?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ACPI: pci_root: simplify list traversals
[linux-2.6/linux-acpi-2.6.git] / drivers / md / md.c
blob641b211fe3feb33e15cb1159e3a76ca26b90c124
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->reconfig_mutex);
363 INIT_LIST_HEAD(&new->disks);
364 INIT_LIST_HEAD(&new->all_mddevs);
365 init_timer(&new->safemode_timer);
366 atomic_set(&new->active, 1);
367 atomic_set(&new->openers, 0);
368 atomic_set(&new->active_io, 0);
369 spin_lock_init(&new->write_lock);
370 init_waitqueue_head(&new->sb_wait);
371 init_waitqueue_head(&new->recovery_wait);
372 new->reshape_position = MaxSector;
373 new->resync_min = 0;
374 new->resync_max = MaxSector;
375 new->level = LEVEL_NONE;
376
377 goto retry;
378 }
379
380 static inline int mddev_lock(mddev_t * mddev)
381 {
382 return mutex_lock_interruptible(&mddev->reconfig_mutex);
383 }
384
385 static inline int mddev_is_locked(mddev_t *mddev)
386 {
387 return mutex_is_locked(&mddev->reconfig_mutex);
388 }
389
390 static inline int mddev_trylock(mddev_t * mddev)
391 {
392 return mutex_trylock(&mddev->reconfig_mutex);
393 }
394
395 static inline void mddev_unlock(mddev_t * mddev)
396 {
397 mutex_unlock(&mddev->reconfig_mutex);
398
399 md_wakeup_thread(mddev->thread);
400 }
401
402 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
403 {
404 mdk_rdev_t *rdev;
405
406 list_for_each_entry(rdev, &mddev->disks, same_set)
407 if (rdev->desc_nr == nr)
408 return rdev;
409
410 return NULL;
411 }
412
413 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
414 {
415 mdk_rdev_t *rdev;
416
417 list_for_each_entry(rdev, &mddev->disks, same_set)
418 if (rdev->bdev->bd_dev == dev)
419 return rdev;
420
421 return NULL;
422 }
423
424 static struct mdk_personality *find_pers(int level, char *clevel)
425 {
426 struct mdk_personality *pers;
427 list_for_each_entry(pers, &pers_list, list) {
428 if (level != LEVEL_NONE && pers->level == level)
429 return pers;
430 if (strcmp(pers->name, clevel)==0)
431 return pers;
432 }
433 return NULL;
434 }
435
436 /* return the offset of the super block in 512byte sectors */
437 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
438 {
439 sector_t num_sectors = bdev->bd_inode->i_size / 512;
440 return MD_NEW_SIZE_SECTORS(num_sectors);
441 }
442
443 static sector_t calc_num_sectors(mdk_rdev_t *rdev, unsigned chunk_size)
444 {
445 sector_t num_sectors = rdev->sb_start;
446
447 if (chunk_size)
448 num_sectors &= ~((sector_t)chunk_size/512 - 1);
449 return num_sectors;
450 }
451
452 static int alloc_disk_sb(mdk_rdev_t * rdev)
453 {
454 if (rdev->sb_page)
455 MD_BUG();
456
457 rdev->sb_page = alloc_page(GFP_KERNEL);
458 if (!rdev->sb_page) {
459 printk(KERN_ALERT "md: out of memory.\n");
460 return -ENOMEM;
461 }
462
463 return 0;
464 }
465
466 static void free_disk_sb(mdk_rdev_t * rdev)
467 {
468 if (rdev->sb_page) {
469 put_page(rdev->sb_page);
470 rdev->sb_loaded = 0;
471 rdev->sb_page = NULL;
472 rdev->sb_start = 0;
473 rdev->sectors = 0;
474 }
475 }
476
477
478 static void super_written(struct bio *bio, int error)
479 {
480 mdk_rdev_t *rdev = bio->bi_private;
481 mddev_t *mddev = rdev->mddev;
482
483 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
484 printk("md: super_written gets error=%d, uptodate=%d\n",
485 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
486 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
487 md_error(mddev, rdev);
488 }
489
490 if (atomic_dec_and_test(&mddev->pending_writes))
491 wake_up(&mddev->sb_wait);
492 bio_put(bio);
493 }
494
495 static void super_written_barrier(struct bio *bio, int error)
496 {
497 struct bio *bio2 = bio->bi_private;
498 mdk_rdev_t *rdev = bio2->bi_private;
499 mddev_t *mddev = rdev->mddev;
500
501 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
502 error == -EOPNOTSUPP) {
503 unsigned long flags;
504 /* barriers don't appear to be supported :-( */
505 set_bit(BarriersNotsupp, &rdev->flags);
506 mddev->barriers_work = 0;
507 spin_lock_irqsave(&mddev->write_lock, flags);
508 bio2->bi_next = mddev->biolist;
509 mddev->biolist = bio2;
510 spin_unlock_irqrestore(&mddev->write_lock, flags);
511 wake_up(&mddev->sb_wait);
512 bio_put(bio);
513 } else {
514 bio_put(bio2);
515 bio->bi_private = rdev;
516 super_written(bio, error);
517 }
518 }
519
520 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
521 sector_t sector, int size, struct page *page)
522 {
523 /* write first size bytes of page to sector of rdev
524 * Increment mddev->pending_writes before returning
525 * and decrement it on completion, waking up sb_wait
526 * if zero is reached.
527 * If an error occurred, call md_error
528 *
529 * As we might need to resubmit the request if BIO_RW_BARRIER
530 * causes ENOTSUPP, we allocate a spare bio...
531 */
532 struct bio *bio = bio_alloc(GFP_NOIO, 1);
533 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG);
534
535 bio->bi_bdev = rdev->bdev;
536 bio->bi_sector = sector;
537 bio_add_page(bio, page, size, 0);
538 bio->bi_private = rdev;
539 bio->bi_end_io = super_written;
540 bio->bi_rw = rw;
541
542 atomic_inc(&mddev->pending_writes);
543 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
544 struct bio *rbio;
545 rw |= (1<<BIO_RW_BARRIER);
546 rbio = bio_clone(bio, GFP_NOIO);
547 rbio->bi_private = bio;
548 rbio->bi_end_io = super_written_barrier;
549 submit_bio(rw, rbio);
550 } else
551 submit_bio(rw, bio);
552 }
553
554 void md_super_wait(mddev_t *mddev)
555 {
556 /* wait for all superblock writes that were scheduled to complete.
557 * if any had to be retried (due to BARRIER problems), retry them
558 */
559 DEFINE_WAIT(wq);
560 for(;;) {
561 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
562 if (atomic_read(&mddev->pending_writes)==0)
563 break;
564 while (mddev->biolist) {
565 struct bio *bio;
566 spin_lock_irq(&mddev->write_lock);
567 bio = mddev->biolist;
568 mddev->biolist = bio->bi_next ;
569 bio->bi_next = NULL;
570 spin_unlock_irq(&mddev->write_lock);
571 submit_bio(bio->bi_rw, bio);
572 }
573 schedule();
574 }
575 finish_wait(&mddev->sb_wait, &wq);
576 }
577
578 static void bi_complete(struct bio *bio, int error)
579 {
580 complete((struct completion*)bio->bi_private);
581 }
582
583 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
584 struct page *page, int rw)
585 {
586 struct bio *bio = bio_alloc(GFP_NOIO, 1);
587 struct completion event;
588 int ret;
589
590 rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
591
592 bio->bi_bdev = bdev;
593 bio->bi_sector = sector;
594 bio_add_page(bio, page, size, 0);
595 init_completion(&event);
596 bio->bi_private = &event;
597 bio->bi_end_io = bi_complete;
598 submit_bio(rw, bio);
599 wait_for_completion(&event);
600
601 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
602 bio_put(bio);
603 return ret;
604 }
605 EXPORT_SYMBOL_GPL(sync_page_io);
606
607 static int read_disk_sb(mdk_rdev_t * rdev, int size)
608 {
609 char b[BDEVNAME_SIZE];
610 if (!rdev->sb_page) {
611 MD_BUG();
612 return -EINVAL;
613 }
614 if (rdev->sb_loaded)
615 return 0;
616
617
618 if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
619 goto fail;
620 rdev->sb_loaded = 1;
621 return 0;
622
623 fail:
624 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
625 bdevname(rdev->bdev,b));
626 return -EINVAL;
627 }
628
629 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
630 {
631 return sb1->set_uuid0 == sb2->set_uuid0 &&
632 sb1->set_uuid1 == sb2->set_uuid1 &&
633 sb1->set_uuid2 == sb2->set_uuid2 &&
634 sb1->set_uuid3 == sb2->set_uuid3;
635 }
636
637 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
638 {
639 int ret;
640 mdp_super_t *tmp1, *tmp2;
641
642 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
643 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
644
645 if (!tmp1 || !tmp2) {
646 ret = 0;
647 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
648 goto abort;
649 }
650
651 *tmp1 = *sb1;
652 *tmp2 = *sb2;
653
654 /*
655 * nr_disks is not constant
656 */
657 tmp1->nr_disks = 0;
658 tmp2->nr_disks = 0;
659
660 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
661 abort:
662 kfree(tmp1);
663 kfree(tmp2);
664 return ret;
665 }
666
667
668 static u32 md_csum_fold(u32 csum)
669 {
670 csum = (csum & 0xffff) + (csum >> 16);
671 return (csum & 0xffff) + (csum >> 16);
672 }
673
674 static unsigned int calc_sb_csum(mdp_super_t * sb)
675 {
676 u64 newcsum = 0;
677 u32 *sb32 = (u32*)sb;
678 int i;
679 unsigned int disk_csum, csum;
680
681 disk_csum = sb->sb_csum;
682 sb->sb_csum = 0;
683
684 for (i = 0; i < MD_SB_BYTES/4 ; i++)
685 newcsum += sb32[i];
686 csum = (newcsum & 0xffffffff) + (newcsum>>32);
687
688
689 #ifdef CONFIG_ALPHA
690 /* This used to use csum_partial, which was wrong for several
691 * reasons including that different results are returned on
692 * different architectures. It isn't critical that we get exactly
693 * the same return value as before (we always csum_fold before
694 * testing, and that removes any differences). However as we
695 * know that csum_partial always returned a 16bit value on
696 * alphas, do a fold to maximise conformity to previous behaviour.
697 */
698 sb->sb_csum = md_csum_fold(disk_csum);
699 #else
700 sb->sb_csum = disk_csum;
701 #endif
702 return csum;
703 }
704
705
706 /*
707 * Handle superblock details.
708 * We want to be able to handle multiple superblock formats
709 * so we have a common interface to them all, and an array of
710 * different handlers.
711 * We rely on user-space to write the initial superblock, and support
712 * reading and updating of superblocks.
713 * Interface methods are:
714 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
715 * loads and validates a superblock on dev.
716 * if refdev != NULL, compare superblocks on both devices
717 * Return:
718 * 0 - dev has a superblock that is compatible with refdev
719 * 1 - dev has a superblock that is compatible and newer than refdev
720 * so dev should be used as the refdev in future
721 * -EINVAL superblock incompatible or invalid
722 * -othererror e.g. -EIO
723 *
724 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
725 * Verify that dev is acceptable into mddev.
726 * The first time, mddev->raid_disks will be 0, and data from
727 * dev should be merged in. Subsequent calls check that dev
728 * is new enough. Return 0 or -EINVAL
729 *
730 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
731 * Update the superblock for rdev with data in mddev
732 * This does not write to disc.
733 *
734 */
735
736 struct super_type {
737 char *name;
738 struct module *owner;
739 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
740 int minor_version);
741 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
742 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
743 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
744 sector_t num_sectors);
745 };
746
747 /*
748 * load_super for 0.90.0
749 */
750 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
751 {
752 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
753 mdp_super_t *sb;
754 int ret;
755
756 /*
757 * Calculate the position of the superblock (512byte sectors),
758 * it's at the end of the disk.
759 *
760 * It also happens to be a multiple of 4Kb.
761 */
762 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
763
764 ret = read_disk_sb(rdev, MD_SB_BYTES);
765 if (ret) return ret;
766
767 ret = -EINVAL;
768
769 bdevname(rdev->bdev, b);
770 sb = (mdp_super_t*)page_address(rdev->sb_page);
771
772 if (sb->md_magic != MD_SB_MAGIC) {
773 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
774 b);
775 goto abort;
776 }
777
778 if (sb->major_version != 0 ||
779 sb->minor_version < 90 ||
780 sb->minor_version > 91) {
781 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
782 sb->major_version, sb->minor_version,
783 b);
784 goto abort;
785 }
786
787 if (sb->raid_disks <= 0)
788 goto abort;
789
790 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
791 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
792 b);
793 goto abort;
794 }
795
796 rdev->preferred_minor = sb->md_minor;
797 rdev->data_offset = 0;
798 rdev->sb_size = MD_SB_BYTES;
799
800 if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
801 if (sb->level != 1 && sb->level != 4
802 && sb->level != 5 && sb->level != 6
803 && sb->level != 10) {
804 /* FIXME use a better test */
805 printk(KERN_WARNING
806 "md: bitmaps not supported for this level.\n");
807 goto abort;
808 }
809 }
810
811 if (sb->level == LEVEL_MULTIPATH)
812 rdev->desc_nr = -1;
813 else
814 rdev->desc_nr = sb->this_disk.number;
815
816 if (!refdev) {
817 ret = 1;
818 } else {
819 __u64 ev1, ev2;
820 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
821 if (!uuid_equal(refsb, sb)) {
822 printk(KERN_WARNING "md: %s has different UUID to %s\n",
823 b, bdevname(refdev->bdev,b2));
824 goto abort;
825 }
826 if (!sb_equal(refsb, sb)) {
827 printk(KERN_WARNING "md: %s has same UUID"
828 " but different superblock to %s\n",
829 b, bdevname(refdev->bdev, b2));
830 goto abort;
831 }
832 ev1 = md_event(sb);
833 ev2 = md_event(refsb);
834 if (ev1 > ev2)
835 ret = 1;
836 else
837 ret = 0;
838 }
839 rdev->sectors = calc_num_sectors(rdev, sb->chunk_size);
840
841 if (rdev->sectors < sb->size * 2 && sb->level > 1)
842 /* "this cannot possibly happen" ... */
843 ret = -EINVAL;
844
845 abort:
846 return ret;
847 }
848
849 /*
850 * validate_super for 0.90.0
851 */
852 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
853 {
854 mdp_disk_t *desc;
855 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
856 __u64 ev1 = md_event(sb);
857
858 rdev->raid_disk = -1;
859 clear_bit(Faulty, &rdev->flags);
860 clear_bit(In_sync, &rdev->flags);
861 clear_bit(WriteMostly, &rdev->flags);
862 clear_bit(BarriersNotsupp, &rdev->flags);
863
864 if (mddev->raid_disks == 0) {
865 mddev->major_version = 0;
866 mddev->minor_version = sb->minor_version;
867 mddev->patch_version = sb->patch_version;
868 mddev->external = 0;
869 mddev->chunk_size = sb->chunk_size;
870 mddev->ctime = sb->ctime;
871 mddev->utime = sb->utime;
872 mddev->level = sb->level;
873 mddev->clevel[0] = 0;
874 mddev->layout = sb->layout;
875 mddev->raid_disks = sb->raid_disks;
876 mddev->dev_sectors = sb->size * 2;
877 mddev->events = ev1;
878 mddev->bitmap_offset = 0;
879 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
880
881 if (mddev->minor_version >= 91) {
882 mddev->reshape_position = sb->reshape_position;
883 mddev->delta_disks = sb->delta_disks;
884 mddev->new_level = sb->new_level;
885 mddev->new_layout = sb->new_layout;
886 mddev->new_chunk = sb->new_chunk;
887 } else {
888 mddev->reshape_position = MaxSector;
889 mddev->delta_disks = 0;
890 mddev->new_level = mddev->level;
891 mddev->new_layout = mddev->layout;
892 mddev->new_chunk = mddev->chunk_size;
893 }
894
895 if (sb->state & (1<<MD_SB_CLEAN))
896 mddev->recovery_cp = MaxSector;
897 else {
898 if (sb->events_hi == sb->cp_events_hi &&
899 sb->events_lo == sb->cp_events_lo) {
900 mddev->recovery_cp = sb->recovery_cp;
901 } else
902 mddev->recovery_cp = 0;
903 }
904
905 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
906 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
907 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
908 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
909
910 mddev->max_disks = MD_SB_DISKS;
911
912 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
913 mddev->bitmap_file == NULL)
914 mddev->bitmap_offset = mddev->default_bitmap_offset;
915
916 } else if (mddev->pers == NULL) {
917 /* Insist on good event counter while assembling */
918 ++ev1;
919 if (ev1 < mddev->events)
920 return -EINVAL;
921 } else if (mddev->bitmap) {
922 /* if adding to array with a bitmap, then we can accept an
923 * older device ... but not too old.
924 */
925 if (ev1 < mddev->bitmap->events_cleared)
926 return 0;
927 } else {
928 if (ev1 < mddev->events)
929 /* just a hot-add of a new device, leave raid_disk at -1 */
930 return 0;
931 }
932
933 if (mddev->level != LEVEL_MULTIPATH) {
934 desc = sb->disks + rdev->desc_nr;
935
936 if (desc->state & (1<<MD_DISK_FAULTY))
937 set_bit(Faulty, &rdev->flags);
938 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
939 desc->raid_disk < mddev->raid_disks */) {
940 set_bit(In_sync, &rdev->flags);
941 rdev->raid_disk = desc->raid_disk;
942 }
943 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
944 set_bit(WriteMostly, &rdev->flags);
945 } else /* MULTIPATH are always insync */
946 set_bit(In_sync, &rdev->flags);
947 return 0;
948 }
949
950 /*
951 * sync_super for 0.90.0
952 */
953 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
954 {
955 mdp_super_t *sb;
956 mdk_rdev_t *rdev2;
957 int next_spare = mddev->raid_disks;
958
959
960 /* make rdev->sb match mddev data..
961 *
962 * 1/ zero out disks
963 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
964 * 3/ any empty disks < next_spare become removed
965 *
966 * disks[0] gets initialised to REMOVED because
967 * we cannot be sure from other fields if it has
968 * been initialised or not.
969 */
970 int i;
971 int active=0, working=0,failed=0,spare=0,nr_disks=0;
972
973 rdev->sb_size = MD_SB_BYTES;
974
975 sb = (mdp_super_t*)page_address(rdev->sb_page);
976
977 memset(sb, 0, sizeof(*sb));
978
979 sb->md_magic = MD_SB_MAGIC;
980 sb->major_version = mddev->major_version;
981 sb->patch_version = mddev->patch_version;
982 sb->gvalid_words = 0; /* ignored */
983 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
984 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
985 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
986 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
987
988 sb->ctime = mddev->ctime;
989 sb->level = mddev->level;
990 sb->size = mddev->dev_sectors / 2;
991 sb->raid_disks = mddev->raid_disks;
992 sb->md_minor = mddev->md_minor;
993 sb->not_persistent = 0;
994 sb->utime = mddev->utime;
995 sb->state = 0;
996 sb->events_hi = (mddev->events>>32);
997 sb->events_lo = (u32)mddev->events;
998
999 if (mddev->reshape_position == MaxSector)
1000 sb->minor_version = 90;
1001 else {
1002 sb->minor_version = 91;
1003 sb->reshape_position = mddev->reshape_position;
1004 sb->new_level = mddev->new_level;
1005 sb->delta_disks = mddev->delta_disks;
1006 sb->new_layout = mddev->new_layout;
1007 sb->new_chunk = mddev->new_chunk;
1008 }
1009 mddev->minor_version = sb->minor_version;
1010 if (mddev->in_sync)
1011 {
1012 sb->recovery_cp = mddev->recovery_cp;
1013 sb->cp_events_hi = (mddev->events>>32);
1014 sb->cp_events_lo = (u32)mddev->events;
1015 if (mddev->recovery_cp == MaxSector)
1016 sb->state = (1<< MD_SB_CLEAN);
1017 } else
1018 sb->recovery_cp = 0;
1019
1020 sb->layout = mddev->layout;
1021 sb->chunk_size = mddev->chunk_size;
1022
1023 if (mddev->bitmap && mddev->bitmap_file == NULL)
1024 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1025
1026 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1027 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1028 mdp_disk_t *d;
1029 int desc_nr;
1030 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
1031 && !test_bit(Faulty, &rdev2->flags))
1032 desc_nr = rdev2->raid_disk;
1033 else
1034 desc_nr = next_spare++;
1035 rdev2->desc_nr = desc_nr;
1036 d = &sb->disks[rdev2->desc_nr];
1037 nr_disks++;
1038 d->number = rdev2->desc_nr;
1039 d->major = MAJOR(rdev2->bdev->bd_dev);
1040 d->minor = MINOR(rdev2->bdev->bd_dev);
1041 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
1042 && !test_bit(Faulty, &rdev2->flags))
1043 d->raid_disk = rdev2->raid_disk;
1044 else
1045 d->raid_disk = rdev2->desc_nr; /* compatibility */
1046 if (test_bit(Faulty, &rdev2->flags))
1047 d->state = (1<<MD_DISK_FAULTY);
1048 else if (test_bit(In_sync, &rdev2->flags)) {
1049 d->state = (1<<MD_DISK_ACTIVE);
1050 d->state |= (1<<MD_DISK_SYNC);
1051 active++;
1052 working++;
1053 } else {
1054 d->state = 0;
1055 spare++;
1056 working++;
1057 }
1058 if (test_bit(WriteMostly, &rdev2->flags))
1059 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1060 }
1061 /* now set the "removed" and "faulty" bits on any missing devices */
1062 for (i=0 ; i < mddev->raid_disks ; i++) {
1063 mdp_disk_t *d = &sb->disks[i];
1064 if (d->state == 0 && d->number == 0) {
1065 d->number = i;
1066 d->raid_disk = i;
1067 d->state = (1<<MD_DISK_REMOVED);
1068 d->state |= (1<<MD_DISK_FAULTY);
1069 failed++;
1070 }
1071 }
1072 sb->nr_disks = nr_disks;
1073 sb->active_disks = active;
1074 sb->working_disks = working;
1075 sb->failed_disks = failed;
1076 sb->spare_disks = spare;
1077
1078 sb->this_disk = sb->disks[rdev->desc_nr];
1079 sb->sb_csum = calc_sb_csum(sb);
1080 }
1081
1082 /*
1083 * rdev_size_change for 0.90.0
1084 */
1085 static unsigned long long
1086 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1087 {
1088 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1089 return 0; /* component must fit device */
1090 if (rdev->mddev->bitmap_offset)
1091 return 0; /* can't move bitmap */
1092 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1093 if (!num_sectors || num_sectors > rdev->sb_start)
1094 num_sectors = rdev->sb_start;
1095 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1096 rdev->sb_page);
1097 md_super_wait(rdev->mddev);
1098 return num_sectors / 2; /* kB for sysfs */
1099 }
1100
1101
1102 /*
1103 * version 1 superblock
1104 */
1105
1106 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1107 {
1108 __le32 disk_csum;
1109 u32 csum;
1110 unsigned long long newcsum;
1111 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1112 __le32 *isuper = (__le32*)sb;
1113 int i;
1114
1115 disk_csum = sb->sb_csum;
1116 sb->sb_csum = 0;
1117 newcsum = 0;
1118 for (i=0; size>=4; size -= 4 )
1119 newcsum += le32_to_cpu(*isuper++);
1120
1121 if (size == 2)
1122 newcsum += le16_to_cpu(*(__le16*) isuper);
1123
1124 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1125 sb->sb_csum = disk_csum;
1126 return cpu_to_le32(csum);
1127 }
1128
1129 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1130 {
1131 struct mdp_superblock_1 *sb;
1132 int ret;
1133 sector_t sb_start;
1134 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1135 int bmask;
1136
1137 /*
1138 * Calculate the position of the superblock in 512byte sectors.
1139 * It is always aligned to a 4K boundary and
1140 * depeding on minor_version, it can be:
1141 * 0: At least 8K, but less than 12K, from end of device
1142 * 1: At start of device
1143 * 2: 4K from start of device.
