?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm/sysctl-2.6
[wandboard.git] / drivers / md / md.c
blob5f154ef1e4befeeef26358d9ae37e424c62cee47
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/hdreg.h>
43 #include <linux/proc_fs.h>
44 #include <linux/random.h>
45 #include <linux/reboot.h>
46 #include <linux/file.h>
47 #include <linux/delay.h>
48 #include <linux/raid/md_p.h>
49 #include <linux/raid/md_u.h>
50 #include "md.h"
51 #include "bitmap.h"
52
53 #define DEBUG 0
54 #define dprintk(x...) ((void)(DEBUG && printk(x)))
55
56
57 #ifndef MODULE
58 static void autostart_arrays(int part);
59 #endif
60
61 static LIST_HEAD(pers_list);
62 static DEFINE_SPINLOCK(pers_lock);
63
64 static void md_print_devices(void);
65
66 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
67
68 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
69
70 /*
71 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
72 * is 1000 KB/sec, so the extra system load does not show up that much.
73 * Increase it if you want to have more _guaranteed_ speed. Note that
74 * the RAID driver will use the maximum available bandwidth if the IO
75 * subsystem is idle. There is also an 'absolute maximum' reconstruction
76 * speed limit - in case reconstruction slows down your system despite
77 * idle IO detection.
78 *
79 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
80 * or /sys/block/mdX/md/sync_speed_{min,max}
81 */
82
83 static int sysctl_speed_limit_min = 1000;
84 static int sysctl_speed_limit_max = 200000;
85 static inline int speed_min(mddev_t *mddev)
86 {
87 return mddev->sync_speed_min ?
88 mddev->sync_speed_min : sysctl_speed_limit_min;
89 }
90
91 static inline int speed_max(mddev_t *mddev)
92 {
93 return mddev->sync_speed_max ?
94 mddev->sync_speed_max : sysctl_speed_limit_max;
95 }
96
97 static struct ctl_table_header *raid_table_header;
98
99 static ctl_table raid_table[] = {
100 {
101 .procname = "speed_limit_min",
102 .data = &sysctl_speed_limit_min,
103 .maxlen = sizeof(int),
104 .mode = S_IRUGO|S_IWUSR,
105 .proc_handler = proc_dointvec,
106 },
107 {
108 .procname = "speed_limit_max",
109 .data = &sysctl_speed_limit_max,
110 .maxlen = sizeof(int),
111 .mode = S_IRUGO|S_IWUSR,
112 .proc_handler = proc_dointvec,
113 },
114 { }
115 };
116
117 static ctl_table raid_dir_table[] = {
118 {
119 .procname = "raid",
120 .maxlen = 0,
121 .mode = S_IRUGO|S_IXUGO,
122 .child = raid_table,
123 },
124 { }
125 };
126
127 static ctl_table raid_root_table[] = {
128 {
129 .procname = "dev",
130 .maxlen = 0,
131 .mode = 0555,
132 .child = raid_dir_table,
133 },
134 { }
135 };
136
137 static const struct block_device_operations md_fops;
138
139 static int start_readonly;
140
141 /*
142 * We have a system wide 'event count' that is incremented
143 * on any 'interesting' event, and readers of /proc/mdstat
144 * can use 'poll' or 'select' to find out when the event
145 * count increases.
146 *
147 * Events are:
148 * start array, stop array, error, add device, remove device,
149 * start build, activate spare
150 */
151 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
152 static atomic_t md_event_count;
153 void md_new_event(mddev_t *mddev)
154 {
155 atomic_inc(&md_event_count);
156 wake_up(&md_event_waiters);
157 }
158 EXPORT_SYMBOL_GPL(md_new_event);
159
160 /* Alternate version that can be called from interrupts
161 * when calling sysfs_notify isn't needed.
162 */
163 static void md_new_event_inintr(mddev_t *mddev)
164 {
165 atomic_inc(&md_event_count);
166 wake_up(&md_event_waiters);
167 }
168
169 /*
170 * Enables to iterate over all existing md arrays
171 * all_mddevs_lock protects this list.
172 */
173 static LIST_HEAD(all_mddevs);
174 static DEFINE_SPINLOCK(all_mddevs_lock);
175
176
177 /*
178 * iterates through all used mddevs in the system.
179 * We take care to grab the all_mddevs_lock whenever navigating
180 * the list, and to always hold a refcount when unlocked.
181 * Any code which breaks out of this loop while own
182 * a reference to the current mddev and must mddev_put it.
183 */
184 #define for_each_mddev(mddev,tmp) \
185 \
186 for (({ spin_lock(&all_mddevs_lock); \
187 tmp = all_mddevs.next; \
188 mddev = NULL;}); \
189 ({ if (tmp != &all_mddevs) \
190 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
191 spin_unlock(&all_mddevs_lock); \
192 if (mddev) mddev_put(mddev); \
193 mddev = list_entry(tmp, mddev_t, all_mddevs); \
194 tmp != &all_mddevs;}); \
195 ({ spin_lock(&all_mddevs_lock); \
196 tmp = tmp->next;}) \
197 )
198
199
200 /* Rather than calling directly into the personality make_request function,
201 * IO requests come here first so that we can check if the device is
202 * being suspended pending a reconfiguration.
203 * We hold a refcount over the call to ->make_request. By the time that
204 * call has finished, the bio has been linked into some internal structure
205 * and so is visible to ->quiesce(), so we don't need the refcount any more.
206 */
207 static int md_make_request(struct request_queue *q, struct bio *bio)
208 {
209 mddev_t *mddev = q->queuedata;
210 int rv;
211 if (mddev == NULL || mddev->pers == NULL) {
212 bio_io_error(bio);
213 return 0;
214 }
215 rcu_read_lock();
216 if (mddev->suspended) {
217 DEFINE_WAIT(__wait);
218 for (;;) {
219 prepare_to_wait(&mddev->sb_wait, &__wait,
220 TASK_UNINTERRUPTIBLE);
221 if (!mddev->suspended)
222 break;
223 rcu_read_unlock();
224 schedule();
225 rcu_read_lock();
226 }
227 finish_wait(&mddev->sb_wait, &__wait);
228 }
229 atomic_inc(&mddev->active_io);
230 rcu_read_unlock();
231 rv = mddev->pers->make_request(q, bio);
232 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
233 wake_up(&mddev->sb_wait);
234
235 return rv;
236 }
237
238 static void mddev_suspend(mddev_t *mddev)
239 {
240 BUG_ON(mddev->suspended);
241 mddev->suspended = 1;
242 synchronize_rcu();
243 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
244 mddev->pers->quiesce(mddev, 1);
245 md_unregister_thread(mddev->thread);
246 mddev->thread = NULL;
247 /* we now know that no code is executing in the personality module,
248 * except possibly the tail end of a ->bi_end_io function, but that
249 * is certain to complete before the module has a chance to get
250 * unloaded
251 */
252 }
253
254 static void mddev_resume(mddev_t *mddev)
255 {
256 mddev->suspended = 0;
257 wake_up(&mddev->sb_wait);
258 mddev->pers->quiesce(mddev, 0);
259 }
260
261 int mddev_congested(mddev_t *mddev, int bits)
262 {
263 return mddev->suspended;
264 }
265 EXPORT_SYMBOL(mddev_congested);
266
267
268 static inline mddev_t *mddev_get(mddev_t *mddev)
269 {
270 atomic_inc(&mddev->active);
271 return mddev;
272 }
273
274 static void mddev_delayed_delete(struct work_struct *ws);
275
276 static void mddev_put(mddev_t *mddev)
277 {
278 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
279 return;
280 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
281 !mddev->hold_active) {
282 list_del(&mddev->all_mddevs);
283 if (mddev->gendisk) {
284 /* we did a probe so need to clean up.
285 * Call schedule_work inside the spinlock
286 * so that flush_scheduled_work() after
287 * mddev_find will succeed in waiting for the
288 * work to be done.
289 */
290 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
291 schedule_work(&mddev->del_work);
292 } else
293 kfree(mddev);
294 }
295 spin_unlock(&all_mddevs_lock);
296 }
297
298 static mddev_t * mddev_find(dev_t unit)
299 {
300 mddev_t *mddev, *new = NULL;
301
302 retry:
303 spin_lock(&all_mddevs_lock);
304
305 if (unit) {
306 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
307 if (mddev->unit == unit) {
308 mddev_get(mddev);
309 spin_unlock(&all_mddevs_lock);
310 kfree(new);
311 return mddev;
312 }
313
314 if (new) {
315 list_add(&new->all_mddevs, &all_mddevs);
316 spin_unlock(&all_mddevs_lock);
317 new->hold_active = UNTIL_IOCTL;
318 return new;
319 }
320 } else if (new) {
321 /* find an unused unit number */
322 static int next_minor = 512;
323 int start = next_minor;
324 int is_free = 0;
325 int dev = 0;
326 while (!is_free) {
327 dev = MKDEV(MD_MAJOR, next_minor);
328 next_minor++;
329 if (next_minor > MINORMASK)
330 next_minor = 0;
331 if (next_minor == start) {
332 /* Oh dear, all in use. */
333 spin_unlock(&all_mddevs_lock);
334 kfree(new);
335 return NULL;
336 }
337
338 is_free = 1;
339 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
340 if (mddev->unit == dev) {
341 is_free = 0;
342 break;
343 }
344 }
345 new->unit = dev;
346 new->md_minor = MINOR(dev);
347 new->hold_active = UNTIL_STOP;
348 list_add(&new->all_mddevs, &all_mddevs);
349 spin_unlock(&all_mddevs_lock);
350 return new;
351 }
352 spin_unlock(&all_mddevs_lock);
353
354 new = kzalloc(sizeof(*new), GFP_KERNEL);
355 if (!new)
356 return NULL;
357
358 new->unit = unit;
359 if (MAJOR(unit) == MD_MAJOR)
360 new->md_minor = MINOR(unit);
361 else
362 new->md_minor = MINOR(unit) >> MdpMinorShift;
363
364 mutex_init(&new->open_mutex);
365 mutex_init(&new->reconfig_mutex);
366 INIT_LIST_HEAD(&new->disks);
367 INIT_LIST_HEAD(&new->all_mddevs);
368 init_timer(&new->safemode_timer);
369 atomic_set(&new->active, 1);
370 atomic_set(&new->openers, 0);
371 atomic_set(&new->active_io, 0);
372 spin_lock_init(&new->write_lock);
373 init_waitqueue_head(&new->sb_wait);
374 init_waitqueue_head(&new->recovery_wait);
375 new->reshape_position = MaxSector;
376 new->resync_min = 0;
377 new->resync_max = MaxSector;
378 new->level = LEVEL_NONE;
379
380 goto retry;
381 }
382
383 static inline int mddev_lock(mddev_t * mddev)
384 {
385 return mutex_lock_interruptible(&mddev->reconfig_mutex);
386 }
387
388 static inline int mddev_is_locked(mddev_t *mddev)
389 {
390 return mutex_is_locked(&mddev->reconfig_mutex);
391 }
392
393 static inline int mddev_trylock(mddev_t * mddev)
394 {
395 return mutex_trylock(&mddev->reconfig_mutex);
396 }
397
398 static inline void mddev_unlock(mddev_t * mddev)
399 {
400 mutex_unlock(&mddev->reconfig_mutex);
401
402 md_wakeup_thread(mddev->thread);
403 }
404
405 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
406 {
407 mdk_rdev_t *rdev;
408
409 list_for_each_entry(rdev, &mddev->disks, same_set)
410 if (rdev->desc_nr == nr)
411 return rdev;
412
413 return NULL;
414 }
415
416 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
417 {
418 mdk_rdev_t *rdev;
419
420 list_for_each_entry(rdev, &mddev->disks, same_set)
421 if (rdev->bdev->bd_dev == dev)
422 return rdev;
423
424 return NULL;
425 }
426
427 static struct mdk_personality *find_pers(int level, char *clevel)
428 {
429 struct mdk_personality *pers;
430 list_for_each_entry(pers, &pers_list, list) {
431 if (level != LEVEL_NONE && pers->level == level)
432 return pers;
433 if (strcmp(pers->name, clevel)==0)
434 return pers;
435 }
436 return NULL;
437 }
438
439 /* return the offset of the super block in 512byte sectors */
440 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
441 {
442 sector_t num_sectors = bdev->bd_inode->i_size / 512;
443 return MD_NEW_SIZE_SECTORS(num_sectors);
444 }
445
446 static int alloc_disk_sb(mdk_rdev_t * rdev)
447 {
448 if (rdev->sb_page)
449 MD_BUG();
450
451 rdev->sb_page = alloc_page(GFP_KERNEL);
452 if (!rdev->sb_page) {
453 printk(KERN_ALERT "md: out of memory.\n");
454 return -ENOMEM;
455 }
456
457 return 0;
458 }
459
460 static void free_disk_sb(mdk_rdev_t * rdev)
461 {
462 if (rdev->sb_page) {
463 put_page(rdev->sb_page);
464 rdev->sb_loaded = 0;
465 rdev->sb_page = NULL;
466 rdev->sb_start = 0;
467 rdev->sectors = 0;
468 }
469 }
470
471
472 static void super_written(struct bio *bio, int error)
473 {
474 mdk_rdev_t *rdev = bio->bi_private;
475 mddev_t *mddev = rdev->mddev;
476
477 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
478 printk("md: super_written gets error=%d, uptodate=%d\n",
479 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
480 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
481 md_error(mddev, rdev);
482 }
483
484 if (atomic_dec_and_test(&mddev->pending_writes))
485 wake_up(&mddev->sb_wait);
486 bio_put(bio);
487 }
488
489 static void super_written_barrier(struct bio *bio, int error)
490 {
491 struct bio *bio2 = bio->bi_private;
492 mdk_rdev_t *rdev = bio2->bi_private;
493 mddev_t *mddev = rdev->mddev;
494
495 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
496 error == -EOPNOTSUPP) {
497 unsigned long flags;
498 /* barriers don't appear to be supported :-( */
499 set_bit(BarriersNotsupp, &rdev->flags);
500 mddev->barriers_work = 0;
501 spin_lock_irqsave(&mddev->write_lock, flags);
502 bio2->bi_next = mddev->biolist;
503 mddev->biolist = bio2;
504 spin_unlock_irqrestore(&mddev->write_lock, flags);
505 wake_up(&mddev->sb_wait);
506 bio_put(bio);
507 } else {
508 bio_put(bio2);
509 bio->bi_private = rdev;
510 super_written(bio, error);
511 }
512 }
513
514 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
515 sector_t sector, int size, struct page *page)
516 {
517 /* write first size bytes of page to sector of rdev
518 * Increment mddev->pending_writes before returning
519 * and decrement it on completion, waking up sb_wait
520 * if zero is reached.
521 * If an error occurred, call md_error
522 *
523 * As we might need to resubmit the request if BIO_RW_BARRIER
524 * causes ENOTSUPP, we allocate a spare bio...
525 */
526 struct bio *bio = bio_alloc(GFP_NOIO, 1);
527 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG);
528
529 bio->bi_bdev = rdev->bdev;
530 bio->bi_sector = sector;
531 bio_add_page(bio, page, size, 0);
532 bio->bi_private = rdev;
533 bio->bi_end_io = super_written;
534 bio->bi_rw = rw;
535
536 atomic_inc(&mddev->pending_writes);
537 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
538 struct bio *rbio;
539 rw |= (1<<BIO_RW_BARRIER);
540 rbio = bio_clone(bio, GFP_NOIO);
541 rbio->bi_private = bio;
542 rbio->bi_end_io = super_written_barrier;
543 submit_bio(rw, rbio);
544 } else
545 submit_bio(rw, bio);
546 }
547
548 void md_super_wait(mddev_t *mddev)
549 {
550 /* wait for all superblock writes that were scheduled to complete.
551 * if any had to be retried (due to BARRIER problems), retry them
552 */
553 DEFINE_WAIT(wq);
554 for(;;) {
555 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
556 if (atomic_read(&mddev->pending_writes)==0)
557 break;
558 while (mddev->biolist) {
559 struct bio *bio;
560 spin_lock_irq(&mddev->write_lock);
561 bio = mddev->biolist;
562 mddev->biolist = bio->bi_next ;
563 bio->bi_next = NULL;
564 spin_unlock_irq(&mddev->write_lock);
565 submit_bio(bio->bi_rw, bio);
566 }
567 schedule();
568 }
569 finish_wait(&mddev->sb_wait, &wq);
570 }
571
572 static void bi_complete(struct bio *bio, int error)
573 {
574 complete((struct completion*)bio->bi_private);
575 }
576
577 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
578 struct page *page, int rw)
579 {
580 struct bio *bio = bio_alloc(GFP_NOIO, 1);
581 struct completion event;
582 int ret;
583
584 rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
585
586 bio->bi_bdev = bdev;
587 bio->bi_sector = sector;
588 bio_add_page(bio, page, size, 0);
589 init_completion(&event);
590 bio->bi_private = &event;
591 bio->bi_end_io = bi_complete;
592 submit_bio(rw, bio);
593 wait_for_completion(&event);
594
595 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
596 bio_put(bio);
597 return ret;
598 }
599 EXPORT_SYMBOL_GPL(sync_page_io);
600
601 static int read_disk_sb(mdk_rdev_t * rdev, int size)
602 {
603 char b[BDEVNAME_SIZE];
604 if (!rdev->sb_page) {
605 MD_BUG();
606 return -EINVAL;
607 }
608 if (rdev->sb_loaded)
609 return 0;
610
611
612 if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
613 goto fail;
614 rdev->sb_loaded = 1;
615 return 0;
616
617 fail:
618 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
619 bdevname(rdev->bdev,b));
620 return -EINVAL;
621 }
622
623 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
624 {
625 return sb1->set_uuid0 == sb2->set_uuid0 &&
626 sb1->set_uuid1 == sb2->set_uuid1 &&
627 sb1->set_uuid2 == sb2->set_uuid2 &&
628 sb1->set_uuid3 == sb2->set_uuid3;
629 }
630
631 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
632 {
633 int ret;
634 mdp_super_t *tmp1, *tmp2;
635
636 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
637 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
638
639 if (!tmp1 || !tmp2) {
640 ret = 0;
641 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
642 goto abort;
643 }
644
645 *tmp1 = *sb1;
646 *tmp2 = *sb2;
647
648 /*
649 * nr_disks is not constant
650 */
651 tmp1->nr_disks = 0;
652 tmp2->nr_disks = 0;
653
654 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
655 abort:
656 kfree(tmp1);
657 kfree(tmp2);
658 return ret;
659 }
660
661
662 static u32 md_csum_fold(u32 csum)
663 {
664 csum = (csum & 0xffff) + (csum >> 16);
665 return (csum & 0xffff) + (csum >> 16);
666 }
667
668 static unsigned int calc_sb_csum(mdp_super_t * sb)
669 {
670 u64 newcsum = 0;
671 u32 *sb32 = (u32*)sb;
672 int i;
673 unsigned int disk_csum, csum;
674
675 disk_csum = sb->sb_csum;
676 sb->sb_csum = 0;
677
678 for (i = 0; i < MD_SB_BYTES/4 ; i++)
679 newcsum += sb32[i];
680 csum = (newcsum & 0xffffffff) + (newcsum>>32);
681
682
683 #ifdef CONFIG_ALPHA
684 /* This used to use csum_partial, which was wrong for several
685 * reasons including that different results are returned on
686 * different architectures. It isn't critical that we get exactly
687 * the same return value as before (we always csum_fold before
688 * testing, and that removes any differences). However as we
689 * know that csum_partial always returned a 16bit value on
690 * alphas, do a fold to maximise conformity to previous behaviour.
691 */
692 sb->sb_csum = md_csum_fold(disk_csum);
693 #else
694 sb->sb_csum = disk_csum;
695 #endif
696 return csum;
697 }
698
699
700 /*
701 * Handle superblock details.
702 * We want to be able to handle multiple superblock formats
703 * so we have a common interface to them all, and an array of
704 * different handlers.
705 * We rely on user-space to write the initial superblock, and support
706 * reading and updating of superblocks.
707 * Interface methods are:
708 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
709 * loads and validates a superblock on dev.
710 * if refdev != NULL, compare superblocks on both devices
711 * Return:
712 * 0 - dev has a superblock that is compatible with refdev
713 * 1 - dev has a superblock that is compatible and newer than refdev
714 * so dev should be used as the refdev in future
715 * -EINVAL superblock incompatible or invalid
716 * -othererror e.g. -EIO
717 *
718 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
719 * Verify that dev is acceptable into mddev.
720 * The first time, mddev->raid_disks will be 0, and data from
721 * dev should be merged in. Subsequent calls check that dev
722 * is new enough. Return 0 or -EINVAL
723 *
724 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
725 * Update the superblock for rdev with data in mddev
726 * This does not write to disc.
727 *
728 */
729
730 struct super_type {
731 char *name;
732 struct module *owner;
733 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
734 int minor_version);
735 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
736 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
737 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
738 sector_t num_sectors);
739 };
740
741 /*
742 * Check that the given mddev has no bitmap.
743 *
744 * This function is called from the run method of all personalities that do not
745 * support bitmaps. It prints an error message and returns non-zero if mddev
746 * has a bitmap. Otherwise, it returns 0.
747 *
748 */
749 int md_check_no_bitmap(mddev_t *mddev)
750 {
751 if (!mddev->bitmap_file && !mddev->bitmap_offset)
752 return 0;
753 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
754 mdname(mddev), mddev->pers->name);
755 return 1;
756 }
757 EXPORT_SYMBOL(md_check_no_bitmap);
758
759 /*
760 * load_super for 0.90.0
761 */
762 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
763 {
764 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
765 mdp_super_t *sb;
766 int ret;
767
768 /*
769 * Calculate the position of the superblock (512byte sectors),
770 * it's at the end of the disk.
771 *
772 * It also happens to be a multiple of 4Kb.
773 */
774 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
775
776 ret = read_disk_sb(rdev, MD_SB_BYTES);
777 if (ret) return ret;
778
779 ret = -EINVAL;
780
781 bdevname(rdev->bdev, b);
782 sb = (mdp_super_t*)page_address(rdev->sb_page);
783
784 if (sb->md_magic != MD_SB_MAGIC) {
785 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
786 b);
787 goto abort;
788 }
789
790 if (sb->major_version != 0 ||
791 sb->minor_version < 90 ||
792 sb->minor_version > 91) {
793 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
794 sb->major_version, sb->minor_version,
795 b);
796 goto abort;
797 }
798
799 if (sb->raid_disks <= 0)
800 goto abort;
801
802 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
803 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
804 b);
805 goto abort;
806 }
807
808 rdev->preferred_minor = sb->md_minor;
809 rdev->data_offset = 0;
810 rdev->sb_size = MD_SB_BYTES;
811
812 if (sb->level == LEVEL_MULTIPATH)
813 rdev->desc_nr = -1;
814 else
815 rdev->desc_nr = sb->this_disk.number;
816
817 if (!refdev) {
818 ret = 1;
819 } else {
820 __u64 ev1, ev2;
821 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
822 if (!uuid_equal(refsb, sb)) {
823 printk(KERN_WARNING "md: %s has different UUID to %s\n",
824 b, bdevname(refdev->bdev,b2));
825 goto abort;
826 }
827 if (!sb_equal(refsb, sb)) {
828 printk(KERN_WARNING "md: %s has same UUID"
829 " but different superblock to %s\n",
830 b, bdevname(refdev->bdev, b2));
831 goto abort;
832 }
833 ev1 = md_event(sb);
834 ev2 = md_event(refsb);
835 if (ev1 > ev2)
836 ret = 1;
837 else
838 ret = 0;
839 }
840 rdev->sectors = rdev->sb_start;
841
842 if (rdev->sectors < sb->size * 2 && sb->level > 1)
843 /* "this cannot possibly happen" ... */
844 ret = -EINVAL;
845
846 abort:
847 return ret;
848 }
849
850 /*
851 * validate_super for 0.90.0
852 */
853 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
854 {
855 mdp_disk_t *desc;
856 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
857 __u64 ev1 = md_event(sb);
858
859 rdev->raid_disk = -1;
860 clear_bit(Faulty, &rdev->flags);
861 clear_bit(In_sync, &rdev->flags);
862 clear_bit(WriteMostly, &rdev->flags);
863 clear_bit(BarriersNotsupp, &rdev->flags);
864
865 if (mddev->raid_disks == 0) {
866 mddev->major_version = 0;
867 mddev->minor_version = sb->minor_version;
868 mddev->patch_version = sb->patch_version;
869 mddev->external = 0;
870 mddev->chunk_sectors = sb->chunk_size >> 9;
871 mddev->ctime = sb->ctime;
872 mddev->utime = sb->utime;
873 mddev->level = sb->level;
874 mddev->clevel[0] = 0;
875 mddev->layout = sb->layout;
876 mddev->raid_disks = sb->raid_disks;
877 mddev->dev_sectors = sb->size * 2;
878 mddev->events = ev1;
879 mddev->bitmap_offset = 0;
880 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
881
882 if (mddev->minor_version >= 91) {
883 mddev->reshape_position = sb->reshape_position;
884 mddev->delta_disks = sb->delta_disks;
885 mddev->new_level = sb->new_level;
886 mddev->new_layout = sb->new_layout;
887 mddev->new_chunk_sectors = sb->new_chunk >> 9;
888 } else {
889 mddev->reshape_position = MaxSector;
890 mddev->delta_disks = 0;
891 mddev->new_level = mddev->level;
892 mddev->new_layout = mddev->layout;
893 mddev->new_chunk_sectors = mddev->chunk_sectors;
894 }
895
896 if (sb->state & (1<<MD_SB_CLEAN))
897 mddev->recovery_cp = MaxSector;
898 else {
899 if (sb->events_hi == sb->cp_events_hi &&
900 sb->events_lo == sb->cp_events_lo) {
901 mddev->recovery_cp = sb->recovery_cp;
902 } else
903 mddev->recovery_cp = 0;
904 }
905
906 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
907 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
908 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
909 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
910
911 mddev->max_disks = MD_SB_DISKS;
912
913 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
914 mddev->bitmap_file == NULL)
915 mddev->bitmap_offset = mddev->default_bitmap_offset;
916
917 } else if (mddev->pers == NULL) {
918 /* Insist on good event counter while assembling */
919 ++ev1;
920 if (ev1 < mddev->events)
921 return -EINVAL;
922 } else if (mddev->bitmap) {
923 /* if adding to array with a bitmap, then we can accept an
924 * older device ... but not too old.
925 */
926 if (ev1 < mddev->bitmap->events_cleared)
927 return 0;
928 } else {
929 if (ev1 < mddev->events)
930 /* just a hot-add of a new device, leave raid_disk at -1 */
931 return 0;
932 }
933
934 if (mddev->level != LEVEL_MULTIPATH) {
935 desc = sb->disks + rdev->desc_nr;
936
937 if (desc->state & (1<<MD_DISK_FAULTY))
938 set_bit(Faulty, &rdev->flags);
939 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
940 desc->raid_disk < mddev->raid_disks */) {
941 set_bit(In_sync, &rdev->flags);
942 rdev->raid_disk = desc->raid_disk;
943 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
944 /* active but not in sync implies recovery up to
945 * reshape position. We don't know exactly where
946 * that is, so set to zero for now */
947 if (mddev->minor_version >= 91) {
948 rdev->recovery_offset = 0;
949 rdev->raid_disk = desc->raid_disk;
950 }
951 }
952 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
953 set_bit(WriteMostly, &rdev->flags);
954 } else /* MULTIPATH are always insync */
955 set_bit(In_sync, &rdev->flags);
956 return 0;
957 }
958
959 /*
960 * sync_super for 0.90.0
961 */
962 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
963 {
964 mdp_super_t *sb;
965 mdk_rdev_t *rdev2;
966 int next_spare = mddev->raid_disks;
967
968
969 /* make rdev->sb match mddev data..
