?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
thinkpad-acpi: drop HKEY event 0x5010
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / md / md.c
blob2938e9ca711e989ce6f086ecf302f55d1e6a18cb
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->ctime == 0 && !mddev->hold_active) {
280 /* Array is not configured at all, and not held active,
281 * so destroy it */
282 list_del(&mddev->all_mddevs);
283 if (mddev->gendisk) {
284 /* we did a probe so need to clean up.
285 * Call schedule_work inside the spinlock
286 * so that flush_scheduled_work() after
287 * mddev_find will succeed in waiting for the
288 * work to be done.
289 */
290 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
291 schedule_work(&mddev->del_work);
292 } else
293 kfree(mddev);
294 }
295 spin_unlock(&all_mddevs_lock);
296 }
297
298 static mddev_t * mddev_find(dev_t unit)
299 {
300 mddev_t *mddev, *new = NULL;
301
302 retry:
303 spin_lock(&all_mddevs_lock);
304
305 if (unit) {
306 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
307 if (mddev->unit == unit) {
308 mddev_get(mddev);
309 spin_unlock(&all_mddevs_lock);
310 kfree(new);
311 return mddev;
312 }
313
314 if (new) {
315 list_add(&new->all_mddevs, &all_mddevs);
316 spin_unlock(&all_mddevs_lock);
317 new->hold_active = UNTIL_IOCTL;
318 return new;
319 }
320 } else if (new) {
321 /* find an unused unit number */
322 static int next_minor = 512;
323 int start = next_minor;
324 int is_free = 0;
325 int dev = 0;
326 while (!is_free) {
327 dev = MKDEV(MD_MAJOR, next_minor);
328 next_minor++;
329 if (next_minor > MINORMASK)
330 next_minor = 0;
331 if (next_minor == start) {
332 /* Oh dear, all in use. */
333 spin_unlock(&all_mddevs_lock);
334 kfree(new);
335 return NULL;
336 }
337
338 is_free = 1;
339 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
340 if (mddev->unit == dev) {
341 is_free = 0;
342 break;
343 }
344 }
345 new->unit = dev;
346 new->md_minor = MINOR(dev);
347 new->hold_active = UNTIL_STOP;
348 list_add(&new->all_mddevs, &all_mddevs);
349 spin_unlock(&all_mddevs_lock);
350 return new;
351 }
352 spin_unlock(&all_mddevs_lock);
353
354 new = kzalloc(sizeof(*new), GFP_KERNEL);
355 if (!new)
356 return NULL;
357
358 new->unit = unit;
359 if (MAJOR(unit) == MD_MAJOR)
360 new->md_minor = MINOR(unit);
361 else
362 new->md_minor = MINOR(unit) >> MdpMinorShift;
363
364 mutex_init(&new->open_mutex);
365 mutex_init(&new->reconfig_mutex);
366 mutex_init(&new->bitmap_mutex);
367 INIT_LIST_HEAD(&new->disks);
368 INIT_LIST_HEAD(&new->all_mddevs);
369 init_timer(&new->safemode_timer);
370 atomic_set(&new->active, 1);
371 atomic_set(&new->openers, 0);
372 atomic_set(&new->active_io, 0);
373 spin_lock_init(&new->write_lock);
374 init_waitqueue_head(&new->sb_wait);
375 init_waitqueue_head(&new->recovery_wait);
376 new->reshape_position = MaxSector;
377 new->resync_min = 0;
378 new->resync_max = MaxSector;
379 new->level = LEVEL_NONE;
380
381 goto retry;
382 }
383
384 static inline int mddev_lock(mddev_t * mddev)
385 {
386 return mutex_lock_interruptible(&mddev->reconfig_mutex);
387 }
388
389 static inline int mddev_is_locked(mddev_t *mddev)
390 {
391 return mutex_is_locked(&mddev->reconfig_mutex);
392 }
393
394 static inline int mddev_trylock(mddev_t * mddev)
395 {
396 return mutex_trylock(&mddev->reconfig_mutex);
397 }
398
399 static inline void mddev_unlock(mddev_t * mddev)
400 {
401 mutex_unlock(&mddev->reconfig_mutex);
402
403 md_wakeup_thread(mddev->thread);
404 }
405
406 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
407 {
408 mdk_rdev_t *rdev;
409
410 list_for_each_entry(rdev, &mddev->disks, same_set)
411 if (rdev->desc_nr == nr)
412 return rdev;
413
414 return NULL;
415 }
416
417 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
418 {
419 mdk_rdev_t *rdev;
420
421 list_for_each_entry(rdev, &mddev->disks, same_set)
422 if (rdev->bdev->bd_dev == dev)
423 return rdev;
424
425 return NULL;
426 }
427
428 static struct mdk_personality *find_pers(int level, char *clevel)
429 {
430 struct mdk_personality *pers;
431 list_for_each_entry(pers, &pers_list, list) {
432 if (level != LEVEL_NONE && pers->level == level)
433 return pers;
434 if (strcmp(pers->name, clevel)==0)
435 return pers;
436 }
437 return NULL;
438 }
439
440 /* return the offset of the super block in 512byte sectors */
441 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
442 {
443 sector_t num_sectors = bdev->bd_inode->i_size / 512;
444 return MD_NEW_SIZE_SECTORS(num_sectors);
445 }
446
447 static int alloc_disk_sb(mdk_rdev_t * rdev)
448 {
449 if (rdev->sb_page)
450 MD_BUG();
451
452 rdev->sb_page = alloc_page(GFP_KERNEL);
453 if (!rdev->sb_page) {
454 printk(KERN_ALERT "md: out of memory.\n");
455 return -ENOMEM;
456 }
457
458 return 0;
459 }
460
461 static void free_disk_sb(mdk_rdev_t * rdev)
462 {
463 if (rdev->sb_page) {
464 put_page(rdev->sb_page);
465 rdev->sb_loaded = 0;
466 rdev->sb_page = NULL;
467 rdev->sb_start = 0;
468 rdev->sectors = 0;
469 }
470 }
471
472
473 static void super_written(struct bio *bio, int error)
474 {
475 mdk_rdev_t *rdev = bio->bi_private;
476 mddev_t *mddev = rdev->mddev;
477
478 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
479 printk("md: super_written gets error=%d, uptodate=%d\n",
480 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
481 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
482 md_error(mddev, rdev);
483 }
484
485 if (atomic_dec_and_test(&mddev->pending_writes))
486 wake_up(&mddev->sb_wait);
487 bio_put(bio);
488 }
489
490 static void super_written_barrier(struct bio *bio, int error)
491 {
492 struct bio *bio2 = bio->bi_private;
493 mdk_rdev_t *rdev = bio2->bi_private;
494 mddev_t *mddev = rdev->mddev;
495
496 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
497 error == -EOPNOTSUPP) {
498 unsigned long flags;
499 /* barriers don't appear to be supported :-( */
500 set_bit(BarriersNotsupp, &rdev->flags);
501 mddev->barriers_work = 0;
502 spin_lock_irqsave(&mddev->write_lock, flags);
503 bio2->bi_next = mddev->biolist;
504 mddev->biolist = bio2;
505 spin_unlock_irqrestore(&mddev->write_lock, flags);
506 wake_up(&mddev->sb_wait);
507 bio_put(bio);
508 } else {
509 bio_put(bio2);
510 bio->bi_private = rdev;
511 super_written(bio, error);
512 }
513 }
514
515 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
516 sector_t sector, int size, struct page *page)
517 {
518 /* write first size bytes of page to sector of rdev
519 * Increment mddev->pending_writes before returning
520 * and decrement it on completion, waking up sb_wait
521 * if zero is reached.
522 * If an error occurred, call md_error
523 *
524 * As we might need to resubmit the request if BIO_RW_BARRIER
525 * causes ENOTSUPP, we allocate a spare bio...
526 */
527 struct bio *bio = bio_alloc(GFP_NOIO, 1);
528 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG);
529
530 bio->bi_bdev = rdev->bdev;
531 bio->bi_sector = sector;
532 bio_add_page(bio, page, size, 0);
533 bio->bi_private = rdev;
534 bio->bi_end_io = super_written;
535 bio->bi_rw = rw;
536
537 atomic_inc(&mddev->pending_writes);
538 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
539 struct bio *rbio;
540 rw |= (1<<BIO_RW_BARRIER);
541 rbio = bio_clone(bio, GFP_NOIO);
542 rbio->bi_private = bio;
543 rbio->bi_end_io = super_written_barrier;
544 submit_bio(rw, rbio);
545 } else
546 submit_bio(rw, bio);
547 }
548
549 void md_super_wait(mddev_t *mddev)
550 {
551 /* wait for all superblock writes that were scheduled to complete.
552 * if any had to be retried (due to BARRIER problems), retry them
553 */
554 DEFINE_WAIT(wq);
555 for(;;) {
556 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
557 if (atomic_read(&mddev->pending_writes)==0)
558 break;
559 while (mddev->biolist) {
560 struct bio *bio;
561 spin_lock_irq(&mddev->write_lock);
562 bio = mddev->biolist;
563 mddev->biolist = bio->bi_next ;
564 bio->bi_next = NULL;
565 spin_unlock_irq(&mddev->write_lock);
566 submit_bio(bio->bi_rw, bio);
567 }
568 schedule();
569 }
570 finish_wait(&mddev->sb_wait, &wq);
571 }
572
573 static void bi_complete(struct bio *bio, int error)
574 {
575 complete((struct completion*)bio->bi_private);
576 }
577
578 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
579 struct page *page, int rw)
580 {
581 struct bio *bio = bio_alloc(GFP_NOIO, 1);
582 struct completion event;
583 int ret;
584
585 rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
586
587 bio->bi_bdev = bdev;
588 bio->bi_sector = sector;
589 bio_add_page(bio, page, size, 0);
590 init_completion(&event);
591 bio->bi_private = &event;
592 bio->bi_end_io = bi_complete;
593 submit_bio(rw, bio);
594 wait_for_completion(&event);
595
596 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
597 bio_put(bio);
598 return ret;
599 }
600 EXPORT_SYMBOL_GPL(sync_page_io);
601
602 static int read_disk_sb(mdk_rdev_t * rdev, int size)
603 {
604 char b[BDEVNAME_SIZE];
605 if (!rdev->sb_page) {
606 MD_BUG();
607 return -EINVAL;
608 }
609 if (rdev->sb_loaded)
610 return 0;
611
612
613 if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
614 goto fail;
615 rdev->sb_loaded = 1;
616 return 0;
617
618 fail:
619 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
620 bdevname(rdev->bdev,b));
621 return -EINVAL;
622 }
623
624 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
625 {
626 return sb1->set_uuid0 == sb2->set_uuid0 &&
627 sb1->set_uuid1 == sb2->set_uuid1 &&
628 sb1->set_uuid2 == sb2->set_uuid2 &&
629 sb1->set_uuid3 == sb2->set_uuid3;
630 }
631
632 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
633 {
634 int ret;
635 mdp_super_t *tmp1, *tmp2;
636
637 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
638 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
639
640 if (!tmp1 || !tmp2) {
641 ret = 0;
642 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
643 goto abort;
644 }
645
646 *tmp1 = *sb1;
647 *tmp2 = *sb2;
648
649 /*
650 * nr_disks is not constant
651 */
652 tmp1->nr_disks = 0;
653 tmp2->nr_disks = 0;
654
655 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
656 abort:
657 kfree(tmp1);
658 kfree(tmp2);
659 return ret;
660 }
661
662
663 static u32 md_csum_fold(u32 csum)
664 {
665 csum = (csum & 0xffff) + (csum >> 16);
666 return (csum & 0xffff) + (csum >> 16);
667 }
668
669 static unsigned int calc_sb_csum(mdp_super_t * sb)
670 {
671 u64 newcsum = 0;
672 u32 *sb32 = (u32*)sb;
673 int i;
674 unsigned int disk_csum, csum;
675
676 disk_csum = sb->sb_csum;
677 sb->sb_csum = 0;
678
679 for (i = 0; i < MD_SB_BYTES/4 ; i++)
680 newcsum += sb32[i];
681 csum = (newcsum & 0xffffffff) + (newcsum>>32);
682
683
684 #ifdef CONFIG_ALPHA
685 /* This used to use csum_partial, which was wrong for several
686 * reasons including that different results are returned on
687 * different architectures. It isn't critical that we get exactly
688 * the same return value as before (we always csum_fold before
689 * testing, and that removes any differences). However as we
690 * know that csum_partial always returned a 16bit value on
691 * alphas, do a fold to maximise conformity to previous behaviour.
692 */
693 sb->sb_csum = md_csum_fold(disk_csum);
694 #else
695 sb->sb_csum = disk_csum;
696 #endif
697 return csum;
698 }
699
700
701 /*
702 * Handle superblock details.
703 * We want to be able to handle multiple superblock formats
704 * so we have a common interface to them all, and an array of
705 * different handlers.
706 * We rely on user-space to write the initial superblock, and support
707 * reading and updating of superblocks.
708 * Interface methods are:
709 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
710 * loads and validates a superblock on dev.
711 * if refdev != NULL, compare superblocks on both devices
712 * Return:
713 * 0 - dev has a superblock that is compatible with refdev
714 * 1 - dev has a superblock that is compatible and newer than refdev
715 * so dev should be used as the refdev in future
716 * -EINVAL superblock incompatible or invalid
717 * -othererror e.g. -EIO
718 *
719 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
720 * Verify that dev is acceptable into mddev.
721 * The first time, mddev->raid_disks will be 0, and data from
722 * dev should be merged in. Subsequent calls check that dev
723 * is new enough. Return 0 or -EINVAL
724 *
725 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
726 * Update the superblock for rdev with data in mddev
727 * This does not write to disc.
728 *
729 */
730
731 struct super_type {
732 char *name;
733 struct module *owner;
734 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
735 int minor_version);
736 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
737 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
738 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
739 sector_t num_sectors);
740 };
741
742 /*
743 * Check that the given mddev has no bitmap.
744 *
745 * This function is called from the run method of all personalities that do not
746 * support bitmaps. It prints an error message and returns non-zero if mddev
747 * has a bitmap. Otherwise, it returns 0.
748 *
749 */
750 int md_check_no_bitmap(mddev_t *mddev)
751 {
752 if (!mddev->bitmap_file && !mddev->bitmap_offset)
753 return 0;
754 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
755 mdname(mddev), mddev->pers->name);
756 return 1;
757 }
758 EXPORT_SYMBOL(md_check_no_bitmap);
759
760 /*
761 * load_super for 0.90.0
762 */
763 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
764 {
765 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
766 mdp_super_t *sb;
767 int ret;
768
769 /*
770 * Calculate the position of the superblock (512byte sectors),
771 * it's at the end of the disk.
772 *
773 * It also happens to be a multiple of 4Kb.
774 */
775 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
776
777 ret = read_disk_sb(rdev, MD_SB_BYTES);
778 if (ret) return ret;
779
780 ret = -EINVAL;
781
782 bdevname(rdev->bdev, b);
783 sb = (mdp_super_t*)page_address(rdev->sb_page);
784
785 if (sb->md_magic != MD_SB_MAGIC) {
786 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
787 b);
788 goto abort;
789 }
790
791 if (sb->major_version != 0 ||
792 sb->minor_version < 90 ||
793 sb->minor_version > 91) {
794 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
795 sb->major_version, sb->minor_version,
796 b);
797 goto abort;
798 }
799
800 if (sb->raid_disks <= 0)
801 goto abort;
802
803 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
804 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
805 b);
806 goto abort;
807 }
808
809 rdev->preferred_minor = sb->md_minor;
810 rdev->data_offset = 0;
811 rdev->sb_size = MD_SB_BYTES;
812
813 if (sb->level == LEVEL_MULTIPATH)
814 rdev->desc_nr = -1;
815 else
816 rdev->desc_nr = sb->this_disk.number;
817
818 if (!refdev) {
819 ret = 1;
820 } else {
821 __u64 ev1, ev2;
822 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
823 if (!uuid_equal(refsb, sb)) {
824 printk(KERN_WARNING "md: %s has different UUID to %s\n",
825 b, bdevname(refdev->bdev,b2));
826 goto abort;
827 }
828 if (!sb_equal(refsb, sb)) {
829 printk(KERN_WARNING "md: %s has same UUID"
830 " but different superblock to %s\n",
831 b, bdevname(refdev->bdev, b2));
832 goto abort;
833 }
834 ev1 = md_event(sb);
835 ev2 = md_event(refsb);
836 if (ev1 > ev2)
837 ret = 1;
838 else
839 ret = 0;
840 }
841 rdev->sectors = rdev->sb_start;
842
843 if (rdev->sectors < sb->size * 2 && sb->level > 1)
844 /* "this cannot possibly happen" ... */
845 ret = -EINVAL;
846
847 abort:
848 return ret;
849 }
850
851 /*
852 * validate_super for 0.90.0
853 */
854 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
855 {
856 mdp_disk_t *desc;
857 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
858 __u64 ev1 = md_event(sb);
859
860 rdev->raid_disk = -1;
861 clear_bit(Faulty, &rdev->flags);
862 clear_bit(In_sync, &rdev->flags);
863 clear_bit(WriteMostly, &rdev->flags);
864 clear_bit(BarriersNotsupp, &rdev->flags);
865
866 if (mddev->raid_disks == 0) {
867 mddev->major_version = 0;
868 mddev->minor_version = sb->minor_version;
869 mddev->patch_version = sb->patch_version;
870 mddev->external = 0;
871 mddev->chunk_sectors = sb->chunk_size >> 9;
872 mddev->ctime = sb->ctime;
873 mddev->utime = sb->utime;
874 mddev->level = sb->level;
875 mddev->clevel[0] = 0;
876 mddev->layout = sb->layout;
877 mddev->raid_disks = sb->raid_disks;
878 mddev->dev_sectors = sb->size * 2;
879 mddev->events = ev1;
880 mddev->bitmap_offset = 0;
881 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
882
883 if (mddev->minor_version >= 91) {
884 mddev->reshape_position = sb->reshape_position;
885 mddev->delta_disks = sb->delta_disks;
886 mddev->new_level = sb->new_level;
887 mddev->new_layout = sb->new_layout;
888 mddev->new_chunk_sectors = sb->new_chunk >> 9;
889 } else {
890 mddev->reshape_position = MaxSector;
891 mddev->delta_disks = 0;
892 mddev->new_level = mddev->level;
893 mddev->new_layout = mddev->layout;
894 mddev->new_chunk_sectors = mddev->chunk_sectors;
895 }
896
897 if (sb->state & (1<<MD_SB_CLEAN))
898 mddev->recovery_cp = MaxSector;
899 else {
900 if (sb->events_hi == sb->cp_events_hi &&
901 sb->events_lo == sb->cp_events_lo) {
902 mddev->recovery_cp = sb->recovery_cp;
903 } else
904 mddev->recovery_cp = 0;
905 }
906
907 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
908 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
909 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
910 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
911
912 mddev->max_disks = MD_SB_DISKS;
913
914 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
915 mddev->bitmap_file == NULL)
916 mddev->bitmap_offset = mddev->default_bitmap_offset;
917
918 } else if (mddev->pers == NULL) {
919 /* Insist on good event counter while assembling */
920 ++ev1;
921 if (ev1 < mddev->events)
922 return -EINVAL;
923 } else if (mddev->bitmap) {
924 /* if adding to array with a bitmap, then we can accept an
925 * older device ... but not too old.
926 */
927 if (ev1 < mddev->bitmap->events_cleared)
928 return 0;
929 } else {
930 if (ev1 < mddev->events)
931 /* just a hot-add of a new device, leave raid_disk at -1 */
932 return 0;
933 }
934
935 if (mddev->level != LEVEL_MULTIPATH) {
936 desc = sb->disks + rdev->desc_nr;
937
938 if (desc->state & (1<<MD_DISK_FAULTY))
939 set_bit(Faulty, &rdev->flags);
940 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
941 desc->raid_disk < mddev->raid_disks */) {
942 set_bit(In_sync, &rdev->flags);
943 rdev->raid_disk = desc->raid_disk;
944 }
945 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
946 set_bit(WriteMostly, &rdev->flags);
947 } else /* MULTIPATH are always insync */
948 set_bit(In_sync, &rdev->flags);
949 return 0;
950 }
951
952 /*
953 * sync_super for 0.90.0
954 */
955 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
956 {
957 mdp_super_t *sb;
958 mdk_rdev_t *rdev2;
959 int next_spare = mddev->raid_disks;
960
961
962 /* make rdev->sb match mddev data..
963 *
964 * 1/ zero out disks
965 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
966 * 3/ any empty disks < next_spare become removed
967 *
968 * disks[0] gets initialised to REMOVED because
969 * we cannot be sure from other fields if it has
970 * been initialised or not.
971 */
972 int i;
973 int active=0, working=0,failed=0,spare=0,nr_disks=0;
974
975 rdev->sb_size = MD_SB_BYTES;
976
977 sb = (mdp_super_t*)page_address(rdev->sb_page);
978
979 memset(sb, 0, sizeof(*sb));
980
981 sb->md_magic = MD_SB_MAGIC;
982 sb->major_version = mddev->major_version;
983 sb->patch_version = mddev->patch_version;
984 sb->gvalid_words = 0; /* ignored */
985 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
986 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
987 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
988 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
989
990 sb->ctime = mddev->ctime;
991 sb->level = mddev->level;
992 sb->size = mddev->dev_sectors / 2;
993 sb->raid_disks = mddev->raid_disks;
994 sb->md_minor = mddev->md_minor;
995 sb->not_persistent = 0;
996 sb->utime = mddev->utime;
997 sb->state = 0;
998 sb->events_hi = (mddev->events>>32);
999 sb->events_lo = (u32)mddev->events;
1000
1001 if (mddev->reshape_position == MaxSector)
1002 sb->minor_version = 90;
1003 else {
1004 sb->minor_version = 91;
1005 sb->reshape_position = mddev->reshape_position;
1006 sb->new_level = mddev->new_level;
1007 sb->delta_disks = mddev->delta_disks;
1008 sb->new_layout = mddev->new_layout;
1009 sb->new_chunk = mddev->new_chunk_sectors << 9;
1010 }
1011 mddev->minor_version = sb->minor_version;
1012 if (mddev->in_sync)
1013 {
1014 sb->recovery_cp = mddev->recovery_cp;
1015 sb->cp_events_hi = (mddev->events>>32);
1016 sb->cp_events_lo = (u32)mddev->events;
1017 if (mddev->recovery_cp == MaxSector)
1018 sb->state = (1<< MD_SB_CLEAN);
1019 } else
1020 sb->recovery_cp = 0;
1021
1022 sb->layout = mddev->layout;
1023 sb->chunk_size = mddev->chunk_sectors << 9;
1024
1025 if (mddev->bitmap && mddev->bitmap_file == NULL)
1026 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1027
1028 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1029 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1030 mdp_disk_t *d;
1031 int desc_nr;
1032 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
1033 && !test_bit(Faulty, &rdev2->flags))
1034 desc_nr = rdev2->raid_disk;
1035 else
1036 desc_nr = next_spare++;
1037 rdev2->desc_nr = desc_nr;
1038 d = &sb->disks[rdev2->desc_nr];
1039 nr_disks++;
1040 d->number = rdev2->desc_nr;
1041 d->major = MAJOR(rdev2->bdev->bd_dev);
1042 d->minor = MINOR(rdev2->bdev->bd_dev);
1043 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
1044 && !test_bit(Faulty, &rdev2->flags))
1045 d->raid_disk = rdev2->raid_disk;
1046 else
1047 d->raid_disk = rdev2->desc_nr; /* compatibility */
1048 if (test_bit(Faulty, &rdev2->flags))
1049 d->state = (1<<MD_DISK_FAULTY);
1050 else if (test_bit(In_sync, &rdev2->flags)) {
1051 d->state = (1<<MD_DISK_ACTIVE);
1052 d->state |= (1<<MD_DISK_SYNC);
1053 active++;
1054 working++;
1055 } else {
1056 d->state = 0;
1057 spare++;
1058 working++;
1059 }
1060 if (test_bit(WriteMostly, &rdev2->flags))
1061 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1062 }
1063 /* now set the "removed" and "faulty" bits on any missing devices */
1064 for (i=0 ; i < mddev->raid_disks ; i++) {
1065 mdp_disk_t *d = &sb->disks[i];
1066 if (d->state == 0 && d->number == 0) {
1067 d->number = i;
1068 d->raid_disk = i;
1069 d->state = (1<<MD_DISK_REMOVED);
1070 d->state |= (1<<MD_DISK_FAULTY);
1071 failed++;
1072 }
1073 }
1074 sb->nr_disks = nr_disks;
1075 sb->active_disks = active;
1076 sb->working_disks = working;
1077 sb->failed_disks = failed;
1078 sb->spare_disks = spare;
1079
1080 sb->this_disk = sb->disks[rdev->desc_nr];
1081 sb->sb_csum = calc_sb_csum(sb);
1082 }
1083
1084 /*
1085 * rdev_size_change for 0.90.0
1086 */
1087 static unsigned long long
1088 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1089 {
1090 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1091 return 0; /* component must fit device */
1092 if (rdev->mddev->bitmap_offset)
1093 return 0; /* can't move bitmap */
1094 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1095 if (!num_sectors || num_sectors > rdev->sb_start)
1096 num_sectors = rdev->sb_start;
1097 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1098 rdev->sb_page);
1099 md_super_wait(rdev->mddev);
1100 return num_sectors / 2; /* kB for sysfs */
1101 }
1102
1103
1104 /*
1105 * version 1 superblock
1106 */
1107
1108 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1109 {
1110 __le32 disk_csum;
1111 u32 csum;
1112 unsigned long long newcsum;
1113 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1114 __le32 *isuper = (__le32*)sb;
1115 int i;
1116
1117 disk_csum = sb->sb_csum;
1118 sb->sb_csum = 0;
1119 newcsum = 0;
1120 for (i=0; size>=4; size -= 4 )
1121 newcsum += le32_to_cpu(*isuper++);
1122
1123 if (size == 2)
1124 newcsum += le16_to_cpu(*(__le16*) isuper);
1125
1126 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1127 sb->sb_csum = disk_csum;
1128 return cpu_to_le32(csum);
1129 }
1130
1131 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1132 {
1133 struct mdp_superblock_1 *sb;
1134 int ret;
1135 sector_t sb_start;
1136 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1137 int bmask;
1138
1139 /*
1140 * Calculate the position of the superblock in 512byte sectors.
