(refsb, sb)) {
1014 printk(KERN_WARNING "md: %s has same UUID"
1015 " but different superblock to %s\n",
1016 b, bdevname(refdev->bdev, b2));
1017 goto abort;
1018 }
1019 ev1 = md_event(sb);
1020 ev2 = md_event(refsb);
1021 if (ev1 > ev2)
1022 ret = 1;
1023 else
1024 ret = 0;
1025 }
1026 rdev->sectors = rdev->sb_start;
1027 /* Limit to 4TB as metadata cannot record more than that.
1028 * (not needed for Linear and RAID0 as metadata doesn't
1029 * record this size)
1030 */
1031 if (IS_ENABLED(CONFIG_LBDAF) && (u64)rdev->sectors >= (2ULL << 32) &&
1032 sb->level >= 1)
1033 rdev->sectors = (sector_t)(2ULL << 32) - 2;
1034
1035 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1036 /* "this cannot possibly happen" ... */
1037 ret = -EINVAL;
1038
1039 abort:
1040 return ret;
1041 }
1042
1043 /*
1044 * validate_super for 0.90.0
1045 */
1046 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1047 {
1048 mdp_disk_t *desc;
1049 mdp_super_t *sb = page_address(rdev->sb_page);
1050 __u64 ev1 = md_event(sb);
1051
1052 rdev->raid_disk = -1;
1053 clear_bit(Faulty, &rdev->flags);
1054 clear_bit(In_sync, &rdev->flags);
1055 clear_bit(Bitmap_sync, &rdev->flags);
1056 clear_bit(WriteMostly, &rdev->flags);
1057
1058 if (mddev->raid_disks == 0) {
1059 mddev->major_version = 0;
1060 mddev->minor_version = sb->minor_version;
1061 mddev->patch_version = sb->patch_version;
1062 mddev->external = 0;
1063 mddev->chunk_sectors = sb->chunk_size >> 9;
1064 mddev->ctime = sb->ctime;
1065 mddev->utime = sb->utime;
1066 mddev->level = sb->level;
1067 mddev->clevel[0] = 0;
1068 mddev->layout = sb->layout;
1069 mddev->raid_disks = sb->raid_disks;
1070 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1071 mddev->events = ev1;
1072 mddev->bitmap_info.offset = 0;
1073 mddev->bitmap_info.space = 0;
1074 /* bitmap can use 60 K after the 4K superblocks */
1075 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1076 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1077 mddev->reshape_backwards = 0;
1078
1079 if (mddev->minor_version >= 91) {
1080 mddev->reshape_position = sb->reshape_position;
1081 mddev->delta_disks = sb->delta_disks;
1082 mddev->new_level = sb->new_level;
1083 mddev->new_layout = sb->new_layout;
1084 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1085 if (mddev->delta_disks < 0)
1086 mddev->reshape_backwards = 1;
1087 } else {
1088 mddev->reshape_position = MaxSector;
1089 mddev->delta_disks = 0;
1090 mddev->new_level = mddev->level;
1091 mddev->new_layout = mddev->layout;
1092 mddev->new_chunk_sectors = mddev->chunk_sectors;
1093 }
1094
1095 if (sb->state & (1<<MD_SB_CLEAN))
1096 mddev->recovery_cp = MaxSector;
1097 else {
1098 if (sb->events_hi == sb->cp_events_hi &&
1099 sb->events_lo == sb->cp_events_lo) {
1100 mddev->recovery_cp = sb->recovery_cp;
1101 } else
1102 mddev->recovery_cp = 0;
1103 }
1104
1105 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1106 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1107 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1108 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1109
1110 mddev->max_disks = MD_SB_DISKS;
1111
1112 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1113 mddev->bitmap_info.file == NULL) {
1114 mddev->bitmap_info.offset =
1115 mddev->bitmap_info.default_offset;
1116 mddev->bitmap_info.space =
1117 mddev->bitmap_info.default_space;
1118 }
1119
1120 } else if (mddev->pers == NULL) {
1121 /* Insist on good event counter while assembling, except
1122 * for spares (which don't need an event count) */
1123 ++ev1;
1124 if (sb->disks[rdev->desc_nr].state & (
1125 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1126 if (ev1 < mddev->events)
1127 return -EINVAL;
1128 } else if (mddev->bitmap) {
1129 /* if adding to array with a bitmap, then we can accept an
1130 * older device ... but not too old.
1131 */
1132 if (ev1 < mddev->bitmap->events_cleared)
1133 return 0;
1134 if (ev1 < mddev->events)
1135 set_bit(Bitmap_sync, &rdev->flags);
1136 } else {
1137 if (ev1 < mddev->events)
1138 /* just a hot-add of a new device, leave raid_disk at -1 */
1139 return 0;
1140 }
1141
1142 if (mddev->level != LEVEL_MULTIPATH) {
1143 desc = sb->disks + rdev->desc_nr;
1144
1145 if (desc->state & (1<<MD_DISK_FAULTY))
1146 set_bit(Faulty, &rdev->flags);
1147 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1148 desc->raid_disk < mddev->raid_disks */) {
1149 set_bit(In_sync, &rdev->flags);
1150 rdev->raid_disk = desc->raid_disk;
1151 rdev->saved_raid_disk = desc->raid_disk;
1152 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1153 /* active but not in sync implies recovery up to
1154 * reshape position. We don't know exactly where
1155 * that is, so set to zero for now */
1156 if (mddev->minor_version >= 91) {
1157 rdev->recovery_offset = 0;
1158 rdev->raid_disk = desc->raid_disk;
1159 }
1160 }
1161 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1162 set_bit(WriteMostly, &rdev->flags);
1163 } else /* MULTIPATH are always insync */
1164 set_bit(In_sync, &rdev->flags);
1165 return 0;
1166 }
1167
1168 /*
1169 * sync_super for 0.90.0
1170 */
1171 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1172 {
1173 mdp_super_t *sb;
1174 struct md_rdev *rdev2;
1175 int next_spare = mddev->raid_disks;
1176
1177 /* make rdev->sb match mddev data..
1178 *
1179 * 1/ zero out disks
1180 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1181 * 3/ any empty disks < next_spare become removed
1182 *
1183 * disks[0] gets initialised to REMOVED because
1184 * we cannot be sure from other fields if it has
1185 * been initialised or not.
1186 */
1187 int i;
1188 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1189
1190 rdev->sb_size = MD_SB_BYTES;
1191
1192 sb = page_address(rdev->sb_page);
1193
1194 memset(sb, 0, sizeof(*sb));
1195
1196 sb->md_magic = MD_SB_MAGIC;
1197 sb->major_version = mddev->major_version;
1198 sb->patch_version = mddev->patch_version;
1199 sb->gvalid_words = 0; /* ignored */
1200 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1201 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1202 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1203 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1204
1205 sb->ctime = mddev->ctime;
1206 sb->level = mddev->level;
1207 sb->size = mddev->dev_sectors / 2;
1208 sb->raid_disks = mddev->raid_disks;
1209 sb->md_minor = mddev->md_minor;
1210 sb->not_persistent = 0;
1211 sb->utime = mddev->utime;
1212 sb->state = 0;
1213 sb->events_hi = (mddev->events>>32);
1214 sb->events_lo = (u32)mddev->events;
1215
1216 if (mddev->reshape_position == MaxSector)
1217 sb->minor_version = 90;
1218 else {
1219 sb->minor_version = 91;
1220 sb->reshape_position = mddev->reshape_position;
1221 sb->new_level = mddev->new_level;
1222 sb->delta_disks = mddev->delta_disks;
1223 sb->new_layout = mddev->new_layout;
1224 sb->new_chunk = mddev->new_chunk_sectors << 9;
1225 }
1226 mddev->minor_version = sb->minor_version;
1227 if (mddev->in_sync)
1228 {
1229 sb->recovery_cp = mddev->recovery_cp;
1230 sb->cp_events_hi = (mddev->events>>32);
1231 sb->cp_events_lo = (u32)mddev->events;
1232 if (mddev->recovery_cp == MaxSector)
1233 sb->state = (1<< MD_SB_CLEAN);
1234 } else
1235 sb->recovery_cp = 0;
1236
1237 sb->layout = mddev->layout;
1238 sb->chunk_size = mddev->chunk_sectors << 9;
1239
1240 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1241 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1242
1243 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1244 rdev_for_each(rdev2, mddev) {
1245 mdp_disk_t *d;
1246 int desc_nr;
1247 int is_active = test_bit(In_sync, &rdev2->flags);
1248
1249 if (rdev2->raid_disk >= 0 &&
1250 sb->minor_version >= 91)
1251 /* we have nowhere to store the recovery_offset,
1252 * but if it is not below the reshape_position,
1253 * we can piggy-back on that.
1254 */
1255 is_active = 1;
1256 if (rdev2->raid_disk < 0 ||
1257 test_bit(Faulty, &rdev2->flags))
1258 is_active = 0;
1259 if (is_active)
1260 desc_nr = rdev2->raid_disk;
1261 else
1262 desc_nr = next_spare++;
1263 rdev2->desc_nr = desc_nr;
1264 d = &sb->disks[rdev2->desc_nr];
1265 nr_disks++;
1266 d->number = rdev2->desc_nr;
1267 d->major = MAJOR(rdev2->bdev->bd_dev);
1268 d->minor = MINOR(rdev2->bdev->bd_dev);
1269 if (is_active)
1270 d->raid_disk = rdev2->raid_disk;
1271 else
1272 d->raid_disk = rdev2->desc_nr; /* compatibility */
1273 if (test_bit(Faulty, &rdev2->flags))
1274 d->state = (1<<MD_DISK_FAULTY);
1275 else if (is_active) {
1276 d->state = (1<<MD_DISK_ACTIVE);
1277 if (test_bit(In_sync, &rdev2->flags))
1278 d->state |= (1<<MD_DISK_SYNC);
1279 active++;
1280 working++;
1281 } else {
1282 d->state = 0;
1283 spare++;
1284 working++;
1285 }
1286 if (test_bit(WriteMostly, &rdev2->flags))
1287 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1288 }
1289 /* now set the "removed" and "faulty" bits on any missing devices */
1290 for (i=0 ; i < mddev->raid_disks ; i++) {
1291 mdp_disk_t *d = &sb->disks[i];
1292 if (d->state == 0 && d->number == 0) {
1293 d->number = i;
1294 d->raid_disk = i;
1295 d->state = (1<<MD_DISK_REMOVED);
1296 d->state |= (1<<MD_DISK_FAULTY);
1297 failed++;
1298 }
1299 }
1300 sb->nr_disks = nr_disks;
1301 sb->active_disks = active;
1302 sb->working_disks = working;
1303 sb->failed_disks = failed;
1304 sb->spare_disks = spare;
1305
1306 sb->this_disk = sb->disks[rdev->desc_nr];
1307 sb->sb_csum = calc_sb_csum(sb);
1308 }
1309
1310 /*
1311 * rdev_size_change for 0.90.0
1312 */
1313 static unsigned long long
1314 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1315 {
1316 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1317 return 0; /* component must fit device */
1318 if (rdev->mddev->bitmap_info.offset)
1319 return 0; /* can't move bitmap */
1320 rdev->sb_start = calc_dev_sboffset(rdev);
1321 if (!num_sectors || num_sectors > rdev->sb_start)
1322 num_sectors = rdev->sb_start;
1323 /* Limit to 4TB as metadata cannot record more than that.
1324 * 4TB == 2^32 KB, or 2*2^32 sectors.
1325 */
1326 if (IS_ENABLED(CONFIG_LBDAF) && (u64)num_sectors >= (2ULL << 32) &&
1327 rdev->mddev->level >= 1)
1328 num_sectors = (sector_t)(2ULL << 32) - 2;
1329 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1330 rdev->sb_page);
1331 md_super_wait(rdev->mddev);
1332 return num_sectors;
1333 }
1334
1335 static int
1336 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1337 {
1338 /* non-zero offset changes not possible with v0.90 */
1339 return new_offset == 0;
1340 }
1341
1342 /*
1343 * version 1 superblock
1344 */
1345
1346 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1347 {
1348 __le32 disk_csum;
1349 u32 csum;
1350 unsigned long long newcsum;
1351 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1352 __le32 *isuper = (__le32*)sb;
1353
1354 disk_csum = sb->sb_csum;
1355 sb->sb_csum = 0;
1356 newcsum = 0;
1357 for (; size >= 4; size -= 4)
1358 newcsum += le32_to_cpu(*isuper++);
1359
1360 if (size == 2)
1361 newcsum += le16_to_cpu(*(__le16*) isuper);
1362
1363 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1364 sb->sb_csum = disk_csum;
1365 return cpu_to_le32(csum);
1366 }
1367
1368 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1369 int acknowledged);
1370 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1371 {
1372 struct mdp_superblock_1 *sb;
1373 int ret;
1374 sector_t sb_start;
1375 sector_t sectors;
1376 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1377 int bmask;
1378
1379 /*
1380 * Calculate the position of the superblock in 512byte sectors.
1381 * It is always aligned to a 4K boundary and
1382 * depeding on minor_version, it can be:
1383 * 0: At least 8K, but less than 12K, from end of device
1384 * 1: At start of device
1385 * 2: 4K from start of device.
1386 */
1387 switch(minor_version) {
1388 case 0:
1389 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1390 sb_start -= 8*2;
1391 sb_start &= ~(sector_t)(4*2-1);
1392 break;
1393 case 1:
1394 sb_start = 0;
1395 break;
1396 case 2:
1397 sb_start = 8;
1398 break;
1399 default:
1400 return -EINVAL;
1401 }
1402 rdev->sb_start = sb_start;
1403
1404 /* superblock is rarely larger than 1K, but it can be larger,
1405 * and it is safe to read 4k, so we do that
1406 */
1407 ret = read_disk_sb(rdev, 4096);
1408 if (ret) return ret;
1409
1410 sb = page_address(rdev->sb_page);
1411
1412 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1413 sb->major_version != cpu_to_le32(1) ||
1414 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1415 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1416 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1417 return -EINVAL;
1418
1419 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1420 printk("md: invalid superblock checksum on %s\n",
1421 bdevname(rdev->bdev,b));
1422 return -EINVAL;
1423 }
1424 if (le64_to_cpu(sb->data_size) < 10) {
1425 printk("md: data_size too small on %s\n",
1426 bdevname(rdev->bdev,b));
1427 return -EINVAL;
1428 }
1429 if (sb->pad0 ||
1430 sb->pad3[0] ||
1431 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1432 /* Some padding is non-zero, might be a new feature */
1433 return -EINVAL;
1434
1435 rdev->preferred_minor = 0xffff;
1436 rdev->data_offset = le64_to_cpu(sb->data_offset);
1437 rdev->new_data_offset = rdev->data_offset;
1438 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1439 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1440 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1441 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1442
1443 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1444 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1445 if (rdev->sb_size & bmask)
1446 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1447
1448 if (minor_version
1449 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1450 return -EINVAL;
1451 if (minor_version
1452 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1453 return -EINVAL;
1454
1455 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1456 rdev->desc_nr = -1;
1457 else
1458 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1459
1460 if (!rdev->bb_page) {
1461 rdev->bb_page = alloc_page(GFP_KERNEL);
1462 if (!rdev->bb_page)
1463 return -ENOMEM;
1464 }
1465 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1466 rdev->badblocks.count == 0) {
1467 /* need to load the bad block list.
1468 * Currently we limit it to one page.
1469 */
1470 s32 offset;
1471 sector_t bb_sector;
1472 u64 *bbp;
1473 int i;
1474 int sectors = le16_to_cpu(sb->bblog_size);
1475 if (sectors > (PAGE_SIZE / 512))
1476 return -EINVAL;
1477 offset = le32_to_cpu(sb->bblog_offset);
1478 if (offset == 0)
1479 return -EINVAL;
1480 bb_sector = (long long)offset;
1481 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1482 rdev->bb_page, READ, true))
1483 return -EIO;
1484 bbp = (u64 *)page_address(rdev->bb_page);
1485 rdev->badblocks.shift = sb->bblog_shift;
1486 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1487 u64 bb = le64_to_cpu(*bbp);
1488 int count = bb & (0x3ff);
1489 u64 sector = bb >> 10;
1490 sector <<= sb->bblog_shift;
1491 count <<= sb->bblog_shift;
1492 if (bb + 1 == 0)
1493 break;
1494 if (md_set_badblocks(&rdev->badblocks,
1495 sector, count, 1) == 0)
1496 return -EINVAL;
1497 }
1498 } else if (sb->bblog_offset != 0)
1499 rdev->badblocks.shift = 0;
1500
1501 if (!refdev) {
1502 ret = 1;
1503 } else {
1504 __u64 ev1, ev2;
1505 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1506
1507 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1508 sb->level != refsb->level ||
1509 sb->layout != refsb->layout ||
1510 sb->chunksize != refsb->chunksize) {
1511 printk(KERN_WARNING "md: %s has strangely different"
1512 " superblock to %s\n",
1513 bdevname(rdev->bdev,b),
1514 bdevname(refdev->bdev,b2));
1515 return -EINVAL;
1516 }
1517 ev1 = le64_to_cpu(sb->events);
1518 ev2 = le64_to_cpu(refsb->events);
1519
1520 if (ev1 > ev2)
1521 ret = 1;
1522 else
1523 ret = 0;
1524 }
1525 if (minor_version) {
1526 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1527 sectors -= rdev->data_offset;
1528 } else
1529 sectors = rdev->sb_start;
1530 if (sectors < le64_to_cpu(sb->data_size))
1531 return -EINVAL;
1532 rdev->sectors = le64_to_cpu(sb->data_size);
1533 return ret;
1534 }
1535
1536 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1537 {
1538 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1539 __u64 ev1 = le64_to_cpu(sb->events);
1540
1541 rdev->raid_disk = -1;
1542 clear_bit(Faulty, &rdev->flags);
1543 clear_bit(In_sync, &rdev->flags);
1544 clear_bit(Bitmap_sync, &rdev->flags);
1545 clear_bit(WriteMostly, &rdev->flags);
1546
1547 if (mddev->raid_disks == 0) {
1548 mddev->major_version = 1;
1549 mddev->patch_version = 0;
1550 mddev->external = 0;
1551 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1552 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1553 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1554 mddev->level = le32_to_cpu(sb->level);
1555 mddev->clevel[0] = 0;
1556 mddev->layout = le32_to_cpu(sb->layout);
1557 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1558 mddev->dev_sectors = le64_to_cpu(sb->size);
1559 mddev->events = ev1;
1560 mddev->bitmap_info.offset = 0;
1561 mddev->bitmap_info.space = 0;
1562 /* Default location for bitmap is 1K after superblock
1563 * using 3K - total of 4K
1564 */
1565 mddev->bitmap_info.default_offset = 1024 >> 9;
1566 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1567 mddev->reshape_backwards = 0;
1568
1569 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1570 memcpy(mddev->uuid, sb->set_uuid, 16);
1571
1572 mddev->max_disks = (4096-256)/2;
1573
1574 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1575 mddev->bitmap_info.file == NULL) {
1576 mddev->bitmap_info.offset =
1577 (__s32)le32_to_cpu(sb->bitmap_offset);
1578 /* Metadata doesn't record how much space is available.
1579 * For 1.0, we assume we can use up to the superblock
1580 * if before, else to 4K beyond superblock.
1581 * For others, assume no change is possible.
1582 */
1583 if (mddev->minor_version > 0)
1584 mddev->bitmap_info.space = 0;
1585 else if (mddev->bitmap_info.offset > 0)
1586 mddev->bitmap_info.space =
1587 8 - mddev->bitmap_info.offset;
1588 else
1589 mddev->bitmap_info.space =
1590 -mddev->bitmap_info.offset;
1591 }
1592
1593 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1594 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1595 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1596 mddev->new_level = le32_to_cpu(sb->new_level);
1597 mddev->new_layout = le32_to_cpu(sb->new_layout);
1598 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1599 if (mddev->delta_disks < 0 ||
1600 (mddev->delta_disks == 0 &&
1601 (le32_to_cpu(sb->feature_map)
1602 & MD_FEATURE_RESHAPE_BACKWARDS)))
1603 mddev->reshape_backwards = 1;
1604 } else {
1605 mddev->reshape_position = MaxSector;
1606 mddev->delta_disks = 0;
1607 mddev->new_level = mddev->level;
1608 mddev->new_layout = mddev->layout;
1609 mddev->new_chunk_sectors = mddev->chunk_sectors;
1610 }
1611
1612 } else if (mddev->pers == NULL) {
1613 /* Insist of good event counter while assembling, except for
1614 * spares (which don't need an event count) */
1615 ++ev1;
1616 if (rdev->desc_nr >= 0 &&
1617 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1618 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1619 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1620 if (ev1 < mddev->events)
1621 return -EINVAL;
1622 } else if (mddev->bitmap) {
1623 /* If adding to array with a bitmap, then we can accept an
1624 * older device, but not too old.
1625 */
1626 if (ev1 < mddev->bitmap->events_cleared)
1627 return 0;
1628 if (ev1 < mddev->events)
1629 set_bit(Bitmap_sync, &rdev->flags);
1630 } else {
1631 if (ev1 < mddev->events)
1632 /* just a hot-add of a new device, leave raid_disk at -1 */
1633 return 0;
1634 }
1635 if (mddev->level != LEVEL_MULTIPATH) {
1636 int role;
1637 if (rdev->desc_nr < 0 ||
1638 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1639 role = MD_DISK_ROLE_SPARE;
1640 rdev->desc_nr = -1;
1641 } else
1642 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1643 switch(role) {
1644 case MD_DISK_ROLE_SPARE: /* spare */
1645 break;
1646 case MD_DISK_ROLE_FAULTY: /* faulty */
1647 set_bit(Faulty, &rdev->flags);
1648 break;
1649 case MD_DISK_ROLE_JOURNAL: /* journal device */
1650 if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1651 /* journal device without journal feature */
1652 printk(KERN_WARNING
1653 "md: journal device provided without journal feature, ignoring the device\n");
1654 return -EINVAL;
1655 }
1656 set_bit(Journal, &rdev->flags);
1657 rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1658 if (mddev->recovery_cp == MaxSector)
1659 set_bit(MD_JOURNAL_CLEAN, &mddev->flags);
1660 rdev->raid_disk = 0;
1661 break;
1662 default:
1663 rdev->saved_raid_disk = role;
1664 if ((le32_to_cpu(sb->feature_map) &
1665 MD_FEATURE_RECOVERY_OFFSET)) {
1666 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1667 if (!(le32_to_cpu(sb->feature_map) &
1668 MD_FEATURE_RECOVERY_BITMAP))
1669 rdev->saved_raid_disk = -1;
1670 } else
1671 set_bit(In_sync, &rdev->flags);
1672 rdev->raid_disk = role;
1673 break;
1674 }
1675 if (sb->devflags & WriteMostly1)
1676 set_bit(WriteMostly, &rdev->flags);
1677 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1678 set_bit(Replacement, &rdev->flags);
1679 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1680 set_bit(MD_HAS_JOURNAL, &mddev->flags);
1681 } else /* MULTIPATH are always insync */
1682 set_bit(In_sync, &rdev->flags);
1683
1684 return 0;
1685 }
1686
1687 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1688 {
1689 struct mdp_superblock_1 *sb;
1690 struct md_rdev *rdev2;
1691 int max_dev, i;
1692 /* make rdev->sb match mddev and rdev data. */
1693
1694 sb = page_address(rdev->sb_page);
1695
1696 sb->feature_map = 0;
1697 sb->pad0 = 0;
1698 sb->recovery_offset = cpu_to_le64(0);
1699 memset(sb->pad3, 0, sizeof(sb->pad3));
1700
1701 sb->utime = cpu_to_le64((__u64)mddev->utime);
1702 sb->events = cpu_to_le64(mddev->events);
1703 if (mddev->in_sync)
1704 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1705 else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
1706 sb->resync_offset = cpu_to_le64(MaxSector);
1707 else
1708 sb->resync_offset = cpu_to_le64(0);
1709
1710 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1711
1712 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1713 sb->size = cpu_to_le64(mddev->dev_sectors);
1714 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1715 sb->level = cpu_to_le32(mddev->level);
1716 sb->layout = cpu_to_le32(mddev->layout);
1717
1718 if (test_bit(WriteMostly, &rdev->flags))
1719 sb->devflags |= WriteMostly1;
1720 else
1721 sb->devflags &= ~WriteMostly1;
1722 sb->data_offset = cpu_to_le64(rdev->data_offset);
1723 sb->data_size = cpu_to_le64(rdev->sectors);
1724
1725 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1726 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1727 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1728 }
1729
1730 if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
1731 !test_bit(In_sync, &rdev->flags)) {
1732 sb->feature_map |=
1733 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1734 sb->recovery_offset =
1735 cpu_to_le64(rdev->recovery_offset);
1736 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1737 sb->feature_map |=
1738 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
1739 }
1740 /* Note: recovery_offset and journal_tail share space */
1741 if (test_bit(Journal, &rdev->flags))
1742 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
1743 if (test_bit(Replacement, &rdev->flags))
1744 sb->feature_map |=
1745 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1746
1747 if (mddev->reshape_position != MaxSector) {
1748 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1749 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1750 sb->new_layout = cpu_to_le32(mddev->new_layout);
1751 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1752 sb->new_level = cpu_to_le32(mddev->new_level);
1753 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1754 if (mddev->delta_disks == 0 &&
1755 mddev->reshape_backwards)
1756 sb->feature_map
1757 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1758 if (rdev->new_data_offset != rdev->data_offset) {
1759 sb->feature_map
1760 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1761 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1762 - rdev->data_offset));
1763 }
1764 }
1765
1766 if (mddev_is_clustered(mddev))
1767 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
1768
1769 if (rdev->badblocks.count == 0)
1770 /* Nothing to do for bad blocks*/ ;
1771 else if (sb->bblog_offset == 0)
1772 /* Cannot record bad blocks on this device */
1773 md_error(mddev, rdev);
1774 else {
1775 struct badblocks *bb = &rdev->badblocks;
1776 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1777 u64 *p = bb->page;
1778 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1779 if (bb->changed) {
1780 unsigned seq;
1781
1782 retry:
1783 seq = read_seqbegin(&bb->lock);
1784
1785 memset(bbp, 0xff, PAGE_SIZE);
1786
1787 for (i = 0 ; i < bb->count ; i++) {
1788 u64 internal_bb = p[i];
1789 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1790 | BB_LEN(internal_bb));
1791 bbp[i] = cpu_to_le64(store_bb);
1792 }
1793 bb->changed = 0;
1794 if (read_seqretry(&bb->lock, seq))
1795 goto retry;
1796
1797 bb->sector = (rdev->sb_start +
1798 (int)le32_to_cpu(sb->bblog_offset));
1799 bb->size = le16_to_cpu(sb->bblog_size);
1800 }
1801 }
1802
1803 max_dev = 0;
1804 rdev_for_each(rdev2, mddev)
1805 if (rdev2->desc_nr+1 > max_dev)
1806 max_dev = rdev2->desc_nr+1;
1807
1808 if (max_dev > le32_to_cpu(sb->max_dev)) {
1809 int bmask;
1810 sb->max_dev = cpu_to_le32(max_dev);
1811 rdev->sb_size = max_dev * 2 + 256;
1812 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1813 if (rdev->sb_size & bmask)
1814 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1815 } else
1816 max_dev = le32_to_cpu(sb->max_dev);
1817
1818 for (i=0; i<max_dev;i++)
1819 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
1820
1821 if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
1822 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
1823
1824 rdev_for_each(rdev2, mddev) {
1825 i = rdev2->desc_nr;
1826 if (test_bit(Faulty, &rdev2->flags))
1827 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
1828 else if (test_bit(In_sync, &rdev2->flags))
1829 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1830 else if (test_bit(Journal, &rdev2->flags))
1831 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
1832 else if (rdev2->raid_disk >= 0)
1833 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1834 else
1835 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
1836 }
1837
1838 sb->sb_csum = calc_sb_1_csum(sb);
1839 }
1840
1841 static unsigned long long
1842 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1843 {
1844 struct mdp_superblock_1 *sb;
1845 sector_t max_sectors;
1846 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1847 return 0; /* component must fit device */
1848 if (rdev->data_offset != rdev->new_data_offset)
1849 return 0; /* too confusing */
1850 if (rdev->sb_start < rdev->data_offset) {
1851 /* minor versions 1 and 2; superblock before data */
1852 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1853 max_sectors -= rdev->data_offset;
1854 if (!num_sectors || num_sectors > max_sectors)
1855 num_sectors = max_sectors;
1856 } else if (rdev->mddev->bitmap_info.offset) {
1857 /* minor version 0 with bitmap we can't move */
1858 return 0;
1859 } else {
1860 /* minor version 0; superblock after data */
1861 sector_t sb_start;
1862 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1863 sb_start &= ~(sector_t)(4*2 - 1);
1864 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1865 if (!num_sectors || num_sectors > max_sectors)
1866 num_sectors = max_sectors;
1867 rdev->sb_start = sb_start;
1868 }
1869 sb = page_address(rdev->sb_page);
1870 sb->data_size = cpu_to_le64(num_sectors);
1871 sb->super_offset = cpu_to_le64(rdev->sb_start);
1872 sb->sb_csum = calc_sb_1_csum(sb);
1873 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1874 rdev->sb_page);
1875 md_super_wait(rdev->mddev);
1876 return num_sectors;
1877
1878 }
1879
1880 static int
1881 super_1_allow_new_offset(struct md_rdev *rdev,
1882 unsigned long long new_offset)
1883 {
1884 /* All necessary checks on new >= old have been done */
1885 struct bitmap *bitmap;
1886 if (new_offset >= rdev->data_offset)
1887 return 1;
1888
1889 /* with 1.0 metadata, there is no metadata to tread on
1890 * so we can always move back */
1891 if (rdev->mddev->minor_version == 0)
1892 return 1;
1893
1894 /* otherwise we must be sure not to step on
1895 * any metadata, so stay:
1896 * 36K beyond start of superblock
1897 * beyond end of badblocks
1898 * beyond write-intent bitmap
1899 */
1900 if (rdev->sb_start + (32+4)*2 > new_offset)
1901 return 0;
1902 bitmap = rdev->mddev->bitmap;
1903 if (bitmap && !rdev->mddev->bitmap_info.file &&
1904 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1905 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
1906 return 0;
1907 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1908 return 0;
1909
1910 return 1;
1911 }
1912
1913 static struct super_type super_types[] = {
1914 [0] = {
1915 .name = "0.90.0",
1916 .owner = THIS_MODULE,
1917 .load_super = super_90_load,
1918 .validate_super = super_90_validate,
1919 .sync_super = super_90_sync,
1920 .rdev_size_change = super_90_rdev_size_change,
1921 .allow_new_offset = super_90_allow_new_offset,
1922 },
1923 [1] = {
1924 .name = "md-1",
1925 .owner = THIS_MODULE,
1926 .load_super = super_1_load,
1927 .validate_super = super_1_validate,
1928 .sync_super = super_1_sync,
1929 .rdev_size_change = super_1_rdev_size_change,
1930 .allow_new_offset = super_1_allow_new_offset,
1931 },
1932 };
1933
1934 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1935 {
1936 if (mddev->sync_super) {
1937 mddev->sync_super(mddev, rdev);
1938 return;
1939 }
1940
1941 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1942
1943 super_types[mddev->major_version].sync_super(mddev, rdev);
1944 }
1945
1946 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1947 {
1948 struct md_rdev *rdev, *rdev2;
1949
1950 rcu_read_lock();
1951 rdev_for_each_rcu(rdev, mddev1) {
1952 if (test_bit(Faulty, &rdev->flags) ||
1953 test_bit(Journal, &rdev->flags) ||
1954 rdev->raid_disk == -1)
1955 continue;
1956 rdev_for_each_rcu(rdev2, mddev2) {
1957 if (test_bit(Faulty, &rdev2->flags) ||
1958 test_bit(Journal, &rdev2->flags) ||
1959 rdev2->raid_disk == -1)
1960 continue;
1961 if (rdev->bdev->bd_contains ==
1962 rdev2->bdev->bd_contains) {
1963 rcu_read_unlock();
1964 return 1;
1965 }
1966 }
1967 }
1968 rcu_read_unlock();
1969 return 0;
1970 }
1971
1972 static LIST_HEAD(pending_raid_disks);
1973
1974 /*
1975 * Try to register data integrity profile for an mddev
1976 *
1977 * This is called when an array is started and after a disk has been kicked
1978 * from the array. It only succeeds if all working and active component devices
1979 * are integrity capable with matching profiles.