970 *
971 * 1/ zero out disks
972 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
973 * 3/ any empty disks < next_spare become removed
974 *
975 * disks[0] gets initialised to REMOVED because
976 * we cannot be sure from other fields if it has
977 * been initialised or not.
978 */
979 int i;
980 int active=0, working=0,failed=0,spare=0,nr_disks=0;
981
982 rdev->sb_size = MD_SB_BYTES;
983
984 sb = (mdp_super_t*)page_address(rdev->sb_page);
985
986 memset(sb, 0, sizeof(*sb));
987
988 sb->md_magic = MD_SB_MAGIC;
989 sb->major_version = mddev->major_version;
990 sb->patch_version = mddev->patch_version;
991 sb->gvalid_words = 0; /* ignored */
992 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
993 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
994 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
995 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
996
997 sb->ctime = mddev->ctime;
998 sb->level = mddev->level;
999 sb->size = mddev->dev_sectors / 2;
1000 sb->raid_disks = mddev->raid_disks;
1001 sb->md_minor = mddev->md_minor;
1002 sb->not_persistent = 0;
1003 sb->utime = mddev->utime;
1004 sb->state = 0;
1005 sb->events_hi = (mddev->events>>32);
1006 sb->events_lo = (u32)mddev->events;
1007
1008 if (mddev->reshape_position == MaxSector)
1009 sb->minor_version = 90;
1010 else {
1011 sb->minor_version = 91;
1012 sb->reshape_position = mddev->reshape_position;
1013 sb->new_level = mddev->new_level;
1014 sb->delta_disks = mddev->delta_disks;
1015 sb->new_layout = mddev->new_layout;
1016 sb->new_chunk = mddev->new_chunk_sectors << 9;
1017 }
1018 mddev->minor_version = sb->minor_version;
1019 if (mddev->in_sync)
1020 {
1021 sb->recovery_cp = mddev->recovery_cp;
1022 sb->cp_events_hi = (mddev->events>>32);
1023 sb->cp_events_lo = (u32)mddev->events;
1024 if (mddev->recovery_cp == MaxSector)
1025 sb->state = (1<< MD_SB_CLEAN);
1026 } else
1027 sb->recovery_cp = 0;
1028
1029 sb->layout = mddev->layout;
1030 sb->chunk_size = mddev->chunk_sectors << 9;
1031
1032 if (mddev->bitmap && mddev->bitmap_file == NULL)
1033 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1034
1035 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1036 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1037 mdp_disk_t *d;
1038 int desc_nr;
1039 int is_active = test_bit(In_sync, &rdev2->flags);
1040
1041 if (rdev2->raid_disk >= 0 &&
1042 sb->minor_version >= 91)
1043 /* we have nowhere to store the recovery_offset,
1044 * but if it is not below the reshape_position,
1045 * we can piggy-back on that.
1046 */
1047 is_active = 1;
1048 if (rdev2->raid_disk < 0 ||
1049 test_bit(Faulty, &rdev2->flags))
1050 is_active = 0;
1051 if (is_active)
1052 desc_nr = rdev2->raid_disk;
1053 else
1054 desc_nr = next_spare++;
1055 rdev2->desc_nr = desc_nr;
1056 d = &sb->disks[rdev2->desc_nr];
1057 nr_disks++;
1058 d->number = rdev2->desc_nr;
1059 d->major = MAJOR(rdev2->bdev->bd_dev);
1060 d->minor = MINOR(rdev2->bdev->bd_dev);
1061 if (is_active)
1062 d->raid_disk = rdev2->raid_disk;
1063 else
1064 d->raid_disk = rdev2->desc_nr; /* compatibility */
1065 if (test_bit(Faulty, &rdev2->flags))
1066 d->state = (1<<MD_DISK_FAULTY);
1067 else if (is_active) {
1068 d->state = (1<<MD_DISK_ACTIVE);
1069 if (test_bit(In_sync, &rdev2->flags))
1070 d->state |= (1<<MD_DISK_SYNC);
1071 active++;
1072 working++;
1073 } else {
1074 d->state = 0;
1075 spare++;
1076 working++;
1077 }
1078 if (test_bit(WriteMostly, &rdev2->flags))
1079 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1080 }
1081 /* now set the "removed" and "faulty" bits on any missing devices */
1082 for (i=0 ; i < mddev->raid_disks ; i++) {
1083 mdp_disk_t *d = &sb->disks[i];
1084 if (d->state == 0 && d->number == 0) {
1085 d->number = i;
1086 d->raid_disk = i;
1087 d->state = (1<<MD_DISK_REMOVED);
1088 d->state |= (1<<MD_DISK_FAULTY);
1089 failed++;
1090 }
1091 }
1092 sb->nr_disks = nr_disks;
1093 sb->active_disks = active;
1094 sb->working_disks = working;
1095 sb->failed_disks = failed;
1096 sb->spare_disks = spare;
1097
1098 sb->this_disk = sb->disks[rdev->desc_nr];
1099 sb->sb_csum = calc_sb_csum(sb);
1100 }
1101
1102 /*
1103 * rdev_size_change for 0.90.0
1104 */
1105 static unsigned long long
1106 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1107 {
1108 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1109 return 0; /* component must fit device */
1110 if (rdev->mddev->bitmap_offset)
1111 return 0; /* can't move bitmap */
1112 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1113 if (!num_sectors || num_sectors > rdev->sb_start)
1114 num_sectors = rdev->sb_start;
1115 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1116 rdev->sb_page);
1117 md_super_wait(rdev->mddev);
1118 return num_sectors / 2; /* kB for sysfs */
1119 }
1120
1121
1122 /*
1123 * version 1 superblock
1124 */
1125
1126 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1127 {
1128 __le32 disk_csum;
1129 u32 csum;
1130 unsigned long long newcsum;
1131 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1132 __le32 *isuper = (__le32*)sb;
1133 int i;
1134
1135 disk_csum = sb->sb_csum;
1136 sb->sb_csum = 0;
1137 newcsum = 0;
1138 for (i=0; size>=4; size -= 4 )
1139 newcsum += le32_to_cpu(*isuper++);
1140
1141 if (size == 2)
1142 newcsum += le16_to_cpu(*(__le16*) isuper);
1143
1144 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1145 sb->sb_csum = disk_csum;
1146 return cpu_to_le32(csum);
1147 }
1148
1149 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1150 {
1151 struct mdp_superblock_1 *sb;
1152 int ret;
1153 sector_t sb_start;
1154 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1155 int bmask;
1156
1157 /*
1158 * Calculate the position of the superblock in 512byte sectors.
1159 * It is always aligned to a 4K boundary and
1160 * depeding on minor_version, it can be:
1161 * 0: At least 8K, but less than 12K, from end of device
1162 * 1: At start of device
1163 * 2: 4K from start of device.
1164 */
1165 switch(minor_version) {
1166 case 0:
1167 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1168 sb_start -= 8*2;
1169 sb_start &= ~(sector_t)(4*2-1);
1170 break;
1171 case 1:
1172 sb_start = 0;
1173 break;
1174 case 2:
1175 sb_start = 8;
1176 break;
1177 default:
1178 return -EINVAL;
1179 }
1180 rdev->sb_start = sb_start;
1181
1182 /* superblock is rarely larger than 1K, but it can be larger,
1183 * and it is safe to read 4k, so we do that
1184 */
1185 ret = read_disk_sb(rdev, 4096);
1186 if (ret) return ret;
1187
1188
1189 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1190
1191 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1192 sb->major_version != cpu_to_le32(1) ||
1193 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1194 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1195 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1196 return -EINVAL;
1197
1198 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1199 printk("md: invalid superblock checksum on %s\n",
1200 bdevname(rdev->bdev,b));
1201 return -EINVAL;
1202 }
1203 if (le64_to_cpu(sb->data_size) < 10) {
1204 printk("md: data_size too small on %s\n",
1205 bdevname(rdev->bdev,b));
1206 return -EINVAL;
1207 }
1208
1209 rdev->preferred_minor = 0xffff;
1210 rdev->data_offset = le64_to_cpu(sb->data_offset);
1211 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1212
1213 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1214 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1215 if (rdev->sb_size & bmask)
1216 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1217
1218 if (minor_version
1219 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1220 return -EINVAL;
1221
1222 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1223 rdev->desc_nr = -1;
1224 else
1225 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1226
1227 if (!refdev) {
1228 ret = 1;
1229 } else {
1230 __u64 ev1, ev2;
1231 struct mdp_superblock_1 *refsb =
1232 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1233
1234 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1235 sb->level != refsb->level ||
1236 sb->layout != refsb->layout ||
1237 sb->chunksize != refsb->chunksize) {
1238 printk(KERN_WARNING "md: %s has strangely different"
1239 " superblock to %s\n",
1240 bdevname(rdev->bdev,b),
1241 bdevname(refdev->bdev,b2));
1242 return -EINVAL;
1243 }
1244 ev1 = le64_to_cpu(sb->events);
1245 ev2 = le64_to_cpu(refsb->events);
1246
1247 if (ev1 > ev2)
1248 ret = 1;
1249 else
1250 ret = 0;
1251 }
1252 if (minor_version)
1253 rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
1254 le64_to_cpu(sb->data_offset);
1255 else
1256 rdev->sectors = rdev->sb_start;
1257 if (rdev->sectors < le64_to_cpu(sb->data_size))
1258 return -EINVAL;
1259 rdev->sectors = le64_to_cpu(sb->data_size);
1260 if (le64_to_cpu(sb->size) > rdev->sectors)
1261 return -EINVAL;
1262 return ret;
1263 }
1264
1265 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1266 {
1267 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1268 __u64 ev1 = le64_to_cpu(sb->events);
1269
1270 rdev->raid_disk = -1;
1271 clear_bit(Faulty, &rdev->flags);
1272 clear_bit(In_sync, &rdev->flags);
1273 clear_bit(WriteMostly, &rdev->flags);
1274 clear_bit(BarriersNotsupp, &rdev->flags);
1275
1276 if (mddev->raid_disks == 0) {
1277 mddev->major_version = 1;
1278 mddev->patch_version = 0;
1279 mddev->external = 0;
1280 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1281 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1282 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1283 mddev->level = le32_to_cpu(sb->level);
1284 mddev->clevel[0] = 0;
1285 mddev->layout = le32_to_cpu(sb->layout);
1286 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1287 mddev->dev_sectors = le64_to_cpu(sb->size);
1288 mddev->events = ev1;
1289 mddev->bitmap_offset = 0;
1290 mddev->default_bitmap_offset = 1024 >> 9;
1291
1292 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1293 memcpy(mddev->uuid, sb->set_uuid, 16);
1294
1295 mddev->max_disks = (4096-256)/2;
1296
1297 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1298 mddev->bitmap_file == NULL )
1299 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1300
1301 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1302 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1303 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1304 mddev->new_level = le32_to_cpu(sb->new_level);
1305 mddev->new_layout = le32_to_cpu(sb->new_layout);
1306 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1307 } else {
1308 mddev->reshape_position = MaxSector;
1309 mddev->delta_disks = 0;
1310 mddev->new_level = mddev->level;
1311 mddev->new_layout = mddev->layout;
1312 mddev->new_chunk_sectors = mddev->chunk_sectors;
1313 }
1314
1315 } else if (mddev->pers == NULL) {
1316 /* Insist of good event counter while assembling */
1317 ++ev1;
1318 if (ev1 < mddev->events)
1319 return -EINVAL;
1320 } else if (mddev->bitmap) {
1321 /* If adding to array with a bitmap, then we can accept an
1322 * older device, but not too old.
1323 */
1324 if (ev1 < mddev->bitmap->events_cleared)
1325 return 0;
1326 } else {
1327 if (ev1 < mddev->events)
1328 /* just a hot-add of a new device, leave raid_disk at -1 */
1329 return 0;
1330 }
1331 if (mddev->level != LEVEL_MULTIPATH) {
1332 int role;
1333 if (rdev->desc_nr < 0 ||
1334 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1335 role = 0xffff;
1336 rdev->desc_nr = -1;
1337 } else
1338 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1339 switch(role) {
1340 case 0xffff: /* spare */
1341 break;
1342 case 0xfffe: /* faulty */
1343 set_bit(Faulty, &rdev->flags);
1344 break;
1345 default:
1346 if ((le32_to_cpu(sb->feature_map) &
1347 MD_FEATURE_RECOVERY_OFFSET))
1348 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1349 else
1350 set_bit(In_sync, &rdev->flags);
1351 rdev->raid_disk = role;
1352 break;
1353 }
1354 if (sb->devflags & WriteMostly1)
1355 set_bit(WriteMostly, &rdev->flags);
1356 } else /* MULTIPATH are always insync */
1357 set_bit(In_sync, &rdev->flags);
1358
1359 return 0;
1360 }
1361
1362 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1363 {
1364 struct mdp_superblock_1 *sb;
1365 mdk_rdev_t *rdev2;
1366 int max_dev, i;
1367 /* make rdev->sb match mddev and rdev data. */
1368
1369 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1370
1371 sb->feature_map = 0;
1372 sb->pad0 = 0;
1373 sb->recovery_offset = cpu_to_le64(0);
1374 memset(sb->pad1, 0, sizeof(sb->pad1));
1375 memset(sb->pad2, 0, sizeof(sb->pad2));
1376 memset(sb->pad3, 0, sizeof(sb->pad3));
1377
1378 sb->utime = cpu_to_le64((__u64)mddev->utime);
1379 sb->events = cpu_to_le64(mddev->events);
1380 if (mddev->in_sync)
1381 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1382 else
1383 sb->resync_offset = cpu_to_le64(0);
1384
1385 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1386
1387 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1388 sb->size = cpu_to_le64(mddev->dev_sectors);
1389 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1390 sb->level = cpu_to_le32(mddev->level);
1391 sb->layout = cpu_to_le32(mddev->layout);
1392
1393 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1394 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1395 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1396 }
1397
1398 if (rdev->raid_disk >= 0 &&
1399 !test_bit(In_sync, &rdev->flags)) {
1400 if (rdev->recovery_offset > 0) {
1401 sb->feature_map |=
1402 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1403 sb->recovery_offset =
1404 cpu_to_le64(rdev->recovery_offset);
1405 }
1406 }
1407
1408 if (mddev->reshape_position != MaxSector) {
1409 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1410 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1411 sb->new_layout = cpu_to_le32(mddev->new_layout);
1412 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1413 sb->new_level = cpu_to_le32(mddev->new_level);
1414 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1415 }
1416
1417 max_dev = 0;
1418 list_for_each_entry(rdev2, &mddev->disks, same_set)
1419 if (rdev2->desc_nr+1 > max_dev)
1420 max_dev = rdev2->desc_nr+1;
1421
1422 if (max_dev > le32_to_cpu(sb->max_dev)) {
1423 int bmask;
1424 sb->max_dev = cpu_to_le32(max_dev);
1425 rdev->sb_size = max_dev * 2 + 256;
1426 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1427 if (rdev->sb_size & bmask)
1428 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1429 }
1430 for (i=0; i<max_dev;i++)
1431 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1432
1433 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1434 i = rdev2->desc_nr;
1435 if (test_bit(Faulty, &rdev2->flags))
1436 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1437 else if (test_bit(In_sync, &rdev2->flags))
1438 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1439 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1440 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1441 else
1442 sb->dev_roles[i] = cpu_to_le16(0xffff);
1443 }
1444
1445 sb->sb_csum = calc_sb_1_csum(sb);
1446 }
1447
1448 static unsigned long long
1449 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1450 {
1451 struct mdp_superblock_1 *sb;
1452 sector_t max_sectors;
1453 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1454 return 0; /* component must fit device */
1455 if (rdev->sb_start < rdev->data_offset) {
1456 /* minor versions 1 and 2; superblock before data */
1457 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1458 max_sectors -= rdev->data_offset;
1459 if (!num_sectors || num_sectors > max_sectors)
1460 num_sectors = max_sectors;
1461 } else if (rdev->mddev->bitmap_offset) {
1462 /* minor version 0 with bitmap we can't move */
1463 return 0;
1464 } else {
1465 /* minor version 0; superblock after data */
1466 sector_t sb_start;
1467 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1468 sb_start &= ~(sector_t)(4*2 - 1);
1469 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1470 if (!num_sectors || num_sectors > max_sectors)
1471 num_sectors = max_sectors;
1472 rdev->sb_start = sb_start;
1473 }
1474 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1475 sb->data_size = cpu_to_le64(num_sectors);
1476 sb->super_offset = rdev->sb_start;
1477 sb->sb_csum = calc_sb_1_csum(sb);
1478 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1479 rdev->sb_page);
1480 md_super_wait(rdev->mddev);
1481 return num_sectors / 2; /* kB for sysfs */
1482 }
1483
1484 static struct super_type super_types[] = {
1485 [0] = {
1486 .name = "0.90.0",
1487 .owner = THIS_MODULE,
1488 .load_super = super_90_load,
1489 .validate_super = super_90_validate,
1490 .sync_super = super_90_sync,
1491 .rdev_size_change = super_90_rdev_size_change,
1492 },
1493 [1] = {
1494 .name = "md-1",
1495 .owner = THIS_MODULE,
1496 .load_super = super_1_load,
1497 .validate_super = super_1_validate,
1498 .sync_super = super_1_sync,
1499 .rdev_size_change = super_1_rdev_size_change,
1500 },
1501 };
1502
1503 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1504 {
1505 mdk_rdev_t *rdev, *rdev2;
1506
1507 rcu_read_lock();
1508 rdev_for_each_rcu(rdev, mddev1)
1509 rdev_for_each_rcu(rdev2, mddev2)
1510 if (rdev->bdev->bd_contains ==
1511 rdev2->bdev->bd_contains) {
1512 rcu_read_unlock();
1513 return 1;
1514 }
1515 rcu_read_unlock();
1516 return 0;
1517 }
1518
1519 static LIST_HEAD(pending_raid_disks);
1520
1521 /*
1522 * Try to register data integrity profile for an mddev
1523 *
1524 * This is called when an array is started and after a disk has been kicked
1525 * from the array. It only succeeds if all working and active component devices
1526 * are integrity capable with matching profiles.
1527 */
1528 int md_integrity_register(mddev_t *mddev)
1529 {
1530 mdk_rdev_t *rdev, *reference = NULL;
1531
1532 if (list_empty(&mddev->disks))
1533 return 0; /* nothing to do */
1534 if (blk_get_integrity(mddev->gendisk))
1535 return 0; /* already registered */
1536 list_for_each_entry(rdev, &mddev->disks, same_set) {
1537 /* skip spares and non-functional disks */
1538 if (test_bit(Faulty, &rdev->flags))
1539 continue;
1540 if (rdev->raid_disk < 0)
1541 continue;
1542 /*
1543 * If at least one rdev is not integrity capable, we can not
1544 * enable data integrity for the md device.
1545 */
1546 if (!bdev_get_integrity(rdev->bdev))
1547 return -EINVAL;
1548 if (!reference) {
1549 /* Use the first rdev as the reference */
1550 reference = rdev;
1551 continue;
1552 }
1553 /* does this rdev's profile match the reference profile? */
1554 if (blk_integrity_compare(reference->bdev->bd_disk,
1555 rdev->bdev->bd_disk) < 0)
1556 return -EINVAL;
1557 }
1558 /*
1559 * All component devices are integrity capable and have matching
1560 * profiles, register the common profile for the md device.
1561 */
1562 if (blk_integrity_register(mddev->gendisk,
1563 bdev_get_integrity(reference->bdev)) != 0) {
1564 printk(KERN_ERR "md: failed to register integrity for %s\n",
1565 mdname(mddev));
1566 return -EINVAL;
1567 }
1568 printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1569 mdname(mddev));
1570 return 0;
1571 }
1572 EXPORT_SYMBOL(md_integrity_register);
1573
1574 /* Disable data integrity if non-capable/non-matching disk is being added */
1575 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
1576 {
1577 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1578 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1579
1580 if (!bi_mddev) /* nothing to do */
1581 return;
1582 if (rdev->raid_disk < 0) /* skip spares */
1583 return;
1584 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1585 rdev->bdev->bd_disk) >= 0)
1586 return;
1587 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1588 blk_integrity_unregister(mddev->gendisk);
1589 }
1590 EXPORT_SYMBOL(md_integrity_add_rdev);
1591
1592 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1593 {
1594 char b[BDEVNAME_SIZE];
1595 struct kobject *ko;
1596 char *s;
1597 int err;
1598
1599 if (rdev->mddev) {
1600 MD_BUG();
1601 return -EINVAL;
1602 }
1603
1604 /* prevent duplicates */
1605 if (find_rdev(mddev, rdev->bdev->bd_dev))
1606 return -EEXIST;
1607
1608 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1609 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1610 rdev->sectors < mddev->dev_sectors)) {
1611 if (mddev->pers) {
1612 /* Cannot change size, so fail
1613 * If mddev->level <= 0, then we don't care
1614 * about aligning sizes (e.g. linear)
1615 */
1616 if (mddev->level > 0)
1617 return -ENOSPC;
1618 } else
1619 mddev->dev_sectors = rdev->sectors;
1620 }
1621
1622 /* Verify rdev->desc_nr is unique.
1623 * If it is -1, assign a free number, else
1624 * check number is not in use
1625 */
1626 if (rdev->desc_nr < 0) {
1627 int choice = 0;
1628 if (mddev->pers) choice = mddev->raid_disks;
1629 while (find_rdev_nr(mddev, choice))
1630 choice++;
1631 rdev->desc_nr = choice;
1632 } else {
1633 if (find_rdev_nr(mddev, rdev->desc_nr))
1634 return -EBUSY;
1635 }
1636 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1637 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1638 mdname(mddev), mddev->max_disks);
1639 return -EBUSY;
1640 }
1641 bdevname(rdev->bdev,b);
1642 while ( (s=strchr(b, '/')) != NULL)
1643 *s = '!';
1644
1645 rdev->mddev = mddev;
1646 printk(KERN_INFO "md: bind<%s>\n", b);
1647
1648 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1649 goto fail;
1650
1651 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1652 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1653 kobject_del(&rdev->kobj);
1654 goto fail;
1655 }
1656 rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, "state");
1657
1658 list_add_rcu(&rdev->same_set, &mddev->disks);
1659 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1660
1661 /* May as well allow recovery to be retried once */
1662 mddev->recovery_disabled = 0;
1663
1664 return 0;
1665
1666 fail:
1667 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1668 b, mdname(mddev));
1669 return err;
1670 }
1671
1672 static void md_delayed_delete(struct work_struct *ws)
1673 {
1674 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1675 kobject_del(&rdev->kobj);
1676 kobject_put(&rdev->kobj);
1677 }
1678
1679 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1680 {
1681 char b[BDEVNAME_SIZE];
1682 if (!rdev->mddev) {
1683 MD_BUG();
1684 return;
1685 }
1686 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1687 list_del_rcu(&rdev->same_set);
1688 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1689 rdev->mddev = NULL;
1690 sysfs_remove_link(&rdev->kobj, "block");
1691 sysfs_put(rdev->sysfs_state);
1692 rdev->sysfs_state = NULL;
1693 /* We need to delay this, otherwise we can deadlock when
1694 * writing to 'remove' to "dev/state". We also need
1695 * to delay it due to rcu usage.
1696 */
1697 synchronize_rcu();
1698 INIT_WORK(&rdev->del_work, md_delayed_delete);
1699 kobject_get(&rdev->kobj);
1700 schedule_work(&rdev->del_work);
1701 }
1702
1703 /*
1704 * prevent the device from being mounted, repartitioned or
1705 * otherwise reused by a RAID array (or any other kernel
1706 * subsystem), by bd_claiming the device.
1707 */
1708 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1709 {
1710 int err = 0;
1711 struct block_device *bdev;
1712 char b[BDEVNAME_SIZE];
1713
1714 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1715 if (IS_ERR(bdev)) {
1716 printk(KERN_ERR "md: could not open %s.\n",
1717 __bdevname(dev, b));
1718 return PTR_ERR(bdev);
1719 }
1720 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1721 if (err) {
1722 printk(KERN_ERR "md: could not bd_claim %s.\n",
1723 bdevname(bdev, b));
1724 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1725 return err;
1726 }
1727 if (!shared)
1728 set_bit(AllReserved, &rdev->flags);
1729 rdev->bdev = bdev;
1730 return err;
1731 }
1732
1733 static void unlock_rdev(mdk_rdev_t *rdev)
1734 {
1735 struct block_device *bdev = rdev->bdev;
1736 rdev->bdev = NULL;
1737 if (!bdev)
1738 MD_BUG();
1739 bd_release(bdev);
1740 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1741 }
1742
1743 void md_autodetect_dev(dev_t dev);
1744
1745 static void export_rdev(mdk_rdev_t * rdev)
1746 {
1747 char b[BDEVNAME_SIZE];
1748 printk(KERN_INFO "md: export_rdev(%s)\n",
1749 bdevname(rdev->bdev,b));
1750 if (rdev->mddev)
1751 MD_BUG();
1752 free_disk_sb(rdev);
1753 #ifndef MODULE
1754 if (test_bit(AutoDetected, &rdev->flags))
1755 md_autodetect_dev(rdev->bdev->bd_dev);
1756 #endif
1757 unlock_rdev(rdev);
1758 kobject_put(&rdev->kobj);
1759 }
1760
1761 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1762 {
1763 unbind_rdev_from_array(rdev);
1764 export_rdev(rdev);
1765 }
1766
1767 static void export_array(mddev_t *mddev)
1768 {
1769 mdk_rdev_t *rdev, *tmp;
1770
1771 rdev_for_each(rdev, tmp, mddev) {
1772 if (!rdev->mddev) {
1773 MD_BUG();
1774 continue;
1775 }
1776 kick_rdev_from_array(rdev);
1777 }
1778 if (!list_empty(&mddev->disks))
1779 MD_BUG();
1780 mddev->raid_disks = 0;
1781 mddev->major_version = 0;
1782 }
1783
1784 static void print_desc(mdp_disk_t *desc)
1785 {
1786 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1787 desc->major,desc->minor,desc->raid_disk,desc->state);
1788 }
1789
1790 static void print_sb_90(mdp_super_t *sb)
1791 {
1792 int i;
1793
1794 printk(KERN_INFO
1795 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1796 sb->major_version, sb->minor_version, sb->patch_version,
1797 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1798 sb->ctime);
1799 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1800 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1801 sb->md_minor, sb->layout, sb->chunk_size);
1802 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1803 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1804 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1805 sb->failed_disks, sb->spare_disks,
1806 sb->sb_csum, (unsigned long)sb->events_lo);
1807
1808 printk(KERN_INFO);
1809 for (i = 0; i < MD_SB_DISKS; i++) {
1810 mdp_disk_t *desc;
1811
1812 desc = sb->disks + i;
1813 if (desc->number || desc->major || desc->minor ||
1814 desc->raid_disk || (desc->state && (desc->state != 4))) {
1815 printk(" D %2d: ", i);
1816 print_desc(desc);
1817 }
1818 }
1819 printk(KERN_INFO "md: THIS: ");
1820 print_desc(&sb->this_disk);
1821 }
1822
1823 static void print_sb_1(struct mdp_superblock_1 *sb)
1824 {
1825 __u8 *uuid;
1826
1827 uuid = sb->set_uuid;
1828 printk(KERN_INFO
1829 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%02x%02x%02x%02x"
1830 ":%02x%02x:%02x%02x:%02x%02x:%02x%02x%02x%02x%02x%02x>\n"
1831 "md: Name: \"%s\" CT:%llu\n",
1832 le32_to_cpu(sb->major_version),
1833 le32_to_cpu(sb->feature_map),
1834 uuid[0], uuid[1], uuid[2], uuid[3],
1835 uuid[4], uuid[5], uuid[6], uuid[7],
1836 uuid[8], uuid[9], uuid[10], uuid[11],
1837 uuid[12], uuid[13], uuid[14], uuid[15],
1838 sb->set_name,
1839 (unsigned long long)le64_to_cpu(sb->ctime)
1840 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
1841
1842 uuid = sb->device_uuid;
1843 printk(KERN_INFO
1844 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
1845 " RO:%llu\n"
1846 "md: Dev:%08x UUID: %02x%02x%02x%02x:%02x%02x:%02x%02x:%02x%02x"
1847 ":%02x%02x%02x%02x%02x%02x\n"
1848 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
1849 "md: (MaxDev:%u) \n",
1850 le32_to_cpu(sb->level),
1851 (unsigned long long)le64_to_cpu(sb->size),
1852 le32_to_cpu(sb->raid_disks),
1853 le32_to_cpu(sb->layout),
1854 le32_to_cpu(sb->chunksize),
1855 (unsigned long long)le64_to_cpu(sb->data_offset),
1856 (unsigned long long)le64_to_cpu(sb->data_size),
1857 (unsigned long long)le64_to_cpu(sb->super_offset),
1858 (unsigned long long)le64_to_cpu(sb->recovery_offset),
1859 le32_to_cpu(sb->dev_number),
1860 uuid[0], uuid[1], uuid[2], uuid[3],
1861 uuid[4], uuid[5], uuid[6], uuid[7],
1862 uuid[8], uuid[9], uuid[10], uuid[11],
1863 uuid[12], uuid[13], uuid[14], uuid[15],
1864 sb->devflags,
1865 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
1866 (unsigned long long)le64_to_cpu(sb->events),
1867 (unsigned long long)le64_to_cpu(sb->resync_offset),
1868 le32_to_cpu(sb->sb_csum),
1869 le32_to_cpu(sb->max_dev)
1870 );
1871 }
1872
1873 static void print_rdev(mdk_rdev_t *rdev, int major_version)
1874 {
1875 char b[BDEVNAME_SIZE];
1876 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
1877 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
1878 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1879 rdev->desc_nr);
1880 if (rdev->sb_loaded) {
1881 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
1882 switch (major_version) {
1883 case 0:
1884 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
1885 break;
1886 case 1:
1887 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
1888 break;
1889 }
1890 } else
1891 printk(KERN_INFO "md: no rdev superblock!\n");
1892 }
1893
1894 static void md_print_devices(void)
1895 {
1896 struct list_head *tmp;
1897 mdk_rdev_t *rdev;
1898 mddev_t *mddev;
1899 char b[BDEVNAME_SIZE];
1900
1901 printk("\n");
1902 printk("md: **********************************\n");
1903 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1904 printk("md: **********************************\n");
1905 for_each_mddev(mddev, tmp) {
1906
1907 if (mddev->bitmap)
1908 bitmap_print_sb(mddev->bitmap);
1909 else
1910 printk("%s: ", mdname(mddev));
1911 list_for_each_entry(rdev, &mddev->disks, same_set)
1912 printk("<%s>", bdevname(rdev->bdev,b));
1913 printk("\n");
1914
1915 list_for_each_entry(rdev, &mddev->disks, same_set)
1916 print_rdev(rdev, mddev->major_version);
1917 }
1918 printk("md: **********************************\n");
1919 printk("\n");
1920 }
1921
1922
1923 static void sync_sbs(mddev_t * mddev, int nospares)
1924 {
1925 /* Update each superblock (in-memory image), but
1926 * if we are allowed to, skip spares which already
1927 * have the right event counter, or have one earlier
1928 * (which would mean they aren't being marked as dirty
1929 * with the rest of the array)
1930 */
1931 mdk_rdev_t *rdev;
1932
1933 /* First make sure individual recovery_offsets are correct */
1934 list_for_each_entry(rdev, &mddev->disks, same_set) {
1935 if (rdev->raid_disk >= 0 &&
1936 !test_bit(In_sync, &rdev->flags) &&
1937 mddev->curr_resync_completed > rdev->recovery_offset)
1938 rdev->recovery_offset = mddev->curr_resync_completed;
1939
1940 }
1941 list_for_each_entry(rdev, &mddev->disks, same_set) {
1942 if (rdev->sb_events == mddev->events ||
1943 (nospares &&
1944 rdev->raid_disk < 0 &&
1945 (rdev->sb_events&1)==0 &&
1946 rdev->sb_events+1 == mddev->events)) {
1947 /* Don't update this superblock */
1948 rdev->sb_loaded = 2;
1949 } else {
1950 super_types[mddev->major_version].
1951 sync_super(mddev, rdev);
1952 rdev->sb_loaded = 1;
1953 }
1954 }
1955 }
1956
1957 static void md_update_sb(mddev_t * mddev, int force_change)
1958 {
1959 mdk_rdev_t *rdev;
1960 int sync_req;
1961 int nospares = 0;
1962
1963 mddev->utime = get_seconds();
1964 if (mddev->external)
1965 return;
1966 repeat:
1967 spin_lock_irq(&mddev->write_lock);
1968
1969 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1970 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1971 force_change = 1;
1972 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1973 /* just a clean<-> dirty transition, possibly leave spares alone,
1974 * though if events isn't the right even/odd, we will have to do
1975 * spares after all
1976 */
1977 nospares = 1;
1978 if (force_change)
1979 nospares = 0;
1980 if (mddev->degraded)
1981 /* If the array is degraded, then skipping spares is both
1982 * dangerous and fairly pointless.
1983 * Dangerous because a device that was removed from the array
1984 * might have a event_count that still looks up-to-date,
1985 * so it can be re-added without a resync.
1986 * Pointless because if there are any spares to skip,
1987 * then a recovery will happen and soon that array won't
1988 * be degraded any more and the spare can go back to sleep then.
1989 */
1990 nospares = 0;
1991
1992 sync_req = mddev->in_sync;
1993
1994 /* If this is just a dirty<->clean transition, and the array is clean
1995 * and 'events' is odd, we can roll back to the previous clean state */
1996 if (nospares
1997 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1998 && (mddev->events & 1)
1999 && mddev->events != 1)
2000 mddev->events--;
2001 else {
2002 /* otherwise we have to go forward and ... */
2003 mddev->events ++;
2004 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
2005 /* .. if the array isn't clean, an 'even' event must also go
2006 * to spares. */
2007 if ((mddev->events&1)==0)
2008 nospares = 0;
2009 } else {
2010 /* otherwise an 'odd' event must go to spares */
2011 if ((mddev->events&1))
2012 nospares = 0;
2013 }
2014 }
2015
2016 if (!mddev->events) {
2017 /*
2018 * oops, this 64-bit counter should never wrap.
2019 * Either we are in around ~1 trillion A.C., assuming
2020 * 1 reboot per second, or we have a bug:
2021 */
2022 MD_BUG();
2023 mddev->events --;
2024 }
2025
2026 /*
2027 * do not write anything to disk if using
2028 * nonpersistent superblocks
2029 */
2030 if (!mddev->persistent) {
2031 if (!mddev->external)
2032 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2033
2034 spin_unlock_irq(&mddev->write_lock);
2035 wake_up(&mddev->sb_wait);
2036 return;
2037 }
2038 sync_sbs(mddev, nospares);
2039 spin_unlock_irq(&mddev->write_lock);
2040
2041 dprintk(KERN_INFO
2042 "md: updating %s RAID superblock on device (in sync %d)\n",
2043 mdname(mddev),mddev->in_sync);
2044
2045 bitmap_update_sb(mddev->bitmap);
2046 list_for_each_entry(rdev, &mddev->disks, same_set) {
2047 char b[BDEVNAME_SIZE];
2048 dprintk(KERN_INFO "md: ");
2049 if (rdev->sb_loaded != 1)
2050 continue; /* no noise on spare devices */
2051 if (test_bit(Faulty, &rdev->flags))
2052 dprintk("(skipping faulty ");
2053
2054 dprintk("%s ", bdevname(rdev->bdev,b));
2055 if (!test_bit(Faulty, &rdev->flags)) {
2056 md_super_write(mddev,rdev,
2057 rdev->sb_start, rdev->sb_size,
2058 rdev->sb_page);
2059 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2060 bdevname(rdev->bdev,b),
2061 (unsigned long long)rdev->sb_start);
2062 rdev->sb_events = mddev->events;
2063
2064 } else
2065 dprintk(")\n");
2066 if (mddev->level == LEVEL_MULTIPATH)
2067 /* only need to write one superblock... */
2068 break;
2069 }
2070 md_super_wait(mddev);
2071 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2072
2073 spin_lock_irq(&mddev->write_lock);
2074 if (mddev->in_sync != sync_req ||
2075 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2076 /* have to write it out again */
2077 spin_unlock_irq(&mddev->write_lock);
2078 goto repeat;
2079 }
2080 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2081 spin_unlock_irq(&mddev->write_lock);
2082 wake_up(&mddev->sb_wait);
2083 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2084 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2085
2086 }
2087
2088 /* words written to sysfs files may, or may not, be \n terminated.
2089 * We want to accept with case. For this we use cmd_match.
2090 */
2091 static int cmd_match(const char *cmd, const char *str)
2092 {
2093 /* See if cmd, written into a sysfs file, matches
2094 * str. They must either be the same, or cmd can
2095 * have a trailing newline
2096 */
2097 while (*cmd && *str && *cmd == *str) {
2098 cmd++;
2099 str++;
2100 }
2101 if (*cmd == '\n')
2102 cmd++;
2103 if (*str || *cmd)
2104 return 0;
2105 return 1;
2106 }
2107
2108 struct rdev_sysfs_entry {
2109 struct attribute attr;
2110 ssize_t (*show)(mdk_rdev_t *, char *);
2111 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2112 };
2113
2114 static ssize_t
2115 state_show(mdk_rdev_t *rdev, char *page)
2116 {
2117 char *sep = "";
2118 size_t len = 0;
2119
2120 if (test_bit(Faulty, &rdev->flags)) {
2121 len+= sprintf(page+len, "%sfaulty",sep);
2122 sep = ",";
2123 }
2124 if (test_bit(In_sync, &rdev->flags)) {
2125 len += sprintf(page+len, "%sin_sync",sep);
2126 sep = ",";
2127 }
2128 if (test_bit(WriteMostly, &rdev->flags)) {
2129 len += sprintf(page+len, "%swrite_mostly",sep);
2130 sep = ",";
2131 }
2132 if (test_bit(Blocked, &rdev->flags)) {
2133 len += sprintf(page+len, "%sblocked", sep);
2134 sep = ",";
2135 }
2136 if (!test_bit(Faulty, &rdev->flags) &&
2137 !test_bit(In_sync, &rdev->flags)) {
2138 len += sprintf(page+len, "%sspare", sep);
2139 sep = ",";
2140 }
2141 return len+sprintf(page+len, "\n");
2142 }
2143
2144 static ssize_t
2145 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2146 {
2147 /* can write
2148 * faulty - simulates and error
2149 * remove - disconnects the device
2150 * writemostly - sets write_mostly
2151 * -writemostly - clears write_mostly
2152 * blocked - sets the Blocked flag
2153 * -blocked - clears the Blocked flag
2154 * insync - sets Insync providing device isn't active
2155 */
2156 int err = -EINVAL;
2157 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2158 md_error(rdev->mddev, rdev);
2159 err = 0;
2160 } else if (cmd_match(buf, "remove")) {
2161 if (rdev->raid_disk >= 0)
2162 err = -EBUSY;
2163 else {
2164 mddev_t *mddev = rdev->mddev;
2165 kick_rdev_from_array(rdev);
2166 if (mddev->pers)
2167 md_update_sb(mddev, 1);
2168 md_new_event(mddev);
2169 err = 0;
2170 }
2171 } else if (cmd_match(buf, "writemostly")) {
2172 set_bit(WriteMostly, &rdev->flags);
2173 err = 0;
2174 } else if (cmd_match(buf, "-writemostly")) {
2175 clear_bit(WriteMostly, &rdev->flags);
2176 err = 0;
2177 } else if (cmd_match(buf, "blocked")) {
2178 set_bit(Blocked, &rdev->flags);
2179 err = 0;
2180 } else if (cmd_match(buf, "-blocked")) {
2181 clear_bit(Blocked, &rdev->flags);
2182 wake_up(&rdev->blocked_wait);
2183 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2184 md_wakeup_thread(rdev->mddev->thread);
2185
2186 err = 0;
2187 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2188 set_bit(In_sync, &rdev->flags);
2189 err = 0;
2190 }
2191 if (!err && rdev->sysfs_state)
2192 sysfs_notify_dirent(rdev->sysfs_state);
2193 return err ? err : len;
2194 }
2195 static struct rdev_sysfs_entry rdev_state =
2196 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2197
2198 static ssize_t
2199 errors_show(mdk_rdev_t *rdev, char *page)
2200 {
2201 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2202 }
2203
2204 static ssize_t
2205 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2206 {
2207 char *e;
2208 unsigned long n = simple_strtoul(buf, &e, 10);
2209 if (*buf && (*e == 0 || *e == '\n')) {
2210 atomic_set(&rdev->corrected_errors, n);
2211 return len;
2212 }
2213 return -EINVAL;
2214 }
2215 static struct rdev_sysfs_entry rdev_errors =
2216 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2217
2218 static ssize_t
2219 slot_show(mdk_rdev_t *rdev, char *page)
2220 {
2221 if (rdev->raid_disk < 0)
2222 return sprintf(page, "none\n");
2223 else
2224 return sprintf(page, "%d\n", rdev->raid_disk);
2225 }
2226
2227 static ssize_t
2228 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2229 {
2230 char *e;
2231 int err;
2232 char nm[20];
2233 int slot = simple_strtoul(buf, &e, 10);
2234 if (strncmp(buf, "none", 4)==0)
2235 slot = -1;
2236 else if (e==buf || (*e && *e!= '\n'))
2237 return -EINVAL;
2238 if (rdev->mddev->pers && slot == -1) {
2239 /* Setting 'slot' on an active array requires also
2240 * updating the 'rd%d' link, and communicating
2241 * with the personality with ->hot_*_disk.
2242 * For now we only support removing
2243 * failed/spare devices. This normally happens automatically,
2244 * but not when the metadata is externally managed.
2245 */
2246 if (rdev->raid_disk == -1)
2247 return -EEXIST;
2248 /* personality does all needed checks */
2249 if (rdev->mddev->pers->hot_add_disk == NULL)
2250 return -EINVAL;
2251 err = rdev->mddev->pers->
2252 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2253 if (err)
2254 return err;
2255 sprintf(nm, "rd%d", rdev->raid_disk);
2256 sysfs_remove_link(&rdev->mddev->kobj, nm);
2257 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2258 md_wakeup_thread(rdev->mddev->thread);
2259 } else if (rdev->mddev->pers) {
2260 mdk_rdev_t *rdev2;
2261 /* Activating a spare .. or possibly reactivating
2262 * if we ever get bitmaps working here.
2263 */
2264
2265 if (rdev->raid_disk != -1)
2266 return -EBUSY;
2267
2268 if (rdev->mddev->pers->hot_add_disk == NULL)
2269 return -EINVAL;
2270
2271 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2272 if (rdev2->raid_disk == slot)
2273 return -EEXIST;
2274
2275 rdev->raid_disk = slot;
2276 if (test_bit(In_sync, &rdev->flags))
2277 rdev->saved_raid_disk = slot;
2278 else
2279 rdev->saved_raid_disk = -1;
2280 err = rdev->mddev->pers->
2281 hot_add_disk(rdev->mddev, rdev);
2282 if (err) {
2283 rdev->raid_disk = -1;
2284 return err;
2285 } else
2286 sysfs_notify_dirent(rdev->sysfs_state);
2287 sprintf(nm, "rd%d", rdev->raid_disk);
2288 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2289 printk(KERN_WARNING
2290 "md: cannot register "
2291 "%s for %s\n",
2292 nm, mdname(rdev->mddev));
2293
2294 /* don't wakeup anyone, leave that to userspace. */
2295 } else {
2296 if (slot >= rdev->mddev->raid_disks)
2297 return -ENOSPC;
2298 rdev->raid_disk = slot;
2299 /* assume it is working */
2300 clear_bit(Faulty, &rdev->flags);
2301 clear_bit(WriteMostly, &rdev->flags);
2302 set_bit(In_sync, &rdev->flags);
2303 sysfs_notify_dirent(rdev->sysfs_state);
2304 }
2305 return len;
2306 }
2307
2308
2309 static struct rdev_sysfs_entry rdev_slot =
2310 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2311
2312 static ssize_t
2313 offset_show(mdk_rdev_t *rdev, char *page)
2314 {
2315 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2316 }
2317
2318 static ssize_t
2319 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2320 {
2321 char *e;
2322 unsigned long long offset = simple_strtoull(buf, &e, 10);
2323 if (e==buf || (*e && *e != '\n'))
2324 return -EINVAL;
2325 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2326 return -EBUSY;
2327 if (rdev->sectors && rdev->mddev->external)
2328 /* Must set offset before size, so overlap checks
2329 * can be sane */
2330 return -EBUSY;
2331 rdev->data_offset = offset;
2332 return len;
2333 }
2334
2335 static struct rdev_sysfs_entry rdev_offset =
2336 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2337
2338 static ssize_t
2339 rdev_size_show(mdk_rdev_t *rdev, char *page)
2340 {
2341 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2342 }
2343
2344 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2345 {
2346 /* check if two start/length pairs overlap */
2347 if (s1+l1 <= s2)
2348 return 0;
2349 if (s2+l2 <= s1)
2350 return 0;
2351 return 1;
2352 }
2353
2354 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2355 {
2356 unsigned long long blocks;
2357 sector_t new;
2358
2359 if (strict_strtoull(buf, 10, &blocks) < 0)
2360 return -EINVAL;
2361
2362 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2363 return -EINVAL; /* sector conversion overflow */
2364
2365 new = blocks * 2;
2366 if (new != blocks * 2)
2367 return -EINVAL; /* unsigned long long to sector_t overflow */
2368
2369 *sectors = new;
2370 return 0;
2371 }
2372
2373 static ssize_t
2374 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2375 {
2376 mddev_t *my_mddev = rdev->mddev;
2377 sector_t oldsectors = rdev->sectors;
2378 sector_t sectors;
2379
2380 if (strict_blocks_to_sectors(buf, &sectors) < 0)
2381 return -EINVAL;
2382 if (my_mddev->pers && rdev->raid_disk >= 0) {
2383 if (my_mddev->persistent) {
2384 sectors = super_types[my_mddev->major_version].
2385 rdev_size_change(rdev, sectors);
2386 if (!sectors)
2387 return -EBUSY;
2388 } else if (!sectors)
2389 sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2390 rdev->data_offset;
2391 }
2392 if (sectors < my_mddev->dev_sectors)
2393 return -EINVAL; /* component must fit device */
2394
2395 rdev->sectors = sectors;
2396 if (sectors > oldsectors && my_mddev->external) {
2397 /* need to check that all other rdevs with the same ->bdev
2398 * do not overlap. We need to unlock the mddev to avoid
2399 * a deadlock. We have already changed rdev->sectors, and if
2400 * we have to change it back, we will have the lock again.
2401 */
2402 mddev_t *mddev;
2403 int overlap = 0;
2404 struct list_head *tmp;
2405
2406 mddev_unlock(my_mddev);
2407 for_each_mddev(mddev, tmp) {
2408 mdk_rdev_t *rdev2;
2409
2410 mddev_lock(mddev);
2411 list_for_each_entry(rdev2, &mddev->disks, same_set)
2412 if (test_bit(AllReserved, &rdev2->flags) ||
2413 (rdev->bdev == rdev2->bdev &&
2414 rdev != rdev2 &&
2415 overlaps(rdev->data_offset, rdev->sectors,
2416 rdev2->data_offset,
2417 rdev2->sectors))) {
2418 overlap = 1;
2419 break;
2420 }
2421 mddev_unlock(mddev);
2422 if (overlap) {
2423 mddev_put(mddev);
2424 break;
2425 }
2426 }
2427 mddev_lock(my_mddev);
2428 if (overlap) {
2429 /* Someone else could have slipped in a size
2430 * change here, but doing so is just silly.
2431 * We put oldsectors back because we *know* it is
2432 * safe, and trust userspace not to race with
2433 * itself
2434 */
2435 rdev->sectors = oldsectors;
2436 return -EBUSY;
2437 }
2438 }
2439 return len;
2440 }
2441
2442 static struct rdev_sysfs_entry rdev_size =
2443 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2444
2445 static struct attribute *rdev_default_attrs[] = {
2446 &rdev_state.attr,
2447 &rdev_errors.attr,
2448 &rdev_slot.attr,
2449 &rdev_offset.attr,
2450 &rdev_size.attr,
2451 NULL,
2452 };
2453 static ssize_t
2454 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2455 {
2456 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2457 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2458 mddev_t *mddev = rdev->mddev;
2459 ssize_t rv;
2460
2461 if (!entry->show)
2462 return -EIO;
2463
2464 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2465 if (!rv) {
2466 if (rdev->mddev == NULL)
2467 rv = -EBUSY;
2468 else
2469 rv = entry->show(rdev, page);
2470 mddev_unlock(mddev);
2471 }
2472 return rv;
2473 }
2474
2475 static ssize_t
2476 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2477 const char *page, size_t length)
2478 {
2479 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2480 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2481 ssize_t rv;
2482 mddev_t *mddev = rdev->mddev;
2483
2484 if (!entry->store)
2485 return -EIO;
2486 if (!capable(CAP_SYS_ADMIN))
2487 return -EACCES;
2488 rv = mddev ? mddev_lock(mddev): -EBUSY;
2489 if (!rv) {
2490 if (rdev->mddev == NULL)
2491 rv = -EBUSY;
2492 else
2493 rv = entry->store(rdev, page, length);
2494 mddev_unlock(mddev);
2495 }
2496 return rv;
2497 }
2498
2499 static void rdev_free(struct kobject *ko)
2500 {
2501 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2502 kfree(rdev);
2503 }
2504 static struct sysfs_ops rdev_sysfs_ops = {
2505 .show = rdev_attr_show,
2506 .store = rdev_attr_store,
2507 };
2508 static struct kobj_type rdev_ktype = {
2509 .release = rdev_free,
2510 .sysfs_ops = &rdev_sysfs_ops,
2511 .default_attrs = rdev_default_attrs,
2512 };
2513
2514 /*
2515 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2516 *
2517 * mark the device faulty if:
2518 *
2519 * - the device is nonexistent (zero size)
2520 * - the device has no valid superblock
2521 *
2522 * a faulty rdev _never_ has rdev->sb set.
2523 */
2524 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2525 {
2526 char b[BDEVNAME_SIZE];
2527 int err;
2528 mdk_rdev_t *rdev;
2529 sector_t size;
2530
2531 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2532 if (!rdev) {
2533 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2534 return ERR_PTR(-ENOMEM);
2535 }
2536
2537 if ((err = alloc_disk_sb(rdev)))
2538 goto abort_free;
2539
2540 err = lock_rdev(rdev, newdev, super_format == -2);
2541 if (err)
2542 goto abort_free;
2543
2544 kobject_init(&rdev->kobj, &rdev_ktype);
2545
2546 rdev->desc_nr = -1;
2547 rdev->saved_raid_disk = -1;
2548 rdev->raid_disk = -1;
2549 rdev->flags = 0;
2550 rdev->data_offset = 0;
2551 rdev->sb_events = 0;
2552 atomic_set(&rdev->nr_pending, 0);
2553 atomic_set(&rdev->read_errors, 0);
2554 atomic_set(&rdev->corrected_errors, 0);
2555
2556 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2557 if (!size) {
2558 printk(KERN_WARNING
2559 "md: %s has zero or unknown size, marking faulty!\n",
2560 bdevname(rdev->bdev,b));
2561 err = -EINVAL;
2562 goto abort_free;
2563 }
2564
2565 if (super_format >= 0) {
2566 err = super_types[super_format].
2567 load_super(rdev, NULL, super_minor);
2568 if (err == -EINVAL) {
2569 printk(KERN_WARNING
2570 "md: %s does not have a valid v%d.%d "
2571 "superblock, not importing!\n",
2572 bdevname(rdev->bdev,b),
2573 super_format, super_minor);
2574 goto abort_free;
2575 }
2576 if (err < 0) {
2577 printk(KERN_WARNING
2578 "md: could not read %s's sb, not importing!\n",
2579 bdevname(rdev->bdev,b));
2580 goto abort_free;
2581 }
2582 }
2583
2584 INIT_LIST_HEAD(&rdev->same_set);
2585 init_waitqueue_head(&rdev->blocked_wait);
2586
2587 return rdev;
2588
2589 abort_free:
2590 if (rdev->sb_page) {
2591 if (rdev->bdev)
2592 unlock_rdev(rdev);
2593 free_disk_sb(rdev);
2594 }
2595 kfree(rdev);
2596 return ERR_PTR(err);
2597 }
2598
2599 /*
2600 * Check a full RAID array for plausibility
2601 */
2602
2603
2604 static void analyze_sbs(mddev_t * mddev)
2605 {
2606 int i;
2607 mdk_rdev_t *rdev, *freshest, *tmp;
2608 char b[BDEVNAME_SIZE];
2609
2610 freshest = NULL;
2611 rdev_for_each(rdev, tmp, mddev)
2612 switch (super_types[mddev->major_version].
2613 load_super(rdev, freshest, mddev->minor_version)) {
2614 case 1:
2615 freshest = rdev;
2616 break;
2617 case 0:
2618 break;
2619 default:
2620 printk( KERN_ERR \
2621 "md: fatal superblock inconsistency in %s"
2622 " -- removing from array\n",
2623 bdevname(rdev->bdev,b));
2624 kick_rdev_from_array(rdev);
2625 }
2626
2627
2628 super_types[mddev->major_version].
2629 validate_super(mddev, freshest);
2630
2631 i = 0;
2632 rdev_for_each(rdev, tmp, mddev) {
2633 if (rdev->desc_nr >= mddev->max_disks ||
2634 i > mddev->max_disks) {
2635 printk(KERN_WARNING
2636 "md: %s: %s: only %d devices permitted\n",
2637 mdname(mddev), bdevname(rdev->bdev, b),
2638 mddev->max_disks);
2639 kick_rdev_from_array(rdev);
2640 continue;
2641 }
2642 if (rdev != freshest)
2643 if (super_types[mddev->major_version].