1144 */
1145 switch(minor_version) {
1146 case 0:
1147 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1148 sb_start -= 8*2;
1149 sb_start &= ~(sector_t)(4*2-1);
1150 break;
1151 case 1:
1152 sb_start = 0;
1153 break;
1154 case 2:
1155 sb_start = 8;
1156 break;
1157 default:
1158 return -EINVAL;
1159 }
1160 rdev->sb_start = sb_start;
1161
1162 /* superblock is rarely larger than 1K, but it can be larger,
1163 * and it is safe to read 4k, so we do that
1164 */
1165 ret = read_disk_sb(rdev, 4096);
1166 if (ret) return ret;
1167
1168
1169 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1170
1171 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1172 sb->major_version != cpu_to_le32(1) ||
1173 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1174 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1175 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1176 return -EINVAL;
1177
1178 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1179 printk("md: invalid superblock checksum on %s\n",
1180 bdevname(rdev->bdev,b));
1181 return -EINVAL;
1182 }
1183 if (le64_to_cpu(sb->data_size) < 10) {
1184 printk("md: data_size too small on %s\n",
1185 bdevname(rdev->bdev,b));
1186 return -EINVAL;
1187 }
1188 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1189 if (sb->level != cpu_to_le32(1) &&
1190 sb->level != cpu_to_le32(4) &&
1191 sb->level != cpu_to_le32(5) &&
1192 sb->level != cpu_to_le32(6) &&
1193 sb->level != cpu_to_le32(10)) {
1194 printk(KERN_WARNING
1195 "md: bitmaps not supported for this level.\n");
1196 return -EINVAL;
1197 }
1198 }
1199
1200 rdev->preferred_minor = 0xffff;
1201 rdev->data_offset = le64_to_cpu(sb->data_offset);
1202 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1203
1204 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1205 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1206 if (rdev->sb_size & bmask)
1207 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1208
1209 if (minor_version
1210 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1211 return -EINVAL;
1212
1213 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1214 rdev->desc_nr = -1;
1215 else
1216 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1217
1218 if (!refdev) {
1219 ret = 1;
1220 } else {
1221 __u64 ev1, ev2;
1222 struct mdp_superblock_1 *refsb =
1223 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1224
1225 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1226 sb->level != refsb->level ||
1227 sb->layout != refsb->layout ||
1228 sb->chunksize != refsb->chunksize) {
1229 printk(KERN_WARNING "md: %s has strangely different"
1230 " superblock to %s\n",
1231 bdevname(rdev->bdev,b),
1232 bdevname(refdev->bdev,b2));
1233 return -EINVAL;
1234 }
1235 ev1 = le64_to_cpu(sb->events);
1236 ev2 = le64_to_cpu(refsb->events);
1237
1238 if (ev1 > ev2)
1239 ret = 1;
1240 else
1241 ret = 0;
1242 }
1243 if (minor_version)
1244 rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
1245 le64_to_cpu(sb->data_offset);
1246 else
1247 rdev->sectors = rdev->sb_start;
1248 if (rdev->sectors < le64_to_cpu(sb->data_size))
1249 return -EINVAL;
1250 rdev->sectors = le64_to_cpu(sb->data_size);
1251 if (le32_to_cpu(sb->chunksize))
1252 rdev->sectors &= ~((sector_t)le32_to_cpu(sb->chunksize) - 1);
1253
1254 if (le64_to_cpu(sb->size) > rdev->sectors)
1255 return -EINVAL;
1256 return ret;
1257 }
1258
1259 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1260 {
1261 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1262 __u64 ev1 = le64_to_cpu(sb->events);
1263
1264 rdev->raid_disk = -1;
1265 clear_bit(Faulty, &rdev->flags);
1266 clear_bit(In_sync, &rdev->flags);
1267 clear_bit(WriteMostly, &rdev->flags);
1268 clear_bit(BarriersNotsupp, &rdev->flags);
1269
1270 if (mddev->raid_disks == 0) {
1271 mddev->major_version = 1;
1272 mddev->patch_version = 0;
1273 mddev->external = 0;
1274 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1275 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1276 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1277 mddev->level = le32_to_cpu(sb->level);
1278 mddev->clevel[0] = 0;
1279 mddev->layout = le32_to_cpu(sb->layout);
1280 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1281 mddev->dev_sectors = le64_to_cpu(sb->size);
1282 mddev->events = ev1;
1283 mddev->bitmap_offset = 0;
1284 mddev->default_bitmap_offset = 1024 >> 9;
1285
1286 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1287 memcpy(mddev->uuid, sb->set_uuid, 16);
1288
1289 mddev->max_disks = (4096-256)/2;
1290
1291 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1292 mddev->bitmap_file == NULL )
1293 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1294
1295 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1296 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1297 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1298 mddev->new_level = le32_to_cpu(sb->new_level);
1299 mddev->new_layout = le32_to_cpu(sb->new_layout);
1300 mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1301 } else {
1302 mddev->reshape_position = MaxSector;
1303 mddev->delta_disks = 0;
1304 mddev->new_level = mddev->level;
1305 mddev->new_layout = mddev->layout;
1306 mddev->new_chunk = mddev->chunk_size;
1307 }
1308
1309 } else if (mddev->pers == NULL) {
1310 /* Insist of good event counter while assembling */
1311 ++ev1;
1312 if (ev1 < mddev->events)
1313 return -EINVAL;
1314 } else if (mddev->bitmap) {
1315 /* If adding to array with a bitmap, then we can accept an
1316 * older device, but not too old.
1317 */
1318 if (ev1 < mddev->bitmap->events_cleared)
1319 return 0;
1320 } else {
1321 if (ev1 < mddev->events)
1322 /* just a hot-add of a new device, leave raid_disk at -1 */
1323 return 0;
1324 }
1325 if (mddev->level != LEVEL_MULTIPATH) {
1326 int role;
1327 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1328 switch(role) {
1329 case 0xffff: /* spare */
1330 break;
1331 case 0xfffe: /* faulty */
1332 set_bit(Faulty, &rdev->flags);
1333 break;
1334 default:
1335 if ((le32_to_cpu(sb->feature_map) &
1336 MD_FEATURE_RECOVERY_OFFSET))
1337 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1338 else
1339 set_bit(In_sync, &rdev->flags);
1340 rdev->raid_disk = role;
1341 break;
1342 }
1343 if (sb->devflags & WriteMostly1)
1344 set_bit(WriteMostly, &rdev->flags);
1345 } else /* MULTIPATH are always insync */
1346 set_bit(In_sync, &rdev->flags);
1347
1348 return 0;
1349 }
1350
1351 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1352 {
1353 struct mdp_superblock_1 *sb;
1354 mdk_rdev_t *rdev2;
1355 int max_dev, i;
1356 /* make rdev->sb match mddev and rdev data. */
1357
1358 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1359
1360 sb->feature_map = 0;
1361 sb->pad0 = 0;
1362 sb->recovery_offset = cpu_to_le64(0);
1363 memset(sb->pad1, 0, sizeof(sb->pad1));
1364 memset(sb->pad2, 0, sizeof(sb->pad2));
1365 memset(sb->pad3, 0, sizeof(sb->pad3));
1366
1367 sb->utime = cpu_to_le64((__u64)mddev->utime);
1368 sb->events = cpu_to_le64(mddev->events);
1369 if (mddev->in_sync)
1370 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1371 else
1372 sb->resync_offset = cpu_to_le64(0);
1373
1374 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1375
1376 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1377 sb->size = cpu_to_le64(mddev->dev_sectors);
1378 sb->chunksize = cpu_to_le32(mddev->chunk_size >> 9);
1379 sb->level = cpu_to_le32(mddev->level);
1380 sb->layout = cpu_to_le32(mddev->layout);
1381
1382 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1383 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1384 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1385 }
1386
1387 if (rdev->raid_disk >= 0 &&
1388 !test_bit(In_sync, &rdev->flags)) {
1389 if (mddev->curr_resync_completed > rdev->recovery_offset)
1390 rdev->recovery_offset = mddev->curr_resync_completed;
1391 if (rdev->recovery_offset > 0) {
1392 sb->feature_map |=
1393 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1394 sb->recovery_offset =
1395 cpu_to_le64(rdev->recovery_offset);
1396 }
1397 }
1398
1399 if (mddev->reshape_position != MaxSector) {
1400 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1401 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1402 sb->new_layout = cpu_to_le32(mddev->new_layout);
1403 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1404 sb->new_level = cpu_to_le32(mddev->new_level);
1405 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1406 }
1407
1408 max_dev = 0;
1409 list_for_each_entry(rdev2, &mddev->disks, same_set)
1410 if (rdev2->desc_nr+1 > max_dev)
1411 max_dev = rdev2->desc_nr+1;
1412
1413 if (max_dev > le32_to_cpu(sb->max_dev))
1414 sb->max_dev = cpu_to_le32(max_dev);
1415 for (i=0; i<max_dev;i++)
1416 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1417
1418 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1419 i = rdev2->desc_nr;
1420 if (test_bit(Faulty, &rdev2->flags))
1421 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1422 else if (test_bit(In_sync, &rdev2->flags))
1423 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1424 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1425 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1426 else
1427 sb->dev_roles[i] = cpu_to_le16(0xffff);
1428 }
1429
1430 sb->sb_csum = calc_sb_1_csum(sb);
1431 }
1432
1433 static unsigned long long
1434 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1435 {
1436 struct mdp_superblock_1 *sb;
1437 sector_t max_sectors;
1438 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1439 return 0; /* component must fit device */
1440 if (rdev->sb_start < rdev->data_offset) {
1441 /* minor versions 1 and 2; superblock before data */
1442 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1443 max_sectors -= rdev->data_offset;
1444 if (!num_sectors || num_sectors > max_sectors)
1445 num_sectors = max_sectors;
1446 } else if (rdev->mddev->bitmap_offset) {
1447 /* minor version 0 with bitmap we can't move */
1448 return 0;
1449 } else {
1450 /* minor version 0; superblock after data */
1451 sector_t sb_start;
1452 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1453 sb_start &= ~(sector_t)(4*2 - 1);
1454 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1455 if (!num_sectors || num_sectors > max_sectors)
1456 num_sectors = max_sectors;
1457 rdev->sb_start = sb_start;
1458 }
1459 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1460 sb->data_size = cpu_to_le64(num_sectors);
1461 sb->super_offset = rdev->sb_start;
1462 sb->sb_csum = calc_sb_1_csum(sb);
1463 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1464 rdev->sb_page);
1465 md_super_wait(rdev->mddev);
1466 return num_sectors / 2; /* kB for sysfs */
1467 }
1468
1469 static struct super_type super_types[] = {
1470 [0] = {
1471 .name = "0.90.0",
1472 .owner = THIS_MODULE,
1473 .load_super = super_90_load,
1474 .validate_super = super_90_validate,
1475 .sync_super = super_90_sync,
1476 .rdev_size_change = super_90_rdev_size_change,
1477 },
1478 [1] = {
1479 .name = "md-1",
1480 .owner = THIS_MODULE,
1481 .load_super = super_1_load,
1482 .validate_super = super_1_validate,
1483 .sync_super = super_1_sync,
1484 .rdev_size_change = super_1_rdev_size_change,
1485 },
1486 };
1487
1488 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1489 {
1490 mdk_rdev_t *rdev, *rdev2;
1491
1492 rcu_read_lock();
1493 rdev_for_each_rcu(rdev, mddev1)
1494 rdev_for_each_rcu(rdev2, mddev2)
1495 if (rdev->bdev->bd_contains ==
1496 rdev2->bdev->bd_contains) {
1497 rcu_read_unlock();
1498 return 1;
1499 }
1500 rcu_read_unlock();
1501 return 0;
1502 }
1503
1504 static LIST_HEAD(pending_raid_disks);
1505
1506 static void md_integrity_check(mdk_rdev_t *rdev, mddev_t *mddev)
1507 {
1508 struct mdk_personality *pers = mddev->pers;
1509 struct gendisk *disk = mddev->gendisk;
1510 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1511 struct blk_integrity *bi_mddev = blk_get_integrity(disk);
1512
1513 /* Data integrity passthrough not supported on RAID 4, 5 and 6 */
1514 if (pers && pers->level >= 4 && pers->level <= 6)
1515 return;
1516
1517 /* If rdev is integrity capable, register profile for mddev */
1518 if (!bi_mddev && bi_rdev) {
1519 if (blk_integrity_register(disk, bi_rdev))
1520 printk(KERN_ERR "%s: %s Could not register integrity!\n",
1521 __func__, disk->disk_name);
1522 else
1523 printk(KERN_NOTICE "Enabling data integrity on %s\n",
1524 disk->disk_name);
1525 return;
1526 }
1527
1528 /* Check that mddev and rdev have matching profiles */
1529 if (blk_integrity_compare(disk, rdev->bdev->bd_disk) < 0) {
1530 printk(KERN_ERR "%s: %s/%s integrity mismatch!\n", __func__,
1531 disk->disk_name, rdev->bdev->bd_disk->disk_name);
1532 printk(KERN_NOTICE "Disabling data integrity on %s\n",
1533 disk->disk_name);
1534 blk_integrity_unregister(disk);
1535 }
1536 }
1537
1538 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1539 {
1540 char b[BDEVNAME_SIZE];
1541 struct kobject *ko;
1542 char *s;
1543 int err;
1544
1545 if (rdev->mddev) {
1546 MD_BUG();
1547 return -EINVAL;
1548 }
1549
1550 /* prevent duplicates */
1551 if (find_rdev(mddev, rdev->bdev->bd_dev))
1552 return -EEXIST;
1553
1554 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1555 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1556 rdev->sectors < mddev->dev_sectors)) {
1557 if (mddev->pers) {
1558 /* Cannot change size, so fail
1559 * If mddev->level <= 0, then we don't care
1560 * about aligning sizes (e.g. linear)
1561 */
1562 if (mddev->level > 0)
1563 return -ENOSPC;
1564 } else
1565 mddev->dev_sectors = rdev->sectors;
1566 }
1567
1568 /* Verify rdev->desc_nr is unique.
1569 * If it is -1, assign a free number, else
1570 * check number is not in use
1571 */
1572 if (rdev->desc_nr < 0) {
1573 int choice = 0;
1574 if (mddev->pers) choice = mddev->raid_disks;
1575 while (find_rdev_nr(mddev, choice))
1576 choice++;
1577 rdev->desc_nr = choice;
1578 } else {
1579 if (find_rdev_nr(mddev, rdev->desc_nr))
1580 return -EBUSY;
1581 }
1582 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1583 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1584 mdname(mddev), mddev->max_disks);
1585 return -EBUSY;
1586 }
1587 bdevname(rdev->bdev,b);
1588 while ( (s=strchr(b, '/')) != NULL)
1589 *s = '!';
1590
1591 rdev->mddev = mddev;
1592 printk(KERN_INFO "md: bind<%s>\n", b);
1593
1594 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1595 goto fail;
1596
1597 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1598 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1599 kobject_del(&rdev->kobj);
1600 goto fail;
1601 }
1602 rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, "state");
1603
1604 list_add_rcu(&rdev->same_set, &mddev->disks);
1605 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1606
1607 /* May as well allow recovery to be retried once */
1608 mddev->recovery_disabled = 0;
1609
1610 md_integrity_check(rdev, mddev);
1611 return 0;
1612
1613 fail:
1614 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1615 b, mdname(mddev));
1616 return err;
1617 }
1618
1619 static void md_delayed_delete(struct work_struct *ws)
1620 {
1621 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1622 kobject_del(&rdev->kobj);
1623 kobject_put(&rdev->kobj);
1624 }
1625
1626 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1627 {
1628 char b[BDEVNAME_SIZE];
1629 if (!rdev->mddev) {
1630 MD_BUG();
1631 return;
1632 }
1633 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1634 list_del_rcu(&rdev->same_set);
1635 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1636 rdev->mddev = NULL;
1637 sysfs_remove_link(&rdev->kobj, "block");
1638 sysfs_put(rdev->sysfs_state);
1639 rdev->sysfs_state = NULL;
1640 /* We need to delay this, otherwise we can deadlock when
1641 * writing to 'remove' to "dev/state". We also need
1642 * to delay it due to rcu usage.
1643 */
1644 synchronize_rcu();
1645 INIT_WORK(&rdev->del_work, md_delayed_delete);
1646 kobject_get(&rdev->kobj);
1647 schedule_work(&rdev->del_work);
1648 }
1649
1650 /*
1651 * prevent the device from being mounted, repartitioned or
1652 * otherwise reused by a RAID array (or any other kernel
1653 * subsystem), by bd_claiming the device.
1654 */
1655 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1656 {
1657 int err = 0;
1658 struct block_device *bdev;
1659 char b[BDEVNAME_SIZE];
1660
1661 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1662 if (IS_ERR(bdev)) {
1663 printk(KERN_ERR "md: could not open %s.\n",
1664 __bdevname(dev, b));
1665 return PTR_ERR(bdev);
1666 }
1667 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1668 if (err) {
1669 printk(KERN_ERR "md: could not bd_claim %s.\n",
1670 bdevname(bdev, b));
1671 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1672 return err;
1673 }
1674 if (!shared)
1675 set_bit(AllReserved, &rdev->flags);
1676 rdev->bdev = bdev;
1677 return err;
1678 }
1679
1680 static void unlock_rdev(mdk_rdev_t *rdev)
1681 {
1682 struct block_device *bdev = rdev->bdev;
1683 rdev->bdev = NULL;
1684 if (!bdev)
1685 MD_BUG();
1686 bd_release(bdev);
1687 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1688 }
1689
1690 void md_autodetect_dev(dev_t dev);
1691
1692 static void export_rdev(mdk_rdev_t * rdev)
1693 {
1694 char b[BDEVNAME_SIZE];
1695 printk(KERN_INFO "md: export_rdev(%s)\n",
1696 bdevname(rdev->bdev,b));
1697 if (rdev->mddev)
1698 MD_BUG();
1699 free_disk_sb(rdev);
1700 #ifndef MODULE
1701 if (test_bit(AutoDetected, &rdev->flags))
1702 md_autodetect_dev(rdev->bdev->bd_dev);
1703 #endif
1704 unlock_rdev(rdev);
1705 kobject_put(&rdev->kobj);
1706 }
1707
1708 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1709 {
1710 unbind_rdev_from_array(rdev);
1711 export_rdev(rdev);
1712 }
1713
1714 static void export_array(mddev_t *mddev)
1715 {
1716 mdk_rdev_t *rdev, *tmp;
1717
1718 rdev_for_each(rdev, tmp, mddev) {
1719 if (!rdev->mddev) {
1720 MD_BUG();
1721 continue;
1722 }
1723 kick_rdev_from_array(rdev);
1724 }
1725 if (!list_empty(&mddev->disks))
1726 MD_BUG();
1727 mddev->raid_disks = 0;
1728 mddev->major_version = 0;
1729 }
1730
1731 static void print_desc(mdp_disk_t *desc)
1732 {
1733 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1734 desc->major,desc->minor,desc->raid_disk,desc->state);
1735 }
1736
1737 static void print_sb_90(mdp_super_t *sb)
1738 {
1739 int i;
1740
1741 printk(KERN_INFO
1742 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1743 sb->major_version, sb->minor_version, sb->patch_version,
1744 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1745 sb->ctime);
1746 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1747 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1748 sb->md_minor, sb->layout, sb->chunk_size);
1749 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1750 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1751 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1752 sb->failed_disks, sb->spare_disks,
1753 sb->sb_csum, (unsigned long)sb->events_lo);
1754
1755 printk(KERN_INFO);
1756 for (i = 0; i < MD_SB_DISKS; i++) {
1757 mdp_disk_t *desc;
1758
1759 desc = sb->disks + i;
1760 if (desc->number || desc->major || desc->minor ||
1761 desc->raid_disk || (desc->state && (desc->state != 4))) {
1762 printk(" D %2d: ", i);
1763 print_desc(desc);
1764 }
1765 }
1766 printk(KERN_INFO "md: THIS: ");
1767 print_desc(&sb->this_disk);
1768 }
1769
1770 static void print_sb_1(struct mdp_superblock_1 *sb)
1771 {
1772 __u8 *uuid;
1773
1774 uuid = sb->set_uuid;
1775 printk(KERN_INFO "md: SB: (V:%u) (F:0x%08x) Array-ID:<%02x%02x%02x%02x"
1776 ":%02x%02x:%02x%02x:%02x%02x:%02x%02x%02x%02x%02x%02x>\n"
1777 KERN_INFO "md: Name: \"%s\" CT:%llu\n",
1778 le32_to_cpu(sb->major_version),
1779 le32_to_cpu(sb->feature_map),
1780 uuid[0], uuid[1], uuid[2], uuid[3],
1781 uuid[4], uuid[5], uuid[6], uuid[7],
1782 uuid[8], uuid[9], uuid[10], uuid[11],
1783 uuid[12], uuid[13], uuid[14], uuid[15],
1784 sb->set_name,
1785 (unsigned long long)le64_to_cpu(sb->ctime)
1786 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
1787
1788 uuid = sb->device_uuid;
1789 printk(KERN_INFO "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
1790 " RO:%llu\n"
1791 KERN_INFO "md: Dev:%08x UUID: %02x%02x%02x%02x:%02x%02x:%02x%02x:%02x%02x"
1792 ":%02x%02x%02x%02x%02x%02x\n"
1793 KERN_INFO "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
1794 KERN_INFO "md: (MaxDev:%u) \n",
1795 le32_to_cpu(sb->level),
1796 (unsigned long long)le64_to_cpu(sb->size),
1797 le32_to_cpu(sb->raid_disks),
1798 le32_to_cpu(sb->layout),
1799 le32_to_cpu(sb->chunksize),
1800 (unsigned long long)le64_to_cpu(sb->data_offset),
1801 (unsigned long long)le64_to_cpu(sb->data_size),
1802 (unsigned long long)le64_to_cpu(sb->super_offset),
1803 (unsigned long long)le64_to_cpu(sb->recovery_offset),
1804 le32_to_cpu(sb->dev_number),
1805 uuid[0], uuid[1], uuid[2], uuid[3],
1806 uuid[4], uuid[5], uuid[6], uuid[7],
1807 uuid[8], uuid[9], uuid[10], uuid[11],
1808 uuid[12], uuid[13], uuid[14], uuid[15],
1809 sb->devflags,
1810 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
1811 (unsigned long long)le64_to_cpu(sb->events),
1812 (unsigned long long)le64_to_cpu(sb->resync_offset),
1813 le32_to_cpu(sb->sb_csum),
1814 le32_to_cpu(sb->max_dev)
1815 );
1816 }
1817
1818 static void print_rdev(mdk_rdev_t *rdev, int major_version)
1819 {
1820 char b[BDEVNAME_SIZE];
1821 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
1822 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
1823 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1824 rdev->desc_nr);
1825 if (rdev->sb_loaded) {
1826 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
1827 switch (major_version) {
1828 case 0:
1829 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
1830 break;
1831 case 1:
1832 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
1833 break;
1834 }
1835 } else
1836 printk(KERN_INFO "md: no rdev superblock!\n");
1837 }
1838
1839 static void md_print_devices(void)
1840 {
1841 struct list_head *tmp;
1842 mdk_rdev_t *rdev;
1843 mddev_t *mddev;
1844 char b[BDEVNAME_SIZE];
1845
1846 printk("\n");
1847 printk("md: **********************************\n");
1848 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1849 printk("md: **********************************\n");
1850 for_each_mddev(mddev, tmp) {
1851
1852 if (mddev->bitmap)
1853 bitmap_print_sb(mddev->bitmap);
1854 else
1855 printk("%s: ", mdname(mddev));
1856 list_for_each_entry(rdev, &mddev->disks, same_set)
1857 printk("<%s>", bdevname(rdev->bdev,b));
1858 printk("\n");
1859
1860 list_for_each_entry(rdev, &mddev->disks, same_set)
1861 print_rdev(rdev, mddev->major_version);
1862 }
1863 printk("md: **********************************\n");
1864 printk("\n");
1865 }
1866
1867
1868 static void sync_sbs(mddev_t * mddev, int nospares)
1869 {
1870 /* Update each superblock (in-memory image), but
1871 * if we are allowed to, skip spares which already
1872 * have the right event counter, or have one earlier
1873 * (which would mean they aren't being marked as dirty
1874 * with the rest of the array)
1875 */
1876 mdk_rdev_t *rdev;
1877
1878 list_for_each_entry(rdev, &mddev->disks, same_set) {
1879 if (rdev->sb_events == mddev->events ||
1880 (nospares &&
1881 rdev->raid_disk < 0 &&
1882 (rdev->sb_events&1)==0 &&
1883 rdev->sb_events+1 == mddev->events)) {
1884 /* Don't update this superblock */
1885 rdev->sb_loaded = 2;
1886 } else {
1887 super_types[mddev->major_version].
1888 sync_super(mddev, rdev);
1889 rdev->sb_loaded = 1;
1890 }
1891 }
1892 }
1893
1894 static void md_update_sb(mddev_t * mddev, int force_change)
1895 {
1896 mdk_rdev_t *rdev;
1897 int sync_req;
1898 int nospares = 0;
1899
1900 if (mddev->external)
1901 return;
1902 repeat:
1903 spin_lock_irq(&mddev->write_lock);
1904
1905 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1906 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1907 force_change = 1;
1908 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1909 /* just a clean<-> dirty transition, possibly leave spares alone,
1910 * though if events isn't the right even/odd, we will have to do
1911 * spares after all
1912 */
1913 nospares = 1;
1914 if (force_change)
1915 nospares = 0;
1916 if (mddev->degraded)
1917 /* If the array is degraded, then skipping spares is both
1918 * dangerous and fairly pointless.
1919 * Dangerous because a device that was removed from the array
1920 * might have a event_count that still looks up-to-date,
1921 * so it can be re-added without a resync.
1922 * Pointless because if there are any spares to skip,
1923 * then a recovery will happen and soon that array won't
1924 * be degraded any more and the spare can go back to sleep then.
1925 */
1926 nospares = 0;
1927
1928 sync_req = mddev->in_sync;
1929 mddev->utime = get_seconds();
1930
1931 /* If this is just a dirty<->clean transition, and the array is clean
1932 * and 'events' is odd, we can roll back to the previous clean state */
1933 if (nospares
1934 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1935 && (mddev->events & 1)
1936 && mddev->events != 1)
1937 mddev->events--;
1938 else {
1939 /* otherwise we have to go forward and ... */
1940 mddev->events ++;
1941 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1942 /* .. if the array isn't clean, insist on an odd 'events' */
1943 if ((mddev->events&1)==0) {
1944 mddev->events++;
1945 nospares = 0;
1946 }
1947 } else {
1948 /* otherwise insist on an even 'events' (for clean states) */
1949 if ((mddev->events&1)) {
1950 mddev->events++;
1951 nospares = 0;
1952 }
1953 }
1954 }
1955
1956 if (!mddev->events) {
1957 /*
1958 * oops, this 64-bit counter should never wrap.
1959 * Either we are in around ~1 trillion A.C., assuming
1960 * 1 reboot per second, or we have a bug:
1961 */
1962 MD_BUG();
1963 mddev->events --;
1964 }
1965
1966 /*
1967 * do not write anything to disk if using
1968 * nonpersistent superblocks
1969 */
1970 if (!mddev->persistent) {
1971 if (!mddev->external)
1972 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1973
1974 spin_unlock_irq(&mddev->write_lock);
1975 wake_up(&mddev->sb_wait);
1976 return;
1977 }
1978 sync_sbs(mddev, nospares);
1979 spin_unlock_irq(&mddev->write_lock);
1980
1981 dprintk(KERN_INFO
1982 "md: updating %s RAID superblock on device (in sync %d)\n",
1983 mdname(mddev),mddev->in_sync);
1984
1985 bitmap_update_sb(mddev->bitmap);
1986 list_for_each_entry(rdev, &mddev->disks, same_set) {
1987 char b[BDEVNAME_SIZE];
1988 dprintk(KERN_INFO "md: ");
1989 if (rdev->sb_loaded != 1)
1990 continue; /* no noise on spare devices */
1991 if (test_bit(Faulty, &rdev->flags))
1992 dprintk("(skipping faulty ");
1993
1994 dprintk("%s ", bdevname(rdev->bdev,b));
1995 if (!test_bit(Faulty, &rdev->flags)) {
1996 md_super_write(mddev,rdev,
1997 rdev->sb_start, rdev->sb_size,
1998 rdev->sb_page);
1999 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2000 bdevname(rdev->bdev,b),
2001 (unsigned long long)rdev->sb_start);
2002 rdev->sb_events = mddev->events;
2003
2004 } else
2005 dprintk(")\n");
2006 if (mddev->level == LEVEL_MULTIPATH)
2007 /* only need to write one superblock... */
2008 break;
2009 }
2010 md_super_wait(mddev);
2011 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2012
2013 spin_lock_irq(&mddev->write_lock);
2014 if (mddev->in_sync != sync_req ||
2015 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2016 /* have to write it out again */
2017 spin_unlock_irq(&mddev->write_lock);
2018 goto repeat;
2019 }
2020 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2021 spin_unlock_irq(&mddev->write_lock);
2022 wake_up(&mddev->sb_wait);
2023 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2024 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2025
2026 }
2027
2028 /* words written to sysfs files may, or may not, be \n terminated.
2029 * We want to accept with case. For this we use cmd_match.
2030 */
2031 static int cmd_match(const char *cmd, const char *str)
2032 {
2033 /* See if cmd, written into a sysfs file, matches
2034 * str. They must either be the same, or cmd can
2035 * have a trailing newline
2036 */
2037 while (*cmd && *str && *cmd == *str) {
2038 cmd++;
2039 str++;
2040 }
2041 if (*cmd == '\n')
2042 cmd++;
2043 if (*str || *cmd)
2044 return 0;
2045 return 1;
2046 }
2047
2048 struct rdev_sysfs_entry {
2049 struct attribute attr;
2050 ssize_t (*show)(mdk_rdev_t *, char *);
2051 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2052 };
2053
2054 static ssize_t
2055 state_show(mdk_rdev_t *rdev, char *page)
2056 {
2057 char *sep = "";
2058 size_t len = 0;
2059
2060 if (test_bit(Faulty, &rdev->flags)) {
2061 len+= sprintf(page+len, "%sfaulty",sep);
2062 sep = ",";
2063 }
2064 if (test_bit(In_sync, &rdev->flags)) {
2065 len += sprintf(page+len, "%sin_sync",sep);
2066 sep = ",";
2067 }
2068 if (test_bit(WriteMostly, &rdev->flags)) {
2069 len += sprintf(page+len, "%swrite_mostly",sep);
2070 sep = ",";
2071 }
2072 if (test_bit(Blocked, &rdev->flags)) {
2073 len += sprintf(page+len, "%sblocked", sep);
2074 sep = ",";
2075 }
2076 if (!test_bit(Faulty, &rdev->flags) &&
2077 !test_bit(In_sync, &rdev->flags)) {
2078 len += sprintf(page+len, "%sspare", sep);
2079 sep = ",";
2080 }
2081 return len+sprintf(page+len, "\n");
2082 }
2083
2084 static ssize_t
2085 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2086 {
2087 /* can write
2088 * faulty - simulates and error
2089 * remove - disconnects the device
2090 * writemostly - sets write_mostly
2091 * -writemostly - clears write_mostly
2092 * blocked - sets the Blocked flag
2093 * -blocked - clears the Blocked flag
2094 * insync - sets Insync providing device isn't active
2095 */
2096 int err = -EINVAL;
2097 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2098 md_error(rdev->mddev, rdev);
2099 err = 0;
2100 } else if (cmd_match(buf, "remove")) {
2101 if (rdev->raid_disk >= 0)
2102 err = -EBUSY;
2103 else {
2104 mddev_t *mddev = rdev->mddev;
2105 kick_rdev_from_array(rdev);
2106 if (mddev->pers)
2107 md_update_sb(mddev, 1);
2108 md_new_event(mddev);
2109 err = 0;
2110 }
2111 } else if (cmd_match(buf, "writemostly")) {
2112 set_bit(WriteMostly, &rdev->flags);
2113 err = 0;
2114 } else if (cmd_match(buf, "-writemostly")) {
2115 clear_bit(WriteMostly, &rdev->flags);
2116 err = 0;
2117 } else if (cmd_match(buf, "blocked")) {
2118 set_bit(Blocked, &rdev->flags);
2119 err = 0;
2120 } else if (cmd_match(buf, "-blocked")) {
2121 clear_bit(Blocked, &rdev->flags);
2122 wake_up(&rdev->blocked_wait);
2123 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2124 md_wakeup_thread(rdev->mddev->thread);
2125
2126 err = 0;
2127 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2128 set_bit(In_sync, &rdev->flags);
2129 err = 0;
2130 }
2131 if (!err && rdev->sysfs_state)
2132 sysfs_notify_dirent(rdev->sysfs_state);
2133 return err ? err : len;
2134 }
2135 static struct rdev_sysfs_entry rdev_state =
2136 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2137
2138 static ssize_t
2139 errors_show(mdk_rdev_t *rdev, char *page)
2140 {
2141 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2142 }
2143
2144 static ssize_t
2145 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2146 {
2147 char *e;
2148 unsigned long n = simple_strtoul(buf, &e, 10);
2149 if (*buf && (*e == 0 || *e == '\n')) {
2150 atomic_set(&rdev->corrected_errors, n);
2151 return len;
2152 }
2153 return -EINVAL;
2154 }
2155 static struct rdev_sysfs_entry rdev_errors =
2156 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2157
2158 static ssize_t
2159 slot_show(mdk_rdev_t *rdev, char *page)
2160 {
2161 if (rdev->raid_disk < 0)
2162 return sprintf(page, "none\n");
2163 else
2164 return sprintf(page, "%d\n", rdev->raid_disk);
2165 }
2166
2167 static ssize_t
2168 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2169 {
2170 char *e;
2171 int err;
2172 char nm[20];
2173 int slot = simple_strtoul(buf, &e, 10);
2174 if (strncmp(buf, "none", 4)==0)
2175 slot = -1;
2176 else if (e==buf || (*e && *e!= '\n'))
2177 return -EINVAL;
2178 if (rdev->mddev->pers && slot == -1) {
2179 /* Setting 'slot' on an active array requires also
2180 * updating the 'rd%d' link, and communicating
2181 * with the personality with ->hot_*_disk.
2182 * For now we only support removing
2183 * failed/spare devices. This normally happens automatically,
2184 * but not when the metadata is externally managed.
2185 */
2186 if (rdev->raid_disk == -1)
2187 return -EEXIST;
2188 /* personality does all needed checks */
2189 if (rdev->mddev->pers->hot_add_disk == NULL)
2190 return -EINVAL;
2191 err = rdev->mddev->pers->
2192 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2193 if (err)
2194 return err;
2195 sprintf(nm, "rd%d", rdev->raid_disk);
2196 sysfs_remove_link(&rdev->mddev->kobj, nm);
2197 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2198 md_wakeup_thread(rdev->mddev->thread);
2199 } else if (rdev->mddev->pers) {
2200 mdk_rdev_t *rdev2;
2201 /* Activating a spare .. or possibly reactivating
2202 * if we ever get bitmaps working here.
2203 */
2204
2205 if (rdev->raid_disk != -1)
2206 return -EBUSY;
2207
2208 if (rdev->mddev->pers->hot_add_disk == NULL)
2209 return -EINVAL;
2210
2211 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2212 if (rdev2->raid_disk == slot)
2213 return -EEXIST;
2214
2215 rdev->raid_disk = slot;
2216 if (test_bit(In_sync, &rdev->flags))
2217 rdev->saved_raid_disk = slot;
2218 else
2219 rdev->saved_raid_disk = -1;
2220 err = rdev->mddev->pers->
2221 hot_add_disk(rdev->mddev, rdev);
2222 if (err) {
2223 rdev->raid_disk = -1;
2224 return err;
2225 } else
2226 sysfs_notify_dirent(rdev->sysfs_state);
2227 sprintf(nm, "rd%d", rdev->raid_disk);
2228 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2229 printk(KERN_WARNING
2230 "md: cannot register "
2231 "%s for %s\n",
2232 nm, mdname(rdev->mddev));
2233
2234 /* don't wakeup anyone, leave that to userspace. */
2235 } else {
2236 if (slot >= rdev->mddev->raid_disks)
2237 return -ENOSPC;
2238 rdev->raid_disk = slot;
2239 /* assume it is working */
2240 clear_bit(Faulty, &rdev->flags);
2241 clear_bit(WriteMostly, &rdev->flags);
2242 set_bit(In_sync, &rdev->flags);
2243 sysfs_notify_dirent(rdev->sysfs_state);
2244 }
2245 return len;
2246 }
2247
2248
2249 static struct rdev_sysfs_entry rdev_slot =
2250 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2251
2252 static ssize_t
2253 offset_show(mdk_rdev_t *rdev, char *page)
2254 {
2255 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2256 }
2257
2258 static ssize_t
2259 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2260 {
2261 char *e;
2262 unsigned long long offset = simple_strtoull(buf, &e, 10);
2263 if (e==buf || (*e && *e != '\n'))
2264 return -EINVAL;
2265 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2266 return -EBUSY;
2267 if (rdev->sectors && rdev->mddev->external)
2268 /* Must set offset before size, so overlap checks
2269 * can be sane */
2270 return -EBUSY;
2271 rdev->data_offset = offset;
2272 return len;
2273 }
2274
2275 static struct rdev_sysfs_entry rdev_offset =
2276 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2277
2278 static ssize_t
2279 rdev_size_show(mdk_rdev_t *rdev, char *page)
2280 {
2281 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2282 }
2283
2284 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2285 {
2286 /* check if two start/length pairs overlap */
2287 if (s1+l1 <= s2)
2288 return 0;
2289 if (s2+l2 <= s1)
2290 return 0;
2291 return 1;
2292 }
2293
2294 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2295 {
2296 unsigned long long blocks;
2297 sector_t new;
2298
2299 if (strict_strtoull(buf, 10, &blocks) < 0)
2300 return -EINVAL;
2301
2302 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2303 return -EINVAL; /* sector conversion overflow */
2304
2305 new = blocks * 2;
2306 if (new != blocks * 2)
2307 return -EINVAL; /* unsigned long long to sector_t overflow */
2308
2309 *sectors = new;
2310 return 0;
2311 }
2312
2313 static ssize_t
2314 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2315 {
2316 mddev_t *my_mddev = rdev->mddev;
2317 sector_t oldsectors = rdev->sectors;
2318 sector_t sectors;
2319
2320 if (strict_blocks_to_sectors(buf, &sectors) < 0)
2321 return -EINVAL;
2322 if (my_mddev->pers && rdev->raid_disk >= 0) {
2323 if (my_mddev->persistent) {
2324 sectors = super_types[my_mddev->major_version].
2325 rdev_size_change(rdev, sectors);
2326 if (!sectors)
2327 return -EBUSY;
2328 } else if (!sectors)
2329 sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2330 rdev->data_offset;
2331 }
2332 if (sectors < my_mddev->dev_sectors)
2333 return -EINVAL; /* component must fit device */
2334
2335 rdev->sectors = sectors;
2336 if (sectors > oldsectors && my_mddev->external) {
2337 /* need to check that all other rdevs with the same ->bdev
2338 * do not overlap. We need to unlock the mddev to avoid
2339 * a deadlock. We have already changed rdev->sectors, and if
2340 * we have to change it back, we will have the lock again.
2341 */
2342 mddev_t *mddev;
2343 int overlap = 0;
2344 struct list_head *tmp;
2345
2346 mddev_unlock(my_mddev);
2347 for_each_mddev(mddev, tmp) {
2348 mdk_rdev_t *rdev2;
2349
2350 mddev_lock(mddev);
2351 list_for_each_entry(rdev2, &mddev->disks, same_set)
2352 if (test_bit(AllReserved, &rdev2->flags) ||
2353 (rdev->bdev == rdev2->bdev &&
2354 rdev != rdev2 &&
2355 overlaps(rdev->data_offset, rdev->sectors,
2356 rdev2->data_offset,
2357 rdev2->sectors))) {
2358 overlap = 1;
2359 break;
2360 }
2361 mddev_unlock(mddev);
2362 if (overlap) {
2363 mddev_put(mddev);
2364 break;
2365 }
2366 }
2367 mddev_lock(my_mddev);
2368 if (overlap) {
2369 /* Someone else could have slipped in a size
2370 * change here, but doing so is just silly.
2371 * We put oldsectors back because we *know* it is
2372 * safe, and trust userspace not to race with
2373 * itself
2374 */
2375 rdev->sectors = oldsectors;
2376 return -EBUSY;
2377 }
2378 }
2379 return len;
2380 }
2381
2382 static struct rdev_sysfs_entry rdev_size =
2383 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2384
2385 static struct attribute *rdev_default_attrs[] = {
2386 &rdev_state.attr,
2387 &rdev_errors.attr,
2388 &rdev_slot.attr,
2389 &rdev_offset.attr,
2390 &rdev_size.attr,
2391 NULL,
2392 };
2393 static ssize_t
2394 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2395 {
2396 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2397 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2398 mddev_t *mddev = rdev->mddev;
2399 ssize_t rv;
2400
2401 if (!entry->show)
2402 return -EIO;
2403
2404 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2405 if (!rv) {
2406 if (rdev->mddev == NULL)
2407 rv = -EBUSY;
2408 else
2409 rv = entry->show(rdev, page);
2410 mddev_unlock(mddev);
2411 }
2412 return rv;
2413 }
2414
2415 static ssize_t
2416 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2417 const char *page, size_t length)
2418 {
2419 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2420 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2421 ssize_t rv;
2422 mddev_t *mddev = rdev->mddev;
2423
2424 if (!entry->store)
2425 return -EIO;
2426 if (!capable(CAP_SYS_ADMIN))
2427 return -EACCES;
2428 rv = mddev ? mddev_lock(mddev): -EBUSY;
2429 if (!rv) {
2430 if (rdev->mddev == NULL)
2431 rv = -EBUSY;
2432 else
2433 rv = entry->store(rdev, page, length);
2434 mddev_unlock(mddev);
2435 }
2436 return rv;
2437 }
2438
2439 static void rdev_free(struct kobject *ko)
2440 {
2441 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2442 kfree(rdev);
2443 }
2444 static struct sysfs_ops rdev_sysfs_ops = {
2445 .show = rdev_attr_show,
2446 .store = rdev_attr_store,
2447 };
2448 static struct kobj_type rdev_ktype = {
2449 .release = rdev_free,
2450 .sysfs_ops = &rdev_sysfs_ops,
2451 .default_attrs = rdev_default_attrs,
2452 };
2453
2454 /*
2455 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2456 *
2457 * mark the device faulty if:
2458 *
2459 * - the device is nonexistent (zero size)
2460 * - the device has no valid superblock
2461 *
2462 * a faulty rdev _never_ has rdev->sb set.
2463 */
2464 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2465 {
2466 char b[BDEVNAME_SIZE];
2467 int err;
2468 mdk_rdev_t *rdev;
2469 sector_t size;
2470
2471 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2472 if (!rdev) {
2473 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2474 return ERR_PTR(-ENOMEM);
2475 }
2476
2477 if ((err = alloc_disk_sb(rdev)))
2478 goto abort_free;
2479
2480 err = lock_rdev(rdev, newdev, super_format == -2);
2481 if (err)
2482 goto abort_free;
2483
2484 kobject_init(&rdev->kobj, &rdev_ktype);
2485
2486 rdev->desc_nr = -1;
2487 rdev->saved_raid_disk = -1;
2488 rdev->raid_disk = -1;
2489 rdev->flags = 0;
2490 rdev->data_offset = 0;
2491 rdev->sb_events = 0;
2492 atomic_set(&rdev->nr_pending, 0);
2493 atomic_set(&rdev->read_errors, 0);
2494 atomic_set(&rdev->corrected_errors, 0);
2495
2496 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2497 if (!size) {
2498 printk(KERN_WARNING
2499 "md: %s has zero or unknown size, marking faulty!\n",
2500 bdevname(rdev->bdev,b));
2501 err = -EINVAL;
2502 goto abort_free;
2503 }
2504
2505 if (super_format >= 0) {
2506 err = super_types[super_format].
2507 load_super(rdev, NULL, super_minor);
2508 if (err == -EINVAL) {
2509 printk(KERN_WARNING
2510 "md: %s does not have a valid v%d.%d "
2511 "superblock, not importing!\n",
2512 bdevname(rdev->bdev,b),
2513 super_format, super_minor);
2514 goto abort_free;
2515 }
2516 if (err < 0) {
2517 printk(KERN_WARNING
2518 "md: could not read %s's sb, not importing!\n",
2519 bdevname(rdev->bdev,b));
2520 goto abort_free;
2521 }
2522 }
2523
2524 INIT_LIST_HEAD(&rdev->same_set);
2525 init_waitqueue_head(&rdev->blocked_wait);
2526
2527 return rdev;
2528
2529 abort_free:
2530 if (rdev->sb_page) {
2531 if (rdev->bdev)
2532 unlock_rdev(rdev);
2533 free_disk_sb(rdev);
2534 }
2535 kfree(rdev);
2536 return ERR_PTR(err);
2537 }
2538
2539 /*
2540 * Check a full RAID array for plausibility
2541 */
2542
2543
2544 static void analyze_sbs(mddev_t * mddev)
2545 {
2546 int i;
2547 mdk_rdev_t *rdev, *freshest, *tmp;
2548 char b[BDEVNAME_SIZE];
2549
2550 freshest = NULL;
2551 rdev_for_each(rdev, tmp, mddev)
2552 switch (super_types[mddev->major_version].
2553 load_super(rdev, freshest, mddev->minor_version)) {
2554 case 1:
2555 freshest = rdev;
2556 break;
2557 case 0:
2558 break;
2559 default:
2560 printk( KERN_ERR \
2561 "md: fatal superblock inconsistency in %s"
2562 " -- removing from array\n",
2563 bdevname(rdev->bdev,b));
2564 kick_rdev_from_array(rdev);
2565 }
2566
2567
2568 super_types[mddev->major_version].
2569 validate_super(mddev, freshest);
2570
2571 i = 0;
2572 rdev_for_each(rdev, tmp, mddev) {
2573 if (rdev->desc_nr >= mddev->max_disks ||
2574 i > mddev->max_disks) {
2575 printk(KERN_WARNING
2576 "md: %s: %s: only %d devices permitted\n",
2577 mdname(mddev), bdevname(rdev->bdev, b),
2578 mddev->max_disks);
2579 kick_rdev_from_array(rdev);
2580 continue;
2581 }
2582 if (rdev != freshest)
2583 if (super_types[mddev->major_version].