1980 */
1981 int md_integrity_register(struct mddev *mddev)
1982 {
1983 struct md_rdev *rdev, *reference = NULL;
1984
1985 if (list_empty(&mddev->disks))
1986 return 0; /* nothing to do */
1987 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1988 return 0; /* shouldn't register, or already is */
1989 rdev_for_each(rdev, mddev) {
1990 /* skip spares and non-functional disks */
1991 if (test_bit(Faulty, &rdev->flags))
1992 continue;
1993 if (rdev->raid_disk < 0)
1994 continue;
1995 if (!reference) {
1996 /* Use the first rdev as the reference */
1997 reference = rdev;
1998 continue;
1999 }
2000 /* does this rdev's profile match the reference profile? */
2001 if (blk_integrity_compare(reference->bdev->bd_disk,
2002 rdev->bdev->bd_disk) < 0)
2003 return -EINVAL;
2004 }
2005 if (!reference || !bdev_get_integrity(reference->bdev))
2006 return 0;
2007 /*
2008 * All component devices are integrity capable and have matching
2009 * profiles, register the common profile for the md device.
2010 */
2011 blk_integrity_register(mddev->gendisk,
2012 bdev_get_integrity(reference->bdev));
2013
2014 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
2015 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
2016 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
2017 mdname(mddev));
2018 return -EINVAL;
2019 }
2020 return 0;
2021 }
2022 EXPORT_SYMBOL(md_integrity_register);
2023
2024 /*
2025 * Attempt to add an rdev, but only if it is consistent with the current
2026 * integrity profile
2027 */
2028 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2029 {
2030 struct blk_integrity *bi_rdev;
2031 struct blk_integrity *bi_mddev;
2032 char name[BDEVNAME_SIZE];
2033
2034 if (!mddev->gendisk)
2035 return 0;
2036
2037 bi_rdev = bdev_get_integrity(rdev->bdev);
2038 bi_mddev = blk_get_integrity(mddev->gendisk);
2039
2040 if (!bi_mddev) /* nothing to do */
2041 return 0;
2042
2043 if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2044 printk(KERN_NOTICE "%s: incompatible integrity profile for %s\n",
2045 mdname(mddev), bdevname(rdev->bdev, name));
2046 return -ENXIO;
2047 }
2048
2049 return 0;
2050 }
2051 EXPORT_SYMBOL(md_integrity_add_rdev);
2052
2053 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2054 {
2055 char b[BDEVNAME_SIZE];
2056 struct kobject *ko;
2057 int err;
2058
2059 /* prevent duplicates */
2060 if (find_rdev(mddev, rdev->bdev->bd_dev))
2061 return -EEXIST;
2062
2063 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2064 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2065 rdev->sectors < mddev->dev_sectors)) {
2066 if (mddev->pers) {
2067 /* Cannot change size, so fail
2068 * If mddev->level <= 0, then we don't care
2069 * about aligning sizes (e.g. linear)
2070 */
2071 if (mddev->level > 0)
2072 return -ENOSPC;
2073 } else
2074 mddev->dev_sectors = rdev->sectors;
2075 }
2076
2077 /* Verify rdev->desc_nr is unique.
2078 * If it is -1, assign a free number, else
2079 * check number is not in use
2080 */
2081 rcu_read_lock();
2082 if (rdev->desc_nr < 0) {
2083 int choice = 0;
2084 if (mddev->pers)
2085 choice = mddev->raid_disks;
2086 while (md_find_rdev_nr_rcu(mddev, choice))
2087 choice++;
2088 rdev->desc_nr = choice;
2089 } else {
2090 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2091 rcu_read_unlock();
2092 return -EBUSY;
2093 }
2094 }
2095 rcu_read_unlock();
2096 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2097 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2098 mdname(mddev), mddev->max_disks);
2099 return -EBUSY;
2100 }
2101 bdevname(rdev->bdev,b);
2102 strreplace(b, '/', '!');
2103
2104 rdev->mddev = mddev;
2105 printk(KERN_INFO "md: bind<%s>\n", b);
2106
2107 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2108 goto fail;
2109
2110 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2111 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2112 /* failure here is OK */;
2113 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2114
2115 list_add_rcu(&rdev->same_set, &mddev->disks);
2116 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2117
2118 /* May as well allow recovery to be retried once */
2119 mddev->recovery_disabled++;
2120
2121 return 0;
2122
2123 fail:
2124 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2125 b, mdname(mddev));
2126 return err;
2127 }
2128
2129 static void md_delayed_delete(struct work_struct *ws)
2130 {
2131 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2132 kobject_del(&rdev->kobj);
2133 kobject_put(&rdev->kobj);
2134 }
2135
2136 static void unbind_rdev_from_array(struct md_rdev *rdev)
2137 {
2138 char b[BDEVNAME_SIZE];
2139
2140 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2141 list_del_rcu(&rdev->same_set);
2142 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2143 rdev->mddev = NULL;
2144 sysfs_remove_link(&rdev->kobj, "block");
2145 sysfs_put(rdev->sysfs_state);
2146 rdev->sysfs_state = NULL;
2147 rdev->badblocks.count = 0;
2148 /* We need to delay this, otherwise we can deadlock when
2149 * writing to 'remove' to "dev/state". We also need
2150 * to delay it due to rcu usage.
2151 */
2152 synchronize_rcu();
2153 INIT_WORK(&rdev->del_work, md_delayed_delete);
2154 kobject_get(&rdev->kobj);
2155 queue_work(md_misc_wq, &rdev->del_work);
2156 }
2157
2158 /*
2159 * prevent the device from being mounted, repartitioned or
2160 * otherwise reused by a RAID array (or any other kernel
2161 * subsystem), by bd_claiming the device.
2162 */
2163 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2164 {
2165 int err = 0;
2166 struct block_device *bdev;
2167 char b[BDEVNAME_SIZE];
2168
2169 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2170 shared ? (struct md_rdev *)lock_rdev : rdev);
2171 if (IS_ERR(bdev)) {
2172 printk(KERN_ERR "md: could not open %s.\n",
2173 __bdevname(dev, b));
2174 return PTR_ERR(bdev);
2175 }
2176 rdev->bdev = bdev;
2177 return err;
2178 }
2179
2180 static void unlock_rdev(struct md_rdev *rdev)
2181 {
2182 struct block_device *bdev = rdev->bdev;
2183 rdev->bdev = NULL;
2184 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2185 }
2186
2187 void md_autodetect_dev(dev_t dev);
2188
2189 static void export_rdev(struct md_rdev *rdev)
2190 {
2191 char b[BDEVNAME_SIZE];
2192
2193 printk(KERN_INFO "md: export_rdev(%s)\n",
2194 bdevname(rdev->bdev,b));
2195 md_rdev_clear(rdev);
2196 #ifndef MODULE
2197 if (test_bit(AutoDetected, &rdev->flags))
2198 md_autodetect_dev(rdev->bdev->bd_dev);
2199 #endif
2200 unlock_rdev(rdev);
2201 kobject_put(&rdev->kobj);
2202 }
2203
2204 void md_kick_rdev_from_array(struct md_rdev *rdev)
2205 {
2206 unbind_rdev_from_array(rdev);
2207 export_rdev(rdev);
2208 }
2209 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2210
2211 static void export_array(struct mddev *mddev)
2212 {
2213 struct md_rdev *rdev;
2214
2215 while (!list_empty(&mddev->disks)) {
2216 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2217 same_set);
2218 md_kick_rdev_from_array(rdev);
2219 }
2220 mddev->raid_disks = 0;
2221 mddev->major_version = 0;
2222 }
2223
2224 static void sync_sbs(struct mddev *mddev, int nospares)
2225 {
2226 /* Update each superblock (in-memory image), but
2227 * if we are allowed to, skip spares which already
2228 * have the right event counter, or have one earlier
2229 * (which would mean they aren't being marked as dirty
2230 * with the rest of the array)
2231 */
2232 struct md_rdev *rdev;
2233 rdev_for_each(rdev, mddev) {
2234 if (rdev->sb_events == mddev->events ||
2235 (nospares &&
2236 rdev->raid_disk < 0 &&
2237 rdev->sb_events+1 == mddev->events)) {
2238 /* Don't update this superblock */
2239 rdev->sb_loaded = 2;
2240 } else {
2241 sync_super(mddev, rdev);
2242 rdev->sb_loaded = 1;
2243 }
2244 }
2245 }
2246
2247 static bool does_sb_need_changing(struct mddev *mddev)
2248 {
2249 struct md_rdev *rdev;
2250 struct mdp_superblock_1 *sb;
2251 int role;
2252
2253 /* Find a good rdev */
2254 rdev_for_each(rdev, mddev)
2255 if ((rdev->raid_disk >= 0) && !test_bit(Faulty, &rdev->flags))
2256 break;
2257
2258 /* No good device found. */
2259 if (!rdev)
2260 return false;
2261
2262 sb = page_address(rdev->sb_page);
2263 /* Check if a device has become faulty or a spare become active */
2264 rdev_for_each(rdev, mddev) {
2265 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2266 /* Device activated? */
2267 if (role == 0xffff && rdev->raid_disk >=0 &&
2268 !test_bit(Faulty, &rdev->flags))
2269 return true;
2270 /* Device turned faulty? */
2271 if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd))
2272 return true;
2273 }
2274
2275 /* Check if any mddev parameters have changed */
2276 if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2277 (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2278 (mddev->layout != le32_to_cpu(sb->layout)) ||
2279 (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2280 (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2281 return true;
2282
2283 return false;
2284 }
2285
2286 void md_update_sb(struct mddev *mddev, int force_change)
2287 {
2288 struct md_rdev *rdev;
2289 int sync_req;
2290 int nospares = 0;
2291 int any_badblocks_changed = 0;
2292 int ret = -1;
2293
2294 if (mddev->ro) {
2295 if (force_change)
2296 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2297 return;
2298 }
2299
2300 if (mddev_is_clustered(mddev)) {
2301 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2302 force_change = 1;
2303 ret = md_cluster_ops->metadata_update_start(mddev);
2304 /* Has someone else has updated the sb */
2305 if (!does_sb_need_changing(mddev)) {
2306 if (ret == 0)
2307 md_cluster_ops->metadata_update_cancel(mddev);
2308 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2309 return;
2310 }
2311 }
2312 repeat:
2313 /* First make sure individual recovery_offsets are correct */
2314 rdev_for_each(rdev, mddev) {
2315 if (rdev->raid_disk >= 0 &&
2316 mddev->delta_disks >= 0 &&
2317 !test_bit(Journal, &rdev->flags) &&
2318 !test_bit(In_sync, &rdev->flags) &&
2319 mddev->curr_resync_completed > rdev->recovery_offset)
2320 rdev->recovery_offset = mddev->curr_resync_completed;
2321
2322 }
2323 if (!mddev->persistent) {
2324 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2325 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2326 if (!mddev->external) {
2327 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2328 rdev_for_each(rdev, mddev) {
2329 if (rdev->badblocks.changed) {
2330 rdev->badblocks.changed = 0;
2331 md_ack_all_badblocks(&rdev->badblocks);
2332 md_error(mddev, rdev);
2333 }
2334 clear_bit(Blocked, &rdev->flags);
2335 clear_bit(BlockedBadBlocks, &rdev->flags);
2336 wake_up(&rdev->blocked_wait);
2337 }
2338 }
2339 wake_up(&mddev->sb_wait);
2340 return;
2341 }
2342
2343 spin_lock(&mddev->lock);
2344
2345 mddev->utime = get_seconds();
2346
2347 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2348 force_change = 1;
2349 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2350 /* just a clean<-> dirty transition, possibly leave spares alone,
2351 * though if events isn't the right even/odd, we will have to do
2352 * spares after all
2353 */
2354 nospares = 1;
2355 if (force_change)
2356 nospares = 0;
2357 if (mddev->degraded)
2358 /* If the array is degraded, then skipping spares is both
2359 * dangerous and fairly pointless.
2360 * Dangerous because a device that was removed from the array
2361 * might have a event_count that still looks up-to-date,
2362 * so it can be re-added without a resync.
2363 * Pointless because if there are any spares to skip,
2364 * then a recovery will happen and soon that array won't
2365 * be degraded any more and the spare can go back to sleep then.
2366 */
2367 nospares = 0;
2368
2369 sync_req = mddev->in_sync;
2370
2371 /* If this is just a dirty<->clean transition, and the array is clean
2372 * and 'events' is odd, we can roll back to the previous clean state */
2373 if (nospares
2374 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2375 && mddev->can_decrease_events
2376 && mddev->events != 1) {
2377 mddev->events--;
2378 mddev->can_decrease_events = 0;
2379 } else {
2380 /* otherwise we have to go forward and ... */
2381 mddev->events ++;
2382 mddev->can_decrease_events = nospares;
2383 }
2384
2385 /*
2386 * This 64-bit counter should never wrap.
2387 * Either we are in around ~1 trillion A.C., assuming
2388 * 1 reboot per second, or we have a bug...
2389 */
2390 WARN_ON(mddev->events == 0);
2391
2392 rdev_for_each(rdev, mddev) {
2393 if (rdev->badblocks.changed)
2394 any_badblocks_changed++;
2395 if (test_bit(Faulty, &rdev->flags))
2396 set_bit(FaultRecorded, &rdev->flags);
2397 }
2398
2399 sync_sbs(mddev, nospares);
2400 spin_unlock(&mddev->lock);
2401
2402 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2403 mdname(mddev), mddev->in_sync);
2404
2405 bitmap_update_sb(mddev->bitmap);
2406 rdev_for_each(rdev, mddev) {
2407 char b[BDEVNAME_SIZE];
2408
2409 if (rdev->sb_loaded != 1)
2410 continue; /* no noise on spare devices */
2411
2412 if (!test_bit(Faulty, &rdev->flags)) {
2413 md_super_write(mddev,rdev,
2414 rdev->sb_start, rdev->sb_size,
2415 rdev->sb_page);
2416 pr_debug("md: (write) %s's sb offset: %llu\n",
2417 bdevname(rdev->bdev, b),
2418 (unsigned long long)rdev->sb_start);
2419 rdev->sb_events = mddev->events;
2420 if (rdev->badblocks.size) {
2421 md_super_write(mddev, rdev,
2422 rdev->badblocks.sector,
2423 rdev->badblocks.size << 9,
2424 rdev->bb_page);
2425 rdev->badblocks.size = 0;
2426 }
2427
2428 } else
2429 pr_debug("md: %s (skipping faulty)\n",
2430 bdevname(rdev->bdev, b));
2431
2432 if (mddev->level == LEVEL_MULTIPATH)
2433 /* only need to write one superblock... */
2434 break;
2435 }
2436 md_super_wait(mddev);
2437 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2438
2439 spin_lock(&mddev->lock);
2440 if (mddev->in_sync != sync_req ||
2441 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2442 /* have to write it out again */
2443 spin_unlock(&mddev->lock);
2444 goto repeat;
2445 }
2446 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2447 spin_unlock(&mddev->lock);
2448 wake_up(&mddev->sb_wait);
2449 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2450 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2451
2452 rdev_for_each(rdev, mddev) {
2453 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2454 clear_bit(Blocked, &rdev->flags);
2455
2456 if (any_badblocks_changed)
2457 md_ack_all_badblocks(&rdev->badblocks);
2458 clear_bit(BlockedBadBlocks, &rdev->flags);
2459 wake_up(&rdev->blocked_wait);
2460 }
2461
2462 if (mddev_is_clustered(mddev) && ret == 0)
2463 md_cluster_ops->metadata_update_finish(mddev);
2464 }
2465 EXPORT_SYMBOL(md_update_sb);
2466
2467 static int add_bound_rdev(struct md_rdev *rdev)
2468 {
2469 struct mddev *mddev = rdev->mddev;
2470 int err = 0;
2471
2472 if (!mddev->pers->hot_remove_disk) {
2473 /* If there is hot_add_disk but no hot_remove_disk
2474 * then added disks for geometry changes,
2475 * and should be added immediately.
2476 */
2477 super_types[mddev->major_version].
2478 validate_super(mddev, rdev);
2479 err = mddev->pers->hot_add_disk(mddev, rdev);
2480 if (err) {
2481 unbind_rdev_from_array(rdev);
2482 export_rdev(rdev);
2483 return err;
2484 }
2485 }
2486 sysfs_notify_dirent_safe(rdev->sysfs_state);
2487
2488 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2489 if (mddev->degraded)
2490 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2491 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2492 md_new_event(mddev);
2493 md_wakeup_thread(mddev->thread);
2494 return 0;
2495 }
2496
2497 /* words written to sysfs files may, or may not, be \n terminated.
2498 * We want to accept with case. For this we use cmd_match.
2499 */
2500 static int cmd_match(const char *cmd, const char *str)
2501 {
2502 /* See if cmd, written into a sysfs file, matches
2503 * str. They must either be the same, or cmd can
2504 * have a trailing newline
2505 */
2506 while (*cmd && *str && *cmd == *str) {
2507 cmd++;
2508 str++;
2509 }
2510 if (*cmd == '\n')
2511 cmd++;
2512 if (*str || *cmd)
2513 return 0;
2514 return 1;
2515 }
2516
2517 struct rdev_sysfs_entry {
2518 struct attribute attr;
2519 ssize_t (*show)(struct md_rdev *, char *);
2520 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2521 };
2522
2523 static ssize_t
2524 state_show(struct md_rdev *rdev, char *page)
2525 {
2526 char *sep = "";
2527 size_t len = 0;
2528 unsigned long flags = ACCESS_ONCE(rdev->flags);
2529
2530 if (test_bit(Faulty, &flags) ||
2531 rdev->badblocks.unacked_exist) {
2532 len+= sprintf(page+len, "%sfaulty",sep);
2533 sep = ",";
2534 }
2535 if (test_bit(In_sync, &flags)) {
2536 len += sprintf(page+len, "%sin_sync",sep);
2537 sep = ",";
2538 }
2539 if (test_bit(Journal, &flags)) {
2540 len += sprintf(page+len, "%sjournal",sep);
2541 sep = ",";
2542 }
2543 if (test_bit(WriteMostly, &flags)) {
2544 len += sprintf(page+len, "%swrite_mostly",sep);
2545 sep = ",";
2546 }
2547 if (test_bit(Blocked, &flags) ||
2548 (rdev->badblocks.unacked_exist
2549 && !test_bit(Faulty, &flags))) {
2550 len += sprintf(page+len, "%sblocked", sep);
2551 sep = ",";
2552 }
2553 if (!test_bit(Faulty, &flags) &&
2554 !test_bit(Journal, &flags) &&
2555 !test_bit(In_sync, &flags)) {
2556 len += sprintf(page+len, "%sspare", sep);
2557 sep = ",";
2558 }
2559 if (test_bit(WriteErrorSeen, &flags)) {
2560 len += sprintf(page+len, "%swrite_error", sep);
2561 sep = ",";
2562 }
2563 if (test_bit(WantReplacement, &flags)) {
2564 len += sprintf(page+len, "%swant_replacement", sep);
2565 sep = ",";
2566 }
2567 if (test_bit(Replacement, &flags)) {
2568 len += sprintf(page+len, "%sreplacement", sep);
2569 sep = ",";
2570 }
2571
2572 return len+sprintf(page+len, "\n");
2573 }
2574
2575 static ssize_t
2576 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2577 {
2578 /* can write
2579 * faulty - simulates an error
2580 * remove - disconnects the device
2581 * writemostly - sets write_mostly
2582 * -writemostly - clears write_mostly
2583 * blocked - sets the Blocked flags
2584 * -blocked - clears the Blocked and possibly simulates an error
2585 * insync - sets Insync providing device isn't active
2586 * -insync - clear Insync for a device with a slot assigned,
2587 * so that it gets rebuilt based on bitmap
2588 * write_error - sets WriteErrorSeen
2589 * -write_error - clears WriteErrorSeen
2590 */
2591 int err = -EINVAL;
2592 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2593 md_error(rdev->mddev, rdev);
2594 if (test_bit(Faulty, &rdev->flags))
2595 err = 0;
2596 else
2597 err = -EBUSY;
2598 } else if (cmd_match(buf, "remove")) {
2599 if (rdev->raid_disk >= 0)
2600 err = -EBUSY;
2601 else {
2602 struct mddev *mddev = rdev->mddev;
2603 err = 0;
2604 if (mddev_is_clustered(mddev))
2605 err = md_cluster_ops->remove_disk(mddev, rdev);
2606
2607 if (err == 0) {
2608 md_kick_rdev_from_array(rdev);
2609 if (mddev->pers)
2610 md_update_sb(mddev, 1);
2611 md_new_event(mddev);
2612 }
2613 }
2614 } else if (cmd_match(buf, "writemostly")) {
2615 set_bit(WriteMostly, &rdev->flags);
2616 err = 0;
2617 } else if (cmd_match(buf, "-writemostly")) {
2618 clear_bit(WriteMostly, &rdev->flags);
2619 err = 0;
2620 } else if (cmd_match(buf, "blocked")) {
2621 set_bit(Blocked, &rdev->flags);
2622 err = 0;
2623 } else if (cmd_match(buf, "-blocked")) {
2624 if (!test_bit(Faulty, &rdev->flags) &&
2625 rdev->badblocks.unacked_exist) {
2626 /* metadata handler doesn't understand badblocks,
2627 * so we need to fail the device
2628 */
2629 md_error(rdev->mddev, rdev);
2630 }
2631 clear_bit(Blocked, &rdev->flags);
2632 clear_bit(BlockedBadBlocks, &rdev->flags);
2633 wake_up(&rdev->blocked_wait);
2634 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2635 md_wakeup_thread(rdev->mddev->thread);
2636
2637 err = 0;
2638 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2639 set_bit(In_sync, &rdev->flags);
2640 err = 0;
2641 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
2642 !test_bit(Journal, &rdev->flags)) {
2643 if (rdev->mddev->pers == NULL) {
2644 clear_bit(In_sync, &rdev->flags);
2645 rdev->saved_raid_disk = rdev->raid_disk;
2646 rdev->raid_disk = -1;
2647 err = 0;
2648 }
2649 } else if (cmd_match(buf, "write_error")) {
2650 set_bit(WriteErrorSeen, &rdev->flags);
2651 err = 0;
2652 } else if (cmd_match(buf, "-write_error")) {
2653 clear_bit(WriteErrorSeen, &rdev->flags);
2654 err = 0;
2655 } else if (cmd_match(buf, "want_replacement")) {
2656 /* Any non-spare device that is not a replacement can
2657 * become want_replacement at any time, but we then need to
2658 * check if recovery is needed.
2659 */
2660 if (rdev->raid_disk >= 0 &&
2661 !test_bit(Journal, &rdev->flags) &&
2662 !test_bit(Replacement, &rdev->flags))
2663 set_bit(WantReplacement, &rdev->flags);
2664 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2665 md_wakeup_thread(rdev->mddev->thread);
2666 err = 0;
2667 } else if (cmd_match(buf, "-want_replacement")) {
2668 /* Clearing 'want_replacement' is always allowed.
2669 * Once replacements starts it is too late though.
2670 */
2671 err = 0;
2672 clear_bit(WantReplacement, &rdev->flags);
2673 } else if (cmd_match(buf, "replacement")) {
2674 /* Can only set a device as a replacement when array has not
2675 * yet been started. Once running, replacement is automatic
2676 * from spares, or by assigning 'slot'.
2677 */
2678 if (rdev->mddev->pers)
2679 err = -EBUSY;
2680 else {
2681 set_bit(Replacement, &rdev->flags);
2682 err = 0;
2683 }
2684 } else if (cmd_match(buf, "-replacement")) {
2685 /* Similarly, can only clear Replacement before start */
2686 if (rdev->mddev->pers)
2687 err = -EBUSY;
2688 else {
2689 clear_bit(Replacement, &rdev->flags);
2690 err = 0;
2691 }
2692 } else if (cmd_match(buf, "re-add")) {
2693 if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1) &&
2694 rdev->saved_raid_disk >= 0) {
2695 /* clear_bit is performed _after_ all the devices
2696 * have their local Faulty bit cleared. If any writes
2697 * happen in the meantime in the local node, they
2698 * will land in the local bitmap, which will be synced
2699 * by this node eventually
2700 */
2701 if (!mddev_is_clustered(rdev->mddev) ||
2702 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
2703 clear_bit(Faulty, &rdev->flags);
2704 err = add_bound_rdev(rdev);
2705 }
2706 } else
2707 err = -EBUSY;
2708 }
2709 if (!err)
2710 sysfs_notify_dirent_safe(rdev->sysfs_state);
2711 return err ? err : len;
2712 }
2713 static struct rdev_sysfs_entry rdev_state =
2714 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
2715
2716 static ssize_t
2717 errors_show(struct md_rdev *rdev, char *page)
2718 {
2719 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2720 }
2721
2722 static ssize_t
2723 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2724 {
2725 unsigned int n;
2726 int rv;
2727
2728 rv = kstrtouint(buf, 10, &n);
2729 if (rv < 0)
2730 return rv;
2731 atomic_set(&rdev->corrected_errors, n);
2732 return len;
2733 }
2734 static struct rdev_sysfs_entry rdev_errors =
2735 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2736
2737 static ssize_t
2738 slot_show(struct md_rdev *rdev, char *page)
2739 {
2740 if (test_bit(Journal, &rdev->flags))
2741 return sprintf(page, "journal\n");
2742 else if (rdev->raid_disk < 0)
2743 return sprintf(page, "none\n");
2744 else
2745 return sprintf(page, "%d\n", rdev->raid_disk);
2746 }
2747
2748 static ssize_t
2749 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2750 {
2751 int slot;
2752 int err;
2753
2754 if (test_bit(Journal, &rdev->flags))
2755 return -EBUSY;
2756 if (strncmp(buf, "none", 4)==0)
2757 slot = -1;
2758 else {
2759 err = kstrtouint(buf, 10, (unsigned int *)&slot);
2760 if (err < 0)
2761 return err;
2762 }
2763 if (rdev->mddev->pers && slot == -1) {
2764 /* Setting 'slot' on an active array requires also
2765 * updating the 'rd%d' link, and communicating
2766 * with the personality with ->hot_*_disk.
2767 * For now we only support removing
2768 * failed/spare devices. This normally happens automatically,
2769 * but not when the metadata is externally managed.
2770 */
2771 if (rdev->raid_disk == -1)
2772 return -EEXIST;
2773 /* personality does all needed checks */
2774 if (rdev->mddev->pers->hot_remove_disk == NULL)
2775 return -EINVAL;
2776 clear_bit(Blocked, &rdev->flags);
2777 remove_and_add_spares(rdev->mddev, rdev);
2778 if (rdev->raid_disk >= 0)
2779 return -EBUSY;
2780 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2781 md_wakeup_thread(rdev->mddev->thread);
2782 } else if (rdev->mddev->pers) {
2783 /* Activating a spare .. or possibly reactivating
2784 * if we ever get bitmaps working here.
2785 */
2786 int err;
2787
2788 if (rdev->raid_disk != -1)
2789 return -EBUSY;
2790
2791 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2792 return -EBUSY;
2793
2794 if (rdev->mddev->pers->hot_add_disk == NULL)
2795 return -EINVAL;
2796
2797 if (slot >= rdev->mddev->raid_disks &&
2798 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2799 return -ENOSPC;
2800
2801 rdev->raid_disk = slot;
2802 if (test_bit(In_sync, &rdev->flags))
2803 rdev->saved_raid_disk = slot;
2804 else
2805 rdev->saved_raid_disk = -1;
2806 clear_bit(In_sync, &rdev->flags);
2807 clear_bit(Bitmap_sync, &rdev->flags);
2808 err = rdev->mddev->pers->
2809 hot_add_disk(rdev->mddev, rdev);
2810 if (err) {
2811 rdev->raid_disk = -1;
2812 return err;
2813 } else
2814 sysfs_notify_dirent_safe(rdev->sysfs_state);
2815 if (sysfs_link_rdev(rdev->mddev, rdev))
2816 /* failure here is OK */;
2817 /* don't wakeup anyone, leave that to userspace. */
2818 } else {
2819 if (slot >= rdev->mddev->raid_disks &&
2820 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2821 return -ENOSPC;
2822 rdev->raid_disk = slot;
2823 /* assume it is working */
2824 clear_bit(Faulty, &rdev->flags);
2825 clear_bit(WriteMostly, &rdev->flags);
2826 set_bit(In_sync, &rdev->flags);
2827 sysfs_notify_dirent_safe(rdev->sysfs_state);
2828 }
2829 return len;
2830 }
2831
2832 static struct rdev_sysfs_entry rdev_slot =
2833 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2834
2835 static ssize_t
2836 offset_show(struct md_rdev *rdev, char *page)
2837 {
2838 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2839 }
2840
2841 static ssize_t
2842 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2843 {
2844 unsigned long long offset;
2845 if (kstrtoull(buf, 10, &offset) < 0)
2846 return -EINVAL;
2847 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2848 return -EBUSY;
2849 if (rdev->sectors && rdev->mddev->external)
2850 /* Must set offset before size, so overlap checks
2851 * can be sane */
2852 return -EBUSY;
2853 rdev->data_offset = offset;
2854 rdev->new_data_offset = offset;
2855 return len;
2856 }
2857
2858 static struct rdev_sysfs_entry rdev_offset =
2859 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2860
2861 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2862 {
2863 return sprintf(page, "%llu\n",
2864 (unsigned long long)rdev->new_data_offset);
2865 }
2866
2867 static ssize_t new_offset_store(struct md_rdev *rdev,
2868 const char *buf, size_t len)
2869 {
2870 unsigned long long new_offset;
2871 struct mddev *mddev = rdev->mddev;
2872
2873 if (kstrtoull(buf, 10, &new_offset) < 0)
2874 return -EINVAL;
2875
2876 if (mddev->sync_thread ||
2877 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
2878 return -EBUSY;
2879 if (new_offset == rdev->data_offset)
2880 /* reset is always permitted */
2881 ;
2882 else if (new_offset > rdev->data_offset) {
2883 /* must not push array size beyond rdev_sectors */
2884 if (new_offset - rdev->data_offset
2885 + mddev->dev_sectors > rdev->sectors)
2886 return -E2BIG;
2887 }
2888 /* Metadata worries about other space details. */
2889
2890 /* decreasing the offset is inconsistent with a backwards
2891 * reshape.
2892 */
2893 if (new_offset < rdev->data_offset &&
2894 mddev->reshape_backwards)
2895 return -EINVAL;
2896 /* Increasing offset is inconsistent with forwards
2897 * reshape. reshape_direction should be set to
2898 * 'backwards' first.
2899 */
2900 if (new_offset > rdev->data_offset &&
2901 !mddev->reshape_backwards)
2902 return -EINVAL;
2903
2904 if (mddev->pers && mddev->persistent &&
2905 !super_types[mddev->major_version]
2906 .allow_new_offset(rdev, new_offset))
2907 return -E2BIG;
2908 rdev->new_data_offset = new_offset;
2909 if (new_offset > rdev->data_offset)
2910 mddev->reshape_backwards = 1;
2911 else if (new_offset < rdev->data_offset)
2912 mddev->reshape_backwards = 0;
2913
2914 return len;
2915 }
2916 static struct rdev_sysfs_entry rdev_new_offset =
2917 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2918
2919 static ssize_t
2920 rdev_size_show(struct md_rdev *rdev, char *page)
2921 {
2922 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2923 }
2924
2925 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2926 {
2927 /* check if two start/length pairs overlap */
2928 if (s1+l1 <= s2)
2929 return 0;
2930 if (s2+l2 <= s1)
2931 return 0;
2932 return 1;
2933 }
2934
2935 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2936 {
2937 unsigned long long blocks;
2938 sector_t new;
2939
2940 if (kstrtoull(buf, 10, &blocks) < 0)
2941 return -EINVAL;
2942
2943 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2944 return -EINVAL; /* sector conversion overflow */
2945
2946 new = blocks * 2;
2947 if (new != blocks * 2)
2948 return -EINVAL; /* unsigned long long to sector_t overflow */
2949
2950 *sectors = new;
2951 return 0;
2952 }
2953
2954 static ssize_t
2955 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
2956 {
2957 struct mddev *my_mddev = rdev->mddev;
2958 sector_t oldsectors = rdev->sectors;
2959 sector_t sectors;
2960
2961 if (test_bit(Journal, &rdev->flags))
2962 return -EBUSY;
2963 if (strict_blocks_to_sectors(buf, &sectors) < 0)
2964 return -EINVAL;
2965 if (rdev->data_offset != rdev->new_data_offset)
2966 return -EINVAL; /* too confusing */
2967 if (my_mddev->pers && rdev->raid_disk >= 0) {
2968 if (my_mddev->persistent) {
2969 sectors = super_types[my_mddev->major_version].
2970 rdev_size_change(rdev, sectors);
2971 if (!sectors)
2972 return -EBUSY;
2973 } else if (!sectors)
2974 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2975 rdev->data_offset;
2976 if (!my_mddev->pers->resize)
2977 /* Cannot change size for RAID0 or Linear etc */
2978 return -EINVAL;
2979 }
2980 if (sectors < my_mddev->dev_sectors)
2981 return -EINVAL; /* component must fit device */
2982
2983 rdev->sectors = sectors;
2984 if (sectors > oldsectors && my_mddev->external) {
2985 /* Need to check that all other rdevs with the same
2986 * ->bdev do not overlap. 'rcu' is sufficient to walk
2987 * the rdev lists safely.
2988 * This check does not provide a hard guarantee, it
2989 * just helps avoid dangerous mistakes.
