]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - drivers/md/md.c
projects / mirror_ubuntu-jammy-kernel.git / blob
? search:


c5a1b0725c9fc4f7d25405050429fefebd26882f
[mirror_ubuntu-jammy-kernel.git] / drivers / md / md.c
1 /*
2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
4
5 completely rewritten, based on the MD driver code from Marc Zyngier
6
7 Changes:
8
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
19
20 Neil Brown <neilb@cse.unsw.edu.au>.
21
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
28 any later version.
29
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33 */
34
35 #include <linux/kthread.h>
36 #include <linux/blkdev.h>
37 #include <linux/sysctl.h>
38 #include <linux/seq_file.h>
39 #include <linux/buffer_head.h> /* for invalidate_bdev */
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/reboot.h>
47 #include <linux/file.h>
48 #include <linux/compat.h>
49 #include <linux/delay.h>
50 #include <linux/raid/md_p.h>
51 #include <linux/raid/md_u.h>
52 #include "md.h"
53 #include "bitmap.h"
54
55 #define DEBUG 0
56 #define dprintk(x...) ((void)(DEBUG && printk(x)))
57
58
59 #ifndef MODULE
60 static void autostart_arrays(int part);
61 #endif
62
63 static LIST_HEAD(pers_list);
64 static DEFINE_SPINLOCK(pers_lock);
65
66 static void md_print_devices(void);
67
68 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
69
70 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
71
72 /*
73 * Default number of read corrections we'll attempt on an rdev
74 * before ejecting it from the array. We divide the read error
75 * count by 2 for every hour elapsed between read errors.
76 */
77 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
78 /*
79 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
80 * is 1000 KB/sec, so the extra system load does not show up that much.
81 * Increase it if you want to have more _guaranteed_ speed. Note that
82 * the RAID driver will use the maximum available bandwidth if the IO
83 * subsystem is idle. There is also an 'absolute maximum' reconstruction
84 * speed limit - in case reconstruction slows down your system despite
85 * idle IO detection.
86 *
87 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
88 * or /sys/block/mdX/md/sync_speed_{min,max}
89 */
90
91 static int sysctl_speed_limit_min = 1000;
92 static int sysctl_speed_limit_max = 200000;
93 static inline int speed_min(mddev_t *mddev)
94 {
95 return mddev->sync_speed_min ?
96 mddev->sync_speed_min : sysctl_speed_limit_min;
97 }
98
99 static inline int speed_max(mddev_t *mddev)
100 {
101 return mddev->sync_speed_max ?
102 mddev->sync_speed_max : sysctl_speed_limit_max;
103 }
104
105 static struct ctl_table_header *raid_table_header;
106
107 static ctl_table raid_table[] = {
108 {
109 .procname = "speed_limit_min",
110 .data = &sysctl_speed_limit_min,
111 .maxlen = sizeof(int),
112 .mode = S_IRUGO|S_IWUSR,
113 .proc_handler = proc_dointvec,
114 },
115 {
116 .procname = "speed_limit_max",
117 .data = &sysctl_speed_limit_max,
118 .maxlen = sizeof(int),
119 .mode = S_IRUGO|S_IWUSR,
120 .proc_handler = proc_dointvec,
121 },
122 { }
123 };
124
125 static ctl_table raid_dir_table[] = {
126 {
127 .procname = "raid",
128 .maxlen = 0,
129 .mode = S_IRUGO|S_IXUGO,
130 .child = raid_table,
131 },
132 { }
133 };
134
135 static ctl_table raid_root_table[] = {
136 {
137 .procname = "dev",
138 .maxlen = 0,
139 .mode = 0555,
140 .child = raid_dir_table,
141 },
142 { }
143 };
144
145 static const struct block_device_operations md_fops;
146
147 static int start_readonly;
148
149 /*
150 * We have a system wide 'event count' that is incremented
151 * on any 'interesting' event, and readers of /proc/mdstat
152 * can use 'poll' or 'select' to find out when the event
153 * count increases.
154 *
155 * Events are:
156 * start array, stop array, error, add device, remove device,
157 * start build, activate spare
158 */
159 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
160 static atomic_t md_event_count;
161 void md_new_event(mddev_t *mddev)
162 {
163 atomic_inc(&md_event_count);
164 wake_up(&md_event_waiters);
165 }
166 EXPORT_SYMBOL_GPL(md_new_event);
167
168 /* Alternate version that can be called from interrupts
169 * when calling sysfs_notify isn't needed.
170 */
171 static void md_new_event_inintr(mddev_t *mddev)
172 {
173 atomic_inc(&md_event_count);
174 wake_up(&md_event_waiters);
175 }
176
177 /*
178 * Enables to iterate over all existing md arrays
179 * all_mddevs_lock protects this list.
180 */
181 static LIST_HEAD(all_mddevs);
182 static DEFINE_SPINLOCK(all_mddevs_lock);
183
184
185 /*
186 * iterates through all used mddevs in the system.
187 * We take care to grab the all_mddevs_lock whenever navigating
188 * the list, and to always hold a refcount when unlocked.
189 * Any code which breaks out of this loop while own
190 * a reference to the current mddev and must mddev_put it.
191 */
192 #define for_each_mddev(mddev,tmp) \
193 \
194 for (({ spin_lock(&all_mddevs_lock); \
195 tmp = all_mddevs.next; \
196 mddev = NULL;}); \
197 ({ if (tmp != &all_mddevs) \
198 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
199 spin_unlock(&all_mddevs_lock); \
200 if (mddev) mddev_put(mddev); \
201 mddev = list_entry(tmp, mddev_t, all_mddevs); \
202 tmp != &all_mddevs;}); \
203 ({ spin_lock(&all_mddevs_lock); \
204 tmp = tmp->next;}) \
205 )
206
207
208 /* Rather than calling directly into the personality make_request function,
209 * IO requests come here first so that we can check if the device is
210 * being suspended pending a reconfiguration.
211 * We hold a refcount over the call to ->make_request. By the time that
212 * call has finished, the bio has been linked into some internal structure
213 * and so is visible to ->quiesce(), so we don't need the refcount any more.
214 */
215 static int md_make_request(struct request_queue *q, struct bio *bio)
216 {
217 mddev_t *mddev = q->queuedata;
218 int rv;
219 if (mddev == NULL || mddev->pers == NULL) {
220 bio_io_error(bio);
221 return 0;
222 }
223 rcu_read_lock();
224 if (mddev->suspended || mddev->barrier) {
225 DEFINE_WAIT(__wait);
226 for (;;) {
227 prepare_to_wait(&mddev->sb_wait, &__wait,
228 TASK_UNINTERRUPTIBLE);
229 if (!mddev->suspended && !mddev->barrier)
230 break;
231 rcu_read_unlock();
232 schedule();
233 rcu_read_lock();
234 }
235 finish_wait(&mddev->sb_wait, &__wait);
236 }
237 atomic_inc(&mddev->active_io);
238 rcu_read_unlock();
239 rv = mddev->pers->make_request(q, bio);
240 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
241 wake_up(&mddev->sb_wait);
242
243 return rv;
244 }
245
246 static void mddev_suspend(mddev_t *mddev)
247 {
248 BUG_ON(mddev->suspended);
249 mddev->suspended = 1;
250 synchronize_rcu();
251 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
252 mddev->pers->quiesce(mddev, 1);
253 md_unregister_thread(mddev->thread);
254 mddev->thread = NULL;
255 /* we now know that no code is executing in the personality module,
256 * except possibly the tail end of a ->bi_end_io function, but that
257 * is certain to complete before the module has a chance to get
258 * unloaded
259 */
260 }
261
262 static void mddev_resume(mddev_t *mddev)
263 {
264 mddev->suspended = 0;
265 wake_up(&mddev->sb_wait);
266 mddev->pers->quiesce(mddev, 0);
267 }
268
269 int mddev_congested(mddev_t *mddev, int bits)
270 {
271 if (mddev->barrier)
272 return 1;
273 return mddev->suspended;
274 }
275 EXPORT_SYMBOL(mddev_congested);
276
277 /*
278 * Generic barrier handling for md
279 */
280
281 #define POST_REQUEST_BARRIER ((void*)1)
282
283 static void md_end_barrier(struct bio *bio, int err)
284 {
285 mdk_rdev_t *rdev = bio->bi_private;
286 mddev_t *mddev = rdev->mddev;
287 if (err == -EOPNOTSUPP && mddev->barrier != POST_REQUEST_BARRIER)
288 set_bit(BIO_EOPNOTSUPP, &mddev->barrier->bi_flags);
289
290 rdev_dec_pending(rdev, mddev);
291
292 if (atomic_dec_and_test(&mddev->flush_pending)) {
293 if (mddev->barrier == POST_REQUEST_BARRIER) {
294 /* This was a post-request barrier */
295 mddev->barrier = NULL;
296 wake_up(&mddev->sb_wait);
297 } else
298 /* The pre-request barrier has finished */
299 schedule_work(&mddev->barrier_work);
300 }
301 bio_put(bio);
302 }
303
304 static void submit_barriers(mddev_t *mddev)
305 {
306 mdk_rdev_t *rdev;
307
308 rcu_read_lock();
309 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
310 if (rdev->raid_disk >= 0 &&
311 !test_bit(Faulty, &rdev->flags)) {
312 /* Take two references, one is dropped
313 * when request finishes, one after
314 * we reclaim rcu_read_lock
315 */
316 struct bio *bi;
317 atomic_inc(&rdev->nr_pending);
318 atomic_inc(&rdev->nr_pending);
319 rcu_read_unlock();
320 bi = bio_alloc(GFP_KERNEL, 0);
321 bi->bi_end_io = md_end_barrier;
322 bi->bi_private = rdev;
323 bi->bi_bdev = rdev->bdev;
324 atomic_inc(&mddev->flush_pending);
325 submit_bio(WRITE_BARRIER, bi);
326 rcu_read_lock();
327 rdev_dec_pending(rdev, mddev);
328 }
329 rcu_read_unlock();
330 }
331
332 static void md_submit_barrier(struct work_struct *ws)
333 {
334 mddev_t *mddev = container_of(ws, mddev_t, barrier_work);
335 struct bio *bio = mddev->barrier;
336
337 atomic_set(&mddev->flush_pending, 1);
338
339 if (test_bit(BIO_EOPNOTSUPP, &bio->bi_flags))
340 bio_endio(bio, -EOPNOTSUPP);
341 else if (bio->bi_size == 0)
342 /* an empty barrier - all done */
343 bio_endio(bio, 0);
344 else {
345 bio->bi_rw &= ~(1<<BIO_RW_BARRIER);
346 if (mddev->pers->make_request(mddev->queue, bio))
347 generic_make_request(bio);
348 mddev->barrier = POST_REQUEST_BARRIER;
349 submit_barriers(mddev);
350 }
351 if (atomic_dec_and_test(&mddev->flush_pending)) {
352 mddev->barrier = NULL;
353 wake_up(&mddev->sb_wait);
354 }
355 }
356
357 void md_barrier_request(mddev_t *mddev, struct bio *bio)
358 {
359 spin_lock_irq(&mddev->write_lock);
360 wait_event_lock_irq(mddev->sb_wait,
361 !mddev->barrier,
362 mddev->write_lock, /*nothing*/);
363 mddev->barrier = bio;
364 spin_unlock_irq(&mddev->write_lock);
365
366 atomic_set(&mddev->flush_pending, 1);
367 INIT_WORK(&mddev->barrier_work, md_submit_barrier);
368
369 submit_barriers(mddev);
370
371 if (atomic_dec_and_test(&mddev->flush_pending))
372 schedule_work(&mddev->barrier_work);
373 }
374 EXPORT_SYMBOL(md_barrier_request);
375
376 static inline mddev_t *mddev_get(mddev_t *mddev)
377 {
378 atomic_inc(&mddev->active);
379 return mddev;
380 }
381
382 static void mddev_delayed_delete(struct work_struct *ws);
383
384 static void mddev_put(mddev_t *mddev)
385 {
386 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
387 return;
388 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
389 mddev->ctime == 0 && !mddev->hold_active) {
390 /* Array is not configured at all, and not held active,
391 * so destroy it */
392 list_del(&mddev->all_mddevs);
393 if (mddev->gendisk) {
394 /* we did a probe so need to clean up.
395 * Call schedule_work inside the spinlock
396 * so that flush_scheduled_work() after
397 * mddev_find will succeed in waiting for the
398 * work to be done.
399 */
400 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
401 schedule_work(&mddev->del_work);
402 } else
403 kfree(mddev);
404 }
405 spin_unlock(&all_mddevs_lock);
406 }
407
408 static mddev_t * mddev_find(dev_t unit)
409 {
410 mddev_t *mddev, *new = NULL;
411
412 retry:
413 spin_lock(&all_mddevs_lock);
414
415 if (unit) {
416 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
417 if (mddev->unit == unit) {
418 mddev_get(mddev);
419 spin_unlock(&all_mddevs_lock);
420 kfree(new);
421 return mddev;
422 }
423
424 if (new) {
425 list_add(&new->all_mddevs, &all_mddevs);
426 spin_unlock(&all_mddevs_lock);
427 new->hold_active = UNTIL_IOCTL;
428 return new;
429 }
430 } else if (new) {
431 /* find an unused unit number */
432 static int next_minor = 512;
433 int start = next_minor;
434 int is_free = 0;
435 int dev = 0;
436 while (!is_free) {
437 dev = MKDEV(MD_MAJOR, next_minor);
438 next_minor++;
439 if (next_minor > MINORMASK)
440 next_minor = 0;
441 if (next_minor == start) {
442 /* Oh dear, all in use. */
443 spin_unlock(&all_mddevs_lock);
444 kfree(new);
445 return NULL;
446 }
447
448 is_free = 1;
449 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
450 if (mddev->unit == dev) {
451 is_free = 0;
452 break;
453 }
454 }
455 new->unit = dev;
456 new->md_minor = MINOR(dev);
457 new->hold_active = UNTIL_STOP;
458 list_add(&new->all_mddevs, &all_mddevs);
459 spin_unlock(&all_mddevs_lock);
460 return new;
461 }
462 spin_unlock(&all_mddevs_lock);
463
464 new = kzalloc(sizeof(*new), GFP_KERNEL);
465 if (!new)
466 return NULL;
467
468 new->unit = unit;
469 if (MAJOR(unit) == MD_MAJOR)
470 new->md_minor = MINOR(unit);
471 else
472 new->md_minor = MINOR(unit) >> MdpMinorShift;
473
474 mutex_init(&new->open_mutex);
475 mutex_init(&new->reconfig_mutex);
476 mutex_init(&new->bitmap_info.mutex);
477 INIT_LIST_HEAD(&new->disks);
478 INIT_LIST_HEAD(&new->all_mddevs);
479 init_timer(&new->safemode_timer);
480 atomic_set(&new->active, 1);
481 atomic_set(&new->openers, 0);
482 atomic_set(&new->active_io, 0);
483 spin_lock_init(&new->write_lock);
484 atomic_set(&new->flush_pending, 0);
485 init_waitqueue_head(&new->sb_wait);
486 init_waitqueue_head(&new->recovery_wait);
487 new->reshape_position = MaxSector;
488 new->resync_min = 0;
489 new->resync_max = MaxSector;
490 new->level = LEVEL_NONE;
491
492 goto retry;
493 }
494
495 static inline int mddev_lock(mddev_t * mddev)
496 {
497 return mutex_lock_interruptible(&mddev->reconfig_mutex);
498 }
499
500 static inline int mddev_is_locked(mddev_t *mddev)
501 {
502 return mutex_is_locked(&mddev->reconfig_mutex);
503 }
504
505 static inline int mddev_trylock(mddev_t * mddev)
506 {
507 return mutex_trylock(&mddev->reconfig_mutex);
508 }
509
510 static struct attribute_group md_redundancy_group;
511
512 static void mddev_unlock(mddev_t * mddev)
513 {
514 if (mddev->to_remove) {
515 /* These cannot be removed under reconfig_mutex as
516 * an access to the files will try to take reconfig_mutex
517 * while holding the file unremovable, which leads to
518 * a deadlock.
519 * So hold open_mutex instead - we are allowed to take
520 * it while holding reconfig_mutex, and md_run can
521 * use it to wait for the remove to complete.
522 */
523 struct attribute_group *to_remove = mddev->to_remove;
524 mddev->to_remove = NULL;
525 mutex_lock(&mddev->open_mutex);
526 mutex_unlock(&mddev->reconfig_mutex);
527
528 if (to_remove != &md_redundancy_group)
529 sysfs_remove_group(&mddev->kobj, to_remove);
530 if (mddev->pers == NULL ||
531 mddev->pers->sync_request == NULL) {
532 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
533 if (mddev->sysfs_action)
534 sysfs_put(mddev->sysfs_action);
535 mddev->sysfs_action = NULL;
536 }
537 mutex_unlock(&mddev->open_mutex);
538 } else
539 mutex_unlock(&mddev->reconfig_mutex);
540
541 md_wakeup_thread(mddev->thread);
542 }
543
544 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
545 {
546 mdk_rdev_t *rdev;
547
548 list_for_each_entry(rdev, &mddev->disks, same_set)
549 if (rdev->desc_nr == nr)
550 return rdev;
551
552 return NULL;
553 }
554
555 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
556 {
557 mdk_rdev_t *rdev;
558
559 list_for_each_entry(rdev, &mddev->disks, same_set)
560 if (rdev->bdev->bd_dev == dev)
561 return rdev;
562
563 return NULL;
564 }
565
566 static struct mdk_personality *find_pers(int level, char *clevel)
567 {
568 struct mdk_personality *pers;
569 list_for_each_entry(pers, &pers_list, list) {
570 if (level != LEVEL_NONE && pers->level == level)
571 return pers;
572 if (strcmp(pers->name, clevel)==0)
573 return pers;
574 }
575 return NULL;
576 }
577
578 /* return the offset of the super block in 512byte sectors */
579 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
580 {
581 sector_t num_sectors = bdev->bd_inode->i_size / 512;
582 return MD_NEW_SIZE_SECTORS(num_sectors);
583 }
584
585 static int alloc_disk_sb(mdk_rdev_t * rdev)
586 {
587 if (rdev->sb_page)
588 MD_BUG();
589
590 rdev->sb_page = alloc_page(GFP_KERNEL);
591 if (!rdev->sb_page) {
592 printk(KERN_ALERT "md: out of memory.\n");
593 return -ENOMEM;
594 }
595
596 return 0;
597 }
598
599 static void free_disk_sb(mdk_rdev_t * rdev)
600 {
601 if (rdev->sb_page) {
602 put_page(rdev->sb_page);
603 rdev->sb_loaded = 0;
604 rdev->sb_page = NULL;
605 rdev->sb_start = 0;
606 rdev->sectors = 0;
607 }
608 }
609
610
611 static void super_written(struct bio *bio, int error)
612 {
613 mdk_rdev_t *rdev = bio->bi_private;
614 mddev_t *mddev = rdev->mddev;
615
616 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
617 printk("md: super_written gets error=%d, uptodate=%d\n",
618 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
619 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
620 md_error(mddev, rdev);
621 }
622
623 if (atomic_dec_and_test(&mddev->pending_writes))
624 wake_up(&mddev->sb_wait);
625 bio_put(bio);
626 }
627
628 static void super_written_barrier(struct bio *bio, int error)
629 {
630 struct bio *bio2 = bio->bi_private;
631 mdk_rdev_t *rdev = bio2->bi_private;
632 mddev_t *mddev = rdev->mddev;
633
634 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
635 error == -EOPNOTSUPP) {
636 unsigned long flags;
637 /* barriers don't appear to be supported :-( */
638 set_bit(BarriersNotsupp, &rdev->flags);
639 mddev->barriers_work = 0;
640 spin_lock_irqsave(&mddev->write_lock, flags);
641 bio2->bi_next = mddev->biolist;
642 mddev->biolist = bio2;
643 spin_unlock_irqrestore(&mddev->write_lock, flags);
644 wake_up(&mddev->sb_wait);
645 bio_put(bio);
646 } else {
647 bio_put(bio2);
648 bio->bi_private = rdev;
649 super_written(bio, error);
650 }
651 }
652
653 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
654 sector_t sector, int size, struct page *page)
655 {
656 /* write first size bytes of page to sector of rdev
657 * Increment mddev->pending_writes before returning
658 * and decrement it on completion, waking up sb_wait
659 * if zero is reached.
660 * If an error occurred, call md_error
661 *
662 * As we might need to resubmit the request if BIO_RW_BARRIER
663 * causes ENOTSUPP, we allocate a spare bio...
664 */
665 struct bio *bio = bio_alloc(GFP_NOIO, 1);
666 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG);
667
668 bio->bi_bdev = rdev->bdev;
669 bio->bi_sector = sector;
670 bio_add_page(bio, page, size, 0);
671 bio->bi_private = rdev;
672 bio->bi_end_io = super_written;
673 bio->bi_rw = rw;
674
675 atomic_inc(&mddev->pending_writes);
676 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
677 struct bio *rbio;
678 rw |= (1<<BIO_RW_BARRIER);
679 rbio = bio_clone(bio, GFP_NOIO);
680 rbio->bi_private = bio;
681 rbio->bi_end_io = super_written_barrier;
682 submit_bio(rw, rbio);
683 } else
684 submit_bio(rw, bio);
685 }
686
687 void md_super_wait(mddev_t *mddev)
688 {
689 /* wait for all superblock writes that were scheduled to complete.
690 * if any had to be retried (due to BARRIER problems), retry them
691 */
692 DEFINE_WAIT(wq);
693 for(;;) {
694 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
695 if (atomic_read(&mddev->pending_writes)==0)
696 break;
697 while (mddev->biolist) {
698 struct bio *bio;
699 spin_lock_irq(&mddev->write_lock);
700 bio = mddev->biolist;
701 mddev->biolist = bio->bi_next ;
702 bio->bi_next = NULL;
703 spin_unlock_irq(&mddev->write_lock);
704 submit_bio(bio->bi_rw, bio);
705 }
706 schedule();
707 }
708 finish_wait(&mddev->sb_wait, &wq);
709 }
710
711 static void bi_complete(struct bio *bio, int error)
712 {
713 complete((struct completion*)bio->bi_private);
714 }
715
716 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
717 struct page *page, int rw)
718 {
719 struct bio *bio = bio_alloc(GFP_NOIO, 1);
720 struct completion event;
721 int ret;
722
723 rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
724
725 bio->bi_bdev = bdev;
726 bio->bi_sector = sector;
727 bio_add_page(bio, page, size, 0);
728 init_completion(&event);
729 bio->bi_private = &event;
730 bio->bi_end_io = bi_complete;
731 submit_bio(rw, bio);
732 wait_for_completion(&event);
733
734 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
735 bio_put(bio);
736 return ret;
737 }
738 EXPORT_SYMBOL_GPL(sync_page_io);
739
740 static int read_disk_sb(mdk_rdev_t * rdev, int size)
741 {
742 char b[BDEVNAME_SIZE];
743 if (!rdev->sb_page) {
744 MD_BUG();
745 return -EINVAL;
746 }
747 if (rdev->sb_loaded)
748 return 0;
749
750
751 if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
752 goto fail;
753 rdev->sb_loaded = 1;
754 return 0;
755
756 fail:
757 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
758 bdevname(rdev->bdev,b));
759 return -EINVAL;
760 }
761
762 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
763 {
764 return sb1->set_uuid0 == sb2->set_uuid0 &&
765 sb1->set_uuid1 == sb2->set_uuid1 &&
766 sb1->set_uuid2 == sb2->set_uuid2 &&
767 sb1->set_uuid3 == sb2->set_uuid3;
768 }
769
770 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
771 {
772 int ret;
773 mdp_super_t *tmp1, *tmp2;
774
775 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
776 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
777
778 if (!tmp1 || !tmp2) {
779 ret = 0;
780 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
781 goto abort;
782 }
783
784 *tmp1 = *sb1;
785 *tmp2 = *sb2;
786
787 /*
788 * nr_disks is not constant
789 */
790 tmp1->nr_disks = 0;
791 tmp2->nr_disks = 0;
792
793 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
794 abort:
795 kfree(tmp1);
796 kfree(tmp2);
797 return ret;
798 }
799
800
801 static u32 md_csum_fold(u32 csum)
802 {
803 csum = (csum & 0xffff) + (csum >> 16);
804 return (csum & 0xffff) + (csum >> 16);
805 }
806
807 static unsigned int calc_sb_csum(mdp_super_t * sb)
808 {
809 u64 newcsum = 0;
810 u32 *sb32 = (u32*)sb;
811 int i;
812 unsigned int disk_csum, csum;
813
814 disk_csum = sb->sb_csum;
815 sb->sb_csum = 0;
816
817 for (i = 0; i < MD_SB_BYTES/4 ; i++)
818 newcsum += sb32[i];
819 csum = (newcsum & 0xffffffff) + (newcsum>>32);
820
821
822 #ifdef CONFIG_ALPHA
823 /* This used to use csum_partial, which was wrong for several
824 * reasons including that different results are returned on
825 * different architectures. It isn't critical that we get exactly
826 * the same return value as before (we always csum_fold before
827 * testing, and that removes any differences). However as we
828 * know that csum_partial always returned a 16bit value on
829 * alphas, do a fold to maximise conformity to previous behaviour.
830 */
831 sb->sb_csum = md_csum_fold(disk_csum);
832 #else
833 sb->sb_csum = disk_csum;
834 #endif
835 return csum;
836 }
837
838
839 /*
840 * Handle superblock details.
841 * We want to be able to handle multiple superblock formats
842 * so we have a common interface to them all, and an array of
843 * different handlers.
844 * We rely on user-space to write the initial superblock, and support
845 * reading and updating of superblocks.
846 * Interface methods are:
847 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
848 * loads and validates a superblock on dev.
849 * if refdev != NULL, compare superblocks on both devices
850 * Return:
851 * 0 - dev has a superblock that is compatible with refdev
852 * 1 - dev has a superblock that is compatible and newer than refdev
853 * so dev should be used as the refdev in future
854 * -EINVAL superblock incompatible or invalid
855 * -othererror e.g. -EIO
856 *
857 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
858 * Verify that dev is acceptable into mddev.
859 * The first time, mddev->raid_disks will be 0, and data from
860 * dev should be merged in. Subsequent calls check that dev
861 * is new enough. Return 0 or -EINVAL
862 *
863 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
864 * Update the superblock for rdev with data in mddev
865 * This does not write to disc.
866 *
867 */
868
869 struct super_type {
870 char *name;
871 struct module *owner;
872 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
873 int minor_version);
874 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
875 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
876 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
877 sector_t num_sectors);
878 };
879
880 /*
881 * Check that the given mddev has no bitmap.
882 *
883 * This function is called from the run method of all personalities that do not
884 * support bitmaps. It prints an error message and returns non-zero if mddev
885 * has a bitmap. Otherwise, it returns 0.
886 *
887 */
888 int md_check_no_bitmap(mddev_t *mddev)
889 {
890 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
891 return 0;
892 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
893 mdname(mddev), mddev->pers->name);
894 return 1;
895 }
896 EXPORT_SYMBOL(md_check_no_bitmap);
897
898 /*
899 * load_super for 0.90.0
900 */
901 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
902 {
903 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
904 mdp_super_t *sb;
905 int ret;
906
907 /*
908 * Calculate the position of the superblock (512byte sectors),
909 * it's at the end of the disk.
910 *
911 * It also happens to be a multiple of 4Kb.
912 */
913 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
914
915 ret = read_disk_sb(rdev, MD_SB_BYTES);
916 if (ret) return ret;
917
918 ret = -EINVAL;
919
920 bdevname(rdev->bdev, b);
921 sb = (mdp_super_t*)page_address(rdev->sb_page);
922
923 if (sb->md_magic != MD_SB_MAGIC) {
924 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
925 b);
926 goto abort;
927 }
928
929 if (sb->major_version != 0 ||
930 sb->minor_version < 90 ||
931 sb->minor_version > 91) {
932 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
933 sb->major_version, sb->minor_version,
934 b);
935 goto abort;
936 }
937
938 if (sb->raid_disks <= 0)
939 goto abort;
940
941 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
942 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
943 b);
944 goto abort;
945 }
946
947 rdev->preferred_minor = sb->md_minor;
948 rdev->data_offset = 0;
949 rdev->sb_size = MD_SB_BYTES;
950
951 if (sb->level == LEVEL_MULTIPATH)
952 rdev->desc_nr = -1;
953 else
954 rdev->desc_nr = sb->this_disk.number;
955
956 if (!refdev) {
957 ret = 1;
958 } else {
959 __u64 ev1, ev2;
960 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
961 if (!uuid_equal(refsb, sb)) {
962 printk(KERN_WARNING "md: %s has different UUID to %s\n",
963 b, bdevname(refdev->bdev,b2));
964 goto abort;
965 }
966 if (!sb_equal(refsb, sb)) {
967 printk(KERN_WARNING "md: %s has same UUID"
968 " but different superblock to %s\n",
969 b, bdevname(refdev->bdev, b2));
970 goto abort;
971 }
972 ev1 = md_event(sb);
973 ev2 = md_event(refsb);
974 if (ev1 > ev2)
975 ret = 1;
976 else
977 ret = 0;
978 }
979 rdev->sectors = rdev->sb_start;
980
981 if (rdev->sectors < sb->size * 2 && sb->level > 1)
982 /* "this cannot possibly happen" ... */
983 ret = -EINVAL;
984
985 abort:
986 return ret;
987 }
988
989 /*
990 * validate_super for 0.90.0
991 */
992 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
993 {
994 mdp_disk_t *desc;
995 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
996 __u64 ev1 = md_event(sb);
997
998 rdev->raid_disk = -1;
999 clear_bit(Faulty, &rdev->flags);
1000 clear_bit(In_sync, &rdev->flags);
1001 clear_bit(WriteMostly, &rdev->flags);
1002 clear_bit(BarriersNotsupp, &rdev->flags);
1003
1004 if (mddev->raid_disks == 0) {
1005 mddev->major_version = 0;
1006 mddev->minor_version = sb->minor_version;
1007 mddev->patch_version = sb->patch_version;
1008 mddev->external = 0;
1009 mddev->chunk_sectors = sb->chunk_size >> 9;
1010 mddev->ctime = sb->ctime;
1011 mddev->utime = sb->utime;
1012 mddev->level = sb->level;
1013 mddev->clevel[0] = 0;
1014 mddev->layout = sb->layout;
1015 mddev->raid_disks = sb->raid_disks;
1016 mddev->dev_sectors = sb->size * 2;
1017 mddev->events = ev1;
1018 mddev->bitmap_info.offset = 0;
1019 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1020
1021 if (mddev->minor_version >= 91) {
1022 mddev->reshape_position = sb->reshape_position;
1023 mddev->delta_disks = sb->delta_disks;
1024 mddev->new_level = sb->new_level;
1025 mddev->new_layout = sb->new_layout;
1026 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1027 } else {
1028 mddev->reshape_position = MaxSector;
1029 mddev->delta_disks = 0;
1030 mddev->new_level = mddev->level;
1031 mddev->new_layout = mddev->layout;
1032 mddev->new_chunk_sectors = mddev->chunk_sectors;
1033 }
1034
1035 if (sb->state & (1<<MD_SB_CLEAN))
1036 mddev->recovery_cp = MaxSector;
1037 else {
1038 if (sb->events_hi == sb->cp_events_hi &&
1039 sb->events_lo == sb->cp_events_lo) {
1040 mddev->recovery_cp = sb->recovery_cp;
1041 } else
1042 mddev->recovery_cp = 0;
1043 }
1044
1045 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1046 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1047 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1048 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1049
1050 mddev->max_disks = MD_SB_DISKS;
1051
1052 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1053 mddev->bitmap_info.file == NULL)
1054 mddev->bitmap_info.offset =
1055 mddev->bitmap_info.default_offset;
1056
1057 } else if (mddev->pers == NULL) {
1058 /* Insist on good event counter while assembling */
1059 ++ev1;
1060 if (ev1 < mddev->events)
1061 return -EINVAL;
1062 } else if (mddev->bitmap) {
1063 /* if adding to array with a bitmap, then we can accept an
1064 * older device ... but not too old.
1065 */
1066 if (ev1 < mddev->bitmap->events_cleared)
1067 return 0;
1068 } else {
1069 if (ev1 < mddev->events)
1070 /* just a hot-add of a new device, leave raid_disk at -1 */
1071 return 0;
1072 }
1073
1074 if (mddev->level != LEVEL_MULTIPATH) {
1075 desc = sb->disks + rdev->desc_nr;
1076
1077 if (desc->state & (1<<MD_DISK_FAULTY))
1078 set_bit(Faulty, &rdev->flags);
1079 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1080 desc->raid_disk < mddev->raid_disks */) {
1081 set_bit(In_sync, &rdev->flags);
1082 rdev->raid_disk = desc->raid_disk;
1083 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1084 /* active but not in sync implies recovery up to
1085 * reshape position. We don't know exactly where
1086 * that is, so set to zero for now */
1087 if (mddev->minor_version >= 91) {
1088 rdev->recovery_offset = 0;
1089 rdev->raid_disk = desc->raid_disk;
1090 }
1091 }
1092 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1093 set_bit(WriteMostly, &rdev->flags);
1094 } else /* MULTIPATH are always insync */
1095 set_bit(In_sync, &rdev->flags);
1096 return 0;
1097 }
1098
1099 /*
1100 * sync_super for 0.90.0
1101 */
1102 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1103 {
1104 mdp_super_t *sb;
1105 mdk_rdev_t *rdev2;
1106 int next_spare = mddev->raid_disks;
1107
1108
1109 /* make rdev->sb match mddev data..
1110 *
1111 * 1/ zero out disks
1112 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1113 * 3/ any empty disks < next_spare become removed
1114 *
1115 * disks[0] gets initialised to REMOVED because
1116 * we cannot be sure from other fields if it has
1117 * been initialised or not.
1118 */
1119 int i;
1120 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1121
1122 rdev->sb_size = MD_SB_BYTES;
1123
1124 sb = (mdp_super_t*)page_address(rdev->sb_page);
1125
1126 memset(sb, 0, sizeof(*sb));
1127
1128 sb->md_magic = MD_SB_MAGIC;
1129 sb->major_version = mddev->major_version;
1130 sb->patch_version = mddev->patch_version;
1131 sb->gvalid_words = 0; /* ignored */
1132 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1133 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1134 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1135 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1136
1137 sb->ctime = mddev->ctime;
1138 sb->level = mddev->level;
1139 sb->size = mddev->dev_sectors / 2;
1140 sb->raid_disks = mddev->raid_disks;
1141 sb->md_minor = mddev->md_minor;
1142 sb->not_persistent = 0;
1143 sb->utime = mddev->utime;
1144 sb->state = 0;
1145 sb->events_hi = (mddev->events>>32);
1146 sb->events_lo = (u32)mddev->events;
1147
1148 if (mddev->reshape_position == MaxSector)
1149 sb->minor_version = 90;
1150 else {
1151 sb->minor_version = 91;
1152 sb->reshape_position = mddev->reshape_position;
1153 sb->new_level = mddev->new_level;
1154 sb->delta_disks = mddev->delta_disks;
1155 sb->new_layout = mddev->new_layout;
1156 sb->new_chunk = mddev->new_chunk_sectors << 9;
1157 }
1158 mddev->minor_version = sb->minor_version;
1159 if (mddev->in_sync)
1160 {
1161 sb->recovery_cp = mddev->recovery_cp;
1162 sb->cp_events_hi = (mddev->events>>32);
1163 sb->cp_events_lo = (u32)mddev->events;
1164 if (mddev->recovery_cp == MaxSector)
1165 sb->state = (1<< MD_SB_CLEAN);
1166 } else
1167 sb->recovery_cp = 0;
1168
1169 sb->layout = mddev->layout;
1170 sb->chunk_size = mddev->chunk_sectors << 9;
1171
1172 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1173 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1174
1175 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1176 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1177 mdp_disk_t *d;
1178 int desc_nr;
1179 int is_active = test_bit(In_sync, &rdev2->flags);
1180
1181 if (rdev2->raid_disk >= 0 &&
1182 sb->minor_version >= 91)
1183 /* we have nowhere to store the recovery_offset,
1184 * but if it is not below the reshape_position,
1185 * we can piggy-back on that.
1186 */
1187 is_active = 1;
1188 if (rdev2->raid_disk < 0 ||
1189 test_bit(Faulty, &rdev2->flags))
1190 is_active = 0;
1191 if (is_active)
1192 desc_nr = rdev2->raid_disk;
1193 else
1194 desc_nr = next_spare++;
1195 rdev2->desc_nr = desc_nr;
1196 d = &sb->disks[rdev2->desc_nr];
1197 nr_disks++;
1198 d->number = rdev2->desc_nr;
1199 d->major = MAJOR(rdev2->bdev->bd_dev);
1200 d->minor = MINOR(rdev2->bdev->bd_dev);
1201 if (is_active)
1202 d->raid_disk = rdev2->raid_disk;
1203 else
1204 d->raid_disk = rdev2->desc_nr; /* compatibility */
1205 if (test_bit(Faulty, &rdev2->flags))
1206 d->state = (1<<MD_DISK_FAULTY);
1207 else if (is_active) {
1208 d->state = (1<<MD_DISK_ACTIVE);
1209 if (test_bit(In_sync, &rdev2->flags))
1210 d->state |= (1<<MD_DISK_SYNC);
1211 active++;
1212 working++;
1213 } else {
1214 d->state = 0;
1215 spare++;
1216 working++;
1217 }
1218 if (test_bit(WriteMostly, &rdev2->flags))
1219 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1220 }
1221 /* now set the "removed" and "faulty" bits on any missing devices */
1222 for (i=0 ; i < mddev->raid_disks ; i++) {
1223 mdp_disk_t *d = &sb->disks[i];
1224 if (d->state == 0 && d->number == 0) {
1225 d->number = i;
1226 d->raid_disk = i;
1227 d->state = (1<<MD_DISK_REMOVED);
1228 d->state |= (1<<MD_DISK_FAULTY);
1229 failed++;
1230 }
1231 }
1232 sb->nr_disks = nr_disks;
1233 sb->active_disks = active;
1234 sb->working_disks = working;
1235 sb->failed_disks = failed;
1236 sb->spare_disks = spare;
1237
1238 sb->this_disk = sb->disks[rdev->desc_nr];
1239 sb->sb_csum = calc_sb_csum(sb);
1240 }
1241
1242 /*
1243 * rdev_size_change for 0.90.0
1244 */
1245 static unsigned long long
1246 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1247 {
1248 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1249 return 0; /* component must fit device */
1250 if (rdev->mddev->bitmap_info.offset)
1251 return 0; /* can't move bitmap */
1252 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1253 if (!num_sectors || num_sectors > rdev->sb_start)
1254 num_sectors = rdev->sb_start;
1255 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1256 rdev->sb_page);
1257 md_super_wait(rdev->mddev);
1258 return num_sectors / 2; /* kB for sysfs */
1259 }
1260
1261
1262 /*
1263 * version 1 superblock
1264 */
1265
1266 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1267 {
1268 __le32 disk_csum;
1269 u32 csum;
1270 unsigned long long newcsum;
1271 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1272 __le32 *isuper = (__le32*)sb;
1273 int i;
1274
1275 disk_csum = sb->sb_csum;
1276 sb->sb_csum = 0;
1277 newcsum = 0;
1278 for (i=0; size>=4; size -= 4 )
1279 newcsum += le32_to_cpu(*isuper++);
1280
1281 if (size == 2)
1282 newcsum += le16_to_cpu(*(__le16*) isuper);
1283
1284 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1285 sb->sb_csum = disk_csum;
1286 return cpu_to_le32(csum);
1287 }
1288
1289 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1290 {
1291 struct mdp_superblock_1 *sb;
1292 int ret;
1293 sector_t sb_start;
1294 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1295 int bmask;
1296
1297 /*
1298 * Calculate the position of the superblock in 512byte sectors.
1299 * It is always aligned to a 4K boundary and
1300 * depeding on minor_version, it can be:
1301 * 0: At least 8K, but less than 12K, from end of device
1302 * 1: At start of device
1303 * 2: 4K from start of device.
1304 */
1305 switch(minor_version) {
1306 case 0:
1307 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1308 sb_start -= 8*2;
1309 sb_start &= ~(sector_t)(4*2-1);
1310 break;
1311 case 1:
1312 sb_start = 0;
1313 break;
1314 case 2:
1315 sb_start = 8;
1316 break;
1317 default:
1318 return -EINVAL;
1319 }
1320 rdev->sb_start = sb_start;
1321
1322 /* superblock is rarely larger than 1K, but it can be larger,
1323 * and it is safe to read 4k, so we do that
1324 */
1325 ret = read_disk_sb(rdev, 4096);
1326 if (ret) return ret;
1327
1328
1329 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1330
1331 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1332 sb->major_version != cpu_to_le32(1) ||
1333 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1334 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1335 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1336 return -EINVAL;
1337
1338 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1339 printk("md: invalid superblock checksum on %s\n",
1340 bdevname(rdev->bdev,b));
1341 return -EINVAL;
1342 }
1343 if (le64_to_cpu(sb->data_size) < 10) {
1344 printk("md: data_size too small on %s\n",
1345 bdevname(rdev->bdev,b));
1346 return -EINVAL;
1347 }
1348
1349 rdev->preferred_minor = 0xffff;
1350 rdev->data_offset = le64_to_cpu(sb->data_offset);
1351 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1352
1353 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1354 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1355 if (rdev->sb_size & bmask)
1356 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1357
1358 if (minor_version
1359 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1360 return -EINVAL;
1361
1362 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1363 rdev->desc_nr = -1;
1364 else
1365 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1366
1367 if (!refdev) {
1368 ret = 1;
1369 } else {
1370 __u64 ev1, ev2;
1371 struct mdp_superblock_1 *refsb =
1372 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1373
1374 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1375 sb->level != refsb->level ||
1376 sb->layout != refsb->layout ||
1377 sb->chunksize != refsb->chunksize) {
1378 printk(KERN_WARNING "md: %s has strangely different"
1379 " superblock to %s\n",
1380 bdevname(rdev->bdev,b),
1381 bdevname(refdev->bdev,b2));
1382 return -EINVAL;
1383 }
1384 ev1 = le64_to_cpu(sb->events);
1385 ev2 = le64_to_cpu(refsb->events);
1386
1387 if (ev1 > ev2)
1388 ret = 1;
1389 else
1390 ret = 0;
1391 }
1392 if (minor_version)
1393 rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
1394 le64_to_cpu(sb->data_offset);
1395 else
1396 rdev->sectors = rdev->sb_start;
1397 if (rdev->sectors < le64_to_cpu(sb->data_size))
1398 return -EINVAL;
1399 rdev->sectors = le64_to_cpu(sb->data_size);
1400 if (le64_to_cpu(sb->size) > rdev->sectors)
1401 return -EINVAL;
1402 return ret;
1403 }
1404
1405 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1406 {
1407 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1408 __u64 ev1 = le64_to_cpu(sb->events);
1409
1410 rdev->raid_disk = -1;
1411 clear_bit(Faulty, &rdev->flags);
1412 clear_bit(In_sync, &rdev->flags);
1413 clear_bit(WriteMostly, &rdev->flags);
1414 clear_bit(BarriersNotsupp, &rdev->flags);
1415
1416 if (mddev->raid_disks == 0) {
1417 mddev->major_version = 1;
1418 mddev->patch_version = 0;
1419 mddev->external = 0;
1420 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1421 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1422 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1423 mddev->level = le32_to_cpu(sb->level);
1424 mddev->clevel[0] = 0;
1425 mddev->layout = le32_to_cpu(sb->layout);
1426 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1427 mddev->dev_sectors = le64_to_cpu(sb->size);
1428 mddev->events = ev1;
1429 mddev->bitmap_info.offset = 0;
1430 mddev->bitmap_info.default_offset = 1024 >> 9;
1431
1432 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1433 memcpy(mddev->uuid, sb->set_uuid, 16);
1434
1435 mddev->max_disks = (4096-256)/2;
1436
1437 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1438 mddev->bitmap_info.file == NULL )
1439 mddev->bitmap_info.offset =
1440 (__s32)le32_to_cpu(sb->bitmap_offset);
1441
1442 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1443 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1444 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1445 mddev->new_level = le32_to_cpu(sb->new_level);
1446 mddev->new_layout = le32_to_cpu(sb->new_layout);
1447 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1448 } else {
1449 mddev->reshape_position = MaxSector;
1450 mddev->delta_disks = 0;
1451 mddev->new_level = mddev->level;
1452 mddev->new_layout = mddev->layout;
1453 mddev->new_chunk_sectors = mddev->chunk_sectors;
1454 }
1455
1456 } else if (mddev->pers == NULL) {
1457 /* Insist of good event counter while assembling */
1458 ++ev1;
1459 if (ev1 < mddev->events)
1460 return -EINVAL;
1461 } else if (mddev->bitmap) {
1462 /* If adding to array with a bitmap, then we can accept an
1463 * older device, but not too old.
1464 */
1465 if (ev1 < mddev->bitmap->events_cleared)
1466 return 0;
1467 } else {
1468 if (ev1 < mddev->events)
1469 /* just a hot-add of a new device, leave raid_disk at -1 */
1470 return 0;
1471 }
1472 if (mddev->level != LEVEL_MULTIPATH) {
1473 int role;
1474 if (rdev->desc_nr < 0 ||
1475 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1476 role = 0xffff;
1477 rdev->desc_nr = -1;
1478 } else
1479 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1480 switch(role) {
1481 case 0xffff: /* spare */
1482 break;
1483 case 0xfffe: /* faulty */
1484 set_bit(Faulty, &rdev->flags);
1485 break;
1486 default:
1487 if ((le32_to_cpu(sb->feature_map) &
1488 MD_FEATURE_RECOVERY_OFFSET))
1489 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1490 else
1491 set_bit(In_sync, &rdev->flags);
1492 rdev->raid_disk = role;
1493 break;
1494 }
1495 if (sb->devflags & WriteMostly1)
1496 set_bit(WriteMostly, &rdev->flags);
1497 } else /* MULTIPATH are always insync */
1498 set_bit(In_sync, &rdev->flags);
1499
1500 return 0;
1501 }
1502
1503 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1504 {
1505 struct mdp_superblock_1 *sb;
1506 mdk_rdev_t *rdev2;
1507 int max_dev, i;
1508 /* make rdev->sb match mddev and rdev data. */
1509
1510 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1511
1512 sb->feature_map = 0;
1513 sb->pad0 = 0;
1514 sb->recovery_offset = cpu_to_le64(0);
1515 memset(sb->pad1, 0, sizeof(sb->pad1));
1516 memset(sb->pad2, 0, sizeof(sb->pad2));
1517 memset(sb->pad3, 0, sizeof(sb->pad3));
1518
1519 sb->utime = cpu_to_le64((__u64)mddev->utime);
1520 sb->events = cpu_to_le64(mddev->events);
1521 if (mddev->in_sync)
1522 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1523 else
1524 sb->resync_offset = cpu_to_le64(0);
1525
1526 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1527
1528 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1529 sb->size = cpu_to_le64(mddev->dev_sectors);
1530 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1531 sb->level = cpu_to_le32(mddev->level);
1532 sb->layout = cpu_to_le32(mddev->layout);
1533
1534 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1535 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1536 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1537 }
1538
1539 if (rdev->raid_disk >= 0 &&
1540 !test_bit(In_sync, &rdev->flags)) {
1541 sb->feature_map |=
1542 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1543 sb->recovery_offset =
1544 cpu_to_le64(rdev->recovery_offset);
1545 }
1546
1547 if (mddev->reshape_position != MaxSector) {
1548 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1549 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1550 sb->new_layout = cpu_to_le32(mddev->new_layout);
1551 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1552 sb->new_level = cpu_to_le32(mddev->new_level);
1553 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1554 }
1555
1556 max_dev = 0;
1557 list_for_each_entry(rdev2, &mddev->disks, same_set)
1558 if (rdev2->desc_nr+1 > max_dev)
1559 max_dev = rdev2->desc_nr+1;
1560
1561 if (max_dev > le32_to_cpu(sb->max_dev)) {
1562 int bmask;
1563 sb->max_dev = cpu_to_le32(max_dev);
1564 rdev->sb_size = max_dev * 2 + 256;
1565 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1566 if (rdev->sb_size & bmask)
1567 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1568 }
1569 for (i=0; i<max_dev;i++)
1570 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1571
1572 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1573 i = rdev2->desc_nr;
1574 if (test_bit(Faulty, &rdev2->flags))
1575 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1576 else if (test_bit(In_sync, &rdev2->flags))
1577 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1578 else if (rdev2->raid_disk >= 0)
1579 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1580 else
1581 sb->dev_roles[i] = cpu_to_le16(0xffff);
1582 }
1583
1584 sb->sb_csum = calc_sb_1_csum(sb);
1585 }
1586
1587 static unsigned long long
1588 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1589 {
1590 struct mdp_superblock_1 *sb;
1591 sector_t max_sectors;
1592 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1593 return 0; /* component must fit device */
1594 if (rdev->sb_start < rdev->data_offset) {
1595 /* minor versions 1 and 2; superblock before data */
1596 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1597 max_sectors -= rdev->data_offset;
1598 if (!num_sectors || num_sectors > max_sectors)
1599 num_sectors = max_sectors;
1600 } else if (rdev->mddev->bitmap_info.offset) {
1601 /* minor version 0 with bitmap we can't move */
1602 return 0;
1603 } else {
1604 /* minor version 0; superblock after data */
1605 sector_t sb_start;
1606 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1607 sb_start &= ~(sector_t)(4*2 - 1);
1608 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1609 if (!num_sectors || num_sectors > max_sectors)
1610 num_sectors = max_sectors;
1611 rdev->sb_start = sb_start;
1612 }
1613 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1614 sb->data_size = cpu_to_le64(num_sectors);
1615 sb->super_offset = rdev->sb_start;
1616 sb->sb_csum = calc_sb_1_csum(sb);
1617 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1618 rdev->sb_page);
1619 md_super_wait(rdev->mddev);
1620 return num_sectors / 2; /* kB for sysfs */
1621 }
1622
1623 static struct super_type super_types[] = {
1624 [0] = {
1625 .name = "0.90.0",
1626 .owner = THIS_MODULE,
1627 .load_super = super_90_load,
1628 .validate_super = super_90_validate,
1629 .sync_super = super_90_sync,
1630 .rdev_size_change = super_90_rdev_size_change,
1631 },
1632 [1] = {
1633 .name = "md-1",
1634 .owner = THIS_MODULE,
1635 .load_super = super_1_load,
1636 .validate_super = super_1_validate,
1637 .sync_super = super_1_sync,
1638 .rdev_size_change = super_1_rdev_size_change,
1639 },
1640 };
1641
1642 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1643 {
1644 mdk_rdev_t *rdev, *rdev2;
1645
1646 rcu_read_lock();
1647 rdev_for_each_rcu(rdev, mddev1)
1648 rdev_for_each_rcu(rdev2, mddev2)
1649 if (rdev->bdev->bd_contains ==
1650 rdev2->bdev->bd_contains) {
1651 rcu_read_unlock();
1652 return 1;
1653 }
1654 rcu_read_unlock();
1655 return 0;
1656 }
1657
1658 static LIST_HEAD(pending_raid_disks);
1659
1660 /*
1661 * Try to register data integrity profile for an mddev
1662 *
1663 * This is called when an array is started and after a disk has been kicked
1664 * from the array. It only succeeds if all working and active component devices
1665 * are integrity capable with matching profiles.
1666 */
1667 int md_integrity_register(mddev_t *mddev)
1668 {
1669 mdk_rdev_t *rdev, *reference = NULL;
1670
1671 if (list_empty(&mddev->disks))
1672 return 0; /* nothing to do */
1673 if (blk_get_integrity(mddev->gendisk))
1674 return 0; /* already registered */
1675 list_for_each_entry(rdev, &mddev->disks, same_set) {
1676 /* skip spares and non-functional disks */
1677 if (test_bit(Faulty, &rdev->flags))
1678 continue;
1679 if (rdev->raid_disk < 0)
1680 continue;
1681 /*
1682 * If at least one rdev is not integrity capable, we can not
1683 * enable data integrity for the md device.
1684 */
1685 if (!bdev_get_integrity(rdev->bdev))
1686 return -EINVAL;
1687 if (!reference) {
1688 /* Use the first rdev as the reference */
1689 reference = rdev;
1690 continue;
1691 }
1692 /* does this rdev's profile match the reference profile? */
1693 if (blk_integrity_compare(reference->bdev->bd_disk,
1694 rdev->bdev->bd_disk) < 0)
1695 return -EINVAL;
1696 }
1697 /*
1698 * All component devices are integrity capable and have matching
1699 * profiles, register the common profile for the md device.
1700 */
1701 if (blk_integrity_register(mddev->gendisk,
1702 bdev_get_integrity(reference->bdev)) != 0) {
1703 printk(KERN_ERR "md: failed to register integrity for %s\n",
1704 mdname(mddev));
1705 return -EINVAL;
1706 }
1707 printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1708 mdname(mddev));
1709 return 0;
1710 }
1711 EXPORT_SYMBOL(md_integrity_register);
1712
1713 /* Disable data integrity if non-capable/non-matching disk is being added */
1714 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
1715 {
1716 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1717 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1718
1719 if (!bi_mddev) /* nothing to do */
1720 return;
1721 if (rdev->raid_disk < 0) /* skip spares */
1722 return;
1723 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1724 rdev->bdev->bd_disk) >= 0)
1725 return;
1726 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1727 blk_integrity_unregister(mddev->gendisk);
1728 }
1729 EXPORT_SYMBOL(md_integrity_add_rdev);
1730
1731 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1732 {
1733 char b[BDEVNAME_SIZE];
1734 struct kobject *ko;
1735 char *s;
1736 int err;
1737
1738 if (rdev->mddev) {
1739 MD_BUG();
1740 return -EINVAL;
1741 }
1742
1743 /* prevent duplicates */
1744 if (find_rdev(mddev, rdev->bdev->bd_dev))
1745 return -EEXIST;
1746
1747 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1748 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1749 rdev->sectors < mddev->dev_sectors)) {
1750 if (mddev->pers) {
1751 /* Cannot change size, so fail
1752 * If mddev->level <= 0, then we don't care
1753 * about aligning sizes (e.g. linear)
1754 */
1755 if (mddev->level > 0)
1756 return -ENOSPC;
1757 } else
1758 mddev->dev_sectors = rdev->sectors;
1759 }
1760
1761 /* Verify rdev->desc_nr is unique.
1762 * If it is -1, assign a free number, else
1763 * check number is not in use
1764 */
1765 if (rdev->desc_nr < 0) {
1766 int choice = 0;
1767 if (mddev->pers) choice = mddev->raid_disks;
1768 while (find_rdev_nr(mddev, choice))
1769 choice++;
1770 rdev->desc_nr = choice;
1771 } else {
1772 if (find_rdev_nr(mddev, rdev->desc_nr))
1773 return -EBUSY;
1774 }
1775 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1776 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1777 mdname(mddev), mddev->max_disks);
1778 return -EBUSY;
1779 }
1780 bdevname(rdev->bdev,b);
1781 while ( (s=strchr(b, '/')) != NULL)
1782 *s = '!';
1783
1784 rdev->mddev = mddev;
1785 printk(KERN_INFO "md: bind<%s>\n", b);
1786
1787 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1788 goto fail;
1789
1790 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1791 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1792 kobject_del(&rdev->kobj);
1793 goto fail;
1794 }
1795 rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, "state");
1796
1797 list_add_rcu(&rdev->same_set, &mddev->disks);
1798 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1799
1800 /* May as well allow recovery to be retried once */
1801 mddev->recovery_disabled = 0;
1802
1803 return 0;
1804
1805 fail:
1806 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1807 b, mdname(mddev));
1808 return err;
1809 }
1810
1811 static void md_delayed_delete(struct work_struct *ws)
1812 {
1813 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1814 kobject_del(&rdev->kobj);
1815 kobject_put(&rdev->kobj);
1816 }
1817
1818 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1819 {
1820 char b[BDEVNAME_SIZE];
1821 if (!rdev->mddev) {
1822 MD_BUG();
1823 return;
1824 }
1825 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1826 list_del_rcu(&rdev->same_set);
1827 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1828 rdev->mddev = NULL;
1829 sysfs_remove_link(&rdev->kobj, "block");
1830 sysfs_put(rdev->sysfs_state);
1831 rdev->sysfs_state = NULL;
1832 /* We need to delay this, otherwise we can deadlock when
1833 * writing to 'remove' to "dev/state". We also need
1834 * to delay it due to rcu usage.