1141 * It is always aligned to a 4K boundary and
1142 * depeding on minor_version, it can be:
1143 * 0: At least 8K, but less than 12K, from end of device
1144 * 1: At start of device
1145 * 2: 4K from start of device.
1146 */
1147 switch(minor_version) {
1148 case 0:
1149 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1150 sb_start -= 8*2;
1151 sb_start &= ~(sector_t)(4*2-1);
1152 break;
1153 case 1:
1154 sb_start = 0;
1155 break;
1156 case 2:
1157 sb_start = 8;
1158 break;
1159 default:
1160 return -EINVAL;
1161 }
1162 rdev->sb_start = sb_start;
1163
1164 /* superblock is rarely larger than 1K, but it can be larger,
1165 * and it is safe to read 4k, so we do that
1166 */
1167 ret = read_disk_sb(rdev, 4096);
1168 if (ret) return ret;
1169
1170
1171 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1172
1173 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1174 sb->major_version != cpu_to_le32(1) ||
1175 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1176 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1177 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1178 return -EINVAL;
1179
1180 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1181 printk("md: invalid superblock checksum on %s\n",
1182 bdevname(rdev->bdev,b));
1183 return -EINVAL;
1184 }
1185 if (le64_to_cpu(sb->data_size) < 10) {
1186 printk("md: data_size too small on %s\n",
1187 bdevname(rdev->bdev,b));
1188 return -EINVAL;
1189 }
1190
1191 rdev->preferred_minor = 0xffff;
1192 rdev->data_offset = le64_to_cpu(sb->data_offset);
1193 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1194
1195 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1196 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1197 if (rdev->sb_size & bmask)
1198 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1199
1200 if (minor_version
1201 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1202 return -EINVAL;
1203
1204 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1205 rdev->desc_nr = -1;
1206 else
1207 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1208
1209 if (!refdev) {
1210 ret = 1;
1211 } else {
1212 __u64 ev1, ev2;
1213 struct mdp_superblock_1 *refsb =
1214 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1215
1216 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1217 sb->level != refsb->level ||
1218 sb->layout != refsb->layout ||
1219 sb->chunksize != refsb->chunksize) {
1220 printk(KERN_WARNING "md: %s has strangely different"
1221 " superblock to %s\n",
1222 bdevname(rdev->bdev,b),
1223 bdevname(refdev->bdev,b2));
1224 return -EINVAL;
1225 }
1226 ev1 = le64_to_cpu(sb->events);
1227 ev2 = le64_to_cpu(refsb->events);
1228
1229 if (ev1 > ev2)
1230 ret = 1;
1231 else
1232 ret = 0;
1233 }
1234 if (minor_version)
1235 rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
1236 le64_to_cpu(sb->data_offset);
1237 else
1238 rdev->sectors = rdev->sb_start;
1239 if (rdev->sectors < le64_to_cpu(sb->data_size))
1240 return -EINVAL;
1241 rdev->sectors = le64_to_cpu(sb->data_size);
1242 if (le64_to_cpu(sb->size) > rdev->sectors)
1243 return -EINVAL;
1244 return ret;
1245 }
1246
1247 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1248 {
1249 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1250 __u64 ev1 = le64_to_cpu(sb->events);
1251
1252 rdev->raid_disk = -1;
1253 clear_bit(Faulty, &rdev->flags);
1254 clear_bit(In_sync, &rdev->flags);
1255 clear_bit(WriteMostly, &rdev->flags);
1256 clear_bit(BarriersNotsupp, &rdev->flags);
1257
1258 if (mddev->raid_disks == 0) {
1259 mddev->major_version = 1;
1260 mddev->patch_version = 0;
1261 mddev->external = 0;
1262 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1263 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1264 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1265 mddev->level = le32_to_cpu(sb->level);
1266 mddev->clevel[0] = 0;
1267 mddev->layout = le32_to_cpu(sb->layout);
1268 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1269 mddev->dev_sectors = le64_to_cpu(sb->size);
1270 mddev->events = ev1;
1271 mddev->bitmap_offset = 0;
1272 mddev->default_bitmap_offset = 1024 >> 9;
1273
1274 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1275 memcpy(mddev->uuid, sb->set_uuid, 16);
1276
1277 mddev->max_disks = (4096-256)/2;
1278
1279 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1280 mddev->bitmap_file == NULL )
1281 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1282
1283 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1284 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1285 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1286 mddev->new_level = le32_to_cpu(sb->new_level);
1287 mddev->new_layout = le32_to_cpu(sb->new_layout);
1288 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1289 } else {
1290 mddev->reshape_position = MaxSector;
1291 mddev->delta_disks = 0;
1292 mddev->new_level = mddev->level;
1293 mddev->new_layout = mddev->layout;
1294 mddev->new_chunk_sectors = mddev->chunk_sectors;
1295 }
1296
1297 } else if (mddev->pers == NULL) {
1298 /* Insist of good event counter while assembling */
1299 ++ev1;
1300 if (ev1 < mddev->events)
1301 return -EINVAL;
1302 } else if (mddev->bitmap) {
1303 /* If adding to array with a bitmap, then we can accept an
1304 * older device, but not too old.
1305 */
1306 if (ev1 < mddev->bitmap->events_cleared)
1307 return 0;
1308 } else {
1309 if (ev1 < mddev->events)
1310 /* just a hot-add of a new device, leave raid_disk at -1 */
1311 return 0;
1312 }
1313 if (mddev->level != LEVEL_MULTIPATH) {
1314 int role;
1315 if (rdev->desc_nr < 0 ||
1316 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1317 role = 0xffff;
1318 rdev->desc_nr = -1;
1319 } else
1320 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1321 switch(role) {
1322 case 0xffff: /* spare */
1323 break;
1324 case 0xfffe: /* faulty */
1325 set_bit(Faulty, &rdev->flags);
1326 break;
1327 default:
1328 if ((le32_to_cpu(sb->feature_map) &
1329 MD_FEATURE_RECOVERY_OFFSET))
1330 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1331 else
1332 set_bit(In_sync, &rdev->flags);
1333 rdev->raid_disk = role;
1334 break;
1335 }
1336 if (sb->devflags & WriteMostly1)
1337 set_bit(WriteMostly, &rdev->flags);
1338 } else /* MULTIPATH are always insync */
1339 set_bit(In_sync, &rdev->flags);
1340
1341 return 0;
1342 }
1343
1344 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1345 {
1346 struct mdp_superblock_1 *sb;
1347 mdk_rdev_t *rdev2;
1348 int max_dev, i;
1349 /* make rdev->sb match mddev and rdev data. */
1350
1351 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1352
1353 sb->feature_map = 0;
1354 sb->pad0 = 0;
1355 sb->recovery_offset = cpu_to_le64(0);
1356 memset(sb->pad1, 0, sizeof(sb->pad1));
1357 memset(sb->pad2, 0, sizeof(sb->pad2));
1358 memset(sb->pad3, 0, sizeof(sb->pad3));
1359
1360 sb->utime = cpu_to_le64((__u64)mddev->utime);
1361 sb->events = cpu_to_le64(mddev->events);
1362 if (mddev->in_sync)
1363 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1364 else
1365 sb->resync_offset = cpu_to_le64(0);
1366
1367 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1368
1369 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1370 sb->size = cpu_to_le64(mddev->dev_sectors);
1371 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1372 sb->level = cpu_to_le32(mddev->level);
1373 sb->layout = cpu_to_le32(mddev->layout);
1374
1375 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1376 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1377 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1378 }
1379
1380 if (rdev->raid_disk >= 0 &&
1381 !test_bit(In_sync, &rdev->flags)) {
1382 if (mddev->curr_resync_completed > rdev->recovery_offset)
1383 rdev->recovery_offset = mddev->curr_resync_completed;
1384 if (rdev->recovery_offset > 0) {
1385 sb->feature_map |=
1386 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1387 sb->recovery_offset =
1388 cpu_to_le64(rdev->recovery_offset);
1389 }
1390 }
1391
1392 if (mddev->reshape_position != MaxSector) {
1393 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1394 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1395 sb->new_layout = cpu_to_le32(mddev->new_layout);
1396 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1397 sb->new_level = cpu_to_le32(mddev->new_level);
1398 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1399 }
1400
1401 max_dev = 0;
1402 list_for_each_entry(rdev2, &mddev->disks, same_set)
1403 if (rdev2->desc_nr+1 > max_dev)
1404 max_dev = rdev2->desc_nr+1;
1405
1406 if (max_dev > le32_to_cpu(sb->max_dev)) {
1407 int bmask;
1408 sb->max_dev = cpu_to_le32(max_dev);
1409 rdev->sb_size = max_dev * 2 + 256;
1410 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1411 if (rdev->sb_size & bmask)
1412 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1413 }
1414 for (i=0; i<max_dev;i++)
1415 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1416
1417 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1418 i = rdev2->desc_nr;
1419 if (test_bit(Faulty, &rdev2->flags))
1420 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1421 else if (test_bit(In_sync, &rdev2->flags))
1422 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1423 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1424 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1425 else
1426 sb->dev_roles[i] = cpu_to_le16(0xffff);
1427 }
1428
1429 sb->sb_csum = calc_sb_1_csum(sb);
1430 }
1431
1432 static unsigned long long
1433 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1434 {
1435 struct mdp_superblock_1 *sb;
1436 sector_t max_sectors;
1437 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1438 return 0; /* component must fit device */
1439 if (rdev->sb_start < rdev->data_offset) {
1440 /* minor versions 1 and 2; superblock before data */
1441 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1442 max_sectors -= rdev->data_offset;
1443 if (!num_sectors || num_sectors > max_sectors)
1444 num_sectors = max_sectors;
1445 } else if (rdev->mddev->bitmap_offset) {
1446 /* minor version 0 with bitmap we can't move */
1447 return 0;
1448 } else {
1449 /* minor version 0; superblock after data */
1450 sector_t sb_start;
1451 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1452 sb_start &= ~(sector_t)(4*2 - 1);
1453 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1454 if (!num_sectors || num_sectors > max_sectors)
1455 num_sectors = max_sectors;
1456 rdev->sb_start = sb_start;
1457 }
1458 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1459 sb->data_size = cpu_to_le64(num_sectors);
1460 sb->super_offset = rdev->sb_start;
1461 sb->sb_csum = calc_sb_1_csum(sb);
1462 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1463 rdev->sb_page);
1464 md_super_wait(rdev->mddev);
1465 return num_sectors / 2; /* kB for sysfs */
1466 }
1467
1468 static struct super_type super_types[] = {
1469 [0] = {
1470 .name = "0.90.0",
1471 .owner = THIS_MODULE,
1472 .load_super = super_90_load,
1473 .validate_super = super_90_validate,
1474 .sync_super = super_90_sync,
1475 .rdev_size_change = super_90_rdev_size_change,
1476 },
1477 [1] = {
1478 .name = "md-1",
1479 .owner = THIS_MODULE,
1480 .load_super = super_1_load,
1481 .validate_super = super_1_validate,
1482 .sync_super = super_1_sync,
1483 .rdev_size_change = super_1_rdev_size_change,
1484 },
1485 };
1486
1487 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1488 {
1489 mdk_rdev_t *rdev, *rdev2;
1490
1491 rcu_read_lock();
1492 rdev_for_each_rcu(rdev, mddev1)
1493 rdev_for_each_rcu(rdev2, mddev2)
1494 if (rdev->bdev->bd_contains ==
1495 rdev2->bdev->bd_contains) {
1496 rcu_read_unlock();
1497 return 1;
1498 }
1499 rcu_read_unlock();
1500 return 0;
1501 }
1502
1503 static LIST_HEAD(pending_raid_disks);
1504
1505 /*
1506 * Try to register data integrity profile for an mddev
1507 *
1508 * This is called when an array is started and after a disk has been kicked
1509 * from the array. It only succeeds if all working and active component devices
1510 * are integrity capable with matching profiles.
1511 */
1512 int md_integrity_register(mddev_t *mddev)
1513 {
1514 mdk_rdev_t *rdev, *reference = NULL;
1515
1516 if (list_empty(&mddev->disks))
1517 return 0; /* nothing to do */
1518 if (blk_get_integrity(mddev->gendisk))
1519 return 0; /* already registered */
1520 list_for_each_entry(rdev, &mddev->disks, same_set) {
1521 /* skip spares and non-functional disks */
1522 if (test_bit(Faulty, &rdev->flags))
1523 continue;
1524 if (rdev->raid_disk < 0)
1525 continue;
1526 /*
1527 * If at least one rdev is not integrity capable, we can not
1528 * enable data integrity for the md device.
1529 */
1530 if (!bdev_get_integrity(rdev->bdev))
1531 return -EINVAL;
1532 if (!reference) {
1533 /* Use the first rdev as the reference */
1534 reference = rdev;
1535 continue;
1536 }
1537 /* does this rdev's profile match the reference profile? */
1538 if (blk_integrity_compare(reference->bdev->bd_disk,
1539 rdev->bdev->bd_disk) < 0)
1540 return -EINVAL;
1541 }
1542 /*
1543 * All component devices are integrity capable and have matching
1544 * profiles, register the common profile for the md device.
1545 */
1546 if (blk_integrity_register(mddev->gendisk,
1547 bdev_get_integrity(reference->bdev)) != 0) {
1548 printk(KERN_ERR "md: failed to register integrity for %s\n",
1549 mdname(mddev));
1550 return -EINVAL;
1551 }
1552 printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1553 mdname(mddev));
1554 return 0;
1555 }
1556 EXPORT_SYMBOL(md_integrity_register);
1557
1558 /* Disable data integrity if non-capable/non-matching disk is being added */
1559 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
1560 {
1561 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1562 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1563
1564 if (!bi_mddev) /* nothing to do */
1565 return;
1566 if (rdev->raid_disk < 0) /* skip spares */
1567 return;
1568 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1569 rdev->bdev->bd_disk) >= 0)
1570 return;
1571 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1572 blk_integrity_unregister(mddev->gendisk);
1573 }
1574 EXPORT_SYMBOL(md_integrity_add_rdev);
1575
1576 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1577 {
1578 char b[BDEVNAME_SIZE];
1579 struct kobject *ko;
1580 char *s;
1581 int err;
1582
1583 if (rdev->mddev) {
1584 MD_BUG();
1585 return -EINVAL;
1586 }
1587
1588 /* prevent duplicates */
1589 if (find_rdev(mddev, rdev->bdev->bd_dev))
1590 return -EEXIST;
1591
1592 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1593 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1594 rdev->sectors < mddev->dev_sectors)) {
1595 if (mddev->pers) {
1596 /* Cannot change size, so fail
1597 * If mddev->level <= 0, then we don't care
1598 * about aligning sizes (e.g. linear)
1599 */
1600 if (mddev->level > 0)
1601 return -ENOSPC;
1602 } else
1603 mddev->dev_sectors = rdev->sectors;
1604 }
1605
1606 /* Verify rdev->desc_nr is unique.
1607 * If it is -1, assign a free number, else
1608 * check number is not in use
1609 */
1610 if (rdev->desc_nr < 0) {
1611 int choice = 0;
1612 if (mddev->pers) choice = mddev->raid_disks;
1613 while (find_rdev_nr(mddev, choice))
1614 choice++;
1615 rdev->desc_nr = choice;
1616 } else {
1617 if (find_rdev_nr(mddev, rdev->desc_nr))
1618 return -EBUSY;
1619 }
1620 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1621 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1622 mdname(mddev), mddev->max_disks);
1623 return -EBUSY;
1624 }
1625 bdevname(rdev->bdev,b);
1626 while ( (s=strchr(b, '/')) != NULL)
1627 *s = '!';
1628
1629 rdev->mddev = mddev;
1630 printk(KERN_INFO "md: bind<%s>\n", b);
1631
1632 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1633 goto fail;
1634
1635 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1636 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1637 kobject_del(&rdev->kobj);
1638 goto fail;
1639 }
1640 rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, "state");
1641
1642 list_add_rcu(&rdev->same_set, &mddev->disks);
1643 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1644
1645 /* May as well allow recovery to be retried once */
1646 mddev->recovery_disabled = 0;
1647
1648 return 0;
1649
1650 fail:
1651 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1652 b, mdname(mddev));
1653 return err;
1654 }
1655
1656 static void md_delayed_delete(struct work_struct *ws)
1657 {
1658 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1659 kobject_del(&rdev->kobj);
1660 kobject_put(&rdev->kobj);
1661 }
1662
1663 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1664 {
1665 char b[BDEVNAME_SIZE];
1666 if (!rdev->mddev) {
1667 MD_BUG();
1668 return;
1669 }
1670 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1671 list_del_rcu(&rdev->same_set);
1672 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1673 rdev->mddev = NULL;
1674 sysfs_remove_link(&rdev->kobj, "block");
1675 sysfs_put(rdev->sysfs_state);
1676 rdev->sysfs_state = NULL;
1677 /* We need to delay this, otherwise we can deadlock when
1678 * writing to 'remove' to "dev/state". We also need
1679 * to delay it due to rcu usage.
1680 */
1681 synchronize_rcu();
1682 INIT_WORK(&rdev->del_work, md_delayed_delete);
1683 kobject_get(&rdev->kobj);
1684 schedule_work(&rdev->del_work);
1685 }
1686
1687 /*
1688 * prevent the device from being mounted, repartitioned or
1689 * otherwise reused by a RAID array (or any other kernel
1690 * subsystem), by bd_claiming the device.
1691 */
1692 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1693 {
1694 int err = 0;
1695 struct block_device *bdev;
1696 char b[BDEVNAME_SIZE];
1697
1698 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1699 if (IS_ERR(bdev)) {
1700 printk(KERN_ERR "md: could not open %s.\n",
1701 __bdevname(dev, b));
1702 return PTR_ERR(bdev);
1703 }
1704 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1705 if (err) {
1706 printk(KERN_ERR "md: could not bd_claim %s.\n",
1707 bdevname(bdev, b));
1708 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1709 return err;
1710 }
1711 if (!shared)
1712 set_bit(AllReserved, &rdev->flags);
1713 rdev->bdev = bdev;
1714 return err;
1715 }
1716
1717 static void unlock_rdev(mdk_rdev_t *rdev)
1718 {
1719 struct block_device *bdev = rdev->bdev;
1720 rdev->bdev = NULL;
1721 if (!bdev)
1722 MD_BUG();
1723 bd_release(bdev);
1724 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1725 }
1726
1727 void md_autodetect_dev(dev_t dev);
1728
1729 static void export_rdev(mdk_rdev_t * rdev)
1730 {
1731 char b[BDEVNAME_SIZE];
1732 printk(KERN_INFO "md: export_rdev(%s)\n",
1733 bdevname(rdev->bdev,b));
1734 if (rdev->mddev)
1735 MD_BUG();
1736 free_disk_sb(rdev);
1737 #ifndef MODULE
1738 if (test_bit(AutoDetected, &rdev->flags))
1739 md_autodetect_dev(rdev->bdev->bd_dev);
1740 #endif
1741 unlock_rdev(rdev);
1742 kobject_put(&rdev->kobj);
1743 }
1744
1745 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1746 {
1747 unbind_rdev_from_array(rdev);
1748 export_rdev(rdev);
1749 }
1750
1751 static void export_array(mddev_t *mddev)
1752 {
1753 mdk_rdev_t *rdev, *tmp;
1754
1755 rdev_for_each(rdev, tmp, mddev) {
1756 if (!rdev->mddev) {
1757 MD_BUG();
1758 continue;
1759 }
1760 kick_rdev_from_array(rdev);
1761 }
1762 if (!list_empty(&mddev->disks))
1763 MD_BUG();
1764 mddev->raid_disks = 0;
1765 mddev->major_version = 0;
1766 }
1767
1768 static void print_desc(mdp_disk_t *desc)
1769 {
1770 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1771 desc->major,desc->minor,desc->raid_disk,desc->state);
1772 }
1773
1774 static void print_sb_90(mdp_super_t *sb)
1775 {
1776 int i;
1777
1778 printk(KERN_INFO
1779 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1780 sb->major_version, sb->minor_version, sb->patch_version,
1781 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1782 sb->ctime);
1783 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1784 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1785 sb->md_minor, sb->layout, sb->chunk_size);
1786 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1787 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1788 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1789 sb->failed_disks, sb->spare_disks,
1790 sb->sb_csum, (unsigned long)sb->events_lo);
1791
1792 printk(KERN_INFO);
1793 for (i = 0; i < MD_SB_DISKS; i++) {
1794 mdp_disk_t *desc;
1795
1796 desc = sb->disks + i;
1797 if (desc->number || desc->major || desc->minor ||
1798 desc->raid_disk || (desc->state && (desc->state != 4))) {
1799 printk(" D %2d: ", i);
1800 print_desc(desc);
1801 }
1802 }
1803 printk(KERN_INFO "md: THIS: ");
1804 print_desc(&sb->this_disk);
1805 }
1806
1807 static void print_sb_1(struct mdp_superblock_1 *sb)
1808 {
1809 __u8 *uuid;
1810
1811 uuid = sb->set_uuid;
1812 printk(KERN_INFO
1813 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%02x%02x%02x%02x"
1814 ":%02x%02x:%02x%02x:%02x%02x:%02x%02x%02x%02x%02x%02x>\n"
1815 "md: Name: \"%s\" CT:%llu\n",
1816 le32_to_cpu(sb->major_version),
1817 le32_to_cpu(sb->feature_map),
1818 uuid[0], uuid[1], uuid[2], uuid[3],
1819 uuid[4], uuid[5], uuid[6], uuid[7],
1820 uuid[8], uuid[9], uuid[10], uuid[11],
1821 uuid[12], uuid[13], uuid[14], uuid[15],
1822 sb->set_name,
1823 (unsigned long long)le64_to_cpu(sb->ctime)
1824 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
1825
1826 uuid = sb->device_uuid;
1827 printk(KERN_INFO
1828 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
1829 " RO:%llu\n"
1830 "md: Dev:%08x UUID: %02x%02x%02x%02x:%02x%02x:%02x%02x:%02x%02x"
1831 ":%02x%02x%02x%02x%02x%02x\n"
1832 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
1833 "md: (MaxDev:%u) \n",
1834 le32_to_cpu(sb->level),
1835 (unsigned long long)le64_to_cpu(sb->size),
1836 le32_to_cpu(sb->raid_disks),
1837 le32_to_cpu(sb->layout),
1838 le32_to_cpu(sb->chunksize),
1839 (unsigned long long)le64_to_cpu(sb->data_offset),
1840 (unsigned long long)le64_to_cpu(sb->data_size),
1841 (unsigned long long)le64_to_cpu(sb->super_offset),
1842 (unsigned long long)le64_to_cpu(sb->recovery_offset),
1843 le32_to_cpu(sb->dev_number),
1844 uuid[0], uuid[1], uuid[2], uuid[3],
1845 uuid[4], uuid[5], uuid[6], uuid[7],
1846 uuid[8], uuid[9], uuid[10], uuid[11],
1847 uuid[12], uuid[13], uuid[14], uuid[15],
1848 sb->devflags,
1849 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
1850 (unsigned long long)le64_to_cpu(sb->events),
1851 (unsigned long long)le64_to_cpu(sb->resync_offset),
1852 le32_to_cpu(sb->sb_csum),
1853 le32_to_cpu(sb->max_dev)
1854 );
1855 }
1856
1857 static void print_rdev(mdk_rdev_t *rdev, int major_version)
1858 {
1859 char b[BDEVNAME_SIZE];
1860 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
1861 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
1862 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1863 rdev->desc_nr);
1864 if (rdev->sb_loaded) {
1865 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
1866 switch (major_version) {
1867 case 0:
1868 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
1869 break;
1870 case 1:
1871 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
1872 break;
1873 }
1874 } else
1875 printk(KERN_INFO "md: no rdev superblock!\n");
1876 }
1877
1878 static void md_print_devices(void)
1879 {
1880 struct list_head *tmp;
1881 mdk_rdev_t *rdev;
1882 mddev_t *mddev;
1883 char b[BDEVNAME_SIZE];
1884
1885 printk("\n");
1886 printk("md: **********************************\n");
1887 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1888 printk("md: **********************************\n");
1889 for_each_mddev(mddev, tmp) {
1890
1891 if (mddev->bitmap)
1892 bitmap_print_sb(mddev->bitmap);
1893 else
1894 printk("%s: ", mdname(mddev));
1895 list_for_each_entry(rdev, &mddev->disks, same_set)
1896 printk("<%s>", bdevname(rdev->bdev,b));
1897 printk("\n");
1898
1899 list_for_each_entry(rdev, &mddev->disks, same_set)
1900 print_rdev(rdev, mddev->major_version);
1901 }
1902 printk("md: **********************************\n");
1903 printk("\n");
1904 }
1905
1906
1907 static void sync_sbs(mddev_t * mddev, int nospares)
1908 {
1909 /* Update each superblock (in-memory image), but
1910 * if we are allowed to, skip spares which already
1911 * have the right event counter, or have one earlier
1912 * (which would mean they aren't being marked as dirty
1913 * with the rest of the array)
1914 */
1915 mdk_rdev_t *rdev;
1916
1917 list_for_each_entry(rdev, &mddev->disks, same_set) {
1918 if (rdev->sb_events == mddev->events ||
1919 (nospares &&
1920 rdev->raid_disk < 0 &&
1921 (rdev->sb_events&1)==0 &&
1922 rdev->sb_events+1 == mddev->events)) {
1923 /* Don't update this superblock */
1924 rdev->sb_loaded = 2;
1925 } else {
1926 super_types[mddev->major_version].
1927 sync_super(mddev, rdev);
1928 rdev->sb_loaded = 1;
1929 }
1930 }
1931 }
1932
1933 static void md_update_sb(mddev_t * mddev, int force_change)
1934 {
1935 mdk_rdev_t *rdev;
1936 int sync_req;
1937 int nospares = 0;
1938
1939 mddev->utime = get_seconds();
1940 if (mddev->external)
1941 return;
1942 repeat:
1943 spin_lock_irq(&mddev->write_lock);
1944
1945 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1946 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1947 force_change = 1;
1948 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1949 /* just a clean<-> dirty transition, possibly leave spares alone,
1950 * though if events isn't the right even/odd, we will have to do
1951 * spares after all
1952 */
1953 nospares = 1;
1954 if (force_change)
1955 nospares = 0;
1956 if (mddev->degraded)
1957 /* If the array is degraded, then skipping spares is both
1958 * dangerous and fairly pointless.
