]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - drivers/md/md.c
projects / mirror_ubuntu-jammy-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
md: Add support for Raid5->Raid0 and Raid10->Raid0 takeover
[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 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2389 md_wakeup_thread(rdev->mddev->thread);
2390 } else if (rdev->mddev->pers) {
2391 mdk_rdev_t *rdev2;
2392 /* Activating a spare .. or possibly reactivating
2393 * if we ever get bitmaps working here.
2394 */
2395
2396 if (rdev->raid_disk != -1)
2397 return -EBUSY;
2398
2399 if (rdev->mddev->pers->hot_add_disk == NULL)
2400 return -EINVAL;
2401
2402 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2403 if (rdev2->raid_disk == slot)
2404 return -EEXIST;
2405
2406 rdev->raid_disk = slot;
2407 if (test_bit(In_sync, &rdev->flags))
2408 rdev->saved_raid_disk = slot;
2409 else
2410 rdev->saved_raid_disk = -1;
2411 err = rdev->mddev->pers->
2412 hot_add_disk(rdev->mddev, rdev);
2413 if (err) {
2414 rdev->raid_disk = -1;
2415 return err;
2416 } else
2417 sysfs_notify_dirent(rdev->sysfs_state);
2418 sprintf(nm, "rd%d", rdev->raid_disk);
2419 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2420 printk(KERN_WARNING
2421 "md: cannot register "
2422 "%s for %s\n",
2423 nm, mdname(rdev->mddev));
2424
2425 /* don't wakeup anyone, leave that to userspace. */
2426 } else {
2427 if (slot >= rdev->mddev->raid_disks)
2428 return -ENOSPC;
2429 rdev->raid_disk = slot;
2430 /* assume it is working */
2431 clear_bit(Faulty, &rdev->flags);
2432 clear_bit(WriteMostly, &rdev->flags);
2433 set_bit(In_sync, &rdev->flags);
2434 sysfs_notify_dirent(rdev->sysfs_state);
2435 }
2436 return len;
2437 }
2438
2439
2440 static struct rdev_sysfs_entry rdev_slot =
2441 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2442
2443 static ssize_t
2444 offset_show(mdk_rdev_t *rdev, char *page)
2445 {
2446 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2447 }
2448
2449 static ssize_t
2450 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2451 {
2452 char *e;
2453 unsigned long long offset = simple_strtoull(buf, &e, 10);
2454 if (e==buf || (*e && *e != '\n'))
2455 return -EINVAL;
2456 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2457 return -EBUSY;
2458 if (rdev->sectors && rdev->mddev->external)
2459 /* Must set offset before size, so overlap checks
2460 * can be sane */
2461 return -EBUSY;
2462 rdev->data_offset = offset;
2463 return len;
2464 }
2465
2466 static struct rdev_sysfs_entry rdev_offset =
2467 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2468
2469 static ssize_t
2470 rdev_size_show(mdk_rdev_t *rdev, char *page)
2471 {
2472 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2473 }
2474
2475 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2476 {
2477 /* check if two start/length pairs overlap */
2478 if (s1+l1 <= s2)
2479 return 0;
2480 if (s2+l2 <= s1)
2481 return 0;
2482 return 1;
2483 }
2484
2485 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2486 {
2487 unsigned long long blocks;
2488 sector_t new;
2489
2490 if (strict_strtoull(buf, 10, &blocks) < 0)
2491 return -EINVAL;
2492
2493 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2494 return -EINVAL; /* sector conversion overflow */
2495
2496 new = blocks * 2;
2497 if (new != blocks * 2)
2498 return -EINVAL; /* unsigned long long to sector_t overflow */
2499
2500 *sectors = new;
2501 return 0;
2502 }
2503
2504 static ssize_t
2505 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2506 {
2507 mddev_t *my_mddev = rdev->mddev;
2508 sector_t oldsectors = rdev->sectors;
2509 sector_t sectors;
2510
2511 if (strict_blocks_to_sectors(buf, &sectors) < 0)
2512 return -EINVAL;
2513 if (my_mddev->pers && rdev->raid_disk >= 0) {
2514 if (my_mddev->persistent) {
2515 sectors = super_types[my_mddev->major_version].
2516 rdev_size_change(rdev, sectors);
2517 if (!sectors)
2518 return -EBUSY;
2519 } else if (!sectors)
2520 sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2521 rdev->data_offset;
2522 }
2523 if (sectors < my_mddev->dev_sectors)
2524 return -EINVAL; /* component must fit device */
2525
2526 rdev->sectors = sectors;
2527 if (sectors > oldsectors && my_mddev->external) {
2528 /* need to check that all other rdevs with the same ->bdev
2529 * do not overlap. We need to unlock the mddev to avoid
2530 * a deadlock. We have already changed rdev->sectors, and if
2531 * we have to change it back, we will have the lock again.
2532 */
2533 mddev_t *mddev;
2534 int overlap = 0;
2535 struct list_head *tmp;
2536
2537 mddev_unlock(my_mddev);
2538 for_each_mddev(mddev, tmp) {
2539 mdk_rdev_t *rdev2;
2540
2541 mddev_lock(mddev);
2542 list_for_each_entry(rdev2, &mddev->disks, same_set)
2543 if (test_bit(AllReserved, &rdev2->flags) ||
2544 (rdev->bdev == rdev2->bdev &&
2545 rdev != rdev2 &&
2546 overlaps(rdev->data_offset, rdev->sectors,
2547 rdev2->data_offset,
2548 rdev2->sectors))) {
2549 overlap = 1;
2550 break;
2551 }
2552 mddev_unlock(mddev);
2553 if (overlap) {
2554 mddev_put(mddev);
2555 break;
2556 }
2557 }
2558 mddev_lock(my_mddev);
2559 if (overlap) {
2560 /* Someone else could have slipped in a size
2561 * change here, but doing so is just silly.
2562 * We put oldsectors back because we *know* it is
2563 * safe, and trust userspace not to race with
2564 * itself
2565 */
2566 rdev->sectors = oldsectors;
2567 return -EBUSY;
2568 }
2569 }
2570 return len;
2571 }
2572
2573 static struct rdev_sysfs_entry rdev_size =
2574 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2575
2576
2577 static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page)
2578 {
2579 unsigned long long recovery_start = rdev->recovery_offset;
2580
2581 if (test_bit(In_sync, &rdev->flags) ||
2582 recovery_start == MaxSector)
2583 return sprintf(page, "none\n");
2584
2585 return sprintf(page, "%llu\n", recovery_start);
2586 }
2587
2588 static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2589 {
2590 unsigned long long recovery_start;
2591
2592 if (cmd_match(buf, "none"))
2593 recovery_start = MaxSector;
2594 else if (strict_strtoull(buf, 10, &recovery_start))
2595 return -EINVAL;
2596
2597 if (rdev->mddev->pers &&
2598 rdev->raid_disk >= 0)
2599 return -EBUSY;
2600
2601 rdev->recovery_offset = recovery_start;
2602 if (recovery_start == MaxSector)
2603 set_bit(In_sync, &rdev->flags);
2604 else
2605 clear_bit(In_sync, &rdev->flags);
2606 return len;
2607 }
2608
2609 static struct rdev_sysfs_entry rdev_recovery_start =
2610 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2611
2612 static struct attribute *rdev_default_attrs[] = {
2613 &rdev_state.attr,
2614 &rdev_errors.attr,
2615 &rdev_slot.attr,
2616 &rdev_offset.attr,
2617 &rdev_size.attr,
2618 &rdev_recovery_start.attr,
2619 NULL,
2620 };
2621 static ssize_t
2622 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2623 {
2624 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2625 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2626 mddev_t *mddev = rdev->mddev;
2627 ssize_t rv;
2628
2629 if (!entry->show)
2630 return -EIO;
2631
2632 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2633 if (!rv) {
2634 if (rdev->mddev == NULL)
2635 rv = -EBUSY;
2636 else
2637 rv = entry->show(rdev, page);
2638 mddev_unlock(mddev);
2639 }
2640 return rv;
2641 }
2642
2643 static ssize_t
2644 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2645 const char *page, size_t length)
2646 {
2647 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2648 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2649 ssize_t rv;
2650 mddev_t *mddev = rdev->mddev;
2651
2652 if (!entry->store)
2653 return -EIO;
2654 if (!capable(CAP_SYS_ADMIN))
2655 return -EACCES;
2656 rv = mddev ? mddev_lock(mddev): -EBUSY;
2657 if (!rv) {
2658 if (rdev->mddev == NULL)
2659 rv = -EBUSY;
2660 else
2661 rv = entry->store(rdev, page, length);
2662 mddev_unlock(mddev);
2663 }
2664 return rv;
2665 }
2666
2667 static void rdev_free(struct kobject *ko)
2668 {
2669 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2670 kfree(rdev);
2671 }
2672 static struct sysfs_ops rdev_sysfs_ops = {
2673 .show = rdev_attr_show,
2674 .store = rdev_attr_store,
2675 };
2676 static struct kobj_type rdev_ktype = {
2677 .release = rdev_free,
2678 .sysfs_ops = &rdev_sysfs_ops,
2679 .default_attrs = rdev_default_attrs,
2680 };
2681
2682 /*
2683 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2684 *
2685 * mark the device faulty if:
2686 *
2687 * - the device is nonexistent (zero size)
2688 * - the device has no valid superblock
2689 *
2690 * a faulty rdev _never_ has rdev->sb set.
2691 */
2692 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2693 {
2694 char b[BDEVNAME_SIZE];
2695 int err;
2696 mdk_rdev_t *rdev;
2697 sector_t size;
2698
2699 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2700 if (!rdev) {
2701 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2702 return ERR_PTR(-ENOMEM);
2703 }
2704
2705 if ((err = alloc_disk_sb(rdev)))
2706 goto abort_free;
2707
2708 err = lock_rdev(rdev, newdev, super_format == -2);
2709 if (err)
2710 goto abort_free;
2711
2712 kobject_init(&rdev->kobj, &rdev_ktype);
2713
2714 rdev->desc_nr = -1;
2715 rdev->saved_raid_disk = -1;
2716 rdev->raid_disk = -1;
2717 rdev->flags = 0;
2718 rdev->data_offset = 0;
2719 rdev->sb_events = 0;
2720 rdev->last_read_error.tv_sec = 0;
2721 rdev->last_read_error.tv_nsec = 0;
2722 atomic_set(&rdev->nr_pending, 0);
2723 atomic_set(&rdev->read_errors, 0);
2724 atomic_set(&rdev->corrected_errors, 0);
2725
2726 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2727 if (!size) {
2728 printk(KERN_WARNING
2729 "md: %s has zero or unknown size, marking faulty!\n",
2730 bdevname(rdev->bdev,b));
2731 err = -EINVAL;
2732 goto abort_free;
2733 }
2734
2735 if (super_format >= 0) {
2736 err = super_types[super_format].
2737 load_super(rdev, NULL, super_minor);
2738 if (err == -EINVAL) {
2739 printk(KERN_WARNING
2740 "md: %s does not have a valid v%d.%d "
2741 "superblock, not importing!\n",
2742 bdevname(rdev->bdev,b),
2743 super_format, super_minor);
2744 goto abort_free;
2745 }
2746 if (err < 0) {
2747 printk(KERN_WARNING
2748 "md: could not read %s's sb, not importing!\n",
2749 bdevname(rdev->bdev,b));
2750 goto abort_free;
2751 }
2752 }
2753
2754 INIT_LIST_HEAD(&rdev->same_set);
2755 init_waitqueue_head(&rdev->blocked_wait);
2756
2757 return rdev;
2758
2759 abort_free:
2760 if (rdev->sb_page) {
2761 if (rdev->bdev)
2762 unlock_rdev(rdev);
2763 free_disk_sb(rdev);
2764 }
2765 kfree(rdev);
2766 return ERR_PTR(err);
2767 }
2768
2769 /*
2770 * Check a full RAID array for plausibility
2771 */
2772
2773
2774 static void analyze_sbs(mddev_t * mddev)
2775 {
2776 int i;
2777 mdk_rdev_t *rdev, *freshest, *tmp;
2778 char b[BDEVNAME_SIZE];
2779
2780 freshest = NULL;
2781 rdev_for_each(rdev, tmp, mddev)
2782 switch (super_types[mddev->major_version].
2783 load_super(rdev, freshest, mddev->minor_version)) {
2784 case 1:
2785 freshest = rdev;
2786 break;
2787 case 0:
2788 break;
2789 default:
2790 printk( KERN_ERR \
2791 "md: fatal superblock inconsistency in %s"
2792 " -- removing from array\n",
2793 bdevname(rdev->bdev,b));
2794 kick_rdev_from_array(rdev);
2795 }
2796
2797
2798 super_types[mddev->major_version].
2799 validate_super(mddev, freshest);
2800
2801 i = 0;
2802 rdev_for_each(rdev, tmp, mddev) {
2803 if (rdev->desc_nr >= mddev->max_disks ||
2804 i > mddev->max_disks) {
2805 printk(KERN_WARNING
2806 "md: %s: %s: only %d devices permitted\n",
2807 mdname(mddev), bdevname(rdev->bdev, b),
2808 mddev->max_disks);
2809 kick_rdev_from_array(rdev);
2810 continue;
2811 }
2812 if (rdev != freshest)
2813 if (super_types[mddev->major_version].
2814 validate_super(mddev, rdev)) {
2815 printk(KERN_WARNING "md: kicking non-fresh %s"
2816 " from array!\n",
2817 bdevname(rdev->bdev,b));
2818 kick_rdev_from_array(rdev);
2819 continue;
2820 }
2821 if (mddev->level == LEVEL_MULTIPATH) {
2822 rdev->desc_nr = i++;
2823 rdev->raid_disk = rdev->desc_nr;
2824 set_bit(In_sync, &rdev->flags);
2825 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
2826 rdev->raid_disk = -1;
2827 clear_bit(In_sync, &rdev->flags);
2828 }
2829 }
2830 }
2831
2832 /* Read a fixed-point number.
2833 * Numbers in sysfs attributes should be in "standard" units where
2834 * possible, so time should be in seconds.
2835 * However we internally use a a much smaller unit such as
2836 * milliseconds or jiffies.
2837 * This function takes a decimal number with a possible fractional
2838 * component, and produces an integer which is the result of
2839 * multiplying that number by 10^'scale'.
2840 * all without any floating-point arithmetic.
2841 */
2842 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
2843 {
2844 unsigned long result = 0;
2845 long decimals = -1;
2846 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
2847 if (*cp == '.')
2848 decimals = 0;
2849 else if (decimals < scale) {
2850 unsigned int value;
2851 value = *cp - '0';
2852 result = result * 10 + value;
2853 if (decimals >= 0)
2854 decimals++;
2855 }
2856 cp++;
2857 }
2858 if (*cp == '\n')
2859 cp++;
2860 if (*cp)
2861 return -EINVAL;
2862 if (decimals < 0)
2863 decimals = 0;
2864 while (decimals < scale) {
2865 result *= 10;
2866 decimals ++;
2867 }
2868 *res = result;
2869 return 0;
2870 }
2871
2872
2873 static void md_safemode_timeout(unsigned long data);
2874
2875 static ssize_t
2876 safe_delay_show(mddev_t *mddev, char *page)
2877 {
2878 int msec = (mddev->safemode_delay*1000)/HZ;
2879 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2880 }
2881 static ssize_t
2882 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2883 {
2884 unsigned long msec;
2885
2886 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
2887 return -EINVAL;
2888 if (msec == 0)
2889 mddev->safemode_delay = 0;
2890 else {
2891 unsigned long old_delay = mddev->safemode_delay;
2892 mddev->safemode_delay = (msec*HZ)/1000;
2893 if (mddev->safemode_delay == 0)
2894 mddev->safemode_delay = 1;
2895 if (mddev->safemode_delay < old_delay)
2896 md_safemode_timeout((unsigned long)mddev);
2897 }
2898 return len;
2899 }
2900 static struct md_sysfs_entry md_safe_delay =
2901 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2902
2903 static ssize_t
2904 level_show(mddev_t *mddev, char *page)
2905 {
2906 struct mdk_personality *p = mddev->pers;
2907 if (p)
2908 return sprintf(page, "%s\n", p->name);
2909 else if (mddev->clevel[0])
2910 return sprintf(page, "%s\n", mddev->clevel);
2911 else if (mddev->level != LEVEL_NONE)
2912 return sprintf(page, "%d\n", mddev->level);
2913 else
2914 return 0;
2915 }
2916
2917 static ssize_t
2918 level_store(mddev_t *mddev, const char *buf, size_t len)
2919 {
2920 char level[16];
2921 ssize_t rv = len;
2922 struct mdk_personality *pers;
2923 void *priv;
2924 mdk_rdev_t *rdev;
2925
2926 if (mddev->pers == NULL) {
2927 if (len == 0)
2928 return 0;
2929 if (len >= sizeof(mddev->clevel))
2930 return -ENOSPC;
2931 strncpy(mddev->clevel, buf, len);
2932 if (mddev->clevel[len-1] == '\n')
2933 len--;
2934 mddev->clevel[len] = 0;
2935 mddev->level = LEVEL_NONE;
2936 return rv;
2937 }
2938
2939 /* request to change the personality. Need to ensure:
2940 * - array is not engaged in resync/recovery/reshape
2941 * - old personality can be suspended
2942 * - new personality will access other array.
2943 */
2944
2945 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
2946 return -EBUSY;
2947
2948 if (!mddev->pers->quiesce) {
2949 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
2950 mdname(mddev), mddev->pers->name);
2951 return -EINVAL;
2952 }
2953
2954 /* Now find the new personality */
2955 if (len == 0 || len >= sizeof(level))
2956 return -EINVAL;
2957 strncpy(level, buf, len);
2958 if (level[len-1] == '\n')
2959 len--;
2960 level[len] = 0;
2961
2962 request_module("md-%s", level);
2963 spin_lock(&pers_lock);
2964 pers = find_pers(LEVEL_NONE, level);
2965 if (!pers || !try_module_get(pers->owner)) {
2966 spin_unlock(&pers_lock);
2967 printk(KERN_WARNING "md: personality %s not loaded\n", level);
2968 return -EINVAL;
2969 }
2970 spin_unlock(&pers_lock);
2971
2972 if (pers == mddev->pers) {
2973 /* Nothing to do! */
2974 module_put(pers->owner);
2975 return rv;
2976 }
2977 if (!pers->takeover) {
2978 module_put(pers->owner);
2979 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
2980 mdname(mddev), level);
2981 return -EINVAL;
2982 }
2983
2984 /* ->takeover must set new_* and/or delta_disks
2985 * if it succeeds, and may set them when it fails.
2986 */
2987 priv = pers->takeover(mddev);
2988 if (IS_ERR(priv)) {
2989 mddev->new_level = mddev->level;
2990 mddev->new_layout = mddev->layout;
2991 mddev->new_chunk_sectors = mddev->chunk_sectors;
2992 mddev->raid_disks -= mddev->delta_disks;
2993 mddev->delta_disks = 0;
2994 module_put(pers->owner);
2995 printk(KERN_WARNING "md: %s: %s would not accept array\n",
2996 mdname(mddev), level);
2997 return PTR_ERR(priv);
2998 }
2999
3000 /* Looks like we have a winner */
3001 mddev_suspend(mddev);
3002 mddev->pers->stop(mddev);
3003
3004 if (mddev->pers->sync_request == NULL &&
3005 pers->sync_request != NULL) {
3006 /* need to add the md_redundancy_group */
3007 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3008 printk(KERN_WARNING
3009 "md: cannot register extra attributes for %s\n",
3010 mdname(mddev));
3011 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3012 }
3013 if (mddev->pers->sync_request != NULL &&
3014 pers->sync_request == NULL) {
3015 /* need to remove the md_redundancy_group */
3016 if (mddev->to_remove == NULL)
3017 mddev->to_remove = &md_redundancy_group;
3018 }
3019
3020 if (mddev->pers->sync_request == NULL &&
3021 mddev->external) {
3022 /* We are converting from a no-redundancy array
3023 * to a redundancy array and metadata is managed
3024 * externally so we need to be sure that writes
3025 * won't block due to a need to transition
3026 * clean->dirty
3027 * until external management is started.
3028 */
3029 mddev->in_sync = 0;
3030 mddev->safemode_delay = 0;
3031 mddev->safemode = 0;
3032 }
3033
3034 module_put(mddev->pers->owner);
3035 /* Invalidate devices that are now superfluous */
3036 list_for_each_entry(rdev, &mddev->disks, same_set)
3037 if (rdev->raid_disk >= mddev->raid_disks) {
3038 rdev->raid_disk = -1;
3039 clear_bit(In_sync, &rdev->flags);
3040 }
3041 mddev->pers = pers;
3042 mddev->private = priv;
3043 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3044 mddev->level = mddev->new_level;
3045 mddev->layout = mddev->new_layout;
3046 mddev->chunk_sectors = mddev->new_chunk_sectors;
3047 mddev->delta_disks = 0;
3048 if (mddev->pers->sync_request == NULL) {
3049 /* this is now an array without redundancy, so
3050 * it must always be in_sync
3051 */
3052 mddev->in_sync = 1;
3053 del_timer_sync(&mddev->safemode_timer);
3054 }
3055 pers->run(mddev);
3056 mddev_resume(mddev);
3057 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3058 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3059 md_wakeup_thread(mddev->thread);
3060 return rv;
3061 }
3062
3063 static struct md_sysfs_entry md_level =
3064 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3065
3066
3067 static ssize_t
3068 layout_show(mddev_t *mddev, char *page)
3069 {
3070 /* just a number, not meaningful for all levels */
3071 if (mddev->reshape_position != MaxSector &&
3072 mddev->layout != mddev->new_layout)
3073 return sprintf(page, "%d (%d)\n",
3074 mddev->new_layout, mddev->layout);
3075 return sprintf(page, "%d\n", mddev->layout);
3076 }
3077
3078 static ssize_t
3079 layout_store(mddev_t *mddev, const char *buf, size_t len)
3080 {
3081 char *e;
3082 unsigned long n = simple_strtoul(buf, &e, 10);
3083
3084 if (!*buf || (*e && *e != '\n'))
3085 return -EINVAL;
3086
3087 if (mddev->pers) {
3088 int err;
3089 if (mddev->pers->check_reshape == NULL)
3090 return -EBUSY;
3091 mddev->new_layout = n;
3092 err = mddev->pers->check_reshape(mddev);
3093 if (err) {
3094 mddev->new_layout = mddev->layout;
3095 return err;
3096 }
3097 } else {
3098 mddev->new_layout = n;
3099 if (mddev->reshape_position == MaxSector)
3100 mddev->layout = n;
3101 }
3102 return len;
3103 }
3104 static struct md_sysfs_entry md_layout =
3105 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3106
3107
3108 static ssize_t
3109 raid_disks_show(mddev_t *mddev, char *page)
3110 {
3111 if (mddev->raid_disks == 0)
3112 return 0;
3113 if (mddev->reshape_position != MaxSector &&
3114 mddev->delta_disks != 0)
3115 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3116 mddev->raid_disks - mddev->delta_disks);
3117 return sprintf(page, "%d\n", mddev->raid_disks);
3118 }
3119
3120 static int update_raid_disks(mddev_t *mddev, int raid_disks);
3121
3122 static ssize_t
3123 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
3124 {
3125 char *e;
3126 int rv = 0;
3127 unsigned long n = simple_strtoul(buf, &e, 10);
3128
3129 if (!*buf || (*e && *e != '\n'))
3130 return -EINVAL;
3131
3132 if (mddev->pers)
3133 rv = update_raid_disks(mddev, n);
3134 else if (mddev->reshape_position != MaxSector) {
3135 int olddisks = mddev->raid_disks - mddev->delta_disks;
3136 mddev->delta_disks = n - olddisks;
3137 mddev->raid_disks = n;
3138 } else
3139 mddev->raid_disks = n;
3140 return rv ? rv : len;
3141 }
3142 static struct md_sysfs_entry md_raid_disks =
3143 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3144
3145 static ssize_t
3146 chunk_size_show(mddev_t *mddev, char *page)
3147 {
3148 if (mddev->reshape_position != MaxSector &&
3149 mddev->chunk_sectors != mddev->new_chunk_sectors)
3150 return sprintf(page, "%d (%d)\n",
3151 mddev->new_chunk_sectors << 9,
3152 mddev->chunk_sectors << 9);
3153 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3154 }
3155
3156 static ssize_t
3157 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
3158 {
3159 char *e;
3160 unsigned long n = simple_strtoul(buf, &e, 10);
3161
3162 if (!*buf || (*e && *e != '\n'))
3163 return -EINVAL;
3164
3165 if (mddev->pers) {
3166 int err;
3167 if (mddev->pers->check_reshape == NULL)
3168 return -EBUSY;
3169 mddev->new_chunk_sectors = n >> 9;
3170 err = mddev->pers->check_reshape(mddev);
3171 if (err) {
3172 mddev->new_chunk_sectors = mddev->chunk_sectors;
3173 return err;
3174 }
3175 } else {
3176 mddev->new_chunk_sectors = n >> 9;
3177 if (mddev->reshape_position == MaxSector)
3178 mddev->chunk_sectors = n >> 9;
3179 }
3180 return len;
3181 }
3182 static struct md_sysfs_entry md_chunk_size =
3183 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3184
3185 static ssize_t
3186 resync_start_show(mddev_t *mddev, char *page)
3187 {
3188 if (mddev->recovery_cp == MaxSector)
3189 return sprintf(page, "none\n");
3190 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3191 }
3192
3193 static ssize_t
3194 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
3195 {
3196 char *e;
3197 unsigned long long n = simple_strtoull(buf, &e, 10);
3198
3199 if (mddev->pers)
3200 return -EBUSY;
3201 if (cmd_match(buf, "none"))
3202 n = MaxSector;
3203 else if (!*buf || (*e && *e != '\n'))
3204 return -EINVAL;
3205
3206 mddev->recovery_cp = n;
3207 return len;
3208 }
3209 static struct md_sysfs_entry md_resync_start =
3210 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3211
3212 /*
3213 * The array state can be:
3214 *
3215 * clear
3216 * No devices, no size, no level
3217 * Equivalent to STOP_ARRAY ioctl
3218 * inactive
3219 * May have some settings, but array is not active
3220 * all IO results in error
3221 * When written, doesn't tear down array, but just stops it
3222 * suspended (not supported yet)
3223 * All IO requests will block. The array can be reconfigured.
3224 * Writing this, if accepted, will block until array is quiescent
3225 * readonly
3226 * no resync can happen. no superblocks get written.
3227 * write requests fail
3228 * read-auto
3229 * like readonly, but behaves like 'clean' on a write request.
3230 *
3231 * clean - no pending writes, but otherwise active.
3232 * When written to inactive array, starts without resync
3233 * If a write request arrives then
3234 * if metadata is known, mark 'dirty' and switch to 'active'.
3235 * if not known, block and switch to write-pending
3236 * If written to an active array that has pending writes, then fails.
3237 * active
3238 * fully active: IO and resync can be happening.
3239 * When written to inactive array, starts with resync
3240 *
3241 * write-pending
3242 * clean, but writes are blocked waiting for 'active' to be written.
3243 *
3244 * active-idle
3245 * like active, but no writes have been seen for a while (100msec).
3246 *
3247 */
3248 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3249 write_pending, active_idle, bad_word};
3250 static char *array_states[] = {
3251 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3252 "write-pending", "active-idle", NULL };
3253
3254 static int match_word(const char *word, char **list)
3255 {
3256 int n;
3257 for (n=0; list[n]; n++)
3258 if (cmd_match(word, list[n]))
3259 break;
3260 return n;
3261 }
3262
3263 static ssize_t
3264 array_state_show(mddev_t *mddev, char *page)
3265 {
3266 enum array_state st = inactive;
3267
3268 if (mddev->pers)
3269 switch(mddev->ro) {
3270 case 1:
3271 st = readonly;
3272 break;
3273 case 2:
3274 st = read_auto;
3275 break;
3276 case 0:
3277 if (mddev->in_sync)
3278 st = clean;
3279 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
3280 st = write_pending;
3281 else if (mddev->safemode)
3282 st = active_idle;
3283 else
3284 st = active;
3285 }
3286 else {
3287 if (list_empty(&mddev->disks) &&
3288 mddev->raid_disks == 0 &&
3289 mddev->dev_sectors == 0)
3290 st = clear;
3291 else
3292 st = inactive;
3293 }
3294 return sprintf(page, "%s\n", array_states[st]);
3295 }
3296
3297 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3298 static int do_md_run(mddev_t * mddev);
3299 static int restart_array(mddev_t *mddev);
3300
3301 static ssize_t
3302 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3303 {
3304 int err = -EINVAL;
3305 enum array_state st = match_word(buf, array_states);
3306 switch(st) {
3307 case bad_word:
3308 break;
3309 case clear:
3310 /* stopping an active array */
3311 if (atomic_read(&mddev->openers) > 0)
3312 return -EBUSY;
3313 err = do_md_stop(mddev, 0, 0);
3314 break;
3315 case inactive:
3316 /* stopping an active array */
3317 if (mddev->pers) {
3318 if (atomic_read(&mddev->openers) > 0)
3319 return -EBUSY;
3320 err = do_md_stop(mddev, 2, 0);
3321 } else
3322 err = 0; /* already inactive */
3323 break;
3324 case suspended:
3325 break; /* not supported yet */
3326 case readonly:
3327 if (mddev->pers)
3328 err = do_md_stop(mddev, 1, 0);
3329 else {
3330 mddev->ro = 1;
3331 set_disk_ro(mddev->gendisk, 1);
3332 err = do_md_run(mddev);
3333 }
3334 break;
3335 case read_auto:
3336 if (mddev->pers) {
3337 if (mddev->ro == 0)
3338 err = do_md_stop(mddev, 1, 0);
3339 else if (mddev->ro == 1)
3340 err = restart_array(mddev);
3341 if (err == 0) {
3342 mddev->ro = 2;
3343 set_disk_ro(mddev->gendisk, 0);
3344 }
3345 } else {
3346 mddev->ro = 2;
3347 err = do_md_run(mddev);
3348 }
3349 break;
3350 case clean:
3351 if (mddev->pers) {
3352 restart_array(mddev);
3353 spin_lock_irq(&mddev->write_lock);
3354 if (atomic_read(&mddev->writes_pending) == 0) {
3355 if (mddev->in_sync == 0) {
3356 mddev->in_sync = 1;
3357 if (mddev->safemode == 1)
3358 mddev->safemode = 0;
3359 if (mddev->persistent)
3360 set_bit(MD_CHANGE_CLEAN,
3361 &mddev->flags);
3362 }
3363 err = 0;
3364 } else
3365 err = -EBUSY;
3366 spin_unlock_irq(&mddev->write_lock);
3367 } else
3368 err = -EINVAL;
3369 break;
3370 case active:
3371 if (mddev->pers) {
3372 restart_array(mddev);
3373 if (mddev->external)
3374 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3375 wake_up(&mddev->sb_wait);
3376 err = 0;
3377 } else {
3378 mddev->ro = 0;
3379 set_disk_ro(mddev->gendisk, 0);
3380 err = do_md_run(mddev);
3381 }
3382 break;
3383 case write_pending:
3384 case active_idle:
3385 /* these cannot be set */
3386 break;
3387 }
3388 if (err)
3389 return err;
3390 else {
3391 sysfs_notify_dirent(mddev->sysfs_state);
3392 return len;
3393 }
3394 }
3395 static struct md_sysfs_entry md_array_state =
3396 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3397
3398 static ssize_t
3399 max_corrected_read_errors_show(mddev_t *mddev, char *page) {
3400 return sprintf(page, "%d\n",
3401 atomic_read(&mddev->max_corr_read_errors));
3402 }
3403
3404 static ssize_t
3405 max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len)
3406 {
3407 char *e;
3408 unsigned long n = simple_strtoul(buf, &e, 10);
3409
3410 if (*buf && (*e == 0 || *e == '\n')) {
3411 atomic_set(&mddev->max_corr_read_errors, n);
3412 return len;
3413 }
3414 return -EINVAL;
3415 }
3416
3417 static struct md_sysfs_entry max_corr_read_errors =
3418 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3419 max_corrected_read_errors_store);
3420
3421 static ssize_t
3422 null_show(mddev_t *mddev, char *page)
3423 {
3424 return -EINVAL;
3425 }
3426
3427 static ssize_t
3428 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3429 {
3430 /* buf must be %d:%d\n? giving major and minor numbers */
3431 /* The new device is added to the array.
3432 * If the array has a persistent superblock, we read the
3433 * superblock to initialise info and check validity.
3434 * Otherwise, only checking done is that in bind_rdev_to_array,
3435 * which mainly checks size.
3436 */
3437 char *e;
3438 int major = simple_strtoul(buf, &e, 10);
3439 int minor;
3440 dev_t dev;
3441 mdk_rdev_t *rdev;
3442 int err;
3443
3444 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3445 return -EINVAL;
3446 minor = simple_strtoul(e+1, &e, 10);
3447 if (*e && *e != '\n')
3448 return -EINVAL;
3449 dev = MKDEV(major, minor);
3450 if (major != MAJOR(dev) ||
3451 minor != MINOR(dev))
3452 return -EOVERFLOW;
3453
3454
3455 if (mddev->persistent) {
3456 rdev = md_import_device(dev, mddev->major_version,
3457 mddev->minor_version);
3458 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3459 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3460 mdk_rdev_t, same_set);
3461 err = super_types[mddev->major_version]
3462 .load_super(rdev, rdev0, mddev->minor_version);
3463 if (err < 0)
3464 goto out;
3465 }
3466 } else if (mddev->external)
3467 rdev = md_import_device(dev, -2, -1);
3468 else
3469 rdev = md_import_device(dev, -1, -1);
3470
3471 if (IS_ERR(rdev))
3472 return PTR_ERR(rdev);
3473 err = bind_rdev_to_array(rdev, mddev);
3474 out:
3475 if (err)
3476 export_rdev(rdev);
3477 return err ? err : len;
3478 }
3479
3480 static struct md_sysfs_entry md_new_device =
3481 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3482
3483 static ssize_t
3484 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3485 {
3486 char *end;
3487 unsigned long chunk, end_chunk;
3488
3489 if (!mddev->bitmap)
3490 goto out;
3491 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3492 while (*buf) {
3493 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3494 if (buf == end) break;
3495 if (*end == '-') { /* range */
3496 buf = end + 1;
3497 end_chunk = simple_strtoul(buf, &end, 0);
3498 if (buf == end) break;
3499 }
3500 if (*end && !isspace(*end)) break;
3501 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3502 buf = skip_spaces(end);
3503 }
3504 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3505 out:
3506 return len;
3507 }
3508
3509 static struct md_sysfs_entry md_bitmap =
3510 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3511
3512 static ssize_t
3513 size_show(mddev_t *mddev, char *page)
3514 {
3515 return sprintf(page, "%llu\n",
3516 (unsigned long long)mddev->dev_sectors / 2);
3517 }
3518
3519 static int update_size(mddev_t *mddev, sector_t num_sectors);
3520
3521 static ssize_t
3522 size_store(mddev_t *mddev, const char *buf, size_t len)
3523 {
3524 /* If array is inactive, we can reduce the component size, but
3525 * not increase it (except from 0).
3526 * If array is active, we can try an on-line resize
3527 */
3528 sector_t sectors;
3529 int err = strict_blocks_to_sectors(buf, &sectors);
3530
3531 if (err < 0)
3532 return err;
3533 if (mddev->pers) {
3534 err = update_size(mddev, sectors);
3535 md_update_sb(mddev, 1);
3536 } else {
3537 if (mddev->dev_sectors == 0 ||
3538 mddev->dev_sectors > sectors)
3539 mddev->dev_sectors = sectors;
3540 else
3541 err = -ENOSPC;
3542 }
3543 return err ? err : len;
3544 }
3545
3546 static struct md_sysfs_entry md_size =
3547 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3548
3549
3550 /* Metdata version.
3551 * This is one of
3552 * 'none' for arrays with no metadata (good luck...)
3553 * 'external' for arrays with externally managed metadata,
3554 * or N.M for internally known formats
3555 */
3556 static ssize_t
3557 metadata_show(mddev_t *mddev, char *page)
3558 {
3559 if (mddev->persistent)
3560 return sprintf(page, "%d.%d\n",
3561 mddev->major_version, mddev->minor_version);
3562 else if (mddev->external)
3563 return sprintf(page, "external:%s\n", mddev->metadata_type);
3564 else
3565 return sprintf(page, "none\n");
3566 }
3567
3568 static ssize_t
3569 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3570 {
3571 int major, minor;
3572 char *e;
3573 /* Changing the details of 'external' metadata is
3574 * always permitted. Otherwise there must be
3575 * no devices attached to the array.
3576 */
3577 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3578 ;
3579 else if (!list_empty(&mddev->disks))
3580 return -EBUSY;
3581
3582 if (cmd_match(buf, "none")) {
3583 mddev->persistent = 0;
3584 mddev->external = 0;
3585 mddev->major_version = 0;
3586 mddev->minor_version = 90;
3587 return len;
3588 }
3589 if (strncmp(buf, "external:", 9) == 0) {
3590 size_t namelen = len-9;
3591 if (namelen >= sizeof(mddev->metadata_type))
3592 namelen = sizeof(mddev->metadata_type)-1;
3593 strncpy(mddev->metadata_type, buf+9, namelen);
3594 mddev->metadata_type[namelen] = 0;
3595 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3596 mddev->metadata_type[--namelen] = 0;
3597 mddev->persistent = 0;
3598 mddev->external = 1;
3599 mddev->major_version = 0;
3600 mddev->minor_version = 90;
3601 return len;
3602 }
3603 major = simple_strtoul(buf, &e, 10);
3604 if (e==buf || *e != '.')
3605 return -EINVAL;
3606 buf = e+1;
3607 minor = simple_strtoul(buf, &e, 10);
3608 if (e==buf || (*e && *e != '\n') )
3609 return -EINVAL;
3610 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3611 return -ENOENT;
3612 mddev->major_version = major;
3613 mddev->minor_version = minor;
3614 mddev->persistent = 1;
3615 mddev->external = 0;
3616 return len;
3617 }
3618
3619 static struct md_sysfs_entry md_metadata =
3620 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3621
3622 static ssize_t
3623 action_show(mddev_t *mddev, char *page)
3624 {
3625 char *type = "idle";
3626 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3627 type = "frozen";
3628 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3629 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3630 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3631 type = "reshape";
3632 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3633 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3634 type = "resync";
3635 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3636 type = "check";
3637 else
3638 type = "repair";
3639 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3640 type = "recover";
3641 }
3642 return sprintf(page, "%s\n", type);
3643 }
3644
3645 static ssize_t
3646 action_store(mddev_t *mddev, const char *page, size_t len)
3647 {
3648 if (!mddev->pers || !mddev->pers->sync_request)
3649 return -EINVAL;
3650
3651 if (cmd_match(page, "frozen"))
3652 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3653 else
3654 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3655
3656 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3657 if (mddev->sync_thread) {
3658 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3659 md_unregister_thread(mddev->sync_thread);
3660 mddev->sync_thread = NULL;
3661 mddev->recovery = 0;
3662 }
3663 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3664 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3665 return -EBUSY;
3666 else if (cmd_match(page, "resync"))
3667 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3668 else if (cmd_match(page, "recover")) {
3669 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3670 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3671 } else if (cmd_match(page, "reshape")) {
3672 int err;
3673 if (mddev->pers->start_reshape == NULL)
3674 return -EINVAL;
3675 err = mddev->pers->start_reshape(mddev);
3676 if (err)
3677 return err;
3678 sysfs_notify(&mddev->kobj, NULL, "degraded");
3679 } else {
3680 if (cmd_match(page, "check"))
3681 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3682 else if (!cmd_match(page, "repair"))
3683 return -EINVAL;
3684 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3685 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3686 }
3687 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3688 md_wakeup_thread(mddev->thread);
3689 sysfs_notify_dirent(mddev->sysfs_action);
3690 return len;
3691 }
3692
3693 static ssize_t
3694 mismatch_cnt_show(mddev_t *mddev, char *page)
3695 {
3696 return sprintf(page, "%llu\n",
3697 (unsigned long long) mddev->resync_mismatches);
3698 }
3699
3700 static struct md_sysfs_entry md_scan_mode =
3701 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3702
3703
3704 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3705
3706 static ssize_t
3707 sync_min_show(mddev_t *mddev, char *page)
3708 {
3709 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3710 mddev->sync_speed_min ? "local": "system");
3711 }
3712
3713 static ssize_t
3714 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3715 {
3716 int min;
3717 char *e;
3718 if (strncmp(buf, "system", 6)==0) {
3719 mddev->sync_speed_min = 0;
3720 return len;
3721 }
3722 min = simple_strtoul(buf, &e, 10);
3723 if (buf == e || (*e && *e != '\n') || min <= 0)
3724 return -EINVAL;
3725 mddev->sync_speed_min = min;
3726 return len;
3727 }
3728
3729 static struct md_sysfs_entry md_sync_min =
3730 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3731
3732 static ssize_t
3733 sync_max_show(mddev_t *mddev, char *page)
3734 {
3735 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3736 mddev->sync_speed_max ? "local": "system");
3737 }
3738
3739 static ssize_t
3740 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3741 {
3742 int max;
3743 char *e;
3744 if (strncmp(buf, "system", 6)==0) {
3745 mddev->sync_speed_max = 0;
3746 return len;
3747 }
3748 max = simple_strtoul(buf, &e, 10);
3749 if (buf == e || (*e && *e != '\n') || max <= 0)
3750 return -EINVAL;
3751 mddev->sync_speed_max = max;
3752 return len;
3753 }
3754
3755 static struct md_sysfs_entry md_sync_max =
3756 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3757
3758 static ssize_t
3759 degraded_show(mddev_t *mddev, char *page)
3760 {
3761 return sprintf(page, "%d\n", mddev->degraded);
3762 }
3763 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3764
3765 static ssize_t
3766 sync_force_parallel_show(mddev_t *mddev, char *page)
3767 {
3768 return sprintf(page, "%d\n", mddev->parallel_resync);
3769 }
3770
3771 static ssize_t
3772 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3773 {
3774 long n;
3775
3776 if (strict_strtol(buf, 10, &n))
3777 return -EINVAL;
3778
3779 if (n != 0 && n != 1)
3780 return -EINVAL;
3781
3782 mddev->parallel_resync = n;
3783
3784 if (mddev->sync_thread)
3785 wake_up(&resync_wait);
3786
3787 return len;
3788 }
3789
3790 /* force parallel resync, even with shared block devices */
3791 static struct md_sysfs_entry md_sync_force_parallel =
3792 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3793 sync_force_parallel_show, sync_force_parallel_store);
3794
3795 static ssize_t
3796 sync_speed_show(mddev_t *mddev, char *page)
3797 {
3798 unsigned long resync, dt, db;
3799 if (mddev->curr_resync == 0)
3800 return sprintf(page, "none\n");
3801 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3802 dt = (jiffies - mddev->resync_mark) / HZ;
3803 if (!dt) dt++;
3804 db = resync - mddev->resync_mark_cnt;
3805 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3806 }
3807
3808 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3809
3810 static ssize_t
3811 sync_completed_show(mddev_t *mddev, char *page)
3812 {
3813 unsigned long max_sectors, resync;
3814
3815 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3816 return sprintf(page, "none\n");
3817
3818 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3819 max_sectors = mddev->resync_max_sectors;
3820 else
3821 max_sectors = mddev->dev_sectors;
3822
3823 resync = mddev->curr_resync_completed;
3824 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3825 }
3826
3827 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3828
3829 static ssize_t
3830 min_sync_show(mddev_t *mddev, char *page)
3831 {
3832 return sprintf(page, "%llu\n",
3833 (unsigned long long)mddev->resync_min);
3834 }
3835 static ssize_t
3836 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3837 {
3838 unsigned long long min;
3839 if (strict_strtoull(buf, 10, &min))
3840 return -EINVAL;
3841 if (min > mddev->resync_max)
3842 return -EINVAL;
3843 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3844 return -EBUSY;
3845
3846 /* Must be a multiple of chunk_size */
3847 if (mddev->chunk_sectors) {
3848 sector_t temp = min;
3849 if (sector_div(temp, mddev->chunk_sectors))
3850 return -EINVAL;
3851 }
3852 mddev->resync_min = min;
3853
3854 return len;
3855 }
3856
3857 static struct md_sysfs_entry md_min_sync =
3858 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3859
3860 static ssize_t
3861 max_sync_show(mddev_t *mddev, char *page)
3862 {
3863 if (mddev->resync_max == MaxSector)
3864 return sprintf(page, "max\n");
3865 else
3866 return sprintf(page, "%llu\n",
3867 (unsigned long long)mddev->resync_max);
3868 }
3869 static ssize_t
3870 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3871 {
3872 if (strncmp(buf, "max", 3) == 0)
3873 mddev->resync_max = MaxSector;
3874 else {
3875 unsigned long long max;
3876 if (strict_strtoull(buf, 10, &max))
3877 return -EINVAL;
3878 if (max < mddev->resync_min)
3879 return -EINVAL;
3880 if (max < mddev->resync_max &&
3881 mddev->ro == 0 &&
3882 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3883 return -EBUSY;
3884
3885 /* Must be a multiple of chunk_size */
3886 if (mddev->chunk_sectors) {
3887 sector_t temp = max;
3888 if (sector_div(temp, mddev->chunk_sectors))
3889 return -EINVAL;
3890 }
3891 mddev->resync_max = max;
3892 }
3893 wake_up(&mddev->recovery_wait);
3894 return len;
3895 }
3896
3897 static struct md_sysfs_entry md_max_sync =
3898 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3899
3900 static ssize_t
3901 suspend_lo_show(mddev_t *mddev, char *page)
3902 {
3903 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3904 }
3905
3906 static ssize_t
3907 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3908 {
3909 char *e;
3910 unsigned long long new = simple_strtoull(buf, &e, 10);
3911
3912 if (mddev->pers == NULL ||
3913 mddev->pers->quiesce == NULL)
3914 return -EINVAL;
3915 if (buf == e || (*e && *e != '\n'))
3916 return -EINVAL;
3917 if (new >= mddev->suspend_hi ||
3918 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3919 mddev->suspend_lo = new;
3920 mddev->pers->quiesce(mddev, 2);
3921 return len;
3922 } else
3923 return -EINVAL;
3924 }
3925 static struct md_sysfs_entry md_suspend_lo =
3926 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3927
3928
3929 static ssize_t
3930 suspend_hi_show(mddev_t *mddev, char *page)
3931 {
3932 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3933 }
3934
3935 static ssize_t
3936 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3937 {
3938 char *e;
3939 unsigned long long new = simple_strtoull(buf, &e, 10);
3940
3941 if (mddev->pers == NULL ||
3942 mddev->pers->quiesce == NULL)
3943 return -EINVAL;
3944 if (buf == e || (*e && *e != '\n'))
3945 return -EINVAL;
3946 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3947 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3948 mddev->suspend_hi = new;
3949 mddev->pers->quiesce(mddev, 1);
3950 mddev->pers->quiesce(mddev, 0);
3951 return len;
3952 } else
3953 return -EINVAL;
3954 }
3955 static struct md_sysfs_entry md_suspend_hi =
3956 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3957
3958 static ssize_t
3959 reshape_position_show(mddev_t *mddev, char *page)
3960 {
3961 if (mddev->reshape_position != MaxSector)
3962 return sprintf(page, "%llu\n",
3963 (unsigned long long)mddev->reshape_position);
3964 strcpy(page, "none\n");
3965 return 5;
3966 }
3967
3968 static ssize_t
3969 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3970 {
3971 char *e;
3972 unsigned long long new = simple_strtoull(buf, &e, 10);
3973 if (mddev->pers)
3974 return -EBUSY;
3975 if (buf == e || (*e && *e != '\n'))
3976 return -EINVAL;
3977 mddev->reshape_position = new;
3978 mddev->delta_disks = 0;
3979 mddev->new_level = mddev->level;
3980 mddev->new_layout = mddev->layout;
3981 mddev->new_chunk_sectors = mddev->chunk_sectors;
3982 return len;
3983 }
3984
3985 static struct md_sysfs_entry md_reshape_position =
3986 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3987 reshape_position_store);
3988
3989 static ssize_t
3990 array_size_show(mddev_t *mddev, char *page)
3991 {
3992 if (mddev->external_size)
3993 return sprintf(page, "%llu\n",
3994 (unsigned long long)mddev->array_sectors/2);
3995 else
3996 return sprintf(page, "default\n");
3997 }
3998
3999 static ssize_t
4000 array_size_store(mddev_t *mddev, const char *buf, size_t len)
4001 {
4002 sector_t sectors;
4003
4004 if (strncmp(buf, "default", 7) == 0) {
4005 if (mddev->pers)
4006 sectors = mddev->pers->size(mddev, 0, 0);
4007 else
4008 sectors = mddev->array_sectors;
4009
4010 mddev->external_size = 0;
4011 } else {
4012 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4013 return -EINVAL;
4014 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4015 return -E2BIG;
4016
4017 mddev->external_size = 1;
4018 }
4019
4020 mddev->array_sectors = sectors;
4021 set_capacity(mddev->gendisk, mddev->array_sectors);
4022 if (mddev->pers)
4023 revalidate_disk(mddev->gendisk);
4024
4025 return len;
4026 }
4027
4028 static struct md_sysfs_entry md_array_size =
4029 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4030 array_size_store);
4031
4032 static struct attribute *md_default_attrs[] = {
4033 &md_level.attr,
4034 &md_layout.attr,
4035 &md_raid_disks.attr,
4036 &md_chunk_size.attr,
4037 &md_size.attr,
4038 &md_resync_start.attr,
4039 &md_metadata.attr,
4040 &md_new_device.attr,
4041 &md_safe_delay.attr,
4042 &md_array_state.attr,
4043 &md_reshape_position.attr,
4044 &md_array_size.attr,
4045 &max_corr_read_errors.attr,
4046 NULL,
4047 };
4048
4049 static struct attribute *md_redundancy_attrs[] = {
4050 &md_scan_mode.attr,
4051 &md_mismatches.attr,
4052 &md_sync_min.attr,
4053 &md_sync_max.attr,
4054 &md_sync_speed.attr,
4055 &md_sync_force_parallel.attr,
4056 &md_sync_completed.attr,
4057 &md_min_sync.attr,
4058 &md_max_sync.attr,
4059 &md_suspend_lo.attr,
4060 &md_suspend_hi.attr,
4061 &md_bitmap.attr,
4062 &md_degraded.attr,
4063 NULL,
4064 };
4065 static struct attribute_group md_redundancy_group = {
4066 .name = NULL,
4067 .attrs = md_redundancy_attrs,
4068 };
4069
4070
4071 static ssize_t
4072 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4073 {
4074 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4075 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4076 ssize_t rv;
4077
4078 if (!entry->show)
4079 return -EIO;
4080 rv = mddev_lock(mddev);
4081 if (!rv) {
4082 rv = entry->show(mddev, page);
4083 mddev_unlock(mddev);
4084 }
4085 return rv;
4086 }
4087
4088 static ssize_t
4089 md_attr_store(struct kobject *kobj, struct attribute *attr,
4090 const char *page, size_t length)
4091 {
4092 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4093 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4094 ssize_t rv;
4095
4096 if (!entry->store)
4097 return -EIO;
4098 if (!capable(CAP_SYS_ADMIN))
4099 return -EACCES;
4100 rv = mddev_lock(mddev);
4101 if (mddev->hold_active == UNTIL_IOCTL)
4102 mddev->hold_active = 0;
4103 if (!rv) {
4104 rv = entry->store(mddev, page, length);
4105 mddev_unlock(mddev);
4106 }
4107 return rv;
4108 }
4109
4110 static void md_free(struct kobject *ko)
4111 {
4112 mddev_t *mddev = container_of(ko, mddev_t, kobj);
4113
4114 if (mddev->sysfs_state)
4115 sysfs_put(mddev->sysfs_state);
4116
4117 if (mddev->gendisk) {
4118 del_gendisk(mddev->gendisk);
4119 put_disk(mddev->gendisk);
4120 }
4121 if (mddev->queue)
4122 blk_cleanup_queue(mddev->queue);
4123
4124 kfree(mddev);
4125 }
4126
4127 static struct sysfs_ops md_sysfs_ops = {
4128 .show = md_attr_show,
4129 .store = md_attr_store,
4130 };
4131 static struct kobj_type md_ktype = {
4132 .release = md_free,
4133 .sysfs_ops = &md_sysfs_ops,
4134 .default_attrs = md_default_attrs,
4135 };
4136
4137 int mdp_major = 0;
4138
4139 static void mddev_delayed_delete(struct work_struct *ws)
4140 {
4141 mddev_t *mddev = container_of(ws, mddev_t, del_work);
4142
4143 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4144 kobject_del(&mddev->kobj);
4145 kobject_put(&mddev->kobj);
4146 }
4147
4148 static int md_alloc(dev_t dev, char *name)
4149 {
4150 static DEFINE_MUTEX(disks_mutex);
4151 mddev_t *mddev = mddev_find(dev);
4152 struct gendisk *disk;
4153 int partitioned;
4154 int shift;
4155 int unit;
4156 int error;
4157
4158 if (!mddev)
4159 return -ENODEV;
4160
4161 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4162 shift = partitioned ? MdpMinorShift : 0;
4163 unit = MINOR(mddev->unit) >> shift;
4164
4165 /* wait for any previous instance if this device
4166 * to be completed removed (mddev_delayed_delete).
4167 */
4168 flush_scheduled_work();
4169
4170 mutex_lock(&disks_mutex);
4171 error = -EEXIST;
4172 if (mddev->gendisk)
4173 goto abort;
4174
4175 if (name) {
4176 /* Need to ensure that 'name' is not a duplicate.
4177 */
4178 mddev_t *mddev2;
4179 spin_lock(&all_mddevs_lock);
4180
4181 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4182 if (mddev2->gendisk &&
4183 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4184 spin_unlock(&all_mddevs_lock);
4185 goto abort;
4186 }
4187 spin_unlock(&all_mddevs_lock);
4188 }
4189
4190 error = -ENOMEM;
4191 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4192 if (!mddev->queue)
4193 goto abort;
4194 mddev->queue->queuedata = mddev;
4195
4196 /* Can be unlocked because the queue is new: no concurrency */
4197 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
4198
4199 blk_queue_make_request(mddev->queue, md_make_request);
4200
4201 disk = alloc_disk(1 << shift);
4202 if (!disk) {
4203 blk_cleanup_queue(mddev->queue);
4204 mddev->queue = NULL;
4205 goto abort;
4206 }
4207 disk->major = MAJOR(mddev->unit);
4208 disk->first_minor = unit << shift;
4209 if (name)
4210 strcpy(disk->disk_name, name);
4211 else if (partitioned)
4212 sprintf(disk->disk_name, "md_d%d", unit);
4213 else
4214 sprintf(disk->disk_name, "md%d", unit);
4215 disk->fops = &md_fops;
4216 disk->private_data = mddev;
4217 disk->queue = mddev->queue;
4218 /* Allow extended partitions. This makes the
4219 * 'mdp' device redundant, but we can't really
4220 * remove it now.
4221 */
4222 disk->flags |= GENHD_FL_EXT_DEVT;
4223 add_disk(disk);
4224 mddev->gendisk = disk;
4225 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4226 &disk_to_dev(disk)->kobj, "%s", "md");
4227 if (error) {
4228 /* This isn't possible, but as kobject_init_and_add is marked
4229 * __must_check, we must do something with the result
4230 */
4231 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4232 disk->disk_name);
4233 error = 0;
4234 }
4235 if (sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4236 printk(KERN_DEBUG "pointless warning\n");
4237 abort:
4238 mutex_unlock(&disks_mutex);
4239 if (!error) {
4240 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4241 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state");
4242 }
4243 mddev_put(mddev);
4244 return error;
4245 }
4246
4247 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4248 {
4249 md_alloc(dev, NULL);
4250 return NULL;
4251 }
4252
4253 static int add_named_array(const char *val, struct kernel_param *kp)
4254 {
4255 /* val must be "md_*" where * is not all digits.
4256 * We allocate an array with a large free minor number, and
4257 * set the name to val. val must not already be an active name.
4258 */
4259 int len = strlen(val);
4260 char buf[DISK_NAME_LEN];
4261
4262 while (len && val[len-1] == '\n')
4263 len--;
4264 if (len >= DISK_NAME_LEN)
4265 return -E2BIG;
4266 strlcpy(buf, val, len+1);
4267 if (strncmp(buf, "md_", 3) != 0)
4268 return -EINVAL;
4269 return md_alloc(0, buf);
4270 }
4271
4272 static void md_safemode_timeout(unsigned long data)
4273 {
4274 mddev_t *mddev = (mddev_t *) data;
4275
4276 if (!atomic_read(&mddev->writes_pending)) {
4277 mddev->safemode = 1;
4278 if (mddev->external)
4279 sysfs_notify_dirent(mddev->sysfs_state);
4280 }
4281 md_wakeup_thread(mddev->thread);
4282 }
4283
4284 static int start_dirty_degraded;
4285
4286 static int do_md_run(mddev_t * mddev)
4287 {
4288 int err;
4289 mdk_rdev_t *rdev;
4290 struct gendisk *disk;
4291 struct mdk_personality *pers;
4292
4293 if (list_empty(&mddev->disks))
4294 /* cannot run an array with no devices.. */
4295 return -EINVAL;
4296
4297 if (mddev->pers)
4298 return -EBUSY;
4299
4300 /* These two calls synchronise us with the
4301 * sysfs_remove_group calls in mddev_unlock,
4302 * so they must have completed.
4303 */
4304 mutex_lock(&mddev->open_mutex);
4305 mutex_unlock(&mddev->open_mutex);
4306
4307 /*
4308 * Analyze all RAID superblock(s)
4309 */
4310 if (!mddev->raid_disks) {
4311 if (!mddev->persistent)
4312 return -EINVAL;
4313 analyze_sbs(mddev);
4314 }
4315
4316 if (mddev->level != LEVEL_NONE)
4317 request_module("md-level-%d", mddev->level);
4318 else if (mddev->clevel[0])
4319 request_module("md-%s", mddev->clevel);
4320
4321 /*
4322 * Drop all container device buffers, from now on
4323 * the only valid external interface is through the md
4324 * device.
4325 */
4326 list_for_each_entry(rdev, &mddev->disks, same_set) {
4327 if (test_bit(Faulty, &rdev->flags))
4328 continue;
4329 sync_blockdev(rdev->bdev);
4330 invalidate_bdev(rdev->bdev);
4331
4332 /* perform some consistency tests on the device.
4333 * We don't want the data to overlap the metadata,
4334 * Internal Bitmap issues have been handled elsewhere.
4335 */
4336 if (rdev->data_offset < rdev->sb_start) {
4337 if (mddev->dev_sectors &&
4338 rdev->data_offset + mddev->dev_sectors
4339 > rdev->sb_start) {
4340 printk("md: %s: data overlaps metadata\n",
4341 mdname(mddev));
4342 return -EINVAL;
4343 }
4344 } else {
4345 if (rdev->sb_start + rdev->sb_size/512
4346 > rdev->data_offset) {
4347 printk("md: %s: metadata overlaps data\n",
4348 mdname(mddev));
4349 return -EINVAL;
4350 }
4351 }
4352 sysfs_notify_dirent(rdev->sysfs_state);
4353 }
4354
4355 disk = mddev->gendisk;
4356
4357 spin_lock(&pers_lock);
4358 pers = find_pers(mddev->level, mddev->clevel);
4359 if (!pers || !try_module_get(pers->owner)) {
4360 spin_unlock(&pers_lock);
4361 if (mddev->level != LEVEL_NONE)
4362 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4363 mddev->level);
4364 else
4365 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4366 mddev->clevel);
4367 return -EINVAL;
4368 }
4369 mddev->pers = pers;
4370 spin_unlock(&pers_lock);
4371 if (mddev->level != pers->level) {
4372 mddev->level = pers->level;
4373 mddev->new_level = pers->level;
4374 }
4375 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4376
4377 if (mddev->reshape_position != MaxSector &&
4378 pers->start_reshape == NULL) {
4379 /* This personality cannot handle reshaping... */
4380 mddev->pers = NULL;
4381 module_put(pers->owner);
4382 return -EINVAL;
4383 }
4384
4385 if (pers->sync_request) {
4386 /* Warn if this is a potentially silly
4387 * configuration.
4388 */
4389 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4390 mdk_rdev_t *rdev2;
4391 int warned = 0;
4392
4393 list_for_each_entry(rdev, &mddev->disks, same_set)
4394 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4395 if (rdev < rdev2 &&
4396 rdev->bdev->bd_contains ==
4397 rdev2->bdev->bd_contains) {
4398 printk(KERN_WARNING
4399 "%s: WARNING: %s appears to be"
4400 " on the same physical disk as"
4401 " %s.\n",
4402 mdname(mddev),
4403 bdevname(rdev->bdev,b),
4404 bdevname(rdev2->bdev,b2));
4405 warned = 1;
4406 }
4407 }
4408
4409 if (warned)
4410 printk(KERN_WARNING
4411 "True protection against single-disk"
4412 " failure might be compromised.\n");
4413 }
4414
4415 mddev->recovery = 0;
4416 /* may be over-ridden by personality */
4417 mddev->resync_max_sectors = mddev->dev_sectors;
4418
4419 mddev->barriers_work = 1;
4420 mddev->ok_start_degraded = start_dirty_degraded;
4421
4422 if (start_readonly && mddev->ro == 0)
4423 mddev->ro = 2; /* read-only, but switch on first write */
4424
4425 err = mddev->pers->run(mddev);
4426 if (err)
4427 printk(KERN_ERR "md: pers->run() failed ...\n");
4428 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4429 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4430 " but 'external_size' not in effect?\n", __func__);
4431 printk(KERN_ERR
4432 "md: invalid array_size %llu > default size %llu\n",
4433 (unsigned long long)mddev->array_sectors / 2,
4434 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4435 err = -EINVAL;
4436 mddev->pers->stop(mddev);
4437 }
4438 if (err == 0 && mddev->pers->sync_request) {
4439 err = bitmap_create(mddev);
4440 if (err) {
4441 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4442 mdname(mddev), err);
4443 mddev->pers->stop(mddev);
4444 }
4445 }
4446 if (err) {
4447 module_put(mddev->pers->owner);
4448 mddev->pers = NULL;
4449 bitmap_destroy(mddev);
4450 return err;
4451 }
4452 if (mddev->pers->sync_request) {
4453 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4454 printk(KERN_WARNING
4455 "md: cannot register extra attributes for %s\n",
4456 mdname(mddev));
4457 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4458 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4459 mddev->ro = 0;
4460
4461 atomic_set(&mddev->writes_pending,0);
4462 atomic_set(&mddev->max_corr_read_errors,
4463 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
4464 mddev->safemode = 0;
4465 mddev->safemode_timer.function = md_safemode_timeout;
4466 mddev->safemode_timer.data = (unsigned long) mddev;
4467 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4468 mddev->in_sync = 1;
4469
4470 list_for_each_entry(rdev, &mddev->disks, same_set)
4471 if (rdev->raid_disk >= 0) {
4472 char nm[20];
4473 sprintf(nm, "rd%d", rdev->raid_disk);
4474 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4475 printk("md: cannot register %s for %s\n",
4476 nm, mdname(mddev));
4477 }
4478
4479 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4480
4481 if (mddev->flags)
4482 md_update_sb(mddev, 0);
4483
4484 set_capacity(disk, mddev->array_sectors);
4485
4486 md_wakeup_thread(mddev->thread);
4487 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4488
4489 revalidate_disk(mddev->gendisk);
4490 mddev->changed = 1;
4491 md_new_event(mddev);
4492 sysfs_notify_dirent(mddev->sysfs_state);
4493 if (mddev->sysfs_action)
4494 sysfs_notify_dirent(mddev->sysfs_action);
4495 sysfs_notify(&mddev->kobj, NULL, "degraded");
4496 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4497 return 0;
4498 }
4499
4500 static int restart_array(mddev_t *mddev)
4501 {
4502 struct gendisk *disk = mddev->gendisk;
4503
4504 /* Complain if it has no devices */
4505 if (list_empty(&mddev->disks))
4506 return -ENXIO;
4507 if (!mddev->pers)
4508 return -EINVAL;
4509 if (!mddev->ro)
4510 return -EBUSY;
4511 mddev->safemode = 0;
4512 mddev->ro = 0;
4513 set_disk_ro(disk, 0);
4514 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4515 mdname(mddev));
4516 /* Kick recovery or resync if necessary */
4517 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4518 md_wakeup_thread(mddev->thread);
4519 md_wakeup_thread(mddev->sync_thread);
4520 sysfs_notify_dirent(mddev->sysfs_state);
4521 return 0;
4522 }
4523
4524 /* similar to deny_write_access, but accounts for our holding a reference
4525 * to the file ourselves */
4526 static int deny_bitmap_write_access(struct file * file)
4527 {
4528 struct inode *inode = file->f_mapping->host;
4529
4530 spin_lock(&inode->i_lock);
4531 if (atomic_read(&inode->i_writecount) > 1) {
4532 spin_unlock(&inode->i_lock);
4533 return -ETXTBSY;
4534 }
4535 atomic_set(&inode->i_writecount, -1);
4536 spin_unlock(&inode->i_lock);
4537
4538 return 0;
4539 }
4540
4541 void restore_bitmap_write_access(struct file *file)
4542 {
4543 struct inode *inode = file->f_mapping->host;
4544
4545 spin_lock(&inode->i_lock);
4546 atomic_set(&inode->i_writecount, 1);
4547 spin_unlock(&inode->i_lock);
4548 }
4549
4550 /* mode:
4551 * 0 - completely stop and dis-assemble array
4552 * 1 - switch to readonly
4553 * 2 - stop but do not disassemble array
4554 */
4555 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4556 {
4557 int err = 0;
4558 struct gendisk *disk = mddev->gendisk;
4559 mdk_rdev_t *rdev;
4560
4561 mutex_lock(&mddev->open_mutex);
4562 if (atomic_read(&mddev->openers) > is_open) {
4563 printk("md: %s still in use.\n",mdname(mddev));
4564 err = -EBUSY;
4565 } else if (mddev->pers) {
4566
4567 if (mddev->sync_thread) {
4568 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4569 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4570 md_unregister_thread(mddev->sync_thread);
4571 mddev->sync_thread = NULL;
4572 }
4573
4574 del_timer_sync(&mddev->safemode_timer);
4575
4576 switch(mode) {
4577 case 1: /* readonly */
4578 err = -ENXIO;
4579 if (mddev->ro==1)
4580 goto out;
4581 mddev->ro = 1;
4582 break;
4583 case 0: /* disassemble */
4584 case 2: /* stop */
4585 bitmap_flush(mddev);
4586 md_super_wait(mddev);
4587 if (mddev->ro)
4588 set_disk_ro(disk, 0);
4589
4590 mddev->pers->stop(mddev);
4591 mddev->queue->merge_bvec_fn = NULL;
4592 mddev->queue->unplug_fn = NULL;
4593 mddev->queue->backing_dev_info.congested_fn = NULL;
4594 module_put(mddev->pers->owner);
4595 if (mddev->pers->sync_request && mddev->to_remove == NULL)
4596 mddev->to_remove = &md_redundancy_group;
4597 mddev->pers = NULL;
4598 /* tell userspace to handle 'inactive' */
4599 sysfs_notify_dirent(mddev->sysfs_state);
4600
4601 list_for_each_entry(rdev, &mddev->disks, same_set)
4602 if (rdev->raid_disk >= 0) {
4603 char nm[20];
4604 sprintf(nm, "rd%d", rdev->raid_disk);
4605 sysfs_remove_link(&mddev->kobj, nm);
4606 }
4607
4608 set_capacity(disk, 0);
4609 mddev->changed = 1;
4610
4611 if (mddev->ro)
4612 mddev->ro = 0;
4613 }
4614 if (!mddev->in_sync || mddev->flags) {
4615 /* mark array as shutdown cleanly */
4616 mddev->in_sync = 1;
4617 md_update_sb(mddev, 1);
4618 }
4619 if (mode == 1)
4620 set_disk_ro(disk, 1);
4621 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4622 err = 0;
4623 }
4624 out:
4625 mutex_unlock(&mddev->open_mutex);
4626 if (err)
4627 return err;
4628 /*
4629 * Free resources if final stop
4630 */
4631 if (mode == 0) {
4632
4633 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4634
4635 bitmap_destroy(mddev);
4636 if (mddev->bitmap_info.file) {
4637 restore_bitmap_write_access(mddev->bitmap_info.file);
4638 fput(mddev->bitmap_info.file);
4639 mddev->bitmap_info.file = NULL;
4640 }
4641 mddev->bitmap_info.offset = 0;
4642
4643 export_array(mddev);
4644
4645 mddev->array_sectors = 0;
4646 mddev->external_size = 0;
4647 mddev->dev_sectors = 0;
4648 mddev->raid_disks = 0;
4649 mddev->recovery_cp = 0;
4650 mddev->resync_min = 0;
4651 mddev->resync_max = MaxSector;
4652 mddev->reshape_position = MaxSector;
4653 mddev->external = 0;
4654 mddev->persistent = 0;
4655 mddev->level = LEVEL_NONE;
4656 mddev->clevel[0] = 0;
4657 mddev->flags = 0;
4658 mddev->ro = 0;
4659 mddev->metadata_type[0] = 0;
4660 mddev->chunk_sectors = 0;
4661 mddev->ctime = mddev->utime = 0;
4662 mddev->layout = 0;
4663 mddev->max_disks = 0;
4664 mddev->events = 0;
4665 mddev->delta_disks = 0;
4666 mddev->new_level = LEVEL_NONE;
4667 mddev->new_layout = 0;
4668 mddev->new_chunk_sectors = 0;
4669 mddev->curr_resync = 0;
4670 mddev->resync_mismatches = 0;
4671 mddev->suspend_lo = mddev->suspend_hi = 0;
4672 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4673 mddev->recovery = 0;
4674 mddev->in_sync = 0;
4675 mddev->changed = 0;
4676 mddev->degraded = 0;
4677 mddev->barriers_work = 0;
4678 mddev->safemode = 0;
4679 mddev->bitmap_info.offset = 0;
4680 mddev->bitmap_info.default_offset = 0;
4681 mddev->bitmap_info.chunksize = 0;
4682 mddev->bitmap_info.daemon_sleep = 0;
4683 mddev->bitmap_info.max_write_behind = 0;
4684 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4685 if (mddev->hold_active == UNTIL_STOP)
4686 mddev->hold_active = 0;
4687
4688 } else if (mddev->pers)
4689 printk(KERN_INFO "md: %s switched to read-only mode.\n",
4690 mdname(mddev));
4691 err = 0;
4692 blk_integrity_unregister(disk);
4693 md_new_event(mddev);
4694 sysfs_notify_dirent(mddev->sysfs_state);
4695 return err;
4696 }
4697
4698 #ifndef MODULE
4699 static void autorun_array(mddev_t *mddev)
4700 {
4701 mdk_rdev_t *rdev;
4702 int err;
4703
4704 if (list_empty(&mddev->disks))
4705 return;
4706
4707 printk(KERN_INFO "md: running: ");
4708
4709 list_for_each_entry(rdev, &mddev->disks, same_set) {
4710 char b[BDEVNAME_SIZE];
4711 printk("<%s>", bdevname(rdev->bdev,b));
4712 }
4713 printk("\n");
4714
4715 err = do_md_run(mddev);
4716 if (err) {
4717 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4718 do_md_stop(mddev, 0, 0);
4719 }
4720 }
4721
4722 /*
4723 * lets try to run arrays based on all disks that have arrived
4724 * until now. (those are in pending_raid_disks)
4725 *
4726 * the method: pick the first pending disk, collect all disks with
4727 * the same UUID, remove all from the pending list and put them into
4728 * the 'same_array' list. Then order this list based on superblock
4729 * update time (freshest comes first), kick out 'old' disks and
4730 * compare superblocks. If everything's fine then run it.
4731 *
4732 * If "unit" is allocated, then bump its reference count
4733 */
4734 static void autorun_devices(int part)
4735 {
4736 mdk_rdev_t *rdev0, *rdev, *tmp;
4737 mddev_t *mddev;
4738 char b[BDEVNAME_SIZE];
4739
4740 printk(KERN_INFO "md: autorun ...\n");
4741 while (!list_empty(&pending_raid_disks)) {
4742 int unit;
4743 dev_t dev;
4744 LIST_HEAD(candidates);
4745 rdev0 = list_entry(pending_raid_disks.next,
4746 mdk_rdev_t, same_set);
4747
4748 printk(KERN_INFO "md: considering %s ...\n",
4749 bdevname(rdev0->bdev,b));
4750 INIT_LIST_HEAD(&candidates);
4751 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4752 if (super_90_load(rdev, rdev0, 0) >= 0) {
4753 printk(KERN_INFO "md: adding %s ...\n",
4754 bdevname(rdev->bdev,b));
4755 list_move(&rdev->same_set, &candidates);
4756 }
4757 /*
4758 * now we have a set of devices, with all of them having
4759 * mostly sane superblocks. It's time to allocate the
4760 * mddev.
4761 */
4762 if (part) {
4763 dev = MKDEV(mdp_major,
4764 rdev0->preferred_minor << MdpMinorShift);
4765 unit = MINOR(dev) >> MdpMinorShift;
4766 } else {
4767 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4768 unit = MINOR(dev);
4769 }
4770 if (rdev0->preferred_minor != unit) {
4771 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4772 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4773 break;
4774 }
4775
4776 md_probe(dev, NULL, NULL);
4777 mddev = mddev_find(dev);
4778 if (!mddev || !mddev->gendisk) {
4779 if (mddev)
4780 mddev_put(mddev);
4781 printk(KERN_ERR
4782 "md: cannot allocate memory for md drive.\n");
4783 break;
4784 }
4785 if (mddev_lock(mddev))
4786 printk(KERN_WARNING "md: %s locked, cannot run\n",
4787 mdname(mddev));
4788 else if (mddev->raid_disks || mddev->major_version
4789 || !list_empty(&mddev->disks)) {
4790 printk(KERN_WARNING
4791 "md: %s already running, cannot run %s\n",
4792 mdname(mddev), bdevname(rdev0->bdev,b));
4793 mddev_unlock(mddev);
4794 } else {
4795 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4796 mddev->persistent = 1;
4797 rdev_for_each_list(rdev, tmp, &candidates) {
4798 list_del_init(&rdev->same_set);
4799 if (bind_rdev_to_array(rdev, mddev))
4800 export_rdev(rdev);
4801 }
4802 autorun_array(mddev);
4803 mddev_unlock(mddev);
4804 }
4805 /* on success, candidates will be empty, on error
4806 * it won't...
4807 */
4808 rdev_for_each_list(rdev, tmp, &candidates) {
4809 list_del_init(&rdev->same_set);
4810 export_rdev(rdev);
4811 }
4812 mddev_put(mddev);
4813 }
4814 printk(KERN_INFO "md: ... autorun DONE.\n");
4815 }
4816 #endif /* !MODULE */
4817
4818 static int get_version(void __user * arg)
4819 {
4820 mdu_version_t ver;
4821
4822 ver.major = MD_MAJOR_VERSION;
4823 ver.minor = MD_MINOR_VERSION;
4824 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4825
4826 if (copy_to_user(arg, &ver, sizeof(ver)))
4827 return -EFAULT;
4828
4829 return 0;
4830 }
4831
4832 static int get_array_info(mddev_t * mddev, void __user * arg)
4833 {
4834 mdu_array_info_t info;
4835 int nr,working,insync,failed,spare;
4836 mdk_rdev_t *rdev;
4837
4838 nr=working=insync=failed=spare=0;
4839 list_for_each_entry(rdev, &mddev->disks, same_set) {
4840 nr++;
4841 if (test_bit(Faulty, &rdev->flags))
4842 failed++;
4843 else {
4844 working++;
4845 if (test_bit(In_sync, &rdev->flags))
4846 insync++;
4847 else
4848 spare++;
4849 }
4850 }
4851
4852 info.major_version = mddev->major_version;
4853 info.minor_version = mddev->minor_version;
4854 info.patch_version = MD_PATCHLEVEL_VERSION;
4855 info.ctime = mddev->ctime;
4856 info.level = mddev->level;
4857 info.size = mddev->dev_sectors / 2;
4858 if (info.size != mddev->dev_sectors / 2) /* overflow */
4859 info.size = -1;
4860 info.nr_disks = nr;
4861 info.raid_disks = mddev->raid_disks;
4862 info.md_minor = mddev->md_minor;
4863 info.not_persistent= !mddev->persistent;
4864
4865 info.utime = mddev->utime;
4866 info.state = 0;
4867 if (mddev->in_sync)
4868 info.state = (1<<MD_SB_CLEAN);
4869 if (mddev->bitmap && mddev->bitmap_info.offset)
4870 info.state = (1<<MD_SB_BITMAP_PRESENT);
4871 info.active_disks = insync;
4872 info.working_disks = working;
4873 info.failed_disks = failed;
4874 info.spare_disks = spare;
4875
4876 info.layout = mddev->layout;
4877 info.chunk_size = mddev->chunk_sectors << 9;
4878
4879 if (copy_to_user(arg, &info, sizeof(info)))
4880 return -EFAULT;
4881
4882 return 0;
4883 }
4884
4885 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4886 {
4887 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4888 char *ptr, *buf = NULL;
4889 int err = -ENOMEM;
4890
4891 if (md_allow_write(mddev))
4892 file = kmalloc(sizeof(*file), GFP_NOIO);
4893 else
4894 file = kmalloc(sizeof(*file), GFP_KERNEL);
4895
4896 if (!file)
4897 goto out;
4898
4899 /* bitmap disabled, zero the first byte and copy out */
4900 if (!mddev->bitmap || !mddev->bitmap->file) {
4901 file->pathname[0] = '\0';
4902 goto copy_out;
4903 }
4904
4905 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4906 if (!buf)
4907 goto out;
4908
4909 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4910 if (IS_ERR(ptr))
4911 goto out;
4912
4913 strcpy(file->pathname, ptr);
4914
4915 copy_out:
4916 err = 0;
4917 if (copy_to_user(arg, file, sizeof(*file)))
4918 err = -EFAULT;
4919 out:
4920 kfree(buf);
4921 kfree(file);
4922 return err;
4923 }
4924
4925 static int get_disk_info(mddev_t * mddev, void __user * arg)
4926 {
4927 mdu_disk_info_t info;
4928 mdk_rdev_t *rdev;
4929
4930 if (copy_from_user(&info, arg, sizeof(info)))
4931 return -EFAULT;
4932
4933 rdev = find_rdev_nr(mddev, info.number);
4934 if (rdev) {
4935 info.major = MAJOR(rdev->bdev->bd_dev);
4936 info.minor = MINOR(rdev->bdev->bd_dev);
4937 info.raid_disk = rdev->raid_disk;
4938 info.state = 0;
4939 if (test_bit(Faulty, &rdev->flags))
4940 info.state |= (1<<MD_DISK_FAULTY);
4941 else if (test_bit(In_sync, &rdev->flags)) {
4942 info.state |= (1<<MD_DISK_ACTIVE);
4943 info.state |= (1<<MD_DISK_SYNC);
4944 }
4945 if (test_bit(WriteMostly, &rdev->flags))
4946 info.state |= (1<<MD_DISK_WRITEMOSTLY);
4947 } else {
4948 info.major = info.minor = 0;
4949 info.raid_disk = -1;
4950 info.state = (1<<MD_DISK_REMOVED);
4951 }
4952
4953 if (copy_to_user(arg, &info, sizeof(info)))
4954 return -EFAULT;
4955
4956 return 0;
4957 }
4958
4959 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4960 {
4961 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4962 mdk_rdev_t *rdev;
4963 dev_t dev = MKDEV(info->major,info->minor);
4964
4965 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4966 return -EOVERFLOW;
4967
4968 if (!mddev->raid_disks) {
4969 int err;
4970 /* expecting a device which has a superblock */
4971 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4972 if (IS_ERR(rdev)) {
4973 printk(KERN_WARNING
4974 "md: md_import_device returned %ld\n",
4975 PTR_ERR(rdev));
4976 return PTR_ERR(rdev);
4977 }
4978 if (!list_empty(&mddev->disks)) {
4979 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4980 mdk_rdev_t, same_set);
4981 err = super_types[mddev->major_version]
4982 .load_super(rdev, rdev0, mddev->minor_version);
4983 if (err < 0) {
4984 printk(KERN_WARNING
4985 "md: %s has different UUID to %s\n",
4986 bdevname(rdev->bdev,b),
4987 bdevname(rdev0->bdev,b2));
4988 export_rdev(rdev);
4989 return -EINVAL;
4990 }
4991 }
4992 err = bind_rdev_to_array(rdev, mddev);
4993 if (err)
4994 export_rdev(rdev);
4995 return err;
4996 }
4997
4998 /*
4999 * add_new_disk can be used once the array is assembled
5000 * to add "hot spares". They must already have a superblock
5001 * written
5002 */
5003 if (mddev->pers) {
5004 int err;
5005 if (!mddev->pers->hot_add_disk) {
5006 printk(KERN_WARNING
5007 "%s: personality does not support diskops!\n",
5008 mdname(mddev));
5009 return -EINVAL;
5010 }
5011 if (mddev->persistent)
5012 rdev = md_import_device(dev, mddev->major_version,
5013 mddev->minor_version);
5014 else
5015 rdev = md_import_device(dev, -1, -1);
5016 if (IS_ERR(rdev)) {
5017 printk(KERN_WARNING
5018 "md: md_import_device returned %ld\n",
5019 PTR_ERR(rdev));
5020 return PTR_ERR(rdev);
5021 }
5022 /* set save_raid_disk if appropriate */
5023 if (!mddev->persistent) {
5024 if (info->state & (1<<MD_DISK_SYNC) &&
5025 info->raid_disk < mddev->raid_disks)
5026 rdev->raid_disk = info->raid_disk;
5027 else
5028 rdev->raid_disk = -1;
5029 } else
5030 super_types[mddev->major_version].
5031 validate_super(mddev, rdev);
5032 rdev->saved_raid_disk = rdev->raid_disk;
5033
5034 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5035 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5036 set_bit(WriteMostly, &rdev->flags);
5037 else
5038 clear_bit(WriteMostly, &rdev->flags);
5039
5040 rdev->raid_disk = -1;
5041 err = bind_rdev_to_array(rdev, mddev);
5042 if (!err && !mddev->pers->hot_remove_disk) {
5043 /* If there is hot_add_disk but no hot_remove_disk
5044 * then added disks for geometry changes,
5045 * and should be added immediately.
5046 */
5047 super_types[mddev->major_version].
5048 validate_super(mddev, rdev);
5049 err = mddev->pers->hot_add_disk(mddev, rdev);
5050 if (err)
5051 unbind_rdev_from_array(rdev);
5052 }
5053 if (err)
5054 export_rdev(rdev);
5055 else
5056 sysfs_notify_dirent(rdev->sysfs_state);
5057
5058 md_update_sb(mddev, 1);
5059 if (mddev->degraded)
5060 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5061 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5062 md_wakeup_thread(mddev->thread);
5063 return err;
5064 }
5065
5066 /* otherwise, add_new_disk is only allowed
5067 * for major_version==0 superblocks
5068 */
5069 if (mddev->major_version != 0) {
5070 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5071 mdname(mddev));
5072 return -EINVAL;
5073 }
5074
5075 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5076 int err;
5077 rdev = md_import_device(dev, -1, 0);
5078 if (IS_ERR(rdev)) {
5079 printk(KERN_WARNING
5080 "md: error, md_import_device() returned %ld\n",
5081 PTR_ERR(rdev));
5082 return PTR_ERR(rdev);
5083 }
5084 rdev->desc_nr = info->number;
5085 if (info->raid_disk < mddev->raid_disks)
5086 rdev->raid_disk = info->raid_disk;
5087 else
5088 rdev->raid_disk = -1;
5089
5090 if (rdev->raid_disk < mddev->raid_disks)
5091 if (info->state & (1<<MD_DISK_SYNC))
5092 set_bit(In_sync, &rdev->flags);
5093
5094 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5095 set_bit(WriteMostly, &rdev->flags);
5096
5097 if (!mddev->persistent) {
5098 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5099 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5100 } else
5101 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5102 rdev->sectors = rdev->sb_start;
5103
5104 err = bind_rdev_to_array(rdev, mddev);
5105 if (err) {
5106 export_rdev(rdev);
5107 return err;
5108 }
5109 }
5110
5111 return 0;
5112 }
5113
5114 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
5115 {
5116 char b[BDEVNAME_SIZE];
5117 mdk_rdev_t *rdev;
5118
5119 rdev = find_rdev(mddev, dev);
5120 if (!rdev)
5121 return -ENXIO;
5122
5123 if (rdev->raid_disk >= 0)
5124 goto busy;
5125
5126 kick_rdev_from_array(rdev);
5127 md_update_sb(mddev, 1);
5128 md_new_event(mddev);
5129
5130 return 0;
5131 busy:
5132 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5133 bdevname(rdev->bdev,b), mdname(mddev));
5134 return -EBUSY;
5135 }
5136
5137 static int hot_add_disk(mddev_t * mddev, dev_t dev)
5138 {
5139 char b[BDEVNAME_SIZE];
5140 int err;
5141 mdk_rdev_t *rdev;
5142
5143 if (!mddev->pers)
5144 return -ENODEV;
5145
5146 if (mddev->major_version != 0) {
5147 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5148 " version-0 superblocks.\n",
5149 mdname(mddev));
5150 return -EINVAL;
5151 }
5152 if (!mddev->pers->hot_add_disk) {
5153 printk(KERN_WARNING
5154 "%s: personality does not support diskops!\n",
5155 mdname(mddev));
5156 return -EINVAL;
5157 }
5158
5159 rdev = md_import_device(dev, -1, 0);
5160 if (IS_ERR(rdev)) {
5161 printk(KERN_WARNING
5162 "md: error, md_import_device() returned %ld\n",
5163 PTR_ERR(rdev));
5164 return -EINVAL;
5165 }
5166
5167 if (mddev->persistent)
5168 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5169 else
5170 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5171
5172 rdev->sectors = rdev->sb_start;
5173
5174 if (test_bit(Faulty, &rdev->flags)) {
5175 printk(KERN_WARNING
5176 "md: can not hot-add faulty %s disk to %s!\n",
5177 bdevname(rdev->bdev,b), mdname(mddev));
5178 err = -EINVAL;
5179 goto abort_export;
5180 }
5181 clear_bit(In_sync, &rdev->flags);
5182 rdev->desc_nr = -1;
5183 rdev->saved_raid_disk = -1;
5184 err = bind_rdev_to_array(rdev, mddev);
5185 if (err)
5186 goto abort_export;
5187
5188 /*
5189 * The rest should better be atomic, we can have disk failures
5190 * noticed in interrupt contexts ...
5191 */
5192
5193 rdev->raid_disk = -1;
5194
5195 md_update_sb(mddev, 1);
5196
5197 /*
5198 * Kick recovery, maybe this spare has to be added to the
5199 * array immediately.
5200 */
5201 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5202 md_wakeup_thread(mddev->thread);
5203 md_new_event(mddev);
5204 return 0;
5205
5206 abort_export:
5207 export_rdev(rdev);
5208 return err;
5209 }
5210
5211 static int set_bitmap_file(mddev_t *mddev, int fd)
5212 {
5213 int err;
5214
5215 if (mddev->pers) {
5216 if (!mddev->pers->quiesce)
5217 return -EBUSY;
5218 if (mddev->recovery || mddev->sync_thread)
5219 return -EBUSY;
5220 /* we should be able to change the bitmap.. */
5221 }
5222
5223
5224 if (fd >= 0) {
5225 if (mddev->bitmap)
5226 return -EEXIST; /* cannot add when bitmap is present */
5227 mddev->bitmap_info.file = fget(fd);
5228
5229 if (mddev->bitmap_info.file == NULL) {
5230 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5231 mdname(mddev));
5232 return -EBADF;
5233 }
5234
5235 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5236 if (err) {
5237 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5238 mdname(mddev));
5239 fput(mddev->bitmap_info.file);
5240 mddev->bitmap_info.file = NULL;
5241 return err;
5242 }
5243 mddev->bitmap_info.offset = 0; /* file overrides offset */
5244 } else if (mddev->bitmap == NULL)
5245 return -ENOENT; /* cannot remove what isn't there */
5246 err = 0;
5247 if (mddev->pers) {
5248 mddev->pers->quiesce(mddev, 1);
5249 if (fd >= 0)
5250 err = bitmap_create(mddev);
5251 if (fd < 0 || err) {
5252 bitmap_destroy(mddev);
5253 fd = -1; /* make sure to put the file */
5254 }
5255 mddev->pers->quiesce(mddev, 0);
5256 }
5257 if (fd < 0) {
5258 if (mddev->bitmap_info.file) {
5259 restore_bitmap_write_access(mddev->bitmap_info.file);
5260 fput(mddev->bitmap_info.file);
5261 }
5262 mddev->bitmap_info.file = NULL;
5263 }
5264
5265 return err;
5266 }
5267
5268 /*
5269 * set_array_info is used two different ways
5270 * The original usage is when creating a new array.
5271 * In this usage, raid_disks is > 0 and it together with
5272 * level, size, not_persistent,layout,chunksize determine the
5273 * shape of the array.
5274 * This will always create an array with a type-0.90.0 superblock.
5275 * The newer usage is when assembling an array.
5276 * In this case raid_disks will be 0, and the major_version field is
5277 * use to determine which style super-blocks are to be found on the devices.
5278 * The minor and patch _version numbers are also kept incase the
5279 * super_block handler wishes to interpret them.
5280 */
5281 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5282 {
5283
5284 if (info->raid_disks == 0) {
5285 /* just setting version number for superblock loading */
5286 if (info->major_version < 0 ||
5287 info->major_version >= ARRAY_SIZE(super_types) ||
5288 super_types[info->major_version].name == NULL) {
5289 /* maybe try to auto-load a module? */
5290 printk(KERN_INFO
5291 "md: superblock version %d not known\n",
5292 info->major_version);
5293 return -EINVAL;
5294 }
5295 mddev->major_version = info->major_version;
5296 mddev->minor_version = info->minor_version;
5297 mddev->patch_version = info->patch_version;
5298 mddev->persistent = !info->not_persistent;
5299 /* ensure mddev_put doesn't delete this now that there
5300 * is some minimal configuration.
5301 */
5302 mddev->ctime = get_seconds();
5303 return 0;
5304 }
5305 mddev->major_version = MD_MAJOR_VERSION;
5306 mddev->minor_version = MD_MINOR_VERSION;
5307 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5308 mddev->ctime = get_seconds();
5309
5310 mddev->level = info->level;
5311 mddev->clevel[0] = 0;
5312 mddev->dev_sectors = 2 * (sector_t)info->size;
5313 mddev->raid_disks = info->raid_disks;
5314 /* don't set md_minor, it is determined by which /dev/md* was
5315 * openned
5316 */
5317 if (info->state & (1<<MD_SB_CLEAN))
5318 mddev->recovery_cp = MaxSector;
5319 else
5320 mddev->recovery_cp = 0;
5321 mddev->persistent = ! info->not_persistent;
5322 mddev->external = 0;
5323
5324 mddev->layout = info->layout;
5325 mddev->chunk_sectors = info->chunk_size >> 9;
5326
5327 mddev->max_disks = MD_SB_DISKS;
5328
5329 if (mddev->persistent)
5330 mddev->flags = 0;
5331 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5332
5333 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5334 mddev->bitmap_info.offset = 0;
5335
5336 mddev->reshape_position = MaxSector;
5337
5338 /*
5339 * Generate a 128 bit UUID
5340 */
5341 get_random_bytes(mddev->uuid, 16);
5342
5343 mddev->new_level = mddev->level;
5344 mddev->new_chunk_sectors = mddev->chunk_sectors;
5345 mddev->new_layout = mddev->layout;
5346 mddev->delta_disks = 0;
5347
5348 return 0;
5349 }
5350
5351 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5352 {
5353 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5354
5355 if (mddev->external_size)
5356 return;
5357
5358 mddev->array_sectors = array_sectors;
5359 }
5360 EXPORT_SYMBOL(md_set_array_sectors);
5361
5362 static int update_size(mddev_t *mddev, sector_t num_sectors)
5363 {
5364 mdk_rdev_t *rdev;
5365 int rv;
5366 int fit = (num_sectors == 0);
5367
5368 if (mddev->pers->resize == NULL)
5369 return -EINVAL;
5370 /* The "num_sectors" is the number of sectors of each device that
5371 * is used. This can only make sense for arrays with redundancy.
5372 * linear and raid0 always use whatever space is available. We can only
5373 * consider changing this number if no resync or reconstruction is
5374 * happening, and if the new size is acceptable. It must fit before the
5375 * sb_start or, if that is <data_offset, it must fit before the size
5376 * of each device. If num_sectors is zero, we find the largest size
5377 * that fits.
5378
5379 */
5380 if (mddev->sync_thread)
5381 return -EBUSY;
5382 if (mddev->bitmap)
5383 /* Sorry, cannot grow a bitmap yet, just remove it,
5384 * grow, and re-add.
5385 */
5386 return -EBUSY;
5387 list_for_each_entry(rdev, &mddev->disks, same_set) {
5388 sector_t avail = rdev->sectors;
5389
5390 if (fit && (num_sectors == 0 || num_sectors > avail))
5391 num_sectors = avail;
5392 if (avail < num_sectors)
5393 return -ENOSPC;
5394 }
5395 rv = mddev->pers->resize(mddev, num_sectors);
5396 if (!rv)
5397 revalidate_disk(mddev->gendisk);
5398 return rv;
5399 }
5400
5401 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5402 {
5403 int rv;
5404 /* change the number of raid disks */
5405 if (mddev->pers->check_reshape == NULL)
5406 return -EINVAL;
5407 if (raid_disks <= 0 ||
5408 raid_disks >= mddev->max_disks)
5409 return -EINVAL;
5410 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5411 return -EBUSY;
5412 mddev->delta_disks = raid_disks - mddev->raid_disks;
5413
5414 rv = mddev->pers->check_reshape(mddev);
5415 return rv;
5416 }
5417
5418
5419 /*
5420 * update_array_info is used to change the configuration of an
5421 * on-line array.
5422 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5423 * fields in the info are checked against the array.
5424 * Any differences that cannot be handled will cause an error.
5425 * Normally, only one change can be managed at a time.
5426 */
5427 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5428 {
5429 int rv = 0;
5430 int cnt = 0;
5431 int state = 0;
5432
5433 /* calculate expected state,ignoring low bits */
5434 if (mddev->bitmap && mddev->bitmap_info.offset)
5435 state |= (1 << MD_SB_BITMAP_PRESENT);
5436
5437 if (mddev->major_version != info->major_version ||
5438 mddev->minor_version != info->minor_version ||
5439 /* mddev->patch_version != info->patch_version || */
5440 mddev->ctime != info->ctime ||
5441 mddev->level != info->level ||
5442 /* mddev->layout != info->layout || */
5443 !mddev->persistent != info->not_persistent||
5444 mddev->chunk_sectors != info->chunk_size >> 9 ||
5445 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5446 ((state^info->state) & 0xfffffe00)
5447 )
5448 return -EINVAL;
5449 /* Check there is only one change */
5450 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5451 cnt++;
5452 if (mddev->raid_disks != info->raid_disks)
5453 cnt++;
5454 if (mddev->layout != info->layout)
5455 cnt++;
5456 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5457 cnt++;
5458 if (cnt == 0)
5459 return 0;
5460 if (cnt > 1)
5461 return -EINVAL;
5462
5463 if (mddev->layout != info->layout) {
5464 /* Change layout
5465 * we don't need to do anything at the md level, the
5466 * personality will take care of it all.
5467 */
5468 if (mddev->pers->check_reshape == NULL)
5469 return -EINVAL;
5470 else {
5471 mddev->new_layout = info->layout;
5472 rv = mddev->pers->check_reshape(mddev);
5473 if (rv)
5474 mddev->new_layout = mddev->layout;
5475 return rv;
5476 }
5477 }
5478 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5479 rv = update_size(mddev, (sector_t)info->size * 2);
5480
5481 if (mddev->raid_disks != info->raid_disks)
5482 rv = update_raid_disks(mddev, info->raid_disks);
5483
5484 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5485 if (mddev->pers->quiesce == NULL)
5486 return -EINVAL;
5487 if (mddev->recovery || mddev->sync_thread)
5488 return -EBUSY;
5489 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5490 /* add the bitmap */
5491 if (mddev->bitmap)
5492 return -EEXIST;
5493 if (mddev->bitmap_info.default_offset == 0)
5494 return -EINVAL;
5495 mddev->bitmap_info.offset =
5496 mddev->bitmap_info.default_offset;
5497 mddev->pers->quiesce(mddev, 1);
5498 rv = bitmap_create(mddev);
5499 if (rv)
5500 bitmap_destroy(mddev);
5501 mddev->pers->quiesce(mddev, 0);
5502 } else {
5503 /* remove the bitmap */
5504 if (!mddev->bitmap)
5505 return -ENOENT;
5506 if (mddev->bitmap->file)
5507 return -EINVAL;
5508 mddev->pers->quiesce(mddev, 1);
5509 bitmap_destroy(mddev);
5510 mddev->pers->quiesce(mddev, 0);
5511 mddev->bitmap_info.offset = 0;
5512 }
5513 }
5514 md_update_sb(mddev, 1);
5515 return rv;
5516 }
5517
5518 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5519 {
5520 mdk_rdev_t *rdev;
5521
5522 if (mddev->pers == NULL)
5523 return -ENODEV;
5524
5525 rdev = find_rdev(mddev, dev);
5526 if (!rdev)
5527 return -ENODEV;
5528
5529 md_error(mddev, rdev);
5530 return 0;
5531 }
5532
5533 /*
5534 * We have a problem here : there is no easy way to give a CHS
5535 * virtual geometry. We currently pretend that we have a 2 heads
5536 * 4 sectors (with a BIG number of cylinders...). This drives
5537 * dosfs just mad... ;-)
5538 */
5539 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5540 {
5541 mddev_t *mddev = bdev->bd_disk->private_data;
5542
5543 geo->heads = 2;
5544 geo->sectors = 4;
5545 geo->cylinders = get_capacity(mddev->gendisk) / 8;
5546 return 0;
5547 }
5548
5549 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5550 unsigned int cmd, unsigned long arg)
5551 {
5552 int err = 0;
5553 void __user *argp = (void __user *)arg;
5554 mddev_t *mddev = NULL;
5555 int ro;
5556
5557 if (!capable(CAP_SYS_ADMIN))
5558 return -EACCES;
5559
5560 /*
5561 * Commands dealing with the RAID driver but not any
5562 * particular array:
5563 */
5564 switch (cmd)
5565 {
5566 case RAID_VERSION:
5567 err = get_version(argp);
5568 goto done;
5569
5570 case PRINT_RAID_DEBUG:
5571 err = 0;
5572 md_print_devices();
5573 goto done;
5574
5575 #ifndef MODULE
5576 case RAID_AUTORUN:
5577 err = 0;
5578 autostart_arrays(arg);
5579 goto done;
5580 #endif
5581 default:;
5582 }
5583
5584 /*
5585 * Commands creating/starting a new array:
5586 */
5587
5588 mddev = bdev->bd_disk->private_data;
5589
5590 if (!mddev) {
5591 BUG();
5592 goto abort;
5593 }
5594
5595 err = mddev_lock(mddev);
5596 if (err) {
5597 printk(KERN_INFO
5598 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5599 err, cmd);
5600 goto abort;
5601 }
5602
5603 switch (cmd)
5604 {
5605 case SET_ARRAY_INFO:
5606 {
5607 mdu_array_info_t info;
5608 if (!arg)
5609 memset(&info, 0, sizeof(info));
5610 else if (copy_from_user(&info, argp, sizeof(info))) {
5611 err = -EFAULT;
5612 goto abort_unlock;
5613 }
5614 if (mddev->pers) {
5615 err = update_array_info(mddev, &info);
5616 if (err) {
5617 printk(KERN_WARNING "md: couldn't update"
5618 " array info. %d\n", err);
5619 goto abort_unlock;
5620 }
5621 goto done_unlock;
5622 }
5623 if (!list_empty(&mddev->disks)) {
5624 printk(KERN_WARNING
5625 "md: array %s already has disks!\n",
5626 mdname(mddev));
5627 err = -EBUSY;
5628 goto abort_unlock;
5629 }
5630 if (mddev->raid_disks) {
5631 printk(KERN_WARNING
5632 "md: array %s already initialised!\n",
5633 mdname(mddev));
5634 err = -EBUSY;
5635 goto abort_unlock;
5636 }
5637 err = set_array_info(mddev, &info);
5638 if (err) {
5639 printk(KERN_WARNING "md: couldn't set"
5640 " array info. %d\n", err);
5641 goto abort_unlock;
5642 }
5643 }
5644 goto done_unlock;
5645
5646 default:;
5647 }
5648
5649 /*
5650 * Commands querying/configuring an existing array:
5651 */
5652 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5653 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5654 if ((!mddev->raid_disks && !mddev->external)
5655 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5656 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5657 && cmd != GET_BITMAP_FILE) {
5658 err = -ENODEV;
5659 goto abort_unlock;
5660 }
5661
5662 /*
5663 * Commands even a read-only array can execute:
5664 */
5665 switch (cmd)
5666 {
5667 case GET_ARRAY_INFO:
5668 err = get_array_info(mddev, argp);
5669 goto done_unlock;
5670
5671 case GET_BITMAP_FILE:
5672 err = get_bitmap_file(mddev, argp);
5673 goto done_unlock;
5674
5675 case GET_DISK_INFO:
5676 err = get_disk_info(mddev, argp);
5677 goto done_unlock;
5678
5679 case RESTART_ARRAY_RW:
5680 err = restart_array(mddev);
5681 goto done_unlock;
5682
5683 case STOP_ARRAY:
5684 err = do_md_stop(mddev, 0, 1);
5685 goto done_unlock;
5686
5687 case STOP_ARRAY_RO:
5688 err = do_md_stop(mddev, 1, 1);
5689 goto done_unlock;
5690
5691 case BLKROSET:
5692 if (get_user(ro, (int __user *)(arg))) {
5693 err = -EFAULT;
5694 goto done_unlock;
5695 }
5696 err = -EINVAL;
5697
5698 /* if the bdev is going readonly the value of mddev->ro
5699 * does not matter, no writes are coming
5700 */
5701 if (ro)
5702 goto done_unlock;
5703
5704 /* are we are already prepared for writes? */
5705 if (mddev->ro != 1)
5706 goto done_unlock;
5707
5708 /* transitioning to readauto need only happen for
5709 * arrays that call md_write_start
5710 */
5711 if (mddev->pers) {
5712 err = restart_array(mddev);
5713 if (err == 0) {
5714 mddev->ro = 2;
5715 set_disk_ro(mddev->gendisk, 0);
5716 }
5717 }
5718 goto done_unlock;
5719 }
5720
5721 /*
5722 * The remaining ioctls are changing the state of the
5723 * superblock, so we do not allow them on read-only arrays.
5724 * However non-MD ioctls (e.g. get-size) will still come through
5725 * here and hit the 'default' below, so only disallow
5726 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5727 */
5728 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5729 if (mddev->ro == 2) {
5730 mddev->ro = 0;
5731 sysfs_notify_dirent(mddev->sysfs_state);
5732 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5733 md_wakeup_thread(mddev->thread);
5734 } else {
5735 err = -EROFS;
5736 goto abort_unlock;
5737 }
5738 }
5739
5740 switch (cmd)
5741 {
5742 case ADD_NEW_DISK:
5743 {
5744 mdu_disk_info_t info;
5745 if (copy_from_user(&info, argp, sizeof(info)))
5746 err = -EFAULT;
5747 else
5748 err = add_new_disk(mddev, &info);
5749 goto done_unlock;
5750 }
5751
5752 case HOT_REMOVE_DISK:
5753 err = hot_remove_disk(mddev, new_decode_dev(arg));
5754 goto done_unlock;
5755
5756 case HOT_ADD_DISK:
5757 err = hot_add_disk(mddev, new_decode_dev(arg));
5758 goto done_unlock;
5759
5760 case SET_DISK_FAULTY:
5761 err = set_disk_faulty(mddev, new_decode_dev(arg));
5762 goto done_unlock;
5763
5764 case RUN_ARRAY:
5765 err = do_md_run(mddev);
5766 goto done_unlock;
5767
5768 case SET_BITMAP_FILE:
5769 err = set_bitmap_file(mddev, (int)arg);
5770 goto done_unlock;
5771
5772 default:
5773 err = -EINVAL;
5774 goto abort_unlock;
5775 }
5776
5777 done_unlock:
5778 abort_unlock:
5779 if (mddev->hold_active == UNTIL_IOCTL &&
5780 err != -EINVAL)
5781 mddev->hold_active = 0;
5782 mddev_unlock(mddev);
5783
5784 return err;
5785 done:
5786 if (err)
5787 MD_BUG();
5788 abort:
5789 return err;
5790 }
5791 #ifdef CONFIG_COMPAT
5792 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
5793 unsigned int cmd, unsigned long arg)
5794 {
5795 switch (cmd) {
5796 case HOT_REMOVE_DISK:
5797 case HOT_ADD_DISK:
5798 case SET_DISK_FAULTY:
5799 case SET_BITMAP_FILE:
5800 /* These take in integer arg, do not convert */
5801 break;
5802 default:
5803 arg = (unsigned long)compat_ptr(arg);
5804 break;
5805 }
5806
5807 return md_ioctl(bdev, mode, cmd, arg);
5808 }
5809 #endif /* CONFIG_COMPAT */
5810
5811 static int md_open(struct block_device *bdev, fmode_t mode)
5812 {
5813 /*
5814 * Succeed if we can lock the mddev, which confirms that
5815 * it isn't being stopped right now.
5816 */
5817 mddev_t *mddev = mddev_find(bdev->bd_dev);
5818 int err;
5819
5820 if (mddev->gendisk != bdev->bd_disk) {
5821 /* we are racing with mddev_put which is discarding this
5822 * bd_disk.
5823 */
5824 mddev_put(mddev);
5825 /* Wait until bdev->bd_disk is definitely gone */
5826 flush_scheduled_work();
5827 /* Then retry the open from the top */
5828 return -ERESTARTSYS;
5829 }
5830 BUG_ON(mddev != bdev->bd_disk->private_data);
5831
5832 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
5833 goto out;
5834
5835 err = 0;
5836 atomic_inc(&mddev->openers);
5837 mutex_unlock(&mddev->open_mutex);
5838
5839 check_disk_change(bdev);
5840 out:
5841 return err;
5842 }
5843
5844 static int md_release(struct gendisk *disk, fmode_t mode)
5845 {
5846 mddev_t *mddev = disk->private_data;
5847
5848 BUG_ON(!mddev);
5849 atomic_dec(&mddev->openers);
5850 mddev_put(mddev);
5851
5852 return 0;
5853 }
5854
5855 static int md_media_changed(struct gendisk *disk)
5856 {
5857 mddev_t *mddev = disk->private_data;
5858
5859 return mddev->changed;
5860 }
5861
5862 static int md_revalidate(struct gendisk *disk)
5863 {
5864 mddev_t *mddev = disk->private_data;
5865
5866 mddev->changed = 0;
5867 return 0;
5868 }
5869 static const struct block_device_operations md_fops =
5870 {
5871 .owner = THIS_MODULE,
5872 .open = md_open,
5873 .release = md_release,
5874 .ioctl = md_ioctl,
5875 #ifdef CONFIG_COMPAT
5876 .compat_ioctl = md_compat_ioctl,
5877 #endif
5878 .getgeo = md_getgeo,
5879 .media_changed = md_media_changed,
5880 .revalidate_disk= md_revalidate,
5881 };
5882
5883 static int md_thread(void * arg)
5884 {
5885 mdk_thread_t *thread = arg;
5886
5887 /*
5888 * md_thread is a 'system-thread', it's priority should be very
5889 * high. We avoid resource deadlocks individually in each
5890 * raid personality. (RAID5 does preallocation) We also use RR and
5891 * the very same RT priority as kswapd, thus we will never get
5892 * into a priority inversion deadlock.
5893 *
5894 * we definitely have to have equal or higher priority than
5895 * bdflush, otherwise bdflush will deadlock if there are too
5896 * many dirty RAID5 blocks.
5897 */
5898
5899 allow_signal(SIGKILL);
5900 while (!kthread_should_stop()) {
5901
5902 /* We need to wait INTERRUPTIBLE so that
5903 * we don't add to the load-average.
5904 * That means we need to be sure no signals are
5905 * pending
5906 */
5907 if (signal_pending(current))
5908 flush_signals(current);
5909
5910 wait_event_interruptible_timeout
5911 (thread->wqueue,
5912 test_bit(THREAD_WAKEUP, &thread->flags)
5913 || kthread_should_stop(),
5914 thread->timeout);
5915
5916 clear_bit(THREAD_WAKEUP, &thread->flags);
5917
5918 thread->run(thread->mddev);
5919 }
5920
5921 return 0;
5922 }
5923
5924 void md_wakeup_thread(mdk_thread_t *thread)
5925 {
5926 if (thread) {
5927 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5928 set_bit(THREAD_WAKEUP, &thread->flags);
5929 wake_up(&thread->wqueue);
5930 }
5931 }
5932
5933 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5934 const char *name)
5935 {
5936 mdk_thread_t *thread;
5937
5938 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5939 if (!thread)
5940 return NULL;
5941
5942 init_waitqueue_head(&thread->wqueue);
5943
5944 thread->run = run;
5945 thread->mddev = mddev;
5946 thread->timeout = MAX_SCHEDULE_TIMEOUT;
5947 thread->tsk = kthread_run(md_thread, thread,
5948 "%s_%s",
5949 mdname(thread->mddev),
5950 name ?: mddev->pers->name);
5951 if (IS_ERR(thread->tsk)) {
5952 kfree(thread);
5953 return NULL;
5954 }
5955 return thread;
5956 }
5957
5958 void md_unregister_thread(mdk_thread_t *thread)
5959 {
5960 if (!thread)
5961 return;
5962 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5963
5964 kthread_stop(thread->tsk);
5965 kfree(thread);
5966 }
5967
5968 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5969 {
5970 if (!mddev) {
5971 MD_BUG();
5972 return;
5973 }
5974
5975 if (!rdev || test_bit(Faulty, &rdev->flags))
5976 return;
5977
5978 if (mddev->external)
5979 set_bit(Blocked, &rdev->flags);
5980 /*
5981 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5982 mdname(mddev),
5983 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5984 __builtin_return_address(0),__builtin_return_address(1),
5985 __builtin_return_address(2),__builtin_return_address(3));
5986 */
5987 if (!mddev->pers)
5988 return;
5989 if (!mddev->pers->error_handler)
5990 return;
5991 mddev->pers->error_handler(mddev,rdev);
5992 if (mddev->degraded)
5993 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5994 sysfs_notify_dirent(rdev->sysfs_state);
5995 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5996 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5997 md_wakeup_thread(mddev->thread);
5998 md_new_event_inintr(mddev);
5999 }
6000
6001 /* seq_file implementation /proc/mdstat */
6002
6003 static void status_unused(struct seq_file *seq)
6004 {
6005 int i = 0;
6006 mdk_rdev_t *rdev;
6007
6008 seq_printf(seq, "unused devices: ");
6009
6010 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6011 char b[BDEVNAME_SIZE];
6012 i++;
6013 seq_printf(seq, "%s ",
6014 bdevname(rdev->bdev,b));
6015 }
6016 if (!i)
6017 seq_printf(seq, "<none>");
6018
6019 seq_printf(seq, "\n");
6020 }
6021
6022
6023 static void status_resync(struct seq_file *seq, mddev_t * mddev)
6024 {
6025 sector_t max_sectors, resync, res;
6026 unsigned long dt, db;
6027 sector_t rt;
6028 int scale;
6029 unsigned int per_milli;
6030
6031 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
6032
6033 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6034 max_sectors = mddev->resync_max_sectors;
6035 else
6036 max_sectors = mddev->dev_sectors;
6037
6038 /*
6039 * Should not happen.
6040 */
6041 if (!max_sectors) {
6042 MD_BUG();
6043 return;
6044 }
6045 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6046 * in a sector_t, and (max_sectors>>scale) will fit in a
6047 * u32, as those are the requirements for sector_div.
6048 * Thus 'scale' must be at least 10
6049 */
6050 scale = 10;
6051 if (sizeof(sector_t) > sizeof(unsigned long)) {
6052 while ( max_sectors/2 > (1ULL<<(scale+32)))
6053 scale++;
6054 }
6055 res = (resync>>scale)*1000;
6056 sector_div(res, (u32)((max_sectors>>scale)+1));
6057
6058 per_milli = res;
6059 {
6060 int i, x = per_milli/50, y = 20-x;
6061 seq_printf(seq, "[");
6062 for (i = 0; i < x; i++)
6063 seq_printf(seq, "=");
6064 seq_printf(seq, ">");
6065 for (i = 0; i < y; i++)
6066 seq_printf(seq, ".");
6067 seq_printf(seq, "] ");
6068 }
6069 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6070 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6071 "reshape" :
6072 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6073 "check" :
6074 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6075 "resync" : "recovery"))),
6076 per_milli/10, per_milli % 10,
6077 (unsigned long long) resync/2,
6078 (unsigned long long) max_sectors/2);
6079
6080 /*
6081 * dt: time from mark until now
6082 * db: blocks written from mark until now
6083 * rt: remaining time
6084 *
6085 * rt is a sector_t, so could be 32bit or 64bit.
6086 * So we divide before multiply in case it is 32bit and close
6087 * to the limit.
6088 * We scale the divisor (db) by 32 to avoid loosing precision
6089 * near the end of resync when the number of remaining sectors
6090 * is close to 'db'.
6091 * We then divide rt by 32 after multiplying by db to compensate.
6092 * The '+1' avoids division by zero if db is very small.
6093 */
6094 dt = ((jiffies - mddev->resync_mark) / HZ);
6095 if (!dt) dt++;
6096 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6097 - mddev->resync_mark_cnt;
6098
6099 rt = max_sectors - resync; /* number of remaining sectors */
6100 sector_div(rt, db/32+1);
6101 rt *= dt;
6102 rt >>= 5;
6103
6104 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6105 ((unsigned long)rt % 60)/6);
6106
6107 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6108 }
6109
6110 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6111 {
6112 struct list_head *tmp;
6113 loff_t l = *pos;
6114 mddev_t *mddev;
6115
6116 if (l >= 0x10000)
6117 return NULL;
6118 if (!l--)
6119 /* header */
6120 return (void*)1;
6121
6122 spin_lock(&all_mddevs_lock);
6123 list_for_each(tmp,&all_mddevs)
6124 if (!l--) {
6125 mddev = list_entry(tmp, mddev_t, all_mddevs);
6126 mddev_get(mddev);
6127 spin_unlock(&all_mddevs_lock);
6128 return mddev;
6129 }
6130 spin_unlock(&all_mddevs_lock);
6131 if (!l--)
6132 return (void*)2;/* tail */
6133 return NULL;
6134 }
6135
6136 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6137 {
6138 struct list_head *tmp;
6139 mddev_t *next_mddev, *mddev = v;
6140
6141 ++*pos;
6142 if (v == (void*)2)
6143 return NULL;
6144
6145 spin_lock(&all_mddevs_lock);
6146 if (v == (void*)1)
6147 tmp = all_mddevs.next;
6148 else
6149 tmp = mddev->all_mddevs.next;
6150 if (tmp != &all_mddevs)
6151 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
6152 else {
6153 next_mddev = (void*)2;
6154 *pos = 0x10000;
6155 }
6156 spin_unlock(&all_mddevs_lock);
6157
6158 if (v != (void*)1)
6159 mddev_put(mddev);
6160 return next_mddev;
6161
6162 }
6163
6164 static void md_seq_stop(struct seq_file *seq, void *v)
6165 {
6166 mddev_t *mddev = v;
6167
6168 if (mddev && v != (void*)1 && v != (void*)2)
6169 mddev_put(mddev);
6170 }
6171
6172 struct mdstat_info {
6173 int event;
6174 };
6175
6176 static int md_seq_show(struct seq_file *seq, void *v)
6177 {
6178 mddev_t *mddev = v;
6179 sector_t sectors;
6180 mdk_rdev_t *rdev;
6181 struct mdstat_info *mi = seq->private;
6182 struct bitmap *bitmap;
6183
6184 if (v == (void*)1) {
6185 struct mdk_personality *pers;
6186 seq_printf(seq, "Personalities : ");
6187 spin_lock(&pers_lock);
6188 list_for_each_entry(pers, &pers_list, list)
6189 seq_printf(seq, "[%s] ", pers->name);
6190
6191 spin_unlock(&pers_lock);
6192 seq_printf(seq, "\n");
6193 mi->event = atomic_read(&md_event_count);
6194 return 0;
6195 }
6196 if (v == (void*)2) {
6197 status_unused(seq);
6198 return 0;
6199 }
6200
6201 if (mddev_lock(mddev) < 0)
6202 return -EINTR;
6203
6204 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6205 seq_printf(seq, "%s : %sactive", mdname(mddev),
6206 mddev->pers ? "" : "in");
6207 if (mddev->pers) {
6208 if (mddev->ro==1)
6209 seq_printf(seq, " (read-only)");
6210 if (mddev->ro==2)
6211 seq_printf(seq, " (auto-read-only)");
6212 seq_printf(seq, " %s", mddev->pers->name);
6213 }
6214
6215 sectors = 0;
6216 list_for_each_entry(rdev, &mddev->disks, same_set) {
6217 char b[BDEVNAME_SIZE];
6218 seq_printf(seq, " %s[%d]",
6219 bdevname(rdev->bdev,b), rdev->desc_nr);
6220 if (test_bit(WriteMostly, &rdev->flags))
6221 seq_printf(seq, "(W)");
6222 if (test_bit(Faulty, &rdev->flags)) {
6223 seq_printf(seq, "(F)");
6224 continue;
6225 } else if (rdev->raid_disk < 0)
6226 seq_printf(seq, "(S)"); /* spare */
6227 sectors += rdev->sectors;
6228 }
6229
6230 if (!list_empty(&mddev->disks)) {
6231 if (mddev->pers)
6232 seq_printf(seq, "\n %llu blocks",
6233 (unsigned long long)
6234 mddev->array_sectors / 2);
6235 else
6236 seq_printf(seq, "\n %llu blocks",
6237 (unsigned long long)sectors / 2);
6238 }
6239 if (mddev->persistent) {
6240 if (mddev->major_version != 0 ||
6241 mddev->minor_version != 90) {
6242 seq_printf(seq," super %d.%d",
6243 mddev->major_version,
6244 mddev->minor_version);
6245 }
6246 } else if (mddev->external)
6247 seq_printf(seq, " super external:%s",
6248 mddev->metadata_type);
6249 else
6250 seq_printf(seq, " super non-persistent");
6251
6252 if (mddev->pers) {
6253 mddev->pers->status(seq, mddev);
6254 seq_printf(seq, "\n ");
6255 if (mddev->pers->sync_request) {
6256 if (mddev->curr_resync > 2) {
6257 status_resync(seq, mddev);
6258 seq_printf(seq, "\n ");
6259 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6260 seq_printf(seq, "\tresync=DELAYED\n ");
6261 else if (mddev->recovery_cp < MaxSector)
6262 seq_printf(seq, "\tresync=PENDING\n ");
6263 }
6264 } else
6265 seq_printf(seq, "\n ");
6266
6267 if ((bitmap = mddev->bitmap)) {
6268 unsigned long chunk_kb;
6269 unsigned long flags;
6270 spin_lock_irqsave(&bitmap->lock, flags);
6271 chunk_kb = mddev->bitmap_info.chunksize >> 10;
6272 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6273 "%lu%s chunk",
6274 bitmap->pages - bitmap->missing_pages,
6275 bitmap->pages,
6276 (bitmap->pages - bitmap->missing_pages)
6277 << (PAGE_SHIFT - 10),
6278 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
6279 chunk_kb ? "KB" : "B");
6280 if (bitmap->file) {
6281 seq_printf(seq, ", file: ");
6282 seq_path(seq, &bitmap->file->f_path, " \t\n");
6283 }
6284
6285 seq_printf(seq, "\n");
6286 spin_unlock_irqrestore(&bitmap->lock, flags);
6287 }
6288
6289 seq_printf(seq, "\n");
6290 }
6291 mddev_unlock(mddev);
6292
6293 return 0;
6294 }
6295
6296 static const struct seq_operations md_seq_ops = {
6297 .start = md_seq_start,
6298 .next = md_seq_next,
6299 .stop = md_seq_stop,
6300 .show = md_seq_show,
6301 };
6302
6303 static int md_seq_open(struct inode *inode, struct file *file)
6304 {
6305 int error;
6306 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6307 if (mi == NULL)
6308 return -ENOMEM;
6309
6310 error = seq_open(file, &md_seq_ops);
6311 if (error)
6312 kfree(mi);
6313 else {
6314 struct seq_file *p = file->private_data;
6315 p->private = mi;
6316 mi->event = atomic_read(&md_event_count);
6317 }
6318 return error;
6319 }
6320
6321 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6322 {
6323 struct seq_file *m = filp->private_data;
6324 struct mdstat_info *mi = m->private;
6325 int mask;
6326
6327 poll_wait(filp, &md_event_waiters, wait);
6328
6329 /* always allow read */
6330 mask = POLLIN | POLLRDNORM;
6331
6332 if (mi->event != atomic_read(&md_event_count))
6333 mask |= POLLERR | POLLPRI;
6334 return mask;
6335 }
6336
6337 static const struct file_operations md_seq_fops = {
6338 .owner = THIS_MODULE,
6339 .open = md_seq_open,
6340 .read = seq_read,
6341 .llseek = seq_lseek,
6342 .release = seq_release_private,
6343 .poll = mdstat_poll,
6344 };
6345
6346 int register_md_personality(struct mdk_personality *p)
6347 {
6348 spin_lock(&pers_lock);
6349 list_add_tail(&p->list, &pers_list);
6350 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6351 spin_unlock(&pers_lock);
6352 return 0;
6353 }
6354
6355 int unregister_md_personality(struct mdk_personality *p)
6356 {
6357 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6358 spin_lock(&pers_lock);
6359 list_del_init(&p->list);
6360 spin_unlock(&pers_lock);
6361 return 0;
6362 }
6363
6364 static int is_mddev_idle(mddev_t *mddev, int init)
6365 {
6366 mdk_rdev_t * rdev;
6367 int idle;
6368 int curr_events;
6369
6370 idle = 1;
6371 rcu_read_lock();
6372 rdev_for_each_rcu(rdev, mddev) {
6373 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6374 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6375 (int)part_stat_read(&disk->part0, sectors[1]) -
6376 atomic_read(&disk->sync_io);
6377 /* sync IO will cause sync_io to increase before the disk_stats
6378 * as sync_io is counted when a request starts, and
6379 * disk_stats is counted when it completes.
6380 * So resync activity will cause curr_events to be smaller than
6381 * when there was no such activity.
6382 * non-sync IO will cause disk_stat to increase without
6383 * increasing sync_io so curr_events will (eventually)
6384 * be larger than it was before. Once it becomes
6385 * substantially larger, the test below will cause
6386 * the array to appear non-idle, and resync will slow
6387 * down.
6388 * If there is a lot of outstanding resync activity when
6389 * we set last_event to curr_events, then all that activity
6390 * completing might cause the array to appear non-idle
6391 * and resync will be slowed down even though there might
6392 * not have been non-resync activity. This will only
6393 * happen once though. 'last_events' will soon reflect
6394 * the state where there is little or no outstanding
6395 * resync requests, and further resync activity will
6396 * always make curr_events less than last_events.
6397 *
6398 */
6399 if (init || curr_events - rdev->last_events > 64) {
6400 rdev->last_events = curr_events;
6401 idle = 0;
6402 }
6403 }
6404 rcu_read_unlock();
6405 return idle;
6406 }
6407
6408 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6409 {
6410 /* another "blocks" (512byte) blocks have been synced */
6411 atomic_sub(blocks, &mddev->recovery_active);
6412 wake_up(&mddev->recovery_wait);
6413 if (!ok) {
6414 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6415 md_wakeup_thread(mddev->thread);
6416 // stop recovery, signal do_sync ....
6417 }
6418 }
6419
6420
6421 /* md_write_start(mddev, bi)
6422 * If we need to update some array metadata (e.g. 'active' flag
6423 * in superblock) before writing, schedule a superblock update
6424 * and wait for it to complete.
6425 */
6426 void md_write_start(mddev_t *mddev, struct bio *bi)
6427 {
6428 int did_change = 0;
6429 if (bio_data_dir(bi) != WRITE)
6430 return;
6431
6432 BUG_ON(mddev->ro == 1);
6433 if (mddev->ro == 2) {
6434 /* need to switch to read/write */
6435 mddev->ro = 0;
6436 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6437 md_wakeup_thread(mddev->thread);
6438 md_wakeup_thread(mddev->sync_thread);
6439 did_change = 1;
6440 }
6441 atomic_inc(&mddev->writes_pending);
6442 if (mddev->safemode == 1)
6443 mddev->safemode = 0;
6444 if (mddev->in_sync) {
6445 spin_lock_irq(&mddev->write_lock);
6446 if (mddev->in_sync) {
6447 mddev->in_sync = 0;
6448 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6449 md_wakeup_thread(mddev->thread);
6450 did_change = 1;
6451 }
6452 spin_unlock_irq(&mddev->write_lock);
6453 }
6454 if (did_change)
6455 sysfs_notify_dirent(mddev->sysfs_state);
6456 wait_event(mddev->sb_wait,
6457 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
6458 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6459 }
6460
6461 void md_write_end(mddev_t *mddev)
6462 {
6463 if (atomic_dec_and_test(&mddev->writes_pending)) {
6464 if (mddev->safemode == 2)
6465 md_wakeup_thread(mddev->thread);
6466 else if (mddev->safemode_delay)
6467 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6468 }
6469 }
6470
6471 /* md_allow_write(mddev)
6472 * Calling this ensures that the array is marked 'active' so that writes
6473 * may proceed without blocking. It is important to call this before
6474 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6475 * Must be called with mddev_lock held.
6476 *
6477 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6478 * is dropped, so return -EAGAIN after notifying userspace.
6479 */
6480 int md_allow_write(mddev_t *mddev)
6481 {
6482 if (!mddev->pers)
6483 return 0;
6484 if (mddev->ro)
6485 return 0;
6486 if (!mddev->pers->sync_request)
6487 return 0;
6488
6489 spin_lock_irq(&mddev->write_lock);
6490 if (mddev->in_sync) {
6491 mddev->in_sync = 0;
6492 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6493 if (mddev->safemode_delay &&
6494 mddev->safemode == 0)
6495 mddev->safemode = 1;
6496 spin_unlock_irq(&mddev->write_lock);
6497 md_update_sb(mddev, 0);
6498 sysfs_notify_dirent(mddev->sysfs_state);
6499 } else
6500 spin_unlock_irq(&mddev->write_lock);
6501
6502 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
6503 return -EAGAIN;
6504 else
6505 return 0;
6506 }
6507 EXPORT_SYMBOL_GPL(md_allow_write);
6508
6509 #define SYNC_MARKS 10
6510 #define SYNC_MARK_STEP (3*HZ)
6511 void md_do_sync(mddev_t *mddev)
6512 {
6513 mddev_t *mddev2;
6514 unsigned int currspeed = 0,
6515 window;
6516 sector_t max_sectors,j, io_sectors;
6517 unsigned long mark[SYNC_MARKS];
6518 sector_t mark_cnt[SYNC_MARKS];
6519 int last_mark,m;
6520 struct list_head *tmp;
6521 sector_t last_check;
6522 int skipped = 0;
6523 mdk_rdev_t *rdev;
6524 char *desc;
6525
6526 /* just incase thread restarts... */
6527 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6528 return;
6529 if (mddev->ro) /* never try to sync a read-only array */
6530 return;
6531
6532 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6533 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6534 desc = "data-check";
6535 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6536 desc = "requested-resync";
6537 else
6538 desc = "resync";
6539 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6540 desc = "reshape";
6541 else
6542 desc = "recovery";
6543
6544 /* we overload curr_resync somewhat here.
6545 * 0 == not engaged in resync at all
6546 * 2 == checking that there is no conflict with another sync
6547 * 1 == like 2, but have yielded to allow conflicting resync to
6548 * commense
6549 * other == active in resync - this many blocks
6550 *
6551 * Before starting a resync we must have set curr_resync to
6552 * 2, and then checked that every "conflicting" array has curr_resync
6553 * less than ours. When we find one that is the same or higher
6554 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6555 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6556 * This will mean we have to start checking from the beginning again.
6557 *
6558 */
6559
6560 do {
6561 mddev->curr_resync = 2;
6562
6563 try_again:
6564 if (kthread_should_stop())
6565 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6566
6567 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6568 goto skip;
6569 for_each_mddev(mddev2, tmp) {
6570 if (mddev2 == mddev)
6571 continue;
6572 if (!mddev->parallel_resync
6573 && mddev2->curr_resync
6574 && match_mddev_units(mddev, mddev2)) {
6575 DEFINE_WAIT(wq);
6576 if (mddev < mddev2 && mddev->curr_resync == 2) {
6577 /* arbitrarily yield */
6578 mddev->curr_resync = 1;
6579 wake_up(&resync_wait);
6580 }
6581 if (mddev > mddev2 && mddev->curr_resync == 1)
6582 /* no need to wait here, we can wait the next
6583 * time 'round when curr_resync == 2
6584 */
6585 continue;
6586 /* We need to wait 'interruptible' so as not to
6587 * contribute to the load average, and not to
6588 * be caught by 'softlockup'
6589 */
6590 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6591 if (!kthread_should_stop() &&
6592 mddev2->curr_resync >= mddev->curr_resync) {
6593 printk(KERN_INFO "md: delaying %s of %s"
6594 " until %s has finished (they"
6595 " share one or more physical units)\n",
6596 desc, mdname(mddev), mdname(mddev2));
6597 mddev_put(mddev2);
6598 if (signal_pending(current))
6599 flush_signals(current);
6600 schedule();
6601 finish_wait(&resync_wait, &wq);
6602 goto try_again;
6603 }
6604 finish_wait(&resync_wait, &wq);
6605 }
6606 }
6607 } while (mddev->curr_resync < 2);
6608
6609 j = 0;
6610 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6611 /* resync follows the size requested by the personality,
6612 * which defaults to physical size, but can be virtual size
6613 */
6614 max_sectors = mddev->resync_max_sectors;
6615 mddev->resync_mismatches = 0;
6616 /* we don't use the checkpoint if there's a bitmap */
6617 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6618 j = mddev->resync_min;
6619 else if (!mddev->bitmap)
6620 j = mddev->recovery_cp;
6621
6622 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6623 max_sectors = mddev->dev_sectors;
6624 else {
6625 /* recovery follows the physical size of devices */
6626 max_sectors = mddev->dev_sectors;
6627 j = MaxSector;
6628 rcu_read_lock();
6629 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6630 if (rdev->raid_disk >= 0 &&
6631 !test_bit(Faulty, &rdev->flags) &&
6632 !test_bit(In_sync, &rdev->flags) &&
6633 rdev->recovery_offset < j)
6634 j = rdev->recovery_offset;
6635 rcu_read_unlock();
6636 }
6637
6638 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6639 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6640 " %d KB/sec/disk.\n", speed_min(mddev));
6641 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6642 "(but not more than %d KB/sec) for %s.\n",
6643 speed_max(mddev), desc);
6644
6645 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6646
6647 io_sectors = 0;
6648 for (m = 0; m < SYNC_MARKS; m++) {
6649 mark[m] = jiffies;
6650 mark_cnt[m] = io_sectors;
6651 }
6652 last_mark = 0;
6653 mddev->resync_mark = mark[last_mark];
6654 mddev->resync_mark_cnt = mark_cnt[last_mark];
6655
6656 /*
6657 * Tune reconstruction:
6658 */
6659 window = 32*(PAGE_SIZE/512);
6660 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6661 window/2,(unsigned long long) max_sectors/2);
6662
6663 atomic_set(&mddev->recovery_active, 0);
6664 last_check = 0;
6665
6666 if (j>2) {
6667 printk(KERN_INFO
6668 "md: resuming %s of %s from checkpoint.\n",
6669 desc, mdname(mddev));
6670 mddev->curr_resync = j;
6671 }
6672 mddev->curr_resync_completed = mddev->curr_resync;
6673
6674 while (j < max_sectors) {
6675 sector_t sectors;
6676
6677 skipped = 0;
6678
6679 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6680 ((mddev->curr_resync > mddev->curr_resync_completed &&
6681 (mddev->curr_resync - mddev->curr_resync_completed)
6682 > (max_sectors >> 4)) ||
6683 (j - mddev->curr_resync_completed)*2
6684 >= mddev->resync_max - mddev->curr_resync_completed
6685 )) {
6686 /* time to update curr_resync_completed */
6687 blk_unplug(mddev->queue);
6688 wait_event(mddev->recovery_wait,
6689 atomic_read(&mddev->recovery_active) == 0);
6690 mddev->curr_resync_completed =
6691 mddev->curr_resync;
6692 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6693 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6694 }
6695
6696 while (j >= mddev->resync_max && !kthread_should_stop()) {
6697 /* As this condition is controlled by user-space,
6698 * we can block indefinitely, so use '_interruptible'
6699 * to avoid triggering warnings.
6700 */
6701 flush_signals(current); /* just in case */
6702 wait_event_interruptible(mddev->recovery_wait,
6703 mddev->resync_max > j
6704 || kthread_should_stop());
6705 }
6706
6707 if (kthread_should_stop())
6708 goto interrupted;
6709
6710 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6711 currspeed < speed_min(mddev));
6712 if (sectors == 0) {
6713 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6714 goto out;
6715 }
6716
6717 if (!skipped) { /* actual IO requested */
6718 io_sectors += sectors;
6719 atomic_add(sectors, &mddev->recovery_active);
6720 }
6721
6722 j += sectors;
6723 if (j>1) mddev->curr_resync = j;
6724 mddev->curr_mark_cnt = io_sectors;
6725 if (last_check == 0)
6726 /* this is the earliers that rebuilt will be
6727 * visible in /proc/mdstat
6728 */
6729 md_new_event(mddev);
6730
6731 if (last_check + window > io_sectors || j == max_sectors)
6732 continue;
6733
6734 last_check = io_sectors;
6735
6736 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6737 break;
6738
6739 repeat:
6740 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6741 /* step marks */
6742 int next = (last_mark+1) % SYNC_MARKS;
6743
6744 mddev->resync_mark = mark[next];
6745 mddev->resync_mark_cnt = mark_cnt[next];
6746 mark[next] = jiffies;
6747 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6748 last_mark = next;
6749 }
6750
6751
6752 if (kthread_should_stop())
6753 goto interrupted;
6754
6755
6756 /*
6757 * this loop exits only if either when we are slower than
6758 * the 'hard' speed limit, or the system was IO-idle for
6759 * a jiffy.
6760 * the system might be non-idle CPU-wise, but we only care
6761 * about not overloading the IO subsystem. (things like an
6762 * e2fsck being done on the RAID array should execute fast)
6763 */
6764 blk_unplug(mddev->queue);
6765 cond_resched();
6766
6767 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6768 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6769
6770 if (currspeed > speed_min(mddev)) {
6771 if ((currspeed > speed_max(mddev)) ||
6772 !is_mddev_idle(mddev, 0)) {
6773 msleep(500);
6774 goto repeat;
6775 }
6776 }
6777 }
6778 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6779 /*
6780 * this also signals 'finished resyncing' to md_stop
6781 */
6782 out:
6783 blk_unplug(mddev->queue);
6784
6785 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6786
6787 /* tell personality that we are finished */
6788 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6789
6790 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6791 mddev->curr_resync > 2) {
6792 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6793 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6794 if (mddev->curr_resync >= mddev->recovery_cp) {
6795 printk(KERN_INFO
6796 "md: checkpointing %s of %s.\n",
6797 desc, mdname(mddev));
6798 mddev->recovery_cp = mddev->curr_resync;
6799 }
6800 } else
6801 mddev->recovery_cp = MaxSector;
6802 } else {
6803 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6804 mddev->curr_resync = MaxSector;
6805 rcu_read_lock();
6806 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6807 if (rdev->raid_disk >= 0 &&
6808 !test_bit(Faulty, &rdev->flags) &&
6809 !test_bit(In_sync, &rdev->flags) &&
6810 rdev->recovery_offset < mddev->curr_resync)
6811 rdev->recovery_offset = mddev->curr_resync;
6812 rcu_read_unlock();
6813 }
6814 }
6815 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6816
6817 skip:
6818 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6819 /* We completed so min/max setting can be forgotten if used. */
6820 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6821 mddev->resync_min = 0;
6822 mddev->resync_max = MaxSector;
6823 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6824 mddev->resync_min = mddev->curr_resync_completed;
6825 mddev->curr_resync = 0;
6826 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6827 mddev->curr_resync_completed = 0;
6828 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6829 wake_up(&resync_wait);
6830 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6831 md_wakeup_thread(mddev->thread);
6832 return;
6833
6834 interrupted:
6835 /*
6836 * got a signal, exit.
6837 */
6838 printk(KERN_INFO
6839 "md: md_do_sync() got signal ... exiting\n");
6840 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6841 goto out;
6842
6843 }
6844 EXPORT_SYMBOL_GPL(md_do_sync);
6845
6846
6847 static int remove_and_add_spares(mddev_t *mddev)
6848 {
6849 mdk_rdev_t *rdev;
6850 int spares = 0;
6851
6852 mddev->curr_resync_completed = 0;
6853
6854 list_for_each_entry(rdev, &mddev->disks, same_set)
6855 if (rdev->raid_disk >= 0 &&
6856 !test_bit(Blocked, &rdev->flags) &&
6857 (test_bit(Faulty, &rdev->flags) ||
6858 ! test_bit(In_sync, &rdev->flags)) &&
6859 atomic_read(&rdev->nr_pending)==0) {
6860 if (mddev->pers->hot_remove_disk(
6861 mddev, rdev->raid_disk)==0) {
6862 char nm[20];
6863 sprintf(nm,"rd%d", rdev->raid_disk);
6864 sysfs_remove_link(&mddev->kobj, nm);
6865 rdev->raid_disk = -1;
6866 }
6867 }
6868
6869 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
6870 list_for_each_entry(rdev, &mddev->disks, same_set) {
6871 if (rdev->raid_disk >= 0 &&
6872 !test_bit(In_sync, &rdev->flags) &&
6873 !test_bit(Blocked, &rdev->flags))
6874 spares++;
6875 if (rdev->raid_disk < 0
6876 && !test_bit(Faulty, &rdev->flags)) {
6877 rdev->recovery_offset = 0;
6878 if (mddev->pers->
6879 hot_add_disk(mddev, rdev) == 0) {
6880 char nm[20];
6881 sprintf(nm, "rd%d", rdev->raid_disk);
6882 if (sysfs_create_link(&mddev->kobj,
6883 &rdev->kobj, nm))
6884 printk(KERN_WARNING
6885 "md: cannot register "
6886 "%s for %s\n",
6887 nm, mdname(mddev));
6888 spares++;
6889 md_new_event(mddev);
6890 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6891 } else
6892 break;
6893 }
6894 }
6895 }
6896 return spares;
6897 }
6898 /*
6899 * This routine is regularly called by all per-raid-array threads to
6900 * deal with generic issues like resync and super-block update.
6901 * Raid personalities that don't have a thread (linear/raid0) do not
6902 * need this as they never do any recovery or update the superblock.
6903 *
6904 * It does not do any resync itself, but rather "forks" off other threads
6905 * to do that as needed.
6906 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6907 * "->recovery" and create a thread at ->sync_thread.
6908 * When the thread finishes it sets MD_RECOVERY_DONE
6909 * and wakeups up this thread which will reap the thread and finish up.
6910 * This thread also removes any faulty devices (with nr_pending == 0).
6911 *
6912 * The overall approach is:
6913 * 1/ if the superblock needs updating, update it.
6914 * 2/ If a recovery thread is running, don't do anything else.
6915 * 3/ If recovery has finished, clean up, possibly marking spares active.
6916 * 4/ If there are any faulty devices, remove them.
6917 * 5/ If array is degraded, try to add spares devices
6918 * 6/ If array has spares or is not in-sync, start a resync thread.
6919 */
6920 void md_check_recovery(mddev_t *mddev)
6921 {
6922 mdk_rdev_t *rdev;
6923
6924
6925 if (mddev->bitmap)
6926 bitmap_daemon_work(mddev);
6927
6928 if (mddev->ro)
6929 return;
6930
6931 if (signal_pending(current)) {
6932 if (mddev->pers->sync_request && !mddev->external) {
6933 printk(KERN_INFO "md: %s in immediate safe mode\n",
6934 mdname(mddev));
6935 mddev->safemode = 2;
6936 }
6937 flush_signals(current);
6938 }
6939
6940 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
6941 return;
6942 if ( ! (
6943 (mddev->flags && !mddev->external) ||
6944 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
6945 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
6946 (mddev->external == 0 && mddev->safemode == 1) ||
6947 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6948 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6949 ))
6950 return;
6951
6952 if (mddev_trylock(mddev)) {
6953 int spares = 0;
6954
6955 if (mddev->ro) {
6956 /* Only thing we do on a ro array is remove
6957 * failed devices.
6958 */
6959 remove_and_add_spares(mddev);
6960 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6961 goto unlock;
6962 }
6963
6964 if (!mddev->external) {
6965 int did_change = 0;
6966 spin_lock_irq(&mddev->write_lock);
6967 if (mddev->safemode &&
6968 !atomic_read(&mddev->writes_pending) &&
6969 !mddev->in_sync &&
6970 mddev->recovery_cp == MaxSector) {
6971 mddev->in_sync = 1;
6972 did_change = 1;
6973 if (mddev->persistent)
6974 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6975 }
6976 if (mddev->safemode == 1)
6977 mddev->safemode = 0;
6978 spin_unlock_irq(&mddev->write_lock);
6979 if (did_change)
6980 sysfs_notify_dirent(mddev->sysfs_state);
6981 }
6982
6983 if (mddev->flags)
6984 md_update_sb(mddev, 0);
6985
6986 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6987 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6988 /* resync/recovery still happening */
6989 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6990 goto unlock;
6991 }
6992 if (mddev->sync_thread) {
6993 /* resync has finished, collect result */
6994 md_unregister_thread(mddev->sync_thread);
6995 mddev->sync_thread = NULL;
6996 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
6997 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
6998 /* success...*/
6999 /* activate any spares */
7000 if (mddev->pers->spare_active(mddev))
7001 sysfs_notify(&mddev->kobj, NULL,
7002 "degraded");
7003 }
7004 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7005 mddev->pers->finish_reshape)
7006 mddev->pers->finish_reshape(mddev);
7007 md_update_sb(mddev, 1);
7008
7009 /* if array is no-longer degraded, then any saved_raid_disk
7010 * information must be scrapped
7011 */
7012 if (!mddev->degraded)
7013 list_for_each_entry(rdev, &mddev->disks, same_set)
7014 rdev->saved_raid_disk = -1;
7015
7016 mddev->recovery = 0;
7017 /* flag recovery needed just to double check */
7018 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7019 sysfs_notify_dirent(mddev->sysfs_action);
7020 md_new_event(mddev);
7021 goto unlock;
7022 }
7023 /* Set RUNNING before clearing NEEDED to avoid
7024 * any transients in the value of "sync_action".
7025 */
7026 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7027 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7028 /* Clear some bits that don't mean anything, but
7029 * might be left set
7030 */
7031 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7032 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7033
7034 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7035 goto unlock;
7036 /* no recovery is running.
7037 * remove any failed drives, then
7038 * add spares if possible.
7039 * Spare are also removed and re-added, to allow
7040 * the personality to fail the re-add.
7041 */
7042
7043 if (mddev->reshape_position != MaxSector) {
7044 if (mddev->pers->check_reshape == NULL ||
7045 mddev->pers->check_reshape(mddev) != 0)
7046 /* Cannot proceed */
7047 goto unlock;
7048 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7049 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7050 } else if ((spares = remove_and_add_spares(mddev))) {
7051 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7052 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7053 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7054 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7055 } else if (mddev->recovery_cp < MaxSector) {
7056 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7057 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7058 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7059 /* nothing to be done ... */
7060 goto unlock;
7061
7062 if (mddev->pers->sync_request) {
7063 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7064 /* We are adding a device or devices to an array
7065 * which has the bitmap stored on all devices.
7066 * So make sure all bitmap pages get written
7067 */
7068 bitmap_write_all(mddev->bitmap);
7069 }
7070 mddev->sync_thread = md_register_thread(md_do_sync,
7071 mddev,
7072 "resync");
7073 if (!mddev->sync_thread) {
7074 printk(KERN_ERR "%s: could not start resync"
7075 " thread...\n",
7076 mdname(mddev));
7077 /* leave the spares where they are, it shouldn't hurt */
7078 mddev->recovery = 0;
7079 } else
7080 md_wakeup_thread(mddev->sync_thread);
7081 sysfs_notify_dirent(mddev->sysfs_action);
7082 md_new_event(mddev);
7083 }
7084 unlock:
7085 if (!mddev->sync_thread) {
7086 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7087 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7088 &mddev->recovery))
7089 if (mddev->sysfs_action)
7090 sysfs_notify_dirent(mddev->sysfs_action);
7091 }
7092 mddev_unlock(mddev);
7093 }
7094 }
7095
7096 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
7097 {
7098 sysfs_notify_dirent(rdev->sysfs_state);
7099 wait_event_timeout(rdev->blocked_wait,
7100 !test_bit(Blocked, &rdev->flags),
7101 msecs_to_jiffies(5000));
7102 rdev_dec_pending(rdev, mddev);
7103 }
7104 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7105
7106 static int md_notify_reboot(struct notifier_block *this,
7107 unsigned long code, void *x)
7108 {
7109 struct list_head *tmp;
7110 mddev_t *mddev;
7111
7112 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
7113
7114 printk(KERN_INFO "md: stopping all md devices.\n");
7115
7116 for_each_mddev(mddev, tmp)
7117 if (mddev_trylock(mddev)) {
7118 /* Force a switch to readonly even array
7119 * appears to still be in use. Hence
7120 * the '100'.
7121 */
7122 do_md_stop(mddev, 1, 100);
7123 mddev_unlock(mddev);
7124 }
7125 /*
7126 * certain more exotic SCSI devices are known to be
7127 * volatile wrt too early system reboots. While the
7128 * right place to handle this issue is the given
7129 * driver, we do want to have a safe RAID driver ...
7130 */
7131 mdelay(1000*1);
7132 }
7133 return NOTIFY_DONE;
7134 }
7135
7136 static struct notifier_block md_notifier = {
7137 .notifier_call = md_notify_reboot,
7138 .next = NULL,
7139 .priority = INT_MAX, /* before any real devices */
7140 };
7141
7142 static void md_geninit(void)
7143 {
7144 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
7145
7146 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
7147 }
7148
7149 static int __init md_init(void)
7150 {
7151 if (register_blkdev(MD_MAJOR, "md"))
7152 return -1;
7153 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
7154 unregister_blkdev(MD_MAJOR, "md");
7155 return -1;
7156 }
7157 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
7158 md_probe, NULL, NULL);
7159 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
7160 md_probe, NULL, NULL);
7161
7162 register_reboot_notifier(&md_notifier);
7163 raid_table_header = register_sysctl_table(raid_root_table);
7164
7165 md_geninit();
7166 return 0;
7167 }
7168
7169
7170 #ifndef MODULE
7171
7172 /*
7173 * Searches all registered partitions for autorun RAID arrays
7174 * at boot time.
7175 */
7176
7177 static LIST_HEAD(all_detected_devices);
7178 struct detected_devices_node {
7179 struct list_head list;
7180 dev_t dev;
7181 };
7182
7183 void md_autodetect_dev(dev_t dev)
7184 {
7185 struct detected_devices_node *node_detected_dev;
7186
7187 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
7188 if (node_detected_dev) {
7189 node_detected_dev->dev = dev;
7190 list_add_tail(&node_detected_dev->list, &all_detected_devices);
7191 } else {
7192 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
7193 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
7194 }
7195 }
7196
7197
7198 static void autostart_arrays(int part)
7199 {
7200 mdk_rdev_t *rdev;
7201 struct detected_devices_node *node_detected_dev;
7202 dev_t dev;
7203 int i_scanned, i_passed;
7204
7205 i_scanned = 0;
7206 i_passed = 0;
7207
7208 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
7209
7210 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
7211 i_scanned++;
7212 node_detected_dev = list_entry(all_detected_devices.next,
7213 struct detected_devices_node, list);
7214 list_del(&node_detected_dev->list);
7215 dev = node_detected_dev->dev;
7216 kfree(node_detected_dev);
7217 rdev = md_import_device(dev,0, 90);
7218 if (IS_ERR(rdev))
7219 continue;
7220
7221 if (test_bit(Faulty, &rdev->flags)) {
7222 MD_BUG();
7223 continue;
7224 }
7225 set_bit(AutoDetected, &rdev->flags);
7226 list_add(&rdev->same_set, &pending_raid_disks);
7227 i_passed++;
7228 }
7229
7230 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
7231 i_scanned, i_passed);
7232
7233 autorun_devices(part);
7234 }
7235
7236 #endif /* !MODULE */
7237
7238 static __exit void md_exit(void)
7239 {
7240 mddev_t *mddev;
7241 struct list_head *tmp;
7242
7243 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
7244 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
7245
7246 unregister_blkdev(MD_MAJOR,"md");
7247 unregister_blkdev(mdp_major, "mdp");
7248 unregister_reboot_notifier(&md_notifier);
7249 unregister_sysctl_table(raid_table_header);
7250 remove_proc_entry("mdstat", NULL);
7251 for_each_mddev(mddev, tmp) {
7252 export_array(mddev);
7253 mddev->hold_active = 0;
7254 }
7255 }
7256
7257 subsys_initcall(md_init);
7258 module_exit(md_exit)
7259
7260 static int get_ro(char *buffer, struct kernel_param *kp)
7261 {
7262 return sprintf(buffer, "%d", start_readonly);
7263 }
7264 static int set_ro(const char *val, struct kernel_param *kp)
7265 {
7266 char *e;
7267 int num = simple_strtoul(val, &e, 10);
7268 if (*val && (*e == '\0' || *e == '\n')) {
7269 start_readonly = num;
7270 return 0;
7271 }
7272 return -EINVAL;
7273 }
7274
7275 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
7276 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
7277
7278 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
7279
7280 EXPORT_SYMBOL(register_md_personality);
7281 EXPORT_SYMBOL(unregister_md_personality);
7282 EXPORT_SYMBOL(md_error);
7283 EXPORT_SYMBOL(md_done_sync);
7284 EXPORT_SYMBOL(md_write_start);
7285 EXPORT_SYMBOL(md_write_end);
7286 EXPORT_SYMBOL(md_register_thread);
7287 EXPORT_SYMBOL(md_unregister_thread);
7288 EXPORT_SYMBOL(md_wakeup_thread);
7289 EXPORT_SYMBOL(md_check_recovery);
7290 MODULE_LICENSE("GPL");
7291 MODULE_DESCRIPTION("MD RAID framework");
7292 MODULE_ALIAS("md");
7293 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Ubuntu Jammy Linux kernel mirror