2990 */
2991 struct mddev *mddev;
2992 int overlap = 0;
2993 struct list_head *tmp;
2994
2995 rcu_read_lock();
2996 for_each_mddev(mddev, tmp) {
2997 struct md_rdev *rdev2;
2998
2999 rdev_for_each(rdev2, mddev)
3000 if (rdev->bdev == rdev2->bdev &&
3001 rdev != rdev2 &&
3002 overlaps(rdev->data_offset, rdev->sectors,
3003 rdev2->data_offset,
3004 rdev2->sectors)) {
3005 overlap = 1;
3006 break;
3007 }
3008 if (overlap) {
3009 mddev_put(mddev);
3010 break;
3011 }
3012 }
3013 rcu_read_unlock();
3014 if (overlap) {
3015 /* Someone else could have slipped in a size
3016 * change here, but doing so is just silly.
3017 * We put oldsectors back because we *know* it is
3018 * safe, and trust userspace not to race with
3019 * itself
3020 */
3021 rdev->sectors = oldsectors;
3022 return -EBUSY;
3023 }
3024 }
3025 return len;
3026 }
3027
3028 static struct rdev_sysfs_entry rdev_size =
3029 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3030
3031 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3032 {
3033 unsigned long long recovery_start = rdev->recovery_offset;
3034
3035 if (test_bit(In_sync, &rdev->flags) ||
3036 recovery_start == MaxSector)
3037 return sprintf(page, "none\n");
3038
3039 return sprintf(page, "%llu\n", recovery_start);
3040 }
3041
3042 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3043 {
3044 unsigned long long recovery_start;
3045
3046 if (cmd_match(buf, "none"))
3047 recovery_start = MaxSector;
3048 else if (kstrtoull(buf, 10, &recovery_start))
3049 return -EINVAL;
3050
3051 if (rdev->mddev->pers &&
3052 rdev->raid_disk >= 0)
3053 return -EBUSY;
3054
3055 rdev->recovery_offset = recovery_start;
3056 if (recovery_start == MaxSector)
3057 set_bit(In_sync, &rdev->flags);
3058 else
3059 clear_bit(In_sync, &rdev->flags);
3060 return len;
3061 }
3062
3063 static struct rdev_sysfs_entry rdev_recovery_start =
3064 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3065
3066 static ssize_t
3067 badblocks_show(struct badblocks *bb, char *page, int unack);
3068 static ssize_t
3069 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
3070
3071 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3072 {
3073 return badblocks_show(&rdev->badblocks, page, 0);
3074 }
3075 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3076 {
3077 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3078 /* Maybe that ack was all we needed */
3079 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3080 wake_up(&rdev->blocked_wait);
3081 return rv;
3082 }
3083 static struct rdev_sysfs_entry rdev_bad_blocks =
3084 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3085
3086 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3087 {
3088 return badblocks_show(&rdev->badblocks, page, 1);
3089 }
3090 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3091 {
3092 return badblocks_store(&rdev->badblocks, page, len, 1);
3093 }
3094 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3095 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3096
3097 static struct attribute *rdev_default_attrs[] = {
3098 &rdev_state.attr,
3099 &rdev_errors.attr,
3100 &rdev_slot.attr,
3101 &rdev_offset.attr,
3102 &rdev_new_offset.attr,
3103 &rdev_size.attr,
3104 &rdev_recovery_start.attr,
3105 &rdev_bad_blocks.attr,
3106 &rdev_unack_bad_blocks.attr,
3107 NULL,
3108 };
3109 static ssize_t
3110 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3111 {
3112 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3113 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3114
3115 if (!entry->show)
3116 return -EIO;
3117 if (!rdev->mddev)
3118 return -EBUSY;
3119 return entry->show(rdev, page);
3120 }
3121
3122 static ssize_t
3123 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3124 const char *page, size_t length)
3125 {
3126 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3127 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3128 ssize_t rv;
3129 struct mddev *mddev = rdev->mddev;
3130
3131 if (!entry->store)
3132 return -EIO;
3133 if (!capable(CAP_SYS_ADMIN))
3134 return -EACCES;
3135 rv = mddev ? mddev_lock(mddev): -EBUSY;
3136 if (!rv) {
3137 if (rdev->mddev == NULL)
3138 rv = -EBUSY;
3139 else
3140 rv = entry->store(rdev, page, length);
3141 mddev_unlock(mddev);
3142 }
3143 return rv;
3144 }
3145
3146 static void rdev_free(struct kobject *ko)
3147 {
3148 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3149 kfree(rdev);
3150 }
3151 static const struct sysfs_ops rdev_sysfs_ops = {
3152 .show = rdev_attr_show,
3153 .store = rdev_attr_store,
3154 };
3155 static struct kobj_type rdev_ktype = {
3156 .release = rdev_free,
3157 .sysfs_ops = &rdev_sysfs_ops,
3158 .default_attrs = rdev_default_attrs,
3159 };
3160
3161 int md_rdev_init(struct md_rdev *rdev)
3162 {
3163 rdev->desc_nr = -1;
3164 rdev->saved_raid_disk = -1;
3165 rdev->raid_disk = -1;
3166 rdev->flags = 0;
3167 rdev->data_offset = 0;
3168 rdev->new_data_offset = 0;
3169 rdev->sb_events = 0;
3170 rdev->last_read_error.tv_sec = 0;
3171 rdev->last_read_error.tv_nsec = 0;
3172 rdev->sb_loaded = 0;
3173 rdev->bb_page = NULL;
3174 atomic_set(&rdev->nr_pending, 0);
3175 atomic_set(&rdev->read_errors, 0);
3176 atomic_set(&rdev->corrected_errors, 0);
3177
3178 INIT_LIST_HEAD(&rdev->same_set);
3179 init_waitqueue_head(&rdev->blocked_wait);
3180
3181 /* Add space to store bad block list.
3182 * This reserves the space even on arrays where it cannot
3183 * be used - I wonder if that matters
3184 */
3185 rdev->badblocks.count = 0;
3186 rdev->badblocks.shift = -1; /* disabled until explicitly enabled */
3187 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3188 seqlock_init(&rdev->badblocks.lock);
3189 if (rdev->badblocks.page == NULL)
3190 return -ENOMEM;
3191
3192 return 0;
3193 }
3194 EXPORT_SYMBOL_GPL(md_rdev_init);
3195 /*
3196 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3197 *
3198 * mark the device faulty if:
3199 *
3200 * - the device is nonexistent (zero size)
3201 * - the device has no valid superblock
3202 *
3203 * a faulty rdev _never_ has rdev->sb set.
3204 */
3205 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3206 {
3207 char b[BDEVNAME_SIZE];
3208 int err;
3209 struct md_rdev *rdev;
3210 sector_t size;
3211
3212 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3213 if (!rdev) {
3214 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3215 return ERR_PTR(-ENOMEM);
3216 }
3217
3218 err = md_rdev_init(rdev);
3219 if (err)
3220 goto abort_free;
3221 err = alloc_disk_sb(rdev);
3222 if (err)
3223 goto abort_free;
3224
3225 err = lock_rdev(rdev, newdev, super_format == -2);
3226 if (err)
3227 goto abort_free;
3228
3229 kobject_init(&rdev->kobj, &rdev_ktype);
3230
3231 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3232 if (!size) {
3233 printk(KERN_WARNING
3234 "md: %s has zero or unknown size, marking faulty!\n",
3235 bdevname(rdev->bdev,b));
3236 err = -EINVAL;
3237 goto abort_free;
3238 }
3239
3240 if (super_format >= 0) {
3241 err = super_types[super_format].
3242 load_super(rdev, NULL, super_minor);
3243 if (err == -EINVAL) {
3244 printk(KERN_WARNING
3245 "md: %s does not have a valid v%d.%d "
3246 "superblock, not importing!\n",
3247 bdevname(rdev->bdev,b),
3248 super_format, super_minor);
3249 goto abort_free;
3250 }
3251 if (err < 0) {
3252 printk(KERN_WARNING
3253 "md: could not read %s's sb, not importing!\n",
3254 bdevname(rdev->bdev,b));
3255 goto abort_free;
3256 }
3257 }
3258
3259 return rdev;
3260
3261 abort_free:
3262 if (rdev->bdev)
3263 unlock_rdev(rdev);
3264 md_rdev_clear(rdev);
3265 kfree(rdev);
3266 return ERR_PTR(err);
3267 }
3268
3269 /*
3270 * Check a full RAID array for plausibility
3271 */
3272
3273 static void analyze_sbs(struct mddev *mddev)
3274 {
3275 int i;
3276 struct md_rdev *rdev, *freshest, *tmp;
3277 char b[BDEVNAME_SIZE];
3278
3279 freshest = NULL;
3280 rdev_for_each_safe(rdev, tmp, mddev)
3281 switch (super_types[mddev->major_version].
3282 load_super(rdev, freshest, mddev->minor_version)) {
3283 case 1:
3284 freshest = rdev;
3285 break;
3286 case 0:
3287 break;
3288 default:
3289 printk( KERN_ERR \
3290 "md: fatal superblock inconsistency in %s"
3291 " -- removing from array\n",
3292 bdevname(rdev->bdev,b));
3293 md_kick_rdev_from_array(rdev);
3294 }
3295
3296 super_types[mddev->major_version].
3297 validate_super(mddev, freshest);
3298
3299 i = 0;
3300 rdev_for_each_safe(rdev, tmp, mddev) {
3301 if (mddev->max_disks &&
3302 (rdev->desc_nr >= mddev->max_disks ||
3303 i > mddev->max_disks)) {
3304 printk(KERN_WARNING
3305 "md: %s: %s: only %d devices permitted\n",
3306 mdname(mddev), bdevname(rdev->bdev, b),
3307 mddev->max_disks);
3308 md_kick_rdev_from_array(rdev);
3309 continue;
3310 }
3311 if (rdev != freshest) {
3312 if (super_types[mddev->major_version].
3313 validate_super(mddev, rdev)) {
3314 printk(KERN_WARNING "md: kicking non-fresh %s"
3315 " from array!\n",
3316 bdevname(rdev->bdev,b));
3317 md_kick_rdev_from_array(rdev);
3318 continue;
3319 }
3320 }
3321 if (mddev->level == LEVEL_MULTIPATH) {
3322 rdev->desc_nr = i++;
3323 rdev->raid_disk = rdev->desc_nr;
3324 set_bit(In_sync, &rdev->flags);
3325 } else if (rdev->raid_disk >=
3326 (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3327 !test_bit(Journal, &rdev->flags)) {
3328 rdev->raid_disk = -1;
3329 clear_bit(In_sync, &rdev->flags);
3330 }
3331 }
3332 }
3333
3334 /* Read a fixed-point number.
3335 * Numbers in sysfs attributes should be in "standard" units where
3336 * possible, so time should be in seconds.
3337 * However we internally use a a much smaller unit such as
3338 * milliseconds or jiffies.
3339 * This function takes a decimal number with a possible fractional
3340 * component, and produces an integer which is the result of
3341 * multiplying that number by 10^'scale'.
3342 * all without any floating-point arithmetic.
3343 */
3344 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3345 {
3346 unsigned long result = 0;
3347 long decimals = -1;
3348 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3349 if (*cp == '.')
3350 decimals = 0;
3351 else if (decimals < scale) {
3352 unsigned int value;
3353 value = *cp - '0';
3354 result = result * 10 + value;
3355 if (decimals >= 0)
3356 decimals++;
3357 }
3358 cp++;
3359 }
3360 if (*cp == '\n')
3361 cp++;
3362 if (*cp)
3363 return -EINVAL;
3364 if (decimals < 0)
3365 decimals = 0;
3366 while (decimals < scale) {
3367 result *= 10;
3368 decimals ++;
3369 }
3370 *res = result;
3371 return 0;
3372 }
3373
3374 static ssize_t
3375 safe_delay_show(struct mddev *mddev, char *page)
3376 {
3377 int msec = (mddev->safemode_delay*1000)/HZ;
3378 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3379 }
3380 static ssize_t
3381 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3382 {
3383 unsigned long msec;
3384
3385 if (mddev_is_clustered(mddev)) {
3386 pr_info("md: Safemode is disabled for clustered mode\n");
3387 return -EINVAL;
3388 }
3389
3390 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3391 return -EINVAL;
3392 if (msec == 0)
3393 mddev->safemode_delay = 0;
3394 else {
3395 unsigned long old_delay = mddev->safemode_delay;
3396 unsigned long new_delay = (msec*HZ)/1000;
3397
3398 if (new_delay == 0)
3399 new_delay = 1;
3400 mddev->safemode_delay = new_delay;
3401 if (new_delay < old_delay || old_delay == 0)
3402 mod_timer(&mddev->safemode_timer, jiffies+1);
3403 }
3404 return len;
3405 }
3406 static struct md_sysfs_entry md_safe_delay =
3407 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3408
3409 static ssize_t
3410 level_show(struct mddev *mddev, char *page)
3411 {
3412 struct md_personality *p;
3413 int ret;
3414 spin_lock(&mddev->lock);
3415 p = mddev->pers;
3416 if (p)
3417 ret = sprintf(page, "%s\n", p->name);
3418 else if (mddev->clevel[0])
3419 ret = sprintf(page, "%s\n", mddev->clevel);
3420 else if (mddev->level != LEVEL_NONE)
3421 ret = sprintf(page, "%d\n", mddev->level);
3422 else
3423 ret = 0;
3424 spin_unlock(&mddev->lock);
3425 return ret;
3426 }
3427
3428 static ssize_t
3429 level_store(struct mddev *mddev, const char *buf, size_t len)
3430 {
3431 char clevel[16];
3432 ssize_t rv;
3433 size_t slen = len;
3434 struct md_personality *pers, *oldpers;
3435 long level;
3436 void *priv, *oldpriv;
3437 struct md_rdev *rdev;
3438
3439 if (slen == 0 || slen >= sizeof(clevel))
3440 return -EINVAL;
3441
3442 rv = mddev_lock(mddev);
3443 if (rv)
3444 return rv;
3445
3446 if (mddev->pers == NULL) {
3447 strncpy(mddev->clevel, buf, slen);
3448 if (mddev->clevel[slen-1] == '\n')
3449 slen--;
3450 mddev->clevel[slen] = 0;
3451 mddev->level = LEVEL_NONE;
3452 rv = len;
3453 goto out_unlock;
3454 }
3455 rv = -EROFS;
3456 if (mddev->ro)
3457 goto out_unlock;
3458
3459 /* request to change the personality. Need to ensure:
3460 * - array is not engaged in resync/recovery/reshape
3461 * - old personality can be suspended
3462 * - new personality will access other array.
3463 */
3464
3465 rv = -EBUSY;
3466 if (mddev->sync_thread ||
3467 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3468 mddev->reshape_position != MaxSector ||
3469 mddev->sysfs_active)
3470 goto out_unlock;
3471
3472 rv = -EINVAL;
3473 if (!mddev->pers->quiesce) {
3474 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3475 mdname(mddev), mddev->pers->name);
3476 goto out_unlock;
3477 }
3478
3479 /* Now find the new personality */
3480 strncpy(clevel, buf, slen);
3481 if (clevel[slen-1] == '\n')
3482 slen--;
3483 clevel[slen] = 0;
3484 if (kstrtol(clevel, 10, &level))
3485 level = LEVEL_NONE;
3486
3487 if (request_module("md-%s", clevel) != 0)
3488 request_module("md-level-%s", clevel);
3489 spin_lock(&pers_lock);
3490 pers = find_pers(level, clevel);
3491 if (!pers || !try_module_get(pers->owner)) {
3492 spin_unlock(&pers_lock);
3493 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3494 rv = -EINVAL;
3495 goto out_unlock;
3496 }
3497 spin_unlock(&pers_lock);
3498
3499 if (pers == mddev->pers) {
3500 /* Nothing to do! */
3501 module_put(pers->owner);
3502 rv = len;
3503 goto out_unlock;
3504 }
3505 if (!pers->takeover) {
3506 module_put(pers->owner);
3507 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3508 mdname(mddev), clevel);
3509 rv = -EINVAL;
3510 goto out_unlock;
3511 }
3512
3513 rdev_for_each(rdev, mddev)
3514 rdev->new_raid_disk = rdev->raid_disk;
3515
3516 /* ->takeover must set new_* and/or delta_disks
3517 * if it succeeds, and may set them when it fails.
3518 */
3519 priv = pers->takeover(mddev);
3520 if (IS_ERR(priv)) {
3521 mddev->new_level = mddev->level;
3522 mddev->new_layout = mddev->layout;
3523 mddev->new_chunk_sectors = mddev->chunk_sectors;
3524 mddev->raid_disks -= mddev->delta_disks;
3525 mddev->delta_disks = 0;
3526 mddev->reshape_backwards = 0;
3527 module_put(pers->owner);
3528 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3529 mdname(mddev), clevel);
3530 rv = PTR_ERR(priv);
3531 goto out_unlock;
3532 }
3533
3534 /* Looks like we have a winner */
3535 mddev_suspend(mddev);
3536 mddev_detach(mddev);
3537
3538 spin_lock(&mddev->lock);
3539 oldpers = mddev->pers;
3540 oldpriv = mddev->private;
3541 mddev->pers = pers;
3542 mddev->private = priv;
3543 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3544 mddev->level = mddev->new_level;
3545 mddev->layout = mddev->new_layout;
3546 mddev->chunk_sectors = mddev->new_chunk_sectors;
3547 mddev->delta_disks = 0;
3548 mddev->reshape_backwards = 0;
3549 mddev->degraded = 0;
3550 spin_unlock(&mddev->lock);
3551
3552 if (oldpers->sync_request == NULL &&
3553 mddev->external) {
3554 /* We are converting from a no-redundancy array
3555 * to a redundancy array and metadata is managed
3556 * externally so we need to be sure that writes
3557 * won't block due to a need to transition
3558 * clean->dirty
3559 * until external management is started.
3560 */
3561 mddev->in_sync = 0;
3562 mddev->safemode_delay = 0;
3563 mddev->safemode = 0;
3564 }
3565
3566 oldpers->free(mddev, oldpriv);
3567
3568 if (oldpers->sync_request == NULL &&
3569 pers->sync_request != NULL) {
3570 /* need to add the md_redundancy_group */
3571 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3572 printk(KERN_WARNING
3573 "md: cannot register extra attributes for %s\n",
3574 mdname(mddev));
3575 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3576 }
3577 if (oldpers->sync_request != NULL &&
3578 pers->sync_request == NULL) {
3579 /* need to remove the md_redundancy_group */
3580 if (mddev->to_remove == NULL)
3581 mddev->to_remove = &md_redundancy_group;
3582 }
3583
3584 rdev_for_each(rdev, mddev) {
3585 if (rdev->raid_disk < 0)
3586 continue;
3587 if (rdev->new_raid_disk >= mddev->raid_disks)
3588 rdev->new_raid_disk = -1;
3589 if (rdev->new_raid_disk == rdev->raid_disk)
3590 continue;
3591 sysfs_unlink_rdev(mddev, rdev);
3592 }
3593 rdev_for_each(rdev, mddev) {
3594 if (rdev->raid_disk < 0)
3595 continue;
3596 if (rdev->new_raid_disk == rdev->raid_disk)
3597 continue;
3598 rdev->raid_disk = rdev->new_raid_disk;
3599 if (rdev->raid_disk < 0)
3600 clear_bit(In_sync, &rdev->flags);
3601 else {
3602 if (sysfs_link_rdev(mddev, rdev))
3603 printk(KERN_WARNING "md: cannot register rd%d"
3604 " for %s after level change\n",
3605 rdev->raid_disk, mdname(mddev));
3606 }
3607 }
3608
3609 if (pers->sync_request == NULL) {
3610 /* this is now an array without redundancy, so
3611 * it must always be in_sync
3612 */
3613 mddev->in_sync = 1;
3614 del_timer_sync(&mddev->safemode_timer);
3615 }
3616 blk_set_stacking_limits(&mddev->queue->limits);
3617 pers->run(mddev);
3618 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3619 mddev_resume(mddev);
3620 if (!mddev->thread)
3621 md_update_sb(mddev, 1);
3622 sysfs_notify(&mddev->kobj, NULL, "level");
3623 md_new_event(mddev);
3624 rv = len;
3625 out_unlock:
3626 mddev_unlock(mddev);
3627 return rv;
3628 }
3629
3630 static struct md_sysfs_entry md_level =
3631 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3632
3633 static ssize_t
3634 layout_show(struct mddev *mddev, char *page)
3635 {
3636 /* just a number, not meaningful for all levels */
3637 if (mddev->reshape_position != MaxSector &&
3638 mddev->layout != mddev->new_layout)
3639 return sprintf(page, "%d (%d)\n",
3640 mddev->new_layout, mddev->layout);
3641 return sprintf(page, "%d\n", mddev->layout);
3642 }
3643
3644 static ssize_t
3645 layout_store(struct mddev *mddev, const char *buf, size_t len)
3646 {
3647 unsigned int n;
3648 int err;
3649
3650 err = kstrtouint(buf, 10, &n);
3651 if (err < 0)
3652 return err;
3653 err = mddev_lock(mddev);
3654 if (err)
3655 return err;
3656
3657 if (mddev->pers) {
3658 if (mddev->pers->check_reshape == NULL)
3659 err = -EBUSY;
3660 else if (mddev->ro)
3661 err = -EROFS;
3662 else {
3663 mddev->new_layout = n;
3664 err = mddev->pers->check_reshape(mddev);
3665 if (err)
3666 mddev->new_layout = mddev->layout;
3667 }
3668 } else {
3669 mddev->new_layout = n;
3670 if (mddev->reshape_position == MaxSector)
3671 mddev->layout = n;
3672 }
3673 mddev_unlock(mddev);
3674 return err ?: len;
3675 }
3676 static struct md_sysfs_entry md_layout =
3677 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3678
3679 static ssize_t
3680 raid_disks_show(struct mddev *mddev, char *page)
3681 {
3682 if (mddev->raid_disks == 0)
3683 return 0;
3684 if (mddev->reshape_position != MaxSector &&
3685 mddev->delta_disks != 0)
3686 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3687 mddev->raid_disks - mddev->delta_disks);
3688 return sprintf(page, "%d\n", mddev->raid_disks);
3689 }
3690
3691 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3692
3693 static ssize_t
3694 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3695 {
3696 unsigned int n;
3697 int err;
3698
3699 err = kstrtouint(buf, 10, &n);
3700 if (err < 0)
3701 return err;
3702
3703 err = mddev_lock(mddev);
3704 if (err)
3705 return err;
3706 if (mddev->pers)
3707 err = update_raid_disks(mddev, n);
3708 else if (mddev->reshape_position != MaxSector) {
3709 struct md_rdev *rdev;
3710 int olddisks = mddev->raid_disks - mddev->delta_disks;
3711
3712 err = -EINVAL;
3713 rdev_for_each(rdev, mddev) {
3714 if (olddisks < n &&
3715 rdev->data_offset < rdev->new_data_offset)
3716 goto out_unlock;
3717 if (olddisks > n &&
3718 rdev->data_offset > rdev->new_data_offset)
3719 goto out_unlock;
3720 }
3721 err = 0;
3722 mddev->delta_disks = n - olddisks;
3723 mddev->raid_disks = n;
3724 mddev->reshape_backwards = (mddev->delta_disks < 0);
3725 } else
3726 mddev->raid_disks = n;
3727 out_unlock:
3728 mddev_unlock(mddev);
3729 return err ? err : len;
3730 }
3731 static struct md_sysfs_entry md_raid_disks =
3732 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3733
3734 static ssize_t
3735 chunk_size_show(struct mddev *mddev, char *page)
3736 {
3737 if (mddev->reshape_position != MaxSector &&
3738 mddev->chunk_sectors != mddev->new_chunk_sectors)
3739 return sprintf(page, "%d (%d)\n",
3740 mddev->new_chunk_sectors << 9,
3741 mddev->chunk_sectors << 9);
3742 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3743 }
3744
3745 static ssize_t
3746 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3747 {
3748 unsigned long n;
3749 int err;
3750
3751 err = kstrtoul(buf, 10, &n);
3752 if (err < 0)
3753 return err;
3754
3755 err = mddev_lock(mddev);
3756 if (err)
3757 return err;
3758 if (mddev->pers) {
3759 if (mddev->pers->check_reshape == NULL)
3760 err = -EBUSY;
3761 else if (mddev->ro)
3762 err = -EROFS;
3763 else {
3764 mddev->new_chunk_sectors = n >> 9;
3765 err = mddev->pers->check_reshape(mddev);
3766 if (err)
3767 mddev->new_chunk_sectors = mddev->chunk_sectors;
3768 }
3769 } else {
3770 mddev->new_chunk_sectors = n >> 9;
3771 if (mddev->reshape_position == MaxSector)
3772 mddev->chunk_sectors = n >> 9;
3773 }
3774 mddev_unlock(mddev);
3775 return err ?: len;
3776 }
3777 static struct md_sysfs_entry md_chunk_size =
3778 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3779
3780 static ssize_t
3781 resync_start_show(struct mddev *mddev, char *page)
3782 {
3783 if (mddev->recovery_cp == MaxSector)
3784 return sprintf(page, "none\n");
3785 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3786 }
3787
3788 static ssize_t
3789 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3790 {
3791 unsigned long long n;
3792 int err;
3793
3794 if (cmd_match(buf, "none"))
3795 n = MaxSector;
3796 else {
3797 err = kstrtoull(buf, 10, &n);
3798 if (err < 0)
3799 return err;
3800 if (n != (sector_t)n)
3801 return -EINVAL;
3802 }
3803
3804 err = mddev_lock(mddev);
3805 if (err)
3806 return err;
3807 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3808 err = -EBUSY;
3809
3810 if (!err) {
3811 mddev->recovery_cp = n;
3812 if (mddev->pers)
3813 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3814 }
3815 mddev_unlock(mddev);
3816 return err ?: len;
3817 }
3818 static struct md_sysfs_entry md_resync_start =
3819 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
3820 resync_start_show, resync_start_store);
3821
3822 /*
3823 * The array state can be:
3824 *
3825 * clear
3826 * No devices, no size, no level
3827 * Equivalent to STOP_ARRAY ioctl
3828 * inactive
3829 * May have some settings, but array is not active
3830 * all IO results in error
3831 * When written, doesn't tear down array, but just stops it
3832 * suspended (not supported yet)
3833 * All IO requests will block. The array can be reconfigured.
3834 * Writing this, if accepted, will block until array is quiescent
3835 * readonly
3836 * no resync can happen. no superblocks get written.
3837 * write requests fail
3838 * read-auto
3839 * like readonly, but behaves like 'clean' on a write request.
3840 *
3841 * clean - no pending writes, but otherwise active.
3842 * When written to inactive array, starts without resync
3843 * If a write request arrives then
3844 * if metadata is known, mark 'dirty' and switch to 'active'.
3845 * if not known, block and switch to write-pending
3846 * If written to an active array that has pending writes, then fails.
3847 * active
3848 * fully active: IO and resync can be happening.
3849 * When written to inactive array, starts with resync
3850 *
3851 * write-pending
3852 * clean, but writes are blocked waiting for 'active' to be written.
3853 *
3854 * active-idle
3855 * like active, but no writes have been seen for a while (100msec).
3856 *
3857 */
3858 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3859 write_pending, active_idle, bad_word};
3860 static char *array_states[] = {
3861 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3862 "write-pending", "active-idle", NULL };
3863
3864 static int match_word(const char *word, char **list)
3865 {
3866 int n;
3867 for (n=0; list[n]; n++)
3868 if (cmd_match(word, list[n]))
3869 break;
3870 return n;
3871 }
3872
3873 static ssize_t
3874 array_state_show(struct mddev *mddev, char *page)
3875 {
3876 enum array_state st = inactive;
3877
3878 if (mddev->pers)
3879 switch(mddev->ro) {
3880 case 1:
3881 st = readonly;
3882 break;
3883 case 2:
3884 st = read_auto;
3885 break;
3886 case 0:
3887 if (mddev->in_sync)
3888 st = clean;
3889 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3890 st = write_pending;
3891 else if (mddev->safemode)
3892 st = active_idle;
3893 else
3894 st = active;
3895 }
3896 else {
3897 if (list_empty(&mddev->disks) &&
3898 mddev->raid_disks == 0 &&
3899 mddev->dev_sectors == 0)
3900 st = clear;
3901 else
3902 st = inactive;
3903 }
3904 return sprintf(page, "%s\n", array_states[st]);
3905 }
3906
3907 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
3908 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
3909 static int do_md_run(struct mddev *mddev);
3910 static int restart_array(struct mddev *mddev);
3911
3912 static ssize_t
3913 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3914 {
3915 int err;
3916 enum array_state st = match_word(buf, array_states);
3917
3918 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
3919 /* don't take reconfig_mutex when toggling between
3920 * clean and active
3921 */
3922 spin_lock(&mddev->lock);
3923 if (st == active) {
3924 restart_array(mddev);
3925 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3926 wake_up(&mddev->sb_wait);
3927 err = 0;
3928 } else /* st == clean */ {
3929 restart_array(mddev);
3930 if (atomic_read(&mddev->writes_pending) == 0) {
3931 if (mddev->in_sync == 0) {
3932 mddev->in_sync = 1;
3933 if (mddev->safemode == 1)
3934 mddev->safemode = 0;
3935 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3936 }
3937 err = 0;
3938 } else
3939 err = -EBUSY;
3940 }
3941 spin_unlock(&mddev->lock);
3942 return err ?: len;
3943 }
3944 err = mddev_lock(mddev);
3945 if (err)
3946 return err;
3947 err = -EINVAL;
3948 switch(st) {
3949 case bad_word:
3950 break;
3951 case clear:
3952 /* stopping an active array */
3953 err = do_md_stop(mddev, 0, NULL);
3954 break;
3955 case inactive:
3956 /* stopping an active array */
3957 if (mddev->pers)
3958 err = do_md_stop(mddev, 2, NULL);
3959 else
3960 err = 0; /* already inactive */
3961 break;
3962 case suspended:
3963 break; /* not supported yet */
3964 case readonly:
3965 if (mddev->pers)
3966 err = md_set_readonly(mddev, NULL);
3967 else {
3968 mddev->ro = 1;
3969 set_disk_ro(mddev->gendisk, 1);
3970 err = do_md_run(mddev);
3971 }
3972 break;
3973 case read_auto:
3974 if (mddev->pers) {
3975 if (mddev->ro == 0)
3976 err = md_set_readonly(mddev, NULL);
3977 else if (mddev->ro == 1)
3978 err = restart_array(mddev);
3979 if (err == 0) {
3980 mddev->ro = 2;
3981 set_disk_ro(mddev->gendisk, 0);
3982 }
3983 } else {
3984 mddev->ro = 2;
3985 err = do_md_run(mddev);
3986 }
3987 break;
3988 case clean:
3989 if (mddev->pers) {
3990 err = restart_array(mddev);
3991 if (err)
3992 break;
3993 spin_lock(&mddev->lock);
3994 if (atomic_read(&mddev->writes_pending) == 0) {
3995 if (mddev->in_sync == 0) {
3996 mddev->in_sync = 1;
3997 if (mddev->safemode == 1)
3998 mddev->safemode = 0;
3999 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
4000 }
4001 err = 0;
4002 } else
4003 err = -EBUSY;
4004 spin_unlock(&mddev->lock);
4005 } else
4006 err = -EINVAL;
4007 break;
4008 case active:
4009 if (mddev->pers) {
4010 err = restart_array(mddev);
4011 if (err)
4012 break;
4013 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
4014 wake_up(&mddev->sb_wait);
4015 err = 0;
4016 } else {
4017 mddev->ro = 0;
4018 set_disk_ro(mddev->gendisk, 0);
4019 err = do_md_run(mddev);
4020 }
4021 break;
4022 case write_pending:
4023 case active_idle:
4024 /* these cannot be set */
4025 break;
4026 }
4027
4028 if (!err) {
4029 if (mddev->hold_active == UNTIL_IOCTL)
4030 mddev->hold_active = 0;
4031 sysfs_notify_dirent_safe(mddev->sysfs_state);
4032 }
4033 mddev_unlock(mddev);
4034 return err ?: len;
4035 }
4036 static struct md_sysfs_entry md_array_state =
4037 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4038
4039 static ssize_t
4040 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4041 return sprintf(page, "%d\n",
4042 atomic_read(&mddev->max_corr_read_errors));
4043 }
4044
4045 static ssize_t
4046 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4047 {
4048 unsigned int n;
4049 int rv;
4050
4051 rv = kstrtouint(buf, 10, &n);
4052 if (rv < 0)
4053 return rv;
4054 atomic_set(&mddev->max_corr_read_errors, n);
4055 return len;
4056 }
4057
4058 static struct md_sysfs_entry max_corr_read_errors =
4059 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4060 max_corrected_read_errors_store);
4061
4062 static ssize_t
4063 null_show(struct mddev *mddev, char *page)
4064 {
4065 return -EINVAL;
4066 }
4067
4068 static ssize_t
4069 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4070 {
4071 /* buf must be %d:%d\n? giving major and minor numbers */
4072 /* The new device is added to the array.
4073 * If the array has a persistent superblock, we read the
4074 * superblock to initialise info and check validity.
4075 * Otherwise, only checking done is that in bind_rdev_to_array,
4076 * which mainly checks size.
4077 */
4078 char *e;
4079 int major = simple_strtoul(buf, &e, 10);
4080 int minor;
4081 dev_t dev;
4082 struct md_rdev *rdev;
4083 int err;
4084
4085 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4086 return -EINVAL;
4087 minor = simple_strtoul(e+1, &e, 10);
4088 if (*e && *e != '\n')
4089 return -EINVAL;
4090 dev = MKDEV(major, minor);
4091 if (major != MAJOR(dev) ||
4092 minor != MINOR(dev))
4093 return -EOVERFLOW;
4094
4095 flush_workqueue(md_misc_wq);
4096
4097 err = mddev_lock(mddev);
4098 if (err)
4099 return err;
4100 if (mddev->persistent) {
4101 rdev = md_import_device(dev, mddev->major_version,
4102 mddev->minor_version);
4103 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4104 struct md_rdev *rdev0
4105 = list_entry(mddev->disks.next,
4106 struct md_rdev, same_set);
4107 err = super_types[mddev->major_version]
4108 .load_super(rdev, rdev0, mddev->minor_version);
4109 if (err < 0)
4110 goto out;
4111 }
4112 } else if (mddev->external)
4113 rdev = md_import_device(dev, -2, -1);
4114 else
4115 rdev = md_import_device(dev, -1, -1);
4116
4117 if (IS_ERR(rdev)) {
4118 mddev_unlock(mddev);
4119 return PTR_ERR(rdev);
4120 }
4121 err = bind_rdev_to_array(rdev, mddev);
4122 out:
4123 if (err)
4124 export_rdev(rdev);
4125 mddev_unlock(mddev);
4126 return err ? err : len;
4127 }
4128
4129 static struct md_sysfs_entry md_new_device =
4130 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4131
4132 static ssize_t
4133 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4134 {
4135 char *end;
4136 unsigned long chunk, end_chunk;
4137 int err;
4138
4139 err = mddev_lock(mddev);
4140 if (err)
4141 return err;
4142 if (!mddev->bitmap)
4143 goto out;
4144 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4145 while (*buf) {
4146 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4147 if (buf == end) break;
4148 if (*end == '-') { /* range */
4149 buf = end + 1;
4150 end_chunk = simple_strtoul(buf, &end, 0);
4151 if (buf == end) break;
4152 }
4153 if (*end && !isspace(*end)) break;
4154 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4155 buf = skip_spaces(end);
4156 }
4157 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4158 out:
4159 mddev_unlock(mddev);
4160 return len;
4161 }
4162
4163 static struct md_sysfs_entry md_bitmap =
4164 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4165
4166 static ssize_t
4167 size_show(struct mddev *mddev, char *page)
4168 {
4169 return sprintf(page, "%llu\n",
4170 (unsigned long long)mddev->dev_sectors / 2);
4171 }
4172
4173 static int update_size(struct mddev *mddev, sector_t num_sectors);
4174
4175 static ssize_t
4176 size_store(struct mddev *mddev, const char *buf, size_t len)
4177 {
4178 /* If array is inactive, we can reduce the component size, but
4179 * not increase it (except from 0).
4180 * If array is active, we can try an on-line resize
4181 */
4182 sector_t sectors;
4183 int err = strict_blocks_to_sectors(buf, &sectors);
4184
4185 if (err < 0)
4186 return err;
4187 err = mddev_lock(mddev);
4188 if (err)
4189 return err;
4190 if (mddev->pers) {
4191 err = update_size(mddev, sectors);
4192 md_update_sb(mddev, 1);
4193 } else {
4194 if (mddev->dev_sectors == 0 ||
4195 mddev->dev_sectors > sectors)
4196 mddev->dev_sectors = sectors;
4197 else
4198 err = -ENOSPC;
4199 }
4200 mddev_unlock(mddev);
4201 return err ? err : len;
4202 }
4203
4204 static struct md_sysfs_entry md_size =
4205 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4206
4207 /* Metadata version.
4208 * This is one of
4209 * 'none' for arrays with no metadata (good luck...)
4210 * 'external' for arrays with externally managed metadata,
4211 * or N.M for internally known formats
4212 */
4213 static ssize_t
4214 metadata_show(struct mddev *mddev, char *page)
4215 {
4216 if (mddev->persistent)
4217 return sprintf(page, "%d.%d\n",
4218 mddev->major_version, mddev->minor_version);
4219 else if (mddev->external)
4220 return sprintf(page, "external:%s\n", mddev->metadata_type);
4221 else
4222 return sprintf(page, "none\n");
4223 }
4224
4225 static ssize_t
4226 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4227 {
4228 int major, minor;
4229 char *e;
4230 int err;
4231 /* Changing the details of 'external' metadata is
4232 * always permitted. Otherwise there must be
4233 * no devices attached to the array.
4234 */
4235
4236 err = mddev_lock(mddev);
4237 if (err)
4238 return err;
4239 err = -EBUSY;
4240 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4241 ;
4242 else if (!list_empty(&mddev->disks))
4243 goto out_unlock;
4244
4245 err = 0;
4246 if (cmd_match(buf, "none")) {
4247 mddev->persistent = 0;
4248 mddev->external = 0;
4249 mddev->major_version = 0;
4250 mddev->minor_version = 90;
4251 goto out_unlock;
4252 }
4253 if (strncmp(buf, "external:", 9) == 0) {
4254 size_t namelen = len-9;
4255 if (namelen >= sizeof(mddev->metadata_type))
4256 namelen = sizeof(mddev->metadata_type)-1;
4257 strncpy(mddev->metadata_type, buf+9, namelen);
4258 mddev->metadata_type[namelen] = 0;
4259 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4260 mddev->metadata_type[--namelen] = 0;
4261 mddev->persistent = 0;
4262 mddev->external = 1;
4263 mddev->major_version = 0;
4264 mddev->minor_version = 90;
4265 goto out_unlock;
4266 }
4267 major = simple_strtoul(buf, &e, 10);
4268 err = -EINVAL;
4269 if (e==buf || *e != '.')
4270 goto out_unlock;
4271 buf = e+1;
4272 minor = simple_strtoul(buf, &e, 10);
4273 if (e==buf || (*e && *e != '\n') )
4274 goto out_unlock;
4275 err = -ENOENT;
4276 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4277 goto out_unlock;
4278 mddev->major_version = major;
4279 mddev->minor_version = minor;
4280 mddev->persistent = 1;
4281 mddev->external = 0;
4282 err = 0;
4283 out_unlock:
4284 mddev_unlock(mddev);
4285 return err ?: len;
4286 }
4287
4288 static struct md_sysfs_entry md_metadata =
4289 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4290
4291 static ssize_t
4292 action_show(struct mddev *mddev, char *page)
4293 {
4294 char *type = "idle";
4295 unsigned long recovery = mddev->recovery;
4296 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4297 type = "frozen";
4298 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4299 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4300 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4301 type = "reshape";
4302 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4303 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4304 type = "resync";
4305 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4306 type = "check";
4307 else
4308 type = "repair";
4309 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4310 type = "recover";
4311 else if (mddev->reshape_position != MaxSector)
4312 type = "reshape";
4313 }
4314 return sprintf(page, "%s\n", type);
4315 }
4316
4317 static ssize_t
4318 action_store(struct mddev *mddev, const char *page, size_t len)
4319 {
4320 if (!mddev->pers || !mddev->pers->sync_request)
4321 return -EINVAL;
4322
4323
4324 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4325 if (cmd_match(page, "frozen"))
4326 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4327 else
4328 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4329 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4330 mddev_lock(mddev) == 0) {
4331 flush_workqueue(md_misc_wq);
4332 if (mddev->sync_thread) {
4333 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4334 md_reap_sync_thread(mddev);
4335 }
4336 mddev_unlock(mddev);
4337 }
4338 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4339 return -EBUSY;
4340 else if (cmd_match(page, "resync"))
4341 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4342 else if (cmd_match(page, "recover")) {
4343 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4344 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4345 } else if (cmd_match(page, "reshape")) {
4346 int err;
4347 if (mddev->pers->start_reshape == NULL)
4348 return -EINVAL;
4349 err = mddev_lock(mddev);
4350 if (!err) {
4351 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4352 err = -EBUSY;
4353 else {
4354 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4355 err = mddev->pers->start_reshape(mddev);
4356 }
4357 mddev_unlock(mddev);
4358 }
4359 if (err)
4360 return err;
4361 sysfs_notify(&mddev->kobj, NULL, "degraded");
4362 } else {
4363 if (cmd_match(page, "check"))
4364 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4365 else if (!cmd_match(page, "repair"))
4366 return -EINVAL;
4367 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4368 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4369 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4370 }
4371 if (mddev->ro == 2) {
4372 /* A write to sync_action is enough to justify
4373 * canceling read-auto mode
4374 */
4375 mddev->ro = 0;
4376 md_wakeup_thread(mddev->sync_thread);
4377 }
4378 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4379 md_wakeup_thread(mddev->thread);
4380 sysfs_notify_dirent_safe(mddev->sysfs_action);
4381 return len;
4382 }
4383
4384 static struct md_sysfs_entry md_scan_mode =
4385 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4386
4387 static ssize_t
4388 last_sync_action_show(struct mddev *mddev, char *page)
4389 {
4390 return sprintf(page, "%s\n", mddev->last_sync_action);
4391 }
4392
4393 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4394
4395 static ssize_t
4396 mismatch_cnt_show(struct mddev *mddev, char *page)
4397 {
4398 return sprintf(page, "%llu\n",
4399 (unsigned long long)
4400 atomic64_read(&mddev->resync_mismatches));
4401 }
4402
4403 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4404
4405 static ssize_t
4406 sync_min_show(struct mddev *mddev, char *page)
4407 {
4408 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4409 mddev->sync_speed_min ? "local": "system");
4410 }
4411
4412 static ssize_t
4413 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4414 {
4415 unsigned int min;
4416 int rv;
4417
4418 if (strncmp(buf, "system", 6)==0) {
4419 min = 0;
4420 } else {
4421 rv = kstrtouint(buf, 10, &min);
4422 if (rv < 0)
4423 return rv;
4424 if (min == 0)
4425 return -EINVAL;
4426 }
4427 mddev->sync_speed_min = min;
4428 return len;
4429 }
4430
4431 static struct md_sysfs_entry md_sync_min =
4432 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4433
4434 static ssize_t
4435 sync_max_show(struct mddev *mddev, char *page)
4436 {
4437 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4438 mddev->sync_speed_max ? "local": "system");
4439 }
4440
4441 static ssize_t
4442 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4443 {
4444 unsigned int max;
4445 int rv;
4446
4447 if (strncmp(buf, "system", 6)==0) {
4448 max = 0;
4449 } else {
4450 rv = kstrtouint(buf, 10, &max);
4451 if (rv < 0)
4452 return rv;
4453 if (max == 0)
4454 return -EINVAL;
4455 }
4456 mddev->sync_speed_max = max;
4457 return len;
4458 }
4459
4460 static struct md_sysfs_entry md_sync_max =
4461 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4462
4463 static ssize_t
4464 degraded_show(struct mddev *mddev, char *page)
4465 {
4466 return sprintf(page, "%d\n", mddev->degraded);
4467 }
4468 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4469
4470 static ssize_t
4471 sync_force_parallel_show(struct mddev *mddev, char *page)
4472 {
4473 return sprintf(page, "%d\n", mddev->parallel_resync);
4474 }
4475
4476 static ssize_t
4477 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4478 {
4479 long n;
4480
4481 if (kstrtol(buf, 10, &n))
4482 return -EINVAL;
4483
4484 if (n != 0 && n != 1)
4485 return -EINVAL;
4486
4487 mddev->parallel_resync = n;
4488
4489 if (mddev->sync_thread)
4490 wake_up(&resync_wait);
4491
4492 return len;
4493 }
4494
4495 /* force parallel resync, even with shared block devices */
4496 static struct md_sysfs_entry md_sync_force_parallel =
4497 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4498 sync_force_parallel_show, sync_force_parallel_store);
4499
4500 static ssize_t
4501 sync_speed_show(struct mddev *mddev, char *page)
4502 {
4503 unsigned long resync, dt, db;
4504 if (mddev->curr_resync == 0)
4505 return sprintf(page, "none\n");
4506 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4507 dt = (jiffies - mddev->resync_mark) / HZ;
4508 if (!dt) dt++;
4509 db = resync - mddev->resync_mark_cnt;
4510 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4511 }
4512
4513 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4514
4515 static ssize_t
4516 sync_completed_show(struct mddev *mddev, char *page)
4517 {
4518 unsigned long long max_sectors, resync;
4519
4520 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4521 return sprintf(page, "none\n");
4522
4523 if (mddev->curr_resync == 1 ||
4524 mddev->curr_resync == 2)
4525 return sprintf(page, "delayed\n");
4526
4527 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4528 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4529 max_sectors = mddev->resync_max_sectors;
4530 else
4531 max_sectors = mddev->dev_sectors;
4532
4533 resync = mddev->curr_resync_completed;
4534 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4535 }
4536
4537 static struct md_sysfs_entry md_sync_completed =
4538 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
4539
4540 static ssize_t
4541 min_sync_show(struct mddev *mddev, char *page)
4542 {
4543 return sprintf(page, "%llu\n",
4544 (unsigned long long)mddev->resync_min);
4545 }
4546 static ssize_t
4547 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4548 {
4549 unsigned long long min;
4550 int err;
4551
4552 if (kstrtoull(buf, 10, &min))
4553 return -EINVAL;
4554
4555 spin_lock(&mddev->lock);
4556 err = -EINVAL;
4557 if (min > mddev->resync_max)
4558 goto out_unlock;
4559
4560 err = -EBUSY;
4561 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4562 goto out_unlock;
4563
4564 /* Round down to multiple of 4K for safety */
4565 mddev->resync_min = round_down(min, 8);
4566 err = 0;
4567
4568 out_unlock:
4569 spin_unlock(&mddev->lock);
4570 return err ?: len;
4571 }
4572
4573 static struct md_sysfs_entry md_min_sync =
4574 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4575
4576 static ssize_t
4577 max_sync_show(struct mddev *mddev, char *page)
4578 {
4579 if (mddev->resync_max == MaxSector)
4580 return sprintf(page, "max\n");
4581 else
4582 return sprintf(page, "%llu\n",
4583 (unsigned long long)mddev->resync_max);
4584 }
4585 static ssize_t
4586 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4587 {
4588 int err;
4589 spin_lock(&mddev->lock);
4590 if (strncmp(buf, "max", 3) == 0)
4591 mddev->resync_max = MaxSector;
4592 else {
4593 unsigned long long max;
4594 int chunk;
4595
4596 err = -EINVAL;
4597 if (kstrtoull(buf, 10, &max))
4598 goto out_unlock;
4599 if (max < mddev->resync_min)
4600 goto out_unlock;
4601
4602 err = -EBUSY;
4603 if (max < mddev->resync_max &&
4604 mddev->ro == 0 &&
4605 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4606 goto out_unlock;
4607
4608 /* Must be a multiple of chunk_size */
4609 chunk = mddev->chunk_sectors;
4610 if (chunk) {
4611 sector_t temp = max;
4612
4613 err = -EINVAL;
4614 if (sector_div(temp, chunk))
4615 goto out_unlock;
4616 }
4617 mddev->resync_max = max;
4618 }
4619 wake_up(&mddev->recovery_wait);
4620 err = 0;
4621 out_unlock:
4622 spin_unlock(&mddev->lock);
4623 return err ?: len;
4624 }
4625
4626 static struct md_sysfs_entry md_max_sync =
4627 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4628
4629 static ssize_t
4630 suspend_lo_show(struct mddev *mddev, char *page)
4631 {
4632 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4633 }
4634
4635 static ssize_t
4636 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4637 {
4638 unsigned long long old, new;
4639 int err;
4640
4641 err = kstrtoull(buf, 10, &new);
4642 if (err < 0)
4643 return err;
4644 if (new != (sector_t)new)
4645 return -EINVAL;
4646
4647 err = mddev_lock(mddev);
4648 if (err)
4649 return err;
4650 err = -EINVAL;
4651 if (mddev->pers == NULL ||
4652 mddev->pers->quiesce == NULL)
4653 goto unlock;
4654 old = mddev->suspend_lo;
4655 mddev->suspend_lo = new;
4656 if (new >= old)
4657 /* Shrinking suspended region */
4658 mddev->pers->quiesce(mddev, 2);
4659 else {
4660 /* Expanding suspended region - need to wait */
4661 mddev->pers->quiesce(mddev, 1);
4662 mddev->pers->quiesce(mddev, 0);
4663 }
4664 err = 0;
4665 unlock:
4666 mddev_unlock(mddev);
4667 return err ?: len;
4668 }
4669 static struct md_sysfs_entry md_suspend_lo =
4670 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4671
4672 static ssize_t
4673 suspend_hi_show(struct mddev *mddev, char *page)
4674 {
4675 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4676 }
4677
4678 static ssize_t
4679 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4680 {
4681 unsigned long long old, new;
4682 int err;
4683
4684 err = kstrtoull(buf, 10, &new);
4685 if (err < 0)
4686 return err;
4687 if (new != (sector_t)new)
4688 return -EINVAL;
4689
4690 err = mddev_lock(mddev);
4691 if (err)
4692 return err;
4693 err = -EINVAL;
4694 if (mddev->pers == NULL ||
4695 mddev->pers->quiesce == NULL)
4696 goto unlock;
4697 old = mddev->suspend_hi;
4698 mddev->suspend_hi = new;
4699 if (new <= old)
4700 /* Shrinking suspended region */
4701 mddev->pers->quiesce(mddev, 2);
4702 else {
4703 /* Expanding suspended region - need to wait */
4704 mddev->pers->quiesce(mddev, 1);
4705 mddev->pers->quiesce(mddev, 0);
4706 }
4707 err = 0;
4708 unlock:
4709 mddev_unlock(mddev);
4710 return err ?: len;
4711 }
4712 static struct md_sysfs_entry md_suspend_hi =
4713 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4714
4715 static ssize_t
4716 reshape_position_show(struct mddev *mddev, char *page)
4717 {
4718 if (mddev->reshape_position != MaxSector)
4719 return sprintf(page, "%llu\n",
4720 (unsigned long long)mddev->reshape_position);
4721 strcpy(page, "none\n");
4722 return 5;
4723 }
4724
4725 static ssize_t
4726 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4727 {
4728 struct md_rdev *rdev;
4729 unsigned long long new;
4730 int err;
4731
4732 err = kstrtoull(buf, 10, &new);
4733 if (err < 0)
4734 return err;
4735 if (new != (sector_t)new)
4736 return -EINVAL;
4737 err = mddev_lock(mddev);
4738 if (err)
4739 return err;
4740 err = -EBUSY;
4741 if (mddev->pers)
4742 goto unlock;
4743 mddev->reshape_position = new;
4744 mddev->delta_disks = 0;
4745 mddev->reshape_backwards = 0;
4746 mddev->new_level = mddev->level;
4747 mddev->new_layout = mddev->layout;
4748 mddev->new_chunk_sectors = mddev->chunk_sectors;
4749 rdev_for_each(rdev, mddev)
4750 rdev->new_data_offset = rdev->data_offset;
4751 err = 0;
4752 unlock:
4753 mddev_unlock(mddev);
4754 return err ?: len;
4755 }
4756
4757 static struct md_sysfs_entry md_reshape_position =
4758 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4759 reshape_position_store);
4760
4761 static ssize_t
4762 reshape_direction_show(struct mddev *mddev, char *page)
4763 {
4764 return sprintf(page, "%s\n",
4765 mddev->reshape_backwards ? "backwards" : "forwards");
4766 }
4767
4768 static ssize_t
4769 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4770 {
4771 int backwards = 0;
4772 int err;
4773
4774 if (cmd_match(buf, "forwards"))
4775 backwards = 0;
4776 else if (cmd_match(buf, "backwards"))
4777 backwards = 1;
4778 else
4779 return -EINVAL;
4780 if (mddev->reshape_backwards == backwards)
4781 return len;
4782
4783 err = mddev_lock(mddev);
4784 if (err)
4785 return err;
4786 /* check if we are allowed to change */
4787 if (mddev->delta_disks)
4788 err = -EBUSY;
4789 else if (mddev->persistent &&
4790 mddev->major_version == 0)
4791 err = -EINVAL;
4792 else
4793 mddev->reshape_backwards = backwards;
4794 mddev_unlock(mddev);
4795 return err ?: len;
4796 }
4797
4798 static struct md_sysfs_entry md_reshape_direction =
4799 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4800 reshape_direction_store);
4801
4802 static ssize_t
4803 array_size_show(struct mddev *mddev, char *page)
4804 {
4805 if (mddev->external_size)
4806 return sprintf(page, "%llu\n",
4807 (unsigned long long)mddev->array_sectors/2);
4808 else
4809 return sprintf(page, "default\n");
4810 }
4811
4812 static ssize_t
4813 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4814 {
4815 sector_t sectors;
4816 int err;
4817
4818 err = mddev_lock(mddev);
4819 if (err)
4820 return err;
4821
4822 if (strncmp(buf, "default", 7) == 0) {
4823 if (mddev->pers)
4824 sectors = mddev->pers->size(mddev, 0, 0);
4825 else
4826 sectors = mddev->array_sectors;
4827
4828 mddev->external_size = 0;
4829 } else {
4830 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4831 err = -EINVAL;
4832 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4833 err = -E2BIG;
4834 else
4835 mddev->external_size = 1;
4836 }
4837
4838 if (!err) {
4839 mddev->array_sectors = sectors;
4840 if (mddev->pers) {
4841 set_capacity(mddev->gendisk, mddev->array_sectors);
4842 revalidate_disk(mddev->gendisk);
4843 }
4844 }
4845 mddev_unlock(mddev);
4846 return err ?: len;
4847 }
4848
4849 static struct md_sysfs_entry md_array_size =
4850 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4851 array_size_store);
4852
4853 static struct attribute *md_default_attrs[] = {
4854 &md_level.attr,
4855 &md_layout.attr,
4856 &md_raid_disks.attr,
4857 &md_chunk_size.attr,
4858 &md_size.attr,
4859 &md_resync_start.attr,
4860 &md_metadata.attr,
4861 &md_new_device.attr,
4862 &md_safe_delay.attr,
4863 &md_array_state.attr,
4864 &md_reshape_position.attr,
4865 &md_reshape_direction.attr,
4866 &md_array_size.attr,
4867 &max_corr_read_errors.attr,
4868 NULL,
4869 };
4870
4871 static struct attribute *md_redundancy_attrs[] = {
4872 &md_scan_mode.attr,
4873 &md_last_scan_mode.attr,
4874 &md_mismatches.attr,
4875 &md_sync_min.attr,
4876 &md_sync_max.attr,
4877 &md_sync_speed.attr,
4878 &md_sync_force_parallel.attr,
4879 &md_sync_completed.attr,
4880 &md_min_sync.attr,
4881 &md_max_sync.attr,
4882 &md_suspend_lo.attr,
4883 &md_suspend_hi.attr,
4884 &md_bitmap.attr,
4885 &md_degraded.attr,
4886 NULL,
4887 };
4888 static struct attribute_group md_redundancy_group = {
4889 .name = NULL,
4890 .attrs = md_redundancy_attrs,
4891 };
4892
4893 static ssize_t
4894 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4895 {
4896 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4897 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4898 ssize_t rv;
4899
4900 if (!entry->show)
4901 return -EIO;
4902 spin_lock(&all_mddevs_lock);
4903 if (list_empty(&mddev->all_mddevs)) {
4904 spin_unlock(&all_mddevs_lock);
4905 return -EBUSY;
4906 }
4907 mddev_get(mddev);
4908 spin_unlock(&all_mddevs_lock);
4909
4910 rv = entry->show(mddev, page);
4911 mddev_put(mddev);
4912 return rv;
4913 }
4914
4915 static ssize_t
4916 md_attr_store(struct kobject *kobj, struct attribute *attr,
4917 const char *page, size_t length)
4918 {
4919 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4920 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4921 ssize_t rv;
4922
4923 if (!entry->store)
4924 return -EIO;
4925 if (!capable(CAP_SYS_ADMIN))
4926 return -EACCES;
4927 spin_lock(&all_mddevs_lock);
4928 if (list_empty(&mddev->all_mddevs)) {
4929 spin_unlock(&all_mddevs_lock);
4930 return -EBUSY;
4931 }
4932 mddev_get(mddev);
4933 spin_unlock(&all_mddevs_lock);
4934 rv = entry->store(mddev, page, length);
4935 mddev_put(mddev);
4936 return rv;
4937 }
4938
4939 static void md_free(struct kobject *ko)
4940 {
4941 struct mddev *mddev = container_of(ko, struct mddev, kobj);
4942
4943 if (mddev->sysfs_state)
4944 sysfs_put(mddev->sysfs_state);
4945
4946 if (mddev->queue)
4947 blk_cleanup_queue(mddev->queue);
4948 if (mddev->gendisk) {
4949 del_gendisk(mddev->gendisk);
4950 put_disk(mddev->gendisk);
4951 }
4952
4953 kfree(mddev);
4954 }
4955
4956 static const struct sysfs_ops md_sysfs_ops = {
4957 .show = md_attr_show,
4958 .store = md_attr_store,
4959 };
4960 static struct kobj_type md_ktype = {
4961 .release = md_free,
4962 .sysfs_ops = &md_sysfs_ops,
4963 .default_attrs = md_default_attrs,
4964 };
4965
4966 int mdp_major = 0;
4967
4968 static void mddev_delayed_delete(struct work_struct *ws)
4969 {
4970 struct mddev *mddev = container_of(ws, struct mddev, del_work);
4971
4972 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4973 kobject_del(&mddev->kobj);
4974 kobject_put(&mddev->kobj);
4975 }
4976
4977 static int md_alloc(dev_t dev, char *name)
4978 {
4979 static DEFINE_MUTEX(disks_mutex);
4980 struct mddev *mddev = mddev_find(dev);
4981 struct gendisk *disk;
4982 int partitioned;
4983 int shift;
4984 int unit;
4985 int error;
4986
4987 if (!mddev)
4988 return -ENODEV;
4989
4990 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4991 shift = partitioned ? MdpMinorShift : 0;
4992 unit = MINOR(mddev->unit) >> shift;
4993
4994 /* wait for any previous instance of this device to be
4995 * completely removed (mddev_delayed_delete).
4996 */
4997 flush_workqueue(md_misc_wq);
4998
4999 mutex_lock(&disks_mutex);
5000 error = -EEXIST;
5001 if (mddev->gendisk)
5002 goto abort;
5003
5004 if (name) {
5005 /* Need to ensure that 'name' is not a duplicate.
5006 */
5007 struct mddev *mddev2;
5008 spin_lock(&all_mddevs_lock);
5009
5010 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5011 if (mddev2->gendisk &&
5012 strcmp(mddev2->gendisk->disk_name, name) == 0) {
5013 spin_unlock(&all_mddevs_lock);
5014 goto abort;
5015 }
5016 spin_unlock(&all_mddevs_lock);
5017 }
5018
5019 error = -ENOMEM;
5020 mddev->queue = blk_alloc_queue(GFP_KERNEL);
5021 if (!mddev->queue)
5022 goto abort;
5023 mddev->queue->queuedata = mddev;
5024
5025 blk_queue_make_request(mddev->queue, md_make_request);
5026 blk_set_stacking_limits(&mddev->queue->limits);
5027
5028 disk = alloc_disk(1 << shift);
5029 if (!disk) {
5030 blk_cleanup_queue(mddev->queue);
5031 mddev->queue = NULL;
5032 goto abort;
5033 }
5034 disk->major = MAJOR(mddev->unit);
5035 disk->first_minor = unit << shift;
5036 if (name)
5037 strcpy(disk->disk_name, name);
5038 else if (partitioned)
5039 sprintf(disk->disk_name, "md_d%d", unit);
5040 else
5041 sprintf(disk->disk_name, "md%d", unit);
5042 disk->fops = &md_fops;
5043 disk->private_data = mddev;
5044 disk->queue = mddev->queue;
5045 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
5046 /* Allow extended partitions. This makes the
5047 * 'mdp' device redundant, but we can't really
5048 * remove it now.
5049 */
5050 disk->flags |= GENHD_FL_EXT_DEVT;
5051 mddev->gendisk = disk;
5052 /* As soon as we call add_disk(), another thread could get
5053 * through to md_open, so make sure it doesn't get too far
5054 */
5055 mutex_lock(&mddev->open_mutex);
5056 add_disk(disk);
5057
5058 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
5059 &disk_to_dev(disk)->kobj, "%s", "md");
5060 if (error) {
5061 /* This isn't possible, but as kobject_init_and_add is marked
5062 * __must_check, we must do something with the result
5063 */
5064 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
5065 disk->disk_name);
5066 error = 0;
5067 }
5068 if (mddev->kobj.sd &&
5069 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
5070 printk(KERN_DEBUG "pointless warning\n");
5071 mutex_unlock(&mddev->open_mutex);
5072 abort:
5073 mutex_unlock(&disks_mutex);
5074 if (!error && mddev->kobj.sd) {
5075 kobject_uevent(&mddev->kobj, KOBJ_ADD);
5076 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5077 }
5078 mddev_put(mddev);
5079 return error;
5080 }
5081
5082 static struct kobject *md_probe(dev_t dev, int *part, void *data)
5083 {
5084 md_alloc(dev, NULL);
5085 return NULL;
5086 }
5087
5088 static int add_named_array(const char *val, struct kernel_param *kp)
5089 {
5090 /* val must be "md_*" where * is not all digits.
5091 * We allocate an array with a large free minor number, and
5092 * set the name to val. val must not already be an active name.
5093 */
5094 int len = strlen(val);
5095 char buf[DISK_NAME_LEN];
5096
5097 while (len && val[len-1] == '\n')
5098 len--;
5099 if (len >= DISK_NAME_LEN)
5100 return -E2BIG;
5101 strlcpy(buf, val, len+1);
5102 if (strncmp(buf, "md_", 3) != 0)
5103 return -EINVAL;
5104 return md_alloc(0, buf);
5105 }
5106
5107 static void md_safemode_timeout(unsigned long data)
5108 {
5109 struct mddev *mddev = (struct mddev *) data;
5110
5111 if (!atomic_read(&mddev->writes_pending)) {
5112 mddev->safemode = 1;
5113 if (mddev->external)
5114 sysfs_notify_dirent_safe(mddev->sysfs_state);
5115 }
5116 md_wakeup_thread(mddev->thread);
5117 }
5118
5119 static int start_dirty_degraded;
5120
5121 int md_run(struct mddev *mddev)
5122 {
5123 int err;
5124 struct md_rdev *rdev;
5125 struct md_personality *pers;
5126
5127 if (list_empty(&mddev->disks))
5128 /* cannot run an array with no devices.. */
5129 return -EINVAL;
5130
5131 if (mddev->pers)
5132 return -EBUSY;
5133 /* Cannot run until previous stop completes properly */
5134 if (mddev->sysfs_active)
5135 return -EBUSY;
5136
5137 /*
5138 * Analyze all RAID superblock(s)
5139 */
5140 if (!mddev->raid_disks) {
5141 if (!mddev->persistent)
5142 return -EINVAL;
5143 analyze_sbs(mddev);
5144 }
5145
5146 if (mddev->level != LEVEL_NONE)
5147 request_module("md-level-%d", mddev->level);
5148 else if (mddev->clevel[0])
5149 request_module("md-%s", mddev->clevel);
5150
5151 /*
5152 * Drop all container device buffers, from now on
5153 * the only valid external interface is through the md
5154 * device.
5155 */
5156 rdev_for_each(rdev, mddev) {
5157 if (test_bit(Faulty, &rdev->flags))
5158 continue;
5159 sync_blockdev(rdev->bdev);
5160 invalidate_bdev(rdev->bdev);
5161
5162 /* perform some consistency tests on the device.
5163 * We don't want the data to overlap the metadata,
5164 * Internal Bitmap issues have been handled elsewhere.
5165 */
5166 if (rdev->meta_bdev) {
5167 /* Nothing to check */;
5168 } else if (rdev->data_offset < rdev->sb_start) {
5169 if (mddev->dev_sectors &&
5170 rdev->data_offset + mddev->dev_sectors
5171 > rdev->sb_start) {
5172 printk("md: %s: data overlaps metadata\n",
5173 mdname(mddev));
5174 return -EINVAL;
5175 }
5176 } else {
5177 if (rdev->sb_start + rdev->sb_size/512
5178 > rdev->data_offset) {
5179 printk("md: %s: metadata overlaps data\n",
5180 mdname(mddev));
5181 return -EINVAL;
5182 }
5183 }
5184 sysfs_notify_dirent_safe(rdev->sysfs_state);
5185 }
5186
5187 if (mddev->bio_set == NULL)
5188 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
5189
5190 spin_lock(&pers_lock);
5191 pers = find_pers(mddev->level, mddev->clevel);
5192 if (!pers || !try_module_get(pers->owner)) {
5193 spin_unlock(&pers_lock);
5194 if (mddev->level != LEVEL_NONE)
5195 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5196 mddev->level);
5197 else
5198 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5199 mddev->clevel);
5200 return -EINVAL;
5201 }
5202 spin_unlock(&pers_lock);
5203 if (mddev->level != pers->level) {
5204 mddev->level = pers->level;
5205 mddev->new_level = pers->level;
5206 }
5207 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5208
5209 if (mddev->reshape_position != MaxSector &&
5210 pers->start_reshape == NULL) {
5211 /* This personality cannot handle reshaping... */
5212 module_put(pers->owner);
5213 return -EINVAL;
5214 }
5215
5216 if (pers->sync_request) {
5217 /* Warn if this is a potentially silly
5218 * configuration.