1959 * Dangerous because a device that was removed from the array
1960 * might have a event_count that still looks up-to-date,
1961 * so it can be re-added without a resync.
1962 * Pointless because if there are any spares to skip,
1963 * then a recovery will happen and soon that array won't
1964 * be degraded any more and the spare can go back to sleep then.
1965 */
1966 nospares = 0;
1967
1968 sync_req = mddev->in_sync;
1969
1970 /* If this is just a dirty<->clean transition, and the array is clean
1971 * and 'events' is odd, we can roll back to the previous clean state */
1972 if (nospares
1973 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1974 && (mddev->events & 1)
1975 && mddev->events != 1)
1976 mddev->events--;
1977 else {
1978 /* otherwise we have to go forward and ... */
1979 mddev->events ++;
1980 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1981 /* .. if the array isn't clean, an 'even' event must also go
1982 * to spares. */
1983 if ((mddev->events&1)==0)
1984 nospares = 0;
1985 } else {
1986 /* otherwise an 'odd' event must go to spares */
1987 if ((mddev->events&1))
1988 nospares = 0;
1989 }
1990 }
1991
1992 if (!mddev->events) {
1993 /*
1994 * oops, this 64-bit counter should never wrap.
1995 * Either we are in around ~1 trillion A.C., assuming
1996 * 1 reboot per second, or we have a bug:
1997 */
1998 MD_BUG();
1999 mddev->events --;
2000 }
2001
2002 /*
2003 * do not write anything to disk if using
2004 * nonpersistent superblocks
2005 */
2006 if (!mddev->persistent) {
2007 if (!mddev->external)
2008 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2009
2010 spin_unlock_irq(&mddev->write_lock);
2011 wake_up(&mddev->sb_wait);
2012 return;
2013 }
2014 sync_sbs(mddev, nospares);
2015 spin_unlock_irq(&mddev->write_lock);
2016
2017 dprintk(KERN_INFO
2018 "md: updating %s RAID superblock on device (in sync %d)\n",
2019 mdname(mddev),mddev->in_sync);
2020
2021 bitmap_update_sb(mddev->bitmap);
2022 list_for_each_entry(rdev, &mddev->disks, same_set) {
2023 char b[BDEVNAME_SIZE];
2024 dprintk(KERN_INFO "md: ");
2025 if (rdev->sb_loaded != 1)
2026 continue; /* no noise on spare devices */
2027 if (test_bit(Faulty, &rdev->flags))
2028 dprintk("(skipping faulty ");
2029
2030 dprintk("%s ", bdevname(rdev->bdev,b));
2031 if (!test_bit(Faulty, &rdev->flags)) {
2032 md_super_write(mddev,rdev,
2033 rdev->sb_start, rdev->sb_size,
2034 rdev->sb_page);
2035 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2036 bdevname(rdev->bdev,b),
2037 (unsigned long long)rdev->sb_start);
2038 rdev->sb_events = mddev->events;
2039
2040 } else
2041 dprintk(")\n");
2042 if (mddev->level == LEVEL_MULTIPATH)
2043 /* only need to write one superblock... */
2044 break;
2045 }
2046 md_super_wait(mddev);
2047 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2048
2049 spin_lock_irq(&mddev->write_lock);
2050 if (mddev->in_sync != sync_req ||
2051 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2052 /* have to write it out again */
2053 spin_unlock_irq(&mddev->write_lock);
2054 goto repeat;
2055 }
2056 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2057 spin_unlock_irq(&mddev->write_lock);
2058 wake_up(&mddev->sb_wait);
2059 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2060 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2061
2062 }
2063
2064 /* words written to sysfs files may, or may not, be \n terminated.
2065 * We want to accept with case. For this we use cmd_match.
2066 */
2067 static int cmd_match(const char *cmd, const char *str)
2068 {
2069 /* See if cmd, written into a sysfs file, matches
2070 * str. They must either be the same, or cmd can
2071 * have a trailing newline
2072 */
2073 while (*cmd && *str && *cmd == *str) {
2074 cmd++;
2075 str++;
2076 }
2077 if (*cmd == '\n')
2078 cmd++;
2079 if (*str || *cmd)
2080 return 0;
2081 return 1;
2082 }
2083
2084 struct rdev_sysfs_entry {
2085 struct attribute attr;
2086 ssize_t (*show)(mdk_rdev_t *, char *);
2087 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2088 };
2089
2090 static ssize_t
2091 state_show(mdk_rdev_t *rdev, char *page)
2092 {
2093 char *sep = "";
2094 size_t len = 0;
2095
2096 if (test_bit(Faulty, &rdev->flags)) {
2097 len+= sprintf(page+len, "%sfaulty",sep);
2098 sep = ",";
2099 }
2100 if (test_bit(In_sync, &rdev->flags)) {
2101 len += sprintf(page+len, "%sin_sync",sep);
2102 sep = ",";
2103 }
2104 if (test_bit(WriteMostly, &rdev->flags)) {
2105 len += sprintf(page+len, "%swrite_mostly",sep);
2106 sep = ",";
2107 }
2108 if (test_bit(Blocked, &rdev->flags)) {
2109 len += sprintf(page+len, "%sblocked", sep);
2110 sep = ",";
2111 }
2112 if (!test_bit(Faulty, &rdev->flags) &&
2113 !test_bit(In_sync, &rdev->flags)) {
2114 len += sprintf(page+len, "%sspare", sep);
2115 sep = ",";
2116 }
2117 return len+sprintf(page+len, "\n");
2118 }
2119
2120 static ssize_t
2121 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2122 {
2123 /* can write
2124 * faulty - simulates and error
2125 * remove - disconnects the device
2126 * writemostly - sets write_mostly
2127 * -writemostly - clears write_mostly
2128 * blocked - sets the Blocked flag
2129 * -blocked - clears the Blocked flag
2130 * insync - sets Insync providing device isn't active
2131 */
2132 int err = -EINVAL;
2133 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2134 md_error(rdev->mddev, rdev);
2135 err = 0;
2136 } else if (cmd_match(buf, "remove")) {
2137 if (rdev->raid_disk >= 0)
2138 err = -EBUSY;
2139 else {
2140 mddev_t *mddev = rdev->mddev;
2141 kick_rdev_from_array(rdev);
2142 if (mddev->pers)
2143 md_update_sb(mddev, 1);
2144 md_new_event(mddev);
2145 err = 0;
2146 }
2147 } else if (cmd_match(buf, "writemostly")) {
2148 set_bit(WriteMostly, &rdev->flags);
2149 err = 0;
2150 } else if (cmd_match(buf, "-writemostly")) {
2151 clear_bit(WriteMostly, &rdev->flags);
2152 err = 0;
2153 } else if (cmd_match(buf, "blocked")) {
2154 set_bit(Blocked, &rdev->flags);
2155 err = 0;
2156 } else if (cmd_match(buf, "-blocked")) {
2157 clear_bit(Blocked, &rdev->flags);
2158 wake_up(&rdev->blocked_wait);
2159 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2160 md_wakeup_thread(rdev->mddev->thread);
2161
2162 err = 0;
2163 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2164 set_bit(In_sync, &rdev->flags);
2165 err = 0;
2166 }
2167 if (!err && rdev->sysfs_state)
2168 sysfs_notify_dirent(rdev->sysfs_state);
2169 return err ? err : len;
2170 }
2171 static struct rdev_sysfs_entry rdev_state =
2172 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2173
2174 static ssize_t
2175 errors_show(mdk_rdev_t *rdev, char *page)
2176 {
2177 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2178 }
2179
2180 static ssize_t
2181 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2182 {
2183 char *e;
2184 unsigned long n = simple_strtoul(buf, &e, 10);
2185 if (*buf && (*e == 0 || *e == '\n')) {
2186 atomic_set(&rdev->corrected_errors, n);
2187 return len;
2188 }
2189 return -EINVAL;
2190 }
2191 static struct rdev_sysfs_entry rdev_errors =
2192 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2193
2194 static ssize_t
2195 slot_show(mdk_rdev_t *rdev, char *page)
2196 {
2197 if (rdev->raid_disk < 0)
2198 return sprintf(page, "none\n");
2199 else
2200 return sprintf(page, "%d\n", rdev->raid_disk);
2201 }
2202
2203 static ssize_t
2204 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2205 {
2206 char *e;
2207 int err;
2208 char nm[20];
2209 int slot = simple_strtoul(buf, &e, 10);
2210 if (strncmp(buf, "none", 4)==0)
2211 slot = -1;
2212 else if (e==buf || (*e && *e!= '\n'))
2213 return -EINVAL;
2214 if (rdev->mddev->pers && slot == -1) {
2215 /* Setting 'slot' on an active array requires also
2216 * updating the 'rd%d' link, and communicating
2217 * with the personality with ->hot_*_disk.
2218 * For now we only support removing
2219 * failed/spare devices. This normally happens automatically,
2220 * but not when the metadata is externally managed.
2221 */
2222 if (rdev->raid_disk == -1)
2223 return -EEXIST;
2224 /* personality does all needed checks */
2225 if (rdev->mddev->pers->hot_add_disk == NULL)
2226 return -EINVAL;
2227 err = rdev->mddev->pers->
2228 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2229 if (err)
2230 return err;
2231 sprintf(nm, "rd%d", rdev->raid_disk);
2232 sysfs_remove_link(&rdev->mddev->kobj, nm);
2233 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2234 md_wakeup_thread(rdev->mddev->thread);
2235 } else if (rdev->mddev->pers) {
2236 mdk_rdev_t *rdev2;
2237 /* Activating a spare .. or possibly reactivating
2238 * if we ever get bitmaps working here.
2239 */
2240
2241 if (rdev->raid_disk != -1)
2242 return -EBUSY;
2243
2244 if (rdev->mddev->pers->hot_add_disk == NULL)
2245 return -EINVAL;
2246
2247 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2248 if (rdev2->raid_disk == slot)
2249 return -EEXIST;
2250
2251 rdev->raid_disk = slot;
2252 if (test_bit(In_sync, &rdev->flags))
2253 rdev->saved_raid_disk = slot;
2254 else
2255 rdev->saved_raid_disk = -1;
2256 err = rdev->mddev->pers->
2257 hot_add_disk(rdev->mddev, rdev);
2258 if (err) {
2259 rdev->raid_disk = -1;
2260 return err;
2261 } else
2262 sysfs_notify_dirent(rdev->sysfs_state);
2263 sprintf(nm, "rd%d", rdev->raid_disk);
2264 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2265 printk(KERN_WARNING
2266 "md: cannot register "
2267 "%s for %s\n",
2268 nm, mdname(rdev->mddev));
2269
2270 /* don't wakeup anyone, leave that to userspace. */
2271 } else {
2272 if (slot >= rdev->mddev->raid_disks)
2273 return -ENOSPC;
2274 rdev->raid_disk = slot;
2275 /* assume it is working */
2276 clear_bit(Faulty, &rdev->flags);
2277 clear_bit(WriteMostly, &rdev->flags);
2278 set_bit(In_sync, &rdev->flags);
2279 sysfs_notify_dirent(rdev->sysfs_state);
2280 }
2281 return len;
2282 }
2283
2284
2285 static struct rdev_sysfs_entry rdev_slot =
2286 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2287
2288 static ssize_t
2289 offset_show(mdk_rdev_t *rdev, char *page)
2290 {
2291 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2292 }
2293
2294 static ssize_t
2295 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2296 {
2297 char *e;
2298 unsigned long long offset = simple_strtoull(buf, &e, 10);
2299 if (e==buf || (*e && *e != '\n'))
2300 return -EINVAL;
2301 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2302 return -EBUSY;
2303 if (rdev->sectors && rdev->mddev->external)
2304 /* Must set offset before size, so overlap checks
2305 * can be sane */
2306 return -EBUSY;
2307 rdev->data_offset = offset;
2308 return len;
2309 }
2310
2311 static struct rdev_sysfs_entry rdev_offset =
2312 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2313
2314 static ssize_t
2315 rdev_size_show(mdk_rdev_t *rdev, char *page)
2316 {
2317 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2318 }
2319
2320 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2321 {
2322 /* check if two start/length pairs overlap */
2323 if (s1+l1 <= s2)
2324 return 0;
2325 if (s2+l2 <= s1)
2326 return 0;
2327 return 1;
2328 }
2329
2330 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2331 {
2332 unsigned long long blocks;
2333 sector_t new;
2334
2335 if (strict_strtoull(buf, 10, &blocks) < 0)
2336 return -EINVAL;
2337
2338 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2339 return -EINVAL; /* sector conversion overflow */
2340
2341 new = blocks * 2;
2342 if (new != blocks * 2)
2343 return -EINVAL; /* unsigned long long to sector_t overflow */
2344
2345 *sectors = new;
2346 return 0;
2347 }
2348
2349 static ssize_t
2350 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2351 {
2352 mddev_t *my_mddev = rdev->mddev;
2353 sector_t oldsectors = rdev->sectors;
2354 sector_t sectors;
2355
2356 if (strict_blocks_to_sectors(buf, &sectors) < 0)
2357 return -EINVAL;
2358 if (my_mddev->pers && rdev->raid_disk >= 0) {
2359 if (my_mddev->persistent) {
2360 sectors = super_types[my_mddev->major_version].
2361 rdev_size_change(rdev, sectors);
2362 if (!sectors)
2363 return -EBUSY;
2364 } else if (!sectors)
2365 sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2366 rdev->data_offset;
2367 }
2368 if (sectors < my_mddev->dev_sectors)
2369 return -EINVAL; /* component must fit device */
2370
2371 rdev->sectors = sectors;
2372 if (sectors > oldsectors && my_mddev->external) {
2373 /* need to check that all other rdevs with the same ->bdev
2374 * do not overlap. We need to unlock the mddev to avoid
2375 * a deadlock. We have already changed rdev->sectors, and if
2376 * we have to change it back, we will have the lock again.
2377 */
2378 mddev_t *mddev;
2379 int overlap = 0;
2380 struct list_head *tmp;
2381
2382 mddev_unlock(my_mddev);
2383 for_each_mddev(mddev, tmp) {
2384 mdk_rdev_t *rdev2;
2385
2386 mddev_lock(mddev);
2387 list_for_each_entry(rdev2, &mddev->disks, same_set)
2388 if (test_bit(AllReserved, &rdev2->flags) ||
2389 (rdev->bdev == rdev2->bdev &&
2390 rdev != rdev2 &&
2391 overlaps(rdev->data_offset, rdev->sectors,
2392 rdev2->data_offset,
2393 rdev2->sectors))) {
2394 overlap = 1;
2395 break;
2396 }
2397 mddev_unlock(mddev);
2398 if (overlap) {
2399 mddev_put(mddev);
2400 break;
2401 }
2402 }
2403 mddev_lock(my_mddev);
2404 if (overlap) {
2405 /* Someone else could have slipped in a size
2406 * change here, but doing so is just silly.
2407 * We put oldsectors back because we *know* it is
2408 * safe, and trust userspace not to race with
2409 * itself
2410 */
2411 rdev->sectors = oldsectors;
2412 return -EBUSY;
2413 }
2414 }
2415 return len;
2416 }
2417
2418 static struct rdev_sysfs_entry rdev_size =
2419 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2420
2421 static struct attribute *rdev_default_attrs[] = {
2422 &rdev_state.attr,
2423 &rdev_errors.attr,
2424 &rdev_slot.attr,
2425 &rdev_offset.attr,
2426 &rdev_size.attr,
2427 NULL,
2428 };
2429 static ssize_t
2430 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2431 {
2432 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2433 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2434 mddev_t *mddev = rdev->mddev;
2435 ssize_t rv;
2436
2437 if (!entry->show)
2438 return -EIO;
2439
2440 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2441 if (!rv) {
2442 if (rdev->mddev == NULL)
2443 rv = -EBUSY;
2444 else
2445 rv = entry->show(rdev, page);
2446 mddev_unlock(mddev);
2447 }
2448 return rv;
2449 }
2450
2451 static ssize_t
2452 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2453 const char *page, size_t length)
2454 {
2455 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2456 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2457 ssize_t rv;
2458 mddev_t *mddev = rdev->mddev;
2459
2460 if (!entry->store)
2461 return -EIO;
2462 if (!capable(CAP_SYS_ADMIN))
2463 return -EACCES;
2464 rv = mddev ? mddev_lock(mddev): -EBUSY;
2465 if (!rv) {
2466 if (rdev->mddev == NULL)
2467 rv = -EBUSY;
2468 else
2469 rv = entry->store(rdev, page, length);
2470 mddev_unlock(mddev);
2471 }
2472 return rv;
2473 }
2474
2475 static void rdev_free(struct kobject *ko)
2476 {
2477 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2478 kfree(rdev);
2479 }
2480 static struct sysfs_ops rdev_sysfs_ops = {
2481 .show = rdev_attr_show,
2482 .store = rdev_attr_store,
2483 };
2484 static struct kobj_type rdev_ktype = {
2485 .release = rdev_free,
2486 .sysfs_ops = &rdev_sysfs_ops,
2487 .default_attrs = rdev_default_attrs,
2488 };
2489
2490 /*
2491 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2492 *
2493 * mark the device faulty if:
2494 *
2495 * - the device is nonexistent (zero size)
2496 * - the device has no valid superblock
2497 *
2498 * a faulty rdev _never_ has rdev->sb set.
2499 */
2500 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2501 {
2502 char b[BDEVNAME_SIZE];
2503 int err;
2504 mdk_rdev_t *rdev;
2505 sector_t size;
2506
2507 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2508 if (!rdev) {
2509 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2510 return ERR_PTR(-ENOMEM);
2511 }
2512
2513 if ((err = alloc_disk_sb(rdev)))
2514 goto abort_free;
2515
2516 err = lock_rdev(rdev, newdev, super_format == -2);
2517 if (err)
2518 goto abort_free;
2519
2520 kobject_init(&rdev->kobj, &rdev_ktype);
2521
2522 rdev->desc_nr = -1;
2523 rdev->saved_raid_disk = -1;
2524 rdev->raid_disk = -1;
2525 rdev->flags = 0;
2526 rdev->data_offset = 0;
2527 rdev->sb_events = 0;
2528 atomic_set(&rdev->nr_pending, 0);
2529 atomic_set(&rdev->read_errors, 0);
2530 atomic_set(&rdev->corrected_errors, 0);
2531
2532 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2533 if (!size) {
2534 printk(KERN_WARNING
2535 "md: %s has zero or unknown size, marking faulty!\n",
2536 bdevname(rdev->bdev,b));
2537 err = -EINVAL;
2538 goto abort_free;
2539 }
2540
2541 if (super_format >= 0) {
2542 err = super_types[super_format].
2543 load_super(rdev, NULL, super_minor);
2544 if (err == -EINVAL) {
2545 printk(KERN_WARNING
2546 "md: %s does not have a valid v%d.%d "
2547 "superblock, not importing!\n",
2548 bdevname(rdev->bdev,b),
2549 super_format, super_minor);
2550 goto abort_free;
2551 }
2552 if (err < 0) {
2553 printk(KERN_WARNING
2554 "md: could not read %s's sb, not importing!\n",
2555 bdevname(rdev->bdev,b));
2556 goto abort_free;
2557 }
2558 }
2559
2560 INIT_LIST_HEAD(&rdev->same_set);
2561 init_waitqueue_head(&rdev->blocked_wait);
2562
2563 return rdev;
2564
2565 abort_free:
2566 if (rdev->sb_page) {
2567 if (rdev->bdev)
2568 unlock_rdev(rdev);
2569 free_disk_sb(rdev);
2570 }
2571 kfree(rdev);
2572 return ERR_PTR(err);
2573 }
2574
2575 /*
2576 * Check a full RAID array for plausibility
2577 */
2578
2579
2580 static void analyze_sbs(mddev_t * mddev)
2581 {
2582 int i;
2583 mdk_rdev_t *rdev, *freshest, *tmp;
2584 char b[BDEVNAME_SIZE];
2585
2586 freshest = NULL;
2587 rdev_for_each(rdev, tmp, mddev)
2588 switch (super_types[mddev->major_version].
2589 load_super(rdev, freshest, mddev->minor_version)) {
2590 case 1:
2591 freshest = rdev;
2592 break;
2593 case 0:
2594 break;
2595 default:
2596 printk( KERN_ERR \
2597 "md: fatal superblock inconsistency in %s"
2598 " -- removing from array\n",
2599 bdevname(rdev->bdev,b));
2600 kick_rdev_from_array(rdev);
2601 }
2602
2603
2604 super_types[mddev->major_version].
2605 validate_super(mddev, freshest);
2606
2607 i = 0;
2608 rdev_for_each(rdev, tmp, mddev) {
2609 if (rdev->desc_nr >= mddev->max_disks ||
2610 i > mddev->max_disks) {
2611 printk(KERN_WARNING
2612 "md: %s: %s: only %d devices permitted\n",
2613 mdname(mddev), bdevname(rdev->bdev, b),
2614 mddev->max_disks);
2615 kick_rdev_from_array(rdev);
2616 continue;
2617 }
2618 if (rdev != freshest)
2619 if (super_types[mddev->major_version].
2620 validate_super(mddev, rdev)) {
2621 printk(KERN_WARNING "md: kicking non-fresh %s"
2622 " from array!\n",
2623 bdevname(rdev->bdev,b));
2624 kick_rdev_from_array(rdev);
2625 continue;
2626 }
2627 if (mddev->level == LEVEL_MULTIPATH) {
2628 rdev->desc_nr = i++;
2629 rdev->raid_disk = rdev->desc_nr;
2630 set_bit(In_sync, &rdev->flags);
2631 } else if (rdev->raid_disk >= mddev->raid_disks) {
2632 rdev->raid_disk = -1;
2633 clear_bit(In_sync, &rdev->flags);
2634 }
2635 }
2636 }
2637
2638 static void md_safemode_timeout(unsigned long data);
2639
2640 static ssize_t
2641 safe_delay_show(mddev_t *mddev, char *page)
2642 {
2643 int msec = (mddev->safemode_delay*1000)/HZ;
2644 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2645 }
2646 static ssize_t
2647 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2648 {
2649 int scale=1;
2650 int dot=0;
2651 int i;
2652 unsigned long msec;
2653 char buf[30];
2654
2655 /* remove a period, and count digits after it */
2656 if (len >= sizeof(buf))
2657 return -EINVAL;
2658 strlcpy(buf, cbuf, sizeof(buf));
2659 for (i=0; i<len; i++) {
2660 if (dot) {
2661 if (isdigit(buf[i])) {
2662 buf[i-1] = buf[i];
2663 scale *= 10;
2664 }
2665 buf[i] = 0;
2666 } else if (buf[i] == '.') {
2667 dot=1;
2668 buf[i] = 0;
2669 }
2670 }
2671 if (strict_strtoul(buf, 10, &msec) < 0)
2672 return -EINVAL;
2673 msec = (msec * 1000) / scale;
2674 if (msec == 0)
2675 mddev->safemode_delay = 0;
2676 else {
2677 unsigned long old_delay = mddev->safemode_delay;
2678 mddev->safemode_delay = (msec*HZ)/1000;
2679 if (mddev->safemode_delay == 0)
2680 mddev->safemode_delay = 1;
2681 if (mddev->safemode_delay < old_delay)
2682 md_safemode_timeout((unsigned long)mddev);
2683 }
2684 return len;
2685 }
2686 static struct md_sysfs_entry md_safe_delay =
2687 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2688
2689 static ssize_t
2690 level_show(mddev_t *mddev, char *page)
2691 {
2692 struct mdk_personality *p = mddev->pers;
2693 if (p)
2694 return sprintf(page, "%s\n", p->name);
2695 else if (mddev->clevel[0])
2696 return sprintf(page, "%s\n", mddev->clevel);
2697 else if (mddev->level != LEVEL_NONE)
2698 return sprintf(page, "%d\n", mddev->level);
2699 else
2700 return 0;
2701 }
2702
2703 static ssize_t
2704 level_store(mddev_t *mddev, const char *buf, size_t len)
2705 {
2706 char level[16];
2707 ssize_t rv = len;
2708 struct mdk_personality *pers;
2709 void *priv;
2710 mdk_rdev_t *rdev;
2711
2712 if (mddev->pers == NULL) {
2713 if (len == 0)
2714 return 0;
2715 if (len >= sizeof(mddev->clevel))
2716 return -ENOSPC;
2717 strncpy(mddev->clevel, buf, len);
2718 if (mddev->clevel[len-1] == '\n')
2719 len--;
2720 mddev->clevel[len] = 0;
2721 mddev->level = LEVEL_NONE;
2722 return rv;
2723 }
2724
2725 /* request to change the personality. Need to ensure:
2726 * - array is not engaged in resync/recovery/reshape
2727 * - old personality can be suspended
2728 * - new personality will access other array.
2729 */
2730
2731 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
2732 return -EBUSY;
2733
2734 if (!mddev->pers->quiesce) {
2735 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
2736 mdname(mddev), mddev->pers->name);
2737 return -EINVAL;
2738 }
2739
2740 /* Now find the new personality */
2741 if (len == 0 || len >= sizeof(level))
2742 return -EINVAL;
2743 strncpy(level, buf, len);
2744 if (level[len-1] == '\n')
2745 len--;
2746 level[len] = 0;
2747
2748 request_module("md-%s", level);
2749 spin_lock(&pers_lock);
2750 pers = find_pers(LEVEL_NONE, level);
2751 if (!pers || !try_module_get(pers->owner)) {
2752 spin_unlock(&pers_lock);
2753 printk(KERN_WARNING "md: personality %s not loaded\n", level);
2754 return -EINVAL;
2755 }
2756 spin_unlock(&pers_lock);
2757
2758 if (pers == mddev->pers) {
2759 /* Nothing to do! */
2760 module_put(pers->owner);
2761 return rv;
2762 }
2763 if (!pers->takeover) {
2764 module_put(pers->owner);
2765 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
2766 mdname(mddev), level);
2767 return -EINVAL;
2768 }
2769
2770 /* ->takeover must set new_* and/or delta_disks
2771 * if it succeeds, and may set them when it fails.
