?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'i2c-fixes-rc6' of git://aeryn.fluff.org.uk/bjdooks/linux
[linux-2.6/mini2440.git] / drivers / md / md.c
blob9dd872000cec843db28fd393f2eb3023725c4bf4
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 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
102 .procname = "speed_limit_min",
103 .data = &sysctl_speed_limit_min,
104 .maxlen = sizeof(int),
105 .mode = S_IRUGO|S_IWUSR,
106 .proc_handler = &proc_dointvec,
107 },
108 {
109 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
110 .procname = "speed_limit_max",
111 .data = &sysctl_speed_limit_max,
112 .maxlen = sizeof(int),
113 .mode = S_IRUGO|S_IWUSR,
114 .proc_handler = &proc_dointvec,
115 },
116 { .ctl_name = 0 }
117 };
118
119 static ctl_table raid_dir_table[] = {
120 {
121 .ctl_name = DEV_RAID,
122 .procname = "raid",
123 .maxlen = 0,
124 .mode = S_IRUGO|S_IXUGO,
125 .child = raid_table,
126 },
127 { .ctl_name = 0 }
128 };
129
130 static ctl_table raid_root_table[] = {
131 {
132 .ctl_name = CTL_DEV,
133 .procname = "dev",
134 .maxlen = 0,
135 .mode = 0555,
136 .child = raid_dir_table,
137 },
138 { .ctl_name = 0 }
139 };
140
141 static struct block_device_operations md_fops;
142
143 static int start_readonly;
144
145 /*
146 * We have a system wide 'event count' that is incremented
147 * on any 'interesting' event, and readers of /proc/mdstat
148 * can use 'poll' or 'select' to find out when the event
149 * count increases.
150 *
151 * Events are:
152 * start array, stop array, error, add device, remove device,
153 * start build, activate spare
154 */
155 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
156 static atomic_t md_event_count;
157 void md_new_event(mddev_t *mddev)
158 {
159 atomic_inc(&md_event_count);
160 wake_up(&md_event_waiters);
161 }
162 EXPORT_SYMBOL_GPL(md_new_event);
163
164 /* Alternate version that can be called from interrupts
165 * when calling sysfs_notify isn't needed.
166 */
167 static void md_new_event_inintr(mddev_t *mddev)
168 {
169 atomic_inc(&md_event_count);
170 wake_up(&md_event_waiters);
171 }
172
173 /*
174 * Enables to iterate over all existing md arrays
175 * all_mddevs_lock protects this list.
176 */
177 static LIST_HEAD(all_mddevs);
178 static DEFINE_SPINLOCK(all_mddevs_lock);
179
180
181 /*
182 * iterates through all used mddevs in the system.
183 * We take care to grab the all_mddevs_lock whenever navigating
184 * the list, and to always hold a refcount when unlocked.
185 * Any code which breaks out of this loop while own
186 * a reference to the current mddev and must mddev_put it.
187 */
188 #define for_each_mddev(mddev,tmp) \
189 \
190 for (({ spin_lock(&all_mddevs_lock); \
191 tmp = all_mddevs.next; \
192 mddev = NULL;}); \
193 ({ if (tmp != &all_mddevs) \
194 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
195 spin_unlock(&all_mddevs_lock); \
196 if (mddev) mddev_put(mddev); \
197 mddev = list_entry(tmp, mddev_t, all_mddevs); \
198 tmp != &all_mddevs;}); \
199 ({ spin_lock(&all_mddevs_lock); \
200 tmp = tmp->next;}) \
201 )
202
203
204 /* Rather than calling directly into the personality make_request function,
205 * IO requests come here first so that we can check if the device is
206 * being suspended pending a reconfiguration.
207 * We hold a refcount over the call to ->make_request. By the time that
208 * call has finished, the bio has been linked into some internal structure
209 * and so is visible to ->quiesce(), so we don't need the refcount any more.
210 */
211 static int md_make_request(struct request_queue *q, struct bio *bio)
212 {
213 mddev_t *mddev = q->queuedata;
214 int rv;
215 if (mddev == NULL || mddev->pers == NULL) {
216 bio_io_error(bio);
217 return 0;
218 }
219 rcu_read_lock();
220 if (mddev->suspended) {
221 DEFINE_WAIT(__wait);
222 for (;;) {
223 prepare_to_wait(&mddev->sb_wait, &__wait,
224 TASK_UNINTERRUPTIBLE);
225 if (!mddev->suspended)
226 break;
227 rcu_read_unlock();
228 schedule();
229 rcu_read_lock();
230 }
231 finish_wait(&mddev->sb_wait, &__wait);
232 }
233 atomic_inc(&mddev->active_io);
234 rcu_read_unlock();
235 rv = mddev->pers->make_request(q, bio);
236 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
237 wake_up(&mddev->sb_wait);
238
239 return rv;
240 }
241
242 static void mddev_suspend(mddev_t *mddev)
243 {
244 BUG_ON(mddev->suspended);
245 mddev->suspended = 1;
246 synchronize_rcu();
247 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
248 mddev->pers->quiesce(mddev, 1);
249 md_unregister_thread(mddev->thread);
250 mddev->thread = NULL;
251 /* we now know that no code is executing in the personality module,
252 * except possibly the tail end of a ->bi_end_io function, but that
253 * is certain to complete before the module has a chance to get
254 * unloaded
255 */
256 }
257
258 static void mddev_resume(mddev_t *mddev)
259 {
260 mddev->suspended = 0;
261 wake_up(&mddev->sb_wait);
262 mddev->pers->quiesce(mddev, 0);
263 }
264
265
266 static inline mddev_t *mddev_get(mddev_t *mddev)
267 {
268 atomic_inc(&mddev->active);
269 return mddev;
270 }
271
272 static void mddev_delayed_delete(struct work_struct *ws);
273
274 static void mddev_put(mddev_t *mddev)
275 {
276 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
277 return;
278 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
279 !mddev->hold_active) {
280 list_del(&mddev->all_mddevs);
281 if (mddev->gendisk) {
282 /* we did a probe so need to clean up.
283 * Call schedule_work inside the spinlock
284 * so that flush_scheduled_work() after
285 * mddev_find will succeed in waiting for the
286 * work to be done.
287 */
288 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
289 schedule_work(&mddev->del_work);
290 } else
291 kfree(mddev);
292 }
293 spin_unlock(&all_mddevs_lock);
294 }
295
296 static mddev_t * mddev_find(dev_t unit)
297 {
298 mddev_t *mddev, *new = NULL;
299
300 retry:
301 spin_lock(&all_mddevs_lock);
302
303 if (unit) {
304 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
305 if (mddev->unit == unit) {
306 mddev_get(mddev);
307 spin_unlock(&all_mddevs_lock);
308 kfree(new);
309 return mddev;
310 }
311
312 if (new) {
313 list_add(&new->all_mddevs, &all_mddevs);
314 spin_unlock(&all_mddevs_lock);
315 new->hold_active = UNTIL_IOCTL;
316 return new;
317 }
318 } else if (new) {
319 /* find an unused unit number */
320 static int next_minor = 512;
321 int start = next_minor;
322 int is_free = 0;
323 int dev = 0;
324 while (!is_free) {
325 dev = MKDEV(MD_MAJOR, next_minor);
326 next_minor++;
327 if (next_minor > MINORMASK)
328 next_minor = 0;
329 if (next_minor == start) {
330 /* Oh dear, all in use. */
331 spin_unlock(&all_mddevs_lock);
332 kfree(new);
333 return NULL;
334 }
335
336 is_free = 1;
337 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
338 if (mddev->unit == dev) {
339 is_free = 0;
340 break;
341 }
342 }
343 new->unit = dev;
344 new->md_minor = MINOR(dev);
345 new->hold_active = UNTIL_STOP;
346 list_add(&new->all_mddevs, &all_mddevs);
347 spin_unlock(&all_mddevs_lock);
348 return new;
349 }
350 spin_unlock(&all_mddevs_lock);
351
352 new = kzalloc(sizeof(*new), GFP_KERNEL);
353 if (!new)
354 return NULL;
355
356 new->unit = unit;
357 if (MAJOR(unit) == MD_MAJOR)
358 new->md_minor = MINOR(unit);
359 else
360 new->md_minor = MINOR(unit) >> MdpMinorShift;
361
362 mutex_init(&new->open_mutex);
363 mutex_init(&new->reconfig_mutex);
364 INIT_LIST_HEAD(&new->disks);
365 INIT_LIST_HEAD(&new->all_mddevs);
366 init_timer(&new->safemode_timer);
367 atomic_set(&new->active, 1);
368 atomic_set(&new->openers, 0);
369 atomic_set(&new->active_io, 0);
370 spin_lock_init(&new->write_lock);
371 init_waitqueue_head(&new->sb_wait);
372 init_waitqueue_head(&new->recovery_wait);
373 new->reshape_position = MaxSector;
374 new->resync_min = 0;
375 new->resync_max = MaxSector;
376 new->level = LEVEL_NONE;
377
378 goto retry;
379 }
380
381 static inline int mddev_lock(mddev_t * mddev)
382 {
383 return mutex_lock_interruptible(&mddev->reconfig_mutex);
384 }
385
386 static inline int mddev_is_locked(mddev_t *mddev)
387 {
388 return mutex_is_locked(&mddev->reconfig_mutex);
389 }
390
391 static inline int mddev_trylock(mddev_t * mddev)
392 {
393 return mutex_trylock(&mddev->reconfig_mutex);
394 }
395
396 static inline void mddev_unlock(mddev_t * mddev)
397 {
398 mutex_unlock(&mddev->reconfig_mutex);
399
400 md_wakeup_thread(mddev->thread);
401 }
402
403 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
404 {
405 mdk_rdev_t *rdev;
406
407 list_for_each_entry(rdev, &mddev->disks, same_set)
408 if (rdev->desc_nr == nr)
409 return rdev;
410
411 return NULL;
412 }
413
414 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
415 {
416 mdk_rdev_t *rdev;
417
418 list_for_each_entry(rdev, &mddev->disks, same_set)
419 if (rdev->bdev->bd_dev == dev)
420 return rdev;
421
422 return NULL;
423 }
424
425 static struct mdk_personality *find_pers(int level, char *clevel)
426 {
427 struct mdk_personality *pers;
428 list_for_each_entry(pers, &pers_list, list) {
429 if (level != LEVEL_NONE && pers->level == level)
430 return pers;
431 if (strcmp(pers->name, clevel)==0)
432 return pers;
433 }
434 return NULL;
435 }
436
437 /* return the offset of the super block in 512byte sectors */
438 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
439 {
440 sector_t num_sectors = bdev->bd_inode->i_size / 512;
441 return MD_NEW_SIZE_SECTORS(num_sectors);
442 }
443
444 static int alloc_disk_sb(mdk_rdev_t * rdev)
445 {
446 if (rdev->sb_page)
447 MD_BUG();
448
449 rdev->sb_page = alloc_page(GFP_KERNEL);
450 if (!rdev->sb_page) {
451 printk(KERN_ALERT "md: out of memory.\n");
452 return -ENOMEM;
453 }
454
455 return 0;
456 }
457
458 static void free_disk_sb(mdk_rdev_t * rdev)
459 {
460 if (rdev->sb_page) {
461 put_page(rdev->sb_page);
462 rdev->sb_loaded = 0;
463 rdev->sb_page = NULL;
464 rdev->sb_start = 0;
465 rdev->sectors = 0;
466 }
467 }
468
469
470 static void super_written(struct bio *bio, int error)
471 {
472 mdk_rdev_t *rdev = bio->bi_private;
473 mddev_t *mddev = rdev->mddev;
474
475 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
476 printk("md: super_written gets error=%d, uptodate=%d\n",
477 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
478 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
479 md_error(mddev, rdev);
480 }
481
482 if (atomic_dec_and_test(&mddev->pending_writes))
483 wake_up(&mddev->sb_wait);
484 bio_put(bio);
485 }
486
487 static void super_written_barrier(struct bio *bio, int error)
488 {
489 struct bio *bio2 = bio->bi_private;
490 mdk_rdev_t *rdev = bio2->bi_private;
491 mddev_t *mddev = rdev->mddev;
492
493 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
494 error == -EOPNOTSUPP) {
495 unsigned long flags;
496 /* barriers don't appear to be supported :-( */
497 set_bit(BarriersNotsupp, &rdev->flags);
498 mddev->barriers_work = 0;
499 spin_lock_irqsave(&mddev->write_lock, flags);
500 bio2->bi_next = mddev->biolist;
501 mddev->biolist = bio2;
502 spin_unlock_irqrestore(&mddev->write_lock, flags);
503 wake_up(&mddev->sb_wait);
504 bio_put(bio);
505 } else {
506 bio_put(bio2);
507 bio->bi_private = rdev;
508 super_written(bio, error);
509 }
510 }
511
512 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
513 sector_t sector, int size, struct page *page)
514 {
515 /* write first size bytes of page to sector of rdev
516 * Increment mddev->pending_writes before returning
517 * and decrement it on completion, waking up sb_wait
518 * if zero is reached.
519 * If an error occurred, call md_error
520 *
521 * As we might need to resubmit the request if BIO_RW_BARRIER
522 * causes ENOTSUPP, we allocate a spare bio...
523 */
524 struct bio *bio = bio_alloc(GFP_NOIO, 1);
525 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG);
526
527 bio->bi_bdev = rdev->bdev;
528 bio->bi_sector = sector;
529 bio_add_page(bio, page, size, 0);
530 bio->bi_private = rdev;
531 bio->bi_end_io = super_written;
532 bio->bi_rw = rw;
533
534 atomic_inc(&mddev->pending_writes);
535 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
536 struct bio *rbio;
537 rw |= (1<<BIO_RW_BARRIER);
538 rbio = bio_clone(bio, GFP_NOIO);
539 rbio->bi_private = bio;
540 rbio->bi_end_io = super_written_barrier;
541 submit_bio(rw, rbio);
542 } else
543 submit_bio(rw, bio);
544 }
545
546 void md_super_wait(mddev_t *mddev)
547 {
548 /* wait for all superblock writes that were scheduled to complete.
549 * if any had to be retried (due to BARRIER problems), retry them
550 */
551 DEFINE_WAIT(wq);
552 for(;;) {
553 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
554 if (atomic_read(&mddev->pending_writes)==0)
555 break;
556 while (mddev->biolist) {
557 struct bio *bio;
558 spin_lock_irq(&mddev->write_lock);
559 bio = mddev->biolist;
560 mddev->biolist = bio->bi_next ;
561 bio->bi_next = NULL;
562 spin_unlock_irq(&mddev->write_lock);
563 submit_bio(bio->bi_rw, bio);
564 }
565 schedule();
566 }
567 finish_wait(&mddev->sb_wait, &wq);
568 }
569
570 static void bi_complete(struct bio *bio, int error)
571 {
572 complete((struct completion*)bio->bi_private);
573 }
574
575 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
576 struct page *page, int rw)
577 {
578 struct bio *bio = bio_alloc(GFP_NOIO, 1);
579 struct completion event;
580 int ret;
581
582 rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
583
584 bio->bi_bdev = bdev;
585 bio->bi_sector = sector;
586 bio_add_page(bio, page, size, 0);
587 init_completion(&event);
588 bio->bi_private = &event;
589 bio->bi_end_io = bi_complete;
590 submit_bio(rw, bio);
591 wait_for_completion(&event);
592
593 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
594 bio_put(bio);
595 return ret;
596 }
597 EXPORT_SYMBOL_GPL(sync_page_io);
598
599 static int read_disk_sb(mdk_rdev_t * rdev, int size)
600 {
601 char b[BDEVNAME_SIZE];
602 if (!rdev->sb_page) {
603 MD_BUG();
604 return -EINVAL;
605 }
606 if (rdev->sb_loaded)
607 return 0;
608
609
610 if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
611 goto fail;
612 rdev->sb_loaded = 1;
613 return 0;
614
615 fail:
616 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
617 bdevname(rdev->bdev,b));
618 return -EINVAL;
619 }
620
621 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
622 {
623 return sb1->set_uuid0 == sb2->set_uuid0 &&
624 sb1->set_uuid1 == sb2->set_uuid1 &&
625 sb1->set_uuid2 == sb2->set_uuid2 &&
626 sb1->set_uuid3 == sb2->set_uuid3;
627 }
628
629 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
630 {
631 int ret;
632 mdp_super_t *tmp1, *tmp2;
633
634 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
635 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
636
637 if (!tmp1 || !tmp2) {
638 ret = 0;
639 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
640 goto abort;
641 }
642
643 *tmp1 = *sb1;
644 *tmp2 = *sb2;
645
646 /*
647 * nr_disks is not constant
648 */
649 tmp1->nr_disks = 0;
650 tmp2->nr_disks = 0;
651
652 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
653 abort:
654 kfree(tmp1);
655 kfree(tmp2);
656 return ret;
657 }
658
659
660 static u32 md_csum_fold(u32 csum)
661 {
662 csum = (csum & 0xffff) + (csum >> 16);
663 return (csum & 0xffff) + (csum >> 16);
664 }
665
666 static unsigned int calc_sb_csum(mdp_super_t * sb)
667 {
668 u64 newcsum = 0;
669 u32 *sb32 = (u32*)sb;
670 int i;
671 unsigned int disk_csum, csum;
672
673 disk_csum = sb->sb_csum;
674 sb->sb_csum = 0;
675
676 for (i = 0; i < MD_SB_BYTES/4 ; i++)
677 newcsum += sb32[i];
678 csum = (newcsum & 0xffffffff) + (newcsum>>32);
679
680
681 #ifdef CONFIG_ALPHA
682 /* This used to use csum_partial, which was wrong for several
683 * reasons including that different results are returned on
684 * different architectures. It isn't critical that we get exactly
685 * the same return value as before (we always csum_fold before
686 * testing, and that removes any differences). However as we
687 * know that csum_partial always returned a 16bit value on
688 * alphas, do a fold to maximise conformity to previous behaviour.
689 */
690 sb->sb_csum = md_csum_fold(disk_csum);
691 #else
692 sb->sb_csum = disk_csum;
693 #endif
694 return csum;
695 }
696
697
698 /*
699 * Handle superblock details.
700 * We want to be able to handle multiple superblock formats
701 * so we have a common interface to them all, and an array of
702 * different handlers.
703 * We rely on user-space to write the initial superblock, and support
704 * reading and updating of superblocks.
705 * Interface methods are:
706 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
707 * loads and validates a superblock on dev.
708 * if refdev != NULL, compare superblocks on both devices
709 * Return:
710 * 0 - dev has a superblock that is compatible with refdev
711 * 1 - dev has a superblock that is compatible and newer than refdev
712 * so dev should be used as the refdev in future
713 * -EINVAL superblock incompatible or invalid
714 * -othererror e.g. -EIO
715 *
716 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
717 * Verify that dev is acceptable into mddev.
718 * The first time, mddev->raid_disks will be 0, and data from
719 * dev should be merged in. Subsequent calls check that dev
720 * is new enough. Return 0 or -EINVAL
721 *
722 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
723 * Update the superblock for rdev with data in mddev
724 * This does not write to disc.
725 *
726 */
727
728 struct super_type {
729 char *name;
730 struct module *owner;
731 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
732 int minor_version);
733 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
734 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
735 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
736 sector_t num_sectors);
737 };
738
739 /*
740 * Check that the given mddev has no bitmap.
741 *
742 * This function is called from the run method of all personalities that do not
743 * support bitmaps. It prints an error message and returns non-zero if mddev
744 * has a bitmap. Otherwise, it returns 0.
745 *
746 */
747 int md_check_no_bitmap(mddev_t *mddev)
748 {
749 if (!mddev->bitmap_file && !mddev->bitmap_offset)
750 return 0;
751 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
752 mdname(mddev), mddev->pers->name);
753 return 1;
754 }
755 EXPORT_SYMBOL(md_check_no_bitmap);
756
757 /*
758 * load_super for 0.90.0
759 */
760 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
761 {
762 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
763 mdp_super_t *sb;
764 int ret;
765
766 /*
767 * Calculate the position of the superblock (512byte sectors),
768 * it's at the end of the disk.
769 *
770 * It also happens to be a multiple of 4Kb.
771 */
772 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
773
774 ret = read_disk_sb(rdev, MD_SB_BYTES);
775 if (ret) return ret;
776
777 ret = -EINVAL;
778
779 bdevname(rdev->bdev, b);
780 sb = (mdp_super_t*)page_address(rdev->sb_page);
781
782 if (sb->md_magic != MD_SB_MAGIC) {
783 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
784 b);
785 goto abort;
786 }
787
788 if (sb->major_version != 0 ||
789 sb->minor_version < 90 ||
790 sb->minor_version > 91) {
791 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
792 sb->major_version, sb->minor_version,
793 b);
794 goto abort;
795 }
796
797 if (sb->raid_disks <= 0)
798 goto abort;
799
800 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
801 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
802 b);
803 goto abort;
804 }
805
806 rdev->preferred_minor = sb->md_minor;
807 rdev->data_offset = 0;
808 rdev->sb_size = MD_SB_BYTES;
809
810 if (sb->level == LEVEL_MULTIPATH)
811 rdev->desc_nr = -1;
812 else
813 rdev->desc_nr = sb->this_disk.number;
814
815 if (!refdev) {
816 ret = 1;
817 } else {
818 __u64 ev1, ev2;
819 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
820 if (!uuid_equal(refsb, sb)) {
821 printk(KERN_WARNING "md: %s has different UUID to %s\n",
822 b, bdevname(refdev->bdev,b2));
823 goto abort;
824 }
825 if (!sb_equal(refsb, sb)) {
826 printk(KERN_WARNING "md: %s has same UUID"
827 " but different superblock to %s\n",
828 b, bdevname(refdev->bdev, b2));
829 goto abort;
830 }
831 ev1 = md_event(sb);
832 ev2 = md_event(refsb);
833 if (ev1 > ev2)
834 ret = 1;
835 else
836 ret = 0;
837 }
838 rdev->sectors = rdev->sb_start;
839
840 if (rdev->sectors < sb->size * 2 && sb->level > 1)
841 /* "this cannot possibly happen" ... */
842 ret = -EINVAL;
843
844 abort:
845 return ret;
846 }
847
848 /*
849 * validate_super for 0.90.0
850 */
851 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
852 {
853 mdp_disk_t *desc;
854 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
855 __u64 ev1 = md_event(sb);
856
857 rdev->raid_disk = -1;
858 clear_bit(Faulty, &rdev->flags);
859 clear_bit(In_sync, &rdev->flags);
860 clear_bit(WriteMostly, &rdev->flags);
861 clear_bit(BarriersNotsupp, &rdev->flags);
862
863 if (mddev->raid_disks == 0) {
864 mddev->major_version = 0;
865 mddev->minor_version = sb->minor_version;
866 mddev->patch_version = sb->patch_version;
867 mddev->external = 0;
868 mddev->chunk_sectors = sb->chunk_size >> 9;
869 mddev->ctime = sb->ctime;
870 mddev->utime = sb->utime;
871 mddev->level = sb->level;
872 mddev->clevel[0] = 0;
873 mddev->layout = sb->layout;
874 mddev->raid_disks = sb->raid_disks;
875 mddev->dev_sectors = sb->size * 2;
876 mddev->events = ev1;
877 mddev->bitmap_offset = 0;
878 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
879
880 if (mddev->minor_version >= 91) {
881 mddev->reshape_position = sb->reshape_position;
882 mddev->delta_disks = sb->delta_disks;
883 mddev->new_level = sb->new_level;
884 mddev->new_layout = sb->new_layout;
885 mddev->new_chunk_sectors = sb->new_chunk >> 9;
886 } else {
887 mddev->reshape_position = MaxSector;
888 mddev->delta_disks = 0;
889 mddev->new_level = mddev->level;
890 mddev->new_layout = mddev->layout;
891 mddev->new_chunk_sectors = mddev->chunk_sectors;
892 }
893
894 if (sb->state & (1<<MD_SB_CLEAN))
895 mddev->recovery_cp = MaxSector;
896 else {
897 if (sb->events_hi == sb->cp_events_hi &&
898 sb->events_lo == sb->cp_events_lo) {
899 mddev->recovery_cp = sb->recovery_cp;
900 } else
901 mddev->recovery_cp = 0;
902 }
903
904 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
905 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
906 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
907 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
908
909 mddev->max_disks = MD_SB_DISKS;
910
911 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
912 mddev->bitmap_file == NULL)
913 mddev->bitmap_offset = mddev->default_bitmap_offset;
914
915 } else if (mddev->pers == NULL) {
916 /* Insist on good event counter while assembling */
917 ++ev1;
918 if (ev1 < mddev->events)
919 return -EINVAL;
920 } else if (mddev->bitmap) {
921 /* if adding to array with a bitmap, then we can accept an
922 * older device ... but not too old.
923 */
924 if (ev1 < mddev->bitmap->events_cleared)
925 return 0;
926 } else {
927 if (ev1 < mddev->events)
928 /* just a hot-add of a new device, leave raid_disk at -1 */
929 return 0;
930 }
931
932 if (mddev->level != LEVEL_MULTIPATH) {
933 desc = sb->disks + rdev->desc_nr;
934
935 if (desc->state & (1<<MD_DISK_FAULTY))
936 set_bit(Faulty, &rdev->flags);
937 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
938 desc->raid_disk < mddev->raid_disks */) {
939 set_bit(In_sync, &rdev->flags);
940 rdev->raid_disk = desc->raid_disk;
941 }
942 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
943 set_bit(WriteMostly, &rdev->flags);
944 } else /* MULTIPATH are always insync */
945 set_bit(In_sync, &rdev->flags);
946 return 0;
947 }
948
949 /*
950 * sync_super for 0.90.0
951 */
952 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
953 {
954 mdp_super_t *sb;
955 mdk_rdev_t *rdev2;
956 int next_spare = mddev->raid_disks;
957
958
959 /* make rdev->sb match mddev data..
960 *
961 * 1/ zero out disks
962 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
963 * 3/ any empty disks < next_spare become removed
964 *
965 * disks[0] gets initialised to REMOVED because
966 * we cannot be sure from other fields if it has
967 * been initialised or not.
968 */
969 int i;
970 int active=0, working=0,failed=0,spare=0,nr_disks=0;
971
972 rdev->sb_size = MD_SB_BYTES;
973
974 sb = (mdp_super_t*)page_address(rdev->sb_page);
975
976 memset(sb, 0, sizeof(*sb));
977
978 sb->md_magic = MD_SB_MAGIC;
979 sb->major_version = mddev->major_version;
980 sb->patch_version = mddev->patch_version;
981 sb->gvalid_words = 0; /* ignored */
982 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
983 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
984 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
985 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
986
987 sb->ctime = mddev->ctime;
988 sb->level = mddev->level;
989 sb->size = mddev->dev_sectors / 2;
990 sb->raid_disks = mddev->raid_disks;
991 sb->md_minor = mddev->md_minor;
992 sb->not_persistent = 0;
993 sb->utime = mddev->utime;
994 sb->state = 0;
995 sb->events_hi = (mddev->events>>32);
996 sb->events_lo = (u32)mddev->events;
997
998 if (mddev->reshape_position == MaxSector)
999 sb->minor_version = 90;
1000 else {
1001 sb->minor_version = 91;
1002 sb->reshape_position = mddev->reshape_position;
1003 sb->new_level = mddev->new_level;
1004 sb->delta_disks = mddev->delta_disks;
1005 sb->new_layout = mddev->new_layout;
1006 sb->new_chunk = mddev->new_chunk_sectors << 9;
1007 }
1008 mddev->minor_version = sb->minor_version;
1009 if (mddev->in_sync)
1010 {
1011 sb->recovery_cp = mddev->recovery_cp;
1012 sb->cp_events_hi = (mddev->events>>32);
1013 sb->cp_events_lo = (u32)mddev->events;
1014 if (mddev->recovery_cp == MaxSector)
1015 sb->state = (1<< MD_SB_CLEAN);
1016 } else
1017 sb->recovery_cp = 0;
1018
1019 sb->layout = mddev->layout;
1020 sb->chunk_size = mddev->chunk_sectors << 9;
1021
1022 if (mddev->bitmap && mddev->bitmap_file == NULL)
1023 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1024
1025 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1026 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1027 mdp_disk_t *d;
1028 int desc_nr;
1029 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
1030 && !test_bit(Faulty, &rdev2->flags))
1031 desc_nr = rdev2->raid_disk;
1032 else
1033 desc_nr = next_spare++;
1034 rdev2->desc_nr = desc_nr;
1035 d = &sb->disks[rdev2->desc_nr];
1036 nr_disks++;
1037 d->number = rdev2->desc_nr;
1038 d->major = MAJOR(rdev2->bdev->bd_dev);
1039 d->minor = MINOR(rdev2->bdev->bd_dev);
1040 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
1041 && !test_bit(Faulty, &rdev2->flags))
1042 d->raid_disk = rdev2->raid_disk;
1043 else
1044 d->raid_disk = rdev2->desc_nr; /* compatibility */
1045 if (test_bit(Faulty, &rdev2->flags))
1046 d->state = (1<<MD_DISK_FAULTY);
1047 else if (test_bit(In_sync, &rdev2->flags)) {
1048 d->state = (1<<MD_DISK_ACTIVE);
1049 d->state |= (1<<MD_DISK_SYNC);
1050 active++;
1051 working++;
1052 } else {
1053 d->state = 0;
1054 spare++;
1055 working++;
1056 }
1057 if (test_bit(WriteMostly, &rdev2->flags))
1058 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1059 }
1060 /* now set the "removed" and "faulty" bits on any missing devices */
1061 for (i=0 ; i < mddev->raid_disks ; i++) {
1062 mdp_disk_t *d = &sb->disks[i];
1063 if (d->state == 0 && d->number == 0) {
1064 d->number = i;
1065 d->raid_disk = i;
1066 d->state = (1<<MD_DISK_REMOVED);
1067 d->state |= (1<<MD_DISK_FAULTY);
1068 failed++;
1069 }
1070 }
1071 sb->nr_disks = nr_disks;
1072 sb->active_disks = active;
1073 sb->working_disks = working;
1074 sb->failed_disks = failed;
1075 sb->spare_disks = spare;
1076
1077 sb->this_disk = sb->disks[rdev->desc_nr];
1078 sb->sb_csum = calc_sb_csum(sb);
1079 }
1080
1081 /*
1082 * rdev_size_change for 0.90.0
1083 */
1084 static unsigned long long
1085 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1086 {
1087 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1088 return 0; /* component must fit device */
1089 if (rdev->mddev->bitmap_offset)
1090 return 0; /* can't move bitmap */
1091 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1092 if (!num_sectors || num_sectors > rdev->sb_start)
1093 num_sectors = rdev->sb_start;
1094 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1095 rdev->sb_page);
1096 md_super_wait(rdev->mddev);
1097 return num_sectors / 2; /* kB for sysfs */
1098 }
1099
1100
1101 /*
1102 * version 1 superblock
1103 */
1104
1105 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1106 {
1107 __le32 disk_csum;
1108 u32 csum;
1109 unsigned long long newcsum;
1110 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1111 __le32 *isuper = (__le32*)sb;
1112 int i;
1113
1114 disk_csum = sb->sb_csum;
1115 sb->sb_csum = 0;
1116 newcsum = 0;
1117 for (i=0; size>=4; size -= 4 )
1118 newcsum += le32_to_cpu(*isuper++);
1119
1120 if (size == 2)
1121 newcsum += le16_to_cpu(*(__le16*) isuper);
1122
1123 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1124 sb->sb_csum = disk_csum;
1125 return cpu_to_le32(csum);
1126 }
1127
1128 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1129 {
1130 struct mdp_superblock_1 *sb;
1131 int ret;
1132 sector_t sb_start;
1133 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1134 int bmask;
1135
1136 /*
1137 * Calculate the position of the superblock in 512byte sectors.
1138 * It is always aligned to a 4K boundary and
1139 * depeding on minor_version, it can be:
1140 * 0: At least 8K, but less than 12K, from end of device
1141 * 1: At start of device
1142 * 2: 4K from start of device.
1143 */
1144 switch(minor_version) {
1145 case 0:
1146 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1147 sb_start -= 8*2;
1148 sb_start &= ~(sector_t)(4*2-1);
1149 break;
1150 case 1:
1151 sb_start = 0;
1152 break;
1153 case 2:
1154 sb_start = 8;
1155 break;
1156 default:
1157 return -EINVAL;
1158 }
1159 rdev->sb_start = sb_start;
1160
1161 /* superblock is rarely larger than 1K, but it can be larger,
1162 * and it is safe to read 4k, so we do that
1163 */
1164 ret = read_disk_sb(rdev, 4096);
1165 if (ret) return ret;
1166
1167
1168 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1169
1170 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1171 sb->major_version != cpu_to_le32(1) ||
1172 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1173 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1174 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1175 return -EINVAL;
1176
1177 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1178 printk("md: invalid superblock checksum on %s\n",
1179 bdevname(rdev->bdev,b));
1180 return -EINVAL;
1181 }
1182 if (le64_to_cpu(sb->data_size) < 10) {
1183 printk("md: data_size too small on %s\n",
1184 bdevname(rdev->bdev,b));
1185 return -EINVAL;
1186 }
1187
1188 rdev->preferred_minor = 0xffff;
1189 rdev->data_offset = le64_to_cpu(sb->data_offset);
1190 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1191
1192 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1193 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1194 if (rdev->sb_size & bmask)
1195 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1196
1197 if (minor_version
1198 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1199 return -EINVAL;
1200
1201 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1202 rdev->desc_nr = -1;
1203 else
1204 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1205
1206 if (!refdev) {
1207 ret = 1;
1208 } else {
1209 __u64 ev1, ev2;
1210 struct mdp_superblock_1 *refsb =
1211 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1212
1213 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1214 sb->level != refsb->level ||
1215 sb->layout != refsb->layout ||
1216 sb->chunksize != refsb->chunksize) {
1217 printk(KERN_WARNING "md: %s has strangely different"
1218 " superblock to %s\n",
1219 bdevname(rdev->bdev,b),
1220 bdevname(refdev->bdev,b2));
1221 return -EINVAL;
1222 }
1223 ev1 = le64_to_cpu(sb->events);
1224 ev2 = le64_to_cpu(refsb->events);
1225
1226 if (ev1 > ev2)
1227 ret = 1;
1228 else
1229 ret = 0;
1230 }
1231 if (minor_version)
1232 rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
1233 le64_to_cpu(sb->data_offset);
1234 else
1235 rdev->sectors = rdev->sb_start;
1236 if (rdev->sectors < le64_to_cpu(sb->data_size))
1237 return -EINVAL;
1238 rdev->sectors = le64_to_cpu(sb->data_size);
1239 if (le64_to_cpu(sb->size) > rdev->sectors)
1240 return -EINVAL;
1241 return ret;
1242 }
1243
1244 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1245 {
1246 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1247 __u64 ev1 = le64_to_cpu(sb->events);
1248
1249 rdev->raid_disk = -1;
1250 clear_bit(Faulty, &rdev->flags);
1251 clear_bit(In_sync, &rdev->flags);
1252 clear_bit(WriteMostly, &rdev->flags);
1253 clear_bit(BarriersNotsupp, &rdev->flags);
1254
1255 if (mddev->raid_disks == 0) {
1256 mddev->major_version = 1;
1257 mddev->patch_version = 0;
1258 mddev->external = 0;
1259 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1260 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1261 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1262 mddev->level = le32_to_cpu(sb->level);
1263 mddev->clevel[0] = 0;
1264 mddev->layout = le32_to_cpu(sb->layout);
1265 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1266 mddev->dev_sectors = le64_to_cpu(sb->size);
1267 mddev->events = ev1;
1268 mddev->bitmap_offset = 0;
1269 mddev->default_bitmap_offset = 1024 >> 9;
1270
1271 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1272 memcpy(mddev->uuid, sb->set_uuid, 16);
1273
1274 mddev->max_disks = (4096-256)/2;
1275
1276 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1277 mddev->bitmap_file == NULL )
1278 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1279
1280 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1281 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1282 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1283 mddev->new_level = le32_to_cpu(sb->new_level);
1284 mddev->new_layout = le32_to_cpu(sb->new_layout);
1285 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1286 } else {
1287 mddev->reshape_position = MaxSector;
1288 mddev->delta_disks = 0;
1289 mddev->new_level = mddev->level;
1290 mddev->new_layout = mddev->layout;
1291 mddev->new_chunk_sectors = mddev->chunk_sectors;
1292 }
1293
1294 } else if (mddev->pers == NULL) {
1295 /* Insist of good event counter while assembling */
1296 ++ev1;
1297 if (ev1 < mddev->events)
1298 return -EINVAL;
1299 } else if (mddev->bitmap) {
1300 /* If adding to array with a bitmap, then we can accept an
1301 * older device, but not too old.
1302 */
1303 if (ev1 < mddev->bitmap->events_cleared)
1304 return 0;
1305 } else {
1306 if (ev1 < mddev->events)
1307 /* just a hot-add of a new device, leave raid_disk at -1 */
1308 return 0;
1309 }
1310 if (mddev->level != LEVEL_MULTIPATH) {
1311 int role;
1312 if (rdev->desc_nr < 0 ||
1313 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1314 role = 0xffff;
1315 rdev->desc_nr = -1;
1316 } else
1317 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1318 switch(role) {
1319 case 0xffff: /* spare */
1320 break;
1321 case 0xfffe: /* faulty */
1322 set_bit(Faulty, &rdev->flags);
1323 break;
1324 default:
1325 if ((le32_to_cpu(sb->feature_map) &
1326 MD_FEATURE_RECOVERY_OFFSET))
1327 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1328 else
1329 set_bit(In_sync, &rdev->flags);
1330 rdev->raid_disk = role;
1331 break;
1332 }
1333 if (sb->devflags & WriteMostly1)
1334 set_bit(WriteMostly, &rdev->flags);
1335 } else /* MULTIPATH are always insync */
1336 set_bit(In_sync, &rdev->flags);
1337
1338 return 0;
1339 }
1340
1341 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1342 {
1343 struct mdp_superblock_1 *sb;
1344 mdk_rdev_t *rdev2;
1345 int max_dev, i;
1346 /* make rdev->sb match mddev and rdev data. */
1347
1348 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1349
1350 sb->feature_map = 0;
1351 sb->pad0 = 0;
1352 sb->recovery_offset = cpu_to_le64(0);
1353 memset(sb->pad1, 0, sizeof(sb->pad1));
1354 memset(sb->pad2, 0, sizeof(sb->pad2));
1355 memset(sb->pad3, 0, sizeof(sb->pad3));
1356
1357 sb->utime = cpu_to_le64((__u64)mddev->utime);
1358 sb->events = cpu_to_le64(mddev->events);
1359 if (mddev->in_sync)
1360 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1361 else
1362 sb->resync_offset = cpu_to_le64(0);
1363
1364 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1365
1366 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1367 sb->size = cpu_to_le64(mddev->dev_sectors);
1368 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1369 sb->level = cpu_to_le32(mddev->level);
1370 sb->layout = cpu_to_le32(mddev->layout);
1371
1372 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1373 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1374 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1375 }
1376
1377 if (rdev->raid_disk >= 0 &&
1378 !test_bit(In_sync, &rdev->flags)) {
1379 if (mddev->curr_resync_completed > rdev->recovery_offset)
1380 rdev->recovery_offset = mddev->curr_resync_completed;
1381 if (rdev->recovery_offset > 0) {
1382 sb->feature_map |=
1383 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1384 sb->recovery_offset =
1385 cpu_to_le64(rdev->recovery_offset);
1386 }
1387 }
1388
1389 if (mddev->reshape_position != MaxSector) {
1390 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1391 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1392 sb->new_layout = cpu_to_le32(mddev->new_layout);
1393 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1394 sb->new_level = cpu_to_le32(mddev->new_level);
1395 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1396 }
1397
1398 max_dev = 0;
1399 list_for_each_entry(rdev2, &mddev->disks, same_set)
1400 if (rdev2->desc_nr+1 > max_dev)
1401 max_dev = rdev2->desc_nr+1;
1402
1403 if (max_dev > le32_to_cpu(sb->max_dev)) {
1404 int bmask;
1405 sb->max_dev = cpu_to_le32(max_dev);
1406 rdev->sb_size = max_dev * 2 + 256;
1407 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1408 if (rdev->sb_size & bmask)
1409 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1410 }
1411 for (i=0; i<max_dev;i++)
1412 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1413
1414 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1415 i = rdev2->desc_nr;
1416 if (test_bit(Faulty, &rdev2->flags))
1417 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1418 else if (test_bit(In_sync, &rdev2->flags))
1419 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1420 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1421 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1422 else
1423 sb->dev_roles[i] = cpu_to_le16(0xffff);
1424 }
1425
1426 sb->sb_csum = calc_sb_1_csum(sb);
1427 }
1428
1429 static unsigned long long
1430 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1431 {
1432 struct mdp_superblock_1 *sb;
1433 sector_t max_sectors;
1434 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1435 return 0; /* component must fit device */
1436 if (rdev->sb_start < rdev->data_offset) {
1437 /* minor versions 1 and 2; superblock before data */
1438 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1439 max_sectors -= rdev->data_offset;
1440 if (!num_sectors || num_sectors > max_sectors)
1441 num_sectors = max_sectors;
1442 } else if (rdev->mddev->bitmap_offset) {
1443 /* minor version 0 with bitmap we can't move */
1444 return 0;
1445 } else {
1446 /* minor version 0; superblock after data */
1447 sector_t sb_start;
1448 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1449 sb_start &= ~(sector_t)(4*2 - 1);
1450 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1451 if (!num_sectors || num_sectors > max_sectors)
1452 num_sectors = max_sectors;
1453 rdev->sb_start = sb_start;
1454 }
1455 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1456 sb->data_size = cpu_to_le64(num_sectors);
1457 sb->super_offset = rdev->sb_start;
1458 sb->sb_csum = calc_sb_1_csum(sb);
1459 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1460 rdev->sb_page);
1461 md_super_wait(rdev->mddev);
1462 return num_sectors / 2; /* kB for sysfs */
1463 }
1464
1465 static struct super_type super_types[] = {
1466 [0] = {
1467 .name = "0.90.0",
1468 .owner = THIS_MODULE,
1469 .load_super = super_90_load,
1470 .validate_super = super_90_validate,
1471 .sync_super = super_90_sync,
1472 .rdev_size_change = super_90_rdev_size_change,
1473 },
1474 [1] = {
1475 .name = "md-1",
1476 .owner = THIS_MODULE,
1477 .load_super = super_1_load,
1478 .validate_super = super_1_validate,
1479 .sync_super = super_1_sync,
1480 .rdev_size_change = super_1_rdev_size_change,
1481 },
1482 };
1483
1484 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1485 {
1486 mdk_rdev_t *rdev, *rdev2;
1487
1488 rcu_read_lock();
1489 rdev_for_each_rcu(rdev, mddev1)
1490 rdev_for_each_rcu(rdev2, mddev2)
1491 if (rdev->bdev->bd_contains ==
1492 rdev2->bdev->bd_contains) {
1493 rcu_read_unlock();
1494 return 1;
1495 }
1496 rcu_read_unlock();
1497 return 0;
1498 }
1499
1500 static LIST_HEAD(pending_raid_disks);
1501
1502 /*
1503 * Try to register data integrity profile for an mddev
1504 *
1505 * This is called when an array is started and after a disk has been kicked
1506 * from the array. It only succeeds if all working and active component devices
1507 * are integrity capable with matching profiles.
1508 */
1509 int md_integrity_register(mddev_t *mddev)
1510 {
1511 mdk_rdev_t *rdev, *reference = NULL;
1512
1513 if (list_empty(&mddev->disks))
1514 return 0; /* nothing to do */
1515 if (blk_get_integrity(mddev->gendisk))
1516 return 0; /* already registered */
1517 list_for_each_entry(rdev, &mddev->disks, same_set) {
1518 /* skip spares and non-functional disks */
1519 if (test_bit(Faulty, &rdev->flags))
1520 continue;
1521 if (rdev->raid_disk < 0)
1522 continue;
1523 /*
1524 * If at least one rdev is not integrity capable, we can not
1525 * enable data integrity for the md device.
1526 */
1527 if (!bdev_get_integrity(rdev->bdev))
1528 return -EINVAL;
1529 if (!reference) {
1530 /* Use the first rdev as the reference */
1531 reference = rdev;
1532 continue;
1533 }
1534 /* does this rdev's profile match the reference profile? */
1535 if (blk_integrity_compare(reference->bdev->bd_disk,
1536 rdev->bdev->bd_disk) < 0)
1537 return -EINVAL;
1538 }
1539 /*
1540 * All component devices are integrity capable and have matching
1541 * profiles, register the common profile for the md device.
1542 */
1543 if (blk_integrity_register(mddev->gendisk,
1544 bdev_get_integrity(reference->bdev)) != 0) {
1545 printk(KERN_ERR "md: failed to register integrity for %s\n",
1546 mdname(mddev));
1547 return -EINVAL;
1548 }
1549 printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1550 mdname(mddev));
1551 return 0;
1552 }
1553 EXPORT_SYMBOL(md_integrity_register);
1554
1555 /* Disable data integrity if non-capable/non-matching disk is being added */
1556 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
1557 {
1558 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1559 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1560
1561 if (!bi_mddev) /* nothing to do */
1562 return;
1563 if (rdev->raid_disk < 0) /* skip spares */
1564 return;
1565 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1566 rdev->bdev->bd_disk) >= 0)
1567 return;
1568 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1569 blk_integrity_unregister(mddev->gendisk);
1570 }
1571 EXPORT_SYMBOL(md_integrity_add_rdev);
1572
1573 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1574 {
1575 char b[BDEVNAME_SIZE];
1576 struct kobject *ko;
1577 char *s;
1578 int err;
1579
1580 if (rdev->mddev) {
1581 MD_BUG();
1582 return -EINVAL;
1583 }
1584
1585 /* prevent duplicates */
1586 if (find_rdev(mddev, rdev->bdev->bd_dev))
1587 return -EEXIST;
1588
1589 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1590 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1591 rdev->sectors < mddev->dev_sectors)) {
1592 if (mddev->pers) {
1593 /* Cannot change size, so fail
1594 * If mddev->level <= 0, then we don't care
1595 * about aligning sizes (e.g. linear)
1596 */
1597 if (mddev->level > 0)
1598 return -ENOSPC;
1599 } else
1600 mddev->dev_sectors = rdev->sectors;
1601 }
1602
1603 /* Verify rdev->desc_nr is unique.
1604 * If it is -1, assign a free number, else
1605 * check number is not in use
1606 */
1607 if (rdev->desc_nr < 0) {
1608 int choice = 0;
1609 if (mddev->pers) choice = mddev->raid_disks;
1610 while (find_rdev_nr(mddev, choice))
1611 choice++;
1612 rdev->desc_nr = choice;
1613 } else {
1614 if (find_rdev_nr(mddev, rdev->desc_nr))
1615 return -EBUSY;
1616 }
1617 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1618 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1619 mdname(mddev), mddev->max_disks);
1620 return -EBUSY;
1621 }
1622 bdevname(rdev->bdev,b);
1623 while ( (s=strchr(b, '/')) != NULL)
1624 *s = '!';
1625
1626 rdev->mddev = mddev;
1627 printk(KERN_INFO "md: bind<%s>\n", b);
1628
1629 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1630 goto fail;
1631
1632 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1633 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1634 kobject_del(&rdev->kobj);
1635 goto fail;
1636 }
1637 rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, "state");
1638
1639 list_add_rcu(&rdev->same_set, &mddev->disks);
1640 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1641
1642 /* May as well allow recovery to be retried once */
1643 mddev->recovery_disabled = 0;
1644
1645 return 0;
1646
1647 fail:
1648 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1649 b, mdname(mddev));
1650 return err;
1651 }
1652
1653 static void md_delayed_delete(struct work_struct *ws)
1654 {
1655 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1656 kobject_del(&rdev->kobj);
1657 kobject_put(&rdev->kobj);
1658 }
1659
1660 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1661 {
1662 char b[BDEVNAME_SIZE];
1663 if (!rdev->mddev) {
1664 MD_BUG();
1665 return;
1666 }
1667 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1668 list_del_rcu(&rdev->same_set);
1669 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1670 rdev->mddev = NULL;
1671 sysfs_remove_link(&rdev->kobj, "block");
1672 sysfs_put(rdev->sysfs_state);
1673 rdev->sysfs_state = NULL;
1674 /* We need to delay this, otherwise we can deadlock when
1675 * writing to 'remove' to "dev/state". We also need
1676 * to delay it due to rcu usage.
1677 */
1678 synchronize_rcu();
1679 INIT_WORK(&rdev->del_work, md_delayed_delete);
1680 kobject_get(&rdev->kobj);
1681 schedule_work(&rdev->del_work);
1682 }
1683
1684 /*
1685 * prevent the device from being mounted, repartitioned or
1686 * otherwise reused by a RAID array (or any other kernel
1687 * subsystem), by bd_claiming the device.
1688 */
1689 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1690 {
1691 int err = 0;
1692 struct block_device *bdev;
1693 char b[BDEVNAME_SIZE];
1694
1695 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1696 if (IS_ERR(bdev)) {
1697 printk(KERN_ERR "md: could not open %s.\n",
1698 __bdevname(dev, b));
1699 return PTR_ERR(bdev);
1700 }
1701 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1702 if (err) {
1703 printk(KERN_ERR "md: could not bd_claim %s.\n",
1704 bdevname(bdev, b));
1705 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1706 return err;
1707 }
1708 if (!shared)
1709 set_bit(AllReserved, &rdev->flags);
1710 rdev->bdev = bdev;
1711 return err;
1712 }
1713
1714 static void unlock_rdev(mdk_rdev_t *rdev)
1715 {
1716 struct block_device *bdev = rdev->bdev;
1717 rdev->bdev = NULL;
1718 if (!bdev)
1719 MD_BUG();
1720 bd_release(bdev);
1721 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1722 }
1723
1724 void md_autodetect_dev(dev_t dev);
1725
1726 static void export_rdev(mdk_rdev_t * rdev)
1727 {
1728 char b[BDEVNAME_SIZE];
1729 printk(KERN_INFO "md: export_rdev(%s)\n",
1730 bdevname(rdev->bdev,b));
1731 if (rdev->mddev)
1732 MD_BUG();
1733 free_disk_sb(rdev);
1734 #ifndef MODULE
1735 if (test_bit(AutoDetected, &rdev->flags))
1736 md_autodetect_dev(rdev->bdev->bd_dev);
1737 #endif
1738 unlock_rdev(rdev);
1739 kobject_put(&rdev->kobj);
1740 }
1741
1742 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1743 {
1744 unbind_rdev_from_array(rdev);
1745 export_rdev(rdev);
1746 }
1747
1748 static void export_array(mddev_t *mddev)
1749 {
1750 mdk_rdev_t *rdev, *tmp;
1751
1752 rdev_for_each(rdev, tmp, mddev) {
1753 if (!rdev->mddev) {
1754 MD_BUG();
1755 continue;
1756 }
1757 kick_rdev_from_array(rdev);
1758 }
1759 if (!list_empty(&mddev->disks))
1760 MD_BUG();
1761 mddev->raid_disks = 0;
1762 mddev->major_version = 0;
1763 }
1764
1765 static void print_desc(mdp_disk_t *desc)
1766 {
1767 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1768 desc->major,desc->minor,desc->raid_disk,desc->state);
1769 }
1770
1771 static void print_sb_90(mdp_super_t *sb)
1772 {
1773 int i;
1774
1775 printk(KERN_INFO
1776 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1777 sb->major_version, sb->minor_version, sb->patch_version,
1778 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1779 sb->ctime);
1780 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1781 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1782 sb->md_minor, sb->layout, sb->chunk_size);
1783 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1784 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1785 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1786 sb->failed_disks, sb->spare_disks,
1787 sb->sb_csum, (unsigned long)sb->events_lo);
1788
1789 printk(KERN_INFO);
1790 for (i = 0; i < MD_SB_DISKS; i++) {
1791 mdp_disk_t *desc;
1792
1793 desc = sb->disks + i;
1794 if (desc->number || desc->major || desc->minor ||
1795 desc->raid_disk || (desc->state && (desc->state != 4))) {
1796 printk(" D %2d: ", i);
1797 print_desc(desc);
1798 }
1799 }
1800 printk(KERN_INFO "md: THIS: ");
1801 print_desc(&sb->this_disk);
1802 }
1803
1804 static void print_sb_1(struct mdp_superblock_1 *sb)
1805 {
1806 __u8 *uuid;
1807
1808 uuid = sb->set_uuid;
1809 printk(KERN_INFO
1810 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%02x%02x%02x%02x"
1811 ":%02x%02x:%02x%02x:%02x%02x:%02x%02x%02x%02x%02x%02x>\n"
1812 "md: Name: \"%s\" CT:%llu\n",
1813 le32_to_cpu(sb->major_version),
1814 le32_to_cpu(sb->feature_map),
1815 uuid[0], uuid[1], uuid[2], uuid[3],
1816 uuid[4], uuid[5], uuid[6], uuid[7],
1817 uuid[8], uuid[9], uuid[10], uuid[11],
1818 uuid[12], uuid[13], uuid[14], uuid[15],
1819 sb->set_name,
1820 (unsigned long long)le64_to_cpu(sb->ctime)
1821 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
1822
1823 uuid = sb->device_uuid;
1824 printk(KERN_INFO
1825 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
1826 " RO:%llu\n"
1827 "md: Dev:%08x UUID: %02x%02x%02x%02x:%02x%02x:%02x%02x:%02x%02x"
1828 ":%02x%02x%02x%02x%02x%02x\n"
1829 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
1830 "md: (MaxDev:%u) \n",
1831 le32_to_cpu(sb->level),
1832 (unsigned long long)le64_to_cpu(sb->size),
1833 le32_to_cpu(sb->raid_disks),
1834 le32_to_cpu(sb->layout),
1835 le32_to_cpu(sb->chunksize),
1836 (unsigned long long)le64_to_cpu(sb->data_offset),
1837 (unsigned long long)le64_to_cpu(sb->data_size),
1838 (unsigned long long)le64_to_cpu(sb->super_offset),
1839 (unsigned long long)le64_to_cpu(sb->recovery_offset),
1840 le32_to_cpu(sb->dev_number),
1841 uuid[0], uuid[1], uuid[2], uuid[3],
1842 uuid[4], uuid[5], uuid[6], uuid[7],
1843 uuid[8], uuid[9], uuid[10], uuid[11],
1844 uuid[12], uuid[13], uuid[14], uuid[15],
1845 sb->devflags,
1846 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
1847 (unsigned long long)le64_to_cpu(sb->events),
1848 (unsigned long long)le64_to_cpu(sb->resync_offset),
1849 le32_to_cpu(sb->sb_csum),
1850 le32_to_cpu(sb->max_dev)
1851 );
1852 }
1853
1854 static void print_rdev(mdk_rdev_t *rdev, int major_version)
1855 {
1856 char b[BDEVNAME_SIZE];
1857 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
1858 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
1859 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1860 rdev->desc_nr);
1861 if (rdev->sb_loaded) {
1862 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
1863 switch (major_version) {
1864 case 0:
1865 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
1866 break;
1867 case 1:
1868 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
1869 break;
1870 }
1871 } else
1872 printk(KERN_INFO "md: no rdev superblock!\n");
1873 }
1874
1875 static void md_print_devices(void)
1876 {
1877 struct list_head *tmp;
1878 mdk_rdev_t *rdev;
1879 mddev_t *mddev;
1880 char b[BDEVNAME_SIZE];
1881
1882 printk("\n");
1883 printk("md: **********************************\n");
1884 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1885 printk("md: **********************************\n");
1886 for_each_mddev(mddev, tmp) {
1887
1888 if (mddev->bitmap)
1889 bitmap_print_sb(mddev->bitmap);
1890 else
1891 printk("%s: ", mdname(mddev));
1892 list_for_each_entry(rdev, &mddev->disks, same_set)
1893 printk("<%s>", bdevname(rdev->bdev,b));
1894 printk("\n");
1895
1896 list_for_each_entry(rdev, &mddev->disks, same_set)
1897 print_rdev(rdev, mddev->major_version);
1898 }
1899 printk("md: **********************************\n");
1900 printk("\n");
1901 }
1902
1903
1904 static void sync_sbs(mddev_t * mddev, int nospares)
1905 {
1906 /* Update each superblock (in-memory image), but
1907 * if we are allowed to, skip spares which already
1908 * have the right event counter, or have one earlier
1909 * (which would mean they aren't being marked as dirty
1910 * with the rest of the array)
1911 */
1912 mdk_rdev_t *rdev;
1913
1914 list_for_each_entry(rdev, &mddev->disks, same_set) {
1915 if (rdev->sb_events == mddev->events ||
1916 (nospares &&
1917 rdev->raid_disk < 0 &&
1918 (rdev->sb_events&1)==0 &&
1919 rdev->sb_events+1 == mddev->events)) {
1920 /* Don't update this superblock */
1921 rdev->sb_loaded = 2;
1922 } else {
1923 super_types[mddev->major_version].
1924 sync_super(mddev, rdev);
1925 rdev->sb_loaded = 1;
1926 }
1927 }
1928 }
1929
1930 static void md_update_sb(mddev_t * mddev, int force_change)
1931 {
1932 mdk_rdev_t *rdev;
1933 int sync_req;
1934 int nospares = 0;
1935
1936 mddev->utime = get_seconds();
1937 if (mddev->external)
1938 return;
1939 repeat:
1940 spin_lock_irq(&mddev->write_lock);
1941
1942 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1943 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1944 force_change = 1;
1945 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1946 /* just a clean<-> dirty transition, possibly leave spares alone,
1947 * though if events isn't the right even/odd, we will have to do
1948 * spares after all
1949 */
1950 nospares = 1;
1951 if (force_change)
1952 nospares = 0;
1953 if (mddev->degraded)
1954 /* If the array is degraded, then skipping spares is both
1955 * dangerous and fairly pointless.
1956 * Dangerous because a device that was removed from the array
1957 * might have a event_count that still looks up-to-date,
1958 * so it can be re-added without a resync.
1959 * Pointless because if there are any spares to skip,
1960 * then a recovery will happen and soon that array won't
1961 * be degraded any more and the spare can go back to sleep then.
1962 */
1963 nospares = 0;
1964
1965 sync_req = mddev->in_sync;
1966
1967 /* If this is just a dirty<->clean transition, and the array is clean
1968 * and 'events' is odd, we can roll back to the previous clean state */
1969 if (nospares
1970 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1971 && (mddev->events & 1)
1972 && mddev->events != 1)
1973 mddev->events--;
1974 else {
1975 /* otherwise we have to go forward and ... */
1976 mddev->events ++;
1977 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1978 /* .. if the array isn't clean, an 'even' event must also go
1979 * to spares. */
1980 if ((mddev->events&1)==0)
1981 nospares = 0;
1982 } else {
1983 /* otherwise an 'odd' event must go to spares */
1984 if ((mddev->events&1))
1985 nospares = 0;
1986 }
1987 }
1988
1989 if (!mddev->events) {
1990 /*
1991 * oops, this 64-bit counter should never wrap.
1992 * Either we are in around ~1 trillion A.C., assuming
1993 * 1 reboot per second, or we have a bug:
1994 */
1995 MD_BUG();
1996 mddev->events --;
1997 }
1998
1999 /*
2000 * do not write anything to disk if using
2001 * nonpersistent superblocks
2002 */
2003 if (!mddev->persistent) {
2004 if (!mddev->external)
2005 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2006
2007 spin_unlock_irq(&mddev->write_lock);
2008 wake_up(&mddev->sb_wait);
2009 return;
2010 }
2011 sync_sbs(mddev, nospares);
2012 spin_unlock_irq(&mddev->write_lock);
2013
2014 dprintk(KERN_INFO
2015 "md: updating %s RAID superblock on device (in sync %d)\n",
2016 mdname(mddev),mddev->in_sync);
2017
2018 bitmap_update_sb(mddev->bitmap);
2019 list_for_each_entry(rdev, &mddev->disks, same_set) {
2020 char b[BDEVNAME_SIZE];
2021 dprintk(KERN_INFO "md: ");
2022 if (rdev->sb_loaded != 1)
2023 continue; /* no noise on spare devices */
2024 if (test_bit(Faulty, &rdev->flags))
2025 dprintk("(skipping faulty ");
2026
2027 dprintk("%s ", bdevname(rdev->bdev,b));
2028 if (!test_bit(Faulty, &rdev->flags)) {
2029 md_super_write(mddev,rdev,
2030 rdev->sb_start, rdev->sb_size,
2031 rdev->sb_page);
2032 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2033 bdevname(rdev->bdev,b),
2034 (unsigned long long)rdev->sb_start);
2035 rdev->sb_events = mddev->events;
2036
2037 } else
2038 dprintk(")\n");
2039 if (mddev->level == LEVEL_MULTIPATH)
2040 /* only need to write one superblock... */
2041 break;
2042 }
2043 md_super_wait(mddev);
2044 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2045
2046 spin_lock_irq(&mddev->write_lock);
2047 if (mddev->in_sync != sync_req ||
2048 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2049 /* have to write it out again */
2050 spin_unlock_irq(&mddev->write_lock);
2051 goto repeat;
2052 }
2053 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2054 spin_unlock_irq(&mddev->write_lock);
2055 wake_up(&mddev->sb_wait);
2056 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2057 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2058
2059 }
2060
2061 /* words written to sysfs files may, or may not, be \n terminated.
2062 * We want to accept with case. For this we use cmd_match.
2063 */
2064 static int cmd_match(const char *cmd, const char *str)
2065 {
2066 /* See if cmd, written into a sysfs file, matches
2067 * str. They must either be the same, or cmd can
2068 * have a trailing newline
2069 */
2070 while (*cmd && *str && *cmd == *str) {
2071 cmd++;
2072 str++;
2073 }
2074 if (*cmd == '\n')
2075 cmd++;
2076 if (*str || *cmd)
2077 return 0;
2078 return 1;
2079 }
2080
2081 struct rdev_sysfs_entry {
2082 struct attribute attr;
2083 ssize_t (*show)(mdk_rdev_t *, char *);
2084 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2085 };
2086
2087 static ssize_t
2088 state_show(mdk_rdev_t *rdev, char *page)
2089 {
2090 char *sep = "";
2091 size_t len = 0;
2092
2093 if (test_bit(Faulty, &rdev->flags)) {
2094 len+= sprintf(page+len, "%sfaulty",sep);
2095 sep = ",";
2096 }
2097 if (test_bit(In_sync, &rdev->flags)) {
2098 len += sprintf(page+len, "%sin_sync",sep);
2099 sep = ",";
2100 }
2101 if (test_bit(WriteMostly, &rdev->flags)) {
2102 len += sprintf(page+len, "%swrite_mostly",sep);
2103 sep = ",";
2104 }
2105 if (test_bit(Blocked, &rdev->flags)) {
2106 len += sprintf(page+len, "%sblocked", sep);
2107 sep = ",";
2108 }
2109 if (!test_bit(Faulty, &rdev->flags) &&
2110 !test_bit(In_sync, &rdev->flags)) {
2111 len += sprintf(page+len, "%sspare", sep);
2112 sep = ",";
2113 }
2114 return len+sprintf(page+len, "\n");
2115 }
2116
2117 static ssize_t
2118 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2119 {
2120 /* can write
2121 * faulty - simulates and error
2122 * remove - disconnects the device
2123 * writemostly - sets write_mostly
2124 * -writemostly - clears write_mostly
2125 * blocked - sets the Blocked flag
2126 * -blocked - clears the Blocked flag
2127 * insync - sets Insync providing device isn't active
2128 */
2129 int err = -EINVAL;
2130 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2131 md_error(rdev->mddev, rdev);
2132 err = 0;
2133 } else if (cmd_match(buf, "remove")) {
2134 if (rdev->raid_disk >= 0)
2135 err = -EBUSY;
2136 else {
2137 mddev_t *mddev = rdev->mddev;
2138 kick_rdev_from_array(rdev);
2139 if (mddev->pers)
2140 md_update_sb(mddev, 1);
2141 md_new_event(mddev);
2142 err = 0;
2143 }
2144 } else if (cmd_match(buf, "writemostly")) {
2145 set_bit(WriteMostly, &rdev->flags);
2146 err = 0;
2147 } else if (cmd_match(buf, "-writemostly")) {
2148 clear_bit(WriteMostly, &rdev->flags);
2149 err = 0;
2150 } else if (cmd_match(buf, "blocked")) {
2151 set_bit(Blocked, &rdev->flags);
2152 err = 0;
2153 } else if (cmd_match(buf, "-blocked")) {
2154 clear_bit(Blocked, &rdev->flags);
2155 wake_up(&rdev->blocked_wait);
2156 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2157 md_wakeup_thread(rdev->mddev->thread);
2158
2159 err = 0;
2160 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2161 set_bit(In_sync, &rdev->flags);
2162 err = 0;
2163 }
2164 if (!err && rdev->sysfs_state)
2165 sysfs_notify_dirent(rdev->sysfs_state);
2166 return err ? err : len;
2167 }
2168 static struct rdev_sysfs_entry rdev_state =
2169 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2170
2171 static ssize_t
2172 errors_show(mdk_rdev_t *rdev, char *page)
2173 {
2174 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2175 }
2176
2177 static ssize_t
2178 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2179 {
2180 char *e;
2181 unsigned long n = simple_strtoul(buf, &e, 10);
2182 if (*buf && (*e == 0 || *e == '\n')) {
2183 atomic_set(&rdev->corrected_errors, n);
2184 return len;
2185 }
2186 return -EINVAL;
2187 }
2188 static struct rdev_sysfs_entry rdev_errors =
2189 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2190
2191 static ssize_t
2192 slot_show(mdk_rdev_t *rdev, char *page)
2193 {
2194 if (rdev->raid_disk < 0)
2195 return sprintf(page, "none\n");
2196 else
2197 return sprintf(page, "%d\n", rdev->raid_disk);
2198 }
2199
2200 static ssize_t
2201 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2202 {
2203 char *e;
2204 int err;
2205 char nm[20];
2206 int slot = simple_strtoul(buf, &e, 10);
2207 if (strncmp(buf, "none", 4)==0)
2208 slot = -1;
2209 else if (e==buf || (*e && *e!= '\n'))
2210 return -EINVAL;
2211 if (rdev->mddev->pers && slot == -1) {
2212 /* Setting 'slot' on an active array requires also
2213 * updating the 'rd%d' link, and communicating
2214 * with the personality with ->hot_*_disk.
2215 * For now we only support removing
2216 * failed/spare devices. This normally happens automatically,
2217 * but not when the metadata is externally managed.
2218 */
2219 if (rdev->raid_disk == -1)
2220 return -EEXIST;
2221 /* personality does all needed checks */
2222 if (rdev->mddev->pers->hot_add_disk == NULL)
2223 return -EINVAL;
2224 err = rdev->mddev->pers->
2225 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2226 if (err)
2227 return err;
2228 sprintf(nm, "rd%d", rdev->raid_disk);
2229 sysfs_remove_link(&rdev->mddev->kobj, nm);
2230 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2231 md_wakeup_thread(rdev->mddev->thread);
2232 } else if (rdev->mddev->pers) {
2233 mdk_rdev_t *rdev2;
2234 /* Activating a spare .. or possibly reactivating
2235 * if we ever get bitmaps working here.
2236 */
2237
2238 if (rdev->raid_disk != -1)
2239 return -EBUSY;
2240
2241 if (rdev->mddev->pers->hot_add_disk == NULL)
2242 return -EINVAL;
2243
2244 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2245 if (rdev2->raid_disk == slot)
2246 return -EEXIST;
2247
2248 rdev->raid_disk = slot;
2249 if (test_bit(In_sync, &rdev->flags))
2250 rdev->saved_raid_disk = slot;
2251 else
2252 rdev->saved_raid_disk = -1;
2253 err = rdev->mddev->pers->
2254 hot_add_disk(rdev->mddev, rdev);
2255 if (err) {
2256 rdev->raid_disk = -1;
2257 return err;
2258 } else
2259 sysfs_notify_dirent(rdev->sysfs_state);
2260 sprintf(nm, "rd%d", rdev->raid_disk);
2261 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2262 printk(KERN_WARNING
2263 "md: cannot register "
2264 "%s for %s\n",
2265 nm, mdname(rdev->mddev));
2266
2267 /* don't wakeup anyone, leave that to userspace. */
2268 } else {
2269 if (slot >= rdev->mddev->raid_disks)
2270 return -ENOSPC;
2271 rdev->raid_disk = slot;
2272 /* assume it is working */
2273 clear_bit(Faulty, &rdev->flags);
2274 clear_bit(WriteMostly, &rdev->flags);
2275 set_bit(In_sync, &rdev->flags);
2276 sysfs_notify_dirent(rdev->sysfs_state);
2277 }
2278 return len;
2279 }
2280
2281
2282 static struct rdev_sysfs_entry rdev_slot =
2283 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2284
2285 static ssize_t
2286 offset_show(mdk_rdev_t *rdev, char *page)
2287 {
2288 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2289 }
2290
2291 static ssize_t
2292 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2293 {
2294 char *e;
2295 unsigned long long offset = simple_strtoull(buf, &e, 10);
2296 if (e==buf || (*e && *e != '\n'))
2297 return -EINVAL;
2298 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2299 return -EBUSY;
2300 if (rdev->sectors && rdev->mddev->external)
2301 /* Must set offset before size, so overlap checks
2302 * can be sane */
2303 return -EBUSY;
2304 rdev->data_offset = offset;
2305 return len;
2306 }
2307
2308 static struct rdev_sysfs_entry rdev_offset =
2309 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2310
2311 static ssize_t
2312 rdev_size_show(mdk_rdev_t *rdev, char *page)
2313 {
2314 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2315 }
2316
2317 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2318 {
2319 /* check if two start/length pairs overlap */
2320 if (s1+l1 <= s2)
2321 return 0;
2322 if (s2+l2 <= s1)
2323 return 0;
2324 return 1;
2325 }
2326
2327 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2328 {
2329 unsigned long long blocks;
2330 sector_t new;
2331
2332 if (strict_strtoull(buf, 10, &blocks) < 0)
2333 return -EINVAL;
2334
2335 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2336 return -EINVAL; /* sector conversion overflow */
2337
2338 new = blocks * 2;
2339 if (new != blocks * 2)
2340 return -EINVAL; /* unsigned long long to sector_t overflow */
2341
2342 *sectors = new;
2343 return 0;
2344 }
2345
2346 static ssize_t
2347 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2348 {
2349 mddev_t *my_mddev = rdev->mddev;
2350 sector_t oldsectors = rdev->sectors;
2351 sector_t sectors;
2352
2353 if (strict_blocks_to_sectors(buf, &sectors) < 0)
2354 return -EINVAL;
2355 if (my_mddev->pers && rdev->raid_disk >= 0) {
2356 if (my_mddev->persistent) {
2357 sectors = super_types[my_mddev->major_version].
2358 rdev_size_change(rdev, sectors);
2359 if (!sectors)
2360 return -EBUSY;
2361 } else if (!sectors)
2362 sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2363 rdev->data_offset;
2364 }
2365 if (sectors < my_mddev->dev_sectors)
2366 return -EINVAL; /* component must fit device */
2367
2368 rdev->sectors = sectors;
2369 if (sectors > oldsectors && my_mddev->external) {
2370 /* need to check that all other rdevs with the same ->bdev
2371 * do not overlap. We need to unlock the mddev to avoid
2372 * a deadlock. We have already changed rdev->sectors, and if
2373 * we have to change it back, we will have the lock again.
2374 */
2375 mddev_t *mddev;
2376 int overlap = 0;
2377 struct list_head *tmp;
2378
2379 mddev_unlock(my_mddev);
2380 for_each_mddev(mddev, tmp) {
2381 mdk_rdev_t *rdev2;
2382
2383 mddev_lock(mddev);
2384 list_for_each_entry(rdev2, &mddev->disks, same_set)
2385 if (test_bit(AllReserved, &rdev2->flags) ||
2386 (rdev->bdev == rdev2->bdev &&
2387 rdev != rdev2 &&
2388 overlaps(rdev->data_offset, rdev->sectors,
2389 rdev2->data_offset,
2390 rdev2->sectors))) {
2391 overlap = 1;
2392 break;
2393 }
2394 mddev_unlock(mddev);
2395 if (overlap) {
2396 mddev_put(mddev);
2397 break;
2398 }
2399 }
2400 mddev_lock(my_mddev);
2401 if (overlap) {
2402 /* Someone else could have slipped in a size
2403 * change here, but doing so is just silly.
2404 * We put oldsectors back because we *know* it is
2405 * safe, and trust userspace not to race with
2406 * itself
2407 */
2408 rdev->sectors = oldsectors;
2409 return -EBUSY;
2410 }
2411 }
2412 return len;
2413 }
2414
2415 static struct rdev_sysfs_entry rdev_size =
2416 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2417
2418 static struct attribute *rdev_default_attrs[] = {
2419 &rdev_state.attr,
2420 &rdev_errors.attr,
2421 &rdev_slot.attr,
2422 &rdev_offset.attr,
2423 &rdev_size.attr,
2424 NULL,
2425 };
2426 static ssize_t
2427 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2428 {
2429 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2430 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2431 mddev_t *mddev = rdev->mddev;
2432 ssize_t rv;
2433
2434 if (!entry->show)
2435 return -EIO;
2436
2437 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2438 if (!rv) {
2439 if (rdev->mddev == NULL)
2440 rv = -EBUSY;
2441 else
2442 rv = entry->show(rdev, page);
2443 mddev_unlock(mddev);
2444 }
2445 return rv;
2446 }
2447
2448 static ssize_t
2449 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2450 const char *page, size_t length)
2451 {
2452 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2453 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2454 ssize_t rv;
2455 mddev_t *mddev = rdev->mddev;
2456
2457 if (!entry->store)
2458 return -EIO;
2459 if (!capable(CAP_SYS_ADMIN))
2460 return -EACCES;
2461 rv = mddev ? mddev_lock(mddev): -EBUSY;
2462 if (!rv) {
2463 if (rdev->mddev == NULL)
2464 rv = -EBUSY;
2465 else
2466 rv = entry->store(rdev, page, length);
2467 mddev_unlock(mddev);
2468 }
2469 return rv;
2470 }
2471
2472 static void rdev_free(struct kobject *ko)
2473 {
2474 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2475 kfree(rdev);
2476 }
2477 static struct sysfs_ops rdev_sysfs_ops = {
2478 .show = rdev_attr_show,
2479 .store = rdev_attr_store,
2480 };
2481 static struct kobj_type rdev_ktype = {
2482 .release = rdev_free,
2483 .sysfs_ops = &rdev_sysfs_ops,
2484 .default_attrs = rdev_default_attrs,
2485 };
2486
2487 /*
2488 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2489 *
2490 * mark the device faulty if:
2491 *
2492 * - the device is nonexistent (zero size)
2493 * - the device has no valid superblock
2494 *
2495 * a faulty rdev _never_ has rdev->sb set.
2496 */
2497 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2498 {
2499 char b[BDEVNAME_SIZE];
2500 int err;
2501 mdk_rdev_t *rdev;
2502 sector_t size;
2503
2504 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2505 if (!rdev) {
2506 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2507 return ERR_PTR(-ENOMEM);
2508 }
2509
2510 if ((err = alloc_disk_sb(rdev)))
2511 goto abort_free;
2512
2513 err = lock_rdev(rdev, newdev, super_format == -2);
2514 if (err)
2515 goto abort_free;
2516
2517 kobject_init(&rdev->kobj, &rdev_ktype);
2518
2519 rdev->desc_nr = -1;
2520 rdev->saved_raid_disk = -1;
2521 rdev->raid_disk = -1;
2522 rdev->flags = 0;
2523 rdev->data_offset = 0;
2524 rdev->sb_events = 0;
2525 atomic_set(&rdev->nr_pending, 0);
2526 atomic_set(&rdev->read_errors, 0);
2527 atomic_set(&rdev->corrected_errors, 0);
2528
2529 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2530 if (!size) {
2531 printk(KERN_WARNING
2532 "md: %s has zero or unknown size, marking faulty!\n",
2533 bdevname(rdev->bdev,b));
2534 err = -EINVAL;
2535 goto abort_free;
2536 }
2537
2538 if (super_format >= 0) {
2539 err = super_types[super_format].
2540 load_super(rdev, NULL, super_minor);
2541 if (err == -EINVAL) {
2542 printk(KERN_WARNING
2543 "md: %s does not have a valid v%d.%d "
2544 "superblock, not importing!\n",
2545 bdevname(rdev->bdev,b),
2546 super_format, super_minor);
2547 goto abort_free;
2548 }
2549 if (err < 0) {
2550 printk(KERN_WARNING
2551 "md: could not read %s's sb, not importing!\n",
2552 bdevname(rdev->bdev,b));
2553 goto abort_free;
2554 }
2555 }
2556
2557 INIT_LIST_HEAD(&rdev->same_set);
2558 init_waitqueue_head(&rdev->blocked_wait);
2559
2560 return rdev;
2561
2562 abort_free:
2563 if (rdev->sb_page) {
2564 if (rdev->bdev)
2565 unlock_rdev(rdev);
2566 free_disk_sb(rdev);
2567 }
2568 kfree(rdev);
2569 return ERR_PTR(err);
2570 }
2571
2572 /*
2573 * Check a full RAID array for plausibility
2574 */
2575
2576
2577 static void analyze_sbs(mddev_t * mddev)
2578 {
2579 int i;
2580 mdk_rdev_t *rdev, *freshest, *tmp;
2581 char b[BDEVNAME_SIZE];
2582
2583 freshest = NULL;
2584 rdev_for_each(rdev, tmp, mddev)
2585 switch (super_types[mddev->major_version].
2586 load_super(rdev, freshest, mddev->minor_version)) {
2587 case 1:
2588 freshest = rdev;
2589 break;
2590 case 0:
2591 break;
2592 default:
2593 printk( KERN_ERR \
2594 "md: fatal superblock inconsistency in %s"
2595 " -- removing from array\n",
2596 bdevname(rdev->bdev,b));
2597 kick_rdev_from_array(rdev);
2598 }
2599
2600
2601 super_types[mddev->major_version].
2602 validate_super(mddev, freshest);
2603
2604 i = 0;
2605 rdev_for_each(rdev, tmp, mddev) {
2606 if (rdev->desc_nr >= mddev->max_disks ||
2607 i > mddev->max_disks) {
2608 printk(KERN_WARNING
2609 "md: %s: %s: only %d devices permitted\n",
2610 mdname(mddev), bdevname(rdev->bdev, b),
2611 mddev->max_disks);
2612 kick_rdev_from_array(rdev);
2613 continue;
2614 }
2615 if (rdev != freshest)
2616 if (super_types[mddev->major_version].
2617 validate_super(mddev, rdev)) {
2618 printk(KERN_WARNING "md: kicking non-fresh %s"
2619 " from array!\n",
2620 bdevname(rdev->bdev,b));
2621 kick_rdev_from_array(rdev);
2622 continue;
2623 }
2624 if (mddev->level == LEVEL_MULTIPATH) {
2625 rdev->desc_nr = i++;
2626 rdev->raid_disk = rdev->desc_nr;
2627 set_bit(In_sync, &rdev->flags);
2628 } else if (rdev->raid_disk >= mddev->raid_disks) {
2629 rdev->raid_disk = -1;
2630 clear_bit(In_sync, &rdev->flags);
2631 }
2632 }
2633 }
2634
2635 static void md_safemode_timeout(unsigned long data);
2636
2637 static ssize_t
2638 safe_delay_show(mddev_t *mddev, char *page)
2639 {
2640 int msec = (mddev->safemode_delay*1000)/HZ;
2641 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2642 }
2643 static ssize_t
2644 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2645 {
2646 int scale=1;
2647 int dot=0;
2648 int i;
2649 unsigned long msec;
2650 char buf[30];
2651
2652 /* remove a period, and count digits after it */
2653 if (len >= sizeof(buf))
2654 return -EINVAL;
2655 strlcpy(buf, cbuf, sizeof(buf));
2656 for (i=0; i<len; i++) {
2657 if (dot) {
2658 if (isdigit(buf[i])) {
2659 buf[i-1] = buf[i];
2660 scale *= 10;
2661 }
2662 buf[i] = 0;
2663 } else if (buf[i] == '.') {
2664 dot=1;
2665 buf[i] = 0;
2666 }
2667 }
2668 if (strict_strtoul(buf, 10, &msec) < 0)
2669 return -EINVAL;
2670 msec = (msec * 1000) / scale;
2671 if (msec == 0)
2672 mddev->safemode_delay = 0;
2673 else {
2674 unsigned long old_delay = mddev->safemode_delay;
2675 mddev->safemode_delay = (msec*HZ)/1000;
2676 if (mddev->safemode_delay == 0)
2677 mddev->safemode_delay = 1;
2678 if (mddev->safemode_delay < old_delay)
2679 md_safemode_timeout((unsigned long)mddev);
2680 }
2681 return len;
2682 }
2683 static struct md_sysfs_entry md_safe_delay =
2684 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2685
2686 static ssize_t
2687 level_show(mddev_t *mddev, char *page)
2688 {
2689 struct mdk_personality *p = mddev->pers;
2690 if (p)
2691 return sprintf(page, "%s\n", p->name);
2692 else if (mddev->clevel[0])
2693 return sprintf(page, "%s\n", mddev->clevel);
2694 else if (mddev->level != LEVEL_NONE)
2695 return sprintf(page, "%d\n", mddev->level);
2696 else
2697 return 0;
2698 }
2699
2700 static ssize_t
2701 level_store(mddev_t *mddev, const char *buf, size_t len)
2702 {
2703 char level[16];
2704 ssize_t rv = len;
2705 struct mdk_personality *pers;
2706 void *priv;
2707 mdk_rdev_t *rdev;
2708
2709 if (mddev->pers == NULL) {
2710 if (len == 0)
2711 return 0;
2712 if (len >= sizeof(mddev->clevel))
2713 return -ENOSPC;
2714 strncpy(mddev->clevel, buf, len);
2715 if (mddev->clevel[len-1] == '\n')
2716 len--;
2717 mddev->clevel[len] = 0;
2718 mddev->level = LEVEL_NONE;
2719 return rv;
2720 }
2721
2722 /* request to change the personality. Need to ensure:
2723 * - array is not engaged in resync/recovery/reshape
2724 * - old personality can be suspended
2725 * - new personality will access other array.
2726 */
2727
2728 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
2729 return -EBUSY;
2730
2731 if (!mddev->pers->quiesce) {
2732 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
2733 mdname(mddev), mddev->pers->name);
2734 return -EINVAL;
2735 }
2736
2737 /* Now find the new personality */
2738 if (len == 0 || len >= sizeof(level))
2739 return -EINVAL;
2740 strncpy(level, buf, len);
2741 if (level[len-1] == '\n')
2742 len--;
2743 level[len] = 0;
2744
2745 request_module("md-%s", level);
2746 spin_lock(&pers_lock);
2747 pers = find_pers(LEVEL_NONE, level);
2748 if (!pers || !try_module_get(pers->owner)) {
2749 spin_unlock(&pers_lock);
2750 printk(KERN_WARNING "md: personality %s not loaded\n", level);
2751 return -EINVAL;
2752 }
2753 spin_unlock(&pers_lock);
2754
2755 if (pers == mddev->pers) {
2756 /* Nothing to do! */
2757 module_put(pers->owner);
2758 return rv;
2759 }
2760 if (!pers->takeover) {
2761 module_put(pers->owner);
2762 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
2763 mdname(mddev), level);
2764 return -EINVAL;
2765 }
2766
2767 /* ->takeover must set new_* and/or delta_disks
2768 * if it succeeds, and may set them when it fails.
2769 */
2770 priv = pers->takeover(mddev);
2771 if (IS_ERR(priv)) {
2772 mddev->new_level = mddev->level;
2773 mddev->new_layout = mddev->layout;
2774 mddev->new_chunk_sectors = mddev->chunk_sectors;
2775 mddev->raid_disks -= mddev->delta_disks;
2776 mddev->delta_disks = 0;
2777 module_put(pers->owner);
2778 printk(KERN_WARNING "md: %s: %s would not accept array\n",
2779 mdname(mddev), level);
2780 return PTR_ERR(priv);
2781 }
2782
2783 /* Looks like we have a winner */
2784 mddev_suspend(mddev);
2785 mddev->pers->stop(mddev);
2786 module_put(mddev->pers->owner);
2787 /* Invalidate devices that are now superfluous */
2788 list_for_each_entry(rdev, &mddev->disks, same_set)
2789 if (rdev->raid_disk >= mddev->raid_disks) {
2790 rdev->raid_disk = -1;
2791 clear_bit(In_sync, &rdev->flags);
2792 }
2793 mddev->pers = pers;
2794 mddev->private = priv;
2795 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
2796 mddev->level = mddev->new_level;
2797 mddev->layout = mddev->new_layout;
2798 mddev->chunk_sectors = mddev->new_chunk_sectors;
2799 mddev->delta_disks = 0;
2800 pers->run(mddev);
2801 mddev_resume(mddev);
2802 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2803 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2804 md_wakeup_thread(mddev->thread);
2805 return rv;
2806 }
2807
2808 static struct md_sysfs_entry md_level =
2809 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2810
2811
2812 static ssize_t
2813 layout_show(mddev_t *mddev, char *page)
2814 {
2815 /* just a number, not meaningful for all levels */
2816 if (mddev->reshape_position != MaxSector &&
2817 mddev->layout != mddev->new_layout)
2818 return sprintf(page, "%d (%d)\n",
2819 mddev->new_layout, mddev->layout);
2820 return sprintf(page, "%d\n", mddev->layout);
2821 }
2822
2823 static ssize_t
2824 layout_store(mddev_t *mddev, const char *buf, size_t len)
2825 {
2826 char *e;
2827 unsigned long n = simple_strtoul(buf, &e, 10);
2828
2829 if (!*buf || (*e && *e != '\n'))
2830 return -EINVAL;
2831
2832 if (mddev->pers) {
2833 int err;
2834 if (mddev->pers->check_reshape == NULL)
2835 return -EBUSY;
2836 mddev->new_layout = n;
2837 err = mddev->pers->check_reshape(mddev);
2838 if (err) {
2839 mddev->new_layout = mddev->layout;
2840 return err;
2841 }
2842 } else {
2843 mddev->new_layout = n;
2844 if (mddev->reshape_position == MaxSector)
2845 mddev->layout = n;
2846 }
2847 return len;
2848 }
2849 static struct md_sysfs_entry md_layout =
2850 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2851
2852
2853 static ssize_t
2854 raid_disks_show(mddev_t *mddev, char *page)
2855 {
2856 if (mddev->raid_disks == 0)
2857 return 0;
2858 if (mddev->reshape_position != MaxSector &&
2859 mddev->delta_disks != 0)
2860 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2861 mddev->raid_disks - mddev->delta_disks);
2862 return sprintf(page, "%d\n", mddev->raid_disks);
2863 }
2864
2865 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2866
2867 static ssize_t
2868 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2869 {
2870 char *e;
2871 int rv = 0;
2872 unsigned long n = simple_strtoul(buf, &e, 10);
2873
2874 if (!*buf || (*e && *e != '\n'))
2875 return -EINVAL;
2876
2877 if (mddev->pers)
2878 rv = update_raid_disks(mddev, n);
2879 else if (mddev->reshape_position != MaxSector) {
2880 int olddisks = mddev->raid_disks - mddev->delta_disks;
2881 mddev->delta_disks = n - olddisks;
2882 mddev->raid_disks = n;
2883 } else
2884 mddev->raid_disks = n;
2885 return rv ? rv : len;
2886 }
2887 static struct md_sysfs_entry md_raid_disks =
2888 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2889
2890 static ssize_t
2891 chunk_size_show(mddev_t *mddev, char *page)
2892 {
2893 if (mddev->reshape_position != MaxSector &&
2894 mddev->chunk_sectors != mddev->new_chunk_sectors)
2895 return sprintf(page, "%d (%d)\n",
2896 mddev->new_chunk_sectors << 9,
2897 mddev->chunk_sectors << 9);
2898 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
2899 }
2900
2901 static ssize_t
2902 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2903 {
2904 char *e;
2905 unsigned long n = simple_strtoul(buf, &e, 10);
2906
2907 if (!*buf || (*e && *e != '\n'))
2908 return -EINVAL;
2909
2910 if (mddev->pers) {
2911 int err;
2912 if (mddev->pers->check_reshape == NULL)
2913 return -EBUSY;
2914 mddev->new_chunk_sectors = n >> 9;
2915 err = mddev->pers->check_reshape(mddev);
2916 if (err) {
2917 mddev->new_chunk_sectors = mddev->chunk_sectors;
2918 return err;
2919 }
2920 } else {
2921 mddev->new_chunk_sectors = n >> 9;
2922 if (mddev->reshape_position == MaxSector)
2923 mddev->chunk_sectors = n >> 9;
2924 }
2925 return len;
2926 }
2927 static struct md_sysfs_entry md_chunk_size =
2928 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2929
2930 static ssize_t
2931 resync_start_show(mddev_t *mddev, char *page)
2932 {
2933 if (mddev->recovery_cp == MaxSector)
2934 return sprintf(page, "none\n");
2935 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2936 }
2937
2938 static ssize_t
2939 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2940 {
2941 char *e;
2942 unsigned long long n = simple_strtoull(buf, &e, 10);
2943
2944 if (mddev->pers)
2945 return -EBUSY;
2946 if (!*buf || (*e && *e != '\n'))
2947 return -EINVAL;
2948
2949 mddev->recovery_cp = n;
2950 return len;
2951 }
2952 static struct md_sysfs_entry md_resync_start =
2953 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2954
2955 /*
2956 * The array state can be:
2957 *
2958 * clear
2959 * No devices, no size, no level
2960 * Equivalent to STOP_ARRAY ioctl
2961 * inactive
2962 * May have some settings, but array is not active
2963 * all IO results in error
2964 * When written, doesn't tear down array, but just stops it
2965 * suspended (not supported yet)
2966 * All IO requests will block. The array can be reconfigured.
2967 * Writing this, if accepted, will block until array is quiescent
2968 * readonly
2969 * no resync can happen. no superblocks get written.
2970 * write requests fail
2971 * read-auto
2972 * like readonly, but behaves like 'clean' on a write request.
2973 *
2974 * clean - no pending writes, but otherwise active.
2975 * When written to inactive array, starts without resync
2976 * If a write request arrives then
2977 * if metadata is known, mark 'dirty' and switch to 'active'.
2978 * if not known, block and switch to write-pending
2979 * If written to an active array that has pending writes, then fails.
2980 * active
2981 * fully active: IO and resync can be happening.
2982 * When written to inactive array, starts with resync
2983 *
2984 * write-pending
2985 * clean, but writes are blocked waiting for 'active' to be written.
2986 *
2987 * active-idle
2988 * like active, but no writes have been seen for a while (100msec).
2989 *
2990 */
2991 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2992 write_pending, active_idle, bad_word};
2993 static char *array_states[] = {
2994 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2995 "write-pending", "active-idle", NULL };
2996
2997 static int match_word(const char *word, char **list)
2998 {
2999 int n;
3000 for (n=0; list[n]; n++)
3001 if (cmd_match(word, list[n]))
3002 break;
3003 return n;
3004 }
3005
3006 static ssize_t
3007 array_state_show(mddev_t *mddev, char *page)
3008 {
3009 enum array_state st = inactive;
3010
3011 if (mddev->pers)
3012 switch(mddev->ro) {
3013 case 1:
3014 st = readonly;
3015 break;
3016 case 2:
3017 st = read_auto;
3018 break;
3019 case 0:
3020 if (mddev->in_sync)
3021 st = clean;
3022 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
3023 st = write_pending;
3024 else if (mddev->safemode)
3025 st = active_idle;
3026 else
3027 st = active;
3028 }
3029 else {
3030 if (list_empty(&mddev->disks) &&
3031 mddev->raid_disks == 0 &&
3032 mddev->dev_sectors == 0)
3033 st = clear;
3034 else
3035 st = inactive;
3036 }
3037 return sprintf(page, "%s\n", array_states[st]);
3038 }
3039
3040 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3041 static int do_md_run(mddev_t * mddev);
3042 static int restart_array(mddev_t *mddev);
3043
3044 static ssize_t
3045 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3046 {
3047 int err = -EINVAL;
3048 enum array_state st = match_word(buf, array_states);
3049 switch(st) {
3050 case bad_word:
3051 break;
3052 case clear:
3053 /* stopping an active array */
3054 if (atomic_read(&mddev->openers) > 0)
3055 return -EBUSY;
3056 err = do_md_stop(mddev, 0, 0);
3057 break;
3058 case inactive:
3059 /* stopping an active array */
3060 if (mddev->pers) {
3061 if (atomic_read(&mddev->openers) > 0)
3062 return -EBUSY;
3063 err = do_md_stop(mddev, 2, 0);
3064 } else
3065 err = 0; /* already inactive */
3066 break;
3067 case suspended:
3068 break; /* not supported yet */
3069 case readonly:
3070 if (mddev->pers)
3071 err = do_md_stop(mddev, 1, 0);
3072 else {
3073 mddev->ro = 1;
3074 set_disk_ro(mddev->gendisk, 1);
3075 err = do_md_run(mddev);
3076 }
3077 break;
3078 case read_auto:
3079 if (mddev->pers) {
3080 if (mddev->ro == 0)
3081 err = do_md_stop(mddev, 1, 0);
3082 else if (mddev->ro == 1)
3083 err = restart_array(mddev);
3084 if (err == 0) {
3085 mddev->ro = 2;
3086 set_disk_ro(mddev->gendisk, 0);
3087 }
3088 } else {
3089 mddev->ro = 2;
3090 err = do_md_run(mddev);
3091 }
3092 break;
3093 case clean:
3094 if (mddev->pers) {
3095 restart_array(mddev);
3096 spin_lock_irq(&mddev->write_lock);
3097 if (atomic_read(&mddev->writes_pending) == 0) {
3098 if (mddev->in_sync == 0) {
3099 mddev->in_sync = 1;
3100 if (mddev->safemode == 1)
3101 mddev->safemode = 0;
3102 if (mddev->persistent)
3103 set_bit(MD_CHANGE_CLEAN,
3104 &mddev->flags);
3105 }
3106 err = 0;
3107 } else
3108 err = -EBUSY;
3109 spin_unlock_irq(&mddev->write_lock);
3110 } else
3111 err = -EINVAL;
3112 break;
3113 case active:
3114 if (mddev->pers) {
3115 restart_array(mddev);
3116 if (mddev->external)
3117 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3118 wake_up(&mddev->sb_wait);
3119 err = 0;
3120 } else {
3121 mddev->ro = 0;
3122 set_disk_ro(mddev->gendisk, 0);
3123 err = do_md_run(mddev);
3124 }
3125 break;
3126 case write_pending:
3127 case active_idle:
3128 /* these cannot be set */
3129 break;
3130 }
3131 if (err)
3132 return err;
3133 else {
3134 sysfs_notify_dirent(mddev->sysfs_state);
3135 return len;
3136 }
3137 }
3138 static struct md_sysfs_entry md_array_state =
3139 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3140
3141 static ssize_t
3142 null_show(mddev_t *mddev, char *page)
3143 {
3144 return -EINVAL;
3145 }
3146
3147 static ssize_t
3148 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3149 {
3150 /* buf must be %d:%d\n? giving major and minor numbers */
3151 /* The new device is added to the array.
3152 * If the array has a persistent superblock, we read the
3153 * superblock to initialise info and check validity.
3154 * Otherwise, only checking done is that in bind_rdev_to_array,
3155 * which mainly checks size.
3156 */
3157 char *e;
3158 int major = simple_strtoul(buf, &e, 10);
3159 int minor;
3160 dev_t dev;
3161 mdk_rdev_t *rdev;
3162 int err;
3163
3164 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3165 return -EINVAL;
3166 minor = simple_strtoul(e+1, &e, 10);
3167 if (*e && *e != '\n')
3168 return -EINVAL;
3169 dev = MKDEV(major, minor);
3170 if (major != MAJOR(dev) ||
3171 minor != MINOR(dev))
3172 return -EOVERFLOW;
3173
3174
3175 if (mddev->persistent) {
3176 rdev = md_import_device(dev, mddev->major_version,
3177 mddev->minor_version);
3178 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3179 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3180 mdk_rdev_t, same_set);
3181 err = super_types[mddev->major_version]
3182 .load_super(rdev, rdev0, mddev->minor_version);
3183 if (err < 0)
3184 goto out;
3185 }
3186 } else if (mddev->external)
3187 rdev = md_import_device(dev, -2, -1);
3188 else
3189 rdev = md_import_device(dev, -1, -1);
3190
3191 if (IS_ERR(rdev))
3192 return PTR_ERR(rdev);
3193 err = bind_rdev_to_array(rdev, mddev);
3194 out:
3195 if (err)
3196 export_rdev(rdev);
3197 return err ? err : len;
3198 }
3199
3200 static struct md_sysfs_entry md_new_device =
3201 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3202
3203 static ssize_t
3204 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3205 {
3206 char *end;
3207 unsigned long chunk, end_chunk;
3208
3209 if (!mddev->bitmap)
3210 goto out;
3211 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3212 while (*buf) {
3213 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3214 if (buf == end) break;
3215 if (*end == '-') { /* range */
3216 buf = end + 1;
3217 end_chunk = simple_strtoul(buf, &end, 0);
3218 if (buf == end) break;
3219 }
3220 if (*end && !isspace(*end)) break;
3221 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3222 buf = end;
3223 while (isspace(*buf)) buf++;
3224 }
3225 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3226 out:
3227 return len;
3228 }
3229
3230 static struct md_sysfs_entry md_bitmap =
3231 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3232
3233 static ssize_t
3234 size_show(mddev_t *mddev, char *page)
3235 {
3236 return sprintf(page, "%llu\n",
3237 (unsigned long long)mddev->dev_sectors / 2);
3238 }
3239
3240 static int update_size(mddev_t *mddev, sector_t num_sectors);
3241
3242 static ssize_t
3243 size_store(mddev_t *mddev, const char *buf, size_t len)
3244 {
3245 /* If array is inactive, we can reduce the component size, but
3246 * not increase it (except from 0).
3247 * If array is active, we can try an on-line resize
3248 */
3249 sector_t sectors;
3250 int err = strict_blocks_to_sectors(buf, &sectors);
3251
3252 if (err < 0)
3253 return err;
3254 if (mddev->pers) {
3255 err = update_size(mddev, sectors);
3256 md_update_sb(mddev, 1);
3257 } else {
3258 if (mddev->dev_sectors == 0 ||
3259 mddev->dev_sectors > sectors)
3260 mddev->dev_sectors = sectors;
3261 else
3262 err = -ENOSPC;
3263 }
3264 return err ? err : len;
3265 }
3266
3267 static struct md_sysfs_entry md_size =
3268 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3269
3270
3271 /* Metdata version.
3272 * This is one of
3273 * 'none' for arrays with no metadata (good luck...)
3274 * 'external' for arrays with externally managed metadata,
3275 * or N.M for internally known formats
3276 */
3277 static ssize_t
3278 metadata_show(mddev_t *mddev, char *page)
3279 {
3280 if (mddev->persistent)
3281 return sprintf(page, "%d.%d\n",
3282 mddev->major_version, mddev->minor_version);
3283 else if (mddev->external)
3284 return sprintf(page, "external:%s\n", mddev->metadata_type);
3285 else
3286 return sprintf(page, "none\n");
3287 }
3288
3289 static ssize_t
3290 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3291 {
3292 int major, minor;
3293 char *e;
3294 /* Changing the details of 'external' metadata is
3295 * always permitted. Otherwise there must be
3296 * no devices attached to the array.
3297 */
3298 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3299 ;
3300 else if (!list_empty(&mddev->disks))
3301 return -EBUSY;
3302
3303 if (cmd_match(buf, "none")) {
3304 mddev->persistent = 0;
3305 mddev->external = 0;
3306 mddev->major_version = 0;
3307 mddev->minor_version = 90;
3308 return len;
3309 }
3310 if (strncmp(buf, "external:", 9) == 0) {
3311 size_t namelen = len-9;
3312 if (namelen >= sizeof(mddev->metadata_type))
3313 namelen = sizeof(mddev->metadata_type)-1;
3314 strncpy(mddev->metadata_type, buf+9, namelen);
3315 mddev->metadata_type[namelen] = 0;
3316 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3317 mddev->metadata_type[--namelen] = 0;
3318 mddev->persistent = 0;
3319 mddev->external = 1;
3320 mddev->major_version = 0;
3321 mddev->minor_version = 90;
3322 return len;
3323 }
3324 major = simple_strtoul(buf, &e, 10);
3325 if (e==buf || *e != '.')
3326 return -EINVAL;
3327 buf = e+1;
3328 minor = simple_strtoul(buf, &e, 10);
3329 if (e==buf || (*e && *e != '\n') )
3330 return -EINVAL;
3331 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3332 return -ENOENT;
3333 mddev->major_version = major;
3334 mddev->minor_version = minor;
3335 mddev->persistent = 1;
3336 mddev->external = 0;
3337 return len;
3338 }
3339
3340 static struct md_sysfs_entry md_metadata =
3341 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3342
3343 static ssize_t
3344 action_show(mddev_t *mddev, char *page)
3345 {
3346 char *type = "idle";
3347 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3348 type = "frozen";
3349 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3350 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3351 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3352 type = "reshape";
3353 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3354 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3355 type = "resync";
3356 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3357 type = "check";
3358 else
3359 type = "repair";
3360 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3361 type = "recover";
3362 }
3363 return sprintf(page, "%s\n", type);
3364 }
3365
3366 static ssize_t
3367 action_store(mddev_t *mddev, const char *page, size_t len)
3368 {
3369 if (!mddev->pers || !mddev->pers->sync_request)
3370 return -EINVAL;
3371
3372 if (cmd_match(page, "frozen"))
3373 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3374 else
3375 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3376
3377 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3378 if (mddev->sync_thread) {
3379 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3380 md_unregister_thread(mddev->sync_thread);
3381 mddev->sync_thread = NULL;
3382 mddev->recovery = 0;
3383 }
3384 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3385 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3386 return -EBUSY;
3387 else if (cmd_match(page, "resync"))
3388 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3389 else if (cmd_match(page, "recover")) {
3390 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3391 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3392 } else if (cmd_match(page, "reshape")) {
3393 int err;
3394 if (mddev->pers->start_reshape == NULL)
3395 return -EINVAL;
3396 err = mddev->pers->start_reshape(mddev);
3397 if (err)
3398 return err;
3399 sysfs_notify(&mddev->kobj, NULL, "degraded");
3400 } else {
3401 if (cmd_match(page, "check"))
3402 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3403 else if (!cmd_match(page, "repair"))
3404 return -EINVAL;
3405 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3406 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3407 }
3408 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3409 md_wakeup_thread(mddev->thread);
3410 sysfs_notify_dirent(mddev->sysfs_action);
3411 return len;
3412 }
3413
3414 static ssize_t
3415 mismatch_cnt_show(mddev_t *mddev, char *page)
3416 {
3417 return sprintf(page, "%llu\n",
3418 (unsigned long long) mddev->resync_mismatches);
3419 }
3420
3421 static struct md_sysfs_entry md_scan_mode =
3422 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3423
3424
3425 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3426
3427 static ssize_t
3428 sync_min_show(mddev_t *mddev, char *page)
3429 {
3430 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3431 mddev->sync_speed_min ? "local": "system");
3432 }
3433
3434 static ssize_t
3435 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3436 {
3437 int min;
3438 char *e;
3439 if (strncmp(buf, "system", 6)==0) {
3440 mddev->sync_speed_min = 0;
3441 return len;
3442 }
3443 min = simple_strtoul(buf, &e, 10);
3444 if (buf == e || (*e && *e != '\n') || min <= 0)
3445 return -EINVAL;
3446 mddev->sync_speed_min = min;
3447 return len;
3448 }
3449
3450 static struct md_sysfs_entry md_sync_min =
3451 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3452
3453 static ssize_t
3454 sync_max_show(mddev_t *mddev, char *page)
3455 {
3456 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3457 mddev->sync_speed_max ? "local": "system");
3458 }
3459
3460 static ssize_t
3461 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3462 {
3463 int max;
3464 char *e;
3465 if (strncmp(buf, "system", 6)==0) {
3466 mddev->sync_speed_max = 0;
3467 return len;
3468 }
3469 max = simple_strtoul(buf, &e, 10);
3470 if (buf == e || (*e && *e != '\n') || max <= 0)
3471 return -EINVAL;
3472 mddev->sync_speed_max = max;
3473 return len;
3474 }
3475
3476 static struct md_sysfs_entry md_sync_max =
3477 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3478
3479 static ssize_t
3480 degraded_show(mddev_t *mddev, char *page)
3481 {
3482 return sprintf(page, "%d\n", mddev->degraded);
3483 }
3484 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3485
3486 static ssize_t
3487 sync_force_parallel_show(mddev_t *mddev, char *page)
3488 {
3489 return sprintf(page, "%d\n", mddev->parallel_resync);
3490 }
3491
3492 static ssize_t
3493 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3494 {
3495 long n;
3496
3497 if (strict_strtol(buf, 10, &n))
3498 return -EINVAL;
3499
3500 if (n != 0 && n != 1)
3501 return -EINVAL;
3502
3503 mddev->parallel_resync = n;
3504
3505 if (mddev->sync_thread)
3506 wake_up(&resync_wait);
3507
3508 return len;
3509 }
3510
3511 /* force parallel resync, even with shared block devices */
3512 static struct md_sysfs_entry md_sync_force_parallel =
3513 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3514 sync_force_parallel_show, sync_force_parallel_store);
3515
3516 static ssize_t
3517 sync_speed_show(mddev_t *mddev, char *page)
3518 {
3519 unsigned long resync, dt, db;
3520 if (mddev->curr_resync == 0)
3521 return sprintf(page, "none\n");
3522 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3523 dt = (jiffies - mddev->resync_mark) / HZ;
3524 if (!dt) dt++;
3525 db = resync - mddev->resync_mark_cnt;
3526 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3527 }
3528
3529 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3530
3531 static ssize_t
3532 sync_completed_show(mddev_t *mddev, char *page)
3533 {
3534 unsigned long max_sectors, resync;
3535
3536 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3537 return sprintf(page, "none\n");
3538
3539 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3540 max_sectors = mddev->resync_max_sectors;
3541 else
3542 max_sectors = mddev->dev_sectors;
3543
3544 resync = mddev->curr_resync_completed;
3545 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3546 }
3547
3548 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3549
3550 static ssize_t
3551 min_sync_show(mddev_t *mddev, char *page)
3552 {
3553 return sprintf(page, "%llu\n",
3554 (unsigned long long)mddev->resync_min);
3555 }
3556 static ssize_t
3557 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3558 {
3559 unsigned long long min;
3560 if (strict_strtoull(buf, 10, &min))
3561 return -EINVAL;
3562 if (min > mddev->resync_max)
3563 return -EINVAL;
3564 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3565 return -EBUSY;
3566
3567 /* Must be a multiple of chunk_size */
3568 if (mddev->chunk_sectors) {
3569 sector_t temp = min;
3570 if (sector_div(temp, mddev->chunk_sectors))
3571 return -EINVAL;
3572 }
3573 mddev->resync_min = min;
3574
3575 return len;
3576 }
3577
3578 static struct md_sysfs_entry md_min_sync =
3579 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3580
3581 static ssize_t
3582 max_sync_show(mddev_t *mddev, char *page)
3583 {
3584 if (mddev->resync_max == MaxSector)
3585 return sprintf(page, "max\n");
3586 else
3587 return sprintf(page, "%llu\n",
3588 (unsigned long long)mddev->resync_max);
3589 }
3590 static ssize_t
3591 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3592 {
3593 if (strncmp(buf, "max", 3) == 0)
3594 mddev->resync_max = MaxSector;
3595 else {
3596 unsigned long long max;
3597 if (strict_strtoull(buf, 10, &max))
3598 return -EINVAL;
3599 if (max < mddev->resync_min)
3600 return -EINVAL;
3601 if (max < mddev->resync_max &&
3602 mddev->ro == 0 &&
3603 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3604 return -EBUSY;
3605
3606 /* Must be a multiple of chunk_size */
3607 if (mddev->chunk_sectors) {
3608 sector_t temp = max;
3609 if (sector_div(temp, mddev->chunk_sectors))
3610 return -EINVAL;
3611 }
3612 mddev->resync_max = max;
3613 }
3614 wake_up(&mddev->recovery_wait);
3615 return len;
3616 }
3617
3618 static struct md_sysfs_entry md_max_sync =
3619 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3620
3621 static ssize_t
3622 suspend_lo_show(mddev_t *mddev, char *page)
3623 {
3624 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3625 }
3626
3627 static ssize_t
3628 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3629 {
3630 char *e;
3631 unsigned long long new = simple_strtoull(buf, &e, 10);
3632
3633 if (mddev->pers == NULL ||
3634 mddev->pers->quiesce == NULL)
3635 return -EINVAL;
3636 if (buf == e || (*e && *e != '\n'))
3637 return -EINVAL;
3638 if (new >= mddev->suspend_hi ||
3639 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3640 mddev->suspend_lo = new;
3641 mddev->pers->quiesce(mddev, 2);
3642 return len;
3643 } else
3644 return -EINVAL;
3645 }
3646 static struct md_sysfs_entry md_suspend_lo =
3647 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3648
3649
3650 static ssize_t
3651 suspend_hi_show(mddev_t *mddev, char *page)
3652 {
3653 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3654 }
3655
3656 static ssize_t
3657 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3658 {
3659 char *e;
3660 unsigned long long new = simple_strtoull(buf, &e, 10);
3661
3662 if (mddev->pers == NULL ||
3663 mddev->pers->quiesce == NULL)
3664 return -EINVAL;
3665 if (buf == e || (*e && *e != '\n'))
3666 return -EINVAL;
3667 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3668 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3669 mddev->suspend_hi = new;
3670 mddev->pers->quiesce(mddev, 1);
3671 mddev->pers->quiesce(mddev, 0);
3672 return len;
3673 } else
3674 return -EINVAL;
3675 }
3676 static struct md_sysfs_entry md_suspend_hi =
3677 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3678
3679 static ssize_t
3680 reshape_position_show(mddev_t *mddev, char *page)
3681 {
3682 if (mddev->reshape_position != MaxSector)
3683 return sprintf(page, "%llu\n",
3684 (unsigned long long)mddev->reshape_position);
3685 strcpy(page, "none\n");
3686 return 5;
3687 }
3688
3689 static ssize_t
3690 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3691 {
3692 char *e;
3693 unsigned long long new = simple_strtoull(buf, &e, 10);
3694 if (mddev->pers)
3695 return -EBUSY;
3696 if (buf == e || (*e && *e != '\n'))
3697 return -EINVAL;
3698 mddev->reshape_position = new;
3699 mddev->delta_disks = 0;
3700 mddev->new_level = mddev->level;
3701 mddev->new_layout = mddev->layout;
3702 mddev->new_chunk_sectors = mddev->chunk_sectors;
3703 return len;
3704 }
3705
3706 static struct md_sysfs_entry md_reshape_position =
3707 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3708 reshape_position_store);
3709
3710 static ssize_t
3711 array_size_show(mddev_t *mddev, char *page)
3712 {
3713 if (mddev->external_size)
3714 return sprintf(page, "%llu\n",
3715 (unsigned long long)mddev->array_sectors/2);
3716 else
3717 return sprintf(page, "default\n");
3718 }
3719
3720 static ssize_t
3721 array_size_store(mddev_t *mddev, const char *buf, size_t len)
3722 {
3723 sector_t sectors;
3724
3725 if (strncmp(buf, "default", 7) == 0) {
3726 if (mddev->pers)
3727 sectors = mddev->pers->size(mddev, 0, 0);
3728 else
3729 sectors = mddev->array_sectors;
3730
3731 mddev->external_size = 0;
3732 } else {
3733 if (strict_blocks_to_sectors(buf, &sectors) < 0)
3734 return -EINVAL;
3735 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
3736 return -E2BIG;
3737
3738 mddev->external_size = 1;
3739 }
3740
3741 mddev->array_sectors = sectors;
3742 set_capacity(mddev->gendisk, mddev->array_sectors);
3743 if (mddev->pers)
3744 revalidate_disk(mddev->gendisk);
3745
3746 return len;
3747 }
3748
3749 static struct md_sysfs_entry md_array_size =
3750 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
3751 array_size_store);
3752
3753 static struct attribute *md_default_attrs[] = {
3754 &md_level.attr,
3755 &md_layout.attr,
3756 &md_raid_disks.attr,
3757 &md_chunk_size.attr,
3758 &md_size.attr,
3759 &md_resync_start.attr,
3760 &md_metadata.attr,
3761 &md_new_device.attr,
3762 &md_safe_delay.attr,
3763 &md_array_state.attr,
3764 &md_reshape_position.attr,
3765 &md_array_size.attr,
3766 NULL,
3767 };
3768
3769 static struct attribute *md_redundancy_attrs[] = {
3770 &md_scan_mode.attr,
3771 &md_mismatches.attr,
3772 &md_sync_min.attr,
3773 &md_sync_max.attr,
3774 &md_sync_speed.attr,
3775 &md_sync_force_parallel.attr,
3776 &md_sync_completed.attr,
3777 &md_min_sync.attr,
3778 &md_max_sync.attr,
3779 &md_suspend_lo.attr,
3780 &md_suspend_hi.attr,
3781 &md_bitmap.attr,
3782 &md_degraded.attr,
3783 NULL,
3784 };
3785 static struct attribute_group md_redundancy_group = {
3786 .name = NULL,
3787 .attrs = md_redundancy_attrs,
3788 };
3789
3790
3791 static ssize_t
3792 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3793 {
3794 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3795 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3796 ssize_t rv;
3797
3798 if (!entry->show)
3799 return -EIO;
3800 rv = mddev_lock(mddev);
3801 if (!rv) {
3802 rv = entry->show(mddev, page);
3803 mddev_unlock(mddev);
3804 }
3805 return rv;
3806 }
3807
3808 static ssize_t
3809 md_attr_store(struct kobject *kobj, struct attribute *attr,
3810 const char *page, size_t length)
3811 {
3812 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3813 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3814 ssize_t rv;
3815
3816 if (!entry->store)
3817 return -EIO;
3818 if (!capable(CAP_SYS_ADMIN))
3819 return -EACCES;
3820 rv = mddev_lock(mddev);
3821 if (mddev->hold_active == UNTIL_IOCTL)
3822 mddev->hold_active = 0;
3823 if (!rv) {
3824 rv = entry->store(mddev, page, length);
3825 mddev_unlock(mddev);
3826 }
3827 return rv;
3828 }
3829
3830 static void md_free(struct kobject *ko)
3831 {
3832 mddev_t *mddev = container_of(ko, mddev_t, kobj);
3833
3834 if (mddev->sysfs_state)
3835 sysfs_put(mddev->sysfs_state);
3836
3837 if (mddev->gendisk) {
3838 del_gendisk(mddev->gendisk);
3839 put_disk(mddev->gendisk);
3840 }
3841 if (mddev->queue)
3842 blk_cleanup_queue(mddev->queue);
3843
3844 kfree(mddev);
3845 }
3846
3847 static struct sysfs_ops md_sysfs_ops = {
3848 .show = md_attr_show,
3849 .store = md_attr_store,
3850 };
3851 static struct kobj_type md_ktype = {
3852 .release = md_free,
3853 .sysfs_ops = &md_sysfs_ops,
3854 .default_attrs = md_default_attrs,
3855 };
3856
3857 int mdp_major = 0;
3858
3859 static void mddev_delayed_delete(struct work_struct *ws)
3860 {
3861 mddev_t *mddev = container_of(ws, mddev_t, del_work);
3862
3863 if (mddev->private == &md_redundancy_group) {
3864 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3865 if (mddev->sysfs_action)
3866 sysfs_put(mddev->sysfs_action);
3867 mddev->sysfs_action = NULL;
3868 mddev->private = NULL;
3869 }
3870 kobject_del(&mddev->kobj);
3871 kobject_put(&mddev->kobj);
3872 }
3873
3874 static int md_alloc(dev_t dev, char *name)
3875 {
3876 static DEFINE_MUTEX(disks_mutex);
3877 mddev_t *mddev = mddev_find(dev);
3878 struct gendisk *disk;
3879 int partitioned;
3880 int shift;
3881 int unit;
3882 int error;
3883
3884 if (!mddev)
3885 return -ENODEV;
3886
3887 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
3888 shift = partitioned ? MdpMinorShift : 0;
3889 unit = MINOR(mddev->unit) >> shift;
3890
3891 /* wait for any previous instance if this device
3892 * to be completed removed (mddev_delayed_delete).
3893 */
3894 flush_scheduled_work();
3895
3896 mutex_lock(&disks_mutex);
3897 error = -EEXIST;
3898 if (mddev->gendisk)
3899 goto abort;
3900
3901 if (name) {
3902 /* Need to ensure that 'name' is not a duplicate.
3903 */
3904 mddev_t *mddev2;
3905 spin_lock(&all_mddevs_lock);
3906
3907 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
3908 if (mddev2->gendisk &&
3909 strcmp(mddev2->gendisk->disk_name, name) == 0) {
3910 spin_unlock(&all_mddevs_lock);
3911 goto abort;
3912 }
3913 spin_unlock(&all_mddevs_lock);
3914 }
3915
3916 error = -ENOMEM;
3917 mddev->queue = blk_alloc_queue(GFP_KERNEL);
3918 if (!mddev->queue)
3919 goto abort;
3920 mddev->queue->queuedata = mddev;
3921
3922 /* Can be unlocked because the queue is new: no concurrency */
3923 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
3924
3925 blk_queue_make_request(mddev->queue, md_make_request);
3926
3927 disk = alloc_disk(1 << shift);
3928 if (!disk) {
3929 blk_cleanup_queue(mddev->queue);
3930 mddev->queue = NULL;
3931 goto abort;
3932 }
3933 disk->major = MAJOR(mddev->unit);
3934 disk->first_minor = unit << shift;
3935 if (name)
3936 strcpy(disk->disk_name, name);
3937 else if (partitioned)
3938 sprintf(disk->disk_name, "md_d%d", unit);
3939 else
3940 sprintf(disk->disk_name, "md%d", unit);
3941 disk->fops = &md_fops;
3942 disk->private_data = mddev;
3943 disk->queue = mddev->queue;
3944 /* Allow extended partitions. This makes the
3945 * 'mdp' device redundant, but we can't really
3946 * remove it now.
3947 */
3948 disk->flags |= GENHD_FL_EXT_DEVT;
3949 add_disk(disk);
3950 mddev->gendisk = disk;
3951 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
3952 &disk_to_dev(disk)->kobj, "%s", "md");
3953 if (error) {
3954 /* This isn't possible, but as kobject_init_and_add is marked
3955 * __must_check, we must do something with the result
3956 */
3957 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3958 disk->disk_name);
3959 error = 0;
3960 }
3961 abort:
3962 mutex_unlock(&disks_mutex);
3963 if (!error) {
3964 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3965 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state");
3966 }
3967 mddev_put(mddev);
3968 return error;
3969 }
3970
3971 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3972 {
3973 md_alloc(dev, NULL);
3974 return NULL;
3975 }
3976
3977 static int add_named_array(const char *val, struct kernel_param *kp)
3978 {
3979 /* val must be "md_*" where * is not all digits.
3980 * We allocate an array with a large free minor number, and
3981 * set the name to val. val must not already be an active name.
3982 */
3983 int len = strlen(val);
3984 char buf[DISK_NAME_LEN];
3985
3986 while (len && val[len-1] == '\n')
3987 len--;
3988 if (len >= DISK_NAME_LEN)
3989 return -E2BIG;
3990 strlcpy(buf, val, len+1);
3991 if (strncmp(buf, "md_", 3) != 0)
3992 return -EINVAL;
3993 return md_alloc(0, buf);
3994 }
3995
3996 static void md_safemode_timeout(unsigned long data)
3997 {
3998 mddev_t *mddev = (mddev_t *) data;
3999
4000 if (!atomic_read(&mddev->writes_pending)) {
4001 mddev->safemode = 1;
4002 if (mddev->external)
4003 sysfs_notify_dirent(mddev->sysfs_state);
4004 }
4005 md_wakeup_thread(mddev->thread);
4006 }
4007
4008 static int start_dirty_degraded;
4009
4010 static int do_md_run(mddev_t * mddev)
4011 {
4012 int err;
4013 mdk_rdev_t *rdev;
4014 struct gendisk *disk;
4015 struct mdk_personality *pers;
4016
4017 if (list_empty(&mddev->disks))
4018 /* cannot run an array with no devices.. */
4019 return -EINVAL;
4020
4021 if (mddev->pers)
4022 return -EBUSY;
4023
4024 /*
4025 * Analyze all RAID superblock(s)
4026 */
4027 if (!mddev->raid_disks) {
4028 if (!mddev->persistent)
4029 return -EINVAL;
4030 analyze_sbs(mddev);
4031 }
4032
4033 if (mddev->level != LEVEL_NONE)
4034 request_module("md-level-%d", mddev->level);
4035 else if (mddev->clevel[0])
4036 request_module("md-%s", mddev->clevel);
4037
4038 /*
4039 * Drop all container device buffers, from now on
4040 * the only valid external interface is through the md
4041 * device.
4042 */
4043 list_for_each_entry(rdev, &mddev->disks, same_set) {
4044 if (test_bit(Faulty, &rdev->flags))
4045 continue;
4046 sync_blockdev(rdev->bdev);
4047 invalidate_bdev(rdev->bdev);
4048
4049 /* perform some consistency tests on the device.
4050 * We don't want the data to overlap the metadata,
4051 * Internal Bitmap issues have been handled elsewhere.
4052 */
4053 if (rdev->data_offset < rdev->sb_start) {
4054 if (mddev->dev_sectors &&
4055 rdev->data_offset + mddev->dev_sectors
4056 > rdev->sb_start) {
4057 printk("md: %s: data overlaps metadata\n",
4058 mdname(mddev));
4059 return -EINVAL;
4060 }
4061 } else {
4062 if (rdev->sb_start + rdev->sb_size/512
4063 > rdev->data_offset) {
4064 printk("md: %s: metadata overlaps data\n",
4065 mdname(mddev));
4066 return -EINVAL;
4067 }
4068 }
4069 sysfs_notify_dirent(rdev->sysfs_state);
4070 }
4071
4072 md_probe(mddev->unit, NULL, NULL);
4073 disk = mddev->gendisk;
4074 if (!disk)
4075 return -ENOMEM;
4076
4077 spin_lock(&pers_lock);
4078 pers = find_pers(mddev->level, mddev->clevel);
4079 if (!pers || !try_module_get(pers->owner)) {
4080 spin_unlock(&pers_lock);
4081 if (mddev->level != LEVEL_NONE)
4082 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4083 mddev->level);
4084 else
4085 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4086 mddev->clevel);
4087 return -EINVAL;
4088 }
4089 mddev->pers = pers;
4090 spin_unlock(&pers_lock);
4091 if (mddev->level != pers->level) {
4092 mddev->level = pers->level;
4093 mddev->new_level = pers->level;
4094 }
4095 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4096
4097 if (mddev->reshape_position != MaxSector &&
4098 pers->start_reshape == NULL) {
4099 /* This personality cannot handle reshaping... */
4100 mddev->pers = NULL;
4101 module_put(pers->owner);
4102 return -EINVAL;
4103 }
4104
4105 if (pers->sync_request) {
4106 /* Warn if this is a potentially silly
4107 * configuration.
4108 */
4109 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4110 mdk_rdev_t *rdev2;
4111 int warned = 0;
4112
4113 list_for_each_entry(rdev, &mddev->disks, same_set)
4114 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4115 if (rdev < rdev2 &&
4116 rdev->bdev->bd_contains ==
4117 rdev2->bdev->bd_contains) {
4118 printk(KERN_WARNING
4119 "%s: WARNING: %s appears to be"
4120 " on the same physical disk as"
4121 " %s.\n",
4122 mdname(mddev),
4123 bdevname(rdev->bdev,b),
4124 bdevname(rdev2->bdev,b2));
4125 warned = 1;
4126 }
4127 }
4128
4129 if (warned)
4130 printk(KERN_WARNING
4131 "True protection against single-disk"
4132 " failure might be compromised.\n");
4133 }
4134
4135 mddev->recovery = 0;
4136 /* may be over-ridden by personality */
4137 mddev->resync_max_sectors = mddev->dev_sectors;
4138
4139 mddev->barriers_work = 1;
4140 mddev->ok_start_degraded = start_dirty_degraded;
4141
4142 if (start_readonly)
4143 mddev->ro = 2; /* read-only, but switch on first write */
4144
4145 err = mddev->pers->run(mddev);
4146 if (err)
4147 printk(KERN_ERR "md: pers->run() failed ...\n");
4148 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4149 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4150 " but 'external_size' not in effect?\n", __func__);
4151 printk(KERN_ERR
4152 "md: invalid array_size %llu > default size %llu\n",
4153 (unsigned long long)mddev->array_sectors / 2,
4154 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4155 err = -EINVAL;
4156 mddev->pers->stop(mddev);
4157 }
4158 if (err == 0 && mddev->pers->sync_request) {
4159 err = bitmap_create(mddev);
4160 if (err) {
4161 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4162 mdname(mddev), err);
4163 mddev->pers->stop(mddev);
4164 }
4165 }
4166 if (err) {
4167 module_put(mddev->pers->owner);
4168 mddev->pers = NULL;
4169 bitmap_destroy(mddev);
4170 return err;
4171 }
4172 if (mddev->pers->sync_request) {
4173 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4174 printk(KERN_WARNING
4175 "md: cannot register extra attributes for %s\n",
4176 mdname(mddev));
4177 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4178 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4179 mddev->ro = 0;
4180
4181 atomic_set(&mddev->writes_pending,0);
4182 mddev->safemode = 0;
4183 mddev->safemode_timer.function = md_safemode_timeout;
4184 mddev->safemode_timer.data = (unsigned long) mddev;
4185 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4186 mddev->in_sync = 1;
4187
4188 list_for_each_entry(rdev, &mddev->disks, same_set)
4189 if (rdev->raid_disk >= 0) {
4190 char nm[20];
4191 sprintf(nm, "rd%d", rdev->raid_disk);
4192 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4193 printk("md: cannot register %s for %s\n",
4194 nm, mdname(mddev));
4195 }
4196
4197 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4198
4199 if (mddev->flags)
4200 md_update_sb(mddev, 0);
4201
4202 set_capacity(disk, mddev->array_sectors);
4203
4204 /* If there is a partially-recovered drive we need to
4205 * start recovery here. If we leave it to md_check_recovery,
4206 * it will remove the drives and not do the right thing
4207 */
4208 if (mddev->degraded && !mddev->sync_thread) {
4209 int spares = 0;
4210 list_for_each_entry(rdev, &mddev->disks, same_set)
4211 if (rdev->raid_disk >= 0 &&
4212 !test_bit(In_sync, &rdev->flags) &&
4213 !test_bit(Faulty, &rdev->flags))
4214 /* complete an interrupted recovery */
4215 spares++;
4216 if (spares && mddev->pers->sync_request) {
4217 mddev->recovery = 0;
4218 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4219 mddev->sync_thread = md_register_thread(md_do_sync,
4220 mddev,
4221 "%s_resync");
4222 if (!mddev->sync_thread) {
4223 printk(KERN_ERR "%s: could not start resync"
4224 " thread...\n",
4225 mdname(mddev));
4226 /* leave the spares where they are, it shouldn't hurt */
4227 mddev->recovery = 0;
4228 }
4229 }
4230 }
4231 md_wakeup_thread(mddev->thread);
4232 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4233
4234 revalidate_disk(mddev->gendisk);
4235 mddev->changed = 1;
4236 md_new_event(mddev);
4237 sysfs_notify_dirent(mddev->sysfs_state);
4238 if (mddev->sysfs_action)
4239 sysfs_notify_dirent(mddev->sysfs_action);
4240 sysfs_notify(&mddev->kobj, NULL, "degraded");
4241 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4242 return 0;
4243 }
4244
4245 static int restart_array(mddev_t *mddev)
4246 {
4247 struct gendisk *disk = mddev->gendisk;
4248
4249 /* Complain if it has no devices */
4250 if (list_empty(&mddev->disks))
4251 return -ENXIO;
4252 if (!mddev->pers)
4253 return -EINVAL;
4254 if (!mddev->ro)
4255 return -EBUSY;
4256 mddev->safemode = 0;
4257 mddev->ro = 0;
4258 set_disk_ro(disk, 0);
4259 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4260 mdname(mddev));
4261 /* Kick recovery or resync if necessary */
4262 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4263 md_wakeup_thread(mddev->thread);
4264 md_wakeup_thread(mddev->sync_thread);
4265 sysfs_notify_dirent(mddev->sysfs_state);
4266 return 0;
4267 }
4268
4269 /* similar to deny_write_access, but accounts for our holding a reference
4270 * to the file ourselves */
4271 static int deny_bitmap_write_access(struct file * file)
4272 {
4273 struct inode *inode = file->f_mapping->host;
4274
4275 spin_lock(&inode->i_lock);
4276 if (atomic_read(&inode->i_writecount) > 1) {
4277 spin_unlock(&inode->i_lock);
4278 return -ETXTBSY;
4279 }
4280 atomic_set(&inode->i_writecount, -1);
4281 spin_unlock(&inode->i_lock);
4282
4283 return 0;
4284 }
4285
4286 static void restore_bitmap_write_access(struct file *file)
4287 {
4288 struct inode *inode = file->f_mapping->host;
4289
4290 spin_lock(&inode->i_lock);
4291 atomic_set(&inode->i_writecount, 1);
4292 spin_unlock(&inode->i_lock);
4293 }
4294
4295 /* mode:
4296 * 0 - completely stop and dis-assemble array
4297 * 1 - switch to readonly
4298 * 2 - stop but do not disassemble array
4299 */
4300 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4301 {
4302 int err = 0;
4303 struct gendisk *disk = mddev->gendisk;
4304 mdk_rdev_t *rdev;
4305
4306 mutex_lock(&mddev->open_mutex);
4307 if (atomic_read(&mddev->openers) > is_open) {
4308 printk("md: %s still in use.\n",mdname(mddev));
4309 err = -EBUSY;
4310 } else if (mddev->pers) {
4311
4312 if (mddev->sync_thread) {
4313 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4314 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4315 md_unregister_thread(mddev->sync_thread);
4316 mddev->sync_thread = NULL;
4317 }
4318
4319 del_timer_sync(&mddev->safemode_timer);
4320
4321 switch(mode) {
4322 case 1: /* readonly */
4323 err = -ENXIO;
4324 if (mddev->ro==1)
4325 goto out;
4326 mddev->ro = 1;
4327 break;
4328 case 0: /* disassemble */
4329 case 2: /* stop */
4330 bitmap_flush(mddev);
4331 md_super_wait(mddev);
4332 if (mddev->ro)
4333 set_disk_ro(disk, 0);
4334
4335 mddev->pers->stop(mddev);
4336 mddev->queue->merge_bvec_fn = NULL;
4337 mddev->queue->unplug_fn = NULL;
4338 mddev->queue->backing_dev_info.congested_fn = NULL;
4339 module_put(mddev->pers->owner);
4340 if (mddev->pers->sync_request)
4341 mddev->private = &md_redundancy_group;
4342 mddev->pers = NULL;
4343 /* tell userspace to handle 'inactive' */
4344 sysfs_notify_dirent(mddev->sysfs_state);
4345
4346 list_for_each_entry(rdev, &mddev->disks, same_set)
4347 if (rdev->raid_disk >= 0) {
4348 char nm[20];
4349 sprintf(nm, "rd%d", rdev->raid_disk);
4350 sysfs_remove_link(&mddev->kobj, nm);
4351 }
4352
4353 set_capacity(disk, 0);
4354 mddev->changed = 1;
4355
4356 if (mddev->ro)
4357 mddev->ro = 0;
4358 }
4359 if (!mddev->in_sync || mddev->flags) {
4360 /* mark array as shutdown cleanly */
4361 mddev->in_sync = 1;
4362 md_update_sb(mddev, 1);
4363 }
4364 if (mode == 1)
4365 set_disk_ro(disk, 1);
4366 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4367 err = 0;
4368 }
4369 out:
4370 mutex_unlock(&mddev->open_mutex);
4371 if (err)
4372 return err;
4373 /*
4374 * Free resources if final stop
4375 */
4376 if (mode == 0) {
4377
4378 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4379
4380 bitmap_destroy(mddev);
4381 if (mddev->bitmap_file) {
4382 restore_bitmap_write_access(mddev->bitmap_file);
4383 fput(mddev->bitmap_file);
4384 mddev->bitmap_file = NULL;
4385 }
4386 mddev->bitmap_offset = 0;
4387
4388 /* make sure all md_delayed_delete calls have finished */
4389 flush_scheduled_work();
4390
4391 export_array(mddev);
4392
4393 mddev->array_sectors = 0;
4394 mddev->external_size = 0;
4395 mddev->dev_sectors = 0;
4396 mddev->raid_disks = 0;
4397 mddev->recovery_cp = 0;
4398 mddev->resync_min = 0;
4399 mddev->resync_max = MaxSector;
4400 mddev->reshape_position = MaxSector;
4401 mddev->external = 0;
4402 mddev->persistent = 0;
4403 mddev->level = LEVEL_NONE;
4404 mddev->clevel[0] = 0;
4405 mddev->flags = 0;
4406 mddev->ro = 0;
4407 mddev->metadata_type[0] = 0;
4408 mddev->chunk_sectors = 0;
4409 mddev->ctime = mddev->utime = 0;
4410 mddev->layout = 0;
4411 mddev->max_disks = 0;
4412 mddev->events = 0;
4413 mddev->delta_disks = 0;
4414 mddev->new_level = LEVEL_NONE;
4415 mddev->new_layout = 0;
4416 mddev->new_chunk_sectors = 0;
4417 mddev->curr_resync = 0;
4418 mddev->resync_mismatches = 0;
4419 mddev->suspend_lo = mddev->suspend_hi = 0;
4420 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4421 mddev->recovery = 0;
4422 mddev->in_sync = 0;
4423 mddev->changed = 0;
4424 mddev->degraded = 0;
4425 mddev->barriers_work = 0;
4426 mddev->safemode = 0;
4427 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4428 if (mddev->hold_active == UNTIL_STOP)
4429 mddev->hold_active = 0;
4430
4431 } else if (mddev->pers)
4432 printk(KERN_INFO "md: %s switched to read-only mode.\n",
4433 mdname(mddev));
4434 err = 0;
4435 blk_integrity_unregister(disk);
4436 md_new_event(mddev);
4437 sysfs_notify_dirent(mddev->sysfs_state);
4438 return err;
4439 }
4440
4441 #ifndef MODULE
4442 static void autorun_array(mddev_t *mddev)
4443 {
4444 mdk_rdev_t *rdev;
4445 int err;
4446
4447 if (list_empty(&mddev->disks))
4448 return;
4449
4450 printk(KERN_INFO "md: running: ");
4451
4452 list_for_each_entry(rdev, &mddev->disks, same_set) {
4453 char b[BDEVNAME_SIZE];
4454 printk("<%s>", bdevname(rdev->bdev,b));
4455 }
4456 printk("\n");
4457
4458 err = do_md_run(mddev);
4459 if (err) {
4460 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4461 do_md_stop(mddev, 0, 0);
4462 }
4463 }
4464
4465 /*
4466 * lets try to run arrays based on all disks that have arrived
4467 * until now. (those are in pending_raid_disks)
4468 *
4469 * the method: pick the first pending disk, collect all disks with
4470 * the same UUID, remove all from the pending list and put them into
4471 * the 'same_array' list. Then order this list based on superblock
4472 * update time (freshest comes first), kick out 'old' disks and
4473 * compare superblocks. If everything's fine then run it.
4474 *
4475 * If "unit" is allocated, then bump its reference count
4476 */
4477 static void autorun_devices(int part)
4478 {
4479 mdk_rdev_t *rdev0, *rdev, *tmp;
4480 mddev_t *mddev;
4481 char b[BDEVNAME_SIZE];
4482
4483 printk(KERN_INFO "md: autorun ...\n");
4484 while (!list_empty(&pending_raid_disks)) {
4485 int unit;
4486 dev_t dev;
4487 LIST_HEAD(candidates);
4488 rdev0 = list_entry(pending_raid_disks.next,
4489 mdk_rdev_t, same_set);
4490
4491 printk(KERN_INFO "md: considering %s ...\n",
4492 bdevname(rdev0->bdev,b));
4493 INIT_LIST_HEAD(&candidates);
4494 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4495 if (super_90_load(rdev, rdev0, 0) >= 0) {
4496 printk(KERN_INFO "md: adding %s ...\n",
4497 bdevname(rdev->bdev,b));
4498 list_move(&rdev->same_set, &candidates);
4499 }
4500 /*
4501 * now we have a set of devices, with all of them having
4502 * mostly sane superblocks. It's time to allocate the
4503 * mddev.
4504 */
4505 if (part) {
4506 dev = MKDEV(mdp_major,
4507 rdev0->preferred_minor << MdpMinorShift);
4508 unit = MINOR(dev) >> MdpMinorShift;
4509 } else {
4510 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4511 unit = MINOR(dev);
4512 }
4513 if (rdev0->preferred_minor != unit) {
4514 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4515 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4516 break;
4517 }
4518
4519 md_probe(dev, NULL, NULL);
4520 mddev = mddev_find(dev);
4521 if (!mddev || !mddev->gendisk) {
4522 if (mddev)
4523 mddev_put(mddev);
4524 printk(KERN_ERR
4525 "md: cannot allocate memory for md drive.\n");
4526 break;
4527 }
4528 if (mddev_lock(mddev))
4529 printk(KERN_WARNING "md: %s locked, cannot run\n",
4530 mdname(mddev));
4531 else if (mddev->raid_disks || mddev->major_version
4532 || !list_empty(&mddev->disks)) {
4533 printk(KERN_WARNING
4534 "md: %s already running, cannot run %s\n",
4535 mdname(mddev), bdevname(rdev0->bdev,b));
4536 mddev_unlock(mddev);
4537 } else {
4538 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4539 mddev->persistent = 1;
4540 rdev_for_each_list(rdev, tmp, &candidates) {
4541 list_del_init(&rdev->same_set);
4542 if (bind_rdev_to_array(rdev, mddev))
4543 export_rdev(rdev);
4544 }
4545 autorun_array(mddev);
4546 mddev_unlock(mddev);
4547 }
4548 /* on success, candidates will be empty, on error
4549 * it won't...
4550 */
4551 rdev_for_each_list(rdev, tmp, &candidates) {
4552 list_del_init(&rdev->same_set);
4553 export_rdev(rdev);
4554 }
4555 mddev_put(mddev);
4556 }
4557 printk(KERN_INFO "md: ... autorun DONE.\n");
4558 }
4559 #endif /* !MODULE */
4560
4561 static int get_version(void __user * arg)
4562 {
4563 mdu_version_t ver;
4564
4565 ver.major = MD_MAJOR_VERSION;
4566 ver.minor = MD_MINOR_VERSION;
4567 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4568
4569 if (copy_to_user(arg, &ver, sizeof(ver)))
4570 return -EFAULT;
4571
4572 return 0;
4573 }
4574
4575 static int get_array_info(mddev_t * mddev, void __user * arg)
4576 {
4577 mdu_array_info_t info;
4578 int nr,working,active,failed,spare;
4579 mdk_rdev_t *rdev;
4580
4581 nr=working=active=failed=spare=0;
4582 list_for_each_entry(rdev, &mddev->disks, same_set) {
4583 nr++;
4584 if (test_bit(Faulty, &rdev->flags))
4585 failed++;
4586 else {
4587 working++;
4588 if (test_bit(In_sync, &rdev->flags))
4589 active++;
4590 else
4591 spare++;
4592 }
4593 }
4594
4595 info.major_version = mddev->major_version;
4596 info.minor_version = mddev->minor_version;
4597 info.patch_version = MD_PATCHLEVEL_VERSION;
4598 info.ctime = mddev->ctime;
4599 info.level = mddev->level;
4600 info.size = mddev->dev_sectors / 2;
4601 if (info.size != mddev->dev_sectors / 2) /* overflow */
4602 info.size = -1;
4603 info.nr_disks = nr;
4604 info.raid_disks = mddev->raid_disks;
4605 info.md_minor = mddev->md_minor;
4606 info.not_persistent= !mddev->persistent;
4607
4608 info.utime = mddev->utime;
4609 info.state = 0;
4610 if (mddev->in_sync)
4611 info.state = (1<<MD_SB_CLEAN);
4612 if (mddev->bitmap && mddev->bitmap_offset)
4613 info.state = (1<<MD_SB_BITMAP_PRESENT);
4614 info.active_disks = active;
4615 info.working_disks = working;
4616 info.failed_disks = failed;
4617 info.spare_disks = spare;
4618
4619 info.layout = mddev->layout;
4620 info.chunk_size = mddev->chunk_sectors << 9;
4621
4622 if (copy_to_user(arg, &info, sizeof(info)))
4623 return -EFAULT;
4624
4625 return 0;
4626 }
4627
4628 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4629 {
4630 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4631 char *ptr, *buf = NULL;
4632 int err = -ENOMEM;
4633
4634 if (md_allow_write(mddev))
4635 file = kmalloc(sizeof(*file), GFP_NOIO);
4636 else
4637 file = kmalloc(sizeof(*file), GFP_KERNEL);
4638
4639 if (!file)
4640 goto out;
4641
4642 /* bitmap disabled, zero the first byte and copy out */
4643 if (!mddev->bitmap || !mddev->bitmap->file) {
4644 file->pathname[0] = '\0';
4645 goto copy_out;
4646 }
4647
4648 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4649 if (!buf)
4650 goto out;
4651
4652 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4653 if (IS_ERR(ptr))
4654 goto out;
4655
4656 strcpy(file->pathname, ptr);
4657
4658 copy_out:
4659 err = 0;
4660 if (copy_to_user(arg, file, sizeof(*file)))
4661 err = -EFAULT;
4662 out:
4663 kfree(buf);
4664 kfree(file);
4665 return err;
4666 }
4667
4668 static int get_disk_info(mddev_t * mddev, void __user * arg)
4669 {
4670 mdu_disk_info_t info;
4671 mdk_rdev_t *rdev;
4672
4673 if (copy_from_user(&info, arg, sizeof(info)))
4674 return -EFAULT;
4675
4676 rdev = find_rdev_nr(mddev, info.number);
4677 if (rdev) {
4678 info.major = MAJOR(rdev->bdev->bd_dev);
4679 info.minor = MINOR(rdev->bdev->bd_dev);
4680 info.raid_disk = rdev->raid_disk;
4681 info.state = 0;
4682 if (test_bit(Faulty, &rdev->flags))
4683 info.state |= (1<<MD_DISK_FAULTY);
4684 else if (test_bit(In_sync, &rdev->flags)) {
4685 info.state |= (1<<MD_DISK_ACTIVE);
4686 info.state |= (1<<MD_DISK_SYNC);
4687 }
4688 if (test_bit(WriteMostly, &rdev->flags))
4689 info.state |= (1<<MD_DISK_WRITEMOSTLY);
4690 } else {
4691 info.major = info.minor = 0;
4692 info.raid_disk = -1;
4693 info.state = (1<<MD_DISK_REMOVED);
4694 }
4695
4696 if (copy_to_user(arg, &info, sizeof(info)))
4697 return -EFAULT;
4698
4699 return 0;
4700 }
4701
4702 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4703 {
4704 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4705 mdk_rdev_t *rdev;
4706 dev_t dev = MKDEV(info->major,info->minor);
4707
4708 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4709 return -EOVERFLOW;
4710
4711 if (!mddev->raid_disks) {
4712 int err;
4713 /* expecting a device which has a superblock */
4714 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4715 if (IS_ERR(rdev)) {
4716 printk(KERN_WARNING
4717 "md: md_import_device returned %ld\n",
4718 PTR_ERR(rdev));
4719 return PTR_ERR(rdev);
4720 }
4721 if (!list_empty(&mddev->disks)) {
4722 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4723 mdk_rdev_t, same_set);
4724 int err = super_types[mddev->major_version]
4725 .load_super(rdev, rdev0, mddev->minor_version);
4726 if (err < 0) {
4727 printk(KERN_WARNING
4728 "md: %s has different UUID to %s\n",
4729 bdevname(rdev->bdev,b),
4730 bdevname(rdev0->bdev,b2));
4731 export_rdev(rdev);
4732 return -EINVAL;
4733 }
4734 }
4735 err = bind_rdev_to_array(rdev, mddev);
4736 if (err)
4737 export_rdev(rdev);
4738 return err;
4739 }
4740
4741 /*
4742 * add_new_disk can be used once the array is assembled
4743 * to add "hot spares". They must already have a superblock
4744 * written
4745 */
4746 if (mddev->pers) {
4747 int err;
4748 if (!mddev->pers->hot_add_disk) {
4749 printk(KERN_WARNING
4750 "%s: personality does not support diskops!\n",
4751 mdname(mddev));
4752 return -EINVAL;
4753 }
4754 if (mddev->persistent)
4755 rdev = md_import_device(dev, mddev->major_version,
4756 mddev->minor_version);
4757 else
4758 rdev = md_import_device(dev, -1, -1);
4759 if (IS_ERR(rdev)) {
4760 printk(KERN_WARNING
4761 "md: md_import_device returned %ld\n",
4762 PTR_ERR(rdev));
4763 return PTR_ERR(rdev);
4764 }
4765 /* set save_raid_disk if appropriate */
4766 if (!mddev->persistent) {
4767 if (info->state & (1<<MD_DISK_SYNC) &&
4768 info->raid_disk < mddev->raid_disks)
4769 rdev->raid_disk = info->raid_disk;
4770 else
4771 rdev->raid_disk = -1;
4772 } else
4773 super_types[mddev->major_version].
4774 validate_super(mddev, rdev);
4775 rdev->saved_raid_disk = rdev->raid_disk;
4776
4777 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4778 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4779 set_bit(WriteMostly, &rdev->flags);
4780 else
4781 clear_bit(WriteMostly, &rdev->flags);
4782
4783 rdev->raid_disk = -1;
4784 err = bind_rdev_to_array(rdev, mddev);
4785 if (!err && !mddev->pers->hot_remove_disk) {
4786 /* If there is hot_add_disk but no hot_remove_disk
4787 * then added disks for geometry changes,
4788 * and should be added immediately.
4789 */
4790 super_types[mddev->major_version].
4791 validate_super(mddev, rdev);
4792 err = mddev->pers->hot_add_disk(mddev, rdev);
4793 if (err)
4794 unbind_rdev_from_array(rdev);
4795 }
4796 if (err)
4797 export_rdev(rdev);
4798 else
4799 sysfs_notify_dirent(rdev->sysfs_state);
4800
4801 md_update_sb(mddev, 1);
4802 if (mddev->degraded)
4803 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4804 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4805 md_wakeup_thread(mddev->thread);
4806 return err;
4807 }
4808
4809 /* otherwise, add_new_disk is only allowed
4810 * for major_version==0 superblocks
4811 */
4812 if (mddev->major_version != 0) {
4813 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4814 mdname(mddev));
4815 return -EINVAL;
4816 }
4817
4818 if (!(info->state & (1<<MD_DISK_FAULTY))) {
4819 int err;
4820 rdev = md_import_device(dev, -1, 0);
4821 if (IS_ERR(rdev)) {
4822 printk(KERN_WARNING
4823 "md: error, md_import_device() returned %ld\n",
4824 PTR_ERR(rdev));
4825 return PTR_ERR(rdev);
4826 }
4827 rdev->desc_nr = info->number;
4828 if (info->raid_disk < mddev->raid_disks)
4829 rdev->raid_disk = info->raid_disk;
4830 else
4831 rdev->raid_disk = -1;
4832
4833 if (rdev->raid_disk < mddev->raid_disks)
4834 if (info->state & (1<<MD_DISK_SYNC))
4835 set_bit(In_sync, &rdev->flags);
4836
4837 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4838 set_bit(WriteMostly, &rdev->flags);
4839
4840 if (!mddev->persistent) {
4841 printk(KERN_INFO "md: nonpersistent superblock ...\n");
4842 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4843 } else
4844 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4845 rdev->sectors = rdev->sb_start;
4846
4847 err = bind_rdev_to_array(rdev, mddev);
4848 if (err) {
4849 export_rdev(rdev);
4850 return err;
4851 }
4852 }
4853
4854 return 0;
4855 }
4856
4857 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4858 {
4859 char b[BDEVNAME_SIZE];
4860 mdk_rdev_t *rdev;
4861
4862 rdev = find_rdev(mddev, dev);
4863 if (!rdev)
4864 return -ENXIO;
4865
4866 if (rdev->raid_disk >= 0)
4867 goto busy;
4868
4869 kick_rdev_from_array(rdev);
4870 md_update_sb(mddev, 1);
4871 md_new_event(mddev);
4872
4873 return 0;
4874 busy:
4875 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4876 bdevname(rdev->bdev,b), mdname(mddev));
4877 return -EBUSY;
4878 }
4879
4880 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4881 {
4882 char b[BDEVNAME_SIZE];
4883 int err;
4884 mdk_rdev_t *rdev;
4885
4886 if (!mddev->pers)
4887 return -ENODEV;
4888
4889 if (mddev->major_version != 0) {
4890 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4891 " version-0 superblocks.\n",
4892 mdname(mddev));
4893 return -EINVAL;
4894 }
4895 if (!mddev->pers->hot_add_disk) {
4896 printk(KERN_WARNING
4897 "%s: personality does not support diskops!\n",
4898 mdname(mddev));
4899 return -EINVAL;
4900 }
4901
4902 rdev = md_import_device(dev, -1, 0);
4903 if (IS_ERR(rdev)) {
4904 printk(KERN_WARNING
4905 "md: error, md_import_device() returned %ld\n",
4906 PTR_ERR(rdev));
4907 return -EINVAL;
4908 }
4909
4910 if (mddev->persistent)
4911 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4912 else
4913 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4914
4915 rdev->sectors = rdev->sb_start;
4916
4917 if (test_bit(Faulty, &rdev->flags)) {
4918 printk(KERN_WARNING
4919 "md: can not hot-add faulty %s disk to %s!\n",
4920 bdevname(rdev->bdev,b), mdname(mddev));
4921 err = -EINVAL;
4922 goto abort_export;
4923 }
4924 clear_bit(In_sync, &rdev->flags);
4925 rdev->desc_nr = -1;
4926 rdev->saved_raid_disk = -1;
4927 err = bind_rdev_to_array(rdev, mddev);
4928 if (err)
4929 goto abort_export;
4930
4931 /*
4932 * The rest should better be atomic, we can have disk failures
4933 * noticed in interrupt contexts ...
4934 */
4935
4936 rdev->raid_disk = -1;
4937
4938 md_update_sb(mddev, 1);
4939
4940 /*
4941 * Kick recovery, maybe this spare has to be added to the
4942 * array immediately.
4943 */
4944 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4945 md_wakeup_thread(mddev->thread);
4946 md_new_event(mddev);
4947 return 0;
4948
4949 abort_export:
4950 export_rdev(rdev);
4951 return err;
4952 }
4953
4954 static int set_bitmap_file(mddev_t *mddev, int fd)
4955 {
4956 int err;
4957
4958 if (mddev->pers) {
4959 if (!mddev->pers->quiesce)
4960 return -EBUSY;
4961 if (mddev->recovery || mddev->sync_thread)
4962 return -EBUSY;
4963 /* we should be able to change the bitmap.. */
4964 }
4965
4966
4967 if (fd >= 0) {
4968 if (mddev->bitmap)
4969 return -EEXIST; /* cannot add when bitmap is present */
4970 mddev->bitmap_file = fget(fd);
4971
4972 if (mddev->bitmap_file == NULL) {
4973 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4974 mdname(mddev));
4975 return -EBADF;
4976 }
4977
4978 err = deny_bitmap_write_access(mddev->bitmap_file);
4979 if (err) {
4980 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4981 mdname(mddev));
4982 fput(mddev->bitmap_file);
4983 mddev->bitmap_file = NULL;
4984 return err;
4985 }
4986 mddev->bitmap_offset = 0; /* file overrides offset */
4987 } else if (mddev->bitmap == NULL)
4988 return -ENOENT; /* cannot remove what isn't there */
4989 err = 0;
4990 if (mddev->pers) {
4991 mddev->pers->quiesce(mddev, 1);
4992 if (fd >= 0)
4993 err = bitmap_create(mddev);
4994 if (fd < 0 || err) {
4995 bitmap_destroy(mddev);
4996 fd = -1; /* make sure to put the file */
4997 }
4998 mddev->pers->quiesce(mddev, 0);
4999 }
5000 if (fd < 0) {
5001 if (mddev->bitmap_file) {
5002 restore_bitmap_write_access(mddev->bitmap_file);
5003 fput(mddev->bitmap_file);
5004 }
5005 mddev->bitmap_file = NULL;
5006 }
5007
5008 return err;
5009 }
5010
5011 /*
5012 * set_array_info is used two different ways
5013 * The original usage is when creating a new array.
5014 * In this usage, raid_disks is > 0 and it together with
5015 * level, size, not_persistent,layout,chunksize determine the
5016 * shape of the array.
5017 * This will always create an array with a type-0.90.0 superblock.
5018 * The newer usage is when assembling an array.
5019 * In this case raid_disks will be 0, and the major_version field is
5020 * use to determine which style super-blocks are to be found on the devices.
5021 * The minor and patch _version numbers are also kept incase the
5022 * super_block handler wishes to interpret them.
5023 */
5024 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5025 {
5026
5027 if (info->raid_disks == 0) {
5028 /* just setting version number for superblock loading */
5029 if (info->major_version < 0 ||
5030 info->major_version >= ARRAY_SIZE(super_types) ||
5031 super_types[info->major_version].name == NULL) {
5032 /* maybe try to auto-load a module? */
5033 printk(KERN_INFO
5034 "md: superblock version %d not known\n",
5035 info->major_version);
5036 return -EINVAL;
5037 }
5038 mddev->major_version = info->major_version;
5039 mddev->minor_version = info->minor_version;
5040 mddev->patch_version = info->patch_version;
5041 mddev->persistent = !info->not_persistent;
5042 return 0;
5043 }
5044 mddev->major_version = MD_MAJOR_VERSION;
5045 mddev->minor_version = MD_MINOR_VERSION;
5046 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5047 mddev->ctime = get_seconds();
5048
5049 mddev->level = info->level;
5050 mddev->clevel[0] = 0;
5051 mddev->dev_sectors = 2 * (sector_t)info->size;
5052 mddev->raid_disks = info->raid_disks;
5053 /* don't set md_minor, it is determined by which /dev/md* was
5054 * openned
5055 */
5056 if (info->state & (1<<MD_SB_CLEAN))
5057 mddev->recovery_cp = MaxSector;
5058 else
5059 mddev->recovery_cp = 0;
5060 mddev->persistent = ! info->not_persistent;
5061 mddev->external = 0;
5062
5063 mddev->layout = info->layout;
5064 mddev->chunk_sectors = info->chunk_size >> 9;
5065
5066 mddev->max_disks = MD_SB_DISKS;
5067
5068 if (mddev->persistent)
5069 mddev->flags = 0;
5070 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5071
5072 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
5073 mddev->bitmap_offset = 0;
5074
5075 mddev->reshape_position = MaxSector;
5076
5077 /*
5078 * Generate a 128 bit UUID
5079 */
5080 get_random_bytes(mddev->uuid, 16);
5081
5082 mddev->new_level = mddev->level;
5083 mddev->new_chunk_sectors = mddev->chunk_sectors;
5084 mddev->new_layout = mddev->layout;
5085 mddev->delta_disks = 0;
5086
5087 return 0;
5088 }
5089
5090 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5091 {
5092 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5093
5094 if (mddev->external_size)
5095 return;
5096
5097 mddev->array_sectors = array_sectors;
5098 }
5099 EXPORT_SYMBOL(md_set_array_sectors);
5100
5101 static int update_size(mddev_t *mddev, sector_t num_sectors)
5102 {
5103 mdk_rdev_t *rdev;
5104 int rv;
5105 int fit = (num_sectors == 0);
5106
5107 if (mddev->pers->resize == NULL)
5108 return -EINVAL;
5109 /* The "num_sectors" is the number of sectors of each device that
5110 * is used. This can only make sense for arrays with redundancy.
5111 * linear and raid0 always use whatever space is available. We can only
5112 * consider changing this number if no resync or reconstruction is
5113 * happening, and if the new size is acceptable. It must fit before the
5114 * sb_start or, if that is <data_offset, it must fit before the size
5115 * of each device. If num_sectors is zero, we find the largest size
5116 * that fits.
5117
5118 */
5119 if (mddev->sync_thread)
5120 return -EBUSY;
5121 if (mddev->bitmap)
5122 /* Sorry, cannot grow a bitmap yet, just remove it,
5123 * grow, and re-add.
5124 */
5125 return -EBUSY;
5126 list_for_each_entry(rdev, &mddev->disks, same_set) {
5127 sector_t avail = rdev->sectors;
5128
5129 if (fit && (num_sectors == 0 || num_sectors > avail))
5130 num_sectors = avail;
5131 if (avail < num_sectors)
5132 return -ENOSPC;
5133 }
5134 rv = mddev->pers->resize(mddev, num_sectors);
5135 if (!rv)
5136 revalidate_disk(mddev->gendisk);
5137 return rv;
5138 }
5139
5140 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5141 {
5142 int rv;
5143 /* change the number of raid disks */
5144 if (mddev->pers->check_reshape == NULL)
5145 return -EINVAL;
5146 if (raid_disks <= 0 ||
5147 raid_disks >= mddev->max_disks)
5148 return -EINVAL;
5149 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5150 return -EBUSY;
5151 mddev->delta_disks = raid_disks - mddev->raid_disks;
5152
5153 rv = mddev->pers->check_reshape(mddev);
5154 return rv;
5155 }
5156
5157
5158 /*
5159 * update_array_info is used to change the configuration of an
5160 * on-line array.
5161 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5162 * fields in the info are checked against the array.
5163 * Any differences that cannot be handled will cause an error.
5164 * Normally, only one change can be managed at a time.
5165 */
5166 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5167 {
5168 int rv = 0;
5169 int cnt = 0;
5170 int state = 0;
5171
5172 /* calculate expected state,ignoring low bits */
5173 if (mddev->bitmap && mddev->bitmap_offset)
5174 state |= (1 << MD_SB_BITMAP_PRESENT);
5175
5176 if (mddev->major_version != info->major_version ||
5177 mddev->minor_version != info->minor_version ||
5178 /* mddev->patch_version != info->patch_version || */
5179 mddev->ctime != info->ctime ||
5180 mddev->level != info->level ||
5181 /* mddev->layout != info->layout || */
5182 !mddev->persistent != info->not_persistent||
5183 mddev->chunk_sectors != info->chunk_size >> 9 ||
5184 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5185 ((state^info->state) & 0xfffffe00)
5186 )
5187 return -EINVAL;
5188 /* Check there is only one change */
5189 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5190 cnt++;
5191 if (mddev->raid_disks != info->raid_disks)
5192 cnt++;
5193 if (mddev->layout != info->layout)
5194 cnt++;
5195 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5196 cnt++;
5197 if (cnt == 0)
5198 return 0;
5199 if (cnt > 1)
5200 return -EINVAL;
5201
5202 if (mddev->layout != info->layout) {
5203 /* Change layout
5204 * we don't need to do anything at the md level, the
5205 * personality will take care of it all.
5206 */
5207 if (mddev->pers->check_reshape == NULL)
5208 return -EINVAL;
5209 else {
5210 mddev->new_layout = info->layout;
5211 rv = mddev->pers->check_reshape(mddev);
5212 if (rv)
5213 mddev->new_layout = mddev->layout;
5214 return rv;
5215 }
5216 }
5217 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5218 rv = update_size(mddev, (sector_t)info->size * 2);
5219
5220 if (mddev->raid_disks != info->raid_disks)
5221 rv = update_raid_disks(mddev, info->raid_disks);
5222
5223 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5224 if (mddev->pers->quiesce == NULL)
5225 return -EINVAL;
5226 if (mddev->recovery || mddev->sync_thread)
5227 return -EBUSY;
5228 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5229 /* add the bitmap */
5230 if (mddev->bitmap)
5231 return -EEXIST;
5232 if (mddev->default_bitmap_offset == 0)
5233 return -EINVAL;
5234 mddev->bitmap_offset = mddev->default_bitmap_offset;
5235 mddev->pers->quiesce(mddev, 1);
5236 rv = bitmap_create(mddev);
5237 if (rv)
5238 bitmap_destroy(mddev);
5239 mddev->pers->quiesce(mddev, 0);
5240 } else {
5241 /* remove the bitmap */
5242 if (!mddev->bitmap)
5243 return -ENOENT;
5244 if (mddev->bitmap->file)
5245 return -EINVAL;
5246 mddev->pers->quiesce(mddev, 1);
5247 bitmap_destroy(mddev);
5248 mddev->pers->quiesce(mddev, 0);
5249 mddev->bitmap_offset = 0;
5250 }
5251 }
5252 md_update_sb(mddev, 1);
5253 return rv;
5254 }
5255
5256 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5257 {
5258 mdk_rdev_t *rdev;
5259
5260 if (mddev->pers == NULL)
5261 return -ENODEV;
5262
5263 rdev = find_rdev(mddev, dev);
5264 if (!rdev)
5265 return -ENODEV;
5266
5267 md_error(mddev, rdev);
5268 return 0;
5269 }
5270
5271 /*
5272 * We have a problem here : there is no easy way to give a CHS
5273 * virtual geometry. We currently pretend that we have a 2 heads
5274 * 4 sectors (with a BIG number of cylinders...). This drives
5275 * dosfs just mad... ;-)
5276 */
5277 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5278 {
5279 mddev_t *mddev = bdev->bd_disk->private_data;
5280
5281 geo->heads = 2;
5282 geo->sectors = 4;
5283 geo->cylinders = get_capacity(mddev->gendisk) / 8;
5284 return 0;
5285 }
5286
5287 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5288 unsigned int cmd, unsigned long arg)
5289 {
5290 int err = 0;
5291 void __user *argp = (void __user *)arg;
5292 mddev_t *mddev = NULL;
5293
5294 if (!capable(CAP_SYS_ADMIN))
5295 return -EACCES;
5296
5297 /*
5298 * Commands dealing with the RAID driver but not any
5299 * particular array:
5300 */
5301 switch (cmd)
5302 {
5303 case RAID_VERSION:
5304 err = get_version(argp);
5305 goto done;
5306
5307 case PRINT_RAID_DEBUG:
5308 err = 0;
5309 md_print_devices();
5310 goto done;
5311
5312 #ifndef MODULE
5313 case RAID_AUTORUN:
5314 err = 0;
5315 autostart_arrays(arg);
5316 goto done;
5317 #endif
5318 default:;
5319 }
5320
5321 /*
5322 * Commands creating/starting a new array:
5323 */
5324
5325 mddev = bdev->bd_disk->private_data;
5326
5327 if (!mddev) {
5328 BUG();
5329 goto abort;
5330 }
5331
5332 err = mddev_lock(mddev);
5333 if (err) {
5334 printk(KERN_INFO
5335 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5336 err, cmd);
5337 goto abort;
5338 }
5339
5340 switch (cmd)
5341 {
5342 case SET_ARRAY_INFO:
5343 {
5344 mdu_array_info_t info;
5345 if (!arg)
5346 memset(&info, 0, sizeof(info));
5347 else if (copy_from_user(&info, argp, sizeof(info))) {
5348 err = -EFAULT;
5349 goto abort_unlock;
5350 }
5351 if (mddev->pers) {
5352 err = update_array_info(mddev, &info);
5353 if (err) {
5354 printk(KERN_WARNING "md: couldn't update"
5355 " array info. %d\n", err);
5356 goto abort_unlock;
5357 }
5358 goto done_unlock;
5359 }
5360 if (!list_empty(&mddev->disks)) {
5361 printk(KERN_WARNING
5362 "md: array %s already has disks!\n",
5363 mdname(mddev));
5364 err = -EBUSY;
5365 goto abort_unlock;
5366 }
5367 if (mddev->raid_disks) {
5368 printk(KERN_WARNING
5369 "md: array %s already initialised!\n",
5370 mdname(mddev));
5371 err = -EBUSY;
5372 goto abort_unlock;
5373 }
5374 err = set_array_info(mddev, &info);
5375 if (err) {
5376 printk(KERN_WARNING "md: couldn't set"
5377 " array info. %d\n", err);
5378 goto abort_unlock;
5379 }
5380 }
5381 goto done_unlock;
5382
5383 default:;
5384 }
5385
5386 /*
5387 * Commands querying/configuring an existing array:
5388 */
5389 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5390 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5391 if ((!mddev->raid_disks && !mddev->external)
5392 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5393 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5394 && cmd != GET_BITMAP_FILE) {
5395 err = -ENODEV;
5396 goto abort_unlock;
5397 }
5398
5399 /*
5400 * Commands even a read-only array can execute:
5401 */
5402 switch (cmd)
5403 {
5404 case GET_ARRAY_INFO:
5405 err = get_array_info(mddev, argp);
5406 goto done_unlock;
5407
5408 case GET_BITMAP_FILE:
5409 err = get_bitmap_file(mddev, argp);
5410 goto done_unlock;
5411
5412 case GET_DISK_INFO:
5413 err = get_disk_info(mddev, argp);
5414 goto done_unlock;
5415
5416 case RESTART_ARRAY_RW:
5417 err = restart_array(mddev);
5418 goto done_unlock;
5419
5420 case STOP_ARRAY:
5421 err = do_md_stop(mddev, 0, 1);
5422 goto done_unlock;
5423
5424 case STOP_ARRAY_RO:
5425 err = do_md_stop(mddev, 1, 1);
5426 goto done_unlock;
5427
5428 }
5429
5430 /*
5431 * The remaining ioctls are changing the state of the
5432 * superblock, so we do not allow them on read-only arrays.
5433 * However non-MD ioctls (e.g. get-size) will still come through
5434 * here and hit the 'default' below, so only disallow
5435 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5436 */
5437 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5438 if (mddev->ro == 2) {
5439 mddev->ro = 0;
5440 sysfs_notify_dirent(mddev->sysfs_state);
5441 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5442 md_wakeup_thread(mddev->thread);
5443 } else {
5444 err = -EROFS;
5445 goto abort_unlock;
5446 }
5447 }
5448
5449 switch (cmd)
5450 {
5451 case ADD_NEW_DISK:
5452 {
5453 mdu_disk_info_t info;
5454 if (copy_from_user(&info, argp, sizeof(info)))
5455 err = -EFAULT;
5456 else
5457 err = add_new_disk(mddev, &info);
5458 goto done_unlock;
5459 }
5460
5461 case HOT_REMOVE_DISK:
5462 err = hot_remove_disk(mddev, new_decode_dev(arg));
5463 goto done_unlock;
5464
5465 case HOT_ADD_DISK:
5466 err = hot_add_disk(mddev, new_decode_dev(arg));
5467 goto done_unlock;
5468
5469 case SET_DISK_FAULTY:
5470 err = set_disk_faulty(mddev, new_decode_dev(arg));
5471 goto done_unlock;
5472
5473 case RUN_ARRAY:
5474 err = do_md_run(mddev);
5475 goto done_unlock;
5476
5477 case SET_BITMAP_FILE:
5478 err = set_bitmap_file(mddev, (int)arg);
5479 goto done_unlock;
5480
5481 default:
5482 err = -EINVAL;
5483 goto abort_unlock;
5484 }
5485
5486 done_unlock:
5487 abort_unlock:
5488 if (mddev->hold_active == UNTIL_IOCTL &&
5489 err != -EINVAL)
5490 mddev->hold_active = 0;
5491 mddev_unlock(mddev);
5492
5493 return err;
5494 done:
5495 if (err)
5496 MD_BUG();
5497 abort:
5498 return err;
5499 }
5500
5501 static int md_open(struct block_device *bdev, fmode_t mode)
5502 {
5503 /*
5504 * Succeed if we can lock the mddev, which confirms that
5505 * it isn't being stopped right now.
5506 */
5507 mddev_t *mddev = mddev_find(bdev->bd_dev);
5508 int err;
5509
5510 if (mddev->gendisk != bdev->bd_disk) {
5511 /* we are racing with mddev_put which is discarding this
5512 * bd_disk.
5513 */
5514 mddev_put(mddev);
5515 /* Wait until bdev->bd_disk is definitely gone */
5516 flush_scheduled_work();
5517 /* Then retry the open from the top */
5518 return -ERESTARTSYS;
5519 }
5520 BUG_ON(mddev != bdev->bd_disk->private_data);
5521
5522 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
5523 goto out;
5524
5525 err = 0;
5526 atomic_inc(&mddev->openers);
5527 mutex_unlock(&mddev->open_mutex);
5528
5529 check_disk_change(bdev);
5530 out:
5531 return err;
5532 }
5533
5534 static int md_release(struct gendisk *disk, fmode_t mode)
5535 {
5536 mddev_t *mddev = disk->private_data;
5537
5538 BUG_ON(!mddev);
5539 atomic_dec(&mddev->openers);
5540 mddev_put(mddev);
5541
5542 return 0;
5543 }
5544
5545 static int md_media_changed(struct gendisk *disk)
5546 {
5547 mddev_t *mddev = disk->private_data;
5548
5549 return mddev->changed;
5550 }
5551
5552 static int md_revalidate(struct gendisk *disk)
5553 {
5554 mddev_t *mddev = disk->private_data;
5555
5556 mddev->changed = 0;
5557 return 0;
5558 }
5559 static struct block_device_operations md_fops =
5560 {
5561 .owner = THIS_MODULE,
5562 .open = md_open,
5563 .release = md_release,
5564 .ioctl = md_ioctl,
5565 .getgeo = md_getgeo,
5566 .media_changed = md_media_changed,
5567 .revalidate_disk= md_revalidate,
5568 };
5569
5570 static int md_thread(void * arg)
5571 {
5572 mdk_thread_t *thread = arg;
5573
5574 /*
5575 * md_thread is a 'system-thread', it's priority should be very
5576 * high. We avoid resource deadlocks individually in each
5577 * raid personality. (RAID5 does preallocation) We also use RR and
5578 * the very same RT priority as kswapd, thus we will never get
5579 * into a priority inversion deadlock.
5580 *
5581 * we definitely have to have equal or higher priority than
5582 * bdflush, otherwise bdflush will deadlock if there are too
5583 * many dirty RAID5 blocks.
5584 */
5585
5586 allow_signal(SIGKILL);
5587 while (!kthread_should_stop()) {
5588
5589 /* We need to wait INTERRUPTIBLE so that
5590 * we don't add to the load-average.
5591 * That means we need to be sure no signals are
5592 * pending
5593 */
5594 if (signal_pending(current))
5595 flush_signals(current);
5596
5597 wait_event_interruptible_timeout
5598 (thread->wqueue,
5599 test_bit(THREAD_WAKEUP, &thread->flags)
5600 || kthread_should_stop(),
5601 thread->timeout);
5602
5603 clear_bit(THREAD_WAKEUP, &thread->flags);
5604
5605 thread->run(thread->mddev);
5606 }
5607
5608 return 0;
5609 }
5610
5611 void md_wakeup_thread(mdk_thread_t *thread)
5612 {
5613 if (thread) {
5614 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5615 set_bit(THREAD_WAKEUP, &thread->flags);
5616 wake_up(&thread->wqueue);
5617 }
5618 }
5619
5620 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5621 const char *name)
5622 {
5623 mdk_thread_t *thread;
5624
5625 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5626 if (!thread)
5627 return NULL;
5628
5629 init_waitqueue_head(&thread->wqueue);
5630
5631 thread->run = run;
5632 thread->mddev = mddev;
5633 thread->timeout = MAX_SCHEDULE_TIMEOUT;
5634 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
5635 if (IS_ERR(thread->tsk)) {
5636 kfree(thread);
5637 return NULL;
5638 }
5639 return thread;
5640 }
5641
5642 void md_unregister_thread(mdk_thread_t *thread)
5643 {
5644 if (!thread)
5645 return;
5646 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5647
5648 kthread_stop(thread->tsk);
5649 kfree(thread);
5650 }
5651
5652 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5653 {
5654 if (!mddev) {
5655 MD_BUG();
5656 return;
5657 }
5658
5659 if (!rdev || test_bit(Faulty, &rdev->flags))
5660 return;
5661
5662 if (mddev->external)
5663 set_bit(Blocked, &rdev->flags);
5664 /*
5665 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5666 mdname(mddev),
5667 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5668 __builtin_return_address(0),__builtin_return_address(1),
5669 __builtin_return_address(2),__builtin_return_address(3));
5670 */
5671 if (!mddev->pers)
5672 return;
5673 if (!mddev->pers->error_handler)
5674 return;
5675 mddev->pers->error_handler(mddev,rdev);
5676 if (mddev->degraded)
5677 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5678 set_bit(StateChanged, &rdev->flags);
5679 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5680 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5681 md_wakeup_thread(mddev->thread);
5682 md_new_event_inintr(mddev);
5683 }
5684
5685 /* seq_file implementation /proc/mdstat */
5686
5687 static void status_unused(struct seq_file *seq)
5688 {
5689 int i = 0;
5690 mdk_rdev_t *rdev;
5691
5692 seq_printf(seq, "unused devices: ");
5693
5694 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
5695 char b[BDEVNAME_SIZE];
5696 i++;
5697 seq_printf(seq, "%s ",
5698 bdevname(rdev->bdev,b));
5699 }
5700 if (!i)
5701 seq_printf(seq, "<none>");
5702
5703 seq_printf(seq, "\n");
5704 }
5705
5706
5707 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5708 {
5709 sector_t max_sectors, resync, res;
5710 unsigned long dt, db;
5711 sector_t rt;
5712 int scale;
5713 unsigned int per_milli;
5714
5715 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
5716
5717 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5718 max_sectors = mddev->resync_max_sectors;
5719 else
5720 max_sectors = mddev->dev_sectors;
5721
5722 /*
5723 * Should not happen.
5724 */
5725 if (!max_sectors) {
5726 MD_BUG();
5727 return;
5728 }
5729 /* Pick 'scale' such that (resync>>scale)*1000 will fit
5730 * in a sector_t, and (max_sectors>>scale) will fit in a
5731 * u32, as those are the requirements for sector_div.
5732 * Thus 'scale' must be at least 10
5733 */
5734 scale = 10;
5735 if (sizeof(sector_t) > sizeof(unsigned long)) {
5736 while ( max_sectors/2 > (1ULL<<(scale+32)))
5737 scale++;
5738 }
5739 res = (resync>>scale)*1000;
5740 sector_div(res, (u32)((max_sectors>>scale)+1));
5741
5742 per_milli = res;
5743 {
5744 int i, x = per_milli/50, y = 20-x;
5745 seq_printf(seq, "[");
5746 for (i = 0; i < x; i++)
5747 seq_printf(seq, "=");
5748 seq_printf(seq, ">");
5749 for (i = 0; i < y; i++)
5750 seq_printf(seq, ".");
5751 seq_printf(seq, "] ");
5752 }
5753 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5754 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5755 "reshape" :
5756 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5757 "check" :
5758 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5759 "resync" : "recovery"))),
5760 per_milli/10, per_milli % 10,
5761 (unsigned long long) resync/2,
5762 (unsigned long long) max_sectors/2);
5763
5764 /*
5765 * dt: time from mark until now
5766 * db: blocks written from mark until now
5767 * rt: remaining time
5768 *
5769 * rt is a sector_t, so could be 32bit or 64bit.
5770 * So we divide before multiply in case it is 32bit and close
5771 * to the limit.
5772 * We scale the divisor (db) by 32 to avoid loosing precision
5773 * near the end of resync when the number of remaining sectors
5774 * is close to 'db'.
5775 * We then divide rt by 32 after multiplying by db to compensate.
5776 * The '+1' avoids division by zero if db is very small.
5777 */
5778 dt = ((jiffies - mddev->resync_mark) / HZ);
5779 if (!dt) dt++;
5780 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5781 - mddev->resync_mark_cnt;
5782
5783 rt = max_sectors - resync; /* number of remaining sectors */
5784 sector_div(rt, db/32+1);
5785 rt *= dt;
5786 rt >>= 5;
5787
5788 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
5789 ((unsigned long)rt % 60)/6);
5790
5791 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5792 }
5793
5794 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5795 {
5796 struct list_head *tmp;
5797 loff_t l = *pos;
5798 mddev_t *mddev;
5799
5800 if (l >= 0x10000)
5801 return NULL;
5802 if (!l--)
5803 /* header */
5804 return (void*)1;
5805
5806 spin_lock(&all_mddevs_lock);
5807 list_for_each(tmp,&all_mddevs)
5808 if (!l--) {
5809 mddev = list_entry(tmp, mddev_t, all_mddevs);
5810 mddev_get(mddev);
5811 spin_unlock(&all_mddevs_lock);
5812 return mddev;
5813 }
5814 spin_unlock(&all_mddevs_lock);
5815 if (!l--)
5816 return (void*)2;/* tail */
5817 return NULL;
5818 }
5819
5820 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5821 {
5822 struct list_head *tmp;
5823 mddev_t *next_mddev, *mddev = v;
5824
5825 ++*pos;
5826 if (v == (void*)2)
5827 return NULL;
5828
5829 spin_lock(&all_mddevs_lock);
5830 if (v == (void*)1)
5831 tmp = all_mddevs.next;
5832 else
5833 tmp = mddev->all_mddevs.next;
5834 if (tmp != &all_mddevs)
5835 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5836 else {
5837 next_mddev = (void*)2;
5838 *pos = 0x10000;
5839 }
5840 spin_unlock(&all_mddevs_lock);
5841
5842 if (v != (void*)1)
5843 mddev_put(mddev);
5844 return next_mddev;
5845
5846 }
5847
5848 static void md_seq_stop(struct seq_file *seq, void *v)
5849 {
5850 mddev_t *mddev = v;
5851
5852 if (mddev && v != (void*)1 && v != (void*)2)
5853 mddev_put(mddev);
5854 }
5855
5856 struct mdstat_info {
5857 int event;
5858 };
5859
5860 static int md_seq_show(struct seq_file *seq, void *v)
5861 {
5862 mddev_t *mddev = v;
5863 sector_t sectors;
5864 mdk_rdev_t *rdev;
5865 struct mdstat_info *mi = seq->private;
5866 struct bitmap *bitmap;
5867
5868 if (v == (void*)1) {
5869 struct mdk_personality *pers;
5870 seq_printf(seq, "Personalities : ");
5871 spin_lock(&pers_lock);
5872 list_for_each_entry(pers, &pers_list, list)
5873 seq_printf(seq, "[%s] ", pers->name);
5874
5875 spin_unlock(&pers_lock);
5876 seq_printf(seq, "\n");
5877 mi->event = atomic_read(&md_event_count);
5878 return 0;
5879 }
5880 if (v == (void*)2) {
5881 status_unused(seq);
5882 return 0;
5883 }
5884
5885 if (mddev_lock(mddev) < 0)
5886 return -EINTR;
5887
5888 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5889 seq_printf(seq, "%s : %sactive", mdname(mddev),
5890 mddev->pers ? "" : "in");
5891 if (mddev->pers) {
5892 if (mddev->ro==1)
5893 seq_printf(seq, " (read-only)");
5894 if (mddev->ro==2)
5895 seq_printf(seq, " (auto-read-only)");
5896 seq_printf(seq, " %s", mddev->pers->name);
5897 }
5898
5899 sectors = 0;
5900 list_for_each_entry(rdev, &mddev->disks, same_set) {
5901 char b[BDEVNAME_SIZE];
5902 seq_printf(seq, " %s[%d]",
5903 bdevname(rdev->bdev,b), rdev->desc_nr);
5904 if (test_bit(WriteMostly, &rdev->flags))
5905 seq_printf(seq, "(W)");
5906 if (test_bit(Faulty, &rdev->flags)) {
5907 seq_printf(seq, "(F)");
5908 continue;
5909 } else if (rdev->raid_disk < 0)
5910 seq_printf(seq, "(S)"); /* spare */
5911 sectors += rdev->sectors;
5912 }
5913
5914 if (!list_empty(&mddev->disks)) {
5915 if (mddev->pers)
5916 seq_printf(seq, "\n %llu blocks",
5917 (unsigned long long)
5918 mddev->array_sectors / 2);
5919 else
5920 seq_printf(seq, "\n %llu blocks",
5921 (unsigned long long)sectors / 2);
5922 }
5923 if (mddev->persistent) {
5924 if (mddev->major_version != 0 ||
5925 mddev->minor_version != 90) {
5926 seq_printf(seq," super %d.%d",
5927 mddev->major_version,
5928 mddev->minor_version);
5929 }
5930 } else if (mddev->external)
5931 seq_printf(seq, " super external:%s",
5932 mddev->metadata_type);
5933 else
5934 seq_printf(seq, " super non-persistent");
5935
5936 if (mddev->pers) {
5937 mddev->pers->status(seq, mddev);
5938 seq_printf(seq, "\n ");
5939 if (mddev->pers->sync_request) {
5940 if (mddev->curr_resync > 2) {
5941 status_resync(seq, mddev);
5942 seq_printf(seq, "\n ");
5943 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5944 seq_printf(seq, "\tresync=DELAYED\n ");
5945 else if (mddev->recovery_cp < MaxSector)
5946 seq_printf(seq, "\tresync=PENDING\n ");
5947 }
5948 } else
5949 seq_printf(seq, "\n ");
5950
5951 if ((bitmap = mddev->bitmap)) {
5952 unsigned long chunk_kb;
5953 unsigned long flags;
5954 spin_lock_irqsave(&bitmap->lock, flags);
5955 chunk_kb = bitmap->chunksize >> 10;
5956 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5957 "%lu%s chunk",
5958 bitmap->pages - bitmap->missing_pages,
5959 bitmap->pages,
5960 (bitmap->pages - bitmap->missing_pages)
5961 << (PAGE_SHIFT - 10),
5962 chunk_kb ? chunk_kb : bitmap->chunksize,
5963 chunk_kb ? "KB" : "B");
5964 if (bitmap->file) {
5965 seq_printf(seq, ", file: ");
5966 seq_path(seq, &bitmap->file->f_path, " \t\n");
5967 }
5968
5969 seq_printf(seq, "\n");
5970 spin_unlock_irqrestore(&bitmap->lock, flags);
5971 }
5972
5973 seq_printf(seq, "\n");
5974 }
5975 mddev_unlock(mddev);
5976
5977 return 0;
5978 }
5979
5980 static const struct seq_operations md_seq_ops = {
5981 .start = md_seq_start,
5982 .next = md_seq_next,
5983 .stop = md_seq_stop,
5984 .show = md_seq_show,
5985 };
5986
5987 static int md_seq_open(struct inode *inode, struct file *file)
5988 {
5989 int error;
5990 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5991 if (mi == NULL)
5992 return -ENOMEM;
5993
5994 error = seq_open(file, &md_seq_ops);
5995 if (error)
5996 kfree(mi);
5997 else {
5998 struct seq_file *p = file->private_data;
5999 p->private = mi;
6000 mi->event = atomic_read(&md_event_count);
6001 }
6002 return error;
6003 }
6004
6005 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6006 {
6007 struct seq_file *m = filp->private_data;
6008 struct mdstat_info *mi = m->private;
6009 int mask;
6010
6011 poll_wait(filp, &md_event_waiters, wait);
6012
6013 /* always allow read */
6014 mask = POLLIN | POLLRDNORM;
6015
6016 if (mi->event != atomic_read(&md_event_count))
6017 mask |= POLLERR | POLLPRI;
6018 return mask;
6019 }
6020
6021 static const struct file_operations md_seq_fops = {
6022 .owner = THIS_MODULE,
6023 .open = md_seq_open,
6024 .read = seq_read,
6025 .llseek = seq_lseek,
6026 .release = seq_release_private,
6027 .poll = mdstat_poll,
6028 };
6029
6030 int register_md_personality(struct mdk_personality *p)
6031 {
6032 spin_lock(&pers_lock);
6033 list_add_tail(&p->list, &pers_list);
6034 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6035 spin_unlock(&pers_lock);
6036 return 0;
6037 }
6038
6039 int unregister_md_personality(struct mdk_personality *p)
6040 {
6041 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6042 spin_lock(&pers_lock);
6043 list_del_init(&p->list);
6044 spin_unlock(&pers_lock);
6045 return 0;
6046 }
6047
6048 static int is_mddev_idle(mddev_t *mddev, int init)
6049 {
6050 mdk_rdev_t * rdev;
6051 int idle;
6052 int curr_events;
6053
6054 idle = 1;
6055 rcu_read_lock();
6056 rdev_for_each_rcu(rdev, mddev) {
6057 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6058 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6059 (int)part_stat_read(&disk->part0, sectors[1]) -
6060 atomic_read(&disk->sync_io);
6061 /* sync IO will cause sync_io to increase before the disk_stats
6062 * as sync_io is counted when a request starts, and
6063 * disk_stats is counted when it completes.
6064 * So resync activity will cause curr_events to be smaller than
6065 * when there was no such activity.
6066 * non-sync IO will cause disk_stat to increase without
6067 * increasing sync_io so curr_events will (eventually)
6068 * be larger than it was before. Once it becomes
6069 * substantially larger, the test below will cause
6070 * the array to appear non-idle, and resync will slow
6071 * down.
6072 * If there is a lot of outstanding resync activity when
6073 * we set last_event to curr_events, then all that activity
6074 * completing might cause the array to appear non-idle
6075 * and resync will be slowed down even though there might
6076 * not have been non-resync activity. This will only
6077 * happen once though. 'last_events' will soon reflect
6078 * the state where there is little or no outstanding
6079 * resync requests, and further resync activity will
6080 * always make curr_events less than last_events.
6081 *
6082 */
6083 if (init || curr_events - rdev->last_events > 64) {
6084 rdev->last_events = curr_events;
6085 idle = 0;
6086 }
6087 }
6088 rcu_read_unlock();
6089 return idle;
6090 }
6091
6092 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6093 {
6094 /* another "blocks" (512byte) blocks have been synced */
6095 atomic_sub(blocks, &mddev->recovery_active);
6096 wake_up(&mddev->recovery_wait);
6097 if (!ok) {
6098 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6099 md_wakeup_thread(mddev->thread);
6100 // stop recovery, signal do_sync ....
6101 }
6102 }
6103
6104
6105 /* md_write_start(mddev, bi)
6106 * If we need to update some array metadata (e.g. 'active' flag
6107 * in superblock) before writing, schedule a superblock update
6108 * and wait for it to complete.
6109 */
6110 void md_write_start(mddev_t *mddev, struct bio *bi)
6111 {
6112 int did_change = 0;
6113 if (bio_data_dir(bi) != WRITE)
6114 return;
6115
6116 BUG_ON(mddev->ro == 1);
6117 if (mddev->ro == 2) {
6118 /* need to switch to read/write */
6119 mddev->ro = 0;
6120 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6121 md_wakeup_thread(mddev->thread);
6122 md_wakeup_thread(mddev->sync_thread);
6123 did_change = 1;
6124 }
6125 atomic_inc(&mddev->writes_pending);
6126 if (mddev->safemode == 1)
6127 mddev->safemode = 0;
6128 if (mddev->in_sync) {
6129 spin_lock_irq(&mddev->write_lock);
6130 if (mddev->in_sync) {
6131 mddev->in_sync = 0;
6132 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6133 md_wakeup_thread(mddev->thread);
6134 did_change = 1;
6135 }
6136 spin_unlock_irq(&mddev->write_lock);
6137 }
6138 if (did_change)
6139 sysfs_notify_dirent(mddev->sysfs_state);
6140 wait_event(mddev->sb_wait,
6141 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
6142 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6143 }
6144
6145 void md_write_end(mddev_t *mddev)
6146 {
6147 if (atomic_dec_and_test(&mddev->writes_pending)) {
6148 if (mddev->safemode == 2)
6149 md_wakeup_thread(mddev->thread);
6150 else if (mddev->safemode_delay)
6151 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6152 }
6153 }
6154
6155 /* md_allow_write(mddev)
6156 * Calling this ensures that the array is marked 'active' so that writes
6157 * may proceed without blocking. It is important to call this before
6158 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6159 * Must be called with mddev_lock held.
6160 *
6161 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6162 * is dropped, so return -EAGAIN after notifying userspace.
6163 */
6164 int md_allow_write(mddev_t *mddev)
6165 {
6166 if (!mddev->pers)
6167 return 0;
6168 if (mddev->ro)
6169 return 0;
6170 if (!mddev->pers->sync_request)
6171 return 0;
6172
6173 spin_lock_irq(&mddev->write_lock);
6174 if (mddev->in_sync) {
6175 mddev->in_sync = 0;
6176 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6177 if (mddev->safemode_delay &&
6178 mddev->safemode == 0)
6179 mddev->safemode = 1;
6180 spin_unlock_irq(&mddev->write_lock);
6181 md_update_sb(mddev, 0);
6182 sysfs_notify_dirent(mddev->sysfs_state);
6183 } else
6184 spin_unlock_irq(&mddev->write_lock);
6185
6186 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
6187 return -EAGAIN;
6188 else
6189 return 0;
6190 }
6191 EXPORT_SYMBOL_GPL(md_allow_write);
6192
6193 #define SYNC_MARKS 10
6194 #define SYNC_MARK_STEP (3*HZ)
6195 void md_do_sync(mddev_t *mddev)
6196 {
6197 mddev_t *mddev2;
6198 unsigned int currspeed = 0,
6199 window;
6200 sector_t max_sectors,j, io_sectors;
6201 unsigned long mark[SYNC_MARKS];
6202 sector_t mark_cnt[SYNC_MARKS];
6203 int last_mark,m;
6204 struct list_head *tmp;
6205 sector_t last_check;
6206 int skipped = 0;
6207 mdk_rdev_t *rdev;
6208 char *desc;
6209
6210 /* just incase thread restarts... */
6211 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6212 return;
6213 if (mddev->ro) /* never try to sync a read-only array */
6214 return;
6215
6216 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6217 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6218 desc = "data-check";
6219 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6220 desc = "requested-resync";
6221 else
6222 desc = "resync";
6223 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6224 desc = "reshape";
6225 else
6226 desc = "recovery";
6227
6228 /* we overload curr_resync somewhat here.
6229 * 0 == not engaged in resync at all
6230 * 2 == checking that there is no conflict with another sync
6231 * 1 == like 2, but have yielded to allow conflicting resync to
6232 * commense
6233 * other == active in resync - this many blocks
6234 *
6235 * Before starting a resync we must have set curr_resync to
6236 * 2, and then checked that every "conflicting" array has curr_resync
6237 * less than ours. When we find one that is the same or higher
6238 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6239 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6240 * This will mean we have to start checking from the beginning again.
6241 *
6242 */
6243
6244 do {
6245 mddev->curr_resync = 2;
6246
6247 try_again:
6248 if (kthread_should_stop()) {
6249 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6250 goto skip;
6251 }
6252 for_each_mddev(mddev2, tmp) {
6253 if (mddev2 == mddev)
6254 continue;
6255 if (!mddev->parallel_resync
6256 && mddev2->curr_resync
6257 && match_mddev_units(mddev, mddev2)) {
6258 DEFINE_WAIT(wq);
6259 if (mddev < mddev2 && mddev->curr_resync == 2) {
6260 /* arbitrarily yield */
6261 mddev->curr_resync = 1;
6262 wake_up(&resync_wait);
6263 }
6264 if (mddev > mddev2 && mddev->curr_resync == 1)
6265 /* no need to wait here, we can wait the next
6266 * time 'round when curr_resync == 2
6267 */
6268 continue;
6269 /* We need to wait 'interruptible' so as not to
6270 * contribute to the load average, and not to
6271 * be caught by 'softlockup'
6272 */
6273 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6274 if (!kthread_should_stop() &&
6275 mddev2->curr_resync >= mddev->curr_resync) {
6276 printk(KERN_INFO "md: delaying %s of %s"
6277 " until %s has finished (they"
6278 " share one or more physical units)\n",
6279 desc, mdname(mddev), mdname(mddev2));
6280 mddev_put(mddev2);
6281 if (signal_pending(current))
6282 flush_signals(current);
6283 schedule();
6284 finish_wait(&resync_wait, &wq);
6285 goto try_again;
6286 }
6287 finish_wait(&resync_wait, &wq);
6288 }
6289 }
6290 } while (mddev->curr_resync < 2);
6291
6292 j = 0;
6293 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6294 /* resync follows the size requested by the personality,
6295 * which defaults to physical size, but can be virtual size
6296 */
6297 max_sectors = mddev->resync_max_sectors;
6298 mddev->resync_mismatches = 0;
6299 /* we don't use the checkpoint if there's a bitmap */
6300 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6301 j = mddev->resync_min;
6302 else if (!mddev->bitmap)
6303 j = mddev->recovery_cp;
6304
6305 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6306 max_sectors = mddev->dev_sectors;
6307 else {
6308 /* recovery follows the physical size of devices */
6309 max_sectors = mddev->dev_sectors;
6310 j = MaxSector;
6311 list_for_each_entry(rdev, &mddev->disks, same_set)
6312 if (rdev->raid_disk >= 0 &&
6313 !test_bit(Faulty, &rdev->flags) &&
6314 !test_bit(In_sync, &rdev->flags) &&
6315 rdev->recovery_offset < j)
6316 j = rdev->recovery_offset;
6317 }
6318
6319 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6320 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6321 " %d KB/sec/disk.\n", speed_min(mddev));
6322 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6323 "(but not more than %d KB/sec) for %s.\n",
6324 speed_max(mddev), desc);
6325
6326 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6327
6328 io_sectors = 0;
6329 for (m = 0; m < SYNC_MARKS; m++) {
6330 mark[m] = jiffies;
6331 mark_cnt[m] = io_sectors;
6332 }
6333 last_mark = 0;
6334 mddev->resync_mark = mark[last_mark];
6335 mddev->resync_mark_cnt = mark_cnt[last_mark];
6336
6337 /*
6338 * Tune reconstruction:
6339 */
6340 window = 32*(PAGE_SIZE/512);
6341 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6342 window/2,(unsigned long long) max_sectors/2);
6343
6344 atomic_set(&mddev->recovery_active, 0);
6345 last_check = 0;
6346
6347 if (j>2) {
6348 printk(KERN_INFO
6349 "md: resuming %s of %s from checkpoint.\n",
6350 desc, mdname(mddev));
6351 mddev->curr_resync = j;
6352 }
6353
6354 while (j < max_sectors) {
6355 sector_t sectors;
6356
6357 skipped = 0;
6358
6359 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6360 ((mddev->curr_resync > mddev->curr_resync_completed &&
6361 (mddev->curr_resync - mddev->curr_resync_completed)
6362 > (max_sectors >> 4)) ||
6363 (j - mddev->curr_resync_completed)*2
6364 >= mddev->resync_max - mddev->curr_resync_completed
6365 )) {
6366 /* time to update curr_resync_completed */
6367 blk_unplug(mddev->queue);
6368 wait_event(mddev->recovery_wait,
6369 atomic_read(&mddev->recovery_active) == 0);
6370 mddev->curr_resync_completed =
6371 mddev->curr_resync;
6372 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6373 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6374 }
6375
6376 while (j >= mddev->resync_max && !kthread_should_stop()) {
6377 /* As this condition is controlled by user-space,
6378 * we can block indefinitely, so use '_interruptible'
6379 * to avoid triggering warnings.
6380 */
6381 flush_signals(current); /* just in case */
6382 wait_event_interruptible(mddev->recovery_wait,
6383 mddev->resync_max > j
6384 || kthread_should_stop());
6385 }
6386
6387 if (kthread_should_stop())
6388 goto interrupted;
6389
6390 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6391 currspeed < speed_min(mddev));
6392 if (sectors == 0) {
6393 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6394 goto out;
6395 }
6396
6397 if (!skipped) { /* actual IO requested */
6398 io_sectors += sectors;
6399 atomic_add(sectors, &mddev->recovery_active);
6400 }
6401
6402 j += sectors;
6403 if (j>1) mddev->curr_resync = j;
6404 mddev->curr_mark_cnt = io_sectors;
6405 if (last_check == 0)
6406 /* this is the earliers that rebuilt will be
6407 * visible in /proc/mdstat
6408 */
6409 md_new_event(mddev);
6410
6411 if (last_check + window > io_sectors || j == max_sectors)
6412 continue;
6413
6414 last_check = io_sectors;
6415
6416 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6417 break;
6418
6419 repeat:
6420 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6421 /* step marks */
6422 int next = (last_mark+1) % SYNC_MARKS;
6423
6424 mddev->resync_mark = mark[next];
6425 mddev->resync_mark_cnt = mark_cnt[next];
6426 mark[next] = jiffies;
6427 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6428 last_mark = next;
6429 }
6430
6431
6432 if (kthread_should_stop())
6433 goto interrupted;
6434
6435
6436 /*
6437 * this loop exits only if either when we are slower than
6438 * the 'hard' speed limit, or the system was IO-idle for
6439 * a jiffy.
6440 * the system might be non-idle CPU-wise, but we only care
6441 * about not overloading the IO subsystem. (things like an
6442 * e2fsck being done on the RAID array should execute fast)
6443 */
6444 blk_unplug(mddev->queue);
6445 cond_resched();
6446
6447 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6448 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6449
6450 if (currspeed > speed_min(mddev)) {
6451 if ((currspeed > speed_max(mddev)) ||
6452 !is_mddev_idle(mddev, 0)) {
6453 msleep(500);
6454 goto repeat;
6455 }
6456 }
6457 }
6458 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6459 /*
6460 * this also signals 'finished resyncing' to md_stop
6461 */
6462 out:
6463 blk_unplug(mddev->queue);
6464
6465 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6466
6467 /* tell personality that we are finished */
6468 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6469
6470 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6471 mddev->curr_resync > 2) {
6472 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6473 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6474 if (mddev->curr_resync >= mddev->recovery_cp) {
6475 printk(KERN_INFO
6476 "md: checkpointing %s of %s.\n",
6477 desc, mdname(mddev));
6478 mddev->recovery_cp = mddev->curr_resync;
6479 }
6480 } else
6481 mddev->recovery_cp = MaxSector;
6482 } else {
6483 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6484 mddev->curr_resync = MaxSector;
6485 list_for_each_entry(rdev, &mddev->disks, same_set)
6486 if (rdev->raid_disk >= 0 &&
6487 !test_bit(Faulty, &rdev->flags) &&
6488 !test_bit(In_sync, &rdev->flags) &&
6489 rdev->recovery_offset < mddev->curr_resync)
6490 rdev->recovery_offset = mddev->curr_resync;
6491 }
6492 }
6493 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6494
6495 skip:
6496 mddev->curr_resync = 0;
6497 mddev->curr_resync_completed = 0;
6498 mddev->resync_min = 0;
6499 mddev->resync_max = MaxSector;
6500 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6501 wake_up(&resync_wait);
6502 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6503 md_wakeup_thread(mddev->thread);
6504 return;
6505
6506 interrupted:
6507 /*
6508 * got a signal, exit.
6509 */
6510 printk(KERN_INFO
6511 "md: md_do_sync() got signal ... exiting\n");
6512 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6513 goto out;
6514
6515 }
6516 EXPORT_SYMBOL_GPL(md_do_sync);
6517
6518
6519 static int remove_and_add_spares(mddev_t *mddev)
6520 {
6521 mdk_rdev_t *rdev;
6522 int spares = 0;
6523
6524 mddev->curr_resync_completed = 0;
6525
6526 list_for_each_entry(rdev, &mddev->disks, same_set)
6527 if (rdev->raid_disk >= 0 &&
6528 !test_bit(Blocked, &rdev->flags) &&
6529 (test_bit(Faulty, &rdev->flags) ||
6530 ! test_bit(In_sync, &rdev->flags)) &&
6531 atomic_read(&rdev->nr_pending)==0) {
6532 if (mddev->pers->hot_remove_disk(
6533 mddev, rdev->raid_disk)==0) {
6534 char nm[20];
6535 sprintf(nm,"rd%d", rdev->raid_disk);
6536 sysfs_remove_link(&mddev->kobj, nm);
6537 rdev->raid_disk = -1;
6538 }
6539 }
6540
6541 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
6542 list_for_each_entry(rdev, &mddev->disks, same_set) {
6543 if (rdev->raid_disk >= 0 &&
6544 !test_bit(In_sync, &rdev->flags) &&
6545 !test_bit(Blocked, &rdev->flags))
6546 spares++;
6547 if (rdev->raid_disk < 0
6548 && !test_bit(Faulty, &rdev->flags)) {
6549 rdev->recovery_offset = 0;
6550 if (mddev->pers->
6551 hot_add_disk(mddev, rdev) == 0) {
6552 char nm[20];
6553 sprintf(nm, "rd%d", rdev->raid_disk);
6554 if (sysfs_create_link(&mddev->kobj,
6555 &rdev->kobj, nm))
6556 printk(KERN_WARNING
6557 "md: cannot register "
6558 "%s for %s\n",
6559 nm, mdname(mddev));
6560 spares++;
6561 md_new_event(mddev);
6562 } else
6563 break;
6564 }
6565 }
6566 }
6567 return spares;
6568 }
6569 /*
6570 * This routine is regularly called by all per-raid-array threads to
6571 * deal with generic issues like resync and super-block update.
6572 * Raid personalities that don't have a thread (linear/raid0) do not
6573 * need this as they never do any recovery or update the superblock.
6574 *
6575 * It does not do any resync itself, but rather "forks" off other threads
6576 * to do that as needed.
6577 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6578 * "->recovery" and create a thread at ->sync_thread.
6579 * When the thread finishes it sets MD_RECOVERY_DONE
6580 * and wakeups up this thread which will reap the thread and finish up.
6581 * This thread also removes any faulty devices (with nr_pending == 0).
6582 *
6583 * The overall approach is:
6584 * 1/ if the superblock needs updating, update it.
6585 * 2/ If a recovery thread is running, don't do anything else.
6586 * 3/ If recovery has finished, clean up, possibly marking spares active.
6587 * 4/ If there are any faulty devices, remove them.
6588 * 5/ If array is degraded, try to add spares devices
6589 * 6/ If array has spares or is not in-sync, start a resync thread.
6590 */
6591 void md_check_recovery(mddev_t *mddev)
6592 {
6593 mdk_rdev_t *rdev;
6594
6595
6596 if (mddev->bitmap)
6597 bitmap_daemon_work(mddev->bitmap);
6598
6599 if (mddev->ro)
6600 return;
6601
6602 if (signal_pending(current)) {
6603 if (mddev->pers->sync_request && !mddev->external) {
6604 printk(KERN_INFO "md: %s in immediate safe mode\n",
6605 mdname(mddev));
6606 mddev->safemode = 2;
6607 }
6608 flush_signals(current);
6609 }
6610
6611 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
6612 return;
6613 if ( ! (
6614 (mddev->flags && !mddev->external) ||
6615 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
6616 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
6617 (mddev->external == 0 && mddev->safemode == 1) ||
6618 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6619 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6620 ))
6621 return;
6622
6623 if (mddev_trylock(mddev)) {
6624 int spares = 0;
6625
6626 if (mddev->ro) {
6627 /* Only thing we do on a ro array is remove
6628 * failed devices.
6629 */
6630 remove_and_add_spares(mddev);
6631 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6632 goto unlock;
6633 }
6634
6635 if (!mddev->external) {
6636 int did_change = 0;
6637 spin_lock_irq(&mddev->write_lock);
6638 if (mddev->safemode &&
6639 !atomic_read(&mddev->writes_pending) &&
6640 !mddev->in_sync &&
6641 mddev->recovery_cp == MaxSector) {
6642 mddev->in_sync = 1;
6643 did_change = 1;
6644 if (mddev->persistent)
6645 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6646 }
6647 if (mddev->safemode == 1)
6648 mddev->safemode = 0;
6649 spin_unlock_irq(&mddev->write_lock);
6650 if (did_change)
6651 sysfs_notify_dirent(mddev->sysfs_state);
6652 }
6653
6654 if (mddev->flags)
6655 md_update_sb(mddev, 0);
6656
6657 list_for_each_entry(rdev, &mddev->disks, same_set)
6658 if (test_and_clear_bit(StateChanged, &rdev->flags))
6659 sysfs_notify_dirent(rdev->sysfs_state);
6660
6661
6662 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6663 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6664 /* resync/recovery still happening */
6665 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6666 goto unlock;
6667 }
6668 if (mddev->sync_thread) {
6669 /* resync has finished, collect result */
6670 md_unregister_thread(mddev->sync_thread);
6671 mddev->sync_thread = NULL;
6672 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
6673 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
6674 /* success...*/
6675 /* activate any spares */
6676 if (mddev->pers->spare_active(mddev))
6677 sysfs_notify(&mddev->kobj, NULL,
6678 "degraded");
6679 }
6680 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6681 mddev->pers->finish_reshape)
6682 mddev->pers->finish_reshape(mddev);
6683 md_update_sb(mddev, 1);
6684
6685 /* if array is no-longer degraded, then any saved_raid_disk
6686 * information must be scrapped
6687 */
6688 if (!mddev->degraded)
6689 list_for_each_entry(rdev, &mddev->disks, same_set)
6690 rdev->saved_raid_disk = -1;
6691
6692 mddev->recovery = 0;
6693 /* flag recovery needed just to double check */
6694 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6695 sysfs_notify_dirent(mddev->sysfs_action);
6696 md_new_event(mddev);
6697 goto unlock;
6698 }
6699 /* Set RUNNING before clearing NEEDED to avoid
6700 * any transients in the value of "sync_action".
6701 */
6702 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6703 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6704 /* Clear some bits that don't mean anything, but
6705 * might be left set
6706 */
6707 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
6708 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
6709
6710 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6711 goto unlock;
6712 /* no recovery is running.
6713 * remove any failed drives, then
6714 * add spares if possible.
6715 * Spare are also removed and re-added, to allow
6716 * the personality to fail the re-add.
6717 */
6718
6719 if (mddev->reshape_position != MaxSector) {
6720 if (mddev->pers->check_reshape == NULL ||
6721 mddev->pers->check_reshape(mddev) != 0)
6722 /* Cannot proceed */
6723 goto unlock;
6724 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6725 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6726 } else if ((spares = remove_and_add_spares(mddev))) {
6727 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6728 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6729 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
6730 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6731 } else if (mddev->recovery_cp < MaxSector) {
6732 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6733 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6734 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6735 /* nothing to be done ... */
6736 goto unlock;
6737
6738 if (mddev->pers->sync_request) {
6739 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6740 /* We are adding a device or devices to an array
6741 * which has the bitmap stored on all devices.
6742 * So make sure all bitmap pages get written
6743 */
6744 bitmap_write_all(mddev->bitmap);
6745 }
6746 mddev->sync_thread = md_register_thread(md_do_sync,
6747 mddev,
6748 "%s_resync");
6749 if (!mddev->sync_thread) {
6750 printk(KERN_ERR "%s: could not start resync"
6751 " thread...\n",
6752 mdname(mddev));
6753 /* leave the spares where they are, it shouldn't hurt */
6754 mddev->recovery = 0;
6755 } else
6756 md_wakeup_thread(mddev->sync_thread);
6757 sysfs_notify_dirent(mddev->sysfs_action);
6758 md_new_event(mddev);
6759 }
6760 unlock:
6761 if (!mddev->sync_thread) {
6762 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6763 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
6764 &mddev->recovery))
6765 if (mddev->sysfs_action)
6766 sysfs_notify_dirent(mddev->sysfs_action);
6767 }
6768 mddev_unlock(mddev);
6769 }
6770 }
6771
6772 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
6773 {
6774 sysfs_notify_dirent(rdev->sysfs_state);
6775 wait_event_timeout(rdev->blocked_wait,
6776 !test_bit(Blocked, &rdev->flags),
6777 msecs_to_jiffies(5000));
6778 rdev_dec_pending(rdev, mddev);
6779 }
6780 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
6781
6782 static int md_notify_reboot(struct notifier_block *this,
6783 unsigned long code, void *x)
6784 {
6785 struct list_head *tmp;
6786 mddev_t *mddev;
6787
6788 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
6789
6790 printk(KERN_INFO "md: stopping all md devices.\n");
6791
6792 for_each_mddev(mddev, tmp)
6793 if (mddev_trylock(mddev)) {
6794 /* Force a switch to readonly even array
6795 * appears to still be in use. Hence
6796 * the '100'.
6797 */
6798 do_md_stop(mddev, 1, 100);
6799 mddev_unlock(mddev);
6800 }
6801 /*
6802 * certain more exotic SCSI devices are known to be
6803 * volatile wrt too early system reboots. While the
6804 * right place to handle this issue is the given
6805 * driver, we do want to have a safe RAID driver ...
6806 */
6807 mdelay(1000*1);
6808 }
6809 return NOTIFY_DONE;
6810 }
6811
6812 static struct notifier_block md_notifier = {
6813 .notifier_call = md_notify_reboot,
6814 .next = NULL,
6815 .priority = INT_MAX, /* before any real devices */
6816 };
6817
6818 static void md_geninit(void)
6819 {
6820 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
6821
6822 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
6823 }
6824
6825 static int __init md_init(void)
6826 {
6827 if (register_blkdev(MD_MAJOR, "md"))
6828 return -1;
6829 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
6830 unregister_blkdev(MD_MAJOR, "md");
6831 return -1;
6832 }
6833 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
6834 md_probe, NULL, NULL);
6835 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
6836 md_probe, NULL, NULL);
6837
6838 register_reboot_notifier(&md_notifier);
6839 raid_table_header = register_sysctl_table(raid_root_table);
6840
6841 md_geninit();
6842 return 0;
6843 }
6844
6845
6846 #ifndef MODULE
6847
6848 /*
6849 * Searches all registered partitions for autorun RAID arrays
6850 * at boot time.
6851 */
6852
6853 static LIST_HEAD(all_detected_devices);
6854 struct detected_devices_node {
6855 struct list_head list;
6856 dev_t dev;
6857 };
6858
6859 void md_autodetect_dev(dev_t dev)
6860 {
6861 struct detected_devices_node *node_detected_dev;
6862
6863 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6864 if (node_detected_dev) {
6865 node_detected_dev->dev = dev;
6866 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6867 } else {
6868 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6869 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6870 }
6871 }
6872
6873
6874 static void autostart_arrays(int part)
6875 {
6876 mdk_rdev_t *rdev;
6877 struct detected_devices_node *node_detected_dev;
6878 dev_t dev;
6879 int i_scanned, i_passed;
6880
6881 i_scanned = 0;
6882 i_passed = 0;
6883
6884 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6885
6886 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6887 i_scanned++;
6888 node_detected_dev = list_entry(all_detected_devices.next,
6889 struct detected_devices_node, list);
6890 list_del(&node_detected_dev->list);
6891 dev = node_detected_dev->dev;
6892 kfree(node_detected_dev);
6893 rdev = md_import_device(dev,0, 90);
6894 if (IS_ERR(rdev))
6895 continue;
6896
6897 if (test_bit(Faulty, &rdev->flags)) {
6898 MD_BUG();
6899 continue;
6900 }
6901 set_bit(AutoDetected, &rdev->flags);
6902 list_add(&rdev->same_set, &pending_raid_disks);
6903 i_passed++;
6904 }
6905
6906 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6907 i_scanned, i_passed);
6908
6909 autorun_devices(part);
6910 }
6911
6912 #endif /* !MODULE */
6913
6914 static __exit void md_exit(void)
6915 {
6916 mddev_t *mddev;
6917 struct list_head *tmp;
6918
6919 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
6920 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6921
6922 unregister_blkdev(MD_MAJOR,"md");
6923 unregister_blkdev(mdp_major, "mdp");
6924 unregister_reboot_notifier(&md_notifier);
6925 unregister_sysctl_table(raid_table_header);
6926 remove_proc_entry("mdstat", NULL);
6927 for_each_mddev(mddev, tmp) {
6928 export_array(mddev);
6929 mddev->hold_active = 0;
6930 }
6931 }
6932
6933 subsys_initcall(md_init);
6934 module_exit(md_exit)
6935
6936 static int get_ro(char *buffer, struct kernel_param *kp)
6937 {
6938 return sprintf(buffer, "%d", start_readonly);
6939 }
6940 static int set_ro(const char *val, struct kernel_param *kp)
6941 {
6942 char *e;
6943 int num = simple_strtoul(val, &e, 10);
6944 if (*val && (*e == '\0' || *e == '\n')) {
6945 start_readonly = num;
6946 return 0;
6947 }
6948 return -EINVAL;
6949 }
6950
6951 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6952 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6953
6954 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
6955
6956 EXPORT_SYMBOL(register_md_personality);
6957 EXPORT_SYMBOL(unregister_md_personality);
6958 EXPORT_SYMBOL(md_error);
6959 EXPORT_SYMBOL(md_done_sync);
6960 EXPORT_SYMBOL(md_write_start);
6961 EXPORT_SYMBOL(md_write_end);
6962 EXPORT_SYMBOL(md_register_thread);
6963 EXPORT_SYMBOL(md_unregister_thread);
6964 EXPORT_SYMBOL(md_wakeup_thread);
6965 EXPORT_SYMBOL(md_check_recovery);
6966 MODULE_LICENSE("GPL");
6967 MODULE_ALIAS("md");
6968 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Support for the Chinese Samsung S3C2440 based development boards