1835 */
1836 synchronize_rcu();
1837 INIT_WORK(&rdev->del_work, md_delayed_delete);
1838 kobject_get(&rdev->kobj);
1839 schedule_work(&rdev->del_work);
1840 }
1841
1842 /*
1843 * prevent the device from being mounted, repartitioned or
1844 * otherwise reused by a RAID array (or any other kernel
1845 * subsystem), by bd_claiming the device.
1846 */
1847 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1848 {
1849 int err = 0;
1850 struct block_device *bdev;
1851 char b[BDEVNAME_SIZE];
1852
1853 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1854 if (IS_ERR(bdev)) {
1855 printk(KERN_ERR "md: could not open %s.\n",
1856 __bdevname(dev, b));
1857 return PTR_ERR(bdev);
1858 }
1859 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1860 if (err) {
1861 printk(KERN_ERR "md: could not bd_claim %s.\n",
1862 bdevname(bdev, b));
1863 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1864 return err;
1865 }
1866 if (!shared)
1867 set_bit(AllReserved, &rdev->flags);
1868 rdev->bdev = bdev;
1869 return err;
1870 }
1871
1872 static void unlock_rdev(mdk_rdev_t *rdev)
1873 {
1874 struct block_device *bdev = rdev->bdev;
1875 rdev->bdev = NULL;
1876 if (!bdev)
1877 MD_BUG();
1878 bd_release(bdev);
1879 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1880 }
1881
1882 void md_autodetect_dev(dev_t dev);
1883
1884 static void export_rdev(mdk_rdev_t * rdev)
1885 {
1886 char b[BDEVNAME_SIZE];
1887 printk(KERN_INFO "md: export_rdev(%s)\n",
1888 bdevname(rdev->bdev,b));
1889 if (rdev->mddev)
1890 MD_BUG();
1891 free_disk_sb(rdev);
1892 #ifndef MODULE
1893 if (test_bit(AutoDetected, &rdev->flags))
1894 md_autodetect_dev(rdev->bdev->bd_dev);
1895 #endif
1896 unlock_rdev(rdev);
1897 kobject_put(&rdev->kobj);
1898 }
1899
1900 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1901 {
1902 unbind_rdev_from_array(rdev);
1903 export_rdev(rdev);
1904 }
1905
1906 static void export_array(mddev_t *mddev)
1907 {
1908 mdk_rdev_t *rdev, *tmp;
1909
1910 rdev_for_each(rdev, tmp, mddev) {
1911 if (!rdev->mddev) {
1912 MD_BUG();
1913 continue;
1914 }
1915 kick_rdev_from_array(rdev);
1916 }
1917 if (!list_empty(&mddev->disks))
1918 MD_BUG();
1919 mddev->raid_disks = 0;
1920 mddev->major_version = 0;
1921 }
1922
1923 static void print_desc(mdp_disk_t *desc)
1924 {
1925 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1926 desc->major,desc->minor,desc->raid_disk,desc->state);
1927 }
1928
1929 static void print_sb_90(mdp_super_t *sb)
1930 {
1931 int i;
1932
1933 printk(KERN_INFO
1934 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1935 sb->major_version, sb->minor_version, sb->patch_version,
1936 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1937 sb->ctime);
1938 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1939 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1940 sb->md_minor, sb->layout, sb->chunk_size);
1941 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1942 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1943 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1944 sb->failed_disks, sb->spare_disks,
1945 sb->sb_csum, (unsigned long)sb->events_lo);
1946
1947 printk(KERN_INFO);
1948 for (i = 0; i < MD_SB_DISKS; i++) {
1949 mdp_disk_t *desc;
1950
1951 desc = sb->disks + i;
1952 if (desc->number || desc->major || desc->minor ||
1953 desc->raid_disk || (desc->state && (desc->state != 4))) {
1954 printk(" D %2d: ", i);
1955 print_desc(desc);
1956 }
1957 }
1958 printk(KERN_INFO "md: THIS: ");
1959 print_desc(&sb->this_disk);
1960 }
1961
1962 static void print_sb_1(struct mdp_superblock_1 *sb)
1963 {
1964 __u8 *uuid;
1965
1966 uuid = sb->set_uuid;
1967 printk(KERN_INFO
1968 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
1969 "md: Name: \"%s\" CT:%llu\n",
1970 le32_to_cpu(sb->major_version),
1971 le32_to_cpu(sb->feature_map),
1972 uuid,
1973 sb->set_name,
1974 (unsigned long long)le64_to_cpu(sb->ctime)
1975 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
1976
1977 uuid = sb->device_uuid;
1978 printk(KERN_INFO
1979 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
1980 " RO:%llu\n"
1981 "md: Dev:%08x UUID: %pU\n"
1982 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
1983 "md: (MaxDev:%u) \n",
1984 le32_to_cpu(sb->level),
1985 (unsigned long long)le64_to_cpu(sb->size),
1986 le32_to_cpu(sb->raid_disks),
1987 le32_to_cpu(sb->layout),
1988 le32_to_cpu(sb->chunksize),
1989 (unsigned long long)le64_to_cpu(sb->data_offset),
1990 (unsigned long long)le64_to_cpu(sb->data_size),
1991 (unsigned long long)le64_to_cpu(sb->super_offset),
1992 (unsigned long long)le64_to_cpu(sb->recovery_offset),
1993 le32_to_cpu(sb->dev_number),
1994 uuid,
1995 sb->devflags,
1996 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
1997 (unsigned long long)le64_to_cpu(sb->events),
1998 (unsigned long long)le64_to_cpu(sb->resync_offset),
1999 le32_to_cpu(sb->sb_csum),
2000 le32_to_cpu(sb->max_dev)
2001 );
2002 }
2003
2004 static void print_rdev(mdk_rdev_t *rdev, int major_version)
2005 {
2006 char b[BDEVNAME_SIZE];
2007 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2008 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2009 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2010 rdev->desc_nr);
2011 if (rdev->sb_loaded) {
2012 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2013 switch (major_version) {
2014 case 0:
2015 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
2016 break;
2017 case 1:
2018 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
2019 break;
2020 }
2021 } else
2022 printk(KERN_INFO "md: no rdev superblock!\n");
2023 }
2024
2025 static void md_print_devices(void)
2026 {
2027 struct list_head *tmp;
2028 mdk_rdev_t *rdev;
2029 mddev_t *mddev;
2030 char b[BDEVNAME_SIZE];
2031
2032 printk("\n");
2033 printk("md: **********************************\n");
2034 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2035 printk("md: **********************************\n");
2036 for_each_mddev(mddev, tmp) {
2037
2038 if (mddev->bitmap)
2039 bitmap_print_sb(mddev->bitmap);
2040 else
2041 printk("%s: ", mdname(mddev));
2042 list_for_each_entry(rdev, &mddev->disks, same_set)
2043 printk("<%s>", bdevname(rdev->bdev,b));
2044 printk("\n");
2045
2046 list_for_each_entry(rdev, &mddev->disks, same_set)
2047 print_rdev(rdev, mddev->major_version);
2048 }
2049 printk("md: **********************************\n");
2050 printk("\n");
2051 }
2052
2053
2054 static void sync_sbs(mddev_t * mddev, int nospares)
2055 {
2056 /* Update each superblock (in-memory image), but
2057 * if we are allowed to, skip spares which already
2058 * have the right event counter, or have one earlier
2059 * (which would mean they aren't being marked as dirty
2060 * with the rest of the array)
2061 */
2062 mdk_rdev_t *rdev;
2063
2064 /* First make sure individual recovery_offsets are correct */
2065 list_for_each_entry(rdev, &mddev->disks, same_set) {
2066 if (rdev->raid_disk >= 0 &&
2067 !test_bit(In_sync, &rdev->flags) &&
2068 mddev->curr_resync_completed > rdev->recovery_offset)
2069 rdev->recovery_offset = mddev->curr_resync_completed;
2070
2071 }
2072 list_for_each_entry(rdev, &mddev->disks, same_set) {
2073 if (rdev->sb_events == mddev->events ||
2074 (nospares &&
2075 rdev->raid_disk < 0 &&
2076 (rdev->sb_events&1)==0 &&
2077 rdev->sb_events+1 == mddev->events)) {
2078 /* Don't update this superblock */
2079 rdev->sb_loaded = 2;
2080 } else {
2081 super_types[mddev->major_version].
2082 sync_super(mddev, rdev);
2083 rdev->sb_loaded = 1;
2084 }
2085 }
2086 }
2087
2088 static void md_update_sb(mddev_t * mddev, int force_change)
2089 {
2090 mdk_rdev_t *rdev;
2091 int sync_req;
2092 int nospares = 0;
2093
2094 mddev->utime = get_seconds();
2095 if (mddev->external)
2096 return;
2097 repeat:
2098 spin_lock_irq(&mddev->write_lock);
2099
2100 set_bit(MD_CHANGE_PENDING, &mddev->flags);
2101 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2102 force_change = 1;
2103 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2104 /* just a clean<-> dirty transition, possibly leave spares alone,
2105 * though if events isn't the right even/odd, we will have to do
2106 * spares after all
2107 */
2108 nospares = 1;
2109 if (force_change)
2110 nospares = 0;
2111 if (mddev->degraded)
2112 /* If the array is degraded, then skipping spares is both
2113 * dangerous and fairly pointless.
2114 * Dangerous because a device that was removed from the array
2115 * might have a event_count that still looks up-to-date,
2116 * so it can be re-added without a resync.
2117 * Pointless because if there are any spares to skip,
2118 * then a recovery will happen and soon that array won't
2119 * be degraded any more and the spare can go back to sleep then.
2120 */
2121 nospares = 0;
2122
2123 sync_req = mddev->in_sync;
2124
2125 /* If this is just a dirty<->clean transition, and the array is clean
2126 * and 'events' is odd, we can roll back to the previous clean state */
2127 if (nospares
2128 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2129 && (mddev->events & 1)
2130 && mddev->events != 1)
2131 mddev->events--;
2132 else {
2133 /* otherwise we have to go forward and ... */
2134 mddev->events ++;
2135 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
2136 /* .. if the array isn't clean, an 'even' event must also go
2137 * to spares. */
2138 if ((mddev->events&1)==0)
2139 nospares = 0;
2140 } else {
2141 /* otherwise an 'odd' event must go to spares */
2142 if ((mddev->events&1))
2143 nospares = 0;
2144 }
2145 }
2146
2147 if (!mddev->events) {
2148 /*
2149 * oops, this 64-bit counter should never wrap.
2150 * Either we are in around ~1 trillion A.C., assuming
2151 * 1 reboot per second, or we have a bug:
2152 */
2153 MD_BUG();
2154 mddev->events --;
2155 }
2156
2157 /*
2158 * do not write anything to disk if using
2159 * nonpersistent superblocks
2160 */
2161 if (!mddev->persistent) {
2162 if (!mddev->external)
2163 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2164
2165 spin_unlock_irq(&mddev->write_lock);
2166 wake_up(&mddev->sb_wait);
2167 return;
2168 }
2169 sync_sbs(mddev, nospares);
2170 spin_unlock_irq(&mddev->write_lock);
2171
2172 dprintk(KERN_INFO
2173 "md: updating %s RAID superblock on device (in sync %d)\n",
2174 mdname(mddev),mddev->in_sync);
2175
2176 bitmap_update_sb(mddev->bitmap);
2177 list_for_each_entry(rdev, &mddev->disks, same_set) {
2178 char b[BDEVNAME_SIZE];
2179 dprintk(KERN_INFO "md: ");
2180 if (rdev->sb_loaded != 1)
2181 continue; /* no noise on spare devices */
2182 if (test_bit(Faulty, &rdev->flags))
2183 dprintk("(skipping faulty ");
2184
2185 dprintk("%s ", bdevname(rdev->bdev,b));
2186 if (!test_bit(Faulty, &rdev->flags)) {
2187 md_super_write(mddev,rdev,
2188 rdev->sb_start, rdev->sb_size,
2189 rdev->sb_page);
2190 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2191 bdevname(rdev->bdev,b),
2192 (unsigned long long)rdev->sb_start);
2193 rdev->sb_events = mddev->events;
2194
2195 } else
2196 dprintk(")\n");
2197 if (mddev->level == LEVEL_MULTIPATH)
2198 /* only need to write one superblock... */
2199 break;
2200 }
2201 md_super_wait(mddev);
2202 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2203
2204 spin_lock_irq(&mddev->write_lock);
2205 if (mddev->in_sync != sync_req ||
2206 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2207 /* have to write it out again */
2208 spin_unlock_irq(&mddev->write_lock);
2209 goto repeat;
2210 }
2211 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2212 spin_unlock_irq(&mddev->write_lock);
2213 wake_up(&mddev->sb_wait);
2214 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2215 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2216
2217 }
2218
2219 /* words written to sysfs files may, or may not, be \n terminated.
2220 * We want to accept with case. For this we use cmd_match.
2221 */
2222 static int cmd_match(const char *cmd, const char *str)
2223 {
2224 /* See if cmd, written into a sysfs file, matches
2225 * str. They must either be the same, or cmd can
2226 * have a trailing newline
2227 */
2228 while (*cmd && *str && *cmd == *str) {
2229 cmd++;
2230 str++;
2231 }
2232 if (*cmd == '\n')
2233 cmd++;
2234 if (*str || *cmd)
2235 return 0;
2236 return 1;
2237 }
2238
2239 struct rdev_sysfs_entry {
2240 struct attribute attr;
2241 ssize_t (*show)(mdk_rdev_t *, char *);
2242 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2243 };
2244
2245 static ssize_t
2246 state_show(mdk_rdev_t *rdev, char *page)
2247 {
2248 char *sep = "";
2249 size_t len = 0;
2250
2251 if (test_bit(Faulty, &rdev->flags)) {
2252 len+= sprintf(page+len, "%sfaulty",sep);
2253 sep = ",";
2254 }
2255 if (test_bit(In_sync, &rdev->flags)) {
2256 len += sprintf(page+len, "%sin_sync",sep);
2257 sep = ",";
2258 }
2259 if (test_bit(WriteMostly, &rdev->flags)) {
2260 len += sprintf(page+len, "%swrite_mostly",sep);
2261 sep = ",";
2262 }
2263 if (test_bit(Blocked, &rdev->flags)) {
2264 len += sprintf(page+len, "%sblocked", sep);
2265 sep = ",";
2266 }
2267 if (!test_bit(Faulty, &rdev->flags) &&
2268 !test_bit(In_sync, &rdev->flags)) {
2269 len += sprintf(page+len, "%sspare", sep);
2270 sep = ",";
2271 }
2272 return len+sprintf(page+len, "\n");
2273 }
2274
2275 static ssize_t
2276 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2277 {
2278 /* can write
2279 * faulty - simulates and error
2280 * remove - disconnects the device
2281 * writemostly - sets write_mostly
2282 * -writemostly - clears write_mostly
2283 * blocked - sets the Blocked flag
2284 * -blocked - clears the Blocked flag
2285 * insync - sets Insync providing device isn't active
2286 */
2287 int err = -EINVAL;
2288 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2289 md_error(rdev->mddev, rdev);
2290 err = 0;
2291 } else if (cmd_match(buf, "remove")) {
2292 if (rdev->raid_disk >= 0)
2293 err = -EBUSY;
2294 else {
2295 mddev_t *mddev = rdev->mddev;
2296 kick_rdev_from_array(rdev);
2297 if (mddev->pers)
2298 md_update_sb(mddev, 1);
2299 md_new_event(mddev);
2300 err = 0;
2301 }
2302 } else if (cmd_match(buf, "writemostly")) {
2303 set_bit(WriteMostly, &rdev->flags);
2304 err = 0;
2305 } else if (cmd_match(buf, "-writemostly")) {
2306 clear_bit(WriteMostly, &rdev->flags);
2307 err = 0;
2308 } else if (cmd_match(buf, "blocked")) {
2309 set_bit(Blocked, &rdev->flags);
2310 err = 0;
2311 } else if (cmd_match(buf, "-blocked")) {
2312 clear_bit(Blocked, &rdev->flags);
2313 wake_up(&rdev->blocked_wait);
2314 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2315 md_wakeup_thread(rdev->mddev->thread);
2316
2317 err = 0;
2318 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2319 set_bit(In_sync, &rdev->flags);
2320 err = 0;
2321 }
2322 if (!err && rdev->sysfs_state)
2323 sysfs_notify_dirent(rdev->sysfs_state);
2324 return err ? err : len;
2325 }
2326 static struct rdev_sysfs_entry rdev_state =
2327 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2328
2329 static ssize_t
2330 errors_show(mdk_rdev_t *rdev, char *page)
2331 {
2332 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2333 }
2334
2335 static ssize_t
2336 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2337 {
2338 char *e;
2339 unsigned long n = simple_strtoul(buf, &e, 10);
2340 if (*buf && (*e == 0 || *e == '\n')) {
2341 atomic_set(&rdev->corrected_errors, n);
2342 return len;
2343 }
2344 return -EINVAL;
2345 }
2346 static struct rdev_sysfs_entry rdev_errors =
2347 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2348
2349 static ssize_t
2350 slot_show(mdk_rdev_t *rdev, char *page)
2351 {
2352 if (rdev->raid_disk < 0)
2353 return sprintf(page, "none\n");
2354 else
2355 return sprintf(page, "%d\n", rdev->raid_disk);
2356 }
2357
2358 static ssize_t
2359 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2360 {
2361 char *e;
2362 int err;
2363 char nm[20];
2364 int slot = simple_strtoul(buf, &e, 10);
2365 if (strncmp(buf, "none", 4)==0)
2366 slot = -1;
2367 else if (e==buf || (*e && *e!= '\n'))
2368 return -EINVAL;
2369 if (rdev->mddev->pers && slot == -1) {
2370 /* Setting 'slot' on an active array requires also
2371 * updating the 'rd%d' link, and communicating
2372 * with the personality with ->hot_*_disk.
2373 * For now we only support removing
2374 * failed/spare devices. This normally happens automatically,
2375 * but not when the metadata is externally managed.
2376 */
2377 if (rdev->raid_disk == -1)
2378 return -EEXIST;
2379 /* personality does all needed checks */
2380 if (rdev->mddev->pers->hot_add_disk == NULL)
2381 return -EINVAL;
2382 err = rdev->mddev->pers->
2383 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2384 if (err)
2385 return err;
2386 sprintf(nm, "rd%d", rdev->raid_disk);
2387 sysfs_remove_link(&rdev->mddev->kobj, nm);
2388 rdev->raid_disk = -1;
2389 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2390 md_wakeup_thread(rdev->mddev->thread);
2391 } else if (rdev->mddev->pers) {
2392 mdk_rdev_t *rdev2;
2393 /* Activating a spare .. or possibly reactivating
2394 * if we ever get bitmaps working here.
2395 */
2396
2397 if (rdev->raid_disk != -1)
2398 return -EBUSY;
2399
2400 if (rdev->mddev->pers->hot_add_disk == NULL)
2401 return -EINVAL;
2402
2403 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2404 if (rdev2->raid_disk == slot)
2405 return -EEXIST;
2406
2407 rdev->raid_disk = slot;
2408 if (test_bit(In_sync, &rdev->flags))
2409 rdev->saved_raid_disk = slot;
2410 else
2411 rdev->saved_raid_disk = -1;
2412 err = rdev->mddev->pers->
2413 hot_add_disk(rdev->mddev, rdev);
2414 if (err) {
2415 rdev->raid_disk = -1;
2416 return err;
2417 } else
2418 sysfs_notify_dirent(rdev->sysfs_state);
2419 sprintf(nm, "rd%d", rdev->raid_disk);
2420 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2421 printk(KERN_WARNING
2422 "md: cannot register "
2423 "%s for %s\n",
2424 nm, mdname(rdev->mddev));
2425
2426 /* don't wakeup anyone, leave that to userspace. */
2427 } else {
2428 if (slot >= rdev->mddev->raid_disks)
2429 return -ENOSPC;
2430 rdev->raid_disk = slot;
2431 /* assume it is working */
2432 clear_bit(Faulty, &rdev->flags);
2433 clear_bit(WriteMostly, &rdev->flags);
2434 set_bit(In_sync, &rdev->flags);
2435 sysfs_notify_dirent(rdev->sysfs_state);
2436 }
2437 return len;
2438 }
2439
2440
2441 static struct rdev_sysfs_entry rdev_slot =
2442 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2443
2444 static ssize_t
2445 offset_show(mdk_rdev_t *rdev, char *page)
2446 {
2447 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2448 }
2449
2450 static ssize_t
2451 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2452 {
2453 char *e;
2454 unsigned long long offset = simple_strtoull(buf, &e, 10);
2455 if (e==buf || (*e && *e != '\n'))
2456 return -EINVAL;
2457 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2458 return -EBUSY;
2459 if (rdev->sectors && rdev->mddev->external)
2460 /* Must set offset before size, so overlap checks
2461 * can be sane */
2462 return -EBUSY;
2463 rdev->data_offset = offset;
2464 return len;
2465 }
2466
2467 static struct rdev_sysfs_entry rdev_offset =
2468 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2469
2470 static ssize_t
2471 rdev_size_show(mdk_rdev_t *rdev, char *page)
2472 {
2473 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2474 }
2475
2476 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2477 {
2478 /* check if two start/length pairs overlap */
2479 if (s1+l1 <= s2)
2480 return 0;
2481 if (s2+l2 <= s1)
2482 return 0;
2483 return 1;
2484 }
2485
2486 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2487 {
2488 unsigned long long blocks;
2489 sector_t new;
2490
2491 if (strict_strtoull(buf, 10, &blocks) < 0)
2492 return -EINVAL;
2493
2494 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2495 return -EINVAL; /* sector conversion overflow */
2496
2497 new = blocks * 2;
2498 if (new != blocks * 2)
2499 return -EINVAL; /* unsigned long long to sector_t overflow */
2500
2501 *sectors = new;
2502 return 0;
2503 }
2504
2505 static ssize_t
2506 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2507 {
2508 mddev_t *my_mddev = rdev->mddev;
2509 sector_t oldsectors = rdev->sectors;
2510 sector_t sectors;
2511
2512 if (strict_blocks_to_sectors(buf, &sectors) < 0)
2513 return -EINVAL;
2514 if (my_mddev->pers && rdev->raid_disk >= 0) {
2515 if (my_mddev->persistent) {
2516 sectors = super_types[my_mddev->major_version].
2517 rdev_size_change(rdev, sectors);
2518 if (!sectors)
2519 return -EBUSY;
2520 } else if (!sectors)
2521 sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2522 rdev->data_offset;
2523 }
2524 if (sectors < my_mddev->dev_sectors)
2525 return -EINVAL; /* component must fit device */
2526
2527 rdev->sectors = sectors;
2528 if (sectors > oldsectors && my_mddev->external) {
2529 /* need to check that all other rdevs with the same ->bdev
2530 * do not overlap. We need to unlock the mddev to avoid
2531 * a deadlock. We have already changed rdev->sectors, and if
2532 * we have to change it back, we will have the lock again.
2533 */
2534 mddev_t *mddev;
2535 int overlap = 0;
2536 struct list_head *tmp;
2537
2538 mddev_unlock(my_mddev);
2539 for_each_mddev(mddev, tmp) {
2540 mdk_rdev_t *rdev2;
2541
2542 mddev_lock(mddev);
2543 list_for_each_entry(rdev2, &mddev->disks, same_set)
2544 if (test_bit(AllReserved, &rdev2->flags) ||
2545 (rdev->bdev == rdev2->bdev &&
2546 rdev != rdev2 &&
2547 overlaps(rdev->data_offset, rdev->sectors,
2548 rdev2->data_offset,
2549 rdev2->sectors))) {
2550 overlap = 1;
2551 break;
2552 }
2553 mddev_unlock(mddev);
2554 if (overlap) {
2555 mddev_put(mddev);
2556 break;
2557 }
2558 }
2559 mddev_lock(my_mddev);
2560 if (overlap) {
2561 /* Someone else could have slipped in a size
2562 * change here, but doing so is just silly.
2563 * We put oldsectors back because we *know* it is
2564 * safe, and trust userspace not to race with
2565 * itself
2566 */
2567 rdev->sectors = oldsectors;
2568 return -EBUSY;
2569 }
2570 }
2571 return len;
2572 }
2573
2574 static struct rdev_sysfs_entry rdev_size =
2575 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2576
2577
2578 static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page)
2579 {
2580 unsigned long long recovery_start = rdev->recovery_offset;
2581
2582 if (test_bit(In_sync, &rdev->flags) ||
2583 recovery_start == MaxSector)
2584 return sprintf(page, "none\n");
2585
2586 return sprintf(page, "%llu\n", recovery_start);
2587 }
2588
2589 static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2590 {
2591 unsigned long long recovery_start;
2592
2593 if (cmd_match(buf, "none"))
2594 recovery_start = MaxSector;
2595 else if (strict_strtoull(buf, 10, &recovery_start))
2596 return -EINVAL;
2597
2598 if (rdev->mddev->pers &&
2599 rdev->raid_disk >= 0)
2600 return -EBUSY;
2601
2602 rdev->recovery_offset = recovery_start;
2603 if (recovery_start == MaxSector)
2604 set_bit(In_sync, &rdev->flags);
2605 else
2606 clear_bit(In_sync, &rdev->flags);
2607 return len;
2608 }
2609
2610 static struct rdev_sysfs_entry rdev_recovery_start =
2611 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2612
2613 static struct attribute *rdev_default_attrs[] = {
2614 &rdev_state.attr,
2615 &rdev_errors.attr,
2616 &rdev_slot.attr,
2617 &rdev_offset.attr,
2618 &rdev_size.attr,
2619 &rdev_recovery_start.attr,
2620 NULL,
2621 };
2622 static ssize_t
2623 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2624 {
2625 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2626 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2627 mddev_t *mddev = rdev->mddev;
2628 ssize_t rv;
2629
2630 if (!entry->show)
2631 return -EIO;
2632
2633 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2634 if (!rv) {
2635 if (rdev->mddev == NULL)
2636 rv = -EBUSY;
2637 else
2638 rv = entry->show(rdev, page);
2639 mddev_unlock(mddev);
2640 }
2641 return rv;
2642 }
2643
2644 static ssize_t
2645 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2646 const char *page, size_t length)
2647 {
2648 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2649 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2650 ssize_t rv;
2651 mddev_t *mddev = rdev->mddev;
2652
2653 if (!entry->store)
2654 return -EIO;
2655 if (!capable(CAP_SYS_ADMIN))
2656 return -EACCES;
2657 rv = mddev ? mddev_lock(mddev): -EBUSY;
2658 if (!rv) {
2659 if (rdev->mddev == NULL)
2660 rv = -EBUSY;
2661 else
2662 rv = entry->store(rdev, page, length);
2663 mddev_unlock(mddev);
2664 }
2665 return rv;
2666 }
2667
2668 static void rdev_free(struct kobject *ko)
2669 {
2670 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2671 kfree(rdev);
2672 }
2673 static struct sysfs_ops rdev_sysfs_ops = {
2674 .show = rdev_attr_show,
2675 .store = rdev_attr_store,
2676 };
2677 static struct kobj_type rdev_ktype = {
2678 .release = rdev_free,
2679 .sysfs_ops = &rdev_sysfs_ops,
2680 .default_attrs = rdev_default_attrs,
2681 };
2682
2683 /*
2684 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2685 *
2686 * mark the device faulty if:
2687 *
2688 * - the device is nonexistent (zero size)
2689 * - the device has no valid superblock
2690 *
2691 * a faulty rdev _never_ has rdev->sb set.
2692 */
2693 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2694 {
2695 char b[BDEVNAME_SIZE];
2696 int err;
2697 mdk_rdev_t *rdev;
2698 sector_t size;
2699
2700 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2701 if (!rdev) {
2702 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2703 return ERR_PTR(-ENOMEM);
2704 }
2705
2706 if ((err = alloc_disk_sb(rdev)))
2707 goto abort_free;
2708
2709 err = lock_rdev(rdev, newdev, super_format == -2);
2710 if (err)
2711 goto abort_free;
2712
2713 kobject_init(&rdev->kobj, &rdev_ktype);
2714
2715 rdev->desc_nr = -1;
2716 rdev->saved_raid_disk = -1;
2717 rdev->raid_disk = -1;
2718 rdev->flags = 0;
2719 rdev->data_offset = 0;
2720 rdev->sb_events = 0;
2721 rdev->last_read_error.tv_sec = 0;
2722 rdev->last_read_error.tv_nsec = 0;
2723 atomic_set(&rdev->nr_pending, 0);
2724 atomic_set(&rdev->read_errors, 0);
2725 atomic_set(&rdev->corrected_errors, 0);
2726
2727 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2728 if (!size) {
2729 printk(KERN_WARNING
2730 "md: %s has zero or unknown size, marking faulty!\n",
2731 bdevname(rdev->bdev,b));
2732 err = -EINVAL;
2733 goto abort_free;
2734 }
2735
2736 if (super_format >= 0) {
2737 err = super_types[super_format].
2738 load_super(rdev, NULL, super_minor);
2739 if (err == -EINVAL) {
2740 printk(KERN_WARNING
2741 "md: %s does not have a valid v%d.%d "
2742 "superblock, not importing!\n",
2743 bdevname(rdev->bdev,b),
2744 super_format, super_minor);
2745 goto abort_free;
2746 }
2747 if (err < 0) {
2748 printk(KERN_WARNING
2749 "md: could not read %s's sb, not importing!\n",
2750 bdevname(rdev->bdev,b));
2751 goto abort_free;
2752 }
2753 }
2754
2755 INIT_LIST_HEAD(&rdev->same_set);
2756 init_waitqueue_head(&rdev->blocked_wait);
2757
2758 return rdev;
2759
2760 abort_free:
2761 if (rdev->sb_page) {
2762 if (rdev->bdev)
2763 unlock_rdev(rdev);
2764 free_disk_sb(rdev);
2765 }
2766 kfree(rdev);
2767 return ERR_PTR(err);
2768 }
2769
2770 /*
2771 * Check a full RAID array for plausibility
2772 */
2773
2774
2775 static void analyze_sbs(mddev_t * mddev)
2776 {
2777 int i;
2778 mdk_rdev_t *rdev, *freshest, *tmp;
2779 char b[BDEVNAME_SIZE];
2780
2781 freshest = NULL;
2782 rdev_for_each(rdev, tmp, mddev)
2783 switch (super_types[mddev->major_version].
2784 load_super(rdev, freshest, mddev->minor_version)) {
2785 case 1:
2786 freshest = rdev;
2787 break;
2788 case 0:
2789 break;
2790 default:
2791 printk( KERN_ERR \
2792 "md: fatal superblock inconsistency in %s"
2793 " -- removing from array\n",
2794 bdevname(rdev->bdev,b));
2795 kick_rdev_from_array(rdev);
2796 }
2797
2798
2799 super_types[mddev->major_version].
2800 validate_super(mddev, freshest);
2801
2802 i = 0;
2803 rdev_for_each(rdev, tmp, mddev) {
2804 if (mddev->max_disks &&
2805 (rdev->desc_nr >= mddev->max_disks ||
2806 i > mddev->max_disks)) {
2807 printk(KERN_WARNING
2808 "md: %s: %s: only %d devices permitted\n",
2809 mdname(mddev), bdevname(rdev->bdev, b),
2810 mddev->max_disks);
2811 kick_rdev_from_array(rdev);
2812 continue;
2813 }
2814 if (rdev != freshest)
2815 if (super_types[mddev->major_version].
2816 validate_super(mddev, rdev)) {
2817 printk(KERN_WARNING "md: kicking non-fresh %s"
2818 " from array!\n",
2819 bdevname(rdev->bdev,b));
2820 kick_rdev_from_array(rdev);
2821 continue;
2822 }
2823 if (mddev->level == LEVEL_MULTIPATH) {
2824 rdev->desc_nr = i++;
2825 rdev->raid_disk = rdev->desc_nr;
2826 set_bit(In_sync, &rdev->flags);
2827 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
2828 rdev->raid_disk = -1;
2829 clear_bit(In_sync, &rdev->flags);
2830 }
2831 }
2832 }
2833
2834 /* Read a fixed-point number.
2835 * Numbers in sysfs attributes should be in "standard" units where
2836 * possible, so time should be in seconds.
2837 * However we internally use a a much smaller unit such as
2838 * milliseconds or jiffies.
2839 * This function takes a decimal number with a possible fractional
2840 * component, and produces an integer which is the result of
2841 * multiplying that number by 10^'scale'.
2842 * all without any floating-point arithmetic.
2843 */
2844 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
2845 {
2846 unsigned long result = 0;
2847 long decimals = -1;
2848 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
2849 if (*cp == '.')
2850 decimals = 0;
2851 else if (decimals < scale) {
2852 unsigned int value;
2853 value = *cp - '0';
2854 result = result * 10 + value;
2855 if (decimals >= 0)
2856 decimals++;
2857 }
2858 cp++;
2859 }
2860 if (*cp == '\n')
2861 cp++;
2862 if (*cp)
2863 return -EINVAL;
2864 if (decimals < 0)
2865 decimals = 0;
2866 while (decimals < scale) {
2867 result *= 10;
2868 decimals ++;
2869 }
2870 *res = result;
2871 return 0;
2872 }
2873
2874
2875 static void md_safemode_timeout(unsigned long data);
2876
2877 static ssize_t
2878 safe_delay_show(mddev_t *mddev, char *page)
2879 {
2880 int msec = (mddev->safemode_delay*1000)/HZ;
2881 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2882 }
2883 static ssize_t
2884 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2885 {
2886 unsigned long msec;
2887
2888 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
2889 return -EINVAL;
2890 if (msec == 0)
2891 mddev->safemode_delay = 0;
2892 else {
2893 unsigned long old_delay = mddev->safemode_delay;
2894 mddev->safemode_delay = (msec*HZ)/1000;
2895 if (mddev->safemode_delay == 0)
2896 mddev->safemode_delay = 1;
2897 if (mddev->safemode_delay < old_delay)
2898 md_safemode_timeout((unsigned long)mddev);
2899 }
2900 return len;
2901 }
2902 static struct md_sysfs_entry md_safe_delay =
2903 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2904
2905 static ssize_t
2906 level_show(mddev_t *mddev, char *page)
2907 {
2908 struct mdk_personality *p = mddev->pers;
2909 if (p)
2910 return sprintf(page, "%s\n", p->name);
2911 else if (mddev->clevel[0])
2912 return sprintf(page, "%s\n", mddev->clevel);
2913 else if (mddev->level != LEVEL_NONE)
2914 return sprintf(page, "%d\n", mddev->level);
2915 else
2916 return 0;
2917 }
2918
2919 static ssize_t
2920 level_store(mddev_t *mddev, const char *buf, size_t len)
2921 {
2922 char level[16];
2923 ssize_t rv = len;
2924 struct mdk_personality *pers;
2925 void *priv;
2926 mdk_rdev_t *rdev;
2927
2928 if (mddev->pers == NULL) {
2929 if (len == 0)
2930 return 0;
2931 if (len >= sizeof(mddev->clevel))
2932 return -ENOSPC;
2933 strncpy(mddev->clevel, buf, len);
2934 if (mddev->clevel[len-1] == '\n')
2935 len--;
2936 mddev->clevel[len] = 0;
2937 mddev->level = LEVEL_NONE;
2938 return rv;
2939 }
2940
2941 /* request to change the personality. Need to ensure:
2942 * - array is not engaged in resync/recovery/reshape
2943 * - old personality can be suspended
2944 * - new personality will access other array.
2945 */
2946
2947 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
2948 return -EBUSY;
2949
2950 if (!mddev->pers->quiesce) {
2951 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
2952 mdname(mddev), mddev->pers->name);
2953 return -EINVAL;
2954 }
2955
2956 /* Now find the new personality */
2957 if (len == 0 || len >= sizeof(level))
2958 return -EINVAL;
2959 strncpy(level, buf, len);
2960 if (level[len-1] == '\n')
2961 len--;
2962 level[len] = 0;
2963
2964 request_module("md-%s", level);
2965 spin_lock(&pers_lock);
2966 pers = find_pers(LEVEL_NONE, level);
2967 if (!pers || !try_module_get(pers->owner)) {
2968 spin_unlock(&pers_lock);
2969 printk(KERN_WARNING "md: personality %s not loaded\n", level);
2970 return -EINVAL;
2971 }
2972 spin_unlock(&pers_lock);
2973
2974 if (pers == mddev->pers) {
2975 /* Nothing to do! */
2976 module_put(pers->owner);
2977 return rv;
2978 }
2979 if (!pers->takeover) {
2980 module_put(pers->owner);
2981 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
2982 mdname(mddev), level);
2983 return -EINVAL;
2984 }
2985
2986 /* ->takeover must set new_* and/or delta_disks
2987 * if it succeeds, and may set them when it fails.
2988 */
2989 priv = pers->takeover(mddev);
2990 if (IS_ERR(priv)) {
2991 mddev->new_level = mddev->level;
2992 mddev->new_layout = mddev->layout;
2993 mddev->new_chunk_sectors = mddev->chunk_sectors;
2994 mddev->raid_disks -= mddev->delta_disks;
2995 mddev->delta_disks = 0;
2996 module_put(pers->owner);
2997 printk(KERN_WARNING "md: %s: %s would not accept array\n",
2998 mdname(mddev), level);
2999 return PTR_ERR(priv);
3000 }
3001
3002 /* Looks like we have a winner */
3003 mddev_suspend(mddev);
3004 mddev->pers->stop(mddev);
3005
3006 if (mddev->pers->sync_request == NULL &&
3007 pers->sync_request != NULL) {
3008 /* need to add the md_redundancy_group */
3009 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3010 printk(KERN_WARNING
3011 "md: cannot register extra attributes for %s\n",
3012 mdname(mddev));
3013 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3014 }
3015 if (mddev->pers->sync_request != NULL &&
3016 pers->sync_request == NULL) {
3017 /* need to remove the md_redundancy_group */
3018 if (mddev->to_remove == NULL)
3019 mddev->to_remove = &md_redundancy_group;
3020 }
3021
3022 if (mddev->pers->sync_request == NULL &&
3023 mddev->external) {
3024 /* We are converting from a no-redundancy array
3025 * to a redundancy array and metadata is managed
3026 * externally so we need to be sure that writes
3027 * won't block due to a need to transition
3028 * clean->dirty
3029 * until external management is started.
3030 */
3031 mddev->in_sync = 0;
3032 mddev->safemode_delay = 0;
3033 mddev->safemode = 0;
3034 }
3035
3036 module_put(mddev->pers->owner);
3037 /* Invalidate devices that are now superfluous */
3038 list_for_each_entry(rdev, &mddev->disks, same_set)
3039 if (rdev->raid_disk >= mddev->raid_disks) {
3040 rdev->raid_disk = -1;
3041 clear_bit(In_sync, &rdev->flags);
3042 }
3043 mddev->pers = pers;
3044 mddev->private = priv;
3045 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3046 mddev->level = mddev->new_level;
3047 mddev->layout = mddev->new_layout;
3048 mddev->chunk_sectors = mddev->new_chunk_sectors;
3049 mddev->delta_disks = 0;
3050 if (mddev->pers->sync_request == NULL) {
3051 /* this is now an array without redundancy, so
3052 * it must always be in_sync
3053 */
3054 mddev->in_sync = 1;
3055 del_timer_sync(&mddev->safemode_timer);
3056 }
3057 pers->run(mddev);
3058 mddev_resume(mddev);
3059 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3060 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3061 md_wakeup_thread(mddev->thread);
3062 sysfs_notify(&mddev->kobj, NULL, "level");
3063 return rv;
3064 }
3065
3066 static struct md_sysfs_entry md_level =
3067 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3068
3069
3070 static ssize_t
3071 layout_show(mddev_t *mddev, char *page)
3072 {
3073 /* just a number, not meaningful for all levels */
3074 if (mddev->reshape_position != MaxSector &&
3075 mddev->layout != mddev->new_layout)
3076 return sprintf(page, "%d (%d)\n",
3077 mddev->new_layout, mddev->layout);
3078 return sprintf(page, "%d\n", mddev->layout);
3079 }
3080
3081 static ssize_t
3082 layout_store(mddev_t *mddev, const char *buf, size_t len)
3083 {
3084 char *e;
3085 unsigned long n = simple_strtoul(buf, &e, 10);
3086
3087 if (!*buf || (*e && *e != '\n'))
3088 return -EINVAL;
3089
3090 if (mddev->pers) {
3091 int err;
3092 if (mddev->pers->check_reshape == NULL)
3093 return -EBUSY;
3094 mddev->new_layout = n;
3095 err = mddev->pers->check_reshape(mddev);
3096 if (err) {
3097 mddev->new_layout = mddev->layout;
3098 return err;
3099 }
3100 } else {
3101 mddev->new_layout = n;
3102 if (mddev->reshape_position == MaxSector)
3103 mddev->layout = n;
3104 }
3105 return len;
3106 }
3107 static struct md_sysfs_entry md_layout =
3108 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3109
3110
3111 static ssize_t
3112 raid_disks_show(mddev_t *mddev, char *page)
3113 {
3114 if (mddev->raid_disks == 0)
3115 return 0;
3116 if (mddev->reshape_position != MaxSector &&
3117 mddev->delta_disks != 0)
3118 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3119 mddev->raid_disks - mddev->delta_disks);
3120 return sprintf(page, "%d\n", mddev->raid_disks);
3121 }
3122
3123 static int update_raid_disks(mddev_t *mddev, int raid_disks);
3124
3125 static ssize_t
3126 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
3127 {
3128 char *e;
3129 int rv = 0;
3130 unsigned long n = simple_strtoul(buf, &e, 10);
3131
3132 if (!*buf || (*e && *e != '\n'))
3133 return -EINVAL;
3134
3135 if (mddev->pers)
3136 rv = update_raid_disks(mddev, n);
3137 else if (mddev->reshape_position != MaxSector) {
3138 int olddisks = mddev->raid_disks - mddev->delta_disks;
3139 mddev->delta_disks = n - olddisks;
3140 mddev->raid_disks = n;
3141 } else
3142 mddev->raid_disks = n;
3143 return rv ? rv : len;
3144 }
3145 static struct md_sysfs_entry md_raid_disks =
3146 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3147
3148 static ssize_t
3149 chunk_size_show(mddev_t *mddev, char *page)
3150 {
3151 if (mddev->reshape_position != MaxSector &&
3152 mddev->chunk_sectors != mddev->new_chunk_sectors)
3153 return sprintf(page, "%d (%d)\n",
3154 mddev->new_chunk_sectors << 9,
3155 mddev->chunk_sectors << 9);
3156 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3157 }
3158
3159 static ssize_t
3160 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
3161 {
3162 char *e;
3163 unsigned long n = simple_strtoul(buf, &e, 10);
3164
3165 if (!*buf || (*e && *e != '\n'))
3166 return -EINVAL;
3167
3168 if (mddev->pers) {
3169 int err;
3170 if (mddev->pers->check_reshape == NULL)
3171 return -EBUSY;
3172 mddev->new_chunk_sectors = n >> 9;
3173 err = mddev->pers->check_reshape(mddev);
3174 if (err) {
3175 mddev->new_chunk_sectors = mddev->chunk_sectors;
3176 return err;
3177 }
3178 } else {
3179 mddev->new_chunk_sectors = n >> 9;
3180 if (mddev->reshape_position == MaxSector)
3181 mddev->chunk_sectors = n >> 9;
3182 }
3183 return len;
3184 }
3185 static struct md_sysfs_entry md_chunk_size =
3186 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3187
3188 static ssize_t
3189 resync_start_show(mddev_t *mddev, char *page)
3190 {
3191 if (mddev->recovery_cp == MaxSector)
3192 return sprintf(page, "none\n");
3193 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3194 }
3195
3196 static ssize_t
3197 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
3198 {
3199 char *e;
3200 unsigned long long n = simple_strtoull(buf, &e, 10);
3201
3202 if (mddev->pers)
3203 return -EBUSY;
3204 if (cmd_match(buf, "none"))
3205 n = MaxSector;
3206 else if (!*buf || (*e && *e != '\n'))
3207 return -EINVAL;
3208
3209 mddev->recovery_cp = n;
3210 return len;
3211 }
3212 static struct md_sysfs_entry md_resync_start =
3213 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3214
3215 /*
3216 * The array state can be:
3217 *
3218 * clear
3219 * No devices, no size, no level
3220 * Equivalent to STOP_ARRAY ioctl
3221 * inactive
3222 * May have some settings, but array is not active
3223 * all IO results in error
3224 * When written, doesn't tear down array, but just stops it
3225 * suspended (not supported yet)
3226 * All IO requests will block. The array can be reconfigured.
3227 * Writing this, if accepted, will block until array is quiescent
3228 * readonly
3229 * no resync can happen. no superblocks get written.
3230 * write requests fail
3231 * read-auto
3232 * like readonly, but behaves like 'clean' on a write request.
3233 *
3234 * clean - no pending writes, but otherwise active.
3235 * When written to inactive array, starts without resync
3236 * If a write request arrives then
3237 * if metadata is known, mark 'dirty' and switch to 'active'.
3238 * if not known, block and switch to write-pending
3239 * If written to an active array that has pending writes, then fails.
3240 * active
3241 * fully active: IO and resync can be happening.
3242 * When written to inactive array, starts with resync
3243 *
3244 * write-pending
3245 * clean, but writes are blocked waiting for 'active' to be written.
3246 *
3247 * active-idle
3248 * like active, but no writes have been seen for a while (100msec).
3249 *
3250 */
3251 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3252 write_pending, active_idle, bad_word};
3253 static char *array_states[] = {
3254 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3255 "write-pending", "active-idle", NULL };
3256
3257 static int match_word(const char *word, char **list)
3258 {
3259 int n;
3260 for (n=0; list[n]; n++)
3261 if (cmd_match(word, list[n]))
3262 break;
3263 return n;
3264 }
3265
3266 static ssize_t
3267 array_state_show(mddev_t *mddev, char *page)
3268 {
3269 enum array_state st = inactive;
3270
3271 if (mddev->pers)
3272 switch(mddev->ro) {
3273 case 1:
3274 st = readonly;
3275 break;
3276 case 2:
3277 st = read_auto;
3278 break;
3279 case 0:
3280 if (mddev->in_sync)
3281 st = clean;
3282 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
3283 st = write_pending;
3284 else if (mddev->safemode)
3285 st = active_idle;
3286 else
3287 st = active;
3288 }
3289 else {
3290 if (list_empty(&mddev->disks) &&
3291 mddev->raid_disks == 0 &&
3292 mddev->dev_sectors == 0)
3293 st = clear;
3294 else
3295 st = inactive;
3296 }
3297 return sprintf(page, "%s\n", array_states[st]);
3298 }
3299
3300 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3301 static int do_md_run(mddev_t * mddev);
3302 static int restart_array(mddev_t *mddev);
3303
3304 static ssize_t
3305 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3306 {
3307 int err = -EINVAL;
3308 enum array_state st = match_word(buf, array_states);
3309 switch(st) {
3310 case bad_word:
3311 break;
3312 case clear:
3313 /* stopping an active array */
3314 if (atomic_read(&mddev->openers) > 0)
3315 return -EBUSY;
3316 err = do_md_stop(mddev, 0, 0);
3317 break;
3318 case inactive:
3319 /* stopping an active array */
3320 if (mddev->pers) {
3321 if (atomic_read(&mddev->openers) > 0)
3322 return -EBUSY;
3323 err = do_md_stop(mddev, 2, 0);
3324 } else
3325 err = 0; /* already inactive */
3326 break;
3327 case suspended:
3328 break; /* not supported yet */
3329 case readonly:
3330 if (mddev->pers)
3331 err = do_md_stop(mddev, 1, 0);
3332 else {
3333 mddev->ro = 1;
3334 set_disk_ro(mddev->gendisk, 1);
3335 err = do_md_run(mddev);
3336 }
3337 break;
3338 case read_auto:
3339 if (mddev->pers) {
3340 if (mddev->ro == 0)
3341 err = do_md_stop(mddev, 1, 0);
3342 else if (mddev->ro == 1)
3343 err = restart_array(mddev);
3344 if (err == 0) {
3345 mddev->ro = 2;
3346 set_disk_ro(mddev->gendisk, 0);
3347 }
3348 } else {
3349 mddev->ro = 2;
3350 err = do_md_run(mddev);
3351 }
3352 break;
3353 case clean:
3354 if (mddev->pers) {
3355 restart_array(mddev);
3356 spin_lock_irq(&mddev->write_lock);
3357 if (atomic_read(&mddev->writes_pending) == 0) {
3358 if (mddev->in_sync == 0) {
3359 mddev->in_sync = 1;
3360 if (mddev->safemode == 1)
3361 mddev->safemode = 0;
3362 if (mddev->persistent)
3363 set_bit(MD_CHANGE_CLEAN,
3364 &mddev->flags);
3365 }
3366 err = 0;
3367 } else
3368 err = -EBUSY;
3369 spin_unlock_irq(&mddev->write_lock);
3370 } else
3371 err = -EINVAL;
3372 break;
3373 case active:
3374 if (mddev->pers) {
3375 restart_array(mddev);
3376 if (mddev->external)
3377 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3378 wake_up(&mddev->sb_wait);
3379 err = 0;
3380 } else {
3381 mddev->ro = 0;
3382 set_disk_ro(mddev->gendisk, 0);
3383 err = do_md_run(mddev);
3384 }
3385 break;
3386 case write_pending:
3387 case active_idle:
3388 /* these cannot be set */
3389 break;
3390 }
3391 if (err)
3392 return err;
3393 else {
3394 sysfs_notify_dirent(mddev->sysfs_state);
3395 return len;
3396 }
3397 }
3398 static struct md_sysfs_entry md_array_state =
3399 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3400
3401 static ssize_t
3402 max_corrected_read_errors_show(mddev_t *mddev, char *page) {
3403 return sprintf(page, "%d\n",
3404 atomic_read(&mddev->max_corr_read_errors));
3405 }
3406
3407 static ssize_t
3408 max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len)
3409 {
3410 char *e;
3411 unsigned long n = simple_strtoul(buf, &e, 10);
3412
3413 if (*buf && (*e == 0 || *e == '\n')) {
3414 atomic_set(&mddev->max_corr_read_errors, n);
3415 return len;
3416 }
3417 return -EINVAL;
3418 }
3419
3420 static struct md_sysfs_entry max_corr_read_errors =
3421 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3422 max_corrected_read_errors_store);
3423
3424 static ssize_t
3425 null_show(mddev_t *mddev, char *page)
3426 {
3427 return -EINVAL;
3428 }
3429
3430 static ssize_t
3431 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3432 {
3433 /* buf must be %d:%d\n? giving major and minor numbers */
3434 /* The new device is added to the array.
3435 * If the array has a persistent superblock, we read the
3436 * superblock to initialise info and check validity.
3437 * Otherwise, only checking done is that in bind_rdev_to_array,
3438 * which mainly checks size.
3439 */
3440 char *e;
3441 int major = simple_strtoul(buf, &e, 10);
3442 int minor;
3443 dev_t dev;
3444 mdk_rdev_t *rdev;
3445 int err;
3446
3447 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3448 return -EINVAL;
3449 minor = simple_strtoul(e+1, &e, 10);
3450 if (*e && *e != '\n')
3451 return -EINVAL;
3452 dev = MKDEV(major, minor);
3453 if (major != MAJOR(dev) ||
3454 minor != MINOR(dev))
3455 return -EOVERFLOW;
3456
3457
3458 if (mddev->persistent) {
3459 rdev = md_import_device(dev, mddev->major_version,
3460 mddev->minor_version);
3461 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3462 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3463 mdk_rdev_t, same_set);
3464 err = super_types[mddev->major_version]
3465 .load_super(rdev, rdev0, mddev->minor_version);
3466 if (err < 0)
3467 goto out;
3468 }
3469 } else if (mddev->external)
3470 rdev = md_import_device(dev, -2, -1);
3471 else
3472 rdev = md_import_device(dev, -1, -1);
3473
3474 if (IS_ERR(rdev))
3475 return PTR_ERR(rdev);
3476 err = bind_rdev_to_array(rdev, mddev);
3477 out:
3478 if (err)
3479 export_rdev(rdev);
3480 return err ? err : len;
3481 }
3482
3483 static struct md_sysfs_entry md_new_device =
3484 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3485
3486 static ssize_t
3487 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3488 {
3489 char *end;
3490 unsigned long chunk, end_chunk;
3491
3492 if (!mddev->bitmap)
3493 goto out;
3494 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3495 while (*buf) {
3496 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3497 if (buf == end) break;
3498 if (*end == '-') { /* range */
3499 buf = end + 1;
3500 end_chunk = simple_strtoul(buf, &end, 0);
3501 if (buf == end) break;
3502 }
3503 if (*end && !isspace(*end)) break;
3504 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3505 buf = skip_spaces(end);
3506 }
3507 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3508 out:
3509 return len;
3510 }
3511
3512 static struct md_sysfs_entry md_bitmap =
3513 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3514
3515 static ssize_t
3516 size_show(mddev_t *mddev, char *page)
3517 {
3518 return sprintf(page, "%llu\n",
3519 (unsigned long long)mddev->dev_sectors / 2);
3520 }
3521
3522 static int update_size(mddev_t *mddev, sector_t num_sectors);
3523
3524 static ssize_t
3525 size_store(mddev_t *mddev, const char *buf, size_t len)
3526 {
3527 /* If array is inactive, we can reduce the component size, but
3528 * not increase it (except from 0).
3529 * If array is active, we can try an on-line resize
3530 */
3531 sector_t sectors;
3532 int err = strict_blocks_to_sectors(buf, &sectors);
3533
3534 if (err < 0)
3535 return err;
3536 if (mddev->pers) {
3537 err = update_size(mddev, sectors);
3538 md_update_sb(mddev, 1);
3539 } else {
3540 if (mddev->dev_sectors == 0 ||
3541 mddev->dev_sectors > sectors)
3542 mddev->dev_sectors = sectors;
3543 else
3544 err = -ENOSPC;
3545 }
3546 return err ? err : len;
3547 }
3548
3549 static struct md_sysfs_entry md_size =
3550 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3551
3552
3553 /* Metdata version.
3554 * This is one of
3555 * 'none' for arrays with no metadata (good luck...)
3556 * 'external' for arrays with externally managed metadata,
3557 * or N.M for internally known formats
3558 */
3559 static ssize_t
3560 metadata_show(mddev_t *mddev, char *page)
3561 {
3562 if (mddev->persistent)
3563 return sprintf(page, "%d.%d\n",
3564 mddev->major_version, mddev->minor_version);
3565 else if (mddev->external)
3566 return sprintf(page, "external:%s\n", mddev->metadata_type);
3567 else
3568 return sprintf(page, "none\n");
3569 }
3570
3571 static ssize_t
3572 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3573 {
3574 int major, minor;
3575 char *e;
3576 /* Changing the details of 'external' metadata is
3577 * always permitted. Otherwise there must be
3578 * no devices attached to the array.
3579 */
3580 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3581 ;
3582 else if (!list_empty(&mddev->disks))
3583 return -EBUSY;
3584
3585 if (cmd_match(buf, "none")) {
3586 mddev->persistent = 0;
3587 mddev->external = 0;
3588 mddev->major_version = 0;
3589 mddev->minor_version = 90;
3590 return len;
3591 }
3592 if (strncmp(buf, "external:", 9) == 0) {
3593 size_t namelen = len-9;
3594 if (namelen >= sizeof(mddev->metadata_type))
3595 namelen = sizeof(mddev->metadata_type)-1;
3596 strncpy(mddev->metadata_type, buf+9, namelen);
3597 mddev->metadata_type[namelen] = 0;
3598 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3599 mddev->metadata_type[--namelen] = 0;
3600 mddev->persistent = 0;
3601 mddev->external = 1;
3602 mddev->major_version = 0;
3603 mddev->minor_version = 90;
3604 return len;
3605 }
3606 major = simple_strtoul(buf, &e, 10);
3607 if (e==buf || *e != '.')