2644 validate_super(mddev, rdev)) {
2645 printk(KERN_WARNING "md: kicking non-fresh %s"
2646 " from array!\n",
2647 bdevname(rdev->bdev,b));
2648 kick_rdev_from_array(rdev);
2649 continue;
2650 }
2651 if (mddev->level == LEVEL_MULTIPATH) {
2652 rdev->desc_nr = i++;
2653 rdev->raid_disk = rdev->desc_nr;
2654 set_bit(In_sync, &rdev->flags);
2655 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
2656 rdev->raid_disk = -1;
2657 clear_bit(In_sync, &rdev->flags);
2658 }
2659 }
2660 }
2661
2662 static void md_safemode_timeout(unsigned long data);
2663
2664 static ssize_t
2665 safe_delay_show(mddev_t *mddev, char *page)
2666 {
2667 int msec = (mddev->safemode_delay*1000)/HZ;
2668 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2669 }
2670 static ssize_t
2671 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2672 {
2673 int scale=1;
2674 int dot=0;
2675 int i;
2676 unsigned long msec;
2677 char buf[30];
2678
2679 /* remove a period, and count digits after it */
2680 if (len >= sizeof(buf))
2681 return -EINVAL;
2682 strlcpy(buf, cbuf, sizeof(buf));
2683 for (i=0; i<len; i++) {
2684 if (dot) {
2685 if (isdigit(buf[i])) {
2686 buf[i-1] = buf[i];
2687 scale *= 10;
2688 }
2689 buf[i] = 0;
2690 } else if (buf[i] == '.') {
2691 dot=1;
2692 buf[i] = 0;
2693 }
2694 }
2695 if (strict_strtoul(buf, 10, &msec) < 0)
2696 return -EINVAL;
2697 msec = (msec * 1000) / scale;
2698 if (msec == 0)
2699 mddev->safemode_delay = 0;
2700 else {
2701 unsigned long old_delay = mddev->safemode_delay;
2702 mddev->safemode_delay = (msec*HZ)/1000;
2703 if (mddev->safemode_delay == 0)
2704 mddev->safemode_delay = 1;
2705 if (mddev->safemode_delay < old_delay)
2706 md_safemode_timeout((unsigned long)mddev);
2707 }
2708 return len;
2709 }
2710 static struct md_sysfs_entry md_safe_delay =
2711 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2712
2713 static ssize_t
2714 level_show(mddev_t *mddev, char *page)
2715 {
2716 struct mdk_personality *p = mddev->pers;
2717 if (p)
2718 return sprintf(page, "%s\n", p->name);
2719 else if (mddev->clevel[0])
2720 return sprintf(page, "%s\n", mddev->clevel);
2721 else if (mddev->level != LEVEL_NONE)
2722 return sprintf(page, "%d\n", mddev->level);
2723 else
2724 return 0;
2725 }
2726
2727 static ssize_t
2728 level_store(mddev_t *mddev, const char *buf, size_t len)
2729 {
2730 char level[16];
2731 ssize_t rv = len;
2732 struct mdk_personality *pers;
2733 void *priv;
2734 mdk_rdev_t *rdev;
2735
2736 if (mddev->pers == NULL) {
2737 if (len == 0)
2738 return 0;
2739 if (len >= sizeof(mddev->clevel))
2740 return -ENOSPC;
2741 strncpy(mddev->clevel, buf, len);
2742 if (mddev->clevel[len-1] == '\n')
2743 len--;
2744 mddev->clevel[len] = 0;
2745 mddev->level = LEVEL_NONE;
2746 return rv;
2747 }
2748
2749 /* request to change the personality. Need to ensure:
2750 * - array is not engaged in resync/recovery/reshape
2751 * - old personality can be suspended
2752 * - new personality will access other array.
2753 */
2754
2755 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
2756 return -EBUSY;
2757
2758 if (!mddev->pers->quiesce) {
2759 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
2760 mdname(mddev), mddev->pers->name);
2761 return -EINVAL;
2762 }
2763
2764 /* Now find the new personality */
2765 if (len == 0 || len >= sizeof(level))
2766 return -EINVAL;
2767 strncpy(level, buf, len);
2768 if (level[len-1] == '\n')
2769 len--;
2770 level[len] = 0;
2771
2772 request_module("md-%s", level);
2773 spin_lock(&pers_lock);
2774 pers = find_pers(LEVEL_NONE, level);
2775 if (!pers || !try_module_get(pers->owner)) {
2776 spin_unlock(&pers_lock);
2777 printk(KERN_WARNING "md: personality %s not loaded\n", level);
2778 return -EINVAL;
2779 }
2780 spin_unlock(&pers_lock);
2781
2782 if (pers == mddev->pers) {
2783 /* Nothing to do! */
2784 module_put(pers->owner);
2785 return rv;
2786 }
2787 if (!pers->takeover) {
2788 module_put(pers->owner);
2789 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
2790 mdname(mddev), level);
2791 return -EINVAL;
2792 }
2793
2794 /* ->takeover must set new_* and/or delta_disks
2795 * if it succeeds, and may set them when it fails.
2796 */
2797 priv = pers->takeover(mddev);
2798 if (IS_ERR(priv)) {
2799 mddev->new_level = mddev->level;
2800 mddev->new_layout = mddev->layout;
2801 mddev->new_chunk_sectors = mddev->chunk_sectors;
2802 mddev->raid_disks -= mddev->delta_disks;
2803 mddev->delta_disks = 0;
2804 module_put(pers->owner);
2805 printk(KERN_WARNING "md: %s: %s would not accept array\n",
2806 mdname(mddev), level);
2807 return PTR_ERR(priv);
2808 }
2809
2810 /* Looks like we have a winner */
2811 mddev_suspend(mddev);
2812 mddev->pers->stop(mddev);
2813 module_put(mddev->pers->owner);
2814 /* Invalidate devices that are now superfluous */
2815 list_for_each_entry(rdev, &mddev->disks, same_set)
2816 if (rdev->raid_disk >= mddev->raid_disks) {
2817 rdev->raid_disk = -1;
2818 clear_bit(In_sync, &rdev->flags);
2819 }
2820 mddev->pers = pers;
2821 mddev->private = priv;
2822 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
2823 mddev->level = mddev->new_level;
2824 mddev->layout = mddev->new_layout;
2825 mddev->chunk_sectors = mddev->new_chunk_sectors;
2826 mddev->delta_disks = 0;
2827 pers->run(mddev);
2828 mddev_resume(mddev);
2829 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2830 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2831 md_wakeup_thread(mddev->thread);
2832 return rv;
2833 }
2834
2835 static struct md_sysfs_entry md_level =
2836 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2837
2838
2839 static ssize_t
2840 layout_show(mddev_t *mddev, char *page)
2841 {
2842 /* just a number, not meaningful for all levels */
2843 if (mddev->reshape_position != MaxSector &&
2844 mddev->layout != mddev->new_layout)
2845 return sprintf(page, "%d (%d)\n",
2846 mddev->new_layout, mddev->layout);
2847 return sprintf(page, "%d\n", mddev->layout);
2848 }
2849
2850 static ssize_t
2851 layout_store(mddev_t *mddev, const char *buf, size_t len)
2852 {
2853 char *e;
2854 unsigned long n = simple_strtoul(buf, &e, 10);
2855
2856 if (!*buf || (*e && *e != '\n'))
2857 return -EINVAL;
2858
2859 if (mddev->pers) {
2860 int err;
2861 if (mddev->pers->check_reshape == NULL)
2862 return -EBUSY;
2863 mddev->new_layout = n;
2864 err = mddev->pers->check_reshape(mddev);
2865 if (err) {
2866 mddev->new_layout = mddev->layout;
2867 return err;
2868 }
2869 } else {
2870 mddev->new_layout = n;
2871 if (mddev->reshape_position == MaxSector)
2872 mddev->layout = n;
2873 }
2874 return len;
2875 }
2876 static struct md_sysfs_entry md_layout =
2877 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2878
2879
2880 static ssize_t
2881 raid_disks_show(mddev_t *mddev, char *page)
2882 {
2883 if (mddev->raid_disks == 0)
2884 return 0;
2885 if (mddev->reshape_position != MaxSector &&
2886 mddev->delta_disks != 0)
2887 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2888 mddev->raid_disks - mddev->delta_disks);
2889 return sprintf(page, "%d\n", mddev->raid_disks);
2890 }
2891
2892 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2893
2894 static ssize_t
2895 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2896 {
2897 char *e;
2898 int rv = 0;
2899 unsigned long n = simple_strtoul(buf, &e, 10);
2900
2901 if (!*buf || (*e && *e != '\n'))
2902 return -EINVAL;
2903
2904 if (mddev->pers)
2905 rv = update_raid_disks(mddev, n);
2906 else if (mddev->reshape_position != MaxSector) {
2907 int olddisks = mddev->raid_disks - mddev->delta_disks;
2908 mddev->delta_disks = n - olddisks;
2909 mddev->raid_disks = n;
2910 } else
2911 mddev->raid_disks = n;
2912 return rv ? rv : len;
2913 }
2914 static struct md_sysfs_entry md_raid_disks =
2915 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2916
2917 static ssize_t
2918 chunk_size_show(mddev_t *mddev, char *page)
2919 {
2920 if (mddev->reshape_position != MaxSector &&
2921 mddev->chunk_sectors != mddev->new_chunk_sectors)
2922 return sprintf(page, "%d (%d)\n",
2923 mddev->new_chunk_sectors << 9,
2924 mddev->chunk_sectors << 9);
2925 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
2926 }
2927
2928 static ssize_t
2929 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2930 {
2931 char *e;
2932 unsigned long n = simple_strtoul(buf, &e, 10);
2933
2934 if (!*buf || (*e && *e != '\n'))
2935 return -EINVAL;
2936
2937 if (mddev->pers) {
2938 int err;
2939 if (mddev->pers->check_reshape == NULL)
2940 return -EBUSY;
2941 mddev->new_chunk_sectors = n >> 9;
2942 err = mddev->pers->check_reshape(mddev);
2943 if (err) {
2944 mddev->new_chunk_sectors = mddev->chunk_sectors;
2945 return err;
2946 }
2947 } else {
2948 mddev->new_chunk_sectors = n >> 9;
2949 if (mddev->reshape_position == MaxSector)
2950 mddev->chunk_sectors = n >> 9;
2951 }
2952 return len;
2953 }
2954 static struct md_sysfs_entry md_chunk_size =
2955 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2956
2957 static ssize_t
2958 resync_start_show(mddev_t *mddev, char *page)
2959 {
2960 if (mddev->recovery_cp == MaxSector)
2961 return sprintf(page, "none\n");
2962 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2963 }
2964
2965 static ssize_t
2966 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2967 {
2968 char *e;
2969 unsigned long long n = simple_strtoull(buf, &e, 10);
2970
2971 if (mddev->pers)
2972 return -EBUSY;
2973 if (!*buf || (*e && *e != '\n'))
2974 return -EINVAL;
2975
2976 mddev->recovery_cp = n;
2977 return len;
2978 }
2979 static struct md_sysfs_entry md_resync_start =
2980 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2981
2982 /*
2983 * The array state can be:
2984 *
2985 * clear
2986 * No devices, no size, no level
2987 * Equivalent to STOP_ARRAY ioctl
2988 * inactive
2989 * May have some settings, but array is not active
2990 * all IO results in error
2991 * When written, doesn't tear down array, but just stops it
2992 * suspended (not supported yet)
2993 * All IO requests will block. The array can be reconfigured.
2994 * Writing this, if accepted, will block until array is quiescent
2995 * readonly
2996 * no resync can happen. no superblocks get written.
2997 * write requests fail
2998 * read-auto
2999 * like readonly, but behaves like 'clean' on a write request.
3000 *
3001 * clean - no pending writes, but otherwise active.
3002 * When written to inactive array, starts without resync
3003 * If a write request arrives then
3004 * if metadata is known, mark 'dirty' and switch to 'active'.
3005 * if not known, block and switch to write-pending
3006 * If written to an active array that has pending writes, then fails.
3007 * active
3008 * fully active: IO and resync can be happening.
3009 * When written to inactive array, starts with resync
3010 *
3011 * write-pending
3012 * clean, but writes are blocked waiting for 'active' to be written.
3013 *
3014 * active-idle
3015 * like active, but no writes have been seen for a while (100msec).
3016 *
3017 */
3018 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3019 write_pending, active_idle, bad_word};
3020 static char *array_states[] = {
3021 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3022 "write-pending", "active-idle", NULL };
3023
3024 static int match_word(const char *word, char **list)
3025 {
3026 int n;
3027 for (n=0; list[n]; n++)
3028 if (cmd_match(word, list[n]))
3029 break;
3030 return n;
3031 }
3032
3033 static ssize_t
3034 array_state_show(mddev_t *mddev, char *page)
3035 {
3036 enum array_state st = inactive;
3037
3038 if (mddev->pers)
3039 switch(mddev->ro) {
3040 case 1:
3041 st = readonly;
3042 break;
3043 case 2:
3044 st = read_auto;
3045 break;
3046 case 0:
3047 if (mddev->in_sync)
3048 st = clean;
3049 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
3050 st = write_pending;
3051 else if (mddev->safemode)
3052 st = active_idle;
3053 else
3054 st = active;
3055 }
3056 else {
3057 if (list_empty(&mddev->disks) &&
3058 mddev->raid_disks == 0 &&
3059 mddev->dev_sectors == 0)
3060 st = clear;
3061 else
3062 st = inactive;
3063 }
3064 return sprintf(page, "%s\n", array_states[st]);
3065 }
3066
3067 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3068 static int do_md_run(mddev_t * mddev);
3069 static int restart_array(mddev_t *mddev);
3070
3071 static ssize_t
3072 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3073 {
3074 int err = -EINVAL;
3075 enum array_state st = match_word(buf, array_states);
3076 switch(st) {
3077 case bad_word:
3078 break;
3079 case clear:
3080 /* stopping an active array */
3081 if (atomic_read(&mddev->openers) > 0)
3082 return -EBUSY;
3083 err = do_md_stop(mddev, 0, 0);
3084 break;
3085 case inactive:
3086 /* stopping an active array */
3087 if (mddev->pers) {
3088 if (atomic_read(&mddev->openers) > 0)
3089 return -EBUSY;
3090 err = do_md_stop(mddev, 2, 0);
3091 } else
3092 err = 0; /* already inactive */
3093 break;
3094 case suspended:
3095 break; /* not supported yet */
3096 case readonly:
3097 if (mddev->pers)
3098 err = do_md_stop(mddev, 1, 0);
3099 else {
3100 mddev->ro = 1;
3101 set_disk_ro(mddev->gendisk, 1);
3102 err = do_md_run(mddev);
3103 }
3104 break;
3105 case read_auto:
3106 if (mddev->pers) {
3107 if (mddev->ro == 0)
3108 err = do_md_stop(mddev, 1, 0);
3109 else if (mddev->ro == 1)
3110 err = restart_array(mddev);
3111 if (err == 0) {
3112 mddev->ro = 2;
3113 set_disk_ro(mddev->gendisk, 0);
3114 }
3115 } else {
3116 mddev->ro = 2;
3117 err = do_md_run(mddev);
3118 }
3119 break;
3120 case clean:
3121 if (mddev->pers) {
3122 restart_array(mddev);
3123 spin_lock_irq(&mddev->write_lock);
3124 if (atomic_read(&mddev->writes_pending) == 0) {
3125 if (mddev->in_sync == 0) {
3126 mddev->in_sync = 1;
3127 if (mddev->safemode == 1)
3128 mddev->safemode = 0;
3129 if (mddev->persistent)
3130 set_bit(MD_CHANGE_CLEAN,
3131 &mddev->flags);
3132 }
3133 err = 0;
3134 } else
3135 err = -EBUSY;
3136 spin_unlock_irq(&mddev->write_lock);
3137 } else
3138 err = -EINVAL;
3139 break;
3140 case active:
3141 if (mddev->pers) {
3142 restart_array(mddev);
3143 if (mddev->external)
3144 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3145 wake_up(&mddev->sb_wait);
3146 err = 0;
3147 } else {
3148 mddev->ro = 0;
3149 set_disk_ro(mddev->gendisk, 0);
3150 err = do_md_run(mddev);
3151 }
3152 break;
3153 case write_pending:
3154 case active_idle:
3155 /* these cannot be set */
3156 break;
3157 }
3158 if (err)
3159 return err;
3160 else {
3161 sysfs_notify_dirent(mddev->sysfs_state);
3162 return len;
3163 }
3164 }
3165 static struct md_sysfs_entry md_array_state =
3166 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3167
3168 static ssize_t
3169 null_show(mddev_t *mddev, char *page)
3170 {
3171 return -EINVAL;
3172 }
3173
3174 static ssize_t
3175 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3176 {
3177 /* buf must be %d:%d\n? giving major and minor numbers */
3178 /* The new device is added to the array.
3179 * If the array has a persistent superblock, we read the
3180 * superblock to initialise info and check validity.
3181 * Otherwise, only checking done is that in bind_rdev_to_array,
3182 * which mainly checks size.
3183 */
3184 char *e;
3185 int major = simple_strtoul(buf, &e, 10);
3186 int minor;
3187 dev_t dev;
3188 mdk_rdev_t *rdev;
3189 int err;
3190
3191 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3192 return -EINVAL;
3193 minor = simple_strtoul(e+1, &e, 10);
3194 if (*e && *e != '\n')
3195 return -EINVAL;
3196 dev = MKDEV(major, minor);
3197 if (major != MAJOR(dev) ||
3198 minor != MINOR(dev))
3199 return -EOVERFLOW;
3200
3201
3202 if (mddev->persistent) {
3203 rdev = md_import_device(dev, mddev->major_version,
3204 mddev->minor_version);
3205 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3206 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3207 mdk_rdev_t, same_set);
3208 err = super_types[mddev->major_version]
3209 .load_super(rdev, rdev0, mddev->minor_version);
3210 if (err < 0)
3211 goto out;
3212 }
3213 } else if (mddev->external)
3214 rdev = md_import_device(dev, -2, -1);
3215 else
3216 rdev = md_import_device(dev, -1, -1);
3217
3218 if (IS_ERR(rdev))
3219 return PTR_ERR(rdev);
3220 err = bind_rdev_to_array(rdev, mddev);
3221 out:
3222 if (err)
3223 export_rdev(rdev);
3224 return err ? err : len;
3225 }
3226
3227 static struct md_sysfs_entry md_new_device =
3228 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3229
3230 static ssize_t
3231 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3232 {
3233 char *end;
3234 unsigned long chunk, end_chunk;
3235
3236 if (!mddev->bitmap)
3237 goto out;
3238 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3239 while (*buf) {
3240 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3241 if (buf == end) break;
3242 if (*end == '-') { /* range */
3243 buf = end + 1;
3244 end_chunk = simple_strtoul(buf, &end, 0);
3245 if (buf == end) break;
3246 }
3247 if (*end && !isspace(*end)) break;
3248 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3249 buf = end;
3250 while (isspace(*buf)) buf++;
3251 }
3252 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3253 out:
3254 return len;
3255 }
3256
3257 static struct md_sysfs_entry md_bitmap =
3258 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3259
3260 static ssize_t
3261 size_show(mddev_t *mddev, char *page)
3262 {
3263 return sprintf(page, "%llu\n",
3264 (unsigned long long)mddev->dev_sectors / 2);
3265 }
3266
3267 static int update_size(mddev_t *mddev, sector_t num_sectors);
3268
3269 static ssize_t
3270 size_store(mddev_t *mddev, const char *buf, size_t len)
3271 {
3272 /* If array is inactive, we can reduce the component size, but
3273 * not increase it (except from 0).
3274 * If array is active, we can try an on-line resize
3275 */
3276 sector_t sectors;
3277 int err = strict_blocks_to_sectors(buf, &sectors);
3278
3279 if (err < 0)
3280 return err;
3281 if (mddev->pers) {
3282 err = update_size(mddev, sectors);
3283 md_update_sb(mddev, 1);
3284 } else {
3285 if (mddev->dev_sectors == 0 ||
3286 mddev->dev_sectors > sectors)
3287 mddev->dev_sectors = sectors;
3288 else
3289 err = -ENOSPC;
3290 }
3291 return err ? err : len;
3292 }
3293
3294 static struct md_sysfs_entry md_size =
3295 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3296
3297
3298 /* Metdata version.
3299 * This is one of
3300 * 'none' for arrays with no metadata (good luck...)
3301 * 'external' for arrays with externally managed metadata,
3302 * or N.M for internally known formats
3303 */
3304 static ssize_t
3305 metadata_show(mddev_t *mddev, char *page)
3306 {
3307 if (mddev->persistent)
3308 return sprintf(page, "%d.%d\n",
3309 mddev->major_version, mddev->minor_version);
3310 else if (mddev->external)
3311 return sprintf(page, "external:%s\n", mddev->metadata_type);
3312 else
3313 return sprintf(page, "none\n");
3314 }
3315
3316 static ssize_t
3317 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3318 {
3319 int major, minor;
3320 char *e;
3321 /* Changing the details of 'external' metadata is
3322 * always permitted. Otherwise there must be
3323 * no devices attached to the array.
3324 */
3325 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3326 ;
3327 else if (!list_empty(&mddev->disks))
3328 return -EBUSY;
3329
3330 if (cmd_match(buf, "none")) {
3331 mddev->persistent = 0;
3332 mddev->external = 0;
3333 mddev->major_version = 0;
3334 mddev->minor_version = 90;
3335 return len;
3336 }
3337 if (strncmp(buf, "external:", 9) == 0) {
3338 size_t namelen = len-9;
3339 if (namelen >= sizeof(mddev->metadata_type))
3340 namelen = sizeof(mddev->metadata_type)-1;
3341 strncpy(mddev->metadata_type, buf+9, namelen);
3342 mddev->metadata_type[namelen] = 0;
3343 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3344 mddev->metadata_type[--namelen] = 0;
3345 mddev->persistent = 0;
3346 mddev->external = 1;
3347 mddev->major_version = 0;
3348 mddev->minor_version = 90;
3349 return len;
3350 }
3351 major = simple_strtoul(buf, &e, 10);
3352 if (e==buf || *e != '.')
3353 return -EINVAL;
3354 buf = e+1;
3355 minor = simple_strtoul(buf, &e, 10);
3356 if (e==buf || (*e && *e != '\n') )
3357 return -EINVAL;
3358 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3359 return -ENOENT;
3360 mddev->major_version = major;
3361 mddev->minor_version = minor;
3362 mddev->persistent = 1;
3363 mddev->external = 0;
3364 return len;
3365 }
3366
3367 static struct md_sysfs_entry md_metadata =
3368 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3369
3370 static ssize_t
3371 action_show(mddev_t *mddev, char *page)
3372 {
3373 char *type = "idle";
3374 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3375 type = "frozen";
3376 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3377 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3378 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3379 type = "reshape";
3380 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3381 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3382 type = "resync";
3383 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3384 type = "check";
3385 else
3386 type = "repair";
3387 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3388 type = "recover";
3389 }
3390 return sprintf(page, "%s\n", type);
3391 }
3392
3393 static ssize_t
3394 action_store(mddev_t *mddev, const char *page, size_t len)
3395 {
3396 if (!mddev->pers || !mddev->pers->sync_request)
3397 return -EINVAL;
3398
3399 if (cmd_match(page, "frozen"))
3400 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3401 else
3402 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3403
3404 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3405 if (mddev->sync_thread) {
3406 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3407 md_unregister_thread(mddev->sync_thread);
3408 mddev->sync_thread = NULL;
3409 mddev->recovery = 0;
3410 }
3411 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3412 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3413 return -EBUSY;
3414 else if (cmd_match(page, "resync"))
3415 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3416 else if (cmd_match(page, "recover")) {
3417 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3418 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3419 } else if (cmd_match(page, "reshape")) {
3420 int err;
3421 if (mddev->pers->start_reshape == NULL)
3422 return -EINVAL;
3423 err = mddev->pers->start_reshape(mddev);
3424 if (err)
3425 return err;
3426 sysfs_notify(&mddev->kobj, NULL, "degraded");
3427 } else {
3428 if (cmd_match(page, "check"))
3429 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3430 else if (!cmd_match(page, "repair"))
3431 return -EINVAL;
3432 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3433 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3434 }
3435 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3436 md_wakeup_thread(mddev->thread);
3437 sysfs_notify_dirent(mddev->sysfs_action);
3438 return len;
3439 }
3440
3441 static ssize_t
3442 mismatch_cnt_show(mddev_t *mddev, char *page)
3443 {
3444 return sprintf(page, "%llu\n",
3445 (unsigned long long) mddev->resync_mismatches);
3446 }
3447
3448 static struct md_sysfs_entry md_scan_mode =
3449 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3450
3451
3452 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3453
3454 static ssize_t
3455 sync_min_show(mddev_t *mddev, char *page)
3456 {
3457 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3458 mddev->sync_speed_min ? "local": "system");
3459 }
3460
3461 static ssize_t
3462 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3463 {
3464 int min;
3465 char *e;
3466 if (strncmp(buf, "system", 6)==0) {
3467 mddev->sync_speed_min = 0;
3468 return len;
3469 }
3470 min = simple_strtoul(buf, &e, 10);
3471 if (buf == e || (*e && *e != '\n') || min <= 0)
3472 return -EINVAL;
3473 mddev->sync_speed_min = min;
3474 return len;
3475 }
3476
3477 static struct md_sysfs_entry md_sync_min =
3478 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3479
3480 static ssize_t
3481 sync_max_show(mddev_t *mddev, char *page)
3482 {
3483 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3484 mddev->sync_speed_max ? "local": "system");
3485 }
3486
3487 static ssize_t
3488 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3489 {
3490 int max;
3491 char *e;
3492 if (strncmp(buf, "system", 6)==0) {
3493 mddev->sync_speed_max = 0;
3494 return len;
3495 }
3496 max = simple_strtoul(buf, &e, 10);
3497 if (buf == e || (*e && *e != '\n') || max <= 0)
3498 return -EINVAL;
3499 mddev->sync_speed_max = max;
3500 return len;
3501 }
3502
3503 static struct md_sysfs_entry md_sync_max =
3504 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3505
3506 static ssize_t
3507 degraded_show(mddev_t *mddev, char *page)
3508 {
3509 return sprintf(page, "%d\n", mddev->degraded);
3510 }
3511 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3512
3513 static ssize_t
3514 sync_force_parallel_show(mddev_t *mddev, char *page)
3515 {
3516 return sprintf(page, "%d\n", mddev->parallel_resync);
3517 }
3518
3519 static ssize_t
3520 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3521 {
3522 long n;
3523
3524 if (strict_strtol(buf, 10, &n))
3525 return -EINVAL;
3526
3527 if (n != 0 && n != 1)
3528 return -EINVAL;
3529
3530 mddev->parallel_resync = n;
3531
3532 if (mddev->sync_thread)
3533 wake_up(&resync_wait);
3534
3535 return len;
3536 }
3537
3538 /* force parallel resync, even with shared block devices */
3539 static struct md_sysfs_entry md_sync_force_parallel =
3540 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3541 sync_force_parallel_show, sync_force_parallel_store);
3542
3543 static ssize_t
3544 sync_speed_show(mddev_t *mddev, char *page)
3545 {
3546 unsigned long resync, dt, db;
3547 if (mddev->curr_resync == 0)
3548 return sprintf(page, "none\n");
3549 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3550 dt = (jiffies - mddev->resync_mark) / HZ;
3551 if (!dt) dt++;
3552 db = resync - mddev->resync_mark_cnt;
3553 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3554 }
3555
3556 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3557
3558 static ssize_t
3559 sync_completed_show(mddev_t *mddev, char *page)
3560 {
3561 unsigned long max_sectors, resync;
3562
3563 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3564 return sprintf(page, "none\n");
3565
3566 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3567 max_sectors = mddev->resync_max_sectors;
3568 else
3569 max_sectors = mddev->dev_sectors;
3570
3571 resync = mddev->curr_resync_completed;
3572 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3573 }
3574
3575 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3576
3577 static ssize_t
3578 min_sync_show(mddev_t *mddev, char *page)
3579 {
3580 return sprintf(page, "%llu\n",
3581 (unsigned long long)mddev->resync_min);
3582 }
3583 static ssize_t
3584 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3585 {
3586 unsigned long long min;
3587 if (strict_strtoull(buf, 10, &min))
3588 return -EINVAL;
3589 if (min > mddev->resync_max)
3590 return -EINVAL;
3591 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3592 return -EBUSY;
3593
3594 /* Must be a multiple of chunk_size */
3595 if (mddev->chunk_sectors) {
3596 sector_t temp = min;
3597 if (sector_div(temp, mddev->chunk_sectors))
3598 return -EINVAL;
3599 }
3600 mddev->resync_min = min;
3601
3602 return len;
3603 }
3604
3605 static struct md_sysfs_entry md_min_sync =
3606 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3607
3608 static ssize_t
3609 max_sync_show(mddev_t *mddev, char *page)
3610 {
3611 if (mddev->resync_max == MaxSector)
3612 return sprintf(page, "max\n");
3613 else
3614 return sprintf(page, "%llu\n",
3615 (unsigned long long)mddev->resync_max);
3616 }
3617 static ssize_t
3618 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3619 {
3620 if (strncmp(buf, "max", 3) == 0)
3621 mddev->resync_max = MaxSector;
3622 else {
3623 unsigned long long max;
3624 if (strict_strtoull(buf, 10, &max))
3625 return -EINVAL;
3626 if (max < mddev->resync_min)
3627 return -EINVAL;
3628 if (max < mddev->resync_max &&
3629 mddev->ro == 0 &&
3630 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3631 return -EBUSY;
3632
3633 /* Must be a multiple of chunk_size */
3634 if (mddev->chunk_sectors) {
3635 sector_t temp = max;
3636 if (sector_div(temp, mddev->chunk_sectors))
3637 return -EINVAL;
3638 }
3639 mddev->resync_max = max;
3640 }
3641 wake_up(&mddev->recovery_wait);
3642 return len;
3643 }
3644
3645 static struct md_sysfs_entry md_max_sync =
3646 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3647
3648 static ssize_t
3649 suspend_lo_show(mddev_t *mddev, char *page)
3650 {
3651 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3652 }
3653
3654 static ssize_t
3655 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3656 {
3657 char *e;
3658 unsigned long long new = simple_strtoull(buf, &e, 10);
3659
3660 if (mddev->pers == NULL ||
3661 mddev->pers->quiesce == NULL)
3662 return -EINVAL;
3663 if (buf == e || (*e && *e != '\n'))
3664 return -EINVAL;
3665 if (new >= mddev->suspend_hi ||
3666 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3667 mddev->suspend_lo = new;
3668 mddev->pers->quiesce(mddev, 2);
3669 return len;
3670 } else
3671 return -EINVAL;
3672 }
3673 static struct md_sysfs_entry md_suspend_lo =
3674 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3675
3676
3677 static ssize_t
3678 suspend_hi_show(mddev_t *mddev, char *page)
3679 {
3680 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3681 }
3682
3683 static ssize_t
3684 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3685 {
3686 char *e;
3687 unsigned long long new = simple_strtoull(buf, &e, 10);
3688
3689 if (mddev->pers == NULL ||
3690 mddev->pers->quiesce == NULL)
3691 return -EINVAL;
3692 if (buf == e || (*e && *e != '\n'))
3693 return -EINVAL;
3694 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3695 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3696 mddev->suspend_hi = new;
3697 mddev->pers->quiesce(mddev, 1);
3698 mddev->pers->quiesce(mddev, 0);
3699 return len;
3700 } else
3701 return -EINVAL;
3702 }
3703 static struct md_sysfs_entry md_suspend_hi =
3704 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3705
3706 static ssize_t
3707 reshape_position_show(mddev_t *mddev, char *page)
3708 {
3709 if (mddev->reshape_position != MaxSector)
3710 return sprintf(page, "%llu\n",
3711 (unsigned long long)mddev->reshape_position);
3712 strcpy(page, "none\n");
3713 return 5;
3714 }
3715
3716 static ssize_t
3717 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3718 {
3719 char *e;
3720 unsigned long long new = simple_strtoull(buf, &e, 10);
3721 if (mddev->pers)
3722 return -EBUSY;
3723 if (buf == e || (*e && *e != '\n'))
3724 return -EINVAL;
3725 mddev->reshape_position = new;
3726 mddev->delta_disks = 0;
3727 mddev->new_level = mddev->level;
3728 mddev->new_layout = mddev->layout;
3729 mddev->new_chunk_sectors = mddev->chunk_sectors;
3730 return len;
3731 }
3732
3733 static struct md_sysfs_entry md_reshape_position =
3734 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3735 reshape_position_store);
3736
3737 static ssize_t
3738 array_size_show(mddev_t *mddev, char *page)
3739 {
3740 if (mddev->external_size)
3741 return sprintf(page, "%llu\n",
3742 (unsigned long long)mddev->array_sectors/2);
3743 else
3744 return sprintf(page, "default\n");
3745 }
3746
3747 static ssize_t
3748 array_size_store(mddev_t *mddev, const char *buf, size_t len)
3749 {
3750 sector_t sectors;
3751
3752 if (strncmp(buf, "default", 7) == 0) {
3753 if (mddev->pers)
3754 sectors = mddev->pers->size(mddev, 0, 0);
3755 else
3756 sectors = mddev->array_sectors;
3757
3758 mddev->external_size = 0;
3759 } else {
3760 if (strict_blocks_to_sectors(buf, &sectors) < 0)
3761 return -EINVAL;
3762 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
3763 return -E2BIG;
3764
3765 mddev->external_size = 1;
3766 }
3767
3768 mddev->array_sectors = sectors;
3769 set_capacity(mddev->gendisk, mddev->array_sectors);
3770 if (mddev->pers)
3771 revalidate_disk(mddev->gendisk);
3772
3773 return len;
3774 }
3775
3776 static struct md_sysfs_entry md_array_size =
3777 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
3778 array_size_store);
3779
3780 static struct attribute *md_default_attrs[] = {
3781 &md_level.attr,
3782 &md_layout.attr,
3783 &md_raid_disks.attr,
3784 &md_chunk_size.attr,
3785 &md_size.attr,
3786 &md_resync_start.attr,
3787 &md_metadata.attr,
3788 &md_new_device.attr,
3789 &md_safe_delay.attr,
3790 &md_array_state.attr,
3791 &md_reshape_position.attr,
3792 &md_array_size.attr,
3793 NULL,
3794 };
3795
3796 static struct attribute *md_redundancy_attrs[] = {
3797 &md_scan_mode.attr,
3798 &md_mismatches.attr,
3799 &md_sync_min.attr,
3800 &md_sync_max.attr,
3801 &md_sync_speed.attr,
3802 &md_sync_force_parallel.attr,
3803 &md_sync_completed.attr,
3804 &md_min_sync.attr,
3805 &md_max_sync.attr,
3806 &md_suspend_lo.attr,
3807 &md_suspend_hi.attr,
3808 &md_bitmap.attr,
3809 &md_degraded.attr,
3810 NULL,
3811 };
3812 static struct attribute_group md_redundancy_group = {
3813 .name = NULL,
3814 .attrs = md_redundancy_attrs,
3815 };
3816
3817
3818 static ssize_t
3819 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3820 {
3821 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3822 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3823 ssize_t rv;
3824
3825 if (!entry->show)
3826 return -EIO;
3827 rv = mddev_lock(mddev);
3828 if (!rv) {
3829 rv = entry->show(mddev, page);
3830 mddev_unlock(mddev);
3831 }
3832 return rv;
3833 }
3834
3835 static ssize_t
3836 md_attr_store(struct kobject *kobj, struct attribute *attr,
3837 const char *page, size_t length)
3838 {
3839 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3840 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3841 ssize_t rv;
3842
3843 if (!entry->store)
3844 return -EIO;
3845 if (!capable(CAP_SYS_ADMIN))
3846 return -EACCES;
3847 rv = mddev_lock(mddev);
3848 if (mddev->hold_active == UNTIL_IOCTL)
3849 mddev->hold_active = 0;
3850 if (!rv) {
3851 rv = entry->store(mddev, page, length);
3852 mddev_unlock(mddev);
3853 }
3854 return rv;
3855 }
3856
3857 static void md_free(struct kobject *ko)
3858 {
3859 mddev_t *mddev = container_of(ko, mddev_t, kobj);
3860
3861 if (mddev->sysfs_state)
3862 sysfs_put(mddev->sysfs_state);
3863
3864 if (mddev->gendisk) {
3865 del_gendisk(mddev->gendisk);
3866 put_disk(mddev->gendisk);
3867 }
3868 if (mddev->queue)
3869 blk_cleanup_queue(mddev->queue);
3870
3871 kfree(mddev);
3872 }
3873
3874 static struct sysfs_ops md_sysfs_ops = {
3875 .show = md_attr_show,
3876 .store = md_attr_store,
3877 };
3878 static struct kobj_type md_ktype = {
3879 .release = md_free,
3880 .sysfs_ops = &md_sysfs_ops,
3881 .default_attrs = md_default_attrs,
3882 };
3883
3884 int mdp_major = 0;
3885
3886 static void mddev_delayed_delete(struct work_struct *ws)
3887 {
3888 mddev_t *mddev = container_of(ws, mddev_t, del_work);
3889
3890 if (mddev->private == &md_redundancy_group) {
3891 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3892 if (mddev->sysfs_action)
3893 sysfs_put(mddev->sysfs_action);
3894 mddev->sysfs_action = NULL;
3895 mddev->private = NULL;
3896 }
3897 kobject_del(&mddev->kobj);
3898 kobject_put(&mddev->kobj);
3899 }
3900
3901 static int md_alloc(dev_t dev, char *name)
3902 {
3903 static DEFINE_MUTEX(disks_mutex);
3904 mddev_t *mddev = mddev_find(dev);
3905 struct gendisk *disk;
3906 int partitioned;
3907 int shift;
3908 int unit;
3909 int error;
3910
3911 if (!mddev)
3912 return -ENODEV;
3913
3914 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
3915 shift = partitioned ? MdpMinorShift : 0;
3916 unit = MINOR(mddev->unit) >> shift;
3917
3918 /* wait for any previous instance if this device
3919 * to be completed removed (mddev_delayed_delete).
3920 */
3921 flush_scheduled_work();
3922
3923 mutex_lock(&disks_mutex);
3924 error = -EEXIST;
3925 if (mddev->gendisk)
3926 goto abort;
3927
3928 if (name) {
3929 /* Need to ensure that 'name' is not a duplicate.
3930 */
3931 mddev_t *mddev2;
3932 spin_lock(&all_mddevs_lock);
3933
3934 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
3935 if (mddev2->gendisk &&
3936 strcmp(mddev2->gendisk->disk_name, name) == 0) {
3937 spin_unlock(&all_mddevs_lock);
3938 goto abort;
3939 }
3940 spin_unlock(&all_mddevs_lock);
3941 }
3942
3943 error = -ENOMEM;
3944 mddev->queue = blk_alloc_queue(GFP_KERNEL);
3945 if (!mddev->queue)
3946 goto abort;
3947 mddev->queue->queuedata = mddev;
3948
3949 /* Can be unlocked because the queue is new: no concurrency */
3950 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
3951
3952 blk_queue_make_request(mddev->queue, md_make_request);
3953
3954 disk = alloc_disk(1 << shift);
3955 if (!disk) {
3956 blk_cleanup_queue(mddev->queue);
3957 mddev->queue = NULL;
3958 goto abort;
3959 }
3960 disk->major = MAJOR(mddev->unit);
3961 disk->first_minor = unit << shift;
3962 if (name)
3963 strcpy(disk->disk_name, name);
3964 else if (partitioned)
3965 sprintf(disk->disk_name, "md_d%d", unit);
3966 else
3967 sprintf(disk->disk_name, "md%d", unit);
3968 disk->fops = &md_fops;
3969 disk->private_data = mddev;
3970 disk->queue = mddev->queue;
3971 /* Allow extended partitions. This makes the
3972 * 'mdp' device redundant, but we can't really
3973 * remove it now.
3974 */
3975 disk->flags |= GENHD_FL_EXT_DEVT;
3976 add_disk(disk);
3977 mddev->gendisk = disk;
3978 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
3979 &disk_to_dev(disk)->kobj, "%s", "md");
3980 if (error) {
3981 /* This isn't possible, but as kobject_init_and_add is marked
3982 * __must_check, we must do something with the result
3983 */
3984 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3985 disk->disk_name);
3986 error = 0;
3987 }
3988 abort:
3989 mutex_unlock(&disks_mutex);
3990 if (!error) {
3991 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3992 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state");
3993 }
3994 mddev_put(mddev);
3995 return error;
3996 }
3997
3998 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3999 {
4000 md_alloc(dev, NULL);
4001 return NULL;
4002 }
4003
4004 static int add_named_array(const char *val, struct kernel_param *kp)
4005 {
4006 /* val must be "md_*" where * is not all digits.
4007 * We allocate an array with a large free minor number, and
4008 * set the name to val. val must not already be an active name.
4009 */
4010 int len = strlen(val);
4011 char buf[DISK_NAME_LEN];
4012
4013 while (len && val[len-1] == '\n')
4014 len--;
4015 if (len >= DISK_NAME_LEN)
4016 return -E2BIG;
4017 strlcpy(buf, val, len+1);
4018 if (strncmp(buf, "md_", 3) != 0)
4019 return -EINVAL;
4020 return md_alloc(0, buf);
4021 }
4022
4023 static void md_safemode_timeout(unsigned long data)
4024 {
4025 mddev_t *mddev = (mddev_t *) data;
4026
4027 if (!atomic_read(&mddev->writes_pending)) {
4028 mddev->safemode = 1;
4029 if (mddev->external)
4030 sysfs_notify_dirent(mddev->sysfs_state);
4031 }
4032 md_wakeup_thread(mddev->thread);
4033 }
4034
4035 static int start_dirty_degraded;
4036
4037 static int do_md_run(mddev_t * mddev)
4038 {
4039 int err;
4040 mdk_rdev_t *rdev;
4041 struct gendisk *disk;
4042 struct mdk_personality *pers;
4043
4044 if (list_empty(&mddev->disks))
4045 /* cannot run an array with no devices.. */
4046 return -EINVAL;
4047
4048 if (mddev->pers)
4049 return -EBUSY;
4050
4051 /*
4052 * Analyze all RAID superblock(s)
4053 */
4054 if (!mddev->raid_disks) {
4055 if (!mddev->persistent)
4056 return -EINVAL;
4057 analyze_sbs(mddev);
4058 }
4059
4060 if (mddev->level != LEVEL_NONE)
4061 request_module("md-level-%d", mddev->level);
4062 else if (mddev->clevel[0])
4063 request_module("md-%s", mddev->clevel);
4064
4065 /*
4066 * Drop all container device buffers, from now on
4067 * the only valid external interface is through the md
4068 * device.
4069 */
4070 list_for_each_entry(rdev, &mddev->disks, same_set) {
4071 if (test_bit(Faulty, &rdev->flags))
4072 continue;
4073 sync_blockdev(rdev->bdev);
4074 invalidate_bdev(rdev->bdev);
4075
4076 /* perform some consistency tests on the device.
4077 * We don't want the data to overlap the metadata,
4078 * Internal Bitmap issues have been handled elsewhere.
4079 */
4080 if (rdev->data_offset < rdev->sb_start) {
4081 if (mddev->dev_sectors &&
4082 rdev->data_offset + mddev->dev_sectors
4083 > rdev->sb_start) {
4084 printk("md: %s: data overlaps metadata\n",
4085 mdname(mddev));
4086 return -EINVAL;
4087 }
4088 } else {
4089 if (rdev->sb_start + rdev->sb_size/512
4090 > rdev->data_offset) {
4091 printk("md: %s: metadata overlaps data\n",
4092 mdname(mddev));
4093 return -EINVAL;
4094 }
4095 }
4096 sysfs_notify_dirent(rdev->sysfs_state);
4097 }
4098
4099 md_probe(mddev->unit, NULL, NULL);
4100 disk = mddev->gendisk;
4101 if (!disk)
4102 return -ENOMEM;
4103
4104 spin_lock(&pers_lock);
4105 pers = find_pers(mddev->level, mddev->clevel);
4106 if (!pers || !try_module_get(pers->owner)) {
4107 spin_unlock(&pers_lock);
4108 if (mddev->level != LEVEL_NONE)
4109 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4110 mddev->level);
4111 else
4112 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4113 mddev->clevel);
4114 return -EINVAL;
4115 }
4116 mddev->pers = pers;
4117 spin_unlock(&pers_lock);
4118 if (mddev->level != pers->level) {
4119 mddev->level = pers->level;
4120 mddev->new_level = pers->level;
4121 }
4122 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4123
4124 if (mddev->reshape_position != MaxSector &&
4125 pers->start_reshape == NULL) {
4126 /* This personality cannot handle reshaping... */
4127 mddev->pers = NULL;
4128 module_put(pers->owner);
4129 return -EINVAL;
4130 }
4131
4132 if (pers->sync_request) {
4133 /* Warn if this is a potentially silly
4134 * configuration.
4135 */
4136 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4137 mdk_rdev_t *rdev2;
4138 int warned = 0;
4139
4140 list_for_each_entry(rdev, &mddev->disks, same_set)
4141 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4142 if (rdev < rdev2 &&
4143 rdev->bdev->bd_contains ==
4144 rdev2->bdev->bd_contains) {
4145 printk(KERN_WARNING
4146 "%s: WARNING: %s appears to be"
4147 " on the same physical disk as"
4148 " %s.\n",
4149 mdname(mddev),
4150 bdevname(rdev->bdev,b),
4151 bdevname(rdev2->bdev,b2));
4152 warned = 1;
4153 }
4154 }
4155
4156 if (warned)
4157 printk(KERN_WARNING
4158 "True protection against single-disk"
4159 " failure might be compromised.\n");
4160 }
4161
4162 mddev->recovery = 0;
4163 /* may be over-ridden by personality */
4164 mddev->resync_max_sectors = mddev->dev_sectors;
4165
4166 mddev->barriers_work = 1;
4167 mddev->ok_start_degraded = start_dirty_degraded;
4168
4169 if (start_readonly)
4170 mddev->ro = 2; /* read-only, but switch on first write */
4171
4172 err = mddev->pers->run(mddev);
4173 if (err)
4174 printk(KERN_ERR "md: pers->run() failed ...\n");
4175 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4176 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4177 " but 'external_size' not in effect?\n", __func__);
4178 printk(KERN_ERR
4179 "md: invalid array_size %llu > default size %llu\n",
4180 (unsigned long long)mddev->array_sectors / 2,
4181 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4182 err = -EINVAL;
4183 mddev->pers->stop(mddev);
4184 }
4185 if (err == 0 && mddev->pers->sync_request) {
4186 err = bitmap_create(mddev);
4187 if (err) {
4188 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4189 mdname(mddev), err);
4190 mddev->pers->stop(mddev);
4191 }
4192 }
4193 if (err) {
4194 module_put(mddev->pers->owner);
4195 mddev->pers = NULL;
4196 bitmap_destroy(mddev);
4197 return err;
4198 }
4199 if (mddev->pers->sync_request) {
4200 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4201 printk(KERN_WARNING
4202 "md: cannot register extra attributes for %s\n",
4203 mdname(mddev));
4204 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4205 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4206 mddev->ro = 0;
4207
4208 atomic_set(&mddev->writes_pending,0);
4209 mddev->safemode = 0;
4210 mddev->safemode_timer.function = md_safemode_timeout;
4211 mddev->safemode_timer.data = (unsigned long) mddev;
4212 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4213 mddev->in_sync = 1;
4214
4215 list_for_each_entry(rdev, &mddev->disks, same_set)
4216 if (rdev->raid_disk >= 0) {
4217 char nm[20];
4218 sprintf(nm, "rd%d", rdev->raid_disk);
4219 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4220 printk("md: cannot register %s for %s\n",
4221 nm, mdname(mddev));
4222 }
4223
4224 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4225
4226 if (mddev->flags)
4227 md_update_sb(mddev, 0);
4228
4229 set_capacity(disk, mddev->array_sectors);
4230
4231 /* If there is a partially-recovered drive we need to
4232 * start recovery here. If we leave it to md_check_recovery,
4233 * it will remove the drives and not do the right thing
4234 */
4235 if (mddev->degraded && !mddev->sync_thread) {
4236 int spares = 0;
4237 list_for_each_entry(rdev, &mddev->disks, same_set)
4238 if (rdev->raid_disk >= 0 &&
4239 !test_bit(In_sync, &rdev->flags) &&
4240 !test_bit(Faulty, &rdev->flags))
4241 /* complete an interrupted recovery */
4242 spares++;
4243 if (spares && mddev->pers->sync_request) {
4244 mddev->recovery = 0;
4245 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4246 mddev->sync_thread = md_register_thread(md_do_sync,
4247 mddev,
4248 "resync");
4249 if (!mddev->sync_thread) {
4250 printk(KERN_ERR "%s: could not start resync"
4251 " thread...\n",
4252 mdname(mddev));
4253 /* leave the spares where they are, it shouldn't hurt */
4254 mddev->recovery = 0;
4255 }
4256 }
4257 }
4258 md_wakeup_thread(mddev->thread);
4259 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4260
4261 revalidate_disk(mddev->gendisk);
4262 mddev->changed = 1;
4263 md_new_event(mddev);
4264 sysfs_notify_dirent(mddev->sysfs_state);
4265 if (mddev->sysfs_action)
4266 sysfs_notify_dirent(mddev->sysfs_action);
4267 sysfs_notify(&mddev->kobj, NULL, "degraded");
4268 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4269 return 0;
4270 }
4271
4272 static int restart_array(mddev_t *mddev)
4273 {
4274 struct gendisk *disk = mddev->gendisk;
4275
4276 /* Complain if it has no devices */
4277 if (list_empty(&mddev->disks))
4278 return -ENXIO;
4279 if (!mddev->pers)
4280 return -EINVAL;
4281 if (!mddev->ro)
4282 return -EBUSY;
4283 mddev->safemode = 0;
4284 mddev->ro = 0;
4285 set_disk_ro(disk, 0);
4286 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4287 mdname(mddev));
4288 /* Kick recovery or resync if necessary */
4289 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4290 md_wakeup_thread(mddev->thread);
4291 md_wakeup_thread(mddev->sync_thread);
4292 sysfs_notify_dirent(mddev->sysfs_state);
4293 return 0;
4294 }
4295
4296 /* similar to deny_write_access, but accounts for our holding a reference
4297 * to the file ourselves */
4298 static int deny_bitmap_write_access(struct file * file)
4299 {
4300 struct inode *inode = file->f_mapping->host;
4301
4302 spin_lock(&inode->i_lock);
4303 if (atomic_read(&inode->i_writecount) > 1) {
4304 spin_unlock(&inode->i_lock);
4305 return -ETXTBSY;
4306 }
4307 atomic_set(&inode->i_writecount, -1);
4308 spin_unlock(&inode->i_lock);
4309
4310 return 0;
4311 }
4312
4313 static void restore_bitmap_write_access(struct file *file)
4314 {
4315 struct inode *inode = file->f_mapping->host;
4316
4317 spin_lock(&inode->i_lock);
4318 atomic_set(&inode->i_writecount, 1);
4319 spin_unlock(&inode->i_lock);
4320 }
4321
4322 /* mode:
4323 * 0 - completely stop and dis-assemble array
4324 * 1 - switch to readonly
4325 * 2 - stop but do not disassemble array
4326 */
4327 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4328 {
4329 int err = 0;
4330 struct gendisk *disk = mddev->gendisk;
4331 mdk_rdev_t *rdev;
4332
4333 mutex_lock(&mddev->open_mutex);
4334 if (atomic_read(&mddev->openers) > is_open) {
4335 printk("md: %s still in use.\n",mdname(mddev));
4336 err = -EBUSY;
4337 } else if (mddev->pers) {
4338
4339 if (mddev->sync_thread) {
4340 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4341 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4342 md_unregister_thread(mddev->sync_thread);
4343 mddev->sync_thread = NULL;
4344 }
4345
4346 del_timer_sync(&mddev->safemode_timer);
4347
4348 switch(mode) {
4349 case 1: /* readonly */
4350 err = -ENXIO;
4351 if (mddev->ro==1)
4352 goto out;
4353 mddev->ro = 1;
4354 break;
4355 case 0: /* disassemble */
4356 case 2: /* stop */
4357 bitmap_flush(mddev);
4358 md_super_wait(mddev);
4359 if (mddev->ro)
4360 set_disk_ro(disk, 0);
4361
4362 mddev->pers->stop(mddev);
4363 mddev->queue->merge_bvec_fn = NULL;
4364 mddev->queue->unplug_fn = NULL;
4365 mddev->queue->backing_dev_info.congested_fn = NULL;
4366 module_put(mddev->pers->owner);
4367 if (mddev->pers->sync_request)
4368 mddev->private = &md_redundancy_group;
4369 mddev->pers = NULL;
4370 /* tell userspace to handle 'inactive' */
4371 sysfs_notify_dirent(mddev->sysfs_state);
4372
4373 list_for_each_entry(rdev, &mddev->disks, same_set)
4374 if (rdev->raid_disk >= 0) {
4375 char nm[20];
4376 sprintf(nm, "rd%d", rdev->raid_disk);
4377 sysfs_remove_link(&mddev->kobj, nm);
4378 }
4379
4380 set_capacity(disk, 0);
4381 mddev->changed = 1;
4382
4383 if (mddev->ro)
4384 mddev->ro = 0;
4385 }
4386 if (!mddev->in_sync || mddev->flags) {
4387 /* mark array as shutdown cleanly */
4388 mddev->in_sync = 1;
4389 md_update_sb(mddev, 1);
4390 }
4391 if (mode == 1)
4392 set_disk_ro(disk, 1);
4393 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4394 err = 0;
4395 }
4396 out:
4397 mutex_unlock(&mddev->open_mutex);
4398 if (err)
4399 return err;
4400 /*
4401 * Free resources if final stop
4402 */
4403 if (mode == 0) {
4404
4405 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4406
4407 bitmap_destroy(mddev);
4408 if (mddev->bitmap_file) {
4409 restore_bitmap_write_access(mddev->bitmap_file);
4410 fput(mddev->bitmap_file);
4411 mddev->bitmap_file = NULL;
4412 }
4413 mddev->bitmap_offset = 0;
4414
4415 /* make sure all md_delayed_delete calls have finished */
4416 flush_scheduled_work();
4417
4418 export_array(mddev);
4419
4420 mddev->array_sectors = 0;
4421 mddev->external_size = 0;
4422 mddev->dev_sectors = 0;
4423 mddev->raid_disks = 0;
4424 mddev->recovery_cp = 0;
4425 mddev->resync_min = 0;
4426 mddev->resync_max = MaxSector;
4427 mddev->reshape_position = MaxSector;
4428 mddev->external = 0;
4429 mddev->persistent = 0;
4430 mddev->level = LEVEL_NONE;
4431 mddev->clevel[0] = 0;
4432 mddev->flags = 0;
4433 mddev->ro = 0;
4434 mddev->metadata_type[0] = 0;
4435 mddev->chunk_sectors = 0;
4436 mddev->ctime = mddev->utime = 0;
4437 mddev->layout = 0;
4438 mddev->max_disks = 0;
4439 mddev->events = 0;
4440 mddev->delta_disks = 0;
4441 mddev->new_level = LEVEL_NONE;
4442 mddev->new_layout = 0;
4443 mddev->new_chunk_sectors = 0;
4444 mddev->curr_resync = 0;
4445 mddev->resync_mismatches = 0;
4446 mddev->suspend_lo = mddev->suspend_hi = 0;
4447 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4448 mddev->recovery = 0;
4449 mddev->in_sync = 0;
4450 mddev->changed = 0;
4451 mddev->degraded = 0;
4452 mddev->barriers_work = 0;
4453 mddev->safemode = 0;
4454 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4455 if (mddev->hold_active == UNTIL_STOP)
4456 mddev->hold_active = 0;
4457
4458 } else if (mddev->pers)
4459 printk(KERN_INFO "md: %s switched to read-only mode.\n",
4460 mdname(mddev));
4461 err = 0;
4462 blk_integrity_unregister(disk);
4463 md_new_event(mddev);
4464 sysfs_notify_dirent(mddev->sysfs_state);
4465 return err;
4466 }
4467
4468 #ifndef MODULE
4469 static void autorun_array(mddev_t *mddev)
4470 {
4471 mdk_rdev_t *rdev;
4472 int err;
4473
4474 if (list_empty(&mddev->disks))
4475 return;
4476
4477 printk(KERN_INFO "md: running: ");
4478
4479 list_for_each_entry(rdev, &mddev->disks, same_set) {
4480 char b[BDEVNAME_SIZE];
4481 printk("<%s>", bdevname(rdev->bdev,b));
4482 }
4483 printk("\n");
4484
4485 err = do_md_run(mddev);
4486 if (err) {
4487 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4488 do_md_stop(mddev, 0, 0);
4489 }
4490 }
4491
4492 /*
4493 * lets try to run arrays based on all disks that have arrived
4494 * until now. (those are in pending_raid_disks)
4495 *
4496 * the method: pick the first pending disk, collect all disks with
4497 * the same UUID, remove all from the pending list and put them into
4498 * the 'same_array' list. Then order this list based on superblock
4499 * update time (freshest comes first), kick out 'old' disks and
4500 * compare superblocks. If everything's fine then run it.
4501 *
4502 * If "unit" is allocated, then bump its reference count
4503 */
4504 static void autorun_devices(int part)
4505 {
4506 mdk_rdev_t *rdev0, *rdev, *tmp;
4507 mddev_t *mddev;
4508 char b[BDEVNAME_SIZE];
4509
4510 printk(KERN_INFO "md: autorun ...\n");
4511 while (!list_empty(&pending_raid_disks)) {
4512 int unit;
4513 dev_t dev;
4514 LIST_HEAD(candidates);
4515 rdev0 = list_entry(pending_raid_disks.next,
4516 mdk_rdev_t, same_set);
4517
4518 printk(KERN_INFO "md: considering %s ...\n",
4519 bdevname(rdev0->bdev,b));
4520 INIT_LIST_HEAD(&candidates);
4521 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4522 if (super_90_load(rdev, rdev0, 0) >= 0) {
4523 printk(KERN_INFO "md: adding %s ...\n",
4524 bdevname(rdev->bdev,b));
4525 list_move(&rdev->same_set, &candidates);
4526 }
4527 /*
4528 * now we have a set of devices, with all of them having
4529 * mostly sane superblocks. It's time to allocate the
4530 * mddev.
4531 */
4532 if (part) {
4533 dev = MKDEV(mdp_major,
4534 rdev0->preferred_minor << MdpMinorShift);
4535 unit = MINOR(dev) >> MdpMinorShift;
4536 } else {
4537 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4538 unit = MINOR(dev);
4539 }
4540 if (rdev0->preferred_minor != unit) {
4541 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4542 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4543 break;
4544 }
4545
4546 md_probe(dev, NULL, NULL);
4547 mddev = mddev_find(dev);
4548 if (!mddev || !mddev->gendisk) {
4549 if (mddev)
4550 mddev_put(mddev);
4551 printk(KERN_ERR
4552 "md: cannot allocate memory for md drive.\n");
4553 break;
4554 }
4555 if (mddev_lock(mddev))
4556 printk(KERN_WARNING "md: %s locked, cannot run\n",
4557 mdname(mddev));
4558 else if (mddev->raid_disks || mddev->major_version
4559 || !list_empty(&mddev->disks)) {
4560 printk(KERN_WARNING
4561 "md: %s already running, cannot run %s\n",
4562 mdname(mddev), bdevname(rdev0->bdev,b));
4563 mddev_unlock(mddev);
4564 } else {
4565 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4566 mddev->persistent = 1;
4567 rdev_for_each_list(rdev, tmp, &candidates) {
4568 list_del_init(&rdev->same_set);
4569 if (bind_rdev_to_array(rdev, mddev))
4570 export_rdev(rdev);
4571 }
4572 autorun_array(mddev);
4573 mddev_unlock(mddev);
4574 }
4575 /* on success, candidates will be empty, on error
4576 * it won't...
4577 */
4578 rdev_for_each_list(rdev, tmp, &candidates) {
4579 list_del_init(&rdev->same_set);
4580 export_rdev(rdev);
4581 }
4582 mddev_put(mddev);
4583 }
4584 printk(KERN_INFO "md: ... autorun DONE.\n");
4585 }
4586 #endif /* !MODULE */
4587
4588 static int get_version(void __user * arg)
4589 {
4590 mdu_version_t ver;
4591
4592 ver.major = MD_MAJOR_VERSION;
4593 ver.minor = MD_MINOR_VERSION;
4594 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4595
4596 if (copy_to_user(arg, &ver, sizeof(ver)))
4597 return -EFAULT;
4598
4599 return 0;
4600 }
4601
4602 static int get_array_info(mddev_t * mddev, void __user * arg)
4603 {
4604 mdu_array_info_t info;
4605 int nr,working,insync,failed,spare;
4606 mdk_rdev_t *rdev;
4607
4608 nr=working=insync=failed=spare=0;
4609 list_for_each_entry(rdev, &mddev->disks, same_set) {
4610 nr++;
4611 if (test_bit(Faulty, &rdev->flags))
4612 failed++;
4613 else {
4614 working++;
4615 if (test_bit(In_sync, &rdev->flags))
4616 insync++;
4617 else
4618 spare++;
4619 }
4620 }
4621
4622 info.major_version = mddev->major_version;
4623 info.minor_version = mddev->minor_version;
4624 info.patch_version = MD_PATCHLEVEL_VERSION;
4625 info.ctime = mddev->ctime;
4626 info.level = mddev->level;
4627 info.size = mddev->dev_sectors / 2;
4628 if (info.size != mddev->dev_sectors / 2) /* overflow */
4629 info.size = -1;
4630 info.nr_disks = nr;
4631 info.raid_disks = mddev->raid_disks;
4632 info.md_minor = mddev->md_minor;
4633 info.not_persistent= !mddev->persistent;
4634
4635 info.utime = mddev->utime;
4636 info.state = 0;
4637 if (mddev->in_sync)
4638 info.state = (1<<MD_SB_CLEAN);
4639 if (mddev->bitmap && mddev->bitmap_offset)
4640 info.state = (1<<MD_SB_BITMAP_PRESENT);
4641 info.active_disks = insync;
4642 info.working_disks = working;
4643 info.failed_disks = failed;
4644 info.spare_disks = spare;
4645
4646 info.layout = mddev->layout;
4647 info.chunk_size = mddev->chunk_sectors << 9;
4648
4649 if (copy_to_user(arg, &info, sizeof(info)))
4650 return -EFAULT;
4651
4652 return 0;
4653 }
4654
4655 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4656 {
4657 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4658 char *ptr, *buf = NULL;
4659 int err = -ENOMEM;
4660
4661 if (md_allow_write(mddev))
4662 file = kmalloc(sizeof(*file), GFP_NOIO);
4663 else
4664 file = kmalloc(sizeof(*file), GFP_KERNEL);
4665
4666 if (!file)
4667 goto out;
4668
4669 /* bitmap disabled, zero the first byte and copy out */
4670 if (!mddev->bitmap || !mddev->bitmap->file) {
4671 file->pathname[0] = '\0';
4672 goto copy_out;
4673 }
4674
4675 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4676 if (!buf)
4677 goto out;
4678
4679 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4680 if (IS_ERR(ptr))
4681 goto out;
4682
4683 strcpy(file->pathname, ptr);
4684
4685 copy_out:
4686 err = 0;
4687 if (copy_to_user(arg, file, sizeof(*file)))
4688 err = -EFAULT;
4689 out:
4690 kfree(buf);
4691 kfree(file);
4692 return err;
4693 }
4694
4695 static int get_disk_info(mddev_t * mddev, void __user * arg)
4696 {
4697 mdu_disk_info_t info;
4698 mdk_rdev_t *rdev;
4699
4700 if (copy_from_user(&info, arg, sizeof(info)))
4701 return -EFAULT;
4702
4703 rdev = find_rdev_nr(mddev, info.number);
4704 if (rdev) {
4705 info.major = MAJOR(rdev->bdev->bd_dev);
4706 info.minor = MINOR(rdev->bdev->bd_dev);
4707 info.raid_disk = rdev->raid_disk;
4708 info.state = 0;
4709 if (test_bit(Faulty, &rdev->flags))
4710 info.state |= (1<<MD_DISK_FAULTY);
4711 else if (test_bit(In_sync, &rdev->flags)) {
4712 info.state |= (1<<MD_DISK_ACTIVE);
4713 info.state |= (1<<MD_DISK_SYNC);
4714 }
4715 if (test_bit(WriteMostly, &rdev->flags))
4716 info.state |= (1<<MD_DISK_WRITEMOSTLY);
4717 } else {
4718 info.major = info.minor = 0;
4719 info.raid_disk = -1;
4720 info.state = (1<<MD_DISK_REMOVED);
4721 }
4722
4723 if (copy_to_user(arg, &info, sizeof(info)))
4724 return -EFAULT;
4725
4726 return 0;
4727 }
4728
4729 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4730 {
4731 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4732 mdk_rdev_t *rdev;
4733 dev_t dev = MKDEV(info->major,info->minor);
4734
4735 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4736 return -EOVERFLOW;
4737
4738 if (!mddev->raid_disks) {
4739 int err;
4740 /* expecting a device which has a superblock */
4741 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4742 if (IS_ERR(rdev)) {
4743 printk(KERN_WARNING
4744 "md: md_import_device returned %ld\n",
4745 PTR_ERR(rdev));
4746 return PTR_ERR(rdev);
4747 }
4748 if (!list_empty(&mddev->disks)) {
4749 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4750 mdk_rdev_t, same_set);
4751 err = super_types[mddev->major_version]
4752 .load_super(rdev, rdev0, mddev->minor_version);
4753 if (err < 0) {
4754 printk(KERN_WARNING
4755 "md: %s has different UUID to %s\n",
4756 bdevname(rdev->bdev,b),
4757 bdevname(rdev0->bdev,b2));
4758 export_rdev(rdev);
4759 return -EINVAL;
4760 }
4761 }
4762 err = bind_rdev_to_array(rdev, mddev);
4763 if (err)
4764 export_rdev(rdev);
4765 return err;
4766 }
4767
4768 /*
4769 * add_new_disk can be used once the array is assembled
4770 * to add "hot spares". They must already have a superblock
4771 * written
4772 */
4773 if (mddev->pers) {
4774 int err;
4775 if (!mddev->pers->hot_add_disk) {
4776 printk(KERN_WARNING
4777 "%s: personality does not support diskops!\n",
4778 mdname(mddev));
4779 return -EINVAL;
4780 }
4781 if (mddev->persistent)
4782 rdev = md_import_device(dev, mddev->major_version,
4783 mddev->minor_version);
4784 else
4785 rdev = md_import_device(dev, -1, -1);
4786 if (IS_ERR(rdev)) {
4787 printk(KERN_WARNING
4788 "md: md_import_device returned %ld\n",
4789 PTR_ERR(rdev));
4790 return PTR_ERR(rdev);
4791 }
4792 /* set save_raid_disk if appropriate */
4793 if (!mddev->persistent) {
4794 if (info->state & (1<<MD_DISK_SYNC) &&
4795 info->raid_disk < mddev->raid_disks)
4796 rdev->raid_disk = info->raid_disk;
4797 else
4798 rdev->raid_disk = -1;
4799 } else
4800 super_types[mddev->major_version].
4801 validate_super(mddev, rdev);
4802 rdev->saved_raid_disk = rdev->raid_disk;
4803
4804 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4805 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4806 set_bit(WriteMostly, &rdev->flags);
4807 else
4808 clear_bit(WriteMostly, &rdev->flags);
4809
4810 rdev->raid_disk = -1;
4811 err = bind_rdev_to_array(rdev, mddev);
4812 if (!err && !mddev->pers->hot_remove_disk) {
4813 /* If there is hot_add_disk but no hot_remove_disk
4814 * then added disks for geometry changes,
4815 * and should be added immediately.
4816 */
4817 super_types[mddev->major_version].
4818 validate_super(mddev, rdev);
4819 err = mddev->pers->hot_add_disk(mddev, rdev);
4820 if (err)
4821 unbind_rdev_from_array(rdev);
4822 }
4823 if (err)
4824 export_rdev(rdev);
4825 else
4826 sysfs_notify_dirent(rdev->sysfs_state);
4827
4828 md_update_sb(mddev, 1);
4829 if (mddev->degraded)
4830 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4831 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4832 md_wakeup_thread(mddev->thread);
4833 return err;
4834 }
4835
4836 /* otherwise, add_new_disk is only allowed
4837 * for major_version==0 superblocks
4838 */
4839 if (mddev->major_version != 0) {
4840 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4841 mdname(mddev));
4842 return -EINVAL;
4843 }
4844
4845 if (!(info->state & (1<<MD_DISK_FAULTY))) {
4846 int err;
4847 rdev = md_import_device(dev, -1, 0);
4848 if (IS_ERR(rdev)) {
4849 printk(KERN_WARNING
4850 "md: error, md_import_device() returned %ld\n",
4851 PTR_ERR(rdev));
4852 return PTR_ERR(rdev);
4853 }
4854 rdev->desc_nr = info->number;
4855 if (info->raid_disk < mddev->raid_disks)
4856 rdev->raid_disk = info->raid_disk;
4857 else
4858 rdev->raid_disk = -1;
4859
4860 if (rdev->raid_disk < mddev->raid_disks)
4861 if (info->state & (1<<MD_DISK_SYNC))
4862 set_bit(In_sync, &rdev->flags);
4863
4864 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4865 set_bit(WriteMostly, &rdev->flags);
4866
4867 if (!mddev->persistent) {
4868 printk(KERN_INFO "md: nonpersistent superblock ...\n");
4869 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4870 } else
4871 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4872 rdev->sectors = rdev->sb_start;
4873
4874 err = bind_rdev_to_array(rdev, mddev);
4875 if (err) {
4876 export_rdev(rdev);
4877 return err;
4878 }
4879 }
4880
4881 return 0;
4882 }
4883
4884 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4885 {
4886 char b[BDEVNAME_SIZE];
4887 mdk_rdev_t *rdev;
4888
4889 rdev = find_rdev(mddev, dev);
4890 if (!rdev)
4891 return -ENXIO;
4892
4893 if (rdev->raid_disk >= 0)
4894 goto busy;
4895
4896 kick_rdev_from_array(rdev);
4897 md_update_sb(mddev, 1);
4898 md_new_event(mddev);
4899
4900 return 0;
4901 busy:
4902 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4903 bdevname(rdev->bdev,b), mdname(mddev));
4904 return -EBUSY;
4905 }
4906
4907 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4908 {
4909 char b[BDEVNAME_SIZE];
4910 int err;
4911 mdk_rdev_t *rdev;
4912
4913 if (!mddev->pers)
4914 return -ENODEV;
4915
4916 if (mddev->major_version != 0) {
4917 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4918 " version-0 superblocks.\n",
4919 mdname(mddev));
4920 return -EINVAL;
4921 }
4922 if (!mddev->pers->hot_add_disk) {
4923 printk(KERN_WARNING
4924 "%s: personality does not support diskops!\n",
4925 mdname(mddev));
4926 return -EINVAL;
4927 }
4928
4929 rdev = md_import_device(dev, -1, 0);
4930 if (IS_ERR(rdev)) {
4931 printk(KERN_WARNING
4932 "md: error, md_import_device() returned %ld\n",
4933 PTR_ERR(rdev));
4934 return -EINVAL;
4935 }
4936
4937 if (mddev->persistent)
4938 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4939 else
4940 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4941
4942 rdev->sectors = rdev->sb_start;
4943
4944 if (test_bit(Faulty, &rdev->flags)) {
4945 printk(KERN_WARNING
4946 "md: can not hot-add faulty %s disk to %s!\n",
4947 bdevname(rdev->bdev,b), mdname(mddev));
4948 err = -EINVAL;
4949 goto abort_export;
4950 }
4951 clear_bit(In_sync, &rdev->flags);
4952 rdev->desc_nr = -1;
4953 rdev->saved_raid_disk = -1;
4954 err = bind_rdev_to_array(rdev, mddev);
4955 if (err)
4956 goto abort_export;
4957
4958 /*
4959 * The rest should better be atomic, we can have disk failures
4960 * noticed in interrupt contexts ...
4961 */
4962
4963 rdev->raid_disk = -1;
4964
4965 md_update_sb(mddev, 1);
4966
4967 /*
4968 * Kick recovery, maybe this spare has to be added to the
4969 * array immediately.
4970 */
4971 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4972 md_wakeup_thread(mddev->thread);
4973 md_new_event(mddev);
4974 return 0;
4975
4976 abort_export:
4977 export_rdev(rdev);
4978 return err;
4979 }
4980
4981 static int set_bitmap_file(mddev_t *mddev, int fd)
4982 {
4983 int err;
4984
4985 if (mddev->pers) {
4986 if (!mddev->pers->quiesce)
4987 return -EBUSY;
4988 if (mddev->recovery || mddev->sync_thread)
4989 return -EBUSY;
4990 /* we should be able to change the bitmap.. */
4991 }
4992
4993
4994 if (fd >= 0) {
4995 if (mddev->bitmap)
4996 return -EEXIST; /* cannot add when bitmap is present */
4997 mddev->bitmap_file = fget(fd);
4998
4999 if (mddev->bitmap_file == NULL) {
5000 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5001 mdname(mddev));
5002 return -EBADF;
5003 }
5004
5005 err = deny_bitmap_write_access(mddev->bitmap_file);
5006 if (err) {
5007 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5008 mdname(mddev));
5009 fput(mddev->bitmap_file);
5010 mddev->bitmap_file = NULL;
5011 return err;
5012 }
5013 mddev->bitmap_offset = 0; /* file overrides offset */
5014 } else if (mddev->bitmap == NULL)
5015 return -ENOENT; /* cannot remove what isn't there */
5016 err = 0;
5017 if (mddev->pers) {
5018 mddev->pers->quiesce(mddev, 1);
5019 if (fd >= 0)
5020 err = bitmap_create(mddev);
5021 if (fd < 0 || err) {
5022 bitmap_destroy(mddev);
5023 fd = -1; /* make sure to put the file */
5024 }
5025 mddev->pers->quiesce(mddev, 0);
5026 }
5027 if (fd < 0) {
5028 if (mddev->bitmap_file) {
5029 restore_bitmap_write_access(mddev->bitmap_file);
5030 fput(mddev->bitmap_file);
5031 }
5032 mddev->bitmap_file = NULL;
5033 }
5034
5035 return err;
5036 }
5037
5038 /*
5039 * set_array_info is used two different ways
5040 * The original usage is when creating a new array.
5041 * In this usage, raid_disks is > 0 and it together with
5042 * level, size, not_persistent,layout,chunksize determine the
5043 * shape of the array.
5044 * This will always create an array with a type-0.90.0 superblock.
5045 * The newer usage is when assembling an array.
5046 * In this case raid_disks will be 0, and the major_version field is
5047 * use to determine which style super-blocks are to be found on the devices.
5048 * The minor and patch _version numbers are also kept incase the
5049 * super_block handler wishes to interpret them.
5050 */
5051 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5052 {
5053
5054 if (info->raid_disks == 0) {
5055 /* just setting version number for superblock loading */
5056 if (info->major_version < 0 ||
5057 info->major_version >= ARRAY_SIZE(super_types) ||
5058 super_types[info->major_version].name == NULL) {
5059 /* maybe try to auto-load a module? */
5060 printk(KERN_INFO
5061 "md: superblock version %d not known\n",
5062 info->major_version);
5063 return -EINVAL;
5064 }
5065 mddev->major_version = info->major_version;
5066 mddev->minor_version = info->minor_version;
5067 mddev->patch_version = info->patch_version;
5068 mddev->persistent = !info->not_persistent;
5069 return 0;
5070 }
5071 mddev->major_version = MD_MAJOR_VERSION;
5072 mddev->minor_version = MD_MINOR_VERSION;
5073 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5074 mddev->ctime = get_seconds();
5075
5076 mddev->level = info->level;
5077 mddev->clevel[0] = 0;
5078 mddev->dev_sectors = 2 * (sector_t)info->size;
5079 mddev->raid_disks = info->raid_disks;
5080 /* don't set md_minor, it is determined by which /dev/md* was
5081 * openned
5082 */
5083 if (info->state & (1<<MD_SB_CLEAN))
5084 mddev->recovery_cp = MaxSector;
5085 else
5086 mddev->recovery_cp = 0;
5087 mddev->persistent = ! info->not_persistent;
5088 mddev->external = 0;
5089
5090 mddev->layout = info->layout;
5091 mddev->chunk_sectors = info->chunk_size >> 9;
5092
5093 mddev->max_disks = MD_SB_DISKS;
5094
5095 if (mddev->persistent)
5096 mddev->flags = 0;
5097 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5098
5099 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
5100 mddev->bitmap_offset = 0;
5101
5102 mddev->reshape_position = MaxSector;
5103
5104 /*
5105 * Generate a 128 bit UUID
5106 */
5107 get_random_bytes(mddev->uuid, 16);
5108
5109 mddev->new_level = mddev->level;
5110 mddev->new_chunk_sectors = mddev->chunk_sectors;
5111 mddev->new_layout = mddev->layout;
5112 mddev->delta_disks = 0;
5113
5114 return 0;
5115 }
5116
5117 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5118 {
5119 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5120
5121 if (mddev->external_size)
5122 return;
5123
5124 mddev->array_sectors = array_sectors;
5125 }
5126 EXPORT_SYMBOL(md_set_array_sectors);
5127
5128 static int update_size(mddev_t *mddev, sector_t num_sectors)
5129 {
5130 mdk_rdev_t *rdev;
5131 int rv;
5132 int fit = (num_sectors == 0);
5133
5134 if (mddev->pers->resize == NULL)
5135 return -EINVAL;
5136 /* The "num_sectors" is the number of sectors of each device that
5137 * is used. This can only make sense for arrays with redundancy.
5138 * linear and raid0 always use whatever space is available. We can only
5139 * consider changing this number if no resync or reconstruction is
5140 * happening, and if the new size is acceptable. It must fit before the
5141 * sb_start or, if that is <data_offset, it must fit before the size
5142 * of each device. If num_sectors is zero, we find the largest size
5143 * that fits.
5144
5145 */
5146 if (mddev->sync_thread)
5147 return -EBUSY;
5148 if (mddev->bitmap)
5149 /* Sorry, cannot grow a bitmap yet, just remove it,
5150 * grow, and re-add.
5151 */
5152 return -EBUSY;
5153 list_for_each_entry(rdev, &mddev->disks, same_set) {
5154 sector_t avail = rdev->sectors;
5155
5156 if (fit && (num_sectors == 0 || num_sectors > avail))
5157 num_sectors = avail;
5158 if (avail < num_sectors)
5159 return -ENOSPC;
5160 }
5161 rv = mddev->pers->resize(mddev, num_sectors);
5162 if (!rv)
5163 revalidate_disk(mddev->gendisk);
5164 return rv;
5165 }
5166
5167 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5168 {
5169 int rv;
5170 /* change the number of raid disks */
5171 if (mddev->pers->check_reshape == NULL)
5172 return -EINVAL;
5173 if (raid_disks <= 0 ||
5174 raid_disks >= mddev->max_disks)
5175 return -EINVAL;
5176 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5177 return -EBUSY;
5178 mddev->delta_disks = raid_disks - mddev->raid_disks;
5179
5180 rv = mddev->pers->check_reshape(mddev);
5181 return rv;
5182 }
5183
5184
5185 /*
5186 * update_array_info is used to change the configuration of an
5187 * on-line array.
5188 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5189 * fields in the info are checked against the array.
5190 * Any differences that cannot be handled will cause an error.
5191 * Normally, only one change can be managed at a time.
5192 */
5193 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5194 {
5195 int rv = 0;
5196 int cnt = 0;
5197 int state = 0;
5198
5199 /* calculate expected state,ignoring low bits */
5200 if (mddev->bitmap && mddev->bitmap_offset)
5201 state |= (1 << MD_SB_BITMAP_PRESENT);
5202
5203 if (mddev->major_version != info->major_version ||
5204 mddev->minor_version != info->minor_version ||
5205 /* mddev->patch_version != info->patch_version || */
5206 mddev->ctime != info->ctime ||
5207 mddev->level != info->level ||
5208 /* mddev->layout != info->layout || */
5209 !mddev->persistent != info->not_persistent||
5210 mddev->chunk_sectors != info->chunk_size >> 9 ||
5211 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5212 ((state^info->state) & 0xfffffe00)
5213 )
5214 return -EINVAL;
5215 /* Check there is only one change */
5216 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5217 cnt++;
5218 if (mddev->raid_disks != info->raid_disks)
5219 cnt++;
5220 if (mddev->layout != info->layout)
5221 cnt++;
5222 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5223 cnt++;
5224 if (cnt == 0)
5225 return 0;
5226 if (cnt > 1)
5227 return -EINVAL;
5228
5229 if (mddev->layout != info->layout) {
5230 /* Change layout
5231 * we don't need to do anything at the md level, the
5232 * personality will take care of it all.
5233 */
5234 if (mddev->pers->check_reshape == NULL)
5235 return -EINVAL;
5236 else {
5237 mddev->new_layout = info->layout;
5238 rv = mddev->pers->check_reshape(mddev);
5239 if (rv)
5240 mddev->new_layout = mddev->layout;
5241 return rv;
5242 }
5243 }
5244 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5245 rv = update_size(mddev, (sector_t)info->size * 2);
5246
5247 if (mddev->raid_disks != info->raid_disks)
5248 rv = update_raid_disks(mddev, info->raid_disks);
5249
5250 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5251 if (mddev->pers->quiesce == NULL)
5252 return -EINVAL;
5253 if (mddev->recovery || mddev->sync_thread)
5254 return -EBUSY;
5255 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5256 /* add the bitmap */
5257 if (mddev->bitmap)
5258 return -EEXIST;
5259 if (mddev->default_bitmap_offset == 0)
5260 return -EINVAL;
5261 mddev->bitmap_offset = mddev->default_bitmap_offset;
5262 mddev->pers->quiesce(mddev, 1);
5263 rv = bitmap_create(mddev);
5264 if (rv)
5265 bitmap_destroy(mddev);
5266 mddev->pers->quiesce(mddev, 0);
5267 } else {
5268 /* remove the bitmap */
5269 if (!mddev->bitmap)
5270 return -ENOENT;
5271 if (mddev->bitmap->file)
5272 return -EINVAL;
5273 mddev->pers->quiesce(mddev, 1);
5274 bitmap_destroy(mddev);
5275 mddev->pers->quiesce(mddev, 0);
5276 mddev->bitmap_offset = 0;
5277 }
5278 }
5279 md_update_sb(mddev, 1);
5280 return rv;
5281 }
5282
5283 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5284 {
5285 mdk_rdev_t *rdev;
5286
5287 if (mddev->pers == NULL)
5288 return -ENODEV;
5289
5290 rdev = find_rdev(mddev, dev);
5291 if (!rdev)
5292 return -ENODEV;
5293
5294 md_error(mddev, rdev);
5295 return 0;
5296 }
5297
5298 /*
5299 * We have a problem here : there is no easy way to give a CHS
5300 * virtual geometry. We currently pretend that we have a 2 heads
5301 * 4 sectors (with a BIG number of cylinders...). This drives
5302 * dosfs just mad... ;-)
5303 */
5304 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5305 {
5306 mddev_t *mddev = bdev->bd_disk->private_data;
5307
5308 geo->heads = 2;
5309 geo->sectors = 4;
5310 geo->cylinders = get_capacity(mddev->gendisk) / 8;
5311 return 0;
5312 }
5313
5314 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5315 unsigned int cmd, unsigned long arg)
5316 {
5317 int err = 0;
5318 void __user *argp = (void __user *)arg;
5319 mddev_t *mddev = NULL;
5320
5321 if (!capable(CAP_SYS_ADMIN))
5322 return -EACCES;
5323
5324 /*
5325 * Commands dealing with the RAID driver but not any
5326 * particular array:
5327 */
5328 switch (cmd)
5329 {
5330 case RAID_VERSION:
5331 err = get_version(argp);
5332 goto done;
5333
5334 case PRINT_RAID_DEBUG:
5335 err = 0;
5336 md_print_devices();
5337 goto done;
5338
5339 #ifndef MODULE
5340 case RAID_AUTORUN:
5341 err = 0;
5342 autostart_arrays(arg);
5343 goto done;
5344 #endif
5345 default:;
5346 }
5347
5348 /*
5349 * Commands creating/starting a new array:
5350 */
5351
5352 mddev = bdev->bd_disk->private_data;
5353
5354 if (!mddev) {
5355 BUG();
5356 goto abort;
5357 }
5358
5359 err = mddev_lock(mddev);
5360 if (err) {
5361 printk(KERN_INFO
5362 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5363 err, cmd);
5364 goto abort;
5365 }
5366
5367 switch (cmd)
5368 {
5369 case SET_ARRAY_INFO:
5370 {
5371 mdu_array_info_t info;
5372 if (!arg)
5373 memset(&info, 0, sizeof(info));
5374 else if (copy_from_user(&info, argp, sizeof(info))) {
5375 err = -EFAULT;
5376 goto abort_unlock;
5377 }
5378 if (mddev->pers) {
5379 err = update_array_info(mddev, &info);
5380 if (err) {
5381 printk(KERN_WARNING "md: couldn't update"
5382 " array info. %d\n", err);
5383 goto abort_unlock;
5384 }
5385 goto done_unlock;
5386 }
5387 if (!list_empty(&mddev->disks)) {
5388 printk(KERN_WARNING
5389 "md: array %s already has disks!\n",
5390 mdname(mddev));
5391 err = -EBUSY;
5392 goto abort_unlock;
5393 }
5394 if (mddev->raid_disks) {
5395 printk(KERN_WARNING
5396 "md: array %s already initialised!\n",
5397 mdname(mddev));
5398 err = -EBUSY;
5399 goto abort_unlock;
5400 }
5401 err = set_array_info(mddev, &info);
5402 if (err) {
5403 printk(KERN_WARNING "md: couldn't set"
5404 " array info. %d\n", err);
5405 goto abort_unlock;
5406 }
5407 }
5408 goto done_unlock;
5409
5410 default:;
5411 }
5412
5413 /*
5414 * Commands querying/configuring an existing array:
5415 */
5416 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5417 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5418 if ((!mddev->raid_disks && !mddev->external)
5419 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5420 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5421 && cmd != GET_BITMAP_FILE) {
5422 err = -ENODEV;
5423 goto abort_unlock;
5424 }
5425
5426 /*
5427 * Commands even a read-only array can execute:
5428 */
5429 switch (cmd)
5430 {
5431 case GET_ARRAY_INFO:
5432 err = get_array_info(mddev, argp);
5433 goto done_unlock;
5434
5435 case GET_BITMAP_FILE:
5436 err = get_bitmap_file(mddev, argp);
5437 goto done_unlock;
5438
5439 case GET_DISK_INFO:
5440 err = get_disk_info(mddev, argp);
5441 goto done_unlock;
5442
5443 case RESTART_ARRAY_RW:
5444 err = restart_array(mddev);
5445 goto done_unlock;
5446
5447 case STOP_ARRAY:
5448 err = do_md_stop(mddev, 0, 1);
5449 goto done_unlock;
5450
5451 case STOP_ARRAY_RO:
5452 err = do_md_stop(mddev, 1, 1);
5453 goto done_unlock;
5454
5455 }
5456
5457 /*
5458 * The remaining ioctls are changing the state of the
5459 * superblock, so we do not allow them on read-only arrays.
5460 * However non-MD ioctls (e.g. get-size) will still come through
5461 * here and hit the 'default' below, so only disallow
5462 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5463 */
5464 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5465 if (mddev->ro == 2) {
5466 mddev->ro = 0;
5467 sysfs_notify_dirent(mddev->sysfs_state);
5468 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5469 md_wakeup_thread(mddev->thread);
5470 } else {
5471 err = -EROFS;
5472 goto abort_unlock;
5473 }
5474 }
5475
5476 switch (cmd)
5477 {
5478 case ADD_NEW_DISK:
5479 {
5480 mdu_disk_info_t info;
5481 if (copy_from_user(&info, argp, sizeof(info)))
5482 err = -EFAULT;
5483 else
5484 err = add_new_disk(mddev, &info);
5485 goto done_unlock;
5486 }
5487
5488 case HOT_REMOVE_DISK:
5489 err = hot_remove_disk(mddev, new_decode_dev(arg));
5490 goto done_unlock;
5491
5492 case HOT_ADD_DISK:
5493 err = hot_add_disk(mddev, new_decode_dev(arg));
5494 goto done_unlock;
5495
5496 case SET_DISK_FAULTY:
5497 err = set_disk_faulty(mddev, new_decode_dev(arg));
5498 goto done_unlock;
5499
5500 case RUN_ARRAY:
5501 err = do_md_run(mddev);
5502 goto done_unlock;
5503
5504 case SET_BITMAP_FILE:
5505 err = set_bitmap_file(mddev, (int)arg);
5506 goto done_unlock;
5507
5508 default:
5509 err = -EINVAL;
5510 goto abort_unlock;
5511 }
5512
5513 done_unlock:
5514 abort_unlock:
5515 if (mddev->hold_active == UNTIL_IOCTL &&
5516 err != -EINVAL)
5517 mddev->hold_active = 0;
5518 mddev_unlock(mddev);
5519
5520 return err;
5521 done:
5522 if (err)
5523 MD_BUG();
5524 abort:
5525 return err;
5526 }
5527
5528 static int md_open(struct block_device *bdev, fmode_t mode)
5529 {
5530 /*
5531 * Succeed if we can lock the mddev, which confirms that
5532 * it isn't being stopped right now.
5533 */
5534 mddev_t *mddev = mddev_find(bdev->bd_dev);
5535 int err;
5536
5537 if (mddev->gendisk != bdev->bd_disk) {
5538 /* we are racing with mddev_put which is discarding this
5539 * bd_disk.
5540 */
5541 mddev_put(mddev);
5542 /* Wait until bdev->bd_disk is definitely gone */
5543 flush_scheduled_work();
5544 /* Then retry the open from the top */
5545 return -ERESTARTSYS;
5546 }
5547 BUG_ON(mddev != bdev->bd_disk->private_data);
5548
5549 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
5550 goto out;
5551
5552 err = 0;
5553 atomic_inc(&mddev->openers);
5554 mutex_unlock(&mddev->open_mutex);
5555
5556 check_disk_change(bdev);
5557 out:
5558 return err;
5559 }
5560
5561 static int md_release(struct gendisk *disk, fmode_t mode)
5562 {
5563 mddev_t *mddev = disk->private_data;
5564
5565 BUG_ON(!mddev);
5566 atomic_dec(&mddev->openers);
5567 mddev_put(mddev);
5568
5569 return 0;
5570 }
5571
5572 static int md_media_changed(struct gendisk *disk)
5573 {
5574 mddev_t *mddev = disk->private_data;
5575
5576 return mddev->changed;
5577 }
5578
5579 static int md_revalidate(struct gendisk *disk)
5580 {
5581 mddev_t *mddev = disk->private_data;
5582
5583 mddev->changed = 0;
5584 return 0;
5585 }
5586 static const struct block_device_operations md_fops =
5587 {
5588 .owner = THIS_MODULE,
5589 .open = md_open,
5590 .release = md_release,
5591 .ioctl = md_ioctl,
5592 .getgeo = md_getgeo,
5593 .media_changed = md_media_changed,
5594 .revalidate_disk= md_revalidate,
5595 };
5596
5597 static int md_thread(void * arg)
5598 {
5599 mdk_thread_t *thread = arg;
5600
5601 /*
5602 * md_thread is a 'system-thread', it's priority should be very
5603 * high. We avoid resource deadlocks individually in each
5604 * raid personality. (RAID5 does preallocation) We also use RR and
5605 * the very same RT priority as kswapd, thus we will never get
5606 * into a priority inversion deadlock.
5607 *
5608 * we definitely have to have equal or higher priority than
5609 * bdflush, otherwise bdflush will deadlock if there are too
5610 * many dirty RAID5 blocks.
5611 */
5612
5613 allow_signal(SIGKILL);
5614 while (!kthread_should_stop()) {
5615
5616 /* We need to wait INTERRUPTIBLE so that
5617 * we don't add to the load-average.
5618 * That means we need to be sure no signals are
5619 * pending
5620 */
5621 if (signal_pending(current))
5622 flush_signals(current);
5623
5624 wait_event_interruptible_timeout
5625 (thread->wqueue,
5626 test_bit(THREAD_WAKEUP, &thread->flags)
5627 || kthread_should_stop(),
5628 thread->timeout);
5629
5630 clear_bit(THREAD_WAKEUP, &thread->flags);
5631
5632 thread->run(thread->mddev);
5633 }
5634
5635 return 0;
5636 }
5637
5638 void md_wakeup_thread(mdk_thread_t *thread)
5639 {
5640 if (thread) {
5641 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5642 set_bit(THREAD_WAKEUP, &thread->flags);
5643 wake_up(&thread->wqueue);
5644 }
5645 }
5646
5647 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5648 const char *name)
5649 {
5650 mdk_thread_t *thread;
5651
5652 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5653 if (!thread)
5654 return NULL;
5655
5656 init_waitqueue_head(&thread->wqueue);
5657
5658 thread->run = run;
5659 thread->mddev = mddev;
5660 thread->timeout = MAX_SCHEDULE_TIMEOUT;
5661 thread->tsk = kthread_run(md_thread, thread,
5662 "%s_%s",
5663 mdname(thread->mddev),
5664 name ?: mddev->pers->name);
5665 if (IS_ERR(thread->tsk)) {
5666 kfree(thread);
5667 return NULL;
5668 }
5669 return thread;
5670 }
5671
5672 void md_unregister_thread(mdk_thread_t *thread)
5673 {
5674 if (!thread)
5675 return;
5676 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5677
5678 kthread_stop(thread->tsk);
5679 kfree(thread);
5680 }
5681
5682 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5683 {
5684 if (!mddev) {
5685 MD_BUG();
5686 return;
5687 }
5688
5689 if (!rdev || test_bit(Faulty, &rdev->flags))
5690 return;
5691
5692 if (mddev->external)
5693 set_bit(Blocked, &rdev->flags);
5694 /*
5695 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5696 mdname(mddev),
5697 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5698 __builtin_return_address(0),__builtin_return_address(1),
5699 __builtin_return_address(2),__builtin_return_address(3));
5700 */
5701 if (!mddev->pers)
5702 return;
5703 if (!mddev->pers->error_handler)
5704 return;
5705 mddev->pers->error_handler(mddev,rdev);
5706 if (mddev->degraded)
5707 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5708 set_bit(StateChanged, &rdev->flags);
5709 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5710 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5711 md_wakeup_thread(mddev->thread);
5712 md_new_event_inintr(mddev);
5713 }
5714
5715 /* seq_file implementation /proc/mdstat */
5716
5717 static void status_unused(struct seq_file *seq)
5718 {
5719 int i = 0;
5720 mdk_rdev_t *rdev;
5721
5722 seq_printf(seq, "unused devices: ");
5723
5724 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
5725 char b[BDEVNAME_SIZE];
5726 i++;
5727 seq_printf(seq, "%s ",
5728 bdevname(rdev->bdev,b));
5729 }
5730 if (!i)
5731 seq_printf(seq, "<none>");
5732
5733 seq_printf(seq, "\n");
5734 }
5735
5736
5737 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5738 {
5739 sector_t max_sectors, resync, res;
5740 unsigned long dt, db;
5741 sector_t rt;
5742 int scale;
5743 unsigned int per_milli;
5744
5745 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
5746
5747 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5748 max_sectors = mddev->resync_max_sectors;
5749 else
5750 max_sectors = mddev->dev_sectors;
5751
5752 /*
5753 * Should not happen.
5754 */
5755 if (!max_sectors) {
5756 MD_BUG();
5757 return;
5758 }
5759 /* Pick 'scale' such that (resync>>scale)*1000 will fit
5760 * in a sector_t, and (max_sectors>>scale) will fit in a
5761 * u32, as those are the requirements for sector_div.
5762 * Thus 'scale' must be at least 10
5763 */
5764 scale = 10;
5765 if (sizeof(sector_t) > sizeof(unsigned long)) {
5766 while ( max_sectors/2 > (1ULL<<(scale+32)))
5767 scale++;
5768 }
5769 res = (resync>>scale)*1000;
5770 sector_div(res, (u32)((max_sectors>>scale)+1));
5771
5772 per_milli = res;
5773 {
5774 int i, x = per_milli/50, y = 20-x;
5775 seq_printf(seq, "[");
5776 for (i = 0; i < x; i++)
5777 seq_printf(seq, "=");
5778 seq_printf(seq, ">");
5779 for (i = 0; i < y; i++)
5780 seq_printf(seq, ".");
5781 seq_printf(seq, "] ");
5782 }
5783 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5784 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5785 "reshape" :
5786 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5787 "check" :
5788 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5789 "resync" : "recovery"))),
5790 per_milli/10, per_milli % 10,
5791 (unsigned long long) resync/2,
5792 (unsigned long long) max_sectors/2);
5793
5794 /*
5795 * dt: time from mark until now
5796 * db: blocks written from mark until now
5797 * rt: remaining time
5798 *
5799 * rt is a sector_t, so could be 32bit or 64bit.
5800 * So we divide before multiply in case it is 32bit and close
5801 * to the limit.
5802 * We scale the divisor (db) by 32 to avoid loosing precision
5803 * near the end of resync when the number of remaining sectors
5804 * is close to 'db'.
5805 * We then divide rt by 32 after multiplying by db to compensate.
5806 * The '+1' avoids division by zero if db is very small.
5807 */
5808 dt = ((jiffies - mddev->resync_mark) / HZ);
5809 if (!dt) dt++;
5810 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5811 - mddev->resync_mark_cnt;
5812
5813 rt = max_sectors - resync; /* number of remaining sectors */
5814 sector_div(rt, db/32+1);
5815 rt *= dt;
5816 rt >>= 5;
5817
5818 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
5819 ((unsigned long)rt % 60)/6);
5820
5821 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5822 }
5823
5824 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5825 {
5826 struct list_head *tmp;
5827 loff_t l = *pos;
5828 mddev_t *mddev;
5829
5830 if (l >= 0x10000)
5831 return NULL;
5832 if (!l--)
5833 /* header */
5834 return (void*)1;
5835
5836 spin_lock(&all_mddevs_lock);
5837 list_for_each(tmp,&all_mddevs)
5838 if (!l--) {
5839 mddev = list_entry(tmp, mddev_t, all_mddevs);
5840 mddev_get(mddev);
5841 spin_unlock(&all_mddevs_lock);
5842 return mddev;
5843 }
5844 spin_unlock(&all_mddevs_lock);
5845 if (!l--)
5846 return (void*)2;/* tail */
5847 return NULL;
5848 }
5849
5850 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5851 {
5852 struct list_head *tmp;
5853 mddev_t *next_mddev, *mddev = v;
5854
5855 ++*pos;
5856 if (v == (void*)2)
5857 return NULL;
5858
5859 spin_lock(&all_mddevs_lock);
5860 if (v == (void*)1)
5861 tmp = all_mddevs.next;
5862 else
5863 tmp = mddev->all_mddevs.next;
5864 if (tmp != &all_mddevs)
5865 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5866 else {
5867 next_mddev = (void*)2;
5868 *pos = 0x10000;
5869 }
5870 spin_unlock(&all_mddevs_lock);
5871
5872 if (v != (void*)1)
5873 mddev_put(mddev);
5874 return next_mddev;
5875
5876 }
5877
5878 static void md_seq_stop(struct seq_file *seq, void *v)
5879 {
5880 mddev_t *mddev = v;
5881
5882 if (mddev && v != (void*)1 && v != (void*)2)
5883 mddev_put(mddev);
5884 }
5885
5886 struct mdstat_info {
5887 int event;
5888 };
5889
5890 static int md_seq_show(struct seq_file *seq, void *v)
5891 {
5892 mddev_t *mddev = v;
5893 sector_t sectors;
5894 mdk_rdev_t *rdev;
5895 struct mdstat_info *mi = seq->private;
5896 struct bitmap *bitmap;
5897
5898 if (v == (void*)1) {
5899 struct mdk_personality *pers;
5900 seq_printf(seq, "Personalities : ");
5901 spin_lock(&pers_lock);
5902 list_for_each_entry(pers, &pers_list, list)
5903 seq_printf(seq, "[%s] ", pers->name);
5904
5905 spin_unlock(&pers_lock);
5906 seq_printf(seq, "\n");
5907 mi->event = atomic_read(&md_event_count);
5908 return 0;
5909 }
5910 if (v == (void*)2) {
5911 status_unused(seq);
5912 return 0;
5913 }
5914
5915 if (mddev_lock(mddev) < 0)
5916 return -EINTR;
5917
5918 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5919 seq_printf(seq, "%s : %sactive", mdname(mddev),
5920 mddev->pers ? "" : "in");
5921 if (mddev->pers) {
5922 if (mddev->ro==1)
5923 seq_printf(seq, " (read-only)");
5924 if (mddev->ro==2)
5925 seq_printf(seq, " (auto-read-only)");
5926 seq_printf(seq, " %s", mddev->pers->name);
5927 }
5928
5929 sectors = 0;
5930 list_for_each_entry(rdev, &mddev->disks, same_set) {
5931 char b[BDEVNAME_SIZE];
5932 seq_printf(seq, " %s[%d]",
5933 bdevname(rdev->bdev,b), rdev->desc_nr);
5934 if (test_bit(WriteMostly, &rdev->flags))
5935 seq_printf(seq, "(W)");
5936 if (test_bit(Faulty, &rdev->flags)) {
5937 seq_printf(seq, "(F)");
5938 continue;
5939 } else if (rdev->raid_disk < 0)
5940 seq_printf(seq, "(S)"); /* spare */
5941 sectors += rdev->sectors;
5942 }
5943
5944 if (!list_empty(&mddev->disks)) {
5945 if (mddev->pers)
5946 seq_printf(seq, "\n %llu blocks",
5947 (unsigned long long)
5948 mddev->array_sectors / 2);
5949 else
5950 seq_printf(seq, "\n %llu blocks",
5951 (unsigned long long)sectors / 2);
5952 }
5953 if (mddev->persistent) {
5954 if (mddev->major_version != 0 ||
5955 mddev->minor_version != 90) {
5956 seq_printf(seq," super %d.%d",
5957 mddev->major_version,
5958 mddev->minor_version);
5959 }
5960 } else if (mddev->external)
5961 seq_printf(seq, " super external:%s",
5962 mddev->metadata_type);
5963 else
5964 seq_printf(seq, " super non-persistent");
5965
5966 if (mddev->pers) {
5967 mddev->pers->status(seq, mddev);
5968 seq_printf(seq, "\n ");
5969 if (mddev->pers->sync_request) {
5970 if (mddev->curr_resync > 2) {
5971 status_resync(seq, mddev);
5972 seq_printf(seq, "\n ");
5973 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5974 seq_printf(seq, "\tresync=DELAYED\n ");
5975 else if (mddev->recovery_cp < MaxSector)
5976 seq_printf(seq, "\tresync=PENDING\n ");
5977 }
5978 } else
5979 seq_printf(seq, "\n ");
5980
5981 if ((bitmap = mddev->bitmap)) {
5982 unsigned long chunk_kb;
5983 unsigned long flags;
5984 spin_lock_irqsave(&bitmap->lock, flags);
5985 chunk_kb = bitmap->chunksize >> 10;
5986 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5987 "%lu%s chunk",
5988 bitmap->pages - bitmap->missing_pages,
5989 bitmap->pages,
5990 (bitmap->pages - bitmap->missing_pages)
5991 << (PAGE_SHIFT - 10),
5992 chunk_kb ? chunk_kb : bitmap->chunksize,
5993 chunk_kb ? "KB" : "B");
5994 if (bitmap->file) {
5995 seq_printf(seq, ", file: ");
5996 seq_path(seq, &bitmap->file->f_path, " \t\n");
5997 }
5998
5999 seq_printf(seq, "\n");
6000 spin_unlock_irqrestore(&bitmap->lock, flags);
6001 }
6002
6003 seq_printf(seq, "\n");
6004 }
6005 mddev_unlock(mddev);
6006
6007 return 0;
6008 }
6009
6010 static const struct seq_operations md_seq_ops = {
6011 .start = md_seq_start,
6012 .next = md_seq_next,
6013 .stop = md_seq_stop,
6014 .show = md_seq_show,
6015 };
6016
6017 static int md_seq_open(struct inode *inode, struct file *file)
6018 {
6019 int error;
6020 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6021 if (mi == NULL)
6022 return -ENOMEM;
6023
6024 error = seq_open(file, &md_seq_ops);
6025 if (error)
6026 kfree(mi);
6027 else {
6028 struct seq_file *p = file->private_data;
6029 p->private = mi;
6030 mi->event = atomic_read(&md_event_count);
6031 }
6032 return error;
6033 }
6034
6035 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6036 {
6037 struct seq_file *m = filp->private_data;
6038 struct mdstat_info *mi = m->private;
6039 int mask;
6040
6041 poll_wait(filp, &md_event_waiters, wait);
6042
6043 /* always allow read */
6044 mask = POLLIN | POLLRDNORM;
6045
6046 if (mi->event != atomic_read(&md_event_count))
6047 mask |= POLLERR | POLLPRI;
6048 return mask;
6049 }
6050
6051 static const struct file_operations md_seq_fops = {
6052 .owner = THIS_MODULE,
6053 .open = md_seq_open,
6054 .read = seq_read,
6055 .llseek = seq_lseek,
6056 .release = seq_release_private,
6057 .poll = mdstat_poll,
6058 };
6059
6060 int register_md_personality(struct mdk_personality *p)
6061 {
6062 spin_lock(&pers_lock);
6063 list_add_tail(&p->list, &pers_list);
6064 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6065 spin_unlock(&pers_lock);
6066 return 0;
6067 }
6068
6069 int unregister_md_personality(struct mdk_personality *p)
6070 {
6071 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6072 spin_lock(&pers_lock);
6073 list_del_init(&p->list);
6074 spin_unlock(&pers_lock);
6075 return 0;
6076 }
6077
6078 static int is_mddev_idle(mddev_t *mddev, int init)
6079 {
6080 mdk_rdev_t * rdev;
6081 int idle;
6082 int curr_events;
6083
6084 idle = 1;
6085 rcu_read_lock();
6086 rdev_for_each_rcu(rdev, mddev) {
6087 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6088 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6089 (int)part_stat_read(&disk->part0, sectors[1]) -
6090 atomic_read(&disk->sync_io);
6091 /* sync IO will cause sync_io to increase before the disk_stats
6092 * as sync_io is counted when a request starts, and
6093 * disk_stats is counted when it completes.
6094 * So resync activity will cause curr_events to be smaller than
6095 * when there was no such activity.
6096 * non-sync IO will cause disk_stat to increase without
6097 * increasing sync_io so curr_events will (eventually)
6098 * be larger than it was before. Once it becomes
6099 * substantially larger, the test below will cause
6100 * the array to appear non-idle, and resync will slow
6101 * down.
6102 * If there is a lot of outstanding resync activity when
6103 * we set last_event to curr_events, then all that activity
6104 * completing might cause the array to appear non-idle
6105 * and resync will be slowed down even though there might
6106 * not have been non-resync activity. This will only
6107 * happen once though. 'last_events' will soon reflect
6108 * the state where there is little or no outstanding
6109 * resync requests, and further resync activity will
6110 * always make curr_events less than last_events.
6111 *
6112 */
6113 if (init || curr_events - rdev->last_events > 64) {
6114 rdev->last_events = curr_events;
6115 idle = 0;
6116 }
6117 }
6118 rcu_read_unlock();
6119 return idle;
6120 }
6121
6122 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6123 {
6124 /* another "blocks" (512byte) blocks have been synced */
6125 atomic_sub(blocks, &mddev->recovery_active);
6126 wake_up(&mddev->recovery_wait);
6127 if (!ok) {
6128 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6129 md_wakeup_thread(mddev->thread);
6130 // stop recovery, signal do_sync ....
6131 }
6132 }
6133
6134
6135 /* md_write_start(mddev, bi)
6136 * If we need to update some array metadata (e.g. 'active' flag
6137 * in superblock) before writing, schedule a superblock update
6138 * and wait for it to complete.
6139 */
6140 void md_write_start(mddev_t *mddev, struct bio *bi)
6141 {
6142 int did_change = 0;
6143 if (bio_data_dir(bi) != WRITE)
6144 return;
6145
6146 BUG_ON(mddev->ro == 1);
6147 if (mddev->ro == 2) {
6148 /* need to switch to read/write */
6149 mddev->ro = 0;
6150 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6151 md_wakeup_thread(mddev->thread);
6152 md_wakeup_thread(mddev->sync_thread);
6153 did_change = 1;
6154 }
6155 atomic_inc(&mddev->writes_pending);
6156 if (mddev->safemode == 1)
6157 mddev->safemode = 0;
6158 if (mddev->in_sync) {
6159 spin_lock_irq(&mddev->write_lock);
6160 if (mddev->in_sync) {
6161 mddev->in_sync = 0;
6162 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6163 md_wakeup_thread(mddev->thread);
6164 did_change = 1;
6165 }
6166 spin_unlock_irq(&mddev->write_lock);
6167 }
6168 if (did_change)
6169 sysfs_notify_dirent(mddev->sysfs_state);
6170 wait_event(mddev->sb_wait,
6171 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
6172 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6173 }
6174
6175 void md_write_end(mddev_t *mddev)
6176 {
6177 if (atomic_dec_and_test(&mddev->writes_pending)) {
6178 if (mddev->safemode == 2)
6179 md_wakeup_thread(mddev->thread);
6180 else if (mddev->safemode_delay)
6181 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6182 }
6183 }
6184
6185 /* md_allow_write(mddev)
6186 * Calling this ensures that the array is marked 'active' so that writes
6187 * may proceed without blocking. It is important to call this before
6188 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6189 * Must be called with mddev_lock held.
6190 *
6191 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6192 * is dropped, so return -EAGAIN after notifying userspace.
6193 */
6194 int md_allow_write(mddev_t *mddev)
6195 {
6196 if (!mddev->pers)
6197 return 0;
6198 if (mddev->ro)
6199 return 0;
6200 if (!mddev->pers->sync_request)
6201 return 0;
6202
6203 spin_lock_irq(&mddev->write_lock);
6204 if (mddev->in_sync) {
6205 mddev->in_sync = 0;
6206 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6207 if (mddev->safemode_delay &&
6208 mddev->safemode == 0)
6209 mddev->safemode = 1;
6210 spin_unlock_irq(&mddev->write_lock);
6211 md_update_sb(mddev, 0);
6212 sysfs_notify_dirent(mddev->sysfs_state);
6213 } else
6214 spin_unlock_irq(&mddev->write_lock);
6215
6216 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
6217 return -EAGAIN;
6218 else
6219 return 0;
6220 }
6221 EXPORT_SYMBOL_GPL(md_allow_write);
6222
6223 #define SYNC_MARKS 10
6224 #define SYNC_MARK_STEP (3*HZ)
6225 void md_do_sync(mddev_t *mddev)
6226 {
6227 mddev_t *mddev2;
6228 unsigned int currspeed = 0,
6229 window;
6230 sector_t max_sectors,j, io_sectors;
6231 unsigned long mark[SYNC_MARKS];
6232 sector_t mark_cnt[SYNC_MARKS];
6233 int last_mark,m;
6234 struct list_head *tmp;
6235 sector_t last_check;
6236 int skipped = 0;
6237 mdk_rdev_t *rdev;
6238 char *desc;
6239
6240 /* just incase thread restarts... */
6241 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6242 return;
6243 if (mddev->ro) /* never try to sync a read-only array */
6244 return;
6245
6246 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6247 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6248 desc = "data-check";
6249 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6250 desc = "requested-resync";
6251 else
6252 desc = "resync";
6253 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6254 desc = "reshape";
6255 else
6256 desc = "recovery";
6257
6258 /* we overload curr_resync somewhat here.
6259 * 0 == not engaged in resync at all
6260 * 2 == checking that there is no conflict with another sync
6261 * 1 == like 2, but have yielded to allow conflicting resync to
6262 * commense
6263 * other == active in resync - this many blocks
6264 *
6265 * Before starting a resync we must have set curr_resync to
6266 * 2, and then checked that every "conflicting" array has curr_resync
6267 * less than ours. When we find one that is the same or higher
6268 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6269 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6270 * This will mean we have to start checking from the beginning again.
6271 *
6272 */
6273
6274 do {
6275 mddev->curr_resync = 2;
6276
6277 try_again:
6278 if (kthread_should_stop()) {
6279 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6280 goto skip;
6281 }
6282 for_each_mddev(mddev2, tmp) {
6283 if (mddev2 == mddev)
6284 continue;
6285 if (!mddev->parallel_resync
6286 && mddev2->curr_resync
6287 && match_mddev_units(mddev, mddev2)) {
6288 DEFINE_WAIT(wq);
6289 if (mddev < mddev2 && mddev->curr_resync == 2) {
6290 /* arbitrarily yield */
6291 mddev->curr_resync = 1;
6292 wake_up(&resync_wait);
6293 }
6294 if (mddev > mddev2 && mddev->curr_resync == 1)
6295 /* no need to wait here, we can wait the next
6296 * time 'round when curr_resync == 2
6297 */
6298 continue;
6299 /* We need to wait 'interruptible' so as not to
6300 * contribute to the load average, and not to
6301 * be caught by 'softlockup'
6302 */
6303 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6304 if (!kthread_should_stop() &&
6305 mddev2->curr_resync >= mddev->curr_resync) {
6306 printk(KERN_INFO "md: delaying %s of %s"
6307 " until %s has finished (they"
6308 " share one or more physical units)\n",
6309 desc, mdname(mddev), mdname(mddev2));
6310 mddev_put(mddev2);
6311 if (signal_pending(current))
6312 flush_signals(current);
6313 schedule();
6314 finish_wait(&resync_wait, &wq);
6315 goto try_again;
6316 }
6317 finish_wait(&resync_wait, &wq);
6318 }
6319 }
6320 } while (mddev->curr_resync < 2);
6321
6322 j = 0;
6323 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6324 /* resync follows the size requested by the personality,
6325 * which defaults to physical size, but can be virtual size
6326 */
6327 max_sectors = mddev->resync_max_sectors;
6328 mddev->resync_mismatches = 0;
6329 /* we don't use the checkpoint if there's a bitmap */
6330 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6331 j = mddev->resync_min;
6332 else if (!mddev->bitmap)
6333 j = mddev->recovery_cp;
6334
6335 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6336 max_sectors = mddev->dev_sectors;
6337 else {
6338 /* recovery follows the physical size of devices */
6339 max_sectors = mddev->dev_sectors;
6340 j = MaxSector;
6341 list_for_each_entry(rdev, &mddev->disks, same_set)
6342 if (rdev->raid_disk >= 0 &&
6343 !test_bit(Faulty, &rdev->flags) &&
6344 !test_bit(In_sync, &rdev->flags) &&
6345 rdev->recovery_offset < j)
6346 j = rdev->recovery_offset;
6347 }
6348
6349 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6350 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6351 " %d KB/sec/disk.\n", speed_min(mddev));
6352 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6353 "(but not more than %d KB/sec) for %s.\n",
6354 speed_max(mddev), desc);
6355
6356 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6357
6358 io_sectors = 0;
6359 for (m = 0; m < SYNC_MARKS; m++) {
6360 mark[m] = jiffies;
6361 mark_cnt[m] = io_sectors;
6362 }
6363 last_mark = 0;
6364 mddev->resync_mark = mark[last_mark];
6365 mddev->resync_mark_cnt = mark_cnt[last_mark];
6366
6367 /*
6368 * Tune reconstruction:
6369 */
6370 window = 32*(PAGE_SIZE/512);
6371 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6372 window/2,(unsigned long long) max_sectors/2);
6373
6374 atomic_set(&mddev->recovery_active, 0);
6375 last_check = 0;
6376
6377 if (j>2) {
6378 printk(KERN_INFO
6379 "md: resuming %s of %s from checkpoint.\n",
6380 desc, mdname(mddev));
6381 mddev->curr_resync = j;
6382 }
6383
6384 while (j < max_sectors) {
6385 sector_t sectors;
6386
6387 skipped = 0;
6388
6389 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6390 ((mddev->curr_resync > mddev->curr_resync_completed &&
6391 (mddev->curr_resync - mddev->curr_resync_completed)
6392 > (max_sectors >> 4)) ||
6393 (j - mddev->curr_resync_completed)*2
6394 >= mddev->resync_max - mddev->curr_resync_completed
6395 )) {
6396 /* time to update curr_resync_completed */
6397 blk_unplug(mddev->queue);
6398 wait_event(mddev->recovery_wait,
6399 atomic_read(&mddev->recovery_active) == 0);
6400 mddev->curr_resync_completed =
6401 mddev->curr_resync;
6402 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6403 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6404 }
6405
6406 while (j >= mddev->resync_max && !kthread_should_stop()) {
6407 /* As this condition is controlled by user-space,
6408 * we can block indefinitely, so use '_interruptible'
6409 * to avoid triggering warnings.
6410 */
6411 flush_signals(current); /* just in case */
6412 wait_event_interruptible(mddev->recovery_wait,
6413 mddev->resync_max > j
6414 || kthread_should_stop());
6415 }
6416
6417 if (kthread_should_stop())
6418 goto interrupted;
6419
6420 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6421 currspeed < speed_min(mddev));
6422 if (sectors == 0) {
6423 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6424 goto out;
6425 }
6426
6427 if (!skipped) { /* actual IO requested */
6428 io_sectors += sectors;
6429 atomic_add(sectors, &mddev->recovery_active);
6430 }
6431
6432 j += sectors;
6433 if (j>1) mddev->curr_resync = j;
6434 mddev->curr_mark_cnt = io_sectors;
6435 if (last_check == 0)
6436 /* this is the earliers that rebuilt will be
6437 * visible in /proc/mdstat
6438 */
6439 md_new_event(mddev);
6440
6441 if (last_check + window > io_sectors || j == max_sectors)
6442 continue;
6443
6444 last_check = io_sectors;
6445
6446 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6447 break;
6448
6449 repeat:
6450 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6451 /* step marks */
6452 int next = (last_mark+1) % SYNC_MARKS;
6453
6454 mddev->resync_mark = mark[next];
6455 mddev->resync_mark_cnt = mark_cnt[next];
6456 mark[next] = jiffies;
6457 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6458 last_mark = next;
6459 }
6460
6461
6462 if (kthread_should_stop())
6463 goto interrupted;
6464
6465
6466 /*
6467 * this loop exits only if either when we are slower than
6468 * the 'hard' speed limit, or the system was IO-idle for
6469 * a jiffy.
6470 * the system might be non-idle CPU-wise, but we only care
6471 * about not overloading the IO subsystem. (things like an
6472 * e2fsck being done on the RAID array should execute fast)
6473 */
6474 blk_unplug(mddev->queue);
6475 cond_resched();
6476
6477 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6478 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6479
6480 if (currspeed > speed_min(mddev)) {
6481 if ((currspeed > speed_max(mddev)) ||
6482 !is_mddev_idle(mddev, 0)) {
6483 msleep(500);
6484 goto repeat;
6485 }
6486 }
6487 }
6488 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6489 /*
6490 * this also signals 'finished resyncing' to md_stop
6491 */
6492 out:
6493 blk_unplug(mddev->queue);
6494
6495 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6496
6497 /* tell personality that we are finished */
6498 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6499
6500 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6501 mddev->curr_resync > 2) {
6502 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6503 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6504 if (mddev->curr_resync >= mddev->recovery_cp) {
6505 printk(KERN_INFO
6506 "md: checkpointing %s of %s.\n",
6507 desc, mdname(mddev));
6508 mddev->recovery_cp = mddev->curr_resync;
6509 }
6510 } else
6511 mddev->recovery_cp = MaxSector;
6512 } else {
6513 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6514 mddev->curr_resync = MaxSector;
6515 list_for_each_entry(rdev, &mddev->disks, same_set)
6516 if (rdev->raid_disk >= 0 &&
6517 !test_bit(Faulty, &rdev->flags) &&
6518 !test_bit(In_sync, &rdev->flags) &&
6519 rdev->recovery_offset < mddev->curr_resync)
6520 rdev->recovery_offset = mddev->curr_resync;
6521 }
6522 }
6523 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6524
6525 skip:
6526 mddev->curr_resync = 0;
6527 mddev->curr_resync_completed = 0;
6528 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6529 /* We completed so max setting can be forgotten. */
6530 mddev->resync_max = MaxSector;
6531 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6532 wake_up(&resync_wait);
6533 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6534 md_wakeup_thread(mddev->thread);
6535 return;
6536
6537 interrupted:
6538 /*
6539 * got a signal, exit.
6540 */
6541 printk(KERN_INFO
6542 "md: md_do_sync() got signal ... exiting\n");
6543 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6544 goto out;
6545
6546 }
6547 EXPORT_SYMBOL_GPL(md_do_sync);
6548
6549
6550 static int remove_and_add_spares(mddev_t *mddev)
6551 {
6552 mdk_rdev_t *rdev;
6553 int spares = 0;
6554
6555 mddev->curr_resync_completed = 0;
6556
6557 list_for_each_entry(rdev, &mddev->disks, same_set)
6558 if (rdev->raid_disk >= 0 &&
6559 !test_bit(Blocked, &rdev->flags) &&
6560 (test_bit(Faulty, &rdev->flags) ||
6561 ! test_bit(In_sync, &rdev->flags)) &&
6562 atomic_read(&rdev->nr_pending)==0) {
6563 if (mddev->pers->hot_remove_disk(
6564 mddev, rdev->raid_disk)==0) {
6565 char nm[20];
6566 sprintf(nm,"rd%d", rdev->raid_disk);
6567 sysfs_remove_link(&mddev->kobj, nm);
6568 rdev->raid_disk = -1;
6569 }
6570 }
6571
6572 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
6573 list_for_each_entry(rdev, &mddev->disks, same_set) {
6574 if (rdev->raid_disk >= 0 &&
6575 !test_bit(In_sync, &rdev->flags) &&
6576 !test_bit(Blocked, &rdev->flags))
6577 spares++;
6578 if (rdev->raid_disk < 0
6579 && !test_bit(Faulty, &rdev->flags)) {
6580 rdev->recovery_offset = 0;
6581 if (mddev->pers->
6582 hot_add_disk(mddev, rdev) == 0) {
6583 char nm[20];
6584 sprintf(nm, "rd%d", rdev->raid_disk);
6585 if (sysfs_create_link(&mddev->kobj,
6586 &rdev->kobj, nm))
6587 printk(KERN_WARNING
6588 "md: cannot register "
6589 "%s for %s\n",
6590 nm, mdname(mddev));
6591 spares++;
6592 md_new_event(mddev);
6593 } else
6594 break;
6595 }
6596 }
6597 }
6598 return spares;
6599 }
6600 /*
6601 * This routine is regularly called by all per-raid-array threads to
6602 * deal with generic issues like resync and super-block update.
6603 * Raid personalities that don't have a thread (linear/raid0) do not
6604 * need this as they never do any recovery or update the superblock.
6605 *
6606 * It does not do any resync itself, but rather "forks" off other threads
6607 * to do that as needed.
6608 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6609 * "->recovery" and create a thread at ->sync_thread.
6610 * When the thread finishes it sets MD_RECOVERY_DONE
6611 * and wakeups up this thread which will reap the thread and finish up.
6612 * This thread also removes any faulty devices (with nr_pending == 0).
6613 *
6614 * The overall approach is:
6615 * 1/ if the superblock needs updating, update it.
6616 * 2/ If a recovery thread is running, don't do anything else.
6617 * 3/ If recovery has finished, clean up, possibly marking spares active.
6618 * 4/ If there are any faulty devices, remove them.
6619 * 5/ If array is degraded, try to add spares devices
6620 * 6/ If array has spares or is not in-sync, start a resync thread.
6621 */
6622 void md_check_recovery(mddev_t *mddev)
6623 {
6624 mdk_rdev_t *rdev;
6625
6626
6627 if (mddev->bitmap)
6628 bitmap_daemon_work(mddev->bitmap);
6629
6630 if (mddev->ro)
6631 return;
6632
6633 if (signal_pending(current)) {
6634 if (mddev->pers->sync_request && !mddev->external) {
6635 printk(KERN_INFO "md: %s in immediate safe mode\n",
6636 mdname(mddev));
6637 mddev->safemode = 2;
6638 }
6639 flush_signals(current);
6640 }
6641
6642 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
6643 return;
6644 if ( ! (
6645 (mddev->flags && !mddev->external) ||
6646 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
6647 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
6648 (mddev->external == 0 && mddev->safemode == 1) ||
6649 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6650 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6651 ))
6652 return;
6653
6654 if (mddev_trylock(mddev)) {
6655 int spares = 0;
6656
6657 if (mddev->ro) {
6658 /* Only thing we do on a ro array is remove
6659 * failed devices.
6660 */
6661 remove_and_add_spares(mddev);
6662 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6663 goto unlock;
6664 }
6665
6666 if (!mddev->external) {
6667 int did_change = 0;
6668 spin_lock_irq(&mddev->write_lock);
6669 if (mddev->safemode &&
6670 !atomic_read(&mddev->writes_pending) &&
6671 !mddev->in_sync &&
6672 mddev->recovery_cp == MaxSector) {
6673 mddev->in_sync = 1;
6674 did_change = 1;
6675 if (mddev->persistent)
6676 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6677 }
6678 if (mddev->safemode == 1)
6679 mddev->safemode = 0;
6680 spin_unlock_irq(&mddev->write_lock);
6681 if (did_change)
6682 sysfs_notify_dirent(mddev->sysfs_state);
6683 }
6684
6685 if (mddev->flags)
6686 md_update_sb(mddev, 0);
6687
6688 list_for_each_entry(rdev, &mddev->disks, same_set)
6689 if (test_and_clear_bit(StateChanged, &rdev->flags))
6690 sysfs_notify_dirent(rdev->sysfs_state);
6691
6692
6693 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6694 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6695 /* resync/recovery still happening */
6696 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6697 goto unlock;
6698 }
6699 if (mddev->sync_thread) {
6700 /* resync has finished, collect result */
6701 md_unregister_thread(mddev->sync_thread);
6702 mddev->sync_thread = NULL;
6703 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
6704 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
6705 /* success...*/
6706 /* activate any spares */
6707 if (mddev->pers->spare_active(mddev))
6708 sysfs_notify(&mddev->kobj, NULL,
6709 "degraded");
6710 }
6711 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6712 mddev->pers->finish_reshape)
6713 mddev->pers->finish_reshape(mddev);
6714 md_update_sb(mddev, 1);
6715
6716 /* if array is no-longer degraded, then any saved_raid_disk
6717 * information must be scrapped
6718 */
6719 if (!mddev->degraded)
6720 list_for_each_entry(rdev, &mddev->disks, same_set)
6721 rdev->saved_raid_disk = -1;
6722
6723 mddev->recovery = 0;
6724 /* flag recovery needed just to double check */
6725 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6726 sysfs_notify_dirent(mddev->sysfs_action);
6727 md_new_event(mddev);
6728 goto unlock;
6729 }
6730 /* Set RUNNING before clearing NEEDED to avoid
6731 * any transients in the value of "sync_action".
6732 */
6733 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6734 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6735 /* Clear some bits that don't mean anything, but
6736 * might be left set
6737 */
6738 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
6739 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
6740
6741 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6742 goto unlock;
6743 /* no recovery is running.
6744 * remove any failed drives, then
6745 * add spares if possible.
6746 * Spare are also removed and re-added, to allow
6747 * the personality to fail the re-add.
6748 */
6749
6750 if (mddev->reshape_position != MaxSector) {
6751 if (mddev->pers->check_reshape == NULL ||
6752 mddev->pers->check_reshape(mddev) != 0)
6753 /* Cannot proceed */
6754 goto unlock;
6755 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6756 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6757 } else if ((spares = remove_and_add_spares(mddev))) {
6758 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6759 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6760 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
6761 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6762 } else if (mddev->recovery_cp < MaxSector) {
6763 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6764 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6765 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6766 /* nothing to be done ... */
6767 goto unlock;
6768
6769 if (mddev->pers->sync_request) {
6770 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6771 /* We are adding a device or devices to an array
6772 * which has the bitmap stored on all devices.
6773 * So make sure all bitmap pages get written
6774 */
6775 bitmap_write_all(mddev->bitmap);
6776 }
6777 mddev->sync_thread = md_register_thread(md_do_sync,
6778 mddev,
6779 "resync");
6780 if (!mddev->sync_thread) {
6781 printk(KERN_ERR "%s: could not start resync"
6782 " thread...\n",
6783 mdname(mddev));
6784 /* leave the spares where they are, it shouldn't hurt */
6785 mddev->recovery = 0;
6786 } else
6787 md_wakeup_thread(mddev->sync_thread);
6788 sysfs_notify_dirent(mddev->sysfs_action);
6789 md_new_event(mddev);
6790 }
6791 unlock:
6792 if (!mddev->sync_thread) {
6793 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6794 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
6795 &mddev->recovery))
6796 if (mddev->sysfs_action)
6797 sysfs_notify_dirent(mddev->sysfs_action);
6798 }
6799 mddev_unlock(mddev);
6800 }
6801 }
6802
6803 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
6804 {
6805 sysfs_notify_dirent(rdev->sysfs_state);
6806 wait_event_timeout(rdev->blocked_wait,
6807 !test_bit(Blocked, &rdev->flags),
6808 msecs_to_jiffies(5000));
6809 rdev_dec_pending(rdev, mddev);
6810 }
6811 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
6812
6813 static int md_notify_reboot(struct notifier_block *this,
6814 unsigned long code, void *x)
6815 {
6816 struct list_head *tmp;
6817 mddev_t *mddev;
6818
6819 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
6820
6821 printk(KERN_INFO "md: stopping all md devices.\n");
6822
6823 for_each_mddev(mddev, tmp)
6824 if (mddev_trylock(mddev)) {
6825 /* Force a switch to readonly even array
6826 * appears to still be in use. Hence
6827 * the '100'.
6828 */
6829 do_md_stop(mddev, 1, 100);
6830 mddev_unlock(mddev);
6831 }
6832 /*
6833 * certain more exotic SCSI devices are known to be
6834 * volatile wrt too early system reboots. While the
6835 * right place to handle this issue is the given
6836 * driver, we do want to have a safe RAID driver ...
6837 */
6838 mdelay(1000*1);
6839 }
6840 return NOTIFY_DONE;
6841 }
6842
6843 static struct notifier_block md_notifier = {
6844 .notifier_call = md_notify_reboot,
6845 .next = NULL,
6846 .priority = INT_MAX, /* before any real devices */
6847 };
6848
6849 static void md_geninit(void)
6850 {
6851 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
6852
6853 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
6854 }
6855
6856 static int __init md_init(void)
6857 {
6858 if (register_blkdev(MD_MAJOR, "md"))
6859 return -1;
6860 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
6861 unregister_blkdev(MD_MAJOR, "md");
6862 return -1;
6863 }
6864 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
6865 md_probe, NULL, NULL);
6866 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
6867 md_probe, NULL, NULL);
6868
6869 register_reboot_notifier(&md_notifier);
6870 raid_table_header = register_sysctl_table(raid_root_table);
6871
6872 md_geninit();
6873 return 0;
6874 }
6875
6876
6877 #ifndef MODULE
6878
6879 /*
6880 * Searches all registered partitions for autorun RAID arrays
6881 * at boot time.
6882 */
6883
6884 static LIST_HEAD(all_detected_devices);
6885 struct detected_devices_node {
6886 struct list_head list;
6887 dev_t dev;
6888 };
6889
6890 void md_autodetect_dev(dev_t dev)
6891 {
6892 struct detected_devices_node *node_detected_dev;
6893
6894 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6895 if (node_detected_dev) {
6896 node_detected_dev->dev = dev;
6897 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6898 } else {
6899 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6900 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6901 }
6902 }
6903
6904
6905 static void autostart_arrays(int part)
6906 {
6907 mdk_rdev_t *rdev;
6908 struct detected_devices_node *node_detected_dev;
6909 dev_t dev;
6910 int i_scanned, i_passed;
6911
6912 i_scanned = 0;
6913 i_passed = 0;
6914
6915 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6916
6917 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6918 i_scanned++;
6919 node_detected_dev = list_entry(all_detected_devices.next,
6920 struct detected_devices_node, list);
6921 list_del(&node_detected_dev->list);
6922 dev = node_detected_dev->dev;
6923 kfree(node_detected_dev);
6924 rdev = md_import_device(dev,0, 90);
6925 if (IS_ERR(rdev))
6926 continue;
6927
6928 if (test_bit(Faulty, &rdev->flags)) {
6929 MD_BUG();
6930 continue;
6931 }
6932 set_bit(AutoDetected, &rdev->flags);
6933 list_add(&rdev->same_set, &pending_raid_disks);
6934 i_passed++;
6935 }
6936
6937 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6938 i_scanned, i_passed);
6939
6940 autorun_devices(part);
6941 }
6942
6943 #endif /* !MODULE */
6944
6945 static __exit void md_exit(void)
6946 {
6947 mddev_t *mddev;
6948 struct list_head *tmp;
6949
6950 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
6951 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6952
6953 unregister_blkdev(MD_MAJOR,"md");
6954 unregister_blkdev(mdp_major, "mdp");
6955 unregister_reboot_notifier(&md_notifier);
6956 unregister_sysctl_table(raid_table_header);
6957 remove_proc_entry("mdstat", NULL);
6958 for_each_mddev(mddev, tmp) {
6959 export_array(mddev);
6960 mddev->hold_active = 0;
6961 }
6962 }
6963
6964 subsys_initcall(md_init);
6965 module_exit(md_exit)
6966
6967 static int get_ro(char *buffer, struct kernel_param *kp)
6968 {
6969 return sprintf(buffer, "%d", start_readonly);
6970 }
6971 static int set_ro(const char *val, struct kernel_param *kp)
6972 {
6973 char *e;
6974 int num = simple_strtoul(val, &e, 10);
6975 if (*val && (*e == '\0' || *e == '\n')) {
6976 start_readonly = num;
6977 return 0;
6978 }
6979 return -EINVAL;
6980 }
6981
6982 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6983 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6984
6985 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
6986
6987 EXPORT_SYMBOL(register_md_personality);
6988 EXPORT_SYMBOL(unregister_md_personality);
6989 EXPORT_SYMBOL(md_error);
6990 EXPORT_SYMBOL(md_done_sync);
6991 EXPORT_SYMBOL(md_write_start);
6992 EXPORT_SYMBOL(md_write_end);
6993 EXPORT_SYMBOL(md_register_thread);
6994 EXPORT_SYMBOL(md_unregister_thread);
6995 EXPORT_SYMBOL(md_wakeup_thread);
6996 EXPORT_SYMBOL(md_check_recovery);
6997 MODULE_LICENSE("GPL");
6998 MODULE_ALIAS("md");
6999 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
WandBoard kernel