?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
init from v2.6.32.60
[mach-moxart.git] / drivers / md / md.c
blob4ce6e2fa4b53de688fa990492742facb6c14a7de
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 const 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 int mddev_congested(mddev_t *mddev, int bits)
266 {
267 return mddev->suspended;
268 }
269 EXPORT_SYMBOL(mddev_congested);
270
271
272 static inline mddev_t *mddev_get(mddev_t *mddev)
273 {
274 atomic_inc(&mddev->active);
275 return mddev;
276 }
277
278 static void mddev_delayed_delete(struct work_struct *ws);
279
280 static void mddev_put(mddev_t *mddev)
281 {
282 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
283 return;
284 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
285 mddev->ctime == 0 && !mddev->hold_active) {
286 /* Array is not configured at all, and not held active,
287 * so destroy it */
288 list_del(&mddev->all_mddevs);
289 if (mddev->gendisk) {
290 /* we did a probe so need to clean up.
291 * Call schedule_work inside the spinlock
292 * so that flush_scheduled_work() after
293 * mddev_find will succeed in waiting for the
294 * work to be done.
295 */
296 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
297 schedule_work(&mddev->del_work);
298 } else
299 kfree(mddev);
300 }
301 spin_unlock(&all_mddevs_lock);
302 }
303
304 static mddev_t * mddev_find(dev_t unit)
305 {
306 mddev_t *mddev, *new = NULL;
307
308 if (unit && MAJOR(unit) != MD_MAJOR)
309 unit &= ~((1<<MdpMinorShift)-1);
310
311 retry:
312 spin_lock(&all_mddevs_lock);
313
314 if (unit) {
315 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
316 if (mddev->unit == unit) {
317 mddev_get(mddev);
318 spin_unlock(&all_mddevs_lock);
319 kfree(new);
320 return mddev;
321 }
322
323 if (new) {
324 list_add(&new->all_mddevs, &all_mddevs);
325 spin_unlock(&all_mddevs_lock);
326 new->hold_active = UNTIL_IOCTL;
327 return new;
328 }
329 } else if (new) {
330 /* find an unused unit number */
331 static int next_minor = 512;
332 int start = next_minor;
333 int is_free = 0;
334 int dev = 0;
335 while (!is_free) {
336 dev = MKDEV(MD_MAJOR, next_minor);
337 next_minor++;
338 if (next_minor > MINORMASK)
339 next_minor = 0;
340 if (next_minor == start) {
341 /* Oh dear, all in use. */
342 spin_unlock(&all_mddevs_lock);
343 kfree(new);
344 return NULL;
345 }
346
347 is_free = 1;
348 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
349 if (mddev->unit == dev) {
350 is_free = 0;
351 break;
352 }
353 }
354 new->unit = dev;
355 new->md_minor = MINOR(dev);
356 new->hold_active = UNTIL_STOP;
357 list_add(&new->all_mddevs, &all_mddevs);
358 spin_unlock(&all_mddevs_lock);
359 return new;
360 }
361 spin_unlock(&all_mddevs_lock);
362
363 new = kzalloc(sizeof(*new), GFP_KERNEL);
364 if (!new)
365 return NULL;
366
367 new->unit = unit;
368 if (MAJOR(unit) == MD_MAJOR)
369 new->md_minor = MINOR(unit);
370 else
371 new->md_minor = MINOR(unit) >> MdpMinorShift;
372
373 mutex_init(&new->open_mutex);
374 mutex_init(&new->reconfig_mutex);
375 mutex_init(&new->bitmap_mutex);
376 INIT_LIST_HEAD(&new->disks);
377 INIT_LIST_HEAD(&new->all_mddevs);
378 init_timer(&new->safemode_timer);
379 atomic_set(&new->active, 1);
380 atomic_set(&new->openers, 0);
381 atomic_set(&new->active_io, 0);
382 spin_lock_init(&new->write_lock);
383 init_waitqueue_head(&new->sb_wait);
384 init_waitqueue_head(&new->recovery_wait);
385 new->reshape_position = MaxSector;
386 new->resync_min = 0;
387 new->resync_max = MaxSector;
388 new->level = LEVEL_NONE;
389
390 goto retry;
391 }
392
393 static inline int mddev_lock(mddev_t * mddev)
394 {
395 return mutex_lock_interruptible(&mddev->reconfig_mutex);
396 }
397
398 static inline int mddev_is_locked(mddev_t *mddev)
399 {
400 return mutex_is_locked(&mddev->reconfig_mutex);
401 }
402
403 static inline int mddev_trylock(mddev_t * mddev)
404 {
405 return mutex_trylock(&mddev->reconfig_mutex);
406 }
407
408 static inline void mddev_unlock(mddev_t * mddev)
409 {
410 mutex_unlock(&mddev->reconfig_mutex);
411
412 md_wakeup_thread(mddev->thread);
413 }
414
415 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
416 {
417 mdk_rdev_t *rdev;
418
419 list_for_each_entry(rdev, &mddev->disks, same_set)
420 if (rdev->desc_nr == nr)
421 return rdev;
422
423 return NULL;
424 }
425
426 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
427 {
428 mdk_rdev_t *rdev;
429
430 list_for_each_entry(rdev, &mddev->disks, same_set)
431 if (rdev->bdev->bd_dev == dev)
432 return rdev;
433
434 return NULL;
435 }
436
437 static struct mdk_personality *find_pers(int level, char *clevel)
438 {
439 struct mdk_personality *pers;
440 list_for_each_entry(pers, &pers_list, list) {
441 if (level != LEVEL_NONE && pers->level == level)
442 return pers;
443 if (strcmp(pers->name, clevel)==0)
444 return pers;
445 }
446 return NULL;
447 }
448
449 /* return the offset of the super block in 512byte sectors */
450 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
451 {
452 sector_t num_sectors = bdev->bd_inode->i_size / 512;
453 return MD_NEW_SIZE_SECTORS(num_sectors);
454 }
455
456 static int alloc_disk_sb(mdk_rdev_t * rdev)
457 {
458 if (rdev->sb_page)
459 MD_BUG();
460
461 rdev->sb_page = alloc_page(GFP_KERNEL);
462 if (!rdev->sb_page) {
463 printk(KERN_ALERT "md: out of memory.\n");
464 return -ENOMEM;
465 }
466
467 return 0;
468 }
469
470 static void free_disk_sb(mdk_rdev_t * rdev)
471 {
472 if (rdev->sb_page) {
473 put_page(rdev->sb_page);
474 rdev->sb_loaded = 0;
475 rdev->sb_page = NULL;
476 rdev->sb_start = 0;
477 rdev->sectors = 0;
478 }
479 }
480
481
482 static void super_written(struct bio *bio, int error)
483 {
484 mdk_rdev_t *rdev = bio->bi_private;
485 mddev_t *mddev = rdev->mddev;
486
487 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
488 printk("md: super_written gets error=%d, uptodate=%d\n",
489 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
490 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
491 md_error(mddev, rdev);
492 }
493
494 if (atomic_dec_and_test(&mddev->pending_writes))
495 wake_up(&mddev->sb_wait);
496 bio_put(bio);
497 }
498
499 static void super_written_barrier(struct bio *bio, int error)
500 {
501 struct bio *bio2 = bio->bi_private;
502 mdk_rdev_t *rdev = bio2->bi_private;
503 mddev_t *mddev = rdev->mddev;
504
505 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
506 error == -EOPNOTSUPP) {
507 unsigned long flags;
508 /* barriers don't appear to be supported :-( */
509 set_bit(BarriersNotsupp, &rdev->flags);
510 mddev->barriers_work = 0;
511 spin_lock_irqsave(&mddev->write_lock, flags);
512 bio2->bi_next = mddev->biolist;
513 mddev->biolist = bio2;
514 spin_unlock_irqrestore(&mddev->write_lock, flags);
515 wake_up(&mddev->sb_wait);
516 bio_put(bio);
517 } else {
518 bio_put(bio2);
519 bio->bi_private = rdev;
520 super_written(bio, error);
521 }
522 }
523
524 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
525 sector_t sector, int size, struct page *page)
526 {
527 /* write first size bytes of page to sector of rdev
528 * Increment mddev->pending_writes before returning
529 * and decrement it on completion, waking up sb_wait
530 * if zero is reached.
531 * If an error occurred, call md_error
532 *
533 * As we might need to resubmit the request if BIO_RW_BARRIER
534 * causes ENOTSUPP, we allocate a spare bio...
535 */
536 struct bio *bio = bio_alloc(GFP_NOIO, 1);
537 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG);
538
539 bio->bi_bdev = rdev->bdev;
540 bio->bi_sector = sector;
541 bio_add_page(bio, page, size, 0);
542 bio->bi_private = rdev;
543 bio->bi_end_io = super_written;
544 bio->bi_rw = rw;
545
546 atomic_inc(&mddev->pending_writes);
547 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
548 struct bio *rbio;
549 rw |= (1<<BIO_RW_BARRIER);
550 rbio = bio_clone(bio, GFP_NOIO);
551 rbio->bi_private = bio;
552 rbio->bi_end_io = super_written_barrier;
553 submit_bio(rw, rbio);
554 } else
555 submit_bio(rw, bio);
556 }
557
558 void md_super_wait(mddev_t *mddev)
559 {
560 /* wait for all superblock writes that were scheduled to complete.
561 * if any had to be retried (due to BARRIER problems), retry them
562 */
563 DEFINE_WAIT(wq);
564 for(;;) {
565 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
566 if (atomic_read(&mddev->pending_writes)==0)
567 break;
568 while (mddev->biolist) {
569 struct bio *bio;
570 spin_lock_irq(&mddev->write_lock);
571 bio = mddev->biolist;
572 mddev->biolist = bio->bi_next ;
573 bio->bi_next = NULL;
574 spin_unlock_irq(&mddev->write_lock);
575 submit_bio(bio->bi_rw, bio);
576 }
577 schedule();
578 }
579 finish_wait(&mddev->sb_wait, &wq);
580 }
581
582 static void bi_complete(struct bio *bio, int error)
583 {
584 complete((struct completion*)bio->bi_private);
585 }
586
587 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
588 struct page *page, int rw)
589 {
590 struct bio *bio = bio_alloc(GFP_NOIO, 1);
591 struct completion event;
592 int ret;
593
594 rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
595
596 bio->bi_bdev = bdev;
597 bio->bi_sector = sector;
598 bio_add_page(bio, page, size, 0);
599 init_completion(&event);
600 bio->bi_private = &event;
601 bio->bi_end_io = bi_complete;
602 submit_bio(rw, bio);
603 wait_for_completion(&event);
604
605 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
606 bio_put(bio);
607 return ret;
608 }
609 EXPORT_SYMBOL_GPL(sync_page_io);
610
611 static int read_disk_sb(mdk_rdev_t * rdev, int size)
612 {
613 char b[BDEVNAME_SIZE];
614 if (!rdev->sb_page) {
615 MD_BUG();
616 return -EINVAL;
617 }
618 if (rdev->sb_loaded)
619 return 0;
620
621
622 if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
623 goto fail;
624 rdev->sb_loaded = 1;
625 return 0;
626
627 fail:
628 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
629 bdevname(rdev->bdev,b));
630 return -EINVAL;
631 }
632
633 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
634 {
635 return sb1->set_uuid0 == sb2->set_uuid0 &&
636 sb1->set_uuid1 == sb2->set_uuid1 &&
637 sb1->set_uuid2 == sb2->set_uuid2 &&
638 sb1->set_uuid3 == sb2->set_uuid3;
639 }
640
641 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
642 {
643 int ret;
644 mdp_super_t *tmp1, *tmp2;
645
646 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
647 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
648
649 if (!tmp1 || !tmp2) {
650 ret = 0;
651 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
652 goto abort;
653 }
654
655 *tmp1 = *sb1;
656 *tmp2 = *sb2;
657
658 /*
659 * nr_disks is not constant
660 */
661 tmp1->nr_disks = 0;
662 tmp2->nr_disks = 0;
663
664 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
665 abort:
666 kfree(tmp1);
667 kfree(tmp2);
668 return ret;
669 }
670
671
672 static u32 md_csum_fold(u32 csum)
673 {
674 csum = (csum & 0xffff) + (csum >> 16);
675 return (csum & 0xffff) + (csum >> 16);
676 }
677
678 static unsigned int calc_sb_csum(mdp_super_t * sb)
679 {
680 u64 newcsum = 0;
681 u32 *sb32 = (u32*)sb;
682 int i;
683 unsigned int disk_csum, csum;
684
685 disk_csum = sb->sb_csum;
686 sb->sb_csum = 0;
687
688 for (i = 0; i < MD_SB_BYTES/4 ; i++)
689 newcsum += sb32[i];
690 csum = (newcsum & 0xffffffff) + (newcsum>>32);
691
692
693 #ifdef CONFIG_ALPHA
694 /* This used to use csum_partial, which was wrong for several
695 * reasons including that different results are returned on
696 * different architectures. It isn't critical that we get exactly
697 * the same return value as before (we always csum_fold before
698 * testing, and that removes any differences). However as we
699 * know that csum_partial always returned a 16bit value on
700 * alphas, do a fold to maximise conformity to previous behaviour.
701 */
702 sb->sb_csum = md_csum_fold(disk_csum);
703 #else
704 sb->sb_csum = disk_csum;
705 #endif
706 return csum;
707 }
708
709
710 /*
711 * Handle superblock details.
712 * We want to be able to handle multiple superblock formats
713 * so we have a common interface to them all, and an array of
714 * different handlers.
715 * We rely on user-space to write the initial superblock, and support
716 * reading and updating of superblocks.
717 * Interface methods are:
718 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
719 * loads and validates a superblock on dev.
720 * if refdev != NULL, compare superblocks on both devices
721 * Return:
722 * 0 - dev has a superblock that is compatible with refdev
723 * 1 - dev has a superblock that is compatible and newer than refdev
724 * so dev should be used as the refdev in future
725 * -EINVAL superblock incompatible or invalid
726 * -othererror e.g. -EIO
727 *
728 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
729 * Verify that dev is acceptable into mddev.
730 * The first time, mddev->raid_disks will be 0, and data from
731 * dev should be merged in. Subsequent calls check that dev
732 * is new enough. Return 0 or -EINVAL
733 *
734 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
735 * Update the superblock for rdev with data in mddev
736 * This does not write to disc.
737 *
738 */
739
740 struct super_type {
741 char *name;
742 struct module *owner;
743 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
744 int minor_version);
745 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
746 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
747 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
748 sector_t num_sectors);
749 };
750
751 /*
752 * Check that the given mddev has no bitmap.
753 *
754 * This function is called from the run method of all personalities that do not
755 * support bitmaps. It prints an error message and returns non-zero if mddev
756 * has a bitmap. Otherwise, it returns 0.
757 *
758 */
759 int md_check_no_bitmap(mddev_t *mddev)
760 {
761 if (!mddev->bitmap_file && !mddev->bitmap_offset)
762 return 0;
763 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
764 mdname(mddev), mddev->pers->name);
765 return 1;
766 }
767 EXPORT_SYMBOL(md_check_no_bitmap);
768
769 /*
770 * load_super for 0.90.0
771 */
772 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
773 {
774 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
775 mdp_super_t *sb;
776 int ret;
777
778 /*
779 * Calculate the position of the superblock (512byte sectors),
780 * it's at the end of the disk.
781 *
782 * It also happens to be a multiple of 4Kb.
783 */
784 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
785
786 ret = read_disk_sb(rdev, MD_SB_BYTES);
787 if (ret) return ret;
788
789 ret = -EINVAL;
790
791 bdevname(rdev->bdev, b);
792 sb = (mdp_super_t*)page_address(rdev->sb_page);
793
794 if (sb->md_magic != MD_SB_MAGIC) {
795 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
796 b);
797 goto abort;
798 }
799
800 if (sb->major_version != 0 ||
801 sb->minor_version < 90 ||
802 sb->minor_version > 91) {
803 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
804 sb->major_version, sb->minor_version,
805 b);
806 goto abort;
807 }
808
809 if (sb->raid_disks <= 0)
810 goto abort;
811
812 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
813 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
814 b);
815 goto abort;
816 }
817
818 rdev->preferred_minor = sb->md_minor;
819 rdev->data_offset = 0;
820 rdev->sb_size = MD_SB_BYTES;
821
822 if (sb->level == LEVEL_MULTIPATH)
823 rdev->desc_nr = -1;
824 else
825 rdev->desc_nr = sb->this_disk.number;
826
827 if (!refdev) {
828 ret = 1;
829 } else {
830 __u64 ev1, ev2;
831 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
832 if (!uuid_equal(refsb, sb)) {
833 printk(KERN_WARNING "md: %s has different UUID to %s\n",
834 b, bdevname(refdev->bdev,b2));
835 goto abort;
836 }
837 if (!sb_equal(refsb, sb)) {
838 printk(KERN_WARNING "md: %s has same UUID"
839 " but different superblock to %s\n",
840 b, bdevname(refdev->bdev, b2));
841 goto abort;
842 }
843 ev1 = md_event(sb);
844 ev2 = md_event(refsb);
845 if (ev1 > ev2)
846 ret = 1;
847 else
848 ret = 0;
849 }
850 rdev->sectors = rdev->sb_start;
851 /* Limit to 4TB as metadata cannot record more than that */
852 if (rdev->sectors >= (2ULL << 32))
853 rdev->sectors = (2ULL << 32) - 2;
854
855 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
856 /* "this cannot possibly happen" ... */
857 ret = -EINVAL;
858
859 abort:
860 return ret;
861 }
862
863 /*
864 * validate_super for 0.90.0
865 */
866 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
867 {
868 mdp_disk_t *desc;
869 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
870 __u64 ev1 = md_event(sb);
871
872 rdev->raid_disk = -1;
873 clear_bit(Faulty, &rdev->flags);
874 clear_bit(In_sync, &rdev->flags);
875 clear_bit(WriteMostly, &rdev->flags);
876 clear_bit(BarriersNotsupp, &rdev->flags);
877
878 if (mddev->raid_disks == 0) {
879 mddev->major_version = 0;
880 mddev->minor_version = sb->minor_version;
881 mddev->patch_version = sb->patch_version;
882 mddev->external = 0;
883 mddev->chunk_sectors = sb->chunk_size >> 9;
884 mddev->ctime = sb->ctime;
885 mddev->utime = sb->utime;
886 mddev->level = sb->level;
887 mddev->clevel[0] = 0;
888 mddev->layout = sb->layout;
889 mddev->raid_disks = sb->raid_disks;
890 mddev->dev_sectors = ((sector_t)sb->size) * 2;
891 mddev->events = ev1;
892 mddev->bitmap_offset = 0;
893 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
894
895 if (mddev->minor_version >= 91) {
896 mddev->reshape_position = sb->reshape_position;
897 mddev->delta_disks = sb->delta_disks;
898 mddev->new_level = sb->new_level;
899 mddev->new_layout = sb->new_layout;
900 mddev->new_chunk_sectors = sb->new_chunk >> 9;
901 } else {
902 mddev->reshape_position = MaxSector;
903 mddev->delta_disks = 0;
904 mddev->new_level = mddev->level;
905 mddev->new_layout = mddev->layout;
906 mddev->new_chunk_sectors = mddev->chunk_sectors;
907 }
908
909 if (sb->state & (1<<MD_SB_CLEAN))
910 mddev->recovery_cp = MaxSector;
911 else {
912 if (sb->events_hi == sb->cp_events_hi &&
913 sb->events_lo == sb->cp_events_lo) {
914 mddev->recovery_cp = sb->recovery_cp;
915 } else
916 mddev->recovery_cp = 0;
917 }
918
919 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
920 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
921 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
922 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
923
924 mddev->max_disks = MD_SB_DISKS;
925
926 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
927 mddev->bitmap_file == NULL)
928 mddev->bitmap_offset = mddev->default_bitmap_offset;
929
930 } else if (mddev->pers == NULL) {
931 /* Insist on good event counter while assembling */
932 ++ev1;
933 if (ev1 < mddev->events)
934 return -EINVAL;
935 } else if (mddev->bitmap) {
936 /* if adding to array with a bitmap, then we can accept an
937 * older device ... but not too old.
938 */
939 if (ev1 < mddev->bitmap->events_cleared)
940 return 0;
941 } else {
942 if (ev1 < mddev->events)
943 /* just a hot-add of a new device, leave raid_disk at -1 */
944 return 0;
945 }
946
947 if (mddev->level != LEVEL_MULTIPATH) {
948 desc = sb->disks + rdev->desc_nr;
949
950 if (desc->state & (1<<MD_DISK_FAULTY))
951 set_bit(Faulty, &rdev->flags);
952 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
953 desc->raid_disk < mddev->raid_disks */) {
954 set_bit(In_sync, &rdev->flags);
955 rdev->raid_disk = desc->raid_disk;
956 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
957 /* active but not in sync implies recovery up to
958 * reshape position. We don't know exactly where
959 * that is, so set to zero for now */
960 if (mddev->minor_version >= 91) {
961 rdev->recovery_offset = 0;
962 rdev->raid_disk = desc->raid_disk;
963 }
964 }
965 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
966 set_bit(WriteMostly, &rdev->flags);
967 } else /* MULTIPATH are always insync */
968 set_bit(In_sync, &rdev->flags);
969 return 0;
970 }
971
972 /*
973 * sync_super for 0.90.0
974 */
975 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
976 {
977 mdp_super_t *sb;
978 mdk_rdev_t *rdev2;
979 int next_spare = mddev->raid_disks;
980
981
982 /* make rdev->sb match mddev data..
983 *
984 * 1/ zero out disks
985 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
986 * 3/ any empty disks < next_spare become removed
987 *
988 * disks[0] gets initialised to REMOVED because
989 * we cannot be sure from other fields if it has
990 * been initialised or not.
991 */
992 int i;
993 int active=0, working=0,failed=0,spare=0,nr_disks=0;
994
995 rdev->sb_size = MD_SB_BYTES;
996
997 sb = (mdp_super_t*)page_address(rdev->sb_page);
998
999 memset(sb, 0, sizeof(*sb));
1000
1001 sb->md_magic = MD_SB_MAGIC;
1002 sb->major_version = mddev->major_version;
1003 sb->patch_version = mddev->patch_version;
1004 sb->gvalid_words = 0; /* ignored */
1005 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1006 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1007 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1008 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1009
1010 sb->ctime = mddev->ctime;
1011 sb->level = mddev->level;
1012 sb->size = mddev->dev_sectors / 2;
1013 sb->raid_disks = mddev->raid_disks;
1014 sb->md_minor = mddev->md_minor;
1015 sb->not_persistent = 0;
1016 sb->utime = mddev->utime;
1017 sb->state = 0;
1018 sb->events_hi = (mddev->events>>32);
1019 sb->events_lo = (u32)mddev->events;
1020
1021 if (mddev->reshape_position == MaxSector)
1022 sb->minor_version = 90;
1023 else {
1024 sb->minor_version = 91;
1025 sb->reshape_position = mddev->reshape_position;
1026 sb->new_level = mddev->new_level;
1027 sb->delta_disks = mddev->delta_disks;
1028 sb->new_layout = mddev->new_layout;
1029 sb->new_chunk = mddev->new_chunk_sectors << 9;
1030 }
1031 mddev->minor_version = sb->minor_version;
1032 if (mddev->in_sync)
1033 {
1034 sb->recovery_cp = mddev->recovery_cp;
1035 sb->cp_events_hi = (mddev->events>>32);
1036 sb->cp_events_lo = (u32)mddev->events;
1037 if (mddev->recovery_cp == MaxSector)
1038 sb->state = (1<< MD_SB_CLEAN);
1039 } else
1040 sb->recovery_cp = 0;
1041
1042 sb->layout = mddev->layout;
1043 sb->chunk_size = mddev->chunk_sectors << 9;
1044
1045 if (mddev->bitmap && mddev->bitmap_file == NULL)
1046 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1047
1048 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1049 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1050 mdp_disk_t *d;
1051 int desc_nr;
1052 int is_active = test_bit(In_sync, &rdev2->flags);
1053
1054 if (rdev2->raid_disk >= 0 &&
1055 sb->minor_version >= 91)
1056 /* we have nowhere to store the recovery_offset,
1057 * but if it is not below the reshape_position,
1058 * we can piggy-back on that.
1059 */
1060 is_active = 1;
1061 if (rdev2->raid_disk < 0 ||
1062 test_bit(Faulty, &rdev2->flags))
1063 is_active = 0;
1064 if (is_active)
1065 desc_nr = rdev2->raid_disk;
1066 else
1067 desc_nr = next_spare++;
1068 rdev2->desc_nr = desc_nr;
1069 d = &sb->disks[rdev2->desc_nr];
1070 nr_disks++;
1071 d->number = rdev2->desc_nr;
1072 d->major = MAJOR(rdev2->bdev->bd_dev);
1073 d->minor = MINOR(rdev2->bdev->bd_dev);
1074 if (is_active)
1075 d->raid_disk = rdev2->raid_disk;
1076 else
1077 d->raid_disk = rdev2->desc_nr; /* compatibility */
1078 if (test_bit(Faulty, &rdev2->flags))
1079 d->state = (1<<MD_DISK_FAULTY);
1080 else if (is_active) {
1081 d->state = (1<<MD_DISK_ACTIVE);
1082 if (test_bit(In_sync, &rdev2->flags))
1083 d->state |= (1<<MD_DISK_SYNC);
1084 active++;
1085 working++;
1086 } else {
1087 d->state = 0;
1088 spare++;
1089 working++;
1090 }
1091 if (test_bit(WriteMostly, &rdev2->flags))
1092 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1093 }
1094 /* now set the "removed" and "faulty" bits on any missing devices */
1095 for (i=0 ; i < mddev->raid_disks ; i++) {
1096 mdp_disk_t *d = &sb->disks[i];
1097 if (d->state == 0 && d->number == 0) {
1098 d->number = i;
1099 d->raid_disk = i;
1100 d->state = (1<<MD_DISK_REMOVED);
1101 d->state |= (1<<MD_DISK_FAULTY);
1102 failed++;
1103 }
1104 }
1105 sb->nr_disks = nr_disks;
1106 sb->active_disks = active;
1107 sb->working_disks = working;
1108 sb->failed_disks = failed;
1109 sb->spare_disks = spare;
1110
1111 sb->this_disk = sb->disks[rdev->desc_nr];
1112 sb->sb_csum = calc_sb_csum(sb);
1113 }
1114
1115 /*
1116 * rdev_size_change for 0.90.0
1117 */
1118 static unsigned long long
1119 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1120 {
1121 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1122 return 0; /* component must fit device */
1123 if (rdev->mddev->bitmap_offset)
1124 return 0; /* can't move bitmap */
1125 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1126 if (!num_sectors || num_sectors > rdev->sb_start)
1127 num_sectors = rdev->sb_start;
1128 /* Limit to 4TB as metadata cannot record more than that.
1129 * 4TB == 2^32 KB, or 2*2^32 sectors.
1130 */
1131 if (num_sectors >= (2ULL << 32))
1132 num_sectors = (2ULL << 32) - 2;
1133 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1134 rdev->sb_page);
1135 md_super_wait(rdev->mddev);
1136 return num_sectors;
1137 }
1138
1139
1140 /*
1141 * version 1 superblock
1142 */
1143
1144 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1145 {
1146 __le32 disk_csum;
1147 u32 csum;
1148 unsigned long long newcsum;
1149 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1150 __le32 *isuper = (__le32*)sb;
1151 int i;
1152
1153 disk_csum = sb->sb_csum;
1154 sb->sb_csum = 0;
1155 newcsum = 0;
1156 for (i=0; size>=4; size -= 4 )
1157 newcsum += le32_to_cpu(*isuper++);
1158
1159 if (size == 2)
1160 newcsum += le16_to_cpu(*(__le16*) isuper);
1161
1162 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1163 sb->sb_csum = disk_csum;
1164 return cpu_to_le32(csum);
1165 }
1166
1167 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1168 {
1169 struct mdp_superblock_1 *sb;
1170 int ret;
1171 sector_t sb_start;
1172 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1173 int bmask;
1174
1175 /*
1176 * Calculate the position of the superblock in 512byte sectors.
1177 * It is always aligned to a 4K boundary and
1178 * depeding on minor_version, it can be:
1179 * 0: At least 8K, but less than 12K, from end of device
1180 * 1: At start of device
1181 * 2: 4K from start of device.
1182 */
1183 switch(minor_version) {
1184 case 0:
1185 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1186 sb_start -= 8*2;
1187 sb_start &= ~(sector_t)(4*2-1);
1188 break;
1189 case 1:
1190 sb_start = 0;
1191 break;
1192 case 2:
1193 sb_start = 8;
1194 break;
1195 default:
1196 return -EINVAL;
1197 }
1198 rdev->sb_start = sb_start;
1199
1200 /* superblock is rarely larger than 1K, but it can be larger,
1201 * and it is safe to read 4k, so we do that
1202 */
1203 ret = read_disk_sb(rdev, 4096);
1204 if (ret) return ret;
1205
1206
1207 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1208
1209 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1210 sb->major_version != cpu_to_le32(1) ||
1211 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1212 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1213 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1214 return -EINVAL;
1215
1216 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1217 printk("md: invalid superblock checksum on %s\n",
1218 bdevname(rdev->bdev,b));
1219 return -EINVAL;
1220 }
1221 if (le64_to_cpu(sb->data_size) < 10) {
1222 printk("md: data_size too small on %s\n",
1223 bdevname(rdev->bdev,b));
1224 return -EINVAL;
1225 }
1226
1227 rdev->preferred_minor = 0xffff;
1228 rdev->data_offset = le64_to_cpu(sb->data_offset);
1229 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1230
1231 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1232 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1233 if (rdev->sb_size & bmask)
1234 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1235
1236 if (minor_version
1237 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1238 return -EINVAL;
1239
1240 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1241 rdev->desc_nr = -1;
1242 else
1243 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1244
1245 if (!refdev) {
1246 ret = 1;
1247 } else {
1248 __u64 ev1, ev2;
1249 struct mdp_superblock_1 *refsb =
1250 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1251
1252 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1253 sb->level != refsb->level ||
1254 sb->layout != refsb->layout ||
1255 sb->chunksize != refsb->chunksize) {
1256 printk(KERN_WARNING "md: %s has strangely different"
1257 " superblock to %s\n",
1258 bdevname(rdev->bdev,b),
1259 bdevname(refdev->bdev,b2));
1260 return -EINVAL;
1261 }
1262 ev1 = le64_to_cpu(sb->events);
1263 ev2 = le64_to_cpu(refsb->events);
1264
1265 if (ev1 > ev2)
1266 ret = 1;
1267 else
1268 ret = 0;
1269 }
1270 if (minor_version)
1271 rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
1272 le64_to_cpu(sb->data_offset);
1273 else
1274 rdev->sectors = rdev->sb_start;
1275 if (rdev->sectors < le64_to_cpu(sb->data_size))
1276 return -EINVAL;
1277 rdev->sectors = le64_to_cpu(sb->data_size);
1278 if (le64_to_cpu(sb->size) > rdev->sectors)
1279 return -EINVAL;
1280 return ret;
1281 }
1282
1283 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1284 {
1285 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1286 __u64 ev1 = le64_to_cpu(sb->events);
1287
1288 rdev->raid_disk = -1;
1289 clear_bit(Faulty, &rdev->flags);
1290 clear_bit(In_sync, &rdev->flags);
1291 clear_bit(WriteMostly, &rdev->flags);
1292 clear_bit(BarriersNotsupp, &rdev->flags);
1293
1294 if (mddev->raid_disks == 0) {
1295 mddev->major_version = 1;
1296 mddev->patch_version = 0;
1297 mddev->external = 0;
1298 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1299 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1300 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1301 mddev->level = le32_to_cpu(sb->level);
1302 mddev->clevel[0] = 0;
1303 mddev->layout = le32_to_cpu(sb->layout);
1304 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1305 mddev->dev_sectors = le64_to_cpu(sb->size);
1306 mddev->events = ev1;
1307 mddev->bitmap_offset = 0;
1308 mddev->default_bitmap_offset = 1024 >> 9;
1309
1310 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1311 memcpy(mddev->uuid, sb->set_uuid, 16);
1312
1313 mddev->max_disks = (4096-256)/2;
1314
1315 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1316 mddev->bitmap_file == NULL )
1317 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1318
1319 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1320 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1321 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1322 mddev->new_level = le32_to_cpu(sb->new_level);
1323 mddev->new_layout = le32_to_cpu(sb->new_layout);
1324 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1325 } else {
1326 mddev->reshape_position = MaxSector;
1327 mddev->delta_disks = 0;
1328 mddev->new_level = mddev->level;
1329 mddev->new_layout = mddev->layout;
1330 mddev->new_chunk_sectors = mddev->chunk_sectors;
1331 }
1332
1333 } else if (mddev->pers == NULL) {
1334 /* Insist of good event counter while assembling */
1335 ++ev1;
1336 if (ev1 < mddev->events)
1337 return -EINVAL;
1338 } else if (mddev->bitmap) {
1339 /* If adding to array with a bitmap, then we can accept an
1340 * older device, but not too old.
1341 */
1342 if (ev1 < mddev->bitmap->events_cleared)
1343 return 0;
1344 } else {
1345 if (ev1 < mddev->events)
1346 /* just a hot-add of a new device, leave raid_disk at -1 */
1347 return 0;
1348 }
1349 if (mddev->level != LEVEL_MULTIPATH) {
1350 int role;
1351 if (rdev->desc_nr < 0 ||
1352 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1353 role = 0xffff;
1354 rdev->desc_nr = -1;
1355 } else
1356 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1357 switch(role) {
1358 case 0xffff: /* spare */
1359 break;
1360 case 0xfffe: /* faulty */
1361 set_bit(Faulty, &rdev->flags);
1362 break;
1363 default:
1364 if ((le32_to_cpu(sb->feature_map) &
1365 MD_FEATURE_RECOVERY_OFFSET))
1366 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1367 else
1368 set_bit(In_sync, &rdev->flags);
1369 rdev->raid_disk = role;
1370 break;
1371 }
1372 if (sb->devflags & WriteMostly1)
1373 set_bit(WriteMostly, &rdev->flags);
1374 } else /* MULTIPATH are always insync */
1375 set_bit(In_sync, &rdev->flags);
1376
1377 return 0;
1378 }
1379
1380 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1381 {
1382 struct mdp_superblock_1 *sb;
1383 mdk_rdev_t *rdev2;
1384 int max_dev, i;
1385 /* make rdev->sb match mddev and rdev data. */
1386
1387 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1388
1389 sb->feature_map = 0;
1390 sb->pad0 = 0;
1391 sb->recovery_offset = cpu_to_le64(0);
1392 memset(sb->pad1, 0, sizeof(sb->pad1));
1393 memset(sb->pad2, 0, sizeof(sb->pad2));
1394 memset(sb->pad3, 0, sizeof(sb->pad3));
1395
1396 sb->utime = cpu_to_le64((__u64)mddev->utime);
1397 sb->events = cpu_to_le64(mddev->events);
1398 if (mddev->in_sync)
1399 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1400 else
1401 sb->resync_offset = cpu_to_le64(0);
1402
1403 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1404
1405 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1406 sb->size = cpu_to_le64(mddev->dev_sectors);
1407 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1408 sb->level = cpu_to_le32(mddev->level);
1409 sb->layout = cpu_to_le32(mddev->layout);
1410
1411 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1412 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1413 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1414 }
1415
1416 if (rdev->raid_disk >= 0 &&
1417 !test_bit(In_sync, &rdev->flags)) {
1418 if (rdev->recovery_offset > 0) {
1419 sb->feature_map |=
1420 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1421 sb->recovery_offset =
1422 cpu_to_le64(rdev->recovery_offset);
1423 }
1424 }
1425
1426 if (mddev->reshape_position != MaxSector) {
1427 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1428 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1429 sb->new_layout = cpu_to_le32(mddev->new_layout);
1430 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1431 sb->new_level = cpu_to_le32(mddev->new_level);
1432 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1433 }
1434
1435 max_dev = 0;
1436 list_for_each_entry(rdev2, &mddev->disks, same_set)
1437 if (rdev2->desc_nr+1 > max_dev)
1438 max_dev = rdev2->desc_nr+1;
1439
1440 if (max_dev > le32_to_cpu(sb->max_dev)) {
1441 int bmask;
1442 sb->max_dev = cpu_to_le32(max_dev);
1443 rdev->sb_size = max_dev * 2 + 256;
1444 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1445 if (rdev->sb_size & bmask)
1446 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1447 }
1448 for (i=0; i<max_dev;i++)
1449 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1450
1451 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1452 i = rdev2->desc_nr;
1453 if (test_bit(Faulty, &rdev2->flags))
1454 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1455 else if (test_bit(In_sync, &rdev2->flags))
1456 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1457 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1458 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1459 else
1460 sb->dev_roles[i] = cpu_to_le16(0xffff);
1461 }
1462
1463 sb->sb_csum = calc_sb_1_csum(sb);
1464 }
1465
1466 static unsigned long long
1467 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1468 {
1469 struct mdp_superblock_1 *sb;
1470 sector_t max_sectors;
1471 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1472 return 0; /* component must fit device */
1473 if (rdev->sb_start < rdev->data_offset) {
1474 /* minor versions 1 and 2; superblock before data */
1475 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1476 max_sectors -= rdev->data_offset;
1477 if (!num_sectors || num_sectors > max_sectors)
1478 num_sectors = max_sectors;
1479 } else if (rdev->mddev->bitmap_offset) {
1480 /* minor version 0 with bitmap we can't move */
1481 return 0;
1482 } else {
1483 /* minor version 0; superblock after data */
1484 sector_t sb_start;
1485 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1486 sb_start &= ~(sector_t)(4*2 - 1);
1487 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1488 if (!num_sectors || num_sectors > max_sectors)
1489 num_sectors = max_sectors;
1490 rdev->sb_start = sb_start;
1491 }
1492 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1493 sb->data_size = cpu_to_le64(num_sectors);
1494 sb->super_offset = rdev->sb_start;
1495 sb->sb_csum = calc_sb_1_csum(sb);
1496 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1497 rdev->sb_page);
1498 md_super_wait(rdev->mddev);
1499 return num_sectors;
1500 }
1501
1502 static struct super_type super_types[] = {
1503 [0] = {
1504 .name = "0.90.0",
1505 .owner = THIS_MODULE,
1506 .load_super = super_90_load,
1507 .validate_super = super_90_validate,
1508 .sync_super = super_90_sync,
1509 .rdev_size_change = super_90_rdev_size_change,
1510 },
1511 [1] = {
1512 .name = "md-1",
1513 .owner = THIS_MODULE,
1514 .load_super = super_1_load,
1515 .validate_super = super_1_validate,
1516 .sync_super = super_1_sync,
1517 .rdev_size_change = super_1_rdev_size_change,
1518 },
1519 };
1520
1521 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1522 {
1523 mdk_rdev_t *rdev, *rdev2;
1524
1525 rcu_read_lock();
1526 rdev_for_each_rcu(rdev, mddev1)
1527 rdev_for_each_rcu(rdev2, mddev2)
1528 if (rdev->bdev->bd_contains ==
1529 rdev2->bdev->bd_contains) {
1530 rcu_read_unlock();
1531 return 1;
1532 }
1533 rcu_read_unlock();
1534 return 0;
1535 }
1536
1537 static LIST_HEAD(pending_raid_disks);
1538
1539 /*
1540 * Try to register data integrity profile for an mddev
1541 *
1542 * This is called when an array is started and after a disk has been kicked
1543 * from the array. It only succeeds if all working and active component devices
1544 * are integrity capable with matching profiles.
1545 */
1546 int md_integrity_register(mddev_t *mddev)
1547 {
1548 mdk_rdev_t *rdev, *reference = NULL;
1549
1550 if (list_empty(&mddev->disks))
1551 return 0; /* nothing to do */
1552 if (blk_get_integrity(mddev->gendisk))
1553 return 0; /* already registered */
1554 list_for_each_entry(rdev, &mddev->disks, same_set) {
1555 /* skip spares and non-functional disks */
1556 if (test_bit(Faulty, &rdev->flags))
1557 continue;
1558 if (rdev->raid_disk < 0)
1559 continue;
1560 /*
1561 * If at least one rdev is not integrity capable, we can not
1562 * enable data integrity for the md device.
1563 */
1564 if (!bdev_get_integrity(rdev->bdev))
1565 return -EINVAL;
1566 if (!reference) {
1567 /* Use the first rdev as the reference */
1568 reference = rdev;
1569 continue;
1570 }
1571 /* does this rdev's profile match the reference profile? */
1572 if (blk_integrity_compare(reference->bdev->bd_disk,
1573 rdev->bdev->bd_disk) < 0)
1574 return -EINVAL;
1575 }
1576 /*
1577 * All component devices are integrity capable and have matching
1578 * profiles, register the common profile for the md device.
1579 */
1580 if (blk_integrity_register(mddev->gendisk,
1581 bdev_get_integrity(reference->bdev)) != 0) {
1582 printk(KERN_ERR "md: failed to register integrity for %s\n",
1583 mdname(mddev));
1584 return -EINVAL;
1585 }
1586 printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1587 mdname(mddev));
1588 return 0;
1589 }
1590 EXPORT_SYMBOL(md_integrity_register);
1591
1592 /* Disable data integrity if non-capable/non-matching disk is being added */
1593 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
1594 {
1595 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1596 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1597
1598 if (!bi_mddev) /* nothing to do */
1599 return;
1600 if (rdev->raid_disk < 0) /* skip spares */
1601 return;
1602 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1603 rdev->bdev->bd_disk) >= 0)
1604 return;
1605 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1606 blk_integrity_unregister(mddev->gendisk);
1607 }
1608 EXPORT_SYMBOL(md_integrity_add_rdev);
1609
1610 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1611 {
1612 char b[BDEVNAME_SIZE];
1613 struct kobject *ko;
1614 char *s;
1615 int err;
1616
1617 if (rdev->mddev) {
1618 MD_BUG();
1619 return -EINVAL;
1620 }
1621
1622 /* prevent duplicates */
1623 if (find_rdev(mddev, rdev->bdev->bd_dev))
1624 return -EEXIST;
1625
1626 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1627 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1628 rdev->sectors < mddev->dev_sectors)) {
1629 if (mddev->pers) {
1630 /* Cannot change size, so fail
1631 * If mddev->level <= 0, then we don't care
1632 * about aligning sizes (e.g. linear)
1633 */
1634 if (mddev->level > 0)
1635 return -ENOSPC;
1636 } else
1637 mddev->dev_sectors = rdev->sectors;
1638 }
1639
1640 /* Verify rdev->desc_nr is unique.
1641 * If it is -1, assign a free number, else
1642 * check number is not in use
1643 */
1644 if (rdev->desc_nr < 0) {
1645 int choice = 0;
1646 if (mddev->pers) choice = mddev->raid_disks;
1647 while (find_rdev_nr(mddev, choice))
1648 choice++;
1649 rdev->desc_nr = choice;
1650 } else {
1651 if (find_rdev_nr(mddev, rdev->desc_nr))
1652 return -EBUSY;
1653 }
1654 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1655 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1656 mdname(mddev), mddev->max_disks);
1657 return -EBUSY;
1658 }
1659 bdevname(rdev->bdev,b);
1660 while ( (s=strchr(b, '/')) != NULL)
1661 *s = '!';
1662
1663 rdev->mddev = mddev;
1664 printk(KERN_INFO "md: bind<%s>\n", b);
1665
1666 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1667 goto fail;
1668
1669 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1670 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1671 kobject_del(&rdev->kobj);
1672 goto fail;
1673 }
1674 rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, "state");
1675
1676 list_add_rcu(&rdev->same_set, &mddev->disks);
1677 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1678
1679 /* May as well allow recovery to be retried once */
1680 mddev->recovery_disabled = 0;
1681
1682 return 0;
1683
1684 fail:
1685 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1686 b, mdname(mddev));
1687 return err;
1688 }
1689
1690 static void md_delayed_delete(struct work_struct *ws)
1691 {
1692 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1693 kobject_del(&rdev->kobj);
1694 kobject_put(&rdev->kobj);
1695 }
1696
1697 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1698 {
1699 char b[BDEVNAME_SIZE];
1700 if (!rdev->mddev) {
1701 MD_BUG();
1702 return;
1703 }
1704 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1705 list_del_rcu(&rdev->same_set);
1706 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1707 rdev->mddev = NULL;
1708 sysfs_remove_link(&rdev->kobj, "block");
1709 sysfs_put(rdev->sysfs_state);
1710 rdev->sysfs_state = NULL;
1711 /* We need to delay this, otherwise we can deadlock when
1712 * writing to 'remove' to "dev/state". We also need
1713 * to delay it due to rcu usage.
1714 */
1715 synchronize_rcu();
1716 INIT_WORK(&rdev->del_work, md_delayed_delete);
1717 kobject_get(&rdev->kobj);
1718 schedule_work(&rdev->del_work);
1719 }
1720
1721 /*
1722 * prevent the device from being mounted, repartitioned or
1723 * otherwise reused by a RAID array (or any other kernel
1724 * subsystem), by bd_claiming the device.
1725 */
1726 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1727 {
1728 int err = 0;
1729 struct block_device *bdev;
1730 char b[BDEVNAME_SIZE];
1731
1732 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1733 if (IS_ERR(bdev)) {
1734 printk(KERN_ERR "md: could not open %s.\n",
1735 __bdevname(dev, b));
1736 return PTR_ERR(bdev);
1737 }
1738 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1739 if (err) {
1740 printk(KERN_ERR "md: could not bd_claim %s.\n",
1741 bdevname(bdev, b));
1742 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1743 return err;
1744 }
1745 if (!shared)
1746 set_bit(AllReserved, &rdev->flags);
1747 rdev->bdev = bdev;
1748 return err;
1749 }
1750
1751 static void unlock_rdev(mdk_rdev_t *rdev)
1752 {
1753 struct block_device *bdev = rdev->bdev;
1754 rdev->bdev = NULL;
1755 if (!bdev)
1756 MD_BUG();
1757 bd_release(bdev);
1758 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1759 }
1760
1761 void md_autodetect_dev(dev_t dev);
1762
1763 static void export_rdev(mdk_rdev_t * rdev)
1764 {
1765 char b[BDEVNAME_SIZE];
1766 printk(KERN_INFO "md: export_rdev(%s)\n",
1767 bdevname(rdev->bdev,b));
1768 if (rdev->mddev)
1769 MD_BUG();
1770 free_disk_sb(rdev);
1771 #ifndef MODULE
1772 if (test_bit(AutoDetected, &rdev->flags))
1773 md_autodetect_dev(rdev->bdev->bd_dev);
1774 #endif
1775 unlock_rdev(rdev);
1776 kobject_put(&rdev->kobj);
1777 }
1778
1779 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1780 {
1781 unbind_rdev_from_array(rdev);
1782 export_rdev(rdev);
1783 }
1784
1785 static void export_array(mddev_t *mddev)
1786 {
1787 mdk_rdev_t *rdev, *tmp;
1788
1789 rdev_for_each(rdev, tmp, mddev) {
1790 if (!rdev->mddev) {
1791 MD_BUG();
1792 continue;
1793 }
1794 kick_rdev_from_array(rdev);
1795 }
1796 if (!list_empty(&mddev->disks))
1797 MD_BUG();
1798 mddev->raid_disks = 0;
1799 mddev->major_version = 0;
1800 }
1801
1802 static void print_desc(mdp_disk_t *desc)
1803 {
1804 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1805 desc->major,desc->minor,desc->raid_disk,desc->state);
1806 }
1807
1808 static void print_sb_90(mdp_super_t *sb)
1809 {
1810 int i;
1811
1812 printk(KERN_INFO
1813 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1814 sb->major_version, sb->minor_version, sb->patch_version,
1815 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1816 sb->ctime);
1817 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1818 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1819 sb->md_minor, sb->layout, sb->chunk_size);
1820 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1821 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1822 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1823 sb->failed_disks, sb->spare_disks,
1824 sb->sb_csum, (unsigned long)sb->events_lo);
1825
1826 printk(KERN_INFO);
1827 for (i = 0; i < MD_SB_DISKS; i++) {
1828 mdp_disk_t *desc;
1829
1830 desc = sb->disks + i;
1831 if (desc->number || desc->major || desc->minor ||
1832 desc->raid_disk || (desc->state && (desc->state != 4))) {
1833 printk(" D %2d: ", i);
1834 print_desc(desc);
1835 }
1836 }
1837 printk(KERN_INFO "md: THIS: ");
1838 print_desc(&sb->this_disk);
1839 }
1840
1841 static void print_sb_1(struct mdp_superblock_1 *sb)
1842 {
1843 __u8 *uuid;
1844
1845 uuid = sb->set_uuid;
1846 printk(KERN_INFO
1847 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%02x%02x%02x%02x"
1848 ":%02x%02x:%02x%02x:%02x%02x:%02x%02x%02x%02x%02x%02x>\n"
1849 "md: Name: \"%s\" CT:%llu\n",
1850 le32_to_cpu(sb->major_version),
1851 le32_to_cpu(sb->feature_map),
1852 uuid[0], uuid[1], uuid[2], uuid[3],
1853 uuid[4], uuid[5], uuid[6], uuid[7],
1854 uuid[8], uuid[9], uuid[10], uuid[11],
1855 uuid[12], uuid[13], uuid[14], uuid[15],
1856 sb->set_name,
1857 (unsigned long long)le64_to_cpu(sb->ctime)
1858 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
1859
1860 uuid = sb->device_uuid;
1861 printk(KERN_INFO
1862 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
1863 " RO:%llu\n"
1864 "md: Dev:%08x UUID: %02x%02x%02x%02x:%02x%02x:%02x%02x:%02x%02x"
1865 ":%02x%02x%02x%02x%02x%02x\n"
1866 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
1867 "md: (MaxDev:%u) \n",
1868 le32_to_cpu(sb->level),
1869 (unsigned long long)le64_to_cpu(sb->size),
1870 le32_to_cpu(sb->raid_disks),
1871 le32_to_cpu(sb->layout),
1872 le32_to_cpu(sb->chunksize),
1873 (unsigned long long)le64_to_cpu(sb->data_offset),
1874 (unsigned long long)le64_to_cpu(sb->data_size),
1875 (unsigned long long)le64_to_cpu(sb->super_offset),
1876 (unsigned long long)le64_to_cpu(sb->recovery_offset),
1877 le32_to_cpu(sb->dev_number),
1878 uuid[0], uuid[1], uuid[2], uuid[3],
1879 uuid[4], uuid[5], uuid[6], uuid[7],
1880 uuid[8], uuid[9], uuid[10], uuid[11],
1881 uuid[12], uuid[13], uuid[14], uuid[15],
1882 sb->devflags,
1883 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
1884 (unsigned long long)le64_to_cpu(sb->events),
1885 (unsigned long long)le64_to_cpu(sb->resync_offset),
1886 le32_to_cpu(sb->sb_csum),
1887 le32_to_cpu(sb->max_dev)
1888 );
1889 }
1890
1891 static void print_rdev(mdk_rdev_t *rdev, int major_version)
1892 {
1893 char b[BDEVNAME_SIZE];
1894 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
1895 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
1896 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1897 rdev->desc_nr);
1898 if (rdev->sb_loaded) {
1899 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
1900 switch (major_version) {
1901 case 0:
1902 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
1903 break;
1904 case 1:
1905 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
1906 break;
1907 }
1908 } else
1909 printk(KERN_INFO "md: no rdev superblock!\n");
1910 }
1911
1912 static void md_print_devices(void)
1913 {
1914 struct list_head *tmp;
1915 mdk_rdev_t *rdev;
1916 mddev_t *mddev;
1917 char b[BDEVNAME_SIZE];
1918
1919 printk("\n");
1920 printk("md: **********************************\n");
1921 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1922 printk("md: **********************************\n");
1923 for_each_mddev(mddev, tmp) {
1924
1925 if (mddev->bitmap)
1926 bitmap_print_sb(mddev->bitmap);
1927 else
1928 printk("%s: ", mdname(mddev));
1929 list_for_each_entry(rdev, &mddev->disks, same_set)
1930 printk("<%s>", bdevname(rdev->bdev,b));
1931 printk("\n");
1932
1933 list_for_each_entry(rdev, &mddev->disks, same_set)
1934 print_rdev(rdev, mddev->major_version);
1935 }
1936 printk("md: **********************************\n");
1937 printk("\n");
1938 }
1939
1940
1941 static void sync_sbs(mddev_t * mddev, int nospares)
1942 {
1943 /* Update each superblock (in-memory image), but
1944 * if we are allowed to, skip spares which already
1945 * have the right event counter, or have one earlier
1946 * (which would mean they aren't being marked as dirty
1947 * with the rest of the array)
1948 */
1949 mdk_rdev_t *rdev;
1950
1951 /* First make sure individual recovery_offsets are correct */
1952 list_for_each_entry(rdev, &mddev->disks, same_set) {
1953 if (rdev->raid_disk >= 0 &&
1954 !test_bit(In_sync, &rdev->flags) &&
1955 mddev->curr_resync_completed > rdev->recovery_offset)
1956 rdev->recovery_offset = mddev->curr_resync_completed;
1957
1958 }
1959 list_for_each_entry(rdev, &mddev->disks, same_set) {
1960 if (rdev->sb_events == mddev->events ||
1961 (nospares &&
1962 rdev->raid_disk < 0 &&
1963 (rdev->sb_events&1)==0 &&
1964 rdev->sb_events+1 == mddev->events)) {
1965 /* Don't update this superblock */
1966 rdev->sb_loaded = 2;
1967 } else {
1968 super_types[mddev->major_version].
1969 sync_super(mddev, rdev);
1970 rdev->sb_loaded = 1;
1971 }
1972 }
1973 }
1974
1975 static void md_update_sb(mddev_t * mddev, int force_change)
1976 {
1977 mdk_rdev_t *rdev;
1978 int sync_req;
1979 int nospares = 0;
1980
1981 mddev->utime = get_seconds();
1982 if (mddev->external)
1983 return;
1984 repeat:
1985 spin_lock_irq(&mddev->write_lock);
1986
1987 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1988 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1989 force_change = 1;
1990 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1991 /* just a clean<-> dirty transition, possibly leave spares alone,
1992 * though if events isn't the right even/odd, we will have to do
1993 * spares after all
1994 */
1995 nospares = 1;
1996 if (force_change)
1997 nospares = 0;
1998 if (mddev->degraded)
1999 /* If the array is degraded, then skipping spares is both
2000 * dangerous and fairly pointless.
2001 * Dangerous because a device that was removed from the array
2002 * might have a event_count that still looks up-to-date,
2003 * so it can be re-added without a resync.
2004 * Pointless because if there are any spares to skip,
2005 * then a recovery will happen and soon that array won't
2006 * be degraded any more and the spare can go back to sleep then.
2007 */
2008 nospares = 0;
2009
2010 sync_req = mddev->in_sync;
2011
2012 /* If this is just a dirty<->clean transition, and the array is clean
2013 * and 'events' is odd, we can roll back to the previous clean state */
2014 if (nospares
2015 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2016 && (mddev->events & 1)
2017 && mddev->events != 1)
2018 mddev->events--;
2019 else {
2020 /* otherwise we have to go forward and ... */
2021 mddev->events ++;
2022 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
2023 /* .. if the array isn't clean, an 'even' event must also go
2024 * to spares. */
2025 if ((mddev->events&1)==0) {
2026 nospares = 0;
2027 sync_req = 2; /* force a second update to get the
2028 * even/odd in sync */
2029 }
2030 } else {
2031 /* otherwise an 'odd' event must go to spares */
2032 if ((mddev->events&1)) {
2033 nospares = 0;
2034 sync_req = 2; /* force a second update to get the
2035 * even/odd in sync */
2036 }
2037 }
2038 }
2039
2040 if (!mddev->events) {
2041 /*
2042 * oops, this 64-bit counter should never wrap.
2043 * Either we are in around ~1 trillion A.C., assuming
2044 * 1 reboot per second, or we have a bug:
2045 */
2046 MD_BUG();
2047 mddev->events --;
2048 }
2049
2050 /*
2051 * do not write anything to disk if using
2052 * nonpersistent superblocks
2053 */
2054 if (!mddev->persistent) {
2055 if (!mddev->external)
2056 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2057
2058 spin_unlock_irq(&mddev->write_lock);
2059 wake_up(&mddev->sb_wait);
2060 return;
2061 }
2062 sync_sbs(mddev, nospares);
2063 spin_unlock_irq(&mddev->write_lock);
2064
2065 dprintk(KERN_INFO
2066 "md: updating %s RAID superblock on device (in sync %d)\n",
2067 mdname(mddev),mddev->in_sync);
2068
2069 bitmap_update_sb(mddev->bitmap);
2070 list_for_each_entry(rdev, &mddev->disks, same_set) {
2071 char b[BDEVNAME_SIZE];
2072 dprintk(KERN_INFO "md: ");
2073 if (rdev->sb_loaded != 1)
2074 continue; /* no noise on spare devices */
2075 if (test_bit(Faulty, &rdev->flags))
2076 dprintk("(skipping faulty ");
2077
2078 dprintk("%s ", bdevname(rdev->bdev,b));
2079 if (!test_bit(Faulty, &rdev->flags)) {
2080 md_super_write(mddev,rdev,
2081 rdev->sb_start, rdev->sb_size,
2082 rdev->sb_page);
2083 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2084 bdevname(rdev->bdev,b),
2085 (unsigned long long)rdev->sb_start);
2086 rdev->sb_events = mddev->events;
2087
2088 } else
2089 dprintk(")\n");
2090 if (mddev->level == LEVEL_MULTIPATH)
2091 /* only need to write one superblock... */
2092 break;
2093 }
2094 md_super_wait(mddev);
2095 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2096
2097 spin_lock_irq(&mddev->write_lock);
2098 if (mddev->in_sync != sync_req ||
2099 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2100 /* have to write it out again */
2101 spin_unlock_irq(&mddev->write_lock);
2102 goto repeat;
2103 }
2104 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2105 spin_unlock_irq(&mddev->write_lock);
2106 wake_up(&mddev->sb_wait);
2107 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2108 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2109
2110 }
2111
2112 /* words written to sysfs files may, or may not, be \n terminated.
2113 * We want to accept with case. For this we use cmd_match.
2114 */
2115 static int cmd_match(const char *cmd, const char *str)
2116 {
2117 /* See if cmd, written into a sysfs file, matches
2118 * str. They must either be the same, or cmd can
2119 * have a trailing newline
2120 */
2121 while (*cmd && *str && *cmd == *str) {
2122 cmd++;
2123 str++;
2124 }
2125 if (*cmd == '\n')
2126 cmd++;
2127 if (*str || *cmd)
2128 return 0;
2129 return 1;
2130 }
2131
2132 struct rdev_sysfs_entry {
2133 struct attribute attr;
2134 ssize_t (*show)(mdk_rdev_t *, char *);
2135 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2136 };
2137
2138 static ssize_t
2139 state_show(mdk_rdev_t *rdev, char *page)
2140 {
2141 char *sep = "";
2142 size_t len = 0;
2143
2144 if (test_bit(Faulty, &rdev->flags)) {
2145 len+= sprintf(page+len, "%sfaulty",sep);
2146 sep = ",";
2147 }
2148 if (test_bit(In_sync, &rdev->flags)) {
2149 len += sprintf(page+len, "%sin_sync",sep);
2150 sep = ",";
2151 }
2152 if (test_bit(WriteMostly, &rdev->flags)) {
2153 len += sprintf(page+len, "%swrite_mostly",sep);
2154 sep = ",";
2155 }
2156 if (test_bit(Blocked, &rdev->flags)) {
2157 len += sprintf(page+len, "%sblocked", sep);
2158 sep = ",";
2159 }
2160 if (!test_bit(Faulty, &rdev->flags) &&
2161 !test_bit(In_sync, &rdev->flags)) {
2162 len += sprintf(page+len, "%sspare", sep);
2163 sep = ",";
2164 }
2165 return len+sprintf(page+len, "\n");
2166 }
2167
2168 static ssize_t
2169 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2170 {
2171 /* can write
2172 * faulty - simulates and error
2173 * remove - disconnects the device
2174 * writemostly - sets write_mostly
2175 * -writemostly - clears write_mostly
2176 * blocked - sets the Blocked flag
2177 * -blocked - clears the Blocked flag
2178 * insync - sets Insync providing device isn't active
2179 */
2180 int err = -EINVAL;
2181 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2182 md_error(rdev->mddev, rdev);
2183 err = 0;
2184 } else if (cmd_match(buf, "remove")) {
2185 if (rdev->raid_disk >= 0)
2186 err = -EBUSY;
2187 else {
2188 mddev_t *mddev = rdev->mddev;
2189 kick_rdev_from_array(rdev);
2190 if (mddev->pers)
2191 md_update_sb(mddev, 1);
2192 md_new_event(mddev);
2193 err = 0;
2194 }
2195 } else if (cmd_match(buf, "writemostly")) {
2196 set_bit(WriteMostly, &rdev->flags);
2197 err = 0;
2198 } else if (cmd_match(buf, "-writemostly")) {
2199 clear_bit(WriteMostly, &rdev->flags);
2200 err = 0;
2201 } else if (cmd_match(buf, "blocked")) {
2202 set_bit(Blocked, &rdev->flags);
2203 err = 0;
2204 } else if (cmd_match(buf, "-blocked")) {
2205 clear_bit(Blocked, &rdev->flags);
2206 wake_up(&rdev->blocked_wait);
2207 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2208 md_wakeup_thread(rdev->mddev->thread);
2209
2210 err = 0;
2211 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2212 set_bit(In_sync, &rdev->flags);
2213 err = 0;
2214 }
2215 if (!err && rdev->sysfs_state)
2216 sysfs_notify_dirent(rdev->sysfs_state);
2217 return err ? err : len;
2218 }
2219 static struct rdev_sysfs_entry rdev_state =
2220 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2221
2222 static ssize_t
2223 errors_show(mdk_rdev_t *rdev, char *page)
2224 {
2225 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2226 }
2227
2228 static ssize_t
2229 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2230 {
2231 char *e;
2232 unsigned long n = simple_strtoul(buf, &e, 10);
2233 if (*buf && (*e == 0 || *e == '\n')) {
2234 atomic_set(&rdev->corrected_errors, n);
2235 return len;
2236 }
2237 return -EINVAL;
2238 }
2239 static struct rdev_sysfs_entry rdev_errors =
2240 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2241
2242 static ssize_t
2243 slot_show(mdk_rdev_t *rdev, char *page)
2244 {
2245 if (rdev->raid_disk < 0)
2246 return sprintf(page, "none\n");
2247 else
2248 return sprintf(page, "%d\n", rdev->raid_disk);
2249 }
2250
2251 static ssize_t
2252 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2253 {
2254 char *e;
2255 int err;
2256 char nm[20];
2257 int slot = simple_strtoul(buf, &e, 10);
2258 if (strncmp(buf, "none", 4)==0)
2259 slot = -1;
2260 else if (e==buf || (*e && *e!= '\n'))
2261 return -EINVAL;
2262 if (rdev->mddev->pers && slot == -1) {
2263 /* Setting 'slot' on an active array requires also
2264 * updating the 'rd%d' link, and communicating
2265 * with the personality with ->hot_*_disk.
2266 * For now we only support removing
2267 * failed/spare devices. This normally happens automatically,
2268 * but not when the metadata is externally managed.
2269 */
2270 if (rdev->raid_disk == -1)
2271 return -EEXIST;
2272 /* personality does all needed checks */
2273 if (rdev->mddev->pers->hot_remove_disk == NULL)
2274 return -EINVAL;
2275 err = rdev->mddev->pers->
2276 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2277 if (err)
2278 return err;
2279 sprintf(nm, "rd%d", rdev->raid_disk);
2280 sysfs_remove_link(&rdev->mddev->kobj, nm);
2281 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2282 md_wakeup_thread(rdev->mddev->thread);
2283 } else if (rdev->mddev->pers) {
2284 mdk_rdev_t *rdev2;
2285 /* Activating a spare .. or possibly reactivating
2286 * if we ever get bitmaps working here.
2287 */
2288
2289 if (rdev->raid_disk != -1)
2290 return -EBUSY;
2291
2292 if (rdev->mddev->pers->hot_add_disk == NULL)
2293 return -EINVAL;
2294
2295 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2296 if (rdev2->raid_disk == slot)
2297 return -EEXIST;
2298
2299 rdev->raid_disk = slot;
2300 if (test_bit(In_sync, &rdev->flags))
2301 rdev->saved_raid_disk = slot;
2302 else
2303 rdev->saved_raid_disk = -1;
2304 err = rdev->mddev->pers->
2305 hot_add_disk(rdev->mddev, rdev);
2306 if (err) {
2307 rdev->raid_disk = -1;
2308 return err;
2309 } else
2310 sysfs_notify_dirent(rdev->sysfs_state);
2311 sprintf(nm, "rd%d", rdev->raid_disk);
2312 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2313 printk(KERN_WARNING
2314 "md: cannot register "
2315 "%s for %s\n",
2316 nm, mdname(rdev->mddev));
2317
2318 /* don't wakeup anyone, leave that to userspace. */
2319 } else {
2320 if (slot >= rdev->mddev->raid_disks)
2321 return -ENOSPC;
2322 rdev->raid_disk = slot;
2323 /* assume it is working */
2324 clear_bit(Faulty, &rdev->flags);
2325 clear_bit(WriteMostly, &rdev->flags);
2326 set_bit(In_sync, &rdev->flags);
2327 sysfs_notify_dirent(rdev->sysfs_state);
2328 }
2329 return len;
2330 }
2331
2332
2333 static struct rdev_sysfs_entry rdev_slot =
2334 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2335
2336 static ssize_t
2337 offset_show(mdk_rdev_t *rdev, char *page)
2338 {
2339 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2340 }
2341
2342 static ssize_t
2343 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2344 {
2345 char *e;
2346 unsigned long long offset = simple_strtoull(buf, &e, 10);
2347 if (e==buf || (*e && *e != '\n'))
2348 return -EINVAL;
2349 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2350 return -EBUSY;
2351 if (rdev->sectors && rdev->mddev->external)
2352 /* Must set offset before size, so overlap checks
2353 * can be sane */
2354 return -EBUSY;
2355 rdev->data_offset = offset;
2356 return len;
2357 }
2358
2359 static struct rdev_sysfs_entry rdev_offset =
2360 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2361
2362 static ssize_t
2363 rdev_size_show(mdk_rdev_t *rdev, char *page)
2364 {
2365 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2366 }
2367
2368 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2369 {
2370 /* check if two start/length pairs overlap */
2371 if (s1+l1 <= s2)
2372 return 0;
2373 if (s2+l2 <= s1)
2374 return 0;
2375 return 1;
2376 }
2377
2378 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2379 {
2380 unsigned long long blocks;
2381 sector_t new;
2382
2383 if (strict_strtoull(buf, 10, &blocks) < 0)
2384 return -EINVAL;
2385
2386 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2387 return -EINVAL; /* sector conversion overflow */
2388
2389 new = blocks * 2;
2390 if (new != blocks * 2)
2391 return -EINVAL; /* unsigned long long to sector_t overflow */
2392
2393 *sectors = new;
2394 return 0;
2395 }
2396
2397 static ssize_t
2398 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2399 {
2400 mddev_t *my_mddev = rdev->mddev;
2401 sector_t oldsectors = rdev->sectors;
2402 sector_t sectors;
2403
2404 if (strict_blocks_to_sectors(buf, &sectors) < 0)
2405 return -EINVAL;
2406 if (my_mddev->pers && rdev->raid_disk >= 0) {
2407 if (my_mddev->persistent) {
2408 sectors = super_types[my_mddev->major_version].
2409 rdev_size_change(rdev, sectors);
2410 if (!sectors)
2411 return -EBUSY;
2412 } else if (!sectors)
2413 sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2414 rdev->data_offset;
2415 }
2416 if (sectors < my_mddev->dev_sectors)
2417 return -EINVAL; /* component must fit device */
2418
2419 rdev->sectors = sectors;
2420 if (sectors > oldsectors && my_mddev->external) {
2421 /* need to check that all other rdevs with the same ->bdev
2422 * do not overlap. We need to unlock the mddev to avoid
2423 * a deadlock. We have already changed rdev->sectors, and if
2424 * we have to change it back, we will have the lock again.
2425 */
2426 mddev_t *mddev;
2427 int overlap = 0;
2428 struct list_head *tmp;
2429
2430 mddev_unlock(my_mddev);
2431 for_each_mddev(mddev, tmp) {
2432 mdk_rdev_t *rdev2;
2433
2434 mddev_lock(mddev);
2435 list_for_each_entry(rdev2, &mddev->disks, same_set)
2436 if (test_bit(AllReserved, &rdev2->flags) ||
2437 (rdev->bdev == rdev2->bdev &&
2438 rdev != rdev2 &&
2439 overlaps(rdev->data_offset, rdev->sectors,
2440 rdev2->data_offset,
2441 rdev2->sectors))) {
2442 overlap = 1;
2443 break;
2444 }
2445 mddev_unlock(mddev);
2446 if (overlap) {
2447 mddev_put(mddev);
2448 break;
2449 }
2450 }
2451 mddev_lock(my_mddev);
2452 if (overlap) {
2453 /* Someone else could have slipped in a size
2454 * change here, but doing so is just silly.
2455 * We put oldsectors back because we *know* it is
2456 * safe, and trust userspace not to race with
2457 * itself
2458 */
2459 rdev->sectors = oldsectors;
2460 return -EBUSY;
2461 }
2462 }
2463 return len;
2464 }
2465
2466 static struct rdev_sysfs_entry rdev_size =
2467 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2468
2469 static struct attribute *rdev_default_attrs[] = {
2470 &rdev_state.attr,
2471 &rdev_errors.attr,
2472 &rdev_slot.attr,
2473 &rdev_offset.attr,
2474 &rdev_size.attr,
2475 NULL,
2476 };
2477 static ssize_t
2478 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2479 {
2480 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2481 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2482 mddev_t *mddev = rdev->mddev;
2483 ssize_t rv;
2484
2485 if (!entry->show)
2486 return -EIO;
2487
2488 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2489 if (!rv) {
2490 if (rdev->mddev == NULL)
2491 rv = -EBUSY;
2492 else
2493 rv = entry->show(rdev, page);
2494 mddev_unlock(mddev);
2495 }
2496 return rv;
2497 }
2498
2499 static ssize_t
2500 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2501 const char *page, size_t length)
2502 {
2503 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2504 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2505 ssize_t rv;
2506 mddev_t *mddev = rdev->mddev;
2507
2508 if (!entry->store)
2509 return -EIO;
2510 if (!capable(CAP_SYS_ADMIN))
2511 return -EACCES;
2512 rv = mddev ? mddev_lock(mddev): -EBUSY;
2513 if (!rv) {
2514 if (rdev->mddev == NULL)
2515 rv = -EBUSY;
2516 else
2517 rv = entry->store(rdev, page, length);
2518 mddev_unlock(mddev);
2519 }
2520 return rv;
2521 }
2522
2523 static void rdev_free(struct kobject *ko)
2524 {
2525 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2526 kfree(rdev);
2527 }
2528 static struct sysfs_ops rdev_sysfs_ops = {
2529 .show = rdev_attr_show,
2530 .store = rdev_attr_store,
2531 };
2532 static struct kobj_type rdev_ktype = {
2533 .release = rdev_free,
2534 .sysfs_ops = &rdev_sysfs_ops,
2535 .default_attrs = rdev_default_attrs,
2536 };
2537
2538 /*
2539 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2540 *
2541 * mark the device faulty if:
2542 *
2543 * - the device is nonexistent (zero size)
2544 * - the device has no valid superblock
2545 *
2546 * a faulty rdev _never_ has rdev->sb set.
2547 */
2548 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2549 {
2550 char b[BDEVNAME_SIZE];
2551 int err;
2552 mdk_rdev_t *rdev;
2553 sector_t size;
2554
2555 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2556 if (!rdev) {
2557 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2558 return ERR_PTR(-ENOMEM);
2559 }
2560
2561 if ((err = alloc_disk_sb(rdev)))
2562 goto abort_free;
2563
2564 err = lock_rdev(rdev, newdev, super_format == -2);
2565 if (err)
2566 goto abort_free;
2567
2568 kobject_init(&rdev->kobj, &rdev_ktype);
2569
2570 rdev->desc_nr = -1;
2571 rdev->saved_raid_disk = -1;
2572 rdev->raid_disk = -1;
2573 rdev->flags = 0;
2574 rdev->data_offset = 0;
2575 rdev->sb_events = 0;
2576 atomic_set(&rdev->nr_pending, 0);
2577 atomic_set(&rdev->read_errors, 0);
2578 atomic_set(&rdev->corrected_errors, 0);
2579
2580 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2581 if (!size) {
2582 printk(KERN_WARNING
2583 "md: %s has zero or unknown size, marking faulty!\n",
2584 bdevname(rdev->bdev,b));
2585 err = -EINVAL;
2586 goto abort_free;
2587 }
2588
2589 if (super_format >= 0) {
2590 err = super_types[super_format].
2591 load_super(rdev, NULL, super_minor);
2592 if (err == -EINVAL) {
2593 printk(KERN_WARNING
2594 "md: %s does not have a valid v%d.%d "
2595 "superblock, not importing!\n",
2596 bdevname(rdev->bdev,b),
2597 super_format, super_minor);
2598 goto abort_free;
2599 }
2600 if (err < 0) {
2601 printk(KERN_WARNING
2602 "md: could not read %s's sb, not importing!\n",
2603 bdevname(rdev->bdev,b));
2604 goto abort_free;
2605 }
2606 }
2607
2608 INIT_LIST_HEAD(&rdev->same_set);
2609 init_waitqueue_head(&rdev->blocked_wait);
2610
2611 return rdev;
2612
2613 abort_free:
2614 if (rdev->sb_page) {
2615 if (rdev->bdev)
2616 unlock_rdev(rdev);
2617 free_disk_sb(rdev);
2618 }
2619 kfree(rdev);
2620 return ERR_PTR(err);
2621 }
2622
2623 /*
2624 * Check a full RAID array for plausibility
2625 */
2626
2627
2628 static void analyze_sbs(mddev_t * mddev)
2629 {
2630 int i;
2631 mdk_rdev_t *rdev, *freshest, *tmp;
2632 char b[BDEVNAME_SIZE];
2633
2634 freshest = NULL;
2635 rdev_for_each(rdev, tmp, mddev)
2636 switch (super_types[mddev->major_version].
2637 load_super(rdev, freshest, mddev->minor_version)) {
2638 case 1:
2639 freshest = rdev;
2640 break;
2641 case 0:
2642 break;
2643 default:
2644 printk( KERN_ERR \
2645 "md: fatal superblock inconsistency in %s"
2646 " -- removing from array\n",
2647 bdevname(rdev->bdev,b));
2648 kick_rdev_from_array(rdev);
2649 }
2650
2651
2652 super_types[mddev->major_version].
2653 validate_super(mddev, freshest);
2654
2655 i = 0;
2656 rdev_for_each(rdev, tmp, mddev) {
2657 if (rdev->desc_nr >= mddev->max_disks ||
2658 i > mddev->max_disks) {
2659 printk(KERN_WARNING
2660 "md: %s: %s: only %d devices permitted\n",
2661 mdname(mddev), bdevname(rdev->bdev, b),
2662 mddev->max_disks);
2663 kick_rdev_from_array(rdev);
2664 continue;
2665 }
2666 if (rdev != freshest)
2667 if (super_types[mddev->major_version].
2668 validate_super(mddev, rdev)) {
2669 printk(KERN_WARNING "md: kicking non-fresh %s"
2670 " from array!\n",
2671 bdevname(rdev->bdev,b));
2672 kick_rdev_from_array(rdev);
2673 continue;
2674 }
2675 if (mddev->level == LEVEL_MULTIPATH) {
2676 rdev->desc_nr = i++;
2677 rdev->raid_disk = rdev->desc_nr;
2678 set_bit(In_sync, &rdev->flags);
2679 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
2680 rdev->raid_disk = -1;
2681 clear_bit(In_sync, &rdev->flags);
2682 }
2683 }
2684 }
2685
2686 static void md_safemode_timeout(unsigned long data);
2687
2688 static ssize_t
2689 safe_delay_show(mddev_t *mddev, char *page)
2690 {
2691 int msec = (mddev->safemode_delay*1000)/HZ;
2692 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2693 }
2694 static ssize_t
2695 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2696 {
2697 int scale=1;
2698 int dot=0;
2699 int i;
2700 unsigned long msec;
2701 char buf[30];
2702
2703 /* remove a period, and count digits after it */
2704 if (len >= sizeof(buf))
2705 return -EINVAL;
2706 strlcpy(buf, cbuf, sizeof(buf));
2707 for (i=0; i<len; i++) {
2708 if (dot) {
2709 if (isdigit(buf[i])) {
2710 buf[i-1] = buf[i];
2711 scale *= 10;
2712 }
2713 buf[i] = 0;
2714 } else if (buf[i] == '.') {
2715 dot=1;
2716 buf[i] = 0;
2717 }
2718 }
2719 if (strict_strtoul(buf, 10, &msec) < 0)
2720 return -EINVAL;
2721 msec = (msec * 1000) / scale;
2722 if (msec == 0)
2723 mddev->safemode_delay = 0;
2724 else {
2725 unsigned long old_delay = mddev->safemode_delay;
2726 mddev->safemode_delay = (msec*HZ)/1000;
2727 if (mddev->safemode_delay == 0)
2728 mddev->safemode_delay = 1;
2729 if (mddev->safemode_delay < old_delay)
2730 md_safemode_timeout((unsigned long)mddev);
2731 }
2732 return len;
2733 }
2734 static struct md_sysfs_entry md_safe_delay =
2735 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2736
2737 static ssize_t
2738 level_show(mddev_t *mddev, char *page)
2739 {
2740 struct mdk_personality *p = mddev->pers;
2741 if (p)
2742 return sprintf(page, "%s\n", p->name);
2743 else if (mddev->clevel[0])
2744 return sprintf(page, "%s\n", mddev->clevel);
2745 else if (mddev->level != LEVEL_NONE)
2746 return sprintf(page, "%d\n", mddev->level);
2747 else
2748 return 0;
2749 }
2750
2751 static ssize_t
2752 level_store(mddev_t *mddev, const char *buf, size_t len)
2753 {
2754 char level[16];
2755 ssize_t rv = len;
2756 struct mdk_personality *pers;
2757 void *priv;
2758 mdk_rdev_t *rdev;
2759
2760 if (mddev->pers == NULL) {
2761 if (len == 0)
2762 return 0;
2763 if (len >= sizeof(mddev->clevel))
2764 return -ENOSPC;
2765 strncpy(mddev->clevel, buf, len);
2766 if (mddev->clevel[len-1] == '\n')
2767 len--;
2768 mddev->clevel[len] = 0;
2769 mddev->level = LEVEL_NONE;
2770 return rv;
2771 }
2772
2773 /* request to change the personality. Need to ensure:
2774 * - array is not engaged in resync/recovery/reshape
2775 * - old personality can be suspended
2776 * - new personality will access other array.
2777 */
2778
2779 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
2780 return -EBUSY;
2781
2782 if (!mddev->pers->quiesce) {
2783 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
2784 mdname(mddev), mddev->pers->name);
2785 return -EINVAL;
2786 }
2787
2788 /* Now find the new personality */
2789 if (len == 0 || len >= sizeof(level))
2790 return -EINVAL;
2791 strncpy(level, buf, len);
2792 if (level[len-1] == '\n')
2793 len--;
2794 level[len] = 0;
2795
2796 request_module("md-%s", level);
2797 spin_lock(&pers_lock);
2798 pers = find_pers(LEVEL_NONE, level);
2799 if (!pers || !try_module_get(pers->owner)) {
2800 spin_unlock(&pers_lock);
2801 printk(KERN_WARNING "md: personality %s not loaded\n", level);
2802 return -EINVAL;
2803 }
2804 spin_unlock(&pers_lock);
2805
2806 if (pers == mddev->pers) {
2807 /* Nothing to do! */
2808 module_put(pers->owner);
2809 return rv;
2810 }
2811 if (!pers->takeover) {
2812 module_put(pers->owner);
2813 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
2814 mdname(mddev), level);
2815 return -EINVAL;
2816 }
2817
2818 /* ->takeover must set new_* and/or delta_disks
2819 * if it succeeds, and may set them when it fails.
2820 */
2821 priv = pers->takeover(mddev);
2822 if (IS_ERR(priv)) {
2823 mddev->new_level = mddev->level;
2824 mddev->new_layout = mddev->layout;
2825 mddev->new_chunk_sectors = mddev->chunk_sectors;
2826 mddev->raid_disks -= mddev->delta_disks;
2827 mddev->delta_disks = 0;
2828 module_put(pers->owner);
2829 printk(KERN_WARNING "md: %s: %s would not accept array\n",
2830 mdname(mddev), level);
2831 return PTR_ERR(priv);
2832 }
2833
2834 /* Looks like we have a winner */
2835 mddev_suspend(mddev);
2836 mddev->pers->stop(mddev);
2837 module_put(mddev->pers->owner);
2838 /* Invalidate devices that are now superfluous */
2839 list_for_each_entry(rdev, &mddev->disks, same_set)
2840 if (rdev->raid_disk >= mddev->raid_disks) {
2841 rdev->raid_disk = -1;
2842 clear_bit(In_sync, &rdev->flags);
2843 }
2844 mddev->pers = pers;
2845 mddev->private = priv;
2846 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
2847 mddev->level = mddev->new_level;
2848 mddev->layout = mddev->new_layout;
2849 mddev->chunk_sectors = mddev->new_chunk_sectors;
2850 mddev->delta_disks = 0;
2851 pers->run(mddev);
2852 mddev_resume(mddev);
2853 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2854 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2855 md_wakeup_thread(mddev->thread);
2856 return rv;
2857 }
2858
2859 static struct md_sysfs_entry md_level =
2860 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2861
2862
2863 static ssize_t
2864 layout_show(mddev_t *mddev, char *page)
2865 {
2866 /* just a number, not meaningful for all levels */
2867 if (mddev->reshape_position != MaxSector &&
2868 mddev->layout != mddev->new_layout)
2869 return sprintf(page, "%d (%d)\n",
2870 mddev->new_layout, mddev->layout);
2871 return sprintf(page, "%d\n", mddev->layout);
2872 }
2873
2874 static ssize_t
2875 layout_store(mddev_t *mddev, const char *buf, size_t len)
2876 {
2877 char *e;
2878 unsigned long n = simple_strtoul(buf, &e, 10);
2879
2880 if (!*buf || (*e && *e != '\n'))
2881 return -EINVAL;
2882
2883 if (mddev->pers) {
2884 int err;
2885 if (mddev->pers->check_reshape == NULL)
2886 return -EBUSY;
2887 mddev->new_layout = n;
2888 err = mddev->pers->check_reshape(mddev);
2889 if (err) {
2890 mddev->new_layout = mddev->layout;
2891 return err;
2892 }
2893 } else {
2894 mddev->new_layout = n;
2895 if (mddev->reshape_position == MaxSector)
2896 mddev->layout = n;
2897 }
2898 return len;
2899 }
2900 static struct md_sysfs_entry md_layout =
2901 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2902
2903
2904 static ssize_t
2905 raid_disks_show(mddev_t *mddev, char *page)
2906 {
2907 if (mddev->raid_disks == 0)
2908 return 0;
2909 if (mddev->reshape_position != MaxSector &&
2910 mddev->delta_disks != 0)
2911 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2912 mddev->raid_disks - mddev->delta_disks);
2913 return sprintf(page, "%d\n", mddev->raid_disks);
2914 }
2915
2916 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2917
2918 static ssize_t
2919 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2920 {
2921 char *e;
2922 int rv = 0;
2923 unsigned long n = simple_strtoul(buf, &e, 10);
2924
2925 if (!*buf || (*e && *e != '\n'))
2926 return -EINVAL;
2927
2928 if (mddev->pers)
2929 rv = update_raid_disks(mddev, n);
2930 else if (mddev->reshape_position != MaxSector) {
2931 int olddisks = mddev->raid_disks - mddev->delta_disks;
2932 mddev->delta_disks = n - olddisks;
2933 mddev->raid_disks = n;
2934 } else
2935 mddev->raid_disks = n;
2936 return rv ? rv : len;
2937 }
2938 static struct md_sysfs_entry md_raid_disks =
2939 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2940
2941 static ssize_t
2942 chunk_size_show(mddev_t *mddev, char *page)
2943 {
2944 if (mddev->reshape_position != MaxSector &&
2945 mddev->chunk_sectors != mddev->new_chunk_sectors)
2946 return sprintf(page, "%d (%d)\n",
2947 mddev->new_chunk_sectors << 9,
2948 mddev->chunk_sectors << 9);
2949 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
2950 }
2951
2952 static ssize_t
2953 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2954 {
2955 char *e;
2956 unsigned long n = simple_strtoul(buf, &e, 10);
2957
2958 if (!*buf || (*e && *e != '\n'))
2959 return -EINVAL;
2960
2961 if (mddev->pers) {
2962 int err;
2963 if (mddev->pers->check_reshape == NULL)
2964 return -EBUSY;
2965 mddev->new_chunk_sectors = n >> 9;
2966 err = mddev->pers->check_reshape(mddev);
2967 if (err) {
2968 mddev->new_chunk_sectors = mddev->chunk_sectors;
2969 return err;
2970 }
2971 } else {
2972 mddev->new_chunk_sectors = n >> 9;
2973 if (mddev->reshape_position == MaxSector)
2974 mddev->chunk_sectors = n >> 9;
2975 }
2976 return len;
2977 }
2978 static struct md_sysfs_entry md_chunk_size =
2979 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2980
2981 static ssize_t
2982 resync_start_show(mddev_t *mddev, char *page)
2983 {
2984 if (mddev->recovery_cp == MaxSector)
2985 return sprintf(page, "none\n");
2986 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2987 }
2988
2989 static ssize_t
2990 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2991 {
2992 char *e;
2993 unsigned long long n = simple_strtoull(buf, &e, 10);
2994
2995 if (mddev->pers)
2996 return -EBUSY;
2997 if (!*buf || (*e && *e != '\n'))
2998 return -EINVAL;
2999
3000 mddev->recovery_cp = n;
3001 return len;
3002 }
3003 static struct md_sysfs_entry md_resync_start =
3004 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3005
3006 /*
3007 * The array state can be:
3008 *
3009 * clear
3010 * No devices, no size, no level
3011 * Equivalent to STOP_ARRAY ioctl
3012 * inactive
3013 * May have some settings, but array is not active
3014 * all IO results in error
3015 * When written, doesn't tear down array, but just stops it
3016 * suspended (not supported yet)
3017 * All IO requests will block. The array can be reconfigured.
3018 * Writing this, if accepted, will block until array is quiescent
3019 * readonly
3020 * no resync can happen. no superblocks get written.
3021 * write requests fail
3022 * read-auto
3023 * like readonly, but behaves like 'clean' on a write request.
3024 *
3025 * clean - no pending writes, but otherwise active.
3026 * When written to inactive array, starts without resync
3027 * If a write request arrives then
3028 * if metadata is known, mark 'dirty' and switch to 'active'.
3029 * if not known, block and switch to write-pending
3030 * If written to an active array that has pending writes, then fails.
3031 * active
3032 * fully active: IO and resync can be happening.
3033 * When written to inactive array, starts with resync
3034 *
3035 * write-pending
3036 * clean, but writes are blocked waiting for 'active' to be written.
3037 *
3038 * active-idle
3039 * like active, but no writes have been seen for a while (100msec).
3040 *
3041 */
3042 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3043 write_pending, active_idle, bad_word};
3044 static char *array_states[] = {
3045 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3046 "write-pending", "active-idle", NULL };
3047
3048 static int match_word(const char *word, char **list)
3049 {
3050 int n;
3051 for (n=0; list[n]; n++)
3052 if (cmd_match(word, list[n]))
3053 break;
3054 return n;
3055 }
3056
3057 static ssize_t
3058 array_state_show(mddev_t *mddev, char *page)
3059 {
3060 enum array_state st = inactive;
3061
3062 if (mddev->pers)
3063 switch(mddev->ro) {
3064 case 1:
3065 st = readonly;
3066 break;
3067 case 2:
3068 st = read_auto;
3069 break;
3070 case 0:
3071 if (mddev->in_sync)
3072 st = clean;
3073 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
3074 st = write_pending;
3075 else if (mddev->safemode)
3076 st = active_idle;
3077 else
3078 st = active;
3079 }
3080 else {
3081 if (list_empty(&mddev->disks) &&
3082 mddev->raid_disks == 0 &&
3083 mddev->dev_sectors == 0)
3084 st = clear;
3085 else
3086 st = inactive;
3087 }
3088 return sprintf(page, "%s\n", array_states[st]);
3089 }
3090
3091 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3092 static int do_md_run(mddev_t * mddev);
3093 static int restart_array(mddev_t *mddev);
3094
3095 static ssize_t
3096 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3097 {
3098 int err = -EINVAL;
3099 enum array_state st = match_word(buf, array_states);
3100 switch(st) {
3101 case bad_word:
3102 break;
3103 case clear:
3104 /* stopping an active array */
3105 if (atomic_read(&mddev->openers) > 0)
3106 return -EBUSY;
3107 err = do_md_stop(mddev, 0, 0);
3108 break;
3109 case inactive:
3110 /* stopping an active array */
3111 if (mddev->pers) {
3112 if (atomic_read(&mddev->openers) > 0)
3113 return -EBUSY;
3114 err = do_md_stop(mddev, 2, 0);
3115 } else
3116 err = 0; /* already inactive */
3117 break;
3118 case suspended:
3119 break; /* not supported yet */
3120 case readonly:
3121 if (mddev->pers)
3122 err = do_md_stop(mddev, 1, 0);
3123 else {
3124 mddev->ro = 1;
3125 set_disk_ro(mddev->gendisk, 1);
3126 err = do_md_run(mddev);
3127 }
3128 break;
3129 case read_auto:
3130 if (mddev->pers) {
3131 if (mddev->ro == 0)
3132 err = do_md_stop(mddev, 1, 0);
3133 else if (mddev->ro == 1)
3134 err = restart_array(mddev);
3135 if (err == 0) {
3136 mddev->ro = 2;
3137 set_disk_ro(mddev->gendisk, 0);
3138 }
3139 } else {
3140 mddev->ro = 2;
3141 err = do_md_run(mddev);
3142 }
3143 break;
3144 case clean:
3145 if (mddev->pers) {
3146 restart_array(mddev);
3147 spin_lock_irq(&mddev->write_lock);
3148 if (atomic_read(&mddev->writes_pending) == 0) {
3149 if (mddev->in_sync == 0) {
3150 mddev->in_sync = 1;
3151 if (mddev->safemode == 1)
3152 mddev->safemode = 0;
3153 if (mddev->persistent)
3154 set_bit(MD_CHANGE_CLEAN,
3155 &mddev->flags);
3156 }
3157 err = 0;
3158 } else
3159 err = -EBUSY;
3160 spin_unlock_irq(&mddev->write_lock);
3161 } else
3162 err = -EINVAL;
3163 break;
3164 case active:
3165 if (mddev->pers) {
3166 restart_array(mddev);
3167 if (mddev->external)
3168 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3169 wake_up(&mddev->sb_wait);
3170 err = 0;
3171 } else {
3172 mddev->ro = 0;
3173 set_disk_ro(mddev->gendisk, 0);
3174 err = do_md_run(mddev);
3175 }
3176 break;
3177 case write_pending:
3178 case active_idle:
3179 /* these cannot be set */
3180 break;
3181 }
3182 if (err)
3183 return err;
3184 else {
3185 sysfs_notify_dirent(mddev->sysfs_state);
3186 return len;
3187 }
3188 }
3189 static struct md_sysfs_entry md_array_state =
3190 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3191
3192 static ssize_t
3193 null_show(mddev_t *mddev, char *page)
3194 {
3195 return -EINVAL;
3196 }
3197
3198 static ssize_t
3199 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3200 {
3201 /* buf must be %d:%d\n? giving major and minor numbers */
3202 /* The new device is added to the array.
3203 * If the array has a persistent superblock, we read the
3204 * superblock to initialise info and check validity.
3205 * Otherwise, only checking done is that in bind_rdev_to_array,
3206 * which mainly checks size.
3207 */
3208 char *e;
3209 int major = simple_strtoul(buf, &e, 10);
3210 int minor;
3211 dev_t dev;
3212 mdk_rdev_t *rdev;
3213 int err;
3214
3215 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3216 return -EINVAL;
3217 minor = simple_strtoul(e+1, &e, 10);
3218 if (*e && *e != '\n')
3219 return -EINVAL;
3220 dev = MKDEV(major, minor);
3221 if (major != MAJOR(dev) ||
3222 minor != MINOR(dev))
3223 return -EOVERFLOW;
3224
3225
3226 if (mddev->persistent) {
3227 rdev = md_import_device(dev, mddev->major_version,
3228 mddev->minor_version);
3229 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3230 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3231 mdk_rdev_t, same_set);
3232 err = super_types[mddev->major_version]
3233 .load_super(rdev, rdev0, mddev->minor_version);
3234 if (err < 0)
3235 goto out;
3236 }
3237 } else if (mddev->external)
3238 rdev = md_import_device(dev, -2, -1);
3239 else
3240 rdev = md_import_device(dev, -1, -1);
3241
3242 if (IS_ERR(rdev))
3243 return PTR_ERR(rdev);
3244 err = bind_rdev_to_array(rdev, mddev);
3245 out:
3246 if (err)
3247 export_rdev(rdev);
3248 return err ? err : len;
3249 }
3250
3251 static struct md_sysfs_entry md_new_device =
3252 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3253
3254 static ssize_t
3255 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3256 {
3257 char *end;
3258 unsigned long chunk, end_chunk;
3259
3260 if (!mddev->bitmap)
3261 goto out;
3262 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3263 while (*buf) {
3264 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3265 if (buf == end) break;
3266 if (*end == '-') { /* range */
3267 buf = end + 1;
3268 end_chunk = simple_strtoul(buf, &end, 0);
3269 if (buf == end) break;
3270 }
3271 if (*end && !isspace(*end)) break;
3272 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3273 buf = end;
3274 while (isspace(*buf)) buf++;
3275 }
3276 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3277 out:
3278 return len;
3279 }
3280
3281 static struct md_sysfs_entry md_bitmap =
3282 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3283
3284 static ssize_t
3285 size_show(mddev_t *mddev, char *page)
3286 {
3287 return sprintf(page, "%llu\n",
3288 (unsigned long long)mddev->dev_sectors / 2);
3289 }
3290
3291 static int update_size(mddev_t *mddev, sector_t num_sectors);
3292
3293 static ssize_t
3294 size_store(mddev_t *mddev, const char *buf, size_t len)
3295 {
3296 /* If array is inactive, we can reduce the component size, but
3297 * not increase it (except from 0).
3298 * If array is active, we can try an on-line resize
3299 */
3300 sector_t sectors;
3301 int err = strict_blocks_to_sectors(buf, &sectors);
3302
3303 if (err < 0)
3304 return err;
3305 if (mddev->pers) {
3306 err = update_size(mddev, sectors);
3307 md_update_sb(mddev, 1);
3308 } else {
3309 if (mddev->dev_sectors == 0 ||
3310 mddev->dev_sectors > sectors)
3311 mddev->dev_sectors = sectors;
3312 else
3313 err = -ENOSPC;
3314 }
3315 return err ? err : len;
3316 }
3317
3318 static struct md_sysfs_entry md_size =
3319 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3320
3321
3322 /* Metdata version.
3323 * This is one of
3324 * 'none' for arrays with no metadata (good luck...)
3325 * 'external' for arrays with externally managed metadata,
3326 * or N.M for internally known formats
3327 */
3328 static ssize_t
3329 metadata_show(mddev_t *mddev, char *page)
3330 {
3331 if (mddev->persistent)
3332 return sprintf(page, "%d.%d\n",
3333 mddev->major_version, mddev->minor_version);
3334 else if (mddev->external)
3335 return sprintf(page, "external:%s\n", mddev->metadata_type);
3336 else
3337 return sprintf(page, "none\n");
3338 }
3339
3340 static ssize_t
3341 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3342 {
3343 int major, minor;
3344 char *e;
3345 /* Changing the details of 'external' metadata is
3346 * always permitted. Otherwise there must be
3347 * no devices attached to the array.
3348 */
3349 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3350 ;
3351 else if (!list_empty(&mddev->disks))
3352 return -EBUSY;
3353
3354 if (cmd_match(buf, "none")) {
3355 mddev->persistent = 0;
3356 mddev->external = 0;
3357 mddev->major_version = 0;
3358 mddev->minor_version = 90;
3359 return len;
3360 }
3361 if (strncmp(buf, "external:", 9) == 0) {
3362 size_t namelen = len-9;
3363 if (namelen >= sizeof(mddev->metadata_type))
3364 namelen = sizeof(mddev->metadata_type)-1;
3365 strncpy(mddev->metadata_type, buf+9, namelen);
3366 mddev->metadata_type[namelen] = 0;
3367 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3368 mddev->metadata_type[--namelen] = 0;
3369 mddev->persistent = 0;
3370 mddev->external = 1;
3371 mddev->major_version = 0;
3372 mddev->minor_version = 90;
3373 return len;
3374 }
3375 major = simple_strtoul(buf, &e, 10);
3376 if (e==buf || *e != '.')
3377 return -EINVAL;
3378 buf = e+1;
3379 minor = simple_strtoul(buf, &e, 10);
3380 if (e==buf || (*e && *e != '\n') )
3381 return -EINVAL;
3382 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3383 return -ENOENT;
3384 mddev->major_version = major;
3385 mddev->minor_version = minor;
3386 mddev->persistent = 1;
3387 mddev->external = 0;
3388 return len;
3389 }
3390
3391 static struct md_sysfs_entry md_metadata =
3392 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3393
3394 static ssize_t
3395 action_show(mddev_t *mddev, char *page)
3396 {
3397 char *type = "idle";
3398 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3399 type = "frozen";
3400 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3401 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3402 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3403 type = "reshape";
3404 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3405 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3406 type = "resync";
3407 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3408 type = "check";
3409 else
3410 type = "repair";
3411 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3412 type = "recover";
3413 }
3414 return sprintf(page, "%s\n", type);
3415 }
3416
3417 static ssize_t
3418 action_store(mddev_t *mddev, const char *page, size_t len)
3419 {
3420 if (!mddev->pers || !mddev->pers->sync_request)
3421 return -EINVAL;
3422
3423 if (cmd_match(page, "frozen"))
3424 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3425 else
3426 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3427
3428 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3429 if (mddev->sync_thread) {
3430 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3431 md_unregister_thread(mddev->sync_thread);
3432 mddev->sync_thread = NULL;
3433 mddev->recovery = 0;
3434 }
3435 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3436 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3437 return -EBUSY;
3438 else if (cmd_match(page, "resync"))
3439 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3440 else if (cmd_match(page, "recover")) {
3441 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3442 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3443 } else if (cmd_match(page, "reshape")) {
3444 int err;
3445 if (mddev->pers->start_reshape == NULL)
3446 return -EINVAL;
3447 err = mddev->pers->start_reshape(mddev);
3448 if (err)
3449 return err;
3450 sysfs_notify(&mddev->kobj, NULL, "degraded");
3451 } else {
3452 if (cmd_match(page, "check"))
3453 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3454 else if (!cmd_match(page, "repair"))
3455 return -EINVAL;
3456 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3457 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3458 }
3459 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3460 md_wakeup_thread(mddev->thread);
3461 sysfs_notify_dirent(mddev->sysfs_action);
3462 return len;
3463 }
3464
3465 static ssize_t
3466 mismatch_cnt_show(mddev_t *mddev, char *page)
3467 {
3468 return sprintf(page, "%llu\n",
3469 (unsigned long long) mddev->resync_mismatches);
3470 }
3471
3472 static struct md_sysfs_entry md_scan_mode =
3473 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3474
3475
3476 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3477
3478 static ssize_t
3479 sync_min_show(mddev_t *mddev, char *page)
3480 {
3481 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3482 mddev->sync_speed_min ? "local": "system");
3483 }
3484
3485 static ssize_t
3486 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3487 {
3488 int min;
3489 char *e;
3490 if (strncmp(buf, "system", 6)==0) {
3491 mddev->sync_speed_min = 0;
3492 return len;
3493 }
3494 min = simple_strtoul(buf, &e, 10);
3495 if (buf == e || (*e && *e != '\n') || min <= 0)
3496 return -EINVAL;
3497 mddev->sync_speed_min = min;
3498 return len;
3499 }
3500
3501 static struct md_sysfs_entry md_sync_min =
3502 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3503
3504 static ssize_t
3505 sync_max_show(mddev_t *mddev, char *page)
3506 {
3507 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3508 mddev->sync_speed_max ? "local": "system");
3509 }
3510
3511 static ssize_t
3512 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3513 {
3514 int max;
3515 char *e;
3516 if (strncmp(buf, "system", 6)==0) {
3517 mddev->sync_speed_max = 0;
3518 return len;
3519 }
3520 max = simple_strtoul(buf, &e, 10);
3521 if (buf == e || (*e && *e != '\n') || max <= 0)
3522 return -EINVAL;
3523 mddev->sync_speed_max = max;
3524 return len;
3525 }
3526
3527 static struct md_sysfs_entry md_sync_max =
3528 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3529
3530 static ssize_t
3531 degraded_show(mddev_t *mddev, char *page)
3532 {
3533 return sprintf(page, "%d\n", mddev->degraded);
3534 }
3535 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3536
3537 static ssize_t
3538 sync_force_parallel_show(mddev_t *mddev, char *page)
3539 {
3540 return sprintf(page, "%d\n", mddev->parallel_resync);
3541 }
3542
3543 static ssize_t
3544 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3545 {
3546 long n;
3547
3548 if (strict_strtol(buf, 10, &n))
3549 return -EINVAL;
3550
3551 if (n != 0 && n != 1)
3552 return -EINVAL;
3553
3554 mddev->parallel_resync = n;
3555
3556 if (mddev->sync_thread)
3557 wake_up(&resync_wait);
3558
3559 return len;
3560 }
3561
3562 /* force parallel resync, even with shared block devices */
3563 static struct md_sysfs_entry md_sync_force_parallel =
3564 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3565 sync_force_parallel_show, sync_force_parallel_store);
3566
3567 static ssize_t
3568 sync_speed_show(mddev_t *mddev, char *page)
3569 {
3570 unsigned long resync, dt, db;
3571 if (mddev->curr_resync == 0)
3572 return sprintf(page, "none\n");
3573 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3574 dt = (jiffies - mddev->resync_mark) / HZ;
3575 if (!dt) dt++;
3576 db = resync - mddev->resync_mark_cnt;
3577 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3578 }
3579
3580 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3581
3582 static ssize_t
3583 sync_completed_show(mddev_t *mddev, char *page)
3584 {
3585 unsigned long max_sectors, resync;
3586
3587 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3588 return sprintf(page, "none\n");
3589
3590 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3591 max_sectors = mddev->resync_max_sectors;
3592 else
3593 max_sectors = mddev->dev_sectors;
3594
3595 resync = mddev->curr_resync_completed;
3596 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3597 }
3598
3599 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3600
3601 static ssize_t
3602 min_sync_show(mddev_t *mddev, char *page)
3603 {
3604 return sprintf(page, "%llu\n",
3605 (unsigned long long)mddev->resync_min);
3606 }
3607 static ssize_t
3608 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3609 {
3610 unsigned long long min;
3611 if (strict_strtoull(buf, 10, &min))
3612 return -EINVAL;
3613 if (min > mddev->resync_max)
3614 return -EINVAL;
3615 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3616 return -EBUSY;
3617
3618 /* Must be a multiple of chunk_size */
3619 if (mddev->chunk_sectors) {
3620 sector_t temp = min;
3621 if (sector_div(temp, mddev->chunk_sectors))
3622 return -EINVAL;
3623 }
3624 mddev->resync_min = min;
3625
3626 return len;
3627 }
3628
3629 static struct md_sysfs_entry md_min_sync =
3630 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3631
3632 static ssize_t
3633 max_sync_show(mddev_t *mddev, char *page)
3634 {
3635 if (mddev->resync_max == MaxSector)
3636 return sprintf(page, "max\n");
3637 else
3638 return sprintf(page, "%llu\n",
3639 (unsigned long long)mddev->resync_max);
3640 }
3641 static ssize_t
3642 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3643 {
3644 if (strncmp(buf, "max", 3) == 0)
3645 mddev->resync_max = MaxSector;
3646 else {
3647 unsigned long long max;
3648 if (strict_strtoull(buf, 10, &max))
3649 return -EINVAL;
3650 if (max < mddev->resync_min)
3651 return -EINVAL;
3652 if (max < mddev->resync_max &&
3653 mddev->ro == 0 &&
3654 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3655 return -EBUSY;
3656
3657 /* Must be a multiple of chunk_size */
3658 if (mddev->chunk_sectors) {
3659 sector_t temp = max;
3660 if (sector_div(temp, mddev->chunk_sectors))
3661 return -EINVAL;
3662 }
3663 mddev->resync_max = max;
3664 }
3665 wake_up(&mddev->recovery_wait);
3666 return len;
3667 }
3668
3669 static struct md_sysfs_entry md_max_sync =
3670 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3671
3672 static ssize_t
3673 suspend_lo_show(mddev_t *mddev, char *page)
3674 {
3675 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3676 }
3677
3678 static ssize_t
3679 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3680 {
3681 char *e;
3682 unsigned long long new = simple_strtoull(buf, &e, 10);
3683
3684 if (mddev->pers == NULL ||
3685 mddev->pers->quiesce == NULL)
3686 return -EINVAL;
3687 if (buf == e || (*e && *e != '\n'))
3688 return -EINVAL;
3689 if (new >= mddev->suspend_hi ||
3690 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3691 mddev->suspend_lo = new;
3692 mddev->pers->quiesce(mddev, 2);
3693 return len;
3694 } else
3695 return -EINVAL;
3696 }
3697 static struct md_sysfs_entry md_suspend_lo =
3698 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3699
3700
3701 static ssize_t
3702 suspend_hi_show(mddev_t *mddev, char *page)
3703 {
3704 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3705 }
3706
3707 static ssize_t
3708 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3709 {
3710 char *e;
3711 unsigned long long new = simple_strtoull(buf, &e, 10);
3712
3713 if (mddev->pers == NULL ||
3714 mddev->pers->quiesce == NULL)
3715 return -EINVAL;
3716 if (buf == e || (*e && *e != '\n'))
3717 return -EINVAL;
3718 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3719 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3720 mddev->suspend_hi = new;
3721 mddev->pers->quiesce(mddev, 1);
3722 mddev->pers->quiesce(mddev, 0);
3723 return len;
3724 } else
3725 return -EINVAL;
3726 }
3727 static struct md_sysfs_entry md_suspend_hi =
3728 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3729
3730 static ssize_t
3731 reshape_position_show(mddev_t *mddev, char *page)
3732 {
3733 if (mddev->reshape_position != MaxSector)
3734 return sprintf(page, "%llu\n",
3735 (unsigned long long)mddev->reshape_position);
3736 strcpy(page, "none\n");
3737 return 5;
3738 }
3739
3740 static ssize_t
3741 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3742 {
3743 char *e;
3744 unsigned long long new = simple_strtoull(buf, &e, 10);
3745 if (mddev->pers)
3746 return -EBUSY;
3747 if (buf == e || (*e && *e != '\n'))
3748 return -EINVAL;
3749 mddev->reshape_position = new;
3750 mddev->delta_disks = 0;
3751 mddev->new_level = mddev->level;
3752 mddev->new_layout = mddev->layout;
3753 mddev->new_chunk_sectors = mddev->chunk_sectors;
3754 return len;
3755 }
3756
3757 static struct md_sysfs_entry md_reshape_position =
3758 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3759 reshape_position_store);
3760
3761 static ssize_t
3762 array_size_show(mddev_t *mddev, char *page)
3763 {
3764 if (mddev->external_size)
3765 return sprintf(page, "%llu\n",
3766 (unsigned long long)mddev->array_sectors/2);
3767 else
3768 return sprintf(page, "default\n");
3769 }
3770
3771 static ssize_t
3772 array_size_store(mddev_t *mddev, const char *buf, size_t len)
3773 {
3774 sector_t sectors;
3775
3776 if (strncmp(buf, "default", 7) == 0) {
3777 if (mddev->pers)
3778 sectors = mddev->pers->size(mddev, 0, 0);
3779 else
3780 sectors = mddev->array_sectors;
3781
3782 mddev->external_size = 0;
3783 } else {
3784 if (strict_blocks_to_sectors(buf, &sectors) < 0)
3785 return -EINVAL;
3786 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
3787 return -E2BIG;
3788
3789 mddev->external_size = 1;
3790 }
3791
3792 mddev->array_sectors = sectors;
3793 set_capacity(mddev->gendisk, mddev->array_sectors);
3794 if (mddev->pers)
3795 revalidate_disk(mddev->gendisk);
3796
3797 return len;
3798 }
3799
3800 static struct md_sysfs_entry md_array_size =
3801 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
3802 array_size_store);
3803
3804 static struct attribute *md_default_attrs[] = {
3805 &md_level.attr,
3806 &md_layout.attr,
3807 &md_raid_disks.attr,
3808 &md_chunk_size.attr,
3809 &md_size.attr,
3810 &md_resync_start.attr,
3811 &md_metadata.attr,
3812 &md_new_device.attr,
3813 &md_safe_delay.attr,
3814 &md_array_state.attr,
3815 &md_reshape_position.attr,
3816 &md_array_size.attr,
3817 NULL,
3818 };
3819
3820 static struct attribute *md_redundancy_attrs[] = {
3821 &md_scan_mode.attr,
3822 &md_mismatches.attr,
3823 &md_sync_min.attr,
3824 &md_sync_max.attr,
3825 &md_sync_speed.attr,
3826 &md_sync_force_parallel.attr,
3827 &md_sync_completed.attr,
3828 &md_min_sync.attr,
3829 &md_max_sync.attr,
3830 &md_suspend_lo.attr,
3831 &md_suspend_hi.attr,
3832 &md_bitmap.attr,
3833 &md_degraded.attr,
3834 NULL,
3835 };
3836 static struct attribute_group md_redundancy_group = {
3837 .name = NULL,
3838 .attrs = md_redundancy_attrs,
3839 };
3840
3841
3842 static ssize_t
3843 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3844 {
3845 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3846 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3847 ssize_t rv;
3848
3849 if (!entry->show)
3850 return -EIO;
3851 rv = mddev_lock(mddev);
3852 if (!rv) {
3853 rv = entry->show(mddev, page);
3854 mddev_unlock(mddev);
3855 }
3856 return rv;
3857 }
3858
3859 static ssize_t
3860 md_attr_store(struct kobject *kobj, struct attribute *attr,
3861 const char *page, size_t length)
3862 {
3863 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3864 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3865 ssize_t rv;
3866
3867 if (!entry->store)
3868 return -EIO;
3869 if (!capable(CAP_SYS_ADMIN))
3870 return -EACCES;
3871 rv = mddev_lock(mddev);
3872 if (mddev->hold_active == UNTIL_IOCTL)
3873 mddev->hold_active = 0;
3874 if (!rv) {
3875 rv = entry->store(mddev, page, length);
3876 mddev_unlock(mddev);
3877 }
3878 return rv;
3879 }
3880
3881 static void md_free(struct kobject *ko)
3882 {
3883 mddev_t *mddev = container_of(ko, mddev_t, kobj);
3884
3885 if (mddev->sysfs_state)
3886 sysfs_put(mddev->sysfs_state);
3887
3888 if (mddev->gendisk) {
3889 del_gendisk(mddev->gendisk);
3890 put_disk(mddev->gendisk);
3891 }
3892 if (mddev->queue)
3893 blk_cleanup_queue(mddev->queue);
3894
3895 kfree(mddev);
3896 }
3897
3898 static struct sysfs_ops md_sysfs_ops = {
3899 .show = md_attr_show,
3900 .store = md_attr_store,
3901 };
3902 static struct kobj_type md_ktype = {
3903 .release = md_free,
3904 .sysfs_ops = &md_sysfs_ops,
3905 .default_attrs = md_default_attrs,
3906 };
3907
3908 int mdp_major = 0;
3909
3910 static void mddev_delayed_delete(struct work_struct *ws)
3911 {
3912 mddev_t *mddev = container_of(ws, mddev_t, del_work);
3913
3914 if (mddev->private == &md_redundancy_group) {
3915 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3916 if (mddev->sysfs_action)
3917 sysfs_put(mddev->sysfs_action);
3918 mddev->sysfs_action = NULL;
3919 mddev->private = NULL;
3920 }
3921 kobject_del(&mddev->kobj);
3922 kobject_put(&mddev->kobj);
3923 }
3924
3925 static int md_alloc(dev_t dev, char *name)
3926 {
3927 static DEFINE_MUTEX(disks_mutex);
3928 mddev_t *mddev = mddev_find(dev);
3929 struct gendisk *disk;
3930 int partitioned;
3931 int shift;
3932 int unit;
3933 int error;
3934
3935 if (!mddev)
3936 return -ENODEV;
3937
3938 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
3939 shift = partitioned ? MdpMinorShift : 0;
3940 unit = MINOR(mddev->unit) >> shift;
3941
3942 /* wait for any previous instance if this device
3943 * to be completed removed (mddev_delayed_delete).
3944 */
3945 flush_scheduled_work();
3946
3947 mutex_lock(&disks_mutex);
3948 error = -EEXIST;
3949 if (mddev->gendisk)
3950 goto abort;
3951
3952 if (name) {
3953 /* Need to ensure that 'name' is not a duplicate.
3954 */
3955 mddev_t *mddev2;
3956 spin_lock(&all_mddevs_lock);
3957
3958 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
3959 if (mddev2->gendisk &&
3960 strcmp(mddev2->gendisk->disk_name, name) == 0) {
3961 spin_unlock(&all_mddevs_lock);
3962 goto abort;
3963 }
3964 spin_unlock(&all_mddevs_lock);
3965 }
3966
3967 error = -ENOMEM;
3968 mddev->queue = blk_alloc_queue(GFP_KERNEL);
3969 if (!mddev->queue)
3970 goto abort;
3971 mddev->queue->queuedata = mddev;
3972
3973 blk_queue_make_request(mddev->queue, md_make_request);
3974
3975 disk = alloc_disk(1 << shift);
3976 if (!disk) {
3977 blk_cleanup_queue(mddev->queue);
3978 mddev->queue = NULL;
3979 goto abort;
3980 }
3981 disk->major = MAJOR(mddev->unit);
3982 disk->first_minor = unit << shift;
3983 if (name)
3984 strcpy(disk->disk_name, name);
3985 else if (partitioned)
3986 sprintf(disk->disk_name, "md_d%d", unit);
3987 else
3988 sprintf(disk->disk_name, "md%d", unit);
3989 disk->fops = &md_fops;
3990 disk->private_data = mddev;
3991 disk->queue = mddev->queue;
3992 /* Allow extended partitions. This makes the
3993 * 'mdp' device redundant, but we can't really
3994 * remove it now.
3995 */
3996 disk->flags |= GENHD_FL_EXT_DEVT;
3997 add_disk(disk);
3998 mddev->gendisk = disk;
3999 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4000 &disk_to_dev(disk)->kobj, "%s", "md");
4001 if (error) {
4002 /* This isn't possible, but as kobject_init_and_add is marked
4003 * __must_check, we must do something with the result
4004 */
4005 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4006 disk->disk_name);
4007 error = 0;
4008 }
4009 abort:
4010 mutex_unlock(&disks_mutex);
4011 if (!error) {
4012 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4013 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state");
4014 }
4015 mddev_put(mddev);
4016 return error;
4017 }
4018
4019 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4020 {
4021 md_alloc(dev, NULL);
4022 return NULL;
4023 }
4024
4025 static int add_named_array(const char *val, struct kernel_param *kp)
4026 {
4027 /* val must be "md_*" where * is not all digits.
4028 * We allocate an array with a large free minor number, and
4029 * set the name to val. val must not already be an active name.
4030 */
4031 int len = strlen(val);
4032 char buf[DISK_NAME_LEN];
4033
4034 while (len && val[len-1] == '\n')
4035 len--;
4036 if (len >= DISK_NAME_LEN)
4037 return -E2BIG;
4038 strlcpy(buf, val, len+1);
4039 if (strncmp(buf, "md_", 3) != 0)
4040 return -EINVAL;
4041 return md_alloc(0, buf);
4042 }
4043
4044 static void md_safemode_timeout(unsigned long data)
4045 {
4046 mddev_t *mddev = (mddev_t *) data;
4047
4048 if (!atomic_read(&mddev->writes_pending)) {
4049 mddev->safemode = 1;
4050 if (mddev->external)
4051 sysfs_notify_dirent(mddev->sysfs_state);
4052 }
4053 md_wakeup_thread(mddev->thread);
4054 }
4055
4056 static int start_dirty_degraded;
4057
4058 static int do_md_run(mddev_t * mddev)
4059 {
4060 int err;
4061 mdk_rdev_t *rdev;
4062 struct gendisk *disk;
4063 struct mdk_personality *pers;
4064
4065 if (list_empty(&mddev->disks))
4066 /* cannot run an array with no devices.. */
4067 return -EINVAL;
4068
4069 if (mddev->pers)
4070 return -EBUSY;
4071
4072 /*
4073 * Analyze all RAID superblock(s)
4074 */
4075 if (!mddev->raid_disks) {
4076 if (!mddev->persistent)
4077 return -EINVAL;
4078 analyze_sbs(mddev);
4079 }
4080
4081 if (mddev->level != LEVEL_NONE)
4082 request_module("md-level-%d", mddev->level);
4083 else if (mddev->clevel[0])
4084 request_module("md-%s", mddev->clevel);
4085
4086 /*
4087 * Drop all container device buffers, from now on
4088 * the only valid external interface is through the md
4089 * device.
4090 */
4091 list_for_each_entry(rdev, &mddev->disks, same_set) {
4092 if (test_bit(Faulty, &rdev->flags))
4093 continue;
4094 sync_blockdev(rdev->bdev);
4095 invalidate_bdev(rdev->bdev);
4096
4097 /* perform some consistency tests on the device.
4098 * We don't want the data to overlap the metadata,
4099 * Internal Bitmap issues have been handled elsewhere.
4100 */
4101 if (rdev->data_offset < rdev->sb_start) {
4102 if (mddev->dev_sectors &&
4103 rdev->data_offset + mddev->dev_sectors
4104 > rdev->sb_start) {
4105 printk("md: %s: data overlaps metadata\n",
4106 mdname(mddev));
4107 return -EINVAL;
4108 }
4109 } else {
4110 if (rdev->sb_start + rdev->sb_size/512
4111 > rdev->data_offset) {
4112 printk("md: %s: metadata overlaps data\n",
4113 mdname(mddev));
4114 return -EINVAL;
4115 }
4116 }
4117 sysfs_notify_dirent(rdev->sysfs_state);
4118 }
4119
4120 md_probe(mddev->unit, NULL, NULL);
4121 disk = mddev->gendisk;
4122 if (!disk)
4123 return -ENOMEM;
4124
4125 spin_lock(&pers_lock);
4126 pers = find_pers(mddev->level, mddev->clevel);
4127 if (!pers || !try_module_get(pers->owner)) {
4128 spin_unlock(&pers_lock);
4129 if (mddev->level != LEVEL_NONE)
4130 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4131 mddev->level);
4132 else
4133 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4134 mddev->clevel);
4135 return -EINVAL;
4136 }
4137 mddev->pers = pers;
4138 spin_unlock(&pers_lock);
4139 if (mddev->level != pers->level) {
4140 mddev->level = pers->level;
4141 mddev->new_level = pers->level;
4142 }
4143 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4144
4145 if (mddev->reshape_position != MaxSector &&
4146 pers->start_reshape == NULL) {
4147 /* This personality cannot handle reshaping... */
4148 mddev->pers = NULL;
4149 module_put(pers->owner);
4150 return -EINVAL;
4151 }
4152
4153 if (pers->sync_request) {
4154 /* Warn if this is a potentially silly
4155 * configuration.
4156 */
4157 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4158 mdk_rdev_t *rdev2;
4159 int warned = 0;
4160
4161 list_for_each_entry(rdev, &mddev->disks, same_set)
4162 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4163 if (rdev < rdev2 &&
4164 rdev->bdev->bd_contains ==
4165 rdev2->bdev->bd_contains) {
4166 printk(KERN_WARNING
4167 "%s: WARNING: %s appears to be"
4168 " on the same physical disk as"
4169 " %s.\n",
4170 mdname(mddev),
4171 bdevname(rdev->bdev,b),
4172 bdevname(rdev2->bdev,b2));
4173 warned = 1;
4174 }
4175 }
4176
4177 if (warned)
4178 printk(KERN_WARNING
4179 "True protection against single-disk"
4180 " failure might be compromised.\n");
4181 }
4182
4183 mddev->recovery = 0;
4184 /* may be over-ridden by personality */
4185 mddev->resync_max_sectors = mddev->dev_sectors;
4186
4187 mddev->barriers_work = 1;
4188 mddev->ok_start_degraded = start_dirty_degraded;
4189
4190 if (start_readonly && mddev->ro == 0)
4191 mddev->ro = 2; /* read-only, but switch on first write */
4192
4193 err = mddev->pers->run(mddev);
4194 if (err)
4195 printk(KERN_ERR "md: pers->run() failed ...\n");
4196 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4197 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4198 " but 'external_size' not in effect?\n", __func__);
4199 printk(KERN_ERR
4200 "md: invalid array_size %llu > default size %llu\n",
4201 (unsigned long long)mddev->array_sectors / 2,
4202 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4203 err = -EINVAL;
4204 mddev->pers->stop(mddev);
4205 }
4206 if (err == 0 && mddev->pers->sync_request) {
4207 err = bitmap_create(mddev);
4208 if (err) {
4209 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4210 mdname(mddev), err);
4211 mddev->pers->stop(mddev);
4212 }
4213 }
4214 if (err) {
4215 module_put(mddev->pers->owner);
4216 mddev->pers = NULL;
4217 bitmap_destroy(mddev);
4218 return err;
4219 }
4220 if (mddev->pers->sync_request) {
4221 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4222 printk(KERN_WARNING
4223 "md: cannot register extra attributes for %s\n",
4224 mdname(mddev));
4225 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4226 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4227 mddev->ro = 0;
4228
4229 atomic_set(&mddev->writes_pending,0);
4230 mddev->safemode = 0;
4231 mddev->safemode_timer.function = md_safemode_timeout;
4232 mddev->safemode_timer.data = (unsigned long) mddev;
4233 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4234 mddev->in_sync = 1;
4235
4236 list_for_each_entry(rdev, &mddev->disks, same_set)
4237 if (rdev->raid_disk >= 0) {
4238 char nm[20];
4239 sprintf(nm, "rd%d", rdev->raid_disk);
4240 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4241 printk("md: cannot register %s for %s\n",
4242 nm, mdname(mddev));
4243 }
4244
4245 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4246
4247 if (mddev->flags)
4248 md_update_sb(mddev, 0);
4249
4250 set_capacity(disk, mddev->array_sectors);
4251
4252 /* If there is a partially-recovered drive we need to
4253 * start recovery here. If we leave it to md_check_recovery,
4254 * it will remove the drives and not do the right thing
4255 */
4256 if (mddev->degraded && !mddev->sync_thread) {
4257 int spares = 0;
4258 list_for_each_entry(rdev, &mddev->disks, same_set)
4259 if (rdev->raid_disk >= 0 &&
4260 !test_bit(In_sync, &rdev->flags) &&
4261 !test_bit(Faulty, &rdev->flags))
4262 /* complete an interrupted recovery */
4263 spares++;
4264 if (spares && mddev->pers->sync_request) {
4265 mddev->recovery = 0;
4266 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4267 mddev->sync_thread = md_register_thread(md_do_sync,
4268 mddev,
4269 "resync");
4270 if (!mddev->sync_thread) {
4271 printk(KERN_ERR "%s: could not start resync"
4272 " thread...\n",
4273 mdname(mddev));
4274 /* leave the spares where they are, it shouldn't hurt */
4275 mddev->recovery = 0;
4276 }
4277 }
4278 }
4279 md_wakeup_thread(mddev->thread);
4280 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4281
4282 revalidate_disk(mddev->gendisk);
4283 mddev->changed = 1;
4284 md_new_event(mddev);
4285 sysfs_notify_dirent(mddev->sysfs_state);
4286 if (mddev->sysfs_action)
4287 sysfs_notify_dirent(mddev->sysfs_action);
4288 sysfs_notify(&mddev->kobj, NULL, "degraded");
4289 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4290 return 0;
4291 }
4292
4293 static int restart_array(mddev_t *mddev)
4294 {
4295 struct gendisk *disk = mddev->gendisk;
4296
4297 /* Complain if it has no devices */
4298 if (list_empty(&mddev->disks))
4299 return -ENXIO;
4300 if (!mddev->pers)
4301 return -EINVAL;
4302 if (!mddev->ro)
4303 return -EBUSY;
4304 mddev->safemode = 0;
4305 mddev->ro = 0;
4306 set_disk_ro(disk, 0);
4307 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4308 mdname(mddev));
4309 /* Kick recovery or resync if necessary */
4310 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4311 md_wakeup_thread(mddev->thread);
4312 md_wakeup_thread(mddev->sync_thread);
4313 sysfs_notify_dirent(mddev->sysfs_state);
4314 return 0;
4315 }
4316
4317 /* similar to deny_write_access, but accounts for our holding a reference
4318 * to the file ourselves */
4319 static int deny_bitmap_write_access(struct file * file)
4320 {
4321 struct inode *inode = file->f_mapping->host;
4322
4323 spin_lock(&inode->i_lock);
4324 if (atomic_read(&inode->i_writecount) > 1) {
4325 spin_unlock(&inode->i_lock);
4326 return -ETXTBSY;
4327 }
4328 atomic_set(&inode->i_writecount, -1);
4329 spin_unlock(&inode->i_lock);
4330
4331 return 0;
4332 }
4333
4334 static void restore_bitmap_write_access(struct file *file)
4335 {
4336 struct inode *inode = file->f_mapping->host;
4337
4338 spin_lock(&inode->i_lock);
4339 atomic_set(&inode->i_writecount, 1);
4340 spin_unlock(&inode->i_lock);
4341 }
4342
4343 /* mode:
4344 * 0 - completely stop and dis-assemble array
4345 * 1 - switch to readonly
4346 * 2 - stop but do not disassemble array
4347 */
4348 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4349 {
4350 int err = 0;
4351 struct gendisk *disk = mddev->gendisk;
4352 mdk_rdev_t *rdev;
4353
4354 mutex_lock(&mddev->open_mutex);
4355 if (atomic_read(&mddev->openers) > is_open) {
4356 printk("md: %s still in use.\n",mdname(mddev));
4357 err = -EBUSY;
4358 } else if (mddev->pers) {
4359
4360 if (mddev->sync_thread) {
4361 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4362 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4363 md_unregister_thread(mddev->sync_thread);
4364 mddev->sync_thread = NULL;
4365 }
4366
4367 del_timer_sync(&mddev->safemode_timer);
4368
4369 switch(mode) {
4370 case 1: /* readonly */
4371 err = -ENXIO;
4372 if (mddev->ro==1)
4373 goto out;
4374 mddev->ro = 1;
4375 break;
4376 case 0: /* disassemble */
4377 case 2: /* stop */
4378 bitmap_flush(mddev);
4379 md_super_wait(mddev);
4380 if (mddev->ro)
4381 set_disk_ro(disk, 0);
4382
4383 mddev->pers->stop(mddev);
4384 mddev->queue->merge_bvec_fn = NULL;
4385 mddev->queue->unplug_fn = NULL;
4386 mddev->queue->backing_dev_info.congested_fn = NULL;
4387 module_put(mddev->pers->owner);
4388 if (mddev->pers->sync_request)
4389 mddev->private = &md_redundancy_group;
4390 mddev->pers = NULL;
4391 /* tell userspace to handle 'inactive' */
4392 sysfs_notify_dirent(mddev->sysfs_state);
4393
4394 list_for_each_entry(rdev, &mddev->disks, same_set)
4395 if (rdev->raid_disk >= 0) {
4396 char nm[20];
4397 sprintf(nm, "rd%d", rdev->raid_disk);
4398 sysfs_remove_link(&mddev->kobj, nm);
4399 }
4400
4401 set_capacity(disk, 0);
4402 mddev->changed = 1;
4403
4404 if (mddev->ro)
4405 mddev->ro = 0;
4406 }
4407 if (!mddev->in_sync || mddev->flags) {
4408 /* mark array as shutdown cleanly */
4409 mddev->in_sync = 1;
4410 md_update_sb(mddev, 1);
4411 }
4412 if (mode == 1)
4413 set_disk_ro(disk, 1);
4414 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4415 err = 0;
4416 }
4417 out:
4418 mutex_unlock(&mddev->open_mutex);
4419 if (err)
4420 return err;
4421 /*
4422 * Free resources if final stop
4423 */
4424 if (mode == 0) {
4425
4426 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4427
4428 bitmap_destroy(mddev);
4429 if (mddev->bitmap_file) {
4430 restore_bitmap_write_access(mddev->bitmap_file);
4431 fput(mddev->bitmap_file);
4432 mddev->bitmap_file = NULL;
4433 }
4434 mddev->bitmap_offset = 0;
4435
4436 /* make sure all md_delayed_delete calls have finished */
4437 flush_scheduled_work();
4438
4439 export_array(mddev);
4440
4441 mddev->array_sectors = 0;
4442 mddev->external_size = 0;
4443 mddev->dev_sectors = 0;
4444 mddev->raid_disks = 0;
4445 mddev->recovery_cp = 0;
4446 mddev->resync_min = 0;
4447 mddev->resync_max = MaxSector;
4448 mddev->reshape_position = MaxSector;
4449 mddev->external = 0;
4450 mddev->persistent = 0;
4451 mddev->level = LEVEL_NONE;
4452 mddev->clevel[0] = 0;
4453 mddev->flags = 0;
4454 mddev->ro = 0;
4455 mddev->metadata_type[0] = 0;
4456 mddev->chunk_sectors = 0;
4457 mddev->ctime = mddev->utime = 0;
4458 mddev->layout = 0;
4459 mddev->max_disks = 0;
4460 mddev->events = 0;
4461 mddev->delta_disks = 0;
4462 mddev->new_level = LEVEL_NONE;
4463 mddev->new_layout = 0;
4464 mddev->new_chunk_sectors = 0;
4465 mddev->curr_resync = 0;
4466 mddev->resync_mismatches = 0;
4467 mddev->suspend_lo = mddev->suspend_hi = 0;
4468 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4469 mddev->recovery = 0;
4470 mddev->in_sync = 0;
4471 mddev->changed = 0;
4472 mddev->degraded = 0;
4473 mddev->barriers_work = 0;
4474 mddev->safemode = 0;
4475 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4476 if (mddev->hold_active == UNTIL_STOP)
4477 mddev->hold_active = 0;
4478
4479 } else if (mddev->pers)
4480 printk(KERN_INFO "md: %s switched to read-only mode.\n",
4481 mdname(mddev));
4482 err = 0;
4483 blk_integrity_unregister(disk);
4484 md_new_event(mddev);
4485 sysfs_notify_dirent(mddev->sysfs_state);
4486 return err;
4487 }
4488
4489 #ifndef MODULE
4490 static void autorun_array(mddev_t *mddev)
4491 {
4492 mdk_rdev_t *rdev;
4493 int err;
4494
4495 if (list_empty(&mddev->disks))
4496 return;
4497
4498 printk(KERN_INFO "md: running: ");
4499
4500 list_for_each_entry(rdev, &mddev->disks, same_set) {
4501 char b[BDEVNAME_SIZE];
4502 printk("<%s>", bdevname(rdev->bdev,b));
4503 }
4504 printk("\n");
4505
4506 err = do_md_run(mddev);
4507 if (err) {
4508 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4509 do_md_stop(mddev, 0, 0);
4510 }
4511 }
4512
4513 /*
4514 * lets try to run arrays based on all disks that have arrived
4515 * until now. (those are in pending_raid_disks)
4516 *
4517 * the method: pick the first pending disk, collect all disks with
4518 * the same UUID, remove all from the pending list and put them into
4519 * the 'same_array' list. Then order this list based on superblock
4520 * update time (freshest comes first), kick out 'old' disks and
4521 * compare superblocks. If everything's fine then run it.
4522 *
4523 * If "unit" is allocated, then bump its reference count
4524 */
4525 static void autorun_devices(int part)
4526 {
4527 mdk_rdev_t *rdev0, *rdev, *tmp;
4528 mddev_t *mddev;
4529 char b[BDEVNAME_SIZE];
4530
4531 printk(KERN_INFO "md: autorun ...\n");
4532 while (!list_empty(&pending_raid_disks)) {
4533 int unit;
4534 dev_t dev;
4535 LIST_HEAD(candidates);
4536 rdev0 = list_entry(pending_raid_disks.next,
4537 mdk_rdev_t, same_set);
4538
4539 printk(KERN_INFO "md: considering %s ...\n",
4540 bdevname(rdev0->bdev,b));
4541 INIT_LIST_HEAD(&candidates);
4542 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4543 if (super_90_load(rdev, rdev0, 0) >= 0) {
4544 printk(KERN_INFO "md: adding %s ...\n",
4545 bdevname(rdev->bdev,b));
4546 list_move(&rdev->same_set, &candidates);
4547 }
4548 /*
4549 * now we have a set of devices, with all of them having
4550 * mostly sane superblocks. It's time to allocate the
4551 * mddev.
4552 */
4553 if (part) {
4554 dev = MKDEV(mdp_major,
4555 rdev0->preferred_minor << MdpMinorShift);
4556 unit = MINOR(dev) >> MdpMinorShift;
4557 } else {
4558 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4559 unit = MINOR(dev);
4560 }
4561 if (rdev0->preferred_minor != unit) {
4562 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4563 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4564 break;
4565 }
4566
4567 md_probe(dev, NULL, NULL);
4568 mddev = mddev_find(dev);
4569 if (!mddev || !mddev->gendisk) {
4570 if (mddev)
4571 mddev_put(mddev);
4572 printk(KERN_ERR
4573 "md: cannot allocate memory for md drive.\n");
4574 break;
4575 }
4576 if (mddev_lock(mddev))
4577 printk(KERN_WARNING "md: %s locked, cannot run\n",
4578 mdname(mddev));
4579 else if (mddev->raid_disks || mddev->major_version
4580 || !list_empty(&mddev->disks)) {
4581 printk(KERN_WARNING
4582 "md: %s already running, cannot run %s\n",
4583 mdname(mddev), bdevname(rdev0->bdev,b));
4584 mddev_unlock(mddev);
4585 } else {
4586 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4587 mddev->persistent = 1;
4588 rdev_for_each_list(rdev, tmp, &candidates) {
4589 list_del_init(&rdev->same_set);
4590 if (bind_rdev_to_array(rdev, mddev))
4591 export_rdev(rdev);
4592 }
4593 autorun_array(mddev);
4594 mddev_unlock(mddev);
4595 }
4596 /* on success, candidates will be empty, on error
4597 * it won't...
4598 */
4599 rdev_for_each_list(rdev, tmp, &candidates) {
4600 list_del_init(&rdev->same_set);
4601 export_rdev(rdev);
4602 }
4603 mddev_put(mddev);
4604 }
4605 printk(KERN_INFO "md: ... autorun DONE.\n");
4606 }
4607 #endif /* !MODULE */
4608
4609 static int get_version(void __user * arg)
4610 {
4611 mdu_version_t ver;
4612
4613 ver.major = MD_MAJOR_VERSION;
4614 ver.minor = MD_MINOR_VERSION;
4615 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4616
4617 if (copy_to_user(arg, &ver, sizeof(ver)))
4618 return -EFAULT;
4619
4620 return 0;
4621 }
4622
4623 static int get_array_info(mddev_t * mddev, void __user * arg)
4624 {
4625 mdu_array_info_t info;
4626 int nr,working,insync,failed,spare;
4627 mdk_rdev_t *rdev;
4628
4629 nr=working=insync=failed=spare=0;
4630 list_for_each_entry(rdev, &mddev->disks, same_set) {
4631 nr++;
4632 if (test_bit(Faulty, &rdev->flags))
4633 failed++;
4634 else {
4635 working++;
4636 if (test_bit(In_sync, &rdev->flags))
4637 insync++;
4638 else
4639 spare++;
4640 }
4641 }
4642
4643 info.major_version = mddev->major_version;
4644 info.minor_version = mddev->minor_version;
4645 info.patch_version = MD_PATCHLEVEL_VERSION;
4646 info.ctime = mddev->ctime;
4647 info.level = mddev->level;
4648 info.size = mddev->dev_sectors / 2;
4649 if (info.size != mddev->dev_sectors / 2) /* overflow */
4650 info.size = -1;
4651 info.nr_disks = nr;
4652 info.raid_disks = mddev->raid_disks;
4653 info.md_minor = mddev->md_minor;
4654 info.not_persistent= !mddev->persistent;
4655
4656 info.utime = mddev->utime;
4657 info.state = 0;
4658 if (mddev->in_sync)
4659 info.state = (1<<MD_SB_CLEAN);
4660 if (mddev->bitmap && mddev->bitmap_offset)
4661 info.state = (1<<MD_SB_BITMAP_PRESENT);
4662 info.active_disks = insync;
4663 info.working_disks = working;
4664 info.failed_disks = failed;
4665 info.spare_disks = spare;
4666
4667 info.layout = mddev->layout;
4668 info.chunk_size = mddev->chunk_sectors << 9;
4669
4670 if (copy_to_user(arg, &info, sizeof(info)))
4671 return -EFAULT;
4672
4673 return 0;
4674 }
4675
4676 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4677 {
4678 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4679 char *ptr, *buf = NULL;
4680 int err = -ENOMEM;
4681
4682 if (md_allow_write(mddev))
4683 file = kmalloc(sizeof(*file), GFP_NOIO);
4684 else
4685 file = kmalloc(sizeof(*file), GFP_KERNEL);
4686
4687 if (!file)
4688 goto out;
4689
4690 /* bitmap disabled, zero the first byte and copy out */
4691 if (!mddev->bitmap || !mddev->bitmap->file) {
4692 file->pathname[0] = '\0';
4693 goto copy_out;
4694 }
4695
4696 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4697 if (!buf)
4698 goto out;
4699
4700 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4701 if (IS_ERR(ptr))
4702 goto out;
4703
4704 strcpy(file->pathname, ptr);
4705
4706 copy_out:
4707 err = 0;
4708 if (copy_to_user(arg, file, sizeof(*file)))
4709 err = -EFAULT;
4710 out:
4711 kfree(buf);
4712 kfree(file);
4713 return err;
4714 }
4715
4716 static int get_disk_info(mddev_t * mddev, void __user * arg)
4717 {
4718 mdu_disk_info_t info;
4719 mdk_rdev_t *rdev;
4720
4721 if (copy_from_user(&info, arg, sizeof(info)))
4722 return -EFAULT;
4723
4724 rdev = find_rdev_nr(mddev, info.number);
4725 if (rdev) {
4726 info.major = MAJOR(rdev->bdev->bd_dev);
4727 info.minor = MINOR(rdev->bdev->bd_dev);
4728 info.raid_disk = rdev->raid_disk;
4729 info.state = 0;
4730 if (test_bit(Faulty, &rdev->flags))
4731 info.state |= (1<<MD_DISK_FAULTY);
4732 else if (test_bit(In_sync, &rdev->flags)) {
4733 info.state |= (1<<MD_DISK_ACTIVE);
4734 info.state |= (1<<MD_DISK_SYNC);
4735 }
4736 if (test_bit(WriteMostly, &rdev->flags))
4737 info.state |= (1<<MD_DISK_WRITEMOSTLY);
4738 } else {
4739 info.major = info.minor = 0;
4740 info.raid_disk = -1;
4741 info.state = (1<<MD_DISK_REMOVED);
4742 }
4743
4744 if (copy_to_user(arg, &info, sizeof(info)))
4745 return -EFAULT;
4746
4747 return 0;
4748 }
4749
4750 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4751 {
4752 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4753 mdk_rdev_t *rdev;
4754 dev_t dev = MKDEV(info->major,info->minor);
4755
4756 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4757 return -EOVERFLOW;
4758
4759 if (!mddev->raid_disks) {
4760 int err;
4761 /* expecting a device which has a superblock */
4762 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4763 if (IS_ERR(rdev)) {
4764 printk(KERN_WARNING
4765 "md: md_import_device returned %ld\n",
4766 PTR_ERR(rdev));
4767 return PTR_ERR(rdev);
4768 }
4769 if (!list_empty(&mddev->disks)) {
4770 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4771 mdk_rdev_t, same_set);
4772 err = super_types[mddev->major_version]
4773 .load_super(rdev, rdev0, mddev->minor_version);
4774 if (err < 0) {
4775 printk(KERN_WARNING
4776 "md: %s has different UUID to %s\n",
4777 bdevname(rdev->bdev,b),
4778 bdevname(rdev0->bdev,b2));
4779 export_rdev(rdev);
4780 return -EINVAL;
4781 }
4782 }
4783 err = bind_rdev_to_array(rdev, mddev);
4784 if (err)
4785 export_rdev(rdev);
4786 return err;
4787 }
4788
4789 /*
4790 * add_new_disk can be used once the array is assembled
4791 * to add "hot spares". They must already have a superblock
4792 * written
4793 */
4794 if (mddev->pers) {
4795 int err;
4796 if (!mddev->pers->hot_add_disk) {
4797 printk(KERN_WARNING
4798 "%s: personality does not support diskops!\n",
4799 mdname(mddev));
4800 return -EINVAL;
4801 }
4802 if (mddev->persistent)
4803 rdev = md_import_device(dev, mddev->major_version,
4804 mddev->minor_version);
4805 else
4806 rdev = md_import_device(dev, -1, -1);
4807 if (IS_ERR(rdev)) {
4808 printk(KERN_WARNING
4809 "md: md_import_device returned %ld\n",
4810 PTR_ERR(rdev));
4811 return PTR_ERR(rdev);
4812 }
4813 /* set saved_raid_disk if appropriate */
4814 if (!mddev->persistent) {
4815 if (info->state & (1<<MD_DISK_SYNC) &&
4816 info->raid_disk < mddev->raid_disks) {
4817 rdev->raid_disk = info->raid_disk;
4818 set_bit(In_sync, &rdev->flags);
4819 } else
4820 rdev->raid_disk = -1;
4821 } else
4822 super_types[mddev->major_version].
4823 validate_super(mddev, rdev);
4824 if (test_bit(In_sync, &rdev->flags))
4825 rdev->saved_raid_disk = rdev->raid_disk;
4826 else
4827 rdev->saved_raid_disk = -1;
4828
4829 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4830 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4831 set_bit(WriteMostly, &rdev->flags);
4832 else
4833 clear_bit(WriteMostly, &rdev->flags);
4834
4835 rdev->raid_disk = -1;
4836 err = bind_rdev_to_array(rdev, mddev);
4837 if (!err && !mddev->pers->hot_remove_disk) {
4838 /* If there is hot_add_disk but no hot_remove_disk
4839 * then added disks for geometry changes,
4840 * and should be added immediately.
4841 */
4842 super_types[mddev->major_version].
4843 validate_super(mddev, rdev);
4844 err = mddev->pers->hot_add_disk(mddev, rdev);
4845 if (err)
4846 unbind_rdev_from_array(rdev);
4847 }
4848 if (err)
4849 export_rdev(rdev);
4850 else
4851 sysfs_notify_dirent(rdev->sysfs_state);
4852
4853 md_update_sb(mddev, 1);
4854 if (mddev->degraded)
4855 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4856 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4857 md_wakeup_thread(mddev->thread);
4858 return err;
4859 }
4860
4861 /* otherwise, add_new_disk is only allowed
4862 * for major_version==0 superblocks
4863 */
4864 if (mddev->major_version != 0) {
4865 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4866 mdname(mddev));
4867 return -EINVAL;
4868 }
4869
4870 if (!(info->state & (1<<MD_DISK_FAULTY))) {
4871 int err;
4872 rdev = md_import_device(dev, -1, 0);
4873 if (IS_ERR(rdev)) {
4874 printk(KERN_WARNING
4875 "md: error, md_import_device() returned %ld\n",
4876 PTR_ERR(rdev));
4877 return PTR_ERR(rdev);
4878 }
4879 rdev->desc_nr = info->number;
4880 if (info->raid_disk < mddev->raid_disks)
4881 rdev->raid_disk = info->raid_disk;
4882 else
4883 rdev->raid_disk = -1;
4884
4885 if (rdev->raid_disk < mddev->raid_disks)
4886 if (info->state & (1<<MD_DISK_SYNC))
4887 set_bit(In_sync, &rdev->flags);
4888
4889 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4890 set_bit(WriteMostly, &rdev->flags);
4891
4892 if (!mddev->persistent) {
4893 printk(KERN_INFO "md: nonpersistent superblock ...\n");
4894 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4895 } else
4896 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4897 rdev->sectors = rdev->sb_start;
4898
4899 err = bind_rdev_to_array(rdev, mddev);
4900 if (err) {
4901 export_rdev(rdev);
4902 return err;
4903 }
4904 }
4905
4906 return 0;
4907 }
4908
4909 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4910 {
4911 char b[BDEVNAME_SIZE];
4912 mdk_rdev_t *rdev;
4913
4914 rdev = find_rdev(mddev, dev);
4915 if (!rdev)
4916 return -ENXIO;
4917
4918 if (rdev->raid_disk >= 0)
4919 goto busy;
4920
4921 kick_rdev_from_array(rdev);
4922 md_update_sb(mddev, 1);
4923 md_new_event(mddev);
4924
4925 return 0;
4926 busy:
4927 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4928 bdevname(rdev->bdev,b), mdname(mddev));
4929 return -EBUSY;
4930 }
4931
4932 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4933 {
4934 char b[BDEVNAME_SIZE];
4935 int err;
4936 mdk_rdev_t *rdev;
4937
4938 if (!mddev->pers)
4939 return -ENODEV;
4940
4941 if (mddev->major_version != 0) {
4942 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4943 " version-0 superblocks.\n",
4944 mdname(mddev));
4945 return -EINVAL;
4946 }
4947 if (!mddev->pers->hot_add_disk) {
4948 printk(KERN_WARNING
4949 "%s: personality does not support diskops!\n",
4950 mdname(mddev));
4951 return -EINVAL;
4952 }
4953
4954 rdev = md_import_device(dev, -1, 0);
4955 if (IS_ERR(rdev)) {
4956 printk(KERN_WARNING
4957 "md: error, md_import_device() returned %ld\n",
4958 PTR_ERR(rdev));
4959 return -EINVAL;
4960 }
4961
4962 if (mddev->persistent)
4963 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4964 else
4965 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4966
4967 rdev->sectors = rdev->sb_start;
4968
4969 if (test_bit(Faulty, &rdev->flags)) {
4970 printk(KERN_WARNING
4971 "md: can not hot-add faulty %s disk to %s!\n",
4972 bdevname(rdev->bdev,b), mdname(mddev));
4973 err = -EINVAL;
4974 goto abort_export;
4975 }
4976 clear_bit(In_sync, &rdev->flags);
4977 rdev->desc_nr = -1;
4978 rdev->saved_raid_disk = -1;
4979 err = bind_rdev_to_array(rdev, mddev);
4980 if (err)
4981 goto abort_export;
4982
4983 /*
4984 * The rest should better be atomic, we can have disk failures
4985 * noticed in interrupt contexts ...
4986 */
4987
4988 rdev->raid_disk = -1;
4989
4990 md_update_sb(mddev, 1);
4991
4992 /*
4993 * Kick recovery, maybe this spare has to be added to the
4994 * array immediately.
4995 */
4996 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4997 md_wakeup_thread(mddev->thread);
4998 md_new_event(mddev);
4999 return 0;
5000
5001 abort_export:
5002 export_rdev(rdev);
5003 return err;
5004 }
5005
5006 static int set_bitmap_file(mddev_t *mddev, int fd)
5007 {
5008 int err;
5009
5010 if (mddev->pers) {
5011 if (!mddev->pers->quiesce)
5012 return -EBUSY;
5013 if (mddev->recovery || mddev->sync_thread)
5014 return -EBUSY;
5015 /* we should be able to change the bitmap.. */
5016 }
5017
5018
5019 if (fd >= 0) {
5020 if (mddev->bitmap)
5021 return -EEXIST; /* cannot add when bitmap is present */
5022 mddev->bitmap_file = fget(fd);
5023
5024 if (mddev->bitmap_file == NULL) {
5025 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5026 mdname(mddev));
5027 return -EBADF;
5028 }
5029
5030 err = deny_bitmap_write_access(mddev->bitmap_file);
5031 if (err) {
5032 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5033 mdname(mddev));
5034 fput(mddev->bitmap_file);
5035 mddev->bitmap_file = NULL;
5036 return err;
5037 }
5038 mddev->bitmap_offset = 0; /* file overrides offset */
5039 } else if (mddev->bitmap == NULL)
5040 return -ENOENT; /* cannot remove what isn't there */
5041 err = 0;
5042 if (mddev->pers) {
5043 mddev->pers->quiesce(mddev, 1);
5044 if (fd >= 0)
5045 err = bitmap_create(mddev);
5046 if (fd < 0 || err) {
5047 bitmap_destroy(mddev);
5048 fd = -1; /* make sure to put the file */
5049 }
5050 mddev->pers->quiesce(mddev, 0);
5051 }
5052 if (fd < 0) {
5053 if (mddev->bitmap_file) {
5054 restore_bitmap_write_access(mddev->bitmap_file);
5055 fput(mddev->bitmap_file);
5056 }
5057 mddev->bitmap_file = NULL;
5058 }
5059
5060 return err;
5061 }
5062
5063 /*
5064 * set_array_info is used two different ways
5065 * The original usage is when creating a new array.
5066 * In this usage, raid_disks is > 0 and it together with
5067 * level, size, not_persistent,layout,chunksize determine the
5068 * shape of the array.
5069 * This will always create an array with a type-0.90.0 superblock.
5070 * The newer usage is when assembling an array.
5071 * In this case raid_disks will be 0, and the major_version field is
5072 * use to determine which style super-blocks are to be found on the devices.
5073 * The minor and patch _version numbers are also kept incase the
5074 * super_block handler wishes to interpret them.
5075 */
5076 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5077 {
5078
5079 if (info->raid_disks == 0) {
5080 /* just setting version number for superblock loading */
5081 if (info->major_version < 0 ||
5082 info->major_version >= ARRAY_SIZE(super_types) ||
5083 super_types[info->major_version].name == NULL) {
5084 /* maybe try to auto-load a module? */
5085 printk(KERN_INFO
5086 "md: superblock version %d not known\n",
5087 info->major_version);
5088 return -EINVAL;
5089 }
5090 mddev->major_version = info->major_version;
5091 mddev->minor_version = info->minor_version;
5092 mddev->patch_version = info->patch_version;
5093 mddev->persistent = !info->not_persistent;
5094 /* ensure mddev_put doesn't delete this now that there
5095 * is some minimal configuration.
5096 */
5097 mddev->ctime = get_seconds();
5098 return 0;
5099 }
5100 mddev->major_version = MD_MAJOR_VERSION;
5101 mddev->minor_version = MD_MINOR_VERSION;
5102 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5103 mddev->ctime = get_seconds();
5104
5105 mddev->level = info->level;
5106 mddev->clevel[0] = 0;
5107 mddev->dev_sectors = 2 * (sector_t)info->size;
5108 mddev->raid_disks = info->raid_disks;
5109 /* don't set md_minor, it is determined by which /dev/md* was
5110 * openned
5111 */
5112 if (info->state & (1<<MD_SB_CLEAN))
5113 mddev->recovery_cp = MaxSector;
5114 else
5115 mddev->recovery_cp = 0;
5116 mddev->persistent = ! info->not_persistent;
5117 mddev->external = 0;
5118
5119 mddev->layout = info->layout;
5120 mddev->chunk_sectors = info->chunk_size >> 9;
5121
5122 mddev->max_disks = MD_SB_DISKS;
5123
5124 if (mddev->persistent)
5125 mddev->flags = 0;
5126 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5127
5128 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
5129 mddev->bitmap_offset = 0;
5130
5131 mddev->reshape_position = MaxSector;
5132
5133 /*
5134 * Generate a 128 bit UUID
5135 */
5136 get_random_bytes(mddev->uuid, 16);
5137
5138 mddev->new_level = mddev->level;
5139 mddev->new_chunk_sectors = mddev->chunk_sectors;
5140 mddev->new_layout = mddev->layout;
5141 mddev->delta_disks = 0;
5142
5143 return 0;
5144 }
5145
5146 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5147 {
5148 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5149
5150 if (mddev->external_size)
5151 return;
5152
5153 mddev->array_sectors = array_sectors;
5154 }
5155 EXPORT_SYMBOL(md_set_array_sectors);
5156
5157 static int update_size(mddev_t *mddev, sector_t num_sectors)
5158 {
5159 mdk_rdev_t *rdev;
5160 int rv;
5161 int fit = (num_sectors == 0);
5162
5163 if (mddev->pers->resize == NULL)
5164 return -EINVAL;
5165 /* The "num_sectors" is the number of sectors of each device that
5166 * is used. This can only make sense for arrays with redundancy.
5167 * linear and raid0 always use whatever space is available. We can only
5168 * consider changing this number if no resync or reconstruction is
5169 * happening, and if the new size is acceptable. It must fit before the
5170 * sb_start or, if that is <data_offset, it must fit before the size
5171 * of each device. If num_sectors is zero, we find the largest size
5172 * that fits.
5173
5174 */
5175 if (mddev->sync_thread)
5176 return -EBUSY;
5177 if (mddev->bitmap)
5178 /* Sorry, cannot grow a bitmap yet, just remove it,
5179 * grow, and re-add.
5180 */
5181 return -EBUSY;
5182 list_for_each_entry(rdev, &mddev->disks, same_set) {
5183 sector_t avail = rdev->sectors;
5184
5185 if (fit && (num_sectors == 0 || num_sectors > avail))
5186 num_sectors = avail;
5187 if (avail < num_sectors)
5188 return -ENOSPC;
5189 }
5190 rv = mddev->pers->resize(mddev, num_sectors);
5191 if (!rv)
5192 revalidate_disk(mddev->gendisk);
5193 return rv;
5194 }
5195
5196 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5197 {
5198 int rv;
5199 /* change the number of raid disks */
5200 if (mddev->pers->check_reshape == NULL)
5201 return -EINVAL;
5202 if (raid_disks <= 0 ||
5203 raid_disks >= mddev->max_disks)
5204 return -EINVAL;
5205 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5206 return -EBUSY;
5207 mddev->delta_disks = raid_disks - mddev->raid_disks;
5208
5209 rv = mddev->pers->check_reshape(mddev);
5210 return rv;
5211 }
5212
5213
5214 /*
5215 * update_array_info is used to change the configuration of an
5216 * on-line array.
5217 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5218 * fields in the info are checked against the array.
5219 * Any differences that cannot be handled will cause an error.
5220 * Normally, only one change can be managed at a time.
5221 */
5222 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5223 {
5224 int rv = 0;
5225 int cnt = 0;
5226 int state = 0;
5227
5228 /* calculate expected state,ignoring low bits */
5229 if (mddev->bitmap && mddev->bitmap_offset)
5230 state |= (1 << MD_SB_BITMAP_PRESENT);
5231
5232 if (mddev->major_version != info->major_version ||
5233 mddev->minor_version != info->minor_version ||
5234 /* mddev->patch_version != info->patch_version || */
5235 mddev->ctime != info->ctime ||
5236 mddev->level != info->level ||
5237 /* mddev->layout != info->layout || */
5238 !mddev->persistent != info->not_persistent||
5239 mddev->chunk_sectors != info->chunk_size >> 9 ||
5240 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5241 ((state^info->state) & 0xfffffe00)
5242 )
5243 return -EINVAL;
5244 /* Check there is only one change */
5245 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5246 cnt++;
5247 if (mddev->raid_disks != info->raid_disks)
5248 cnt++;
5249 if (mddev->layout != info->layout)
5250 cnt++;
5251 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5252 cnt++;
5253 if (cnt == 0)
5254 return 0;
5255 if (cnt > 1)
5256 return -EINVAL;
5257
5258 if (mddev->layout != info->layout) {
5259 /* Change layout
5260 * we don't need to do anything at the md level, the
5261 * personality will take care of it all.
5262 */
5263 if (mddev->pers->check_reshape == NULL)
5264 return -EINVAL;
5265 else {
5266 mddev->new_layout = info->layout;
5267 rv = mddev->pers->check_reshape(mddev);
5268 if (rv)
5269 mddev->new_layout = mddev->layout;
5270 return rv;
5271 }
5272 }
5273 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5274 rv = update_size(mddev, (sector_t)info->size * 2);
5275
5276 if (mddev->raid_disks != info->raid_disks)
5277 rv = update_raid_disks(mddev, info->raid_disks);
5278
5279 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5280 if (mddev->pers->quiesce == NULL)
5281 return -EINVAL;
5282 if (mddev->recovery || mddev->sync_thread)
5283 return -EBUSY;
5284 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5285 /* add the bitmap */
5286 if (mddev->bitmap)
5287 return -EEXIST;
5288 if (mddev->default_bitmap_offset == 0)
5289 return -EINVAL;
5290 mddev->bitmap_offset = mddev->default_bitmap_offset;
5291 mddev->pers->quiesce(mddev, 1);
5292 rv = bitmap_create(mddev);
5293 if (rv)
5294 bitmap_destroy(mddev);
5295 mddev->pers->quiesce(mddev, 0);
5296 } else {
5297 /* remove the bitmap */
5298 if (!mddev->bitmap)
5299 return -ENOENT;
5300 if (mddev->bitmap->file)
5301 return -EINVAL;
5302 mddev->pers->quiesce(mddev, 1);
5303 bitmap_destroy(mddev);
5304 mddev->pers->quiesce(mddev, 0);
5305 mddev->bitmap_offset = 0;
5306 }
5307 }
5308 md_update_sb(mddev, 1);
5309 return rv;
5310 }
5311
5312 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5313 {
5314 mdk_rdev_t *rdev;
5315
5316 if (mddev->pers == NULL)
5317 return -ENODEV;
5318
5319 rdev = find_rdev(mddev, dev);
5320 if (!rdev)
5321 return -ENODEV;
5322
5323 md_error(mddev, rdev);
5324 return 0;
5325 }
5326
5327 /*
5328 * We have a problem here : there is no easy way to give a CHS
5329 * virtual geometry. We currently pretend that we have a 2 heads
5330 * 4 sectors (with a BIG number of cylinders...). This drives
5331 * dosfs just mad... ;-)
5332 */
5333 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5334 {
5335 mddev_t *mddev = bdev->bd_disk->private_data;
5336
5337 geo->heads = 2;
5338 geo->sectors = 4;
5339 geo->cylinders = get_capacity(mddev->gendisk) / 8;
5340 return 0;
5341 }
5342
5343 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5344 unsigned int cmd, unsigned long arg)
5345 {
5346 int err = 0;
5347 void __user *argp = (void __user *)arg;
5348 mddev_t *mddev = NULL;
5349 int ro;
5350
5351 if (!capable(CAP_SYS_ADMIN))
5352 return -EACCES;
5353
5354 /*
5355 * Commands dealing with the RAID driver but not any
5356 * particular array:
5357 */
5358 switch (cmd)
5359 {
5360 case RAID_VERSION:
5361 err = get_version(argp);
5362 goto done;
5363
5364 case PRINT_RAID_DEBUG:
5365 err = 0;
5366 md_print_devices();
5367 goto done;
5368
5369 #ifndef MODULE
5370 case RAID_AUTORUN:
5371 err = 0;
5372 autostart_arrays(arg);
5373 goto done;
5374 #endif
5375 default:;
5376 }
5377
5378 /*
5379 * Commands creating/starting a new array:
5380 */
5381
5382 mddev = bdev->bd_disk->private_data;
5383
5384 if (!mddev) {
5385 BUG();
5386 goto abort;
5387 }
5388
5389 err = mddev_lock(mddev);
5390 if (err) {
5391 printk(KERN_INFO
5392 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5393 err, cmd);
5394 goto abort;
5395 }
5396
5397 switch (cmd)
5398 {
5399 case SET_ARRAY_INFO:
5400 {
5401 mdu_array_info_t info;
5402 if (!arg)
5403 memset(&info, 0, sizeof(info));
5404 else if (copy_from_user(&info, argp, sizeof(info))) {
5405 err = -EFAULT;
5406 goto abort_unlock;
5407 }
5408 if (mddev->pers) {
5409 err = update_array_info(mddev, &info);
5410 if (err) {
5411 printk(KERN_WARNING "md: couldn't update"
5412 " array info. %d\n", err);
5413 goto abort_unlock;
5414 }
5415 goto done_unlock;
5416 }
5417 if (!list_empty(&mddev->disks)) {
5418 printk(KERN_WARNING
5419 "md: array %s already has disks!\n",
5420 mdname(mddev));
5421 err = -EBUSY;
5422 goto abort_unlock;
5423 }
5424 if (mddev->raid_disks) {
5425 printk(KERN_WARNING
5426 "md: array %s already initialised!\n",
5427 mdname(mddev));
5428 err = -EBUSY;
5429 goto abort_unlock;
5430 }
5431 err = set_array_info(mddev, &info);
5432 if (err) {
5433 printk(KERN_WARNING "md: couldn't set"
5434 " array info. %d\n", err);
5435 goto abort_unlock;
5436 }
5437 }
5438 goto done_unlock;
5439
5440 default:;
5441 }
5442
5443 /*
5444 * Commands querying/configuring an existing array:
5445 */
5446 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5447 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5448 if ((!mddev->raid_disks && !mddev->external)
5449 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5450 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5451 && cmd != GET_BITMAP_FILE) {
5452 err = -ENODEV;
5453 goto abort_unlock;
5454 }
5455
5456 /*
5457 * Commands even a read-only array can execute:
5458 */
5459 switch (cmd)
5460 {
5461 case GET_ARRAY_INFO:
5462 err = get_array_info(mddev, argp);
5463 goto done_unlock;
5464
5465 case GET_BITMAP_FILE:
5466 err = get_bitmap_file(mddev, argp);
5467 goto done_unlock;
5468
5469 case GET_DISK_INFO:
5470 err = get_disk_info(mddev, argp);
5471 goto done_unlock;
5472
5473 case RESTART_ARRAY_RW:
5474 err = restart_array(mddev);
5475 goto done_unlock;
5476
5477 case STOP_ARRAY:
5478 err = do_md_stop(mddev, 0, 1);
5479 goto done_unlock;
5480
5481 case STOP_ARRAY_RO:
5482 err = do_md_stop(mddev, 1, 1);
5483 goto done_unlock;
5484
5485 case BLKROSET:
5486 if (get_user(ro, (int __user *)(arg))) {
5487 err = -EFAULT;
5488 goto done_unlock;
5489 }
5490 err = -EINVAL;
5491
5492 /* if the bdev is going readonly the value of mddev->ro
5493 * does not matter, no writes are coming
5494 */
5495 if (ro)
5496 goto done_unlock;
5497
5498 /* are we are already prepared for writes? */
5499 if (mddev->ro != 1)
5500 goto done_unlock;
5501
5502 /* transitioning to readauto need only happen for
5503 * arrays that call md_write_start
5504 */
5505 if (mddev->pers) {
5506 err = restart_array(mddev);
5507 if (err == 0) {
5508 mddev->ro = 2;
5509 set_disk_ro(mddev->gendisk, 0);
5510 }
5511 }
5512 goto done_unlock;
5513 }
5514
5515 /*
5516 * The remaining ioctls are changing the state of the
5517 * superblock, so we do not allow them on read-only arrays.
5518 * However non-MD ioctls (e.g. get-size) will still come through
5519 * here and hit the 'default' below, so only disallow
5520 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5521 */
5522 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5523 if (mddev->ro == 2) {
5524 mddev->ro = 0;
5525 sysfs_notify_dirent(mddev->sysfs_state);
5526 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5527 md_wakeup_thread(mddev->thread);
5528 } else {
5529 err = -EROFS;
5530 goto abort_unlock;
5531 }
5532 }
5533
5534 switch (cmd)
5535 {
5536 case ADD_NEW_DISK:
5537 {
5538 mdu_disk_info_t info;
5539 if (copy_from_user(&info, argp, sizeof(info)))
5540 err = -EFAULT;
5541 else
5542 err = add_new_disk(mddev, &info);
5543 goto done_unlock;
5544 }
5545
5546 case HOT_REMOVE_DISK:
5547 err = hot_remove_disk(mddev, new_decode_dev(arg));
5548 goto done_unlock;
5549
5550 case HOT_ADD_DISK:
5551 err = hot_add_disk(mddev, new_decode_dev(arg));
5552 goto done_unlock;
5553
5554 case SET_DISK_FAULTY:
5555 err = set_disk_faulty(mddev, new_decode_dev(arg));
5556 goto done_unlock;
5557
5558 case RUN_ARRAY:
5559 err = do_md_run(mddev);
5560 goto done_unlock;
5561
5562 case SET_BITMAP_FILE:
5563 err = set_bitmap_file(mddev, (int)arg);
5564 goto done_unlock;
5565
5566 default:
5567 err = -EINVAL;
5568 goto abort_unlock;
5569 }
5570
5571 done_unlock:
5572 abort_unlock:
5573 if (mddev->hold_active == UNTIL_IOCTL &&
5574 err != -EINVAL)
5575 mddev->hold_active = 0;
5576 mddev_unlock(mddev);
5577
5578 return err;
5579 done:
5580 if (err)
5581 MD_BUG();
5582 abort:
5583 return err;
5584 }
5585
5586 static int md_open(struct block_device *bdev, fmode_t mode)
5587 {
5588 /*
5589 * Succeed if we can lock the mddev, which confirms that
5590 * it isn't being stopped right now.
5591 */
5592 mddev_t *mddev = mddev_find(bdev->bd_dev);
5593 int err;
5594
5595 if (mddev->gendisk != bdev->bd_disk) {
5596 /* we are racing with mddev_put which is discarding this
5597 * bd_disk.
5598 */
5599 mddev_put(mddev);
5600 /* Wait until bdev->bd_disk is definitely gone */
5601 flush_scheduled_work();
5602 /* Then retry the open from the top */
5603 return -ERESTARTSYS;
5604 }
5605 BUG_ON(mddev != bdev->bd_disk->private_data);
5606
5607 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
5608 goto out;
5609
5610 err = 0;
5611 atomic_inc(&mddev->openers);
5612 mutex_unlock(&mddev->open_mutex);
5613
5614 check_disk_change(bdev);
5615 out:
5616 return err;
5617 }
5618
5619 static int md_release(struct gendisk *disk, fmode_t mode)
5620 {
5621 mddev_t *mddev = disk->private_data;
5622
5623 BUG_ON(!mddev);
5624 atomic_dec(&mddev->openers);
5625 mddev_put(mddev);
5626
5627 return 0;
5628 }
5629
5630 static int md_media_changed(struct gendisk *disk)
5631 {
5632 mddev_t *mddev = disk->private_data;
5633
5634 return mddev->changed;
5635 }
5636
5637 static int md_revalidate(struct gendisk *disk)
5638 {
5639 mddev_t *mddev = disk->private_data;
5640
5641 mddev->changed = 0;
5642 return 0;
5643 }
5644 static const struct block_device_operations md_fops =
5645 {
5646 .owner = THIS_MODULE,
5647 .open = md_open,
5648 .release = md_release,
5649 .ioctl = md_ioctl,
5650 .getgeo = md_getgeo,
5651 .media_changed = md_media_changed,
5652 .revalidate_disk= md_revalidate,
5653 };
5654
5655 static int md_thread(void * arg)
5656 {
5657 mdk_thread_t *thread = arg;
5658
5659 /*
5660 * md_thread is a 'system-thread', it's priority should be very
5661 * high. We avoid resource deadlocks individually in each
5662 * raid personality. (RAID5 does preallocation) We also use RR and
5663 * the very same RT priority as kswapd, thus we will never get
5664 * into a priority inversion deadlock.
5665 *
5666 * we definitely have to have equal or higher priority than
5667 * bdflush, otherwise bdflush will deadlock if there are too
5668 * many dirty RAID5 blocks.
5669 */
5670
5671 allow_signal(SIGKILL);
5672 while (!kthread_should_stop()) {
5673
5674 /* We need to wait INTERRUPTIBLE so that
5675 * we don't add to the load-average.
5676 * That means we need to be sure no signals are
5677 * pending
5678 */
5679 if (signal_pending(current))
5680 flush_signals(current);
5681
5682 wait_event_interruptible_timeout
5683 (thread->wqueue,
5684 test_bit(THREAD_WAKEUP, &thread->flags)
5685 || kthread_should_stop(),
5686 thread->timeout);
5687
5688 clear_bit(THREAD_WAKEUP, &thread->flags);
5689
5690 thread->run(thread->mddev);
5691 }
5692
5693 return 0;
5694 }
5695
5696 void md_wakeup_thread(mdk_thread_t *thread)
5697 {
5698 if (thread) {
5699 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5700 set_bit(THREAD_WAKEUP, &thread->flags);
5701 wake_up(&thread->wqueue);
5702 }
5703 }
5704
5705 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5706 const char *name)
5707 {
5708 mdk_thread_t *thread;
5709
5710 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5711 if (!thread)
5712 return NULL;
5713
5714 init_waitqueue_head(&thread->wqueue);
5715
5716 thread->run = run;
5717 thread->mddev = mddev;
5718 thread->timeout = MAX_SCHEDULE_TIMEOUT;
5719 thread->tsk = kthread_run(md_thread, thread,
5720 "%s_%s",
5721 mdname(thread->mddev),
5722 name ?: mddev->pers->name);
5723 if (IS_ERR(thread->tsk)) {
5724 kfree(thread);
5725 return NULL;
5726 }
5727 return thread;
5728 }
5729
5730 void md_unregister_thread(mdk_thread_t *thread)
5731 {
5732 if (!thread)
5733 return;
5734 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5735
5736 kthread_stop(thread->tsk);
5737 kfree(thread);
5738 }
5739
5740 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5741 {
5742 if (!mddev) {
5743 MD_BUG();
5744 return;
5745 }
5746
5747 if (!rdev || test_bit(Faulty, &rdev->flags))
5748 return;
5749
5750 if (mddev->external)
5751 set_bit(Blocked, &rdev->flags);
5752 /*
5753 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5754 mdname(mddev),
5755 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5756 __builtin_return_address(0),__builtin_return_address(1),
5757 __builtin_return_address(2),__builtin_return_address(3));
5758 */
5759 if (!mddev->pers)
5760 return;
5761 if (!mddev->pers->error_handler)
5762 return;
5763 mddev->pers->error_handler(mddev,rdev);
5764 if (mddev->degraded)
5765 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5766 set_bit(StateChanged, &rdev->flags);
5767 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5768 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5769 md_wakeup_thread(mddev->thread);
5770 md_new_event_inintr(mddev);
5771 }
5772
5773 /* seq_file implementation /proc/mdstat */
5774
5775 static void status_unused(struct seq_file *seq)
5776 {
5777 int i = 0;
5778 mdk_rdev_t *rdev;
5779
5780 seq_printf(seq, "unused devices: ");
5781
5782 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
5783 char b[BDEVNAME_SIZE];
5784 i++;
5785 seq_printf(seq, "%s ",
5786 bdevname(rdev->bdev,b));
5787 }
5788 if (!i)
5789 seq_printf(seq, "<none>");
5790
5791 seq_printf(seq, "\n");
5792 }
5793
5794
5795 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5796 {
5797 sector_t max_sectors, resync, res;
5798 unsigned long dt, db;
5799 sector_t rt;
5800 int scale;
5801 unsigned int per_milli;
5802
5803 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
5804
5805 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5806 max_sectors = mddev->resync_max_sectors;
5807 else
5808 max_sectors = mddev->dev_sectors;
5809
5810 /*
5811 * Should not happen.
5812 */
5813 if (!max_sectors) {
5814 MD_BUG();
5815 return;
5816 }
5817 /* Pick 'scale' such that (resync>>scale)*1000 will fit
5818 * in a sector_t, and (max_sectors>>scale) will fit in a
5819 * u32, as those are the requirements for sector_div.
5820 * Thus 'scale' must be at least 10
5821 */
5822 scale = 10;
5823 if (sizeof(sector_t) > sizeof(unsigned long)) {
5824 while ( max_sectors/2 > (1ULL<<(scale+32)))
5825 scale++;
5826 }
5827 res = (resync>>scale)*1000;
5828 sector_div(res, (u32)((max_sectors>>scale)+1));
5829
5830 per_milli = res;
5831 {
5832 int i, x = per_milli/50, y = 20-x;
5833 seq_printf(seq, "[");
5834 for (i = 0; i < x; i++)
5835 seq_printf(seq, "=");
5836 seq_printf(seq, ">");
5837 for (i = 0; i < y; i++)
5838 seq_printf(seq, ".");
5839 seq_printf(seq, "] ");
5840 }
5841 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5842 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5843 "reshape" :
5844 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5845 "check" :
5846 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5847 "resync" : "recovery"))),
5848 per_milli/10, per_milli % 10,
5849 (unsigned long long) resync/2,
5850 (unsigned long long) max_sectors/2);
5851
5852 /*
5853 * dt: time from mark until now
5854 * db: blocks written from mark until now
5855 * rt: remaining time
5856 *
5857 * rt is a sector_t, so could be 32bit or 64bit.
5858 * So we divide before multiply in case it is 32bit and close
5859 * to the limit.
5860 * We scale the divisor (db) by 32 to avoid loosing precision
5861 * near the end of resync when the number of remaining sectors
5862 * is close to 'db'.
5863 * We then divide rt by 32 after multiplying by db to compensate.
5864 * The '+1' avoids division by zero if db is very small.
5865 */
5866 dt = ((jiffies - mddev->resync_mark) / HZ);
5867 if (!dt) dt++;
5868 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5869 - mddev->resync_mark_cnt;
5870
5871 rt = max_sectors - resync; /* number of remaining sectors */
5872 sector_div(rt, db/32+1);
5873 rt *= dt;
5874 rt >>= 5;
5875
5876 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
5877 ((unsigned long)rt % 60)/6);
5878
5879 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5880 }
5881
5882 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5883 {
5884 struct list_head *tmp;
5885 loff_t l = *pos;
5886 mddev_t *mddev;
5887
5888 if (l >= 0x10000)
5889 return NULL;
5890 if (!l--)
5891 /* header */
5892 return (void*)1;
5893
5894 spin_lock(&all_mddevs_lock);
5895 list_for_each(tmp,&all_mddevs)
5896 if (!l--) {
5897 mddev = list_entry(tmp, mddev_t, all_mddevs);
5898 mddev_get(mddev);
5899 spin_unlock(&all_mddevs_lock);
5900 return mddev;
5901 }
5902 spin_unlock(&all_mddevs_lock);
5903 if (!l--)
5904 return (void*)2;/* tail */
5905 return NULL;
5906 }
5907
5908 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5909 {
5910 struct list_head *tmp;
5911 mddev_t *next_mddev, *mddev = v;
5912
5913 ++*pos;
5914 if (v == (void*)2)
5915 return NULL;
5916
5917 spin_lock(&all_mddevs_lock);
5918 if (v == (void*)1)
5919 tmp = all_mddevs.next;
5920 else
5921 tmp = mddev->all_mddevs.next;
5922 if (tmp != &all_mddevs)
5923 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5924 else {
5925 next_mddev = (void*)2;
5926 *pos = 0x10000;
5927 }
5928 spin_unlock(&all_mddevs_lock);
5929
5930 if (v != (void*)1)
5931 mddev_put(mddev);
5932 return next_mddev;
5933
5934 }
5935
5936 static void md_seq_stop(struct seq_file *seq, void *v)
5937 {
5938 mddev_t *mddev = v;
5939
5940 if (mddev && v != (void*)1 && v != (void*)2)
5941 mddev_put(mddev);
5942 }
5943
5944 struct mdstat_info {
5945 int event;
5946 };
5947
5948 static int md_seq_show(struct seq_file *seq, void *v)
5949 {
5950 mddev_t *mddev = v;
5951 sector_t sectors;
5952 mdk_rdev_t *rdev;
5953 struct mdstat_info *mi = seq->private;
5954 struct bitmap *bitmap;
5955
5956 if (v == (void*)1) {
5957 struct mdk_personality *pers;
5958 seq_printf(seq, "Personalities : ");
5959 spin_lock(&pers_lock);
5960 list_for_each_entry(pers, &pers_list, list)
5961 seq_printf(seq, "[%s] ", pers->name);
5962
5963 spin_unlock(&pers_lock);
5964 seq_printf(seq, "\n");
5965 mi->event = atomic_read(&md_event_count);
5966 return 0;
5967 }
5968 if (v == (void*)2) {
5969 status_unused(seq);
5970 return 0;
5971 }
5972
5973 if (mddev_lock(mddev) < 0)
5974 return -EINTR;
5975
5976 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5977 seq_printf(seq, "%s : %sactive", mdname(mddev),
5978 mddev->pers ? "" : "in");
5979 if (mddev->pers) {
5980 if (mddev->ro==1)
5981 seq_printf(seq, " (read-only)");
5982 if (mddev->ro==2)
5983 seq_printf(seq, " (auto-read-only)");
5984 seq_printf(seq, " %s", mddev->pers->name);
5985 }
5986
5987 sectors = 0;
5988 list_for_each_entry(rdev, &mddev->disks, same_set) {
5989 char b[BDEVNAME_SIZE];
5990 seq_printf(seq, " %s[%d]",
5991 bdevname(rdev->bdev,b), rdev->desc_nr);
5992 if (test_bit(WriteMostly, &rdev->flags))
5993 seq_printf(seq, "(W)");
5994 if (test_bit(Faulty, &rdev->flags)) {
5995 seq_printf(seq, "(F)");
5996 continue;
5997 } else if (rdev->raid_disk < 0)
5998 seq_printf(seq, "(S)"); /* spare */
5999 sectors += rdev->sectors;
6000 }
6001
6002 if (!list_empty(&mddev->disks)) {
6003 if (mddev->pers)
6004 seq_printf(seq, "\n %llu blocks",
6005 (unsigned long long)
6006 mddev->array_sectors / 2);
6007 else
6008 seq_printf(seq, "\n %llu blocks",
6009 (unsigned long long)sectors / 2);
6010 }
6011 if (mddev->persistent) {
6012 if (mddev->major_version != 0 ||
6013 mddev->minor_version != 90) {
6014 seq_printf(seq," super %d.%d",
6015 mddev->major_version,
6016 mddev->minor_version);
6017 }
6018 } else if (mddev->external)
6019 seq_printf(seq, " super external:%s",
6020 mddev->metadata_type);
6021 else
6022 seq_printf(seq, " super non-persistent");
6023
6024 if (mddev->pers) {
6025 mddev->pers->status(seq, mddev);
6026 seq_printf(seq, "\n ");
6027 if (mddev->pers->sync_request) {
6028 if (mddev->curr_resync > 2) {
6029 status_resync(seq, mddev);
6030 seq_printf(seq, "\n ");
6031 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6032 seq_printf(seq, "\tresync=DELAYED\n ");
6033 else if (mddev->recovery_cp < MaxSector)
6034 seq_printf(seq, "\tresync=PENDING\n ");
6035 }
6036 } else
6037 seq_printf(seq, "\n ");
6038
6039 if ((bitmap = mddev->bitmap)) {
6040 unsigned long chunk_kb;
6041 unsigned long flags;
6042 spin_lock_irqsave(&bitmap->lock, flags);
6043 chunk_kb = bitmap->chunksize >> 10;
6044 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6045 "%lu%s chunk",
6046 bitmap->pages - bitmap->missing_pages,
6047 bitmap->pages,
6048 (bitmap->pages - bitmap->missing_pages)
6049 << (PAGE_SHIFT - 10),
6050 chunk_kb ? chunk_kb : bitmap->chunksize,
6051 chunk_kb ? "KB" : "B");
6052 if (bitmap->file) {
6053 seq_printf(seq, ", file: ");
6054 seq_path(seq, &bitmap->file->f_path, " \t\n");
6055 }
6056
6057 seq_printf(seq, "\n");
6058 spin_unlock_irqrestore(&bitmap->lock, flags);
6059 }
6060
6061 seq_printf(seq, "\n");
6062 }
6063 mddev_unlock(mddev);
6064
6065 return 0;
6066 }
6067
6068 static const struct seq_operations md_seq_ops = {
6069 .start = md_seq_start,
6070 .next = md_seq_next,
6071 .stop = md_seq_stop,
6072 .show = md_seq_show,
6073 };
6074
6075 static int md_seq_open(struct inode *inode, struct file *file)
6076 {
6077 int error;
6078 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6079 if (mi == NULL)
6080 return -ENOMEM;
6081
6082 error = seq_open(file, &md_seq_ops);
6083 if (error)
6084 kfree(mi);
6085 else {
6086 struct seq_file *p = file->private_data;
6087 p->private = mi;
6088 mi->event = atomic_read(&md_event_count);
6089 }
6090 return error;
6091 }
6092
6093 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6094 {
6095 struct seq_file *m = filp->private_data;
6096 struct mdstat_info *mi = m->private;
6097 int mask;
6098
6099 poll_wait(filp, &md_event_waiters, wait);
6100
6101 /* always allow read */
6102 mask = POLLIN | POLLRDNORM;
6103
6104 if (mi->event != atomic_read(&md_event_count))
6105 mask |= POLLERR | POLLPRI;
6106 return mask;
6107 }
6108
6109 static const struct file_operations md_seq_fops = {
6110 .owner = THIS_MODULE,
6111 .open = md_seq_open,
6112 .read = seq_read,
6113 .llseek = seq_lseek,
6114 .release = seq_release_private,
6115 .poll = mdstat_poll,
6116 };
6117
6118 int register_md_personality(struct mdk_personality *p)
6119 {
6120 spin_lock(&pers_lock);
6121 list_add_tail(&p->list, &pers_list);
6122 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6123 spin_unlock(&pers_lock);
6124 return 0;
6125 }
6126
6127 int unregister_md_personality(struct mdk_personality *p)
6128 {
6129 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6130 spin_lock(&pers_lock);
6131 list_del_init(&p->list);
6132 spin_unlock(&pers_lock);
6133 return 0;
6134 }
6135
6136 static int is_mddev_idle(mddev_t *mddev, int init)
6137 {
6138 mdk_rdev_t * rdev;
6139 int idle;
6140 int curr_events;
6141
6142 idle = 1;
6143 rcu_read_lock();
6144 rdev_for_each_rcu(rdev, mddev) {
6145 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6146 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6147 (int)part_stat_read(&disk->part0, sectors[1]) -
6148 atomic_read(&disk->sync_io);
6149 /* sync IO will cause sync_io to increase before the disk_stats
6150 * as sync_io is counted when a request starts, and
6151 * disk_stats is counted when it completes.
6152 * So resync activity will cause curr_events to be smaller than
6153 * when there was no such activity.
6154 * non-sync IO will cause disk_stat to increase without
6155 * increasing sync_io so curr_events will (eventually)
6156 * be larger than it was before. Once it becomes
6157 * substantially larger, the test below will cause
6158 * the array to appear non-idle, and resync will slow
6159 * down.
6160 * If there is a lot of outstanding resync activity when
6161 * we set last_event to curr_events, then all that activity
6162 * completing might cause the array to appear non-idle
6163 * and resync will be slowed down even though there might
6164 * not have been non-resync activity. This will only
6165 * happen once though. 'last_events' will soon reflect
6166 * the state where there is little or no outstanding
6167 * resync requests, and further resync activity will
6168 * always make curr_events less than last_events.
6169 *
6170 */
6171 if (init || curr_events - rdev->last_events > 64) {
6172 rdev->last_events = curr_events;
6173 idle = 0;
6174 }
6175 }
6176 rcu_read_unlock();
6177 return idle;
6178 }
6179
6180 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6181 {
6182 /* another "blocks" (512byte) blocks have been synced */
6183 atomic_sub(blocks, &mddev->recovery_active);
6184 wake_up(&mddev->recovery_wait);
6185 if (!ok) {
6186 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6187 md_wakeup_thread(mddev->thread);
6188 // stop recovery, signal do_sync ....
6189 }
6190 }
6191
6192
6193 /* md_write_start(mddev, bi)
6194 * If we need to update some array metadata (e.g. 'active' flag
6195 * in superblock) before writing, schedule a superblock update
6196 * and wait for it to complete.
6197 */
6198 void md_write_start(mddev_t *mddev, struct bio *bi)
6199 {
6200 int did_change = 0;
6201 if (bio_data_dir(bi) != WRITE)
6202 return;
6203
6204 BUG_ON(mddev->ro == 1);
6205 if (mddev->ro == 2) {
6206 /* need to switch to read/write */
6207 mddev->ro = 0;
6208 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6209 md_wakeup_thread(mddev->thread);
6210 md_wakeup_thread(mddev->sync_thread);
6211 did_change = 1;
6212 }
6213 atomic_inc(&mddev->writes_pending);
6214 if (mddev->safemode == 1)
6215 mddev->safemode = 0;
6216 if (mddev->in_sync) {
6217 spin_lock_irq(&mddev->write_lock);
6218 if (mddev->in_sync) {
6219 mddev->in_sync = 0;
6220 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6221 md_wakeup_thread(mddev->thread);
6222 did_change = 1;
6223 }
6224 spin_unlock_irq(&mddev->write_lock);
6225 }
6226 if (did_change)
6227 sysfs_notify_dirent(mddev->sysfs_state);
6228 wait_event(mddev->sb_wait,
6229 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
6230 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6231 }
6232
6233 void md_write_end(mddev_t *mddev)
6234 {
6235 if (atomic_dec_and_test(&mddev->writes_pending)) {
6236 if (mddev->safemode == 2)
6237 md_wakeup_thread(mddev->thread);
6238 else if (mddev->safemode_delay)
6239 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6240 }
6241 }
6242
6243 /* md_allow_write(mddev)
6244 * Calling this ensures that the array is marked 'active' so that writes
6245 * may proceed without blocking. It is important to call this before
6246 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6247 * Must be called with mddev_lock held.
6248 *
6249 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6250 * is dropped, so return -EAGAIN after notifying userspace.
6251 */
6252 int md_allow_write(mddev_t *mddev)
6253 {
6254 if (!mddev->pers)
6255 return 0;
6256 if (mddev->ro)
6257 return 0;
6258 if (!mddev->pers->sync_request)
6259 return 0;
6260
6261 spin_lock_irq(&mddev->write_lock);
6262 if (mddev->in_sync) {
6263 mddev->in_sync = 0;
6264 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6265 if (mddev->safemode_delay &&
6266 mddev->safemode == 0)
6267 mddev->safemode = 1;
6268 spin_unlock_irq(&mddev->write_lock);
6269 md_update_sb(mddev, 0);
6270 sysfs_notify_dirent(mddev->sysfs_state);
6271 } else
6272 spin_unlock_irq(&mddev->write_lock);
6273
6274 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
6275 return -EAGAIN;
6276 else
6277 return 0;
6278 }
6279 EXPORT_SYMBOL_GPL(md_allow_write);
6280
6281 #define SYNC_MARKS 10
6282 #define SYNC_MARK_STEP (3*HZ)
6283 void md_do_sync(mddev_t *mddev)
6284 {
6285 mddev_t *mddev2;
6286 unsigned int currspeed = 0,
6287 window;
6288 sector_t max_sectors,j, io_sectors;
6289 unsigned long mark[SYNC_MARKS];
6290 sector_t mark_cnt[SYNC_MARKS];
6291 int last_mark,m;
6292 struct list_head *tmp;
6293 sector_t last_check;
6294 int skipped = 0;
6295 mdk_rdev_t *rdev;
6296 char *desc;
6297
6298 /* just incase thread restarts... */
6299 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6300 return;
6301 if (mddev->ro) /* never try to sync a read-only array */
6302 return;
6303
6304 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6305 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6306 desc = "data-check";
6307 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6308 desc = "requested-resync";
6309 else
6310 desc = "resync";
6311 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6312 desc = "reshape";
6313 else
6314 desc = "recovery";
6315
6316 /* we overload curr_resync somewhat here.
6317 * 0 == not engaged in resync at all
6318 * 2 == checking that there is no conflict with another sync
6319 * 1 == like 2, but have yielded to allow conflicting resync to
6320 * commense
6321 * other == active in resync - this many blocks
6322 *
6323 * Before starting a resync we must have set curr_resync to
6324 * 2, and then checked that every "conflicting" array has curr_resync
6325 * less than ours. When we find one that is the same or higher
6326 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6327 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6328 * This will mean we have to start checking from the beginning again.
6329 *
6330 */
6331
6332 do {
6333 mddev->curr_resync = 2;
6334
6335 try_again:
6336 if (kthread_should_stop()) {
6337 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6338 goto skip;
6339 }
6340 for_each_mddev(mddev2, tmp) {
6341 if (mddev2 == mddev)
6342 continue;
6343 if (!mddev->parallel_resync
6344 && mddev2->curr_resync
6345 && match_mddev_units(mddev, mddev2)) {
6346 DEFINE_WAIT(wq);
6347 if (mddev < mddev2 && mddev->curr_resync == 2) {
6348 /* arbitrarily yield */
6349 mddev->curr_resync = 1;
6350 wake_up(&resync_wait);
6351 }
6352 if (mddev > mddev2 && mddev->curr_resync == 1)
6353 /* no need to wait here, we can wait the next
6354 * time 'round when curr_resync == 2
6355 */
6356 continue;
6357 /* We need to wait 'interruptible' so as not to
6358 * contribute to the load average, and not to
6359 * be caught by 'softlockup'
6360 */
6361 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6362 if (!kthread_should_stop() &&
6363 mddev2->curr_resync >= mddev->curr_resync) {
6364 printk(KERN_INFO "md: delaying %s of %s"
6365 " until %s has finished (they"
6366 " share one or more physical units)\n",
6367 desc, mdname(mddev), mdname(mddev2));
6368 mddev_put(mddev2);
6369 if (signal_pending(current))
6370 flush_signals(current);
6371 schedule();
6372 finish_wait(&resync_wait, &wq);
6373 goto try_again;
6374 }
6375 finish_wait(&resync_wait, &wq);
6376 }
6377 }
6378 } while (mddev->curr_resync < 2);
6379
6380 j = 0;
6381 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6382 /* resync follows the size requested by the personality,
6383 * which defaults to physical size, but can be virtual size
6384 */
6385 max_sectors = mddev->resync_max_sectors;
6386 mddev->resync_mismatches = 0;
6387 /* we don't use the checkpoint if there's a bitmap */
6388 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6389 j = mddev->resync_min;
6390 else if (!mddev->bitmap)
6391 j = mddev->recovery_cp;
6392
6393 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6394 max_sectors = mddev->dev_sectors;
6395 else {
6396 /* recovery follows the physical size of devices */
6397 max_sectors = mddev->dev_sectors;
6398 j = MaxSector;
6399 list_for_each_entry(rdev, &mddev->disks, same_set)
6400 if (rdev->raid_disk >= 0 &&
6401 !test_bit(Faulty, &rdev->flags) &&
6402 !test_bit(In_sync, &rdev->flags) &&
6403 rdev->recovery_offset < j)
6404 j = rdev->recovery_offset;
6405 }
6406
6407 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6408 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6409 " %d KB/sec/disk.\n", speed_min(mddev));
6410 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6411 "(but not more than %d KB/sec) for %s.\n",
6412 speed_max(mddev), desc);
6413
6414 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6415
6416 io_sectors = 0;
6417 for (m = 0; m < SYNC_MARKS; m++) {
6418 mark[m] = jiffies;
6419 mark_cnt[m] = io_sectors;
6420 }
6421 last_mark = 0;
6422 mddev->resync_mark = mark[last_mark];
6423 mddev->resync_mark_cnt = mark_cnt[last_mark];
6424
6425 /*
6426 * Tune reconstruction:
6427 */
6428 window = 32*(PAGE_SIZE/512);
6429 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6430 window/2,(unsigned long long) max_sectors/2);
6431
6432 atomic_set(&mddev->recovery_active, 0);
6433 last_check = 0;
6434
6435 if (j>2) {
6436 printk(KERN_INFO
6437 "md: resuming %s of %s from checkpoint.\n",
6438 desc, mdname(mddev));
6439 mddev->curr_resync = j;
6440 }
6441
6442 while (j < max_sectors) {
6443 sector_t sectors;
6444
6445 skipped = 0;
6446
6447 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6448 ((mddev->curr_resync > mddev->curr_resync_completed &&
6449 (mddev->curr_resync - mddev->curr_resync_completed)
6450 > (max_sectors >> 4)) ||
6451 (j - mddev->curr_resync_completed)*2
6452 >= mddev->resync_max - mddev->curr_resync_completed
6453 )) {
6454 /* time to update curr_resync_completed */
6455 blk_unplug(mddev->queue);
6456 wait_event(mddev->recovery_wait,
6457 atomic_read(&mddev->recovery_active) == 0);
6458 mddev->curr_resync_completed =
6459 mddev->curr_resync;
6460 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6461 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6462 }
6463
6464 while (j >= mddev->resync_max && !kthread_should_stop()) {
6465 /* As this condition is controlled by user-space,
6466 * we can block indefinitely, so use '_interruptible'
6467 * to avoid triggering warnings.
6468 */
6469 flush_signals(current); /* just in case */
6470 wait_event_interruptible(mddev->recovery_wait,
6471 mddev->resync_max > j
6472 || kthread_should_stop());
6473 }
6474
6475 if (kthread_should_stop())
6476 goto interrupted;
6477
6478 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6479 currspeed < speed_min(mddev));
6480 if (sectors == 0) {
6481 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6482 goto out;
6483 }
6484
6485 if (!skipped) { /* actual IO requested */
6486 io_sectors += sectors;
6487 atomic_add(sectors, &mddev->recovery_active);
6488 }
6489
6490 j += sectors;
6491 if (j>1) mddev->curr_resync = j;
6492 mddev->curr_mark_cnt = io_sectors;
6493 if (last_check == 0)
6494 /* this is the earliers that rebuilt will be
6495 * visible in /proc/mdstat
6496 */
6497 md_new_event(mddev);
6498
6499 if (last_check + window > io_sectors || j == max_sectors)
6500 continue;
6501
6502 last_check = io_sectors;
6503
6504 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6505 break;
6506
6507 repeat:
6508 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6509 /* step marks */
6510 int next = (last_mark+1) % SYNC_MARKS;
6511
6512 mddev->resync_mark = mark[next];
6513 mddev->resync_mark_cnt = mark_cnt[next];
6514 mark[next] = jiffies;
6515 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6516 last_mark = next;
6517 }
6518
6519
6520 if (kthread_should_stop())
6521 goto interrupted;
6522
6523
6524 /*
6525 * this loop exits only if either when we are slower than
6526 * the 'hard' speed limit, or the system was IO-idle for
6527 * a jiffy.
6528 * the system might be non-idle CPU-wise, but we only care
6529 * about not overloading the IO subsystem. (things like an
6530 * e2fsck being done on the RAID array should execute fast)
6531 */
6532 blk_unplug(mddev->queue);
6533 cond_resched();
6534
6535 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6536 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6537
6538 if (currspeed > speed_min(mddev)) {
6539 if ((currspeed > speed_max(mddev)) ||
6540 !is_mddev_idle(mddev, 0)) {
6541 msleep(500);
6542 goto repeat;
6543 }
6544 }
6545 }
6546 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6547 /*
6548 * this also signals 'finished resyncing' to md_stop
6549 */
6550 out:
6551 blk_unplug(mddev->queue);
6552
6553 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6554
6555 /* tell personality that we are finished */
6556 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6557
6558 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6559 mddev->curr_resync > 2) {
6560 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6561 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6562 if (mddev->curr_resync >= mddev->recovery_cp) {
6563 printk(KERN_INFO
6564 "md: checkpointing %s of %s.\n",
6565 desc, mdname(mddev));
6566 mddev->recovery_cp = mddev->curr_resync;
6567 }
6568 } else
6569 mddev->recovery_cp = MaxSector;
6570 } else {
6571 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6572 mddev->curr_resync = MaxSector;
6573 list_for_each_entry(rdev, &mddev->disks, same_set)
6574 if (rdev->raid_disk >= 0 &&
6575 !test_bit(Faulty, &rdev->flags) &&
6576 !test_bit(In_sync, &rdev->flags) &&
6577 rdev->recovery_offset < mddev->curr_resync)
6578 rdev->recovery_offset = mddev->curr_resync;
6579 }
6580 }
6581 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6582
6583 skip:
6584 mddev->curr_resync = 0;
6585 mddev->curr_resync_completed = 0;
6586 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6587 /* We completed so max setting can be forgotten. */
6588 mddev->resync_max = MaxSector;
6589 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6590 wake_up(&resync_wait);
6591 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6592 md_wakeup_thread(mddev->thread);
6593 return;
6594
6595 interrupted:
6596 /*
6597 * got a signal, exit.
6598 */
6599 printk(KERN_INFO
6600 "md: md_do_sync() got signal ... exiting\n");
6601 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6602 goto out;
6603
6604 }
6605 EXPORT_SYMBOL_GPL(md_do_sync);
6606
6607
6608 static int remove_and_add_spares(mddev_t *mddev)
6609 {
6610 mdk_rdev_t *rdev;
6611 int spares = 0;
6612
6613 mddev->curr_resync_completed = 0;
6614
6615 list_for_each_entry(rdev, &mddev->disks, same_set)
6616 if (rdev->raid_disk >= 0 &&
6617 !test_bit(Blocked, &rdev->flags) &&
6618 (test_bit(Faulty, &rdev->flags) ||
6619 ! test_bit(In_sync, &rdev->flags)) &&
6620 atomic_read(&rdev->nr_pending)==0) {
6621 if (mddev->pers->hot_remove_disk(
6622 mddev, rdev->raid_disk)==0) {
6623 char nm[20];
6624 sprintf(nm,"rd%d", rdev->raid_disk);
6625 sysfs_remove_link(&mddev->kobj, nm);
6626 rdev->raid_disk = -1;
6627 }
6628 }
6629
6630 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
6631 list_for_each_entry(rdev, &mddev->disks, same_set) {
6632 if (rdev->raid_disk >= 0 &&
6633 !test_bit(In_sync, &rdev->flags) &&
6634 !test_bit(Faulty, &rdev->flags) &&
6635 !test_bit(Blocked, &rdev->flags))
6636 spares++;
6637 if (rdev->raid_disk < 0
6638 && !test_bit(Faulty, &rdev->flags)) {
6639 rdev->recovery_offset = 0;
6640 if (mddev->pers->
6641 hot_add_disk(mddev, rdev) == 0) {
6642 char nm[20];
6643 sprintf(nm, "rd%d", rdev->raid_disk);
6644 if (sysfs_create_link(&mddev->kobj,
6645 &rdev->kobj, nm))
6646 printk(KERN_WARNING
6647 "md: cannot register "
6648 "%s for %s\n",
6649 nm, mdname(mddev));
6650 spares++;
6651 md_new_event(mddev);
6652 } else
6653 break;
6654 }
6655 }
6656 }
6657 return spares;
6658 }
6659 /*
6660 * This routine is regularly called by all per-raid-array threads to
6661 * deal with generic issues like resync and super-block update.
6662 * Raid personalities that don't have a thread (linear/raid0) do not
6663 * need this as they never do any recovery or update the superblock.
6664 *
6665 * It does not do any resync itself, but rather "forks" off other threads
6666 * to do that as needed.
6667 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6668 * "->recovery" and create a thread at ->sync_thread.
6669 * When the thread finishes it sets MD_RECOVERY_DONE
6670 * and wakeups up this thread which will reap the thread and finish up.
6671 * This thread also removes any faulty devices (with nr_pending == 0).
6672 *
6673 * The overall approach is:
6674 * 1/ if the superblock needs updating, update it.
6675 * 2/ If a recovery thread is running, don't do anything else.
6676 * 3/ If recovery has finished, clean up, possibly marking spares active.
6677 * 4/ If there are any faulty devices, remove them.
6678 * 5/ If array is degraded, try to add spares devices
6679 * 6/ If array has spares or is not in-sync, start a resync thread.
6680 */
6681 void md_check_recovery(mddev_t *mddev)
6682 {
6683 mdk_rdev_t *rdev;
6684
6685
6686 if (mddev->bitmap)
6687 bitmap_daemon_work(mddev);
6688
6689 if (mddev->ro)
6690 return;
6691
6692 if (signal_pending(current)) {
6693 if (mddev->pers->sync_request && !mddev->external) {
6694 printk(KERN_INFO "md: %s in immediate safe mode\n",
6695 mdname(mddev));
6696 mddev->safemode = 2;
6697 }
6698 flush_signals(current);
6699 }
6700
6701 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
6702 return;
6703 if ( ! (
6704 (mddev->flags && !mddev->external) ||
6705 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
6706 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
6707 (mddev->external == 0 && mddev->safemode == 1) ||
6708 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6709 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6710 ))
6711 return;
6712
6713 if (mddev_trylock(mddev)) {
6714 int spares = 0;
6715
6716 if (mddev->ro) {
6717 /* Only thing we do on a ro array is remove
6718 * failed devices.
6719 */
6720 remove_and_add_spares(mddev);
6721 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6722 goto unlock;
6723 }
6724
6725 if (!mddev->external) {
6726 int did_change = 0;
6727 spin_lock_irq(&mddev->write_lock);
6728 if (mddev->safemode &&
6729 !atomic_read(&mddev->writes_pending) &&
6730 !mddev->in_sync &&
6731 mddev->recovery_cp == MaxSector) {
6732 mddev->in_sync = 1;
6733 did_change = 1;
6734 if (mddev->persistent)
6735 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6736 }
6737 if (mddev->safemode == 1)
6738 mddev->safemode = 0;
6739 spin_unlock_irq(&mddev->write_lock);
6740 if (did_change)
6741 sysfs_notify_dirent(mddev->sysfs_state);
6742 }
6743
6744 if (mddev->flags)
6745 md_update_sb(mddev, 0);
6746
6747 list_for_each_entry(rdev, &mddev->disks, same_set)
6748 if (test_and_clear_bit(StateChanged, &rdev->flags))
6749 sysfs_notify_dirent(rdev->sysfs_state);
6750
6751
6752 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6753 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6754 /* resync/recovery still happening */
6755 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6756 goto unlock;
6757 }
6758 if (mddev->sync_thread) {
6759 /* resync has finished, collect result */
6760 md_unregister_thread(mddev->sync_thread);
6761 mddev->sync_thread = NULL;
6762 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
6763 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
6764 /* success...*/
6765 /* activate any spares */
6766 if (mddev->pers->spare_active(mddev))
6767 sysfs_notify(&mddev->kobj, NULL,
6768 "degraded");
6769 }
6770 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6771 mddev->pers->finish_reshape)
6772 mddev->pers->finish_reshape(mddev);
6773 md_update_sb(mddev, 1);
6774
6775 /* if array is no-longer degraded, then any saved_raid_disk
6776 * information must be scrapped
6777 */
6778 if (!mddev->degraded)
6779 list_for_each_entry(rdev, &mddev->disks, same_set)
6780 rdev->saved_raid_disk = -1;
6781
6782 mddev->recovery = 0;
6783 /* flag recovery needed just to double check */
6784 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6785 sysfs_notify_dirent(mddev->sysfs_action);
6786 md_new_event(mddev);
6787 goto unlock;
6788 }
6789 /* Set RUNNING before clearing NEEDED to avoid
6790 * any transients in the value of "sync_action".
6791 */
6792 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6793 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6794 /* Clear some bits that don't mean anything, but
6795 * might be left set
6796 */
6797 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
6798 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
6799
6800 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6801 goto unlock;
6802 /* no recovery is running.
6803 * remove any failed drives, then
6804 * add spares if possible.
6805 * Spare are also removed and re-added, to allow
6806 * the personality to fail the re-add.
6807 */
6808
6809 if (mddev->reshape_position != MaxSector) {
6810 if (mddev->pers->check_reshape == NULL ||
6811 mddev->pers->check_reshape(mddev) != 0)
6812 /* Cannot proceed */
6813 goto unlock;
6814 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6815 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6816 } else if ((spares = remove_and_add_spares(mddev))) {
6817 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6818 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6819 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
6820 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6821 } else if (mddev->recovery_cp < MaxSector) {
6822 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6823 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6824 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6825 /* nothing to be done ... */
6826 goto unlock;
6827
6828 if (mddev->pers->sync_request) {
6829 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6830 /* We are adding a device or devices to an array
6831 * which has the bitmap stored on all devices.
6832 * So make sure all bitmap pages get written
6833 */
6834 bitmap_write_all(mddev->bitmap);
6835 }
6836 mddev->sync_thread = md_register_thread(md_do_sync,
6837 mddev,
6838 "resync");
6839 if (!mddev->sync_thread) {
6840 printk(KERN_ERR "%s: could not start resync"
6841 " thread...\n",
6842 mdname(mddev));
6843 /* leave the spares where they are, it shouldn't hurt */
6844 mddev->recovery = 0;
6845 } else
6846 md_wakeup_thread(mddev->sync_thread);
6847 sysfs_notify_dirent(mddev->sysfs_action);
6848 md_new_event(mddev);
6849 }
6850 unlock:
6851 if (!mddev->sync_thread) {
6852 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6853 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
6854 &mddev->recovery))
6855 if (mddev->sysfs_action)
6856 sysfs_notify_dirent(mddev->sysfs_action);
6857 }
6858 mddev_unlock(mddev);
6859 }
6860 }
6861
6862 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
6863 {
6864 sysfs_notify_dirent(rdev->sysfs_state);
6865 wait_event_timeout(rdev->blocked_wait,
6866 !test_bit(Blocked, &rdev->flags),
6867 msecs_to_jiffies(5000));
6868 rdev_dec_pending(rdev, mddev);
6869 }
6870 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
6871
6872 static int md_notify_reboot(struct notifier_block *this,
6873 unsigned long code, void *x)
6874 {
6875 struct list_head *tmp;
6876 mddev_t *mddev;
6877
6878 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
6879
6880 printk(KERN_INFO "md: stopping all md devices.\n");
6881
6882 for_each_mddev(mddev, tmp)
6883 if (mddev_trylock(mddev)) {
6884 /* Force a switch to readonly even array
6885 * appears to still be in use. Hence
6886 * the '100'.
6887 */
6888 do_md_stop(mddev, 1, 100);
6889 mddev_unlock(mddev);
6890 }
6891 /*
6892 * certain more exotic SCSI devices are known to be
6893 * volatile wrt too early system reboots. While the
6894 * right place to handle this issue is the given
6895 * driver, we do want to have a safe RAID driver ...
6896 */
6897 mdelay(1000*1);
6898 }
6899 return NOTIFY_DONE;
6900 }
6901
6902 static struct notifier_block md_notifier = {
6903 .notifier_call = md_notify_reboot,
6904 .next = NULL,
6905 .priority = INT_MAX, /* before any real devices */
6906 };
6907
6908 static void md_geninit(void)
6909 {
6910 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
6911
6912 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
6913 }
6914
6915 static int __init md_init(void)
6916 {
6917 if (register_blkdev(MD_MAJOR, "md"))
6918 return -1;
6919 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
6920 unregister_blkdev(MD_MAJOR, "md");
6921 return -1;
6922 }
6923 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
6924 md_probe, NULL, NULL);
6925 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
6926 md_probe, NULL, NULL);
6927
6928 register_reboot_notifier(&md_notifier);
6929 raid_table_header = register_sysctl_table(raid_root_table);
6930
6931 md_geninit();
6932 return 0;
6933 }
6934
6935
6936 #ifndef MODULE
6937
6938 /*
6939 * Searches all registered partitions for autorun RAID arrays
6940 * at boot time.
6941 */
6942
6943 static LIST_HEAD(all_detected_devices);
6944 struct detected_devices_node {
6945 struct list_head list;
6946 dev_t dev;
6947 };
6948
6949 void md_autodetect_dev(dev_t dev)
6950 {
6951 struct detected_devices_node *node_detected_dev;
6952
6953 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6954 if (node_detected_dev) {
6955 node_detected_dev->dev = dev;
6956 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6957 } else {
6958 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6959 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6960 }
6961 }
6962
6963
6964 static void autostart_arrays(int part)
6965 {
6966 mdk_rdev_t *rdev;
6967 struct detected_devices_node *node_detected_dev;
6968 dev_t dev;
6969 int i_scanned, i_passed;
6970
6971 i_scanned = 0;
6972 i_passed = 0;
6973
6974 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6975
6976 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6977 i_scanned++;
6978 node_detected_dev = list_entry(all_detected_devices.next,
6979 struct detected_devices_node, list);
6980 list_del(&node_detected_dev->list);
6981 dev = node_detected_dev->dev;
6982 kfree(node_detected_dev);
6983 rdev = md_import_device(dev,0, 90);
6984 if (IS_ERR(rdev))
6985 continue;
6986
6987 if (test_bit(Faulty, &rdev->flags)) {
6988 MD_BUG();
6989 continue;
6990 }
6991 set_bit(AutoDetected, &rdev->flags);
6992 list_add(&rdev->same_set, &pending_raid_disks);
6993 i_passed++;
6994 }
6995
6996 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6997 i_scanned, i_passed);
6998
6999 autorun_devices(part);
7000 }
7001
7002 #endif /* !MODULE */
7003
7004 static __exit void md_exit(void)
7005 {
7006 mddev_t *mddev;
7007 struct list_head *tmp;
7008
7009 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
7010 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
7011
7012 unregister_blkdev(MD_MAJOR,"md");
7013 unregister_blkdev(mdp_major, "mdp");
7014 unregister_reboot_notifier(&md_notifier);
7015 unregister_sysctl_table(raid_table_header);
7016 remove_proc_entry("mdstat", NULL);
7017 for_each_mddev(mddev, tmp) {
7018 export_array(mddev);
7019 mddev->hold_active = 0;
7020 }
7021 }
7022
7023 subsys_initcall(md_init);
7024 module_exit(md_exit)
7025
7026 static int get_ro(char *buffer, struct kernel_param *kp)
7027 {
7028 return sprintf(buffer, "%d", start_readonly);
7029 }
7030 static int set_ro(const char *val, struct kernel_param *kp)
7031 {
7032 char *e;
7033 int num = simple_strtoul(val, &e, 10);
7034 if (*val && (*e == '\0' || *e == '\n')) {
7035 start_readonly = num;
7036 return 0;
7037 }
7038 return -EINVAL;
7039 }
7040
7041 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
7042 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
7043
7044 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
7045
7046 EXPORT_SYMBOL(register_md_personality);
7047 EXPORT_SYMBOL(unregister_md_personality);
7048 EXPORT_SYMBOL(md_error);
7049 EXPORT_SYMBOL(md_done_sync);
7050 EXPORT_SYMBOL(md_write_start);
7051 EXPORT_SYMBOL(md_write_end);
7052 EXPORT_SYMBOL(md_register_thread);
7053 EXPORT_SYMBOL(md_unregister_thread);
7054 EXPORT_SYMBOL(md_wakeup_thread);
7055 EXPORT_SYMBOL(md_check_recovery);
7056 MODULE_LICENSE("GPL");
7057 MODULE_ALIAS("md");
7058 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Linux MOXA ART SoC ARM port