3608 return -EINVAL;
3609 buf = e+1;
3610 minor = simple_strtoul(buf, &e, 10);
3611 if (e==buf || (*e && *e != '\n') )
3612 return -EINVAL;
3613 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3614 return -ENOENT;
3615 mddev->major_version = major;
3616 mddev->minor_version = minor;
3617 mddev->persistent = 1;
3618 mddev->external = 0;
3619 return len;
3620 }
3621
3622 static struct md_sysfs_entry md_metadata =
3623 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3624
3625 static ssize_t
3626 action_show(mddev_t *mddev, char *page)
3627 {
3628 char *type = "idle";
3629 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3630 type = "frozen";
3631 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3632 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3633 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3634 type = "reshape";
3635 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3636 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3637 type = "resync";
3638 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3639 type = "check";
3640 else
3641 type = "repair";
3642 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3643 type = "recover";
3644 }
3645 return sprintf(page, "%s\n", type);
3646 }
3647
3648 static ssize_t
3649 action_store(mddev_t *mddev, const char *page, size_t len)
3650 {
3651 if (!mddev->pers || !mddev->pers->sync_request)
3652 return -EINVAL;
3653
3654 if (cmd_match(page, "frozen"))
3655 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3656 else
3657 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3658
3659 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3660 if (mddev->sync_thread) {
3661 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3662 md_unregister_thread(mddev->sync_thread);
3663 mddev->sync_thread = NULL;
3664 mddev->recovery = 0;
3665 }
3666 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3667 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3668 return -EBUSY;
3669 else if (cmd_match(page, "resync"))
3670 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3671 else if (cmd_match(page, "recover")) {
3672 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3673 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3674 } else if (cmd_match(page, "reshape")) {
3675 int err;
3676 if (mddev->pers->start_reshape == NULL)
3677 return -EINVAL;
3678 err = mddev->pers->start_reshape(mddev);
3679 if (err)
3680 return err;
3681 sysfs_notify(&mddev->kobj, NULL, "degraded");
3682 } else {
3683 if (cmd_match(page, "check"))
3684 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3685 else if (!cmd_match(page, "repair"))
3686 return -EINVAL;
3687 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3688 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3689 }
3690 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3691 md_wakeup_thread(mddev->thread);
3692 sysfs_notify_dirent(mddev->sysfs_action);
3693 return len;
3694 }
3695
3696 static ssize_t
3697 mismatch_cnt_show(mddev_t *mddev, char *page)
3698 {
3699 return sprintf(page, "%llu\n",
3700 (unsigned long long) mddev->resync_mismatches);
3701 }
3702
3703 static struct md_sysfs_entry md_scan_mode =
3704 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3705
3706
3707 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3708
3709 static ssize_t
3710 sync_min_show(mddev_t *mddev, char *page)
3711 {
3712 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3713 mddev->sync_speed_min ? "local": "system");
3714 }
3715
3716 static ssize_t
3717 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3718 {
3719 int min;
3720 char *e;
3721 if (strncmp(buf, "system", 6)==0) {
3722 mddev->sync_speed_min = 0;
3723 return len;
3724 }
3725 min = simple_strtoul(buf, &e, 10);
3726 if (buf == e || (*e && *e != '\n') || min <= 0)
3727 return -EINVAL;
3728 mddev->sync_speed_min = min;
3729 return len;
3730 }
3731
3732 static struct md_sysfs_entry md_sync_min =
3733 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3734
3735 static ssize_t
3736 sync_max_show(mddev_t *mddev, char *page)
3737 {
3738 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3739 mddev->sync_speed_max ? "local": "system");
3740 }
3741
3742 static ssize_t
3743 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3744 {
3745 int max;
3746 char *e;
3747 if (strncmp(buf, "system", 6)==0) {
3748 mddev->sync_speed_max = 0;
3749 return len;
3750 }
3751 max = simple_strtoul(buf, &e, 10);
3752 if (buf == e || (*e && *e != '\n') || max <= 0)
3753 return -EINVAL;
3754 mddev->sync_speed_max = max;
3755 return len;
3756 }
3757
3758 static struct md_sysfs_entry md_sync_max =
3759 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3760
3761 static ssize_t
3762 degraded_show(mddev_t *mddev, char *page)
3763 {
3764 return sprintf(page, "%d\n", mddev->degraded);
3765 }
3766 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3767
3768 static ssize_t
3769 sync_force_parallel_show(mddev_t *mddev, char *page)
3770 {
3771 return sprintf(page, "%d\n", mddev->parallel_resync);
3772 }
3773
3774 static ssize_t
3775 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3776 {
3777 long n;
3778
3779 if (strict_strtol(buf, 10, &n))
3780 return -EINVAL;
3781
3782 if (n != 0 && n != 1)
3783 return -EINVAL;
3784
3785 mddev->parallel_resync = n;
3786
3787 if (mddev->sync_thread)
3788 wake_up(&resync_wait);
3789
3790 return len;
3791 }
3792
3793 /* force parallel resync, even with shared block devices */
3794 static struct md_sysfs_entry md_sync_force_parallel =
3795 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3796 sync_force_parallel_show, sync_force_parallel_store);
3797
3798 static ssize_t
3799 sync_speed_show(mddev_t *mddev, char *page)
3800 {
3801 unsigned long resync, dt, db;
3802 if (mddev->curr_resync == 0)
3803 return sprintf(page, "none\n");
3804 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3805 dt = (jiffies - mddev->resync_mark) / HZ;
3806 if (!dt) dt++;
3807 db = resync - mddev->resync_mark_cnt;
3808 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3809 }
3810
3811 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3812
3813 static ssize_t
3814 sync_completed_show(mddev_t *mddev, char *page)
3815 {
3816 unsigned long max_sectors, resync;
3817
3818 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3819 return sprintf(page, "none\n");
3820
3821 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3822 max_sectors = mddev->resync_max_sectors;
3823 else
3824 max_sectors = mddev->dev_sectors;
3825
3826 resync = mddev->curr_resync_completed;
3827 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3828 }
3829
3830 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3831
3832 static ssize_t
3833 min_sync_show(mddev_t *mddev, char *page)
3834 {
3835 return sprintf(page, "%llu\n",
3836 (unsigned long long)mddev->resync_min);
3837 }
3838 static ssize_t
3839 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3840 {
3841 unsigned long long min;
3842 if (strict_strtoull(buf, 10, &min))
3843 return -EINVAL;
3844 if (min > mddev->resync_max)
3845 return -EINVAL;
3846 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3847 return -EBUSY;
3848
3849 /* Must be a multiple of chunk_size */
3850 if (mddev->chunk_sectors) {
3851 sector_t temp = min;
3852 if (sector_div(temp, mddev->chunk_sectors))
3853 return -EINVAL;
3854 }
3855 mddev->resync_min = min;
3856
3857 return len;
3858 }
3859
3860 static struct md_sysfs_entry md_min_sync =
3861 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3862
3863 static ssize_t
3864 max_sync_show(mddev_t *mddev, char *page)
3865 {
3866 if (mddev->resync_max == MaxSector)
3867 return sprintf(page, "max\n");
3868 else
3869 return sprintf(page, "%llu\n",
3870 (unsigned long long)mddev->resync_max);
3871 }
3872 static ssize_t
3873 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3874 {
3875 if (strncmp(buf, "max", 3) == 0)
3876 mddev->resync_max = MaxSector;
3877 else {
3878 unsigned long long max;
3879 if (strict_strtoull(buf, 10, &max))
3880 return -EINVAL;
3881 if (max < mddev->resync_min)
3882 return -EINVAL;
3883 if (max < mddev->resync_max &&
3884 mddev->ro == 0 &&
3885 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3886 return -EBUSY;
3887
3888 /* Must be a multiple of chunk_size */
3889 if (mddev->chunk_sectors) {
3890 sector_t temp = max;
3891 if (sector_div(temp, mddev->chunk_sectors))
3892 return -EINVAL;
3893 }
3894 mddev->resync_max = max;
3895 }
3896 wake_up(&mddev->recovery_wait);
3897 return len;
3898 }
3899
3900 static struct md_sysfs_entry md_max_sync =
3901 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3902
3903 static ssize_t
3904 suspend_lo_show(mddev_t *mddev, char *page)
3905 {
3906 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3907 }
3908
3909 static ssize_t
3910 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3911 {
3912 char *e;
3913 unsigned long long new = simple_strtoull(buf, &e, 10);
3914
3915 if (mddev->pers == NULL ||
3916 mddev->pers->quiesce == NULL)
3917 return -EINVAL;
3918 if (buf == e || (*e && *e != '\n'))
3919 return -EINVAL;
3920 if (new >= mddev->suspend_hi ||
3921 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3922 mddev->suspend_lo = new;
3923 mddev->pers->quiesce(mddev, 2);
3924 return len;
3925 } else
3926 return -EINVAL;
3927 }
3928 static struct md_sysfs_entry md_suspend_lo =
3929 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3930
3931
3932 static ssize_t
3933 suspend_hi_show(mddev_t *mddev, char *page)
3934 {
3935 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3936 }
3937
3938 static ssize_t
3939 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3940 {
3941 char *e;
3942 unsigned long long new = simple_strtoull(buf, &e, 10);
3943
3944 if (mddev->pers == NULL ||
3945 mddev->pers->quiesce == NULL)
3946 return -EINVAL;
3947 if (buf == e || (*e && *e != '\n'))
3948 return -EINVAL;
3949 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3950 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3951 mddev->suspend_hi = new;
3952 mddev->pers->quiesce(mddev, 1);
3953 mddev->pers->quiesce(mddev, 0);
3954 return len;
3955 } else
3956 return -EINVAL;
3957 }
3958 static struct md_sysfs_entry md_suspend_hi =
3959 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3960
3961 static ssize_t
3962 reshape_position_show(mddev_t *mddev, char *page)
3963 {
3964 if (mddev->reshape_position != MaxSector)
3965 return sprintf(page, "%llu\n",
3966 (unsigned long long)mddev->reshape_position);
3967 strcpy(page, "none\n");
3968 return 5;
3969 }
3970
3971 static ssize_t
3972 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3973 {
3974 char *e;
3975 unsigned long long new = simple_strtoull(buf, &e, 10);
3976 if (mddev->pers)
3977 return -EBUSY;
3978 if (buf == e || (*e && *e != '\n'))
3979 return -EINVAL;
3980 mddev->reshape_position = new;
3981 mddev->delta_disks = 0;
3982 mddev->new_level = mddev->level;
3983 mddev->new_layout = mddev->layout;
3984 mddev->new_chunk_sectors = mddev->chunk_sectors;
3985 return len;
3986 }
3987
3988 static struct md_sysfs_entry md_reshape_position =
3989 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3990 reshape_position_store);
3991
3992 static ssize_t
3993 array_size_show(mddev_t *mddev, char *page)
3994 {
3995 if (mddev->external_size)
3996 return sprintf(page, "%llu\n",
3997 (unsigned long long)mddev->array_sectors/2);
3998 else
3999 return sprintf(page, "default\n");
4000 }
4001
4002 static ssize_t
4003 array_size_store(mddev_t *mddev, const char *buf, size_t len)
4004 {
4005 sector_t sectors;
4006
4007 if (strncmp(buf, "default", 7) == 0) {
4008 if (mddev->pers)
4009 sectors = mddev->pers->size(mddev, 0, 0);
4010 else
4011 sectors = mddev->array_sectors;
4012
4013 mddev->external_size = 0;
4014 } else {
4015 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4016 return -EINVAL;
4017 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4018 return -E2BIG;
4019
4020 mddev->external_size = 1;
4021 }
4022
4023 mddev->array_sectors = sectors;
4024 set_capacity(mddev->gendisk, mddev->array_sectors);
4025 if (mddev->pers)
4026 revalidate_disk(mddev->gendisk);
4027
4028 return len;
4029 }
4030
4031 static struct md_sysfs_entry md_array_size =
4032 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4033 array_size_store);
4034
4035 static struct attribute *md_default_attrs[] = {
4036 &md_level.attr,
4037 &md_layout.attr,
4038 &md_raid_disks.attr,
4039 &md_chunk_size.attr,
4040 &md_size.attr,
4041 &md_resync_start.attr,
4042 &md_metadata.attr,
4043 &md_new_device.attr,
4044 &md_safe_delay.attr,
4045 &md_array_state.attr,
4046 &md_reshape_position.attr,
4047 &md_array_size.attr,
4048 &max_corr_read_errors.attr,
4049 NULL,
4050 };
4051
4052 static struct attribute *md_redundancy_attrs[] = {
4053 &md_scan_mode.attr,
4054 &md_mismatches.attr,
4055 &md_sync_min.attr,
4056 &md_sync_max.attr,
4057 &md_sync_speed.attr,
4058 &md_sync_force_parallel.attr,
4059 &md_sync_completed.attr,
4060 &md_min_sync.attr,
4061 &md_max_sync.attr,
4062 &md_suspend_lo.attr,
4063 &md_suspend_hi.attr,
4064 &md_bitmap.attr,
4065 &md_degraded.attr,
4066 NULL,
4067 };
4068 static struct attribute_group md_redundancy_group = {
4069 .name = NULL,
4070 .attrs = md_redundancy_attrs,
4071 };
4072
4073
4074 static ssize_t
4075 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4076 {
4077 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4078 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4079 ssize_t rv;
4080
4081 if (!entry->show)
4082 return -EIO;
4083 rv = mddev_lock(mddev);
4084 if (!rv) {
4085 rv = entry->show(mddev, page);
4086 mddev_unlock(mddev);
4087 }
4088 return rv;
4089 }
4090
4091 static ssize_t
4092 md_attr_store(struct kobject *kobj, struct attribute *attr,
4093 const char *page, size_t length)
4094 {
4095 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4096 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4097 ssize_t rv;
4098
4099 if (!entry->store)
4100 return -EIO;
4101 if (!capable(CAP_SYS_ADMIN))
4102 return -EACCES;
4103 rv = mddev_lock(mddev);
4104 if (mddev->hold_active == UNTIL_IOCTL)
4105 mddev->hold_active = 0;
4106 if (!rv) {
4107 rv = entry->store(mddev, page, length);
4108 mddev_unlock(mddev);
4109 }
4110 return rv;
4111 }
4112
4113 static void md_free(struct kobject *ko)
4114 {
4115 mddev_t *mddev = container_of(ko, mddev_t, kobj);
4116
4117 if (mddev->sysfs_state)
4118 sysfs_put(mddev->sysfs_state);
4119
4120 if (mddev->gendisk) {
4121 del_gendisk(mddev->gendisk);
4122 put_disk(mddev->gendisk);
4123 }
4124 if (mddev->queue)
4125 blk_cleanup_queue(mddev->queue);
4126
4127 kfree(mddev);
4128 }
4129
4130 static struct sysfs_ops md_sysfs_ops = {
4131 .show = md_attr_show,
4132 .store = md_attr_store,
4133 };
4134 static struct kobj_type md_ktype = {
4135 .release = md_free,
4136 .sysfs_ops = &md_sysfs_ops,
4137 .default_attrs = md_default_attrs,
4138 };
4139
4140 int mdp_major = 0;
4141
4142 static void mddev_delayed_delete(struct work_struct *ws)
4143 {
4144 mddev_t *mddev = container_of(ws, mddev_t, del_work);
4145
4146 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4147 kobject_del(&mddev->kobj);
4148 kobject_put(&mddev->kobj);
4149 }
4150
4151 static int md_alloc(dev_t dev, char *name)
4152 {
4153 static DEFINE_MUTEX(disks_mutex);
4154 mddev_t *mddev = mddev_find(dev);
4155 struct gendisk *disk;
4156 int partitioned;
4157 int shift;
4158 int unit;
4159 int error;
4160
4161 if (!mddev)
4162 return -ENODEV;
4163
4164 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4165 shift = partitioned ? MdpMinorShift : 0;
4166 unit = MINOR(mddev->unit) >> shift;
4167
4168 /* wait for any previous instance if this device
4169 * to be completed removed (mddev_delayed_delete).
4170 */
4171 flush_scheduled_work();
4172
4173 mutex_lock(&disks_mutex);
4174 error = -EEXIST;
4175 if (mddev->gendisk)
4176 goto abort;
4177
4178 if (name) {
4179 /* Need to ensure that 'name' is not a duplicate.
4180 */
4181 mddev_t *mddev2;
4182 spin_lock(&all_mddevs_lock);
4183
4184 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4185 if (mddev2->gendisk &&
4186 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4187 spin_unlock(&all_mddevs_lock);
4188 goto abort;
4189 }
4190 spin_unlock(&all_mddevs_lock);
4191 }
4192
4193 error = -ENOMEM;
4194 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4195 if (!mddev->queue)
4196 goto abort;
4197 mddev->queue->queuedata = mddev;
4198
4199 /* Can be unlocked because the queue is new: no concurrency */
4200 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
4201
4202 blk_queue_make_request(mddev->queue, md_make_request);
4203
4204 disk = alloc_disk(1 << shift);
4205 if (!disk) {
4206 blk_cleanup_queue(mddev->queue);
4207 mddev->queue = NULL;
4208 goto abort;
4209 }
4210 disk->major = MAJOR(mddev->unit);
4211 disk->first_minor = unit << shift;
4212 if (name)
4213 strcpy(disk->disk_name, name);
4214 else if (partitioned)
4215 sprintf(disk->disk_name, "md_d%d", unit);
4216 else
4217 sprintf(disk->disk_name, "md%d", unit);
4218 disk->fops = &md_fops;
4219 disk->private_data = mddev;
4220 disk->queue = mddev->queue;
4221 /* Allow extended partitions. This makes the
4222 * 'mdp' device redundant, but we can't really
4223 * remove it now.
4224 */
4225 disk->flags |= GENHD_FL_EXT_DEVT;
4226 add_disk(disk);
4227 mddev->gendisk = disk;
4228 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4229 &disk_to_dev(disk)->kobj, "%s", "md");
4230 if (error) {
4231 /* This isn't possible, but as kobject_init_and_add is marked
4232 * __must_check, we must do something with the result
4233 */
4234 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4235 disk->disk_name);
4236 error = 0;
4237 }
4238 if (sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4239 printk(KERN_DEBUG "pointless warning\n");
4240 abort:
4241 mutex_unlock(&disks_mutex);
4242 if (!error) {
4243 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4244 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state");
4245 }
4246 mddev_put(mddev);
4247 return error;
4248 }
4249
4250 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4251 {
4252 md_alloc(dev, NULL);
4253 return NULL;
4254 }
4255
4256 static int add_named_array(const char *val, struct kernel_param *kp)
4257 {
4258 /* val must be "md_*" where * is not all digits.
4259 * We allocate an array with a large free minor number, and
4260 * set the name to val. val must not already be an active name.
4261 */
4262 int len = strlen(val);
4263 char buf[DISK_NAME_LEN];
4264
4265 while (len && val[len-1] == '\n')
4266 len--;
4267 if (len >= DISK_NAME_LEN)
4268 return -E2BIG;
4269 strlcpy(buf, val, len+1);
4270 if (strncmp(buf, "md_", 3) != 0)
4271 return -EINVAL;
4272 return md_alloc(0, buf);
4273 }
4274
4275 static void md_safemode_timeout(unsigned long data)
4276 {
4277 mddev_t *mddev = (mddev_t *) data;
4278
4279 if (!atomic_read(&mddev->writes_pending)) {
4280 mddev->safemode = 1;
4281 if (mddev->external)
4282 sysfs_notify_dirent(mddev->sysfs_state);
4283 }
4284 md_wakeup_thread(mddev->thread);
4285 }
4286
4287 static int start_dirty_degraded;
4288
4289 static int do_md_run(mddev_t * mddev)
4290 {
4291 int err;
4292 mdk_rdev_t *rdev;
4293 struct gendisk *disk;
4294 struct mdk_personality *pers;
4295
4296 if (list_empty(&mddev->disks))
4297 /* cannot run an array with no devices.. */
4298 return -EINVAL;
4299
4300 if (mddev->pers)
4301 return -EBUSY;
4302
4303 /* These two calls synchronise us with the
4304 * sysfs_remove_group calls in mddev_unlock,
4305 * so they must have completed.
4306 */
4307 mutex_lock(&mddev->open_mutex);
4308 mutex_unlock(&mddev->open_mutex);
4309
4310 /*
4311 * Analyze all RAID superblock(s)
4312 */
4313 if (!mddev->raid_disks) {
4314 if (!mddev->persistent)
4315 return -EINVAL;
4316 analyze_sbs(mddev);
4317 }
4318
4319 if (mddev->level != LEVEL_NONE)
4320 request_module("md-level-%d", mddev->level);
4321 else if (mddev->clevel[0])
4322 request_module("md-%s", mddev->clevel);
4323
4324 /*
4325 * Drop all container device buffers, from now on
4326 * the only valid external interface is through the md
4327 * device.
4328 */
4329 list_for_each_entry(rdev, &mddev->disks, same_set) {
4330 if (test_bit(Faulty, &rdev->flags))
4331 continue;
4332 sync_blockdev(rdev->bdev);
4333 invalidate_bdev(rdev->bdev);
4334
4335 /* perform some consistency tests on the device.
4336 * We don't want the data to overlap the metadata,
4337 * Internal Bitmap issues have been handled elsewhere.
4338 */
4339 if (rdev->data_offset < rdev->sb_start) {
4340 if (mddev->dev_sectors &&
4341 rdev->data_offset + mddev->dev_sectors
4342 > rdev->sb_start) {
4343 printk("md: %s: data overlaps metadata\n",
4344 mdname(mddev));
4345 return -EINVAL;
4346 }
4347 } else {
4348 if (rdev->sb_start + rdev->sb_size/512
4349 > rdev->data_offset) {
4350 printk("md: %s: metadata overlaps data\n",
4351 mdname(mddev));
4352 return -EINVAL;
4353 }
4354 }
4355 sysfs_notify_dirent(rdev->sysfs_state);
4356 }
4357
4358 disk = mddev->gendisk;
4359
4360 spin_lock(&pers_lock);
4361 pers = find_pers(mddev->level, mddev->clevel);
4362 if (!pers || !try_module_get(pers->owner)) {
4363 spin_unlock(&pers_lock);
4364 if (mddev->level != LEVEL_NONE)
4365 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4366 mddev->level);
4367 else
4368 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4369 mddev->clevel);
4370 return -EINVAL;
4371 }
4372 mddev->pers = pers;
4373 spin_unlock(&pers_lock);
4374 if (mddev->level != pers->level) {
4375 mddev->level = pers->level;
4376 mddev->new_level = pers->level;
4377 }
4378 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4379
4380 if (mddev->reshape_position != MaxSector &&
4381 pers->start_reshape == NULL) {
4382 /* This personality cannot handle reshaping... */
4383 mddev->pers = NULL;
4384 module_put(pers->owner);
4385 return -EINVAL;
4386 }
4387
4388 if (pers->sync_request) {
4389 /* Warn if this is a potentially silly
4390 * configuration.
4391 */
4392 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4393 mdk_rdev_t *rdev2;
4394 int warned = 0;
4395
4396 list_for_each_entry(rdev, &mddev->disks, same_set)
4397 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4398 if (rdev < rdev2 &&
4399 rdev->bdev->bd_contains ==
4400 rdev2->bdev->bd_contains) {
4401 printk(KERN_WARNING
4402 "%s: WARNING: %s appears to be"
4403 " on the same physical disk as"
4404 " %s.\n",
4405 mdname(mddev),
4406 bdevname(rdev->bdev,b),
4407 bdevname(rdev2->bdev,b2));
4408 warned = 1;
4409 }
4410 }
4411
4412 if (warned)
4413 printk(KERN_WARNING
4414 "True protection against single-disk"
4415 " failure might be compromised.\n");
4416 }
4417
4418 mddev->recovery = 0;
4419 /* may be over-ridden by personality */
4420 mddev->resync_max_sectors = mddev->dev_sectors;
4421
4422 mddev->barriers_work = 1;
4423 mddev->ok_start_degraded = start_dirty_degraded;
4424
4425 if (start_readonly && mddev->ro == 0)
4426 mddev->ro = 2; /* read-only, but switch on first write */
4427
4428 err = mddev->pers->run(mddev);
4429 if (err)
4430 printk(KERN_ERR "md: pers->run() failed ...\n");
4431 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4432 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4433 " but 'external_size' not in effect?\n", __func__);
4434 printk(KERN_ERR
4435 "md: invalid array_size %llu > default size %llu\n",
4436 (unsigned long long)mddev->array_sectors / 2,
4437 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4438 err = -EINVAL;
4439 mddev->pers->stop(mddev);
4440 }
4441 if (err == 0 && mddev->pers->sync_request) {
4442 err = bitmap_create(mddev);
4443 if (err) {
4444 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4445 mdname(mddev), err);
4446 mddev->pers->stop(mddev);
4447 }
4448 }
4449 if (err) {
4450 module_put(mddev->pers->owner);
4451 mddev->pers = NULL;
4452 bitmap_destroy(mddev);
4453 return err;
4454 }
4455 if (mddev->pers->sync_request) {
4456 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4457 printk(KERN_WARNING
4458 "md: cannot register extra attributes for %s\n",
4459 mdname(mddev));
4460 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4461 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4462 mddev->ro = 0;
4463
4464 atomic_set(&mddev->writes_pending,0);
4465 atomic_set(&mddev->max_corr_read_errors,
4466 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
4467 mddev->safemode = 0;
4468 mddev->safemode_timer.function = md_safemode_timeout;
4469 mddev->safemode_timer.data = (unsigned long) mddev;
4470 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4471 mddev->in_sync = 1;
4472
4473 list_for_each_entry(rdev, &mddev->disks, same_set)
4474 if (rdev->raid_disk >= 0) {
4475 char nm[20];
4476 sprintf(nm, "rd%d", rdev->raid_disk);
4477 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4478 printk("md: cannot register %s for %s\n",
4479 nm, mdname(mddev));
4480 }
4481
4482 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4483
4484 if (mddev->flags)
4485 md_update_sb(mddev, 0);
4486
4487 set_capacity(disk, mddev->array_sectors);
4488
4489 md_wakeup_thread(mddev->thread);
4490 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4491
4492 revalidate_disk(mddev->gendisk);
4493 mddev->changed = 1;
4494 md_new_event(mddev);
4495 sysfs_notify_dirent(mddev->sysfs_state);
4496 if (mddev->sysfs_action)
4497 sysfs_notify_dirent(mddev->sysfs_action);
4498 sysfs_notify(&mddev->kobj, NULL, "degraded");
4499 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4500 return 0;
4501 }
4502
4503 static int restart_array(mddev_t *mddev)
4504 {
4505 struct gendisk *disk = mddev->gendisk;
4506
4507 /* Complain if it has no devices */
4508 if (list_empty(&mddev->disks))
4509 return -ENXIO;
4510 if (!mddev->pers)
4511 return -EINVAL;
4512 if (!mddev->ro)
4513 return -EBUSY;
4514 mddev->safemode = 0;
4515 mddev->ro = 0;
4516 set_disk_ro(disk, 0);
4517 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4518 mdname(mddev));
4519 /* Kick recovery or resync if necessary */
4520 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4521 md_wakeup_thread(mddev->thread);
4522 md_wakeup_thread(mddev->sync_thread);
4523 sysfs_notify_dirent(mddev->sysfs_state);
4524 return 0;
4525 }
4526
4527 /* similar to deny_write_access, but accounts for our holding a reference
4528 * to the file ourselves */
4529 static int deny_bitmap_write_access(struct file * file)
4530 {
4531 struct inode *inode = file->f_mapping->host;
4532
4533 spin_lock(&inode->i_lock);
4534 if (atomic_read(&inode->i_writecount) > 1) {
4535 spin_unlock(&inode->i_lock);
4536 return -ETXTBSY;
4537 }
4538 atomic_set(&inode->i_writecount, -1);
4539 spin_unlock(&inode->i_lock);
4540
4541 return 0;
4542 }
4543
4544 void restore_bitmap_write_access(struct file *file)
4545 {
4546 struct inode *inode = file->f_mapping->host;
4547
4548 spin_lock(&inode->i_lock);
4549 atomic_set(&inode->i_writecount, 1);
4550 spin_unlock(&inode->i_lock);
4551 }
4552
4553 /* mode:
4554 * 0 - completely stop and dis-assemble array
4555 * 1 - switch to readonly
4556 * 2 - stop but do not disassemble array
4557 */
4558 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4559 {
4560 int err = 0;
4561 struct gendisk *disk = mddev->gendisk;
4562 mdk_rdev_t *rdev;
4563
4564 mutex_lock(&mddev->open_mutex);
4565 if (atomic_read(&mddev->openers) > is_open) {
4566 printk("md: %s still in use.\n",mdname(mddev));
4567 err = -EBUSY;
4568 } else if (mddev->pers) {
4569
4570 if (mddev->sync_thread) {
4571 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4572 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4573 md_unregister_thread(mddev->sync_thread);
4574 mddev->sync_thread = NULL;
4575 }
4576
4577 del_timer_sync(&mddev->safemode_timer);
4578
4579 switch(mode) {
4580 case 1: /* readonly */
4581 err = -ENXIO;
4582 if (mddev->ro==1)
4583 goto out;
4584 mddev->ro = 1;
4585 break;
4586 case 0: /* disassemble */
4587 case 2: /* stop */
4588 bitmap_flush(mddev);
4589 md_super_wait(mddev);
4590 if (mddev->ro)
4591 set_disk_ro(disk, 0);
4592
4593 mddev->pers->stop(mddev);
4594 mddev->queue->merge_bvec_fn = NULL;
4595 mddev->queue->unplug_fn = NULL;
4596 mddev->queue->backing_dev_info.congested_fn = NULL;
4597 module_put(mddev->pers->owner);
4598 if (mddev->pers->sync_request && mddev->to_remove == NULL)
4599 mddev->to_remove = &md_redundancy_group;
4600 mddev->pers = NULL;
4601 /* tell userspace to handle 'inactive' */
4602 sysfs_notify_dirent(mddev->sysfs_state);
4603
4604 list_for_each_entry(rdev, &mddev->disks, same_set)
4605 if (rdev->raid_disk >= 0) {
4606 char nm[20];
4607 sprintf(nm, "rd%d", rdev->raid_disk);
4608 sysfs_remove_link(&mddev->kobj, nm);
4609 }
4610
4611 set_capacity(disk, 0);
4612 mddev->changed = 1;
4613
4614 if (mddev->ro)
4615 mddev->ro = 0;
4616 }
4617 if (!mddev->in_sync || mddev->flags) {
4618 /* mark array as shutdown cleanly */
4619 mddev->in_sync = 1;
4620 md_update_sb(mddev, 1);
4621 }
4622 if (mode == 1)
4623 set_disk_ro(disk, 1);
4624 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4625 err = 0;
4626 }
4627 out:
4628 mutex_unlock(&mddev->open_mutex);
4629 if (err)
4630 return err;
4631 /*
4632 * Free resources if final stop
4633 */
4634 if (mode == 0) {
4635
4636 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4637
4638 bitmap_destroy(mddev);
4639 if (mddev->bitmap_info.file) {
4640 restore_bitmap_write_access(mddev->bitmap_info.file);
4641 fput(mddev->bitmap_info.file);
4642 mddev->bitmap_info.file = NULL;
4643 }
4644 mddev->bitmap_info.offset = 0;
4645
4646 export_array(mddev);
4647
4648 mddev->array_sectors = 0;
4649 mddev->external_size = 0;
4650 mddev->dev_sectors = 0;
4651 mddev->raid_disks = 0;
4652 mddev->recovery_cp = 0;
4653 mddev->resync_min = 0;
4654 mddev->resync_max = MaxSector;
4655 mddev->reshape_position = MaxSector;
4656 mddev->external = 0;
4657 mddev->persistent = 0;
4658 mddev->level = LEVEL_NONE;
4659 mddev->clevel[0] = 0;
4660 mddev->flags = 0;
4661 mddev->ro = 0;
4662 mddev->metadata_type[0] = 0;
4663 mddev->chunk_sectors = 0;
4664 mddev->ctime = mddev->utime = 0;
4665 mddev->layout = 0;
4666 mddev->max_disks = 0;
4667 mddev->events = 0;
4668 mddev->delta_disks = 0;
4669 mddev->new_level = LEVEL_NONE;
4670 mddev->new_layout = 0;
4671 mddev->new_chunk_sectors = 0;
4672 mddev->curr_resync = 0;
4673 mddev->resync_mismatches = 0;
4674 mddev->suspend_lo = mddev->suspend_hi = 0;
4675 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4676 mddev->recovery = 0;
4677 mddev->in_sync = 0;
4678 mddev->changed = 0;
4679 mddev->degraded = 0;
4680 mddev->barriers_work = 0;
4681 mddev->safemode = 0;
4682 mddev->bitmap_info.offset = 0;
4683 mddev->bitmap_info.default_offset = 0;
4684 mddev->bitmap_info.chunksize = 0;
4685 mddev->bitmap_info.daemon_sleep = 0;
4686 mddev->bitmap_info.max_write_behind = 0;
4687 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4688 if (mddev->hold_active == UNTIL_STOP)
4689 mddev->hold_active = 0;
4690
4691 } else if (mddev->pers)
4692 printk(KERN_INFO "md: %s switched to read-only mode.\n",
4693 mdname(mddev));
4694 err = 0;
4695 blk_integrity_unregister(disk);
4696 md_new_event(mddev);
4697 sysfs_notify_dirent(mddev->sysfs_state);
4698 return err;
4699 }
4700
4701 #ifndef MODULE
4702 static void autorun_array(mddev_t *mddev)
4703 {
4704 mdk_rdev_t *rdev;
4705 int err;
4706
4707 if (list_empty(&mddev->disks))
4708 return;
4709
4710 printk(KERN_INFO "md: running: ");
4711
4712 list_for_each_entry(rdev, &mddev->disks, same_set) {
4713 char b[BDEVNAME_SIZE];
4714 printk("<%s>", bdevname(rdev->bdev,b));
4715 }
4716 printk("\n");
4717
4718 err = do_md_run(mddev);
4719 if (err) {
4720 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4721 do_md_stop(mddev, 0, 0);
4722 }
4723 }
4724
4725 /*
4726 * lets try to run arrays based on all disks that have arrived
4727 * until now. (those are in pending_raid_disks)
4728 *
4729 * the method: pick the first pending disk, collect all disks with
4730 * the same UUID, remove all from the pending list and put them into
4731 * the 'same_array' list. Then order this list based on superblock
4732 * update time (freshest comes first), kick out 'old' disks and
4733 * compare superblocks. If everything's fine then run it.
4734 *
4735 * If "unit" is allocated, then bump its reference count
4736 */
4737 static void autorun_devices(int part)
4738 {
4739 mdk_rdev_t *rdev0, *rdev, *tmp;
4740 mddev_t *mddev;
4741 char b[BDEVNAME_SIZE];
4742
4743 printk(KERN_INFO "md: autorun ...\n");
4744 while (!list_empty(&pending_raid_disks)) {
4745 int unit;
4746 dev_t dev;
4747 LIST_HEAD(candidates);
4748 rdev0 = list_entry(pending_raid_disks.next,
4749 mdk_rdev_t, same_set);
4750
4751 printk(KERN_INFO "md: considering %s ...\n",
4752 bdevname(rdev0->bdev,b));
4753 INIT_LIST_HEAD(&candidates);
4754 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4755 if (super_90_load(rdev, rdev0, 0) >= 0) {
4756 printk(KERN_INFO "md: adding %s ...\n",
4757 bdevname(rdev->bdev,b));
4758 list_move(&rdev->same_set, &candidates);
4759 }
4760 /*
4761 * now we have a set of devices, with all of them having
4762 * mostly sane superblocks. It's time to allocate the
4763 * mddev.
4764 */
4765 if (part) {
4766 dev = MKDEV(mdp_major,
4767 rdev0->preferred_minor << MdpMinorShift);
4768 unit = MINOR(dev) >> MdpMinorShift;
4769 } else {
4770 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4771 unit = MINOR(dev);
4772 }
4773 if (rdev0->preferred_minor != unit) {
4774 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4775 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4776 break;
4777 }
4778
4779 md_probe(dev, NULL, NULL);
4780 mddev = mddev_find(dev);
4781 if (!mddev || !mddev->gendisk) {
4782 if (mddev)
4783 mddev_put(mddev);
4784 printk(KERN_ERR
4785 "md: cannot allocate memory for md drive.\n");
4786 break;
4787 }
4788 if (mddev_lock(mddev))
4789 printk(KERN_WARNING "md: %s locked, cannot run\n",
4790 mdname(mddev));
4791 else if (mddev->raid_disks || mddev->major_version
4792 || !list_empty(&mddev->disks)) {
4793 printk(KERN_WARNING
4794 "md: %s already running, cannot run %s\n",
4795 mdname(mddev), bdevname(rdev0->bdev,b));
4796 mddev_unlock(mddev);
4797 } else {
4798 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4799 mddev->persistent = 1;
4800 rdev_for_each_list(rdev, tmp, &candidates) {
4801 list_del_init(&rdev->same_set);
4802 if (bind_rdev_to_array(rdev, mddev))
4803 export_rdev(rdev);
4804 }
4805 autorun_array(mddev);
4806 mddev_unlock(mddev);
4807 }
4808 /* on success, candidates will be empty, on error
4809 * it won't...
4810 */
4811 rdev_for_each_list(rdev, tmp, &candidates) {
4812 list_del_init(&rdev->same_set);
4813 export_rdev(rdev);
4814 }
4815 mddev_put(mddev);
4816 }
4817 printk(KERN_INFO "md: ... autorun DONE.\n");
4818 }
4819 #endif /* !MODULE */
4820
4821 static int get_version(void __user * arg)
4822 {
4823 mdu_version_t ver;
4824
4825 ver.major = MD_MAJOR_VERSION;
4826 ver.minor = MD_MINOR_VERSION;
4827 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4828
4829 if (copy_to_user(arg, &ver, sizeof(ver)))
4830 return -EFAULT;
4831
4832 return 0;
4833 }
4834
4835 static int get_array_info(mddev_t * mddev, void __user * arg)
4836 {
4837 mdu_array_info_t info;
4838 int nr,working,insync,failed,spare;
4839 mdk_rdev_t *rdev;
4840
4841 nr=working=insync=failed=spare=0;
4842 list_for_each_entry(rdev, &mddev->disks, same_set) {
4843 nr++;
4844 if (test_bit(Faulty, &rdev->flags))
4845 failed++;
4846 else {
4847 working++;
4848 if (test_bit(In_sync, &rdev->flags))
4849 insync++;
4850 else
4851 spare++;
4852 }
4853 }
4854
4855 info.major_version = mddev->major_version;
4856 info.minor_version = mddev->minor_version;
4857 info.patch_version = MD_PATCHLEVEL_VERSION;
4858 info.ctime = mddev->ctime;
4859 info.level = mddev->level;
4860 info.size = mddev->dev_sectors / 2;
4861 if (info.size != mddev->dev_sectors / 2) /* overflow */
4862 info.size = -1;
4863 info.nr_disks = nr;
4864 info.raid_disks = mddev->raid_disks;
4865 info.md_minor = mddev->md_minor;
4866 info.not_persistent= !mddev->persistent;
4867
4868 info.utime = mddev->utime;
4869 info.state = 0;
4870 if (mddev->in_sync)
4871 info.state = (1<<MD_SB_CLEAN);
4872 if (mddev->bitmap && mddev->bitmap_info.offset)
4873 info.state = (1<<MD_SB_BITMAP_PRESENT);
4874 info.active_disks = insync;
4875 info.working_disks = working;
4876 info.failed_disks = failed;
4877 info.spare_disks = spare;
4878
4879 info.layout = mddev->layout;
4880 info.chunk_size = mddev->chunk_sectors << 9;
4881
4882 if (copy_to_user(arg, &info, sizeof(info)))
4883 return -EFAULT;
4884
4885 return 0;
4886 }
4887
4888 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4889 {
4890 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4891 char *ptr, *buf = NULL;
4892 int err = -ENOMEM;
4893
4894 if (md_allow_write(mddev))
4895 file = kmalloc(sizeof(*file), GFP_NOIO);
4896 else
4897 file = kmalloc(sizeof(*file), GFP_KERNEL);
4898
4899 if (!file)
4900 goto out;
4901
4902 /* bitmap disabled, zero the first byte and copy out */
4903 if (!mddev->bitmap || !mddev->bitmap->file) {
4904 file->pathname[0] = '\0';
4905 goto copy_out;
4906 }
4907
4908 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4909 if (!buf)
4910 goto out;
4911
4912 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4913 if (IS_ERR(ptr))
4914 goto out;
4915
4916 strcpy(file->pathname, ptr);
4917
4918 copy_out:
4919 err = 0;
4920 if (copy_to_user(arg, file, sizeof(*file)))
4921 err = -EFAULT;
4922 out:
4923 kfree(buf);
4924 kfree(file);
4925 return err;
4926 }
4927
4928 static int get_disk_info(mddev_t * mddev, void __user * arg)
4929 {
4930 mdu_disk_info_t info;
4931 mdk_rdev_t *rdev;
4932
4933 if (copy_from_user(&info, arg, sizeof(info)))
4934 return -EFAULT;
4935
4936 rdev = find_rdev_nr(mddev, info.number);
4937 if (rdev) {
4938 info.major = MAJOR(rdev->bdev->bd_dev);
4939 info.minor = MINOR(rdev->bdev->bd_dev);
4940 info.raid_disk = rdev->raid_disk;
4941 info.state = 0;
4942 if (test_bit(Faulty, &rdev->flags))
4943 info.state |= (1<<MD_DISK_FAULTY);
4944 else if (test_bit(In_sync, &rdev->flags)) {
4945 info.state |= (1<<MD_DISK_ACTIVE);
4946 info.state |= (1<<MD_DISK_SYNC);
4947 }
4948 if (test_bit(WriteMostly, &rdev->flags))
4949 info.state |= (1<<MD_DISK_WRITEMOSTLY);
4950 } else {
4951 info.major = info.minor = 0;
4952 info.raid_disk = -1;
4953 info.state = (1<<MD_DISK_REMOVED);
4954 }
4955
4956 if (copy_to_user(arg, &info, sizeof(info)))
4957 return -EFAULT;
4958
4959 return 0;
4960 }
4961
4962 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4963 {
4964 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4965 mdk_rdev_t *rdev;
4966 dev_t dev = MKDEV(info->major,info->minor);
4967
4968 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4969 return -EOVERFLOW;
4970
4971 if (!mddev->raid_disks) {
4972 int err;
4973 /* expecting a device which has a superblock */
4974 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4975 if (IS_ERR(rdev)) {
4976 printk(KERN_WARNING
4977 "md: md_import_device returned %ld\n",
4978 PTR_ERR(rdev));
4979 return PTR_ERR(rdev);
4980 }
4981 if (!list_empty(&mddev->disks)) {
4982 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4983 mdk_rdev_t, same_set);
4984 err = super_types[mddev->major_version]
4985 .load_super(rdev, rdev0, mddev->minor_version);
4986 if (err < 0) {
4987 printk(KERN_WARNING
4988 "md: %s has different UUID to %s\n",
4989 bdevname(rdev->bdev,b),
4990 bdevname(rdev0->bdev,b2));
4991 export_rdev(rdev);
4992 return -EINVAL;
4993 }
4994 }
4995 err = bind_rdev_to_array(rdev, mddev);
4996 if (err)
4997 export_rdev(rdev);
4998 return err;
4999 }
5000
5001 /*
5002 * add_new_disk can be used once the array is assembled
5003 * to add "hot spares". They must already have a superblock
5004 * written
5005 */
5006 if (mddev->pers) {
5007 int err;
5008 if (!mddev->pers->hot_add_disk) {
5009 printk(KERN_WARNING
5010 "%s: personality does not support diskops!\n",
5011 mdname(mddev));
5012 return -EINVAL;
5013 }
5014 if (mddev->persistent)
5015 rdev = md_import_device(dev, mddev->major_version,
5016 mddev->minor_version);
5017 else
5018 rdev = md_import_device(dev, -1, -1);
5019 if (IS_ERR(rdev)) {
5020 printk(KERN_WARNING
5021 "md: md_import_device returned %ld\n",
5022 PTR_ERR(rdev));
5023 return PTR_ERR(rdev);
5024 }
5025 /* set save_raid_disk if appropriate */
5026 if (!mddev->persistent) {
5027 if (info->state & (1<<MD_DISK_SYNC) &&
5028 info->raid_disk < mddev->raid_disks)
5029 rdev->raid_disk = info->raid_disk;
5030 else
5031 rdev->raid_disk = -1;
5032 } else
5033 super_types[mddev->major_version].
5034 validate_super(mddev, rdev);
5035 rdev->saved_raid_disk = rdev->raid_disk;
5036
5037 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5038 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5039 set_bit(WriteMostly, &rdev->flags);
5040 else
5041 clear_bit(WriteMostly, &rdev->flags);
5042
5043 rdev->raid_disk = -1;
5044 err = bind_rdev_to_array(rdev, mddev);
5045 if (!err && !mddev->pers->hot_remove_disk) {
5046 /* If there is hot_add_disk but no hot_remove_disk
5047 * then added disks for geometry changes,
5048 * and should be added immediately.
5049 */
5050 super_types[mddev->major_version].
5051 validate_super(mddev, rdev);
5052 err = mddev->pers->hot_add_disk(mddev, rdev);
5053 if (err)
5054 unbind_rdev_from_array(rdev);
5055 }
5056 if (err)
5057 export_rdev(rdev);
5058 else
5059 sysfs_notify_dirent(rdev->sysfs_state);
5060
5061 md_update_sb(mddev, 1);
5062 if (mddev->degraded)
5063 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5064 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5065 md_wakeup_thread(mddev->thread);
5066 return err;
5067 }
5068
5069 /* otherwise, add_new_disk is only allowed
5070 * for major_version==0 superblocks
5071 */
5072 if (mddev->major_version != 0) {
5073 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5074 mdname(mddev));
5075 return -EINVAL;
5076 }
5077
5078 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5079 int err;
5080 rdev = md_import_device(dev, -1, 0);
5081 if (IS_ERR(rdev)) {
5082 printk(KERN_WARNING
5083 "md: error, md_import_device() returned %ld\n",
5084 PTR_ERR(rdev));
5085 return PTR_ERR(rdev);
5086 }
5087 rdev->desc_nr = info->number;
5088 if (info->raid_disk < mddev->raid_disks)
5089 rdev->raid_disk = info->raid_disk;
5090 else
5091 rdev->raid_disk = -1;
5092
5093 if (rdev->raid_disk < mddev->raid_disks)
5094 if (info->state & (1<<MD_DISK_SYNC))
5095 set_bit(In_sync, &rdev->flags);
5096
5097 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5098 set_bit(WriteMostly, &rdev->flags);
5099
5100 if (!mddev->persistent) {
5101 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5102 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5103 } else
5104 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5105 rdev->sectors = rdev->sb_start;
5106
5107 err = bind_rdev_to_array(rdev, mddev);
5108 if (err) {
5109 export_rdev(rdev);
5110 return err;
5111 }
5112 }
5113
5114 return 0;
5115 }
5116
5117 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
5118 {
5119 char b[BDEVNAME_SIZE];
5120 mdk_rdev_t *rdev;
5121
5122 rdev = find_rdev(mddev, dev);
5123 if (!rdev)
5124 return -ENXIO;
5125
5126 if (rdev->raid_disk >= 0)
5127 goto busy;
5128
5129 kick_rdev_from_array(rdev);
5130 md_update_sb(mddev, 1);
5131 md_new_event(mddev);
5132
5133 return 0;
5134 busy:
5135 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5136 bdevname(rdev->bdev,b), mdname(mddev));
5137 return -EBUSY;
5138 }
5139
5140 static int hot_add_disk(mddev_t * mddev, dev_t dev)
5141 {
5142 char b[BDEVNAME_SIZE];
5143 int err;
5144 mdk_rdev_t *rdev;
5145
5146 if (!mddev->pers)
5147 return -ENODEV;
5148
5149 if (mddev->major_version != 0) {
5150 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5151 " version-0 superblocks.\n",
5152 mdname(mddev));
5153 return -EINVAL;
5154 }
5155 if (!mddev->pers->hot_add_disk) {
5156 printk(KERN_WARNING
5157 "%s: personality does not support diskops!\n",
5158 mdname(mddev));
5159 return -EINVAL;
5160 }
5161
5162 rdev = md_import_device(dev, -1, 0);
5163 if (IS_ERR(rdev)) {
5164 printk(KERN_WARNING
5165 "md: error, md_import_device() returned %ld\n",
5166 PTR_ERR(rdev));
5167 return -EINVAL;
5168 }
5169
5170 if (mddev->persistent)
5171 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5172 else
5173 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5174
5175 rdev->sectors = rdev->sb_start;
5176
5177 if (test_bit(Faulty, &rdev->flags)) {
5178 printk(KERN_WARNING
5179 "md: can not hot-add faulty %s disk to %s!\n",
5180 bdevname(rdev->bdev,b), mdname(mddev));
5181 err = -EINVAL;
5182 goto abort_export;
5183 }
5184 clear_bit(In_sync, &rdev->flags);
5185 rdev->desc_nr = -1;
5186 rdev->saved_raid_disk = -1;
5187 err = bind_rdev_to_array(rdev, mddev);
5188 if (err)
5189 goto abort_export;
5190
5191 /*
5192 * The rest should better be atomic, we can have disk failures
5193 * noticed in interrupt contexts ...
5194 */
5195
5196 rdev->raid_disk = -1;
5197
5198 md_update_sb(mddev, 1);
5199
5200 /*
5201 * Kick recovery, maybe this spare has to be added to the
5202 * array immediately.
5203 */
5204 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5205 md_wakeup_thread(mddev->thread);
5206 md_new_event(mddev);
5207 return 0;
5208
5209 abort_export:
5210 export_rdev(rdev);
5211 return err;
5212 }
5213
5214 static int set_bitmap_file(mddev_t *mddev, int fd)
5215 {
5216 int err;
5217
5218 if (mddev->pers) {
5219 if (!mddev->pers->quiesce)
5220 return -EBUSY;
5221 if (mddev->recovery || mddev->sync_thread)
5222 return -EBUSY;
5223 /* we should be able to change the bitmap.. */
5224 }
5225
5226
5227 if (fd >= 0) {
5228 if (mddev->bitmap)
5229 return -EEXIST; /* cannot add when bitmap is present */
5230 mddev->bitmap_info.file = fget(fd);
5231
5232 if (mddev->bitmap_info.file == NULL) {
5233 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5234 mdname(mddev));
5235 return -EBADF;
5236 }
5237
5238 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5239 if (err) {
5240 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5241 mdname(mddev));
5242 fput(mddev->bitmap_info.file);
5243 mddev->bitmap_info.file = NULL;
5244 return err;
5245 }
5246 mddev->bitmap_info.offset = 0; /* file overrides offset */
5247 } else if (mddev->bitmap == NULL)
5248 return -ENOENT; /* cannot remove what isn't there */
5249 err = 0;
5250 if (mddev->pers) {
5251 mddev->pers->quiesce(mddev, 1);
5252 if (fd >= 0)
5253 err = bitmap_create(mddev);
5254 if (fd < 0 || err) {
5255 bitmap_destroy(mddev);
5256 fd = -1; /* make sure to put the file */
5257 }
5258 mddev->pers->quiesce(mddev, 0);
5259 }
5260 if (fd < 0) {
5261 if (mddev->bitmap_info.file) {
5262 restore_bitmap_write_access(mddev->bitmap_info.file);
5263 fput(mddev->bitmap_info.file);
5264 }
5265 mddev->bitmap_info.file = NULL;
5266 }
5267
5268 return err;
5269 }
5270
5271 /*
5272 * set_array_info is used two different ways
5273 * The original usage is when creating a new array.
5274 * In this usage, raid_disks is > 0 and it together with
5275 * level, size, not_persistent,layout,chunksize determine the
5276 * shape of the array.
5277 * This will always create an array with a type-0.90.0 superblock.
5278 * The newer usage is when assembling an array.
5279 * In this case raid_disks will be 0, and the major_version field is
5280 * use to determine which style super-blocks are to be found on the devices.
5281 * The minor and patch _version numbers are also kept incase the
5282 * super_block handler wishes to interpret them.
5283 */
5284 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5285 {
5286
5287 if (info->raid_disks == 0) {
5288 /* just setting version number for superblock loading */
5289 if (info->major_version < 0 ||
5290 info->major_version >= ARRAY_SIZE(super_types) ||
5291 super_types[info->major_version].name == NULL) {
5292 /* maybe try to auto-load a module? */
5293 printk(KERN_INFO
5294 "md: superblock version %d not known\n",
5295 info->major_version);
5296 return -EINVAL;
5297 }
5298 mddev->major_version = info->major_version;
5299 mddev->minor_version = info->minor_version;
5300 mddev->patch_version = info->patch_version;
5301 mddev->persistent = !info->not_persistent;
5302 /* ensure mddev_put doesn't delete this now that there
5303 * is some minimal configuration.
5304 */
5305 mddev->ctime = get_seconds();
5306 return 0;
5307 }
5308 mddev->major_version = MD_MAJOR_VERSION;
5309 mddev->minor_version = MD_MINOR_VERSION;
5310 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5311 mddev->ctime = get_seconds();
5312
5313 mddev->level = info->level;
5314 mddev->clevel[0] = 0;
5315 mddev->dev_sectors = 2 * (sector_t)info->size;
5316 mddev->raid_disks = info->raid_disks;
5317 /* don't set md_minor, it is determined by which /dev/md* was
5318 * openned
5319 */
5320 if (info->state & (1<<MD_SB_CLEAN))
5321 mddev->recovery_cp = MaxSector;
5322 else
5323 mddev->recovery_cp = 0;
5324 mddev->persistent = ! info->not_persistent;
5325 mddev->external = 0;
5326
5327 mddev->layout = info->layout;
5328 mddev->chunk_sectors = info->chunk_size >> 9;
5329
5330 mddev->max_disks = MD_SB_DISKS;
5331
5332 if (mddev->persistent)
5333 mddev->flags = 0;
5334 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5335
5336 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5337 mddev->bitmap_info.offset = 0;
5338
5339 mddev->reshape_position = MaxSector;
5340
5341 /*
5342 * Generate a 128 bit UUID
5343 */
5344 get_random_bytes(mddev->uuid, 16);
5345
5346 mddev->new_level = mddev->level;
5347 mddev->new_chunk_sectors = mddev->chunk_sectors;
5348 mddev->new_layout = mddev->layout;
5349 mddev->delta_disks = 0;
5350
5351 return 0;
5352 }
5353
5354 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5355 {
5356 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5357
5358 if (mddev->external_size)
5359 return;
5360
5361 mddev->array_sectors = array_sectors;
5362 }
5363 EXPORT_SYMBOL(md_set_array_sectors);
5364
5365 static int update_size(mddev_t *mddev, sector_t num_sectors)
5366 {
5367 mdk_rdev_t *rdev;
5368 int rv;
5369 int fit = (num_sectors == 0);
5370
5371 if (mddev->pers->resize == NULL)
5372 return -EINVAL;
5373 /* The "num_sectors" is the number of sectors of each device that
5374 * is used. This can only make sense for arrays with redundancy.
5375 * linear and raid0 always use whatever space is available. We can only
5376 * consider changing this number if no resync or reconstruction is
5377 * happening, and if the new size is acceptable. It must fit before the
5378 * sb_start or, if that is <data_offset, it must fit before the size
5379 * of each device. If num_sectors is zero, we find the largest size
5380 * that fits.
5381
5382 */
5383 if (mddev->sync_thread)
5384 return -EBUSY;
5385 if (mddev->bitmap)
5386 /* Sorry, cannot grow a bitmap yet, just remove it,
5387 * grow, and re-add.
5388 */
5389 return -EBUSY;
5390 list_for_each_entry(rdev, &mddev->disks, same_set) {
5391 sector_t avail = rdev->sectors;
5392
5393 if (fit && (num_sectors == 0 || num_sectors > avail))
5394 num_sectors = avail;
5395 if (avail < num_sectors)
5396 return -ENOSPC;
5397 }
5398 rv = mddev->pers->resize(mddev, num_sectors);
5399 if (!rv)
5400 revalidate_disk(mddev->gendisk);
5401 return rv;
5402 }
5403
5404 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5405 {
5406 int rv;
5407 /* change the number of raid disks */
5408 if (mddev->pers->check_reshape == NULL)
5409 return -EINVAL;
5410 if (raid_disks <= 0 ||
5411 (mddev->max_disks && raid_disks >= mddev->max_disks))
5412 return -EINVAL;
5413 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5414 return -EBUSY;
5415 mddev->delta_disks = raid_disks - mddev->raid_disks;
5416
5417 rv = mddev->pers->check_reshape(mddev);
5418 return rv;
5419 }
5420
5421
5422 /*
5423 * update_array_info is used to change the configuration of an
5424 * on-line array.
5425 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5426 * fields in the info are checked against the array.
5427 * Any differences that cannot be handled will cause an error.
5428 * Normally, only one change can be managed at a time.
5429 */
5430 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5431 {
5432 int rv = 0;
5433 int cnt = 0;
5434 int state = 0;
5435
5436 /* calculate expected state,ignoring low bits */
5437 if (mddev->bitmap && mddev->bitmap_info.offset)
5438 state |= (1 << MD_SB_BITMAP_PRESENT);
5439
5440 if (mddev->major_version != info->major_version ||
5441 mddev->minor_version != info->minor_version ||
5442 /* mddev->patch_version != info->patch_version || */
5443 mddev->ctime != info->ctime ||
5444 mddev->level != info->level ||
5445 /* mddev->layout != info->layout || */
5446 !mddev->persistent != info->not_persistent||
5447 mddev->chunk_sectors != info->chunk_size >> 9 ||
5448 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5449 ((state^info->state) & 0xfffffe00)
5450 )
5451 return -EINVAL;
5452 /* Check there is only one change */
5453 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5454 cnt++;
5455 if (mddev->raid_disks != info->raid_disks)
5456 cnt++;
5457 if (mddev->layout != info->layout)
5458 cnt++;
5459 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5460 cnt++;
5461 if (cnt == 0)
5462 return 0;
5463 if (cnt > 1)
5464 return -EINVAL;
5465
5466 if (mddev->layout != info->layout) {
5467 /* Change layout
5468 * we don't need to do anything at the md level, the
5469 * personality will take care of it all.
5470 */
5471 if (mddev->pers->check_reshape == NULL)
5472 return -EINVAL;
5473 else {
5474 mddev->new_layout = info->layout;
5475 rv = mddev->pers->check_reshape(mddev);
5476 if (rv)
5477 mddev->new_layout = mddev->layout;
5478 return rv;
5479 }
5480 }
5481 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5482 rv = update_size(mddev, (sector_t)info->size * 2);
5483
5484 if (mddev->raid_disks != info->raid_disks)
5485 rv = update_raid_disks(mddev, info->raid_disks);
5486
5487 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5488 if (mddev->pers->quiesce == NULL)
5489 return -EINVAL;
5490 if (mddev->recovery || mddev->sync_thread)
5491 return -EBUSY;
5492 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5493 /* add the bitmap */
5494 if (mddev->bitmap)
5495 return -EEXIST;
5496 if (mddev->bitmap_info.default_offset == 0)
5497 return -EINVAL;
5498 mddev->bitmap_info.offset =
5499 mddev->bitmap_info.default_offset;
5500 mddev->pers->quiesce(mddev, 1);
5501 rv = bitmap_create(mddev);
5502 if (rv)
5503 bitmap_destroy(mddev);
5504 mddev->pers->quiesce(mddev, 0);
5505 } else {
5506 /* remove the bitmap */
5507 if (!mddev->bitmap)
5508 return -ENOENT;
5509 if (mddev->bitmap->file)
5510 return -EINVAL;
5511 mddev->pers->quiesce(mddev, 1);
5512 bitmap_destroy(mddev);
5513 mddev->pers->quiesce(mddev, 0);
5514 mddev->bitmap_info.offset = 0;
5515 }
5516 }
5517 md_update_sb(mddev, 1);
5518 return rv;
5519 }
5520
5521 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5522 {
5523 mdk_rdev_t *rdev;
5524
5525 if (mddev->pers == NULL)
5526 return -ENODEV;
5527
5528 rdev = find_rdev(mddev, dev);
5529 if (!rdev)
5530 return -ENODEV;
5531
5532 md_error(mddev, rdev);
5533 return 0;
5534 }
5535
5536 /*
5537 * We have a problem here : there is no easy way to give a CHS
5538 * virtual geometry. We currently pretend that we have a 2 heads
5539 * 4 sectors (with a BIG number of cylinders...). This drives
5540 * dosfs just mad... ;-)
5541 */
5542 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5543 {
5544 mddev_t *mddev = bdev->bd_disk->private_data;
5545
5546 geo->heads = 2;
5547 geo->sectors = 4;
5548 geo->cylinders = get_capacity(mddev->gendisk) / 8;
5549 return 0;
5550 }
5551
5552 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5553 unsigned int cmd, unsigned long arg)
5554 {
5555 int err = 0;
5556 void __user *argp = (void __user *)arg;
5557 mddev_t *mddev = NULL;
5558 int ro;
5559
5560 if (!capable(CAP_SYS_ADMIN))
5561 return -EACCES;
5562
5563 /*
5564 * Commands dealing with the RAID driver but not any
5565 * particular array:
5566 */
5567 switch (cmd)
5568 {
5569 case RAID_VERSION:
5570 err = get_version(argp);
5571 goto done;
5572
5573 case PRINT_RAID_DEBUG:
5574 err = 0;
5575 md_print_devices();
5576 goto done;
5577
5578 #ifndef MODULE
5579 case RAID_AUTORUN:
5580 err = 0;
5581 autostart_arrays(arg);
5582 goto done;
5583 #endif
5584 default:;
5585 }
5586
5587 /*
5588 * Commands creating/starting a new array:
5589 */
5590
5591 mddev = bdev->bd_disk->private_data;
5592
5593 if (!mddev) {
5594 BUG();
5595 goto abort;
5596 }
5597
5598 err = mddev_lock(mddev);
5599 if (err) {
5600 printk(KERN_INFO
5601 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5602 err, cmd);
5603 goto abort;
5604 }
5605
5606 switch (cmd)
5607 {
5608 case SET_ARRAY_INFO:
5609 {
5610 mdu_array_info_t info;
5611 if (!arg)
5612 memset(&info, 0, sizeof(info));
5613 else if (copy_from_user(&info, argp, sizeof(info))) {
5614 err = -EFAULT;
5615 goto abort_unlock;
5616 }
5617 if (mddev->pers) {
5618 err = update_array_info(mddev, &info);
5619 if (err) {
5620 printk(KERN_WARNING "md: couldn't update"
5621 " array info. %d\n", err);
5622 goto abort_unlock;
5623 }
5624 goto done_unlock;
5625 }
5626 if (!list_empty(&mddev->disks)) {
5627 printk(KERN_WARNING
5628 "md: array %s already has disks!\n",
5629 mdname(mddev));
5630 err = -EBUSY;
5631 goto abort_unlock;
5632 }
5633 if (mddev->raid_disks) {
5634 printk(KERN_WARNING
5635 "md: array %s already initialised!\n",
5636 mdname(mddev));
5637 err = -EBUSY;
5638 goto abort_unlock;
5639 }
5640 err = set_array_info(mddev, &info);
5641 if (err) {
5642 printk(KERN_WARNING "md: couldn't set"
5643 " array info. %d\n", err);
5644 goto abort_unlock;
5645 }
5646 }
5647 goto done_unlock;
5648
5649 default:;
5650 }
5651
5652 /*
5653 * Commands querying/configuring an existing array:
5654 */
5655 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5656 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5657 if ((!mddev->raid_disks && !mddev->external)
5658 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5659 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5660 && cmd != GET_BITMAP_FILE) {
5661 err = -ENODEV;
5662 goto abort_unlock;
5663 }
5664
5665 /*
5666 * Commands even a read-only array can execute:
5667 */
5668 switch (cmd)
5669 {
5670 case GET_ARRAY_INFO:
5671 err = get_array_info(mddev, argp);
5672 goto done_unlock;
5673
5674 case GET_BITMAP_FILE:
5675 err = get_bitmap_file(mddev, argp);
5676 goto done_unlock;
5677
5678 case GET_DISK_INFO:
5679 err = get_disk_info(mddev, argp);
5680 goto done_unlock;
5681
5682 case RESTART_ARRAY_RW:
5683 err = restart_array(mddev);
5684 goto done_unlock;
5685
5686 case STOP_ARRAY:
5687 err = do_md_stop(mddev, 0, 1);
5688 goto done_unlock;
5689
5690 case STOP_ARRAY_RO:
5691 err = do_md_stop(mddev, 1, 1);
5692 goto done_unlock;
5693
5694 case BLKROSET:
5695 if (get_user(ro, (int __user *)(arg))) {
5696 err = -EFAULT;
5697 goto done_unlock;
5698 }
5699 err = -EINVAL;
5700
5701 /* if the bdev is going readonly the value of mddev->ro
5702 * does not matter, no writes are coming
5703 */
5704 if (ro)
5705 goto done_unlock;
5706
5707 /* are we are already prepared for writes? */
5708 if (mddev->ro != 1)
5709 goto done_unlock;
5710
5711 /* transitioning to readauto need only happen for
5712 * arrays that call md_write_start
5713 */
5714 if (mddev->pers) {
5715 err = restart_array(mddev);
5716 if (err == 0) {
5717 mddev->ro = 2;
5718 set_disk_ro(mddev->gendisk, 0);
5719 }
5720 }
5721 goto done_unlock;
5722 }
5723
5724 /*
5725 * The remaining ioctls are changing the state of the
5726 * superblock, so we do not allow them on read-only arrays.
5727 * However non-MD ioctls (e.g. get-size) will still come through
5728 * here and hit the 'default' below, so only disallow
5729 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5730 */
5731 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5732 if (mddev->ro == 2) {
5733 mddev->ro = 0;
5734 sysfs_notify_dirent(mddev->sysfs_state);
5735 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5736 md_wakeup_thread(mddev->thread);
5737 } else {
5738 err = -EROFS;
5739 goto abort_unlock;
5740 }
5741 }
5742
5743 switch (cmd)
5744 {
5745 case ADD_NEW_DISK:
5746 {
5747 mdu_disk_info_t info;
5748 if (copy_from_user(&info, argp, sizeof(info)))
5749 err = -EFAULT;
5750 else
5751 err = add_new_disk(mddev, &info);
5752 goto done_unlock;
5753 }
5754
5755 case HOT_REMOVE_DISK:
5756 err = hot_remove_disk(mddev, new_decode_dev(arg));
5757 goto done_unlock;
5758
5759 case HOT_ADD_DISK:
5760 err = hot_add_disk(mddev, new_decode_dev(arg));
5761 goto done_unlock;
5762
5763 case SET_DISK_FAULTY:
5764 err = set_disk_faulty(mddev, new_decode_dev(arg));
5765 goto done_unlock;
5766
5767 case RUN_ARRAY:
5768 err = do_md_run(mddev);
5769 goto done_unlock;
5770
5771 case SET_BITMAP_FILE:
5772 err = set_bitmap_file(mddev, (int)arg);
5773 goto done_unlock;
5774
5775 default:
5776 err = -EINVAL;
5777 goto abort_unlock;
5778 }
5779
5780 done_unlock:
5781 abort_unlock:
5782 if (mddev->hold_active == UNTIL_IOCTL &&
5783 err != -EINVAL)
5784 mddev->hold_active = 0;
5785 mddev_unlock(mddev);
5786
5787 return err;
5788 done:
5789 if (err)
5790 MD_BUG();
5791 abort:
5792 return err;
5793 }
5794 #ifdef CONFIG_COMPAT
5795 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
5796 unsigned int cmd, unsigned long arg)
5797 {
5798 switch (cmd) {
5799 case HOT_REMOVE_DISK:
5800 case HOT_ADD_DISK:
5801 case SET_DISK_FAULTY:
5802 case SET_BITMAP_FILE:
5803 /* These take in integer arg, do not convert */
5804 break;
5805 default:
5806 arg = (unsigned long)compat_ptr(arg);
5807 break;
5808 }
5809
5810 return md_ioctl(bdev, mode, cmd, arg);
5811 }
5812 #endif /* CONFIG_COMPAT */
5813
5814 static int md_open(struct block_device *bdev, fmode_t mode)
5815 {
5816 /*
5817 * Succeed if we can lock the mddev, which confirms that
5818 * it isn't being stopped right now.
5819 */
5820 mddev_t *mddev = mddev_find(bdev->bd_dev);
5821 int err;
5822
5823 if (mddev->gendisk != bdev->bd_disk) {
5824 /* we are racing with mddev_put which is discarding this
5825 * bd_disk.
5826 */
5827 mddev_put(mddev);
5828 /* Wait until bdev->bd_disk is definitely gone */
5829 flush_scheduled_work();
5830 /* Then retry the open from the top */
5831 return -ERESTARTSYS;
5832 }
5833 BUG_ON(mddev != bdev->bd_disk->private_data);
5834
5835 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
5836 goto out;
5837
5838 err = 0;
5839 atomic_inc(&mddev->openers);
5840 mutex_unlock(&mddev->open_mutex);
5841
5842 check_disk_change(bdev);
5843 out:
5844 return err;
5845 }
5846
5847 static int md_release(struct gendisk *disk, fmode_t mode)
5848 {
5849 mddev_t *mddev = disk->private_data;
5850
5851 BUG_ON(!mddev);
5852 atomic_dec(&mddev->openers);
5853 mddev_put(mddev);
5854
5855 return 0;
5856 }
5857
5858 static int md_media_changed(struct gendisk *disk)
5859 {
5860 mddev_t *mddev = disk->private_data;
5861
5862 return mddev->changed;
5863 }
5864
5865 static int md_revalidate(struct gendisk *disk)
5866 {
5867 mddev_t *mddev = disk->private_data;
5868
5869 mddev->changed = 0;
5870 return 0;
5871 }
5872 static const struct block_device_operations md_fops =
5873 {
5874 .owner = THIS_MODULE,
5875 .open = md_open,
5876 .release = md_release,
5877 .ioctl = md_ioctl,
5878 #ifdef CONFIG_COMPAT
5879 .compat_ioctl = md_compat_ioctl,
5880 #endif
5881 .getgeo = md_getgeo,
5882 .media_changed = md_media_changed,
5883 .revalidate_disk= md_revalidate,
5884 };
5885
5886 static int md_thread(void * arg)
5887 {
5888 mdk_thread_t *thread = arg;
5889
5890 /*
5891 * md_thread is a 'system-thread', it's priority should be very
5892 * high. We avoid resource deadlocks individually in each
5893 * raid personality. (RAID5 does preallocation) We also use RR and
5894 * the very same RT priority as kswapd, thus we will never get
5895 * into a priority inversion deadlock.
5896 *
5897 * we definitely have to have equal or higher priority than
5898 * bdflush, otherwise bdflush will deadlock if there are too
5899 * many dirty RAID5 blocks.
5900 */
5901
5902 allow_signal(SIGKILL);
5903 while (!kthread_should_stop()) {
5904
5905 /* We need to wait INTERRUPTIBLE so that
5906 * we don't add to the load-average.
5907 * That means we need to be sure no signals are
5908 * pending
5909 */
5910 if (signal_pending(current))
5911 flush_signals(current);
5912
5913 wait_event_interruptible_timeout
5914 (thread->wqueue,
5915 test_bit(THREAD_WAKEUP, &thread->flags)
5916 || kthread_should_stop(),
5917 thread->timeout);
5918
5919 clear_bit(THREAD_WAKEUP, &thread->flags);
5920
5921 thread->run(thread->mddev);
5922 }
5923
5924 return 0;
5925 }
5926
5927 void md_wakeup_thread(mdk_thread_t *thread)
5928 {
5929 if (thread) {
5930 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5931 set_bit(THREAD_WAKEUP, &thread->flags);
5932 wake_up(&thread->wqueue);
5933 }
5934 }
5935
5936 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5937 const char *name)
5938 {
5939 mdk_thread_t *thread;
5940
5941 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5942 if (!thread)
5943 return NULL;
5944
5945 init_waitqueue_head(&thread->wqueue);
5946
5947 thread->run = run;
5948 thread->mddev = mddev;
5949 thread->timeout = MAX_SCHEDULE_TIMEOUT;
5950 thread->tsk = kthread_run(md_thread, thread,
5951 "%s_%s",
5952 mdname(thread->mddev),
5953 name ?: mddev->pers->name);
5954 if (IS_ERR(thread->tsk)) {
5955 kfree(thread);
5956 return NULL;
5957 }
5958 return thread;
5959 }
5960
5961 void md_unregister_thread(mdk_thread_t *thread)
5962 {
5963 if (!thread)
5964 return;
5965 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5966
5967 kthread_stop(thread->tsk);
5968 kfree(thread);
5969 }
5970
5971 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5972 {
5973 if (!mddev) {
5974 MD_BUG();
5975 return;
5976 }
5977
5978 if (!rdev || test_bit(Faulty, &rdev->flags))
5979 return;
5980
5981 if (mddev->external)
5982 set_bit(Blocked, &rdev->flags);
5983 /*
5984 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5985 mdname(mddev),
5986 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5987 __builtin_return_address(0),__builtin_return_address(1),
5988 __builtin_return_address(2),__builtin_return_address(3));
5989 */
5990 if (!mddev->pers)
5991 return;
5992 if (!mddev->pers->error_handler)
5993 return;
5994 mddev->pers->error_handler(mddev,rdev);
5995 if (mddev->degraded)
5996 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5997 sysfs_notify_dirent(rdev->sysfs_state);
5998 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5999 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6000 md_wakeup_thread(mddev->thread);
6001 md_new_event_inintr(mddev);
6002 }
6003
6004 /* seq_file implementation /proc/mdstat */
6005
6006 static void status_unused(struct seq_file *seq)
6007 {
6008 int i = 0;
6009 mdk_rdev_t *rdev;
6010
6011 seq_printf(seq, "unused devices: ");
6012
6013 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6014 char b[BDEVNAME_SIZE];
6015 i++;
6016 seq_printf(seq, "%s ",
6017 bdevname(rdev->bdev,b));
6018 }
6019 if (!i)
6020 seq_printf(seq, "<none>");
6021
6022 seq_printf(seq, "\n");
6023 }
6024
6025
6026 static void status_resync(struct seq_file *seq, mddev_t * mddev)
6027 {
6028 sector_t max_sectors, resync, res;
6029 unsigned long dt, db;
6030 sector_t rt;
6031 int scale;
6032 unsigned int per_milli;
6033
6034 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
6035
6036 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6037 max_sectors = mddev->resync_max_sectors;
6038 else
6039 max_sectors = mddev->dev_sectors;
6040
6041 /*
6042 * Should not happen.
6043 */
6044 if (!max_sectors) {
6045 MD_BUG();
6046 return;
6047 }
6048 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6049 * in a sector_t, and (max_sectors>>scale) will fit in a
6050 * u32, as those are the requirements for sector_div.
6051 * Thus 'scale' must be at least 10
6052 */
6053 scale = 10;
6054 if (sizeof(sector_t) > sizeof(unsigned long)) {
6055 while ( max_sectors/2 > (1ULL<<(scale+32)))
6056 scale++;
6057 }
6058 res = (resync>>scale)*1000;
6059 sector_div(res, (u32)((max_sectors>>scale)+1));
6060
6061 per_milli = res;
6062 {
6063 int i, x = per_milli/50, y = 20-x;
6064 seq_printf(seq, "[");
6065 for (i = 0; i < x; i++)
6066 seq_printf(seq, "=");
6067 seq_printf(seq, ">");
6068 for (i = 0; i < y; i++)
6069 seq_printf(seq, ".");
6070 seq_printf(seq, "] ");
6071 }
6072 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6073 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6074 "reshape" :
6075 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6076 "check" :
6077 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6078 "resync" : "recovery"))),
6079 per_milli/10, per_milli % 10,
6080 (unsigned long long) resync/2,
6081 (unsigned long long) max_sectors/2);
6082
6083 /*
6084 * dt: time from mark until now
6085 * db: blocks written from mark until now
6086 * rt: remaining time
6087 *
6088 * rt is a sector_t, so could be 32bit or 64bit.
6089 * So we divide before multiply in case it is 32bit and close
6090 * to the limit.
6091 * We scale the divisor (db) by 32 to avoid loosing precision
6092 * near the end of resync when the number of remaining sectors
6093 * is close to 'db'.
6094 * We then divide rt by 32 after multiplying by db to compensate.
6095 * The '+1' avoids division by zero if db is very small.
6096 */
6097 dt = ((jiffies - mddev->resync_mark) / HZ);
6098 if (!dt) dt++;
6099 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6100 - mddev->resync_mark_cnt;
6101
6102 rt = max_sectors - resync; /* number of remaining sectors */
6103 sector_div(rt, db/32+1);
6104 rt *= dt;
6105 rt >>= 5;
6106
6107 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6108 ((unsigned long)rt % 60)/6);
6109
6110 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6111 }
6112
6113 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6114 {
6115 struct list_head *tmp;
6116 loff_t l = *pos;
6117 mddev_t *mddev;
6118
6119 if (l >= 0x10000)
6120 return NULL;
6121 if (!l--)
6122 /* header */
6123 return (void*)1;
6124
6125 spin_lock(&all_mddevs_lock);
6126 list_for_each(tmp,&all_mddevs)
6127 if (!l--) {
6128 mddev = list_entry(tmp, mddev_t, all_mddevs);
6129 mddev_get(mddev);
6130 spin_unlock(&all_mddevs_lock);
6131 return mddev;
6132 }
6133 spin_unlock(&all_mddevs_lock);
6134 if (!l--)
6135 return (void*)2;/* tail */
6136 return NULL;
6137 }
6138
6139 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6140 {
6141 struct list_head *tmp;
6142 mddev_t *next_mddev, *mddev = v;
6143
6144 ++*pos;
6145 if (v == (void*)2)
6146 return NULL;
6147
6148 spin_lock(&all_mddevs_lock);
6149 if (v == (void*)1)
6150 tmp = all_mddevs.next;
6151 else
6152 tmp = mddev->all_mddevs.next;
6153 if (tmp != &all_mddevs)
6154 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
6155 else {
6156 next_mddev = (void*)2;
6157 *pos = 0x10000;
6158 }
6159 spin_unlock(&all_mddevs_lock);
6160
6161 if (v != (void*)1)
6162 mddev_put(mddev);
6163 return next_mddev;
6164
6165 }
6166
6167 static void md_seq_stop(struct seq_file *seq, void *v)
6168 {
6169 mddev_t *mddev = v;
6170
6171 if (mddev && v != (void*)1 && v != (void*)2)
6172 mddev_put(mddev);
6173 }
6174
6175 struct mdstat_info {
6176 int event;
6177 };
6178
6179 static int md_seq_show(struct seq_file *seq, void *v)
6180 {
6181 mddev_t *mddev = v;
6182 sector_t sectors;
6183 mdk_rdev_t *rdev;
6184 struct mdstat_info *mi = seq->private;
6185 struct bitmap *bitmap;
6186
6187 if (v == (void*)1) {
6188 struct mdk_personality *pers;
6189 seq_printf(seq, "Personalities : ");
6190 spin_lock(&pers_lock);
6191 list_for_each_entry(pers, &pers_list, list)
6192 seq_printf(seq, "[%s] ", pers->name);
6193
6194 spin_unlock(&pers_lock);
6195 seq_printf(seq, "\n");
6196 mi->event = atomic_read(&md_event_count);
6197 return 0;
6198 }
6199 if (v == (void*)2) {
6200 status_unused(seq);
6201 return 0;
6202 }
6203
6204 if (mddev_lock(mddev) < 0)
6205 return -EINTR;
6206
6207 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6208 seq_printf(seq, "%s : %sactive", mdname(mddev),
6209 mddev->pers ? "" : "in");
6210 if (mddev->pers) {
6211 if (mddev->ro==1)
6212 seq_printf(seq, " (read-only)");
6213 if (mddev->ro==2)
6214 seq_printf(seq, " (auto-read-only)");
6215 seq_printf(seq, " %s", mddev->pers->name);
6216 }
6217
6218 sectors = 0;
6219 list_for_each_entry(rdev, &mddev->disks, same_set) {
6220 char b[BDEVNAME_SIZE];
6221 seq_printf(seq, " %s[%d]",
6222 bdevname(rdev->bdev,b), rdev->desc_nr);
6223 if (test_bit(WriteMostly, &rdev->flags))
6224 seq_printf(seq, "(W)");
6225 if (test_bit(Faulty, &rdev->flags)) {
6226 seq_printf(seq, "(F)");
6227 continue;
6228 } else if (rdev->raid_disk < 0)
6229 seq_printf(seq, "(S)"); /* spare */
6230 sectors += rdev->sectors;
6231 }
6232
6233 if (!list_empty(&mddev->disks)) {
6234 if (mddev->pers)
6235 seq_printf(seq, "\n %llu blocks",
6236 (unsigned long long)
6237 mddev->array_sectors / 2);
6238 else
6239 seq_printf(seq, "\n %llu blocks",
6240 (unsigned long long)sectors / 2);
6241 }
6242 if (mddev->persistent) {
6243 if (mddev->major_version != 0 ||
6244 mddev->minor_version != 90) {
6245 seq_printf(seq," super %d.%d",
6246 mddev->major_version,
6247 mddev->minor_version);
6248 }
6249 } else if (mddev->external)
6250 seq_printf(seq, " super external:%s",
6251 mddev->metadata_type);
6252 else
6253 seq_printf(seq, " super non-persistent");
6254
6255 if (mddev->pers) {
6256 mddev->pers->status(seq, mddev);
6257 seq_printf(seq, "\n ");
6258 if (mddev->pers->sync_request) {
6259 if (mddev->curr_resync > 2) {
6260 status_resync(seq, mddev);
6261 seq_printf(seq, "\n ");
6262 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6263 seq_printf(seq, "\tresync=DELAYED\n ");
6264 else if (mddev->recovery_cp < MaxSector)
6265 seq_printf(seq, "\tresync=PENDING\n ");
6266 }
6267 } else
6268 seq_printf(seq, "\n ");
6269
6270 if ((bitmap = mddev->bitmap)) {
6271 unsigned long chunk_kb;
6272 unsigned long flags;
6273 spin_lock_irqsave(&bitmap->lock, flags);
6274 chunk_kb = mddev->bitmap_info.chunksize >> 10;
6275 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6276 "%lu%s chunk",
6277 bitmap->pages - bitmap->missing_pages,
6278 bitmap->pages,
6279 (bitmap->pages - bitmap->missing_pages)
6280 << (PAGE_SHIFT - 10),
6281 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
6282 chunk_kb ? "KB" : "B");
6283 if (bitmap->file) {
6284 seq_printf(seq, ", file: ");
6285 seq_path(seq, &bitmap->file->f_path, " \t\n");
6286 }
6287
6288 seq_printf(seq, "\n");
6289 spin_unlock_irqrestore(&bitmap->lock, flags);
6290 }
6291
6292 seq_printf(seq, "\n");
6293 }
6294 mddev_unlock(mddev);
6295
6296 return 0;
6297 }
6298
6299 static const struct seq_operations md_seq_ops = {
6300 .start = md_seq_start,
6301 .next = md_seq_next,
6302 .stop = md_seq_stop,
6303 .show = md_seq_show,
6304 };
6305
6306 static int md_seq_open(struct inode *inode, struct file *file)
6307 {
6308 int error;
6309 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6310 if (mi == NULL)
6311 return -ENOMEM;
6312
6313 error = seq_open(file, &md_seq_ops);
6314 if (error)
6315 kfree(mi);
6316 else {
6317 struct seq_file *p = file->private_data;
6318 p->private = mi;
6319 mi->event = atomic_read(&md_event_count);
6320 }
6321 return error;
6322 }
6323
6324 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6325 {
6326 struct seq_file *m = filp->private_data;
6327 struct mdstat_info *mi = m->private;
6328 int mask;
6329
6330 poll_wait(filp, &md_event_waiters, wait);
6331
6332 /* always allow read */
6333 mask = POLLIN | POLLRDNORM;
6334
6335 if (mi->event != atomic_read(&md_event_count))
6336 mask |= POLLERR | POLLPRI;
6337 return mask;
6338 }
6339
6340 static const struct file_operations md_seq_fops = {
6341 .owner = THIS_MODULE,
6342 .open = md_seq_open,
6343 .read = seq_read,
6344 .llseek = seq_lseek,
6345 .release = seq_release_private,
6346 .poll = mdstat_poll,
6347 };
6348
6349 int register_md_personality(struct mdk_personality *p)
6350 {
6351 spin_lock(&pers_lock);
6352 list_add_tail(&p->list, &pers_list);
6353 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6354 spin_unlock(&pers_lock);
6355 return 0;
6356 }
6357
6358 int unregister_md_personality(struct mdk_personality *p)
6359 {
6360 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6361 spin_lock(&pers_lock);
6362 list_del_init(&p->list);
6363 spin_unlock(&pers_lock);
6364 return 0;
6365 }
6366
6367 static int is_mddev_idle(mddev_t *mddev, int init)
6368 {
6369 mdk_rdev_t * rdev;
6370 int idle;
6371 int curr_events;
6372
6373 idle = 1;
6374 rcu_read_lock();
6375 rdev_for_each_rcu(rdev, mddev) {
6376 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6377 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6378 (int)part_stat_read(&disk->part0, sectors[1]) -
6379 atomic_read(&disk->sync_io);
6380 /* sync IO will cause sync_io to increase before the disk_stats
6381 * as sync_io is counted when a request starts, and
6382 * disk_stats is counted when it completes.
6383 * So resync activity will cause curr_events to be smaller than
6384 * when there was no such activity.
6385 * non-sync IO will cause disk_stat to increase without
6386 * increasing sync_io so curr_events will (eventually)
6387 * be larger than it was before. Once it becomes
6388 * substantially larger, the test below will cause
6389 * the array to appear non-idle, and resync will slow
6390 * down.
6391 * If there is a lot of outstanding resync activity when
6392 * we set last_event to curr_events, then all that activity
6393 * completing might cause the array to appear non-idle
6394 * and resync will be slowed down even though there might
6395 * not have been non-resync activity. This will only
6396 * happen once though. 'last_events' will soon reflect
6397 * the state where there is little or no outstanding
6398 * resync requests, and further resync activity will
6399 * always make curr_events less than last_events.
6400 *
6401 */
6402 if (init || curr_events - rdev->last_events > 64) {
6403 rdev->last_events = curr_events;
6404 idle = 0;
6405 }
6406 }
6407 rcu_read_unlock();
6408 return idle;
6409 }
6410
6411 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6412 {
6413 /* another "blocks" (512byte) blocks have been synced */
6414 atomic_sub(blocks, &mddev->recovery_active);
6415 wake_up(&mddev->recovery_wait);
6416 if (!ok) {
6417 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6418 md_wakeup_thread(mddev->thread);
6419 // stop recovery, signal do_sync ....
6420 }
6421 }
6422
6423
6424 /* md_write_start(mddev, bi)
6425 * If we need to update some array metadata (e.g. 'active' flag
6426 * in superblock) before writing, schedule a superblock update
6427 * and wait for it to complete.
6428 */
6429 void md_write_start(mddev_t *mddev, struct bio *bi)
6430 {
6431 int did_change = 0;
6432 if (bio_data_dir(bi) != WRITE)
6433 return;
6434
6435 BUG_ON(mddev->ro == 1);
6436 if (mddev->ro == 2) {
6437 /* need to switch to read/write */
6438 mddev->ro = 0;
6439 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6440 md_wakeup_thread(mddev->thread);
6441 md_wakeup_thread(mddev->sync_thread);
6442 did_change = 1;
6443 }
6444 atomic_inc(&mddev->writes_pending);
6445 if (mddev->safemode == 1)
6446 mddev->safemode = 0;
6447 if (mddev->in_sync) {
6448 spin_lock_irq(&mddev->write_lock);
6449 if (mddev->in_sync) {
6450 mddev->in_sync = 0;
6451 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6452 md_wakeup_thread(mddev->thread);
6453 did_change = 1;
6454 }
6455 spin_unlock_irq(&mddev->write_lock);
6456 }
6457 if (did_change)
6458 sysfs_notify_dirent(mddev->sysfs_state);
6459 wait_event(mddev->sb_wait,
6460 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
6461 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6462 }
6463
6464 void md_write_end(mddev_t *mddev)
6465 {
6466 if (atomic_dec_and_test(&mddev->writes_pending)) {
6467 if (mddev->safemode == 2)
6468 md_wakeup_thread(mddev->thread);
6469 else if (mddev->safemode_delay)
6470 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6471 }
6472 }
6473
6474 /* md_allow_write(mddev)
6475 * Calling this ensures that the array is marked 'active' so that writes
6476 * may proceed without blocking. It is important to call this before
6477 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6478 * Must be called with mddev_lock held.
6479 *
6480 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6481 * is dropped, so return -EAGAIN after notifying userspace.
6482 */
6483 int md_allow_write(mddev_t *mddev)
6484 {
6485 if (!mddev->pers)
6486 return 0;
6487 if (mddev->ro)
6488 return 0;
6489 if (!mddev->pers->sync_request)
6490 return 0;
6491
6492 spin_lock_irq(&mddev->write_lock);
6493 if (mddev->in_sync) {
6494 mddev->in_sync = 0;
6495 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6496 if (mddev->safemode_delay &&
6497 mddev->safemode == 0)
6498 mddev->safemode = 1;
6499 spin_unlock_irq(&mddev->write_lock);
6500 md_update_sb(mddev, 0);
6501 sysfs_notify_dirent(mddev->sysfs_state);
6502 } else
6503 spin_unlock_irq(&mddev->write_lock);
6504
6505 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
6506 return -EAGAIN;
6507 else
6508 return 0;
6509 }
6510 EXPORT_SYMBOL_GPL(md_allow_write);
6511
6512 #define SYNC_MARKS 10
6513 #define SYNC_MARK_STEP (3*HZ)
6514 void md_do_sync(mddev_t *mddev)
6515 {
6516 mddev_t *mddev2;
6517 unsigned int currspeed = 0,
6518 window;
6519 sector_t max_sectors,j, io_sectors;
6520 unsigned long mark[SYNC_MARKS];
6521 sector_t mark_cnt[SYNC_MARKS];
6522 int last_mark,m;
6523 struct list_head *tmp;
6524 sector_t last_check;
6525 int skipped = 0;
6526 mdk_rdev_t *rdev;
6527 char *desc;
6528
6529 /* just incase thread restarts... */
6530 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6531 return;
6532 if (mddev->ro) /* never try to sync a read-only array */
6533 return;
6534
6535 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6536 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6537 desc = "data-check";
6538 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6539 desc = "requested-resync";
6540 else
6541 desc = "resync";
6542 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6543 desc = "reshape";
6544 else
6545 desc = "recovery";
6546
6547 /* we overload curr_resync somewhat here.
6548 * 0 == not engaged in resync at all
6549 * 2 == checking that there is no conflict with another sync
6550 * 1 == like 2, but have yielded to allow conflicting resync to
6551 * commense
6552 * other == active in resync - this many blocks
6553 *
6554 * Before starting a resync we must have set curr_resync to
6555 * 2, and then checked that every "conflicting" array has curr_resync
6556 * less than ours. When we find one that is the same or higher
6557 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6558 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6559 * This will mean we have to start checking from the beginning again.
6560 *
6561 */
6562
6563 do {
6564 mddev->curr_resync = 2;
6565
6566 try_again:
6567 if (kthread_should_stop())
6568 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6569
6570 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6571 goto skip;
6572 for_each_mddev(mddev2, tmp) {
6573 if (mddev2 == mddev)
6574 continue;
6575 if (!mddev->parallel_resync
6576 && mddev2->curr_resync
6577 && match_mddev_units(mddev, mddev2)) {
6578 DEFINE_WAIT(wq);
6579 if (mddev < mddev2 && mddev->curr_resync == 2) {
6580 /* arbitrarily yield */
6581 mddev->curr_resync = 1;
6582 wake_up(&resync_wait);
6583 }
6584 if (mddev > mddev2 && mddev->curr_resync == 1)
6585 /* no need to wait here, we can wait the next
6586 * time 'round when curr_resync == 2
6587 */
6588 continue;
6589 /* We need to wait 'interruptible' so as not to
6590 * contribute to the load average, and not to
6591 * be caught by 'softlockup'
6592 */
6593 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6594 if (!kthread_should_stop() &&
6595 mddev2->curr_resync >= mddev->curr_resync) {
6596 printk(KERN_INFO "md: delaying %s of %s"
6597 " until %s has finished (they"
6598 " share one or more physical units)\n",
6599 desc, mdname(mddev), mdname(mddev2));
6600 mddev_put(mddev2);
6601 if (signal_pending(current))
6602 flush_signals(current);
6603 schedule();
6604 finish_wait(&resync_wait, &wq);
6605 goto try_again;
6606 }
6607 finish_wait(&resync_wait, &wq);
6608 }
6609 }
6610 } while (mddev->curr_resync < 2);
6611
6612 j = 0;
6613 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6614 /* resync follows the size requested by the personality,
6615 * which defaults to physical size, but can be virtual size
6616 */
6617 max_sectors = mddev->resync_max_sectors;
6618 mddev->resync_mismatches = 0;
6619 /* we don't use the checkpoint if there's a bitmap */
6620 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6621 j = mddev->resync_min;
6622 else if (!mddev->bitmap)
6623 j = mddev->recovery_cp;
6624
6625 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6626 max_sectors = mddev->dev_sectors;
6627 else {
6628 /* recovery follows the physical size of devices */
6629 max_sectors = mddev->dev_sectors;
6630 j = MaxSector;
6631 rcu_read_lock();
6632 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6633 if (rdev->raid_disk >= 0 &&
6634 !test_bit(Faulty, &rdev->flags) &&
6635 !test_bit(In_sync, &rdev->flags) &&
6636 rdev->recovery_offset < j)
6637 j = rdev->recovery_offset;
6638 rcu_read_unlock();
6639 }
6640
6641 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6642 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6643 " %d KB/sec/disk.\n", speed_min(mddev));
6644 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6645 "(but not more than %d KB/sec) for %s.\n",
6646 speed_max(mddev), desc);
6647
6648 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6649
6650 io_sectors = 0;
6651 for (m = 0; m < SYNC_MARKS; m++) {
6652 mark[m] = jiffies;
6653 mark_cnt[m] = io_sectors;
6654 }
6655 last_mark = 0;
6656 mddev->resync_mark = mark[last_mark];
6657 mddev->resync_mark_cnt = mark_cnt[last_mark];
6658
6659 /*
6660 * Tune reconstruction:
6661 */
6662 window = 32*(PAGE_SIZE/512);
6663 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6664 window/2,(unsigned long long) max_sectors/2);
6665
6666 atomic_set(&mddev->recovery_active, 0);
6667 last_check = 0;
6668
6669 if (j>2) {
6670 printk(KERN_INFO
6671 "md: resuming %s of %s from checkpoint.\n",
6672 desc, mdname(mddev));
6673 mddev->curr_resync = j;
6674 }
6675 mddev->curr_resync_completed = mddev->curr_resync;
6676
6677 while (j < max_sectors) {
6678 sector_t sectors;
6679
6680 skipped = 0;
6681
6682 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6683 ((mddev->curr_resync > mddev->curr_resync_completed &&
6684 (mddev->curr_resync - mddev->curr_resync_completed)
6685 > (max_sectors >> 4)) ||
6686 (j - mddev->curr_resync_completed)*2
6687 >= mddev->resync_max - mddev->curr_resync_completed
6688 )) {
6689 /* time to update curr_resync_completed */
6690 blk_unplug(mddev->queue);
6691 wait_event(mddev->recovery_wait,
6692 atomic_read(&mddev->recovery_active) == 0);
6693 mddev->curr_resync_completed =
6694 mddev->curr_resync;
6695 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6696 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6697 }
6698
6699 while (j >= mddev->resync_max && !kthread_should_stop()) {
6700 /* As this condition is controlled by user-space,
6701 * we can block indefinitely, so use '_interruptible'
6702 * to avoid triggering warnings.
6703 */
6704 flush_signals(current); /* just in case */
6705 wait_event_interruptible(mddev->recovery_wait,
6706 mddev->resync_max > j
6707 || kthread_should_stop());
6708 }
6709
6710 if (kthread_should_stop())
6711 goto interrupted;
6712
6713 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6714 currspeed < speed_min(mddev));
6715 if (sectors == 0) {
6716 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6717 goto out;
6718 }
6719
6720 if (!skipped) { /* actual IO requested */
6721 io_sectors += sectors;
6722 atomic_add(sectors, &mddev->recovery_active);
6723 }
6724
6725 j += sectors;
6726 if (j>1) mddev->curr_resync = j;
6727 mddev->curr_mark_cnt = io_sectors;
6728 if (last_check == 0)
6729 /* this is the earliers that rebuilt will be
6730 * visible in /proc/mdstat
6731 */
6732 md_new_event(mddev);
6733
6734 if (last_check + window > io_sectors || j == max_sectors)
6735 continue;
6736
6737 last_check = io_sectors;
6738
6739 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6740 break;
6741
6742 repeat:
6743 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6744 /* step marks */
6745 int next = (last_mark+1) % SYNC_MARKS;
6746
6747 mddev->resync_mark = mark[next];
6748 mddev->resync_mark_cnt = mark_cnt[next];
6749 mark[next] = jiffies;
6750 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6751 last_mark = next;
6752 }
6753
6754
6755 if (kthread_should_stop())
6756 goto interrupted;
6757
6758
6759 /*
6760 * this loop exits only if either when we are slower than
6761 * the 'hard' speed limit, or the system was IO-idle for
6762 * a jiffy.
6763 * the system might be non-idle CPU-wise, but we only care
6764 * about not overloading the IO subsystem. (things like an
6765 * e2fsck being done on the RAID array should execute fast)
6766 */
6767 blk_unplug(mddev->queue);
6768 cond_resched();
6769
6770 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6771 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6772
6773 if (currspeed > speed_min(mddev)) {
6774 if ((currspeed > speed_max(mddev)) ||
6775 !is_mddev_idle(mddev, 0)) {
6776 msleep(500);
6777 goto repeat;
6778 }
6779 }
6780 }
6781 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6782 /*
6783 * this also signals 'finished resyncing' to md_stop
6784 */
6785 out:
6786 blk_unplug(mddev->queue);
6787
6788 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6789
6790 /* tell personality that we are finished */
6791 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6792
6793 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6794 mddev->curr_resync > 2) {
6795 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6796 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6797 if (mddev->curr_resync >= mddev->recovery_cp) {
6798 printk(KERN_INFO
6799 "md: checkpointing %s of %s.\n",
6800 desc, mdname(mddev));
6801 mddev->recovery_cp = mddev->curr_resync;
6802 }
6803 } else
6804 mddev->recovery_cp = MaxSector;
6805 } else {
6806 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6807 mddev->curr_resync = MaxSector;
6808 rcu_read_lock();
6809 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6810 if (rdev->raid_disk >= 0 &&
6811 !test_bit(Faulty, &rdev->flags) &&
6812 !test_bit(In_sync, &rdev->flags) &&
6813 rdev->recovery_offset < mddev->curr_resync)
6814 rdev->recovery_offset = mddev->curr_resync;
6815 rcu_read_unlock();
6816 }
6817 }
6818 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6819
6820 skip:
6821 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6822 /* We completed so min/max setting can be forgotten if used. */
6823 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6824 mddev->resync_min = 0;
6825 mddev->resync_max = MaxSector;
6826 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6827 mddev->resync_min = mddev->curr_resync_completed;
6828 mddev->curr_resync = 0;
6829 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6830 mddev->curr_resync_completed = 0;
6831 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6832 wake_up(&resync_wait);
6833 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6834 md_wakeup_thread(mddev->thread);
6835 return;
6836
6837 interrupted:
6838 /*
6839 * got a signal, exit.
6840 */
6841 printk(KERN_INFO
6842 "md: md_do_sync() got signal ... exiting\n");
6843 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6844 goto out;
6845
6846 }
6847 EXPORT_SYMBOL_GPL(md_do_sync);
6848
6849
6850 static int remove_and_add_spares(mddev_t *mddev)
6851 {
6852 mdk_rdev_t *rdev;
6853 int spares = 0;
6854
6855 mddev->curr_resync_completed = 0;
6856
6857 list_for_each_entry(rdev, &mddev->disks, same_set)
6858 if (rdev->raid_disk >= 0 &&
6859 !test_bit(Blocked, &rdev->flags) &&
6860 (test_bit(Faulty, &rdev->flags) ||
6861 ! test_bit(In_sync, &rdev->flags)) &&
6862 atomic_read(&rdev->nr_pending)==0) {
6863 if (mddev->pers->hot_remove_disk(
6864 mddev, rdev->raid_disk)==0) {
6865 char nm[20];
6866 sprintf(nm,"rd%d", rdev->raid_disk);
6867 sysfs_remove_link(&mddev->kobj, nm);
6868 rdev->raid_disk = -1;
6869 }
6870 }
6871
6872 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
6873 list_for_each_entry(rdev, &mddev->disks, same_set) {
6874 if (rdev->raid_disk >= 0 &&
6875 !test_bit(In_sync, &rdev->flags) &&
6876 !test_bit(Blocked, &rdev->flags))
6877 spares++;
6878 if (rdev->raid_disk < 0
6879 && !test_bit(Faulty, &rdev->flags)) {
6880 rdev->recovery_offset = 0;
6881 if (mddev->pers->
6882 hot_add_disk(mddev, rdev) == 0) {
6883 char nm[20];
6884 sprintf(nm, "rd%d", rdev->raid_disk);
6885 if (sysfs_create_link(&mddev->kobj,
6886 &rdev->kobj, nm))
6887 printk(KERN_WARNING
6888 "md: cannot register "
6889 "%s for %s\n",
6890 nm, mdname(mddev));
6891 spares++;
6892 md_new_event(mddev);
6893 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6894 } else
6895 break;
6896 }
6897 }
6898 }
6899 return spares;
6900 }
6901 /*
6902 * This routine is regularly called by all per-raid-array threads to
6903 * deal with generic issues like resync and super-block update.
6904 * Raid personalities that don't have a thread (linear/raid0) do not
6905 * need this as they never do any recovery or update the superblock.
6906 *
6907 * It does not do any resync itself, but rather "forks" off other threads
6908 * to do that as needed.
6909 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6910 * "->recovery" and create a thread at ->sync_thread.
6911 * When the thread finishes it sets MD_RECOVERY_DONE
6912 * and wakeups up this thread which will reap the thread and finish up.
6913 * This thread also removes any faulty devices (with nr_pending == 0).
6914 *
6915 * The overall approach is:
6916 * 1/ if the superblock needs updating, update it.
6917 * 2/ If a recovery thread is running, don't do anything else.
6918 * 3/ If recovery has finished, clean up, possibly marking spares active.
6919 * 4/ If there are any faulty devices, remove them.
6920 * 5/ If array is degraded, try to add spares devices
6921 * 6/ If array has spares or is not in-sync, start a resync thread.
6922 */
6923 void md_check_recovery(mddev_t *mddev)
6924 {
6925 mdk_rdev_t *rdev;
6926
6927
6928 if (mddev->bitmap)
6929 bitmap_daemon_work(mddev);
6930
6931 if (mddev->ro)
6932 return;
6933
6934 if (signal_pending(current)) {
6935 if (mddev->pers->sync_request && !mddev->external) {
6936 printk(KERN_INFO "md: %s in immediate safe mode\n",
6937 mdname(mddev));
6938 mddev->safemode = 2;
6939 }
6940 flush_signals(current);
6941 }
6942
6943 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
6944 return;
6945 if ( ! (
6946 (mddev->flags && !mddev->external) ||
6947 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
6948 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
6949 (mddev->external == 0 && mddev->safemode == 1) ||
6950 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6951 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6952 ))
6953 return;
6954
6955 if (mddev_trylock(mddev)) {
6956 int spares = 0;
6957
6958 if (mddev->ro) {
6959 /* Only thing we do on a ro array is remove
6960 * failed devices.
6961 */
6962 remove_and_add_spares(mddev);
6963 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6964 goto unlock;
6965 }
6966
6967 if (!mddev->external) {
6968 int did_change = 0;
6969 spin_lock_irq(&mddev->write_lock);
6970 if (mddev->safemode &&
6971 !atomic_read(&mddev->writes_pending) &&
6972 !mddev->in_sync &&
6973 mddev->recovery_cp == MaxSector) {
6974 mddev->in_sync = 1;
6975 did_change = 1;
6976 if (mddev->persistent)
6977 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6978 }
6979 if (mddev->safemode == 1)
6980 mddev->safemode = 0;
6981 spin_unlock_irq(&mddev->write_lock);
6982 if (did_change)
6983 sysfs_notify_dirent(mddev->sysfs_state);
6984 }
6985
6986 if (mddev->flags)
6987 md_update_sb(mddev, 0);
6988
6989 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6990 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6991 /* resync/recovery still happening */
6992 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6993 goto unlock;
6994 }
6995 if (mddev->sync_thread) {
6996 /* resync has finished, collect result */
6997 md_unregister_thread(mddev->sync_thread);
6998 mddev->sync_thread = NULL;
6999 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7000 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7001 /* success...*/
7002 /* activate any spares */
7003 if (mddev->pers->spare_active(mddev))
7004 sysfs_notify(&mddev->kobj, NULL,
7005 "degraded");
7006 }
7007 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7008 mddev->pers->finish_reshape)
7009 mddev->pers->finish_reshape(mddev);
7010 md_update_sb(mddev, 1);
7011
7012 /* if array is no-longer degraded, then any saved_raid_disk
7013 * information must be scrapped
7014 */
7015 if (!mddev->degraded)
7016 list_for_each_entry(rdev, &mddev->disks, same_set)
7017 rdev->saved_raid_disk = -1;
7018
7019 mddev->recovery = 0;
7020 /* flag recovery needed just to double check */
7021 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7022 sysfs_notify_dirent(mddev->sysfs_action);
7023 md_new_event(mddev);
7024 goto unlock;
7025 }
7026 /* Set RUNNING before clearing NEEDED to avoid
7027 * any transients in the value of "sync_action".
7028 */
7029 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7030 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7031 /* Clear some bits that don't mean anything, but
7032 * might be left set
7033 */
7034 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7035 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7036
7037 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7038 goto unlock;
7039 /* no recovery is running.
7040 * remove any failed drives, then
7041 * add spares if possible.
7042 * Spare are also removed and re-added, to allow
7043 * the personality to fail the re-add.
7044 */
7045
7046 if (mddev->reshape_position != MaxSector) {
7047 if (mddev->pers->check_reshape == NULL ||
7048 mddev->pers->check_reshape(mddev) != 0)
7049 /* Cannot proceed */
7050 goto unlock;
7051 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7052 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7053 } else if ((spares = remove_and_add_spares(mddev))) {
7054 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7055 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7056 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7057 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7058 } else if (mddev->recovery_cp < MaxSector) {
7059 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7060 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7061 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7062 /* nothing to be done ... */
7063 goto unlock;
7064
7065 if (mddev->pers->sync_request) {
7066 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7067 /* We are adding a device or devices to an array
7068 * which has the bitmap stored on all devices.
7069 * So make sure all bitmap pages get written
7070 */
7071 bitmap_write_all(mddev->bitmap);
7072 }
7073 mddev->sync_thread = md_register_thread(md_do_sync,
7074 mddev,
7075 "resync");
7076 if (!mddev->sync_thread) {
7077 printk(KERN_ERR "%s: could not start resync"
7078 " thread...\n",
7079 mdname(mddev));
7080 /* leave the spares where they are, it shouldn't hurt */
7081 mddev->recovery = 0;
7082 } else
7083 md_wakeup_thread(mddev->sync_thread);
7084 sysfs_notify_dirent(mddev->sysfs_action);
7085 md_new_event(mddev);
7086 }
7087 unlock:
7088 if (!mddev->sync_thread) {
7089 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7090 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7091 &mddev->recovery))
7092 if (mddev->sysfs_action)
7093 sysfs_notify_dirent(mddev->sysfs_action);
7094 }
7095 mddev_unlock(mddev);
7096 }
7097 }
7098
7099 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
7100 {
7101 sysfs_notify_dirent(rdev->sysfs_state);
7102 wait_event_timeout(rdev->blocked_wait,
7103 !test_bit(Blocked, &rdev->flags),
7104 msecs_to_jiffies(5000));
7105 rdev_dec_pending(rdev, mddev);
7106 }
7107 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7108
7109 static int md_notify_reboot(struct notifier_block *this,
7110 unsigned long code, void *x)
7111 {
7112 struct list_head *tmp;
7113 mddev_t *mddev;
7114
7115 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
7116
7117 printk(KERN_INFO "md: stopping all md devices.\n");
7118
7119 for_each_mddev(mddev, tmp)
7120 if (mddev_trylock(mddev)) {
7121 /* Force a switch to readonly even array
7122 * appears to still be in use. Hence
7123 * the '100'.
7124 */
7125 do_md_stop(mddev, 1, 100);
7126 mddev_unlock(mddev);
7127 }
7128 /*
7129 * certain more exotic SCSI devices are known to be
7130 * volatile wrt too early system reboots. While the
7131 * right place to handle this issue is the given
7132 * driver, we do want to have a safe RAID driver ...
7133 */
7134 mdelay(1000*1);
7135 }
7136 return NOTIFY_DONE;
7137 }
7138
7139 static struct notifier_block md_notifier = {
7140 .notifier_call = md_notify_reboot,
7141 .next = NULL,
7142 .priority = INT_MAX, /* before any real devices */
7143 };
7144
7145 static void md_geninit(void)
7146 {
7147 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
7148
7149 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
7150 }
7151
7152 static int __init md_init(void)
7153 {
7154 if (register_blkdev(MD_MAJOR, "md"))
7155 return -1;
7156 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
7157 unregister_blkdev(MD_MAJOR, "md");
7158 return -1;
7159 }
7160 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
7161 md_probe, NULL, NULL);
7162 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
7163 md_probe, NULL, NULL);
7164
7165 register_reboot_notifier(&md_notifier);
7166 raid_table_header = register_sysctl_table(raid_root_table);
7167
7168 md_geninit();
7169 return 0;
7170 }
7171
7172
7173 #ifndef MODULE
7174
7175 /*
7176 * Searches all registered partitions for autorun RAID arrays
7177 * at boot time.
7178 */
7179
7180 static LIST_HEAD(all_detected_devices);
7181 struct detected_devices_node {
7182 struct list_head list;
7183 dev_t dev;
7184 };
7185
7186 void md_autodetect_dev(dev_t dev)
7187 {
7188 struct detected_devices_node *node_detected_dev;
7189
7190 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
7191 if (node_detected_dev) {
7192 node_detected_dev->dev = dev;
7193 list_add_tail(&node_detected_dev->list, &all_detected_devices);
7194 } else {
7195 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
7196 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
7197 }
7198 }
7199
7200
7201 static void autostart_arrays(int part)
7202 {
7203 mdk_rdev_t *rdev;
7204 struct detected_devices_node *node_detected_dev;
7205 dev_t dev;
7206 int i_scanned, i_passed;
7207
7208 i_scanned = 0;
7209 i_passed = 0;
7210
7211 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
7212
7213 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
7214 i_scanned++;
7215 node_detected_dev = list_entry(all_detected_devices.next,
7216 struct detected_devices_node, list);
7217 list_del(&node_detected_dev->list);
7218 dev = node_detected_dev->dev;
7219 kfree(node_detected_dev);
7220 rdev = md_import_device(dev,0, 90);
7221 if (IS_ERR(rdev))
7222 continue;
7223
7224 if (test_bit(Faulty, &rdev->flags)) {
7225 MD_BUG();
7226 continue;
7227 }
7228 set_bit(AutoDetected, &rdev->flags);
7229 list_add(&rdev->same_set, &pending_raid_disks);
7230 i_passed++;
7231 }
7232
7233 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
7234 i_scanned, i_passed);
7235
7236 autorun_devices(part);
7237 }
7238
7239 #endif /* !MODULE */
7240
7241 static __exit void md_exit(void)
7242 {
7243 mddev_t *mddev;
7244 struct list_head *tmp;
7245
7246 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
7247 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
7248
7249 unregister_blkdev(MD_MAJOR,"md");
7250 unregister_blkdev(mdp_major, "mdp");
7251 unregister_reboot_notifier(&md_notifier);
7252 unregister_sysctl_table(raid_table_header);
7253 remove_proc_entry("mdstat", NULL);
7254 for_each_mddev(mddev, tmp) {
7255 export_array(mddev);
7256 mddev->hold_active = 0;
7257 }
7258 }
7259
7260 subsys_initcall(md_init);
7261 module_exit(md_exit)
7262
7263 static int get_ro(char *buffer, struct kernel_param *kp)
7264 {
7265 return sprintf(buffer, "%d", start_readonly);
7266 }
7267 static int set_ro(const char *val, struct kernel_param *kp)
7268 {
7269 char *e;
7270 int num = simple_strtoul(val, &e, 10);
7271 if (*val && (*e == '\0' || *e == '\n')) {
7272 start_readonly = num;
7273 return 0;
7274 }
7275 return -EINVAL;
7276 }
7277
7278 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
7279 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
7280
7281 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
7282
7283 EXPORT_SYMBOL(register_md_personality);
7284 EXPORT_SYMBOL(unregister_md_personality);
7285 EXPORT_SYMBOL(md_error);
7286 EXPORT_SYMBOL(md_done_sync);
7287 EXPORT_SYMBOL(md_write_start);
7288 EXPORT_SYMBOL(md_write_end);
7289 EXPORT_SYMBOL(md_register_thread);
7290 EXPORT_SYMBOL(md_unregister_thread);
7291 EXPORT_SYMBOL(md_wakeup_thread);
7292 EXPORT_SYMBOL(md_check_recovery);
7293 MODULE_LICENSE("GPL");
7294 MODULE_DESCRIPTION("MD RAID framework");
7295 MODULE_ALIAS("md");
7296 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Ubuntu Jammy Linux kernel mirror