2584 validate_super(mddev, rdev)) {
2585 printk(KERN_WARNING "md: kicking non-fresh %s"
2586 " from array!\n",
2587 bdevname(rdev->bdev,b));
2588 kick_rdev_from_array(rdev);
2589 continue;
2590 }
2591 if (mddev->level == LEVEL_MULTIPATH) {
2592 rdev->desc_nr = i++;
2593 rdev->raid_disk = rdev->desc_nr;
2594 set_bit(In_sync, &rdev->flags);
2595 } else if (rdev->raid_disk >= mddev->raid_disks) {
2596 rdev->raid_disk = -1;
2597 clear_bit(In_sync, &rdev->flags);
2598 }
2599 }
2600
2601
2602
2603 if (mddev->recovery_cp != MaxSector &&
2604 mddev->level >= 1)
2605 printk(KERN_ERR "md: %s: raid array is not clean"
2606 " -- starting background reconstruction\n",
2607 mdname(mddev));
2608
2609 }
2610
2611 static void md_safemode_timeout(unsigned long data);
2612
2613 static ssize_t
2614 safe_delay_show(mddev_t *mddev, char *page)
2615 {
2616 int msec = (mddev->safemode_delay*1000)/HZ;
2617 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2618 }
2619 static ssize_t
2620 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2621 {
2622 int scale=1;
2623 int dot=0;
2624 int i;
2625 unsigned long msec;
2626 char buf[30];
2627
2628 /* remove a period, and count digits after it */
2629 if (len >= sizeof(buf))
2630 return -EINVAL;
2631 strlcpy(buf, cbuf, sizeof(buf));
2632 for (i=0; i<len; i++) {
2633 if (dot) {
2634 if (isdigit(buf[i])) {
2635 buf[i-1] = buf[i];
2636 scale *= 10;
2637 }
2638 buf[i] = 0;
2639 } else if (buf[i] == '.') {
2640 dot=1;
2641 buf[i] = 0;
2642 }
2643 }
2644 if (strict_strtoul(buf, 10, &msec) < 0)
2645 return -EINVAL;
2646 msec = (msec * 1000) / scale;
2647 if (msec == 0)
2648 mddev->safemode_delay = 0;
2649 else {
2650 unsigned long old_delay = mddev->safemode_delay;
2651 mddev->safemode_delay = (msec*HZ)/1000;
2652 if (mddev->safemode_delay == 0)
2653 mddev->safemode_delay = 1;
2654 if (mddev->safemode_delay < old_delay)
2655 md_safemode_timeout((unsigned long)mddev);
2656 }
2657 return len;
2658 }
2659 static struct md_sysfs_entry md_safe_delay =
2660 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2661
2662 static ssize_t
2663 level_show(mddev_t *mddev, char *page)
2664 {
2665 struct mdk_personality *p = mddev->pers;
2666 if (p)
2667 return sprintf(page, "%s\n", p->name);
2668 else if (mddev->clevel[0])
2669 return sprintf(page, "%s\n", mddev->clevel);
2670 else if (mddev->level != LEVEL_NONE)
2671 return sprintf(page, "%d\n", mddev->level);
2672 else
2673 return 0;
2674 }
2675
2676 static ssize_t
2677 level_store(mddev_t *mddev, const char *buf, size_t len)
2678 {
2679 char level[16];
2680 ssize_t rv = len;
2681 struct mdk_personality *pers;
2682 void *priv;
2683
2684 if (mddev->pers == NULL) {
2685 if (len == 0)
2686 return 0;
2687 if (len >= sizeof(mddev->clevel))
2688 return -ENOSPC;
2689 strncpy(mddev->clevel, buf, len);
2690 if (mddev->clevel[len-1] == '\n')
2691 len--;
2692 mddev->clevel[len] = 0;
2693 mddev->level = LEVEL_NONE;
2694 return rv;
2695 }
2696
2697 /* request to change the personality. Need to ensure:
2698 * - array is not engaged in resync/recovery/reshape
2699 * - old personality can be suspended
2700 * - new personality will access other array.
2701 */
2702
2703 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
2704 return -EBUSY;
2705
2706 if (!mddev->pers->quiesce) {
2707 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
2708 mdname(mddev), mddev->pers->name);
2709 return -EINVAL;
2710 }
2711
2712 /* Now find the new personality */
2713 if (len == 0 || len >= sizeof(level))
2714 return -EINVAL;
2715 strncpy(level, buf, len);
2716 if (level[len-1] == '\n')
2717 len--;
2718 level[len] = 0;
2719
2720 request_module("md-%s", level);
2721 spin_lock(&pers_lock);
2722 pers = find_pers(LEVEL_NONE, level);
2723 if (!pers || !try_module_get(pers->owner)) {
2724 spin_unlock(&pers_lock);
2725 printk(KERN_WARNING "md: personality %s not loaded\n", level);
2726 return -EINVAL;
2727 }
2728 spin_unlock(&pers_lock);
2729
2730 if (pers == mddev->pers) {
2731 /* Nothing to do! */
2732 module_put(pers->owner);
2733 return rv;
2734 }
2735 if (!pers->takeover) {
2736 module_put(pers->owner);
2737 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
2738 mdname(mddev), level);
2739 return -EINVAL;
2740 }
2741
2742 /* ->takeover must set new_* and/or delta_disks
2743 * if it succeeds, and may set them when it fails.
2744 */
2745 priv = pers->takeover(mddev);
2746 if (IS_ERR(priv)) {
2747 mddev->new_level = mddev->level;
2748 mddev->new_layout = mddev->layout;
2749 mddev->new_chunk = mddev->chunk_size;
2750 mddev->raid_disks -= mddev->delta_disks;
2751 mddev->delta_disks = 0;
2752 module_put(pers->owner);
2753 printk(KERN_WARNING "md: %s: %s would not accept array\n",
2754 mdname(mddev), level);
2755 return PTR_ERR(priv);
2756 }
2757
2758 /* Looks like we have a winner */
2759 mddev_suspend(mddev);
2760 mddev->pers->stop(mddev);
2761 module_put(mddev->pers->owner);
2762 mddev->pers = pers;
2763 mddev->private = priv;
2764 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
2765 mddev->level = mddev->new_level;
2766 mddev->layout = mddev->new_layout;
2767 mddev->chunk_size = mddev->new_chunk;
2768 mddev->delta_disks = 0;
2769 pers->run(mddev);
2770 mddev_resume(mddev);
2771 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2772 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2773 md_wakeup_thread(mddev->thread);
2774 return rv;
2775 }
2776
2777 static struct md_sysfs_entry md_level =
2778 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2779
2780
2781 static ssize_t
2782 layout_show(mddev_t *mddev, char *page)
2783 {
2784 /* just a number, not meaningful for all levels */
2785 if (mddev->reshape_position != MaxSector &&
2786 mddev->layout != mddev->new_layout)
2787 return sprintf(page, "%d (%d)\n",
2788 mddev->new_layout, mddev->layout);
2789 return sprintf(page, "%d\n", mddev->layout);
2790 }
2791
2792 static ssize_t
2793 layout_store(mddev_t *mddev, const char *buf, size_t len)
2794 {
2795 char *e;
2796 unsigned long n = simple_strtoul(buf, &e, 10);
2797
2798 if (!*buf || (*e && *e != '\n'))
2799 return -EINVAL;
2800
2801 if (mddev->pers) {
2802 int err;
2803 if (mddev->pers->reconfig == NULL)
2804 return -EBUSY;
2805 err = mddev->pers->reconfig(mddev, n, -1);
2806 if (err)
2807 return err;
2808 } else {
2809 mddev->new_layout = n;
2810 if (mddev->reshape_position == MaxSector)
2811 mddev->layout = n;
2812 }
2813 return len;
2814 }
2815 static struct md_sysfs_entry md_layout =
2816 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2817
2818
2819 static ssize_t
2820 raid_disks_show(mddev_t *mddev, char *page)
2821 {
2822 if (mddev->raid_disks == 0)
2823 return 0;
2824 if (mddev->reshape_position != MaxSector &&
2825 mddev->delta_disks != 0)
2826 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2827 mddev->raid_disks - mddev->delta_disks);
2828 return sprintf(page, "%d\n", mddev->raid_disks);
2829 }
2830
2831 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2832
2833 static ssize_t
2834 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2835 {
2836 char *e;
2837 int rv = 0;
2838 unsigned long n = simple_strtoul(buf, &e, 10);
2839
2840 if (!*buf || (*e && *e != '\n'))
2841 return -EINVAL;
2842
2843 if (mddev->pers)
2844 rv = update_raid_disks(mddev, n);
2845 else if (mddev->reshape_position != MaxSector) {
2846 int olddisks = mddev->raid_disks - mddev->delta_disks;
2847 mddev->delta_disks = n - olddisks;
2848 mddev->raid_disks = n;
2849 } else
2850 mddev->raid_disks = n;
2851 return rv ? rv : len;
2852 }
2853 static struct md_sysfs_entry md_raid_disks =
2854 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2855
2856 static ssize_t
2857 chunk_size_show(mddev_t *mddev, char *page)
2858 {
2859 if (mddev->reshape_position != MaxSector &&
2860 mddev->chunk_size != mddev->new_chunk)
2861 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2862 mddev->chunk_size);
2863 return sprintf(page, "%d\n", mddev->chunk_size);
2864 }
2865
2866 static ssize_t
2867 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2868 {
2869 char *e;
2870 unsigned long n = simple_strtoul(buf, &e, 10);
2871
2872 if (!*buf || (*e && *e != '\n'))
2873 return -EINVAL;
2874
2875 if (mddev->pers) {
2876 int err;
2877 if (mddev->pers->reconfig == NULL)
2878 return -EBUSY;
2879 err = mddev->pers->reconfig(mddev, -1, n);
2880 if (err)
2881 return err;
2882 } else {
2883 mddev->new_chunk = n;
2884 if (mddev->reshape_position == MaxSector)
2885 mddev->chunk_size = n;
2886 }
2887 return len;
2888 }
2889 static struct md_sysfs_entry md_chunk_size =
2890 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2891
2892 static ssize_t
2893 resync_start_show(mddev_t *mddev, char *page)
2894 {
2895 if (mddev->recovery_cp == MaxSector)
2896 return sprintf(page, "none\n");
2897 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2898 }
2899
2900 static ssize_t
2901 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2902 {
2903 char *e;
2904 unsigned long long n = simple_strtoull(buf, &e, 10);
2905
2906 if (mddev->pers)
2907 return -EBUSY;
2908 if (!*buf || (*e && *e != '\n'))
2909 return -EINVAL;
2910
2911 mddev->recovery_cp = n;
2912 return len;
2913 }
2914 static struct md_sysfs_entry md_resync_start =
2915 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2916
2917 /*
2918 * The array state can be:
2919 *
2920 * clear
2921 * No devices, no size, no level
2922 * Equivalent to STOP_ARRAY ioctl
2923 * inactive
2924 * May have some settings, but array is not active
2925 * all IO results in error
2926 * When written, doesn't tear down array, but just stops it
2927 * suspended (not supported yet)
2928 * All IO requests will block. The array can be reconfigured.
2929 * Writing this, if accepted, will block until array is quiescent
2930 * readonly
2931 * no resync can happen. no superblocks get written.
2932 * write requests fail
2933 * read-auto
2934 * like readonly, but behaves like 'clean' on a write request.
2935 *
2936 * clean - no pending writes, but otherwise active.
2937 * When written to inactive array, starts without resync
2938 * If a write request arrives then
2939 * if metadata is known, mark 'dirty' and switch to 'active'.
2940 * if not known, block and switch to write-pending
2941 * If written to an active array that has pending writes, then fails.
2942 * active
2943 * fully active: IO and resync can be happening.
2944 * When written to inactive array, starts with resync
2945 *
2946 * write-pending
2947 * clean, but writes are blocked waiting for 'active' to be written.
2948 *
2949 * active-idle
2950 * like active, but no writes have been seen for a while (100msec).
2951 *
2952 */
2953 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2954 write_pending, active_idle, bad_word};
2955 static char *array_states[] = {
2956 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2957 "write-pending", "active-idle", NULL };
2958
2959 static int match_word(const char *word, char **list)
2960 {
2961 int n;
2962 for (n=0; list[n]; n++)
2963 if (cmd_match(word, list[n]))
2964 break;
2965 return n;
2966 }
2967
2968 static ssize_t
2969 array_state_show(mddev_t *mddev, char *page)
2970 {
2971 enum array_state st = inactive;
2972
2973 if (mddev->pers)
2974 switch(mddev->ro) {
2975 case 1:
2976 st = readonly;
2977 break;
2978 case 2:
2979 st = read_auto;
2980 break;
2981 case 0:
2982 if (mddev->in_sync)
2983 st = clean;
2984 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2985 st = write_pending;
2986 else if (mddev->safemode)
2987 st = active_idle;
2988 else
2989 st = active;
2990 }
2991 else {
2992 if (list_empty(&mddev->disks) &&
2993 mddev->raid_disks == 0 &&
2994 mddev->dev_sectors == 0)
2995 st = clear;
2996 else
2997 st = inactive;
2998 }
2999 return sprintf(page, "%s\n", array_states[st]);
3000 }
3001
3002 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3003 static int do_md_run(mddev_t * mddev);
3004 static int restart_array(mddev_t *mddev);
3005
3006 static ssize_t
3007 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3008 {
3009 int err = -EINVAL;
3010 enum array_state st = match_word(buf, array_states);
3011 switch(st) {
3012 case bad_word:
3013 break;
3014 case clear:
3015 /* stopping an active array */
3016 if (atomic_read(&mddev->openers) > 0)
3017 return -EBUSY;
3018 err = do_md_stop(mddev, 0, 0);
3019 break;
3020 case inactive:
3021 /* stopping an active array */
3022 if (mddev->pers) {
3023 if (atomic_read(&mddev->openers) > 0)
3024 return -EBUSY;
3025 err = do_md_stop(mddev, 2, 0);
3026 } else
3027 err = 0; /* already inactive */
3028 break;
3029 case suspended:
3030 break; /* not supported yet */
3031 case readonly:
3032 if (mddev->pers)
3033 err = do_md_stop(mddev, 1, 0);
3034 else {
3035 mddev->ro = 1;
3036 set_disk_ro(mddev->gendisk, 1);
3037 err = do_md_run(mddev);
3038 }
3039 break;
3040 case read_auto:
3041 if (mddev->pers) {
3042 if (mddev->ro == 0)
3043 err = do_md_stop(mddev, 1, 0);
3044 else if (mddev->ro == 1)
3045 err = restart_array(mddev);
3046 if (err == 0) {
3047 mddev->ro = 2;
3048 set_disk_ro(mddev->gendisk, 0);
3049 }
3050 } else {
3051 mddev->ro = 2;
3052 err = do_md_run(mddev);
3053 }
3054 break;
3055 case clean:
3056 if (mddev->pers) {
3057 restart_array(mddev);
3058 spin_lock_irq(&mddev->write_lock);
3059 if (atomic_read(&mddev->writes_pending) == 0) {
3060 if (mddev->in_sync == 0) {
3061 mddev->in_sync = 1;
3062 if (mddev->safemode == 1)
3063 mddev->safemode = 0;
3064 if (mddev->persistent)
3065 set_bit(MD_CHANGE_CLEAN,
3066 &mddev->flags);
3067 }
3068 err = 0;
3069 } else
3070 err = -EBUSY;
3071 spin_unlock_irq(&mddev->write_lock);
3072 } else
3073 err = -EINVAL;
3074 break;
3075 case active:
3076 if (mddev->pers) {
3077 restart_array(mddev);
3078 if (mddev->external)
3079 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3080 wake_up(&mddev->sb_wait);
3081 err = 0;
3082 } else {
3083 mddev->ro = 0;
3084 set_disk_ro(mddev->gendisk, 0);
3085 err = do_md_run(mddev);
3086 }
3087 break;
3088 case write_pending:
3089 case active_idle:
3090 /* these cannot be set */
3091 break;
3092 }
3093 if (err)
3094 return err;
3095 else {
3096 sysfs_notify_dirent(mddev->sysfs_state);
3097 return len;
3098 }
3099 }
3100 static struct md_sysfs_entry md_array_state =
3101 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3102
3103 static ssize_t
3104 null_show(mddev_t *mddev, char *page)
3105 {
3106 return -EINVAL;
3107 }
3108
3109 static ssize_t
3110 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3111 {
3112 /* buf must be %d:%d\n? giving major and minor numbers */
3113 /* The new device is added to the array.
3114 * If the array has a persistent superblock, we read the
3115 * superblock to initialise info and check validity.
3116 * Otherwise, only checking done is that in bind_rdev_to_array,
3117 * which mainly checks size.
3118 */
3119 char *e;
3120 int major = simple_strtoul(buf, &e, 10);
3121 int minor;
3122 dev_t dev;
3123 mdk_rdev_t *rdev;
3124 int err;
3125
3126 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3127 return -EINVAL;
3128 minor = simple_strtoul(e+1, &e, 10);
3129 if (*e && *e != '\n')
3130 return -EINVAL;
3131 dev = MKDEV(major, minor);
3132 if (major != MAJOR(dev) ||
3133 minor != MINOR(dev))
3134 return -EOVERFLOW;
3135
3136
3137 if (mddev->persistent) {
3138 rdev = md_import_device(dev, mddev->major_version,
3139 mddev->minor_version);
3140 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3141 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3142 mdk_rdev_t, same_set);
3143 err = super_types[mddev->major_version]
3144 .load_super(rdev, rdev0, mddev->minor_version);
3145 if (err < 0)
3146 goto out;
3147 }
3148 } else if (mddev->external)
3149 rdev = md_import_device(dev, -2, -1);
3150 else
3151 rdev = md_import_device(dev, -1, -1);
3152
3153 if (IS_ERR(rdev))
3154 return PTR_ERR(rdev);
3155 err = bind_rdev_to_array(rdev, mddev);
3156 out:
3157 if (err)
3158 export_rdev(rdev);
3159 return err ? err : len;
3160 }
3161
3162 static struct md_sysfs_entry md_new_device =
3163 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3164
3165 static ssize_t
3166 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3167 {
3168 char *end;
3169 unsigned long chunk, end_chunk;
3170
3171 if (!mddev->bitmap)
3172 goto out;
3173 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3174 while (*buf) {
3175 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3176 if (buf == end) break;
3177 if (*end == '-') { /* range */
3178 buf = end + 1;
3179 end_chunk = simple_strtoul(buf, &end, 0);
3180 if (buf == end) break;
3181 }
3182 if (*end && !isspace(*end)) break;
3183 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3184 buf = end;
3185 while (isspace(*buf)) buf++;
3186 }
3187 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3188 out:
3189 return len;
3190 }
3191
3192 static struct md_sysfs_entry md_bitmap =
3193 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3194
3195 static ssize_t
3196 size_show(mddev_t *mddev, char *page)
3197 {
3198 return sprintf(page, "%llu\n",
3199 (unsigned long long)mddev->dev_sectors / 2);
3200 }
3201
3202 static int update_size(mddev_t *mddev, sector_t num_sectors);
3203
3204 static ssize_t
3205 size_store(mddev_t *mddev, const char *buf, size_t len)
3206 {
3207 /* If array is inactive, we can reduce the component size, but
3208 * not increase it (except from 0).
3209 * If array is active, we can try an on-line resize
3210 */
3211 sector_t sectors;
3212 int err = strict_blocks_to_sectors(buf, &sectors);
3213
3214 if (err < 0)
3215 return err;
3216 if (mddev->pers) {
3217 err = update_size(mddev, sectors);
3218 md_update_sb(mddev, 1);
3219 } else {
3220 if (mddev->dev_sectors == 0 ||
3221 mddev->dev_sectors > sectors)
3222 mddev->dev_sectors = sectors;
3223 else
3224 err = -ENOSPC;
3225 }
3226 return err ? err : len;
3227 }
3228
3229 static struct md_sysfs_entry md_size =
3230 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3231
3232
3233 /* Metdata version.
3234 * This is one of
3235 * 'none' for arrays with no metadata (good luck...)
3236 * 'external' for arrays with externally managed metadata,
3237 * or N.M for internally known formats
3238 */
3239 static ssize_t
3240 metadata_show(mddev_t *mddev, char *page)
3241 {
3242 if (mddev->persistent)
3243 return sprintf(page, "%d.%d\n",
3244 mddev->major_version, mddev->minor_version);
3245 else if (mddev->external)
3246 return sprintf(page, "external:%s\n", mddev->metadata_type);
3247 else
3248 return sprintf(page, "none\n");
3249 }
3250
3251 static ssize_t
3252 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3253 {
3254 int major, minor;
3255 char *e;
3256 /* Changing the details of 'external' metadata is
3257 * always permitted. Otherwise there must be
3258 * no devices attached to the array.
3259 */
3260 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3261 ;
3262 else if (!list_empty(&mddev->disks))
3263 return -EBUSY;
3264
3265 if (cmd_match(buf, "none")) {
3266 mddev->persistent = 0;
3267 mddev->external = 0;
3268 mddev->major_version = 0;
3269 mddev->minor_version = 90;
3270 return len;
3271 }
3272 if (strncmp(buf, "external:", 9) == 0) {
3273 size_t namelen = len-9;
3274 if (namelen >= sizeof(mddev->metadata_type))
3275 namelen = sizeof(mddev->metadata_type)-1;
3276 strncpy(mddev->metadata_type, buf+9, namelen);
3277 mddev->metadata_type[namelen] = 0;
3278 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3279 mddev->metadata_type[--namelen] = 0;
3280 mddev->persistent = 0;
3281 mddev->external = 1;
3282 mddev->major_version = 0;
3283 mddev->minor_version = 90;
3284 return len;
3285 }
3286 major = simple_strtoul(buf, &e, 10);
3287 if (e==buf || *e != '.')
3288 return -EINVAL;
3289 buf = e+1;
3290 minor = simple_strtoul(buf, &e, 10);
3291 if (e==buf || (*e && *e != '\n') )
3292 return -EINVAL;
3293 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3294 return -ENOENT;
3295 mddev->major_version = major;
3296 mddev->minor_version = minor;
3297 mddev->persistent = 1;
3298 mddev->external = 0;
3299 return len;
3300 }
3301
3302 static struct md_sysfs_entry md_metadata =
3303 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3304
3305 static ssize_t
3306 action_show(mddev_t *mddev, char *page)
3307 {
3308 char *type = "idle";
3309 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3310 type = "frozen";
3311 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3312 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3313 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3314 type = "reshape";
3315 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3316 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3317 type = "resync";
3318 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3319 type = "check";
3320 else
3321 type = "repair";
3322 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3323 type = "recover";
3324 }
3325 return sprintf(page, "%s\n", type);
3326 }
3327
3328 static ssize_t
3329 action_store(mddev_t *mddev, const char *page, size_t len)
3330 {
3331 if (!mddev->pers || !mddev->pers->sync_request)
3332 return -EINVAL;
3333
3334 if (cmd_match(page, "frozen"))
3335 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3336 else
3337 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3338
3339 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3340 if (mddev->sync_thread) {
3341 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3342 md_unregister_thread(mddev->sync_thread);
3343 mddev->sync_thread = NULL;
3344 mddev->recovery = 0;
3345 }
3346 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3347 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3348 return -EBUSY;
3349 else if (cmd_match(page, "resync"))
3350 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3351 else if (cmd_match(page, "recover")) {
3352 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3353 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3354 } else if (cmd_match(page, "reshape")) {
3355 int err;
3356 if (mddev->pers->start_reshape == NULL)
3357 return -EINVAL;
3358 err = mddev->pers->start_reshape(mddev);
3359 if (err)
3360 return err;
3361 sysfs_notify(&mddev->kobj, NULL, "degraded");
3362 } else {
3363 if (cmd_match(page, "check"))
3364 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3365 else if (!cmd_match(page, "repair"))
3366 return -EINVAL;
3367 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3368 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3369 }
3370 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3371 md_wakeup_thread(mddev->thread);
3372 sysfs_notify_dirent(mddev->sysfs_action);
3373 return len;
3374 }
3375
3376 static ssize_t
3377 mismatch_cnt_show(mddev_t *mddev, char *page)
3378 {
3379 return sprintf(page, "%llu\n",
3380 (unsigned long long) mddev->resync_mismatches);
3381 }
3382
3383 static struct md_sysfs_entry md_scan_mode =
3384 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3385
3386
3387 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3388
3389 static ssize_t
3390 sync_min_show(mddev_t *mddev, char *page)
3391 {
3392 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3393 mddev->sync_speed_min ? "local": "system");
3394 }
3395
3396 static ssize_t
3397 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3398 {
3399 int min;
3400 char *e;
3401 if (strncmp(buf, "system", 6)==0) {
3402 mddev->sync_speed_min = 0;
3403 return len;
3404 }
3405 min = simple_strtoul(buf, &e, 10);
3406 if (buf == e || (*e && *e != '\n') || min <= 0)
3407 return -EINVAL;
3408 mddev->sync_speed_min = min;
3409 return len;
3410 }
3411
3412 static struct md_sysfs_entry md_sync_min =
3413 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3414
3415 static ssize_t
3416 sync_max_show(mddev_t *mddev, char *page)
3417 {
3418 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3419 mddev->sync_speed_max ? "local": "system");
3420 }
3421
3422 static ssize_t
3423 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3424 {
3425 int max;
3426 char *e;
3427 if (strncmp(buf, "system", 6)==0) {
3428 mddev->sync_speed_max = 0;
3429 return len;
3430 }
3431 max = simple_strtoul(buf, &e, 10);
3432 if (buf == e || (*e && *e != '\n') || max <= 0)
3433 return -EINVAL;
3434 mddev->sync_speed_max = max;
3435 return len;
3436 }
3437
3438 static struct md_sysfs_entry md_sync_max =
3439 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3440
3441 static ssize_t
3442 degraded_show(mddev_t *mddev, char *page)
3443 {
3444 return sprintf(page, "%d\n", mddev->degraded);
3445 }
3446 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3447
3448 static ssize_t
3449 sync_force_parallel_show(mddev_t *mddev, char *page)
3450 {
3451 return sprintf(page, "%d\n", mddev->parallel_resync);
3452 }
3453
3454 static ssize_t
3455 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3456 {
3457 long n;
3458
3459 if (strict_strtol(buf, 10, &n))
3460 return -EINVAL;
3461
3462 if (n != 0 && n != 1)
3463 return -EINVAL;
3464
3465 mddev->parallel_resync = n;
3466
3467 if (mddev->sync_thread)
3468 wake_up(&resync_wait);
3469
3470 return len;
3471 }
3472
3473 /* force parallel resync, even with shared block devices */
3474 static struct md_sysfs_entry md_sync_force_parallel =
3475 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3476 sync_force_parallel_show, sync_force_parallel_store);
3477
3478 static ssize_t
3479 sync_speed_show(mddev_t *mddev, char *page)
3480 {
3481 unsigned long resync, dt, db;
3482 if (mddev->curr_resync == 0)
3483 return sprintf(page, "none\n");
3484 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3485 dt = (jiffies - mddev->resync_mark) / HZ;
3486 if (!dt) dt++;
3487 db = resync - mddev->resync_mark_cnt;
3488 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3489 }
3490
3491 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3492
3493 static ssize_t
3494 sync_completed_show(mddev_t *mddev, char *page)
3495 {
3496 unsigned long max_sectors, resync;
3497
3498 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3499 return sprintf(page, "none\n");
3500
3501 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3502 max_sectors = mddev->resync_max_sectors;
3503 else
3504 max_sectors = mddev->dev_sectors;
3505
3506 resync = mddev->curr_resync_completed;
3507 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3508 }
3509
3510 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3511
3512 static ssize_t
3513 min_sync_show(mddev_t *mddev, char *page)
3514 {
3515 return sprintf(page, "%llu\n",
3516 (unsigned long long)mddev->resync_min);
3517 }
3518 static ssize_t
3519 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3520 {
3521 unsigned long long min;
3522 if (strict_strtoull(buf, 10, &min))
3523 return -EINVAL;
3524 if (min > mddev->resync_max)
3525 return -EINVAL;
3526 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3527 return -EBUSY;
3528
3529 /* Must be a multiple of chunk_size */
3530 if (mddev->chunk_size) {
3531 if (min & (sector_t)((mddev->chunk_size>>9)-1))
3532 return -EINVAL;
3533 }
3534 mddev->resync_min = min;
3535
3536 return len;
3537 }
3538
3539 static struct md_sysfs_entry md_min_sync =
3540 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3541
3542 static ssize_t
3543 max_sync_show(mddev_t *mddev, char *page)
3544 {
3545 if (mddev->resync_max == MaxSector)
3546 return sprintf(page, "max\n");
3547 else
3548 return sprintf(page, "%llu\n",
3549 (unsigned long long)mddev->resync_max);
3550 }
3551 static ssize_t
3552 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3553 {
3554 if (strncmp(buf, "max", 3) == 0)
3555 mddev->resync_max = MaxSector;
3556 else {
3557 unsigned long long max;
3558 if (strict_strtoull(buf, 10, &max))
3559 return -EINVAL;
3560 if (max < mddev->resync_min)
3561 return -EINVAL;
3562 if (max < mddev->resync_max &&
3563 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3564 return -EBUSY;
3565
3566 /* Must be a multiple of chunk_size */
3567 if (mddev->chunk_size) {
3568 if (max & (sector_t)((mddev->chunk_size>>9)-1))
3569 return -EINVAL;
3570 }
3571 mddev->resync_max = max;
3572 }
3573 wake_up(&mddev->recovery_wait);
3574 return len;
3575 }
3576
3577 static struct md_sysfs_entry md_max_sync =
3578 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3579
3580 static ssize_t
3581 suspend_lo_show(mddev_t *mddev, char *page)
3582 {
3583 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3584 }
3585
3586 static ssize_t
3587 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3588 {
3589 char *e;
3590 unsigned long long new = simple_strtoull(buf, &e, 10);
3591
3592 if (mddev->pers->quiesce == NULL)
3593 return -EINVAL;
3594 if (buf == e || (*e && *e != '\n'))
3595 return -EINVAL;
3596 if (new >= mddev->suspend_hi ||
3597 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3598 mddev->suspend_lo = new;
3599 mddev->pers->quiesce(mddev, 2);
3600 return len;
3601 } else
3602 return -EINVAL;
3603 }
3604 static struct md_sysfs_entry md_suspend_lo =
3605 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3606
3607
3608 static ssize_t
3609 suspend_hi_show(mddev_t *mddev, char *page)
3610 {
3611 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3612 }
3613
3614 static ssize_t
3615 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3616 {
3617 char *e;
3618 unsigned long long new = simple_strtoull(buf, &e, 10);
3619
3620 if (mddev->pers->quiesce == NULL)
3621 return -EINVAL;
3622 if (buf == e || (*e && *e != '\n'))
3623 return -EINVAL;
3624 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3625 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3626 mddev->suspend_hi = new;
3627 mddev->pers->quiesce(mddev, 1);
3628 mddev->pers->quiesce(mddev, 0);
3629 return len;
3630 } else
3631 return -EINVAL;
3632 }
3633 static struct md_sysfs_entry md_suspend_hi =
3634 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3635
3636 static ssize_t
3637 reshape_position_show(mddev_t *mddev, char *page)
3638 {
3639 if (mddev->reshape_position != MaxSector)
3640 return sprintf(page, "%llu\n",
3641 (unsigned long long)mddev->reshape_position);
3642 strcpy(page, "none\n");
3643 return 5;
3644 }
3645
3646 static ssize_t
3647 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3648 {
3649 char *e;
3650 unsigned long long new = simple_strtoull(buf, &e, 10);
3651 if (mddev->pers)
3652 return -EBUSY;
3653 if (buf == e || (*e && *e != '\n'))
3654 return -EINVAL;
3655 mddev->reshape_position = new;
3656 mddev->delta_disks = 0;
3657 mddev->new_level = mddev->level;
3658 mddev->new_layout = mddev->layout;
3659 mddev->new_chunk = mddev->chunk_size;
3660 return len;
3661 }
3662
3663 static struct md_sysfs_entry md_reshape_position =
3664 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3665 reshape_position_store);
3666
3667 static ssize_t
3668 array_size_show(mddev_t *mddev, char *page)
3669 {
3670 if (mddev->external_size)
3671 return sprintf(page, "%llu\n",
3672 (unsigned long long)mddev->array_sectors/2);
3673 else
3674 return sprintf(page, "default\n");
3675 }
3676
3677 static ssize_t
3678 array_size_store(mddev_t *mddev, const char *buf, size_t len)
3679 {
3680 sector_t sectors;
3681
3682 if (strncmp(buf, "default", 7) == 0) {
3683 if (mddev->pers)
3684 sectors = mddev->pers->size(mddev, 0, 0);
3685 else
3686 sectors = mddev->array_sectors;
3687
3688 mddev->external_size = 0;
3689 } else {
3690 if (strict_blocks_to_sectors(buf, &sectors) < 0)
3691 return -EINVAL;
3692 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
3693 return -E2BIG;
3694
3695 mddev->external_size = 1;
3696 }
3697
3698 mddev->array_sectors = sectors;
3699 set_capacity(mddev->gendisk, mddev->array_sectors);
3700 if (mddev->pers) {
3701 struct block_device *bdev = bdget_disk(mddev->gendisk, 0);
3702
3703 if (bdev) {
3704 mutex_lock(&bdev->bd_inode->i_mutex);
3705 i_size_write(bdev->bd_inode,
3706 (loff_t)mddev->array_sectors << 9);
3707 mutex_unlock(&bdev->bd_inode->i_mutex);
3708 bdput(bdev);
3709 }
3710 }
3711
3712 return len;
3713 }
3714
3715 static struct md_sysfs_entry md_array_size =
3716 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
3717 array_size_store);
3718
3719 static struct attribute *md_default_attrs[] = {
3720 &md_level.attr,
3721 &md_layout.attr,
3722 &md_raid_disks.attr,
3723 &md_chunk_size.attr,
3724 &md_size.attr,
3725 &md_resync_start.attr,
3726 &md_metadata.attr,
3727 &md_new_device.attr,
3728 &md_safe_delay.attr,
3729 &md_array_state.attr,
3730 &md_reshape_position.attr,
3731 &md_array_size.attr,
3732 NULL,
3733 };
3734
3735 static struct attribute *md_redundancy_attrs[] = {
3736 &md_scan_mode.attr,
3737 &md_mismatches.attr,
3738 &md_sync_min.attr,
3739 &md_sync_max.attr,
3740 &md_sync_speed.attr,
3741 &md_sync_force_parallel.attr,
3742 &md_sync_completed.attr,
3743 &md_min_sync.attr,
3744 &md_max_sync.attr,
3745 &md_suspend_lo.attr,
3746 &md_suspend_hi.attr,
3747 &md_bitmap.attr,
3748 &md_degraded.attr,
3749 NULL,
3750 };
3751 static struct attribute_group md_redundancy_group = {
3752 .name = NULL,
3753 .attrs = md_redundancy_attrs,
3754 };
3755
3756
3757 static ssize_t
3758 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3759 {
3760 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3761 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3762 ssize_t rv;
3763
3764 if (!entry->show)
3765 return -EIO;
3766 rv = mddev_lock(mddev);
3767 if (!rv) {
3768 rv = entry->show(mddev, page);
3769 mddev_unlock(mddev);
3770 }
3771 return rv;
3772 }
3773
3774 static ssize_t
3775 md_attr_store(struct kobject *kobj, struct attribute *attr,
3776 const char *page, size_t length)
3777 {
3778 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3779 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3780 ssize_t rv;
3781
3782 if (!entry->store)
3783 return -EIO;
3784 if (!capable(CAP_SYS_ADMIN))
3785 return -EACCES;
3786 rv = mddev_lock(mddev);
3787 if (mddev->hold_active == UNTIL_IOCTL)
3788 mddev->hold_active = 0;
3789 if (!rv) {
3790 rv = entry->store(mddev, page, length);
3791 mddev_unlock(mddev);
3792 }
3793 return rv;
3794 }
3795
3796 static void md_free(struct kobject *ko)
3797 {
3798 mddev_t *mddev = container_of(ko, mddev_t, kobj);
3799
3800 if (mddev->sysfs_state)
3801 sysfs_put(mddev->sysfs_state);
3802
3803 if (mddev->gendisk) {
3804 del_gendisk(mddev->gendisk);
3805 put_disk(mddev->gendisk);
3806 }
3807 if (mddev->queue)
3808 blk_cleanup_queue(mddev->queue);
3809
3810 kfree(mddev);
3811 }
3812
3813 static struct sysfs_ops md_sysfs_ops = {
3814 .show = md_attr_show,
3815 .store = md_attr_store,
3816 };
3817 static struct kobj_type md_ktype = {
3818 .release = md_free,
3819 .sysfs_ops = &md_sysfs_ops,
3820 .default_attrs = md_default_attrs,
3821 };
3822
3823 int mdp_major = 0;
3824
3825 static void mddev_delayed_delete(struct work_struct *ws)
3826 {
3827 mddev_t *mddev = container_of(ws, mddev_t, del_work);
3828
3829 if (mddev->private == &md_redundancy_group) {
3830 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3831 if (mddev->sysfs_action)
3832 sysfs_put(mddev->sysfs_action);
3833 mddev->sysfs_action = NULL;
3834 mddev->private = NULL;
3835 }
3836 kobject_del(&mddev->kobj);
3837 kobject_put(&mddev->kobj);
3838 }
3839
3840 static int md_alloc(dev_t dev, char *name)
3841 {
3842 static DEFINE_MUTEX(disks_mutex);
3843 mddev_t *mddev = mddev_find(dev);
3844 struct gendisk *disk;
3845 int partitioned;
3846 int shift;
3847 int unit;
3848 int error;
3849
3850 if (!mddev)
3851 return -ENODEV;
3852
3853 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
3854 shift = partitioned ? MdpMinorShift : 0;
3855 unit = MINOR(mddev->unit) >> shift;
3856
3857 /* wait for any previous instance if this device
3858 * to be completed removed (mddev_delayed_delete).
3859 */
3860 flush_scheduled_work();
3861
3862 mutex_lock(&disks_mutex);
3863 if (mddev->gendisk) {
3864 mutex_unlock(&disks_mutex);
3865 mddev_put(mddev);
3866 return -EEXIST;
3867 }
3868
3869 if (name) {
3870 /* Need to ensure that 'name' is not a duplicate.
3871 */
3872 mddev_t *mddev2;
3873 spin_lock(&all_mddevs_lock);
3874
3875 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
3876 if (mddev2->gendisk &&
3877 strcmp(mddev2->gendisk->disk_name, name) == 0) {
3878 spin_unlock(&all_mddevs_lock);
3879 return -EEXIST;
3880 }
3881 spin_unlock(&all_mddevs_lock);
3882 }
3883
3884 mddev->queue = blk_alloc_queue(GFP_KERNEL);
3885 if (!mddev->queue) {
3886 mutex_unlock(&disks_mutex);
3887 mddev_put(mddev);
3888 return -ENOMEM;
3889 }
3890 mddev->queue->queuedata = mddev;
3891
3892 /* Can be unlocked because the queue is new: no concurrency */
3893 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
3894
3895 blk_queue_make_request(mddev->queue, md_make_request);
3896
3897 disk = alloc_disk(1 << shift);
3898 if (!disk) {
3899 mutex_unlock(&disks_mutex);
3900 blk_cleanup_queue(mddev->queue);
3901 mddev->queue = NULL;
3902 mddev_put(mddev);
3903 return -ENOMEM;
3904 }
3905 disk->major = MAJOR(mddev->unit);
3906 disk->first_minor = unit << shift;
3907 if (name)
3908 strcpy(disk->disk_name, name);
3909 else if (partitioned)
3910 sprintf(disk->disk_name, "md_d%d", unit);
3911 else
3912 sprintf(disk->disk_name, "md%d", unit);
3913 disk->fops = &md_fops;
3914 disk->private_data = mddev;
3915 disk->queue = mddev->queue;
3916 /* Allow extended partitions. This makes the
3917 * 'mdp' device redundant, but we can't really
3918 * remove it now.
3919 */
3920 disk->flags |= GENHD_FL_EXT_DEVT;
3921 add_disk(disk);
3922 mddev->gendisk = disk;
3923 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
3924 &disk_to_dev(disk)->kobj, "%s", "md");
3925 mutex_unlock(&disks_mutex);
3926 if (error)
3927 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3928 disk->disk_name);
3929 else {
3930 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3931 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state");
3932 }
3933 mddev_put(mddev);
3934 return 0;
3935 }
3936
3937 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3938 {
3939 md_alloc(dev, NULL);
3940 return NULL;
3941 }
3942
3943 static int add_named_array(const char *val, struct kernel_param *kp)
3944 {
3945 /* val must be "md_*" where * is not all digits.
3946 * We allocate an array with a large free minor number, and
3947 * set the name to val. val must not already be an active name.
3948 */
3949 int len = strlen(val);
3950 char buf[DISK_NAME_LEN];
3951
3952 while (len && val[len-1] == '\n')
3953 len--;
3954 if (len >= DISK_NAME_LEN)
3955 return -E2BIG;
3956 strlcpy(buf, val, len+1);
3957 if (strncmp(buf, "md_", 3) != 0)
3958 return -EINVAL;
3959 return md_alloc(0, buf);
3960 }
3961
3962 static void md_safemode_timeout(unsigned long data)
3963 {
3964 mddev_t *mddev = (mddev_t *) data;
3965
3966 if (!atomic_read(&mddev->writes_pending)) {
3967 mddev->safemode = 1;
3968 if (mddev->external)
3969 sysfs_notify_dirent(mddev->sysfs_state);
3970 }
3971 md_wakeup_thread(mddev->thread);
3972 }
3973
3974 static int start_dirty_degraded;
3975
3976 static int do_md_run(mddev_t * mddev)
3977 {
3978 int err;
3979 int chunk_size;
3980 mdk_rdev_t *rdev;
3981 struct gendisk *disk;
3982 struct mdk_personality *pers;
3983 char b[BDEVNAME_SIZE];
3984
3985 if (list_empty(&mddev->disks))
3986 /* cannot run an array with no devices.. */
3987 return -EINVAL;
3988
3989 if (mddev->pers)
3990 return -EBUSY;
3991
3992 /*
3993 * Analyze all RAID superblock(s)
3994 */
3995 if (!mddev->raid_disks) {
3996 if (!mddev->persistent)
3997 return -EINVAL;
3998 analyze_sbs(mddev);
3999 }
4000
4001 chunk_size = mddev->chunk_size;
4002
4003 if (chunk_size) {
4004 if (chunk_size > MAX_CHUNK_SIZE) {
4005 printk(KERN_ERR "too big chunk_size: %d > %d\n",
4006 chunk_size, MAX_CHUNK_SIZE);
4007 return -EINVAL;
4008 }
4009 /*
4010 * chunk-size has to be a power of 2
4011 */
4012 if ( (1 << ffz(~chunk_size)) != chunk_size) {
4013 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
4014 return -EINVAL;
4015 }
4016
4017 /* devices must have minimum size of one chunk */
4018 list_for_each_entry(rdev, &mddev->disks, same_set) {
4019 if (test_bit(Faulty, &rdev->flags))
4020 continue;
4021 if (rdev->sectors < chunk_size / 512) {
4022 printk(KERN_WARNING
4023 "md: Dev %s smaller than chunk_size:"
4024 " %llu < %d\n",
4025 bdevname(rdev->bdev,b),
4026 (unsigned long long)rdev->sectors,
4027 chunk_size / 512);
4028 return -EINVAL;
4029 }
4030 }
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 (pers->level >= 4 && pers->level <= 6)
4098 /* Cannot support integrity (yet) */
4099 blk_integrity_unregister(mddev->gendisk);
4100
4101 if (mddev->reshape_position != MaxSector &&
4102 pers->start_reshape == NULL) {
4103 /* This personality cannot handle reshaping... */
4104 mddev->pers = NULL;
4105 module_put(pers->owner);
4106 return -EINVAL;
4107 }
4108
4109 if (pers->sync_request) {
4110 /* Warn if this is a potentially silly
4111 * configuration.
4112 */
4113 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4114 mdk_rdev_t *rdev2;
4115 int warned = 0;
4116
4117 list_for_each_entry(rdev, &mddev->disks, same_set)
4118 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4119 if (rdev < rdev2 &&
4120 rdev->bdev->bd_contains ==
4121 rdev2->bdev->bd_contains) {
4122 printk(KERN_WARNING
4123 "%s: WARNING: %s appears to be"
4124 " on the same physical disk as"
4125 " %s.\n",
4126 mdname(mddev),
4127 bdevname(rdev->bdev,b),
4128 bdevname(rdev2->bdev,b2));
4129 warned = 1;
4130 }
4131 }
4132
4133 if (warned)
4134 printk(KERN_WARNING
4135 "True protection against single-disk"
4136 " failure might be compromised.\n");
4137 }
4138
4139 mddev->recovery = 0;
4140 /* may be over-ridden by personality */
4141 mddev->resync_max_sectors = mddev->dev_sectors;
4142
4143 mddev->barriers_work = 1;
4144 mddev->ok_start_degraded = start_dirty_degraded;
4145
4146 if (start_readonly)
4147 mddev->ro = 2; /* read-only, but switch on first write */
4148
4149 err = mddev->pers->run(mddev);
4150 if (err)
4151 printk(KERN_ERR "md: pers->run() failed ...\n");
4152 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4153 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4154 " but 'external_size' not in effect?\n", __func__);
4155 printk(KERN_ERR
4156 "md: invalid array_size %llu > default size %llu\n",
4157 (unsigned long long)mddev->array_sectors / 2,
4158 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4159 err = -EINVAL;
4160 mddev->pers->stop(mddev);
4161 }
4162 if (err == 0 && mddev->pers->sync_request) {
4163 err = bitmap_create(mddev);
4164 if (err) {
4165 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4166 mdname(mddev), err);
4167 mddev->pers->stop(mddev);
4168 }
4169 }
4170 if (err) {
4171 module_put(mddev->pers->owner);
4172 mddev->pers = NULL;
4173 bitmap_destroy(mddev);
4174 return err;
4175 }
4176 if (mddev->pers->sync_request) {
4177 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4178 printk(KERN_WARNING
4179 "md: cannot register extra attributes for %s\n",
4180 mdname(mddev));
4181 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4182 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4183 mddev->ro = 0;
4184
4185 atomic_set(&mddev->writes_pending,0);
4186 mddev->safemode = 0;
4187 mddev->safemode_timer.function = md_safemode_timeout;
4188 mddev->safemode_timer.data = (unsigned long) mddev;
4189 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4190 mddev->in_sync = 1;
4191
4192 list_for_each_entry(rdev, &mddev->disks, same_set)
4193 if (rdev->raid_disk >= 0) {
4194 char nm[20];
4195 sprintf(nm, "rd%d", rdev->raid_disk);
4196 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4197 printk("md: cannot register %s for %s\n",
4198 nm, mdname(mddev));
4199 }
4200
4201 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4202
4203 if (mddev->flags)
4204 md_update_sb(mddev, 0);
4205
4206 set_capacity(disk, mddev->array_sectors);
4207
4208 /* If there is a partially-recovered drive we need to
4209 * start recovery here. If we leave it to md_check_recovery,
4210 * it will remove the drives and not do the right thing
4211 */
4212 if (mddev->degraded && !mddev->sync_thread) {
4213 int spares = 0;
4214 list_for_each_entry(rdev, &mddev->disks, same_set)
4215 if (rdev->raid_disk >= 0 &&
4216 !test_bit(In_sync, &rdev->flags) &&
4217 !test_bit(Faulty, &rdev->flags))
4218 /* complete an interrupted recovery */
4219 spares++;
4220 if (spares && mddev->pers->sync_request) {
4221 mddev->recovery = 0;
4222 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4223 mddev->sync_thread = md_register_thread(md_do_sync,
4224 mddev,
4225 "%s_resync");
4226 if (!mddev->sync_thread) {
4227 printk(KERN_ERR "%s: could not start resync"
4228 " thread...\n",
4229 mdname(mddev));
4230 /* leave the spares where they are, it shouldn't hurt */
4231 mddev->recovery = 0;
4232 }
4233 }
4234 }
4235 md_wakeup_thread(mddev->thread);
4236 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4237
4238 mddev->changed = 1;
4239 md_new_event(mddev);
4240 sysfs_notify_dirent(mddev->sysfs_state);
4241 if (mddev->sysfs_action)
4242 sysfs_notify_dirent(mddev->sysfs_action);
4243 sysfs_notify(&mddev->kobj, NULL, "degraded");
4244 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4245 return 0;
4246 }
4247
4248 static int restart_array(mddev_t *mddev)
4249 {
4250 struct gendisk *disk = mddev->gendisk;
4251
4252 /* Complain if it has no devices */
4253 if (list_empty(&mddev->disks))
4254 return -ENXIO;
4255 if (!mddev->pers)
4256 return -EINVAL;
4257 if (!mddev->ro)
4258 return -EBUSY;
4259 mddev->safemode = 0;
4260 mddev->ro = 0;
4261 set_disk_ro(disk, 0);
4262 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4263 mdname(mddev));
4264 /* Kick recovery or resync if necessary */
4265 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4266 md_wakeup_thread(mddev->thread);
4267 md_wakeup_thread(mddev->sync_thread);
4268 sysfs_notify_dirent(mddev->sysfs_state);
4269 return 0;
4270 }
4271
4272 /* similar to deny_write_access, but accounts for our holding a reference
4273 * to the file ourselves */
4274 static int deny_bitmap_write_access(struct file * file)
4275 {
4276 struct inode *inode = file->f_mapping->host;
4277
4278 spin_lock(&inode->i_lock);
4279 if (atomic_read(&inode->i_writecount) > 1) {
4280 spin_unlock(&inode->i_lock);
4281 return -ETXTBSY;
4282 }
4283 atomic_set(&inode->i_writecount, -1);
4284 spin_unlock(&inode->i_lock);
4285
4286 return 0;
4287 }
4288
4289 static void restore_bitmap_write_access(struct file *file)
4290 {
4291 struct inode *inode = file->f_mapping->host;
4292
4293 spin_lock(&inode->i_lock);
4294 atomic_set(&inode->i_writecount, 1);
4295 spin_unlock(&inode->i_lock);
4296 }
4297
4298 /* mode:
4299 * 0 - completely stop and dis-assemble array
4300 * 1 - switch to readonly
4301 * 2 - stop but do not disassemble array
4302 */
4303 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4304 {
4305 int err = 0;
4306 struct gendisk *disk = mddev->gendisk;
4307 mdk_rdev_t *rdev;
4308
4309 if (atomic_read(&mddev->openers) > is_open) {
4310 printk("md: %s still in use.\n",mdname(mddev));
4311 return -EBUSY;
4312 }
4313
4314 if (mddev->pers) {
4315
4316 if (mddev->sync_thread) {
4317 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4318 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4319 md_unregister_thread(mddev->sync_thread);
4320 mddev->sync_thread = NULL;
4321 }
4322
4323 del_timer_sync(&mddev->safemode_timer);
4324
4325 switch(mode) {
4326 case 1: /* readonly */
4327 err = -ENXIO;
4328 if (mddev->ro==1)
4329 goto out;
4330 mddev->ro = 1;
4331 break;
4332 case 0: /* disassemble */
4333 case 2: /* stop */
4334 bitmap_flush(mddev);
4335 md_super_wait(mddev);
4336 if (mddev->ro)
4337 set_disk_ro(disk, 0);
4338
4339 mddev->pers->stop(mddev);
4340 mddev->queue->merge_bvec_fn = NULL;
4341 mddev->queue->unplug_fn = NULL;
4342 mddev->queue->backing_dev_info.congested_fn = NULL;
4343 module_put(mddev->pers->owner);
4344 if (mddev->pers->sync_request)
4345 mddev->private = &md_redundancy_group;
4346 mddev->pers = NULL;
4347 /* tell userspace to handle 'inactive' */
4348 sysfs_notify_dirent(mddev->sysfs_state);
4349
4350 list_for_each_entry(rdev, &mddev->disks, same_set)
4351 if (rdev->raid_disk >= 0) {
4352 char nm[20];
4353 sprintf(nm, "rd%d", rdev->raid_disk);
4354 sysfs_remove_link(&mddev->kobj, nm);
4355 }
4356
4357 set_capacity(disk, 0);
4358 mddev->changed = 1;
4359
4360 if (mddev->ro)
4361 mddev->ro = 0;
4362 }
4363 if (!mddev->in_sync || mddev->flags) {
4364 /* mark array as shutdown cleanly */
4365 mddev->in_sync = 1;
4366 md_update_sb(mddev, 1);
4367 }
4368 if (mode == 1)
4369 set_disk_ro(disk, 1);
4370 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4371 }
4372
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_size = 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 = 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 out:
4439 return err;
4440 }
4441
4442 #ifndef MODULE
4443 static void autorun_array(mddev_t *mddev)
4444 {
4445 mdk_rdev_t *rdev;
4446 int err;
4447
4448 if (list_empty(&mddev->disks))
4449 return;
4450
4451 printk(KERN_INFO "md: running: ");
4452
4453 list_for_each_entry(rdev, &mddev->disks, same_set) {
4454 char b[BDEVNAME_SIZE];
4455 printk("<%s>", bdevname(rdev->bdev,b));
4456 }
4457 printk("\n");
4458
4459 err = do_md_run(mddev);
4460 if (err) {
4461 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4462 do_md_stop(mddev, 0, 0);
4463 }
4464 }
4465
4466 /*
4467 * lets try to run arrays based on all disks that have arrived
4468 * until now. (those are in pending_raid_disks)
4469 *
4470 * the method: pick the first pending disk, collect all disks with
4471 * the same UUID, remove all from the pending list and put them into
4472 * the 'same_array' list. Then order this list based on superblock
4473 * update time (freshest comes first), kick out 'old' disks and
4474 * compare superblocks. If everything's fine then run it.
4475 *
4476 * If "unit" is allocated, then bump its reference count
4477 */
4478 static void autorun_devices(int part)
4479 {
4480 mdk_rdev_t *rdev0, *rdev, *tmp;
4481 mddev_t *mddev;
4482 char b[BDEVNAME_SIZE];
4483
4484 printk(KERN_INFO "md: autorun ...\n");
4485 while (!list_empty(&pending_raid_disks)) {
4486 int unit;
4487 dev_t dev;
4488 LIST_HEAD(candidates);
4489 rdev0 = list_entry(pending_raid_disks.next,
4490 mdk_rdev_t, same_set);
4491
4492 printk(KERN_INFO "md: considering %s ...\n",
4493 bdevname(rdev0->bdev,b));
4494 INIT_LIST_HEAD(&candidates);
4495 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4496 if (super_90_load(rdev, rdev0, 0) >= 0) {
4497 printk(KERN_INFO "md: adding %s ...\n",
4498 bdevname(rdev->bdev,b));
4499 list_move(&rdev->same_set, &candidates);
4500 }
4501 /*
4502 * now we have a set of devices, with all of them having
4503 * mostly sane superblocks. It's time to allocate the
4504 * mddev.
4505 */
4506 if (part) {
4507 dev = MKDEV(mdp_major,
4508 rdev0->preferred_minor << MdpMinorShift);
4509 unit = MINOR(dev) >> MdpMinorShift;
4510 } else {
4511 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4512 unit = MINOR(dev);
4513 }
4514 if (rdev0->preferred_minor != unit) {
4515 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4516 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4517 break;
4518 }
4519
4520 md_probe(dev, NULL, NULL);
4521 mddev = mddev_find(dev);
4522 if (!mddev || !mddev->gendisk) {
4523 if (mddev)
4524 mddev_put(mddev);
4525 printk(KERN_ERR
4526 "md: cannot allocate memory for md drive.\n");
4527 break;
4528 }
4529 if (mddev_lock(mddev))
4530 printk(KERN_WARNING "md: %s locked, cannot run\n",
4531 mdname(mddev));
4532 else if (mddev->raid_disks || mddev->major_version
4533 || !list_empty(&mddev->disks)) {
4534 printk(KERN_WARNING
4535 "md: %s already running, cannot run %s\n",
4536 mdname(mddev), bdevname(rdev0->bdev,b));
4537 mddev_unlock(mddev);
4538 } else {
4539 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4540 mddev->persistent = 1;
4541 rdev_for_each_list(rdev, tmp, &candidates) {
4542 list_del_init(&rdev->same_set);
4543 if (bind_rdev_to_array(rdev, mddev))
4544 export_rdev(rdev);
4545 }
4546 autorun_array(mddev);
4547 mddev_unlock(mddev);
4548 }
4549 /* on success, candidates will be empty, on error
4550 * it won't...
4551 */
4552 rdev_for_each_list(rdev, tmp, &candidates) {
4553 list_del_init(&rdev->same_set);
4554 export_rdev(rdev);
4555 }
4556 mddev_put(mddev);
4557 }
4558 printk(KERN_INFO "md: ... autorun DONE.\n");
4559 }
4560 #endif /* !MODULE */
4561
4562 static int get_version(void __user * arg)
4563 {
4564 mdu_version_t ver;
4565
4566 ver.major = MD_MAJOR_VERSION;
4567 ver.minor = MD_MINOR_VERSION;
4568 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4569
4570 if (copy_to_user(arg, &ver, sizeof(ver)))
4571 return -EFAULT;
4572
4573 return 0;
4574 }
4575
4576 static int get_array_info(mddev_t * mddev, void __user * arg)
4577 {
4578 mdu_array_info_t info;
4579 int nr,working,active,failed,spare;
4580 mdk_rdev_t *rdev;
4581
4582 nr=working=active=failed=spare=0;
4583 list_for_each_entry(rdev, &mddev->disks, same_set) {
4584 nr++;
4585 if (test_bit(Faulty, &rdev->flags))
4586 failed++;
4587 else {
4588 working++;
4589 if (test_bit(In_sync, &rdev->flags))
4590 active++;
4591 else
4592 spare++;
4593 }
4594 }
4595
4596 info.major_version = mddev->major_version;
4597 info.minor_version = mddev->minor_version;
4598 info.patch_version = MD_PATCHLEVEL_VERSION;
4599 info.ctime = mddev->ctime;
4600 info.level = mddev->level;
4601 info.size = mddev->dev_sectors / 2;
4602 if (info.size != mddev->dev_sectors / 2) /* overflow */
4603 info.size = -1;
4604 info.nr_disks = nr;
4605 info.raid_disks = mddev->raid_disks;
4606 info.md_minor = mddev->md_minor;
4607 info.not_persistent= !mddev->persistent;
4608
4609 info.utime = mddev->utime;
4610 info.state = 0;
4611 if (mddev->in_sync)
4612 info.state = (1<<MD_SB_CLEAN);
4613 if (mddev->bitmap && mddev->bitmap_offset)
4614 info.state = (1<<MD_SB_BITMAP_PRESENT);
4615 info.active_disks = active;
4616 info.working_disks = working;
4617 info.failed_disks = failed;
4618 info.spare_disks = spare;
4619
4620 info.layout = mddev->layout;
4621 info.chunk_size = mddev->chunk_size;
4622
4623 if (copy_to_user(arg, &info, sizeof(info)))
4624 return -EFAULT;
4625
4626 return 0;
4627 }
4628
4629 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4630 {
4631 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4632 char *ptr, *buf = NULL;
4633 int err = -ENOMEM;
4634
4635 if (md_allow_write(mddev))
4636 file = kmalloc(sizeof(*file), GFP_NOIO);
4637 else
4638 file = kmalloc(sizeof(*file), GFP_KERNEL);
4639
4640 if (!file)
4641 goto out;
4642
4643 /* bitmap disabled, zero the first byte and copy out */
4644 if (!mddev->bitmap || !mddev->bitmap->file) {
4645 file->pathname[0] = '\0';
4646 goto copy_out;
4647 }
4648
4649 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4650 if (!buf)
4651 goto out;
4652
4653 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4654 if (IS_ERR(ptr))
4655 goto out;
4656
4657 strcpy(file->pathname, ptr);
4658
4659 copy_out:
4660 err = 0;
4661 if (copy_to_user(arg, file, sizeof(*file)))
4662 err = -EFAULT;
4663 out:
4664 kfree(buf);
4665 kfree(file);
4666 return err;
4667 }
4668
4669 static int get_disk_info(mddev_t * mddev, void __user * arg)
4670 {
4671 mdu_disk_info_t info;
4672 mdk_rdev_t *rdev;
4673
4674 if (copy_from_user(&info, arg, sizeof(info)))
4675 return -EFAULT;
4676
4677 rdev = find_rdev_nr(mddev, info.number);
4678 if (rdev) {
4679 info.major = MAJOR(rdev->bdev->bd_dev);
4680 info.minor = MINOR(rdev->bdev->bd_dev);
4681 info.raid_disk = rdev->raid_disk;
4682 info.state = 0;
4683 if (test_bit(Faulty, &rdev->flags))
4684 info.state |= (1<<MD_DISK_FAULTY);
4685 else if (test_bit(In_sync, &rdev->flags)) {
4686 info.state |= (1<<MD_DISK_ACTIVE);
4687 info.state |= (1<<MD_DISK_SYNC);
4688 }
4689 if (test_bit(WriteMostly, &rdev->flags))
4690 info.state |= (1<<MD_DISK_WRITEMOSTLY);
4691 } else {
4692 info.major = info.minor = 0;
4693 info.raid_disk = -1;
4694 info.state = (1<<MD_DISK_REMOVED);
4695 }
4696
4697 if (copy_to_user(arg, &info, sizeof(info)))
4698 return -EFAULT;
4699
4700 return 0;
4701 }
4702
4703 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4704 {
4705 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4706 mdk_rdev_t *rdev;
4707 dev_t dev = MKDEV(info->major,info->minor);
4708
4709 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4710 return -EOVERFLOW;
4711
4712 if (!mddev->raid_disks) {
4713 int err;
4714 /* expecting a device which has a superblock */
4715 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4716 if (IS_ERR(rdev)) {
4717 printk(KERN_WARNING
4718 "md: md_import_device returned %ld\n",
4719 PTR_ERR(rdev));
4720 return PTR_ERR(rdev);
4721 }
4722 if (!list_empty(&mddev->disks)) {
4723 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4724 mdk_rdev_t, same_set);
4725 int err = super_types[mddev->major_version]
4726 .load_super(rdev, rdev0, mddev->minor_version);
4727 if (err < 0) {
4728 printk(KERN_WARNING
4729 "md: %s has different UUID to %s\n",
4730 bdevname(rdev->bdev,b),
4731 bdevname(rdev0->bdev,b2));
4732 export_rdev(rdev);
4733 return -EINVAL;
4734 }
4735 }
4736 err = bind_rdev_to_array(rdev, mddev);
4737 if (err)
4738 export_rdev(rdev);
4739 return err;
4740 }
4741
4742 /*
4743 * add_new_disk can be used once the array is assembled
4744 * to add "hot spares". They must already have a superblock
4745 * written
4746 */
4747 if (mddev->pers) {
4748 int err;
4749 if (!mddev->pers->hot_add_disk) {
4750 printk(KERN_WARNING
4751 "%s: personality does not support diskops!\n",
4752 mdname(mddev));
4753 return -EINVAL;
4754 }
4755 if (mddev->persistent)
4756 rdev = md_import_device(dev, mddev->major_version,
4757 mddev->minor_version);
4758 else
4759 rdev = md_import_device(dev, -1, -1);
4760 if (IS_ERR(rdev)) {
4761 printk(KERN_WARNING
4762 "md: md_import_device returned %ld\n",
4763 PTR_ERR(rdev));
4764 return PTR_ERR(rdev);
4765 }
4766 /* set save_raid_disk if appropriate */
4767 if (!mddev->persistent) {
4768 if (info->state & (1<<MD_DISK_SYNC) &&
4769 info->raid_disk < mddev->raid_disks)
4770 rdev->raid_disk = info->raid_disk;
4771 else
4772 rdev->raid_disk = -1;
4773 } else
4774 super_types[mddev->major_version].
4775 validate_super(mddev, rdev);
4776 rdev->saved_raid_disk = rdev->raid_disk;
4777
4778 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4779 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4780 set_bit(WriteMostly, &rdev->flags);
4781 else
4782 clear_bit(WriteMostly, &rdev->flags);
4783
4784 rdev->raid_disk = -1;
4785 err = bind_rdev_to_array(rdev, mddev);
4786 if (!err && !mddev->pers->hot_remove_disk) {
4787 /* If there is hot_add_disk but no hot_remove_disk
4788 * then added disks for geometry changes,
4789 * and should be added immediately.
4790 */
4791 super_types[mddev->major_version].
4792 validate_super(mddev, rdev);
4793 err = mddev->pers->hot_add_disk(mddev, rdev);
4794 if (err)
4795 unbind_rdev_from_array(rdev);
4796 }
4797 if (err)
4798 export_rdev(rdev);
4799 else
4800 sysfs_notify_dirent(rdev->sysfs_state);
4801
4802 md_update_sb(mddev, 1);
4803 if (mddev->degraded)
4804 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4805 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4806 md_wakeup_thread(mddev->thread);
4807 return err;
4808 }
4809
4810 /* otherwise, add_new_disk is only allowed
4811 * for major_version==0 superblocks
4812 */
4813 if (mddev->major_version != 0) {
4814 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4815 mdname(mddev));
4816 return -EINVAL;
4817 }
4818
4819 if (!(info->state & (1<<MD_DISK_FAULTY))) {
4820 int err;
4821 rdev = md_import_device(dev, -1, 0);
4822 if (IS_ERR(rdev)) {
4823 printk(KERN_WARNING
4824 "md: error, md_import_device() returned %ld\n",
4825 PTR_ERR(rdev));
4826 return PTR_ERR(rdev);
4827 }
4828 rdev->desc_nr = info->number;
4829 if (info->raid_disk < mddev->raid_disks)
4830 rdev->raid_disk = info->raid_disk;
4831 else
4832 rdev->raid_disk = -1;
4833
4834 if (rdev->raid_disk < mddev->raid_disks)
4835 if (info->state & (1<<MD_DISK_SYNC))
4836 set_bit(In_sync, &rdev->flags);
4837
4838 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4839 set_bit(WriteMostly, &rdev->flags);
4840
4841 if (!mddev->persistent) {
4842 printk(KERN_INFO "md: nonpersistent superblock ...\n");
4843 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4844 } else
4845 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4846 rdev->sectors = calc_num_sectors(rdev, mddev->chunk_size);
4847
4848 err = bind_rdev_to_array(rdev, mddev);
4849 if (err) {
4850 export_rdev(rdev);
4851 return err;
4852 }
4853 }
4854
4855 return 0;
4856 }
4857
4858 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4859 {
4860 char b[BDEVNAME_SIZE];
4861 mdk_rdev_t *rdev;
4862
4863 rdev = find_rdev(mddev, dev);
4864 if (!rdev)
4865 return -ENXIO;
4866
4867 if (rdev->raid_disk >= 0)
4868 goto busy;
4869
4870 kick_rdev_from_array(rdev);
4871 md_update_sb(mddev, 1);
4872 md_new_event(mddev);
4873
4874 return 0;
4875 busy:
4876 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4877 bdevname(rdev->bdev,b), mdname(mddev));
4878 return -EBUSY;
4879 }
4880
4881 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4882 {
4883 char b[BDEVNAME_SIZE];
4884 int err;
4885 mdk_rdev_t *rdev;
4886
4887 if (!mddev->pers)
4888 return -ENODEV;
4889
4890 if (mddev->major_version != 0) {
4891 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4892 " version-0 superblocks.\n",
4893 mdname(mddev));
4894 return -EINVAL;
4895 }
4896 if (!mddev->pers->hot_add_disk) {
4897 printk(KERN_WARNING
4898 "%s: personality does not support diskops!\n",
4899 mdname(mddev));
4900 return -EINVAL;
4901 }
4902
4903 rdev = md_import_device(dev, -1, 0);
4904 if (IS_ERR(rdev)) {
4905 printk(KERN_WARNING
4906 "md: error, md_import_device() returned %ld\n",
4907 PTR_ERR(rdev));
4908 return -EINVAL;
4909 }
4910
4911 if (mddev->persistent)
4912 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4913 else
4914 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4915
4916 rdev->sectors = calc_num_sectors(rdev, mddev->chunk_size);
4917
4918 if (test_bit(Faulty, &rdev->flags)) {
4919 printk(KERN_WARNING
4920 "md: can not hot-add faulty %s disk to %s!\n",
4921 bdevname(rdev->bdev,b), mdname(mddev));
4922 err = -EINVAL;
4923 goto abort_export;
4924 }
4925 clear_bit(In_sync, &rdev->flags);
4926 rdev->desc_nr = -1;
4927 rdev->saved_raid_disk = -1;
4928 err = bind_rdev_to_array(rdev, mddev);
4929 if (err)
4930 goto abort_export;
4931
4932 /*
4933 * The rest should better be atomic, we can have disk failures
4934 * noticed in interrupt contexts ...
4935 */
4936
4937 rdev->raid_disk = -1;
4938
4939 md_update_sb(mddev, 1);
4940
4941 /*
4942 * Kick recovery, maybe this spare has to be added to the
4943 * array immediately.
4944 */
4945 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4946 md_wakeup_thread(mddev->thread);
4947 md_new_event(mddev);
4948 return 0;
4949
4950 abort_export:
4951 export_rdev(rdev);
4952 return err;
4953 }
4954
4955 static int set_bitmap_file(mddev_t *mddev, int fd)
4956 {
4957 int err;
4958
4959 if (mddev->pers) {
4960 if (!mddev->pers->quiesce)
4961 return -EBUSY;
4962 if (mddev->recovery || mddev->sync_thread)
4963 return -EBUSY;
4964 /* we should be able to change the bitmap.. */
4965 }
4966
4967
4968 if (fd >= 0) {
4969 if (mddev->bitmap)
4970 return -EEXIST; /* cannot add when bitmap is present */
4971 mddev->bitmap_file = fget(fd);
4972
4973 if (mddev->bitmap_file == NULL) {
4974 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4975 mdname(mddev));
4976 return -EBADF;
4977 }
4978
4979 err = deny_bitmap_write_access(mddev->bitmap_file);
4980 if (err) {
4981 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4982 mdname(mddev));
4983 fput(mddev->bitmap_file);
4984 mddev->bitmap_file = NULL;
4985 return err;
4986 }
4987 mddev->bitmap_offset = 0; /* file overrides offset */
4988 } else if (mddev->bitmap == NULL)
4989 return -ENOENT; /* cannot remove what isn't there */
4990 err = 0;
4991 if (mddev->pers) {
4992 mddev->pers->quiesce(mddev, 1);
4993 if (fd >= 0)
4994 err = bitmap_create(mddev);
4995 if (fd < 0 || err) {
4996 bitmap_destroy(mddev);
4997 fd = -1; /* make sure to put the file */
4998 }
4999 mddev->pers->quiesce(mddev, 0);
5000 }
5001 if (fd < 0) {
5002 if (mddev->bitmap_file) {
5003 restore_bitmap_write_access(mddev->bitmap_file);
5004 fput(mddev->bitmap_file);
5005 }
5006 mddev->bitmap_file = NULL;
5007 }
5008
5009 return err;
5010 }
5011
5012 /*
5013 * set_array_info is used two different ways
5014 * The original usage is when creating a new array.
5015 * In this usage, raid_disks is > 0 and it together with
5016 * level, size, not_persistent,layout,chunksize determine the
5017 * shape of the array.
5018 * This will always create an array with a type-0.90.0 superblock.
5019 * The newer usage is when assembling an array.
5020 * In this case raid_disks will be 0, and the major_version field is
5021 * use to determine which style super-blocks are to be found on the devices.
5022 * The minor and patch _version numbers are also kept incase the
5023 * super_block handler wishes to interpret them.
5024 */
5025 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5026 {
5027
5028 if (info->raid_disks == 0) {
5029 /* just setting version number for superblock loading */
5030 if (info->major_version < 0 ||
5031 info->major_version >= ARRAY_SIZE(super_types) ||
5032 super_types[info->major_version].name == NULL) {
5033 /* maybe try to auto-load a module? */
5034 printk(KERN_INFO
5035 "md: superblock version %d not known\n",
5036 info->major_version);
5037 return -EINVAL;
5038 }
5039 mddev->major_version = info->major_version;
5040 mddev->minor_version = info->minor_version;
5041 mddev->patch_version = info->patch_version;
5042 mddev->persistent = !info->not_persistent;
5043 return 0;
5044 }
5045 mddev->major_version = MD_MAJOR_VERSION;
5046 mddev->minor_version = MD_MINOR_VERSION;
5047 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5048 mddev->ctime = get_seconds();
5049
5050 mddev->level = info->level;
5051 mddev->clevel[0] = 0;
5052 mddev->dev_sectors = 2 * (sector_t)info->size;
5053 mddev->raid_disks = info->raid_disks;
5054 /* don't set md_minor, it is determined by which /dev/md* was
5055 * openned
5056 */
5057 if (info->state & (1<<MD_SB_CLEAN))
5058 mddev->recovery_cp = MaxSector;
5059 else
5060 mddev->recovery_cp = 0;
5061 mddev->persistent = ! info->not_persistent;
5062 mddev->external = 0;
5063
5064 mddev->layout = info->layout;
5065 mddev->chunk_size = info->chunk_size;
5066
5067 mddev->max_disks = MD_SB_DISKS;
5068
5069 if (mddev->persistent)
5070 mddev->flags = 0;
5071 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5072
5073 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
5074 mddev->bitmap_offset = 0;
5075
5076 mddev->reshape_position = MaxSector;
5077
5078 /*
5079 * Generate a 128 bit UUID
5080 */
5081 get_random_bytes(mddev->uuid, 16);
5082
5083 mddev->new_level = mddev->level;
5084 mddev->new_chunk = mddev->chunk_size;
5085 mddev->new_layout = mddev->layout;
5086 mddev->delta_disks = 0;
5087
5088 return 0;
5089 }
5090
5091 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5092 {
5093 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5094
5095 if (mddev->external_size)
5096 return;
5097
5098 mddev->array_sectors = array_sectors;
5099 }
5100 EXPORT_SYMBOL(md_set_array_sectors);
5101
5102 static int update_size(mddev_t *mddev, sector_t num_sectors)
5103 {
5104 mdk_rdev_t *rdev;
5105 int rv;
5106 int fit = (num_sectors == 0);
5107
5108 if (mddev->pers->resize == NULL)
5109 return -EINVAL;
5110 /* The "num_sectors" is the number of sectors of each device that
5111 * is used. This can only make sense for arrays with redundancy.
5112 * linear and raid0 always use whatever space is available. We can only
5113 * consider changing this number if no resync or reconstruction is
5114 * happening, and if the new size is acceptable. It must fit before the
5115 * sb_start or, if that is <data_offset, it must fit before the size
5116 * of each device. If num_sectors is zero, we find the largest size
5117 * that fits.
5118
5119 */
5120 if (mddev->sync_thread)
5121 return -EBUSY;
5122 if (mddev->bitmap)
5123 /* Sorry, cannot grow a bitmap yet, just remove it,
5124 * grow, and re-add.
5125 */
5126 return -EBUSY;
5127 list_for_each_entry(rdev, &mddev->disks, same_set) {
5128 sector_t avail = rdev->sectors;
5129
5130 if (fit && (num_sectors == 0 || num_sectors > avail))
5131 num_sectors = avail;
5132 if (avail < num_sectors)
5133 return -ENOSPC;
5134 }
5135 rv = mddev->pers->resize(mddev, num_sectors);
5136 if (!rv) {
5137 struct block_device *bdev;
5138
5139 bdev = bdget_disk(mddev->gendisk, 0);
5140 if (bdev) {
5141 mutex_lock(&bdev->bd_inode->i_mutex);
5142 i_size_write(bdev->bd_inode,
5143 (loff_t)mddev->array_sectors << 9);
5144 mutex_unlock(&bdev->bd_inode->i_mutex);
5145 bdput(bdev);
5146 }
5147 }
5148 return rv;
5149 }
5150
5151 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5152 {
5153 int rv;
5154 /* change the number of raid disks */
5155 if (mddev->pers->check_reshape == NULL)
5156 return -EINVAL;
5157 if (raid_disks <= 0 ||
5158 raid_disks >= mddev->max_disks)
5159 return -EINVAL;
5160 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5161 return -EBUSY;
5162 mddev->delta_disks = raid_disks - mddev->raid_disks;
5163
5164 rv = mddev->pers->check_reshape(mddev);
5165 return rv;
5166 }
5167
5168
5169 /*
5170 * update_array_info is used to change the configuration of an
5171 * on-line array.
5172 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5173 * fields in the info are checked against the array.
5174 * Any differences that cannot be handled will cause an error.
5175 * Normally, only one change can be managed at a time.
5176 */
5177 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5178 {
5179 int rv = 0;
5180 int cnt = 0;
5181 int state = 0;
5182
5183 /* calculate expected state,ignoring low bits */
5184 if (mddev->bitmap && mddev->bitmap_offset)
5185 state |= (1 << MD_SB_BITMAP_PRESENT);
5186
5187 if (mddev->major_version != info->major_version ||
5188 mddev->minor_version != info->minor_version ||
5189 /* mddev->patch_version != info->patch_version || */
5190 mddev->ctime != info->ctime ||
5191 mddev->level != info->level ||
5192 /* mddev->layout != info->layout || */
5193 !mddev->persistent != info->not_persistent||
5194 mddev->chunk_size != info->chunk_size ||
5195 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5196 ((state^info->state) & 0xfffffe00)
5197 )
5198 return -EINVAL;
5199 /* Check there is only one change */
5200 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5201 cnt++;
5202 if (mddev->raid_disks != info->raid_disks)
5203 cnt++;
5204 if (mddev->layout != info->layout)
5205 cnt++;
5206 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5207 cnt++;
5208 if (cnt == 0)
5209 return 0;
5210 if (cnt > 1)
5211 return -EINVAL;
5212
5213 if (mddev->layout != info->layout) {
5214 /* Change layout
5215 * we don't need to do anything at the md level, the
5216 * personality will take care of it all.
5217 */
5218 if (mddev->pers->reconfig == NULL)
5219 return -EINVAL;
5220 else
5221 return mddev->pers->reconfig(mddev, info->layout, -1);
5222 }
5223 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5224 rv = update_size(mddev, (sector_t)info->size * 2);
5225
5226 if (mddev->raid_disks != info->raid_disks)
5227 rv = update_raid_disks(mddev, info->raid_disks);
5228
5229 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5230 if (mddev->pers->quiesce == NULL)
5231 return -EINVAL;
5232 if (mddev->recovery || mddev->sync_thread)
5233 return -EBUSY;
5234 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5235 /* add the bitmap */
5236 if (mddev->bitmap)
5237 return -EEXIST;
5238 if (mddev->default_bitmap_offset == 0)
5239 return -EINVAL;
5240 mddev->bitmap_offset = mddev->default_bitmap_offset;
5241 mddev->pers->quiesce(mddev, 1);
5242 rv = bitmap_create(mddev);
5243 if (rv)
5244 bitmap_destroy(mddev);
5245 mddev->pers->quiesce(mddev, 0);
5246 } else {
5247 /* remove the bitmap */
5248 if (!mddev->bitmap)
5249 return -ENOENT;
5250 if (mddev->bitmap->file)
5251 return -EINVAL;
5252 mddev->pers->quiesce(mddev, 1);
5253 bitmap_destroy(mddev);
5254 mddev->pers->quiesce(mddev, 0);
5255 mddev->bitmap_offset = 0;
5256 }
5257 }
5258 md_update_sb(mddev, 1);
5259 return rv;
5260 }
5261
5262 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5263 {
5264 mdk_rdev_t *rdev;
5265
5266 if (mddev->pers == NULL)
5267 return -ENODEV;
5268
5269 rdev = find_rdev(mddev, dev);
5270 if (!rdev)
5271 return -ENODEV;
5272
5273 md_error(mddev, rdev);
5274 return 0;
5275 }
5276
5277 /*
5278 * We have a problem here : there is no easy way to give a CHS
5279 * virtual geometry. We currently pretend that we have a 2 heads
5280 * 4 sectors (with a BIG number of cylinders...). This drives
5281 * dosfs just mad... ;-)
5282 */
5283 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5284 {
5285 mddev_t *mddev = bdev->bd_disk->private_data;
5286
5287 geo->heads = 2;
5288 geo->sectors = 4;
5289 geo->cylinders = get_capacity(mddev->gendisk) / 8;
5290 return 0;
5291 }
5292
5293 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5294 unsigned int cmd, unsigned long arg)
5295 {
5296 int err = 0;
5297 void __user *argp = (void __user *)arg;
5298 mddev_t *mddev = NULL;
5299
5300 if (!capable(CAP_SYS_ADMIN))
5301 return -EACCES;
5302
5303 /*
5304 * Commands dealing with the RAID driver but not any
5305 * particular array:
5306 */
5307 switch (cmd)
5308 {
5309 case RAID_VERSION:
5310 err = get_version(argp);
5311 goto done;
5312
5313 case PRINT_RAID_DEBUG:
5314 err = 0;
5315 md_print_devices();
5316 goto done;
5317
5318 #ifndef MODULE
5319 case RAID_AUTORUN:
5320 err = 0;
5321 autostart_arrays(arg);
5322 goto done;
5323 #endif
5324 default:;
5325 }
5326
5327 /*
5328 * Commands creating/starting a new array:
5329 */
5330
5331 mddev = bdev->bd_disk->private_data;
5332
5333 if (!mddev) {
5334 BUG();
5335 goto abort;
5336 }
5337
5338 err = mddev_lock(mddev);
5339 if (err) {
5340 printk(KERN_INFO
5341 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5342 err, cmd);
5343 goto abort;
5344 }
5345
5346 switch (cmd)
5347 {
5348 case SET_ARRAY_INFO:
5349 {
5350 mdu_array_info_t info;
5351 if (!arg)
5352 memset(&info, 0, sizeof(info));
5353 else if (copy_from_user(&info, argp, sizeof(info))) {
5354 err = -EFAULT;
5355 goto abort_unlock;
5356 }
5357 if (mddev->pers) {
5358 err = update_array_info(mddev, &info);
5359 if (err) {
5360 printk(KERN_WARNING "md: couldn't update"
5361 " array info. %d\n", err);
5362 goto abort_unlock;
5363 }
5364 goto done_unlock;
5365 }
5366 if (!list_empty(&mddev->disks)) {
5367 printk(KERN_WARNING
5368 "md: array %s already has disks!\n",
5369 mdname(mddev));
5370 err = -EBUSY;
5371 goto abort_unlock;
5372 }
5373 if (mddev->raid_disks) {
5374 printk(KERN_WARNING
5375 "md: array %s already initialised!\n",
5376 mdname(mddev));
5377 err = -EBUSY;
5378 goto abort_unlock;
5379 }
5380 err = set_array_info(mddev, &info);
5381 if (err) {
5382 printk(KERN_WARNING "md: couldn't set"
5383 " array info. %d\n", err);
5384 goto abort_unlock;
5385 }
5386 }
5387 goto done_unlock;
5388
5389 default:;
5390 }
5391
5392 /*
5393 * Commands querying/configuring an existing array:
5394 */
5395 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5396 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5397 if ((!mddev->raid_disks && !mddev->external)
5398 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5399 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5400 && cmd != GET_BITMAP_FILE) {
5401 err = -ENODEV;
5402 goto abort_unlock;
5403 }
5404
5405 /*
5406 * Commands even a read-only array can execute:
5407 */
5408 switch (cmd)
5409 {
5410 case GET_ARRAY_INFO:
5411 err = get_array_info(mddev, argp);
5412 goto done_unlock;
5413
5414 case GET_BITMAP_FILE:
5415 err = get_bitmap_file(mddev, argp);
5416 goto done_unlock;
5417
5418 case GET_DISK_INFO:
5419 err = get_disk_info(mddev, argp);
5420 goto done_unlock;
5421
5422 case RESTART_ARRAY_RW:
5423 err = restart_array(mddev);
5424 goto done_unlock;
5425
5426 case STOP_ARRAY:
5427 err = do_md_stop(mddev, 0, 1);
5428 goto done_unlock;
5429
5430 case STOP_ARRAY_RO:
5431 err = do_md_stop(mddev, 1, 1);
5432 goto done_unlock;
5433
5434 }
5435
5436 /*
5437 * The remaining ioctls are changing the state of the
5438 * superblock, so we do not allow them on read-only arrays.
5439 * However non-MD ioctls (e.g. get-size) will still come through
5440 * here and hit the 'default' below, so only disallow
5441 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5442 */
5443 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5444 if (mddev->ro == 2) {
5445 mddev->ro = 0;
5446 sysfs_notify_dirent(mddev->sysfs_state);
5447 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5448 md_wakeup_thread(mddev->thread);
5449 } else {
5450 err = -EROFS;
5451 goto abort_unlock;
5452 }
5453 }
5454
5455 switch (cmd)
5456 {
5457 case ADD_NEW_DISK:
5458 {
5459 mdu_disk_info_t info;
5460 if (copy_from_user(&info, argp, sizeof(info)))
5461 err = -EFAULT;
5462 else
5463 err = add_new_disk(mddev, &info);
5464 goto done_unlock;
5465 }
5466
5467 case HOT_REMOVE_DISK:
5468 err = hot_remove_disk(mddev, new_decode_dev(arg));
5469 goto done_unlock;
5470
5471 case HOT_ADD_DISK:
5472 err = hot_add_disk(mddev, new_decode_dev(arg));
5473 goto done_unlock;
5474
5475 case SET_DISK_FAULTY:
5476 err = set_disk_faulty(mddev, new_decode_dev(arg));
5477 goto done_unlock;
5478
5479 case RUN_ARRAY:
5480 err = do_md_run(mddev);
5481 goto done_unlock;
5482
5483 case SET_BITMAP_FILE:
5484 err = set_bitmap_file(mddev, (int)arg);
5485 goto done_unlock;
5486
5487 default:
5488 err = -EINVAL;
5489 goto abort_unlock;
5490 }
5491
5492 done_unlock:
5493 abort_unlock:
5494 if (mddev->hold_active == UNTIL_IOCTL &&
5495 err != -EINVAL)
5496 mddev->hold_active = 0;
5497 mddev_unlock(mddev);
5498
5499 return err;
5500 done:
5501 if (err)
5502 MD_BUG();
5503 abort:
5504 return err;
5505 }
5506
5507 static int md_open(struct block_device *bdev, fmode_t mode)
5508 {
5509 /*
5510 * Succeed if we can lock the mddev, which confirms that
5511 * it isn't being stopped right now.
5512 */
5513 mddev_t *mddev = mddev_find(bdev->bd_dev);
5514 int err;
5515
5516 if (mddev->gendisk != bdev->bd_disk) {
5517 /* we are racing with mddev_put which is discarding this
5518 * bd_disk.
5519 */
5520 mddev_put(mddev);
5521 /* Wait until bdev->bd_disk is definitely gone */
5522 flush_scheduled_work();
5523 /* Then retry the open from the top */
5524 return -ERESTARTSYS;
5525 }
5526 BUG_ON(mddev != bdev->bd_disk->private_data);
5527
5528 if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
5529 goto out;
5530
5531 err = 0;
5532 atomic_inc(&mddev->openers);
5533 mddev_unlock(mddev);
5534
5535 check_disk_change(bdev);
5536 out:
5537 return err;
5538 }
5539
5540 static int md_release(struct gendisk *disk, fmode_t mode)
5541 {
5542 mddev_t *mddev = disk->private_data;
5543
5544 BUG_ON(!mddev);
5545 atomic_dec(&mddev->openers);
5546 mddev_put(mddev);
5547
5548 return 0;
5549 }
5550
5551 static int md_media_changed(struct gendisk *disk)
5552 {
5553 mddev_t *mddev = disk->private_data;
5554
5555 return mddev->changed;
5556 }
5557
5558 static int md_revalidate(struct gendisk *disk)
5559 {
5560 mddev_t *mddev = disk->private_data;
5561
5562 mddev->changed = 0;
5563 return 0;
5564 }
5565 static struct block_device_operations md_fops =
5566 {
5567 .owner = THIS_MODULE,
5568 .open = md_open,
5569 .release = md_release,
5570 .ioctl = md_ioctl,
5571 .getgeo = md_getgeo,
5572 .media_changed = md_media_changed,
5573 .revalidate_disk= md_revalidate,
5574 };
5575
5576 static int md_thread(void * arg)
5577 {
5578 mdk_thread_t *thread = arg;
5579
5580 /*
5581 * md_thread is a 'system-thread', it's priority should be very
5582 * high. We avoid resource deadlocks individually in each
5583 * raid personality. (RAID5 does preallocation) We also use RR and
5584 * the very same RT priority as kswapd, thus we will never get
5585 * into a priority inversion deadlock.
5586 *
5587 * we definitely have to have equal or higher priority than
5588 * bdflush, otherwise bdflush will deadlock if there are too
5589 * many dirty RAID5 blocks.
5590 */
5591
5592 allow_signal(SIGKILL);
5593 while (!kthread_should_stop()) {
5594
5595 /* We need to wait INTERRUPTIBLE so that
5596 * we don't add to the load-average.
5597 * That means we need to be sure no signals are
5598 * pending
5599 */
5600 if (signal_pending(current))
5601 flush_signals(current);
5602
5603 wait_event_interruptible_timeout
5604 (thread->wqueue,
5605 test_bit(THREAD_WAKEUP, &thread->flags)
5606 || kthread_should_stop(),
5607 thread->timeout);
5608
5609 clear_bit(THREAD_WAKEUP, &thread->flags);
5610
5611 thread->run(thread->mddev);
5612 }
5613
5614 return 0;
5615 }
5616
5617 void md_wakeup_thread(mdk_thread_t *thread)
5618 {
5619 if (thread) {
5620 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5621 set_bit(THREAD_WAKEUP, &thread->flags);
5622 wake_up(&thread->wqueue);
5623 }
5624 }
5625
5626 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5627 const char *name)
5628 {
5629 mdk_thread_t *thread;
5630
5631 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5632 if (!thread)
5633 return NULL;
5634
5635 init_waitqueue_head(&thread->wqueue);
5636
5637 thread->run = run;
5638 thread->mddev = mddev;
5639 thread->timeout = MAX_SCHEDULE_TIMEOUT;
5640 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
5641 if (IS_ERR(thread->tsk)) {
5642 kfree(thread);
5643 return NULL;
5644 }
5645 return thread;
5646 }
5647
5648 void md_unregister_thread(mdk_thread_t *thread)
5649 {
5650 if (!thread)
5651 return;
5652 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5653
5654 kthread_stop(thread->tsk);
5655 kfree(thread);
5656 }
5657
5658 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5659 {
5660 if (!mddev) {
5661 MD_BUG();
5662 return;
5663 }
5664
5665 if (!rdev || test_bit(Faulty, &rdev->flags))
5666 return;
5667
5668 if (mddev->external)
5669 set_bit(Blocked, &rdev->flags);
5670 /*
5671 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5672 mdname(mddev),
5673 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5674 __builtin_return_address(0),__builtin_return_address(1),
5675 __builtin_return_address(2),__builtin_return_address(3));
5676 */
5677 if (!mddev->pers)
5678 return;
5679 if (!mddev->pers->error_handler)
5680 return;
5681 mddev->pers->error_handler(mddev,rdev);
5682 if (mddev->degraded)
5683 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5684 set_bit(StateChanged, &rdev->flags);
5685 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5686 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5687 md_wakeup_thread(mddev->thread);
5688 md_new_event_inintr(mddev);
5689 }
5690
5691 /* seq_file implementation /proc/mdstat */
5692
5693 static void status_unused(struct seq_file *seq)
5694 {
5695 int i = 0;
5696 mdk_rdev_t *rdev;
5697
5698 seq_printf(seq, "unused devices: ");
5699
5700 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
5701 char b[BDEVNAME_SIZE];
5702 i++;
5703 seq_printf(seq, "%s ",
5704 bdevname(rdev->bdev,b));
5705 }
5706 if (!i)
5707 seq_printf(seq, "<none>");
5708
5709 seq_printf(seq, "\n");
5710 }
5711
5712
5713 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5714 {
5715 sector_t max_sectors, resync, res;
5716 unsigned long dt, db;
5717 sector_t rt;
5718 int scale;
5719 unsigned int per_milli;
5720
5721 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
5722
5723 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5724 max_sectors = mddev->resync_max_sectors;
5725 else
5726 max_sectors = mddev->dev_sectors;
5727
5728 /*
5729 * Should not happen.
5730 */
5731 if (!max_sectors) {
5732 MD_BUG();
5733 return;
5734 }
5735 /* Pick 'scale' such that (resync>>scale)*1000 will fit
5736 * in a sector_t, and (max_sectors>>scale) will fit in a
5737 * u32, as those are the requirements for sector_div.
5738 * Thus 'scale' must be at least 10
5739 */
5740 scale = 10;
5741 if (sizeof(sector_t) > sizeof(unsigned long)) {
5742 while ( max_sectors/2 > (1ULL<<(scale+32)))
5743 scale++;
5744 }
5745 res = (resync>>scale)*1000;
5746 sector_div(res, (u32)((max_sectors>>scale)+1));
5747
5748 per_milli = res;
5749 {
5750 int i, x = per_milli/50, y = 20-x;
5751 seq_printf(seq, "[");
5752 for (i = 0; i < x; i++)
5753 seq_printf(seq, "=");
5754 seq_printf(seq, ">");
5755 for (i = 0; i < y; i++)
5756 seq_printf(seq, ".");
5757 seq_printf(seq, "] ");
5758 }
5759 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5760 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5761 "reshape" :
5762 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5763 "check" :
5764 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5765 "resync" : "recovery"))),
5766 per_milli/10, per_milli % 10,
5767 (unsigned long long) resync/2,
5768 (unsigned long long) max_sectors/2);
5769
5770 /*
5771 * dt: time from mark until now
5772 * db: blocks written from mark until now
5773 * rt: remaining time
5774 *
5775 * rt is a sector_t, so could be 32bit or 64bit.
5776 * So we divide before multiply in case it is 32bit and close
5777 * to the limit.
5778 * We scale the divisor (db) by 32 to avoid loosing precision
5779 * near the end of resync when the number of remaining sectors
5780 * is close to 'db'.
5781 * We then divide rt by 32 after multiplying by db to compensate.
5782 * The '+1' avoids division by zero if db is very small.
5783 */
5784 dt = ((jiffies - mddev->resync_mark) / HZ);
5785 if (!dt) dt++;
5786 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5787 - mddev->resync_mark_cnt;
5788
5789 rt = max_sectors - resync; /* number of remaining sectors */
5790 sector_div(rt, db/32+1);
5791 rt *= dt;
5792 rt >>= 5;
5793
5794 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
5795 ((unsigned long)rt % 60)/6);
5796
5797 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5798 }
5799
5800 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5801 {
5802 struct list_head *tmp;
5803 loff_t l = *pos;
5804 mddev_t *mddev;
5805
5806 if (l >= 0x10000)
5807 return NULL;
5808 if (!l--)
5809 /* header */
5810 return (void*)1;
5811
5812 spin_lock(&all_mddevs_lock);
5813 list_for_each(tmp,&all_mddevs)
5814 if (!l--) {
5815 mddev = list_entry(tmp, mddev_t, all_mddevs);
5816 mddev_get(mddev);
5817 spin_unlock(&all_mddevs_lock);
5818 return mddev;
5819 }
5820 spin_unlock(&all_mddevs_lock);
5821 if (!l--)
5822 return (void*)2;/* tail */
5823 return NULL;
5824 }
5825
5826 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5827 {
5828 struct list_head *tmp;
5829 mddev_t *next_mddev, *mddev = v;
5830
5831 ++*pos;
5832 if (v == (void*)2)
5833 return NULL;
5834
5835 spin_lock(&all_mddevs_lock);
5836 if (v == (void*)1)
5837 tmp = all_mddevs.next;
5838 else
5839 tmp = mddev->all_mddevs.next;
5840 if (tmp != &all_mddevs)
5841 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5842 else {
5843 next_mddev = (void*)2;
5844 *pos = 0x10000;
5845 }
5846 spin_unlock(&all_mddevs_lock);
5847
5848 if (v != (void*)1)
5849 mddev_put(mddev);
5850 return next_mddev;
5851
5852 }
5853
5854 static void md_seq_stop(struct seq_file *seq, void *v)
5855 {
5856 mddev_t *mddev = v;
5857
5858 if (mddev && v != (void*)1 && v != (void*)2)
5859 mddev_put(mddev);
5860 }
5861
5862 struct mdstat_info {
5863 int event;
5864 };
5865
5866 static int md_seq_show(struct seq_file *seq, void *v)
5867 {
5868 mddev_t *mddev = v;
5869 sector_t sectors;
5870 mdk_rdev_t *rdev;
5871 struct mdstat_info *mi = seq->private;
5872 struct bitmap *bitmap;
5873
5874 if (v == (void*)1) {
5875 struct mdk_personality *pers;
5876 seq_printf(seq, "Personalities : ");
5877 spin_lock(&pers_lock);
5878 list_for_each_entry(pers, &pers_list, list)
5879 seq_printf(seq, "[%s] ", pers->name);
5880
5881 spin_unlock(&pers_lock);
5882 seq_printf(seq, "\n");
5883 mi->event = atomic_read(&md_event_count);
5884 return 0;
5885 }
5886 if (v == (void*)2) {
5887 status_unused(seq);
5888 return 0;
5889 }
5890
5891 if (mddev_lock(mddev) < 0)
5892 return -EINTR;
5893
5894 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5895 seq_printf(seq, "%s : %sactive", mdname(mddev),
5896 mddev->pers ? "" : "in");
5897 if (mddev->pers) {
5898 if (mddev->ro==1)
5899 seq_printf(seq, " (read-only)");
5900 if (mddev->ro==2)
5901 seq_printf(seq, " (auto-read-only)");
5902 seq_printf(seq, " %s", mddev->pers->name);
5903 }
5904
5905 sectors = 0;
5906 list_for_each_entry(rdev, &mddev->disks, same_set) {
5907 char b[BDEVNAME_SIZE];
5908 seq_printf(seq, " %s[%d]",
5909 bdevname(rdev->bdev,b), rdev->desc_nr);
5910 if (test_bit(WriteMostly, &rdev->flags))
5911 seq_printf(seq, "(W)");
5912 if (test_bit(Faulty, &rdev->flags)) {
5913 seq_printf(seq, "(F)");
5914 continue;
5915 } else if (rdev->raid_disk < 0)
5916 seq_printf(seq, "(S)"); /* spare */
5917 sectors += rdev->sectors;
5918 }
5919
5920 if (!list_empty(&mddev->disks)) {
5921 if (mddev->pers)
5922 seq_printf(seq, "\n %llu blocks",
5923 (unsigned long long)
5924 mddev->array_sectors / 2);
5925 else
5926 seq_printf(seq, "\n %llu blocks",
5927 (unsigned long long)sectors / 2);
5928 }
5929 if (mddev->persistent) {
5930 if (mddev->major_version != 0 ||
5931 mddev->minor_version != 90) {
5932 seq_printf(seq," super %d.%d",
5933 mddev->major_version,
5934 mddev->minor_version);
5935 }
5936 } else if (mddev->external)
5937 seq_printf(seq, " super external:%s",
5938 mddev->metadata_type);
5939 else
5940 seq_printf(seq, " super non-persistent");
5941
5942 if (mddev->pers) {
5943 mddev->pers->status(seq, mddev);
5944 seq_printf(seq, "\n ");
5945 if (mddev->pers->sync_request) {
5946 if (mddev->curr_resync > 2) {
5947 status_resync(seq, mddev);
5948 seq_printf(seq, "\n ");
5949 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5950 seq_printf(seq, "\tresync=DELAYED\n ");
5951 else if (mddev->recovery_cp < MaxSector)
5952 seq_printf(seq, "\tresync=PENDING\n ");
5953 }
5954 } else
5955 seq_printf(seq, "\n ");
5956
5957 if ((bitmap = mddev->bitmap)) {
5958 unsigned long chunk_kb;
5959 unsigned long flags;
5960 spin_lock_irqsave(&bitmap->lock, flags);
5961 chunk_kb = bitmap->chunksize >> 10;
5962 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5963 "%lu%s chunk",
5964 bitmap->pages - bitmap->missing_pages,
5965 bitmap->pages,
5966 (bitmap->pages - bitmap->missing_pages)
5967 << (PAGE_SHIFT - 10),
5968 chunk_kb ? chunk_kb : bitmap->chunksize,
5969 chunk_kb ? "KB" : "B");
5970 if (bitmap->file) {
5971 seq_printf(seq, ", file: ");
5972 seq_path(seq, &bitmap->file->f_path, " \t\n");
5973 }
5974
5975 seq_printf(seq, "\n");
5976 spin_unlock_irqrestore(&bitmap->lock, flags);
5977 }
5978
5979 seq_printf(seq, "\n");
5980 }
5981 mddev_unlock(mddev);
5982
5983 return 0;
5984 }
5985
5986 static const struct seq_operations md_seq_ops = {
5987 .start = md_seq_start,
5988 .next = md_seq_next,
5989 .stop = md_seq_stop,
5990 .show = md_seq_show,
5991 };
5992
5993 static int md_seq_open(struct inode *inode, struct file *file)
5994 {
5995 int error;
5996 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5997 if (mi == NULL)
5998 return -ENOMEM;
5999
6000 error = seq_open(file, &md_seq_ops);
6001 if (error)
6002 kfree(mi);
6003 else {
6004 struct seq_file *p = file->private_data;
6005 p->private = mi;
6006 mi->event = atomic_read(&md_event_count);
6007 }
6008 return error;
6009 }
6010
6011 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6012 {
6013 struct seq_file *m = filp->private_data;
6014 struct mdstat_info *mi = m->private;
6015 int mask;
6016
6017 poll_wait(filp, &md_event_waiters, wait);
6018
6019 /* always allow read */
6020 mask = POLLIN | POLLRDNORM;
6021
6022 if (mi->event != atomic_read(&md_event_count))
6023 mask |= POLLERR | POLLPRI;
6024 return mask;
6025 }
6026
6027 static const struct file_operations md_seq_fops = {
6028 .owner = THIS_MODULE,
6029 .open = md_seq_open,
6030 .read = seq_read,
6031 .llseek = seq_lseek,
6032 .release = seq_release_private,
6033 .poll = mdstat_poll,
6034 };
6035
6036 int register_md_personality(struct mdk_personality *p)
6037 {
6038 spin_lock(&pers_lock);
6039 list_add_tail(&p->list, &pers_list);
6040 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6041 spin_unlock(&pers_lock);
6042 return 0;
6043 }
6044
6045 int unregister_md_personality(struct mdk_personality *p)
6046 {
6047 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6048 spin_lock(&pers_lock);
6049 list_del_init(&p->list);
6050 spin_unlock(&pers_lock);
6051 return 0;
6052 }
6053
6054 static int is_mddev_idle(mddev_t *mddev, int init)
6055 {
6056 mdk_rdev_t * rdev;
6057 int idle;
6058 int curr_events;
6059
6060 idle = 1;
6061 rcu_read_lock();
6062 rdev_for_each_rcu(rdev, mddev) {
6063 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6064 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6065 (int)part_stat_read(&disk->part0, sectors[1]) -
6066 atomic_read(&disk->sync_io);
6067 /* sync IO will cause sync_io to increase before the disk_stats
6068 * as sync_io is counted when a request starts, and
6069 * disk_stats is counted when it completes.
6070 * So resync activity will cause curr_events to be smaller than
6071 * when there was no such activity.
6072 * non-sync IO will cause disk_stat to increase without
6073 * increasing sync_io so curr_events will (eventually)
6074 * be larger than it was before. Once it becomes
6075 * substantially larger, the test below will cause
6076 * the array to appear non-idle, and resync will slow
6077 * down.
6078 * If there is a lot of outstanding resync activity when
6079 * we set last_event to curr_events, then all that activity
6080 * completing might cause the array to appear non-idle
6081 * and resync will be slowed down even though there might
6082 * not have been non-resync activity. This will only
6083 * happen once though. 'last_events' will soon reflect
6084 * the state where there is little or no outstanding
6085 * resync requests, and further resync activity will
6086 * always make curr_events less than last_events.
6087 *
6088 */
6089 if (init || curr_events - rdev->last_events > 64) {
6090 rdev->last_events = curr_events;
6091 idle = 0;
6092 }
6093 }
6094 rcu_read_unlock();
6095 return idle;
6096 }
6097
6098 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6099 {
6100 /* another "blocks" (512byte) blocks have been synced */
6101 atomic_sub(blocks, &mddev->recovery_active);
6102 wake_up(&mddev->recovery_wait);
6103 if (!ok) {
6104 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6105 md_wakeup_thread(mddev->thread);
6106 // stop recovery, signal do_sync ....
6107 }
6108 }
6109
6110
6111 /* md_write_start(mddev, bi)
6112 * If we need to update some array metadata (e.g. 'active' flag
6113 * in superblock) before writing, schedule a superblock update
6114 * and wait for it to complete.
6115 */
6116 void md_write_start(mddev_t *mddev, struct bio *bi)
6117 {
6118 int did_change = 0;
6119 if (bio_data_dir(bi) != WRITE)
6120 return;
6121
6122 BUG_ON(mddev->ro == 1);
6123 if (mddev->ro == 2) {
6124 /* need to switch to read/write */
6125 mddev->ro = 0;
6126 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6127 md_wakeup_thread(mddev->thread);
6128 md_wakeup_thread(mddev->sync_thread);
6129 did_change = 1;
6130 }
6131 atomic_inc(&mddev->writes_pending);
6132 if (mddev->safemode == 1)
6133 mddev->safemode = 0;
6134 if (mddev->in_sync) {
6135 spin_lock_irq(&mddev->write_lock);
6136 if (mddev->in_sync) {
6137 mddev->in_sync = 0;
6138 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6139 md_wakeup_thread(mddev->thread);
6140 did_change = 1;
6141 }
6142 spin_unlock_irq(&mddev->write_lock);
6143 }
6144 if (did_change)
6145 sysfs_notify_dirent(mddev->sysfs_state);
6146 wait_event(mddev->sb_wait,
6147 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
6148 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6149 }
6150
6151 void md_write_end(mddev_t *mddev)
6152 {
6153 if (atomic_dec_and_test(&mddev->writes_pending)) {
6154 if (mddev->safemode == 2)
6155 md_wakeup_thread(mddev->thread);
6156 else if (mddev->safemode_delay)
6157 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6158 }
6159 }
6160
6161 /* md_allow_write(mddev)
6162 * Calling this ensures that the array is marked 'active' so that writes
6163 * may proceed without blocking. It is important to call this before
6164 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6165 * Must be called with mddev_lock held.
6166 *
6167 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6168 * is dropped, so return -EAGAIN after notifying userspace.
6169 */
6170 int md_allow_write(mddev_t *mddev)
6171 {
6172 if (!mddev->pers)
6173 return 0;
6174 if (mddev->ro)
6175 return 0;
6176 if (!mddev->pers->sync_request)
6177 return 0;
6178
6179 spin_lock_irq(&mddev->write_lock);
6180 if (mddev->in_sync) {
6181 mddev->in_sync = 0;
6182 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6183 if (mddev->safemode_delay &&
6184 mddev->safemode == 0)
6185 mddev->safemode = 1;
6186 spin_unlock_irq(&mddev->write_lock);
6187 md_update_sb(mddev, 0);
6188 sysfs_notify_dirent(mddev->sysfs_state);
6189 } else
6190 spin_unlock_irq(&mddev->write_lock);
6191
6192 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
6193 return -EAGAIN;
6194 else
6195 return 0;
6196 }
6197 EXPORT_SYMBOL_GPL(md_allow_write);
6198
6199 #define SYNC_MARKS 10
6200 #define SYNC_MARK_STEP (3*HZ)
6201 void md_do_sync(mddev_t *mddev)
6202 {
6203 mddev_t *mddev2;
6204 unsigned int currspeed = 0,
6205 window;
6206 sector_t max_sectors,j, io_sectors;
6207 unsigned long mark[SYNC_MARKS];
6208 sector_t mark_cnt[SYNC_MARKS];
6209 int last_mark,m;
6210 struct list_head *tmp;
6211 sector_t last_check;
6212 int skipped = 0;
6213 mdk_rdev_t *rdev;
6214 char *desc;
6215
6216 /* just incase thread restarts... */
6217 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6218 return;
6219 if (mddev->ro) /* never try to sync a read-only array */
6220 return;
6221
6222 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6223 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6224 desc = "data-check";
6225 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6226 desc = "requested-resync";
6227 else
6228 desc = "resync";
6229 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6230 desc = "reshape";
6231 else
6232 desc = "recovery";
6233
6234 /* we overload curr_resync somewhat here.
6235 * 0 == not engaged in resync at all
6236 * 2 == checking that there is no conflict with another sync
6237 * 1 == like 2, but have yielded to allow conflicting resync to
6238 * commense
6239 * other == active in resync - this many blocks
6240 *
6241 * Before starting a resync we must have set curr_resync to
6242 * 2, and then checked that every "conflicting" array has curr_resync
6243 * less than ours. When we find one that is the same or higher
6244 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6245 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6246 * This will mean we have to start checking from the beginning again.
6247 *
6248 */
6249
6250 do {
6251 mddev->curr_resync = 2;
6252
6253 try_again:
6254 if (kthread_should_stop()) {
6255 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6256 goto skip;
6257 }
6258 for_each_mddev(mddev2, tmp) {
6259 if (mddev2 == mddev)
6260 continue;
6261 if (!mddev->parallel_resync
6262 && mddev2->curr_resync
6263 && match_mddev_units(mddev, mddev2)) {
6264 DEFINE_WAIT(wq);
6265 if (mddev < mddev2 && mddev->curr_resync == 2) {
6266 /* arbitrarily yield */
6267 mddev->curr_resync = 1;
6268 wake_up(&resync_wait);
6269 }
6270 if (mddev > mddev2 && mddev->curr_resync == 1)
6271 /* no need to wait here, we can wait the next
6272 * time 'round when curr_resync == 2
6273 */
6274 continue;
6275 /* We need to wait 'interruptible' so as not to
6276 * contribute to the load average, and not to
6277 * be caught by 'softlockup'
6278 */
6279 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6280 if (!kthread_should_stop() &&
6281 mddev2->curr_resync >= mddev->curr_resync) {
6282 printk(KERN_INFO "md: delaying %s of %s"
6283 " until %s has finished (they"
6284 " share one or more physical units)\n",
6285 desc, mdname(mddev), mdname(mddev2));
6286 mddev_put(mddev2);
6287 if (signal_pending(current))
6288 flush_signals(current);
6289 schedule();
6290 finish_wait(&resync_wait, &wq);
6291 goto try_again;
6292 }
6293 finish_wait(&resync_wait, &wq);
6294 }
6295 }
6296 } while (mddev->curr_resync < 2);
6297
6298 j = 0;
6299 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6300 /* resync follows the size requested by the personality,
6301 * which defaults to physical size, but can be virtual size
6302 */
6303 max_sectors = mddev->resync_max_sectors;
6304 mddev->resync_mismatches = 0;
6305 /* we don't use the checkpoint if there's a bitmap */
6306 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6307 j = mddev->resync_min;
6308 else if (!mddev->bitmap)
6309 j = mddev->recovery_cp;
6310
6311 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6312 max_sectors = mddev->dev_sectors;
6313 else {
6314 /* recovery follows the physical size of devices */
6315 max_sectors = mddev->dev_sectors;
6316 j = MaxSector;
6317 list_for_each_entry(rdev, &mddev->disks, same_set)
6318 if (rdev->raid_disk >= 0 &&
6319 !test_bit(Faulty, &rdev->flags) &&
6320 !test_bit(In_sync, &rdev->flags) &&
6321 rdev->recovery_offset < j)
6322 j = rdev->recovery_offset;
6323 }
6324
6325 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6326 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6327 " %d KB/sec/disk.\n", speed_min(mddev));
6328 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6329 "(but not more than %d KB/sec) for %s.\n",
6330 speed_max(mddev), desc);
6331
6332 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6333
6334 io_sectors = 0;
6335 for (m = 0; m < SYNC_MARKS; m++) {
6336 mark[m] = jiffies;
6337 mark_cnt[m] = io_sectors;
6338 }
6339 last_mark = 0;
6340 mddev->resync_mark = mark[last_mark];
6341 mddev->resync_mark_cnt = mark_cnt[last_mark];
6342
6343 /*
6344 * Tune reconstruction:
6345 */
6346 window = 32*(PAGE_SIZE/512);
6347 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6348 window/2,(unsigned long long) max_sectors/2);
6349
6350 atomic_set(&mddev->recovery_active, 0);
6351 last_check = 0;
6352
6353 if (j>2) {
6354 printk(KERN_INFO
6355 "md: resuming %s of %s from checkpoint.\n",
6356 desc, mdname(mddev));
6357 mddev->curr_resync = j;
6358 }
6359
6360 while (j < max_sectors) {
6361 sector_t sectors;
6362
6363 skipped = 0;
6364
6365 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6366 ((mddev->curr_resync > mddev->curr_resync_completed &&
6367 (mddev->curr_resync - mddev->curr_resync_completed)
6368 > (max_sectors >> 4)) ||
6369 (j - mddev->curr_resync_completed)*2
6370 >= mddev->resync_max - mddev->curr_resync_completed
6371 )) {
6372 /* time to update curr_resync_completed */
6373 blk_unplug(mddev->queue);
6374 wait_event(mddev->recovery_wait,
6375 atomic_read(&mddev->recovery_active) == 0);
6376 mddev->curr_resync_completed =
6377 mddev->curr_resync;
6378 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6379 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6380 }
6381
6382 if (j >= mddev->resync_max)
6383 wait_event(mddev->recovery_wait,
6384 mddev->resync_max > j
6385 || kthread_should_stop());
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(mddev) != 0)
6721 /* Cannot proceed */
6722 goto unlock;
6723 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6724 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6725 } else if ((spares = remove_and_add_spares(mddev))) {
6726 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6727 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6728 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
6729 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6730 } else if (mddev->recovery_cp < MaxSector) {
6731 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6732 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6733 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6734 /* nothing to be done ... */
6735 goto unlock;
6736
6737 if (mddev->pers->sync_request) {
6738 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6739 /* We are adding a device or devices to an array
6740 * which has the bitmap stored on all devices.
6741 * So make sure all bitmap pages get written
6742 */
6743 bitmap_write_all(mddev->bitmap);
6744 }
6745 mddev->sync_thread = md_register_thread(md_do_sync,
6746 mddev,
6747 "%s_resync");
6748 if (!mddev->sync_thread) {
6749 printk(KERN_ERR "%s: could not start resync"
6750 " thread...\n",
6751 mdname(mddev));
6752 /* leave the spares where they are, it shouldn't hurt */
6753 mddev->recovery = 0;
6754 } else
6755 md_wakeup_thread(mddev->sync_thread);
6756 sysfs_notify_dirent(mddev->sysfs_action);
6757 md_new_event(mddev);
6758 }
6759 unlock:
6760 if (!mddev->sync_thread) {
6761 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6762 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
6763 &mddev->recovery))
6764 if (mddev->sysfs_action)
6765 sysfs_notify_dirent(mddev->sysfs_action);
6766 }
6767 mddev_unlock(mddev);
6768 }
6769 }
6770
6771 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
6772 {
6773 sysfs_notify_dirent(rdev->sysfs_state);
6774 wait_event_timeout(rdev->blocked_wait,
6775 !test_bit(Blocked, &rdev->flags),
6776 msecs_to_jiffies(5000));
6777 rdev_dec_pending(rdev, mddev);
6778 }
6779 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
6780
6781 static int md_notify_reboot(struct notifier_block *this,
6782 unsigned long code, void *x)
6783 {
6784 struct list_head *tmp;
6785 mddev_t *mddev;
6786
6787 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
6788
6789 printk(KERN_INFO "md: stopping all md devices.\n");
6790
6791 for_each_mddev(mddev, tmp)
6792 if (mddev_trylock(mddev)) {
6793 /* Force a switch to readonly even array
6794 * appears to still be in use. Hence
6795 * the '100'.
6796 */
6797 do_md_stop(mddev, 1, 100);
6798 mddev_unlock(mddev);
6799 }
6800 /*
6801 * certain more exotic SCSI devices are known to be
6802 * volatile wrt too early system reboots. While the
6803 * right place to handle this issue is the given
6804 * driver, we do want to have a safe RAID driver ...
6805 */
6806 mdelay(1000*1);
6807 }
6808 return NOTIFY_DONE;
6809 }
6810
6811 static struct notifier_block md_notifier = {
6812 .notifier_call = md_notify_reboot,
6813 .next = NULL,
6814 .priority = INT_MAX, /* before any real devices */
6815 };
6816
6817 static void md_geninit(void)
6818 {
6819 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
6820
6821 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
6822 }
6823
6824 static int __init md_init(void)
6825 {
6826 if (register_blkdev(MD_MAJOR, "md"))
6827 return -1;
6828 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
6829 unregister_blkdev(MD_MAJOR, "md");
6830 return -1;
6831 }
6832 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
6833 md_probe, NULL, NULL);
6834 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
6835 md_probe, NULL, NULL);
6836
6837 register_reboot_notifier(&md_notifier);
6838 raid_table_header = register_sysctl_table(raid_root_table);
6839
6840 md_geninit();
6841 return 0;
6842 }
6843
6844
6845 #ifndef MODULE
6846
6847 /*
6848 * Searches all registered partitions for autorun RAID arrays
6849 * at boot time.
6850 */
6851
6852 static LIST_HEAD(all_detected_devices);
6853 struct detected_devices_node {
6854 struct list_head list;
6855 dev_t dev;
6856 };
6857
6858 void md_autodetect_dev(dev_t dev)
6859 {
6860 struct detected_devices_node *node_detected_dev;
6861
6862 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6863 if (node_detected_dev) {
6864 node_detected_dev->dev = dev;
6865 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6866 } else {
6867 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6868 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6869 }
6870 }
6871
6872
6873 static void autostart_arrays(int part)
6874 {
6875 mdk_rdev_t *rdev;
6876 struct detected_devices_node *node_detected_dev;
6877 dev_t dev;
6878 int i_scanned, i_passed;
6879
6880 i_scanned = 0;
6881 i_passed = 0;
6882
6883 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6884
6885 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6886 i_scanned++;
6887 node_detected_dev = list_entry(all_detected_devices.next,
6888 struct detected_devices_node, list);
6889 list_del(&node_detected_dev->list);
6890 dev = node_detected_dev->dev;
6891 kfree(node_detected_dev);
6892 rdev = md_import_device(dev,0, 90);
6893 if (IS_ERR(rdev))
6894 continue;
6895
6896 if (test_bit(Faulty, &rdev->flags)) {
6897 MD_BUG();
6898 continue;
6899 }
6900 set_bit(AutoDetected, &rdev->flags);
6901 list_add(&rdev->same_set, &pending_raid_disks);
6902 i_passed++;
6903 }
6904
6905 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6906 i_scanned, i_passed);
6907
6908 autorun_devices(part);
6909 }
6910
6911 #endif /* !MODULE */
6912
6913 static __exit void md_exit(void)
6914 {
6915 mddev_t *mddev;
6916 struct list_head *tmp;
6917
6918 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
6919 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6920
6921 unregister_blkdev(MD_MAJOR,"md");
6922 unregister_blkdev(mdp_major, "mdp");
6923 unregister_reboot_notifier(&md_notifier);
6924 unregister_sysctl_table(raid_table_header);
6925 remove_proc_entry("mdstat", NULL);
6926 for_each_mddev(mddev, tmp) {
6927 export_array(mddev);
6928 mddev->hold_active = 0;
6929 }
6930 }
6931
6932 subsys_initcall(md_init);
6933 module_exit(md_exit)
6934
6935 static int get_ro(char *buffer, struct kernel_param *kp)
6936 {
6937 return sprintf(buffer, "%d", start_readonly);
6938 }
6939 static int set_ro(const char *val, struct kernel_param *kp)
6940 {
6941 char *e;
6942 int num = simple_strtoul(val, &e, 10);
6943 if (*val && (*e == '\0' || *e == '\n')) {
6944 start_readonly = num;
6945 return 0;
6946 }
6947 return -EINVAL;
6948 }
6949
6950 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6951 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6952
6953 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
6954
6955 EXPORT_SYMBOL(register_md_personality);
6956 EXPORT_SYMBOL(unregister_md_personality);
6957 EXPORT_SYMBOL(md_error);
6958 EXPORT_SYMBOL(md_done_sync);
6959 EXPORT_SYMBOL(md_write_start);
6960 EXPORT_SYMBOL(md_write_end);
6961 EXPORT_SYMBOL(md_register_thread);
6962 EXPORT_SYMBOL(md_unregister_thread);
6963 EXPORT_SYMBOL(md_wakeup_thread);
6964 EXPORT_SYMBOL(md_check_recovery);
6965 MODULE_LICENSE("GPL");
6966 MODULE_ALIAS("md");
6967 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Linux ACPI git tree (mirror)