5219 */
5220 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5221 struct md_rdev *rdev2;
5222 int warned = 0;
5223
5224 rdev_for_each(rdev, mddev)
5225 rdev_for_each(rdev2, mddev) {
5226 if (rdev < rdev2 &&
5227 rdev->bdev->bd_contains ==
5228 rdev2->bdev->bd_contains) {
5229 printk(KERN_WARNING
5230 "%s: WARNING: %s appears to be"
5231 " on the same physical disk as"
5232 " %s.\n",
5233 mdname(mddev),
5234 bdevname(rdev->bdev,b),
5235 bdevname(rdev2->bdev,b2));
5236 warned = 1;
5237 }
5238 }
5239
5240 if (warned)
5241 printk(KERN_WARNING
5242 "True protection against single-disk"
5243 " failure might be compromised.\n");
5244 }
5245
5246 mddev->recovery = 0;
5247 /* may be over-ridden by personality */
5248 mddev->resync_max_sectors = mddev->dev_sectors;
5249
5250 mddev->ok_start_degraded = start_dirty_degraded;
5251
5252 if (start_readonly && mddev->ro == 0)
5253 mddev->ro = 2; /* read-only, but switch on first write */
5254
5255 err = pers->run(mddev);
5256 if (err)
5257 printk(KERN_ERR "md: pers->run() failed ...\n");
5258 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5259 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5260 " but 'external_size' not in effect?\n", __func__);
5261 printk(KERN_ERR
5262 "md: invalid array_size %llu > default size %llu\n",
5263 (unsigned long long)mddev->array_sectors / 2,
5264 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5265 err = -EINVAL;
5266 }
5267 if (err == 0 && pers->sync_request &&
5268 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5269 struct bitmap *bitmap;
5270
5271 bitmap = bitmap_create(mddev, -1);
5272 if (IS_ERR(bitmap)) {
5273 err = PTR_ERR(bitmap);
5274 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5275 mdname(mddev), err);
5276 } else
5277 mddev->bitmap = bitmap;
5278
5279 }
5280 if (err) {
5281 mddev_detach(mddev);
5282 if (mddev->private)
5283 pers->free(mddev, mddev->private);
5284 mddev->private = NULL;
5285 module_put(pers->owner);
5286 bitmap_destroy(mddev);
5287 return err;
5288 }
5289 if (mddev->queue) {
5290 mddev->queue->backing_dev_info.congested_data = mddev;
5291 mddev->queue->backing_dev_info.congested_fn = md_congested;
5292 }
5293 if (pers->sync_request) {
5294 if (mddev->kobj.sd &&
5295 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5296 printk(KERN_WARNING
5297 "md: cannot register extra attributes for %s\n",
5298 mdname(mddev));
5299 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5300 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5301 mddev->ro = 0;
5302
5303 atomic_set(&mddev->writes_pending,0);
5304 atomic_set(&mddev->max_corr_read_errors,
5305 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5306 mddev->safemode = 0;
5307 if (mddev_is_clustered(mddev))
5308 mddev->safemode_delay = 0;
5309 else
5310 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5311 mddev->in_sync = 1;
5312 smp_wmb();
5313 spin_lock(&mddev->lock);
5314 mddev->pers = pers;
5315 mddev->ready = 1;
5316 spin_unlock(&mddev->lock);
5317 rdev_for_each(rdev, mddev)
5318 if (rdev->raid_disk >= 0)
5319 if (sysfs_link_rdev(mddev, rdev))
5320 /* failure here is OK */;
5321
5322 if (mddev->degraded && !mddev->ro)
5323 /* This ensures that recovering status is reported immediately
5324 * via sysfs - until a lack of spares is confirmed.
5325 */
5326 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5327 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5328
5329 if (mddev->flags & MD_UPDATE_SB_FLAGS)
5330 md_update_sb(mddev, 0);
5331
5332 md_new_event(mddev);
5333 sysfs_notify_dirent_safe(mddev->sysfs_state);
5334 sysfs_notify_dirent_safe(mddev->sysfs_action);
5335 sysfs_notify(&mddev->kobj, NULL, "degraded");
5336 return 0;
5337 }
5338 EXPORT_SYMBOL_GPL(md_run);
5339
5340 static int do_md_run(struct mddev *mddev)
5341 {
5342 int err;
5343
5344 err = md_run(mddev);
5345 if (err)
5346 goto out;
5347 err = bitmap_load(mddev);
5348 if (err) {
5349 bitmap_destroy(mddev);
5350 goto out;
5351 }
5352
5353 if (mddev_is_clustered(mddev))
5354 md_allow_write(mddev);
5355
5356 md_wakeup_thread(mddev->thread);
5357 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5358
5359 set_capacity(mddev->gendisk, mddev->array_sectors);
5360 revalidate_disk(mddev->gendisk);
5361 mddev->changed = 1;
5362 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5363 out:
5364 return err;
5365 }
5366
5367 static int restart_array(struct mddev *mddev)
5368 {
5369 struct gendisk *disk = mddev->gendisk;
5370
5371 /* Complain if it has no devices */
5372 if (list_empty(&mddev->disks))
5373 return -ENXIO;
5374 if (!mddev->pers)
5375 return -EINVAL;
5376 if (!mddev->ro)
5377 return -EBUSY;
5378 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5379 struct md_rdev *rdev;
5380 bool has_journal = false;
5381
5382 rcu_read_lock();
5383 rdev_for_each_rcu(rdev, mddev) {
5384 if (test_bit(Journal, &rdev->flags) &&
5385 !test_bit(Faulty, &rdev->flags)) {
5386 has_journal = true;
5387 break;
5388 }
5389 }
5390 rcu_read_unlock();
5391
5392 /* Don't restart rw with journal missing/faulty */
5393 if (!has_journal)
5394 return -EINVAL;
5395 }
5396
5397 mddev->safemode = 0;
5398 mddev->ro = 0;
5399 set_disk_ro(disk, 0);
5400 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5401 mdname(mddev));
5402 /* Kick recovery or resync if necessary */
5403 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5404 md_wakeup_thread(mddev->thread);
5405 md_wakeup_thread(mddev->sync_thread);
5406 sysfs_notify_dirent_safe(mddev->sysfs_state);
5407 return 0;
5408 }
5409
5410 static void md_clean(struct mddev *mddev)
5411 {
5412 mddev->array_sectors = 0;
5413 mddev->external_size = 0;
5414 mddev->dev_sectors = 0;
5415 mddev->raid_disks = 0;
5416 mddev->recovery_cp = 0;
5417 mddev->resync_min = 0;
5418 mddev->resync_max = MaxSector;
5419 mddev->reshape_position = MaxSector;
5420 mddev->external = 0;
5421 mddev->persistent = 0;
5422 mddev->level = LEVEL_NONE;
5423 mddev->clevel[0] = 0;
5424 mddev->flags = 0;
5425 mddev->ro = 0;
5426 mddev->metadata_type[0] = 0;
5427 mddev->chunk_sectors = 0;
5428 mddev->ctime = mddev->utime = 0;
5429 mddev->layout = 0;
5430 mddev->max_disks = 0;
5431 mddev->events = 0;
5432 mddev->can_decrease_events = 0;
5433 mddev->delta_disks = 0;
5434 mddev->reshape_backwards = 0;
5435 mddev->new_level = LEVEL_NONE;
5436 mddev->new_layout = 0;
5437 mddev->new_chunk_sectors = 0;
5438 mddev->curr_resync = 0;
5439 atomic64_set(&mddev->resync_mismatches, 0);
5440 mddev->suspend_lo = mddev->suspend_hi = 0;
5441 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5442 mddev->recovery = 0;
5443 mddev->in_sync = 0;
5444 mddev->changed = 0;
5445 mddev->degraded = 0;
5446 mddev->safemode = 0;
5447 mddev->private = NULL;
5448 mddev->bitmap_info.offset = 0;
5449 mddev->bitmap_info.default_offset = 0;
5450 mddev->bitmap_info.default_space = 0;
5451 mddev->bitmap_info.chunksize = 0;
5452 mddev->bitmap_info.daemon_sleep = 0;
5453 mddev->bitmap_info.max_write_behind = 0;
5454 }
5455
5456 static void __md_stop_writes(struct mddev *mddev)
5457 {
5458 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5459 flush_workqueue(md_misc_wq);
5460 if (mddev->sync_thread) {
5461 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5462 md_reap_sync_thread(mddev);
5463 }
5464
5465 del_timer_sync(&mddev->safemode_timer);
5466
5467 bitmap_flush(mddev);
5468 md_super_wait(mddev);
5469
5470 if (mddev->ro == 0 &&
5471 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
5472 (mddev->flags & MD_UPDATE_SB_FLAGS))) {
5473 /* mark array as shutdown cleanly */
5474 if (!mddev_is_clustered(mddev))
5475 mddev->in_sync = 1;
5476 md_update_sb(mddev, 1);
5477 }
5478 }
5479
5480 void md_stop_writes(struct mddev *mddev)
5481 {
5482 mddev_lock_nointr(mddev);
5483 __md_stop_writes(mddev);
5484 mddev_unlock(mddev);
5485 }
5486 EXPORT_SYMBOL_GPL(md_stop_writes);
5487
5488 static void mddev_detach(struct mddev *mddev)
5489 {
5490 struct bitmap *bitmap = mddev->bitmap;
5491 /* wait for behind writes to complete */
5492 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
5493 printk(KERN_INFO "md:%s: behind writes in progress - waiting to stop.\n",
5494 mdname(mddev));
5495 /* need to kick something here to make sure I/O goes? */
5496 wait_event(bitmap->behind_wait,
5497 atomic_read(&bitmap->behind_writes) == 0);
5498 }
5499 if (mddev->pers && mddev->pers->quiesce) {
5500 mddev->pers->quiesce(mddev, 1);
5501 mddev->pers->quiesce(mddev, 0);
5502 }
5503 md_unregister_thread(&mddev->thread);
5504 if (mddev->queue)
5505 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5506 }
5507
5508 static void __md_stop(struct mddev *mddev)
5509 {
5510 struct md_personality *pers = mddev->pers;
5511 mddev_detach(mddev);
5512 /* Ensure ->event_work is done */
5513 flush_workqueue(md_misc_wq);
5514 spin_lock(&mddev->lock);
5515 mddev->ready = 0;
5516 mddev->pers = NULL;
5517 spin_unlock(&mddev->lock);
5518 pers->free(mddev, mddev->private);
5519 mddev->private = NULL;
5520 if (pers->sync_request && mddev->to_remove == NULL)
5521 mddev->to_remove = &md_redundancy_group;
5522 module_put(pers->owner);
5523 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5524 }
5525
5526 void md_stop(struct mddev *mddev)
5527 {
5528 /* stop the array and free an attached data structures.
5529 * This is called from dm-raid
5530 */
5531 __md_stop(mddev);
5532 bitmap_destroy(mddev);
5533 if (mddev->bio_set)
5534 bioset_free(mddev->bio_set);
5535 }
5536
5537 EXPORT_SYMBOL_GPL(md_stop);
5538
5539 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5540 {
5541 int err = 0;
5542 int did_freeze = 0;
5543
5544 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5545 did_freeze = 1;
5546 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5547 md_wakeup_thread(mddev->thread);
5548 }
5549 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5550 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5551 if (mddev->sync_thread)
5552 /* Thread might be blocked waiting for metadata update
5553 * which will now never happen */
5554 wake_up_process(mddev->sync_thread->tsk);
5555
5556 if (mddev->external && test_bit(MD_CHANGE_PENDING, &mddev->flags))
5557 return -EBUSY;
5558 mddev_unlock(mddev);
5559 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5560 &mddev->recovery));
5561 wait_event(mddev->sb_wait,
5562 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5563 mddev_lock_nointr(mddev);
5564
5565 mutex_lock(&mddev->open_mutex);
5566 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5567 mddev->sync_thread ||
5568 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5569 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5570 printk("md: %s still in use.\n",mdname(mddev));
5571 if (did_freeze) {
5572 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5573 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5574 md_wakeup_thread(mddev->thread);
5575 }
5576 err = -EBUSY;
5577 goto out;
5578 }
5579 if (mddev->pers) {
5580 __md_stop_writes(mddev);
5581
5582 err = -ENXIO;
5583 if (mddev->ro==1)
5584 goto out;
5585 mddev->ro = 1;
5586 set_disk_ro(mddev->gendisk, 1);
5587 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5588 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5589 md_wakeup_thread(mddev->thread);
5590 sysfs_notify_dirent_safe(mddev->sysfs_state);
5591 err = 0;
5592 }
5593 out:
5594 mutex_unlock(&mddev->open_mutex);
5595 return err;
5596 }
5597
5598 /* mode:
5599 * 0 - completely stop and dis-assemble array
5600 * 2 - stop but do not disassemble array
5601 */
5602 static int do_md_stop(struct mddev *mddev, int mode,
5603 struct block_device *bdev)
5604 {
5605 struct gendisk *disk = mddev->gendisk;
5606 struct md_rdev *rdev;
5607 int did_freeze = 0;
5608
5609 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5610 did_freeze = 1;
5611 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5612 md_wakeup_thread(mddev->thread);
5613 }
5614 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5615 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5616 if (mddev->sync_thread)
5617 /* Thread might be blocked waiting for metadata update
5618 * which will now never happen */
5619 wake_up_process(mddev->sync_thread->tsk);
5620
5621 mddev_unlock(mddev);
5622 wait_event(resync_wait, (mddev->sync_thread == NULL &&
5623 !test_bit(MD_RECOVERY_RUNNING,
5624 &mddev->recovery)));
5625 mddev_lock_nointr(mddev);
5626
5627 mutex_lock(&mddev->open_mutex);
5628 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5629 mddev->sysfs_active ||
5630 mddev->sync_thread ||
5631 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5632 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5633 printk("md: %s still in use.\n",mdname(mddev));
5634 mutex_unlock(&mddev->open_mutex);
5635 if (did_freeze) {
5636 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5637 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5638 md_wakeup_thread(mddev->thread);
5639 }
5640 return -EBUSY;
5641 }
5642 if (mddev->pers) {
5643 if (mddev->ro)
5644 set_disk_ro(disk, 0);
5645
5646 __md_stop_writes(mddev);
5647 __md_stop(mddev);
5648 mddev->queue->backing_dev_info.congested_fn = NULL;
5649
5650 /* tell userspace to handle 'inactive' */
5651 sysfs_notify_dirent_safe(mddev->sysfs_state);
5652
5653 rdev_for_each(rdev, mddev)
5654 if (rdev->raid_disk >= 0)
5655 sysfs_unlink_rdev(mddev, rdev);
5656
5657 set_capacity(disk, 0);
5658 mutex_unlock(&mddev->open_mutex);
5659 mddev->changed = 1;
5660 revalidate_disk(disk);
5661
5662 if (mddev->ro)
5663 mddev->ro = 0;
5664 } else
5665 mutex_unlock(&mddev->open_mutex);
5666 /*
5667 * Free resources if final stop
5668 */
5669 if (mode == 0) {
5670 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5671
5672 bitmap_destroy(mddev);
5673 if (mddev->bitmap_info.file) {
5674 struct file *f = mddev->bitmap_info.file;
5675 spin_lock(&mddev->lock);
5676 mddev->bitmap_info.file = NULL;
5677 spin_unlock(&mddev->lock);
5678 fput(f);
5679 }
5680 mddev->bitmap_info.offset = 0;
5681
5682 export_array(mddev);
5683
5684 md_clean(mddev);
5685 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5686 if (mddev->hold_active == UNTIL_STOP)
5687 mddev->hold_active = 0;
5688 }
5689 md_new_event(mddev);
5690 sysfs_notify_dirent_safe(mddev->sysfs_state);
5691 return 0;
5692 }
5693
5694 #ifndef MODULE
5695 static void autorun_array(struct mddev *mddev)
5696 {
5697 struct md_rdev *rdev;
5698 int err;
5699
5700 if (list_empty(&mddev->disks))
5701 return;
5702
5703 printk(KERN_INFO "md: running: ");
5704
5705 rdev_for_each(rdev, mddev) {
5706 char b[BDEVNAME_SIZE];
5707 printk("<%s>", bdevname(rdev->bdev,b));
5708 }
5709 printk("\n");
5710
5711 err = do_md_run(mddev);
5712 if (err) {
5713 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
5714 do_md_stop(mddev, 0, NULL);
5715 }
5716 }
5717
5718 /*
5719 * lets try to run arrays based on all disks that have arrived
5720 * until now. (those are in pending_raid_disks)
5721 *
5722 * the method: pick the first pending disk, collect all disks with
5723 * the same UUID, remove all from the pending list and put them into
5724 * the 'same_array' list. Then order this list based on superblock
5725 * update time (freshest comes first), kick out 'old' disks and
5726 * compare superblocks. If everything's fine then run it.
5727 *
5728 * If "unit" is allocated, then bump its reference count
5729 */
5730 static void autorun_devices(int part)
5731 {
5732 struct md_rdev *rdev0, *rdev, *tmp;
5733 struct mddev *mddev;
5734 char b[BDEVNAME_SIZE];
5735
5736 printk(KERN_INFO "md: autorun ...\n");
5737 while (!list_empty(&pending_raid_disks)) {
5738 int unit;
5739 dev_t dev;
5740 LIST_HEAD(candidates);
5741 rdev0 = list_entry(pending_raid_disks.next,
5742 struct md_rdev, same_set);
5743
5744 printk(KERN_INFO "md: considering %s ...\n",
5745 bdevname(rdev0->bdev,b));
5746 INIT_LIST_HEAD(&candidates);
5747 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5748 if (super_90_load(rdev, rdev0, 0) >= 0) {
5749 printk(KERN_INFO "md: adding %s ...\n",
5750 bdevname(rdev->bdev,b));
5751 list_move(&rdev->same_set, &candidates);
5752 }
5753 /*
5754 * now we have a set of devices, with all of them having
5755 * mostly sane superblocks. It's time to allocate the
5756 * mddev.
5757 */
5758 if (part) {
5759 dev = MKDEV(mdp_major,
5760 rdev0->preferred_minor << MdpMinorShift);
5761 unit = MINOR(dev) >> MdpMinorShift;
5762 } else {
5763 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5764 unit = MINOR(dev);
5765 }
5766 if (rdev0->preferred_minor != unit) {
5767 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5768 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5769 break;
5770 }
5771
5772 md_probe(dev, NULL, NULL);
5773 mddev = mddev_find(dev);
5774 if (!mddev || !mddev->gendisk) {
5775 if (mddev)
5776 mddev_put(mddev);
5777 printk(KERN_ERR
5778 "md: cannot allocate memory for md drive.\n");
5779 break;
5780 }
5781 if (mddev_lock(mddev))
5782 printk(KERN_WARNING "md: %s locked, cannot run\n",
5783 mdname(mddev));
5784 else if (mddev->raid_disks || mddev->major_version
5785 || !list_empty(&mddev->disks)) {
5786 printk(KERN_WARNING
5787 "md: %s already running, cannot run %s\n",
5788 mdname(mddev), bdevname(rdev0->bdev,b));
5789 mddev_unlock(mddev);
5790 } else {
5791 printk(KERN_INFO "md: created %s\n", mdname(mddev));
5792 mddev->persistent = 1;
5793 rdev_for_each_list(rdev, tmp, &candidates) {
5794 list_del_init(&rdev->same_set);
5795 if (bind_rdev_to_array(rdev, mddev))
5796 export_rdev(rdev);
5797 }
5798 autorun_array(mddev);
5799 mddev_unlock(mddev);
5800 }
5801 /* on success, candidates will be empty, on error
5802 * it won't...
5803 */
5804 rdev_for_each_list(rdev, tmp, &candidates) {
5805 list_del_init(&rdev->same_set);
5806 export_rdev(rdev);
5807 }
5808 mddev_put(mddev);
5809 }
5810 printk(KERN_INFO "md: ... autorun DONE.\n");
5811 }
5812 #endif /* !MODULE */
5813
5814 static int get_version(void __user *arg)
5815 {
5816 mdu_version_t ver;
5817
5818 ver.major = MD_MAJOR_VERSION;
5819 ver.minor = MD_MINOR_VERSION;
5820 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5821
5822 if (copy_to_user(arg, &ver, sizeof(ver)))
5823 return -EFAULT;
5824
5825 return 0;
5826 }
5827
5828 static int get_array_info(struct mddev *mddev, void __user *arg)
5829 {
5830 mdu_array_info_t info;
5831 int nr,working,insync,failed,spare;
5832 struct md_rdev *rdev;
5833
5834 nr = working = insync = failed = spare = 0;
5835 rcu_read_lock();
5836 rdev_for_each_rcu(rdev, mddev) {
5837 nr++;
5838 if (test_bit(Faulty, &rdev->flags))
5839 failed++;
5840 else {
5841 working++;
5842 if (test_bit(In_sync, &rdev->flags))
5843 insync++;
5844 else
5845 spare++;
5846 }
5847 }
5848 rcu_read_unlock();
5849
5850 info.major_version = mddev->major_version;
5851 info.minor_version = mddev->minor_version;
5852 info.patch_version = MD_PATCHLEVEL_VERSION;
5853 info.ctime = mddev->ctime;
5854 info.level = mddev->level;
5855 info.size = mddev->dev_sectors / 2;
5856 if (info.size != mddev->dev_sectors / 2) /* overflow */
5857 info.size = -1;
5858 info.nr_disks = nr;
5859 info.raid_disks = mddev->raid_disks;
5860 info.md_minor = mddev->md_minor;
5861 info.not_persistent= !mddev->persistent;
5862
5863 info.utime = mddev->utime;
5864 info.state = 0;
5865 if (mddev->in_sync)
5866 info.state = (1<<MD_SB_CLEAN);
5867 if (mddev->bitmap && mddev->bitmap_info.offset)
5868 info.state |= (1<<MD_SB_BITMAP_PRESENT);
5869 if (mddev_is_clustered(mddev))
5870 info.state |= (1<<MD_SB_CLUSTERED);
5871 info.active_disks = insync;
5872 info.working_disks = working;
5873 info.failed_disks = failed;
5874 info.spare_disks = spare;
5875
5876 info.layout = mddev->layout;
5877 info.chunk_size = mddev->chunk_sectors << 9;
5878
5879 if (copy_to_user(arg, &info, sizeof(info)))
5880 return -EFAULT;
5881
5882 return 0;
5883 }
5884
5885 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
5886 {
5887 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5888 char *ptr;
5889 int err;
5890
5891 file = kzalloc(sizeof(*file), GFP_NOIO);
5892 if (!file)
5893 return -ENOMEM;
5894
5895 err = 0;
5896 spin_lock(&mddev->lock);
5897 /* bitmap enabled */
5898 if (mddev->bitmap_info.file) {
5899 ptr = file_path(mddev->bitmap_info.file, file->pathname,
5900 sizeof(file->pathname));
5901 if (IS_ERR(ptr))
5902 err = PTR_ERR(ptr);
5903 else
5904 memmove(file->pathname, ptr,
5905 sizeof(file->pathname)-(ptr-file->pathname));
5906 }
5907 spin_unlock(&mddev->lock);
5908
5909 if (err == 0 &&
5910 copy_to_user(arg, file, sizeof(*file)))
5911 err = -EFAULT;
5912
5913 kfree(file);
5914 return err;
5915 }
5916
5917 static int get_disk_info(struct mddev *mddev, void __user * arg)
5918 {
5919 mdu_disk_info_t info;
5920 struct md_rdev *rdev;
5921
5922 if (copy_from_user(&info, arg, sizeof(info)))
5923 return -EFAULT;
5924
5925 rcu_read_lock();
5926 rdev = md_find_rdev_nr_rcu(mddev, info.number);
5927 if (rdev) {
5928 info.major = MAJOR(rdev->bdev->bd_dev);
5929 info.minor = MINOR(rdev->bdev->bd_dev);
5930 info.raid_disk = rdev->raid_disk;
5931 info.state = 0;
5932 if (test_bit(Faulty, &rdev->flags))
5933 info.state |= (1<<MD_DISK_FAULTY);
5934 else if (test_bit(In_sync, &rdev->flags)) {
5935 info.state |= (1<<MD_DISK_ACTIVE);
5936 info.state |= (1<<MD_DISK_SYNC);
5937 }
5938 if (test_bit(Journal, &rdev->flags))
5939 info.state |= (1<<MD_DISK_JOURNAL);
5940 if (test_bit(WriteMostly, &rdev->flags))
5941 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5942 } else {
5943 info.major = info.minor = 0;
5944 info.raid_disk = -1;
5945 info.state = (1<<MD_DISK_REMOVED);
5946 }
5947 rcu_read_unlock();
5948
5949 if (copy_to_user(arg, &info, sizeof(info)))
5950 return -EFAULT;
5951
5952 return 0;
5953 }
5954
5955 static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
5956 {
5957 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5958 struct md_rdev *rdev;
5959 dev_t dev = MKDEV(info->major,info->minor);
5960
5961 if (mddev_is_clustered(mddev) &&
5962 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
5963 pr_err("%s: Cannot add to clustered mddev.\n",
5964 mdname(mddev));
5965 return -EINVAL;
5966 }
5967
5968 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5969 return -EOVERFLOW;
5970
5971 if (!mddev->raid_disks) {
5972 int err;
5973 /* expecting a device which has a superblock */
5974 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5975 if (IS_ERR(rdev)) {
5976 printk(KERN_WARNING
5977 "md: md_import_device returned %ld\n",
5978 PTR_ERR(rdev));
5979 return PTR_ERR(rdev);
5980 }
5981 if (!list_empty(&mddev->disks)) {
5982 struct md_rdev *rdev0
5983 = list_entry(mddev->disks.next,
5984 struct md_rdev, same_set);
5985 err = super_types[mddev->major_version]
5986 .load_super(rdev, rdev0, mddev->minor_version);
5987 if (err < 0) {
5988 printk(KERN_WARNING
5989 "md: %s has different UUID to %s\n",
5990 bdevname(rdev->bdev,b),
5991 bdevname(rdev0->bdev,b2));
5992 export_rdev(rdev);
5993 return -EINVAL;
5994 }
5995 }
5996 err = bind_rdev_to_array(rdev, mddev);
5997 if (err)
5998 export_rdev(rdev);
5999 return err;
6000 }
6001
6002 /*
6003 * add_new_disk can be used once the array is assembled
6004 * to add "hot spares". They must already have a superblock
6005 * written
6006 */
6007 if (mddev->pers) {
6008 int err;
6009 if (!mddev->pers->hot_add_disk) {
6010 printk(KERN_WARNING
6011 "%s: personality does not support diskops!\n",
6012 mdname(mddev));
6013 return -EINVAL;
6014 }
6015 if (mddev->persistent)
6016 rdev = md_import_device(dev, mddev->major_version,
6017 mddev->minor_version);
6018 else
6019 rdev = md_import_device(dev, -1, -1);
6020 if (IS_ERR(rdev)) {
6021 printk(KERN_WARNING
6022 "md: md_import_device returned %ld\n",
6023 PTR_ERR(rdev));
6024 return PTR_ERR(rdev);
6025 }
6026 /* set saved_raid_disk if appropriate */
6027 if (!mddev->persistent) {
6028 if (info->state & (1<<MD_DISK_SYNC) &&
6029 info->raid_disk < mddev->raid_disks) {
6030 rdev->raid_disk = info->raid_disk;
6031 set_bit(In_sync, &rdev->flags);
6032 clear_bit(Bitmap_sync, &rdev->flags);
6033 } else
6034 rdev->raid_disk = -1;
6035 rdev->saved_raid_disk = rdev->raid_disk;
6036 } else
6037 super_types[mddev->major_version].
6038 validate_super(mddev, rdev);
6039 if ((info->state & (1<<MD_DISK_SYNC)) &&
6040 rdev->raid_disk != info->raid_disk) {
6041 /* This was a hot-add request, but events doesn't
6042 * match, so reject it.
6043 */
6044 export_rdev(rdev);
6045 return -EINVAL;
6046 }
6047
6048 clear_bit(In_sync, &rdev->flags); /* just to be sure */
6049 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6050 set_bit(WriteMostly, &rdev->flags);
6051 else
6052 clear_bit(WriteMostly, &rdev->flags);
6053
6054 if (info->state & (1<<MD_DISK_JOURNAL))
6055 set_bit(Journal, &rdev->flags);
6056 /*
6057 * check whether the device shows up in other nodes
6058 */
6059 if (mddev_is_clustered(mddev)) {
6060 if (info->state & (1 << MD_DISK_CANDIDATE))
6061 set_bit(Candidate, &rdev->flags);
6062 else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6063 /* --add initiated by this node */
6064 err = md_cluster_ops->add_new_disk(mddev, rdev);
6065 if (err) {
6066 export_rdev(rdev);
6067 return err;
6068 }
6069 }
6070 }
6071
6072 rdev->raid_disk = -1;
6073 err = bind_rdev_to_array(rdev, mddev);
6074
6075 if (err)
6076 export_rdev(rdev);
6077
6078 if (mddev_is_clustered(mddev)) {
6079 if (info->state & (1 << MD_DISK_CANDIDATE))
6080 md_cluster_ops->new_disk_ack(mddev, (err == 0));
6081 else {
6082 if (err)
6083 md_cluster_ops->add_new_disk_cancel(mddev);
6084 else
6085 err = add_bound_rdev(rdev);
6086 }
6087
6088 } else if (!err)
6089 err = add_bound_rdev(rdev);
6090
6091 return err;
6092 }
6093
6094 /* otherwise, add_new_disk is only allowed
6095 * for major_version==0 superblocks
6096 */
6097 if (mddev->major_version != 0) {
6098 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
6099 mdname(mddev));
6100 return -EINVAL;
6101 }
6102
6103 if (!(info->state & (1<<MD_DISK_FAULTY))) {
6104 int err;
6105 rdev = md_import_device(dev, -1, 0);
6106 if (IS_ERR(rdev)) {
6107 printk(KERN_WARNING
6108 "md: error, md_import_device() returned %ld\n",
6109 PTR_ERR(rdev));
6110 return PTR_ERR(rdev);
6111 }
6112 rdev->desc_nr = info->number;
6113 if (info->raid_disk < mddev->raid_disks)
6114 rdev->raid_disk = info->raid_disk;
6115 else
6116 rdev->raid_disk = -1;
6117
6118 if (rdev->raid_disk < mddev->raid_disks)
6119 if (info->state & (1<<MD_DISK_SYNC))
6120 set_bit(In_sync, &rdev->flags);
6121
6122 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6123 set_bit(WriteMostly, &rdev->flags);
6124
6125 if (!mddev->persistent) {
6126 printk(KERN_INFO "md: nonpersistent superblock ...\n");
6127 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6128 } else
6129 rdev->sb_start = calc_dev_sboffset(rdev);
6130 rdev->sectors = rdev->sb_start;
6131
6132 err = bind_rdev_to_array(rdev, mddev);
6133 if (err) {
6134 export_rdev(rdev);
6135 return err;
6136 }
6137 }
6138
6139 return 0;
6140 }
6141
6142 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6143 {
6144 char b[BDEVNAME_SIZE];
6145 struct md_rdev *rdev;
6146 int ret = -1;
6147
6148 if (!mddev->pers)
6149 return -ENODEV;
6150
6151 rdev = find_rdev(mddev, dev);
6152 if (!rdev)
6153 return -ENXIO;
6154
6155 if (mddev_is_clustered(mddev))
6156 ret = md_cluster_ops->metadata_update_start(mddev);
6157
6158 if (rdev->raid_disk < 0)
6159 goto kick_rdev;
6160
6161 clear_bit(Blocked, &rdev->flags);
6162 remove_and_add_spares(mddev, rdev);
6163
6164 if (rdev->raid_disk >= 0)
6165 goto busy;
6166
6167 kick_rdev:
6168 if (mddev_is_clustered(mddev) && ret == 0)
6169 md_cluster_ops->remove_disk(mddev, rdev);
6170
6171 md_kick_rdev_from_array(rdev);
6172 md_update_sb(mddev, 1);
6173 md_new_event(mddev);
6174
6175 return 0;
6176 busy:
6177 if (mddev_is_clustered(mddev) && ret == 0)
6178 md_cluster_ops->metadata_update_cancel(mddev);
6179
6180 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
6181 bdevname(rdev->bdev,b), mdname(mddev));
6182 return -EBUSY;
6183 }
6184
6185 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6186 {
6187 char b[BDEVNAME_SIZE];
6188 int err;
6189 struct md_rdev *rdev;
6190
6191 if (!mddev->pers)
6192 return -ENODEV;
6193
6194 if (mddev->major_version != 0) {
6195 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
6196 " version-0 superblocks.\n",
6197 mdname(mddev));
6198 return -EINVAL;
6199 }
6200 if (!mddev->pers->hot_add_disk) {
6201 printk(KERN_WARNING
6202 "%s: personality does not support diskops!\n",
6203 mdname(mddev));
6204 return -EINVAL;
6205 }
6206
6207 rdev = md_import_device(dev, -1, 0);
6208 if (IS_ERR(rdev)) {
6209 printk(KERN_WARNING
6210 "md: error, md_import_device() returned %ld\n",
6211 PTR_ERR(rdev));
6212 return -EINVAL;
6213 }
6214
6215 if (mddev->persistent)
6216 rdev->sb_start = calc_dev_sboffset(rdev);
6217 else
6218 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6219
6220 rdev->sectors = rdev->sb_start;
6221
6222 if (test_bit(Faulty, &rdev->flags)) {
6223 printk(KERN_WARNING
6224 "md: can not hot-add faulty %s disk to %s!\n",
6225 bdevname(rdev->bdev,b), mdname(mddev));
6226 err = -EINVAL;
6227 goto abort_export;
6228 }
6229
6230 clear_bit(In_sync, &rdev->flags);
6231 rdev->desc_nr = -1;
6232 rdev->saved_raid_disk = -1;
6233 err = bind_rdev_to_array(rdev, mddev);
6234 if (err)
6235 goto abort_export;
6236
6237 /*
6238 * The rest should better be atomic, we can have disk failures
6239 * noticed in interrupt contexts ...
6240 */
6241
6242 rdev->raid_disk = -1;
6243
6244 md_update_sb(mddev, 1);
6245 /*
6246 * Kick recovery, maybe this spare has to be added to the
6247 * array immediately.
6248 */
6249 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6250 md_wakeup_thread(mddev->thread);
6251 md_new_event(mddev);
6252 return 0;
6253
6254 abort_export:
6255 export_rdev(rdev);
6256 return err;
6257 }
6258
6259 static int set_bitmap_file(struct mddev *mddev, int fd)
6260 {
6261 int err = 0;
6262
6263 if (mddev->pers) {
6264 if (!mddev->pers->quiesce || !mddev->thread)
6265 return -EBUSY;
6266 if (mddev->recovery || mddev->sync_thread)
6267 return -EBUSY;
6268 /* we should be able to change the bitmap.. */
6269 }
6270
6271 if (fd >= 0) {
6272 struct inode *inode;
6273 struct file *f;
6274
6275 if (mddev->bitmap || mddev->bitmap_info.file)
6276 return -EEXIST; /* cannot add when bitmap is present */
6277 f = fget(fd);
6278
6279 if (f == NULL) {
6280 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
6281 mdname(mddev));
6282 return -EBADF;
6283 }
6284
6285 inode = f->f_mapping->host;
6286 if (!S_ISREG(inode->i_mode)) {
6287 printk(KERN_ERR "%s: error: bitmap file must be a regular file\n",
6288 mdname(mddev));
6289 err = -EBADF;
6290 } else if (!(f->f_mode & FMODE_WRITE)) {
6291 printk(KERN_ERR "%s: error: bitmap file must open for write\n",
6292 mdname(mddev));
6293 err = -EBADF;
6294 } else if (atomic_read(&inode->i_writecount) != 1) {
6295 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
6296 mdname(mddev));
6297 err = -EBUSY;
6298 }
6299 if (err) {
6300 fput(f);
6301 return err;
6302 }
6303 mddev->bitmap_info.file = f;
6304 mddev->bitmap_info.offset = 0; /* file overrides offset */
6305 } else if (mddev->bitmap == NULL)
6306 return -ENOENT; /* cannot remove what isn't there */
6307 err = 0;
6308 if (mddev->pers) {
6309 mddev->pers->quiesce(mddev, 1);
6310 if (fd >= 0) {
6311 struct bitmap *bitmap;
6312
6313 bitmap = bitmap_create(mddev, -1);
6314 if (!IS_ERR(bitmap)) {
6315 mddev->bitmap = bitmap;
6316 err = bitmap_load(mddev);
6317 } else
6318 err = PTR_ERR(bitmap);
6319 }
6320 if (fd < 0 || err) {
6321 bitmap_destroy(mddev);
6322 fd = -1; /* make sure to put the file */
6323 }
6324 mddev->pers->quiesce(mddev, 0);
6325 }
6326 if (fd < 0) {
6327 struct file *f = mddev->bitmap_info.file;
6328 if (f) {
6329 spin_lock(&mddev->lock);
6330 mddev->bitmap_info.file = NULL;
6331 spin_unlock(&mddev->lock);
6332 fput(f);
6333 }
6334 }
6335
6336 return err;
6337 }
6338
6339 /*
6340 * set_array_info is used two different ways
6341 * The original usage is when creating a new array.
6342 * In this usage, raid_disks is > 0 and it together with
6343 * level, size, not_persistent,layout,chunksize determine the
6344 * shape of the array.
6345 * This will always create an array with a type-0.90.0 superblock.
6346 * The newer usage is when assembling an array.
6347 * In this case raid_disks will be 0, and the major_version field is
6348 * use to determine which style super-blocks are to be found on the devices.
6349 * The minor and patch _version numbers are also kept incase the
6350 * super_block handler wishes to interpret them.
6351 */
6352 static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
6353 {
6354
6355 if (info->raid_disks == 0) {
6356 /* just setting version number for superblock loading */
6357 if (info->major_version < 0 ||
6358 info->major_version >= ARRAY_SIZE(super_types) ||
6359 super_types[info->major_version].name == NULL) {
6360 /* maybe try to auto-load a module? */
6361 printk(KERN_INFO
6362 "md: superblock version %d not known\n",
6363 info->major_version);
6364 return -EINVAL;
6365 }
6366 mddev->major_version = info->major_version;
6367 mddev->minor_version = info->minor_version;
6368 mddev->patch_version = info->patch_version;
6369 mddev->persistent = !info->not_persistent;
6370 /* ensure mddev_put doesn't delete this now that there
6371 * is some minimal configuration.
6372 */
6373 mddev->ctime = get_seconds();
6374 return 0;
6375 }
6376 mddev->major_version = MD_MAJOR_VERSION;
6377 mddev->minor_version = MD_MINOR_VERSION;
6378 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6379 mddev->ctime = get_seconds();
6380
6381 mddev->level = info->level;
6382 mddev->clevel[0] = 0;
6383 mddev->dev_sectors = 2 * (sector_t)info->size;
6384 mddev->raid_disks = info->raid_disks;
6385 /* don't set md_minor, it is determined by which /dev/md* was
6386 * openned
6387 */
6388 if (info->state & (1<<MD_SB_CLEAN))
6389 mddev->recovery_cp = MaxSector;
6390 else
6391 mddev->recovery_cp = 0;
6392 mddev->persistent = ! info->not_persistent;
6393 mddev->external = 0;
6394
6395 mddev->layout = info->layout;
6396 mddev->chunk_sectors = info->chunk_size >> 9;
6397
6398 mddev->max_disks = MD_SB_DISKS;
6399
6400 if (mddev->persistent)
6401 mddev->flags = 0;
6402 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6403
6404 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6405 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6406 mddev->bitmap_info.offset = 0;
6407
6408 mddev->reshape_position = MaxSector;
6409
6410 /*
6411 * Generate a 128 bit UUID
6412 */
6413 get_random_bytes(mddev->uuid, 16);
6414
6415 mddev->new_level = mddev->level;
6416 mddev->new_chunk_sectors = mddev->chunk_sectors;
6417 mddev->new_layout = mddev->layout;
6418 mddev->delta_disks = 0;
6419 mddev->reshape_backwards = 0;
6420
6421 return 0;
6422 }
6423
6424 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6425 {
6426 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6427
6428 if (mddev->external_size)
6429 return;
6430
6431 mddev->array_sectors = array_sectors;
6432 }
6433 EXPORT_SYMBOL(md_set_array_sectors);
6434
6435 static int update_size(struct mddev *mddev, sector_t num_sectors)
6436 {
6437 struct md_rdev *rdev;
6438 int rv;
6439 int fit = (num_sectors == 0);
6440
6441 if (mddev->pers->resize == NULL)
6442 return -EINVAL;
6443 /* The "num_sectors" is the number of sectors of each device that
6444 * is used. This can only make sense for arrays with redundancy.
6445 * linear and raid0 always use whatever space is available. We can only
6446 * consider changing this number if no resync or reconstruction is
6447 * happening, and if the new size is acceptable. It must fit before the
6448 * sb_start or, if that is <data_offset, it must fit before the size
6449 * of each device. If num_sectors is zero, we find the largest size
6450 * that fits.
6451 */
6452 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6453 mddev->sync_thread)
6454 return -EBUSY;
6455 if (mddev->ro)
6456 return -EROFS;
6457
6458 rdev_for_each(rdev, mddev) {
6459 sector_t avail = rdev->sectors;
6460
6461 if (fit && (num_sectors == 0 || num_sectors > avail))
6462 num_sectors = avail;
6463 if (avail < num_sectors)
6464 return -ENOSPC;
6465 }
6466 rv = mddev->pers->resize(mddev, num_sectors);
6467 if (!rv)
6468 revalidate_disk(mddev->gendisk);
6469 return rv;
6470 }
6471
6472 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6473 {
6474 int rv;
6475 struct md_rdev *rdev;
6476 /* change the number of raid disks */
6477 if (mddev->pers->check_reshape == NULL)
6478 return -EINVAL;
6479 if (mddev->ro)
6480 return -EROFS;
6481 if (raid_disks <= 0 ||
6482 (mddev->max_disks && raid_disks >= mddev->max_disks))
6483 return -EINVAL;
6484 if (mddev->sync_thread ||
6485 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6486 mddev->reshape_position != MaxSector)
6487 return -EBUSY;
6488
6489 rdev_for_each(rdev, mddev) {
6490 if (mddev->raid_disks < raid_disks &&
6491 rdev->data_offset < rdev->new_data_offset)
6492 return -EINVAL;
6493 if (mddev->raid_disks > raid_disks &&
6494 rdev->data_offset > rdev->new_data_offset)
6495 return -EINVAL;
6496 }
6497
6498 mddev->delta_disks = raid_disks - mddev->raid_disks;
6499 if (mddev->delta_disks < 0)
6500 mddev->reshape_backwards = 1;
6501 else if (mddev->delta_disks > 0)
6502 mddev->reshape_backwards = 0;
6503
6504 rv = mddev->pers->check_reshape(mddev);
6505 if (rv < 0) {
6506 mddev->delta_disks = 0;
6507 mddev->reshape_backwards = 0;
6508 }
6509 return rv;
6510 }
6511
6512 /*
6513 * update_array_info is used to change the configuration of an
6514 * on-line array.
6515 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6516 * fields in the info are checked against the array.
6517 * Any differences that cannot be handled will cause an error.
6518 * Normally, only one change can be managed at a time.
6519 */
6520 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6521 {
6522 int rv = 0;
6523 int cnt = 0;
6524 int state = 0;
6525
6526 /* calculate expected state,ignoring low bits */
6527 if (mddev->bitmap && mddev->bitmap_info.offset)
6528 state |= (1 << MD_SB_BITMAP_PRESENT);
6529
6530 if (mddev->major_version != info->major_version ||
6531 mddev->minor_version != info->minor_version ||
6532 /* mddev->patch_version != info->patch_version || */
6533 mddev->ctime != info->ctime ||
6534 mddev->level != info->level ||
6535 /* mddev->layout != info->layout || */
6536 mddev->persistent != !info->not_persistent ||
6537 mddev->chunk_sectors != info->chunk_size >> 9 ||
6538 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6539 ((state^info->state) & 0xfffffe00)
6540 )
6541 return -EINVAL;
6542 /* Check there is only one change */
6543 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6544 cnt++;
6545 if (mddev->raid_disks != info->raid_disks)
6546 cnt++;
6547 if (mddev->layout != info->layout)
6548 cnt++;
6549 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6550 cnt++;
6551 if (cnt == 0)
6552 return 0;
6553 if (cnt > 1)
6554 return -EINVAL;
6555
6556 if (mddev->layout != info->layout) {
6557 /* Change layout
6558 * we don't need to do anything at the md level, the
6559 * personality will take care of it all.
6560 */
6561 if (mddev->pers->check_reshape == NULL)
6562 return -EINVAL;
6563 else {
6564 mddev->new_layout = info->layout;
6565 rv = mddev->pers->check_reshape(mddev);
6566 if (rv)
6567 mddev->new_layout = mddev->layout;
6568 return rv;
6569 }
6570 }
6571 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6572 rv = update_size(mddev, (sector_t)info->size * 2);
6573
6574 if (mddev->raid_disks != info->raid_disks)
6575 rv = update_raid_disks(mddev, info->raid_disks);
6576
6577 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6578 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
6579 rv = -EINVAL;
6580 goto err;
6581 }
6582 if (mddev->recovery || mddev->sync_thread) {
6583 rv = -EBUSY;
6584 goto err;
6585 }
6586 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6587 struct bitmap *bitmap;
6588 /* add the bitmap */
6589 if (mddev->bitmap) {
6590 rv = -EEXIST;
6591 goto err;
6592 }
6593 if (mddev->bitmap_info.default_offset == 0) {
6594 rv = -EINVAL;
6595 goto err;
6596 }
6597 mddev->bitmap_info.offset =
6598 mddev->bitmap_info.default_offset;
6599 mddev->bitmap_info.space =
6600 mddev->bitmap_info.default_space;
6601 mddev->pers->quiesce(mddev, 1);
6602 bitmap = bitmap_create(mddev, -1);
6603 if (!IS_ERR(bitmap)) {
6604 mddev->bitmap = bitmap;
6605 rv = bitmap_load(mddev);
6606 } else
6607 rv = PTR_ERR(bitmap);
6608 if (rv)
6609 bitmap_destroy(mddev);
6610 mddev->pers->quiesce(mddev, 0);
6611 } else {
6612 /* remove the bitmap */
6613 if (!mddev->bitmap) {
6614 rv = -ENOENT;
6615 goto err;
6616 }
6617 if (mddev->bitmap->storage.file) {
6618 rv = -EINVAL;
6619 goto err;
6620 }
6621 mddev->pers->quiesce(mddev, 1);
6622 bitmap_destroy(mddev);
6623 mddev->pers->quiesce(mddev, 0);
6624 mddev->bitmap_info.offset = 0;
6625 }
6626 }
6627 md_update_sb(mddev, 1);
6628 return rv;
6629 err:
6630 return rv;
6631 }
6632
6633 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6634 {
6635 struct md_rdev *rdev;
6636 int err = 0;
6637
6638 if (mddev->pers == NULL)
6639 return -ENODEV;
6640
6641 rcu_read_lock();
6642 rdev = find_rdev_rcu(mddev, dev);
6643 if (!rdev)
6644 err = -ENODEV;
6645 else {
6646 md_error(mddev, rdev);
6647 if (!test_bit(Faulty, &rdev->flags))
6648 err = -EBUSY;
6649 }
6650 rcu_read_unlock();
6651 return err;
6652 }
6653
6654 /*
6655 * We have a problem here : there is no easy way to give a CHS
6656 * virtual geometry. We currently pretend that we have a 2 heads
6657 * 4 sectors (with a BIG number of cylinders...). This drives
6658 * dosfs just mad... ;-)
6659 */
6660 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6661 {
6662 struct mddev *mddev = bdev->bd_disk->private_data;
6663
6664 geo->heads = 2;
6665 geo->sectors = 4;
6666 geo->cylinders = mddev->array_sectors / 8;
6667 return 0;
6668 }
6669
6670 static inline bool md_ioctl_valid(unsigned int cmd)
6671 {
6672 switch (cmd) {
6673 case ADD_NEW_DISK:
6674 case BLKROSET:
6675 case GET_ARRAY_INFO:
6676 case GET_BITMAP_FILE:
6677 case GET_DISK_INFO:
6678 case HOT_ADD_DISK:
6679 case HOT_REMOVE_DISK:
6680 case RAID_AUTORUN:
6681 case RAID_VERSION:
6682 case RESTART_ARRAY_RW:
6683 case RUN_ARRAY:
6684 case SET_ARRAY_INFO:
6685 case SET_BITMAP_FILE:
6686 case SET_DISK_FAULTY:
6687 case STOP_ARRAY:
6688 case STOP_ARRAY_RO:
6689 case CLUSTERED_DISK_NACK:
6690 return true;
6691 default:
6692 return false;
6693 }
6694 }
6695
6696 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6697 unsigned int cmd, unsigned long arg)
6698 {
6699 int err = 0;
6700 void __user *argp = (void __user *)arg;
6701 struct mddev *mddev = NULL;
6702 int ro;
6703
6704 if (!md_ioctl_valid(cmd))
6705 return -ENOTTY;
6706
6707 switch (cmd) {
6708 case RAID_VERSION:
6709 case GET_ARRAY_INFO:
6710 case GET_DISK_INFO:
6711 break;
6712 default:
6713 if (!capable(CAP_SYS_ADMIN))
6714 return -EACCES;
6715 }
6716
6717 /*
6718 * Commands dealing with the RAID driver but not any
6719 * particular array:
6720 */
6721 switch (cmd) {
6722 case RAID_VERSION:
6723 err = get_version(argp);
6724 goto out;
6725
6726 #ifndef MODULE
6727 case RAID_AUTORUN:
6728 err = 0;
6729 autostart_arrays(arg);
6730 goto out;
6731 #endif
6732 default:;
6733 }
6734
6735 /*
6736 * Commands creating/starting a new array:
6737 */
6738
6739 mddev = bdev->bd_disk->private_data;
6740
6741 if (!mddev) {
6742 BUG();
6743 goto out;
6744 }
6745
6746 /* Some actions do not requires the mutex */
6747 switch (cmd) {
6748 case GET_ARRAY_INFO:
6749 if (!mddev->raid_disks && !mddev->external)
6750 err = -ENODEV;
6751 else
6752 err = get_array_info(mddev, argp);
6753 goto out;
6754
6755 case GET_DISK_INFO:
6756 if (!mddev->raid_disks && !mddev->external)
6757 err = -ENODEV;
6758 else
6759 err = get_disk_info(mddev, argp);
6760 goto out;
6761
6762 case SET_DISK_FAULTY:
6763 err = set_disk_faulty(mddev, new_decode_dev(arg));
6764 goto out;
6765
6766 case GET_BITMAP_FILE:
6767 err = get_bitmap_file(mddev, argp);
6768 goto out;
6769
6770 }
6771
6772 if (cmd == ADD_NEW_DISK)
6773 /* need to ensure md_delayed_delete() has completed */
6774 flush_workqueue(md_misc_wq);
6775
6776 if (cmd == HOT_REMOVE_DISK)
6777 /* need to ensure recovery thread has run */
6778 wait_event_interruptible_timeout(mddev->sb_wait,
6779 !test_bit(MD_RECOVERY_NEEDED,
6780 &mddev->recovery),
6781 msecs_to_jiffies(5000));
6782 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6783 /* Need to flush page cache, and ensure no-one else opens
6784 * and writes
6785 */
6786 mutex_lock(&mddev->open_mutex);
6787 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
6788 mutex_unlock(&mddev->open_mutex);
6789 err = -EBUSY;
6790 goto out;
6791 }
6792 set_bit(MD_STILL_CLOSED, &mddev->flags);
6793 mutex_unlock(&mddev->open_mutex);
6794 sync_blockdev(bdev);
6795 }
6796 err = mddev_lock(mddev);
6797 if (err) {
6798 printk(KERN_INFO
6799 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6800 err, cmd);
6801 goto out;
6802 }
6803
6804 if (cmd == SET_ARRAY_INFO) {
6805 mdu_array_info_t info;
6806 if (!arg)
6807 memset(&info, 0, sizeof(info));
6808 else if (copy_from_user(&info, argp, sizeof(info))) {
6809 err = -EFAULT;
6810 goto unlock;
6811 }
6812 if (mddev->pers) {
6813 err = update_array_info(mddev, &info);
6814 if (err) {
6815 printk(KERN_WARNING "md: couldn't update"
6816 " array info. %d\n", err);
6817 goto unlock;
6818 }
6819 goto unlock;
6820 }
6821 if (!list_empty(&mddev->disks)) {
6822 printk(KERN_WARNING
6823 "md: array %s already has disks!\n",
6824 mdname(mddev));
6825 err = -EBUSY;
6826 goto unlock;
6827 }
6828 if (mddev->raid_disks) {
6829 printk(KERN_WARNING
6830 "md: array %s already initialised!\n",
6831 mdname(mddev));
6832 err = -EBUSY;
6833 goto unlock;
6834 }
6835 err = set_array_info(mddev, &info);
6836 if (err) {
6837 printk(KERN_WARNING "md: couldn't set"
6838 " array info. %d\n", err);
6839 goto unlock;
6840 }
6841 goto unlock;
6842 }
6843
6844 /*
6845 * Commands querying/configuring an existing array:
6846 */
6847 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6848 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6849 if ((!mddev->raid_disks && !mddev->external)
6850 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6851 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6852 && cmd != GET_BITMAP_FILE) {
6853 err = -ENODEV;
6854 goto unlock;
6855 }
6856
6857 /*
6858 * Commands even a read-only array can execute:
6859 */
6860 switch (cmd) {
6861 case RESTART_ARRAY_RW:
6862 err = restart_array(mddev);
6863 goto unlock;
6864
6865 case STOP_ARRAY:
6866 err = do_md_stop(mddev, 0, bdev);
6867 goto unlock;
6868
6869 case STOP_ARRAY_RO:
6870 err = md_set_readonly(mddev, bdev);
6871 goto unlock;
6872
6873 case HOT_REMOVE_DISK:
6874 err = hot_remove_disk(mddev, new_decode_dev(arg));
6875 goto unlock;
6876
6877 case ADD_NEW_DISK:
6878 /* We can support ADD_NEW_DISK on read-only arrays
6879 * on if we are re-adding a preexisting device.
6880 * So require mddev->pers and MD_DISK_SYNC.
6881 */
6882 if (mddev->pers) {
6883 mdu_disk_info_t info;
6884 if (copy_from_user(&info, argp, sizeof(info)))
6885 err = -EFAULT;
6886 else if (!(info.state & (1<<MD_DISK_SYNC)))
6887 /* Need to clear read-only for this */
6888 break;
6889 else
6890 err = add_new_disk(mddev, &info);
6891 goto unlock;
6892 }
6893 break;
6894
6895 case BLKROSET:
6896 if (get_user(ro, (int __user *)(arg))) {
6897 err = -EFAULT;
6898 goto unlock;
6899 }
6900 err = -EINVAL;
6901
6902 /* if the bdev is going readonly the value of mddev->ro
6903 * does not matter, no writes are coming
6904 */
6905 if (ro)
6906 goto unlock;
6907
6908 /* are we are already prepared for writes? */
6909 if (mddev->ro != 1)
6910 goto unlock;
6911
6912 /* transitioning to readauto need only happen for
6913 * arrays that call md_write_start
6914 */
6915 if (mddev->pers) {
6916 err = restart_array(mddev);
6917 if (err == 0) {
6918 mddev->ro = 2;
6919 set_disk_ro(mddev->gendisk, 0);
6920 }
6921 }
6922 goto unlock;
6923 }
6924
6925 /*
6926 * The remaining ioctls are changing the state of the
6927 * superblock, so we do not allow them on read-only arrays.
6928 */
6929 if (mddev->ro && mddev->pers) {
6930 if (mddev->ro == 2) {
6931 mddev->ro = 0;
6932 sysfs_notify_dirent_safe(mddev->sysfs_state);
6933 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6934 /* mddev_unlock will wake thread */
6935 /* If a device failed while we were read-only, we
6936 * need to make sure the metadata is updated now.
6937 */
6938 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6939 mddev_unlock(mddev);
6940 wait_event(mddev->sb_wait,
6941 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6942 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6943 mddev_lock_nointr(mddev);
6944 }
6945 } else {
6946 err = -EROFS;
6947 goto unlock;
6948 }
6949 }
6950
6951 switch (cmd) {
6952 case ADD_NEW_DISK:
6953 {
6954 mdu_disk_info_t info;
6955 if (copy_from_user(&info, argp, sizeof(info)))
6956 err = -EFAULT;
6957 else
6958 err = add_new_disk(mddev, &info);
6959 goto unlock;
6960 }
6961
6962 case CLUSTERED_DISK_NACK:
6963 if (mddev_is_clustered(mddev))
6964 md_cluster_ops->new_disk_ack(mddev, false);
6965 else
6966 err = -EINVAL;
6967 goto unlock;
6968
6969 case HOT_ADD_DISK:
6970 err = hot_add_disk(mddev, new_decode_dev(arg));
6971 goto unlock;
6972
6973 case RUN_ARRAY:
6974 err = do_md_run(mddev);
6975 goto unlock;
6976
6977 case SET_BITMAP_FILE:
6978 err = set_bitmap_file(mddev, (int)arg);
6979 goto unlock;
6980
6981 default:
6982 err = -EINVAL;
6983 goto unlock;
6984 }
6985
6986 unlock:
6987 if (mddev->hold_active == UNTIL_IOCTL &&
6988 err != -EINVAL)
6989 mddev->hold_active = 0;
6990 mddev_unlock(mddev);
6991 out:
6992 return err;
6993 }
6994 #ifdef CONFIG_COMPAT
6995 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6996 unsigned int cmd, unsigned long arg)
6997 {
6998 switch (cmd) {
6999 case HOT_REMOVE_DISK:
7000 case HOT_ADD_DISK:
7001 case SET_DISK_FAULTY:
7002 case SET_BITMAP_FILE:
7003 /* These take in integer arg, do not convert */
7004 break;
7005 default:
7006 arg = (unsigned long)compat_ptr(arg);
7007 break;
7008 }
7009
7010 return md_ioctl(bdev, mode, cmd, arg);
7011 }
7012 #endif /* CONFIG_COMPAT */
7013
7014 static int md_open(struct block_device *bdev, fmode_t mode)
7015 {
7016 /*
7017 * Succeed if we can lock the mddev, which confirms that
7018 * it isn't being stopped right now.
7019 */
7020 struct mddev *mddev = mddev_find(bdev->bd_dev);
7021 int err;
7022
7023 if (!mddev)
7024 return -ENODEV;
7025
7026 if (mddev->gendisk != bdev->bd_disk) {
7027 /* we are racing with mddev_put which is discarding this
7028 * bd_disk.
7029 */
7030 mddev_put(mddev);
7031 /* Wait until bdev->bd_disk is definitely gone */
7032 flush_workqueue(md_misc_wq);
7033 /* Then retry the open from the top */
7034 return -ERESTARTSYS;
7035 }
7036 BUG_ON(mddev != bdev->bd_disk->private_data);
7037
7038 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
7039 goto out;
7040
7041 err = 0;
7042 atomic_inc(&mddev->openers);
7043 clear_bit(MD_STILL_CLOSED, &mddev->flags);
7044 mutex_unlock(&mddev->open_mutex);
7045
7046 check_disk_change(bdev);
7047 out:
7048 return err;
7049 }
7050
7051 static void md_release(struct gendisk *disk, fmode_t mode)
7052 {
7053 struct mddev *mddev = disk->private_data;
7054
7055 BUG_ON(!mddev);
7056 atomic_dec(&mddev->openers);
7057 mddev_put(mddev);
7058 }
7059
7060 static int md_media_changed(struct gendisk *disk)
7061 {
7062 struct mddev *mddev = disk->private_data;
7063
7064 return mddev->changed;
7065 }
7066
7067 static int md_revalidate(struct gendisk *disk)
7068 {
7069 struct mddev *mddev = disk->private_data;
7070
7071 mddev->changed = 0;
7072 return 0;
7073 }
7074 static const struct block_device_operations md_fops =
7075 {
7076 .owner = THIS_MODULE,
7077 .open = md_open,
7078 .release = md_release,
7079 .ioctl = md_ioctl,
7080 #ifdef CONFIG_COMPAT
7081 .compat_ioctl = md_compat_ioctl,
7082 #endif
7083 .getgeo = md_getgeo,
7084 .media_changed = md_media_changed,
7085 .revalidate_disk= md_revalidate,
7086 };
7087
7088 static int md_thread(void *arg)
7089 {
7090 struct md_thread *thread = arg;
7091
7092 /*
7093 * md_thread is a 'system-thread', it's priority should be very
7094 * high. We avoid resource deadlocks individually in each
7095 * raid personality. (RAID5 does preallocation) We also use RR and
7096 * the very same RT priority as kswapd, thus we will never get
7097 * into a priority inversion deadlock.
7098 *
7099 * we definitely have to have equal or higher priority than
7100 * bdflush, otherwise bdflush will deadlock if there are too
7101 * many dirty RAID5 blocks.
7102 */
7103
7104 allow_signal(SIGKILL);
7105 while (!kthread_should_stop()) {
7106
7107 /* We need to wait INTERRUPTIBLE so that
7108 * we don't add to the load-average.
7109 * That means we need to be sure no signals are
7110 * pending
7111 */
7112 if (signal_pending(current))
7113 flush_signals(current);
7114
7115 wait_event_interruptible_timeout
7116 (thread->wqueue,
7117 test_bit(THREAD_WAKEUP, &thread->flags)
7118 || kthread_should_stop(),
7119 thread->timeout);
7120
7121 clear_bit(THREAD_WAKEUP, &thread->flags);
7122 if (!kthread_should_stop())
7123 thread->run(thread);
7124 }
7125
7126 return 0;
7127 }
7128
7129 void md_wakeup_thread(struct md_thread *thread)
7130 {
7131 if (thread) {
7132 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
7133 set_bit(THREAD_WAKEUP, &thread->flags);
7134 wake_up(&thread->wqueue);
7135 }
7136 }
7137 EXPORT_SYMBOL(md_wakeup_thread);
7138
7139 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7140 struct mddev *mddev, const char *name)
7141 {
7142 struct md_thread *thread;
7143
7144 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7145 if (!thread)
7146 return NULL;
7147
7148 init_waitqueue_head(&thread->wqueue);
7149
7150 thread->run = run;
7151 thread->mddev = mddev;
7152 thread->timeout = MAX_SCHEDULE_TIMEOUT;
7153 thread->tsk = kthread_run(md_thread, thread,
7154 "%s_%s",
7155 mdname(thread->mddev),
7156 name);
7157 if (IS_ERR(thread->tsk)) {
7158 kfree(thread);
7159 return NULL;
7160 }
7161 return thread;
7162 }
7163 EXPORT_SYMBOL(md_register_thread);
7164
7165 void md_unregister_thread(struct md_thread **threadp)
7166 {
7167 struct md_thread *thread = *threadp;
7168 if (!thread)
7169 return;
7170 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7171 /* Locking ensures that mddev_unlock does not wake_up a
7172 * non-existent thread
7173 */
7174 spin_lock(&pers_lock);
7175 *threadp = NULL;
7176 spin_unlock(&pers_lock);
7177
7178 kthread_stop(thread->tsk);
7179 kfree(thread);
7180 }
7181 EXPORT_SYMBOL(md_unregister_thread);
7182
7183 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7184 {
7185 if (!rdev || test_bit(Faulty, &rdev->flags))
7186 return;
7187
7188 if (!mddev->pers || !mddev->pers->error_handler)
7189 return;
7190 mddev->pers->error_handler(mddev,rdev);
7191 if (mddev->degraded)
7192 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7193 sysfs_notify_dirent_safe(rdev->sysfs_state);
7194 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7195 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7196 md_wakeup_thread(mddev->thread);
7197 if (mddev->event_work.func)
7198 queue_work(md_misc_wq, &mddev->event_work);
7199 md_new_event_inintr(mddev);
7200 }
7201 EXPORT_SYMBOL(md_error);
7202
7203 /* seq_file implementation /proc/mdstat */
7204
7205 static void status_unused(struct seq_file *seq)
7206 {
7207 int i = 0;
7208 struct md_rdev *rdev;
7209
7210 seq_printf(seq, "unused devices: ");
7211
7212 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
7213 char b[BDEVNAME_SIZE];
7214 i++;
7215 seq_printf(seq, "%s ",
7216 bdevname(rdev->bdev,b));
7217 }
7218 if (!i)
7219 seq_printf(seq, "<none>");
7220
7221 seq_printf(seq, "\n");
7222 }
7223
7224 static int status_resync(struct seq_file *seq, struct mddev *mddev)
7225 {
7226 sector_t max_sectors, resync, res;
7227 unsigned long dt, db;
7228 sector_t rt;
7229 int scale;
7230 unsigned int per_milli;
7231
7232 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7233 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7234 max_sectors = mddev->resync_max_sectors;
7235 else
7236 max_sectors = mddev->dev_sectors;
7237
7238 resync = mddev->curr_resync;
7239 if (resync <= 3) {
7240 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7241 /* Still cleaning up */
7242 resync = max_sectors;
7243 } else
7244 resync -= atomic_read(&mddev->recovery_active);
7245
7246 if (resync == 0) {
7247 if (mddev->recovery_cp < MaxSector) {
7248 seq_printf(seq, "\tresync=PENDING");
7249 return 1;
7250 }
7251 return 0;
7252 }
7253 if (resync < 3) {
7254 seq_printf(seq, "\tresync=DELAYED");
7255 return 1;
7256 }
7257
7258 WARN_ON(max_sectors == 0);
7259 /* Pick 'scale' such that (resync>>scale)*1000 will fit
7260 * in a sector_t, and (max_sectors>>scale) will fit in a
7261 * u32, as those are the requirements for sector_div.
7262 * Thus 'scale' must be at least 10
7263 */
7264 scale = 10;
7265 if (sizeof(sector_t) > sizeof(unsigned long)) {
7266 while ( max_sectors/2 > (1ULL<<(scale+32)))
7267 scale++;
7268 }
7269 res = (resync>>scale)*1000;
7270 sector_div(res, (u32)((max_sectors>>scale)+1));
7271
7272 per_milli = res;
7273 {
7274 int i, x = per_milli/50, y = 20-x;
7275 seq_printf(seq, "[");
7276 for (i = 0; i < x; i++)
7277 seq_printf(seq, "=");
7278 seq_printf(seq, ">");
7279 for (i = 0; i < y; i++)
7280 seq_printf(seq, ".");
7281 seq_printf(seq, "] ");
7282 }
7283 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
7284 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
7285 "reshape" :
7286 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
7287 "check" :
7288 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
7289 "resync" : "recovery"))),
7290 per_milli/10, per_milli % 10,
7291 (unsigned long long) resync/2,
7292 (unsigned long long) max_sectors/2);
7293
7294 /*
7295 * dt: time from mark until now
7296 * db: blocks written from mark until now
7297 * rt: remaining time
7298 *
7299 * rt is a sector_t, so could be 32bit or 64bit.
7300 * So we divide before multiply in case it is 32bit and close
7301 * to the limit.
7302 * We scale the divisor (db) by 32 to avoid losing precision
7303 * near the end of resync when the number of remaining sectors
7304 * is close to 'db'.
7305 * We then divide rt by 32 after multiplying by db to compensate.
7306 * The '+1' avoids division by zero if db is very small.
7307 */
7308 dt = ((jiffies - mddev->resync_mark) / HZ);
7309 if (!dt) dt++;
7310 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
7311 - mddev->resync_mark_cnt;
7312
7313 rt = max_sectors - resync; /* number of remaining sectors */
7314 sector_div(rt, db/32+1);
7315 rt *= dt;
7316 rt >>= 5;
7317
7318 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7319 ((unsigned long)rt % 60)/6);
7320
7321 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
7322 return 1;
7323 }
7324
7325 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7326 {
7327 struct list_head *tmp;
7328 loff_t l = *pos;
7329 struct mddev *mddev;
7330
7331 if (l >= 0x10000)
7332 return NULL;
7333 if (!l--)
7334 /* header */
7335 return (void*)1;
7336
7337 spin_lock(&all_mddevs_lock);
7338 list_for_each(tmp,&all_mddevs)
7339 if (!l--) {
7340 mddev = list_entry(tmp, struct mddev, all_mddevs);
7341 mddev_get(mddev);
7342 spin_unlock(&all_mddevs_lock);
7343 return mddev;
7344 }
7345 spin_unlock(&all_mddevs_lock);
7346 if (!l--)
7347 return (void*)2;/* tail */
7348 return NULL;
7349 }
7350
7351 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7352 {
7353 struct list_head *tmp;
7354 struct mddev *next_mddev, *mddev = v;
7355
7356 ++*pos;
7357 if (v == (void*)2)
7358 return NULL;
7359
7360 spin_lock(&all_mddevs_lock);
7361 if (v == (void*)1)
7362 tmp = all_mddevs.next;
7363 else
7364 tmp = mddev->all_mddevs.next;
7365 if (tmp != &all_mddevs)
7366 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
7367 else {
7368 next_mddev = (void*)2;
7369 *pos = 0x10000;
7370 }
7371 spin_unlock(&all_mddevs_lock);
7372
7373 if (v != (void*)1)
7374 mddev_put(mddev);
7375 return next_mddev;
7376
7377 }
7378
7379 static void md_seq_stop(struct seq_file *seq, void *v)
7380 {
7381 struct mddev *mddev = v;
7382
7383 if (mddev && v != (void*)1 && v != (void*)2)
7384 mddev_put(mddev);
7385 }
7386
7387 static int md_seq_show(struct seq_file *seq, void *v)
7388 {
7389 struct mddev *mddev = v;
7390 sector_t sectors;
7391 struct md_rdev *rdev;
7392
7393 if (v == (void*)1) {
7394 struct md_personality *pers;
7395 seq_printf(seq, "Personalities : ");
7396 spin_lock(&pers_lock);
7397 list_for_each_entry(pers, &pers_list, list)
7398 seq_printf(seq, "[%s] ", pers->name);
7399
7400 spin_unlock(&pers_lock);
7401 seq_printf(seq, "\n");
7402 seq->poll_event = atomic_read(&md_event_count);
7403 return 0;
7404 }
7405 if (v == (void*)2) {
7406 status_unused(seq);
7407 return 0;
7408 }
7409
7410 spin_lock(&mddev->lock);
7411 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7412 seq_printf(seq, "%s : %sactive", mdname(mddev),
7413 mddev->pers ? "" : "in");
7414 if (mddev->pers) {
7415 if (mddev->ro==1)
7416 seq_printf(seq, " (read-only)");
7417 if (mddev->ro==2)
7418 seq_printf(seq, " (auto-read-only)");
7419 seq_printf(seq, " %s", mddev->pers->name);
7420 }
7421
7422 sectors = 0;
7423 rcu_read_lock();
7424 rdev_for_each_rcu(rdev, mddev) {
7425 char b[BDEVNAME_SIZE];
7426 seq_printf(seq, " %s[%d]",
7427 bdevname(rdev->bdev,b), rdev->desc_nr);
7428 if (test_bit(WriteMostly, &rdev->flags))
7429 seq_printf(seq, "(W)");
7430 if (test_bit(Journal, &rdev->flags))
7431 seq_printf(seq, "(J)");
7432 if (test_bit(Faulty, &rdev->flags)) {
7433 seq_printf(seq, "(F)");
7434 continue;
7435 }
7436 if (rdev->raid_disk < 0)
7437 seq_printf(seq, "(S)"); /* spare */
7438 if (test_bit(Replacement, &rdev->flags))
7439 seq_printf(seq, "(R)");
7440 sectors += rdev->sectors;
7441 }
7442 rcu_read_unlock();
7443
7444 if (!list_empty(&mddev->disks)) {
7445 if (mddev->pers)
7446 seq_printf(seq, "\n %llu blocks",
7447 (unsigned long long)
7448 mddev->array_sectors / 2);
7449 else
7450 seq_printf(seq, "\n %llu blocks",
7451 (unsigned long long)sectors / 2);
7452 }
7453 if (mddev->persistent) {
7454 if (mddev->major_version != 0 ||
7455 mddev->minor_version != 90) {
7456 seq_printf(seq," super %d.%d",
7457 mddev->major_version,
7458 mddev->minor_version);
7459 }
7460 } else if (mddev->external)
7461 seq_printf(seq, " super external:%s",
7462 mddev->metadata_type);
7463 else
7464 seq_printf(seq, " super non-persistent");
7465
7466 if (mddev->pers) {
7467 mddev->pers->status(seq, mddev);
7468 seq_printf(seq, "\n ");
7469 if (mddev->pers->sync_request) {
7470 if (status_resync(seq, mddev))
7471 seq_printf(seq, "\n ");
7472 }
7473 } else
7474 seq_printf(seq, "\n ");
7475
7476 bitmap_status(seq, mddev->bitmap);
7477
7478 seq_printf(seq, "\n");
7479 }
7480 spin_unlock(&mddev->lock);
7481
7482 return 0;
7483 }
7484
7485 static const struct seq_operations md_seq_ops = {
7486 .start = md_seq_start,
7487 .next = md_seq_next,
7488 .stop = md_seq_stop,
7489 .show = md_seq_show,
7490 };
7491
7492 static int md_seq_open(struct inode *inode, struct file *file)
7493 {
7494 struct seq_file *seq;
7495 int error;
7496
7497 error = seq_open(file, &md_seq_ops);
7498 if (error)
7499 return error;
7500
7501 seq = file->private_data;
7502 seq->poll_event = atomic_read(&md_event_count);
7503 return error;
7504 }
7505
7506 static int md_unloading;
7507 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7508 {
7509 struct seq_file *seq = filp->private_data;
7510 int mask;
7511
7512 if (md_unloading)
7513 return POLLIN|POLLRDNORM|POLLERR|POLLPRI;
7514 poll_wait(filp, &md_event_waiters, wait);
7515
7516 /* always allow read */
7517 mask = POLLIN | POLLRDNORM;
7518
7519 if (seq->poll_event != atomic_read(&md_event_count))
7520 mask |= POLLERR | POLLPRI;
7521 return mask;
7522 }
7523
7524 static const struct file_operations md_seq_fops = {
7525 .owner = THIS_MODULE,
7526 .open = md_seq_open,
7527 .read = seq_read,
7528 .llseek = seq_lseek,
7529 .release = seq_release_private,
7530 .poll = mdstat_poll,
7531 };
7532
7533 int register_md_personality(struct md_personality *p)
7534 {
7535 printk(KERN_INFO "md: %s personality registered for level %d\n",
7536 p->name, p->level);
7537 spin_lock(&pers_lock);
7538 list_add_tail(&p->list, &pers_list);
7539 spin_unlock(&pers_lock);
7540 return 0;
7541 }
7542 EXPORT_SYMBOL(register_md_personality);
7543
7544 int unregister_md_personality(struct md_personality *p)
7545 {
7546 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
7547 spin_lock(&pers_lock);
7548 list_del_init(&p->list);
7549 spin_unlock(&pers_lock);
7550 return 0;
7551 }
7552 EXPORT_SYMBOL(unregister_md_personality);
7553
7554 int register_md_cluster_operations(struct md_cluster_operations *ops,
7555 struct module *module)
7556 {
7557 int ret = 0;
7558 spin_lock(&pers_lock);
7559 if (md_cluster_ops != NULL)
7560 ret = -EALREADY;
7561 else {
7562 md_cluster_ops = ops;
7563 md_cluster_mod = module;
7564 }
7565 spin_unlock(&pers_lock);
7566 return ret;
7567 }
7568 EXPORT_SYMBOL(register_md_cluster_operations);
7569
7570 int unregister_md_cluster_operations(void)
7571 {
7572 spin_lock(&pers_lock);
7573 md_cluster_ops = NULL;
7574 spin_unlock(&pers_lock);
7575 return 0;
7576 }
7577 EXPORT_SYMBOL(unregister_md_cluster_operations);
7578
7579 int md_setup_cluster(struct mddev *mddev, int nodes)
7580 {
7581 if (!md_cluster_ops)
7582 request_module("md-cluster");
7583 spin_lock(&pers_lock);
7584 /* ensure module won't be unloaded */
7585 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
7586 pr_err("can't find md-cluster module or get it's reference.\n");
7587 spin_unlock(&pers_lock);
7588 return -ENOENT;
7589 }
7590 spin_unlock(&pers_lock);
7591
7592 return md_cluster_ops->join(mddev, nodes);
7593 }
7594
7595 void md_cluster_stop(struct mddev *mddev)
7596 {
7597 if (!md_cluster_ops)
7598 return;
7599 md_cluster_ops->leave(mddev);
7600 module_put(md_cluster_mod);
7601 }
7602
7603 static int is_mddev_idle(struct mddev *mddev, int init)
7604 {
7605 struct md_rdev *rdev;
7606 int idle;
7607 int curr_events;
7608
7609 idle = 1;
7610 rcu_read_lock();
7611 rdev_for_each_rcu(rdev, mddev) {
7612 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7613 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7614 (int)part_stat_read(&disk->part0, sectors[1]) -
7615 atomic_read(&disk->sync_io);
7616 /* sync IO will cause sync_io to increase before the disk_stats
7617 * as sync_io is counted when a request starts, and
7618 * disk_stats is counted when it completes.
7619 * So resync activity will cause curr_events to be smaller than
7620 * when there was no such activity.
7621 * non-sync IO will cause disk_stat to increase without
7622 * increasing sync_io so curr_events will (eventually)
7623 * be larger than it was before. Once it becomes
7624 * substantially larger, the test below will cause
7625 * the array to appear non-idle, and resync will slow
7626 * down.
7627 * If there is a lot of outstanding resync activity when
7628 * we set last_event to curr_events, then all that activity
7629 * completing might cause the array to appear non-idle
7630 * and resync will be slowed down even though there might
7631 * not have been non-resync activity. This will only
7632 * happen once though. 'last_events' will soon reflect
7633 * the state where there is little or no outstanding
7634 * resync requests, and further resync activity will
7635 * always make curr_events less than last_events.
7636 *
7637 */
7638 if (init || curr_events - rdev->last_events > 64) {
7639 rdev->last_events = curr_events;
7640 idle = 0;
7641 }
7642 }
7643 rcu_read_unlock();
7644 return idle;
7645 }
7646
7647 void md_done_sync(struct mddev *mddev, int blocks, int ok)
7648 {
7649 /* another "blocks" (512byte) blocks have been synced */
7650 atomic_sub(blocks, &mddev->recovery_active);
7651 wake_up(&mddev->recovery_wait);
7652 if (!ok) {
7653 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7654 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
7655 md_wakeup_thread(mddev->thread);
7656 // stop recovery, signal do_sync ....
7657 }
7658 }
7659 EXPORT_SYMBOL(md_done_sync);
7660
7661 /* md_write_start(mddev, bi)
7662 * If we need to update some array metadata (e.g. 'active' flag
7663 * in superblock) before writing, schedule a superblock update
7664 * and wait for it to complete.
7665 */
7666 void md_write_start(struct mddev *mddev, struct bio *bi)
7667 {
7668 int did_change = 0;
7669 if (bio_data_dir(bi) != WRITE)
7670 return;
7671
7672 BUG_ON(mddev->ro == 1);
7673 if (mddev->ro == 2) {
7674 /* need to switch to read/write */
7675 mddev->ro = 0;
7676 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7677 md_wakeup_thread(mddev->thread);
7678 md_wakeup_thread(mddev->sync_thread);
7679 did_change = 1;
7680 }
7681 atomic_inc(&mddev->writes_pending);
7682 if (mddev->safemode == 1)
7683 mddev->safemode = 0;
7684 if (mddev->in_sync) {
7685 spin_lock(&mddev->lock);
7686 if (mddev->in_sync) {
7687 mddev->in_sync = 0;
7688 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7689 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7690 md_wakeup_thread(mddev->thread);
7691 did_change = 1;
7692 }
7693 spin_unlock(&mddev->lock);
7694 }
7695 if (did_change)
7696 sysfs_notify_dirent_safe(mddev->sysfs_state);
7697 wait_event(mddev->sb_wait,
7698 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
7699 }
7700 EXPORT_SYMBOL(md_write_start);
7701
7702 void md_write_end(struct mddev *mddev)
7703 {
7704 if (atomic_dec_and_test(&mddev->writes_pending)) {
7705 if (mddev->safemode == 2)
7706 md_wakeup_thread(mddev->thread);
7707 else if (mddev->safemode_delay)
7708 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7709 }
7710 }
7711 EXPORT_SYMBOL(md_write_end);
7712
7713 /* md_allow_write(mddev)
7714 * Calling this ensures that the array is marked 'active' so that writes
7715 * may proceed without blocking. It is important to call this before
7716 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7717 * Must be called with mddev_lock held.
7718 *
7719 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7720 * is dropped, so return -EAGAIN after notifying userspace.
7721 */
7722 int md_allow_write(struct mddev *mddev)
7723 {
7724 if (!mddev->pers)
7725 return 0;
7726 if (mddev->ro)
7727 return 0;
7728 if (!mddev->pers->sync_request)
7729 return 0;
7730
7731 spin_lock(&mddev->lock);
7732 if (mddev->in_sync) {
7733 mddev->in_sync = 0;
7734 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7735 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7736 if (mddev->safemode_delay &&
7737 mddev->safemode == 0)
7738 mddev->safemode = 1;
7739 spin_unlock(&mddev->lock);
7740 md_update_sb(mddev, 0);
7741 sysfs_notify_dirent_safe(mddev->sysfs_state);
7742 } else
7743 spin_unlock(&mddev->lock);
7744
7745 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
7746 return -EAGAIN;
7747 else
7748 return 0;
7749 }
7750 EXPORT_SYMBOL_GPL(md_allow_write);
7751
7752 #define SYNC_MARKS 10
7753 #define SYNC_MARK_STEP (3*HZ)
7754 #define UPDATE_FREQUENCY (5*60*HZ)
7755 void md_do_sync(struct md_thread *thread)
7756 {
7757 struct mddev *mddev = thread->mddev;
7758 struct mddev *mddev2;
7759 unsigned int currspeed = 0,
7760 window;
7761 sector_t max_sectors,j, io_sectors, recovery_done;
7762 unsigned long mark[SYNC_MARKS];
7763 unsigned long update_time;
7764 sector_t mark_cnt[SYNC_MARKS];
7765 int last_mark,m;
7766 struct list_head *tmp;
7767 sector_t last_check;
7768 int skipped = 0;
7769 struct md_rdev *rdev;
7770 char *desc, *action = NULL;
7771 struct blk_plug plug;
7772 bool cluster_resync_finished = false;
7773
7774 /* just incase thread restarts... */
7775 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7776 return;
7777 if (mddev->ro) {/* never try to sync a read-only array */
7778 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7779 return;
7780 }
7781
7782 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7783 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
7784 desc = "data-check";
7785 action = "check";
7786 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7787 desc = "requested-resync";
7788 action = "repair";
7789 } else
7790 desc = "resync";
7791 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7792 desc = "reshape";
7793 else
7794 desc = "recovery";
7795
7796 mddev->last_sync_action = action ?: desc;
7797
7798 /* we overload curr_resync somewhat here.
7799 * 0 == not engaged in resync at all
7800 * 2 == checking that there is no conflict with another sync
7801 * 1 == like 2, but have yielded to allow conflicting resync to
7802 * commense
7803 * other == active in resync - this many blocks
7804 *
7805 * Before starting a resync we must have set curr_resync to
7806 * 2, and then checked that every "conflicting" array has curr_resync
7807 * less than ours. When we find one that is the same or higher
7808 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7809 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7810 * This will mean we have to start checking from the beginning again.
7811 *
7812 */
7813
7814 do {
7815 mddev->curr_resync = 2;
7816
7817 try_again:
7818 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7819 goto skip;
7820 for_each_mddev(mddev2, tmp) {
7821 if (mddev2 == mddev)
7822 continue;
7823 if (!mddev->parallel_resync
7824 && mddev2->curr_resync
7825 && match_mddev_units(mddev, mddev2)) {
7826 DEFINE_WAIT(wq);
7827 if (mddev < mddev2 && mddev->curr_resync == 2) {
7828 /* arbitrarily yield */
7829 mddev->curr_resync = 1;
7830 wake_up(&resync_wait);
7831 }
7832 if (mddev > mddev2 && mddev->curr_resync == 1)
7833 /* no need to wait here, we can wait the next
7834 * time 'round when curr_resync == 2
7835 */
7836 continue;
7837 /* We need to wait 'interruptible' so as not to
7838 * contribute to the load average, and not to
7839 * be caught by 'softlockup'
7840 */
7841 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7842 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7843 mddev2->curr_resync >= mddev->curr_resync) {
7844 printk(KERN_INFO "md: delaying %s of %s"
7845 " until %s has finished (they"
7846 " share one or more physical units)\n",
7847 desc, mdname(mddev), mdname(mddev2));
7848 mddev_put(mddev2);
7849 if (signal_pending(current))
7850 flush_signals(current);
7851 schedule();
7852 finish_wait(&resync_wait, &wq);
7853 goto try_again;
7854 }
7855 finish_wait(&resync_wait, &wq);
7856 }
7857 }
7858 } while (mddev->curr_resync < 2);
7859
7860 j = 0;
7861 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7862 /* resync follows the size requested by the personality,
7863 * which defaults to physical size, but can be virtual size
7864 */
7865 max_sectors = mddev->resync_max_sectors;
7866 atomic64_set(&mddev->resync_mismatches, 0);
7867 /* we don't use the checkpoint if there's a bitmap */
7868 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7869 j = mddev->resync_min;
7870 else if (!mddev->bitmap)
7871 j = mddev->recovery_cp;
7872
7873 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7874 max_sectors = mddev->resync_max_sectors;
7875 else {
7876 /* recovery follows the physical size of devices */
7877 max_sectors = mddev->dev_sectors;
7878 j = MaxSector;
7879 rcu_read_lock();
7880 rdev_for_each_rcu(rdev, mddev)
7881 if (rdev->raid_disk >= 0 &&
7882 !test_bit(Journal, &rdev->flags) &&
7883 !test_bit(Faulty, &rdev->flags) &&
7884 !test_bit(In_sync, &rdev->flags) &&
7885 rdev->recovery_offset < j)
7886 j = rdev->recovery_offset;
7887 rcu_read_unlock();
7888
7889 /* If there is a bitmap, we need to make sure all
7890 * writes that started before we added a spare
7891 * complete before we start doing a recovery.
7892 * Otherwise the write might complete and (via
7893 * bitmap_endwrite) set a bit in the bitmap after the
7894 * recovery has checked that bit and skipped that
7895 * region.
7896 */
7897 if (mddev->bitmap) {
7898 mddev->pers->quiesce(mddev, 1);
7899 mddev->pers->quiesce(mddev, 0);
7900 }
7901 }
7902
7903 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7904 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7905 " %d KB/sec/disk.\n", speed_min(mddev));
7906 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
7907 "(but not more than %d KB/sec) for %s.\n",
7908 speed_max(mddev), desc);
7909
7910 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
7911
7912 io_sectors = 0;
7913 for (m = 0; m < SYNC_MARKS; m++) {
7914 mark[m] = jiffies;
7915 mark_cnt[m] = io_sectors;
7916 }
7917 last_mark = 0;
7918 mddev->resync_mark = mark[last_mark];
7919 mddev->resync_mark_cnt = mark_cnt[last_mark];
7920
7921 /*
7922 * Tune reconstruction:
7923 */
7924 window = 32*(PAGE_SIZE/512);
7925 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7926 window/2, (unsigned long long)max_sectors/2);
7927
7928 atomic_set(&mddev->recovery_active, 0);
7929 last_check = 0;
7930
7931 if (j>2) {
7932 printk(KERN_INFO
7933 "md: resuming %s of %s from checkpoint.\n",
7934 desc, mdname(mddev));
7935 mddev->curr_resync = j;
7936 } else
7937 mddev->curr_resync = 3; /* no longer delayed */
7938 mddev->curr_resync_completed = j;
7939 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7940 md_new_event(mddev);
7941 update_time = jiffies;
7942
7943 blk_start_plug(&plug);
7944 while (j < max_sectors) {
7945 sector_t sectors;
7946
7947 skipped = 0;
7948
7949 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7950 ((mddev->curr_resync > mddev->curr_resync_completed &&
7951 (mddev->curr_resync - mddev->curr_resync_completed)
7952 > (max_sectors >> 4)) ||
7953 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7954 (j - mddev->curr_resync_completed)*2
7955 >= mddev->resync_max - mddev->curr_resync_completed ||
7956 mddev->curr_resync_completed > mddev->resync_max
7957 )) {
7958 /* time to update curr_resync_completed */
7959 wait_event(mddev->recovery_wait,
7960 atomic_read(&mddev->recovery_active) == 0);
7961 mddev->curr_resync_completed = j;
7962 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7963 j > mddev->recovery_cp)
7964 mddev->recovery_cp = j;
7965 update_time = jiffies;
7966 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7967 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7968 }
7969
7970 while (j >= mddev->resync_max &&
7971 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7972 /* As this condition is controlled by user-space,
7973 * we can block indefinitely, so use '_interruptible'
7974 * to avoid triggering warnings.
7975 */
7976 flush_signals(current); /* just in case */
7977 wait_event_interruptible(mddev->recovery_wait,
7978 mddev->resync_max > j
7979 || test_bit(MD_RECOVERY_INTR,
7980 &mddev->recovery));
7981 }
7982
7983 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7984 break;
7985
7986 sectors = mddev->pers->sync_request(mddev, j, &skipped);
7987 if (sectors == 0) {
7988 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7989 break;
7990 }
7991
7992 if (!skipped) { /* actual IO requested */
7993 io_sectors += sectors;
7994 atomic_add(sectors, &mddev->recovery_active);
7995 }
7996
7997 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7998 break;
7999
8000 j += sectors;
8001 if (j > max_sectors)
8002 /* when skipping, extra large numbers can be returned. */
8003 j = max_sectors;
8004 if (j > 2)
8005 mddev->curr_resync = j;
8006 mddev->curr_mark_cnt = io_sectors;
8007 if (last_check == 0)
8008 /* this is the earliest that rebuild will be
8009 * visible in /proc/mdstat
8010 */
8011 md_new_event(mddev);
8012
8013 if (last_check + window > io_sectors || j == max_sectors)
8014 continue;
8015
8016 last_check = io_sectors;
8017 repeat:
8018 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8019 /* step marks */
8020 int next = (last_mark+1) % SYNC_MARKS;
8021
8022 mddev->resync_mark = mark[next];
8023 mddev->resync_mark_cnt = mark_cnt[next];
8024 mark[next] = jiffies;
8025 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
8026 last_mark = next;
8027 }
8028
8029 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8030 break;
8031
8032 /*
8033 * this loop exits only if either when we are slower than
8034 * the 'hard' speed limit, or the system was IO-idle for
8035 * a jiffy.
8036 * the system might be non-idle CPU-wise, but we only care
8037 * about not overloading the IO subsystem. (things like an
8038 * e2fsck being done on the RAID array should execute fast)
8039 */
8040 cond_resched();
8041
8042 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8043 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
8044 /((jiffies-mddev->resync_mark)/HZ +1) +1;
8045
8046 if (currspeed > speed_min(mddev)) {
8047 if (currspeed > speed_max(mddev)) {
8048 msleep(500);
8049 goto repeat;
8050 }
8051 if (!is_mddev_idle(mddev, 0)) {
8052 /*
8053 * Give other IO more of a chance.
8054 * The faster the devices, the less we wait.
8055 */
8056 wait_event(mddev->recovery_wait,
8057 !atomic_read(&mddev->recovery_active));
8058 }
8059 }
8060 }
8061 printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc,
8062 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
8063 ? "interrupted" : "done");
8064 /*
8065 * this also signals 'finished resyncing' to md_stop
8066 */
8067 blk_finish_plug(&plug);
8068 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
8069
8070 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8071 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8072 mddev->curr_resync > 2) {
8073 mddev->curr_resync_completed = mddev->curr_resync;
8074 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8075 }
8076 /* tell personality and other nodes that we are finished */
8077 if (mddev_is_clustered(mddev)) {
8078 md_cluster_ops->resync_finish(mddev);
8079 cluster_resync_finished = true;
8080 }
8081 mddev->pers->sync_request(mddev, max_sectors, &skipped);
8082
8083 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
8084 mddev->curr_resync > 2) {
8085 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8086 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8087 if (mddev->curr_resync >= mddev->recovery_cp) {
8088 printk(KERN_INFO
8089 "md: checkpointing %s of %s.\n",
8090 desc, mdname(mddev));
8091 if (test_bit(MD_RECOVERY_ERROR,
8092 &mddev->recovery))
8093 mddev->recovery_cp =
8094 mddev->curr_resync_completed;
8095 else
8096 mddev->recovery_cp =
8097 mddev->curr_resync;
8098 }
8099 } else
8100 mddev->recovery_cp = MaxSector;
8101 } else {
8102 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8103 mddev->curr_resync = MaxSector;
8104 rcu_read_lock();
8105 rdev_for_each_rcu(rdev, mddev)
8106 if (rdev->raid_disk >= 0 &&
8107 mddev->delta_disks >= 0 &&
8108 !test_bit(Journal, &rdev->flags) &&
8109 !test_bit(Faulty, &rdev->flags) &&
8110 !test_bit(In_sync, &rdev->flags) &&
8111 rdev->recovery_offset < mddev->curr_resync)
8112 rdev->recovery_offset = mddev->curr_resync;
8113 rcu_read_unlock();
8114 }
8115 }
8116 skip:
8117 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8118
8119 if (mddev_is_clustered(mddev) &&
8120 test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8121 !cluster_resync_finished)
8122 md_cluster_ops->resync_finish(mddev);
8123
8124 spin_lock(&mddev->lock);
8125 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8126 /* We completed so min/max setting can be forgotten if used. */
8127 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8128 mddev->resync_min = 0;
8129 mddev->resync_max = MaxSector;
8130 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8131 mddev->resync_min = mddev->curr_resync_completed;
8132 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
8133 mddev->curr_resync = 0;
8134 spin_unlock(&mddev->lock);
8135
8136 wake_up(&resync_wait);
8137 md_wakeup_thread(mddev->thread);
8138 return;
8139 }
8140 EXPORT_SYMBOL_GPL(md_do_sync);
8141
8142 static int remove_and_add_spares(struct mddev *mddev,
8143 struct md_rdev *this)
8144 {
8145 struct md_rdev *rdev;
8146 int spares = 0;
8147 int removed = 0;
8148
8149 rdev_for_each(rdev, mddev)
8150 if ((this == NULL || rdev == this) &&
8151 rdev->raid_disk >= 0 &&
8152 !test_bit(Blocked, &rdev->flags) &&
8153 (test_bit(Faulty, &rdev->flags) ||
8154 (!test_bit(In_sync, &rdev->flags) &&
8155 !test_bit(Journal, &rdev->flags))) &&
8156 atomic_read(&rdev->nr_pending)==0) {
8157 if (mddev->pers->hot_remove_disk(
8158 mddev, rdev) == 0) {
8159 sysfs_unlink_rdev(mddev, rdev);
8160 rdev->saved_raid_disk = rdev->raid_disk;
8161 rdev->raid_disk = -1;
8162 removed++;
8163 }
8164 }
8165 if (removed && mddev->kobj.sd)
8166 sysfs_notify(&mddev->kobj, NULL, "degraded");
8167
8168 if (this && removed)
8169 goto no_add;
8170
8171 rdev_for_each(rdev, mddev) {
8172 if (this && this != rdev)
8173 continue;
8174 if (test_bit(Candidate, &rdev->flags))
8175 continue;
8176 if (rdev->raid_disk >= 0 &&
8177 !test_bit(In_sync, &rdev->flags) &&
8178 !test_bit(Journal, &rdev->flags) &&
8179 !test_bit(Faulty, &rdev->flags))
8180 spares++;
8181 if (rdev->raid_disk >= 0)
8182 continue;
8183 if (test_bit(Faulty, &rdev->flags))
8184 continue;
8185 if (test_bit(Journal, &rdev->flags))
8186 continue;
8187 if (mddev->ro &&
8188 ! (rdev->saved_raid_disk >= 0 &&
8189 !test_bit(Bitmap_sync, &rdev->flags)))
8190 continue;
8191
8192 rdev->recovery_offset = 0;
8193 if (mddev->pers->
8194 hot_add_disk(mddev, rdev) == 0) {
8195 if (sysfs_link_rdev(mddev, rdev))
8196 /* failure here is OK */;
8197 spares++;
8198 md_new_event(mddev);
8199 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8200 }
8201 }
8202 no_add:
8203 if (removed)
8204 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8205 return spares;
8206 }
8207
8208 static void md_start_sync(struct work_struct *ws)
8209 {
8210 struct mddev *mddev = container_of(ws, struct mddev, del_work);
8211 int ret = 0;
8212
8213 if (mddev_is_clustered(mddev)) {
8214 ret = md_cluster_ops->resync_start(mddev);
8215 if (ret) {
8216 mddev->sync_thread = NULL;
8217 goto out;
8218 }
8219 }
8220
8221 mddev->sync_thread = md_register_thread(md_do_sync,
8222 mddev,
8223 "resync");
8224 out:
8225 if (!mddev->sync_thread) {
8226 if (!(mddev_is_clustered(mddev) && ret == -EAGAIN))
8227 printk(KERN_ERR "%s: could not start resync"
8228 " thread...\n",
8229 mdname(mddev));
8230 /* leave the spares where they are, it shouldn't hurt */
8231 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8232 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8233 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8234 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8235 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8236 wake_up(&resync_wait);
8237 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8238 &mddev->recovery))
8239 if (mddev->sysfs_action)
8240 sysfs_notify_dirent_safe(mddev->sysfs_action);
8241 } else
8242 md_wakeup_thread(mddev->sync_thread);
8243 sysfs_notify_dirent_safe(mddev->sysfs_action);
8244 md_new_event(mddev);
8245 }
8246
8247 /*
8248 * This routine is regularly called by all per-raid-array threads to
8249 * deal with generic issues like resync and super-block update.
8250 * Raid personalities that don't have a thread (linear/raid0) do not
8251 * need this as they never do any recovery or update the superblock.
8252 *
8253 * It does not do any resync itself, but rather "forks" off other threads
8254 * to do that as needed.
8255 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8256 * "->recovery" and create a thread at ->sync_thread.
8257 * When the thread finishes it sets MD_RECOVERY_DONE
8258 * and wakeups up this thread which will reap the thread and finish up.
8259 * This thread also removes any faulty devices (with nr_pending == 0).
8260 *
8261 * The overall approach is:
8262 * 1/ if the superblock needs updating, update it.
8263 * 2/ If a recovery thread is running, don't do anything else.
8264 * 3/ If recovery has finished, clean up, possibly marking spares active.
8265 * 4/ If there are any faulty devices, remove them.
8266 * 5/ If array is degraded, try to add spares devices
8267 * 6/ If array has spares or is not in-sync, start a resync thread.
8268 */
8269 void md_check_recovery(struct mddev *mddev)
8270 {
8271 if (mddev->suspended)
8272 return;
8273
8274 if (mddev->bitmap)
8275 bitmap_daemon_work(mddev);
8276
8277 if (signal_pending(current)) {
8278 if (mddev->pers->sync_request && !mddev->external) {
8279 printk(KERN_INFO "md: %s in immediate safe mode\n",
8280 mdname(mddev));
8281 mddev->safemode = 2;
8282 }
8283 flush_signals(current);
8284 }
8285
8286 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
8287 return;
8288 if ( ! (
8289 (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) ||
8290 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8291 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8292 (mddev->external == 0 && mddev->safemode == 1) ||
8293 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
8294 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
8295 ))
8296 return;
8297
8298 if (mddev_trylock(mddev)) {
8299 int spares = 0;
8300
8301 if (mddev->ro) {
8302 struct md_rdev *rdev;
8303 if (!mddev->external && mddev->in_sync)
8304 /* 'Blocked' flag not needed as failed devices
8305 * will be recorded if array switched to read/write.
8306 * Leaving it set will prevent the device
8307 * from being removed.
8308 */
8309 rdev_for_each(rdev, mddev)
8310 clear_bit(Blocked, &rdev->flags);
8311 /* On a read-only array we can:
8312 * - remove failed devices
8313 * - add already-in_sync devices if the array itself
8314 * is in-sync.
8315 * As we only add devices that are already in-sync,
8316 * we can activate the spares immediately.
8317 */
8318 remove_and_add_spares(mddev, NULL);
8319 /* There is no thread, but we need to call
8320 * ->spare_active and clear saved_raid_disk
8321 */
8322 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8323 md_reap_sync_thread(mddev);
8324 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8325 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8326 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
8327 goto unlock;
8328 }
8329
8330 if (!mddev->external) {
8331 int did_change = 0;
8332 spin_lock(&mddev->lock);
8333 if (mddev->safemode &&
8334 !atomic_read(&mddev->writes_pending) &&
8335 !mddev->in_sync &&
8336 mddev->recovery_cp == MaxSector) {
8337 mddev->in_sync = 1;
8338 did_change = 1;
8339 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
8340 }
8341 if (mddev->safemode == 1)
8342 mddev->safemode = 0;
8343 spin_unlock(&mddev->lock);
8344 if (did_change)
8345 sysfs_notify_dirent_safe(mddev->sysfs_state);
8346 }
8347
8348 if (mddev->flags & MD_UPDATE_SB_FLAGS)
8349 md_update_sb(mddev, 0);
8350
8351 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
8352 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
8353 /* resync/recovery still happening */
8354 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8355 goto unlock;
8356 }
8357 if (mddev->sync_thread) {
8358 md_reap_sync_thread(mddev);
8359 goto unlock;
8360 }
8361 /* Set RUNNING before clearing NEEDED to avoid
8362 * any transients in the value of "sync_action".
8363 */
8364 mddev->curr_resync_completed = 0;
8365 spin_lock(&mddev->lock);
8366 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8367 spin_unlock(&mddev->lock);
8368 /* Clear some bits that don't mean anything, but
8369 * might be left set
8370 */
8371 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
8372 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8373
8374 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8375 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
8376 goto not_running;
8377 /* no recovery is running.
8378 * remove any failed drives, then
8379 * add spares if possible.
8380 * Spares are also removed and re-added, to allow
8381 * the personality to fail the re-add.
8382 */
8383
8384 if (mddev->reshape_position != MaxSector) {
8385 if (mddev->pers->check_reshape == NULL ||
8386 mddev->pers->check_reshape(mddev) != 0)
8387 /* Cannot proceed */
8388 goto not_running;
8389 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8390 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8391 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
8392 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8393 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8394 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8395 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8396 } else if (mddev->recovery_cp < MaxSector) {
8397 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8398 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8399 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
8400 /* nothing to be done ... */
8401 goto not_running;
8402
8403 if (mddev->pers->sync_request) {
8404 if (spares) {
8405 /* We are adding a device or devices to an array
8406 * which has the bitmap stored on all devices.
8407 * So make sure all bitmap pages get written
8408 */
8409 bitmap_write_all(mddev->bitmap);
8410 }
8411 INIT_WORK(&mddev->del_work, md_start_sync);
8412 queue_work(md_misc_wq, &mddev->del_work);
8413 goto unlock;
8414 }
8415 not_running:
8416 if (!mddev->sync_thread) {
8417 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8418 wake_up(&resync_wait);
8419 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8420 &mddev->recovery))
8421 if (mddev->sysfs_action)
8422 sysfs_notify_dirent_safe(mddev->sysfs_action);
8423 }
8424 unlock:
8425 wake_up(&mddev->sb_wait);
8426 mddev_unlock(mddev);
8427 }
8428 }
8429 EXPORT_SYMBOL(md_check_recovery);
8430
8431 void md_reap_sync_thread(struct mddev *mddev)
8432 {
8433 struct md_rdev *rdev;
8434
8435 /* resync has finished, collect result */
8436 md_unregister_thread(&mddev->sync_thread);
8437 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8438 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8439 /* success...*/
8440 /* activate any spares */
8441 if (mddev->pers->spare_active(mddev)) {
8442 sysfs_notify(&mddev->kobj, NULL,
8443 "degraded");
8444 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8445 }
8446 }
8447 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8448 mddev->pers->finish_reshape)
8449 mddev->pers->finish_reshape(mddev);
8450
8451 /* If array is no-longer degraded, then any saved_raid_disk
8452 * information must be scrapped.
8453 */
8454 if (!mddev->degraded)
8455 rdev_for_each(rdev, mddev)
8456 rdev->saved_raid_disk = -1;
8457
8458 md_update_sb(mddev, 1);
8459 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8460 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8461 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8462 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8463 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8464 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8465 wake_up(&resync_wait);
8466 /* flag recovery needed just to double check */
8467 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8468 sysfs_notify_dirent_safe(mddev->sysfs_action);
8469 md_new_event(mddev);
8470 if (mddev->event_work.func)
8471 queue_work(md_misc_wq, &mddev->event_work);
8472 }
8473 EXPORT_SYMBOL(md_reap_sync_thread);
8474
8475 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
8476 {
8477 sysfs_notify_dirent_safe(rdev->sysfs_state);
8478 wait_event_timeout(rdev->blocked_wait,
8479 !test_bit(Blocked, &rdev->flags) &&
8480 !test_bit(BlockedBadBlocks, &rdev->flags),
8481 msecs_to_jiffies(5000));
8482 rdev_dec_pending(rdev, mddev);
8483 }
8484 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8485
8486 void md_finish_reshape(struct mddev *mddev)
8487 {
8488 /* called be personality module when reshape completes. */
8489 struct md_rdev *rdev;
8490
8491 rdev_for_each(rdev, mddev) {
8492 if (rdev->data_offset > rdev->new_data_offset)
8493 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8494 else
8495 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8496 rdev->data_offset = rdev->new_data_offset;
8497 }
8498 }
8499 EXPORT_SYMBOL(md_finish_reshape);
8500
8501 /* Bad block management.
8502 * We can record which blocks on each device are 'bad' and so just
8503 * fail those blocks, or that stripe, rather than the whole device.
8504 * Entries in the bad-block table are 64bits wide. This comprises:
8505 * Length of bad-range, in sectors: 0-511 for lengths 1-512
8506 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
8507 * A 'shift' can be set so that larger blocks are tracked and
8508 * consequently larger devices can be covered.
8509 * 'Acknowledged' flag - 1 bit. - the most significant bit.
8510 *
8511 * Locking of the bad-block table uses a seqlock so md_is_badblock
8512 * might need to retry if it is very unlucky.
8513 * We will sometimes want to check for bad blocks in a bi_end_io function,
8514 * so we use the write_seqlock_irq variant.
8515 *
8516 * When looking for a bad block we specify a range and want to
8517 * know if any block in the range is bad. So we binary-search
8518 * to the last range that starts at-or-before the given endpoint,
8519 * (or "before the sector after the target range")
8520 * then see if it ends after the given start.
8521 * We return
8522 * 0 if there are no known bad blocks in the range
8523 * 1 if there are known bad block which are all acknowledged
8524 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
8525 * plus the start/length of the first bad section we overlap.
8526 */
8527 int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
8528 sector_t *first_bad, int *bad_sectors)
8529 {
8530 int hi;
8531 int lo;
8532 u64 *p = bb->page;
8533 int rv;
8534 sector_t target = s + sectors;
8535 unsigned seq;
8536
8537 if (bb->shift > 0) {
8538 /* round the start down, and the end up */
8539 s >>= bb->shift;
8540 target += (1<<bb->shift) - 1;
8541 target >>= bb->shift;
8542 sectors = target - s;
8543 }
8544 /* 'target' is now the first block after the bad range */
8545
8546 retry:
8547 seq = read_seqbegin(&bb->lock);
8548 lo = 0;
8549 rv = 0;
8550 hi = bb->count;
8551
8552 /* Binary search between lo and hi for 'target'
8553 * i.e. for the last range that starts before 'target'
8554 */
8555 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8556 * are known not to be the last range before target.
8557 * VARIANT: hi-lo is the number of possible
8558 * ranges, and decreases until it reaches 1
8559 */
8560 while (hi - lo > 1) {
8561 int mid = (lo + hi) / 2;
8562 sector_t a = BB_OFFSET(p[mid]);
8563 if (a < target)
8564 /* This could still be the one, earlier ranges
8565 * could not. */
8566 lo = mid;
8567 else
8568 /* This and later ranges are definitely out. */
8569 hi = mid;
8570 }
8571 /* 'lo' might be the last that started before target, but 'hi' isn't */
8572 if (hi > lo) {
8573 /* need to check all range that end after 's' to see if
8574 * any are unacknowledged.
8575 */
8576 while (lo >= 0 &&
8577 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8578 if (BB_OFFSET(p[lo]) < target) {
8579 /* starts before the end, and finishes after
8580 * the start, so they must overlap
8581 */
8582 if (rv != -1 && BB_ACK(p[lo]))
8583 rv = 1;
8584 else
8585 rv = -1;
8586 *first_bad = BB_OFFSET(p[lo]);
8587 *bad_sectors = BB_LEN(p[lo]);
8588 }
8589 lo--;
8590 }
8591 }
8592
8593 if (read_seqretry(&bb->lock, seq))
8594 goto retry;
8595
8596 return rv;
8597 }
8598 EXPORT_SYMBOL_GPL(md_is_badblock);
8599
8600 /*
8601 * Add a range of bad blocks to the table.
8602 * This might extend the table, or might contract it
8603 * if two adjacent ranges can be merged.
8604 * We binary-search to find the 'insertion' point, then
8605 * decide how best to handle it.
8606 */
8607 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8608 int acknowledged)
8609 {
8610 u64 *p;
8611 int lo, hi;
8612 int rv = 1;
8613 unsigned long flags;
8614
8615 if (bb->shift < 0)
8616 /* badblocks are disabled */
8617 return 0;
8618
8619 if (bb->shift) {
8620 /* round the start down, and the end up */
8621 sector_t next = s + sectors;
8622 s >>= bb->shift;
8623 next += (1<<bb->shift) - 1;
8624 next >>= bb->shift;
8625 sectors = next - s;
8626 }
8627
8628 write_seqlock_irqsave(&bb->lock, flags);
8629
8630 p = bb->page;
8631 lo = 0;
8632 hi = bb->count;
8633 /* Find the last range that starts at-or-before 's' */
8634 while (hi - lo > 1) {
8635 int mid = (lo + hi) / 2;
8636 sector_t a = BB_OFFSET(p[mid]);
8637 if (a <= s)
8638 lo = mid;
8639 else
8640 hi = mid;
8641 }
8642 if (hi > lo && BB_OFFSET(p[lo]) > s)
8643 hi = lo;
8644
8645 if (hi > lo) {
8646 /* we found a range that might merge with the start
8647 * of our new range
8648 */
8649 sector_t a = BB_OFFSET(p[lo]);
8650 sector_t e = a + BB_LEN(p[lo]);
8651 int ack = BB_ACK(p[lo]);
8652 if (e >= s) {
8653 /* Yes, we can merge with a previous range */
8654 if (s == a && s + sectors >= e)
8655 /* new range covers old */
8656 ack = acknowledged;
8657 else
8658 ack = ack && acknowledged;
8659
8660 if (e < s + sectors)
8661 e = s + sectors;
8662 if (e - a <= BB_MAX_LEN) {
8663 p[lo] = BB_MAKE(a, e-a, ack);
8664 s = e;
8665 } else {
8666 /* does not all fit in one range,
8667 * make p[lo] maximal
8668 */
8669 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8670 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8671 s = a + BB_MAX_LEN;
8672 }
8673 sectors = e - s;
8674 }
8675 }
8676 if (sectors && hi < bb->count) {
8677 /* 'hi' points to the first range that starts after 's'.
8678 * Maybe we can merge with the start of that range */
8679 sector_t a = BB_OFFSET(p[hi]);
8680 sector_t e = a + BB_LEN(p[hi]);
8681 int ack = BB_ACK(p[hi]);
8682 if (a <= s + sectors) {
8683 /* merging is possible */
8684 if (e <= s + sectors) {
8685 /* full overlap */
8686 e = s + sectors;
8687 ack = acknowledged;
8688 } else
8689 ack = ack && acknowledged;
8690
8691 a = s;
8692 if (e - a <= BB_MAX_LEN) {
8693 p[hi] = BB_MAKE(a, e-a, ack);
8694 s = e;
8695 } else {
8696 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8697 s = a + BB_MAX_LEN;
8698 }
8699 sectors = e - s;
8700 lo = hi;
8701 hi++;
8702 }
8703 }
8704 if (sectors == 0 && hi < bb->count) {
8705 /* we might be able to combine lo and hi */
8706 /* Note: 's' is at the end of 'lo' */
8707 sector_t a = BB_OFFSET(p[hi]);
8708 int lolen = BB_LEN(p[lo]);
8709 int hilen = BB_LEN(p[hi]);
8710 int newlen = lolen + hilen - (s - a);
8711 if (s >= a && newlen < BB_MAX_LEN) {
8712 /* yes, we can combine them */
8713 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8714 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8715 memmove(p + hi, p + hi + 1,
8716 (bb->count - hi - 1) * 8);
8717 bb->count--;
8718 }
8719 }
8720 while (sectors) {
8721 /* didn't merge (it all).
8722 * Need to add a range just before 'hi' */
8723 if (bb->count >= MD_MAX_BADBLOCKS) {
8724 /* No room for more */
8725 rv = 0;
8726 break;
8727 } else {
8728 int this_sectors = sectors;
8729 memmove(p + hi + 1, p + hi,
8730 (bb->count - hi) * 8);
8731 bb->count++;
8732
8733 if (this_sectors > BB_MAX_LEN)
8734 this_sectors = BB_MAX_LEN;
8735 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8736 sectors -= this_sectors;
8737 s += this_sectors;
8738 }
8739 }
8740
8741 bb->changed = 1;
8742 if (!acknowledged)
8743 bb->unacked_exist = 1;
8744 write_sequnlock_irqrestore(&bb->lock, flags);
8745
8746 return rv;
8747 }
8748
8749 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8750 int is_new)
8751 {
8752 int rv;
8753 if (is_new)
8754 s += rdev->new_data_offset;
8755 else
8756 s += rdev->data_offset;
8757 rv = md_set_badblocks(&rdev->badblocks,
8758 s, sectors, 0);
8759 if (rv) {
8760 /* Make sure they get written out promptly */
8761 sysfs_notify_dirent_safe(rdev->sysfs_state);
8762 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8763 set_bit(MD_CHANGE_PENDING, &rdev->mddev->flags);
8764 md_wakeup_thread(rdev->mddev->thread);
8765 }
8766 return rv;
8767 }
8768 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8769
8770 /*
8771 * Remove a range of bad blocks from the table.
8772 * This may involve extending the table if we spilt a region,
8773 * but it must not fail. So if the table becomes full, we just
8774 * drop the remove request.
8775 */
8776 static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8777 {
8778 u64 *p;
8779 int lo, hi;
8780 sector_t target = s + sectors;
8781 int rv = 0;
8782
8783 if (bb->shift > 0) {
8784 /* When clearing we round the start up and the end down.
8785 * This should not matter as the shift should align with
8786 * the block size and no rounding should ever be needed.
8787 * However it is better the think a block is bad when it
8788 * isn't than to think a block is not bad when it is.
8789 */
8790 s += (1<<bb->shift) - 1;
8791 s >>= bb->shift;
8792 target >>= bb->shift;
8793 sectors = target - s;
8794 }
8795
8796 write_seqlock_irq(&bb->lock);
8797
8798 p = bb->page;
8799 lo = 0;
8800 hi = bb->count;
8801 /* Find the last range that starts before 'target' */
8802 while (hi - lo > 1) {
8803 int mid = (lo + hi) / 2;
8804 sector_t a = BB_OFFSET(p[mid]);
8805 if (a < target)
8806 lo = mid;
8807 else
8808 hi = mid;
8809 }
8810 if (hi > lo) {
8811 /* p[lo] is the last range that could overlap the
8812 * current range. Earlier ranges could also overlap,
8813 * but only this one can overlap the end of the range.
8814 */
8815 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8816 /* Partial overlap, leave the tail of this range */
8817 int ack = BB_ACK(p[lo]);
8818 sector_t a = BB_OFFSET(p[lo]);
8819 sector_t end = a + BB_LEN(p[lo]);
8820
8821 if (a < s) {
8822 /* we need to split this range */
8823 if (bb->count >= MD_MAX_BADBLOCKS) {
8824 rv = -ENOSPC;
8825 goto out;
8826 }
8827 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8828 bb->count++;
8829 p[lo] = BB_MAKE(a, s-a, ack);
8830 lo++;
8831 }
8832 p[lo] = BB_MAKE(target, end - target, ack);
8833 /* there is no longer an overlap */
8834 hi = lo;
8835 lo--;
8836 }
8837 while (lo >= 0 &&
8838 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8839 /* This range does overlap */
8840 if (BB_OFFSET(p[lo]) < s) {
8841 /* Keep the early parts of this range. */
8842 int ack = BB_ACK(p[lo]);
8843 sector_t start = BB_OFFSET(p[lo]);
8844 p[lo] = BB_MAKE(start, s - start, ack);
8845 /* now low doesn't overlap, so.. */
8846 break;
8847 }
8848 lo--;
8849 }
8850 /* 'lo' is strictly before, 'hi' is strictly after,
8851 * anything between needs to be discarded
8852 */
8853 if (hi - lo > 1) {
8854 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8855 bb->count -= (hi - lo - 1);
8856 }
8857 }
8858
8859 bb->changed = 1;
8860 out:
8861 write_sequnlock_irq(&bb->lock);
8862 return rv;
8863 }
8864
8865 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8866 int is_new)
8867 {
8868 if (is_new)
8869 s += rdev->new_data_offset;
8870 else
8871 s += rdev->data_offset;
8872 return md_clear_badblocks(&rdev->badblocks,
8873 s, sectors);
8874 }
8875 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8876
8877 /*
8878 * Acknowledge all bad blocks in a list.
8879 * This only succeeds if ->changed is clear. It is used by
8880 * in-kernel metadata updates
8881 */
8882 void md_ack_all_badblocks(struct badblocks *bb)
8883 {
8884 if (bb->page == NULL || bb->changed)
8885 /* no point even trying */
8886 return;
8887 write_seqlock_irq(&bb->lock);
8888
8889 if (bb->changed == 0 && bb->unacked_exist) {
8890 u64 *p = bb->page;
8891 int i;
8892 for (i = 0; i < bb->count ; i++) {
8893 if (!BB_ACK(p[i])) {
8894 sector_t start = BB_OFFSET(p[i]);
8895 int len = BB_LEN(p[i]);
8896 p[i] = BB_MAKE(start, len, 1);
8897 }
8898 }
8899 bb->unacked_exist = 0;
8900 }
8901 write_sequnlock_irq(&bb->lock);
8902 }
8903 EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8904
8905 /* sysfs access to bad-blocks list.
8906 * We present two files.
8907 * 'bad-blocks' lists sector numbers and lengths of ranges that
8908 * are recorded as bad. The list is truncated to fit within
8909 * the one-page limit of sysfs.
8910 * Writing "sector length" to this file adds an acknowledged
8911 * bad block list.
8912 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8913 * been acknowledged. Writing to this file adds bad blocks
8914 * without acknowledging them. This is largely for testing.
8915 */
8916
8917 static ssize_t
8918 badblocks_show(struct badblocks *bb, char *page, int unack)
8919 {
8920 size_t len;
8921 int i;
8922 u64 *p = bb->page;
8923 unsigned seq;
8924
8925 if (bb->shift < 0)
8926 return 0;
8927
8928 retry:
8929 seq = read_seqbegin(&bb->lock);
8930
8931 len = 0;
8932 i = 0;
8933
8934 while (len < PAGE_SIZE && i < bb->count) {
8935 sector_t s = BB_OFFSET(p[i]);
8936 unsigned int length = BB_LEN(p[i]);
8937 int ack = BB_ACK(p[i]);
8938 i++;
8939
8940 if (unack && ack)
8941 continue;
8942
8943 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8944 (unsigned long long)s << bb->shift,
8945 length << bb->shift);
8946 }
8947 if (unack && len == 0)
8948 bb->unacked_exist = 0;
8949
8950 if (read_seqretry(&bb->lock, seq))
8951 goto retry;
8952
8953 return len;
8954 }
8955
8956 #define DO_DEBUG 1
8957
8958 static ssize_t
8959 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8960 {
8961 unsigned long long sector;
8962 int length;
8963 char newline;
8964 #ifdef DO_DEBUG
8965 /* Allow clearing via sysfs *only* for testing/debugging.
8966 * Normally only a successful write may clear a badblock
8967 */
8968 int clear = 0;
8969 if (page[0] == '-') {
8970 clear = 1;
8971 page++;
8972 }
8973 #endif /* DO_DEBUG */
8974
8975 switch (sscanf(page, "%llu %d%c", &sector, &length, &newline)) {
8976 case 3:
8977 if (newline != '\n')
8978 return -EINVAL;
8979 case 2:
8980 if (length <= 0)
8981 return -EINVAL;
8982 break;
8983 default:
8984 return -EINVAL;
8985 }
8986
8987 #ifdef DO_DEBUG
8988 if (clear) {
8989 md_clear_badblocks(bb, sector, length);
8990 return len;
8991 }
8992 #endif /* DO_DEBUG */
8993 if (md_set_badblocks(bb, sector, length, !unack))
8994 return len;
8995 else
8996 return -ENOSPC;
8997 }
8998
8999 static int md_notify_reboot(struct notifier_block *this,
9000 unsigned long code, void *x)
9001 {
9002 struct list_head *tmp;
9003 struct mddev *mddev;
9004 int need_delay = 0;
9005
9006 for_each_mddev(mddev, tmp) {
9007 if (mddev_trylock(mddev)) {
9008 if (mddev->pers)
9009 __md_stop_writes(mddev);
9010 if (mddev->persistent)
9011 mddev->safemode = 2;
9012 mddev_unlock(mddev);
9013 }
9014 need_delay = 1;
9015 }
9016 /*
9017 * certain more exotic SCSI devices are known to be
9018 * volatile wrt too early system reboots. While the
9019 * right place to handle this issue is the given
9020 * driver, we do want to have a safe RAID driver ...
9021 */
9022 if (need_delay)
9023 mdelay(1000*1);
9024
9025 return NOTIFY_DONE;
9026 }
9027
9028 static struct notifier_block md_notifier = {
9029 .notifier_call = md_notify_reboot,
9030 .next = NULL,
9031 .priority = INT_MAX, /* before any real devices */
9032 };
9033
9034 static void md_geninit(void)
9035 {
9036 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
9037
9038 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
9039 }
9040
9041 static int __init md_init(void)
9042 {
9043 int ret = -ENOMEM;
9044
9045 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
9046 if (!md_wq)
9047 goto err_wq;
9048
9049 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
9050 if (!md_misc_wq)
9051 goto err_misc_wq;
9052
9053 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
9054 goto err_md;
9055
9056 if ((ret = register_blkdev(0, "mdp")) < 0)
9057 goto err_mdp;
9058 mdp_major = ret;
9059
9060 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
9061 md_probe, NULL, NULL);
9062 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
9063 md_probe, NULL, NULL);
9064
9065 register_reboot_notifier(&md_notifier);
9066 raid_table_header = register_sysctl_table(raid_root_table);
9067
9068 md_geninit();
9069 return 0;
9070
9071 err_mdp:
9072 unregister_blkdev(MD_MAJOR, "md");
9073 err_md:
9074 destroy_workqueue(md_misc_wq);
9075 err_misc_wq:
9076 destroy_workqueue(md_wq);
9077 err_wq:
9078 return ret;
9079 }
9080
9081 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
9082 {
9083 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
9084 struct md_rdev *rdev2;
9085 int role, ret;
9086 char b[BDEVNAME_SIZE];
9087
9088 /* Check for change of roles in the active devices */
9089 rdev_for_each(rdev2, mddev) {
9090 if (test_bit(Faulty, &rdev2->flags))
9091 continue;
9092
9093 /* Check if the roles changed */
9094 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
9095
9096 if (test_bit(Candidate, &rdev2->flags)) {
9097 if (role == 0xfffe) {
9098 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b));
9099 md_kick_rdev_from_array(rdev2);
9100 continue;
9101 }
9102 else
9103 clear_bit(Candidate, &rdev2->flags);
9104 }
9105
9106 if (role != rdev2->raid_disk) {
9107 /* got activated */
9108 if (rdev2->raid_disk == -1 && role != 0xffff) {
9109 rdev2->saved_raid_disk = role;
9110 ret = remove_and_add_spares(mddev, rdev2);
9111 pr_info("Activated spare: %s\n",
9112 bdevname(rdev2->bdev,b));
9113 continue;
9114 }
9115 /* device faulty
9116 * We just want to do the minimum to mark the disk
9117 * as faulty. The recovery is performed by the
9118 * one who initiated the error.
9119 */
9120 if ((role == 0xfffe) || (role == 0xfffd)) {
9121 md_error(mddev, rdev2);
9122 clear_bit(Blocked, &rdev2->flags);
9123 }
9124 }
9125 }
9126
9127 if (mddev->raid_disks != le32_to_cpu(sb->raid_disks))
9128 update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
9129
9130 /* Finally set the event to be up to date */
9131 mddev->events = le64_to_cpu(sb->events);
9132 }
9133
9134 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
9135 {
9136 int err;
9137 struct page *swapout = rdev->sb_page;
9138 struct mdp_superblock_1 *sb;
9139
9140 /* Store the sb page of the rdev in the swapout temporary
9141 * variable in case we err in the future
9142 */
9143 rdev->sb_page = NULL;
9144 alloc_disk_sb(rdev);
9145 ClearPageUptodate(rdev->sb_page);
9146 rdev->sb_loaded = 0;
9147 err = super_types[mddev->major_version].load_super(rdev, NULL, mddev->minor_version);
9148
9149 if (err < 0) {
9150 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
9151 __func__, __LINE__, rdev->desc_nr, err);
9152 put_page(rdev->sb_page);
9153 rdev->sb_page = swapout;
9154 rdev->sb_loaded = 1;
9155 return err;
9156 }
9157
9158 sb = page_address(rdev->sb_page);
9159 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
9160 * is not set
9161 */
9162
9163 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
9164 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
9165
9166 /* The other node finished recovery, call spare_active to set
9167 * device In_sync and mddev->degraded
9168 */
9169 if (rdev->recovery_offset == MaxSector &&
9170 !test_bit(In_sync, &rdev->flags) &&
9171 mddev->pers->spare_active(mddev))
9172 sysfs_notify(&mddev->kobj, NULL, "degraded");
9173
9174 put_page(swapout);
9175 return 0;
9176 }
9177
9178 void md_reload_sb(struct mddev *mddev, int nr)
9179 {
9180 struct md_rdev *rdev;
9181 int err;
9182
9183 /* Find the rdev */
9184 rdev_for_each_rcu(rdev, mddev) {
9185 if (rdev->desc_nr == nr)
9186 break;
9187 }
9188
9189 if (!rdev || rdev->desc_nr != nr) {
9190 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
9191 return;
9192 }
9193
9194 err = read_rdev(mddev, rdev);
9195 if (err < 0)
9196 return;
9197
9198 check_sb_changes(mddev, rdev);
9199
9200 /* Read all rdev's to update recovery_offset */
9201 rdev_for_each_rcu(rdev, mddev)
9202 read_rdev(mddev, rdev);
9203 }
9204 EXPORT_SYMBOL(md_reload_sb);
9205
9206 #ifndef MODULE
9207
9208 /*
9209 * Searches all registered partitions for autorun RAID arrays
9210 * at boot time.
9211 */
9212
9213 static LIST_HEAD(all_detected_devices);
9214 struct detected_devices_node {
9215 struct list_head list;
9216 dev_t dev;
9217 };
9218
9219 void md_autodetect_dev(dev_t dev)
9220 {
9221 struct detected_devices_node *node_detected_dev;
9222
9223 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
9224 if (node_detected_dev) {
9225 node_detected_dev->dev = dev;
9226 list_add_tail(&node_detected_dev->list, &all_detected_devices);
9227 } else {
9228 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
9229 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
9230 }
9231 }
9232
9233 static void autostart_arrays(int part)
9234 {
9235 struct md_rdev *rdev;
9236 struct detected_devices_node *node_detected_dev;
9237 dev_t dev;
9238 int i_scanned, i_passed;
9239
9240 i_scanned = 0;
9241 i_passed = 0;
9242
9243 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
9244
9245 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
9246 i_scanned++;
9247 node_detected_dev = list_entry(all_detected_devices.next,
9248 struct detected_devices_node, list);
9249 list_del(&node_detected_dev->list);
9250 dev = node_detected_dev->dev;
9251 kfree(node_detected_dev);
9252 rdev = md_import_device(dev,0, 90);
9253 if (IS_ERR(rdev))
9254 continue;
9255
9256 if (test_bit(Faulty, &rdev->flags))
9257 continue;
9258
9259 set_bit(AutoDetected, &rdev->flags);
9260 list_add(&rdev->same_set, &pending_raid_disks);
9261 i_passed++;
9262 }
9263
9264 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
9265 i_scanned, i_passed);
9266
9267 autorun_devices(part);
9268 }
9269
9270 #endif /* !MODULE */
9271
9272 static __exit void md_exit(void)
9273 {
9274 struct mddev *mddev;
9275 struct list_head *tmp;
9276 int delay = 1;
9277
9278 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
9279 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
9280
9281 unregister_blkdev(MD_MAJOR,"md");
9282 unregister_blkdev(mdp_major, "mdp");
9283 unregister_reboot_notifier(&md_notifier);
9284 unregister_sysctl_table(raid_table_header);
9285
9286 /* We cannot unload the modules while some process is
9287 * waiting for us in select() or poll() - wake them up
9288 */
9289 md_unloading = 1;
9290 while (waitqueue_active(&md_event_waiters)) {
9291 /* not safe to leave yet */
9292 wake_up(&md_event_waiters);
9293 msleep(delay);
9294 delay += delay;
9295 }
9296 remove_proc_entry("mdstat", NULL);
9297
9298 for_each_mddev(mddev, tmp) {
9299 export_array(mddev);
9300 mddev->hold_active = 0;
9301 }
9302 destroy_workqueue(md_misc_wq);
9303 destroy_workqueue(md_wq);
9304 }
9305
9306 subsys_initcall(md_init);
9307 module_exit(md_exit)
9308
9309 static int get_ro(char *buffer, struct kernel_param *kp)
9310 {
9311 return sprintf(buffer, "%d", start_readonly);
9312 }
9313 static int set_ro(const char *val, struct kernel_param *kp)
9314 {
9315 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
9316 }
9317
9318 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9319 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
9320 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
9321
9322 MODULE_LICENSE("GPL");
9323 MODULE_DESCRIPTION("MD RAID framework");
9324 MODULE_ALIAS("md");
9325 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Linux with added FPC-III support