2772 */
2773 priv = pers->takeover(mddev);
2774 if (IS_ERR(priv)) {
2775 mddev->new_level = mddev->level;
2776 mddev->new_layout = mddev->layout;
2777 mddev->new_chunk_sectors = mddev->chunk_sectors;
2778 mddev->raid_disks -= mddev->delta_disks;
2779 mddev->delta_disks = 0;
2780 module_put(pers->owner);
2781 printk(KERN_WARNING "md: %s: %s would not accept array\n",
2782 mdname(mddev), level);
2783 return PTR_ERR(priv);
2784 }
2785
2786 /* Looks like we have a winner */
2787 mddev_suspend(mddev);
2788 mddev->pers->stop(mddev);
2789 module_put(mddev->pers->owner);
2790 /* Invalidate devices that are now superfluous */
2791 list_for_each_entry(rdev, &mddev->disks, same_set)
2792 if (rdev->raid_disk >= mddev->raid_disks) {
2793 rdev->raid_disk = -1;
2794 clear_bit(In_sync, &rdev->flags);
2795 }
2796 mddev->pers = pers;
2797 mddev->private = priv;
2798 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
2799 mddev->level = mddev->new_level;
2800 mddev->layout = mddev->new_layout;
2801 mddev->chunk_sectors = mddev->new_chunk_sectors;
2802 mddev->delta_disks = 0;
2803 pers->run(mddev);
2804 mddev_resume(mddev);
2805 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2806 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2807 md_wakeup_thread(mddev->thread);
2808 return rv;
2809 }
2810
2811 static struct md_sysfs_entry md_level =
2812 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2813
2814
2815 static ssize_t
2816 layout_show(mddev_t *mddev, char *page)
2817 {
2818 /* just a number, not meaningful for all levels */
2819 if (mddev->reshape_position != MaxSector &&
2820 mddev->layout != mddev->new_layout)
2821 return sprintf(page, "%d (%d)\n",
2822 mddev->new_layout, mddev->layout);
2823 return sprintf(page, "%d\n", mddev->layout);
2824 }
2825
2826 static ssize_t
2827 layout_store(mddev_t *mddev, const char *buf, size_t len)
2828 {
2829 char *e;
2830 unsigned long n = simple_strtoul(buf, &e, 10);
2831
2832 if (!*buf || (*e && *e != '\n'))
2833 return -EINVAL;
2834
2835 if (mddev->pers) {
2836 int err;
2837 if (mddev->pers->check_reshape == NULL)
2838 return -EBUSY;
2839 mddev->new_layout = n;
2840 err = mddev->pers->check_reshape(mddev);
2841 if (err) {
2842 mddev->new_layout = mddev->layout;
2843 return err;
2844 }
2845 } else {
2846 mddev->new_layout = n;
2847 if (mddev->reshape_position == MaxSector)
2848 mddev->layout = n;
2849 }
2850 return len;
2851 }
2852 static struct md_sysfs_entry md_layout =
2853 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2854
2855
2856 static ssize_t
2857 raid_disks_show(mddev_t *mddev, char *page)
2858 {
2859 if (mddev->raid_disks == 0)
2860 return 0;
2861 if (mddev->reshape_position != MaxSector &&
2862 mddev->delta_disks != 0)
2863 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2864 mddev->raid_disks - mddev->delta_disks);
2865 return sprintf(page, "%d\n", mddev->raid_disks);
2866 }
2867
2868 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2869
2870 static ssize_t
2871 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2872 {
2873 char *e;
2874 int rv = 0;
2875 unsigned long n = simple_strtoul(buf, &e, 10);
2876
2877 if (!*buf || (*e && *e != '\n'))
2878 return -EINVAL;
2879
2880 if (mddev->pers)
2881 rv = update_raid_disks(mddev, n);
2882 else if (mddev->reshape_position != MaxSector) {
2883 int olddisks = mddev->raid_disks - mddev->delta_disks;
2884 mddev->delta_disks = n - olddisks;
2885 mddev->raid_disks = n;
2886 } else
2887 mddev->raid_disks = n;
2888 return rv ? rv : len;
2889 }
2890 static struct md_sysfs_entry md_raid_disks =
2891 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2892
2893 static ssize_t
2894 chunk_size_show(mddev_t *mddev, char *page)
2895 {
2896 if (mddev->reshape_position != MaxSector &&
2897 mddev->chunk_sectors != mddev->new_chunk_sectors)
2898 return sprintf(page, "%d (%d)\n",
2899 mddev->new_chunk_sectors << 9,
2900 mddev->chunk_sectors << 9);
2901 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
2902 }
2903
2904 static ssize_t
2905 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2906 {
2907 char *e;
2908 unsigned long n = simple_strtoul(buf, &e, 10);
2909
2910 if (!*buf || (*e && *e != '\n'))
2911 return -EINVAL;
2912
2913 if (mddev->pers) {
2914 int err;
2915 if (mddev->pers->check_reshape == NULL)
2916 return -EBUSY;
2917 mddev->new_chunk_sectors = n >> 9;
2918 err = mddev->pers->check_reshape(mddev);
2919 if (err) {
2920 mddev->new_chunk_sectors = mddev->chunk_sectors;
2921 return err;
2922 }
2923 } else {
2924 mddev->new_chunk_sectors = n >> 9;
2925 if (mddev->reshape_position == MaxSector)
2926 mddev->chunk_sectors = n >> 9;
2927 }
2928 return len;
2929 }
2930 static struct md_sysfs_entry md_chunk_size =
2931 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2932
2933 static ssize_t
2934 resync_start_show(mddev_t *mddev, char *page)
2935 {
2936 if (mddev->recovery_cp == MaxSector)
2937 return sprintf(page, "none\n");
2938 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2939 }
2940
2941 static ssize_t
2942 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2943 {
2944 char *e;
2945 unsigned long long n = simple_strtoull(buf, &e, 10);
2946
2947 if (mddev->pers)
2948 return -EBUSY;
2949 if (!*buf || (*e && *e != '\n'))
2950 return -EINVAL;
2951
2952 mddev->recovery_cp = n;
2953 return len;
2954 }
2955 static struct md_sysfs_entry md_resync_start =
2956 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2957
2958 /*
2959 * The array state can be:
2960 *
2961 * clear
2962 * No devices, no size, no level
2963 * Equivalent to STOP_ARRAY ioctl
2964 * inactive
2965 * May have some settings, but array is not active
2966 * all IO results in error
2967 * When written, doesn't tear down array, but just stops it
2968 * suspended (not supported yet)
2969 * All IO requests will block. The array can be reconfigured.
2970 * Writing this, if accepted, will block until array is quiescent
2971 * readonly
2972 * no resync can happen. no superblocks get written.
2973 * write requests fail
2974 * read-auto
2975 * like readonly, but behaves like 'clean' on a write request.
2976 *
2977 * clean - no pending writes, but otherwise active.
2978 * When written to inactive array, starts without resync
2979 * If a write request arrives then
2980 * if metadata is known, mark 'dirty' and switch to 'active'.
2981 * if not known, block and switch to write-pending
2982 * If written to an active array that has pending writes, then fails.
2983 * active
2984 * fully active: IO and resync can be happening.
2985 * When written to inactive array, starts with resync
2986 *
2987 * write-pending
2988 * clean, but writes are blocked waiting for 'active' to be written.
2989 *
2990 * active-idle
2991 * like active, but no writes have been seen for a while (100msec).
2992 *
2993 */
2994 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2995 write_pending, active_idle, bad_word};
2996 static char *array_states[] = {
2997 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2998 "write-pending", "active-idle", NULL };
2999
3000 static int match_word(const char *word, char **list)
3001 {
3002 int n;
3003 for (n=0; list[n]; n++)
3004 if (cmd_match(word, list[n]))
3005 break;
3006 return n;
3007 }
3008
3009 static ssize_t
3010 array_state_show(mddev_t *mddev, char *page)
3011 {
3012 enum array_state st = inactive;
3013
3014 if (mddev->pers)
3015 switch(mddev->ro) {
3016 case 1:
3017 st = readonly;
3018 break;
3019 case 2:
3020 st = read_auto;
3021 break;
3022 case 0:
3023 if (mddev->in_sync)
3024 st = clean;
3025 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
3026 st = write_pending;
3027 else if (mddev->safemode)
3028 st = active_idle;
3029 else
3030 st = active;
3031 }
3032 else {
3033 if (list_empty(&mddev->disks) &&
3034 mddev->raid_disks == 0 &&
3035 mddev->dev_sectors == 0)
3036 st = clear;
3037 else
3038 st = inactive;
3039 }
3040 return sprintf(page, "%s\n", array_states[st]);
3041 }
3042
3043 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3044 static int do_md_run(mddev_t * mddev);
3045 static int restart_array(mddev_t *mddev);
3046
3047 static ssize_t
3048 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3049 {
3050 int err = -EINVAL;
3051 enum array_state st = match_word(buf, array_states);
3052 switch(st) {
3053 case bad_word:
3054 break;
3055 case clear:
3056 /* stopping an active array */
3057 if (atomic_read(&mddev->openers) > 0)
3058 return -EBUSY;
3059 err = do_md_stop(mddev, 0, 0);
3060 break;
3061 case inactive:
3062 /* stopping an active array */
3063 if (mddev->pers) {
3064 if (atomic_read(&mddev->openers) > 0)
3065 return -EBUSY;
3066 err = do_md_stop(mddev, 2, 0);
3067 } else
3068 err = 0; /* already inactive */
3069 break;
3070 case suspended:
3071 break; /* not supported yet */
3072 case readonly:
3073 if (mddev->pers)
3074 err = do_md_stop(mddev, 1, 0);
3075 else {
3076 mddev->ro = 1;
3077 set_disk_ro(mddev->gendisk, 1);
3078 err = do_md_run(mddev);
3079 }
3080 break;
3081 case read_auto:
3082 if (mddev->pers) {
3083 if (mddev->ro == 0)
3084 err = do_md_stop(mddev, 1, 0);
3085 else if (mddev->ro == 1)
3086 err = restart_array(mddev);
3087 if (err == 0) {
3088 mddev->ro = 2;
3089 set_disk_ro(mddev->gendisk, 0);
3090 }
3091 } else {
3092 mddev->ro = 2;
3093 err = do_md_run(mddev);
3094 }
3095 break;
3096 case clean:
3097 if (mddev->pers) {
3098 restart_array(mddev);
3099 spin_lock_irq(&mddev->write_lock);
3100 if (atomic_read(&mddev->writes_pending) == 0) {
3101 if (mddev->in_sync == 0) {
3102 mddev->in_sync = 1;
3103 if (mddev->safemode == 1)
3104 mddev->safemode = 0;
3105 if (mddev->persistent)
3106 set_bit(MD_CHANGE_CLEAN,
3107 &mddev->flags);
3108 }
3109 err = 0;
3110 } else
3111 err = -EBUSY;
3112 spin_unlock_irq(&mddev->write_lock);
3113 } else
3114 err = -EINVAL;
3115 break;
3116 case active:
3117 if (mddev->pers) {
3118 restart_array(mddev);
3119 if (mddev->external)
3120 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3121 wake_up(&mddev->sb_wait);
3122 err = 0;
3123 } else {
3124 mddev->ro = 0;
3125 set_disk_ro(mddev->gendisk, 0);
3126 err = do_md_run(mddev);
3127 }
3128 break;
3129 case write_pending:
3130 case active_idle:
3131 /* these cannot be set */
3132 break;
3133 }
3134 if (err)
3135 return err;
3136 else {
3137 sysfs_notify_dirent(mddev->sysfs_state);
3138 return len;
3139 }
3140 }
3141 static struct md_sysfs_entry md_array_state =
3142 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3143
3144 static ssize_t
3145 null_show(mddev_t *mddev, char *page)
3146 {
3147 return -EINVAL;
3148 }
3149
3150 static ssize_t
3151 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3152 {
3153 /* buf must be %d:%d\n? giving major and minor numbers */
3154 /* The new device is added to the array.
3155 * If the array has a persistent superblock, we read the
3156 * superblock to initialise info and check validity.
3157 * Otherwise, only checking done is that in bind_rdev_to_array,
3158 * which mainly checks size.
3159 */
3160 char *e;
3161 int major = simple_strtoul(buf, &e, 10);
3162 int minor;
3163 dev_t dev;
3164 mdk_rdev_t *rdev;
3165 int err;
3166
3167 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3168 return -EINVAL;
3169 minor = simple_strtoul(e+1, &e, 10);
3170 if (*e && *e != '\n')
3171 return -EINVAL;
3172 dev = MKDEV(major, minor);
3173 if (major != MAJOR(dev) ||
3174 minor != MINOR(dev))
3175 return -EOVERFLOW;
3176
3177
3178 if (mddev->persistent) {
3179 rdev = md_import_device(dev, mddev->major_version,
3180 mddev->minor_version);
3181 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3182 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3183 mdk_rdev_t, same_set);
3184 err = super_types[mddev->major_version]
3185 .load_super(rdev, rdev0, mddev->minor_version);
3186 if (err < 0)
3187 goto out;
3188 }
3189 } else if (mddev->external)
3190 rdev = md_import_device(dev, -2, -1);
3191 else
3192 rdev = md_import_device(dev, -1, -1);
3193
3194 if (IS_ERR(rdev))
3195 return PTR_ERR(rdev);
3196 err = bind_rdev_to_array(rdev, mddev);
3197 out:
3198 if (err)
3199 export_rdev(rdev);
3200 return err ? err : len;
3201 }
3202
3203 static struct md_sysfs_entry md_new_device =
3204 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3205
3206 static ssize_t
3207 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3208 {
3209 char *end;
3210 unsigned long chunk, end_chunk;
3211
3212 if (!mddev->bitmap)
3213 goto out;
3214 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3215 while (*buf) {
3216 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3217 if (buf == end) break;
3218 if (*end == '-') { /* range */
3219 buf = end + 1;
3220 end_chunk = simple_strtoul(buf, &end, 0);
3221 if (buf == end) break;
3222 }
3223 if (*end && !isspace(*end)) break;
3224 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3225 buf = end;
3226 while (isspace(*buf)) buf++;
3227 }
3228 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3229 out:
3230 return len;
3231 }
3232
3233 static struct md_sysfs_entry md_bitmap =
3234 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3235
3236 static ssize_t
3237 size_show(mddev_t *mddev, char *page)
3238 {
3239 return sprintf(page, "%llu\n",
3240 (unsigned long long)mddev->dev_sectors / 2);
3241 }
3242
3243 static int update_size(mddev_t *mddev, sector_t num_sectors);
3244
3245 static ssize_t
3246 size_store(mddev_t *mddev, const char *buf, size_t len)
3247 {
3248 /* If array is inactive, we can reduce the component size, but
3249 * not increase it (except from 0).
3250 * If array is active, we can try an on-line resize
3251 */
3252 sector_t sectors;
3253 int err = strict_blocks_to_sectors(buf, &sectors);
3254
3255 if (err < 0)
3256 return err;
3257 if (mddev->pers) {
3258 err = update_size(mddev, sectors);
3259 md_update_sb(mddev, 1);
3260 } else {
3261 if (mddev->dev_sectors == 0 ||
3262 mddev->dev_sectors > sectors)
3263 mddev->dev_sectors = sectors;
3264 else
3265 err = -ENOSPC;
3266 }
3267 return err ? err : len;
3268 }
3269
3270 static struct md_sysfs_entry md_size =
3271 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3272
3273
3274 /* Metdata version.
3275 * This is one of
3276 * 'none' for arrays with no metadata (good luck...)
3277 * 'external' for arrays with externally managed metadata,
3278 * or N.M for internally known formats
3279 */
3280 static ssize_t
3281 metadata_show(mddev_t *mddev, char *page)
3282 {
3283 if (mddev->persistent)
3284 return sprintf(page, "%d.%d\n",
3285 mddev->major_version, mddev->minor_version);
3286 else if (mddev->external)
3287 return sprintf(page, "external:%s\n", mddev->metadata_type);
3288 else
3289 return sprintf(page, "none\n");
3290 }
3291
3292 static ssize_t
3293 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3294 {
3295 int major, minor;
3296 char *e;
3297 /* Changing the details of 'external' metadata is
3298 * always permitted. Otherwise there must be
3299 * no devices attached to the array.
3300 */
3301 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3302 ;
3303 else if (!list_empty(&mddev->disks))
3304 return -EBUSY;
3305
3306 if (cmd_match(buf, "none")) {
3307 mddev->persistent = 0;
3308 mddev->external = 0;
3309 mddev->major_version = 0;
3310 mddev->minor_version = 90;
3311 return len;
3312 }
3313 if (strncmp(buf, "external:", 9) == 0) {
3314 size_t namelen = len-9;
3315 if (namelen >= sizeof(mddev->metadata_type))
3316 namelen = sizeof(mddev->metadata_type)-1;
3317 strncpy(mddev->metadata_type, buf+9, namelen);
3318 mddev->metadata_type[namelen] = 0;
3319 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3320 mddev->metadata_type[--namelen] = 0;
3321 mddev->persistent = 0;
3322 mddev->external = 1;
3323 mddev->major_version = 0;
3324 mddev->minor_version = 90;
3325 return len;
3326 }
3327 major = simple_strtoul(buf, &e, 10);
3328 if (e==buf || *e != '.')
3329 return -EINVAL;
3330 buf = e+1;
3331 minor = simple_strtoul(buf, &e, 10);
3332 if (e==buf || (*e && *e != '\n') )
3333 return -EINVAL;
3334 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3335 return -ENOENT;
3336 mddev->major_version = major;
3337 mddev->minor_version = minor;
3338 mddev->persistent = 1;
3339 mddev->external = 0;
3340 return len;
3341 }
3342
3343 static struct md_sysfs_entry md_metadata =
3344 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3345
3346 static ssize_t
3347 action_show(mddev_t *mddev, char *page)
3348 {
3349 char *type = "idle";
3350 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3351 type = "frozen";
3352 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3353 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3354 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3355 type = "reshape";
3356 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3357 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3358 type = "resync";
3359 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3360 type = "check";
3361 else
3362 type = "repair";
3363 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3364 type = "recover";
3365 }
3366 return sprintf(page, "%s\n", type);
3367 }
3368
3369 static ssize_t
3370 action_store(mddev_t *mddev, const char *page, size_t len)
3371 {
3372 if (!mddev->pers || !mddev->pers->sync_request)
3373 return -EINVAL;
3374
3375 if (cmd_match(page, "frozen"))
3376 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3377 else
3378 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3379
3380 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3381 if (mddev->sync_thread) {
3382 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3383 md_unregister_thread(mddev->sync_thread);
3384 mddev->sync_thread = NULL;
3385 mddev->recovery = 0;
3386 }
3387 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3388 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3389 return -EBUSY;
3390 else if (cmd_match(page, "resync"))
3391 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3392 else if (cmd_match(page, "recover")) {
3393 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3394 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3395 } else if (cmd_match(page, "reshape")) {
3396 int err;
3397 if (mddev->pers->start_reshape == NULL)
3398 return -EINVAL;
3399 err = mddev->pers->start_reshape(mddev);
3400 if (err)
3401 return err;
3402 sysfs_notify(&mddev->kobj, NULL, "degraded");
3403 } else {
3404 if (cmd_match(page, "check"))
3405 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3406 else if (!cmd_match(page, "repair"))
3407 return -EINVAL;
3408 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3409 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3410 }
3411 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3412 md_wakeup_thread(mddev->thread);
3413 sysfs_notify_dirent(mddev->sysfs_action);
3414 return len;
3415 }
3416
3417 static ssize_t
3418 mismatch_cnt_show(mddev_t *mddev, char *page)
3419 {
3420 return sprintf(page, "%llu\n",
3421 (unsigned long long) mddev->resync_mismatches);
3422 }
3423
3424 static struct md_sysfs_entry md_scan_mode =
3425 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3426
3427
3428 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3429
3430 static ssize_t
3431 sync_min_show(mddev_t *mddev, char *page)
3432 {
3433 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3434 mddev->sync_speed_min ? "local": "system");
3435 }
3436
3437 static ssize_t
3438 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3439 {
3440 int min;
3441 char *e;
3442 if (strncmp(buf, "system", 6)==0) {
3443 mddev->sync_speed_min = 0;
3444 return len;
3445 }
3446 min = simple_strtoul(buf, &e, 10);
3447 if (buf == e || (*e && *e != '\n') || min <= 0)
3448 return -EINVAL;
3449 mddev->sync_speed_min = min;
3450 return len;
3451 }
3452
3453 static struct md_sysfs_entry md_sync_min =
3454 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3455
3456 static ssize_t
3457 sync_max_show(mddev_t *mddev, char *page)
3458 {
3459 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3460 mddev->sync_speed_max ? "local": "system");
3461 }
3462
3463 static ssize_t
3464 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3465 {
3466 int max;
3467 char *e;
3468 if (strncmp(buf, "system", 6)==0) {
3469 mddev->sync_speed_max = 0;
3470 return len;
3471 }
3472 max = simple_strtoul(buf, &e, 10);
3473 if (buf == e || (*e && *e != '\n') || max <= 0)
3474 return -EINVAL;
3475 mddev->sync_speed_max = max;
3476 return len;
3477 }
3478
3479 static struct md_sysfs_entry md_sync_max =
3480 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3481
3482 static ssize_t
3483 degraded_show(mddev_t *mddev, char *page)
3484 {
3485 return sprintf(page, "%d\n", mddev->degraded);
3486 }
3487 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3488
3489 static ssize_t
3490 sync_force_parallel_show(mddev_t *mddev, char *page)
3491 {
3492 return sprintf(page, "%d\n", mddev->parallel_resync);
3493 }
3494
3495 static ssize_t
3496 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3497 {
3498 long n;
3499
3500 if (strict_strtol(buf, 10, &n))
3501 return -EINVAL;
3502
3503 if (n != 0 && n != 1)
3504 return -EINVAL;
3505
3506 mddev->parallel_resync = n;
3507
3508 if (mddev->sync_thread)
3509 wake_up(&resync_wait);
3510
3511 return len;
3512 }
3513
3514 /* force parallel resync, even with shared block devices */
3515 static struct md_sysfs_entry md_sync_force_parallel =
3516 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3517 sync_force_parallel_show, sync_force_parallel_store);
3518
3519 static ssize_t
3520 sync_speed_show(mddev_t *mddev, char *page)
3521 {
3522 unsigned long resync, dt, db;
3523 if (mddev->curr_resync == 0)
3524 return sprintf(page, "none\n");
3525 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3526 dt = (jiffies - mddev->resync_mark) / HZ;
3527 if (!dt) dt++;
3528 db = resync - mddev->resync_mark_cnt;
3529 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3530 }
3531
3532 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3533
3534 static ssize_t
3535 sync_completed_show(mddev_t *mddev, char *page)
3536 {
3537 unsigned long max_sectors, resync;
3538
3539 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3540 return sprintf(page, "none\n");
3541
3542 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3543 max_sectors = mddev->resync_max_sectors;
3544 else
3545 max_sectors = mddev->dev_sectors;
3546
3547 resync = mddev->curr_resync_completed;
3548 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3549 }
3550
3551 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3552
3553 static ssize_t
3554 min_sync_show(mddev_t *mddev, char *page)
3555 {
3556 return sprintf(page, "%llu\n",
3557 (unsigned long long)mddev->resync_min);
3558 }
3559 static ssize_t
3560 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3561 {
3562 unsigned long long min;
3563 if (strict_strtoull(buf, 10, &min))
3564 return -EINVAL;
3565 if (min > mddev->resync_max)
3566 return -EINVAL;
3567 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3568 return -EBUSY;
3569
3570 /* Must be a multiple of chunk_size */
3571 if (mddev->chunk_sectors) {
3572 sector_t temp = min;
3573 if (sector_div(temp, mddev->chunk_sectors))
3574 return -EINVAL;
3575 }
3576 mddev->resync_min = min;
3577
3578 return len;
3579 }
3580
3581 static struct md_sysfs_entry md_min_sync =
3582 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3583
3584 static ssize_t
3585 max_sync_show(mddev_t *mddev, char *page)
3586 {
3587 if (mddev->resync_max == MaxSector)
3588 return sprintf(page, "max\n");
3589 else
3590 return sprintf(page, "%llu\n",
3591 (unsigned long long)mddev->resync_max);
3592 }
3593 static ssize_t
3594 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3595 {
3596 if (strncmp(buf, "max", 3) == 0)
3597 mddev->resync_max = MaxSector;
3598 else {
3599 unsigned long long max;
3600 if (strict_strtoull(buf, 10, &max))
3601 return -EINVAL;
3602 if (max < mddev->resync_min)
3603 return -EINVAL;
3604 if (max < mddev->resync_max &&
3605 mddev->ro == 0 &&
3606 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3607 return -EBUSY;
3608
3609 /* Must be a multiple of chunk_size */
3610 if (mddev->chunk_sectors) {
3611 sector_t temp = max;
3612 if (sector_div(temp, mddev->chunk_sectors))
3613 return -EINVAL;
3614 }
3615 mddev->resync_max = max;
3616 }
3617 wake_up(&mddev->recovery_wait);
3618 return len;
3619 }
3620
3621 static struct md_sysfs_entry md_max_sync =
3622 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3623
3624 static ssize_t
3625 suspend_lo_show(mddev_t *mddev, char *page)
3626 {
3627 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3628 }
3629
3630 static ssize_t
3631 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3632 {
3633 char *e;
3634 unsigned long long new = simple_strtoull(buf, &e, 10);
3635
3636 if (mddev->pers == NULL ||
3637 mddev->pers->quiesce == NULL)
3638 return -EINVAL;
3639 if (buf == e || (*e && *e != '\n'))
3640 return -EINVAL;
3641 if (new >= mddev->suspend_hi ||
3642 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3643 mddev->suspend_lo = new;
3644 mddev->pers->quiesce(mddev, 2);
3645 return len;
3646 } else
3647 return -EINVAL;
3648 }
3649 static struct md_sysfs_entry md_suspend_lo =
3650 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3651
3652
3653 static ssize_t
3654 suspend_hi_show(mddev_t *mddev, char *page)
3655 {
3656 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3657 }
3658
3659 static ssize_t
3660 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3661 {
3662 char *e;
3663 unsigned long long new = simple_strtoull(buf, &e, 10);
3664
3665 if (mddev->pers == NULL ||
3666 mddev->pers->quiesce == NULL)
3667 return -EINVAL;
3668 if (buf == e || (*e && *e != '\n'))
3669 return -EINVAL;
3670 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3671 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3672 mddev->suspend_hi = new;
3673 mddev->pers->quiesce(mddev, 1);
3674 mddev->pers->quiesce(mddev, 0);
3675 return len;
3676 } else
3677 return -EINVAL;
3678 }
3679 static struct md_sysfs_entry md_suspend_hi =
3680 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3681
3682 static ssize_t
3683 reshape_position_show(mddev_t *mddev, char *page)
3684 {
3685 if (mddev->reshape_position != MaxSector)
3686 return sprintf(page, "%llu\n",
3687 (unsigned long long)mddev->reshape_position);
3688 strcpy(page, "none\n");
3689 return 5;
3690 }
3691
3692 static ssize_t
3693 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3694 {
3695 char *e;
3696 unsigned long long new = simple_strtoull(buf, &e, 10);
3697 if (mddev->pers)
3698 return -EBUSY;
3699 if (buf == e || (*e && *e != '\n'))
3700 return -EINVAL;
3701 mddev->reshape_position = new;
3702 mddev->delta_disks = 0;
3703 mddev->new_level = mddev->level;
3704 mddev->new_layout = mddev->layout;
3705 mddev->new_chunk_sectors = mddev->chunk_sectors;
3706 return len;
3707 }
3708
3709 static struct md_sysfs_entry md_reshape_position =
3710 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3711 reshape_position_store);
3712
3713 static ssize_t
3714 array_size_show(mddev_t *mddev, char *page)
3715 {
3716 if (mddev->external_size)
3717 return sprintf(page, "%llu\n",
3718 (unsigned long long)mddev->array_sectors/2);
3719 else
3720 return sprintf(page, "default\n");
3721 }
3722
3723 static ssize_t
3724 array_size_store(mddev_t *mddev, const char *buf, size_t len)
3725 {
3726 sector_t sectors;
3727
3728 if (strncmp(buf, "default", 7) == 0) {
3729 if (mddev->pers)
3730 sectors = mddev->pers->size(mddev, 0, 0);
3731 else
3732 sectors = mddev->array_sectors;
3733
3734 mddev->external_size = 0;
3735 } else {
3736 if (strict_blocks_to_sectors(buf, &sectors) < 0)
3737 return -EINVAL;
3738 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
3739 return -E2BIG;
3740
3741 mddev->external_size = 1;
3742 }
3743
3744 mddev->array_sectors = sectors;
3745 set_capacity(mddev->gendisk, mddev->array_sectors);
3746 if (mddev->pers)
3747 revalidate_disk(mddev->gendisk);
3748
3749 return len;
3750 }
3751
3752 static struct md_sysfs_entry md_array_size =
3753 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
3754 array_size_store);
3755
3756 static struct attribute *md_default_attrs[] = {
3757 &md_level.attr,
3758 &md_layout.attr,
3759 &md_raid_disks.attr,
3760 &md_chunk_size.attr,
3761 &md_size.attr,
3762 &md_resync_start.attr,
3763 &md_metadata.attr,
3764 &md_new_device.attr,
3765 &md_safe_delay.attr,
3766 &md_array_state.attr,
3767 &md_reshape_position.attr,
3768 &md_array_size.attr,
3769 NULL,
3770 };
3771
3772 static struct attribute *md_redundancy_attrs[] = {
3773 &md_scan_mode.attr,
3774 &md_mismatches.attr,
3775 &md_sync_min.attr,
3776 &md_sync_max.attr,
3777 &md_sync_speed.attr,
3778 &md_sync_force_parallel.attr,
3779 &md_sync_completed.attr,
3780 &md_min_sync.attr,
3781 &md_max_sync.attr,
3782 &md_suspend_lo.attr,
3783 &md_suspend_hi.attr,
3784 &md_bitmap.attr,
3785 &md_degraded.attr,
3786 NULL,
3787 };
3788 static struct attribute_group md_redundancy_group = {
3789 .name = NULL,
3790 .attrs = md_redundancy_attrs,
3791 };
3792
3793
3794 static ssize_t
3795 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3796 {
3797 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3798 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3799 ssize_t rv;
3800
3801 if (!entry->show)
3802 return -EIO;
3803 rv = mddev_lock(mddev);
3804 if (!rv) {
3805 rv = entry->show(mddev, page);
3806 mddev_unlock(mddev);
3807 }
3808 return rv;
3809 }
3810
3811 static ssize_t
3812 md_attr_store(struct kobject *kobj, struct attribute *attr,
3813 const char *page, size_t length)
3814 {
3815 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3816 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3817 ssize_t rv;
3818
3819 if (!entry->store)
3820 return -EIO;
3821 if (!capable(CAP_SYS_ADMIN))
3822 return -EACCES;
3823 rv = mddev_lock(mddev);
3824 if (mddev->hold_active == UNTIL_IOCTL)
3825 mddev->hold_active = 0;
3826 if (!rv) {
3827 rv = entry->store(mddev, page, length);
3828 mddev_unlock(mddev);
3829 }
3830 return rv;
3831 }
3832
3833 static void md_free(struct kobject *ko)
3834 {
3835 mddev_t *mddev = container_of(ko, mddev_t, kobj);
3836
3837 if (mddev->sysfs_state)
3838 sysfs_put(mddev->sysfs_state);
3839
3840 if (mddev->gendisk) {
3841 del_gendisk(mddev->gendisk);
3842 put_disk(mddev->gendisk);
3843 }
3844 if (mddev->queue)
3845 blk_cleanup_queue(mddev->queue);
3846
3847 kfree(mddev);
3848 }
3849
3850 static struct sysfs_ops md_sysfs_ops = {
3851 .show = md_attr_show,
3852 .store = md_attr_store,
3853 };
3854 static struct kobj_type md_ktype = {
3855 .release = md_free,
3856 .sysfs_ops = &md_sysfs_ops,
3857 .default_attrs = md_default_attrs,
3858 };
3859
3860 int mdp_major = 0;
3861
3862 static void mddev_delayed_delete(struct work_struct *ws)
3863 {
3864 mddev_t *mddev = container_of(ws, mddev_t, del_work);
3865
3866 if (mddev->private == &md_redundancy_group) {
3867 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3868 if (mddev->sysfs_action)
3869 sysfs_put(mddev->sysfs_action);
3870 mddev->sysfs_action = NULL;
3871 mddev->private = NULL;
3872 }
3873 kobject_del(&mddev->kobj);
3874 kobject_put(&mddev->kobj);
3875 }
3876
3877 static int md_alloc(dev_t dev, char *name)
3878 {
3879 static DEFINE_MUTEX(disks_mutex);
3880 mddev_t *mddev = mddev_find(dev);
3881 struct gendisk *disk;
3882 int partitioned;
3883 int shift;
3884 int unit;
3885 int error;
3886
3887 if (!mddev)
3888 return -ENODEV;
3889
3890 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
3891 shift = partitioned ? MdpMinorShift : 0;
3892 unit = MINOR(mddev->unit) >> shift;
3893
3894 /* wait for any previous instance if this device
3895 * to be completed removed (mddev_delayed_delete).
3896 */
3897 flush_scheduled_work();
3898
3899 mutex_lock(&disks_mutex);
3900 error = -EEXIST;
3901 if (mddev->gendisk)
3902 goto abort;
3903
3904 if (name) {
3905 /* Need to ensure that 'name' is not a duplicate.
3906 */
3907 mddev_t *mddev2;
3908 spin_lock(&all_mddevs_lock);
3909
3910 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
3911 if (mddev2->gendisk &&
3912 strcmp(mddev2->gendisk->disk_name, name) == 0) {
3913 spin_unlock(&all_mddevs_lock);
3914 goto abort;
3915 }
3916 spin_unlock(&all_mddevs_lock);
3917 }
3918
3919 error = -ENOMEM;
3920 mddev->queue = blk_alloc_queue(GFP_KERNEL);
3921 if (!mddev->queue)
3922 goto abort;
3923 mddev->queue->queuedata = mddev;
3924
3925 /* Can be unlocked because the queue is new: no concurrency */
3926 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
3927
3928 blk_queue_make_request(mddev->queue, md_make_request);
3929
3930 disk = alloc_disk(1 << shift);
3931 if (!disk) {
3932 blk_cleanup_queue(mddev->queue);
3933 mddev->queue = NULL;
3934 goto abort;
3935 }
3936 disk->major = MAJOR(mddev->unit);
3937 disk->first_minor = unit << shift;
3938 if (name)
3939 strcpy(disk->disk_name, name);
3940 else if (partitioned)
3941 sprintf(disk->disk_name, "md_d%d", unit);
3942 else
3943 sprintf(disk->disk_name, "md%d", unit);
3944 disk->fops = &md_fops;
3945 disk->private_data = mddev;
3946 disk->queue = mddev->queue;
3947 /* Allow extended partitions. This makes the
3948 * 'mdp' device redundant, but we can't really
3949 * remove it now.
3950 */
3951 disk->flags |= GENHD_FL_EXT_DEVT;
3952 add_disk(disk);
3953 mddev->gendisk = disk;
3954 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
3955 &disk_to_dev(disk)->kobj, "%s", "md");
3956 if (error) {
3957 /* This isn't possible, but as kobject_init_and_add is marked
3958 * __must_check, we must do something with the result
3959 */
3960 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3961 disk->disk_name);
3962 error = 0;
3963 }
3964 abort:
3965 mutex_unlock(&disks_mutex);
3966 if (!error) {
3967 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3968 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state");
3969 }
3970 mddev_put(mddev);
3971 return error;
3972 }
3973
3974 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3975 {
3976 md_alloc(dev, NULL);
3977 return NULL;
3978 }
3979
3980 static int add_named_array(const char *val, struct kernel_param *kp)
3981 {
3982 /* val must be "md_*" where * is not all digits.
3983 * We allocate an array with a large free minor number, and
3984 * set the name to val. val must not already be an active name.
3985 */
3986 int len = strlen(val);
3987 char buf[DISK_NAME_LEN];
3988
3989 while (len && val[len-1] == '\n')
3990 len--;
3991 if (len >= DISK_NAME_LEN)
3992 return -E2BIG;
3993 strlcpy(buf, val, len+1);
3994 if (strncmp(buf, "md_", 3) != 0)
3995 return -EINVAL;
3996 return md_alloc(0, buf);
3997 }
3998
3999 static void md_safemode_timeout(unsigned long data)
4000 {
4001 mddev_t *mddev = (mddev_t *) data;
4002
4003 if (!atomic_read(&mddev->writes_pending)) {
4004 mddev->safemode = 1;
4005 if (mddev->external)
4006 sysfs_notify_dirent(mddev->sysfs_state);
4007 }
4008 md_wakeup_thread(mddev->thread);
4009 }
4010
4011 static int start_dirty_degraded;
4012
4013 static int do_md_run(mddev_t * mddev)
4014 {
4015 int err;
4016 mdk_rdev_t *rdev;
4017 struct gendisk *disk;
4018 struct mdk_personality *pers;
4019
4020 if (list_empty(&mddev->disks))
4021 /* cannot run an array with no devices.. */
4022 return -EINVAL;
4023
4024 if (mddev->pers)
4025 return -EBUSY;
4026
4027 /*
4028 * Analyze all RAID superblock(s)
4029 */
4030 if (!mddev->raid_disks) {
4031 if (!mddev->persistent)
4032 return -EINVAL;
4033 analyze_sbs(mddev);
4034 }
4035
4036 if (mddev->level != LEVEL_NONE)
4037 request_module("md-level-%d", mddev->level);
4038 else if (mddev->clevel[0])
4039 request_module("md-%s", mddev->clevel);
4040
4041 /*
4042 * Drop all container device buffers, from now on
4043 * the only valid external interface is through the md
4044 * device.
4045 */
4046 list_for_each_entry(rdev, &mddev->disks, same_set) {
4047 if (test_bit(Faulty, &rdev->flags))
4048 continue;
4049 sync_blockdev(rdev->bdev);
4050 invalidate_bdev(rdev->bdev);
4051
4052 /* perform some consistency tests on the device.
4053 * We don't want the data to overlap the metadata,
4054 * Internal Bitmap issues have been handled elsewhere.
4055 */
4056 if (rdev->data_offset < rdev->sb_start) {
4057 if (mddev->dev_sectors &&
4058 rdev->data_offset + mddev->dev_sectors
4059 > rdev->sb_start) {
4060 printk("md: %s: data overlaps metadata\n",
4061 mdname(mddev));
4062 return -EINVAL;
4063 }
4064 } else {
4065 if (rdev->sb_start + rdev->sb_size/512
4066 > rdev->data_offset) {
4067 printk("md: %s: metadata overlaps data\n",
4068 mdname(mddev));
4069 return -EINVAL;
4070 }
4071 }
4072 sysfs_notify_dirent(rdev->sysfs_state);
4073 }
4074
4075 md_probe(mddev->unit, NULL, NULL);
4076 disk = mddev->gendisk;
4077 if (!disk)
4078 return -ENOMEM;
4079
4080 spin_lock(&pers_lock);
4081 pers = find_pers(mddev->level, mddev->clevel);
4082 if (!pers || !try_module_get(pers->owner)) {
4083 spin_unlock(&pers_lock);
4084 if (mddev->level != LEVEL_NONE)
4085 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4086 mddev->level);
4087 else
4088 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4089 mddev->clevel);
4090 return -EINVAL;
4091 }
4092 mddev->pers = pers;
4093 spin_unlock(&pers_lock);
4094 if (mddev->level != pers->level) {
4095 mddev->level = pers->level;
4096 mddev->new_level = pers->level;
4097 }
4098 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4099
4100 if (mddev->reshape_position != MaxSector &&
4101 pers->start_reshape == NULL) {
4102 /* This personality cannot handle reshaping... */
4103 mddev->pers = NULL;
4104 module_put(pers->owner);
4105 return -EINVAL;
4106 }
4107
4108 if (pers->sync_request) {
4109 /* Warn if this is a potentially silly
4110 * configuration.
4111 */
4112 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4113 mdk_rdev_t *rdev2;
4114 int warned = 0;
4115
4116 list_for_each_entry(rdev, &mddev->disks, same_set)
4117 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4118 if (rdev < rdev2 &&
4119 rdev->bdev->bd_contains ==
4120 rdev2->bdev->bd_contains) {
4121 printk(KERN_WARNING
4122 "%s: WARNING: %s appears to be"
4123 " on the same physical disk as"
4124 " %s.\n",
4125 mdname(mddev),
4126 bdevname(rdev->bdev,b),
4127 bdevname(rdev2->bdev,b2));
4128 warned = 1;
4129 }
4130 }
4131
4132 if (warned)
4133 printk(KERN_WARNING
4134 "True protection against single-disk"
4135 " failure might be compromised.\n");
4136 }
4137
4138 mddev->recovery = 0;
4139 /* may be over-ridden by personality */
4140 mddev->resync_max_sectors = mddev->dev_sectors;
4141
4142 mddev->barriers_work = 1;
4143 mddev->ok_start_degraded = start_dirty_degraded;
4144
4145 if (start_readonly)
4146 mddev->ro = 2; /* read-only, but switch on first write */
4147
4148 err = mddev->pers->run(mddev);
4149 if (err)
4150 printk(KERN_ERR "md: pers->run() failed ...\n");
4151 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4152 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4153 " but 'external_size' not in effect?\n", __func__);
4154 printk(KERN_ERR
4155 "md: invalid array_size %llu > default size %llu\n",
4156 (unsigned long long)mddev->array_sectors / 2,
4157 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4158 err = -EINVAL;
4159 mddev->pers->stop(mddev);
4160 }
4161 if (err == 0 && mddev->pers->sync_request) {
4162 err = bitmap_create(mddev);
4163 if (err) {
4164 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4165 mdname(mddev), err);
4166 mddev->pers->stop(mddev);
4167 }
4168 }
4169 if (err) {
4170 module_put(mddev->pers->owner);
4171 mddev->pers = NULL;
4172 bitmap_destroy(mddev);
4173 return err;
4174 }
4175 if (mddev->pers->sync_request) {
4176 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4177 printk(KERN_WARNING
4178 "md: cannot register extra attributes for %s\n",
4179 mdname(mddev));
4180 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4181 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4182 mddev->ro = 0;
4183
4184 atomic_set(&mddev->writes_pending,0);
4185 mddev->safemode = 0;
4186 mddev->safemode_timer.function = md_safemode_timeout;
4187 mddev->safemode_timer.data = (unsigned long) mddev;
4188 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4189 mddev->in_sync = 1;
4190
4191 list_for_each_entry(rdev, &mddev->disks, same_set)
4192 if (rdev->raid_disk >= 0) {
4193 char nm[20];
4194 sprintf(nm, "rd%d", rdev->raid_disk);
4195 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4196 printk("md: cannot register %s for %s\n",
4197 nm, mdname(mddev));
4198 }
4199
4200 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4201
4202 if (mddev->flags)
4203 md_update_sb(mddev, 0);
4204
4205 set_capacity(disk, mddev->array_sectors);
4206
4207 /* If there is a partially-recovered drive we need to
4208 * start recovery here. If we leave it to md_check_recovery,
4209 * it will remove the drives and not do the right thing
4210 */
4211 if (mddev->degraded && !mddev->sync_thread) {
4212 int spares = 0;
4213 list_for_each_entry(rdev, &mddev->disks, same_set)
4214 if (rdev->raid_disk >= 0 &&
4215 !test_bit(In_sync, &rdev->flags) &&
4216 !test_bit(Faulty, &rdev->flags))
4217 /* complete an interrupted recovery */
4218 spares++;
4219 if (spares && mddev->pers->sync_request) {
4220 mddev->recovery = 0;
4221 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4222 mddev->sync_thread = md_register_thread(md_do_sync,
4223 mddev,
4224 "%s_resync");
4225 if (!mddev->sync_thread) {
4226 printk(KERN_ERR "%s: could not start resync"
4227 " thread...\n",
4228 mdname(mddev));
4229 /* leave the spares where they are, it shouldn't hurt */
4230 mddev->recovery = 0;
4231 }
4232 }
4233 }
4234 md_wakeup_thread(mddev->thread);
4235 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4236
4237 revalidate_disk(mddev->gendisk);
4238 mddev->changed = 1;
4239 md_new_event(mddev);
4240 sysfs_notify_dirent(mddev->sysfs_state);
4241 if (mddev->sysfs_action)
4242 sysfs_notify_dirent(mddev->sysfs_action);
4243 sysfs_notify(&mddev->kobj, NULL, "degraded");
4244 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4245 return 0;
4246 }
4247
4248 static int restart_array(mddev_t *mddev)
4249 {
4250 struct gendisk *disk = mddev->gendisk;
4251
4252 /* Complain if it has no devices */
4253 if (list_empty(&mddev->disks))
4254 return -ENXIO;
4255 if (!mddev->pers)
4256 return -EINVAL;
4257 if (!mddev->ro)
4258 return -EBUSY;
4259 mddev->safemode = 0;
4260 mddev->ro = 0;
4261 set_disk_ro(disk, 0);
4262 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4263 mdname(mddev));
4264 /* Kick recovery or resync if necessary */
4265 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4266 md_wakeup_thread(mddev->thread);
4267 md_wakeup_thread(mddev->sync_thread);
4268 sysfs_notify_dirent(mddev->sysfs_state);
4269 return 0;
4270 }
4271
4272 /* similar to deny_write_access, but accounts for our holding a reference
4273 * to the file ourselves */
4274 static int deny_bitmap_write_access(struct file * file)
4275 {
4276 struct inode *inode = file->f_mapping->host;
4277
4278 spin_lock(&inode->i_lock);
4279 if (atomic_read(&inode->i_writecount) > 1) {
4280 spin_unlock(&inode->i_lock);
4281 return -ETXTBSY;
4282 }
4283 atomic_set(&inode->i_writecount, -1);
4284 spin_unlock(&inode->i_lock);
4285
4286 return 0;
4287 }
4288
4289 static void restore_bitmap_write_access(struct file *file)
4290 {
4291 struct inode *inode = file->f_mapping->host;
4292
4293 spin_lock(&inode->i_lock);
4294 atomic_set(&inode->i_writecount, 1);
4295 spin_unlock(&inode->i_lock);
4296 }
4297
4298 /* mode:
4299 * 0 - completely stop and dis-assemble array
4300 * 1 - switch to readonly
4301 * 2 - stop but do not disassemble array
4302 */
4303 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4304 {
4305 int err = 0;
4306 struct gendisk *disk = mddev->gendisk;
4307 mdk_rdev_t *rdev;
4308
4309 mutex_lock(&mddev->open_mutex);
4310 if (atomic_read(&mddev->openers) > is_open) {
4311 printk("md: %s still in use.\n",mdname(mddev));
4312 err = -EBUSY;
4313 } else if (mddev->pers) {
4314
4315 if (mddev->sync_thread) {
4316 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4317 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4318 md_unregister_thread(mddev->sync_thread);
4319 mddev->sync_thread = NULL;
4320 }
4321
4322 del_timer_sync(&mddev->safemode_timer);
4323
4324 switch(mode) {
4325 case 1: /* readonly */
4326 err = -ENXIO;
4327 if (mddev->ro==1)
4328 goto out;
4329 mddev->ro = 1;
4330 break;
4331 case 0: /* disassemble */
4332 case 2: /* stop */
4333 bitmap_flush(mddev);
4334 md_super_wait(mddev);
4335 if (mddev->ro)
4336 set_disk_ro(disk, 0);
4337
4338 mddev->pers->stop(mddev);
4339 mddev->queue->merge_bvec_fn = NULL;
4340 mddev->queue->unplug_fn = NULL;
4341 mddev->queue->backing_dev_info.congested_fn = NULL;
4342 module_put(mddev->pers->owner);
4343 if (mddev->pers->sync_request)
4344 mddev->private = &md_redundancy_group;
4345 mddev->pers = NULL;
4346 /* tell userspace to handle 'inactive' */
4347 sysfs_notify_dirent(mddev->sysfs_state);
4348
4349 list_for_each_entry(rdev, &mddev->disks, same_set)
4350 if (rdev->raid_disk >= 0) {
4351 char nm[20];
4352 sprintf(nm, "rd%d", rdev->raid_disk);
4353 sysfs_remove_link(&mddev->kobj, nm);
4354 }
4355
4356 set_capacity(disk, 0);
4357 mddev->changed = 1;
4358
4359 if (mddev->ro)
4360 mddev->ro = 0;
4361 }
4362 if (!mddev->in_sync || mddev->flags) {
4363 /* mark array as shutdown cleanly */
4364 mddev->in_sync = 1;
4365 md_update_sb(mddev, 1);
4366 }
4367 if (mode == 1)
4368 set_disk_ro(disk, 1);
4369 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4370 err = 0;
4371 }
4372 out:
4373 mutex_unlock(&mddev->open_mutex);
4374 if (err)
4375 return err;
4376 /*
4377 * Free resources if final stop
4378 */
4379 if (mode == 0) {
4380
4381 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4382
4383 bitmap_destroy(mddev);
4384 if (mddev->bitmap_file) {
4385 restore_bitmap_write_access(mddev->bitmap_file);
4386 fput(mddev->bitmap_file);
4387 mddev->bitmap_file = NULL;
4388 }
4389 mddev->bitmap_offset = 0;
4390
4391 /* make sure all md_delayed_delete calls have finished */
4392 flush_scheduled_work();
4393
4394 export_array(mddev);
4395
4396 mddev->array_sectors = 0;
4397 mddev->external_size = 0;
4398 mddev->dev_sectors = 0;
4399 mddev->raid_disks = 0;
4400 mddev->recovery_cp = 0;
4401 mddev->resync_min = 0;
4402 mddev->resync_max = MaxSector;
4403 mddev->reshape_position = MaxSector;
4404 mddev->external = 0;
4405 mddev->persistent = 0;
4406 mddev->level = LEVEL_NONE;
4407 mddev->clevel[0] = 0;
4408 mddev->flags = 0;
4409 mddev->ro = 0;
4410 mddev->metadata_type[0] = 0;
4411 mddev->chunk_sectors = 0;
4412 mddev->ctime = mddev->utime = 0;
4413 mddev->layout = 0;
4414 mddev->max_disks = 0;
4415 mddev->events = 0;
4416 mddev->delta_disks = 0;
4417 mddev->new_level = LEVEL_NONE;
4418 mddev->new_layout = 0;
4419 mddev->new_chunk_sectors = 0;
4420 mddev->curr_resync = 0;
4421 mddev->resync_mismatches = 0;
4422 mddev->suspend_lo = mddev->suspend_hi = 0;
4423 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4424 mddev->recovery = 0;
4425 mddev->in_sync = 0;
4426 mddev->changed = 0;
4427 mddev->degraded = 0;
4428 mddev->barriers_work = 0;
4429 mddev->safemode = 0;
4430 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4431 if (mddev->hold_active == UNTIL_STOP)
4432 mddev->hold_active = 0;
4433
4434 } else if (mddev->pers)
4435 printk(KERN_INFO "md: %s switched to read-only mode.\n",
4436 mdname(mddev));
4437 err = 0;
4438 blk_integrity_unregister(disk);
4439 md_new_event(mddev);
4440 sysfs_notify_dirent(mddev->sysfs_state);
4441 return err;
4442 }
4443
4444 #ifndef MODULE
4445 static void autorun_array(mddev_t *mddev)
4446 {
4447 mdk_rdev_t *rdev;
4448 int err;
4449
4450 if (list_empty(&mddev->disks))
4451 return;
4452
4453 printk(KERN_INFO "md: running: ");
4454
4455 list_for_each_entry(rdev, &mddev->disks, same_set) {
4456 char b[BDEVNAME_SIZE];
4457 printk("<%s>", bdevname(rdev->bdev,b));
4458 }
4459 printk("\n");
4460
4461 err = do_md_run(mddev);
4462 if (err) {
4463 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4464 do_md_stop(mddev, 0, 0);
4465 }
4466 }
4467
4468 /*
4469 * lets try to run arrays based on all disks that have arrived
4470 * until now. (those are in pending_raid_disks)
4471 *
4472 * the method: pick the first pending disk, collect all disks with
4473 * the same UUID, remove all from the pending list and put them into
4474 * the 'same_array' list. Then order this list based on superblock
4475 * update time (freshest comes first), kick out 'old' disks and
4476 * compare superblocks. If everything's fine then run it.
4477 *
4478 * If "unit" is allocated, then bump its reference count
4479 */
4480 static void autorun_devices(int part)
4481 {
4482 mdk_rdev_t *rdev0, *rdev, *tmp;
4483 mddev_t *mddev;
4484 char b[BDEVNAME_SIZE];
4485
4486 printk(KERN_INFO "md: autorun ...\n");
4487 while (!list_empty(&pending_raid_disks)) {
4488 int unit;
4489 dev_t dev;
4490 LIST_HEAD(candidates);
4491 rdev0 = list_entry(pending_raid_disks.next,
4492 mdk_rdev_t, same_set);
4493
4494 printk(KERN_INFO "md: considering %s ...\n",
4495 bdevname(rdev0->bdev,b));
4496 INIT_LIST_HEAD(&candidates);
4497 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4498 if (super_90_load(rdev, rdev0, 0) >= 0) {
4499 printk(KERN_INFO "md: adding %s ...\n",
4500 bdevname(rdev->bdev,b));
4501 list_move(&rdev->same_set, &candidates);
4502 }
4503 /*
4504 * now we have a set of devices, with all of them having
4505 * mostly sane superblocks. It's time to allocate the
4506 * mddev.
4507 */
4508 if (part) {
4509 dev = MKDEV(mdp_major,
4510 rdev0->preferred_minor << MdpMinorShift);
4511 unit = MINOR(dev) >> MdpMinorShift;
4512 } else {
4513 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4514 unit = MINOR(dev);
4515 }
4516 if (rdev0->preferred_minor != unit) {
4517 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4518 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4519 break;
4520 }
4521
4522 md_probe(dev, NULL, NULL);
4523 mddev = mddev_find(dev);
4524 if (!mddev || !mddev->gendisk) {
4525 if (mddev)
4526 mddev_put(mddev);
4527 printk(KERN_ERR
4528 "md: cannot allocate memory for md drive.\n");
4529 break;
4530 }
4531 if (mddev_lock(mddev))
4532 printk(KERN_WARNING "md: %s locked, cannot run\n",
4533 mdname(mddev));
4534 else if (mddev->raid_disks || mddev->major_version
4535 || !list_empty(&mddev->disks)) {
4536 printk(KERN_WARNING
4537 "md: %s already running, cannot run %s\n",
4538 mdname(mddev), bdevname(rdev0->bdev,b));
4539 mddev_unlock(mddev);
4540 } else {
4541 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4542 mddev->persistent = 1;
4543 rdev_for_each_list(rdev, tmp, &candidates) {
4544 list_del_init(&rdev->same_set);
4545 if (bind_rdev_to_array(rdev, mddev))
4546 export_rdev(rdev);
4547 }
4548 autorun_array(mddev);
4549 mddev_unlock(mddev);
4550 }
4551 /* on success, candidates will be empty, on error
4552 * it won't...
4553 */
4554 rdev_for_each_list(rdev, tmp, &candidates) {
4555 list_del_init(&rdev->same_set);
4556 export_rdev(rdev);
4557 }
4558 mddev_put(mddev);
4559 }
4560 printk(KERN_INFO "md: ... autorun DONE.\n");
4561 }
4562 #endif /* !MODULE */
4563
4564 static int get_version(void __user * arg)
4565 {
4566 mdu_version_t ver;
4567
4568 ver.major = MD_MAJOR_VERSION;
4569 ver.minor = MD_MINOR_VERSION;
4570 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4571
4572 if (copy_to_user(arg, &ver, sizeof(ver)))
4573 return -EFAULT;
4574
4575 return 0;
4576 }
4577
4578 static int get_array_info(mddev_t * mddev, void __user * arg)
4579 {
4580 mdu_array_info_t info;
4581 int nr,working,active,failed,spare;
4582 mdk_rdev_t *rdev;
4583
4584 nr=working=active=failed=spare=0;
4585 list_for_each_entry(rdev, &mddev->disks, same_set) {
4586 nr++;
4587 if (test_bit(Faulty, &rdev->flags))
4588 failed++;
4589 else {
4590 working++;
4591 if (test_bit(In_sync, &rdev->flags))
4592 active++;
4593 else
4594 spare++;
4595 }
4596 }
4597
4598 info.major_version = mddev->major_version;
4599 info.minor_version = mddev->minor_version;
4600 info.patch_version = MD_PATCHLEVEL_VERSION;
4601 info.ctime = mddev->ctime;
4602 info.level = mddev->level;
4603 info.size = mddev->dev_sectors / 2;
4604 if (info.size != mddev->dev_sectors / 2) /* overflow */
4605 info.size = -1;
4606 info.nr_disks = nr;
4607 info.raid_disks = mddev->raid_disks;
4608 info.md_minor = mddev->md_minor;
4609 info.not_persistent= !mddev->persistent;
4610
4611 info.utime = mddev->utime;
4612 info.state = 0;
4613 if (mddev->in_sync)
4614 info.state = (1<<MD_SB_CLEAN);
4615 if (mddev->bitmap && mddev->bitmap_offset)
4616 info.state = (1<<MD_SB_BITMAP_PRESENT);
4617 info.active_disks = active;
4618 info.working_disks = working;
4619 info.failed_disks = failed;
4620 info.spare_disks = spare;
4621
4622 info.layout = mddev->layout;
4623 info.chunk_size = mddev->chunk_sectors << 9;
4624
4625 if (copy_to_user(arg, &info, sizeof(info)))
4626 return -EFAULT;
4627
4628 return 0;
4629 }
4630
4631 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4632 {
4633 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4634 char *ptr, *buf = NULL;
4635 int err = -ENOMEM;
4636
4637 if (md_allow_write(mddev))
4638 file = kmalloc(sizeof(*file), GFP_NOIO);
4639 else
4640 file = kmalloc(sizeof(*file), GFP_KERNEL);
4641
4642 if (!file)
4643 goto out;
4644
4645 /* bitmap disabled, zero the first byte and copy out */
4646 if (!mddev->bitmap || !mddev->bitmap->file) {
4647 file->pathname[0] = '\0';
4648 goto copy_out;
4649 }
4650
4651 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4652 if (!buf)
4653 goto out;
4654
4655 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4656 if (IS_ERR(ptr))
4657 goto out;
4658
4659 strcpy(file->pathname, ptr);
4660
4661 copy_out:
4662 err = 0;
4663 if (copy_to_user(arg, file, sizeof(*file)))
4664 err = -EFAULT;
4665 out:
4666 kfree(buf);
4667 kfree(file);
4668 return err;
4669 }
4670
4671 static int get_disk_info(mddev_t * mddev, void __user * arg)
4672 {
4673 mdu_disk_info_t info;
4674 mdk_rdev_t *rdev;
4675
4676 if (copy_from_user(&info, arg, sizeof(info)))
4677 return -EFAULT;
4678
4679 rdev = find_rdev_nr(mddev, info.number);
4680 if (rdev) {
4681 info.major = MAJOR(rdev->bdev->bd_dev);
4682 info.minor = MINOR(rdev->bdev->bd_dev);
4683 info.raid_disk = rdev->raid_disk;
4684 info.state = 0;
4685 if (test_bit(Faulty, &rdev->flags))
4686 info.state |= (1<<MD_DISK_FAULTY);
4687 else if (test_bit(In_sync, &rdev->flags)) {
4688 info.state |= (1<<MD_DISK_ACTIVE);
4689 info.state |= (1<<MD_DISK_SYNC);
4690 }
4691 if (test_bit(WriteMostly, &rdev->flags))
4692 info.state |= (1<<MD_DISK_WRITEMOSTLY);
4693 } else {
4694 info.major = info.minor = 0;
4695 info.raid_disk = -1;
4696 info.state = (1<<MD_DISK_REMOVED);
4697 }
4698
4699 if (copy_to_user(arg, &info, sizeof(info)))
4700 return -EFAULT;
4701
4702 return 0;
4703 }
4704
4705 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4706 {
4707 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4708 mdk_rdev_t *rdev;
4709 dev_t dev = MKDEV(info->major,info->minor);
4710
4711 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4712 return -EOVERFLOW;
4713
4714 if (!mddev->raid_disks) {
4715 int err;
4716 /* expecting a device which has a superblock */
4717 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4718 if (IS_ERR(rdev)) {
4719 printk(KERN_WARNING
4720 "md: md_import_device returned %ld\n",
4721 PTR_ERR(rdev));
4722 return PTR_ERR(rdev);
4723 }
4724 if (!list_empty(&mddev->disks)) {
4725 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4726 mdk_rdev_t, same_set);
4727 int err = super_types[mddev->major_version]
4728 .load_super(rdev, rdev0, mddev->minor_version);
4729 if (err < 0) {
4730 printk(KERN_WARNING
4731 "md: %s has different UUID to %s\n",
4732 bdevname(rdev->bdev,b),
4733 bdevname(rdev0->bdev,b2));
4734 export_rdev(rdev);
4735 return -EINVAL;
4736 }
4737 }
4738 err = bind_rdev_to_array(rdev, mddev);
4739 if (err)
4740 export_rdev(rdev);
4741 return err;
4742 }
4743
4744 /*
4745 * add_new_disk can be used once the array is assembled
4746 * to add "hot spares". They must already have a superblock
4747 * written
4748 */
4749 if (mddev->pers) {
4750 int err;
4751 if (!mddev->pers->hot_add_disk) {
4752 printk(KERN_WARNING
4753 "%s: personality does not support diskops!\n",
4754 mdname(mddev));
4755 return -EINVAL;
4756 }
4757 if (mddev->persistent)
4758 rdev = md_import_device(dev, mddev->major_version,
4759 mddev->minor_version);
4760 else
4761 rdev = md_import_device(dev, -1, -1);
4762 if (IS_ERR(rdev)) {
4763 printk(KERN_WARNING
4764 "md: md_import_device returned %ld\n",
4765 PTR_ERR(rdev));
4766 return PTR_ERR(rdev);
4767 }
4768 /* set save_raid_disk if appropriate */
4769 if (!mddev->persistent) {
4770 if (info->state & (1<<MD_DISK_SYNC) &&
4771 info->raid_disk < mddev->raid_disks)
4772 rdev->raid_disk = info->raid_disk;
4773 else
4774 rdev->raid_disk = -1;
4775 } else
4776 super_types[mddev->major_version].
4777 validate_super(mddev, rdev);
4778 rdev->saved_raid_disk = rdev->raid_disk;
4779
4780 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4781 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4782 set_bit(WriteMostly, &rdev->flags);
4783 else
4784 clear_bit(WriteMostly, &rdev->flags);
4785
4786 rdev->raid_disk = -1;
4787 err = bind_rdev_to_array(rdev, mddev);
4788 if (!err && !mddev->pers->hot_remove_disk) {
4789 /* If there is hot_add_disk but no hot_remove_disk
4790 * then added disks for geometry changes,
4791 * and should be added immediately.
4792 */
4793 super_types[mddev->major_version].
4794 validate_super(mddev, rdev);
4795 err = mddev->pers->hot_add_disk(mddev, rdev);
4796 if (err)
4797 unbind_rdev_from_array(rdev);
4798 }
4799 if (err)
4800 export_rdev(rdev);
4801 else
4802 sysfs_notify_dirent(rdev->sysfs_state);
4803
4804 md_update_sb(mddev, 1);
4805 if (mddev->degraded)
4806 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4807 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4808 md_wakeup_thread(mddev->thread);
4809 return err;
4810 }
4811
4812 /* otherwise, add_new_disk is only allowed
4813 * for major_version==0 superblocks
4814 */
4815 if (mddev->major_version != 0) {
4816 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4817 mdname(mddev));
4818 return -EINVAL;
4819 }
4820
4821 if (!(info->state & (1<<MD_DISK_FAULTY))) {
4822 int err;
4823 rdev = md_import_device(dev, -1, 0);
4824 if (IS_ERR(rdev)) {
4825 printk(KERN_WARNING
4826 "md: error, md_import_device() returned %ld\n",
4827 PTR_ERR(rdev));
4828 return PTR_ERR(rdev);
4829 }
4830 rdev->desc_nr = info->number;
4831 if (info->raid_disk < mddev->raid_disks)
4832 rdev->raid_disk = info->raid_disk;
4833 else
4834 rdev->raid_disk = -1;
4835
4836 if (rdev->raid_disk < mddev->raid_disks)
4837 if (info->state & (1<<MD_DISK_SYNC))
4838 set_bit(In_sync, &rdev->flags);
4839
4840 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4841 set_bit(WriteMostly, &rdev->flags);
4842
4843 if (!mddev->persistent) {
4844 printk(KERN_INFO "md: nonpersistent superblock ...\n");
4845 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4846 } else
4847 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4848 rdev->sectors = rdev->sb_start;
4849
4850 err = bind_rdev_to_array(rdev, mddev);
4851 if (err) {
4852 export_rdev(rdev);
4853 return err;
4854 }
4855 }
4856
4857 return 0;
4858 }
4859
4860 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4861 {
4862 char b[BDEVNAME_SIZE];
4863 mdk_rdev_t *rdev;
4864
4865 rdev = find_rdev(mddev, dev);
4866 if (!rdev)
4867 return -ENXIO;
4868
4869 if (rdev->raid_disk >= 0)
4870 goto busy;
4871
4872 kick_rdev_from_array(rdev);
4873 md_update_sb(mddev, 1);
4874 md_new_event(mddev);
4875
4876 return 0;
4877 busy:
4878 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4879 bdevname(rdev->bdev,b), mdname(mddev));
4880 return -EBUSY;
4881 }
4882
4883 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4884 {
4885 char b[BDEVNAME_SIZE];
4886 int err;
4887 mdk_rdev_t *rdev;
4888
4889 if (!mddev->pers)
4890 return -ENODEV;
4891
4892 if (mddev->major_version != 0) {
4893 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4894 " version-0 superblocks.\n",
4895 mdname(mddev));
4896 return -EINVAL;
4897 }
4898 if (!mddev->pers->hot_add_disk) {
4899 printk(KERN_WARNING
4900 "%s: personality does not support diskops!\n",
4901 mdname(mddev));
4902 return -EINVAL;
4903 }
4904
4905 rdev = md_import_device(dev, -1, 0);
4906 if (IS_ERR(rdev)) {
4907 printk(KERN_WARNING
4908 "md: error, md_import_device() returned %ld\n",
4909 PTR_ERR(rdev));
4910 return -EINVAL;
4911 }
4912
4913 if (mddev->persistent)
4914 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4915 else
4916 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4917
4918 rdev->sectors = rdev->sb_start;
4919
4920 if (test_bit(Faulty, &rdev->flags)) {
4921 printk(KERN_WARNING
4922 "md: can not hot-add faulty %s disk to %s!\n",
4923 bdevname(rdev->bdev,b), mdname(mddev));
4924 err = -EINVAL;
4925 goto abort_export;
4926 }
4927 clear_bit(In_sync, &rdev->flags);
4928 rdev->desc_nr = -1;
4929 rdev->saved_raid_disk = -1;
4930 err = bind_rdev_to_array(rdev, mddev);
4931 if (err)
4932 goto abort_export;
4933
4934 /*
4935 * The rest should better be atomic, we can have disk failures
4936 * noticed in interrupt contexts ...
4937 */
4938
4939 rdev->raid_disk = -1;
4940
4941 md_update_sb(mddev, 1);
4942
4943 /*
4944 * Kick recovery, maybe this spare has to be added to the
4945 * array immediately.
4946 */
4947 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4948 md_wakeup_thread(mddev->thread);
4949 md_new_event(mddev);
4950 return 0;
4951
4952 abort_export:
4953 export_rdev(rdev);
4954 return err;
4955 }
4956
4957 static int set_bitmap_file(mddev_t *mddev, int fd)
4958 {
4959 int err;
4960
4961 if (mddev->pers) {
4962 if (!mddev->pers->quiesce)
4963 return -EBUSY;
4964 if (mddev->recovery || mddev->sync_thread)
4965 return -EBUSY;
4966 /* we should be able to change the bitmap.. */
4967 }
4968
4969
4970 if (fd >= 0) {
4971 if (mddev->bitmap)
4972 return -EEXIST; /* cannot add when bitmap is present */
4973 mddev->bitmap_file = fget(fd);
4974
4975 if (mddev->bitmap_file == NULL) {
4976 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4977 mdname(mddev));
4978 return -EBADF;
4979 }
4980
4981 err = deny_bitmap_write_access(mddev->bitmap_file);
4982 if (err) {
4983 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4984 mdname(mddev));
4985 fput(mddev->bitmap_file);
4986 mddev->bitmap_file = NULL;
4987 return err;
4988 }
4989 mddev->bitmap_offset = 0; /* file overrides offset */
4990 } else if (mddev->bitmap == NULL)
4991 return -ENOENT; /* cannot remove what isn't there */
4992 err = 0;
4993 if (mddev->pers) {
4994 mddev->pers->quiesce(mddev, 1);
4995 if (fd >= 0)
4996 err = bitmap_create(mddev);
4997 if (fd < 0 || err) {
4998 bitmap_destroy(mddev);
4999 fd = -1; /* make sure to put the file */
5000 }
5001 mddev->pers->quiesce(mddev, 0);
5002 }
5003 if (fd < 0) {
5004 if (mddev->bitmap_file) {
5005 restore_bitmap_write_access(mddev->bitmap_file);
5006 fput(mddev->bitmap_file);
5007 }
5008 mddev->bitmap_file = NULL;
5009 }
5010
5011 return err;
5012 }
5013
5014 /*
5015 * set_array_info is used two different ways
5016 * The original usage is when creating a new array.
5017 * In this usage, raid_disks is > 0 and it together with
5018 * level, size, not_persistent,layout,chunksize determine the
5019 * shape of the array.
5020 * This will always create an array with a type-0.90.0 superblock.
5021 * The newer usage is when assembling an array.
5022 * In this case raid_disks will be 0, and the major_version field is
5023 * use to determine which style super-blocks are to be found on the devices.
5024 * The minor and patch _version numbers are also kept incase the
5025 * super_block handler wishes to interpret them.
5026 */
5027 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5028 {
5029
5030 if (info->raid_disks == 0) {
5031 /* just setting version number for superblock loading */
5032 if (info->major_version < 0 ||
5033 info->major_version >= ARRAY_SIZE(super_types) ||
5034 super_types[info->major_version].name == NULL) {
5035 /* maybe try to auto-load a module? */
5036 printk(KERN_INFO
5037 "md: superblock version %d not known\n",
5038 info->major_version);
5039 return -EINVAL;
5040 }
5041 mddev->major_version = info->major_version;
5042 mddev->minor_version = info->minor_version;
5043 mddev->patch_version = info->patch_version;
5044 mddev->persistent = !info->not_persistent;
5045 /* ensure mddev_put doesn't delete this now that there
5046 * is some minimal configuration.
5047 */
5048 mddev->ctime = get_seconds();
5049 return 0;
5050 }
5051 mddev->major_version = MD_MAJOR_VERSION;
5052 mddev->minor_version = MD_MINOR_VERSION;
5053 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5054 mddev->ctime = get_seconds();
5055
5056 mddev->level = info->level;
5057 mddev->clevel[0] = 0;
5058 mddev->dev_sectors = 2 * (sector_t)info->size;
5059 mddev->raid_disks = info->raid_disks;
5060 /* don't set md_minor, it is determined by which /dev/md* was
5061 * openned
5062 */
5063 if (info->state & (1<<MD_SB_CLEAN))
5064 mddev->recovery_cp = MaxSector;
5065 else
5066 mddev->recovery_cp = 0;
5067 mddev->persistent = ! info->not_persistent;
5068 mddev->external = 0;
5069
5070 mddev->layout = info->layout;
5071 mddev->chunk_sectors = info->chunk_size >> 9;
5072
5073 mddev->max_disks = MD_SB_DISKS;
5074
5075 if (mddev->persistent)
5076 mddev->flags = 0;
5077 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5078
5079 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
5080 mddev->bitmap_offset = 0;
5081
5082 mddev->reshape_position = MaxSector;
5083
5084 /*
5085 * Generate a 128 bit UUID
5086 */
5087 get_random_bytes(mddev->uuid, 16);
5088
5089 mddev->new_level = mddev->level;
5090 mddev->new_chunk_sectors = mddev->chunk_sectors;
5091 mddev->new_layout = mddev->layout;
5092 mddev->delta_disks = 0;
5093
5094 return 0;
5095 }
5096
5097 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5098 {
5099 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5100
5101 if (mddev->external_size)
5102 return;
5103
5104 mddev->array_sectors = array_sectors;
5105 }
5106 EXPORT_SYMBOL(md_set_array_sectors);
5107
5108 static int update_size(mddev_t *mddev, sector_t num_sectors)
5109 {
5110 mdk_rdev_t *rdev;
5111 int rv;
5112 int fit = (num_sectors == 0);
5113
5114 if (mddev->pers->resize == NULL)
5115 return -EINVAL;
5116 /* The "num_sectors" is the number of sectors of each device that
5117 * is used. This can only make sense for arrays with redundancy.
5118 * linear and raid0 always use whatever space is available. We can only
5119 * consider changing this number if no resync or reconstruction is
5120 * happening, and if the new size is acceptable. It must fit before the
5121 * sb_start or, if that is <data_offset, it must fit before the size
5122 * of each device. If num_sectors is zero, we find the largest size
5123 * that fits.
5124
5125 */
5126 if (mddev->sync_thread)
5127 return -EBUSY;
5128 if (mddev->bitmap)
5129 /* Sorry, cannot grow a bitmap yet, just remove it,
5130 * grow, and re-add.
5131 */
5132 return -EBUSY;
5133 list_for_each_entry(rdev, &mddev->disks, same_set) {
5134 sector_t avail = rdev->sectors;
5135
5136 if (fit && (num_sectors == 0 || num_sectors > avail))
5137 num_sectors = avail;
5138 if (avail < num_sectors)
5139 return -ENOSPC;
5140 }
5141 rv = mddev->pers->resize(mddev, num_sectors);
5142 if (!rv)
5143 revalidate_disk(mddev->gendisk);
5144 return rv;
5145 }
5146
5147 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5148 {
5149 int rv;
5150 /* change the number of raid disks */
5151 if (mddev->pers->check_reshape == NULL)
5152 return -EINVAL;
5153 if (raid_disks <= 0 ||
5154 raid_disks >= mddev->max_disks)
5155 return -EINVAL;
5156 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5157 return -EBUSY;
5158 mddev->delta_disks = raid_disks - mddev->raid_disks;
5159
5160 rv = mddev->pers->check_reshape(mddev);
5161 return rv;
5162 }
5163
5164
5165 /*
5166 * update_array_info is used to change the configuration of an
5167 * on-line array.
5168 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5169 * fields in the info are checked against the array.
5170 * Any differences that cannot be handled will cause an error.
5171 * Normally, only one change can be managed at a time.
5172 */
5173 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5174 {
5175 int rv = 0;
5176 int cnt = 0;
5177 int state = 0;
5178
5179 /* calculate expected state,ignoring low bits */
5180 if (mddev->bitmap && mddev->bitmap_offset)
5181 state |= (1 << MD_SB_BITMAP_PRESENT);
5182
5183 if (mddev->major_version != info->major_version ||
5184 mddev->minor_version != info->minor_version ||
5185 /* mddev->patch_version != info->patch_version || */
5186 mddev->ctime != info->ctime ||
5187 mddev->level != info->level ||
5188 /* mddev->layout != info->layout || */
5189 !mddev->persistent != info->not_persistent||
5190 mddev->chunk_sectors != info->chunk_size >> 9 ||
5191 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5192 ((state^info->state) & 0xfffffe00)
5193 )
5194 return -EINVAL;
5195 /* Check there is only one change */
5196 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5197 cnt++;
5198 if (mddev->raid_disks != info->raid_disks)
5199 cnt++;
5200 if (mddev->layout != info->layout)
5201 cnt++;
5202 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5203 cnt++;
5204 if (cnt == 0)
5205 return 0;
5206 if (cnt > 1)
5207 return -EINVAL;
5208
5209 if (mddev->layout != info->layout) {
5210 /* Change layout
5211 * we don't need to do anything at the md level, the
5212 * personality will take care of it all.
5213 */
5214 if (mddev->pers->check_reshape == NULL)
5215 return -EINVAL;
5216 else {
5217 mddev->new_layout = info->layout;
5218 rv = mddev->pers->check_reshape(mddev);
5219 if (rv)
5220 mddev->new_layout = mddev->layout;
5221 return rv;
5222 }
5223 }
5224 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5225 rv = update_size(mddev, (sector_t)info->size * 2);
5226
5227 if (mddev->raid_disks != info->raid_disks)
5228 rv = update_raid_disks(mddev, info->raid_disks);
5229
5230 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5231 if (mddev->pers->quiesce == NULL)
5232 return -EINVAL;
5233 if (mddev->recovery || mddev->sync_thread)
5234 return -EBUSY;
5235 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5236 /* add the bitmap */
5237 if (mddev->bitmap)
5238 return -EEXIST;
5239 if (mddev->default_bitmap_offset == 0)
5240 return -EINVAL;
5241 mddev->bitmap_offset = mddev->default_bitmap_offset;
5242 mddev->pers->quiesce(mddev, 1);
5243 rv = bitmap_create(mddev);
5244 if (rv)
5245 bitmap_destroy(mddev);
5246 mddev->pers->quiesce(mddev, 0);
5247 } else {
5248 /* remove the bitmap */
5249 if (!mddev->bitmap)
5250 return -ENOENT;
5251 if (mddev->bitmap->file)
5252 return -EINVAL;
5253 mddev->pers->quiesce(mddev, 1);
5254 bitmap_destroy(mddev);
5255 mddev->pers->quiesce(mddev, 0);
5256 mddev->bitmap_offset = 0;
5257 }
5258 }
5259 md_update_sb(mddev, 1);
5260 return rv;
5261 }
5262
5263 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5264 {
5265 mdk_rdev_t *rdev;
5266
5267 if (mddev->pers == NULL)
5268 return -ENODEV;
5269
5270 rdev = find_rdev(mddev, dev);
5271 if (!rdev)
5272 return -ENODEV;
5273
5274 md_error(mddev, rdev);
5275 return 0;
5276 }
5277
5278 /*
5279 * We have a problem here : there is no easy way to give a CHS
5280 * virtual geometry. We currently pretend that we have a 2 heads
5281 * 4 sectors (with a BIG number of cylinders...). This drives
5282 * dosfs just mad... ;-)
5283 */
5284 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5285 {
5286 mddev_t *mddev = bdev->bd_disk->private_data;
5287
5288 geo->heads = 2;
5289 geo->sectors = 4;
5290 geo->cylinders = get_capacity(mddev->gendisk) / 8;
5291 return 0;
5292 }
5293
5294 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5295 unsigned int cmd, unsigned long arg)
5296 {
5297 int err = 0;
5298 void __user *argp = (void __user *)arg;
5299 mddev_t *mddev = NULL;
5300
5301 if (!capable(CAP_SYS_ADMIN))
5302 return -EACCES;
5303
5304 /*
5305 * Commands dealing with the RAID driver but not any
5306 * particular array:
5307 */
5308 switch (cmd)
5309 {
5310 case RAID_VERSION:
5311 err = get_version(argp);
5312 goto done;
5313
5314 case PRINT_RAID_DEBUG:
5315 err = 0;
5316 md_print_devices();
5317 goto done;
5318
5319 #ifndef MODULE
5320 case RAID_AUTORUN:
5321 err = 0;
5322 autostart_arrays(arg);
5323 goto done;
5324 #endif
5325 default:;
5326 }
5327
5328 /*
5329 * Commands creating/starting a new array:
5330 */
5331
5332 mddev = bdev->bd_disk->private_data;
5333
5334 if (!mddev) {
5335 BUG();
5336 goto abort;
5337 }
5338
5339 err = mddev_lock(mddev);
5340 if (err) {
5341 printk(KERN_INFO
5342 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5343 err, cmd);
5344 goto abort;
5345 }
5346
5347 switch (cmd)
5348 {
5349 case SET_ARRAY_INFO:
5350 {
5351 mdu_array_info_t info;
5352 if (!arg)
5353 memset(&info, 0, sizeof(info));
5354 else if (copy_from_user(&info, argp, sizeof(info))) {
5355 err = -EFAULT;
5356 goto abort_unlock;
5357 }
5358 if (mddev->pers) {
5359 err = update_array_info(mddev, &info);
5360 if (err) {
5361 printk(KERN_WARNING "md: couldn't update"
5362 " array info. %d\n", err);
5363 goto abort_unlock;
5364 }
5365 goto done_unlock;
5366 }
5367 if (!list_empty(&mddev->disks)) {
5368 printk(KERN_WARNING
5369 "md: array %s already has disks!\n",
5370 mdname(mddev));
5371 err = -EBUSY;
5372 goto abort_unlock;
5373 }
5374 if (mddev->raid_disks) {
5375 printk(KERN_WARNING
5376 "md: array %s already initialised!\n",
5377 mdname(mddev));
5378 err = -EBUSY;
5379 goto abort_unlock;
5380 }
5381 err = set_array_info(mddev, &info);
5382 if (err) {
5383 printk(KERN_WARNING "md: couldn't set"
5384 " array info. %d\n", err);
5385 goto abort_unlock;
5386 }
5387 }
5388 goto done_unlock;
5389
5390 default:;
5391 }
5392
5393 /*
5394 * Commands querying/configuring an existing array:
5395 */
5396 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5397 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5398 if ((!mddev->raid_disks && !mddev->external)
5399 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5400 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5401 && cmd != GET_BITMAP_FILE) {
5402 err = -ENODEV;
5403 goto abort_unlock;
5404 }
5405
5406 /*
5407 * Commands even a read-only array can execute:
5408 */
5409 switch (cmd)
5410 {
5411 case GET_ARRAY_INFO:
5412 err = get_array_info(mddev, argp);
5413 goto done_unlock;
5414
5415 case GET_BITMAP_FILE:
5416 err = get_bitmap_file(mddev, argp);
5417 goto done_unlock;
5418
5419 case GET_DISK_INFO:
5420 err = get_disk_info(mddev, argp);
5421 goto done_unlock;
5422
5423 case RESTART_ARRAY_RW:
5424 err = restart_array(mddev);
5425 goto done_unlock;
5426
5427 case STOP_ARRAY:
5428 err = do_md_stop(mddev, 0, 1);
5429 goto done_unlock;
5430
5431 case STOP_ARRAY_RO:
5432 err = do_md_stop(mddev, 1, 1);
5433 goto done_unlock;
5434
5435 }
5436
5437 /*
5438 * The remaining ioctls are changing the state of the
5439 * superblock, so we do not allow them on read-only arrays.
5440 * However non-MD ioctls (e.g. get-size) will still come through
5441 * here and hit the 'default' below, so only disallow
5442 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5443 */
5444 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5445 if (mddev->ro == 2) {
5446 mddev->ro = 0;
5447 sysfs_notify_dirent(mddev->sysfs_state);
5448 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5449 md_wakeup_thread(mddev->thread);
5450 } else {
5451 err = -EROFS;
5452 goto abort_unlock;
5453 }
5454 }
5455
5456 switch (cmd)
5457 {
5458 case ADD_NEW_DISK:
5459 {
5460 mdu_disk_info_t info;
5461 if (copy_from_user(&info, argp, sizeof(info)))
5462 err = -EFAULT;
5463 else
5464 err = add_new_disk(mddev, &info);
5465 goto done_unlock;
5466 }
5467
5468 case HOT_REMOVE_DISK:
5469 err = hot_remove_disk(mddev, new_decode_dev(arg));
5470 goto done_unlock;
5471
5472 case HOT_ADD_DISK:
5473 err = hot_add_disk(mddev, new_decode_dev(arg));
5474 goto done_unlock;
5475
5476 case SET_DISK_FAULTY:
5477 err = set_disk_faulty(mddev, new_decode_dev(arg));
5478 goto done_unlock;
5479
5480 case RUN_ARRAY:
5481 err = do_md_run(mddev);
5482 goto done_unlock;
5483
5484 case SET_BITMAP_FILE:
5485 err = set_bitmap_file(mddev, (int)arg);
5486 goto done_unlock;
5487
5488 default:
5489 err = -EINVAL;
5490 goto abort_unlock;
5491 }
5492
5493 done_unlock:
5494 abort_unlock:
5495 if (mddev->hold_active == UNTIL_IOCTL &&
5496 err != -EINVAL)
5497 mddev->hold_active = 0;
5498 mddev_unlock(mddev);
5499
5500 return err;
5501 done:
5502 if (err)
5503 MD_BUG();
5504 abort:
5505 return err;
5506 }
5507
5508 static int md_open(struct block_device *bdev, fmode_t mode)
5509 {
5510 /*
5511 * Succeed if we can lock the mddev, which confirms that
5512 * it isn't being stopped right now.
5513 */
5514 mddev_t *mddev = mddev_find(bdev->bd_dev);
5515 int err;
5516
5517 if (mddev->gendisk != bdev->bd_disk) {
5518 /* we are racing with mddev_put which is discarding this
5519 * bd_disk.
5520 */
5521 mddev_put(mddev);
5522 /* Wait until bdev->bd_disk is definitely gone */
5523 flush_scheduled_work();
5524 /* Then retry the open from the top */
5525 return -ERESTARTSYS;
5526 }
5527 BUG_ON(mddev != bdev->bd_disk->private_data);
5528
5529 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
5530 goto out;
5531
5532 err = 0;
5533 atomic_inc(&mddev->openers);
5534 mutex_unlock(&mddev->open_mutex);
5535
5536 check_disk_change(bdev);
5537 out:
5538 return err;
5539 }
5540
5541 static int md_release(struct gendisk *disk, fmode_t mode)
5542 {
5543 mddev_t *mddev = disk->private_data;
5544
5545 BUG_ON(!mddev);
5546 atomic_dec(&mddev->openers);
5547 mddev_put(mddev);
5548
5549 return 0;
5550 }
5551
5552 static int md_media_changed(struct gendisk *disk)
5553 {
5554 mddev_t *mddev = disk->private_data;
5555
5556 return mddev->changed;
5557 }
5558
5559 static int md_revalidate(struct gendisk *disk)
5560 {
5561 mddev_t *mddev = disk->private_data;
5562
5563 mddev->changed = 0;
5564 return 0;
5565 }
5566 static struct block_device_operations md_fops =
5567 {
5568 .owner = THIS_MODULE,
5569 .open = md_open,
5570 .release = md_release,
5571 .ioctl = md_ioctl,
5572 .getgeo = md_getgeo,
5573 .media_changed = md_media_changed,
5574 .revalidate_disk= md_revalidate,
5575 };
5576
5577 static int md_thread(void * arg)
5578 {
5579 mdk_thread_t *thread = arg;
5580
5581 /*
5582 * md_thread is a 'system-thread', it's priority should be very
5583 * high. We avoid resource deadlocks individually in each
5584 * raid personality. (RAID5 does preallocation) We also use RR and
5585 * the very same RT priority as kswapd, thus we will never get
5586 * into a priority inversion deadlock.
5587 *
5588 * we definitely have to have equal or higher priority than
5589 * bdflush, otherwise bdflush will deadlock if there are too
5590 * many dirty RAID5 blocks.
5591 */
5592
5593 allow_signal(SIGKILL);
5594 while (!kthread_should_stop()) {
5595
5596 /* We need to wait INTERRUPTIBLE so that
5597 * we don't add to the load-average.
5598 * That means we need to be sure no signals are
5599 * pending
5600 */
5601 if (signal_pending(current))
5602 flush_signals(current);
5603
5604 wait_event_interruptible_timeout
5605 (thread->wqueue,
5606 test_bit(THREAD_WAKEUP, &thread->flags)
5607 || kthread_should_stop(),
5608 thread->timeout);
5609
5610 clear_bit(THREAD_WAKEUP, &thread->flags);
5611
5612 thread->run(thread->mddev);
5613 }
5614
5615 return 0;
5616 }
5617
5618 void md_wakeup_thread(mdk_thread_t *thread)
5619 {
5620 if (thread) {
5621 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5622 set_bit(THREAD_WAKEUP, &thread->flags);
5623 wake_up(&thread->wqueue);
5624 }
5625 }
5626
5627 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5628 const char *name)
5629 {
5630 mdk_thread_t *thread;
5631
5632 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5633 if (!thread)
5634 return NULL;
5635
5636 init_waitqueue_head(&thread->wqueue);
5637
5638 thread->run = run;
5639 thread->mddev = mddev;
5640 thread->timeout = MAX_SCHEDULE_TIMEOUT;
5641 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
5642 if (IS_ERR(thread->tsk)) {
5643 kfree(thread);
5644 return NULL;
5645 }
5646 return thread;
5647 }
5648
5649 void md_unregister_thread(mdk_thread_t *thread)
5650 {
5651 if (!thread)
5652 return;
5653 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5654
5655 kthread_stop(thread->tsk);
5656 kfree(thread);
5657 }
5658
5659 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5660 {
5661 if (!mddev) {
5662 MD_BUG();
5663 return;
5664 }
5665
5666 if (!rdev || test_bit(Faulty, &rdev->flags))
5667 return;
5668
5669 if (mddev->external)
5670 set_bit(Blocked, &rdev->flags);
5671 /*
5672 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5673 mdname(mddev),
5674 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5675 __builtin_return_address(0),__builtin_return_address(1),
5676 __builtin_return_address(2),__builtin_return_address(3));
5677 */
5678 if (!mddev->pers)
5679 return;
5680 if (!mddev->pers->error_handler)
5681 return;
5682 mddev->pers->error_handler(mddev,rdev);
5683 if (mddev->degraded)
5684 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5685 set_bit(StateChanged, &rdev->flags);
5686 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5687 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5688 md_wakeup_thread(mddev->thread);
5689 md_new_event_inintr(mddev);
5690 }
5691
5692 /* seq_file implementation /proc/mdstat */
5693
5694 static void status_unused(struct seq_file *seq)
5695 {
5696 int i = 0;
5697 mdk_rdev_t *rdev;
5698
5699 seq_printf(seq, "unused devices: ");
5700
5701 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
5702 char b[BDEVNAME_SIZE];
5703 i++;
5704 seq_printf(seq, "%s ",
5705 bdevname(rdev->bdev,b));
5706 }
5707 if (!i)
5708 seq_printf(seq, "<none>");
5709
5710 seq_printf(seq, "\n");
5711 }
5712
5713
5714 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5715 {
5716 sector_t max_sectors, resync, res;
5717 unsigned long dt, db;
5718 sector_t rt;
5719 int scale;
5720 unsigned int per_milli;
5721
5722 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
5723
5724 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5725 max_sectors = mddev->resync_max_sectors;
5726 else
5727 max_sectors = mddev->dev_sectors;
5728
5729 /*
5730 * Should not happen.
5731 */
5732 if (!max_sectors) {
5733 MD_BUG();
5734 return;
5735 }
5736 /* Pick 'scale' such that (resync>>scale)*1000 will fit
5737 * in a sector_t, and (max_sectors>>scale) will fit in a
5738 * u32, as those are the requirements for sector_div.
5739 * Thus 'scale' must be at least 10
5740 */
5741 scale = 10;
5742 if (sizeof(sector_t) > sizeof(unsigned long)) {
5743 while ( max_sectors/2 > (1ULL<<(scale+32)))
5744 scale++;
5745 }
5746 res = (resync>>scale)*1000;
5747 sector_div(res, (u32)((max_sectors>>scale)+1));
5748
5749 per_milli = res;
5750 {
5751 int i, x = per_milli/50, y = 20-x;
5752 seq_printf(seq, "[");
5753 for (i = 0; i < x; i++)
5754 seq_printf(seq, "=");
5755 seq_printf(seq, ">");
5756 for (i = 0; i < y; i++)
5757 seq_printf(seq, ".");
5758 seq_printf(seq, "] ");
5759 }
5760 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5761 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5762 "reshape" :
5763 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5764 "check" :
5765 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5766 "resync" : "recovery"))),
5767 per_milli/10, per_milli % 10,
5768 (unsigned long long) resync/2,
5769 (unsigned long long) max_sectors/2);
5770
5771 /*
5772 * dt: time from mark until now
5773 * db: blocks written from mark until now
5774 * rt: remaining time
5775 *
5776 * rt is a sector_t, so could be 32bit or 64bit.
5777 * So we divide before multiply in case it is 32bit and close
5778 * to the limit.
5779 * We scale the divisor (db) by 32 to avoid loosing precision
5780 * near the end of resync when the number of remaining sectors
5781 * is close to 'db'.
5782 * We then divide rt by 32 after multiplying by db to compensate.
5783 * The '+1' avoids division by zero if db is very small.
5784 */
5785 dt = ((jiffies - mddev->resync_mark) / HZ);
5786 if (!dt) dt++;
5787 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5788 - mddev->resync_mark_cnt;
5789
5790 rt = max_sectors - resync; /* number of remaining sectors */
5791 sector_div(rt, db/32+1);
5792 rt *= dt;
5793 rt >>= 5;
5794
5795 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
5796 ((unsigned long)rt % 60)/6);
5797
5798 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5799 }
5800
5801 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5802 {
5803 struct list_head *tmp;
5804 loff_t l = *pos;
5805 mddev_t *mddev;
5806
5807 if (l >= 0x10000)
5808 return NULL;
5809 if (!l--)
5810 /* header */
5811 return (void*)1;
5812
5813 spin_lock(&all_mddevs_lock);
5814 list_for_each(tmp,&all_mddevs)
5815 if (!l--) {
5816 mddev = list_entry(tmp, mddev_t, all_mddevs);
5817 mddev_get(mddev);
5818 spin_unlock(&all_mddevs_lock);
5819 return mddev;
5820 }
5821 spin_unlock(&all_mddevs_lock);
5822 if (!l--)
5823 return (void*)2;/* tail */
5824 return NULL;
5825 }
5826
5827 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5828 {
5829 struct list_head *tmp;
5830 mddev_t *next_mddev, *mddev = v;
5831
5832 ++*pos;
5833 if (v == (void*)2)
5834 return NULL;
5835
5836 spin_lock(&all_mddevs_lock);
5837 if (v == (void*)1)
5838 tmp = all_mddevs.next;
5839 else
5840 tmp = mddev->all_mddevs.next;
5841 if (tmp != &all_mddevs)
5842 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5843 else {
5844 next_mddev = (void*)2;
5845 *pos = 0x10000;
5846 }
5847 spin_unlock(&all_mddevs_lock);
5848
5849 if (v != (void*)1)
5850 mddev_put(mddev);
5851 return next_mddev;
5852
5853 }
5854
5855 static void md_seq_stop(struct seq_file *seq, void *v)
5856 {
5857 mddev_t *mddev = v;
5858
5859 if (mddev && v != (void*)1 && v != (void*)2)
5860 mddev_put(mddev);
5861 }
5862
5863 struct mdstat_info {
5864 int event;
5865 };
5866
5867 static int md_seq_show(struct seq_file *seq, void *v)
5868 {
5869 mddev_t *mddev = v;
5870 sector_t sectors;
5871 mdk_rdev_t *rdev;
5872 struct mdstat_info *mi = seq->private;
5873 struct bitmap *bitmap;
5874
5875 if (v == (void*)1) {
5876 struct mdk_personality *pers;
5877 seq_printf(seq, "Personalities : ");
5878 spin_lock(&pers_lock);
5879 list_for_each_entry(pers, &pers_list, list)
5880 seq_printf(seq, "[%s] ", pers->name);
5881
5882 spin_unlock(&pers_lock);
5883 seq_printf(seq, "\n");
5884 mi->event = atomic_read(&md_event_count);
5885 return 0;
5886 }
5887 if (v == (void*)2) {
5888 status_unused(seq);
5889 return 0;
5890 }
5891
5892 if (mddev_lock(mddev) < 0)
5893 return -EINTR;
5894
5895 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5896 seq_printf(seq, "%s : %sactive", mdname(mddev),
5897 mddev->pers ? "" : "in");
5898 if (mddev->pers) {
5899 if (mddev->ro==1)
5900 seq_printf(seq, " (read-only)");
5901 if (mddev->ro==2)
5902 seq_printf(seq, " (auto-read-only)");
5903 seq_printf(seq, " %s", mddev->pers->name);
5904 }
5905
5906 sectors = 0;
5907 list_for_each_entry(rdev, &mddev->disks, same_set) {
5908 char b[BDEVNAME_SIZE];
5909 seq_printf(seq, " %s[%d]",
5910 bdevname(rdev->bdev,b), rdev->desc_nr);
5911 if (test_bit(WriteMostly, &rdev->flags))
5912 seq_printf(seq, "(W)");
5913 if (test_bit(Faulty, &rdev->flags)) {
5914 seq_printf(seq, "(F)");
5915 continue;
5916 } else if (rdev->raid_disk < 0)
5917 seq_printf(seq, "(S)"); /* spare */
5918 sectors += rdev->sectors;
5919 }
5920
5921 if (!list_empty(&mddev->disks)) {
5922 if (mddev->pers)
5923 seq_printf(seq, "\n %llu blocks",
5924 (unsigned long long)
5925 mddev->array_sectors / 2);
5926 else
5927 seq_printf(seq, "\n %llu blocks",
5928 (unsigned long long)sectors / 2);
5929 }
5930 if (mddev->persistent) {
5931 if (mddev->major_version != 0 ||
5932 mddev->minor_version != 90) {
5933 seq_printf(seq," super %d.%d",
5934 mddev->major_version,
5935 mddev->minor_version);
5936 }
5937 } else if (mddev->external)
5938 seq_printf(seq, " super external:%s",
5939 mddev->metadata_type);
5940 else
5941 seq_printf(seq, " super non-persistent");
5942
5943 if (mddev->pers) {
5944 mddev->pers->status(seq, mddev);
5945 seq_printf(seq, "\n ");
5946 if (mddev->pers->sync_request) {
5947 if (mddev->curr_resync > 2) {
5948 status_resync(seq, mddev);
5949 seq_printf(seq, "\n ");
5950 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5951 seq_printf(seq, "\tresync=DELAYED\n ");
5952 else if (mddev->recovery_cp < MaxSector)
5953 seq_printf(seq, "\tresync=PENDING\n ");
5954 }
5955 } else
5956 seq_printf(seq, "\n ");
5957
5958 if ((bitmap = mddev->bitmap)) {
5959 unsigned long chunk_kb;
5960 unsigned long flags;
5961 spin_lock_irqsave(&bitmap->lock, flags);
5962 chunk_kb = bitmap->chunksize >> 10;
5963 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5964 "%lu%s chunk",
5965 bitmap->pages - bitmap->missing_pages,
5966 bitmap->pages,
5967 (bitmap->pages - bitmap->missing_pages)
5968 << (PAGE_SHIFT - 10),
5969 chunk_kb ? chunk_kb : bitmap->chunksize,
5970 chunk_kb ? "KB" : "B");
5971 if (bitmap->file) {
5972 seq_printf(seq, ", file: ");
5973 seq_path(seq, &bitmap->file->f_path, " \t\n");
5974 }
5975
5976 seq_printf(seq, "\n");
5977 spin_unlock_irqrestore(&bitmap->lock, flags);
5978 }
5979
5980 seq_printf(seq, "\n");
5981 }
5982 mddev_unlock(mddev);
5983
5984 return 0;
5985 }
5986
5987 static const struct seq_operations md_seq_ops = {
5988 .start = md_seq_start,
5989 .next = md_seq_next,
5990 .stop = md_seq_stop,
5991 .show = md_seq_show,
5992 };
5993
5994 static int md_seq_open(struct inode *inode, struct file *file)
5995 {
5996 int error;
5997 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5998 if (mi == NULL)
5999 return -ENOMEM;
6000
6001 error = seq_open(file, &md_seq_ops);
6002 if (error)
6003 kfree(mi);
6004 else {
6005 struct seq_file *p = file->private_data;
6006 p->private = mi;
6007 mi->event = atomic_read(&md_event_count);
6008 }
6009 return error;
6010 }
6011
6012 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6013 {
6014 struct seq_file *m = filp->private_data;
6015 struct mdstat_info *mi = m->private;
6016 int mask;
6017
6018 poll_wait(filp, &md_event_waiters, wait);
6019
6020 /* always allow read */
6021 mask = POLLIN | POLLRDNORM;
6022
6023 if (mi->event != atomic_read(&md_event_count))
6024 mask |= POLLERR | POLLPRI;
6025 return mask;
6026 }
6027
6028 static const struct file_operations md_seq_fops = {
6029 .owner = THIS_MODULE,
6030 .open = md_seq_open,
6031 .read = seq_read,
6032 .llseek = seq_lseek,
6033 .release = seq_release_private,
6034 .poll = mdstat_poll,
6035 };
6036
6037 int register_md_personality(struct mdk_personality *p)
6038 {
6039 spin_lock(&pers_lock);
6040 list_add_tail(&p->list, &pers_list);
6041 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6042 spin_unlock(&pers_lock);
6043 return 0;
6044 }
6045
6046 int unregister_md_personality(struct mdk_personality *p)
6047 {
6048 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6049 spin_lock(&pers_lock);
6050 list_del_init(&p->list);
6051 spin_unlock(&pers_lock);
6052 return 0;
6053 }
6054
6055 static int is_mddev_idle(mddev_t *mddev, int init)
6056 {
6057 mdk_rdev_t * rdev;
6058 int idle;
6059 int curr_events;
6060
6061 idle = 1;
6062 rcu_read_lock();
6063 rdev_for_each_rcu(rdev, mddev) {
6064 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6065 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6066 (int)part_stat_read(&disk->part0, sectors[1]) -
6067 atomic_read(&disk->sync_io);
6068 /* sync IO will cause sync_io to increase before the disk_stats
6069 * as sync_io is counted when a request starts, and
6070 * disk_stats is counted when it completes.
6071 * So resync activity will cause curr_events to be smaller than
6072 * when there was no such activity.
6073 * non-sync IO will cause disk_stat to increase without
6074 * increasing sync_io so curr_events will (eventually)
6075 * be larger than it was before. Once it becomes
6076 * substantially larger, the test below will cause
6077 * the array to appear non-idle, and resync will slow
6078 * down.
6079 * If there is a lot of outstanding resync activity when
6080 * we set last_event to curr_events, then all that activity
6081 * completing might cause the array to appear non-idle
6082 * and resync will be slowed down even though there might
6083 * not have been non-resync activity. This will only
6084 * happen once though. 'last_events' will soon reflect
6085 * the state where there is little or no outstanding
6086 * resync requests, and further resync activity will
6087 * always make curr_events less than last_events.
6088 *
6089 */
6090 if (init || curr_events - rdev->last_events > 64) {
6091 rdev->last_events = curr_events;
6092 idle = 0;
6093 }
6094 }
6095 rcu_read_unlock();
6096 return idle;
6097 }
6098
6099 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6100 {
6101 /* another "blocks" (512byte) blocks have been synced */
6102 atomic_sub(blocks, &mddev->recovery_active);
6103 wake_up(&mddev->recovery_wait);
6104 if (!ok) {
6105 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6106 md_wakeup_thread(mddev->thread);
6107 // stop recovery, signal do_sync ....
6108 }
6109 }
6110
6111
6112 /* md_write_start(mddev, bi)
6113 * If we need to update some array metadata (e.g. 'active' flag
6114 * in superblock) before writing, schedule a superblock update
6115 * and wait for it to complete.
6116 */
6117 void md_write_start(mddev_t *mddev, struct bio *bi)
6118 {
6119 int did_change = 0;
6120 if (bio_data_dir(bi) != WRITE)
6121 return;
6122
6123 BUG_ON(mddev->ro == 1);
6124 if (mddev->ro == 2) {
6125 /* need to switch to read/write */
6126 mddev->ro = 0;
6127 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6128 md_wakeup_thread(mddev->thread);
6129 md_wakeup_thread(mddev->sync_thread);
6130 did_change = 1;
6131 }
6132 atomic_inc(&mddev->writes_pending);
6133 if (mddev->safemode == 1)
6134 mddev->safemode = 0;
6135 if (mddev->in_sync) {
6136 spin_lock_irq(&mddev->write_lock);
6137 if (mddev->in_sync) {
6138 mddev->in_sync = 0;
6139 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6140 md_wakeup_thread(mddev->thread);
6141 did_change = 1;
6142 }
6143 spin_unlock_irq(&mddev->write_lock);
6144 }
6145 if (did_change)
6146 sysfs_notify_dirent(mddev->sysfs_state);
6147 wait_event(mddev->sb_wait,
6148 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
6149 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6150 }
6151
6152 void md_write_end(mddev_t *mddev)
6153 {
6154 if (atomic_dec_and_test(&mddev->writes_pending)) {
6155 if (mddev->safemode == 2)
6156 md_wakeup_thread(mddev->thread);
6157 else if (mddev->safemode_delay)
6158 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6159 }
6160 }
6161
6162 /* md_allow_write(mddev)
6163 * Calling this ensures that the array is marked 'active' so that writes
6164 * may proceed without blocking. It is important to call this before
6165 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6166 * Must be called with mddev_lock held.
6167 *
6168 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6169 * is dropped, so return -EAGAIN after notifying userspace.
6170 */
6171 int md_allow_write(mddev_t *mddev)
6172 {
6173 if (!mddev->pers)
6174 return 0;
6175 if (mddev->ro)
6176 return 0;
6177 if (!mddev->pers->sync_request)
6178 return 0;
6179
6180 spin_lock_irq(&mddev->write_lock);
6181 if (mddev->in_sync) {
6182 mddev->in_sync = 0;
6183 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6184 if (mddev->safemode_delay &&
6185 mddev->safemode == 0)
6186 mddev->safemode = 1;
6187 spin_unlock_irq(&mddev->write_lock);
6188 md_update_sb(mddev, 0);
6189 sysfs_notify_dirent(mddev->sysfs_state);
6190 } else
6191 spin_unlock_irq(&mddev->write_lock);
6192
6193 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
6194 return -EAGAIN;
6195 else
6196 return 0;
6197 }
6198 EXPORT_SYMBOL_GPL(md_allow_write);
6199
6200 #define SYNC_MARKS 10
6201 #define SYNC_MARK_STEP (3*HZ)
6202 void md_do_sync(mddev_t *mddev)
6203 {
6204 mddev_t *mddev2;
6205 unsigned int currspeed = 0,
6206 window;
6207 sector_t max_sectors,j, io_sectors;
6208 unsigned long mark[SYNC_MARKS];
6209 sector_t mark_cnt[SYNC_MARKS];
6210 int last_mark,m;
6211 struct list_head *tmp;
6212 sector_t last_check;
6213 int skipped = 0;
6214 mdk_rdev_t *rdev;
6215 char *desc;
6216
6217 /* just incase thread restarts... */
6218 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6219 return;
6220 if (mddev->ro) /* never try to sync a read-only array */
6221 return;
6222
6223 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6224 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6225 desc = "data-check";
6226 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6227 desc = "requested-resync";
6228 else
6229 desc = "resync";
6230 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6231 desc = "reshape";
6232 else
6233 desc = "recovery";
6234
6235 /* we overload curr_resync somewhat here.
6236 * 0 == not engaged in resync at all
6237 * 2 == checking that there is no conflict with another sync
6238 * 1 == like 2, but have yielded to allow conflicting resync to
6239 * commense
6240 * other == active in resync - this many blocks
6241 *
6242 * Before starting a resync we must have set curr_resync to
6243 * 2, and then checked that every "conflicting" array has curr_resync
6244 * less than ours. When we find one that is the same or higher
6245 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6246 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6247 * This will mean we have to start checking from the beginning again.
6248 *
6249 */
6250
6251 do {
6252 mddev->curr_resync = 2;
6253
6254 try_again:
6255 if (kthread_should_stop()) {
6256 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6257 goto skip;
6258 }
6259 for_each_mddev(mddev2, tmp) {
6260 if (mddev2 == mddev)
6261 continue;
6262 if (!mddev->parallel_resync
6263 && mddev2->curr_resync
6264 && match_mddev_units(mddev, mddev2)) {
6265 DEFINE_WAIT(wq);
6266 if (mddev < mddev2 && mddev->curr_resync == 2) {
6267 /* arbitrarily yield */
6268 mddev->curr_resync = 1;
6269 wake_up(&resync_wait);
6270 }
6271 if (mddev > mddev2 && mddev->curr_resync == 1)
6272 /* no need to wait here, we can wait the next
6273 * time 'round when curr_resync == 2
6274 */
6275 continue;
6276 /* We need to wait 'interruptible' so as not to
6277 * contribute to the load average, and not to
6278 * be caught by 'softlockup'
6279 */
6280 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6281 if (!kthread_should_stop() &&
6282 mddev2->curr_resync >= mddev->curr_resync) {
6283 printk(KERN_INFO "md: delaying %s of %s"
6284 " until %s has finished (they"
6285 " share one or more physical units)\n",
6286 desc, mdname(mddev), mdname(mddev2));
6287 mddev_put(mddev2);
6288 if (signal_pending(current))
6289 flush_signals(current);
6290 schedule();
6291 finish_wait(&resync_wait, &wq);
6292 goto try_again;
6293 }
6294 finish_wait(&resync_wait, &wq);
6295 }
6296 }
6297 } while (mddev->curr_resync < 2);
6298
6299 j = 0;
6300 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6301 /* resync follows the size requested by the personality,
6302 * which defaults to physical size, but can be virtual size
6303 */
6304 max_sectors = mddev->resync_max_sectors;
6305 mddev->resync_mismatches = 0;
6306 /* we don't use the checkpoint if there's a bitmap */
6307 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6308 j = mddev->resync_min;
6309 else if (!mddev->bitmap)
6310 j = mddev->recovery_cp;
6311
6312 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6313 max_sectors = mddev->dev_sectors;
6314 else {
6315 /* recovery follows the physical size of devices */
6316 max_sectors = mddev->dev_sectors;
6317 j = MaxSector;
6318 list_for_each_entry(rdev, &mddev->disks, same_set)
6319 if (rdev->raid_disk >= 0 &&
6320 !test_bit(Faulty, &rdev->flags) &&
6321 !test_bit(In_sync, &rdev->flags) &&
6322 rdev->recovery_offset < j)
6323 j = rdev->recovery_offset;
6324 }
6325
6326 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6327 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6328 " %d KB/sec/disk.\n", speed_min(mddev));
6329 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6330 "(but not more than %d KB/sec) for %s.\n",
6331 speed_max(mddev), desc);
6332
6333 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6334
6335 io_sectors = 0;
6336 for (m = 0; m < SYNC_MARKS; m++) {
6337 mark[m] = jiffies;
6338 mark_cnt[m] = io_sectors;
6339 }
6340 last_mark = 0;
6341 mddev->resync_mark = mark[last_mark];
6342 mddev->resync_mark_cnt = mark_cnt[last_mark];
6343
6344 /*
6345 * Tune reconstruction:
6346 */
6347 window = 32*(PAGE_SIZE/512);
6348 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6349 window/2,(unsigned long long) max_sectors/2);
6350
6351 atomic_set(&mddev->recovery_active, 0);
6352 last_check = 0;
6353
6354 if (j>2) {
6355 printk(KERN_INFO
6356 "md: resuming %s of %s from checkpoint.\n",
6357 desc, mdname(mddev));
6358 mddev->curr_resync = j;
6359 }
6360
6361 while (j < max_sectors) {
6362 sector_t sectors;
6363
6364 skipped = 0;
6365
6366 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6367 ((mddev->curr_resync > mddev->curr_resync_completed &&
6368 (mddev->curr_resync - mddev->curr_resync_completed)
6369 > (max_sectors >> 4)) ||
6370 (j - mddev->curr_resync_completed)*2
6371 >= mddev->resync_max - mddev->curr_resync_completed
6372 )) {
6373 /* time to update curr_resync_completed */
6374 blk_unplug(mddev->queue);
6375 wait_event(mddev->recovery_wait,
6376 atomic_read(&mddev->recovery_active) == 0);
6377 mddev->curr_resync_completed =
6378 mddev->curr_resync;
6379 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6380 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6381 }
6382
6383 while (j >= mddev->resync_max && !kthread_should_stop()) {
6384 /* As this condition is controlled by user-space,
6385 * we can block indefinitely, so use '_interruptible'
6386 * to avoid triggering warnings.
6387 */
6388 flush_signals(current); /* just in case */
6389 wait_event_interruptible(mddev->recovery_wait,
6390 mddev->resync_max > j
6391 || kthread_should_stop());
6392 }
6393
6394 if (kthread_should_stop())
6395 goto interrupted;
6396
6397 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6398 currspeed < speed_min(mddev));
6399 if (sectors == 0) {
6400 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6401 goto out;
6402 }
6403
6404 if (!skipped) { /* actual IO requested */
6405 io_sectors += sectors;
6406 atomic_add(sectors, &mddev->recovery_active);
6407 }
6408
6409 j += sectors;
6410 if (j>1) mddev->curr_resync = j;
6411 mddev->curr_mark_cnt = io_sectors;
6412 if (last_check == 0)
6413 /* this is the earliers that rebuilt will be
6414 * visible in /proc/mdstat
6415 */
6416 md_new_event(mddev);
6417
6418 if (last_check + window > io_sectors || j == max_sectors)
6419 continue;
6420
6421 last_check = io_sectors;
6422
6423 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6424 break;
6425
6426 repeat:
6427 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6428 /* step marks */
6429 int next = (last_mark+1) % SYNC_MARKS;
6430
6431 mddev->resync_mark = mark[next];
6432 mddev->resync_mark_cnt = mark_cnt[next];
6433 mark[next] = jiffies;
6434 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6435 last_mark = next;
6436 }
6437
6438
6439 if (kthread_should_stop())
6440 goto interrupted;
6441
6442
6443 /*
6444 * this loop exits only if either when we are slower than
6445 * the 'hard' speed limit, or the system was IO-idle for
6446 * a jiffy.
6447 * the system might be non-idle CPU-wise, but we only care
6448 * about not overloading the IO subsystem. (things like an
6449 * e2fsck being done on the RAID array should execute fast)
6450 */
6451 blk_unplug(mddev->queue);
6452 cond_resched();
6453
6454 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6455 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6456
6457 if (currspeed > speed_min(mddev)) {
6458 if ((currspeed > speed_max(mddev)) ||
6459 !is_mddev_idle(mddev, 0)) {
6460 msleep(500);
6461 goto repeat;
6462 }
6463 }
6464 }
6465 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6466 /*
6467 * this also signals 'finished resyncing' to md_stop
6468 */
6469 out:
6470 blk_unplug(mddev->queue);
6471
6472 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6473
6474 /* tell personality that we are finished */
6475 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6476
6477 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6478 mddev->curr_resync > 2) {
6479 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6480 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6481 if (mddev->curr_resync >= mddev->recovery_cp) {
6482 printk(KERN_INFO
6483 "md: checkpointing %s of %s.\n",
6484 desc, mdname(mddev));
6485 mddev->recovery_cp = mddev->curr_resync;
6486 }
6487 } else
6488 mddev->recovery_cp = MaxSector;
6489 } else {
6490 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6491 mddev->curr_resync = MaxSector;
6492 list_for_each_entry(rdev, &mddev->disks, same_set)
6493 if (rdev->raid_disk >= 0 &&
6494 !test_bit(Faulty, &rdev->flags) &&
6495 !test_bit(In_sync, &rdev->flags) &&
6496 rdev->recovery_offset < mddev->curr_resync)
6497 rdev->recovery_offset = mddev->curr_resync;
6498 }
6499 }
6500 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6501
6502 skip:
6503 mddev->curr_resync = 0;
6504 mddev->curr_resync_completed = 0;
6505 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6506 /* We completed so max setting can be forgotten. */
6507 mddev->resync_max = MaxSector;
6508 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6509 wake_up(&resync_wait);
6510 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6511 md_wakeup_thread(mddev->thread);
6512 return;
6513
6514 interrupted:
6515 /*
6516 * got a signal, exit.
6517 */
6518 printk(KERN_INFO
6519 "md: md_do_sync() got signal ... exiting\n");
6520 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6521 goto out;
6522
6523 }
6524 EXPORT_SYMBOL_GPL(md_do_sync);
6525
6526
6527 static int remove_and_add_spares(mddev_t *mddev)
6528 {
6529 mdk_rdev_t *rdev;
6530 int spares = 0;
6531
6532 mddev->curr_resync_completed = 0;
6533
6534 list_for_each_entry(rdev, &mddev->disks, same_set)
6535 if (rdev->raid_disk >= 0 &&
6536 !test_bit(Blocked, &rdev->flags) &&
6537 (test_bit(Faulty, &rdev->flags) ||
6538 ! test_bit(In_sync, &rdev->flags)) &&
6539 atomic_read(&rdev->nr_pending)==0) {
6540 if (mddev->pers->hot_remove_disk(
6541 mddev, rdev->raid_disk)==0) {
6542 char nm[20];
6543 sprintf(nm,"rd%d", rdev->raid_disk);
6544 sysfs_remove_link(&mddev->kobj, nm);
6545 rdev->raid_disk = -1;
6546 }
6547 }
6548
6549 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
6550 list_for_each_entry(rdev, &mddev->disks, same_set) {
6551 if (rdev->raid_disk >= 0 &&
6552 !test_bit(In_sync, &rdev->flags) &&
6553 !test_bit(Blocked, &rdev->flags))
6554 spares++;
6555 if (rdev->raid_disk < 0
6556 && !test_bit(Faulty, &rdev->flags)) {
6557 rdev->recovery_offset = 0;
6558 if (mddev->pers->
6559 hot_add_disk(mddev, rdev) == 0) {
6560 char nm[20];
6561 sprintf(nm, "rd%d", rdev->raid_disk);
6562 if (sysfs_create_link(&mddev->kobj,
6563 &rdev->kobj, nm))
6564 printk(KERN_WARNING
6565 "md: cannot register "
6566 "%s for %s\n",
6567 nm, mdname(mddev));
6568 spares++;
6569 md_new_event(mddev);
6570 } else
6571 break;
6572 }
6573 }
6574 }
6575 return spares;
6576 }
6577 /*
6578 * This routine is regularly called by all per-raid-array threads to
6579 * deal with generic issues like resync and super-block update.
6580 * Raid personalities that don't have a thread (linear/raid0) do not
6581 * need this as they never do any recovery or update the superblock.
6582 *
6583 * It does not do any resync itself, but rather "forks" off other threads
6584 * to do that as needed.
6585 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6586 * "->recovery" and create a thread at ->sync_thread.
6587 * When the thread finishes it sets MD_RECOVERY_DONE
6588 * and wakeups up this thread which will reap the thread and finish up.
6589 * This thread also removes any faulty devices (with nr_pending == 0).
6590 *
6591 * The overall approach is:
6592 * 1/ if the superblock needs updating, update it.
6593 * 2/ If a recovery thread is running, don't do anything else.
6594 * 3/ If recovery has finished, clean up, possibly marking spares active.
6595 * 4/ If there are any faulty devices, remove them.
6596 * 5/ If array is degraded, try to add spares devices
6597 * 6/ If array has spares or is not in-sync, start a resync thread.
6598 */
6599 void md_check_recovery(mddev_t *mddev)
6600 {
6601 mdk_rdev_t *rdev;
6602
6603
6604 if (mddev->bitmap)
6605 bitmap_daemon_work(mddev);
6606
6607 if (mddev->ro)
6608 return;
6609
6610 if (signal_pending(current)) {
6611 if (mddev->pers->sync_request && !mddev->external) {
6612 printk(KERN_INFO "md: %s in immediate safe mode\n",
6613 mdname(mddev));
6614 mddev->safemode = 2;
6615 }
6616 flush_signals(current);
6617 }
6618
6619 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
6620 return;
6621 if ( ! (
6622 (mddev->flags && !mddev->external) ||
6623 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
6624 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
6625 (mddev->external == 0 && mddev->safemode == 1) ||
6626 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6627 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6628 ))
6629 return;
6630
6631 if (mddev_trylock(mddev)) {
6632 int spares = 0;
6633
6634 if (mddev->ro) {
6635 /* Only thing we do on a ro array is remove
6636 * failed devices.
6637 */
6638 remove_and_add_spares(mddev);
6639 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6640 goto unlock;
6641 }
6642
6643 if (!mddev->external) {
6644 int did_change = 0;
6645 spin_lock_irq(&mddev->write_lock);
6646 if (mddev->safemode &&
6647 !atomic_read(&mddev->writes_pending) &&
6648 !mddev->in_sync &&
6649 mddev->recovery_cp == MaxSector) {
6650 mddev->in_sync = 1;
6651 did_change = 1;
6652 if (mddev->persistent)
6653 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6654 }
6655 if (mddev->safemode == 1)
6656 mddev->safemode = 0;
6657 spin_unlock_irq(&mddev->write_lock);
6658 if (did_change)
6659 sysfs_notify_dirent(mddev->sysfs_state);
6660 }
6661
6662 if (mddev->flags)
6663 md_update_sb(mddev, 0);
6664
6665 list_for_each_entry(rdev, &mddev->disks, same_set)
6666 if (test_and_clear_bit(StateChanged, &rdev->flags))
6667 sysfs_notify_dirent(rdev->sysfs_state);
6668
6669
6670 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6671 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6672 /* resync/recovery still happening */
6673 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6674 goto unlock;
6675 }
6676 if (mddev->sync_thread) {
6677 /* resync has finished, collect result */
6678 md_unregister_thread(mddev->sync_thread);
6679 mddev->sync_thread = NULL;
6680 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
6681 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
6682 /* success...*/
6683 /* activate any spares */
6684 if (mddev->pers->spare_active(mddev))
6685 sysfs_notify(&mddev->kobj, NULL,
6686 "degraded");
6687 }
6688 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6689 mddev->pers->finish_reshape)
6690 mddev->pers->finish_reshape(mddev);
6691 md_update_sb(mddev, 1);
6692
6693 /* if array is no-longer degraded, then any saved_raid_disk
6694 * information must be scrapped
6695 */
6696 if (!mddev->degraded)
6697 list_for_each_entry(rdev, &mddev->disks, same_set)
6698 rdev->saved_raid_disk = -1;
6699
6700 mddev->recovery = 0;
6701 /* flag recovery needed just to double check */
6702 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6703 sysfs_notify_dirent(mddev->sysfs_action);
6704 md_new_event(mddev);
6705 goto unlock;
6706 }
6707 /* Set RUNNING before clearing NEEDED to avoid
6708 * any transients in the value of "sync_action".
6709 */
6710 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6711 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6712 /* Clear some bits that don't mean anything, but
6713 * might be left set
6714 */
6715 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
6716 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
6717
6718 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6719 goto unlock;
6720 /* no recovery is running.
6721 * remove any failed drives, then
6722 * add spares if possible.
6723 * Spare are also removed and re-added, to allow
6724 * the personality to fail the re-add.
6725 */
6726
6727 if (mddev->reshape_position != MaxSector) {
6728 if (mddev->pers->check_reshape == NULL ||
6729 mddev->pers->check_reshape(mddev) != 0)
6730 /* Cannot proceed */
6731 goto unlock;
6732 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6733 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6734 } else if ((spares = remove_and_add_spares(mddev))) {
6735 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6736 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6737 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
6738 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6739 } else if (mddev->recovery_cp < MaxSector) {
6740 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6741 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6742 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6743 /* nothing to be done ... */
6744 goto unlock;
6745
6746 if (mddev->pers->sync_request) {
6747 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6748 /* We are adding a device or devices to an array
6749 * which has the bitmap stored on all devices.
6750 * So make sure all bitmap pages get written
6751 */
6752 bitmap_write_all(mddev->bitmap);
6753 }
6754 mddev->sync_thread = md_register_thread(md_do_sync,
6755 mddev,
6756 "%s_resync");
6757 if (!mddev->sync_thread) {
6758 printk(KERN_ERR "%s: could not start resync"
6759 " thread...\n",
6760 mdname(mddev));
6761 /* leave the spares where they are, it shouldn't hurt */
6762 mddev->recovery = 0;
6763 } else
6764 md_wakeup_thread(mddev->sync_thread);
6765 sysfs_notify_dirent(mddev->sysfs_action);
6766 md_new_event(mddev);
6767 }
6768 unlock:
6769 if (!mddev->sync_thread) {
6770 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6771 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
6772 &mddev->recovery))
6773 if (mddev->sysfs_action)
6774 sysfs_notify_dirent(mddev->sysfs_action);
6775 }
6776 mddev_unlock(mddev);
6777 }
6778 }
6779
6780 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
6781 {
6782 sysfs_notify_dirent(rdev->sysfs_state);
6783 wait_event_timeout(rdev->blocked_wait,
6784 !test_bit(Blocked, &rdev->flags),
6785 msecs_to_jiffies(5000));
6786 rdev_dec_pending(rdev, mddev);
6787 }
6788 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
6789
6790 static int md_notify_reboot(struct notifier_block *this,
6791 unsigned long code, void *x)
6792 {
6793 struct list_head *tmp;
6794 mddev_t *mddev;
6795
6796 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
6797
6798 printk(KERN_INFO "md: stopping all md devices.\n");
6799
6800 for_each_mddev(mddev, tmp)
6801 if (mddev_trylock(mddev)) {
6802 /* Force a switch to readonly even array
6803 * appears to still be in use. Hence
6804 * the '100'.
6805 */
6806 do_md_stop(mddev, 1, 100);
6807 mddev_unlock(mddev);
6808 }
6809 /*
6810 * certain more exotic SCSI devices are known to be
6811 * volatile wrt too early system reboots. While the
6812 * right place to handle this issue is the given
6813 * driver, we do want to have a safe RAID driver ...
6814 */
6815 mdelay(1000*1);
6816 }
6817 return NOTIFY_DONE;
6818 }
6819
6820 static struct notifier_block md_notifier = {
6821 .notifier_call = md_notify_reboot,
6822 .next = NULL,
6823 .priority = INT_MAX, /* before any real devices */
6824 };
6825
6826 static void md_geninit(void)
6827 {
6828 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
6829
6830 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
6831 }
6832
6833 static int __init md_init(void)
6834 {
6835 if (register_blkdev(MD_MAJOR, "md"))
6836 return -1;
6837 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
6838 unregister_blkdev(MD_MAJOR, "md");
6839 return -1;
6840 }
6841 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
6842 md_probe, NULL, NULL);
6843 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
6844 md_probe, NULL, NULL);
6845
6846 register_reboot_notifier(&md_notifier);
6847 raid_table_header = register_sysctl_table(raid_root_table);
6848
6849 md_geninit();
6850 return 0;
6851 }
6852
6853
6854 #ifndef MODULE
6855
6856 /*
6857 * Searches all registered partitions for autorun RAID arrays
6858 * at boot time.
6859 */
6860
6861 static LIST_HEAD(all_detected_devices);
6862 struct detected_devices_node {
6863 struct list_head list;
6864 dev_t dev;
6865 };
6866
6867 void md_autodetect_dev(dev_t dev)
6868 {
6869 struct detected_devices_node *node_detected_dev;
6870
6871 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6872 if (node_detected_dev) {
6873 node_detected_dev->dev = dev;
6874 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6875 } else {
6876 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6877 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6878 }
6879 }
6880
6881
6882 static void autostart_arrays(int part)
6883 {
6884 mdk_rdev_t *rdev;
6885 struct detected_devices_node *node_detected_dev;
6886 dev_t dev;
6887 int i_scanned, i_passed;
6888
6889 i_scanned = 0;
6890 i_passed = 0;
6891
6892 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6893
6894 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6895 i_scanned++;
6896 node_detected_dev = list_entry(all_detected_devices.next,
6897 struct detected_devices_node, list);
6898 list_del(&node_detected_dev->list);
6899 dev = node_detected_dev->dev;
6900 kfree(node_detected_dev);
6901 rdev = md_import_device(dev,0, 90);
6902 if (IS_ERR(rdev))
6903 continue;
6904
6905 if (test_bit(Faulty, &rdev->flags)) {
6906 MD_BUG();
6907 continue;
6908 }
6909 set_bit(AutoDetected, &rdev->flags);
6910 list_add(&rdev->same_set, &pending_raid_disks);
6911 i_passed++;
6912 }
6913
6914 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6915 i_scanned, i_passed);
6916
6917 autorun_devices(part);
6918 }
6919
6920 #endif /* !MODULE */
6921
6922 static __exit void md_exit(void)
6923 {
6924 mddev_t *mddev;
6925 struct list_head *tmp;
6926
6927 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
6928 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6929
6930 unregister_blkdev(MD_MAJOR,"md");
6931 unregister_blkdev(mdp_major, "mdp");
6932 unregister_reboot_notifier(&md_notifier);
6933 unregister_sysctl_table(raid_table_header);
6934 remove_proc_entry("mdstat", NULL);
6935 for_each_mddev(mddev, tmp) {
6936 export_array(mddev);
6937 mddev->hold_active = 0;
6938 }
6939 }
6940
6941 subsys_initcall(md_init);
6942 module_exit(md_exit)
6943
6944 static int get_ro(char *buffer, struct kernel_param *kp)
6945 {
6946 return sprintf(buffer, "%d", start_readonly);
6947 }
6948 static int set_ro(const char *val, struct kernel_param *kp)
6949 {
6950 char *e;
6951 int num = simple_strtoul(val, &e, 10);
6952 if (*val && (*e == '\0' || *e == '\n')) {
6953 start_readonly = num;
6954 return 0;
6955 }
6956 return -EINVAL;
6957 }
6958
6959 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6960 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6961
6962 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
6963
6964 EXPORT_SYMBOL(register_md_personality);
6965 EXPORT_SYMBOL(unregister_md_personality);
6966 EXPORT_SYMBOL(md_error);
6967 EXPORT_SYMBOL(md_done_sync);
6968 EXPORT_SYMBOL(md_write_start);
6969 EXPORT_SYMBOL(md_write_end);
6970 EXPORT_SYMBOL(md_register_thread);
6971 EXPORT_SYMBOL(md_unregister_thread);
6972 EXPORT_SYMBOL(md_wakeup_thread);
6973 EXPORT_SYMBOL(md_check_recovery);
6974 MODULE_LICENSE("GPL");
6975 MODULE_ALIAS("md");
6976 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree