?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
md: when a level change reduces the number of devices, remove the excess.
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / md / md.c
blob6021dea894da5b2e87506986e8c11a6ec57105a5
1 /*
2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
4
5 completely rewritten, based on the MD driver code from Marc Zyngier
6
7 Changes:
8
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
19
20 Neil Brown <neilb@cse.unsw.edu.au>.
21
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
28 any later version.
29
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33 */
34
35 #include <linux/kthread.h>
36 #include <linux/blkdev.h>
37 #include <linux/sysctl.h>
38 #include <linux/seq_file.h>
39 #include <linux/buffer_head.h> /* for invalidate_bdev */
40 #include <linux/poll.h>
41 #include <linux/ctype.h>
42 #include <linux/hdreg.h>
43 #include <linux/proc_fs.h>
44 #include <linux/random.h>
45 #include <linux/reboot.h>
46 #include <linux/file.h>
47 #include <linux/delay.h>
48 #include <linux/raid/md_p.h>
49 #include <linux/raid/md_u.h>
50 #include "md.h"
51 #include "bitmap.h"
52
53 #define DEBUG 0
54 #define dprintk(x...) ((void)(DEBUG && printk(x)))
55
56
57 #ifndef MODULE
58 static void autostart_arrays(int part);
59 #endif
60
61 static LIST_HEAD(pers_list);
62 static DEFINE_SPINLOCK(pers_lock);
63
64 static void md_print_devices(void);
65
66 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
67
68 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
69
70 /*
71 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
72 * is 1000 KB/sec, so the extra system load does not show up that much.
73 * Increase it if you want to have more _guaranteed_ speed. Note that
74 * the RAID driver will use the maximum available bandwidth if the IO
75 * subsystem is idle. There is also an 'absolute maximum' reconstruction
76 * speed limit - in case reconstruction slows down your system despite
77 * idle IO detection.
78 *
79 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
80 * or /sys/block/mdX/md/sync_speed_{min,max}
81 */
82
83 static int sysctl_speed_limit_min = 1000;
84 static int sysctl_speed_limit_max = 200000;
85 static inline int speed_min(mddev_t *mddev)
86 {
87 return mddev->sync_speed_min ?
88 mddev->sync_speed_min : sysctl_speed_limit_min;
89 }
90
91 static inline int speed_max(mddev_t *mddev)
92 {
93 return mddev->sync_speed_max ?
94 mddev->sync_speed_max : sysctl_speed_limit_max;
95 }
96
97 static struct ctl_table_header *raid_table_header;
98
99 static ctl_table raid_table[] = {
100 {
101 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
102 .procname = "speed_limit_min",
103 .data = &sysctl_speed_limit_min,
104 .maxlen = sizeof(int),
105 .mode = S_IRUGO|S_IWUSR,
106 .proc_handler = &proc_dointvec,
107 },
108 {
109 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
110 .procname = "speed_limit_max",
111 .data = &sysctl_speed_limit_max,
112 .maxlen = sizeof(int),
113 .mode = S_IRUGO|S_IWUSR,
114 .proc_handler = &proc_dointvec,
115 },
116 { .ctl_name = 0 }
117 };
118
119 static ctl_table raid_dir_table[] = {
120 {
121 .ctl_name = DEV_RAID,
122 .procname = "raid",
123 .maxlen = 0,
124 .mode = S_IRUGO|S_IXUGO,
125 .child = raid_table,
126 },
127 { .ctl_name = 0 }
128 };
129
130 static ctl_table raid_root_table[] = {
131 {
132 .ctl_name = CTL_DEV,
133 .procname = "dev",
134 .maxlen = 0,
135 .mode = 0555,
136 .child = raid_dir_table,
137 },
138 { .ctl_name = 0 }
139 };
140
141 static struct block_device_operations md_fops;
142
143 static int start_readonly;
144
145 /*
146 * We have a system wide 'event count' that is incremented
147 * on any 'interesting' event, and readers of /proc/mdstat
148 * can use 'poll' or 'select' to find out when the event
149 * count increases.
150 *
151 * Events are:
152 * start array, stop array, error, add device, remove device,
153 * start build, activate spare
154 */
155 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
156 static atomic_t md_event_count;
157 void md_new_event(mddev_t *mddev)
158 {
159 atomic_inc(&md_event_count);
160 wake_up(&md_event_waiters);
161 }
162 EXPORT_SYMBOL_GPL(md_new_event);
163
164 /* Alternate version that can be called from interrupts
165 * when calling sysfs_notify isn't needed.
166 */
167 static void md_new_event_inintr(mddev_t *mddev)
168 {
169 atomic_inc(&md_event_count);
170 wake_up(&md_event_waiters);
171 }
172
173 /*
174 * Enables to iterate over all existing md arrays
175 * all_mddevs_lock protects this list.
176 */
177 static LIST_HEAD(all_mddevs);
178 static DEFINE_SPINLOCK(all_mddevs_lock);
179
180
181 /*
182 * iterates through all used mddevs in the system.
183 * We take care to grab the all_mddevs_lock whenever navigating
184 * the list, and to always hold a refcount when unlocked.
185 * Any code which breaks out of this loop while own
186 * a reference to the current mddev and must mddev_put it.
187 */
188 #define for_each_mddev(mddev,tmp) \
189 \
190 for (({ spin_lock(&all_mddevs_lock); \
191 tmp = all_mddevs.next; \
192 mddev = NULL;}); \
193 ({ if (tmp != &all_mddevs) \
194 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
195 spin_unlock(&all_mddevs_lock); \
196 if (mddev) mddev_put(mddev); \
197 mddev = list_entry(tmp, mddev_t, all_mddevs); \
198 tmp != &all_mddevs;}); \
199 ({ spin_lock(&all_mddevs_lock); \
200 tmp = tmp->next;}) \
201 )
202
203
204 /* Rather than calling directly into the personality make_request function,
205 * IO requests come here first so that we can check if the device is
206 * being suspended pending a reconfiguration.
207 * We hold a refcount over the call to ->make_request. By the time that
208 * call has finished, the bio has been linked into some internal structure
209 * and so is visible to ->quiesce(), so we don't need the refcount any more.
210 */
211 static int md_make_request(struct request_queue *q, struct bio *bio)
212 {
213 mddev_t *mddev = q->queuedata;
214 int rv;
215 if (mddev == NULL || mddev->pers == NULL) {
216 bio_io_error(bio);
217 return 0;
218 }
219 rcu_read_lock();
220 if (mddev->suspended) {
221 DEFINE_WAIT(__wait);
222 for (;;) {
223 prepare_to_wait(&mddev->sb_wait, &__wait,
224 TASK_UNINTERRUPTIBLE);
225 if (!mddev->suspended)
226 break;
227 rcu_read_unlock();
228 schedule();
229 rcu_read_lock();
230 }
231 finish_wait(&mddev->sb_wait, &__wait);
232 }
233 atomic_inc(&mddev->active_io);
234 rcu_read_unlock();
235 rv = mddev->pers->make_request(q, bio);
236 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
237 wake_up(&mddev->sb_wait);
238
239 return rv;
240 }
241
242 static void mddev_suspend(mddev_t *mddev)
243 {
244 BUG_ON(mddev->suspended);
245 mddev->suspended = 1;
246 synchronize_rcu();
247 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
248 mddev->pers->quiesce(mddev, 1);
249 md_unregister_thread(mddev->thread);
250 mddev->thread = NULL;
251 /* we now know that no code is executing in the personality module,
252 * except possibly the tail end of a ->bi_end_io function, but that
253 * is certain to complete before the module has a chance to get
254 * unloaded
255 */
256 }
257
258 static void mddev_resume(mddev_t *mddev)
259 {
260 mddev->suspended = 0;
261 wake_up(&mddev->sb_wait);
262 mddev->pers->quiesce(mddev, 0);
263 }
264
265
266 static inline mddev_t *mddev_get(mddev_t *mddev)
267 {
268 atomic_inc(&mddev->active);
269 return mddev;
270 }
271
272 static void mddev_delayed_delete(struct work_struct *ws);
273
274 static void mddev_put(mddev_t *mddev)
275 {
276 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
277 return;
278 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
279 !mddev->hold_active) {
280 list_del(&mddev->all_mddevs);
281 if (mddev->gendisk) {
282 /* we did a probe so need to clean up.
283 * Call schedule_work inside the spinlock
284 * so that flush_scheduled_work() after
285 * mddev_find will succeed in waiting for the
286 * work to be done.
287 */
288 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
289 schedule_work(&mddev->del_work);
290 } else
291 kfree(mddev);
292 }
293 spin_unlock(&all_mddevs_lock);
294 }
295
296 static mddev_t * mddev_find(dev_t unit)
297 {
298 mddev_t *mddev, *new = NULL;
299
300 retry:
301 spin_lock(&all_mddevs_lock);
302
303 if (unit) {
304 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
305 if (mddev->unit == unit) {
306 mddev_get(mddev);
307 spin_unlock(&all_mddevs_lock);
308 kfree(new);
309 return mddev;
310 }
311
312 if (new) {
313 list_add(&new->all_mddevs, &all_mddevs);
314 spin_unlock(&all_mddevs_lock);
315 new->hold_active = UNTIL_IOCTL;
316 return new;
317 }
318 } else if (new) {
319 /* find an unused unit number */
320 static int next_minor = 512;
321 int start = next_minor;
322 int is_free = 0;
323 int dev = 0;
324 while (!is_free) {
325 dev = MKDEV(MD_MAJOR, next_minor);
326 next_minor++;
327 if (next_minor > MINORMASK)
328 next_minor = 0;
329 if (next_minor == start) {
330 /* Oh dear, all in use. */
331 spin_unlock(&all_mddevs_lock);
332 kfree(new);
333 return NULL;
334 }
335
336 is_free = 1;
337 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
338 if (mddev->unit == dev) {
339 is_free = 0;
340 break;
341 }
342 }
343 new->unit = dev;
344 new->md_minor = MINOR(dev);
345 new->hold_active = UNTIL_STOP;
346 list_add(&new->all_mddevs, &all_mddevs);
347 spin_unlock(&all_mddevs_lock);
348 return new;
349 }
350 spin_unlock(&all_mddevs_lock);
351
352 new = kzalloc(sizeof(*new), GFP_KERNEL);
353 if (!new)
354 return NULL;
355
356 new->unit = unit;
357 if (MAJOR(unit) == MD_MAJOR)
358 new->md_minor = MINOR(unit);
359 else
360 new->md_minor = MINOR(unit) >> MdpMinorShift;
361
362 mutex_init(&new->reconfig_mutex);
363 INIT_LIST_HEAD(&new->disks);
364 INIT_LIST_HEAD(&new->all_mddevs);
365 init_timer(&new->safemode_timer);
366 atomic_set(&new->active, 1);
367 atomic_set(&new->openers, 0);
368 atomic_set(&new->active_io, 0);
369 spin_lock_init(&new->write_lock);
370 init_waitqueue_head(&new->sb_wait);
371 init_waitqueue_head(&new->recovery_wait);
372 new->reshape_position = MaxSector;
373 new->resync_min = 0;
374 new->resync_max = MaxSector;
375 new->level = LEVEL_NONE;
376
377 goto retry;
378 }
379
380 static inline int mddev_lock(mddev_t * mddev)
381 {
382 return mutex_lock_interruptible(&mddev->reconfig_mutex);
383 }
384
385 static inline int mddev_is_locked(mddev_t *mddev)
386 {
387 return mutex_is_locked(&mddev->reconfig_mutex);
388 }
389
390 static inline int mddev_trylock(mddev_t * mddev)
391 {
392 return mutex_trylock(&mddev->reconfig_mutex);
393 }
394
395 static inline void mddev_unlock(mddev_t * mddev)
396 {
397 mutex_unlock(&mddev->reconfig_mutex);
398
399 md_wakeup_thread(mddev->thread);
400 }
401
402 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
403 {
404 mdk_rdev_t *rdev;
405
406 list_for_each_entry(rdev, &mddev->disks, same_set)
407 if (rdev->desc_nr == nr)
408 return rdev;
409
410 return NULL;
411 }
412
413 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
414 {
415 mdk_rdev_t *rdev;
416
417 list_for_each_entry(rdev, &mddev->disks, same_set)
418 if (rdev->bdev->bd_dev == dev)
419 return rdev;
420
421 return NULL;
422 }
423
424 static struct mdk_personality *find_pers(int level, char *clevel)
425 {
426 struct mdk_personality *pers;
427 list_for_each_entry(pers, &pers_list, list) {
428 if (level != LEVEL_NONE && pers->level == level)
429 return pers;
430 if (strcmp(pers->name, clevel)==0)
431 return pers;
432 }
433 return NULL;
434 }
435
436 /* return the offset of the super block in 512byte sectors */
437 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
438 {
439 sector_t num_sectors = bdev->bd_inode->i_size / 512;
440 return MD_NEW_SIZE_SECTORS(num_sectors);
441 }
442
443 static sector_t calc_num_sectors(mdk_rdev_t *rdev, unsigned chunk_size)
444 {
445 sector_t num_sectors = rdev->sb_start;
446
447 if (chunk_size)
448 num_sectors &= ~((sector_t)chunk_size/512 - 1);
449 return num_sectors;
450 }
451
452 static int alloc_disk_sb(mdk_rdev_t * rdev)
453 {
454 if (rdev->sb_page)
455 MD_BUG();
456
457 rdev->sb_page = alloc_page(GFP_KERNEL);
458 if (!rdev->sb_page) {
459 printk(KERN_ALERT "md: out of memory.\n");
460 return -ENOMEM;
461 }
462
463 return 0;
464 }
465
466 static void free_disk_sb(mdk_rdev_t * rdev)
467 {
468 if (rdev->sb_page) {
469 put_page(rdev->sb_page);
470 rdev->sb_loaded = 0;
471 rdev->sb_page = NULL;
472 rdev->sb_start = 0;
473 rdev->sectors = 0;
474 }
475 }
476
477
478 static void super_written(struct bio *bio, int error)
479 {
480 mdk_rdev_t *rdev = bio->bi_private;
481 mddev_t *mddev = rdev->mddev;
482
483 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
484 printk("md: super_written gets error=%d, uptodate=%d\n",
485 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
486 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
487 md_error(mddev, rdev);
488 }
489
490 if (atomic_dec_and_test(&mddev->pending_writes))
491 wake_up(&mddev->sb_wait);
492 bio_put(bio);
493 }
494
495 static void super_written_barrier(struct bio *bio, int error)
496 {
497 struct bio *bio2 = bio->bi_private;
498 mdk_rdev_t *rdev = bio2->bi_private;
499 mddev_t *mddev = rdev->mddev;
500
501 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
502 error == -EOPNOTSUPP) {
503 unsigned long flags;
504 /* barriers don't appear to be supported :-( */
505 set_bit(BarriersNotsupp, &rdev->flags);
506 mddev->barriers_work = 0;
507 spin_lock_irqsave(&mddev->write_lock, flags);
508 bio2->bi_next = mddev->biolist;
509 mddev->biolist = bio2;
510 spin_unlock_irqrestore(&mddev->write_lock, flags);
511 wake_up(&mddev->sb_wait);
512 bio_put(bio);
513 } else {
514 bio_put(bio2);
515 bio->bi_private = rdev;
516 super_written(bio, error);
517 }
518 }
519
520 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
521 sector_t sector, int size, struct page *page)
522 {
523 /* write first size bytes of page to sector of rdev
524 * Increment mddev->pending_writes before returning
525 * and decrement it on completion, waking up sb_wait
526 * if zero is reached.
527 * If an error occurred, call md_error
528 *
529 * As we might need to resubmit the request if BIO_RW_BARRIER
530 * causes ENOTSUPP, we allocate a spare bio...
531 */
532 struct bio *bio = bio_alloc(GFP_NOIO, 1);
533 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG);
534
535 bio->bi_bdev = rdev->bdev;
536 bio->bi_sector = sector;
537 bio_add_page(bio, page, size, 0);
538 bio->bi_private = rdev;
539 bio->bi_end_io = super_written;
540 bio->bi_rw = rw;
541
542 atomic_inc(&mddev->pending_writes);
543 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
544 struct bio *rbio;
545 rw |= (1<<BIO_RW_BARRIER);
546 rbio = bio_clone(bio, GFP_NOIO);
547 rbio->bi_private = bio;
548 rbio->bi_end_io = super_written_barrier;
549 submit_bio(rw, rbio);
550 } else
551 submit_bio(rw, bio);
552 }
553
554 void md_super_wait(mddev_t *mddev)
555 {
556 /* wait for all superblock writes that were scheduled to complete.
557 * if any had to be retried (due to BARRIER problems), retry them
558 */
559 DEFINE_WAIT(wq);
560 for(;;) {
561 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
562 if (atomic_read(&mddev->pending_writes)==0)
563 break;
564 while (mddev->biolist) {
565 struct bio *bio;
566 spin_lock_irq(&mddev->write_lock);
567 bio = mddev->biolist;
568 mddev->biolist = bio->bi_next ;
569 bio->bi_next = NULL;
570 spin_unlock_irq(&mddev->write_lock);
571 submit_bio(bio->bi_rw, bio);
572 }
573 schedule();
574 }
575 finish_wait(&mddev->sb_wait, &wq);
576 }
577
578 static void bi_complete(struct bio *bio, int error)
579 {
580 complete((struct completion*)bio->bi_private);
581 }
582
583 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
584 struct page *page, int rw)
585 {
586 struct bio *bio = bio_alloc(GFP_NOIO, 1);
587 struct completion event;
588 int ret;
589
590 rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
591
592 bio->bi_bdev = bdev;
593 bio->bi_sector = sector;
594 bio_add_page(bio, page, size, 0);
595 init_completion(&event);
596 bio->bi_private = &event;
597 bio->bi_end_io = bi_complete;
598 submit_bio(rw, bio);
599 wait_for_completion(&event);
600
601 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
602 bio_put(bio);
603 return ret;
604 }
605 EXPORT_SYMBOL_GPL(sync_page_io);
606
607 static int read_disk_sb(mdk_rdev_t * rdev, int size)
608 {
609 char b[BDEVNAME_SIZE];
610 if (!rdev->sb_page) {
611 MD_BUG();
612 return -EINVAL;
613 }
614 if (rdev->sb_loaded)
615 return 0;
616
617
618 if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
619 goto fail;
620 rdev->sb_loaded = 1;
621 return 0;
622
623 fail:
624 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
625 bdevname(rdev->bdev,b));
626 return -EINVAL;
627 }
628
629 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
630 {
631 return sb1->set_uuid0 == sb2->set_uuid0 &&
632 sb1->set_uuid1 == sb2->set_uuid1 &&
633 sb1->set_uuid2 == sb2->set_uuid2 &&
634 sb1->set_uuid3 == sb2->set_uuid3;
635 }
636
637 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
638 {
639 int ret;
640 mdp_super_t *tmp1, *tmp2;
641
642 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
643 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
644
645 if (!tmp1 || !tmp2) {
646 ret = 0;
647 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
648 goto abort;
649 }
650
651 *tmp1 = *sb1;
652 *tmp2 = *sb2;
653
654 /*
655 * nr_disks is not constant
656 */
657 tmp1->nr_disks = 0;
658 tmp2->nr_disks = 0;
659
660 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
661 abort:
662 kfree(tmp1);
663 kfree(tmp2);
664 return ret;
665 }
666
667
668 static u32 md_csum_fold(u32 csum)
669 {
670 csum = (csum & 0xffff) + (csum >> 16);
671 return (csum & 0xffff) + (csum >> 16);
672 }
673
674 static unsigned int calc_sb_csum(mdp_super_t * sb)
675 {
676 u64 newcsum = 0;
677 u32 *sb32 = (u32*)sb;
678 int i;
679 unsigned int disk_csum, csum;
680
681 disk_csum = sb->sb_csum;
682 sb->sb_csum = 0;
683
684 for (i = 0; i < MD_SB_BYTES/4 ; i++)
685 newcsum += sb32[i];
686 csum = (newcsum & 0xffffffff) + (newcsum>>32);
687
688
689 #ifdef CONFIG_ALPHA
690 /* This used to use csum_partial, which was wrong for several
691 * reasons including that different results are returned on
692 * different architectures. It isn't critical that we get exactly
693 * the same return value as before (we always csum_fold before
694 * testing, and that removes any differences). However as we
695 * know that csum_partial always returned a 16bit value on
696 * alphas, do a fold to maximise conformity to previous behaviour.
697 */
698 sb->sb_csum = md_csum_fold(disk_csum);
699 #else
700 sb->sb_csum = disk_csum;
701 #endif
702 return csum;
703 }
704
705
706 /*
707 * Handle superblock details.
708 * We want to be able to handle multiple superblock formats
709 * so we have a common interface to them all, and an array of
710 * different handlers.
711 * We rely on user-space to write the initial superblock, and support
712 * reading and updating of superblocks.
713 * Interface methods are:
714 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
715 * loads and validates a superblock on dev.
716 * if refdev != NULL, compare superblocks on both devices
717 * Return:
718 * 0 - dev has a superblock that is compatible with refdev
719 * 1 - dev has a superblock that is compatible and newer than refdev
720 * so dev should be used as the refdev in future
721 * -EINVAL superblock incompatible or invalid
722 * -othererror e.g. -EIO
723 *
724 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
725 * Verify that dev is acceptable into mddev.
726 * The first time, mddev->raid_disks will be 0, and data from
727 * dev should be merged in. Subsequent calls check that dev
728 * is new enough. Return 0 or -EINVAL
729 *
730 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
731 * Update the superblock for rdev with data in mddev
732 * This does not write to disc.
733 *
734 */
735
736 struct super_type {
737 char *name;
738 struct module *owner;
739 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
740 int minor_version);
741 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
742 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
743 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
744 sector_t num_sectors);
745 };
746
747 /*
748 * load_super for 0.90.0
749 */
750 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
751 {
752 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
753 mdp_super_t *sb;
754 int ret;
755
756 /*
757 * Calculate the position of the superblock (512byte sectors),
758 * it's at the end of the disk.
759 *
760 * It also happens to be a multiple of 4Kb.
761 */
762 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
763
764 ret = read_disk_sb(rdev, MD_SB_BYTES);
765 if (ret) return ret;
766
767 ret = -EINVAL;
768
769 bdevname(rdev->bdev, b);
770 sb = (mdp_super_t*)page_address(rdev->sb_page);
771
772 if (sb->md_magic != MD_SB_MAGIC) {
773 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
774 b);
775 goto abort;
776 }
777
778 if (sb->major_version != 0 ||
779 sb->minor_version < 90 ||
780 sb->minor_version > 91) {
781 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
782 sb->major_version, sb->minor_version,
783 b);
784 goto abort;
785 }
786
787 if (sb->raid_disks <= 0)
788 goto abort;
789
790 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
791 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
792 b);
793 goto abort;
794 }
795
796 rdev->preferred_minor = sb->md_minor;
797 rdev->data_offset = 0;
798 rdev->sb_size = MD_SB_BYTES;
799
800 if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
801 if (sb->level != 1 && sb->level != 4
802 && sb->level != 5 && sb->level != 6
803 && sb->level != 10) {
804 /* FIXME use a better test */
805 printk(KERN_WARNING
806 "md: bitmaps not supported for this level.\n");
807 goto abort;
808 }
809 }
810
811 if (sb->level == LEVEL_MULTIPATH)
812 rdev->desc_nr = -1;
813 else
814 rdev->desc_nr = sb->this_disk.number;
815
816 if (!refdev) {
817 ret = 1;
818 } else {
819 __u64 ev1, ev2;
820 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
821 if (!uuid_equal(refsb, sb)) {
822 printk(KERN_WARNING "md: %s has different UUID to %s\n",
823 b, bdevname(refdev->bdev,b2));
824 goto abort;
825 }
826 if (!sb_equal(refsb, sb)) {
827 printk(KERN_WARNING "md: %s has same UUID"
828 " but different superblock to %s\n",
829 b, bdevname(refdev->bdev, b2));
830 goto abort;
831 }
832 ev1 = md_event(sb);
833 ev2 = md_event(refsb);
834 if (ev1 > ev2)
835 ret = 1;
836 else
837 ret = 0;
838 }
839 rdev->sectors = calc_num_sectors(rdev, sb->chunk_size);
840
841 if (rdev->sectors < sb->size * 2 && sb->level > 1)
842 /* "this cannot possibly happen" ... */
843 ret = -EINVAL;
844
845 abort:
846 return ret;
847 }
848
849 /*
850 * validate_super for 0.90.0
851 */
852 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
853 {
854 mdp_disk_t *desc;
855 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
856 __u64 ev1 = md_event(sb);
857
858 rdev->raid_disk = -1;
859 clear_bit(Faulty, &rdev->flags);
860 clear_bit(In_sync, &rdev->flags);
861 clear_bit(WriteMostly, &rdev->flags);
862 clear_bit(BarriersNotsupp, &rdev->flags);
863
864 if (mddev->raid_disks == 0) {
865 mddev->major_version = 0;
866 mddev->minor_version = sb->minor_version;
867 mddev->patch_version = sb->patch_version;
868 mddev->external = 0;
869 mddev->chunk_size = sb->chunk_size;
870 mddev->ctime = sb->ctime;
871 mddev->utime = sb->utime;
872 mddev->level = sb->level;
873 mddev->clevel[0] = 0;
874 mddev->layout = sb->layout;
875 mddev->raid_disks = sb->raid_disks;
876 mddev->dev_sectors = sb->size * 2;
877 mddev->events = ev1;
878 mddev->bitmap_offset = 0;
879 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
880
881 if (mddev->minor_version >= 91) {
882 mddev->reshape_position = sb->reshape_position;
883 mddev->delta_disks = sb->delta_disks;
884 mddev->new_level = sb->new_level;
885 mddev->new_layout = sb->new_layout;
886 mddev->new_chunk = sb->new_chunk;
887 } else {
888 mddev->reshape_position = MaxSector;
889 mddev->delta_disks = 0;
890 mddev->new_level = mddev->level;
891 mddev->new_layout = mddev->layout;
892 mddev->new_chunk = mddev->chunk_size;
893 }
894
895 if (sb->state & (1<<MD_SB_CLEAN))
896 mddev->recovery_cp = MaxSector;
897 else {
898 if (sb->events_hi == sb->cp_events_hi &&
899 sb->events_lo == sb->cp_events_lo) {
900 mddev->recovery_cp = sb->recovery_cp;
901 } else
902 mddev->recovery_cp = 0;
903 }
904
905 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
906 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
907 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
908 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
909
910 mddev->max_disks = MD_SB_DISKS;
911
912 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
913 mddev->bitmap_file == NULL)
914 mddev->bitmap_offset = mddev->default_bitmap_offset;
915
916 } else if (mddev->pers == NULL) {
917 /* Insist on good event counter while assembling */
918 ++ev1;
919 if (ev1 < mddev->events)
920 return -EINVAL;
921 } else if (mddev->bitmap) {
922 /* if adding to array with a bitmap, then we can accept an
923 * older device ... but not too old.
924 */
925 if (ev1 < mddev->bitmap->events_cleared)
926 return 0;
927 } else {
928 if (ev1 < mddev->events)
929 /* just a hot-add of a new device, leave raid_disk at -1 */
930 return 0;
931 }
932
933 if (mddev->level != LEVEL_MULTIPATH) {
934 desc = sb->disks + rdev->desc_nr;
935
936 if (desc->state & (1<<MD_DISK_FAULTY))
937 set_bit(Faulty, &rdev->flags);
938 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
939 desc->raid_disk < mddev->raid_disks */) {
940 set_bit(In_sync, &rdev->flags);
941 rdev->raid_disk = desc->raid_disk;
942 }
943 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
944 set_bit(WriteMostly, &rdev->flags);
945 } else /* MULTIPATH are always insync */
946 set_bit(In_sync, &rdev->flags);
947 return 0;
948 }
949
950 /*
951 * sync_super for 0.90.0
952 */
953 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
954 {
955 mdp_super_t *sb;
956 mdk_rdev_t *rdev2;
957 int next_spare = mddev->raid_disks;
958
959
960 /* make rdev->sb match mddev data..
961 *
962 * 1/ zero out disks
963 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
964 * 3/ any empty disks < next_spare become removed
965 *
966 * disks[0] gets initialised to REMOVED because
967 * we cannot be sure from other fields if it has
968 * been initialised or not.
969 */
970 int i;
971 int active=0, working=0,failed=0,spare=0,nr_disks=0;
972
973 rdev->sb_size = MD_SB_BYTES;
974
975 sb = (mdp_super_t*)page_address(rdev->sb_page);
976
977 memset(sb, 0, sizeof(*sb));
978
979 sb->md_magic = MD_SB_MAGIC;
980 sb->major_version = mddev->major_version;
981 sb->patch_version = mddev->patch_version;
982 sb->gvalid_words = 0; /* ignored */
983 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
984 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
985 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
986 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
987
988 sb->ctime = mddev->ctime;
989 sb->level = mddev->level;
990 sb->size = mddev->dev_sectors / 2;
991 sb->raid_disks = mddev->raid_disks;
992 sb->md_minor = mddev->md_minor;
993 sb->not_persistent = 0;
994 sb->utime = mddev->utime;
995 sb->state = 0;
996 sb->events_hi = (mddev->events>>32);
997 sb->events_lo = (u32)mddev->events;
998
999 if (mddev->reshape_position == MaxSector)
1000 sb->minor_version = 90;
1001 else {
1002 sb->minor_version = 91;
1003 sb->reshape_position = mddev->reshape_position;
1004 sb->new_level = mddev->new_level;
1005 sb->delta_disks = mddev->delta_disks;
1006 sb->new_layout = mddev->new_layout;
1007 sb->new_chunk = mddev->new_chunk;
1008 }
1009 mddev->minor_version = sb->minor_version;
1010 if (mddev->in_sync)
1011 {
1012 sb->recovery_cp = mddev->recovery_cp;
1013 sb->cp_events_hi = (mddev->events>>32);
1014 sb->cp_events_lo = (u32)mddev->events;
1015 if (mddev->recovery_cp == MaxSector)
1016 sb->state = (1<< MD_SB_CLEAN);
1017 } else
1018 sb->recovery_cp = 0;
1019
1020 sb->layout = mddev->layout;
1021 sb->chunk_size = mddev->chunk_size;
1022
1023 if (mddev->bitmap && mddev->bitmap_file == NULL)
1024 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1025
1026 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1027 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1028 mdp_disk_t *d;
1029 int desc_nr;
1030 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
1031 && !test_bit(Faulty, &rdev2->flags))
1032 desc_nr = rdev2->raid_disk;
1033 else
1034 desc_nr = next_spare++;
1035 rdev2->desc_nr = desc_nr;
1036 d = &sb->disks[rdev2->desc_nr];
1037 nr_disks++;
1038 d->number = rdev2->desc_nr;
1039 d->major = MAJOR(rdev2->bdev->bd_dev);
1040 d->minor = MINOR(rdev2->bdev->bd_dev);
1041 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
1042 && !test_bit(Faulty, &rdev2->flags))
1043 d->raid_disk = rdev2->raid_disk;
1044 else
1045 d->raid_disk = rdev2->desc_nr; /* compatibility */
1046 if (test_bit(Faulty, &rdev2->flags))
1047 d->state = (1<<MD_DISK_FAULTY);
1048 else if (test_bit(In_sync, &rdev2->flags)) {
1049 d->state = (1<<MD_DISK_ACTIVE);
1050 d->state |= (1<<MD_DISK_SYNC);
1051 active++;
1052 working++;
1053 } else {
1054 d->state = 0;
1055 spare++;
1056 working++;
1057 }
1058 if (test_bit(WriteMostly, &rdev2->flags))
1059 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1060 }
1061 /* now set the "removed" and "faulty" bits on any missing devices */
1062 for (i=0 ; i < mddev->raid_disks ; i++) {
1063 mdp_disk_t *d = &sb->disks[i];
1064 if (d->state == 0 && d->number == 0) {
1065 d->number = i;
1066 d->raid_disk = i;
1067 d->state = (1<<MD_DISK_REMOVED);
1068 d->state |= (1<<MD_DISK_FAULTY);
1069 failed++;
1070 }
1071 }
1072 sb->nr_disks = nr_disks;
1073 sb->active_disks = active;
1074 sb->working_disks = working;
1075 sb->failed_disks = failed;
1076 sb->spare_disks = spare;
1077
1078 sb->this_disk = sb->disks[rdev->desc_nr];
1079 sb->sb_csum = calc_sb_csum(sb);
1080 }
1081
1082 /*
1083 * rdev_size_change for 0.90.0
1084 */
1085 static unsigned long long
1086 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1087 {
1088 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1089 return 0; /* component must fit device */
1090 if (rdev->mddev->bitmap_offset)
1091 return 0; /* can't move bitmap */
1092 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1093 if (!num_sectors || num_sectors > rdev->sb_start)
1094 num_sectors = rdev->sb_start;
1095 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1096 rdev->sb_page);
1097 md_super_wait(rdev->mddev);
1098 return num_sectors / 2; /* kB for sysfs */
1099 }
1100
1101
1102 /*
1103 * version 1 superblock
1104 */
1105
1106 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1107 {
1108 __le32 disk_csum;
1109 u32 csum;
1110 unsigned long long newcsum;
1111 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1112 __le32 *isuper = (__le32*)sb;
1113 int i;
1114
1115 disk_csum = sb->sb_csum;
1116 sb->sb_csum = 0;
1117 newcsum = 0;
1118 for (i=0; size>=4; size -= 4 )
1119 newcsum += le32_to_cpu(*isuper++);
1120
1121 if (size == 2)
1122 newcsum += le16_to_cpu(*(__le16*) isuper);
1123
1124 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1125 sb->sb_csum = disk_csum;
1126 return cpu_to_le32(csum);
1127 }
1128
1129 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1130 {
1131 struct mdp_superblock_1 *sb;
1132 int ret;
1133 sector_t sb_start;
1134 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1135 int bmask;
1136
1137 /*
1138 * Calculate the position of the superblock in 512byte sectors.
1139 * It is always aligned to a 4K boundary and
1140 * depeding on minor_version, it can be:
1141 * 0: At least 8K, but less than 12K, from end of device
1142 * 1: At start of device
1143 * 2: 4K from start of device.
1144 */
1145 switch(minor_version) {
1146 case 0:
1147 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1148 sb_start -= 8*2;
1149 sb_start &= ~(sector_t)(4*2-1);
1150 break;
1151 case 1:
1152 sb_start = 0;
1153 break;
1154 case 2:
1155 sb_start = 8;
1156 break;
1157 default:
1158 return -EINVAL;
1159 }
1160 rdev->sb_start = sb_start;
1161
1162 /* superblock is rarely larger than 1K, but it can be larger,
1163 * and it is safe to read 4k, so we do that
1164 */
1165 ret = read_disk_sb(rdev, 4096);
1166 if (ret) return ret;
1167
1168
1169 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1170
1171 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1172 sb->major_version != cpu_to_le32(1) ||
1173 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1174 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1175 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1176 return -EINVAL;
1177
1178 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1179 printk("md: invalid superblock checksum on %s\n",
1180 bdevname(rdev->bdev,b));
1181 return -EINVAL;
1182 }
1183 if (le64_to_cpu(sb->data_size) < 10) {
1184 printk("md: data_size too small on %s\n",
1185 bdevname(rdev->bdev,b));
1186 return -EINVAL;
1187 }
1188 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1189 if (sb->level != cpu_to_le32(1) &&
1190 sb->level != cpu_to_le32(4) &&
1191 sb->level != cpu_to_le32(5) &&
1192 sb->level != cpu_to_le32(6) &&
1193 sb->level != cpu_to_le32(10)) {
1194 printk(KERN_WARNING
1195 "md: bitmaps not supported for this level.\n");
1196 return -EINVAL;
1197 }
1198 }
1199
1200 rdev->preferred_minor = 0xffff;
1201 rdev->data_offset = le64_to_cpu(sb->data_offset);
1202 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1203
1204 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1205 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1206 if (rdev->sb_size & bmask)
1207 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1208
1209 if (minor_version
1210 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1211 return -EINVAL;
1212
1213 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1214 rdev->desc_nr = -1;
1215 else
1216 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1217
1218 if (!refdev) {
1219 ret = 1;
1220 } else {
1221 __u64 ev1, ev2;
1222 struct mdp_superblock_1 *refsb =
1223 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1224
1225 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1226 sb->level != refsb->level ||
1227 sb->layout != refsb->layout ||
1228 sb->chunksize != refsb->chunksize) {
1229 printk(KERN_WARNING "md: %s has strangely different"
1230 " superblock to %s\n",
1231 bdevname(rdev->bdev,b),
1232 bdevname(refdev->bdev,b2));
1233 return -EINVAL;
1234 }
1235 ev1 = le64_to_cpu(sb->events);
1236 ev2 = le64_to_cpu(refsb->events);
1237
1238 if (ev1 > ev2)
1239 ret = 1;
1240 else
1241 ret = 0;
1242 }
1243 if (minor_version)
1244 rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
1245 le64_to_cpu(sb->data_offset);
1246 else
1247 rdev->sectors = rdev->sb_start;
1248 if (rdev->sectors < le64_to_cpu(sb->data_size))
1249 return -EINVAL;
1250 rdev->sectors = le64_to_cpu(sb->data_size);
1251 if (le32_to_cpu(sb->chunksize))
1252 rdev->sectors &= ~((sector_t)le32_to_cpu(sb->chunksize) - 1);
1253
1254 if (le64_to_cpu(sb->size) > rdev->sectors)
1255 return -EINVAL;
1256 return ret;
1257 }
1258
1259 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1260 {
1261 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1262 __u64 ev1 = le64_to_cpu(sb->events);
1263
1264 rdev->raid_disk = -1;
1265 clear_bit(Faulty, &rdev->flags);
1266 clear_bit(In_sync, &rdev->flags);
1267 clear_bit(WriteMostly, &rdev->flags);
1268 clear_bit(BarriersNotsupp, &rdev->flags);
1269
1270 if (mddev->raid_disks == 0) {
1271 mddev->major_version = 1;
1272 mddev->patch_version = 0;
1273 mddev->external = 0;
1274 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1275 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1276 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1277 mddev->level = le32_to_cpu(sb->level);
1278 mddev->clevel[0] = 0;
1279 mddev->layout = le32_to_cpu(sb->layout);
1280 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1281 mddev->dev_sectors = le64_to_cpu(sb->size);
1282 mddev->events = ev1;
1283 mddev->bitmap_offset = 0;
1284 mddev->default_bitmap_offset = 1024 >> 9;
1285
1286 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1287 memcpy(mddev->uuid, sb->set_uuid, 16);
1288
1289 mddev->max_disks = (4096-256)/2;
1290
1291 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1292 mddev->bitmap_file == NULL )
1293 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1294
1295 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1296 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1297 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1298 mddev->new_level = le32_to_cpu(sb->new_level);
1299 mddev->new_layout = le32_to_cpu(sb->new_layout);
1300 mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1301 } else {
1302 mddev->reshape_position = MaxSector;
1303 mddev->delta_disks = 0;
1304 mddev->new_level = mddev->level;
1305 mddev->new_layout = mddev->layout;
1306 mddev->new_chunk = mddev->chunk_size;
1307 }
1308
1309 } else if (mddev->pers == NULL) {
1310 /* Insist of good event counter while assembling */
1311 ++ev1;
1312 if (ev1 < mddev->events)
1313 return -EINVAL;
1314 } else if (mddev->bitmap) {
1315 /* If adding to array with a bitmap, then we can accept an
1316 * older device, but not too old.
1317 */
1318 if (ev1 < mddev->bitmap->events_cleared)
1319 return 0;
1320 } else {
1321 if (ev1 < mddev->events)
1322 /* just a hot-add of a new device, leave raid_disk at -1 */
1323 return 0;
1324 }
1325 if (mddev->level != LEVEL_MULTIPATH) {
1326 int role;
1327 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1328 switch(role) {
1329 case 0xffff: /* spare */
1330 break;
1331 case 0xfffe: /* faulty */
1332 set_bit(Faulty, &rdev->flags);
1333 break;
1334 default:
1335 if ((le32_to_cpu(sb->feature_map) &
1336 MD_FEATURE_RECOVERY_OFFSET))
1337 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1338 else
1339 set_bit(In_sync, &rdev->flags);
1340 rdev->raid_disk = role;
1341 break;
1342 }
1343 if (sb->devflags & WriteMostly1)
1344 set_bit(WriteMostly, &rdev->flags);
1345 } else /* MULTIPATH are always insync */
1346 set_bit(In_sync, &rdev->flags);
1347
1348 return 0;
1349 }
1350
1351 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1352 {
1353 struct mdp_superblock_1 *sb;
1354 mdk_rdev_t *rdev2;
1355 int max_dev, i;
1356 /* make rdev->sb match mddev and rdev data. */
1357
1358 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1359
1360 sb->feature_map = 0;
1361 sb->pad0 = 0;
1362 sb->recovery_offset = cpu_to_le64(0);
1363 memset(sb->pad1, 0, sizeof(sb->pad1));
1364 memset(sb->pad2, 0, sizeof(sb->pad2));
1365 memset(sb->pad3, 0, sizeof(sb->pad3));
1366
1367 sb->utime = cpu_to_le64((__u64)mddev->utime);
1368 sb->events = cpu_to_le64(mddev->events);
1369 if (mddev->in_sync)
1370 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1371 else
1372 sb->resync_offset = cpu_to_le64(0);
1373
1374 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1375
1376 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1377 sb->size = cpu_to_le64(mddev->dev_sectors);
1378 sb->chunksize = cpu_to_le32(mddev->chunk_size >> 9);
1379 sb->level = cpu_to_le32(mddev->level);
1380 sb->layout = cpu_to_le32(mddev->layout);
1381
1382 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1383 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1384 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1385 }
1386
1387 if (rdev->raid_disk >= 0 &&
1388 !test_bit(In_sync, &rdev->flags)) {
1389 if (mddev->curr_resync_completed > rdev->recovery_offset)
1390 rdev->recovery_offset = mddev->curr_resync_completed;
1391 if (rdev->recovery_offset > 0) {
1392 sb->feature_map |=
1393 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1394 sb->recovery_offset =
1395 cpu_to_le64(rdev->recovery_offset);
1396 }
1397 }
1398
1399 if (mddev->reshape_position != MaxSector) {
1400 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1401 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1402 sb->new_layout = cpu_to_le32(mddev->new_layout);
1403 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1404 sb->new_level = cpu_to_le32(mddev->new_level);
1405 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1406 }
1407
1408 max_dev = 0;
1409 list_for_each_entry(rdev2, &mddev->disks, same_set)
1410 if (rdev2->desc_nr+1 > max_dev)
1411 max_dev = rdev2->desc_nr+1;
1412
1413 if (max_dev > le32_to_cpu(sb->max_dev))
1414 sb->max_dev = cpu_to_le32(max_dev);
1415 for (i=0; i<max_dev;i++)
1416 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1417
1418 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1419 i = rdev2->desc_nr;
1420 if (test_bit(Faulty, &rdev2->flags))
1421 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1422 else if (test_bit(In_sync, &rdev2->flags))
1423 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1424 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1425 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1426 else
1427 sb->dev_roles[i] = cpu_to_le16(0xffff);
1428 }
1429
1430 sb->sb_csum = calc_sb_1_csum(sb);
1431 }
1432
1433 static unsigned long long
1434 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1435 {
1436 struct mdp_superblock_1 *sb;
1437 sector_t max_sectors;
1438 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1439 return 0; /* component must fit device */
1440 if (rdev->sb_start < rdev->data_offset) {
1441 /* minor versions 1 and 2; superblock before data */
1442 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1443 max_sectors -= rdev->data_offset;
1444 if (!num_sectors || num_sectors > max_sectors)
1445 num_sectors = max_sectors;
1446 } else if (rdev->mddev->bitmap_offset) {
1447 /* minor version 0 with bitmap we can't move */
1448 return 0;
1449 } else {
1450 /* minor version 0; superblock after data */
1451 sector_t sb_start;
1452 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1453 sb_start &= ~(sector_t)(4*2 - 1);
1454 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1455 if (!num_sectors || num_sectors > max_sectors)
1456 num_sectors = max_sectors;
1457 rdev->sb_start = sb_start;
1458 }
1459 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1460 sb->data_size = cpu_to_le64(num_sectors);
1461 sb->super_offset = rdev->sb_start;
1462 sb->sb_csum = calc_sb_1_csum(sb);
1463 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1464 rdev->sb_page);
1465 md_super_wait(rdev->mddev);
1466 return num_sectors / 2; /* kB for sysfs */
1467 }
1468
1469 static struct super_type super_types[] = {
1470 [0] = {
1471 .name = "0.90.0",
1472 .owner = THIS_MODULE,
1473 .load_super = super_90_load,
1474 .validate_super = super_90_validate,
1475 .sync_super = super_90_sync,
1476 .rdev_size_change = super_90_rdev_size_change,
1477 },
1478 [1] = {
1479 .name = "md-1",
1480 .owner = THIS_MODULE,
1481 .load_super = super_1_load,
1482 .validate_super = super_1_validate,
1483 .sync_super = super_1_sync,
1484 .rdev_size_change = super_1_rdev_size_change,
1485 },
1486 };
1487
1488 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1489 {
1490 mdk_rdev_t *rdev, *rdev2;
1491
1492 rcu_read_lock();
1493 rdev_for_each_rcu(rdev, mddev1)
1494 rdev_for_each_rcu(rdev2, mddev2)
1495 if (rdev->bdev->bd_contains ==
1496 rdev2->bdev->bd_contains) {
1497 rcu_read_unlock();
1498 return 1;
1499 }
1500 rcu_read_unlock();
1501 return 0;
1502 }
1503
1504 static LIST_HEAD(pending_raid_disks);
1505
1506 static void md_integrity_check(mdk_rdev_t *rdev, mddev_t *mddev)
1507 {
1508 struct mdk_personality *pers = mddev->pers;
1509 struct gendisk *disk = mddev->gendisk;
1510 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1511 struct blk_integrity *bi_mddev = blk_get_integrity(disk);
1512
1513 /* Data integrity passthrough not supported on RAID 4, 5 and 6 */
1514 if (pers && pers->level >= 4 && pers->level <= 6)
1515 return;
1516
1517 /* If rdev is integrity capable, register profile for mddev */
1518 if (!bi_mddev && bi_rdev) {
1519 if (blk_integrity_register(disk, bi_rdev))
1520 printk(KERN_ERR "%s: %s Could not register integrity!\n",
1521 __func__, disk->disk_name);
1522 else
1523 printk(KERN_NOTICE "Enabling data integrity on %s\n",
1524 disk->disk_name);
1525 return;
1526 }
1527
1528 /* Check that mddev and rdev have matching profiles */
1529 if (blk_integrity_compare(disk, rdev->bdev->bd_disk) < 0) {
1530 printk(KERN_ERR "%s: %s/%s integrity mismatch!\n", __func__,
1531 disk->disk_name, rdev->bdev->bd_disk->disk_name);
1532 printk(KERN_NOTICE "Disabling data integrity on %s\n",
1533 disk->disk_name);
1534 blk_integrity_unregister(disk);
1535 }
1536 }
1537
1538 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1539 {
1540 char b[BDEVNAME_SIZE];
1541 struct kobject *ko;
1542 char *s;
1543 int err;
1544
1545 if (rdev->mddev) {
1546 MD_BUG();
1547 return -EINVAL;
1548 }
1549
1550 /* prevent duplicates */
1551 if (find_rdev(mddev, rdev->bdev->bd_dev))
1552 return -EEXIST;
1553
1554 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1555 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1556 rdev->sectors < mddev->dev_sectors)) {
1557 if (mddev->pers) {
1558 /* Cannot change size, so fail
1559 * If mddev->level <= 0, then we don't care
1560 * about aligning sizes (e.g. linear)
1561 */
1562 if (mddev->level > 0)
1563 return -ENOSPC;
1564 } else
1565 mddev->dev_sectors = rdev->sectors;
1566 }
1567
1568 /* Verify rdev->desc_nr is unique.
1569 * If it is -1, assign a free number, else
1570 * check number is not in use
1571 */
1572 if (rdev->desc_nr < 0) {
1573 int choice = 0;
1574 if (mddev->pers) choice = mddev->raid_disks;
1575 while (find_rdev_nr(mddev, choice))
1576 choice++;
1577 rdev->desc_nr = choice;
1578 } else {
1579 if (find_rdev_nr(mddev, rdev->desc_nr))
1580 return -EBUSY;
1581 }
1582 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1583 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1584 mdname(mddev), mddev->max_disks);
1585 return -EBUSY;
1586 }
1587 bdevname(rdev->bdev,b);
1588 while ( (s=strchr(b, '/')) != NULL)
1589 *s = '!';
1590
1591 rdev->mddev = mddev;
1592 printk(KERN_INFO "md: bind<%s>\n", b);
1593
1594 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1595 goto fail;
1596
1597 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1598 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1599 kobject_del(&rdev->kobj);
1600 goto fail;
1601 }
1602 rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, "state");
1603
1604 list_add_rcu(&rdev->same_set, &mddev->disks);
1605 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1606
1607 /* May as well allow recovery to be retried once */
1608 mddev->recovery_disabled = 0;
1609
1610 md_integrity_check(rdev, mddev);
1611 return 0;
1612
1613 fail:
1614 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1615 b, mdname(mddev));
1616 return err;
1617 }
1618
1619 static void md_delayed_delete(struct work_struct *ws)
1620 {
1621 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1622 kobject_del(&rdev->kobj);
1623 kobject_put(&rdev->kobj);
1624 }
1625
1626 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1627 {
1628 char b[BDEVNAME_SIZE];
1629 if (!rdev->mddev) {
1630 MD_BUG();
1631 return;
1632 }
1633 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1634 list_del_rcu(&rdev->same_set);
1635 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1636 rdev->mddev = NULL;
1637 sysfs_remove_link(&rdev->kobj, "block");
1638 sysfs_put(rdev->sysfs_state);
1639 rdev->sysfs_state = NULL;
1640 /* We need to delay this, otherwise we can deadlock when
1641 * writing to 'remove' to "dev/state". We also need
1642 * to delay it due to rcu usage.
1643 */
1644 synchronize_rcu();
1645 INIT_WORK(&rdev->del_work, md_delayed_delete);
1646 kobject_get(&rdev->kobj);
1647 schedule_work(&rdev->del_work);
1648 }
1649
1650 /*
1651 * prevent the device from being mounted, repartitioned or
1652 * otherwise reused by a RAID array (or any other kernel
1653 * subsystem), by bd_claiming the device.
1654 */
1655 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1656 {
1657 int err = 0;
1658 struct block_device *bdev;
1659 char b[BDEVNAME_SIZE];
1660
1661 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1662 if (IS_ERR(bdev)) {
1663 printk(KERN_ERR "md: could not open %s.\n",
1664 __bdevname(dev, b));
1665 return PTR_ERR(bdev);
1666 }
1667 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1668 if (err) {
1669 printk(KERN_ERR "md: could not bd_claim %s.\n",
1670 bdevname(bdev, b));
1671 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1672 return err;
1673 }
1674 if (!shared)
1675 set_bit(AllReserved, &rdev->flags);
1676 rdev->bdev = bdev;
1677 return err;
1678 }
1679
1680 static void unlock_rdev(mdk_rdev_t *rdev)
1681 {
1682 struct block_device *bdev = rdev->bdev;
1683 rdev->bdev = NULL;
1684 if (!bdev)
1685 MD_BUG();
1686 bd_release(bdev);
1687 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1688 }
1689
1690 void md_autodetect_dev(dev_t dev);
1691
1692 static void export_rdev(mdk_rdev_t * rdev)
1693 {
1694 char b[BDEVNAME_SIZE];
1695 printk(KERN_INFO "md: export_rdev(%s)\n",
1696 bdevname(rdev->bdev,b));
1697 if (rdev->mddev)
1698 MD_BUG();
1699 free_disk_sb(rdev);
1700 #ifndef MODULE
1701 if (test_bit(AutoDetected, &rdev->flags))
1702 md_autodetect_dev(rdev->bdev->bd_dev);
1703 #endif
1704 unlock_rdev(rdev);
1705 kobject_put(&rdev->kobj);
1706 }
1707
1708 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1709 {
1710 unbind_rdev_from_array(rdev);
1711 export_rdev(rdev);
1712 }
1713
1714 static void export_array(mddev_t *mddev)
1715 {
1716 mdk_rdev_t *rdev, *tmp;
1717
1718 rdev_for_each(rdev, tmp, mddev) {
1719 if (!rdev->mddev) {
1720 MD_BUG();
1721 continue;
1722 }
1723 kick_rdev_from_array(rdev);
1724 }
1725 if (!list_empty(&mddev->disks))
1726 MD_BUG();
1727 mddev->raid_disks = 0;
1728 mddev->major_version = 0;
1729 }
1730
1731 static void print_desc(mdp_disk_t *desc)
1732 {
1733 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1734 desc->major,desc->minor,desc->raid_disk,desc->state);
1735 }
1736
1737 static void print_sb_90(mdp_super_t *sb)
1738 {
1739 int i;
1740
1741 printk(KERN_INFO
1742 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1743 sb->major_version, sb->minor_version, sb->patch_version,
1744 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1745 sb->ctime);
1746 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1747 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1748 sb->md_minor, sb->layout, sb->chunk_size);
1749 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1750 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1751 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1752 sb->failed_disks, sb->spare_disks,
1753 sb->sb_csum, (unsigned long)sb->events_lo);
1754
1755 printk(KERN_INFO);
1756 for (i = 0; i < MD_SB_DISKS; i++) {
1757 mdp_disk_t *desc;
1758
1759 desc = sb->disks + i;
1760 if (desc->number || desc->major || desc->minor ||
1761 desc->raid_disk || (desc->state && (desc->state != 4))) {
1762 printk(" D %2d: ", i);
1763 print_desc(desc);
1764 }
1765 }
1766 printk(KERN_INFO "md: THIS: ");
1767 print_desc(&sb->this_disk);
1768 }
1769
1770 static void print_sb_1(struct mdp_superblock_1 *sb)
1771 {
1772 __u8 *uuid;
1773
1774 uuid = sb->set_uuid;
1775 printk(KERN_INFO "md: SB: (V:%u) (F:0x%08x) Array-ID:<%02x%02x%02x%02x"
1776 ":%02x%02x:%02x%02x:%02x%02x:%02x%02x%02x%02x%02x%02x>\n"
1777 KERN_INFO "md: Name: \"%s\" CT:%llu\n",
1778 le32_to_cpu(sb->major_version),
1779 le32_to_cpu(sb->feature_map),
1780 uuid[0], uuid[1], uuid[2], uuid[3],
1781 uuid[4], uuid[5], uuid[6], uuid[7],
1782 uuid[8], uuid[9], uuid[10], uuid[11],
1783 uuid[12], uuid[13], uuid[14], uuid[15],
1784 sb->set_name,
1785 (unsigned long long)le64_to_cpu(sb->ctime)
1786 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
1787
1788 uuid = sb->device_uuid;
1789 printk(KERN_INFO "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
1790 " RO:%llu\n"
1791 KERN_INFO "md: Dev:%08x UUID: %02x%02x%02x%02x:%02x%02x:%02x%02x:%02x%02x"
1792 ":%02x%02x%02x%02x%02x%02x\n"
1793 KERN_INFO "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
1794 KERN_INFO "md: (MaxDev:%u) \n",
1795 le32_to_cpu(sb->level),
1796 (unsigned long long)le64_to_cpu(sb->size),
1797 le32_to_cpu(sb->raid_disks),
1798 le32_to_cpu(sb->layout),
1799 le32_to_cpu(sb->chunksize),
1800 (unsigned long long)le64_to_cpu(sb->data_offset),
1801 (unsigned long long)le64_to_cpu(sb->data_size),
1802 (unsigned long long)le64_to_cpu(sb->super_offset),
1803 (unsigned long long)le64_to_cpu(sb->recovery_offset),
1804 le32_to_cpu(sb->dev_number),
1805 uuid[0], uuid[1], uuid[2], uuid[3],
1806 uuid[4], uuid[5], uuid[6], uuid[7],
1807 uuid[8], uuid[9], uuid[10], uuid[11],
1808 uuid[12], uuid[13], uuid[14], uuid[15],
1809 sb->devflags,
1810 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
1811 (unsigned long long)le64_to_cpu(sb->events),
1812 (unsigned long long)le64_to_cpu(sb->resync_offset),
1813 le32_to_cpu(sb->sb_csum),
1814 le32_to_cpu(sb->max_dev)
1815 );
1816 }
1817
1818 static void print_rdev(mdk_rdev_t *rdev, int major_version)
1819 {
1820 char b[BDEVNAME_SIZE];
1821 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
1822 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
1823 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1824 rdev->desc_nr);
1825 if (rdev->sb_loaded) {
1826 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
1827 switch (major_version) {
1828 case 0:
1829 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
1830 break;
1831 case 1:
1832 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
1833 break;
1834 }
1835 } else
1836 printk(KERN_INFO "md: no rdev superblock!\n");
1837 }
1838
1839 static void md_print_devices(void)
1840 {
1841 struct list_head *tmp;
1842 mdk_rdev_t *rdev;
1843 mddev_t *mddev;
1844 char b[BDEVNAME_SIZE];
1845
1846 printk("\n");
1847 printk("md: **********************************\n");
1848 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1849 printk("md: **********************************\n");
1850 for_each_mddev(mddev, tmp) {
1851
1852 if (mddev->bitmap)
1853 bitmap_print_sb(mddev->bitmap);
1854 else
1855 printk("%s: ", mdname(mddev));
1856 list_for_each_entry(rdev, &mddev->disks, same_set)
1857 printk("<%s>", bdevname(rdev->bdev,b));
1858 printk("\n");
1859
1860 list_for_each_entry(rdev, &mddev->disks, same_set)
1861 print_rdev(rdev, mddev->major_version);
1862 }
1863 printk("md: **********************************\n");
1864 printk("\n");
1865 }
1866
1867
1868 static void sync_sbs(mddev_t * mddev, int nospares)
1869 {
1870 /* Update each superblock (in-memory image), but
1871 * if we are allowed to, skip spares which already
1872 * have the right event counter, or have one earlier
1873 * (which would mean they aren't being marked as dirty
1874 * with the rest of the array)
1875 */
1876 mdk_rdev_t *rdev;
1877
1878 list_for_each_entry(rdev, &mddev->disks, same_set) {
1879 if (rdev->sb_events == mddev->events ||
1880 (nospares &&
1881 rdev->raid_disk < 0 &&
1882 (rdev->sb_events&1)==0 &&
1883 rdev->sb_events+1 == mddev->events)) {
1884 /* Don't update this superblock */
1885 rdev->sb_loaded = 2;
1886 } else {
1887 super_types[mddev->major_version].
1888 sync_super(mddev, rdev);
1889 rdev->sb_loaded = 1;
1890 }
1891 }
1892 }
1893
1894 static void md_update_sb(mddev_t * mddev, int force_change)
1895 {
1896 mdk_rdev_t *rdev;
1897 int sync_req;
1898 int nospares = 0;
1899
1900 if (mddev->external)
1901 return;
1902 repeat:
1903 spin_lock_irq(&mddev->write_lock);
1904
1905 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1906 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1907 force_change = 1;
1908 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1909 /* just a clean<-> dirty transition, possibly leave spares alone,
1910 * though if events isn't the right even/odd, we will have to do
1911 * spares after all
1912 */
1913 nospares = 1;
1914 if (force_change)
1915 nospares = 0;
1916 if (mddev->degraded)
1917 /* If the array is degraded, then skipping spares is both
1918 * dangerous and fairly pointless.
1919 * Dangerous because a device that was removed from the array
1920 * might have a event_count that still looks up-to-date,
1921 * so it can be re-added without a resync.
1922 * Pointless because if there are any spares to skip,
1923 * then a recovery will happen and soon that array won't
1924 * be degraded any more and the spare can go back to sleep then.
1925 */
1926 nospares = 0;
1927
1928 sync_req = mddev->in_sync;
1929 mddev->utime = get_seconds();
1930
1931 /* If this is just a dirty<->clean transition, and the array is clean
1932 * and 'events' is odd, we can roll back to the previous clean state */
1933 if (nospares
1934 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1935 && (mddev->events & 1)
1936 && mddev->events != 1)
1937 mddev->events--;
1938 else {
1939 /* otherwise we have to go forward and ... */
1940 mddev->events ++;
1941 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1942 /* .. if the array isn't clean, insist on an odd 'events' */
1943 if ((mddev->events&1)==0) {
1944 mddev->events++;
1945 nospares = 0;
1946 }
1947 } else {
1948 /* otherwise insist on an even 'events' (for clean states) */
1949 if ((mddev->events&1)) {
1950 mddev->events++;
1951 nospares = 0;
1952 }
1953 }
1954 }
1955
1956 if (!mddev->events) {
1957 /*
1958 * oops, this 64-bit counter should never wrap.
1959 * Either we are in around ~1 trillion A.C., assuming
1960 * 1 reboot per second, or we have a bug:
1961 */
1962 MD_BUG();
1963 mddev->events --;
1964 }
1965
1966 /*
1967 * do not write anything to disk if using
1968 * nonpersistent superblocks
1969 */
1970 if (!mddev->persistent) {
1971 if (!mddev->external)
1972 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1973
1974 spin_unlock_irq(&mddev->write_lock);
1975 wake_up(&mddev->sb_wait);
1976 return;
1977 }
1978 sync_sbs(mddev, nospares);
1979 spin_unlock_irq(&mddev->write_lock);
1980
1981 dprintk(KERN_INFO
1982 "md: updating %s RAID superblock on device (in sync %d)\n",
1983 mdname(mddev),mddev->in_sync);
1984
1985 bitmap_update_sb(mddev->bitmap);
1986 list_for_each_entry(rdev, &mddev->disks, same_set) {
1987 char b[BDEVNAME_SIZE];
1988 dprintk(KERN_INFO "md: ");
1989 if (rdev->sb_loaded != 1)
1990 continue; /* no noise on spare devices */
1991 if (test_bit(Faulty, &rdev->flags))
1992 dprintk("(skipping faulty ");
1993
1994 dprintk("%s ", bdevname(rdev->bdev,b));
1995 if (!test_bit(Faulty, &rdev->flags)) {
1996 md_super_write(mddev,rdev,
1997 rdev->sb_start, rdev->sb_size,
1998 rdev->sb_page);
1999 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2000 bdevname(rdev->bdev,b),
2001 (unsigned long long)rdev->sb_start);
2002 rdev->sb_events = mddev->events;
2003
2004 } else
2005 dprintk(")\n");
2006 if (mddev->level == LEVEL_MULTIPATH)
2007 /* only need to write one superblock... */
2008 break;
2009 }
2010 md_super_wait(mddev);
2011 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2012
2013 spin_lock_irq(&mddev->write_lock);
2014 if (mddev->in_sync != sync_req ||
2015 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2016 /* have to write it out again */
2017 spin_unlock_irq(&mddev->write_lock);
2018 goto repeat;
2019 }
2020 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2021 spin_unlock_irq(&mddev->write_lock);
2022 wake_up(&mddev->sb_wait);
2023 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2024 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2025
2026 }
2027
2028 /* words written to sysfs files may, or may not, be \n terminated.
2029 * We want to accept with case. For this we use cmd_match.
2030 */
2031 static int cmd_match(const char *cmd, const char *str)
2032 {
2033 /* See if cmd, written into a sysfs file, matches
2034 * str. They must either be the same, or cmd can
2035 * have a trailing newline
2036 */
2037 while (*cmd && *str && *cmd == *str) {
2038 cmd++;
2039 str++;
2040 }
2041 if (*cmd == '\n')
2042 cmd++;
2043 if (*str || *cmd)
2044 return 0;
2045 return 1;
2046 }
2047
2048 struct rdev_sysfs_entry {
2049 struct attribute attr;
2050 ssize_t (*show)(mdk_rdev_t *, char *);
2051 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2052 };
2053
2054 static ssize_t
2055 state_show(mdk_rdev_t *rdev, char *page)
2056 {
2057 char *sep = "";
2058 size_t len = 0;
2059
2060 if (test_bit(Faulty, &rdev->flags)) {
2061 len+= sprintf(page+len, "%sfaulty",sep);
2062 sep = ",";
2063 }
2064 if (test_bit(In_sync, &rdev->flags)) {
2065 len += sprintf(page+len, "%sin_sync",sep);
2066 sep = ",";
2067 }
2068 if (test_bit(WriteMostly, &rdev->flags)) {
2069 len += sprintf(page+len, "%swrite_mostly",sep);
2070 sep = ",";
2071 }
2072 if (test_bit(Blocked, &rdev->flags)) {
2073 len += sprintf(page+len, "%sblocked", sep);
2074 sep = ",";
2075 }
2076 if (!test_bit(Faulty, &rdev->flags) &&
2077 !test_bit(In_sync, &rdev->flags)) {
2078 len += sprintf(page+len, "%sspare", sep);
2079 sep = ",";
2080 }
2081 return len+sprintf(page+len, "\n");
2082 }
2083
2084 static ssize_t
2085 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2086 {
2087 /* can write
2088 * faulty - simulates and error
2089 * remove - disconnects the device
2090 * writemostly - sets write_mostly
2091 * -writemostly - clears write_mostly
2092 * blocked - sets the Blocked flag
2093 * -blocked - clears the Blocked flag
2094 * insync - sets Insync providing device isn't active
2095 */
2096 int err = -EINVAL;
2097 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2098 md_error(rdev->mddev, rdev);
2099 err = 0;
2100 } else if (cmd_match(buf, "remove")) {
2101 if (rdev->raid_disk >= 0)
2102 err = -EBUSY;
2103 else {
2104 mddev_t *mddev = rdev->mddev;
2105 kick_rdev_from_array(rdev);
2106 if (mddev->pers)
2107 md_update_sb(mddev, 1);
2108 md_new_event(mddev);
2109 err = 0;
2110 }
2111 } else if (cmd_match(buf, "writemostly")) {
2112 set_bit(WriteMostly, &rdev->flags);
2113 err = 0;
2114 } else if (cmd_match(buf, "-writemostly")) {
2115 clear_bit(WriteMostly, &rdev->flags);
2116 err = 0;
2117 } else if (cmd_match(buf, "blocked")) {
2118 set_bit(Blocked, &rdev->flags);
2119 err = 0;
2120 } else if (cmd_match(buf, "-blocked")) {
2121 clear_bit(Blocked, &rdev->flags);
2122 wake_up(&rdev->blocked_wait);
2123 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2124 md_wakeup_thread(rdev->mddev->thread);
2125
2126 err = 0;
2127 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2128 set_bit(In_sync, &rdev->flags);
2129 err = 0;
2130 }
2131 if (!err && rdev->sysfs_state)
2132 sysfs_notify_dirent(rdev->sysfs_state);
2133 return err ? err : len;
2134 }
2135 static struct rdev_sysfs_entry rdev_state =
2136 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2137
2138 static ssize_t
2139 errors_show(mdk_rdev_t *rdev, char *page)
2140 {
2141 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2142 }
2143
2144 static ssize_t
2145 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2146 {
2147 char *e;
2148 unsigned long n = simple_strtoul(buf, &e, 10);
2149 if (*buf && (*e == 0 || *e == '\n')) {
2150 atomic_set(&rdev->corrected_errors, n);
2151 return len;
2152 }
2153 return -EINVAL;
2154 }
2155 static struct rdev_sysfs_entry rdev_errors =
2156 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2157
2158 static ssize_t
2159 slot_show(mdk_rdev_t *rdev, char *page)
2160 {
2161 if (rdev->raid_disk < 0)
2162 return sprintf(page, "none\n");
2163 else
2164 return sprintf(page, "%d\n", rdev->raid_disk);
2165 }
2166
2167 static ssize_t
2168 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2169 {
2170 char *e;
2171 int err;
2172 char nm[20];
2173 int slot = simple_strtoul(buf, &e, 10);
2174 if (strncmp(buf, "none", 4)==0)
2175 slot = -1;
2176 else if (e==buf || (*e && *e!= '\n'))
2177 return -EINVAL;
2178 if (rdev->mddev->pers && slot == -1) {
2179 /* Setting 'slot' on an active array requires also
2180 * updating the 'rd%d' link, and communicating
2181 * with the personality with ->hot_*_disk.
2182 * For now we only support removing
2183 * failed/spare devices. This normally happens automatically,
2184 * but not when the metadata is externally managed.
2185 */
2186 if (rdev->raid_disk == -1)
2187 return -EEXIST;
2188 /* personality does all needed checks */
2189 if (rdev->mddev->pers->hot_add_disk == NULL)
2190 return -EINVAL;
2191 err = rdev->mddev->pers->
2192 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2193 if (err)
2194 return err;
2195 sprintf(nm, "rd%d", rdev->raid_disk);
2196 sysfs_remove_link(&rdev->mddev->kobj, nm);
2197 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2198 md_wakeup_thread(rdev->mddev->thread);
2199 } else if (rdev->mddev->pers) {
2200 mdk_rdev_t *rdev2;
2201 /* Activating a spare .. or possibly reactivating
2202 * if we ever get bitmaps working here.
2203 */
2204
2205 if (rdev->raid_disk != -1)
2206 return -EBUSY;
2207
2208 if (rdev->mddev->pers->hot_add_disk == NULL)
2209 return -EINVAL;
2210
2211 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2212 if (rdev2->raid_disk == slot)
2213 return -EEXIST;
2214
2215 rdev->raid_disk = slot;
2216 if (test_bit(In_sync, &rdev->flags))
2217 rdev->saved_raid_disk = slot;
2218 else
2219 rdev->saved_raid_disk = -1;
2220 err = rdev->mddev->pers->
2221 hot_add_disk(rdev->mddev, rdev);
2222 if (err) {
2223 rdev->raid_disk = -1;
2224 return err;
2225 } else
2226 sysfs_notify_dirent(rdev->sysfs_state);
2227 sprintf(nm, "rd%d", rdev->raid_disk);
2228 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2229 printk(KERN_WARNING
2230 "md: cannot register "
2231 "%s for %s\n",
2232 nm, mdname(rdev->mddev));
2233
2234 /* don't wakeup anyone, leave that to userspace. */
2235 } else {
2236 if (slot >= rdev->mddev->raid_disks)
2237 return -ENOSPC;
2238 rdev->raid_disk = slot;
2239 /* assume it is working */
2240 clear_bit(Faulty, &rdev->flags);
2241 clear_bit(WriteMostly, &rdev->flags);
2242 set_bit(In_sync, &rdev->flags);
2243 sysfs_notify_dirent(rdev->sysfs_state);
2244 }
2245 return len;
2246 }
2247
2248
2249 static struct rdev_sysfs_entry rdev_slot =
2250 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2251
2252 static ssize_t
2253 offset_show(mdk_rdev_t *rdev, char *page)
2254 {
2255 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2256 }
2257
2258 static ssize_t
2259 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2260 {
2261 char *e;
2262 unsigned long long offset = simple_strtoull(buf, &e, 10);
2263 if (e==buf || (*e && *e != '\n'))
2264 return -EINVAL;
2265 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2266 return -EBUSY;
2267 if (rdev->sectors && rdev->mddev->external)
2268 /* Must set offset before size, so overlap checks
2269 * can be sane */
2270 return -EBUSY;
2271 rdev->data_offset = offset;
2272 return len;
2273 }
2274
2275 static struct rdev_sysfs_entry rdev_offset =
2276 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2277
2278 static ssize_t
2279 rdev_size_show(mdk_rdev_t *rdev, char *page)
2280 {
2281 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2282 }
2283
2284 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2285 {
2286 /* check if two start/length pairs overlap */
2287 if (s1+l1 <= s2)
2288 return 0;
2289 if (s2+l2 <= s1)
2290 return 0;
2291 return 1;
2292 }
2293
2294 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2295 {
2296 unsigned long long blocks;
2297 sector_t new;
2298
2299 if (strict_strtoull(buf, 10, &blocks) < 0)
2300 return -EINVAL;
2301
2302 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2303 return -EINVAL; /* sector conversion overflow */
2304
2305 new = blocks * 2;
2306 if (new != blocks * 2)
2307 return -EINVAL; /* unsigned long long to sector_t overflow */
2308
2309 *sectors = new;
2310 return 0;
2311 }
2312
2313 static ssize_t
2314 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2315 {
2316 mddev_t *my_mddev = rdev->mddev;
2317 sector_t oldsectors = rdev->sectors;
2318 sector_t sectors;
2319
2320 if (strict_blocks_to_sectors(buf, &sectors) < 0)
2321 return -EINVAL;
2322 if (my_mddev->pers && rdev->raid_disk >= 0) {
2323 if (my_mddev->persistent) {
2324 sectors = super_types[my_mddev->major_version].
2325 rdev_size_change(rdev, sectors);
2326 if (!sectors)
2327 return -EBUSY;
2328 } else if (!sectors)
2329 sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2330 rdev->data_offset;
2331 }
2332 if (sectors < my_mddev->dev_sectors)
2333 return -EINVAL; /* component must fit device */
2334
2335 rdev->sectors = sectors;
2336 if (sectors > oldsectors && my_mddev->external) {
2337 /* need to check that all other rdevs with the same ->bdev
2338 * do not overlap. We need to unlock the mddev to avoid
2339 * a deadlock. We have already changed rdev->sectors, and if
2340 * we have to change it back, we will have the lock again.
2341 */
2342 mddev_t *mddev;
2343 int overlap = 0;
2344 struct list_head *tmp;
2345
2346 mddev_unlock(my_mddev);
2347 for_each_mddev(mddev, tmp) {
2348 mdk_rdev_t *rdev2;
2349
2350 mddev_lock(mddev);
2351 list_for_each_entry(rdev2, &mddev->disks, same_set)
2352 if (test_bit(AllReserved, &rdev2->flags) ||
2353 (rdev->bdev == rdev2->bdev &&
2354 rdev != rdev2 &&
2355 overlaps(rdev->data_offset, rdev->sectors,
2356 rdev2->data_offset,
2357 rdev2->sectors))) {
2358 overlap = 1;
2359 break;
2360 }
2361 mddev_unlock(mddev);
2362 if (overlap) {
2363 mddev_put(mddev);
2364 break;
2365 }
2366 }
2367 mddev_lock(my_mddev);
2368 if (overlap) {
2369 /* Someone else could have slipped in a size
2370 * change here, but doing so is just silly.
2371 * We put oldsectors back because we *know* it is
2372 * safe, and trust userspace not to race with
2373 * itself
2374 */
2375 rdev->sectors = oldsectors;
2376 return -EBUSY;
2377 }
2378 }
2379 return len;
2380 }
2381
2382 static struct rdev_sysfs_entry rdev_size =
2383 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2384
2385 static struct attribute *rdev_default_attrs[] = {
2386 &rdev_state.attr,
2387 &rdev_errors.attr,
2388 &rdev_slot.attr,
2389 &rdev_offset.attr,
2390 &rdev_size.attr,
2391 NULL,
2392 };
2393 static ssize_t
2394 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2395 {
2396 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2397 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2398 mddev_t *mddev = rdev->mddev;
2399 ssize_t rv;
2400
2401 if (!entry->show)
2402 return -EIO;
2403
2404 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2405 if (!rv) {
2406 if (rdev->mddev == NULL)
2407 rv = -EBUSY;
2408 else
2409 rv = entry->show(rdev, page);
2410 mddev_unlock(mddev);
2411 }
2412 return rv;
2413 }
2414
2415 static ssize_t
2416 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2417 const char *page, size_t length)
2418 {
2419 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2420 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2421 ssize_t rv;
2422 mddev_t *mddev = rdev->mddev;
2423
2424 if (!entry->store)
2425 return -EIO;
2426 if (!capable(CAP_SYS_ADMIN))
2427 return -EACCES;
2428 rv = mddev ? mddev_lock(mddev): -EBUSY;
2429 if (!rv) {
2430 if (rdev->mddev == NULL)
2431 rv = -EBUSY;
2432 else
2433 rv = entry->store(rdev, page, length);
2434 mddev_unlock(mddev);
2435 }
2436 return rv;
2437 }
2438
2439 static void rdev_free(struct kobject *ko)
2440 {
2441 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2442 kfree(rdev);
2443 }
2444 static struct sysfs_ops rdev_sysfs_ops = {
2445 .show = rdev_attr_show,
2446 .store = rdev_attr_store,
2447 };
2448 static struct kobj_type rdev_ktype = {
2449 .release = rdev_free,
2450 .sysfs_ops = &rdev_sysfs_ops,
2451 .default_attrs = rdev_default_attrs,
2452 };
2453
2454 /*
2455 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2456 *
2457 * mark the device faulty if:
2458 *
2459 * - the device is nonexistent (zero size)
2460 * - the device has no valid superblock
2461 *
2462 * a faulty rdev _never_ has rdev->sb set.
2463 */
2464 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2465 {
2466 char b[BDEVNAME_SIZE];
2467 int err;
2468 mdk_rdev_t *rdev;
2469 sector_t size;
2470
2471 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2472 if (!rdev) {
2473 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2474 return ERR_PTR(-ENOMEM);
2475 }
2476
2477 if ((err = alloc_disk_sb(rdev)))
2478 goto abort_free;
2479
2480 err = lock_rdev(rdev, newdev, super_format == -2);
2481 if (err)
2482 goto abort_free;
2483
2484 kobject_init(&rdev->kobj, &rdev_ktype);
2485
2486 rdev->desc_nr = -1;
2487 rdev->saved_raid_disk = -1;
2488 rdev->raid_disk = -1;
2489 rdev->flags = 0;
2490 rdev->data_offset = 0;
2491 rdev->sb_events = 0;
2492 atomic_set(&rdev->nr_pending, 0);
2493 atomic_set(&rdev->read_errors, 0);
2494 atomic_set(&rdev->corrected_errors, 0);
2495
2496 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2497 if (!size) {
2498 printk(KERN_WARNING
2499 "md: %s has zero or unknown size, marking faulty!\n",
2500 bdevname(rdev->bdev,b));
2501 err = -EINVAL;
2502 goto abort_free;
2503 }
2504
2505 if (super_format >= 0) {
2506 err = super_types[super_format].
2507 load_super(rdev, NULL, super_minor);
2508 if (err == -EINVAL) {
2509 printk(KERN_WARNING
2510 "md: %s does not have a valid v%d.%d "
2511 "superblock, not importing!\n",
2512 bdevname(rdev->bdev,b),
2513 super_format, super_minor);
2514 goto abort_free;
2515 }
2516 if (err < 0) {
2517 printk(KERN_WARNING
2518 "md: could not read %s's sb, not importing!\n",
2519 bdevname(rdev->bdev,b));
2520 goto abort_free;
2521 }
2522 }
2523
2524 INIT_LIST_HEAD(&rdev->same_set);
2525 init_waitqueue_head(&rdev->blocked_wait);
2526
2527 return rdev;
2528
2529 abort_free:
2530 if (rdev->sb_page) {
2531 if (rdev->bdev)
2532 unlock_rdev(rdev);
2533 free_disk_sb(rdev);
2534 }
2535 kfree(rdev);
2536 return ERR_PTR(err);
2537 }
2538
2539 /*
2540 * Check a full RAID array for plausibility
2541 */
2542
2543
2544 static void analyze_sbs(mddev_t * mddev)
2545 {
2546 int i;
2547 mdk_rdev_t *rdev, *freshest, *tmp;
2548 char b[BDEVNAME_SIZE];
2549
2550 freshest = NULL;
2551 rdev_for_each(rdev, tmp, mddev)
2552 switch (super_types[mddev->major_version].
2553 load_super(rdev, freshest, mddev->minor_version)) {
2554 case 1:
2555 freshest = rdev;
2556 break;
2557 case 0:
2558 break;
2559 default:
2560 printk( KERN_ERR \
2561 "md: fatal superblock inconsistency in %s"
2562 " -- removing from array\n",
2563 bdevname(rdev->bdev,b));
2564 kick_rdev_from_array(rdev);
2565 }
2566
2567
2568 super_types[mddev->major_version].
2569 validate_super(mddev, freshest);
2570
2571 i = 0;
2572 rdev_for_each(rdev, tmp, mddev) {
2573 if (rdev->desc_nr >= mddev->max_disks ||
2574 i > mddev->max_disks) {
2575 printk(KERN_WARNING
2576 "md: %s: %s: only %d devices permitted\n",
2577 mdname(mddev), bdevname(rdev->bdev, b),
2578 mddev->max_disks);
2579 kick_rdev_from_array(rdev);
2580 continue;
2581 }
2582 if (rdev != freshest)
2583 if (super_types[mddev->major_version].
2584 validate_super(mddev, rdev)) {
2585 printk(KERN_WARNING "md: kicking non-fresh %s"
2586 " from array!\n",
2587 bdevname(rdev->bdev,b));
2588 kick_rdev_from_array(rdev);
2589 continue;
2590 }
2591 if (mddev->level == LEVEL_MULTIPATH) {
2592 rdev->desc_nr = i++;
2593 rdev->raid_disk = rdev->desc_nr;
2594 set_bit(In_sync, &rdev->flags);
2595 } else if (rdev->raid_disk >= mddev->raid_disks) {
2596 rdev->raid_disk = -1;
2597 clear_bit(In_sync, &rdev->flags);
2598 }
2599 }
2600
2601
2602
2603 if (mddev->recovery_cp != MaxSector &&
2604 mddev->level >= 1)
2605 printk(KERN_ERR "md: %s: raid array is not clean"
2606 " -- starting background reconstruction\n",
2607 mdname(mddev));
2608
2609 }
2610
2611 static void md_safemode_timeout(unsigned long data);
2612
2613 static ssize_t
2614 safe_delay_show(mddev_t *mddev, char *page)
2615 {
2616 int msec = (mddev->safemode_delay*1000)/HZ;
2617 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2618 }
2619 static ssize_t
2620 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2621 {
2622 int scale=1;
2623 int dot=0;
2624 int i;
2625 unsigned long msec;
2626 char buf[30];
2627
2628 /* remove a period, and count digits after it */
2629 if (len >= sizeof(buf))
2630 return -EINVAL;
2631 strlcpy(buf, cbuf, sizeof(buf));
2632 for (i=0; i<len; i++) {
2633 if (dot) {
2634 if (isdigit(buf[i])) {
2635 buf[i-1] = buf[i];
2636 scale *= 10;
2637 }
2638 buf[i] = 0;
2639 } else if (buf[i] == '.') {
2640 dot=1;
2641 buf[i] = 0;
2642 }
2643 }
2644 if (strict_strtoul(buf, 10, &msec) < 0)
2645 return -EINVAL;
2646 msec = (msec * 1000) / scale;
2647 if (msec == 0)
2648 mddev->safemode_delay = 0;
2649 else {
2650 unsigned long old_delay = mddev->safemode_delay;
2651 mddev->safemode_delay = (msec*HZ)/1000;
2652 if (mddev->safemode_delay == 0)
2653 mddev->safemode_delay = 1;
2654 if (mddev->safemode_delay < old_delay)
2655 md_safemode_timeout((unsigned long)mddev);
2656 }
2657 return len;
2658 }
2659 static struct md_sysfs_entry md_safe_delay =
2660 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2661
2662 static ssize_t
2663 level_show(mddev_t *mddev, char *page)
2664 {
2665 struct mdk_personality *p = mddev->pers;
2666 if (p)
2667 return sprintf(page, "%s\n", p->name);
2668 else if (mddev->clevel[0])
2669 return sprintf(page, "%s\n", mddev->clevel);
2670 else if (mddev->level != LEVEL_NONE)
2671 return sprintf(page, "%d\n", mddev->level);
2672 else
2673 return 0;
2674 }
2675
2676 static ssize_t
2677 level_store(mddev_t *mddev, const char *buf, size_t len)
2678 {
2679 char level[16];
2680 ssize_t rv = len;
2681 struct mdk_personality *pers;
2682 void *priv;
2683 mdk_rdev_t *rdev;
2684
2685 if (mddev->pers == NULL) {
2686 if (len == 0)
2687 return 0;
2688 if (len >= sizeof(mddev->clevel))
2689 return -ENOSPC;
2690 strncpy(mddev->clevel, buf, len);
2691 if (mddev->clevel[len-1] == '\n')
2692 len--;
2693 mddev->clevel[len] = 0;
2694 mddev->level = LEVEL_NONE;
2695 return rv;
2696 }
2697
2698 /* request to change the personality. Need to ensure:
2699 * - array is not engaged in resync/recovery/reshape
2700 * - old personality can be suspended
2701 * - new personality will access other array.
2702 */
2703
2704 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
2705 return -EBUSY;
2706
2707 if (!mddev->pers->quiesce) {
2708 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
2709 mdname(mddev), mddev->pers->name);
2710 return -EINVAL;
2711 }
2712
2713 /* Now find the new personality */
2714 if (len == 0 || len >= sizeof(level))
2715 return -EINVAL;
2716 strncpy(level, buf, len);
2717 if (level[len-1] == '\n')
2718 len--;
2719 level[len] = 0;
2720
2721 request_module("md-%s", level);
2722 spin_lock(&pers_lock);
2723 pers = find_pers(LEVEL_NONE, level);
2724 if (!pers || !try_module_get(pers->owner)) {
2725 spin_unlock(&pers_lock);
2726 printk(KERN_WARNING "md: personality %s not loaded\n", level);
2727 return -EINVAL;
2728 }
2729 spin_unlock(&pers_lock);
2730
2731 if (pers == mddev->pers) {
2732 /* Nothing to do! */
2733 module_put(pers->owner);
2734 return rv;
2735 }
2736 if (!pers->takeover) {
2737 module_put(pers->owner);
2738 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
2739 mdname(mddev), level);
2740 return -EINVAL;
2741 }
2742
2743 /* ->takeover must set new_* and/or delta_disks
2744 * if it succeeds, and may set them when it fails.
2745 */
2746 priv = pers->takeover(mddev);
2747 if (IS_ERR(priv)) {
2748 mddev->new_level = mddev->level;
2749 mddev->new_layout = mddev->layout;
2750 mddev->new_chunk = mddev->chunk_size;
2751 mddev->raid_disks -= mddev->delta_disks;
2752 mddev->delta_disks = 0;
2753 module_put(pers->owner);
2754 printk(KERN_WARNING "md: %s: %s would not accept array\n",
2755 mdname(mddev), level);
2756 return PTR_ERR(priv);
2757 }
2758
2759 /* Looks like we have a winner */
2760 mddev_suspend(mddev);
2761 mddev->pers->stop(mddev);
2762 module_put(mddev->pers->owner);
2763 /* Invalidate devices that are now superfluous */
2764 list_for_each_entry(rdev, &mddev->disks, same_set)
2765 if (rdev->raid_disk >= mddev->raid_disks) {
2766 rdev->raid_disk = -1;
2767 clear_bit(In_sync, &rdev->flags);
2768 }
2769 mddev->pers = pers;
2770 mddev->private = priv;
2771 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
2772 mddev->level = mddev->new_level;
2773 mddev->layout = mddev->new_layout;
2774 mddev->chunk_size = mddev->new_chunk;
2775 mddev->delta_disks = 0;
2776 pers->run(mddev);
2777 mddev_resume(mddev);
2778 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2779 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2780 md_wakeup_thread(mddev->thread);
2781 return rv;
2782 }
2783
2784 static struct md_sysfs_entry md_level =
2785 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2786
2787
2788 static ssize_t
2789 layout_show(mddev_t *mddev, char *page)
2790 {
2791 /* just a number, not meaningful for all levels */
2792 if (mddev->reshape_position != MaxSector &&
2793 mddev->layout != mddev->new_layout)
2794 return sprintf(page, "%d (%d)\n",
2795 mddev->new_layout, mddev->layout);
2796 return sprintf(page, "%d\n", mddev->layout);
2797 }
2798
2799 static ssize_t
2800 layout_store(mddev_t *mddev, const char *buf, size_t len)
2801 {
2802 char *e;
2803 unsigned long n = simple_strtoul(buf, &e, 10);
2804
2805 if (!*buf || (*e && *e != '\n'))
2806 return -EINVAL;
2807
2808 if (mddev->pers) {
2809 int err;
2810 if (mddev->pers->reconfig == NULL)
2811 return -EBUSY;
2812 err = mddev->pers->reconfig(mddev, n, -1);
2813 if (err)
2814 return err;
2815 } else {
2816 mddev->new_layout = n;
2817 if (mddev->reshape_position == MaxSector)
2818 mddev->layout = n;
2819 }
2820 return len;
2821 }
2822 static struct md_sysfs_entry md_layout =
2823 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2824
2825
2826 static ssize_t
2827 raid_disks_show(mddev_t *mddev, char *page)
2828 {
2829 if (mddev->raid_disks == 0)
2830 return 0;
2831 if (mddev->reshape_position != MaxSector &&
2832 mddev->delta_disks != 0)
2833 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2834 mddev->raid_disks - mddev->delta_disks);
2835 return sprintf(page, "%d\n", mddev->raid_disks);
2836 }
2837
2838 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2839
2840 static ssize_t
2841 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2842 {
2843 char *e;
2844 int rv = 0;
2845 unsigned long n = simple_strtoul(buf, &e, 10);
2846
2847 if (!*buf || (*e && *e != '\n'))
2848 return -EINVAL;
2849
2850 if (mddev->pers)
2851 rv = update_raid_disks(mddev, n);
2852 else if (mddev->reshape_position != MaxSector) {
2853 int olddisks = mddev->raid_disks - mddev->delta_disks;
2854 mddev->delta_disks = n - olddisks;
2855 mddev->raid_disks = n;
2856 } else
2857 mddev->raid_disks = n;
2858 return rv ? rv : len;
2859 }
2860 static struct md_sysfs_entry md_raid_disks =
2861 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2862
2863 static ssize_t
2864 chunk_size_show(mddev_t *mddev, char *page)
2865 {
2866 if (mddev->reshape_position != MaxSector &&
2867 mddev->chunk_size != mddev->new_chunk)
2868 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2869 mddev->chunk_size);
2870 return sprintf(page, "%d\n", mddev->chunk_size);
2871 }
2872
2873 static ssize_t
2874 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2875 {
2876 char *e;
2877 unsigned long n = simple_strtoul(buf, &e, 10);
2878
2879 if (!*buf || (*e && *e != '\n'))
2880 return -EINVAL;
2881
2882 if (mddev->pers) {
2883 int err;
2884 if (mddev->pers->reconfig == NULL)
2885 return -EBUSY;
2886 err = mddev->pers->reconfig(mddev, -1, n);
2887 if (err)
2888 return err;
2889 } else {
2890 mddev->new_chunk = n;
2891 if (mddev->reshape_position == MaxSector)
2892 mddev->chunk_size = n;
2893 }
2894 return len;
2895 }
2896 static struct md_sysfs_entry md_chunk_size =
2897 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2898
2899 static ssize_t
2900 resync_start_show(mddev_t *mddev, char *page)
2901 {
2902 if (mddev->recovery_cp == MaxSector)
2903 return sprintf(page, "none\n");
2904 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2905 }
2906
2907 static ssize_t
2908 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2909 {
2910 char *e;
2911 unsigned long long n = simple_strtoull(buf, &e, 10);
2912
2913 if (mddev->pers)
2914 return -EBUSY;
2915 if (!*buf || (*e && *e != '\n'))
2916 return -EINVAL;
2917
2918 mddev->recovery_cp = n;
2919 return len;
2920 }
2921 static struct md_sysfs_entry md_resync_start =
2922 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2923
2924 /*
2925 * The array state can be:
2926 *
2927 * clear
2928 * No devices, no size, no level
2929 * Equivalent to STOP_ARRAY ioctl
2930 * inactive
2931 * May have some settings, but array is not active
2932 * all IO results in error
2933 * When written, doesn't tear down array, but just stops it
2934 * suspended (not supported yet)
2935 * All IO requests will block. The array can be reconfigured.
2936 * Writing this, if accepted, will block until array is quiescent
2937 * readonly
2938 * no resync can happen. no superblocks get written.
2939 * write requests fail
2940 * read-auto
2941 * like readonly, but behaves like 'clean' on a write request.
2942 *
2943 * clean - no pending writes, but otherwise active.
2944 * When written to inactive array, starts without resync
2945 * If a write request arrives then
2946 * if metadata is known, mark 'dirty' and switch to 'active'.
2947 * if not known, block and switch to write-pending
2948 * If written to an active array that has pending writes, then fails.
2949 * active
2950 * fully active: IO and resync can be happening.
2951 * When written to inactive array, starts with resync
2952 *
2953 * write-pending
2954 * clean, but writes are blocked waiting for 'active' to be written.
2955 *
2956 * active-idle
2957 * like active, but no writes have been seen for a while (100msec).
2958 *
2959 */
2960 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2961 write_pending, active_idle, bad_word};
2962 static char *array_states[] = {
2963 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2964 "write-pending", "active-idle", NULL };
2965
2966 static int match_word(const char *word, char **list)
2967 {
2968 int n;
2969 for (n=0; list[n]; n++)
2970 if (cmd_match(word, list[n]))
2971 break;
2972 return n;
2973 }
2974
2975 static ssize_t
2976 array_state_show(mddev_t *mddev, char *page)
2977 {
2978 enum array_state st = inactive;
2979
2980 if (mddev->pers)
2981 switch(mddev->ro) {
2982 case 1:
2983 st = readonly;
2984 break;
2985 case 2:
2986 st = read_auto;
2987 break;
2988 case 0:
2989 if (mddev->in_sync)
2990 st = clean;
2991 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2992 st = write_pending;
2993 else if (mddev->safemode)
2994 st = active_idle;
2995 else
2996 st = active;
2997 }
2998 else {
2999 if (list_empty(&mddev->disks) &&
3000 mddev->raid_disks == 0 &&
3001 mddev->dev_sectors == 0)
3002 st = clear;
3003 else
3004 st = inactive;
3005 }
3006 return sprintf(page, "%s\n", array_states[st]);
3007 }
3008
3009 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3010 static int do_md_run(mddev_t * mddev);
3011 static int restart_array(mddev_t *mddev);
3012
3013 static ssize_t
3014 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3015 {
3016 int err = -EINVAL;
3017 enum array_state st = match_word(buf, array_states);
3018 switch(st) {
3019 case bad_word:
3020 break;
3021 case clear:
3022 /* stopping an active array */
3023 if (atomic_read(&mddev->openers) > 0)
3024 return -EBUSY;
3025 err = do_md_stop(mddev, 0, 0);
3026 break;
3027 case inactive:
3028 /* stopping an active array */
3029 if (mddev->pers) {
3030 if (atomic_read(&mddev->openers) > 0)
3031 return -EBUSY;
3032 err = do_md_stop(mddev, 2, 0);
3033 } else
3034 err = 0; /* already inactive */
3035 break;
3036 case suspended:
3037 break; /* not supported yet */
3038 case readonly:
3039 if (mddev->pers)
3040 err = do_md_stop(mddev, 1, 0);
3041 else {
3042 mddev->ro = 1;
3043 set_disk_ro(mddev->gendisk, 1);
3044 err = do_md_run(mddev);
3045 }
3046 break;
3047 case read_auto:
3048 if (mddev->pers) {
3049 if (mddev->ro == 0)
3050 err = do_md_stop(mddev, 1, 0);
3051 else if (mddev->ro == 1)
3052 err = restart_array(mddev);
3053 if (err == 0) {
3054 mddev->ro = 2;
3055 set_disk_ro(mddev->gendisk, 0);
3056 }
3057 } else {
3058 mddev->ro = 2;
3059 err = do_md_run(mddev);
3060 }
3061 break;
3062 case clean:
3063 if (mddev->pers) {
3064 restart_array(mddev);
3065 spin_lock_irq(&mddev->write_lock);
3066 if (atomic_read(&mddev->writes_pending) == 0) {
3067 if (mddev->in_sync == 0) {
3068 mddev->in_sync = 1;
3069 if (mddev->safemode == 1)
3070 mddev->safemode = 0;
3071 if (mddev->persistent)
3072 set_bit(MD_CHANGE_CLEAN,
3073 &mddev->flags);
3074 }
3075 err = 0;
3076 } else
3077 err = -EBUSY;
3078 spin_unlock_irq(&mddev->write_lock);
3079 } else
3080 err = -EINVAL;
3081 break;
3082 case active:
3083 if (mddev->pers) {
3084 restart_array(mddev);
3085 if (mddev->external)
3086 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3087 wake_up(&mddev->sb_wait);
3088 err = 0;
3089 } else {
3090 mddev->ro = 0;
3091 set_disk_ro(mddev->gendisk, 0);
3092 err = do_md_run(mddev);
3093 }
3094 break;
3095 case write_pending:
3096 case active_idle:
3097 /* these cannot be set */
3098 break;
3099 }
3100 if (err)
3101 return err;
3102 else {
3103 sysfs_notify_dirent(mddev->sysfs_state);
3104 return len;
3105 }
3106 }
3107 static struct md_sysfs_entry md_array_state =
3108 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3109
3110 static ssize_t
3111 null_show(mddev_t *mddev, char *page)
3112 {
3113 return -EINVAL;
3114 }
3115
3116 static ssize_t
3117 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3118 {
3119 /* buf must be %d:%d\n? giving major and minor numbers */
3120 /* The new device is added to the array.
3121 * If the array has a persistent superblock, we read the
3122 * superblock to initialise info and check validity.
3123 * Otherwise, only checking done is that in bind_rdev_to_array,
3124 * which mainly checks size.
3125 */
3126 char *e;
3127 int major = simple_strtoul(buf, &e, 10);
3128 int minor;
3129 dev_t dev;
3130 mdk_rdev_t *rdev;
3131 int err;
3132
3133 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3134 return -EINVAL;
3135 minor = simple_strtoul(e+1, &e, 10);
3136 if (*e && *e != '\n')
3137 return -EINVAL;
3138 dev = MKDEV(major, minor);
3139 if (major != MAJOR(dev) ||
3140 minor != MINOR(dev))
3141 return -EOVERFLOW;
3142
3143
3144 if (mddev->persistent) {
3145 rdev = md_import_device(dev, mddev->major_version,
3146 mddev->minor_version);
3147 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3148 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3149 mdk_rdev_t, same_set);
3150 err = super_types[mddev->major_version]
3151 .load_super(rdev, rdev0, mddev->minor_version);
3152 if (err < 0)
3153 goto out;
3154 }
3155 } else if (mddev->external)
3156 rdev = md_import_device(dev, -2, -1);
3157 else
3158 rdev = md_import_device(dev, -1, -1);
3159
3160 if (IS_ERR(rdev))
3161 return PTR_ERR(rdev);
3162 err = bind_rdev_to_array(rdev, mddev);
3163 out:
3164 if (err)
3165 export_rdev(rdev);
3166 return err ? err : len;
3167 }
3168
3169 static struct md_sysfs_entry md_new_device =
3170 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3171
3172 static ssize_t
3173 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3174 {
3175 char *end;
3176 unsigned long chunk, end_chunk;
3177
3178 if (!mddev->bitmap)
3179 goto out;
3180 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3181 while (*buf) {
3182 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3183 if (buf == end) break;
3184 if (*end == '-') { /* range */
3185 buf = end + 1;
3186 end_chunk = simple_strtoul(buf, &end, 0);
3187 if (buf == end) break;
3188 }
3189 if (*end && !isspace(*end)) break;
3190 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3191 buf = end;
3192 while (isspace(*buf)) buf++;
3193 }
3194 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3195 out:
3196 return len;
3197 }
3198
3199 static struct md_sysfs_entry md_bitmap =
3200 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3201
3202 static ssize_t
3203 size_show(mddev_t *mddev, char *page)
3204 {
3205 return sprintf(page, "%llu\n",
3206 (unsigned long long)mddev->dev_sectors / 2);
3207 }
3208
3209 static int update_size(mddev_t *mddev, sector_t num_sectors);
3210
3211 static ssize_t
3212 size_store(mddev_t *mddev, const char *buf, size_t len)
3213 {
3214 /* If array is inactive, we can reduce the component size, but
3215 * not increase it (except from 0).
3216 * If array is active, we can try an on-line resize
3217 */
3218 sector_t sectors;
3219 int err = strict_blocks_to_sectors(buf, &sectors);
3220
3221 if (err < 0)
3222 return err;
3223 if (mddev->pers) {
3224 err = update_size(mddev, sectors);
3225 md_update_sb(mddev, 1);
3226 } else {
3227 if (mddev->dev_sectors == 0 ||
3228 mddev->dev_sectors > sectors)
3229 mddev->dev_sectors = sectors;
3230 else
3231 err = -ENOSPC;
3232 }
3233 return err ? err : len;
3234 }
3235
3236 static struct md_sysfs_entry md_size =
3237 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3238
3239
3240 /* Metdata version.
3241 * This is one of
3242 * 'none' for arrays with no metadata (good luck...)
3243 * 'external' for arrays with externally managed metadata,
3244 * or N.M for internally known formats
3245 */
3246 static ssize_t
3247 metadata_show(mddev_t *mddev, char *page)
3248 {
3249 if (mddev->persistent)
3250 return sprintf(page, "%d.%d\n",
3251 mddev->major_version, mddev->minor_version);
3252 else if (mddev->external)
3253 return sprintf(page, "external:%s\n", mddev->metadata_type);
3254 else
3255 return sprintf(page, "none\n");
3256 }
3257
3258 static ssize_t
3259 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3260 {
3261 int major, minor;
3262 char *e;
3263 /* Changing the details of 'external' metadata is
3264 * always permitted. Otherwise there must be
3265 * no devices attached to the array.
3266 */
3267 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3268 ;
3269 else if (!list_empty(&mddev->disks))
3270 return -EBUSY;
3271
3272 if (cmd_match(buf, "none")) {
3273 mddev->persistent = 0;
3274 mddev->external = 0;
3275 mddev->major_version = 0;
3276 mddev->minor_version = 90;
3277 return len;
3278 }
3279 if (strncmp(buf, "external:", 9) == 0) {
3280 size_t namelen = len-9;
3281 if (namelen >= sizeof(mddev->metadata_type))
3282 namelen = sizeof(mddev->metadata_type)-1;
3283 strncpy(mddev->metadata_type, buf+9, namelen);
3284 mddev->metadata_type[namelen] = 0;
3285 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3286 mddev->metadata_type[--namelen] = 0;
3287 mddev->persistent = 0;
3288 mddev->external = 1;
3289 mddev->major_version = 0;
3290 mddev->minor_version = 90;
3291 return len;
3292 }
3293 major = simple_strtoul(buf, &e, 10);
3294 if (e==buf || *e != '.')
3295 return -EINVAL;
3296 buf = e+1;
3297 minor = simple_strtoul(buf, &e, 10);
3298 if (e==buf || (*e && *e != '\n') )
3299 return -EINVAL;
3300 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3301 return -ENOENT;
3302 mddev->major_version = major;
3303 mddev->minor_version = minor;
3304 mddev->persistent = 1;
3305 mddev->external = 0;
3306 return len;
3307 }
3308
3309 static struct md_sysfs_entry md_metadata =
3310 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3311
3312 static ssize_t
3313 action_show(mddev_t *mddev, char *page)
3314 {
3315 char *type = "idle";
3316 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3317 type = "frozen";
3318 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3319 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3320 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3321 type = "reshape";
3322 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3323 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3324 type = "resync";
3325 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3326 type = "check";
3327 else
3328 type = "repair";
3329 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3330 type = "recover";
3331 }
3332 return sprintf(page, "%s\n", type);
3333 }
3334
3335 static ssize_t
3336 action_store(mddev_t *mddev, const char *page, size_t len)
3337 {
3338 if (!mddev->pers || !mddev->pers->sync_request)
3339 return -EINVAL;
3340
3341 if (cmd_match(page, "frozen"))
3342 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3343 else
3344 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3345
3346 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3347 if (mddev->sync_thread) {
3348 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3349 md_unregister_thread(mddev->sync_thread);
3350 mddev->sync_thread = NULL;
3351 mddev->recovery = 0;
3352 }
3353 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3354 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3355 return -EBUSY;
3356 else if (cmd_match(page, "resync"))
3357 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3358 else if (cmd_match(page, "recover")) {
3359 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3360 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3361 } else if (cmd_match(page, "reshape")) {
3362 int err;
3363 if (mddev->pers->start_reshape == NULL)
3364 return -EINVAL;
3365 err = mddev->pers->start_reshape(mddev);
3366 if (err)
3367 return err;
3368 sysfs_notify(&mddev->kobj, NULL, "degraded");
3369 } else {
3370 if (cmd_match(page, "check"))
3371 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3372 else if (!cmd_match(page, "repair"))
3373 return -EINVAL;
3374 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3375 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3376 }
3377 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3378 md_wakeup_thread(mddev->thread);
3379 sysfs_notify_dirent(mddev->sysfs_action);
3380 return len;
3381 }
3382
3383 static ssize_t
3384 mismatch_cnt_show(mddev_t *mddev, char *page)
3385 {
3386 return sprintf(page, "%llu\n",
3387 (unsigned long long) mddev->resync_mismatches);
3388 }
3389
3390 static struct md_sysfs_entry md_scan_mode =
3391 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3392
3393
3394 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3395
3396 static ssize_t
3397 sync_min_show(mddev_t *mddev, char *page)
3398 {
3399 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3400 mddev->sync_speed_min ? "local": "system");
3401 }
3402
3403 static ssize_t
3404 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3405 {
3406 int min;
3407 char *e;
3408 if (strncmp(buf, "system", 6)==0) {
3409 mddev->sync_speed_min = 0;
3410 return len;
3411 }
3412 min = simple_strtoul(buf, &e, 10);
3413 if (buf == e || (*e && *e != '\n') || min <= 0)
3414 return -EINVAL;
3415 mddev->sync_speed_min = min;
3416 return len;
3417 }
3418
3419 static struct md_sysfs_entry md_sync_min =
3420 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3421
3422 static ssize_t
3423 sync_max_show(mddev_t *mddev, char *page)
3424 {
3425 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3426 mddev->sync_speed_max ? "local": "system");
3427 }
3428
3429 static ssize_t
3430 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3431 {
3432 int max;
3433 char *e;
3434 if (strncmp(buf, "system", 6)==0) {
3435 mddev->sync_speed_max = 0;
3436 return len;
3437 }
3438 max = simple_strtoul(buf, &e, 10);
3439 if (buf == e || (*e && *e != '\n') || max <= 0)
3440 return -EINVAL;
3441 mddev->sync_speed_max = max;
3442 return len;
3443 }
3444
3445 static struct md_sysfs_entry md_sync_max =
3446 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3447
3448 static ssize_t
3449 degraded_show(mddev_t *mddev, char *page)
3450 {
3451 return sprintf(page, "%d\n", mddev->degraded);
3452 }
3453 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3454
3455 static ssize_t
3456 sync_force_parallel_show(mddev_t *mddev, char *page)
3457 {
3458 return sprintf(page, "%d\n", mddev->parallel_resync);
3459 }
3460
3461 static ssize_t
3462 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3463 {
3464 long n;
3465
3466 if (strict_strtol(buf, 10, &n))
3467 return -EINVAL;
3468
3469 if (n != 0 && n != 1)
3470 return -EINVAL;
3471
3472 mddev->parallel_resync = n;
3473
3474 if (mddev->sync_thread)
3475 wake_up(&resync_wait);
3476
3477 return len;
3478 }
3479
3480 /* force parallel resync, even with shared block devices */
3481 static struct md_sysfs_entry md_sync_force_parallel =
3482 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3483 sync_force_parallel_show, sync_force_parallel_store);
3484
3485 static ssize_t
3486 sync_speed_show(mddev_t *mddev, char *page)
3487 {
3488 unsigned long resync, dt, db;
3489 if (mddev->curr_resync == 0)
3490 return sprintf(page, "none\n");
3491 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3492 dt = (jiffies - mddev->resync_mark) / HZ;
3493 if (!dt) dt++;
3494 db = resync - mddev->resync_mark_cnt;
3495 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3496 }
3497
3498 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3499
3500 static ssize_t
3501 sync_completed_show(mddev_t *mddev, char *page)
3502 {
3503 unsigned long max_sectors, resync;
3504
3505 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3506 return sprintf(page, "none\n");
3507
3508 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3509 max_sectors = mddev->resync_max_sectors;
3510 else
3511 max_sectors = mddev->dev_sectors;
3512
3513 resync = mddev->curr_resync_completed;
3514 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3515 }
3516
3517 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3518
3519 static ssize_t
3520 min_sync_show(mddev_t *mddev, char *page)
3521 {
3522 return sprintf(page, "%llu\n",
3523 (unsigned long long)mddev->resync_min);
3524 }
3525 static ssize_t
3526 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3527 {
3528 unsigned long long min;
3529 if (strict_strtoull(buf, 10, &min))
3530 return -EINVAL;
3531 if (min > mddev->resync_max)
3532 return -EINVAL;
3533 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3534 return -EBUSY;
3535
3536 /* Must be a multiple of chunk_size */
3537 if (mddev->chunk_size) {
3538 if (min & (sector_t)((mddev->chunk_size>>9)-1))
3539 return -EINVAL;
3540 }
3541 mddev->resync_min = min;
3542
3543 return len;
3544 }
3545
3546 static struct md_sysfs_entry md_min_sync =
3547 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3548
3549 static ssize_t
3550 max_sync_show(mddev_t *mddev, char *page)
3551 {
3552 if (mddev->resync_max == MaxSector)
3553 return sprintf(page, "max\n");
3554 else
3555 return sprintf(page, "%llu\n",
3556 (unsigned long long)mddev->resync_max);
3557 }
3558 static ssize_t
3559 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3560 {
3561 if (strncmp(buf, "max", 3) == 0)
3562 mddev->resync_max = MaxSector;
3563 else {
3564 unsigned long long max;
3565 if (strict_strtoull(buf, 10, &max))
3566 return -EINVAL;
3567 if (max < mddev->resync_min)
3568 return -EINVAL;
3569 if (max < mddev->resync_max &&
3570 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3571 return -EBUSY;
3572
3573 /* Must be a multiple of chunk_size */
3574 if (mddev->chunk_size) {
3575 if (max & (sector_t)((mddev->chunk_size>>9)-1))
3576 return -EINVAL;
3577 }
3578 mddev->resync_max = max;
3579 }
3580 wake_up(&mddev->recovery_wait);
3581 return len;
3582 }
3583
3584 static struct md_sysfs_entry md_max_sync =
3585 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3586
3587 static ssize_t
3588 suspend_lo_show(mddev_t *mddev, char *page)
3589 {
3590 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3591 }
3592
3593 static ssize_t
3594 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3595 {
3596 char *e;
3597 unsigned long long new = simple_strtoull(buf, &e, 10);
3598
3599 if (mddev->pers == NULL ||
3600 mddev->pers->quiesce == NULL)
3601 return -EINVAL;
3602 if (buf == e || (*e && *e != '\n'))
3603 return -EINVAL;
3604 if (new >= mddev->suspend_hi ||
3605 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3606 mddev->suspend_lo = new;
3607 mddev->pers->quiesce(mddev, 2);
3608 return len;
3609 } else
3610 return -EINVAL;
3611 }
3612 static struct md_sysfs_entry md_suspend_lo =
3613 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3614
3615
3616 static ssize_t
3617 suspend_hi_show(mddev_t *mddev, char *page)
3618 {
3619 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3620 }
3621
3622 static ssize_t
3623 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3624 {
3625 char *e;
3626 unsigned long long new = simple_strtoull(buf, &e, 10);
3627
3628 if (mddev->pers == NULL ||
3629 mddev->pers->quiesce == NULL)
3630 return -EINVAL;
3631 if (buf == e || (*e && *e != '\n'))
3632 return -EINVAL;
3633 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3634 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3635 mddev->suspend_hi = new;
3636 mddev->pers->quiesce(mddev, 1);
3637 mddev->pers->quiesce(mddev, 0);
3638 return len;
3639 } else
3640 return -EINVAL;
3641 }
3642 static struct md_sysfs_entry md_suspend_hi =
3643 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3644
3645 static ssize_t
3646 reshape_position_show(mddev_t *mddev, char *page)
3647 {
3648 if (mddev->reshape_position != MaxSector)
3649 return sprintf(page, "%llu\n",
3650 (unsigned long long)mddev->reshape_position);
3651 strcpy(page, "none\n");
3652 return 5;
3653 }
3654
3655 static ssize_t
3656 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3657 {
3658 char *e;
3659 unsigned long long new = simple_strtoull(buf, &e, 10);
3660 if (mddev->pers)
3661 return -EBUSY;
3662 if (buf == e || (*e && *e != '\n'))
3663 return -EINVAL;
3664 mddev->reshape_position = new;
3665 mddev->delta_disks = 0;
3666 mddev->new_level = mddev->level;
3667 mddev->new_layout = mddev->layout;
3668 mddev->new_chunk = mddev->chunk_size;
3669 return len;
3670 }
3671
3672 static struct md_sysfs_entry md_reshape_position =
3673 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3674 reshape_position_store);
3675
3676 static ssize_t
3677 array_size_show(mddev_t *mddev, char *page)
3678 {
3679 if (mddev->external_size)
3680 return sprintf(page, "%llu\n",
3681 (unsigned long long)mddev->array_sectors/2);
3682 else
3683 return sprintf(page, "default\n");
3684 }
3685
3686 static ssize_t
3687 array_size_store(mddev_t *mddev, const char *buf, size_t len)
3688 {
3689 sector_t sectors;
3690
3691 if (strncmp(buf, "default", 7) == 0) {
3692 if (mddev->pers)
3693 sectors = mddev->pers->size(mddev, 0, 0);
3694 else
3695 sectors = mddev->array_sectors;
3696
3697 mddev->external_size = 0;
3698 } else {
3699 if (strict_blocks_to_sectors(buf, &sectors) < 0)
3700 return -EINVAL;
3701 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
3702 return -E2BIG;
3703
3704 mddev->external_size = 1;
3705 }
3706
3707 mddev->array_sectors = sectors;
3708 set_capacity(mddev->gendisk, mddev->array_sectors);
3709 if (mddev->pers) {
3710 struct block_device *bdev = bdget_disk(mddev->gendisk, 0);
3711
3712 if (bdev) {
3713 mutex_lock(&bdev->bd_inode->i_mutex);
3714 i_size_write(bdev->bd_inode,
3715 (loff_t)mddev->array_sectors << 9);
3716 mutex_unlock(&bdev->bd_inode->i_mutex);
3717 bdput(bdev);
3718 }
3719 }
3720
3721 return len;
3722 }
3723
3724 static struct md_sysfs_entry md_array_size =
3725 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
3726 array_size_store);
3727
3728 static struct attribute *md_default_attrs[] = {
3729 &md_level.attr,
3730 &md_layout.attr,
3731 &md_raid_disks.attr,
3732 &md_chunk_size.attr,
3733 &md_size.attr,
3734 &md_resync_start.attr,
3735 &md_metadata.attr,
3736 &md_new_device.attr,
3737 &md_safe_delay.attr,
3738 &md_array_state.attr,
3739 &md_reshape_position.attr,
3740 &md_array_size.attr,
3741 NULL,
3742 };
3743
3744 static struct attribute *md_redundancy_attrs[] = {
3745 &md_scan_mode.attr,
3746 &md_mismatches.attr,
3747 &md_sync_min.attr,
3748 &md_sync_max.attr,
3749 &md_sync_speed.attr,
3750 &md_sync_force_parallel.attr,
3751 &md_sync_completed.attr,
3752 &md_min_sync.attr,
3753 &md_max_sync.attr,
3754 &md_suspend_lo.attr,
3755 &md_suspend_hi.attr,
3756 &md_bitmap.attr,
3757 &md_degraded.attr,
3758 NULL,
3759 };
3760 static struct attribute_group md_redundancy_group = {
3761 .name = NULL,
3762 .attrs = md_redundancy_attrs,
3763 };
3764
3765
3766 static ssize_t
3767 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3768 {
3769 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3770 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3771 ssize_t rv;
3772
3773 if (!entry->show)
3774 return -EIO;
3775 rv = mddev_lock(mddev);
3776 if (!rv) {
3777 rv = entry->show(mddev, page);
3778 mddev_unlock(mddev);
3779 }
3780 return rv;
3781 }
3782
3783 static ssize_t
3784 md_attr_store(struct kobject *kobj, struct attribute *attr,
3785 const char *page, size_t length)
3786 {
3787 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3788 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3789 ssize_t rv;
3790
3791 if (!entry->store)
3792 return -EIO;
3793 if (!capable(CAP_SYS_ADMIN))
3794 return -EACCES;
3795 rv = mddev_lock(mddev);
3796 if (mddev->hold_active == UNTIL_IOCTL)
3797 mddev->hold_active = 0;
3798 if (!rv) {
3799 rv = entry->store(mddev, page, length);
3800 mddev_unlock(mddev);
3801 }
3802 return rv;
3803 }
3804
3805 static void md_free(struct kobject *ko)
3806 {
3807 mddev_t *mddev = container_of(ko, mddev_t, kobj);
3808
3809 if (mddev->sysfs_state)
3810 sysfs_put(mddev->sysfs_state);
3811
3812 if (mddev->gendisk) {
3813 del_gendisk(mddev->gendisk);
3814 put_disk(mddev->gendisk);
3815 }
3816 if (mddev->queue)
3817 blk_cleanup_queue(mddev->queue);
3818
3819 kfree(mddev);
3820 }
3821
3822 static struct sysfs_ops md_sysfs_ops = {
3823 .show = md_attr_show,
3824 .store = md_attr_store,
3825 };
3826 static struct kobj_type md_ktype = {
3827 .release = md_free,
3828 .sysfs_ops = &md_sysfs_ops,
3829 .default_attrs = md_default_attrs,
3830 };
3831
3832 int mdp_major = 0;
3833
3834 static void mddev_delayed_delete(struct work_struct *ws)
3835 {
3836 mddev_t *mddev = container_of(ws, mddev_t, del_work);
3837
3838 if (mddev->private == &md_redundancy_group) {
3839 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3840 if (mddev->sysfs_action)
3841 sysfs_put(mddev->sysfs_action);
3842 mddev->sysfs_action = NULL;
3843 mddev->private = NULL;
3844 }
3845 kobject_del(&mddev->kobj);
3846 kobject_put(&mddev->kobj);
3847 }
3848
3849 static int md_alloc(dev_t dev, char *name)
3850 {
3851 static DEFINE_MUTEX(disks_mutex);
3852 mddev_t *mddev = mddev_find(dev);
3853 struct gendisk *disk;
3854 int partitioned;
3855 int shift;
3856 int unit;
3857 int error;
3858
3859 if (!mddev)
3860 return -ENODEV;
3861
3862 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
3863 shift = partitioned ? MdpMinorShift : 0;
3864 unit = MINOR(mddev->unit) >> shift;
3865
3866 /* wait for any previous instance if this device
3867 * to be completed removed (mddev_delayed_delete).
3868 */
3869 flush_scheduled_work();
3870
3871 mutex_lock(&disks_mutex);
3872 if (mddev->gendisk) {
3873 mutex_unlock(&disks_mutex);
3874 mddev_put(mddev);
3875 return -EEXIST;
3876 }
3877
3878 if (name) {
3879 /* Need to ensure that 'name' is not a duplicate.
3880 */
3881 mddev_t *mddev2;
3882 spin_lock(&all_mddevs_lock);
3883
3884 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
3885 if (mddev2->gendisk &&
3886 strcmp(mddev2->gendisk->disk_name, name) == 0) {
3887 spin_unlock(&all_mddevs_lock);
3888 mutex_unlock(&disks_mutex);
3889 mddev_put(mddev);
3890 return -EEXIST;
3891 }
3892 spin_unlock(&all_mddevs_lock);
3893 }
3894
3895 mddev->queue = blk_alloc_queue(GFP_KERNEL);
3896 if (!mddev->queue) {
3897 mutex_unlock(&disks_mutex);
3898 mddev_put(mddev);
3899 return -ENOMEM;
3900 }
3901 mddev->queue->queuedata = mddev;
3902
3903 /* Can be unlocked because the queue is new: no concurrency */
3904 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
3905
3906 blk_queue_make_request(mddev->queue, md_make_request);
3907
3908 disk = alloc_disk(1 << shift);
3909 if (!disk) {
3910 mutex_unlock(&disks_mutex);
3911 blk_cleanup_queue(mddev->queue);
3912 mddev->queue = NULL;
3913 mddev_put(mddev);
3914 return -ENOMEM;
3915 }
3916 disk->major = MAJOR(mddev->unit);
3917 disk->first_minor = unit << shift;
3918 if (name)
3919 strcpy(disk->disk_name, name);
3920 else if (partitioned)
3921 sprintf(disk->disk_name, "md_d%d", unit);
3922 else
3923 sprintf(disk->disk_name, "md%d", unit);
3924 disk->fops = &md_fops;
3925 disk->private_data = mddev;
3926 disk->queue = mddev->queue;
3927 /* Allow extended partitions. This makes the
3928 * 'mdp' device redundant, but we can't really
3929 * remove it now.
3930 */
3931 disk->flags |= GENHD_FL_EXT_DEVT;
3932 add_disk(disk);
3933 mddev->gendisk = disk;
3934 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
3935 &disk_to_dev(disk)->kobj, "%s", "md");
3936 mutex_unlock(&disks_mutex);
3937 if (error)
3938 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3939 disk->disk_name);
3940 else {
3941 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3942 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state");
3943 }
3944 mddev_put(mddev);
3945 return 0;
3946 }
3947
3948 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3949 {
3950 md_alloc(dev, NULL);
3951 return NULL;
3952 }
3953
3954 static int add_named_array(const char *val, struct kernel_param *kp)
3955 {
3956 /* val must be "md_*" where * is not all digits.
3957 * We allocate an array with a large free minor number, and
3958 * set the name to val. val must not already be an active name.
3959 */
3960 int len = strlen(val);
3961 char buf[DISK_NAME_LEN];
3962
3963 while (len && val[len-1] == '\n')
3964 len--;
3965 if (len >= DISK_NAME_LEN)
3966 return -E2BIG;
3967 strlcpy(buf, val, len+1);
3968 if (strncmp(buf, "md_", 3) != 0)
3969 return -EINVAL;
3970 return md_alloc(0, buf);
3971 }
3972
3973 static void md_safemode_timeout(unsigned long data)
3974 {
3975 mddev_t *mddev = (mddev_t *) data;
3976
3977 if (!atomic_read(&mddev->writes_pending)) {
3978 mddev->safemode = 1;
3979 if (mddev->external)
3980 sysfs_notify_dirent(mddev->sysfs_state);
3981 }
3982 md_wakeup_thread(mddev->thread);
3983 }
3984
3985 static int start_dirty_degraded;
3986
3987 static int do_md_run(mddev_t * mddev)
3988 {
3989 int err;
3990 int chunk_size;
3991 mdk_rdev_t *rdev;
3992 struct gendisk *disk;
3993 struct mdk_personality *pers;
3994 char b[BDEVNAME_SIZE];
3995
3996 if (list_empty(&mddev->disks))
3997 /* cannot run an array with no devices.. */
3998 return -EINVAL;
3999
4000 if (mddev->pers)
4001 return -EBUSY;
4002
4003 /*
4004 * Analyze all RAID superblock(s)
4005 */
4006 if (!mddev->raid_disks) {
4007 if (!mddev->persistent)
4008 return -EINVAL;
4009 analyze_sbs(mddev);
4010 }
4011
4012 chunk_size = mddev->chunk_size;
4013
4014 if (chunk_size) {
4015 if (chunk_size > MAX_CHUNK_SIZE) {
4016 printk(KERN_ERR "too big chunk_size: %d > %d\n",
4017 chunk_size, MAX_CHUNK_SIZE);
4018 return -EINVAL;
4019 }
4020 /*
4021 * chunk-size has to be a power of 2
4022 */
4023 if ( (1 << ffz(~chunk_size)) != chunk_size) {
4024 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
4025 return -EINVAL;
4026 }
4027
4028 /* devices must have minimum size of one chunk */
4029 list_for_each_entry(rdev, &mddev->disks, same_set) {
4030 if (test_bit(Faulty, &rdev->flags))
4031 continue;
4032 if (rdev->sectors < chunk_size / 512) {
4033 printk(KERN_WARNING
4034 "md: Dev %s smaller than chunk_size:"
4035 " %llu < %d\n",
4036 bdevname(rdev->bdev,b),
4037 (unsigned long long)rdev->sectors,
4038 chunk_size / 512);
4039 return -EINVAL;
4040 }
4041 }
4042 }
4043
4044 if (mddev->level != LEVEL_NONE)
4045 request_module("md-level-%d", mddev->level);
4046 else if (mddev->clevel[0])
4047 request_module("md-%s", mddev->clevel);
4048
4049 /*
4050 * Drop all container device buffers, from now on
4051 * the only valid external interface is through the md
4052 * device.
4053 */
4054 list_for_each_entry(rdev, &mddev->disks, same_set) {
4055 if (test_bit(Faulty, &rdev->flags))
4056 continue;
4057 sync_blockdev(rdev->bdev);
4058 invalidate_bdev(rdev->bdev);
4059
4060 /* perform some consistency tests on the device.
4061 * We don't want the data to overlap the metadata,
4062 * Internal Bitmap issues have been handled elsewhere.
4063 */
4064 if (rdev->data_offset < rdev->sb_start) {
4065 if (mddev->dev_sectors &&
4066 rdev->data_offset + mddev->dev_sectors
4067 > rdev->sb_start) {
4068 printk("md: %s: data overlaps metadata\n",
4069 mdname(mddev));
4070 return -EINVAL;
4071 }
4072 } else {
4073 if (rdev->sb_start + rdev->sb_size/512
4074 > rdev->data_offset) {
4075 printk("md: %s: metadata overlaps data\n",
4076 mdname(mddev));
4077 return -EINVAL;
4078 }
4079 }
4080 sysfs_notify_dirent(rdev->sysfs_state);
4081 }
4082
4083 md_probe(mddev->unit, NULL, NULL);
4084 disk = mddev->gendisk;
4085 if (!disk)
4086 return -ENOMEM;
4087
4088 spin_lock(&pers_lock);
4089 pers = find_pers(mddev->level, mddev->clevel);
4090 if (!pers || !try_module_get(pers->owner)) {
4091 spin_unlock(&pers_lock);
4092 if (mddev->level != LEVEL_NONE)
4093 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4094 mddev->level);
4095 else
4096 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4097 mddev->clevel);
4098 return -EINVAL;
4099 }
4100 mddev->pers = pers;
4101 spin_unlock(&pers_lock);
4102 if (mddev->level != pers->level) {
4103 mddev->level = pers->level;
4104 mddev->new_level = pers->level;
4105 }
4106 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4107
4108 if (pers->level >= 4 && pers->level <= 6)
4109 /* Cannot support integrity (yet) */
4110 blk_integrity_unregister(mddev->gendisk);
4111
4112 if (mddev->reshape_position != MaxSector &&
4113 pers->start_reshape == NULL) {
4114 /* This personality cannot handle reshaping... */
4115 mddev->pers = NULL;
4116 module_put(pers->owner);
4117 return -EINVAL;
4118 }
4119
4120 if (pers->sync_request) {
4121 /* Warn if this is a potentially silly
4122 * configuration.
4123 */
4124 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4125 mdk_rdev_t *rdev2;
4126 int warned = 0;
4127
4128 list_for_each_entry(rdev, &mddev->disks, same_set)
4129 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4130 if (rdev < rdev2 &&
4131 rdev->bdev->bd_contains ==
4132 rdev2->bdev->bd_contains) {
4133 printk(KERN_WARNING
4134 "%s: WARNING: %s appears to be"
4135 " on the same physical disk as"
4136 " %s.\n",
4137 mdname(mddev),
4138 bdevname(rdev->bdev,b),
4139 bdevname(rdev2->bdev,b2));
4140 warned = 1;
4141 }
4142 }
4143
4144 if (warned)
4145 printk(KERN_WARNING
4146 "True protection against single-disk"
4147 " failure might be compromised.\n");
4148 }
4149
4150 mddev->recovery = 0;
4151 /* may be over-ridden by personality */
4152 mddev->resync_max_sectors = mddev->dev_sectors;
4153
4154 mddev->barriers_work = 1;
4155 mddev->ok_start_degraded = start_dirty_degraded;
4156
4157 if (start_readonly)
4158 mddev->ro = 2; /* read-only, but switch on first write */
4159
4160 err = mddev->pers->run(mddev);
4161 if (err)
4162 printk(KERN_ERR "md: pers->run() failed ...\n");
4163 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4164 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4165 " but 'external_size' not in effect?\n", __func__);
4166 printk(KERN_ERR
4167 "md: invalid array_size %llu > default size %llu\n",
4168 (unsigned long long)mddev->array_sectors / 2,
4169 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4170 err = -EINVAL;
4171 mddev->pers->stop(mddev);
4172 }
4173 if (err == 0 && mddev->pers->sync_request) {
4174 err = bitmap_create(mddev);
4175 if (err) {
4176 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4177 mdname(mddev), err);
4178 mddev->pers->stop(mddev);
4179 }
4180 }
4181 if (err) {
4182 module_put(mddev->pers->owner);
4183 mddev->pers = NULL;
4184 bitmap_destroy(mddev);
4185 return err;
4186 }
4187 if (mddev->pers->sync_request) {
4188 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4189 printk(KERN_WARNING
4190 "md: cannot register extra attributes for %s\n",
4191 mdname(mddev));
4192 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4193 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4194 mddev->ro = 0;
4195
4196 atomic_set(&mddev->writes_pending,0);
4197 mddev->safemode = 0;
4198 mddev->safemode_timer.function = md_safemode_timeout;
4199 mddev->safemode_timer.data = (unsigned long) mddev;
4200 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4201 mddev->in_sync = 1;
4202
4203 list_for_each_entry(rdev, &mddev->disks, same_set)
4204 if (rdev->raid_disk >= 0) {
4205 char nm[20];
4206 sprintf(nm, "rd%d", rdev->raid_disk);
4207 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4208 printk("md: cannot register %s for %s\n",
4209 nm, mdname(mddev));
4210 }
4211
4212 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4213
4214 if (mddev->flags)
4215 md_update_sb(mddev, 0);
4216
4217 set_capacity(disk, mddev->array_sectors);
4218
4219 /* If there is a partially-recovered drive we need to
4220 * start recovery here. If we leave it to md_check_recovery,
4221 * it will remove the drives and not do the right thing
4222 */
4223 if (mddev->degraded && !mddev->sync_thread) {
4224 int spares = 0;
4225 list_for_each_entry(rdev, &mddev->disks, same_set)
4226 if (rdev->raid_disk >= 0 &&
4227 !test_bit(In_sync, &rdev->flags) &&
4228 !test_bit(Faulty, &rdev->flags))
4229 /* complete an interrupted recovery */
4230 spares++;
4231 if (spares && mddev->pers->sync_request) {
4232 mddev->recovery = 0;
4233 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4234 mddev->sync_thread = md_register_thread(md_do_sync,
4235 mddev,
4236 "%s_resync");
4237 if (!mddev->sync_thread) {
4238 printk(KERN_ERR "%s: could not start resync"
4239 " thread...\n",
4240 mdname(mddev));
4241 /* leave the spares where they are, it shouldn't hurt */
4242 mddev->recovery = 0;
4243 }
4244 }
4245 }
4246 md_wakeup_thread(mddev->thread);
4247 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4248
4249 mddev->changed = 1;
4250 md_new_event(mddev);
4251 sysfs_notify_dirent(mddev->sysfs_state);
4252 if (mddev->sysfs_action)
4253 sysfs_notify_dirent(mddev->sysfs_action);
4254 sysfs_notify(&mddev->kobj, NULL, "degraded");
4255 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4256 return 0;
4257 }
4258
4259 static int restart_array(mddev_t *mddev)
4260 {
4261 struct gendisk *disk = mddev->gendisk;
4262
4263 /* Complain if it has no devices */
4264 if (list_empty(&mddev->disks))
4265 return -ENXIO;
4266 if (!mddev->pers)
4267 return -EINVAL;
4268 if (!mddev->ro)
4269 return -EBUSY;
4270 mddev->safemode = 0;
4271 mddev->ro = 0;
4272 set_disk_ro(disk, 0);
4273 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4274 mdname(mddev));
4275 /* Kick recovery or resync if necessary */
4276 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4277 md_wakeup_thread(mddev->thread);
4278 md_wakeup_thread(mddev->sync_thread);
4279 sysfs_notify_dirent(mddev->sysfs_state);
4280 return 0;
4281 }
4282
4283 /* similar to deny_write_access, but accounts for our holding a reference
4284 * to the file ourselves */
4285 static int deny_bitmap_write_access(struct file * file)
4286 {
4287 struct inode *inode = file->f_mapping->host;
4288
4289 spin_lock(&inode->i_lock);
4290 if (atomic_read(&inode->i_writecount) > 1) {
4291 spin_unlock(&inode->i_lock);
4292 return -ETXTBSY;
4293 }
4294 atomic_set(&inode->i_writecount, -1);
4295 spin_unlock(&inode->i_lock);
4296
4297 return 0;
4298 }
4299
4300 static void restore_bitmap_write_access(struct file *file)
4301 {
4302 struct inode *inode = file->f_mapping->host;
4303
4304 spin_lock(&inode->i_lock);
4305 atomic_set(&inode->i_writecount, 1);
4306 spin_unlock(&inode->i_lock);
4307 }
4308
4309 /* mode:
4310 * 0 - completely stop and dis-assemble array
4311 * 1 - switch to readonly
4312 * 2 - stop but do not disassemble array
4313 */
4314 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4315 {
4316 int err = 0;
4317 struct gendisk *disk = mddev->gendisk;
4318 mdk_rdev_t *rdev;
4319
4320 if (atomic_read(&mddev->openers) > is_open) {
4321 printk("md: %s still in use.\n",mdname(mddev));
4322 return -EBUSY;
4323 }
4324
4325 if (mddev->pers) {
4326
4327 if (mddev->sync_thread) {
4328 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4329 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4330 md_unregister_thread(mddev->sync_thread);
4331 mddev->sync_thread = NULL;
4332 }
4333
4334 del_timer_sync(&mddev->safemode_timer);
4335
4336 switch(mode) {
4337 case 1: /* readonly */
4338 err = -ENXIO;
4339 if (mddev->ro==1)
4340 goto out;
4341 mddev->ro = 1;
4342 break;
4343 case 0: /* disassemble */
4344 case 2: /* stop */
4345 bitmap_flush(mddev);
4346 md_super_wait(mddev);
4347 if (mddev->ro)
4348 set_disk_ro(disk, 0);
4349
4350 mddev->pers->stop(mddev);
4351 mddev->queue->merge_bvec_fn = NULL;
4352 mddev->queue->unplug_fn = NULL;
4353 mddev->queue->backing_dev_info.congested_fn = NULL;
4354 module_put(mddev->pers->owner);
4355 if (mddev->pers->sync_request)
4356 mddev->private = &md_redundancy_group;
4357 mddev->pers = NULL;
4358 /* tell userspace to handle 'inactive' */
4359 sysfs_notify_dirent(mddev->sysfs_state);
4360
4361 list_for_each_entry(rdev, &mddev->disks, same_set)
4362 if (rdev->raid_disk >= 0) {
4363 char nm[20];
4364 sprintf(nm, "rd%d", rdev->raid_disk);
4365 sysfs_remove_link(&mddev->kobj, nm);
4366 }
4367
4368 set_capacity(disk, 0);
4369 mddev->changed = 1;
4370
4371 if (mddev->ro)
4372 mddev->ro = 0;
4373 }
4374 if (!mddev->in_sync || mddev->flags) {
4375 /* mark array as shutdown cleanly */
4376 mddev->in_sync = 1;
4377 md_update_sb(mddev, 1);
4378 }
4379 if (mode == 1)
4380 set_disk_ro(disk, 1);
4381 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4382 }
4383
4384 /*
4385 * Free resources if final stop
4386 */
4387 if (mode == 0) {
4388
4389 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4390
4391 bitmap_destroy(mddev);
4392 if (mddev->bitmap_file) {
4393 restore_bitmap_write_access(mddev->bitmap_file);
4394 fput(mddev->bitmap_file);
4395 mddev->bitmap_file = NULL;
4396 }
4397 mddev->bitmap_offset = 0;
4398
4399 /* make sure all md_delayed_delete calls have finished */
4400 flush_scheduled_work();
4401
4402 export_array(mddev);
4403
4404 mddev->array_sectors = 0;
4405 mddev->external_size = 0;
4406 mddev->dev_sectors = 0;
4407 mddev->raid_disks = 0;
4408 mddev->recovery_cp = 0;
4409 mddev->resync_min = 0;
4410 mddev->resync_max = MaxSector;
4411 mddev->reshape_position = MaxSector;
4412 mddev->external = 0;
4413 mddev->persistent = 0;
4414 mddev->level = LEVEL_NONE;
4415 mddev->clevel[0] = 0;
4416 mddev->flags = 0;
4417 mddev->ro = 0;
4418 mddev->metadata_type[0] = 0;
4419 mddev->chunk_size = 0;
4420 mddev->ctime = mddev->utime = 0;
4421 mddev->layout = 0;
4422 mddev->max_disks = 0;
4423 mddev->events = 0;
4424 mddev->delta_disks = 0;
4425 mddev->new_level = LEVEL_NONE;
4426 mddev->new_layout = 0;
4427 mddev->new_chunk = 0;
4428 mddev->curr_resync = 0;
4429 mddev->resync_mismatches = 0;
4430 mddev->suspend_lo = mddev->suspend_hi = 0;
4431 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4432 mddev->recovery = 0;
4433 mddev->in_sync = 0;
4434 mddev->changed = 0;
4435 mddev->degraded = 0;
4436 mddev->barriers_work = 0;
4437 mddev->safemode = 0;
4438 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4439 if (mddev->hold_active == UNTIL_STOP)
4440 mddev->hold_active = 0;
4441
4442 } else if (mddev->pers)
4443 printk(KERN_INFO "md: %s switched to read-only mode.\n",
4444 mdname(mddev));
4445 err = 0;
4446 blk_integrity_unregister(disk);
4447 md_new_event(mddev);
4448 sysfs_notify_dirent(mddev->sysfs_state);
4449 out:
4450 return err;
4451 }
4452
4453 #ifndef MODULE
4454 static void autorun_array(mddev_t *mddev)
4455 {
4456 mdk_rdev_t *rdev;
4457 int err;
4458
4459 if (list_empty(&mddev->disks))
4460 return;
4461
4462 printk(KERN_INFO "md: running: ");
4463
4464 list_for_each_entry(rdev, &mddev->disks, same_set) {
4465 char b[BDEVNAME_SIZE];
4466 printk("<%s>", bdevname(rdev->bdev,b));
4467 }
4468 printk("\n");
4469
4470 err = do_md_run(mddev);
4471 if (err) {
4472 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4473 do_md_stop(mddev, 0, 0);
4474 }
4475 }
4476
4477 /*
4478 * lets try to run arrays based on all disks that have arrived
4479 * until now. (those are in pending_raid_disks)
4480 *
4481 * the method: pick the first pending disk, collect all disks with
4482 * the same UUID, remove all from the pending list and put them into
4483 * the 'same_array' list. Then order this list based on superblock
4484 * update time (freshest comes first), kick out 'old' disks and
4485 * compare superblocks. If everything's fine then run it.
4486 *
4487 * If "unit" is allocated, then bump its reference count
4488 */
4489 static void autorun_devices(int part)
4490 {
4491 mdk_rdev_t *rdev0, *rdev, *tmp;
4492 mddev_t *mddev;
4493 char b[BDEVNAME_SIZE];
4494
4495 printk(KERN_INFO "md: autorun ...\n");
4496 while (!list_empty(&pending_raid_disks)) {
4497 int unit;
4498 dev_t dev;
4499 LIST_HEAD(candidates);
4500 rdev0 = list_entry(pending_raid_disks.next,
4501 mdk_rdev_t, same_set);
4502
4503 printk(KERN_INFO "md: considering %s ...\n",
4504 bdevname(rdev0->bdev,b));
4505 INIT_LIST_HEAD(&candidates);
4506 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4507 if (super_90_load(rdev, rdev0, 0) >= 0) {
4508 printk(KERN_INFO "md: adding %s ...\n",
4509 bdevname(rdev->bdev,b));
4510 list_move(&rdev->same_set, &candidates);
4511 }
4512 /*
4513 * now we have a set of devices, with all of them having
4514 * mostly sane superblocks. It's time to allocate the
4515 * mddev.
4516 */
4517 if (part) {
4518 dev = MKDEV(mdp_major,
4519 rdev0->preferred_minor << MdpMinorShift);
4520 unit = MINOR(dev) >> MdpMinorShift;
4521 } else {
4522 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4523 unit = MINOR(dev);
4524 }
4525 if (rdev0->preferred_minor != unit) {
4526 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4527 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4528 break;
4529 }
4530
4531 md_probe(dev, NULL, NULL);
4532 mddev = mddev_find(dev);
4533 if (!mddev || !mddev->gendisk) {
4534 if (mddev)
4535 mddev_put(mddev);
4536 printk(KERN_ERR
4537 "md: cannot allocate memory for md drive.\n");
4538 break;
4539 }
4540 if (mddev_lock(mddev))
4541 printk(KERN_WARNING "md: %s locked, cannot run\n",
4542 mdname(mddev));
4543 else if (mddev->raid_disks || mddev->major_version
4544 || !list_empty(&mddev->disks)) {
4545 printk(KERN_WARNING
4546 "md: %s already running, cannot run %s\n",
4547 mdname(mddev), bdevname(rdev0->bdev,b));
4548 mddev_unlock(mddev);
4549 } else {
4550 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4551 mddev->persistent = 1;
4552 rdev_for_each_list(rdev, tmp, &candidates) {
4553 list_del_init(&rdev->same_set);
4554 if (bind_rdev_to_array(rdev, mddev))
4555 export_rdev(rdev);
4556 }
4557 autorun_array(mddev);
4558 mddev_unlock(mddev);
4559 }
4560 /* on success, candidates will be empty, on error
4561 * it won't...
4562 */
4563 rdev_for_each_list(rdev, tmp, &candidates) {
4564 list_del_init(&rdev->same_set);
4565 export_rdev(rdev);
4566 }
4567 mddev_put(mddev);
4568 }
4569 printk(KERN_INFO "md: ... autorun DONE.\n");
4570 }
4571 #endif /* !MODULE */
4572
4573 static int get_version(void __user * arg)
4574 {
4575 mdu_version_t ver;
4576
4577 ver.major = MD_MAJOR_VERSION;
4578 ver.minor = MD_MINOR_VERSION;
4579 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4580
4581 if (copy_to_user(arg, &ver, sizeof(ver)))
4582 return -EFAULT;
4583
4584 return 0;
4585 }
4586
4587 static int get_array_info(mddev_t * mddev, void __user * arg)
4588 {
4589 mdu_array_info_t info;
4590 int nr,working,active,failed,spare;
4591 mdk_rdev_t *rdev;
4592
4593 nr=working=active=failed=spare=0;
4594 list_for_each_entry(rdev, &mddev->disks, same_set) {
4595 nr++;
4596 if (test_bit(Faulty, &rdev->flags))
4597 failed++;
4598 else {
4599 working++;
4600 if (test_bit(In_sync, &rdev->flags))
4601 active++;
4602 else
4603 spare++;
4604 }
4605 }
4606
4607 info.major_version = mddev->major_version;
4608 info.minor_version = mddev->minor_version;
4609 info.patch_version = MD_PATCHLEVEL_VERSION;
4610 info.ctime = mddev->ctime;
4611 info.level = mddev->level;
4612 info.size = mddev->dev_sectors / 2;
4613 if (info.size != mddev->dev_sectors / 2) /* overflow */
4614 info.size = -1;
4615 info.nr_disks = nr;
4616 info.raid_disks = mddev->raid_disks;
4617 info.md_minor = mddev->md_minor;
4618 info.not_persistent= !mddev->persistent;
4619
4620 info.utime = mddev->utime;
4621 info.state = 0;
4622 if (mddev->in_sync)
4623 info.state = (1<<MD_SB_CLEAN);
4624 if (mddev->bitmap && mddev->bitmap_offset)
4625 info.state = (1<<MD_SB_BITMAP_PRESENT);
4626 info.active_disks = active;
4627 info.working_disks = working;
4628 info.failed_disks = failed;
4629 info.spare_disks = spare;
4630
4631 info.layout = mddev->layout;
4632 info.chunk_size = mddev->chunk_size;
4633
4634 if (copy_to_user(arg, &info, sizeof(info)))
4635 return -EFAULT;
4636
4637 return 0;
4638 }
4639
4640 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4641 {
4642 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4643 char *ptr, *buf = NULL;
4644 int err = -ENOMEM;
4645
4646 if (md_allow_write(mddev))
4647 file = kmalloc(sizeof(*file), GFP_NOIO);
4648 else
4649 file = kmalloc(sizeof(*file), GFP_KERNEL);
4650
4651 if (!file)
4652 goto out;
4653
4654 /* bitmap disabled, zero the first byte and copy out */
4655 if (!mddev->bitmap || !mddev->bitmap->file) {
4656 file->pathname[0] = '\0';
4657 goto copy_out;
4658 }
4659
4660 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4661 if (!buf)
4662 goto out;
4663
4664 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4665 if (IS_ERR(ptr))
4666 goto out;
4667
4668 strcpy(file->pathname, ptr);
4669
4670 copy_out:
4671 err = 0;
4672 if (copy_to_user(arg, file, sizeof(*file)))
4673 err = -EFAULT;
4674 out:
4675 kfree(buf);
4676 kfree(file);
4677 return err;
4678 }
4679
4680 static int get_disk_info(mddev_t * mddev, void __user * arg)
4681 {
4682 mdu_disk_info_t info;
4683 mdk_rdev_t *rdev;
4684
4685 if (copy_from_user(&info, arg, sizeof(info)))
4686 return -EFAULT;
4687
4688 rdev = find_rdev_nr(mddev, info.number);
4689 if (rdev) {
4690 info.major = MAJOR(rdev->bdev->bd_dev);
4691 info.minor = MINOR(rdev->bdev->bd_dev);
4692 info.raid_disk = rdev->raid_disk;
4693 info.state = 0;
4694 if (test_bit(Faulty, &rdev->flags))
4695 info.state |= (1<<MD_DISK_FAULTY);
4696 else if (test_bit(In_sync, &rdev->flags)) {
4697 info.state |= (1<<MD_DISK_ACTIVE);
4698 info.state |= (1<<MD_DISK_SYNC);
4699 }
4700 if (test_bit(WriteMostly, &rdev->flags))
4701 info.state |= (1<<MD_DISK_WRITEMOSTLY);
4702 } else {
4703 info.major = info.minor = 0;
4704 info.raid_disk = -1;
4705 info.state = (1<<MD_DISK_REMOVED);
4706 }
4707
4708 if (copy_to_user(arg, &info, sizeof(info)))
4709 return -EFAULT;
4710
4711 return 0;
4712 }
4713
4714 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4715 {
4716 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4717 mdk_rdev_t *rdev;
4718 dev_t dev = MKDEV(info->major,info->minor);
4719
4720 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4721 return -EOVERFLOW;
4722
4723 if (!mddev->raid_disks) {
4724 int err;
4725 /* expecting a device which has a superblock */
4726 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4727 if (IS_ERR(rdev)) {
4728 printk(KERN_WARNING
4729 "md: md_import_device returned %ld\n",
4730 PTR_ERR(rdev));
4731 return PTR_ERR(rdev);
4732 }
4733 if (!list_empty(&mddev->disks)) {
4734 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4735 mdk_rdev_t, same_set);
4736 int err = super_types[mddev->major_version]
4737 .load_super(rdev, rdev0, mddev->minor_version);
4738 if (err < 0) {
4739 printk(KERN_WARNING
4740 "md: %s has different UUID to %s\n",
4741 bdevname(rdev->bdev,b),
4742 bdevname(rdev0->bdev,b2));
4743 export_rdev(rdev);
4744 return -EINVAL;
4745 }
4746 }
4747 err = bind_rdev_to_array(rdev, mddev);
4748 if (err)
4749 export_rdev(rdev);
4750 return err;
4751 }
4752
4753 /*
4754 * add_new_disk can be used once the array is assembled
4755 * to add "hot spares". They must already have a superblock
4756 * written
4757 */
4758 if (mddev->pers) {
4759 int err;
4760 if (!mddev->pers->hot_add_disk) {
4761 printk(KERN_WARNING
4762 "%s: personality does not support diskops!\n",
4763 mdname(mddev));
4764 return -EINVAL;
4765 }
4766 if (mddev->persistent)
4767 rdev = md_import_device(dev, mddev->major_version,
4768 mddev->minor_version);
4769 else
4770 rdev = md_import_device(dev, -1, -1);
4771 if (IS_ERR(rdev)) {
4772 printk(KERN_WARNING
4773 "md: md_import_device returned %ld\n",
4774 PTR_ERR(rdev));
4775 return PTR_ERR(rdev);
4776 }
4777 /* set save_raid_disk if appropriate */
4778 if (!mddev->persistent) {
4779 if (info->state & (1<<MD_DISK_SYNC) &&
4780 info->raid_disk < mddev->raid_disks)
4781 rdev->raid_disk = info->raid_disk;
4782 else
4783 rdev->raid_disk = -1;
4784 } else
4785 super_types[mddev->major_version].
4786 validate_super(mddev, rdev);
4787 rdev->saved_raid_disk = rdev->raid_disk;
4788
4789 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4790 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4791 set_bit(WriteMostly, &rdev->flags);
4792 else
4793 clear_bit(WriteMostly, &rdev->flags);
4794
4795 rdev->raid_disk = -1;
4796 err = bind_rdev_to_array(rdev, mddev);
4797 if (!err && !mddev->pers->hot_remove_disk) {
4798 /* If there is hot_add_disk but no hot_remove_disk
4799 * then added disks for geometry changes,
4800 * and should be added immediately.
4801 */
4802 super_types[mddev->major_version].
4803 validate_super(mddev, rdev);
4804 err = mddev->pers->hot_add_disk(mddev, rdev);
4805 if (err)
4806 unbind_rdev_from_array(rdev);
4807 }
4808 if (err)
4809 export_rdev(rdev);
4810 else
4811 sysfs_notify_dirent(rdev->sysfs_state);
4812
4813 md_update_sb(mddev, 1);
4814 if (mddev->degraded)
4815 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4816 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4817 md_wakeup_thread(mddev->thread);
4818 return err;
4819 }
4820
4821 /* otherwise, add_new_disk is only allowed
4822 * for major_version==0 superblocks
4823 */
4824 if (mddev->major_version != 0) {
4825 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4826 mdname(mddev));
4827 return -EINVAL;
4828 }
4829
4830 if (!(info->state & (1<<MD_DISK_FAULTY))) {
4831 int err;
4832 rdev = md_import_device(dev, -1, 0);
4833 if (IS_ERR(rdev)) {
4834 printk(KERN_WARNING
4835 "md: error, md_import_device() returned %ld\n",
4836 PTR_ERR(rdev));
4837 return PTR_ERR(rdev);
4838 }
4839 rdev->desc_nr = info->number;
4840 if (info->raid_disk < mddev->raid_disks)
4841 rdev->raid_disk = info->raid_disk;
4842 else
4843 rdev->raid_disk = -1;
4844
4845 if (rdev->raid_disk < mddev->raid_disks)
4846 if (info->state & (1<<MD_DISK_SYNC))
4847 set_bit(In_sync, &rdev->flags);
4848
4849 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4850 set_bit(WriteMostly, &rdev->flags);
4851
4852 if (!mddev->persistent) {
4853 printk(KERN_INFO "md: nonpersistent superblock ...\n");
4854 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4855 } else
4856 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4857 rdev->sectors = calc_num_sectors(rdev, mddev->chunk_size);
4858
4859 err = bind_rdev_to_array(rdev, mddev);
4860 if (err) {
4861 export_rdev(rdev);
4862 return err;
4863 }
4864 }
4865
4866 return 0;
4867 }
4868
4869 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4870 {
4871 char b[BDEVNAME_SIZE];
4872 mdk_rdev_t *rdev;
4873
4874 rdev = find_rdev(mddev, dev);
4875 if (!rdev)
4876 return -ENXIO;
4877
4878 if (rdev->raid_disk >= 0)
4879 goto busy;
4880
4881 kick_rdev_from_array(rdev);
4882 md_update_sb(mddev, 1);
4883 md_new_event(mddev);
4884
4885 return 0;
4886 busy:
4887 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4888 bdevname(rdev->bdev,b), mdname(mddev));
4889 return -EBUSY;
4890 }
4891
4892 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4893 {
4894 char b[BDEVNAME_SIZE];
4895 int err;
4896 mdk_rdev_t *rdev;
4897
4898 if (!mddev->pers)
4899 return -ENODEV;
4900
4901 if (mddev->major_version != 0) {
4902 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4903 " version-0 superblocks.\n",
4904 mdname(mddev));
4905 return -EINVAL;
4906 }
4907 if (!mddev->pers->hot_add_disk) {
4908 printk(KERN_WARNING
4909 "%s: personality does not support diskops!\n",
4910 mdname(mddev));
4911 return -EINVAL;
4912 }
4913
4914 rdev = md_import_device(dev, -1, 0);
4915 if (IS_ERR(rdev)) {
4916 printk(KERN_WARNING
4917 "md: error, md_import_device() returned %ld\n",
4918 PTR_ERR(rdev));
4919 return -EINVAL;
4920 }
4921
4922 if (mddev->persistent)
4923 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4924 else
4925 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4926
4927 rdev->sectors = calc_num_sectors(rdev, mddev->chunk_size);
4928
4929 if (test_bit(Faulty, &rdev->flags)) {
4930 printk(KERN_WARNING
4931 "md: can not hot-add faulty %s disk to %s!\n",
4932 bdevname(rdev->bdev,b), mdname(mddev));
4933 err = -EINVAL;
4934 goto abort_export;
4935 }
4936 clear_bit(In_sync, &rdev->flags);
4937 rdev->desc_nr = -1;
4938 rdev->saved_raid_disk = -1;
4939 err = bind_rdev_to_array(rdev, mddev);
4940 if (err)
4941 goto abort_export;
4942
4943 /*
4944 * The rest should better be atomic, we can have disk failures
4945 * noticed in interrupt contexts ...
4946 */
4947
4948 rdev->raid_disk = -1;
4949
4950 md_update_sb(mddev, 1);
4951
4952 /*
4953 * Kick recovery, maybe this spare has to be added to the
4954 * array immediately.
4955 */
4956 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4957 md_wakeup_thread(mddev->thread);
4958 md_new_event(mddev);
4959 return 0;
4960
4961 abort_export:
4962 export_rdev(rdev);
4963 return err;
4964 }
4965
4966 static int set_bitmap_file(mddev_t *mddev, int fd)
4967 {
4968 int err;
4969
4970 if (mddev->pers) {
4971 if (!mddev->pers->quiesce)
4972 return -EBUSY;
4973 if (mddev->recovery || mddev->sync_thread)
4974 return -EBUSY;
4975 /* we should be able to change the bitmap.. */
4976 }
4977
4978
4979 if (fd >= 0) {
4980 if (mddev->bitmap)
4981 return -EEXIST; /* cannot add when bitmap is present */
4982 mddev->bitmap_file = fget(fd);
4983
4984 if (mddev->bitmap_file == NULL) {
4985 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4986 mdname(mddev));
4987 return -EBADF;
4988 }
4989
4990 err = deny_bitmap_write_access(mddev->bitmap_file);
4991 if (err) {
4992 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4993 mdname(mddev));
4994 fput(mddev->bitmap_file);
4995 mddev->bitmap_file = NULL;
4996 return err;
4997 }
4998 mddev->bitmap_offset = 0; /* file overrides offset */
4999 } else if (mddev->bitmap == NULL)
5000 return -ENOENT; /* cannot remove what isn't there */
5001 err = 0;
5002 if (mddev->pers) {
5003 mddev->pers->quiesce(mddev, 1);
5004 if (fd >= 0)
5005 err = bitmap_create(mddev);
5006 if (fd < 0 || err) {
5007 bitmap_destroy(mddev);
5008 fd = -1; /* make sure to put the file */
5009 }
5010 mddev->pers->quiesce(mddev, 0);
5011 }
5012 if (fd < 0) {
5013 if (mddev->bitmap_file) {
5014 restore_bitmap_write_access(mddev->bitmap_file);
5015 fput(mddev->bitmap_file);
5016 }
5017 mddev->bitmap_file = NULL;
5018 }
5019
5020 return err;
5021 }
5022
5023 /*
5024 * set_array_info is used two different ways
5025 * The original usage is when creating a new array.
5026 * In this usage, raid_disks is > 0 and it together with
5027 * level, size, not_persistent,layout,chunksize determine the
5028 * shape of the array.
5029 * This will always create an array with a type-0.90.0 superblock.
5030 * The newer usage is when assembling an array.
5031 * In this case raid_disks will be 0, and the major_version field is
5032 * use to determine which style super-blocks are to be found on the devices.
5033 * The minor and patch _version numbers are also kept incase the
5034 * super_block handler wishes to interpret them.
5035 */
5036 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5037 {
5038
5039 if (info->raid_disks == 0) {
5040 /* just setting version number for superblock loading */
5041 if (info->major_version < 0 ||
5042 info->major_version >= ARRAY_SIZE(super_types) ||
5043 super_types[info->major_version].name == NULL) {
5044 /* maybe try to auto-load a module? */
5045 printk(KERN_INFO
5046 "md: superblock version %d not known\n",
5047 info->major_version);
5048 return -EINVAL;
5049 }
5050 mddev->major_version = info->major_version;
5051 mddev->minor_version = info->minor_version;
5052 mddev->patch_version = info->patch_version;
5053 mddev->persistent = !info->not_persistent;
5054 return 0;
5055 }
5056 mddev->major_version = MD_MAJOR_VERSION;
5057 mddev->minor_version = MD_MINOR_VERSION;
5058 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5059 mddev->ctime = get_seconds();
5060
5061 mddev->level = info->level;
5062 mddev->clevel[0] = 0;
5063 mddev->dev_sectors = 2 * (sector_t)info->size;
5064 mddev->raid_disks = info->raid_disks;
5065 /* don't set md_minor, it is determined by which /dev/md* was
5066 * openned
5067 */
5068 if (info->state & (1<<MD_SB_CLEAN))
5069 mddev->recovery_cp = MaxSector;
5070 else
5071 mddev->recovery_cp = 0;
5072 mddev->persistent = ! info->not_persistent;
5073 mddev->external = 0;
5074
5075 mddev->layout = info->layout;
5076 mddev->chunk_size = info->chunk_size;
5077
5078 mddev->max_disks = MD_SB_DISKS;
5079
5080 if (mddev->persistent)
5081 mddev->flags = 0;
5082 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5083
5084 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
5085 mddev->bitmap_offset = 0;
5086
5087 mddev->reshape_position = MaxSector;
5088
5089 /*
5090 * Generate a 128 bit UUID
5091 */
5092 get_random_bytes(mddev->uuid, 16);
5093
5094 mddev->new_level = mddev->level;
5095 mddev->new_chunk = mddev->chunk_size;
5096 mddev->new_layout = mddev->layout;
5097 mddev->delta_disks = 0;
5098
5099 return 0;
5100 }
5101
5102 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5103 {
5104 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5105
5106 if (mddev->external_size)
5107 return;
5108
5109 mddev->array_sectors = array_sectors;
5110 }
5111 EXPORT_SYMBOL(md_set_array_sectors);
5112
5113 static int update_size(mddev_t *mddev, sector_t num_sectors)
5114 {
5115 mdk_rdev_t *rdev;
5116 int rv;
5117 int fit = (num_sectors == 0);
5118
5119 if (mddev->pers->resize == NULL)
5120 return -EINVAL;
5121 /* The "num_sectors" is the number of sectors of each device that
5122 * is used. This can only make sense for arrays with redundancy.
5123 * linear and raid0 always use whatever space is available. We can only
5124 * consider changing this number if no resync or reconstruction is
5125 * happening, and if the new size is acceptable. It must fit before the
5126 * sb_start or, if that is <data_offset, it must fit before the size
5127 * of each device. If num_sectors is zero, we find the largest size
5128 * that fits.
5129
5130 */
5131 if (mddev->sync_thread)
5132 return -EBUSY;
5133 if (mddev->bitmap)
5134 /* Sorry, cannot grow a bitmap yet, just remove it,
5135 * grow, and re-add.
5136 */
5137 return -EBUSY;
5138 list_for_each_entry(rdev, &mddev->disks, same_set) {
5139 sector_t avail = rdev->sectors;
5140
5141 if (fit && (num_sectors == 0 || num_sectors > avail))
5142 num_sectors = avail;
5143 if (avail < num_sectors)
5144 return -ENOSPC;
5145 }
5146 rv = mddev->pers->resize(mddev, num_sectors);
5147 if (!rv) {
5148 struct block_device *bdev;
5149
5150 bdev = bdget_disk(mddev->gendisk, 0);
5151 if (bdev) {
5152 mutex_lock(&bdev->bd_inode->i_mutex);
5153 i_size_write(bdev->bd_inode,
5154 (loff_t)mddev->array_sectors << 9);
5155 mutex_unlock(&bdev->bd_inode->i_mutex);
5156 bdput(bdev);
5157 }
5158 }
5159 return rv;
5160 }
5161
5162 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5163 {
5164 int rv;
5165 /* change the number of raid disks */
5166 if (mddev->pers->check_reshape == NULL)
5167 return -EINVAL;
5168 if (raid_disks <= 0 ||
5169 raid_disks >= mddev->max_disks)
5170 return -EINVAL;
5171 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5172 return -EBUSY;
5173 mddev->delta_disks = raid_disks - mddev->raid_disks;
5174
5175 rv = mddev->pers->check_reshape(mddev);
5176 return rv;
5177 }
5178
5179
5180 /*
5181 * update_array_info is used to change the configuration of an
5182 * on-line array.
5183 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5184 * fields in the info are checked against the array.
5185 * Any differences that cannot be handled will cause an error.
5186 * Normally, only one change can be managed at a time.
5187 */
5188 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5189 {
5190 int rv = 0;
5191 int cnt = 0;
5192 int state = 0;
5193
5194 /* calculate expected state,ignoring low bits */
5195 if (mddev->bitmap && mddev->bitmap_offset)
5196 state |= (1 << MD_SB_BITMAP_PRESENT);
5197
5198 if (mddev->major_version != info->major_version ||
5199 mddev->minor_version != info->minor_version ||
5200 /* mddev->patch_version != info->patch_version || */
5201 mddev->ctime != info->ctime ||
5202 mddev->level != info->level ||
5203 /* mddev->layout != info->layout || */
5204 !mddev->persistent != info->not_persistent||
5205 mddev->chunk_size != info->chunk_size ||
5206 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5207 ((state^info->state) & 0xfffffe00)
5208 )
5209 return -EINVAL;
5210 /* Check there is only one change */
5211 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5212 cnt++;
5213 if (mddev->raid_disks != info->raid_disks)
5214 cnt++;
5215 if (mddev->layout != info->layout)
5216 cnt++;
5217 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5218 cnt++;
5219 if (cnt == 0)
5220 return 0;
5221 if (cnt > 1)
5222 return -EINVAL;
5223
5224 if (mddev->layout != info->layout) {
5225 /* Change layout
5226 * we don't need to do anything at the md level, the
5227 * personality will take care of it all.
5228 */
5229 if (mddev->pers->reconfig == NULL)
5230 return -EINVAL;
5231 else
5232 return mddev->pers->reconfig(mddev, info->layout, -1);
5233 }
5234 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5235 rv = update_size(mddev, (sector_t)info->size * 2);
5236
5237 if (mddev->raid_disks != info->raid_disks)
5238 rv = update_raid_disks(mddev, info->raid_disks);
5239
5240 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5241 if (mddev->pers->quiesce == NULL)
5242 return -EINVAL;
5243 if (mddev->recovery || mddev->sync_thread)
5244 return -EBUSY;
5245 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5246 /* add the bitmap */
5247 if (mddev->bitmap)
5248 return -EEXIST;
5249 if (mddev->default_bitmap_offset == 0)
5250 return -EINVAL;
5251 mddev->bitmap_offset = mddev->default_bitmap_offset;
5252 mddev->pers->quiesce(mddev, 1);
5253 rv = bitmap_create(mddev);
5254 if (rv)
5255 bitmap_destroy(mddev);
5256 mddev->pers->quiesce(mddev, 0);
5257 } else {
5258 /* remove the bitmap */
5259 if (!mddev->bitmap)
5260 return -ENOENT;
5261 if (mddev->bitmap->file)
5262 return -EINVAL;
5263 mddev->pers->quiesce(mddev, 1);
5264 bitmap_destroy(mddev);
5265 mddev->pers->quiesce(mddev, 0);
5266 mddev->bitmap_offset = 0;
5267 }
5268 }
5269 md_update_sb(mddev, 1);
5270 return rv;
5271 }
5272
5273 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5274 {
5275 mdk_rdev_t *rdev;
5276
5277 if (mddev->pers == NULL)
5278 return -ENODEV;
5279
5280 rdev = find_rdev(mddev, dev);
5281 if (!rdev)
5282 return -ENODEV;
5283
5284 md_error(mddev, rdev);
5285 return 0;
5286 }
5287
5288 /*
5289 * We have a problem here : there is no easy way to give a CHS
5290 * virtual geometry. We currently pretend that we have a 2 heads
5291 * 4 sectors (with a BIG number of cylinders...). This drives
5292 * dosfs just mad... ;-)
5293 */
5294 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5295 {
5296 mddev_t *mddev = bdev->bd_disk->private_data;
5297
5298 geo->heads = 2;
5299 geo->sectors = 4;
5300 geo->cylinders = get_capacity(mddev->gendisk) / 8;
5301 return 0;
5302 }
5303
5304 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5305 unsigned int cmd, unsigned long arg)
5306 {
5307 int err = 0;
5308 void __user *argp = (void __user *)arg;
5309 mddev_t *mddev = NULL;
5310
5311 if (!capable(CAP_SYS_ADMIN))
5312 return -EACCES;
5313
5314 /*
5315 * Commands dealing with the RAID driver but not any
5316 * particular array:
5317 */
5318 switch (cmd)
5319 {
5320 case RAID_VERSION:
5321 err = get_version(argp);
5322 goto done;
5323
5324 case PRINT_RAID_DEBUG:
5325 err = 0;
5326 md_print_devices();
5327 goto done;
5328
5329 #ifndef MODULE
5330 case RAID_AUTORUN:
5331 err = 0;
5332 autostart_arrays(arg);
5333 goto done;
5334 #endif
5335 default:;
5336 }
5337
5338 /*
5339 * Commands creating/starting a new array:
5340 */
5341
5342 mddev = bdev->bd_disk->private_data;
5343
5344 if (!mddev) {
5345 BUG();
5346 goto abort;
5347 }
5348
5349 err = mddev_lock(mddev);
5350 if (err) {
5351 printk(KERN_INFO
5352 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5353 err, cmd);
5354 goto abort;
5355 }
5356
5357 switch (cmd)
5358 {
5359 case SET_ARRAY_INFO:
5360 {
5361 mdu_array_info_t info;
5362 if (!arg)
5363 memset(&info, 0, sizeof(info));
5364 else if (copy_from_user(&info, argp, sizeof(info))) {
5365 err = -EFAULT;
5366 goto abort_unlock;
5367 }
5368 if (mddev->pers) {
5369 err = update_array_info(mddev, &info);
5370 if (err) {
5371 printk(KERN_WARNING "md: couldn't update"
5372 " array info. %d\n", err);
5373 goto abort_unlock;
5374 }
5375 goto done_unlock;
5376 }
5377 if (!list_empty(&mddev->disks)) {
5378 printk(KERN_WARNING
5379 "md: array %s already has disks!\n",
5380 mdname(mddev));
5381 err = -EBUSY;
5382 goto abort_unlock;
5383 }
5384 if (mddev->raid_disks) {
5385 printk(KERN_WARNING
5386 "md: array %s already initialised!\n",
5387 mdname(mddev));
5388 err = -EBUSY;
5389 goto abort_unlock;
5390 }
5391 err = set_array_info(mddev, &info);
5392 if (err) {
5393 printk(KERN_WARNING "md: couldn't set"
5394 " array info. %d\n", err);
5395 goto abort_unlock;
5396 }
5397 }
5398 goto done_unlock;
5399
5400 default:;
5401 }
5402
5403 /*
5404 * Commands querying/configuring an existing array:
5405 */
5406 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5407 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5408 if ((!mddev->raid_disks && !mddev->external)
5409 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5410 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5411 && cmd != GET_BITMAP_FILE) {
5412 err = -ENODEV;
5413 goto abort_unlock;
5414 }
5415
5416 /*
5417 * Commands even a read-only array can execute:
5418 */
5419 switch (cmd)
5420 {
5421 case GET_ARRAY_INFO:
5422 err = get_array_info(mddev, argp);
5423 goto done_unlock;
5424
5425 case GET_BITMAP_FILE:
5426 err = get_bitmap_file(mddev, argp);
5427 goto done_unlock;
5428
5429 case GET_DISK_INFO:
5430 err = get_disk_info(mddev, argp);
5431 goto done_unlock;
5432
5433 case RESTART_ARRAY_RW:
5434 err = restart_array(mddev);
5435 goto done_unlock;
5436
5437 case STOP_ARRAY:
5438 err = do_md_stop(mddev, 0, 1);
5439 goto done_unlock;
5440
5441 case STOP_ARRAY_RO:
5442 err = do_md_stop(mddev, 1, 1);
5443 goto done_unlock;
5444
5445 }
5446
5447 /*
5448 * The remaining ioctls are changing the state of the
5449 * superblock, so we do not allow them on read-only arrays.
5450 * However non-MD ioctls (e.g. get-size) will still come through
5451 * here and hit the 'default' below, so only disallow
5452 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5453 */
5454 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5455 if (mddev->ro == 2) {
5456 mddev->ro = 0;
5457 sysfs_notify_dirent(mddev->sysfs_state);
5458 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5459 md_wakeup_thread(mddev->thread);
5460 } else {
5461 err = -EROFS;
5462 goto abort_unlock;
5463 }
5464 }
5465
5466 switch (cmd)
5467 {
5468 case ADD_NEW_DISK:
5469 {
5470 mdu_disk_info_t info;
5471 if (copy_from_user(&info, argp, sizeof(info)))
5472 err = -EFAULT;
5473 else
5474 err = add_new_disk(mddev, &info);
5475 goto done_unlock;
5476 }
5477
5478 case HOT_REMOVE_DISK:
5479 err = hot_remove_disk(mddev, new_decode_dev(arg));
5480 goto done_unlock;
5481
5482 case HOT_ADD_DISK:
5483 err = hot_add_disk(mddev, new_decode_dev(arg));
5484 goto done_unlock;
5485
5486 case SET_DISK_FAULTY:
5487 err = set_disk_faulty(mddev, new_decode_dev(arg));
5488 goto done_unlock;
5489
5490 case RUN_ARRAY:
5491 err = do_md_run(mddev);
5492 goto done_unlock;
5493
5494 case SET_BITMAP_FILE:
5495 err = set_bitmap_file(mddev, (int)arg);
5496 goto done_unlock;
5497
5498 default:
5499 err = -EINVAL;
5500 goto abort_unlock;
5501 }
5502
5503 done_unlock:
5504 abort_unlock:
5505 if (mddev->hold_active == UNTIL_IOCTL &&
5506 err != -EINVAL)
5507 mddev->hold_active = 0;
5508 mddev_unlock(mddev);
5509
5510 return err;
5511 done:
5512 if (err)
5513 MD_BUG();
5514 abort:
5515 return err;
5516 }
5517
5518 static int md_open(struct block_device *bdev, fmode_t mode)
5519 {
5520 /*
5521 * Succeed if we can lock the mddev, which confirms that
5522 * it isn't being stopped right now.
5523 */
5524 mddev_t *mddev = mddev_find(bdev->bd_dev);
5525 int err;
5526
5527 if (mddev->gendisk != bdev->bd_disk) {
5528 /* we are racing with mddev_put which is discarding this
5529 * bd_disk.
5530 */
5531 mddev_put(mddev);
5532 /* Wait until bdev->bd_disk is definitely gone */
5533 flush_scheduled_work();
5534 /* Then retry the open from the top */
5535 return -ERESTARTSYS;
5536 }
5537 BUG_ON(mddev != bdev->bd_disk->private_data);
5538
5539 if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
5540 goto out;
5541
5542 err = 0;
5543 atomic_inc(&mddev->openers);
5544 mddev_unlock(mddev);
5545
5546 check_disk_change(bdev);
5547 out:
5548 return err;
5549 }
5550
5551 static int md_release(struct gendisk *disk, fmode_t mode)
5552 {
5553 mddev_t *mddev = disk->private_data;
5554
5555 BUG_ON(!mddev);
5556 atomic_dec(&mddev->openers);
5557 mddev_put(mddev);
5558
5559 return 0;
5560 }
5561
5562 static int md_media_changed(struct gendisk *disk)
5563 {
5564 mddev_t *mddev = disk->private_data;
5565
5566 return mddev->changed;
5567 }
5568
5569 static int md_revalidate(struct gendisk *disk)
5570 {
5571 mddev_t *mddev = disk->private_data;
5572
5573 mddev->changed = 0;
5574 return 0;
5575 }
5576 static struct block_device_operations md_fops =
5577 {
5578 .owner = THIS_MODULE,
5579 .open = md_open,
5580 .release = md_release,
5581 .ioctl = md_ioctl,
5582 .getgeo = md_getgeo,
5583 .media_changed = md_media_changed,
5584 .revalidate_disk= md_revalidate,
5585 };
5586
5587 static int md_thread(void * arg)
5588 {
5589 mdk_thread_t *thread = arg;
5590
5591 /*
5592 * md_thread is a 'system-thread', it's priority should be very
5593 * high. We avoid resource deadlocks individually in each
5594 * raid personality. (RAID5 does preallocation) We also use RR and
5595 * the very same RT priority as kswapd, thus we will never get
5596 * into a priority inversion deadlock.
5597 *
5598 * we definitely have to have equal or higher priority than
5599 * bdflush, otherwise bdflush will deadlock if there are too
5600 * many dirty RAID5 blocks.
5601 */
5602
5603 allow_signal(SIGKILL);
5604 while (!kthread_should_stop()) {
5605
5606 /* We need to wait INTERRUPTIBLE so that
5607 * we don't add to the load-average.
5608 * That means we need to be sure no signals are
5609 * pending
5610 */
5611 if (signal_pending(current))
5612 flush_signals(current);
5613
5614 wait_event_interruptible_timeout
5615 (thread->wqueue,
5616 test_bit(THREAD_WAKEUP, &thread->flags)
5617 || kthread_should_stop(),
5618 thread->timeout);
5619
5620 clear_bit(THREAD_WAKEUP, &thread->flags);
5621
5622 thread->run(thread->mddev);
5623 }
5624
5625 return 0;
5626 }
5627
5628 void md_wakeup_thread(mdk_thread_t *thread)
5629 {
5630 if (thread) {
5631 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5632 set_bit(THREAD_WAKEUP, &thread->flags);
5633 wake_up(&thread->wqueue);
5634 }
5635 }
5636
5637 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5638 const char *name)
5639 {
5640 mdk_thread_t *thread;
5641
5642 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5643 if (!thread)
5644 return NULL;
5645
5646 init_waitqueue_head(&thread->wqueue);
5647
5648 thread->run = run;
5649 thread->mddev = mddev;
5650 thread->timeout = MAX_SCHEDULE_TIMEOUT;
5651 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
5652 if (IS_ERR(thread->tsk)) {
5653 kfree(thread);
5654 return NULL;
5655 }
5656 return thread;
5657 }
5658
5659 void md_unregister_thread(mdk_thread_t *thread)
5660 {
5661 if (!thread)
5662 return;
5663 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5664
5665 kthread_stop(thread->tsk);
5666 kfree(thread);
5667 }
5668
5669 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5670 {
5671 if (!mddev) {
5672 MD_BUG();
5673 return;
5674 }
5675
5676 if (!rdev || test_bit(Faulty, &rdev->flags))
5677 return;
5678
5679 if (mddev->external)
5680 set_bit(Blocked, &rdev->flags);
5681 /*
5682 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5683 mdname(mddev),
5684 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5685 __builtin_return_address(0),__builtin_return_address(1),
5686 __builtin_return_address(2),__builtin_return_address(3));
5687 */
5688 if (!mddev->pers)
5689 return;
5690 if (!mddev->pers->error_handler)
5691 return;
5692 mddev->pers->error_handler(mddev,rdev);
5693 if (mddev->degraded)
5694 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5695 set_bit(StateChanged, &rdev->flags);
5696 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5697 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5698 md_wakeup_thread(mddev->thread);
5699 md_new_event_inintr(mddev);
5700 }
5701
5702 /* seq_file implementation /proc/mdstat */
5703
5704 static void status_unused(struct seq_file *seq)
5705 {
5706 int i = 0;
5707 mdk_rdev_t *rdev;
5708
5709 seq_printf(seq, "unused devices: ");
5710
5711 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
5712 char b[BDEVNAME_SIZE];
5713 i++;
5714 seq_printf(seq, "%s ",
5715 bdevname(rdev->bdev,b));
5716 }
5717 if (!i)
5718 seq_printf(seq, "<none>");
5719
5720 seq_printf(seq, "\n");
5721 }
5722
5723
5724 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5725 {
5726 sector_t max_sectors, resync, res;
5727 unsigned long dt, db;
5728 sector_t rt;
5729 int scale;
5730 unsigned int per_milli;
5731
5732 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
5733
5734 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5735 max_sectors = mddev->resync_max_sectors;
5736 else
5737 max_sectors = mddev->dev_sectors;
5738
5739 /*
5740 * Should not happen.
5741 */
5742 if (!max_sectors) {
5743 MD_BUG();
5744 return;
5745 }
5746 /* Pick 'scale' such that (resync>>scale)*1000 will fit
5747 * in a sector_t, and (max_sectors>>scale) will fit in a
5748 * u32, as those are the requirements for sector_div.
5749 * Thus 'scale' must be at least 10
5750 */
5751 scale = 10;
5752 if (sizeof(sector_t) > sizeof(unsigned long)) {
5753 while ( max_sectors/2 > (1ULL<<(scale+32)))
5754 scale++;
5755 }
5756 res = (resync>>scale)*1000;
5757 sector_div(res, (u32)((max_sectors>>scale)+1));
5758
5759 per_milli = res;
5760 {
5761 int i, x = per_milli/50, y = 20-x;
5762 seq_printf(seq, "[");
5763 for (i = 0; i < x; i++)
5764 seq_printf(seq, "=");
5765 seq_printf(seq, ">");
5766 for (i = 0; i < y; i++)
5767 seq_printf(seq, ".");
5768 seq_printf(seq, "] ");
5769 }
5770 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5771 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5772 "reshape" :
5773 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5774 "check" :
5775 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5776 "resync" : "recovery"))),
5777 per_milli/10, per_milli % 10,
5778 (unsigned long long) resync/2,
5779 (unsigned long long) max_sectors/2);
5780
5781 /*
5782 * dt: time from mark until now
5783 * db: blocks written from mark until now
5784 * rt: remaining time
5785 *
5786 * rt is a sector_t, so could be 32bit or 64bit.
5787 * So we divide before multiply in case it is 32bit and close
5788 * to the limit.
5789 * We scale the divisor (db) by 32 to avoid loosing precision
5790 * near the end of resync when the number of remaining sectors
5791 * is close to 'db'.
5792 * We then divide rt by 32 after multiplying by db to compensate.
5793 * The '+1' avoids division by zero if db is very small.
5794 */
5795 dt = ((jiffies - mddev->resync_mark) / HZ);
5796 if (!dt) dt++;
5797 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5798 - mddev->resync_mark_cnt;
5799
5800 rt = max_sectors - resync; /* number of remaining sectors */
5801 sector_div(rt, db/32+1);
5802 rt *= dt;
5803 rt >>= 5;
5804
5805 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
5806 ((unsigned long)rt % 60)/6);
5807
5808 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5809 }
5810
5811 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5812 {
5813 struct list_head *tmp;
5814 loff_t l = *pos;
5815 mddev_t *mddev;
5816
5817 if (l >= 0x10000)
5818 return NULL;
5819 if (!l--)
5820 /* header */
5821 return (void*)1;
5822
5823 spin_lock(&all_mddevs_lock);
5824 list_for_each(tmp,&all_mddevs)
5825 if (!l--) {
5826 mddev = list_entry(tmp, mddev_t, all_mddevs);
5827 mddev_get(mddev);
5828 spin_unlock(&all_mddevs_lock);
5829 return mddev;
5830 }
5831 spin_unlock(&all_mddevs_lock);
5832 if (!l--)
5833 return (void*)2;/* tail */
5834 return NULL;
5835 }
5836
5837 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5838 {
5839 struct list_head *tmp;
5840 mddev_t *next_mddev, *mddev = v;
5841
5842 ++*pos;
5843 if (v == (void*)2)
5844 return NULL;
5845
5846 spin_lock(&all_mddevs_lock);
5847 if (v == (void*)1)
5848 tmp = all_mddevs.next;
5849 else
5850 tmp = mddev->all_mddevs.next;
5851 if (tmp != &all_mddevs)
5852 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5853 else {
5854 next_mddev = (void*)2;
5855 *pos = 0x10000;
5856 }
5857 spin_unlock(&all_mddevs_lock);
5858
5859 if (v != (void*)1)
5860 mddev_put(mddev);
5861 return next_mddev;
5862
5863 }
5864
5865 static void md_seq_stop(struct seq_file *seq, void *v)
5866 {
5867 mddev_t *mddev = v;
5868
5869 if (mddev && v != (void*)1 && v != (void*)2)
5870 mddev_put(mddev);
5871 }
5872
5873 struct mdstat_info {
5874 int event;
5875 };
5876
5877 static int md_seq_show(struct seq_file *seq, void *v)
5878 {
5879 mddev_t *mddev = v;
5880 sector_t sectors;
5881 mdk_rdev_t *rdev;
5882 struct mdstat_info *mi = seq->private;
5883 struct bitmap *bitmap;
5884
5885 if (v == (void*)1) {
5886 struct mdk_personality *pers;
5887 seq_printf(seq, "Personalities : ");
5888 spin_lock(&pers_lock);
5889 list_for_each_entry(pers, &pers_list, list)
5890 seq_printf(seq, "[%s] ", pers->name);
5891
5892 spin_unlock(&pers_lock);
5893 seq_printf(seq, "\n");
5894 mi->event = atomic_read(&md_event_count);
5895 return 0;
5896 }
5897 if (v == (void*)2) {
5898 status_unused(seq);
5899 return 0;
5900 }
5901
5902 if (mddev_lock(mddev) < 0)
5903 return -EINTR;
5904
5905 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5906 seq_printf(seq, "%s : %sactive", mdname(mddev),
5907 mddev->pers ? "" : "in");
5908 if (mddev->pers) {
5909 if (mddev->ro==1)
5910 seq_printf(seq, " (read-only)");
5911 if (mddev->ro==2)
5912 seq_printf(seq, " (auto-read-only)");
5913 seq_printf(seq, " %s", mddev->pers->name);
5914 }
5915
5916 sectors = 0;
5917 list_for_each_entry(rdev, &mddev->disks, same_set) {
5918 char b[BDEVNAME_SIZE];
5919 seq_printf(seq, " %s[%d]",
5920 bdevname(rdev->bdev,b), rdev->desc_nr);
5921 if (test_bit(WriteMostly, &rdev->flags))
5922 seq_printf(seq, "(W)");
5923 if (test_bit(Faulty, &rdev->flags)) {
5924 seq_printf(seq, "(F)");
5925 continue;
5926 } else if (rdev->raid_disk < 0)
5927 seq_printf(seq, "(S)"); /* spare */
5928 sectors += rdev->sectors;
5929 }
5930
5931 if (!list_empty(&mddev->disks)) {
5932 if (mddev->pers)
5933 seq_printf(seq, "\n %llu blocks",
5934 (unsigned long long)
5935 mddev->array_sectors / 2);
5936 else
5937 seq_printf(seq, "\n %llu blocks",
5938 (unsigned long long)sectors / 2);
5939 }
5940 if (mddev->persistent) {
5941 if (mddev->major_version != 0 ||
5942 mddev->minor_version != 90) {
5943 seq_printf(seq," super %d.%d",
5944 mddev->major_version,
5945 mddev->minor_version);
5946 }
5947 } else if (mddev->external)
5948 seq_printf(seq, " super external:%s",
5949 mddev->metadata_type);
5950 else
5951 seq_printf(seq, " super non-persistent");
5952
5953 if (mddev->pers) {
5954 mddev->pers->status(seq, mddev);
5955 seq_printf(seq, "\n ");
5956 if (mddev->pers->sync_request) {
5957 if (mddev->curr_resync > 2) {
5958 status_resync(seq, mddev);
5959 seq_printf(seq, "\n ");
5960 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5961 seq_printf(seq, "\tresync=DELAYED\n ");
5962 else if (mddev->recovery_cp < MaxSector)
5963 seq_printf(seq, "\tresync=PENDING\n ");
5964 }
5965 } else
5966 seq_printf(seq, "\n ");
5967
5968 if ((bitmap = mddev->bitmap)) {
5969 unsigned long chunk_kb;
5970 unsigned long flags;
5971 spin_lock_irqsave(&bitmap->lock, flags);
5972 chunk_kb = bitmap->chunksize >> 10;
5973 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5974 "%lu%s chunk",
5975 bitmap->pages - bitmap->missing_pages,
5976 bitmap->pages,
5977 (bitmap->pages - bitmap->missing_pages)
5978 << (PAGE_SHIFT - 10),
5979 chunk_kb ? chunk_kb : bitmap->chunksize,
5980 chunk_kb ? "KB" : "B");
5981 if (bitmap->file) {
5982 seq_printf(seq, ", file: ");
5983 seq_path(seq, &bitmap->file->f_path, " \t\n");
5984 }
5985
5986 seq_printf(seq, "\n");
5987 spin_unlock_irqrestore(&bitmap->lock, flags);
5988 }
5989
5990 seq_printf(seq, "\n");
5991 }
5992 mddev_unlock(mddev);
5993
5994 return 0;
5995 }
5996
5997 static const struct seq_operations md_seq_ops = {
5998 .start = md_seq_start,
5999 .next = md_seq_next,
6000 .stop = md_seq_stop,
6001 .show = md_seq_show,
6002 };
6003
6004 static int md_seq_open(struct inode *inode, struct file *file)
6005 {
6006 int error;
6007 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6008 if (mi == NULL)
6009 return -ENOMEM;
6010
6011 error = seq_open(file, &md_seq_ops);
6012 if (error)
6013 kfree(mi);
6014 else {
6015 struct seq_file *p = file->private_data;
6016 p->private = mi;
6017 mi->event = atomic_read(&md_event_count);
6018 }
6019 return error;
6020 }
6021
6022 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6023 {
6024 struct seq_file *m = filp->private_data;
6025 struct mdstat_info *mi = m->private;
6026 int mask;
6027
6028 poll_wait(filp, &md_event_waiters, wait);
6029
6030 /* always allow read */
6031 mask = POLLIN | POLLRDNORM;
6032
6033 if (mi->event != atomic_read(&md_event_count))
6034 mask |= POLLERR | POLLPRI;
6035 return mask;
6036 }
6037
6038 static const struct file_operations md_seq_fops = {
6039 .owner = THIS_MODULE,
6040 .open = md_seq_open,
6041 .read = seq_read,
6042 .llseek = seq_lseek,
6043 .release = seq_release_private,
6044 .poll = mdstat_poll,
6045 };
6046
6047 int register_md_personality(struct mdk_personality *p)
6048 {
6049 spin_lock(&pers_lock);
6050 list_add_tail(&p->list, &pers_list);
6051 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6052 spin_unlock(&pers_lock);
6053 return 0;
6054 }
6055
6056 int unregister_md_personality(struct mdk_personality *p)
6057 {
6058 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6059 spin_lock(&pers_lock);
6060 list_del_init(&p->list);
6061 spin_unlock(&pers_lock);
6062 return 0;
6063 }
6064
6065 static int is_mddev_idle(mddev_t *mddev, int init)
6066 {
6067 mdk_rdev_t * rdev;
6068 int idle;
6069 int curr_events;
6070
6071 idle = 1;
6072 rcu_read_lock();
6073 rdev_for_each_rcu(rdev, mddev) {
6074 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6075 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6076 (int)part_stat_read(&disk->part0, sectors[1]) -
6077 atomic_read(&disk->sync_io);
6078 /* sync IO will cause sync_io to increase before the disk_stats
6079 * as sync_io is counted when a request starts, and
6080 * disk_stats is counted when it completes.
6081 * So resync activity will cause curr_events to be smaller than
6082 * when there was no such activity.
6083 * non-sync IO will cause disk_stat to increase without
6084 * increasing sync_io so curr_events will (eventually)
6085 * be larger than it was before. Once it becomes
6086 * substantially larger, the test below will cause
6087 * the array to appear non-idle, and resync will slow
6088 * down.
6089 * If there is a lot of outstanding resync activity when
6090 * we set last_event to curr_events, then all that activity
6091 * completing might cause the array to appear non-idle
6092 * and resync will be slowed down even though there might
6093 * not have been non-resync activity. This will only
6094 * happen once though. 'last_events' will soon reflect
6095 * the state where there is little or no outstanding
6096 * resync requests, and further resync activity will
6097 * always make curr_events less than last_events.
6098 *
6099 */
6100 if (init || curr_events - rdev->last_events > 64) {
6101 rdev->last_events = curr_events;
6102 idle = 0;
6103 }
6104 }
6105 rcu_read_unlock();
6106 return idle;
6107 }
6108
6109 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6110 {
6111 /* another "blocks" (512byte) blocks have been synced */
6112 atomic_sub(blocks, &mddev->recovery_active);
6113 wake_up(&mddev->recovery_wait);
6114 if (!ok) {
6115 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6116 md_wakeup_thread(mddev->thread);
6117 // stop recovery, signal do_sync ....
6118 }
6119 }
6120
6121
6122 /* md_write_start(mddev, bi)
6123 * If we need to update some array metadata (e.g. 'active' flag
6124 * in superblock) before writing, schedule a superblock update
6125 * and wait for it to complete.
6126 */
6127 void md_write_start(mddev_t *mddev, struct bio *bi)
6128 {
6129 int did_change = 0;
6130 if (bio_data_dir(bi) != WRITE)
6131 return;
6132
6133 BUG_ON(mddev->ro == 1);
6134 if (mddev->ro == 2) {
6135 /* need to switch to read/write */
6136 mddev->ro = 0;
6137 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6138 md_wakeup_thread(mddev->thread);
6139 md_wakeup_thread(mddev->sync_thread);
6140 did_change = 1;
6141 }
6142 atomic_inc(&mddev->writes_pending);
6143 if (mddev->safemode == 1)
6144 mddev->safemode = 0;
6145 if (mddev->in_sync) {
6146 spin_lock_irq(&mddev->write_lock);
6147 if (mddev->in_sync) {
6148 mddev->in_sync = 0;
6149 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6150 md_wakeup_thread(mddev->thread);
6151 did_change = 1;
6152 }
6153 spin_unlock_irq(&mddev->write_lock);
6154 }
6155 if (did_change)
6156 sysfs_notify_dirent(mddev->sysfs_state);
6157 wait_event(mddev->sb_wait,
6158 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
6159 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6160 }
6161
6162 void md_write_end(mddev_t *mddev)
6163 {
6164 if (atomic_dec_and_test(&mddev->writes_pending)) {
6165 if (mddev->safemode == 2)
6166 md_wakeup_thread(mddev->thread);
6167 else if (mddev->safemode_delay)
6168 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6169 }
6170 }
6171
6172 /* md_allow_write(mddev)
6173 * Calling this ensures that the array is marked 'active' so that writes
6174 * may proceed without blocking. It is important to call this before
6175 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6176 * Must be called with mddev_lock held.
6177 *
6178 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6179 * is dropped, so return -EAGAIN after notifying userspace.
6180 */
6181 int md_allow_write(mddev_t *mddev)
6182 {
6183 if (!mddev->pers)
6184 return 0;
6185 if (mddev->ro)
6186 return 0;
6187 if (!mddev->pers->sync_request)
6188 return 0;
6189
6190 spin_lock_irq(&mddev->write_lock);
6191 if (mddev->in_sync) {
6192 mddev->in_sync = 0;
6193 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6194 if (mddev->safemode_delay &&
6195 mddev->safemode == 0)
6196 mddev->safemode = 1;
6197 spin_unlock_irq(&mddev->write_lock);
6198 md_update_sb(mddev, 0);
6199 sysfs_notify_dirent(mddev->sysfs_state);
6200 } else
6201 spin_unlock_irq(&mddev->write_lock);
6202
6203 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
6204 return -EAGAIN;
6205 else
6206 return 0;
6207 }
6208 EXPORT_SYMBOL_GPL(md_allow_write);
6209
6210 #define SYNC_MARKS 10
6211 #define SYNC_MARK_STEP (3*HZ)
6212 void md_do_sync(mddev_t *mddev)
6213 {
6214 mddev_t *mddev2;
6215 unsigned int currspeed = 0,
6216 window;
6217 sector_t max_sectors,j, io_sectors;
6218 unsigned long mark[SYNC_MARKS];
6219 sector_t mark_cnt[SYNC_MARKS];
6220 int last_mark,m;
6221 struct list_head *tmp;
6222 sector_t last_check;
6223 int skipped = 0;
6224 mdk_rdev_t *rdev;
6225 char *desc;
6226
6227 /* just incase thread restarts... */
6228 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6229 return;
6230 if (mddev->ro) /* never try to sync a read-only array */
6231 return;
6232
6233 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6234 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6235 desc = "data-check";
6236 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6237 desc = "requested-resync";
6238 else
6239 desc = "resync";
6240 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6241 desc = "reshape";
6242 else
6243 desc = "recovery";
6244
6245 /* we overload curr_resync somewhat here.
6246 * 0 == not engaged in resync at all
6247 * 2 == checking that there is no conflict with another sync
6248 * 1 == like 2, but have yielded to allow conflicting resync to
6249 * commense
6250 * other == active in resync - this many blocks
6251 *
6252 * Before starting a resync we must have set curr_resync to
6253 * 2, and then checked that every "conflicting" array has curr_resync
6254 * less than ours. When we find one that is the same or higher
6255 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6256 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6257 * This will mean we have to start checking from the beginning again.
6258 *
6259 */
6260
6261 do {
6262 mddev->curr_resync = 2;
6263
6264 try_again:
6265 if (kthread_should_stop()) {
6266 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6267 goto skip;
6268 }
6269 for_each_mddev(mddev2, tmp) {
6270 if (mddev2 == mddev)
6271 continue;
6272 if (!mddev->parallel_resync
6273 && mddev2->curr_resync
6274 && match_mddev_units(mddev, mddev2)) {
6275 DEFINE_WAIT(wq);
6276 if (mddev < mddev2 && mddev->curr_resync == 2) {
6277 /* arbitrarily yield */
6278 mddev->curr_resync = 1;
6279 wake_up(&resync_wait);
6280 }
6281 if (mddev > mddev2 && mddev->curr_resync == 1)
6282 /* no need to wait here, we can wait the next
6283 * time 'round when curr_resync == 2
6284 */
6285 continue;
6286 /* We need to wait 'interruptible' so as not to
6287 * contribute to the load average, and not to
6288 * be caught by 'softlockup'
6289 */
6290 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6291 if (!kthread_should_stop() &&
6292 mddev2->curr_resync >= mddev->curr_resync) {
6293 printk(KERN_INFO "md: delaying %s of %s"
6294 " until %s has finished (they"
6295 " share one or more physical units)\n",
6296 desc, mdname(mddev), mdname(mddev2));
6297 mddev_put(mddev2);
6298 if (signal_pending(current))
6299 flush_signals(current);
6300 schedule();
6301 finish_wait(&resync_wait, &wq);
6302 goto try_again;
6303 }
6304 finish_wait(&resync_wait, &wq);
6305 }
6306 }
6307 } while (mddev->curr_resync < 2);
6308
6309 j = 0;
6310 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6311 /* resync follows the size requested by the personality,
6312 * which defaults to physical size, but can be virtual size
6313 */
6314 max_sectors = mddev->resync_max_sectors;
6315 mddev->resync_mismatches = 0;
6316 /* we don't use the checkpoint if there's a bitmap */
6317 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6318 j = mddev->resync_min;
6319 else if (!mddev->bitmap)
6320 j = mddev->recovery_cp;
6321
6322 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6323 max_sectors = mddev->dev_sectors;
6324 else {
6325 /* recovery follows the physical size of devices */
6326 max_sectors = mddev->dev_sectors;
6327 j = MaxSector;
6328 list_for_each_entry(rdev, &mddev->disks, same_set)
6329 if (rdev->raid_disk >= 0 &&
6330 !test_bit(Faulty, &rdev->flags) &&
6331 !test_bit(In_sync, &rdev->flags) &&
6332 rdev->recovery_offset < j)
6333 j = rdev->recovery_offset;
6334 }
6335
6336 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6337 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6338 " %d KB/sec/disk.\n", speed_min(mddev));
6339 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6340 "(but not more than %d KB/sec) for %s.\n",
6341 speed_max(mddev), desc);
6342
6343 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6344
6345 io_sectors = 0;
6346 for (m = 0; m < SYNC_MARKS; m++) {
6347 mark[m] = jiffies;
6348 mark_cnt[m] = io_sectors;
6349 }
6350 last_mark = 0;
6351 mddev->resync_mark = mark[last_mark];
6352 mddev->resync_mark_cnt = mark_cnt[last_mark];
6353
6354 /*
6355 * Tune reconstruction:
6356 */
6357 window = 32*(PAGE_SIZE/512);
6358 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6359 window/2,(unsigned long long) max_sectors/2);
6360
6361 atomic_set(&mddev->recovery_active, 0);
6362 last_check = 0;
6363
6364 if (j>2) {
6365 printk(KERN_INFO
6366 "md: resuming %s of %s from checkpoint.\n",
6367 desc, mdname(mddev));
6368 mddev->curr_resync = j;
6369 }
6370
6371 while (j < max_sectors) {
6372 sector_t sectors;
6373
6374 skipped = 0;
6375
6376 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6377 ((mddev->curr_resync > mddev->curr_resync_completed &&
6378 (mddev->curr_resync - mddev->curr_resync_completed)
6379 > (max_sectors >> 4)) ||
6380 (j - mddev->curr_resync_completed)*2
6381 >= mddev->resync_max - mddev->curr_resync_completed
6382 )) {
6383 /* time to update curr_resync_completed */
6384 blk_unplug(mddev->queue);
6385 wait_event(mddev->recovery_wait,
6386 atomic_read(&mddev->recovery_active) == 0);
6387 mddev->curr_resync_completed =
6388 mddev->curr_resync;
6389 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6390 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6391 }
6392
6393 if (j >= mddev->resync_max)
6394 wait_event(mddev->recovery_wait,
6395 mddev->resync_max > j
6396 || kthread_should_stop());
6397
6398 if (kthread_should_stop())
6399 goto interrupted;
6400
6401 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6402 currspeed < speed_min(mddev));
6403 if (sectors == 0) {
6404 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6405 goto out;
6406 }
6407
6408 if (!skipped) { /* actual IO requested */
6409 io_sectors += sectors;
6410 atomic_add(sectors, &mddev->recovery_active);
6411 }
6412
6413 j += sectors;
6414 if (j>1) mddev->curr_resync = j;
6415 mddev->curr_mark_cnt = io_sectors;
6416 if (last_check == 0)
6417 /* this is the earliers that rebuilt will be
6418 * visible in /proc/mdstat
6419 */
6420 md_new_event(mddev);
6421
6422 if (last_check + window > io_sectors || j == max_sectors)
6423 continue;
6424
6425 last_check = io_sectors;
6426
6427 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6428 break;
6429
6430 repeat:
6431 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6432 /* step marks */
6433 int next = (last_mark+1) % SYNC_MARKS;
6434
6435 mddev->resync_mark = mark[next];
6436 mddev->resync_mark_cnt = mark_cnt[next];
6437 mark[next] = jiffies;
6438 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6439 last_mark = next;
6440 }
6441
6442
6443 if (kthread_should_stop())
6444 goto interrupted;
6445
6446
6447 /*
6448 * this loop exits only if either when we are slower than
6449 * the 'hard' speed limit, or the system was IO-idle for
6450 * a jiffy.
6451 * the system might be non-idle CPU-wise, but we only care
6452 * about not overloading the IO subsystem. (things like an
6453 * e2fsck being done on the RAID array should execute fast)
6454 */
6455 blk_unplug(mddev->queue);
6456 cond_resched();
6457
6458 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6459 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6460
6461 if (currspeed > speed_min(mddev)) {
6462 if ((currspeed > speed_max(mddev)) ||
6463 !is_mddev_idle(mddev, 0)) {
6464 msleep(500);
6465 goto repeat;
6466 }
6467 }
6468 }
6469 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6470 /*
6471 * this also signals 'finished resyncing' to md_stop
6472 */
6473 out:
6474 blk_unplug(mddev->queue);
6475
6476 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6477
6478 /* tell personality that we are finished */
6479 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6480
6481 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6482 mddev->curr_resync > 2) {
6483 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6484 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6485 if (mddev->curr_resync >= mddev->recovery_cp) {
6486 printk(KERN_INFO
6487 "md: checkpointing %s of %s.\n",
6488 desc, mdname(mddev));
6489 mddev->recovery_cp = mddev->curr_resync;
6490 }
6491 } else
6492 mddev->recovery_cp = MaxSector;
6493 } else {
6494 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6495 mddev->curr_resync = MaxSector;
6496 list_for_each_entry(rdev, &mddev->disks, same_set)
6497 if (rdev->raid_disk >= 0 &&
6498 !test_bit(Faulty, &rdev->flags) &&
6499 !test_bit(In_sync, &rdev->flags) &&
6500 rdev->recovery_offset < mddev->curr_resync)
6501 rdev->recovery_offset = mddev->curr_resync;
6502 }
6503 }
6504 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6505
6506 skip:
6507 mddev->curr_resync = 0;
6508 mddev->curr_resync_completed = 0;
6509 mddev->resync_min = 0;
6510 mddev->resync_max = MaxSector;
6511 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6512 wake_up(&resync_wait);
6513 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6514 md_wakeup_thread(mddev->thread);
6515 return;
6516
6517 interrupted:
6518 /*
6519 * got a signal, exit.
6520 */
6521 printk(KERN_INFO
6522 "md: md_do_sync() got signal ... exiting\n");
6523 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6524 goto out;
6525
6526 }
6527 EXPORT_SYMBOL_GPL(md_do_sync);
6528
6529
6530 static int remove_and_add_spares(mddev_t *mddev)
6531 {
6532 mdk_rdev_t *rdev;
6533 int spares = 0;
6534
6535 mddev->curr_resync_completed = 0;
6536
6537 list_for_each_entry(rdev, &mddev->disks, same_set)
6538 if (rdev->raid_disk >= 0 &&
6539 !test_bit(Blocked, &rdev->flags) &&
6540 (test_bit(Faulty, &rdev->flags) ||
6541 ! test_bit(In_sync, &rdev->flags)) &&
6542 atomic_read(&rdev->nr_pending)==0) {
6543 if (mddev->pers->hot_remove_disk(
6544 mddev, rdev->raid_disk)==0) {
6545 char nm[20];
6546 sprintf(nm,"rd%d", rdev->raid_disk);
6547 sysfs_remove_link(&mddev->kobj, nm);
6548 rdev->raid_disk = -1;
6549 }
6550 }
6551
6552 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
6553 list_for_each_entry(rdev, &mddev->disks, same_set) {
6554 if (rdev->raid_disk >= 0 &&
6555 !test_bit(In_sync, &rdev->flags) &&
6556 !test_bit(Blocked, &rdev->flags))
6557 spares++;
6558 if (rdev->raid_disk < 0
6559 && !test_bit(Faulty, &rdev->flags)) {
6560 rdev->recovery_offset = 0;
6561 if (mddev->pers->
6562 hot_add_disk(mddev, rdev) == 0) {
6563 char nm[20];
6564 sprintf(nm, "rd%d", rdev->raid_disk);
6565 if (sysfs_create_link(&mddev->kobj,
6566 &rdev->kobj, nm))
6567 printk(KERN_WARNING
6568 "md: cannot register "
6569 "%s for %s\n",
6570 nm, mdname(mddev));
6571 spares++;
6572 md_new_event(mddev);
6573 } else
6574 break;
6575 }
6576 }
6577 }
6578 return spares;
6579 }
6580 /*
6581 * This routine is regularly called by all per-raid-array threads to
6582 * deal with generic issues like resync and super-block update.
6583 * Raid personalities that don't have a thread (linear/raid0) do not
6584 * need this as they never do any recovery or update the superblock.
6585 *
6586 * It does not do any resync itself, but rather "forks" off other threads
6587 * to do that as needed.
6588 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6589 * "->recovery" and create a thread at ->sync_thread.
6590 * When the thread finishes it sets MD_RECOVERY_DONE
6591 * and wakeups up this thread which will reap the thread and finish up.
6592 * This thread also removes any faulty devices (with nr_pending == 0).
6593 *
6594 * The overall approach is:
6595 * 1/ if the superblock needs updating, update it.
6596 * 2/ If a recovery thread is running, don't do anything else.
6597 * 3/ If recovery has finished, clean up, possibly marking spares active.
6598 * 4/ If there are any faulty devices, remove them.
6599 * 5/ If array is degraded, try to add spares devices
6600 * 6/ If array has spares or is not in-sync, start a resync thread.
6601 */
6602 void md_check_recovery(mddev_t *mddev)
6603 {
6604 mdk_rdev_t *rdev;
6605
6606
6607 if (mddev->bitmap)
6608 bitmap_daemon_work(mddev->bitmap);
6609
6610 if (mddev->ro)
6611 return;
6612
6613 if (signal_pending(current)) {
6614 if (mddev->pers->sync_request && !mddev->external) {
6615 printk(KERN_INFO "md: %s in immediate safe mode\n",
6616 mdname(mddev));
6617 mddev->safemode = 2;
6618 }
6619 flush_signals(current);
6620 }
6621
6622 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
6623 return;
6624 if ( ! (
6625 (mddev->flags && !mddev->external) ||
6626 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
6627 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
6628 (mddev->external == 0 && mddev->safemode == 1) ||
6629 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6630 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6631 ))
6632 return;
6633
6634 if (mddev_trylock(mddev)) {
6635 int spares = 0;
6636
6637 if (mddev->ro) {
6638 /* Only thing we do on a ro array is remove
6639 * failed devices.
6640 */
6641 remove_and_add_spares(mddev);
6642 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6643 goto unlock;
6644 }
6645
6646 if (!mddev->external) {
6647 int did_change = 0;
6648 spin_lock_irq(&mddev->write_lock);
6649 if (mddev->safemode &&
6650 !atomic_read(&mddev->writes_pending) &&
6651 !mddev->in_sync &&
6652 mddev->recovery_cp == MaxSector) {
6653 mddev->in_sync = 1;
6654 did_change = 1;
6655 if (mddev->persistent)
6656 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6657 }
6658 if (mddev->safemode == 1)
6659 mddev->safemode = 0;
6660 spin_unlock_irq(&mddev->write_lock);
6661 if (did_change)
6662 sysfs_notify_dirent(mddev->sysfs_state);
6663 }
6664
6665 if (mddev->flags)
6666 md_update_sb(mddev, 0);
6667
6668 list_for_each_entry(rdev, &mddev->disks, same_set)
6669 if (test_and_clear_bit(StateChanged, &rdev->flags))
6670 sysfs_notify_dirent(rdev->sysfs_state);
6671
6672
6673 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6674 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6675 /* resync/recovery still happening */
6676 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6677 goto unlock;
6678 }
6679 if (mddev->sync_thread) {
6680 /* resync has finished, collect result */
6681 md_unregister_thread(mddev->sync_thread);
6682 mddev->sync_thread = NULL;
6683 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
6684 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
6685 /* success...*/
6686 /* activate any spares */
6687 if (mddev->pers->spare_active(mddev))
6688 sysfs_notify(&mddev->kobj, NULL,
6689 "degraded");
6690 }
6691 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6692 mddev->pers->finish_reshape)
6693 mddev->pers->finish_reshape(mddev);
6694 md_update_sb(mddev, 1);
6695
6696 /* if array is no-longer degraded, then any saved_raid_disk
6697 * information must be scrapped
6698 */
6699 if (!mddev->degraded)
6700 list_for_each_entry(rdev, &mddev->disks, same_set)
6701 rdev->saved_raid_disk = -1;
6702
6703 mddev->recovery = 0;
6704 /* flag recovery needed just to double check */
6705 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6706 sysfs_notify_dirent(mddev->sysfs_action);
6707 md_new_event(mddev);
6708 goto unlock;
6709 }
6710 /* Set RUNNING before clearing NEEDED to avoid
6711 * any transients in the value of "sync_action".
6712 */
6713 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6714 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6715 /* Clear some bits that don't mean anything, but
6716 * might be left set
6717 */
6718 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
6719 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
6720
6721 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6722 goto unlock;
6723 /* no recovery is running.
6724 * remove any failed drives, then
6725 * add spares if possible.
6726 * Spare are also removed and re-added, to allow
6727 * the personality to fail the re-add.
6728 */
6729
6730 if (mddev->reshape_position != MaxSector) {
6731 if (mddev->pers->check_reshape(mddev) != 0)
6732 /* Cannot proceed */
6733 goto unlock;
6734 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6735 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6736 } else if ((spares = remove_and_add_spares(mddev))) {
6737 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6738 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6739 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
6740 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6741 } else if (mddev->recovery_cp < MaxSector) {
6742 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6743 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6744 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6745 /* nothing to be done ... */
6746 goto unlock;
6747
6748 if (mddev->pers->sync_request) {
6749 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6750 /* We are adding a device or devices to an array
6751 * which has the bitmap stored on all devices.
6752 * So make sure all bitmap pages get written
6753 */
6754 bitmap_write_all(mddev->bitmap);
6755 }
6756 mddev->sync_thread = md_register_thread(md_do_sync,
6757 mddev,
6758 "%s_resync");
6759 if (!mddev->sync_thread) {
6760 printk(KERN_ERR "%s: could not start resync"
6761 " thread...\n",
6762 mdname(mddev));
6763 /* leave the spares where they are, it shouldn't hurt */
6764 mddev->recovery = 0;
6765 } else
6766 md_wakeup_thread(mddev->sync_thread);
6767 sysfs_notify_dirent(mddev->sysfs_action);
6768 md_new_event(mddev);
6769 }
6770 unlock:
6771 if (!mddev->sync_thread) {
6772 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6773 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
6774 &mddev->recovery))
6775 if (mddev->sysfs_action)
6776 sysfs_notify_dirent(mddev->sysfs_action);
6777 }
6778 mddev_unlock(mddev);
6779 }
6780 }
6781
6782 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
6783 {
6784 sysfs_notify_dirent(rdev->sysfs_state);
6785 wait_event_timeout(rdev->blocked_wait,
6786 !test_bit(Blocked, &rdev->flags),
6787 msecs_to_jiffies(5000));
6788 rdev_dec_pending(rdev, mddev);
6789 }
6790 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
6791
6792 static int md_notify_reboot(struct notifier_block *this,
6793 unsigned long code, void *x)
6794 {
6795 struct list_head *tmp;
6796 mddev_t *mddev;
6797
6798 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
6799
6800 printk(KERN_INFO "md: stopping all md devices.\n");
6801
6802 for_each_mddev(mddev, tmp)
6803 if (mddev_trylock(mddev)) {
6804 /* Force a switch to readonly even array
6805 * appears to still be in use. Hence
6806 * the '100'.
6807 */
6808 do_md_stop(mddev, 1, 100);
6809 mddev_unlock(mddev);
6810 }
6811 /*
6812 * certain more exotic SCSI devices are known to be
6813 * volatile wrt too early system reboots. While the
6814 * right place to handle this issue is the given
6815 * driver, we do want to have a safe RAID driver ...
6816 */
6817 mdelay(1000*1);
6818 }
6819 return NOTIFY_DONE;
6820 }
6821
6822 static struct notifier_block md_notifier = {
6823 .notifier_call = md_notify_reboot,
6824 .next = NULL,
6825 .priority = INT_MAX, /* before any real devices */
6826 };
6827
6828 static void md_geninit(void)
6829 {
6830 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
6831
6832 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
6833 }
6834
6835 static int __init md_init(void)
6836 {
6837 if (register_blkdev(MD_MAJOR, "md"))
6838 return -1;
6839 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
6840 unregister_blkdev(MD_MAJOR, "md");
6841 return -1;
6842 }
6843 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
6844 md_probe, NULL, NULL);
6845 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
6846 md_probe, NULL, NULL);
6847
6848 register_reboot_notifier(&md_notifier);
6849 raid_table_header = register_sysctl_table(raid_root_table);
6850
6851 md_geninit();
6852 return 0;
6853 }
6854
6855
6856 #ifndef MODULE
6857
6858 /*
6859 * Searches all registered partitions for autorun RAID arrays
6860 * at boot time.
6861 */
6862
6863 static LIST_HEAD(all_detected_devices);
6864 struct detected_devices_node {
6865 struct list_head list;
6866 dev_t dev;
6867 };
6868
6869 void md_autodetect_dev(dev_t dev)
6870 {
6871 struct detected_devices_node *node_detected_dev;
6872
6873 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6874 if (node_detected_dev) {
6875 node_detected_dev->dev = dev;
6876 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6877 } else {
6878 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6879 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6880 }
6881 }
6882
6883
6884 static void autostart_arrays(int part)
6885 {
6886 mdk_rdev_t *rdev;
6887 struct detected_devices_node *node_detected_dev;
6888 dev_t dev;
6889 int i_scanned, i_passed;
6890
6891 i_scanned = 0;
6892 i_passed = 0;
6893
6894 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6895
6896 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6897 i_scanned++;
6898 node_detected_dev = list_entry(all_detected_devices.next,
6899 struct detected_devices_node, list);
6900 list_del(&node_detected_dev->list);
6901 dev = node_detected_dev->dev;
6902 kfree(node_detected_dev);
6903 rdev = md_import_device(dev,0, 90);
6904 if (IS_ERR(rdev))
6905 continue;
6906
6907 if (test_bit(Faulty, &rdev->flags)) {
6908 MD_BUG();
6909 continue;
6910 }
6911 set_bit(AutoDetected, &rdev->flags);
6912 list_add(&rdev->same_set, &pending_raid_disks);
6913 i_passed++;
6914 }
6915
6916 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6917 i_scanned, i_passed);
6918
6919 autorun_devices(part);
6920 }
6921
6922 #endif /* !MODULE */
6923
6924 static __exit void md_exit(void)
6925 {
6926 mddev_t *mddev;
6927 struct list_head *tmp;
6928
6929 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
6930 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6931
6932 unregister_blkdev(MD_MAJOR,"md");
6933 unregister_blkdev(mdp_major, "mdp");
6934 unregister_reboot_notifier(&md_notifier);
6935 unregister_sysctl_table(raid_table_header);
6936 remove_proc_entry("mdstat", NULL);
6937 for_each_mddev(mddev, tmp) {
6938 export_array(mddev);
6939 mddev->hold_active = 0;
6940 }
6941 }
6942
6943 subsys_initcall(md_init);
6944 module_exit(md_exit)
6945
6946 static int get_ro(char *buffer, struct kernel_param *kp)
6947 {
6948 return sprintf(buffer, "%d", start_readonly);
6949 }
6950 static int set_ro(const char *val, struct kernel_param *kp)
6951 {
6952 char *e;
6953 int num = simple_strtoul(val, &e, 10);
6954 if (*val && (*e == '\0' || *e == '\n')) {
6955 start_readonly = num;
6956 return 0;
6957 }
6958 return -EINVAL;
6959 }
6960
6961 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6962 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6963
6964 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
6965
6966 EXPORT_SYMBOL(register_md_personality);
6967 EXPORT_SYMBOL(unregister_md_personality);
6968 EXPORT_SYMBOL(md_error);
6969 EXPORT_SYMBOL(md_done_sync);
6970 EXPORT_SYMBOL(md_write_start);
6971 EXPORT_SYMBOL(md_write_end);
6972 EXPORT_SYMBOL(md_register_thread);
6973 EXPORT_SYMBOL(md_unregister_thread);
6974 EXPORT_SYMBOL(md_wakeup_thread);
6975 EXPORT_SYMBOL(md_check_recovery);
6976 MODULE_LICENSE("GPL");
6977 MODULE_ALIAS("md");
6978 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree