?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
m68k: ptrace fixes
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / md / md.c
blobb182f86a19dda9c71d12553e9d6891f2e24d0cfe
1 /*
2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
4
5 completely rewritten, based on the MD driver code from Marc Zyngier
6
7 Changes:
8
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
19
20 Neil Brown <neilb@cse.unsw.edu.au>.
21
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
28 any later version.
29
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33 */
34
35 #include <linux/kthread.h>
36 #include <linux/blkdev.h>
37 #include <linux/sysctl.h>
38 #include <linux/seq_file.h>
39 #include <linux/buffer_head.h> /* for invalidate_bdev */
40 #include <linux/poll.h>
41 #include <linux/ctype.h>
42 #include <linux/hdreg.h>
43 #include <linux/proc_fs.h>
44 #include <linux/random.h>
45 #include <linux/reboot.h>
46 #include <linux/file.h>
47 #include <linux/delay.h>
48 #include <linux/raid/md_p.h>
49 #include <linux/raid/md_u.h>
50 #include "md.h"
51 #include "bitmap.h"
52
53 #define DEBUG 0
54 #define dprintk(x...) ((void)(DEBUG && printk(x)))
55
56
57 #ifndef MODULE
58 static void autostart_arrays(int part);
59 #endif
60
61 static LIST_HEAD(pers_list);
62 static DEFINE_SPINLOCK(pers_lock);
63
64 static void md_print_devices(void);
65
66 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
67
68 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
69
70 /*
71 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
72 * is 1000 KB/sec, so the extra system load does not show up that much.
73 * Increase it if you want to have more _guaranteed_ speed. Note that
74 * the RAID driver will use the maximum available bandwidth if the IO
75 * subsystem is idle. There is also an 'absolute maximum' reconstruction
76 * speed limit - in case reconstruction slows down your system despite
77 * idle IO detection.
78 *
79 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
80 * or /sys/block/mdX/md/sync_speed_{min,max}
81 */
82
83 static int sysctl_speed_limit_min = 1000;
84 static int sysctl_speed_limit_max = 200000;
85 static inline int speed_min(mddev_t *mddev)
86 {
87 return mddev->sync_speed_min ?
88 mddev->sync_speed_min : sysctl_speed_limit_min;
89 }
90
91 static inline int speed_max(mddev_t *mddev)
92 {
93 return mddev->sync_speed_max ?
94 mddev->sync_speed_max : sysctl_speed_limit_max;
95 }
96
97 static struct ctl_table_header *raid_table_header;
98
99 static ctl_table raid_table[] = {
100 {
101 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
102 .procname = "speed_limit_min",
103 .data = &sysctl_speed_limit_min,
104 .maxlen = sizeof(int),
105 .mode = S_IRUGO|S_IWUSR,
106 .proc_handler = &proc_dointvec,
107 },
108 {
109 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
110 .procname = "speed_limit_max",
111 .data = &sysctl_speed_limit_max,
112 .maxlen = sizeof(int),
113 .mode = S_IRUGO|S_IWUSR,
114 .proc_handler = &proc_dointvec,
115 },
116 { .ctl_name = 0 }
117 };
118
119 static ctl_table raid_dir_table[] = {
120 {
121 .ctl_name = DEV_RAID,
122 .procname = "raid",
123 .maxlen = 0,
124 .mode = S_IRUGO|S_IXUGO,
125 .child = raid_table,
126 },
127 { .ctl_name = 0 }
128 };
129
130 static ctl_table raid_root_table[] = {
131 {
132 .ctl_name = CTL_DEV,
133 .procname = "dev",
134 .maxlen = 0,
135 .mode = 0555,
136 .child = raid_dir_table,
137 },
138 { .ctl_name = 0 }
139 };
140
141 static const struct block_device_operations md_fops;
142
143 static int start_readonly;
144
145 /*
146 * We have a system wide 'event count' that is incremented
147 * on any 'interesting' event, and readers of /proc/mdstat
148 * can use 'poll' or 'select' to find out when the event
149 * count increases.
150 *
151 * Events are:
152 * start array, stop array, error, add device, remove device,
153 * start build, activate spare
154 */
155 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
156 static atomic_t md_event_count;
157 void md_new_event(mddev_t *mddev)
158 {
159 atomic_inc(&md_event_count);
160 wake_up(&md_event_waiters);
161 }
162 EXPORT_SYMBOL_GPL(md_new_event);
163
164 /* Alternate version that can be called from interrupts
165 * when calling sysfs_notify isn't needed.
166 */
167 static void md_new_event_inintr(mddev_t *mddev)
168 {
169 atomic_inc(&md_event_count);
170 wake_up(&md_event_waiters);
171 }
172
173 /*
174 * Enables to iterate over all existing md arrays
175 * all_mddevs_lock protects this list.
176 */
177 static LIST_HEAD(all_mddevs);
178 static DEFINE_SPINLOCK(all_mddevs_lock);
179
180
181 /*
182 * iterates through all used mddevs in the system.
183 * We take care to grab the all_mddevs_lock whenever navigating
184 * the list, and to always hold a refcount when unlocked.
185 * Any code which breaks out of this loop while own
186 * a reference to the current mddev and must mddev_put it.
187 */
188 #define for_each_mddev(mddev,tmp) \
189 \
190 for (({ spin_lock(&all_mddevs_lock); \
191 tmp = all_mddevs.next; \
192 mddev = NULL;}); \
193 ({ if (tmp != &all_mddevs) \
194 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
195 spin_unlock(&all_mddevs_lock); \
196 if (mddev) mddev_put(mddev); \
197 mddev = list_entry(tmp, mddev_t, all_mddevs); \
198 tmp != &all_mddevs;}); \
199 ({ spin_lock(&all_mddevs_lock); \
200 tmp = tmp->next;}) \
201 )
202
203
204 /* Rather than calling directly into the personality make_request function,
205 * IO requests come here first so that we can check if the device is
206 * being suspended pending a reconfiguration.
207 * We hold a refcount over the call to ->make_request. By the time that
208 * call has finished, the bio has been linked into some internal structure
209 * and so is visible to ->quiesce(), so we don't need the refcount any more.
210 */
211 static int md_make_request(struct request_queue *q, struct bio *bio)
212 {
213 mddev_t *mddev = q->queuedata;
214 int rv;
215 if (mddev == NULL || mddev->pers == NULL) {
216 bio_io_error(bio);
217 return 0;
218 }
219 rcu_read_lock();
220 if (mddev->suspended) {
221 DEFINE_WAIT(__wait);
222 for (;;) {
223 prepare_to_wait(&mddev->sb_wait, &__wait,
224 TASK_UNINTERRUPTIBLE);
225 if (!mddev->suspended)
226 break;
227 rcu_read_unlock();
228 schedule();
229 rcu_read_lock();
230 }
231 finish_wait(&mddev->sb_wait, &__wait);
232 }
233 atomic_inc(&mddev->active_io);
234 rcu_read_unlock();
235 rv = mddev->pers->make_request(q, bio);
236 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
237 wake_up(&mddev->sb_wait);
238
239 return rv;
240 }
241
242 static void mddev_suspend(mddev_t *mddev)
243 {
244 BUG_ON(mddev->suspended);
245 mddev->suspended = 1;
246 synchronize_rcu();
247 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
248 mddev->pers->quiesce(mddev, 1);
249 md_unregister_thread(mddev->thread);
250 mddev->thread = NULL;
251 /* we now know that no code is executing in the personality module,
252 * except possibly the tail end of a ->bi_end_io function, but that
253 * is certain to complete before the module has a chance to get
254 * unloaded
255 */
256 }
257
258 static void mddev_resume(mddev_t *mddev)
259 {
260 mddev->suspended = 0;
261 wake_up(&mddev->sb_wait);
262 mddev->pers->quiesce(mddev, 0);
263 }
264
265 int mddev_congested(mddev_t *mddev, int bits)
266 {
267 return mddev->suspended;
268 }
269 EXPORT_SYMBOL(mddev_congested);
270
271
272 static inline mddev_t *mddev_get(mddev_t *mddev)
273 {
274 atomic_inc(&mddev->active);
275 return mddev;
276 }
277
278 static void mddev_delayed_delete(struct work_struct *ws);
279
280 static void mddev_put(mddev_t *mddev)
281 {
282 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
283 return;
284 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
285 !mddev->hold_active) {
286 list_del(&mddev->all_mddevs);
287 if (mddev->gendisk) {
288 /* we did a probe so need to clean up.
289 * Call schedule_work inside the spinlock
290 * so that flush_scheduled_work() after
291 * mddev_find will succeed in waiting for the
292 * work to be done.
293 */
294 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
295 schedule_work(&mddev->del_work);
296 } else
297 kfree(mddev);
298 }
299 spin_unlock(&all_mddevs_lock);
300 }
301
302 static mddev_t * mddev_find(dev_t unit)
303 {
304 mddev_t *mddev, *new = NULL;
305
306 retry:
307 spin_lock(&all_mddevs_lock);
308
309 if (unit) {
310 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
311 if (mddev->unit == unit) {
312 mddev_get(mddev);
313 spin_unlock(&all_mddevs_lock);
314 kfree(new);
315 return mddev;
316 }
317
318 if (new) {
319 list_add(&new->all_mddevs, &all_mddevs);
320 spin_unlock(&all_mddevs_lock);
321 new->hold_active = UNTIL_IOCTL;
322 return new;
323 }
324 } else if (new) {
325 /* find an unused unit number */
326 static int next_minor = 512;
327 int start = next_minor;
328 int is_free = 0;
329 int dev = 0;
330 while (!is_free) {
331 dev = MKDEV(MD_MAJOR, next_minor);
332 next_minor++;
333 if (next_minor > MINORMASK)
334 next_minor = 0;
335 if (next_minor == start) {
336 /* Oh dear, all in use. */
337 spin_unlock(&all_mddevs_lock);
338 kfree(new);
339 return NULL;
340 }
341
342 is_free = 1;
343 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
344 if (mddev->unit == dev) {
345 is_free = 0;
346 break;
347 }
348 }
349 new->unit = dev;
350 new->md_minor = MINOR(dev);
351 new->hold_active = UNTIL_STOP;
352 list_add(&new->all_mddevs, &all_mddevs);
353 spin_unlock(&all_mddevs_lock);
354 return new;
355 }
356 spin_unlock(&all_mddevs_lock);
357
358 new = kzalloc(sizeof(*new), GFP_KERNEL);
359 if (!new)
360 return NULL;
361
362 new->unit = unit;
363 if (MAJOR(unit) == MD_MAJOR)
364 new->md_minor = MINOR(unit);
365 else
366 new->md_minor = MINOR(unit) >> MdpMinorShift;
367
368 mutex_init(&new->open_mutex);
369 mutex_init(&new->reconfig_mutex);
370 INIT_LIST_HEAD(&new->disks);
371 INIT_LIST_HEAD(&new->all_mddevs);
372 init_timer(&new->safemode_timer);
373 atomic_set(&new->active, 1);
374 atomic_set(&new->openers, 0);
375 atomic_set(&new->active_io, 0);
376 spin_lock_init(&new->write_lock);
377 init_waitqueue_head(&new->sb_wait);
378 init_waitqueue_head(&new->recovery_wait);
379 new->reshape_position = MaxSector;
380 new->resync_min = 0;
381 new->resync_max = MaxSector;
382 new->level = LEVEL_NONE;
383
384 goto retry;
385 }
386
387 static inline int mddev_lock(mddev_t * mddev)
388 {
389 return mutex_lock_interruptible(&mddev->reconfig_mutex);
390 }
391
392 static inline int mddev_is_locked(mddev_t *mddev)
393 {
394 return mutex_is_locked(&mddev->reconfig_mutex);
395 }
396
397 static inline int mddev_trylock(mddev_t * mddev)
398 {
399 return mutex_trylock(&mddev->reconfig_mutex);
400 }
401
402 static inline void mddev_unlock(mddev_t * mddev)
403 {
404 mutex_unlock(&mddev->reconfig_mutex);
405
406 md_wakeup_thread(mddev->thread);
407 }
408
409 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
410 {
411 mdk_rdev_t *rdev;
412
413 list_for_each_entry(rdev, &mddev->disks, same_set)
414 if (rdev->desc_nr == nr)
415 return rdev;
416
417 return NULL;
418 }
419
420 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
421 {
422 mdk_rdev_t *rdev;
423
424 list_for_each_entry(rdev, &mddev->disks, same_set)
425 if (rdev->bdev->bd_dev == dev)
426 return rdev;
427
428 return NULL;
429 }
430
431 static struct mdk_personality *find_pers(int level, char *clevel)
432 {
433 struct mdk_personality *pers;
434 list_for_each_entry(pers, &pers_list, list) {
435 if (level != LEVEL_NONE && pers->level == level)
436 return pers;
437 if (strcmp(pers->name, clevel)==0)
438 return pers;
439 }
440 return NULL;
441 }
442
443 /* return the offset of the super block in 512byte sectors */
444 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
445 {
446 sector_t num_sectors = bdev->bd_inode->i_size / 512;
447 return MD_NEW_SIZE_SECTORS(num_sectors);
448 }
449
450 static int alloc_disk_sb(mdk_rdev_t * rdev)
451 {
452 if (rdev->sb_page)
453 MD_BUG();
454
455 rdev->sb_page = alloc_page(GFP_KERNEL);
456 if (!rdev->sb_page) {
457 printk(KERN_ALERT "md: out of memory.\n");
458 return -ENOMEM;
459 }
460
461 return 0;
462 }
463
464 static void free_disk_sb(mdk_rdev_t * rdev)
465 {
466 if (rdev->sb_page) {
467 put_page(rdev->sb_page);
468 rdev->sb_loaded = 0;
469 rdev->sb_page = NULL;
470 rdev->sb_start = 0;
471 rdev->sectors = 0;
472 }
473 }
474
475
476 static void super_written(struct bio *bio, int error)
477 {
478 mdk_rdev_t *rdev = bio->bi_private;
479 mddev_t *mddev = rdev->mddev;
480
481 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
482 printk("md: super_written gets error=%d, uptodate=%d\n",
483 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
484 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
485 md_error(mddev, rdev);
486 }
487
488 if (atomic_dec_and_test(&mddev->pending_writes))
489 wake_up(&mddev->sb_wait);
490 bio_put(bio);
491 }
492
493 static void super_written_barrier(struct bio *bio, int error)
494 {
495 struct bio *bio2 = bio->bi_private;
496 mdk_rdev_t *rdev = bio2->bi_private;
497 mddev_t *mddev = rdev->mddev;
498
499 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
500 error == -EOPNOTSUPP) {
501 unsigned long flags;
502 /* barriers don't appear to be supported :-( */
503 set_bit(BarriersNotsupp, &rdev->flags);
504 mddev->barriers_work = 0;
505 spin_lock_irqsave(&mddev->write_lock, flags);
506 bio2->bi_next = mddev->biolist;
507 mddev->biolist = bio2;
508 spin_unlock_irqrestore(&mddev->write_lock, flags);
509 wake_up(&mddev->sb_wait);
510 bio_put(bio);
511 } else {
512 bio_put(bio2);
513 bio->bi_private = rdev;
514 super_written(bio, error);
515 }
516 }
517
518 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
519 sector_t sector, int size, struct page *page)
520 {
521 /* write first size bytes of page to sector of rdev
522 * Increment mddev->pending_writes before returning
523 * and decrement it on completion, waking up sb_wait
524 * if zero is reached.
525 * If an error occurred, call md_error
526 *
527 * As we might need to resubmit the request if BIO_RW_BARRIER
528 * causes ENOTSUPP, we allocate a spare bio...
529 */
530 struct bio *bio = bio_alloc(GFP_NOIO, 1);
531 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG);
532
533 bio->bi_bdev = rdev->bdev;
534 bio->bi_sector = sector;
535 bio_add_page(bio, page, size, 0);
536 bio->bi_private = rdev;
537 bio->bi_end_io = super_written;
538 bio->bi_rw = rw;
539
540 atomic_inc(&mddev->pending_writes);
541 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
542 struct bio *rbio;
543 rw |= (1<<BIO_RW_BARRIER);
544 rbio = bio_clone(bio, GFP_NOIO);
545 rbio->bi_private = bio;
546 rbio->bi_end_io = super_written_barrier;
547 submit_bio(rw, rbio);
548 } else
549 submit_bio(rw, bio);
550 }
551
552 void md_super_wait(mddev_t *mddev)
553 {
554 /* wait for all superblock writes that were scheduled to complete.
555 * if any had to be retried (due to BARRIER problems), retry them
556 */
557 DEFINE_WAIT(wq);
558 for(;;) {
559 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
560 if (atomic_read(&mddev->pending_writes)==0)
561 break;
562 while (mddev->biolist) {
563 struct bio *bio;
564 spin_lock_irq(&mddev->write_lock);
565 bio = mddev->biolist;
566 mddev->biolist = bio->bi_next ;
567 bio->bi_next = NULL;
568 spin_unlock_irq(&mddev->write_lock);
569 submit_bio(bio->bi_rw, bio);
570 }
571 schedule();
572 }
573 finish_wait(&mddev->sb_wait, &wq);
574 }
575
576 static void bi_complete(struct bio *bio, int error)
577 {
578 complete((struct completion*)bio->bi_private);
579 }
580
581 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
582 struct page *page, int rw)
583 {
584 struct bio *bio = bio_alloc(GFP_NOIO, 1);
585 struct completion event;
586 int ret;
587
588 rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
589
590 bio->bi_bdev = bdev;
591 bio->bi_sector = sector;
592 bio_add_page(bio, page, size, 0);
593 init_completion(&event);
594 bio->bi_private = &event;
595 bio->bi_end_io = bi_complete;
596 submit_bio(rw, bio);
597 wait_for_completion(&event);
598
599 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
600 bio_put(bio);
601 return ret;
602 }
603 EXPORT_SYMBOL_GPL(sync_page_io);
604
605 static int read_disk_sb(mdk_rdev_t * rdev, int size)
606 {
607 char b[BDEVNAME_SIZE];
608 if (!rdev->sb_page) {
609 MD_BUG();
610 return -EINVAL;
611 }
612 if (rdev->sb_loaded)
613 return 0;
614
615
616 if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
617 goto fail;
618 rdev->sb_loaded = 1;
619 return 0;
620
621 fail:
622 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
623 bdevname(rdev->bdev,b));
624 return -EINVAL;
625 }
626
627 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
628 {
629 return sb1->set_uuid0 == sb2->set_uuid0 &&
630 sb1->set_uuid1 == sb2->set_uuid1 &&
631 sb1->set_uuid2 == sb2->set_uuid2 &&
632 sb1->set_uuid3 == sb2->set_uuid3;
633 }
634
635 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
636 {
637 int ret;
638 mdp_super_t *tmp1, *tmp2;
639
640 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
641 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
642
643 if (!tmp1 || !tmp2) {
644 ret = 0;
645 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
646 goto abort;
647 }
648
649 *tmp1 = *sb1;
650 *tmp2 = *sb2;
651
652 /*
653 * nr_disks is not constant
654 */
655 tmp1->nr_disks = 0;
656 tmp2->nr_disks = 0;
657
658 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
659 abort:
660 kfree(tmp1);
661 kfree(tmp2);
662 return ret;
663 }
664
665
666 static u32 md_csum_fold(u32 csum)
667 {
668 csum = (csum & 0xffff) + (csum >> 16);
669 return (csum & 0xffff) + (csum >> 16);
670 }
671
672 static unsigned int calc_sb_csum(mdp_super_t * sb)
673 {
674 u64 newcsum = 0;
675 u32 *sb32 = (u32*)sb;
676 int i;
677 unsigned int disk_csum, csum;
678
679 disk_csum = sb->sb_csum;
680 sb->sb_csum = 0;
681
682 for (i = 0; i < MD_SB_BYTES/4 ; i++)
683 newcsum += sb32[i];
684 csum = (newcsum & 0xffffffff) + (newcsum>>32);
685
686
687 #ifdef CONFIG_ALPHA
688 /* This used to use csum_partial, which was wrong for several
689 * reasons including that different results are returned on
690 * different architectures. It isn't critical that we get exactly
691 * the same return value as before (we always csum_fold before
692 * testing, and that removes any differences). However as we
693 * know that csum_partial always returned a 16bit value on
694 * alphas, do a fold to maximise conformity to previous behaviour.
695 */
696 sb->sb_csum = md_csum_fold(disk_csum);
697 #else
698 sb->sb_csum = disk_csum;
699 #endif
700 return csum;
701 }
702
703
704 /*
705 * Handle superblock details.
706 * We want to be able to handle multiple superblock formats
707 * so we have a common interface to them all, and an array of
708 * different handlers.
709 * We rely on user-space to write the initial superblock, and support
710 * reading and updating of superblocks.
711 * Interface methods are:
712 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
713 * loads and validates a superblock on dev.
714 * if refdev != NULL, compare superblocks on both devices
715 * Return:
716 * 0 - dev has a superblock that is compatible with refdev
717 * 1 - dev has a superblock that is compatible and newer than refdev
718 * so dev should be used as the refdev in future
719 * -EINVAL superblock incompatible or invalid
720 * -othererror e.g. -EIO
721 *
722 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
723 * Verify that dev is acceptable into mddev.
724 * The first time, mddev->raid_disks will be 0, and data from
725 * dev should be merged in. Subsequent calls check that dev
726 * is new enough. Return 0 or -EINVAL
727 *
728 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
729 * Update the superblock for rdev with data in mddev
730 * This does not write to disc.
731 *
732 */
733
734 struct super_type {
735 char *name;
736 struct module *owner;
737 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
738 int minor_version);
739 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
740 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
741 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
742 sector_t num_sectors);
743 };
744
745 /*
746 * Check that the given mddev has no bitmap.
747 *
748 * This function is called from the run method of all personalities that do not
749 * support bitmaps. It prints an error message and returns non-zero if mddev
750 * has a bitmap. Otherwise, it returns 0.
751 *
752 */
753 int md_check_no_bitmap(mddev_t *mddev)
754 {
755 if (!mddev->bitmap_file && !mddev->bitmap_offset)
756 return 0;
757 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
758 mdname(mddev), mddev->pers->name);
759 return 1;
760 }
761 EXPORT_SYMBOL(md_check_no_bitmap);
762
763 /*
764 * load_super for 0.90.0
765 */
766 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
767 {
768 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
769 mdp_super_t *sb;
770 int ret;
771
772 /*
773 * Calculate the position of the superblock (512byte sectors),
774 * it's at the end of the disk.
775 *
776 * It also happens to be a multiple of 4Kb.
777 */
778 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
779
780 ret = read_disk_sb(rdev, MD_SB_BYTES);
781 if (ret) return ret;
782
783 ret = -EINVAL;
784
785 bdevname(rdev->bdev, b);
786 sb = (mdp_super_t*)page_address(rdev->sb_page);
787
788 if (sb->md_magic != MD_SB_MAGIC) {
789 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
790 b);
791 goto abort;
792 }
793
794 if (sb->major_version != 0 ||
795 sb->minor_version < 90 ||
796 sb->minor_version > 91) {
797 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
798 sb->major_version, sb->minor_version,
799 b);
800 goto abort;
801 }
802
803 if (sb->raid_disks <= 0)
804 goto abort;
805
806 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
807 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
808 b);
809 goto abort;
810 }
811
812 rdev->preferred_minor = sb->md_minor;
813 rdev->data_offset = 0;
814 rdev->sb_size = MD_SB_BYTES;
815
816 if (sb->level == LEVEL_MULTIPATH)
817 rdev->desc_nr = -1;
818 else
819 rdev->desc_nr = sb->this_disk.number;
820
821 if (!refdev) {
822 ret = 1;
823 } else {
824 __u64 ev1, ev2;
825 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
826 if (!uuid_equal(refsb, sb)) {
827 printk(KERN_WARNING "md: %s has different UUID to %s\n",
828 b, bdevname(refdev->bdev,b2));
829 goto abort;
830 }
831 if (!sb_equal(refsb, sb)) {
832 printk(KERN_WARNING "md: %s has same UUID"
833 " but different superblock to %s\n",
834 b, bdevname(refdev->bdev, b2));
835 goto abort;
836 }
837 ev1 = md_event(sb);
838 ev2 = md_event(refsb);
839 if (ev1 > ev2)
840 ret = 1;
841 else
842 ret = 0;
843 }
844 rdev->sectors = rdev->sb_start;
845
846 if (rdev->sectors < sb->size * 2 && sb->level > 1)
847 /* "this cannot possibly happen" ... */
848 ret = -EINVAL;
849
850 abort:
851 return ret;
852 }
853
854 /*
855 * validate_super for 0.90.0
856 */
857 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
858 {
859 mdp_disk_t *desc;
860 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
861 __u64 ev1 = md_event(sb);
862
863 rdev->raid_disk = -1;
864 clear_bit(Faulty, &rdev->flags);
865 clear_bit(In_sync, &rdev->flags);
866 clear_bit(WriteMostly, &rdev->flags);
867 clear_bit(BarriersNotsupp, &rdev->flags);
868
869 if (mddev->raid_disks == 0) {
870 mddev->major_version = 0;
871 mddev->minor_version = sb->minor_version;
872 mddev->patch_version = sb->patch_version;
873 mddev->external = 0;
874 mddev->chunk_sectors = sb->chunk_size >> 9;
875 mddev->ctime = sb->ctime;
876 mddev->utime = sb->utime;
877 mddev->level = sb->level;
878 mddev->clevel[0] = 0;
879 mddev->layout = sb->layout;
880 mddev->raid_disks = sb->raid_disks;
881 mddev->dev_sectors = sb->size * 2;
882 mddev->events = ev1;
883 mddev->bitmap_offset = 0;
884 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
885
886 if (mddev->minor_version >= 91) {
887 mddev->reshape_position = sb->reshape_position;
888 mddev->delta_disks = sb->delta_disks;
889 mddev->new_level = sb->new_level;
890 mddev->new_layout = sb->new_layout;
891 mddev->new_chunk_sectors = sb->new_chunk >> 9;
892 } else {
893 mddev->reshape_position = MaxSector;
894 mddev->delta_disks = 0;
895 mddev->new_level = mddev->level;
896 mddev->new_layout = mddev->layout;
897 mddev->new_chunk_sectors = mddev->chunk_sectors;
898 }
899
900 if (sb->state & (1<<MD_SB_CLEAN))
901 mddev->recovery_cp = MaxSector;
902 else {
903 if (sb->events_hi == sb->cp_events_hi &&
904 sb->events_lo == sb->cp_events_lo) {
905 mddev->recovery_cp = sb->recovery_cp;
906 } else
907 mddev->recovery_cp = 0;
908 }
909
910 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
911 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
912 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
913 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
914
915 mddev->max_disks = MD_SB_DISKS;
916
917 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
918 mddev->bitmap_file == NULL)
919 mddev->bitmap_offset = mddev->default_bitmap_offset;
920
921 } else if (mddev->pers == NULL) {
922 /* Insist on good event counter while assembling */
923 ++ev1;
924 if (ev1 < mddev->events)
925 return -EINVAL;
926 } else if (mddev->bitmap) {
927 /* if adding to array with a bitmap, then we can accept an
928 * older device ... but not too old.
929 */
930 if (ev1 < mddev->bitmap->events_cleared)
931 return 0;
932 } else {
933 if (ev1 < mddev->events)
934 /* just a hot-add of a new device, leave raid_disk at -1 */
935 return 0;
936 }
937
938 if (mddev->level != LEVEL_MULTIPATH) {
939 desc = sb->disks + rdev->desc_nr;
940
941 if (desc->state & (1<<MD_DISK_FAULTY))
942 set_bit(Faulty, &rdev->flags);
943 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
944 desc->raid_disk < mddev->raid_disks */) {
945 set_bit(In_sync, &rdev->flags);
946 rdev->raid_disk = desc->raid_disk;
947 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
948 /* active but not in sync implies recovery up to
949 * reshape position. We don't know exactly where
950 * that is, so set to zero for now */
951 if (mddev->minor_version >= 91) {
952 rdev->recovery_offset = 0;
953 rdev->raid_disk = desc->raid_disk;
954 }
955 }
956 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
957 set_bit(WriteMostly, &rdev->flags);
958 } else /* MULTIPATH are always insync */
959 set_bit(In_sync, &rdev->flags);
960 return 0;
961 }
962
963 /*
964 * sync_super for 0.90.0
965 */
966 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
967 {
968 mdp_super_t *sb;
969 mdk_rdev_t *rdev2;
970 int next_spare = mddev->raid_disks;
971
972
973 /* make rdev->sb match mddev data..
974 *
975 * 1/ zero out disks
976 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
977 * 3/ any empty disks < next_spare become removed
978 *
979 * disks[0] gets initialised to REMOVED because
980 * we cannot be sure from other fields if it has
981 * been initialised or not.
982 */
983 int i;
984 int active=0, working=0,failed=0,spare=0,nr_disks=0;
985
986 rdev->sb_size = MD_SB_BYTES;
987
988 sb = (mdp_super_t*)page_address(rdev->sb_page);
989
990 memset(sb, 0, sizeof(*sb));
991
992 sb->md_magic = MD_SB_MAGIC;
993 sb->major_version = mddev->major_version;
994 sb->patch_version = mddev->patch_version;
995 sb->gvalid_words = 0; /* ignored */
996 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
997 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
998 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
999 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1000
1001 sb->ctime = mddev->ctime;
1002 sb->level = mddev->level;
1003 sb->size = mddev->dev_sectors / 2;
1004 sb->raid_disks = mddev->raid_disks;
1005 sb->md_minor = mddev->md_minor;
1006 sb->not_persistent = 0;
1007 sb->utime = mddev->utime;
1008 sb->state = 0;
1009 sb->events_hi = (mddev->events>>32);
1010 sb->events_lo = (u32)mddev->events;
1011
1012 if (mddev->reshape_position == MaxSector)
1013 sb->minor_version = 90;
1014 else {
1015 sb->minor_version = 91;
1016 sb->reshape_position = mddev->reshape_position;
1017 sb->new_level = mddev->new_level;
1018 sb->delta_disks = mddev->delta_disks;
1019 sb->new_layout = mddev->new_layout;
1020 sb->new_chunk = mddev->new_chunk_sectors << 9;
1021 }
1022 mddev->minor_version = sb->minor_version;
1023 if (mddev->in_sync)
1024 {
1025 sb->recovery_cp = mddev->recovery_cp;
1026 sb->cp_events_hi = (mddev->events>>32);
1027 sb->cp_events_lo = (u32)mddev->events;
1028 if (mddev->recovery_cp == MaxSector)
1029 sb->state = (1<< MD_SB_CLEAN);
1030 } else
1031 sb->recovery_cp = 0;
1032
1033 sb->layout = mddev->layout;
1034 sb->chunk_size = mddev->chunk_sectors << 9;
1035
1036 if (mddev->bitmap && mddev->bitmap_file == NULL)
1037 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1038
1039 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1040 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1041 mdp_disk_t *d;
1042 int desc_nr;
1043 int is_active = test_bit(In_sync, &rdev2->flags);
1044
1045 if (rdev2->raid_disk >= 0 &&
1046 sb->minor_version >= 91)
1047 /* we have nowhere to store the recovery_offset,
1048 * but if it is not below the reshape_position,
1049 * we can piggy-back on that.
1050 */
1051 is_active = 1;
1052 if (rdev2->raid_disk < 0 ||
1053 test_bit(Faulty, &rdev2->flags))
1054 is_active = 0;
1055 if (is_active)
1056 desc_nr = rdev2->raid_disk;
1057 else
1058 desc_nr = next_spare++;
1059 rdev2->desc_nr = desc_nr;
1060 d = &sb->disks[rdev2->desc_nr];
1061 nr_disks++;
1062 d->number = rdev2->desc_nr;
1063 d->major = MAJOR(rdev2->bdev->bd_dev);
1064 d->minor = MINOR(rdev2->bdev->bd_dev);
1065 if (is_active)
1066 d->raid_disk = rdev2->raid_disk;
1067 else
1068 d->raid_disk = rdev2->desc_nr; /* compatibility */
1069 if (test_bit(Faulty, &rdev2->flags))
1070 d->state = (1<<MD_DISK_FAULTY);
1071 else if (is_active) {
1072 d->state = (1<<MD_DISK_ACTIVE);
1073 if (test_bit(In_sync, &rdev2->flags))
1074 d->state |= (1<<MD_DISK_SYNC);
1075 active++;
1076 working++;
1077 } else {
1078 d->state = 0;
1079 spare++;
1080 working++;
1081 }
1082 if (test_bit(WriteMostly, &rdev2->flags))
1083 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1084 }
1085 /* now set the "removed" and "faulty" bits on any missing devices */
1086 for (i=0 ; i < mddev->raid_disks ; i++) {
1087 mdp_disk_t *d = &sb->disks[i];
1088 if (d->state == 0 && d->number == 0) {
1089 d->number = i;
1090 d->raid_disk = i;
1091 d->state = (1<<MD_DISK_REMOVED);
1092 d->state |= (1<<MD_DISK_FAULTY);
1093 failed++;
1094 }
1095 }
1096 sb->nr_disks = nr_disks;
1097 sb->active_disks = active;
1098 sb->working_disks = working;
1099 sb->failed_disks = failed;
1100 sb->spare_disks = spare;
1101
1102 sb->this_disk = sb->disks[rdev->desc_nr];
1103 sb->sb_csum = calc_sb_csum(sb);
1104 }
1105
1106 /*
1107 * rdev_size_change for 0.90.0
1108 */
1109 static unsigned long long
1110 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1111 {
1112 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1113 return 0; /* component must fit device */
1114 if (rdev->mddev->bitmap_offset)
1115 return 0; /* can't move bitmap */
1116 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1117 if (!num_sectors || num_sectors > rdev->sb_start)
1118 num_sectors = rdev->sb_start;
1119 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1120 rdev->sb_page);
1121 md_super_wait(rdev->mddev);
1122 return num_sectors / 2; /* kB for sysfs */
1123 }
1124
1125
1126 /*
1127 * version 1 superblock
1128 */
1129
1130 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1131 {
1132 __le32 disk_csum;
1133 u32 csum;
1134 unsigned long long newcsum;
1135 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1136 __le32 *isuper = (__le32*)sb;
1137 int i;
1138
1139 disk_csum = sb->sb_csum;
1140 sb->sb_csum = 0;
1141 newcsum = 0;
1142 for (i=0; size>=4; size -= 4 )
1143 newcsum += le32_to_cpu(*isuper++);
1144
1145 if (size == 2)
1146 newcsum += le16_to_cpu(*(__le16*) isuper);
1147
1148 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1149 sb->sb_csum = disk_csum;
1150 return cpu_to_le32(csum);
1151 }
1152
1153 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1154 {
1155 struct mdp_superblock_1 *sb;
1156 int ret;
1157 sector_t sb_start;
1158 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1159 int bmask;
1160
1161 /*
1162 * Calculate the position of the superblock in 512byte sectors.
1163 * It is always aligned to a 4K boundary and
1164 * depeding on minor_version, it can be:
1165 * 0: At least 8K, but less than 12K, from end of device
1166 * 1: At start of device
1167 * 2: 4K from start of device.
1168 */
1169 switch(minor_version) {
1170 case 0:
1171 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1172 sb_start -= 8*2;
1173 sb_start &= ~(sector_t)(4*2-1);
1174 break;
1175 case 1:
1176 sb_start = 0;
1177 break;
1178 case 2:
1179 sb_start = 8;
1180 break;
1181 default:
1182 return -EINVAL;
1183 }
1184 rdev->sb_start = sb_start;
1185
1186 /* superblock is rarely larger than 1K, but it can be larger,
1187 * and it is safe to read 4k, so we do that
1188 */
1189 ret = read_disk_sb(rdev, 4096);
1190 if (ret) return ret;
1191
1192
1193 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1194
1195 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1196 sb->major_version != cpu_to_le32(1) ||
1197 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1198 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1199 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1200 return -EINVAL;
1201
1202 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1203 printk("md: invalid superblock checksum on %s\n",
1204 bdevname(rdev->bdev,b));
1205 return -EINVAL;
1206 }
1207 if (le64_to_cpu(sb->data_size) < 10) {
1208 printk("md: data_size too small on %s\n",
1209 bdevname(rdev->bdev,b));
1210 return -EINVAL;
1211 }
1212
1213 rdev->preferred_minor = 0xffff;
1214 rdev->data_offset = le64_to_cpu(sb->data_offset);
1215 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1216
1217 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1218 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1219 if (rdev->sb_size & bmask)
1220 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1221
1222 if (minor_version
1223 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1224 return -EINVAL;
1225
1226 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1227 rdev->desc_nr = -1;
1228 else
1229 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1230
1231 if (!refdev) {
1232 ret = 1;
1233 } else {
1234 __u64 ev1, ev2;
1235 struct mdp_superblock_1 *refsb =
1236 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1237
1238 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1239 sb->level != refsb->level ||
1240 sb->layout != refsb->layout ||
1241 sb->chunksize != refsb->chunksize) {
1242 printk(KERN_WARNING "md: %s has strangely different"
1243 " superblock to %s\n",
1244 bdevname(rdev->bdev,b),
1245 bdevname(refdev->bdev,b2));
1246 return -EINVAL;
1247 }
1248 ev1 = le64_to_cpu(sb->events);
1249 ev2 = le64_to_cpu(refsb->events);
1250
1251 if (ev1 > ev2)
1252 ret = 1;
1253 else
1254 ret = 0;
1255 }
1256 if (minor_version)
1257 rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
1258 le64_to_cpu(sb->data_offset);
1259 else
1260 rdev->sectors = rdev->sb_start;
1261 if (rdev->sectors < le64_to_cpu(sb->data_size))
1262 return -EINVAL;
1263 rdev->sectors = le64_to_cpu(sb->data_size);
1264 if (le64_to_cpu(sb->size) > rdev->sectors)
1265 return -EINVAL;
1266 return ret;
1267 }
1268
1269 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1270 {
1271 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1272 __u64 ev1 = le64_to_cpu(sb->events);
1273
1274 rdev->raid_disk = -1;
1275 clear_bit(Faulty, &rdev->flags);
1276 clear_bit(In_sync, &rdev->flags);
1277 clear_bit(WriteMostly, &rdev->flags);
1278 clear_bit(BarriersNotsupp, &rdev->flags);
1279
1280 if (mddev->raid_disks == 0) {
1281 mddev->major_version = 1;
1282 mddev->patch_version = 0;
1283 mddev->external = 0;
1284 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1285 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1286 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1287 mddev->level = le32_to_cpu(sb->level);
1288 mddev->clevel[0] = 0;
1289 mddev->layout = le32_to_cpu(sb->layout);
1290 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1291 mddev->dev_sectors = le64_to_cpu(sb->size);
1292 mddev->events = ev1;
1293 mddev->bitmap_offset = 0;
1294 mddev->default_bitmap_offset = 1024 >> 9;
1295
1296 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1297 memcpy(mddev->uuid, sb->set_uuid, 16);
1298
1299 mddev->max_disks = (4096-256)/2;
1300
1301 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1302 mddev->bitmap_file == NULL )
1303 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1304
1305 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1306 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1307 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1308 mddev->new_level = le32_to_cpu(sb->new_level);
1309 mddev->new_layout = le32_to_cpu(sb->new_layout);
1310 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1311 } else {
1312 mddev->reshape_position = MaxSector;
1313 mddev->delta_disks = 0;
1314 mddev->new_level = mddev->level;
1315 mddev->new_layout = mddev->layout;
1316 mddev->new_chunk_sectors = mddev->chunk_sectors;
1317 }
1318
1319 } else if (mddev->pers == NULL) {
1320 /* Insist of good event counter while assembling */
1321 ++ev1;
1322 if (ev1 < mddev->events)
1323 return -EINVAL;
1324 } else if (mddev->bitmap) {
1325 /* If adding to array with a bitmap, then we can accept an
1326 * older device, but not too old.
1327 */
1328 if (ev1 < mddev->bitmap->events_cleared)
1329 return 0;
1330 } else {
1331 if (ev1 < mddev->events)
1332 /* just a hot-add of a new device, leave raid_disk at -1 */
1333 return 0;
1334 }
1335 if (mddev->level != LEVEL_MULTIPATH) {
1336 int role;
1337 if (rdev->desc_nr < 0 ||
1338 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1339 role = 0xffff;
1340 rdev->desc_nr = -1;
1341 } else
1342 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1343 switch(role) {
1344 case 0xffff: /* spare */
1345 break;
1346 case 0xfffe: /* faulty */
1347 set_bit(Faulty, &rdev->flags);
1348 break;
1349 default:
1350 if ((le32_to_cpu(sb->feature_map) &
1351 MD_FEATURE_RECOVERY_OFFSET))
1352 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1353 else
1354 set_bit(In_sync, &rdev->flags);
1355 rdev->raid_disk = role;
1356 break;
1357 }
1358 if (sb->devflags & WriteMostly1)
1359 set_bit(WriteMostly, &rdev->flags);
1360 } else /* MULTIPATH are always insync */
1361 set_bit(In_sync, &rdev->flags);
1362
1363 return 0;
1364 }
1365
1366 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1367 {
1368 struct mdp_superblock_1 *sb;
1369 mdk_rdev_t *rdev2;
1370 int max_dev, i;
1371 /* make rdev->sb match mddev and rdev data. */
1372
1373 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1374
1375 sb->feature_map = 0;
1376 sb->pad0 = 0;
1377 sb->recovery_offset = cpu_to_le64(0);
1378 memset(sb->pad1, 0, sizeof(sb->pad1));
1379 memset(sb->pad2, 0, sizeof(sb->pad2));
1380 memset(sb->pad3, 0, sizeof(sb->pad3));
1381
1382 sb->utime = cpu_to_le64((__u64)mddev->utime);
1383 sb->events = cpu_to_le64(mddev->events);
1384 if (mddev->in_sync)
1385 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1386 else
1387 sb->resync_offset = cpu_to_le64(0);
1388
1389 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1390
1391 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1392 sb->size = cpu_to_le64(mddev->dev_sectors);
1393 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1394 sb->level = cpu_to_le32(mddev->level);
1395 sb->layout = cpu_to_le32(mddev->layout);
1396
1397 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1398 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1399 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1400 }
1401
1402 if (rdev->raid_disk >= 0 &&
1403 !test_bit(In_sync, &rdev->flags)) {
1404 if (rdev->recovery_offset > 0) {
1405 sb->feature_map |=
1406 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1407 sb->recovery_offset =
1408 cpu_to_le64(rdev->recovery_offset);
1409 }
1410 }
1411
1412 if (mddev->reshape_position != MaxSector) {
1413 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1414 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1415 sb->new_layout = cpu_to_le32(mddev->new_layout);
1416 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1417 sb->new_level = cpu_to_le32(mddev->new_level);
1418 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1419 }
1420
1421 max_dev = 0;
1422 list_for_each_entry(rdev2, &mddev->disks, same_set)
1423 if (rdev2->desc_nr+1 > max_dev)
1424 max_dev = rdev2->desc_nr+1;
1425
1426 if (max_dev > le32_to_cpu(sb->max_dev)) {
1427 int bmask;
1428 sb->max_dev = cpu_to_le32(max_dev);
1429 rdev->sb_size = max_dev * 2 + 256;
1430 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1431 if (rdev->sb_size & bmask)
1432 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1433 }
1434 for (i=0; i<max_dev;i++)
1435 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1436
1437 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1438 i = rdev2->desc_nr;
1439 if (test_bit(Faulty, &rdev2->flags))
1440 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1441 else if (test_bit(In_sync, &rdev2->flags))
1442 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1443 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1444 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1445 else
1446 sb->dev_roles[i] = cpu_to_le16(0xffff);
1447 }
1448
1449 sb->sb_csum = calc_sb_1_csum(sb);
1450 }
1451
1452 static unsigned long long
1453 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1454 {
1455 struct mdp_superblock_1 *sb;
1456 sector_t max_sectors;
1457 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1458 return 0; /* component must fit device */
1459 if (rdev->sb_start < rdev->data_offset) {
1460 /* minor versions 1 and 2; superblock before data */
1461 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1462 max_sectors -= rdev->data_offset;
1463 if (!num_sectors || num_sectors > max_sectors)
1464 num_sectors = max_sectors;
1465 } else if (rdev->mddev->bitmap_offset) {
1466 /* minor version 0 with bitmap we can't move */
1467 return 0;
1468 } else {
1469 /* minor version 0; superblock after data */
1470 sector_t sb_start;
1471 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1472 sb_start &= ~(sector_t)(4*2 - 1);
1473 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1474 if (!num_sectors || num_sectors > max_sectors)
1475 num_sectors = max_sectors;
1476 rdev->sb_start = sb_start;
1477 }
1478 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1479 sb->data_size = cpu_to_le64(num_sectors);
1480 sb->super_offset = rdev->sb_start;
1481 sb->sb_csum = calc_sb_1_csum(sb);
1482 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1483 rdev->sb_page);
1484 md_super_wait(rdev->mddev);
1485 return num_sectors / 2; /* kB for sysfs */
1486 }
1487
1488 static struct super_type super_types[] = {
1489 [0] = {
1490 .name = "0.90.0",
1491 .owner = THIS_MODULE,
1492 .load_super = super_90_load,
1493 .validate_super = super_90_validate,
1494 .sync_super = super_90_sync,
1495 .rdev_size_change = super_90_rdev_size_change,
1496 },
1497 [1] = {
1498 .name = "md-1",
1499 .owner = THIS_MODULE,
1500 .load_super = super_1_load,
1501 .validate_super = super_1_validate,
1502 .sync_super = super_1_sync,
1503 .rdev_size_change = super_1_rdev_size_change,
1504 },
1505 };
1506
1507 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1508 {
1509 mdk_rdev_t *rdev, *rdev2;
1510
1511 rcu_read_lock();
1512 rdev_for_each_rcu(rdev, mddev1)
1513 rdev_for_each_rcu(rdev2, mddev2)
1514 if (rdev->bdev->bd_contains ==
1515 rdev2->bdev->bd_contains) {
1516 rcu_read_unlock();
1517 return 1;
1518 }
1519 rcu_read_unlock();
1520 return 0;
1521 }
1522
1523 static LIST_HEAD(pending_raid_disks);
1524
1525 /*
1526 * Try to register data integrity profile for an mddev
1527 *
1528 * This is called when an array is started and after a disk has been kicked
1529 * from the array. It only succeeds if all working and active component devices
1530 * are integrity capable with matching profiles.
1531 */
1532 int md_integrity_register(mddev_t *mddev)
1533 {
1534 mdk_rdev_t *rdev, *reference = NULL;
1535
1536 if (list_empty(&mddev->disks))
1537 return 0; /* nothing to do */
1538 if (blk_get_integrity(mddev->gendisk))
1539 return 0; /* already registered */
1540 list_for_each_entry(rdev, &mddev->disks, same_set) {
1541 /* skip spares and non-functional disks */
1542 if (test_bit(Faulty, &rdev->flags))
1543 continue;
1544 if (rdev->raid_disk < 0)
1545 continue;
1546 /*
1547 * If at least one rdev is not integrity capable, we can not
1548 * enable data integrity for the md device.
1549 */
1550 if (!bdev_get_integrity(rdev->bdev))
1551 return -EINVAL;
1552 if (!reference) {
1553 /* Use the first rdev as the reference */
1554 reference = rdev;
1555 continue;
1556 }
1557 /* does this rdev's profile match the reference profile? */
1558 if (blk_integrity_compare(reference->bdev->bd_disk,
1559 rdev->bdev->bd_disk) < 0)
1560 return -EINVAL;
1561 }
1562 /*
1563 * All component devices are integrity capable and have matching
1564 * profiles, register the common profile for the md device.
1565 */
1566 if (blk_integrity_register(mddev->gendisk,
1567 bdev_get_integrity(reference->bdev)) != 0) {
1568 printk(KERN_ERR "md: failed to register integrity for %s\n",
1569 mdname(mddev));
1570 return -EINVAL;
1571 }
1572 printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1573 mdname(mddev));
1574 return 0;
1575 }
1576 EXPORT_SYMBOL(md_integrity_register);
1577
1578 /* Disable data integrity if non-capable/non-matching disk is being added */
1579 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
1580 {
1581 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1582 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1583
1584 if (!bi_mddev) /* nothing to do */
1585 return;
1586 if (rdev->raid_disk < 0) /* skip spares */
1587 return;
1588 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1589 rdev->bdev->bd_disk) >= 0)
1590 return;
1591 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1592 blk_integrity_unregister(mddev->gendisk);
1593 }
1594 EXPORT_SYMBOL(md_integrity_add_rdev);
1595
1596 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1597 {
1598 char b[BDEVNAME_SIZE];
1599 struct kobject *ko;
1600 char *s;
1601 int err;
1602
1603 if (rdev->mddev) {
1604 MD_BUG();
1605 return -EINVAL;
1606 }
1607
1608 /* prevent duplicates */
1609 if (find_rdev(mddev, rdev->bdev->bd_dev))
1610 return -EEXIST;
1611
1612 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1613 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1614 rdev->sectors < mddev->dev_sectors)) {
1615 if (mddev->pers) {
1616 /* Cannot change size, so fail
1617 * If mddev->level <= 0, then we don't care
1618 * about aligning sizes (e.g. linear)
1619 */
1620 if (mddev->level > 0)
1621 return -ENOSPC;
1622 } else
1623 mddev->dev_sectors = rdev->sectors;
1624 }
1625
1626 /* Verify rdev->desc_nr is unique.
1627 * If it is -1, assign a free number, else
1628 * check number is not in use
1629 */
1630 if (rdev->desc_nr < 0) {
1631 int choice = 0;
1632 if (mddev->pers) choice = mddev->raid_disks;
1633 while (find_rdev_nr(mddev, choice))
1634 choice++;
1635 rdev->desc_nr = choice;
1636 } else {
1637 if (find_rdev_nr(mddev, rdev->desc_nr))
1638 return -EBUSY;
1639 }
1640 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1641 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1642 mdname(mddev), mddev->max_disks);
1643 return -EBUSY;
1644 }
1645 bdevname(rdev->bdev,b);
1646 while ( (s=strchr(b, '/')) != NULL)
1647 *s = '!';
1648
1649 rdev->mddev = mddev;
1650 printk(KERN_INFO "md: bind<%s>\n", b);
1651
1652 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1653 goto fail;
1654
1655 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1656 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1657 kobject_del(&rdev->kobj);
1658 goto fail;
1659 }
1660 rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, "state");
1661
1662 list_add_rcu(&rdev->same_set, &mddev->disks);
1663 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1664
1665 /* May as well allow recovery to be retried once */
1666 mddev->recovery_disabled = 0;
1667
1668 return 0;
1669
1670 fail:
1671 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1672 b, mdname(mddev));
1673 return err;
1674 }
1675
1676 static void md_delayed_delete(struct work_struct *ws)
1677 {
1678 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1679 kobject_del(&rdev->kobj);
1680 kobject_put(&rdev->kobj);
1681 }
1682
1683 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1684 {
1685 char b[BDEVNAME_SIZE];
1686 if (!rdev->mddev) {
1687 MD_BUG();
1688 return;
1689 }
1690 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1691 list_del_rcu(&rdev->same_set);
1692 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1693 rdev->mddev = NULL;
1694 sysfs_remove_link(&rdev->kobj, "block");
1695 sysfs_put(rdev->sysfs_state);
1696 rdev->sysfs_state = NULL;
1697 /* We need to delay this, otherwise we can deadlock when
1698 * writing to 'remove' to "dev/state". We also need
1699 * to delay it due to rcu usage.
1700 */
1701 synchronize_rcu();
1702 INIT_WORK(&rdev->del_work, md_delayed_delete);
1703 kobject_get(&rdev->kobj);
1704 schedule_work(&rdev->del_work);
1705 }
1706
1707 /*
1708 * prevent the device from being mounted, repartitioned or
1709 * otherwise reused by a RAID array (or any other kernel
1710 * subsystem), by bd_claiming the device.
1711 */
1712 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1713 {
1714 int err = 0;
1715 struct block_device *bdev;
1716 char b[BDEVNAME_SIZE];
1717
1718 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1719 if (IS_ERR(bdev)) {
1720 printk(KERN_ERR "md: could not open %s.\n",
1721 __bdevname(dev, b));
1722 return PTR_ERR(bdev);
1723 }
1724 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1725 if (err) {
1726 printk(KERN_ERR "md: could not bd_claim %s.\n",
1727 bdevname(bdev, b));
1728 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1729 return err;
1730 }
1731 if (!shared)
1732 set_bit(AllReserved, &rdev->flags);
1733 rdev->bdev = bdev;
1734 return err;
1735 }
1736
1737 static void unlock_rdev(mdk_rdev_t *rdev)
1738 {
1739 struct block_device *bdev = rdev->bdev;
1740 rdev->bdev = NULL;
1741 if (!bdev)
1742 MD_BUG();
1743 bd_release(bdev);
1744 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1745 }
1746
1747 void md_autodetect_dev(dev_t dev);
1748
1749 static void export_rdev(mdk_rdev_t * rdev)
1750 {
1751 char b[BDEVNAME_SIZE];
1752 printk(KERN_INFO "md: export_rdev(%s)\n",
1753 bdevname(rdev->bdev,b));
1754 if (rdev->mddev)
1755 MD_BUG();
1756 free_disk_sb(rdev);
1757 #ifndef MODULE
1758 if (test_bit(AutoDetected, &rdev->flags))
1759 md_autodetect_dev(rdev->bdev->bd_dev);
1760 #endif
1761 unlock_rdev(rdev);
1762 kobject_put(&rdev->kobj);
1763 }
1764
1765 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1766 {
1767 unbind_rdev_from_array(rdev);
1768 export_rdev(rdev);
1769 }
1770
1771 static void export_array(mddev_t *mddev)
1772 {
1773 mdk_rdev_t *rdev, *tmp;
1774
1775 rdev_for_each(rdev, tmp, mddev) {
1776 if (!rdev->mddev) {
1777 MD_BUG();
1778 continue;
1779 }
1780 kick_rdev_from_array(rdev);
1781 }
1782 if (!list_empty(&mddev->disks))
1783 MD_BUG();
1784 mddev->raid_disks = 0;
1785 mddev->major_version = 0;
1786 }
1787
1788 static void print_desc(mdp_disk_t *desc)
1789 {
1790 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1791 desc->major,desc->minor,desc->raid_disk,desc->state);
1792 }
1793
1794 static void print_sb_90(mdp_super_t *sb)
1795 {
1796 int i;
1797
1798 printk(KERN_INFO
1799 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1800 sb->major_version, sb->minor_version, sb->patch_version,
1801 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1802 sb->ctime);
1803 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1804 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1805 sb->md_minor, sb->layout, sb->chunk_size);
1806 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1807 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1808 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1809 sb->failed_disks, sb->spare_disks,
1810 sb->sb_csum, (unsigned long)sb->events_lo);
1811
1812 printk(KERN_INFO);
1813 for (i = 0; i < MD_SB_DISKS; i++) {
1814 mdp_disk_t *desc;
1815
1816 desc = sb->disks + i;
1817 if (desc->number || desc->major || desc->minor ||
1818 desc->raid_disk || (desc->state && (desc->state != 4))) {
1819 printk(" D %2d: ", i);
1820 print_desc(desc);
1821 }
1822 }
1823 printk(KERN_INFO "md: THIS: ");
1824 print_desc(&sb->this_disk);
1825 }
1826
1827 static void print_sb_1(struct mdp_superblock_1 *sb)
1828 {
1829 __u8 *uuid;
1830
1831 uuid = sb->set_uuid;
1832 printk(KERN_INFO
1833 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%02x%02x%02x%02x"
1834 ":%02x%02x:%02x%02x:%02x%02x:%02x%02x%02x%02x%02x%02x>\n"
1835 "md: Name: \"%s\" CT:%llu\n",
1836 le32_to_cpu(sb->major_version),
1837 le32_to_cpu(sb->feature_map),
1838 uuid[0], uuid[1], uuid[2], uuid[3],
1839 uuid[4], uuid[5], uuid[6], uuid[7],
1840 uuid[8], uuid[9], uuid[10], uuid[11],
1841 uuid[12], uuid[13], uuid[14], uuid[15],
1842 sb->set_name,
1843 (unsigned long long)le64_to_cpu(sb->ctime)
1844 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
1845
1846 uuid = sb->device_uuid;
1847 printk(KERN_INFO
1848 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
1849 " RO:%llu\n"
1850 "md: Dev:%08x UUID: %02x%02x%02x%02x:%02x%02x:%02x%02x:%02x%02x"
1851 ":%02x%02x%02x%02x%02x%02x\n"
1852 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
1853 "md: (MaxDev:%u) \n",
1854 le32_to_cpu(sb->level),
1855 (unsigned long long)le64_to_cpu(sb->size),
1856 le32_to_cpu(sb->raid_disks),
1857 le32_to_cpu(sb->layout),
1858 le32_to_cpu(sb->chunksize),
1859 (unsigned long long)le64_to_cpu(sb->data_offset),
1860 (unsigned long long)le64_to_cpu(sb->data_size),
1861 (unsigned long long)le64_to_cpu(sb->super_offset),
1862 (unsigned long long)le64_to_cpu(sb->recovery_offset),
1863 le32_to_cpu(sb->dev_number),
1864 uuid[0], uuid[1], uuid[2], uuid[3],
1865 uuid[4], uuid[5], uuid[6], uuid[7],
1866 uuid[8], uuid[9], uuid[10], uuid[11],
1867 uuid[12], uuid[13], uuid[14], uuid[15],
1868 sb->devflags,
1869 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
1870 (unsigned long long)le64_to_cpu(sb->events),
1871 (unsigned long long)le64_to_cpu(sb->resync_offset),
1872 le32_to_cpu(sb->sb_csum),
1873 le32_to_cpu(sb->max_dev)
1874 );
1875 }
1876
1877 static void print_rdev(mdk_rdev_t *rdev, int major_version)
1878 {
1879 char b[BDEVNAME_SIZE];
1880 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
1881 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
1882 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1883 rdev->desc_nr);
1884 if (rdev->sb_loaded) {
1885 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
1886 switch (major_version) {
1887 case 0:
1888 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
1889 break;
1890 case 1:
1891 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
1892 break;
1893 }
1894 } else
1895 printk(KERN_INFO "md: no rdev superblock!\n");
1896 }
1897
1898 static void md_print_devices(void)
1899 {
1900 struct list_head *tmp;
1901 mdk_rdev_t *rdev;
1902 mddev_t *mddev;
1903 char b[BDEVNAME_SIZE];
1904
1905 printk("\n");
1906 printk("md: **********************************\n");
1907 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1908 printk("md: **********************************\n");
1909 for_each_mddev(mddev, tmp) {
1910
1911 if (mddev->bitmap)
1912 bitmap_print_sb(mddev->bitmap);
1913 else
1914 printk("%s: ", mdname(mddev));
1915 list_for_each_entry(rdev, &mddev->disks, same_set)
1916 printk("<%s>", bdevname(rdev->bdev,b));
1917 printk("\n");
1918
1919 list_for_each_entry(rdev, &mddev->disks, same_set)
1920 print_rdev(rdev, mddev->major_version);
1921 }
1922 printk("md: **********************************\n");
1923 printk("\n");
1924 }
1925
1926
1927 static void sync_sbs(mddev_t * mddev, int nospares)
1928 {
1929 /* Update each superblock (in-memory image), but
1930 * if we are allowed to, skip spares which already
1931 * have the right event counter, or have one earlier
1932 * (which would mean they aren't being marked as dirty
1933 * with the rest of the array)
1934 */
1935 mdk_rdev_t *rdev;
1936
1937 /* First make sure individual recovery_offsets are correct */
1938 list_for_each_entry(rdev, &mddev->disks, same_set) {
1939 if (rdev->raid_disk >= 0 &&
1940 !test_bit(In_sync, &rdev->flags) &&
1941 mddev->curr_resync_completed > rdev->recovery_offset)
1942 rdev->recovery_offset = mddev->curr_resync_completed;
1943
1944 }
1945 list_for_each_entry(rdev, &mddev->disks, same_set) {
1946 if (rdev->sb_events == mddev->events ||
1947 (nospares &&
1948 rdev->raid_disk < 0 &&
1949 (rdev->sb_events&1)==0 &&
1950 rdev->sb_events+1 == mddev->events)) {
1951 /* Don't update this superblock */
1952 rdev->sb_loaded = 2;
1953 } else {
1954 super_types[mddev->major_version].
1955 sync_super(mddev, rdev);
1956 rdev->sb_loaded = 1;
1957 }
1958 }
1959 }
1960
1961 static void md_update_sb(mddev_t * mddev, int force_change)
1962 {
1963 mdk_rdev_t *rdev;
1964 int sync_req;
1965 int nospares = 0;
1966
1967 mddev->utime = get_seconds();
1968 if (mddev->external)
1969 return;
1970 repeat:
1971 spin_lock_irq(&mddev->write_lock);
1972
1973 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1974 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1975 force_change = 1;
1976 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1977 /* just a clean<-> dirty transition, possibly leave spares alone,
1978 * though if events isn't the right even/odd, we will have to do
1979 * spares after all
1980 */
1981 nospares = 1;
1982 if (force_change)
1983 nospares = 0;
1984 if (mddev->degraded)
1985 /* If the array is degraded, then skipping spares is both
1986 * dangerous and fairly pointless.
1987 * Dangerous because a device that was removed from the array
1988 * might have a event_count that still looks up-to-date,
1989 * so it can be re-added without a resync.
1990 * Pointless because if there are any spares to skip,
1991 * then a recovery will happen and soon that array won't
1992 * be degraded any more and the spare can go back to sleep then.
1993 */
1994 nospares = 0;
1995
1996 sync_req = mddev->in_sync;
1997
1998 /* If this is just a dirty<->clean transition, and the array is clean
1999 * and 'events' is odd, we can roll back to the previous clean state */
2000 if (nospares
2001 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2002 && (mddev->events & 1)
2003 && mddev->events != 1)
2004 mddev->events--;
2005 else {
2006 /* otherwise we have to go forward and ... */
2007 mddev->events ++;
2008 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
2009 /* .. if the array isn't clean, an 'even' event must also go
2010 * to spares. */
2011 if ((mddev->events&1)==0)
2012 nospares = 0;
2013 } else {
2014 /* otherwise an 'odd' event must go to spares */
2015 if ((mddev->events&1))
2016 nospares = 0;
2017 }
2018 }
2019
2020 if (!mddev->events) {
2021 /*
2022 * oops, this 64-bit counter should never wrap.
2023 * Either we are in around ~1 trillion A.C., assuming
2024 * 1 reboot per second, or we have a bug:
2025 */
2026 MD_BUG();
2027 mddev->events --;
2028 }
2029
2030 /*
2031 * do not write anything to disk if using
2032 * nonpersistent superblocks
2033 */
2034 if (!mddev->persistent) {
2035 if (!mddev->external)
2036 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2037
2038 spin_unlock_irq(&mddev->write_lock);
2039 wake_up(&mddev->sb_wait);
2040 return;
2041 }
2042 sync_sbs(mddev, nospares);
2043 spin_unlock_irq(&mddev->write_lock);
2044
2045 dprintk(KERN_INFO
2046 "md: updating %s RAID superblock on device (in sync %d)\n",
2047 mdname(mddev),mddev->in_sync);
2048
2049 bitmap_update_sb(mddev->bitmap);
2050 list_for_each_entry(rdev, &mddev->disks, same_set) {
2051 char b[BDEVNAME_SIZE];
2052 dprintk(KERN_INFO "md: ");
2053 if (rdev->sb_loaded != 1)
2054 continue; /* no noise on spare devices */
2055 if (test_bit(Faulty, &rdev->flags))
2056 dprintk("(skipping faulty ");
2057
2058 dprintk("%s ", bdevname(rdev->bdev,b));
2059 if (!test_bit(Faulty, &rdev->flags)) {
2060 md_super_write(mddev,rdev,
2061 rdev->sb_start, rdev->sb_size,
2062 rdev->sb_page);
2063 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2064 bdevname(rdev->bdev,b),
2065 (unsigned long long)rdev->sb_start);
2066 rdev->sb_events = mddev->events;
2067
2068 } else
2069 dprintk(")\n");
2070 if (mddev->level == LEVEL_MULTIPATH)
2071 /* only need to write one superblock... */
2072 break;
2073 }
2074 md_super_wait(mddev);
2075 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2076
2077 spin_lock_irq(&mddev->write_lock);
2078 if (mddev->in_sync != sync_req ||
2079 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2080 /* have to write it out again */
2081 spin_unlock_irq(&mddev->write_lock);
2082 goto repeat;
2083 }
2084 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2085 spin_unlock_irq(&mddev->write_lock);
2086 wake_up(&mddev->sb_wait);
2087 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2088 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2089
2090 }
2091
2092 /* words written to sysfs files may, or may not, be \n terminated.
2093 * We want to accept with case. For this we use cmd_match.
2094 */
2095 static int cmd_match(const char *cmd, const char *str)
2096 {
2097 /* See if cmd, written into a sysfs file, matches
2098 * str. They must either be the same, or cmd can
2099 * have a trailing newline
2100 */
2101 while (*cmd && *str && *cmd == *str) {
2102 cmd++;
2103 str++;
2104 }
2105 if (*cmd == '\n')
2106 cmd++;
2107 if (*str || *cmd)
2108 return 0;
2109 return 1;
2110 }
2111
2112 struct rdev_sysfs_entry {
2113 struct attribute attr;
2114 ssize_t (*show)(mdk_rdev_t *, char *);
2115 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2116 };
2117
2118 static ssize_t
2119 state_show(mdk_rdev_t *rdev, char *page)
2120 {
2121 char *sep = "";
2122 size_t len = 0;
2123
2124 if (test_bit(Faulty, &rdev->flags)) {
2125 len+= sprintf(page+len, "%sfaulty",sep);
2126 sep = ",";
2127 }
2128 if (test_bit(In_sync, &rdev->flags)) {
2129 len += sprintf(page+len, "%sin_sync",sep);
2130 sep = ",";
2131 }
2132 if (test_bit(WriteMostly, &rdev->flags)) {
2133 len += sprintf(page+len, "%swrite_mostly",sep);
2134 sep = ",";
2135 }
2136 if (test_bit(Blocked, &rdev->flags)) {
2137 len += sprintf(page+len, "%sblocked", sep);
2138 sep = ",";
2139 }
2140 if (!test_bit(Faulty, &rdev->flags) &&
2141 !test_bit(In_sync, &rdev->flags)) {
2142 len += sprintf(page+len, "%sspare", sep);
2143 sep = ",";
2144 }
2145 return len+sprintf(page+len, "\n");
2146 }
2147
2148 static ssize_t
2149 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2150 {
2151 /* can write
2152 * faulty - simulates and error
2153 * remove - disconnects the device
2154 * writemostly - sets write_mostly
2155 * -writemostly - clears write_mostly
2156 * blocked - sets the Blocked flag
2157 * -blocked - clears the Blocked flag
2158 * insync - sets Insync providing device isn't active
2159 */
2160 int err = -EINVAL;
2161 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2162 md_error(rdev->mddev, rdev);
2163 err = 0;
2164 } else if (cmd_match(buf, "remove")) {
2165 if (rdev->raid_disk >= 0)
2166 err = -EBUSY;
2167 else {
2168 mddev_t *mddev = rdev->mddev;
2169 kick_rdev_from_array(rdev);
2170 if (mddev->pers)
2171 md_update_sb(mddev, 1);
2172 md_new_event(mddev);
2173 err = 0;
2174 }
2175 } else if (cmd_match(buf, "writemostly")) {
2176 set_bit(WriteMostly, &rdev->flags);
2177 err = 0;
2178 } else if (cmd_match(buf, "-writemostly")) {
2179 clear_bit(WriteMostly, &rdev->flags);
2180 err = 0;
2181 } else if (cmd_match(buf, "blocked")) {
2182 set_bit(Blocked, &rdev->flags);
2183 err = 0;
2184 } else if (cmd_match(buf, "-blocked")) {
2185 clear_bit(Blocked, &rdev->flags);
2186 wake_up(&rdev->blocked_wait);
2187 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2188 md_wakeup_thread(rdev->mddev->thread);
2189
2190 err = 0;
2191 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2192 set_bit(In_sync, &rdev->flags);
2193 err = 0;
2194 }
2195 if (!err && rdev->sysfs_state)
2196 sysfs_notify_dirent(rdev->sysfs_state);
2197 return err ? err : len;
2198 }
2199 static struct rdev_sysfs_entry rdev_state =
2200 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2201
2202 static ssize_t
2203 errors_show(mdk_rdev_t *rdev, char *page)
2204 {
2205 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2206 }
2207
2208 static ssize_t
2209 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2210 {
2211 char *e;
2212 unsigned long n = simple_strtoul(buf, &e, 10);
2213 if (*buf && (*e == 0 || *e == '\n')) {
2214 atomic_set(&rdev->corrected_errors, n);
2215 return len;
2216 }
2217 return -EINVAL;
2218 }
2219 static struct rdev_sysfs_entry rdev_errors =
2220 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2221
2222 static ssize_t
2223 slot_show(mdk_rdev_t *rdev, char *page)
2224 {
2225 if (rdev->raid_disk < 0)
2226 return sprintf(page, "none\n");
2227 else
2228 return sprintf(page, "%d\n", rdev->raid_disk);
2229 }
2230
2231 static ssize_t
2232 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2233 {
2234 char *e;
2235 int err;
2236 char nm[20];
2237 int slot = simple_strtoul(buf, &e, 10);
2238 if (strncmp(buf, "none", 4)==0)
2239 slot = -1;
2240 else if (e==buf || (*e && *e!= '\n'))
2241 return -EINVAL;
2242 if (rdev->mddev->pers && slot == -1) {
2243 /* Setting 'slot' on an active array requires also
2244 * updating the 'rd%d' link, and communicating
2245 * with the personality with ->hot_*_disk.
2246 * For now we only support removing
2247 * failed/spare devices. This normally happens automatically,
2248 * but not when the metadata is externally managed.
2249 */
2250 if (rdev->raid_disk == -1)
2251 return -EEXIST;
2252 /* personality does all needed checks */
2253 if (rdev->mddev->pers->hot_add_disk == NULL)
2254 return -EINVAL;
2255 err = rdev->mddev->pers->
2256 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2257 if (err)
2258 return err;
2259 sprintf(nm, "rd%d", rdev->raid_disk);
2260 sysfs_remove_link(&rdev->mddev->kobj, nm);
2261 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2262 md_wakeup_thread(rdev->mddev->thread);
2263 } else if (rdev->mddev->pers) {
2264 mdk_rdev_t *rdev2;
2265 /* Activating a spare .. or possibly reactivating
2266 * if we ever get bitmaps working here.
2267 */
2268
2269 if (rdev->raid_disk != -1)
2270 return -EBUSY;
2271
2272 if (rdev->mddev->pers->hot_add_disk == NULL)
2273 return -EINVAL;
2274
2275 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2276 if (rdev2->raid_disk == slot)
2277 return -EEXIST;
2278
2279 rdev->raid_disk = slot;
2280 if (test_bit(In_sync, &rdev->flags))
2281 rdev->saved_raid_disk = slot;
2282 else
2283 rdev->saved_raid_disk = -1;
2284 err = rdev->mddev->pers->
2285 hot_add_disk(rdev->mddev, rdev);
2286 if (err) {
2287 rdev->raid_disk = -1;
2288 return err;
2289 } else
2290 sysfs_notify_dirent(rdev->sysfs_state);
2291 sprintf(nm, "rd%d", rdev->raid_disk);
2292 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2293 printk(KERN_WARNING
2294 "md: cannot register "
2295 "%s for %s\n",
2296 nm, mdname(rdev->mddev));
2297
2298 /* don't wakeup anyone, leave that to userspace. */
2299 } else {
2300 if (slot >= rdev->mddev->raid_disks)
2301 return -ENOSPC;
2302 rdev->raid_disk = slot;
2303 /* assume it is working */
2304 clear_bit(Faulty, &rdev->flags);
2305 clear_bit(WriteMostly, &rdev->flags);
2306 set_bit(In_sync, &rdev->flags);
2307 sysfs_notify_dirent(rdev->sysfs_state);
2308 }
2309 return len;
2310 }
2311
2312
2313 static struct rdev_sysfs_entry rdev_slot =
2314 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2315
2316 static ssize_t
2317 offset_show(mdk_rdev_t *rdev, char *page)
2318 {
2319 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2320 }
2321
2322 static ssize_t
2323 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2324 {
2325 char *e;
2326 unsigned long long offset = simple_strtoull(buf, &e, 10);
2327 if (e==buf || (*e && *e != '\n'))
2328 return -EINVAL;
2329 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2330 return -EBUSY;
2331 if (rdev->sectors && rdev->mddev->external)
2332 /* Must set offset before size, so overlap checks
2333 * can be sane */
2334 return -EBUSY;
2335 rdev->data_offset = offset;
2336 return len;
2337 }
2338
2339 static struct rdev_sysfs_entry rdev_offset =
2340 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2341
2342 static ssize_t
2343 rdev_size_show(mdk_rdev_t *rdev, char *page)
2344 {
2345 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2346 }
2347
2348 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2349 {
2350 /* check if two start/length pairs overlap */
2351 if (s1+l1 <= s2)
2352 return 0;
2353 if (s2+l2 <= s1)
2354 return 0;
2355 return 1;
2356 }
2357
2358 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2359 {
2360 unsigned long long blocks;
2361 sector_t new;
2362
2363 if (strict_strtoull(buf, 10, &blocks) < 0)
2364 return -EINVAL;
2365
2366 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2367 return -EINVAL; /* sector conversion overflow */
2368
2369 new = blocks * 2;
2370 if (new != blocks * 2)
2371 return -EINVAL; /* unsigned long long to sector_t overflow */
2372
2373 *sectors = new;
2374 return 0;
2375 }
2376
2377 static ssize_t
2378 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2379 {
2380 mddev_t *my_mddev = rdev->mddev;
2381 sector_t oldsectors = rdev->sectors;
2382 sector_t sectors;
2383
2384 if (strict_blocks_to_sectors(buf, &sectors) < 0)
2385 return -EINVAL;
2386 if (my_mddev->pers && rdev->raid_disk >= 0) {
2387 if (my_mddev->persistent) {
2388 sectors = super_types[my_mddev->major_version].
2389 rdev_size_change(rdev, sectors);
2390 if (!sectors)
2391 return -EBUSY;
2392 } else if (!sectors)
2393 sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2394 rdev->data_offset;
2395 }
2396 if (sectors < my_mddev->dev_sectors)
2397 return -EINVAL; /* component must fit device */
2398
2399 rdev->sectors = sectors;
2400 if (sectors > oldsectors && my_mddev->external) {
2401 /* need to check that all other rdevs with the same ->bdev
2402 * do not overlap. We need to unlock the mddev to avoid
2403 * a deadlock. We have already changed rdev->sectors, and if
2404 * we have to change it back, we will have the lock again.
2405 */
2406 mddev_t *mddev;
2407 int overlap = 0;
2408 struct list_head *tmp;
2409
2410 mddev_unlock(my_mddev);
2411 for_each_mddev(mddev, tmp) {
2412 mdk_rdev_t *rdev2;
2413
2414 mddev_lock(mddev);
2415 list_for_each_entry(rdev2, &mddev->disks, same_set)
2416 if (test_bit(AllReserved, &rdev2->flags) ||
2417 (rdev->bdev == rdev2->bdev &&
2418 rdev != rdev2 &&
2419 overlaps(rdev->data_offset, rdev->sectors,
2420 rdev2->data_offset,
2421 rdev2->sectors))) {
2422 overlap = 1;
2423 break;
2424 }
2425 mddev_unlock(mddev);
2426 if (overlap) {
2427 mddev_put(mddev);
2428 break;
2429 }
2430 }
2431 mddev_lock(my_mddev);
2432 if (overlap) {
2433 /* Someone else could have slipped in a size
2434 * change here, but doing so is just silly.
2435 * We put oldsectors back because we *know* it is
2436 * safe, and trust userspace not to race with
2437 * itself
2438 */
2439 rdev->sectors = oldsectors;
2440 return -EBUSY;
2441 }
2442 }
2443 return len;
2444 }
2445
2446 static struct rdev_sysfs_entry rdev_size =
2447 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2448
2449 static struct attribute *rdev_default_attrs[] = {
2450 &rdev_state.attr,
2451 &rdev_errors.attr,
2452 &rdev_slot.attr,
2453 &rdev_offset.attr,
2454 &rdev_size.attr,
2455 NULL,
2456 };
2457 static ssize_t
2458 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2459 {
2460 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2461 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2462 mddev_t *mddev = rdev->mddev;
2463 ssize_t rv;
2464
2465 if (!entry->show)
2466 return -EIO;
2467
2468 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2469 if (!rv) {
2470 if (rdev->mddev == NULL)
2471 rv = -EBUSY;
2472 else
2473 rv = entry->show(rdev, page);
2474 mddev_unlock(mddev);
2475 }
2476 return rv;
2477 }
2478
2479 static ssize_t
2480 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2481 const char *page, size_t length)
2482 {
2483 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2484 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2485 ssize_t rv;
2486 mddev_t *mddev = rdev->mddev;
2487
2488 if (!entry->store)
2489 return -EIO;
2490 if (!capable(CAP_SYS_ADMIN))
2491 return -EACCES;
2492 rv = mddev ? mddev_lock(mddev): -EBUSY;
2493 if (!rv) {
2494 if (rdev->mddev == NULL)
2495 rv = -EBUSY;
2496 else
2497 rv = entry->store(rdev, page, length);
2498 mddev_unlock(mddev);
2499 }
2500 return rv;
2501 }
2502
2503 static void rdev_free(struct kobject *ko)
2504 {
2505 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2506 kfree(rdev);
2507 }
2508 static struct sysfs_ops rdev_sysfs_ops = {
2509 .show = rdev_attr_show,
2510 .store = rdev_attr_store,
2511 };
2512 static struct kobj_type rdev_ktype = {
2513 .release = rdev_free,
2514 .sysfs_ops = &rdev_sysfs_ops,
2515 .default_attrs = rdev_default_attrs,
2516 };
2517
2518 /*
2519 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2520 *
2521 * mark the device faulty if:
2522 *
2523 * - the device is nonexistent (zero size)
2524 * - the device has no valid superblock
2525 *
2526 * a faulty rdev _never_ has rdev->sb set.
2527 */
2528 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2529 {
2530 char b[BDEVNAME_SIZE];
2531 int err;
2532 mdk_rdev_t *rdev;
2533 sector_t size;
2534
2535 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2536 if (!rdev) {
2537 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2538 return ERR_PTR(-ENOMEM);
2539 }
2540
2541 if ((err = alloc_disk_sb(rdev)))
2542 goto abort_free;
2543
2544 err = lock_rdev(rdev, newdev, super_format == -2);
2545 if (err)
2546 goto abort_free;
2547
2548 kobject_init(&rdev->kobj, &rdev_ktype);
2549
2550 rdev->desc_nr = -1;
2551 rdev->saved_raid_disk = -1;
2552 rdev->raid_disk = -1;
2553 rdev->flags = 0;
2554 rdev->data_offset = 0;
2555 rdev->sb_events = 0;
2556 atomic_set(&rdev->nr_pending, 0);
2557 atomic_set(&rdev->read_errors, 0);
2558 atomic_set(&rdev->corrected_errors, 0);
2559
2560 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2561 if (!size) {
2562 printk(KERN_WARNING
2563 "md: %s has zero or unknown size, marking faulty!\n",
2564 bdevname(rdev->bdev,b));
2565 err = -EINVAL;
2566 goto abort_free;
2567 }
2568
2569 if (super_format >= 0) {
2570 err = super_types[super_format].
2571 load_super(rdev, NULL, super_minor);
2572 if (err == -EINVAL) {
2573 printk(KERN_WARNING
2574 "md: %s does not have a valid v%d.%d "
2575 "superblock, not importing!\n",
2576 bdevname(rdev->bdev,b),
2577 super_format, super_minor);
2578 goto abort_free;
2579 }
2580 if (err < 0) {
2581 printk(KERN_WARNING
2582 "md: could not read %s's sb, not importing!\n",
2583 bdevname(rdev->bdev,b));
2584 goto abort_free;
2585 }
2586 }
2587
2588 INIT_LIST_HEAD(&rdev->same_set);
2589 init_waitqueue_head(&rdev->blocked_wait);
2590
2591 return rdev;
2592
2593 abort_free:
2594 if (rdev->sb_page) {
2595 if (rdev->bdev)
2596 unlock_rdev(rdev);
2597 free_disk_sb(rdev);
2598 }
2599 kfree(rdev);
2600 return ERR_PTR(err);
2601 }
2602
2603 /*
2604 * Check a full RAID array for plausibility
2605 */
2606
2607
2608 static void analyze_sbs(mddev_t * mddev)
2609 {
2610 int i;
2611 mdk_rdev_t *rdev, *freshest, *tmp;
2612 char b[BDEVNAME_SIZE];
2613
2614 freshest = NULL;
2615 rdev_for_each(rdev, tmp, mddev)
2616 switch (super_types[mddev->major_version].
2617 load_super(rdev, freshest, mddev->minor_version)) {
2618 case 1:
2619 freshest = rdev;
2620 break;
2621 case 0:
2622 break;
2623 default:
2624 printk( KERN_ERR \
2625 "md: fatal superblock inconsistency in %s"
2626 " -- removing from array\n",
2627 bdevname(rdev->bdev,b));
2628 kick_rdev_from_array(rdev);
2629 }
2630
2631
2632 super_types[mddev->major_version].
2633 validate_super(mddev, freshest);
2634
2635 i = 0;
2636 rdev_for_each(rdev, tmp, mddev) {
2637 if (rdev->desc_nr >= mddev->max_disks ||
2638 i > mddev->max_disks) {
2639 printk(KERN_WARNING
2640 "md: %s: %s: only %d devices permitted\n",
2641 mdname(mddev), bdevname(rdev->bdev, b),
2642 mddev->max_disks);
2643 kick_rdev_from_array(rdev);
2644 continue;
2645 }
2646 if (rdev != freshest)
2647 if (super_types[mddev->major_version].
2648 validate_super(mddev, rdev)) {
2649 printk(KERN_WARNING "md: kicking non-fresh %s"
2650 " from array!\n",
2651 bdevname(rdev->bdev,b));
2652 kick_rdev_from_array(rdev);
2653 continue;
2654 }
2655 if (mddev->level == LEVEL_MULTIPATH) {
2656 rdev->desc_nr = i++;
2657 rdev->raid_disk = rdev->desc_nr;
2658 set_bit(In_sync, &rdev->flags);
2659 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
2660 rdev->raid_disk = -1;
2661 clear_bit(In_sync, &rdev->flags);
2662 }
2663 }
2664 }
2665
2666 static void md_safemode_timeout(unsigned long data);
2667
2668 static ssize_t
2669 safe_delay_show(mddev_t *mddev, char *page)
2670 {
2671 int msec = (mddev->safemode_delay*1000)/HZ;
2672 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2673 }
2674 static ssize_t
2675 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2676 {
2677 int scale=1;
2678 int dot=0;
2679 int i;
2680 unsigned long msec;
2681 char buf[30];
2682
2683 /* remove a period, and count digits after it */
2684 if (len >= sizeof(buf))
2685 return -EINVAL;
2686 strlcpy(buf, cbuf, sizeof(buf));
2687 for (i=0; i<len; i++) {
2688 if (dot) {
2689 if (isdigit(buf[i])) {
2690 buf[i-1] = buf[i];
2691 scale *= 10;
2692 }
2693 buf[i] = 0;
2694 } else if (buf[i] == '.') {
2695 dot=1;
2696 buf[i] = 0;
2697 }
2698 }
2699 if (strict_strtoul(buf, 10, &msec) < 0)
2700 return -EINVAL;
2701 msec = (msec * 1000) / scale;
2702 if (msec == 0)
2703 mddev->safemode_delay = 0;
2704 else {
2705 unsigned long old_delay = mddev->safemode_delay;
2706 mddev->safemode_delay = (msec*HZ)/1000;
2707 if (mddev->safemode_delay == 0)
2708 mddev->safemode_delay = 1;
2709 if (mddev->safemode_delay < old_delay)
2710 md_safemode_timeout((unsigned long)mddev);
2711 }
2712 return len;
2713 }
2714 static struct md_sysfs_entry md_safe_delay =
2715 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2716
2717 static ssize_t
2718 level_show(mddev_t *mddev, char *page)
2719 {
2720 struct mdk_personality *p = mddev->pers;
2721 if (p)
2722 return sprintf(page, "%s\n", p->name);
2723 else if (mddev->clevel[0])
2724 return sprintf(page, "%s\n", mddev->clevel);
2725 else if (mddev->level != LEVEL_NONE)
2726 return sprintf(page, "%d\n", mddev->level);
2727 else
2728 return 0;
2729 }
2730
2731 static ssize_t
2732 level_store(mddev_t *mddev, const char *buf, size_t len)
2733 {
2734 char level[16];
2735 ssize_t rv = len;
2736 struct mdk_personality *pers;
2737 void *priv;
2738 mdk_rdev_t *rdev;
2739
2740 if (mddev->pers == NULL) {
2741 if (len == 0)
2742 return 0;
2743 if (len >= sizeof(mddev->clevel))
2744 return -ENOSPC;
2745 strncpy(mddev->clevel, buf, len);
2746 if (mddev->clevel[len-1] == '\n')
2747 len--;
2748 mddev->clevel[len] = 0;
2749 mddev->level = LEVEL_NONE;
2750 return rv;
2751 }
2752
2753 /* request to change the personality. Need to ensure:
2754 * - array is not engaged in resync/recovery/reshape
2755 * - old personality can be suspended
2756 * - new personality will access other array.
2757 */
2758
2759 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
2760 return -EBUSY;
2761
2762 if (!mddev->pers->quiesce) {
2763 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
2764 mdname(mddev), mddev->pers->name);
2765 return -EINVAL;
2766 }
2767
2768 /* Now find the new personality */
2769 if (len == 0 || len >= sizeof(level))
2770 return -EINVAL;
2771 strncpy(level, buf, len);
2772 if (level[len-1] == '\n')
2773 len--;
2774 level[len] = 0;
2775
2776 request_module("md-%s", level);
2777 spin_lock(&pers_lock);
2778 pers = find_pers(LEVEL_NONE, level);
2779 if (!pers || !try_module_get(pers->owner)) {
2780 spin_unlock(&pers_lock);
2781 printk(KERN_WARNING "md: personality %s not loaded\n", level);
2782 return -EINVAL;
2783 }
2784 spin_unlock(&pers_lock);
2785
2786 if (pers == mddev->pers) {
2787 /* Nothing to do! */
2788 module_put(pers->owner);
2789 return rv;
2790 }
2791 if (!pers->takeover) {
2792 module_put(pers->owner);
2793 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
2794 mdname(mddev), level);
2795 return -EINVAL;
2796 }
2797
2798 /* ->takeover must set new_* and/or delta_disks
2799 * if it succeeds, and may set them when it fails.
2800 */
2801 priv = pers->takeover(mddev);
2802 if (IS_ERR(priv)) {
2803 mddev->new_level = mddev->level;
2804 mddev->new_layout = mddev->layout;
2805 mddev->new_chunk_sectors = mddev->chunk_sectors;
2806 mddev->raid_disks -= mddev->delta_disks;
2807 mddev->delta_disks = 0;
2808 module_put(pers->owner);
2809 printk(KERN_WARNING "md: %s: %s would not accept array\n",
2810 mdname(mddev), level);
2811 return PTR_ERR(priv);
2812 }
2813
2814 /* Looks like we have a winner */
2815 mddev_suspend(mddev);
2816 mddev->pers->stop(mddev);
2817 module_put(mddev->pers->owner);
2818 /* Invalidate devices that are now superfluous */
2819 list_for_each_entry(rdev, &mddev->disks, same_set)
2820 if (rdev->raid_disk >= mddev->raid_disks) {
2821 rdev->raid_disk = -1;
2822 clear_bit(In_sync, &rdev->flags);
2823 }
2824 mddev->pers = pers;
2825 mddev->private = priv;
2826 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
2827 mddev->level = mddev->new_level;
2828 mddev->layout = mddev->new_layout;
2829 mddev->chunk_sectors = mddev->new_chunk_sectors;
2830 mddev->delta_disks = 0;
2831 pers->run(mddev);
2832 mddev_resume(mddev);
2833 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2834 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2835 md_wakeup_thread(mddev->thread);
2836 return rv;
2837 }
2838
2839 static struct md_sysfs_entry md_level =
2840 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2841
2842
2843 static ssize_t
2844 layout_show(mddev_t *mddev, char *page)
2845 {
2846 /* just a number, not meaningful for all levels */
2847 if (mddev->reshape_position != MaxSector &&
2848 mddev->layout != mddev->new_layout)
2849 return sprintf(page, "%d (%d)\n",
2850 mddev->new_layout, mddev->layout);
2851 return sprintf(page, "%d\n", mddev->layout);
2852 }
2853
2854 static ssize_t
2855 layout_store(mddev_t *mddev, const char *buf, size_t len)
2856 {
2857 char *e;
2858 unsigned long n = simple_strtoul(buf, &e, 10);
2859
2860 if (!*buf || (*e && *e != '\n'))
2861 return -EINVAL;
2862
2863 if (mddev->pers) {
2864 int err;
2865 if (mddev->pers->check_reshape == NULL)
2866 return -EBUSY;
2867 mddev->new_layout = n;
2868 err = mddev->pers->check_reshape(mddev);
2869 if (err) {
2870 mddev->new_layout = mddev->layout;
2871 return err;
2872 }
2873 } else {
2874 mddev->new_layout = n;
2875 if (mddev->reshape_position == MaxSector)
2876 mddev->layout = n;
2877 }
2878 return len;
2879 }
2880 static struct md_sysfs_entry md_layout =
2881 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2882
2883
2884 static ssize_t
2885 raid_disks_show(mddev_t *mddev, char *page)
2886 {
2887 if (mddev->raid_disks == 0)
2888 return 0;
2889 if (mddev->reshape_position != MaxSector &&
2890 mddev->delta_disks != 0)
2891 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2892 mddev->raid_disks - mddev->delta_disks);
2893 return sprintf(page, "%d\n", mddev->raid_disks);
2894 }
2895
2896 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2897
2898 static ssize_t
2899 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2900 {
2901 char *e;
2902 int rv = 0;
2903 unsigned long n = simple_strtoul(buf, &e, 10);
2904
2905 if (!*buf || (*e && *e != '\n'))
2906 return -EINVAL;
2907
2908 if (mddev->pers)
2909 rv = update_raid_disks(mddev, n);
2910 else if (mddev->reshape_position != MaxSector) {
2911 int olddisks = mddev->raid_disks - mddev->delta_disks;
2912 mddev->delta_disks = n - olddisks;
2913 mddev->raid_disks = n;
2914 } else
2915 mddev->raid_disks = n;
2916 return rv ? rv : len;
2917 }
2918 static struct md_sysfs_entry md_raid_disks =
2919 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2920
2921 static ssize_t
2922 chunk_size_show(mddev_t *mddev, char *page)
2923 {
2924 if (mddev->reshape_position != MaxSector &&
2925 mddev->chunk_sectors != mddev->new_chunk_sectors)
2926 return sprintf(page, "%d (%d)\n",
2927 mddev->new_chunk_sectors << 9,
2928 mddev->chunk_sectors << 9);
2929 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
2930 }
2931
2932 static ssize_t
2933 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2934 {
2935 char *e;
2936 unsigned long n = simple_strtoul(buf, &e, 10);
2937
2938 if (!*buf || (*e && *e != '\n'))
2939 return -EINVAL;
2940
2941 if (mddev->pers) {
2942 int err;
2943 if (mddev->pers->check_reshape == NULL)
2944 return -EBUSY;
2945 mddev->new_chunk_sectors = n >> 9;
2946 err = mddev->pers->check_reshape(mddev);
2947 if (err) {
2948 mddev->new_chunk_sectors = mddev->chunk_sectors;
2949 return err;
2950 }
2951 } else {
2952 mddev->new_chunk_sectors = n >> 9;
2953 if (mddev->reshape_position == MaxSector)
2954 mddev->chunk_sectors = n >> 9;
2955 }
2956 return len;
2957 }
2958 static struct md_sysfs_entry md_chunk_size =
2959 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2960
2961 static ssize_t
2962 resync_start_show(mddev_t *mddev, char *page)
2963 {
2964 if (mddev->recovery_cp == MaxSector)
2965 return sprintf(page, "none\n");
2966 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2967 }
2968
2969 static ssize_t
2970 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2971 {
2972 char *e;
2973 unsigned long long n = simple_strtoull(buf, &e, 10);
2974
2975 if (mddev->pers)
2976 return -EBUSY;
2977 if (!*buf || (*e && *e != '\n'))
2978 return -EINVAL;
2979
2980 mddev->recovery_cp = n;
2981 return len;
2982 }
2983 static struct md_sysfs_entry md_resync_start =
2984 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2985
2986 /*
2987 * The array state can be:
2988 *
2989 * clear
2990 * No devices, no size, no level
2991 * Equivalent to STOP_ARRAY ioctl
2992 * inactive
2993 * May have some settings, but array is not active
2994 * all IO results in error
2995 * When written, doesn't tear down array, but just stops it
2996 * suspended (not supported yet)
2997 * All IO requests will block. The array can be reconfigured.
2998 * Writing this, if accepted, will block until array is quiescent
2999 * readonly
3000 * no resync can happen. no superblocks get written.
3001 * write requests fail
3002 * read-auto
3003 * like readonly, but behaves like 'clean' on a write request.
3004 *
3005 * clean - no pending writes, but otherwise active.
3006 * When written to inactive array, starts without resync
3007 * If a write request arrives then
3008 * if metadata is known, mark 'dirty' and switch to 'active'.
3009 * if not known, block and switch to write-pending
3010 * If written to an active array that has pending writes, then fails.
3011 * active
3012 * fully active: IO and resync can be happening.
3013 * When written to inactive array, starts with resync
3014 *
3015 * write-pending
3016 * clean, but writes are blocked waiting for 'active' to be written.
3017 *
3018 * active-idle
3019 * like active, but no writes have been seen for a while (100msec).
3020 *
3021 */
3022 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3023 write_pending, active_idle, bad_word};
3024 static char *array_states[] = {
3025 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3026 "write-pending", "active-idle", NULL };
3027
3028 static int match_word(const char *word, char **list)
3029 {
3030 int n;
3031 for (n=0; list[n]; n++)
3032 if (cmd_match(word, list[n]))
3033 break;
3034 return n;
3035 }
3036
3037 static ssize_t
3038 array_state_show(mddev_t *mddev, char *page)
3039 {
3040 enum array_state st = inactive;
3041
3042 if (mddev->pers)
3043 switch(mddev->ro) {
3044 case 1:
3045 st = readonly;
3046 break;
3047 case 2:
3048 st = read_auto;
3049 break;
3050 case 0:
3051 if (mddev->in_sync)
3052 st = clean;
3053 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
3054 st = write_pending;
3055 else if (mddev->safemode)
3056 st = active_idle;
3057 else
3058 st = active;
3059 }
3060 else {
3061 if (list_empty(&mddev->disks) &&
3062 mddev->raid_disks == 0 &&
3063 mddev->dev_sectors == 0)
3064 st = clear;
3065 else
3066 st = inactive;
3067 }
3068 return sprintf(page, "%s\n", array_states[st]);
3069 }
3070
3071 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3072 static int do_md_run(mddev_t * mddev);
3073 static int restart_array(mddev_t *mddev);
3074
3075 static ssize_t
3076 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3077 {
3078 int err = -EINVAL;
3079 enum array_state st = match_word(buf, array_states);
3080 switch(st) {
3081 case bad_word:
3082 break;
3083 case clear:
3084 /* stopping an active array */
3085 if (atomic_read(&mddev->openers) > 0)
3086 return -EBUSY;
3087 err = do_md_stop(mddev, 0, 0);
3088 break;
3089 case inactive:
3090 /* stopping an active array */
3091 if (mddev->pers) {
3092 if (atomic_read(&mddev->openers) > 0)
3093 return -EBUSY;
3094 err = do_md_stop(mddev, 2, 0);
3095 } else
3096 err = 0; /* already inactive */
3097 break;
3098 case suspended:
3099 break; /* not supported yet */
3100 case readonly:
3101 if (mddev->pers)
3102 err = do_md_stop(mddev, 1, 0);
3103 else {
3104 mddev->ro = 1;
3105 set_disk_ro(mddev->gendisk, 1);
3106 err = do_md_run(mddev);
3107 }
3108 break;
3109 case read_auto:
3110 if (mddev->pers) {
3111 if (mddev->ro == 0)
3112 err = do_md_stop(mddev, 1, 0);
3113 else if (mddev->ro == 1)
3114 err = restart_array(mddev);
3115 if (err == 0) {
3116 mddev->ro = 2;
3117 set_disk_ro(mddev->gendisk, 0);
3118 }
3119 } else {
3120 mddev->ro = 2;
3121 err = do_md_run(mddev);
3122 }
3123 break;
3124 case clean:
3125 if (mddev->pers) {
3126 restart_array(mddev);
3127 spin_lock_irq(&mddev->write_lock);
3128 if (atomic_read(&mddev->writes_pending) == 0) {
3129 if (mddev->in_sync == 0) {
3130 mddev->in_sync = 1;
3131 if (mddev->safemode == 1)
3132 mddev->safemode = 0;
3133 if (mddev->persistent)
3134 set_bit(MD_CHANGE_CLEAN,
3135 &mddev->flags);
3136 }
3137 err = 0;
3138 } else
3139 err = -EBUSY;
3140 spin_unlock_irq(&mddev->write_lock);
3141 } else
3142 err = -EINVAL;
3143 break;
3144 case active:
3145 if (mddev->pers) {
3146 restart_array(mddev);
3147 if (mddev->external)
3148 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3149 wake_up(&mddev->sb_wait);
3150 err = 0;
3151 } else {
3152 mddev->ro = 0;
3153 set_disk_ro(mddev->gendisk, 0);
3154 err = do_md_run(mddev);
3155 }
3156 break;
3157 case write_pending:
3158 case active_idle:
3159 /* these cannot be set */
3160 break;
3161 }
3162 if (err)
3163 return err;
3164 else {
3165 sysfs_notify_dirent(mddev->sysfs_state);
3166 return len;
3167 }
3168 }
3169 static struct md_sysfs_entry md_array_state =
3170 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3171
3172 static ssize_t
3173 null_show(mddev_t *mddev, char *page)
3174 {
3175 return -EINVAL;
3176 }
3177
3178 static ssize_t
3179 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3180 {
3181 /* buf must be %d:%d\n? giving major and minor numbers */
3182 /* The new device is added to the array.
3183 * If the array has a persistent superblock, we read the
3184 * superblock to initialise info and check validity.
3185 * Otherwise, only checking done is that in bind_rdev_to_array,
3186 * which mainly checks size.
3187 */
3188 char *e;
3189 int major = simple_strtoul(buf, &e, 10);
3190 int minor;
3191 dev_t dev;
3192 mdk_rdev_t *rdev;
3193 int err;
3194
3195 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3196 return -EINVAL;
3197 minor = simple_strtoul(e+1, &e, 10);
3198 if (*e && *e != '\n')
3199 return -EINVAL;
3200 dev = MKDEV(major, minor);
3201 if (major != MAJOR(dev) ||
3202 minor != MINOR(dev))
3203 return -EOVERFLOW;
3204
3205
3206 if (mddev->persistent) {
3207 rdev = md_import_device(dev, mddev->major_version,
3208 mddev->minor_version);
3209 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3210 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3211 mdk_rdev_t, same_set);
3212 err = super_types[mddev->major_version]
3213 .load_super(rdev, rdev0, mddev->minor_version);
3214 if (err < 0)
3215 goto out;
3216 }
3217 } else if (mddev->external)
3218 rdev = md_import_device(dev, -2, -1);
3219 else
3220 rdev = md_import_device(dev, -1, -1);
3221
3222 if (IS_ERR(rdev))
3223 return PTR_ERR(rdev);
3224 err = bind_rdev_to_array(rdev, mddev);
3225 out:
3226 if (err)
3227 export_rdev(rdev);
3228 return err ? err : len;
3229 }
3230
3231 static struct md_sysfs_entry md_new_device =
3232 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3233
3234 static ssize_t
3235 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3236 {
3237 char *end;
3238 unsigned long chunk, end_chunk;
3239
3240 if (!mddev->bitmap)
3241 goto out;
3242 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3243 while (*buf) {
3244 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3245 if (buf == end) break;
3246 if (*end == '-') { /* range */
3247 buf = end + 1;
3248 end_chunk = simple_strtoul(buf, &end, 0);
3249 if (buf == end) break;
3250 }
3251 if (*end && !isspace(*end)) break;
3252 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3253 buf = end;
3254 while (isspace(*buf)) buf++;
3255 }
3256 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3257 out:
3258 return len;
3259 }
3260
3261 static struct md_sysfs_entry md_bitmap =
3262 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3263
3264 static ssize_t
3265 size_show(mddev_t *mddev, char *page)
3266 {
3267 return sprintf(page, "%llu\n",
3268 (unsigned long long)mddev->dev_sectors / 2);
3269 }
3270
3271 static int update_size(mddev_t *mddev, sector_t num_sectors);
3272
3273 static ssize_t
3274 size_store(mddev_t *mddev, const char *buf, size_t len)
3275 {
3276 /* If array is inactive, we can reduce the component size, but
3277 * not increase it (except from 0).
3278 * If array is active, we can try an on-line resize
3279 */
3280 sector_t sectors;
3281 int err = strict_blocks_to_sectors(buf, &sectors);
3282
3283 if (err < 0)
3284 return err;
3285 if (mddev->pers) {
3286 err = update_size(mddev, sectors);
3287 md_update_sb(mddev, 1);
3288 } else {
3289 if (mddev->dev_sectors == 0 ||
3290 mddev->dev_sectors > sectors)
3291 mddev->dev_sectors = sectors;
3292 else
3293 err = -ENOSPC;
3294 }
3295 return err ? err : len;
3296 }
3297
3298 static struct md_sysfs_entry md_size =
3299 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3300
3301
3302 /* Metdata version.
3303 * This is one of
3304 * 'none' for arrays with no metadata (good luck...)
3305 * 'external' for arrays with externally managed metadata,
3306 * or N.M for internally known formats
3307 */
3308 static ssize_t
3309 metadata_show(mddev_t *mddev, char *page)
3310 {
3311 if (mddev->persistent)
3312 return sprintf(page, "%d.%d\n",
3313 mddev->major_version, mddev->minor_version);
3314 else if (mddev->external)
3315 return sprintf(page, "external:%s\n", mddev->metadata_type);
3316 else
3317 return sprintf(page, "none\n");
3318 }
3319
3320 static ssize_t
3321 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3322 {
3323 int major, minor;
3324 char *e;
3325 /* Changing the details of 'external' metadata is
3326 * always permitted. Otherwise there must be
3327 * no devices attached to the array.
3328 */
3329 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3330 ;
3331 else if (!list_empty(&mddev->disks))
3332 return -EBUSY;
3333
3334 if (cmd_match(buf, "none")) {
3335 mddev->persistent = 0;
3336 mddev->external = 0;
3337 mddev->major_version = 0;
3338 mddev->minor_version = 90;
3339 return len;
3340 }
3341 if (strncmp(buf, "external:", 9) == 0) {
3342 size_t namelen = len-9;
3343 if (namelen >= sizeof(mddev->metadata_type))
3344 namelen = sizeof(mddev->metadata_type)-1;
3345 strncpy(mddev->metadata_type, buf+9, namelen);
3346 mddev->metadata_type[namelen] = 0;
3347 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3348 mddev->metadata_type[--namelen] = 0;
3349 mddev->persistent = 0;
3350 mddev->external = 1;
3351 mddev->major_version = 0;
3352 mddev->minor_version = 90;
3353 return len;
3354 }
3355 major = simple_strtoul(buf, &e, 10);
3356 if (e==buf || *e != '.')
3357 return -EINVAL;
3358 buf = e+1;
3359 minor = simple_strtoul(buf, &e, 10);
3360 if (e==buf || (*e && *e != '\n') )
3361 return -EINVAL;
3362 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3363 return -ENOENT;
3364 mddev->major_version = major;
3365 mddev->minor_version = minor;
3366 mddev->persistent = 1;
3367 mddev->external = 0;
3368 return len;
3369 }
3370
3371 static struct md_sysfs_entry md_metadata =
3372 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3373
3374 static ssize_t
3375 action_show(mddev_t *mddev, char *page)
3376 {
3377 char *type = "idle";
3378 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3379 type = "frozen";
3380 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3381 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3382 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3383 type = "reshape";
3384 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3385 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3386 type = "resync";
3387 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3388 type = "check";
3389 else
3390 type = "repair";
3391 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3392 type = "recover";
3393 }
3394 return sprintf(page, "%s\n", type);
3395 }
3396
3397 static ssize_t
3398 action_store(mddev_t *mddev, const char *page, size_t len)
3399 {
3400 if (!mddev->pers || !mddev->pers->sync_request)
3401 return -EINVAL;
3402
3403 if (cmd_match(page, "frozen"))
3404 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3405 else
3406 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3407
3408 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3409 if (mddev->sync_thread) {
3410 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3411 md_unregister_thread(mddev->sync_thread);
3412 mddev->sync_thread = NULL;
3413 mddev->recovery = 0;
3414 }
3415 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3416 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3417 return -EBUSY;
3418 else if (cmd_match(page, "resync"))
3419 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3420 else if (cmd_match(page, "recover")) {
3421 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3422 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3423 } else if (cmd_match(page, "reshape")) {
3424 int err;
3425 if (mddev->pers->start_reshape == NULL)
3426 return -EINVAL;
3427 err = mddev->pers->start_reshape(mddev);
3428 if (err)
3429 return err;
3430 sysfs_notify(&mddev->kobj, NULL, "degraded");
3431 } else {
3432 if (cmd_match(page, "check"))
3433 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3434 else if (!cmd_match(page, "repair"))
3435 return -EINVAL;
3436 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3437 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3438 }
3439 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3440 md_wakeup_thread(mddev->thread);
3441 sysfs_notify_dirent(mddev->sysfs_action);
3442 return len;
3443 }
3444
3445 static ssize_t
3446 mismatch_cnt_show(mddev_t *mddev, char *page)
3447 {
3448 return sprintf(page, "%llu\n",
3449 (unsigned long long) mddev->resync_mismatches);
3450 }
3451
3452 static struct md_sysfs_entry md_scan_mode =
3453 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3454
3455
3456 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3457
3458 static ssize_t
3459 sync_min_show(mddev_t *mddev, char *page)
3460 {
3461 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3462 mddev->sync_speed_min ? "local": "system");
3463 }
3464
3465 static ssize_t
3466 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3467 {
3468 int min;
3469 char *e;
3470 if (strncmp(buf, "system", 6)==0) {
3471 mddev->sync_speed_min = 0;
3472 return len;
3473 }
3474 min = simple_strtoul(buf, &e, 10);
3475 if (buf == e || (*e && *e != '\n') || min <= 0)
3476 return -EINVAL;
3477 mddev->sync_speed_min = min;
3478 return len;
3479 }
3480
3481 static struct md_sysfs_entry md_sync_min =
3482 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3483
3484 static ssize_t
3485 sync_max_show(mddev_t *mddev, char *page)
3486 {
3487 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3488 mddev->sync_speed_max ? "local": "system");
3489 }
3490
3491 static ssize_t
3492 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3493 {
3494 int max;
3495 char *e;
3496 if (strncmp(buf, "system", 6)==0) {
3497 mddev->sync_speed_max = 0;
3498 return len;
3499 }
3500 max = simple_strtoul(buf, &e, 10);
3501 if (buf == e || (*e && *e != '\n') || max <= 0)
3502 return -EINVAL;
3503 mddev->sync_speed_max = max;
3504 return len;
3505 }
3506
3507 static struct md_sysfs_entry md_sync_max =
3508 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3509
3510 static ssize_t
3511 degraded_show(mddev_t *mddev, char *page)
3512 {
3513 return sprintf(page, "%d\n", mddev->degraded);
3514 }
3515 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3516
3517 static ssize_t
3518 sync_force_parallel_show(mddev_t *mddev, char *page)
3519 {
3520 return sprintf(page, "%d\n", mddev->parallel_resync);
3521 }
3522
3523 static ssize_t
3524 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3525 {
3526 long n;
3527
3528 if (strict_strtol(buf, 10, &n))
3529 return -EINVAL;
3530
3531 if (n != 0 && n != 1)
3532 return -EINVAL;
3533
3534 mddev->parallel_resync = n;
3535
3536 if (mddev->sync_thread)
3537 wake_up(&resync_wait);
3538
3539 return len;
3540 }
3541
3542 /* force parallel resync, even with shared block devices */
3543 static struct md_sysfs_entry md_sync_force_parallel =
3544 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3545 sync_force_parallel_show, sync_force_parallel_store);
3546
3547 static ssize_t
3548 sync_speed_show(mddev_t *mddev, char *page)
3549 {
3550 unsigned long resync, dt, db;
3551 if (mddev->curr_resync == 0)
3552 return sprintf(page, "none\n");
3553 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3554 dt = (jiffies - mddev->resync_mark) / HZ;
3555 if (!dt) dt++;
3556 db = resync - mddev->resync_mark_cnt;
3557 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3558 }
3559
3560 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3561
3562 static ssize_t
3563 sync_completed_show(mddev_t *mddev, char *page)
3564 {
3565 unsigned long max_sectors, resync;
3566
3567 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3568 return sprintf(page, "none\n");
3569
3570 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3571 max_sectors = mddev->resync_max_sectors;
3572 else
3573 max_sectors = mddev->dev_sectors;
3574
3575 resync = mddev->curr_resync_completed;
3576 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3577 }
3578
3579 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3580
3581 static ssize_t
3582 min_sync_show(mddev_t *mddev, char *page)
3583 {
3584 return sprintf(page, "%llu\n",
3585 (unsigned long long)mddev->resync_min);
3586 }
3587 static ssize_t
3588 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3589 {
3590 unsigned long long min;
3591 if (strict_strtoull(buf, 10, &min))
3592 return -EINVAL;
3593 if (min > mddev->resync_max)
3594 return -EINVAL;
3595 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3596 return -EBUSY;
3597
3598 /* Must be a multiple of chunk_size */
3599 if (mddev->chunk_sectors) {
3600 sector_t temp = min;
3601 if (sector_div(temp, mddev->chunk_sectors))
3602 return -EINVAL;
3603 }
3604 mddev->resync_min = min;
3605
3606 return len;
3607 }
3608
3609 static struct md_sysfs_entry md_min_sync =
3610 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3611
3612 static ssize_t
3613 max_sync_show(mddev_t *mddev, char *page)
3614 {
3615 if (mddev->resync_max == MaxSector)
3616 return sprintf(page, "max\n");
3617 else
3618 return sprintf(page, "%llu\n",
3619 (unsigned long long)mddev->resync_max);
3620 }
3621 static ssize_t
3622 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3623 {
3624 if (strncmp(buf, "max", 3) == 0)
3625 mddev->resync_max = MaxSector;
3626 else {
3627 unsigned long long max;
3628 if (strict_strtoull(buf, 10, &max))
3629 return -EINVAL;
3630 if (max < mddev->resync_min)
3631 return -EINVAL;
3632 if (max < mddev->resync_max &&
3633 mddev->ro == 0 &&
3634 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3635 return -EBUSY;
3636
3637 /* Must be a multiple of chunk_size */
3638 if (mddev->chunk_sectors) {
3639 sector_t temp = max;
3640 if (sector_div(temp, mddev->chunk_sectors))
3641 return -EINVAL;
3642 }
3643 mddev->resync_max = max;
3644 }
3645 wake_up(&mddev->recovery_wait);
3646 return len;
3647 }
3648
3649 static struct md_sysfs_entry md_max_sync =
3650 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3651
3652 static ssize_t
3653 suspend_lo_show(mddev_t *mddev, char *page)
3654 {
3655 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3656 }
3657
3658 static ssize_t
3659 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3660 {
3661 char *e;
3662 unsigned long long new = simple_strtoull(buf, &e, 10);
3663
3664 if (mddev->pers == NULL ||
3665 mddev->pers->quiesce == NULL)
3666 return -EINVAL;
3667 if (buf == e || (*e && *e != '\n'))
3668 return -EINVAL;
3669 if (new >= mddev->suspend_hi ||
3670 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3671 mddev->suspend_lo = new;
3672 mddev->pers->quiesce(mddev, 2);
3673 return len;
3674 } else
3675 return -EINVAL;
3676 }
3677 static struct md_sysfs_entry md_suspend_lo =
3678 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3679
3680
3681 static ssize_t
3682 suspend_hi_show(mddev_t *mddev, char *page)
3683 {
3684 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3685 }
3686
3687 static ssize_t
3688 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3689 {
3690 char *e;
3691 unsigned long long new = simple_strtoull(buf, &e, 10);
3692
3693 if (mddev->pers == NULL ||
3694 mddev->pers->quiesce == NULL)
3695 return -EINVAL;
3696 if (buf == e || (*e && *e != '\n'))
3697 return -EINVAL;
3698 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3699 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3700 mddev->suspend_hi = new;
3701 mddev->pers->quiesce(mddev, 1);
3702 mddev->pers->quiesce(mddev, 0);
3703 return len;
3704 } else
3705 return -EINVAL;
3706 }
3707 static struct md_sysfs_entry md_suspend_hi =
3708 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3709
3710 static ssize_t
3711 reshape_position_show(mddev_t *mddev, char *page)
3712 {
3713 if (mddev->reshape_position != MaxSector)
3714 return sprintf(page, "%llu\n",
3715 (unsigned long long)mddev->reshape_position);
3716 strcpy(page, "none\n");
3717 return 5;
3718 }
3719
3720 static ssize_t
3721 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3722 {
3723 char *e;
3724 unsigned long long new = simple_strtoull(buf, &e, 10);
3725 if (mddev->pers)
3726 return -EBUSY;
3727 if (buf == e || (*e && *e != '\n'))
3728 return -EINVAL;
3729 mddev->reshape_position = new;
3730 mddev->delta_disks = 0;
3731 mddev->new_level = mddev->level;
3732 mddev->new_layout = mddev->layout;
3733 mddev->new_chunk_sectors = mddev->chunk_sectors;
3734 return len;
3735 }
3736
3737 static struct md_sysfs_entry md_reshape_position =
3738 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3739 reshape_position_store);
3740
3741 static ssize_t
3742 array_size_show(mddev_t *mddev, char *page)
3743 {
3744 if (mddev->external_size)
3745 return sprintf(page, "%llu\n",
3746 (unsigned long long)mddev->array_sectors/2);
3747 else
3748 return sprintf(page, "default\n");
3749 }
3750
3751 static ssize_t
3752 array_size_store(mddev_t *mddev, const char *buf, size_t len)
3753 {
3754 sector_t sectors;
3755
3756 if (strncmp(buf, "default", 7) == 0) {
3757 if (mddev->pers)
3758 sectors = mddev->pers->size(mddev, 0, 0);
3759 else
3760 sectors = mddev->array_sectors;
3761
3762 mddev->external_size = 0;
3763 } else {
3764 if (strict_blocks_to_sectors(buf, &sectors) < 0)
3765 return -EINVAL;
3766 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
3767 return -E2BIG;
3768
3769 mddev->external_size = 1;
3770 }
3771
3772 mddev->array_sectors = sectors;
3773 set_capacity(mddev->gendisk, mddev->array_sectors);
3774 if (mddev->pers)
3775 revalidate_disk(mddev->gendisk);
3776
3777 return len;
3778 }
3779
3780 static struct md_sysfs_entry md_array_size =
3781 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
3782 array_size_store);
3783
3784 static struct attribute *md_default_attrs[] = {
3785 &md_level.attr,
3786 &md_layout.attr,
3787 &md_raid_disks.attr,
3788 &md_chunk_size.attr,
3789 &md_size.attr,
3790 &md_resync_start.attr,
3791 &md_metadata.attr,
3792 &md_new_device.attr,
3793 &md_safe_delay.attr,
3794 &md_array_state.attr,
3795 &md_reshape_position.attr,
3796 &md_array_size.attr,
3797 NULL,
3798 };
3799
3800 static struct attribute *md_redundancy_attrs[] = {
3801 &md_scan_mode.attr,
3802 &md_mismatches.attr,
3803 &md_sync_min.attr,
3804 &md_sync_max.attr,
3805 &md_sync_speed.attr,
3806 &md_sync_force_parallel.attr,
3807 &md_sync_completed.attr,
3808 &md_min_sync.attr,
3809 &md_max_sync.attr,
3810 &md_suspend_lo.attr,
3811 &md_suspend_hi.attr,
3812 &md_bitmap.attr,
3813 &md_degraded.attr,
3814 NULL,
3815 };
3816 static struct attribute_group md_redundancy_group = {
3817 .name = NULL,
3818 .attrs = md_redundancy_attrs,
3819 };
3820
3821
3822 static ssize_t
3823 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3824 {
3825 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3826 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3827 ssize_t rv;
3828
3829 if (!entry->show)
3830 return -EIO;
3831 rv = mddev_lock(mddev);
3832 if (!rv) {
3833 rv = entry->show(mddev, page);
3834 mddev_unlock(mddev);
3835 }
3836 return rv;
3837 }
3838
3839 static ssize_t
3840 md_attr_store(struct kobject *kobj, struct attribute *attr,
3841 const char *page, size_t length)
3842 {
3843 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3844 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3845 ssize_t rv;
3846
3847 if (!entry->store)
3848 return -EIO;
3849 if (!capable(CAP_SYS_ADMIN))
3850 return -EACCES;
3851 rv = mddev_lock(mddev);
3852 if (mddev->hold_active == UNTIL_IOCTL)
3853 mddev->hold_active = 0;
3854 if (!rv) {
3855 rv = entry->store(mddev, page, length);
3856 mddev_unlock(mddev);
3857 }
3858 return rv;
3859 }
3860
3861 static void md_free(struct kobject *ko)
3862 {
3863 mddev_t *mddev = container_of(ko, mddev_t, kobj);
3864
3865 if (mddev->sysfs_state)
3866 sysfs_put(mddev->sysfs_state);
3867
3868 if (mddev->gendisk) {
3869 del_gendisk(mddev->gendisk);
3870 put_disk(mddev->gendisk);
3871 }
3872 if (mddev->queue)
3873 blk_cleanup_queue(mddev->queue);
3874
3875 kfree(mddev);
3876 }
3877
3878 static struct sysfs_ops md_sysfs_ops = {
3879 .show = md_attr_show,
3880 .store = md_attr_store,
3881 };
3882 static struct kobj_type md_ktype = {
3883 .release = md_free,
3884 .sysfs_ops = &md_sysfs_ops,
3885 .default_attrs = md_default_attrs,
3886 };
3887
3888 int mdp_major = 0;
3889
3890 static void mddev_delayed_delete(struct work_struct *ws)
3891 {
3892 mddev_t *mddev = container_of(ws, mddev_t, del_work);
3893
3894 if (mddev->private == &md_redundancy_group) {
3895 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3896 if (mddev->sysfs_action)
3897 sysfs_put(mddev->sysfs_action);
3898 mddev->sysfs_action = NULL;
3899 mddev->private = NULL;
3900 }
3901 kobject_del(&mddev->kobj);
3902 kobject_put(&mddev->kobj);
3903 }
3904
3905 static int md_alloc(dev_t dev, char *name)
3906 {
3907 static DEFINE_MUTEX(disks_mutex);
3908 mddev_t *mddev = mddev_find(dev);
3909 struct gendisk *disk;
3910 int partitioned;
3911 int shift;
3912 int unit;
3913 int error;
3914
3915 if (!mddev)
3916 return -ENODEV;
3917
3918 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
3919 shift = partitioned ? MdpMinorShift : 0;
3920 unit = MINOR(mddev->unit) >> shift;
3921
3922 /* wait for any previous instance if this device
3923 * to be completed removed (mddev_delayed_delete).
3924 */
3925 flush_scheduled_work();
3926
3927 mutex_lock(&disks_mutex);
3928 error = -EEXIST;
3929 if (mddev->gendisk)
3930 goto abort;
3931
3932 if (name) {
3933 /* Need to ensure that 'name' is not a duplicate.
3934 */
3935 mddev_t *mddev2;
3936 spin_lock(&all_mddevs_lock);
3937
3938 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
3939 if (mddev2->gendisk &&
3940 strcmp(mddev2->gendisk->disk_name, name) == 0) {
3941 spin_unlock(&all_mddevs_lock);
3942 goto abort;
3943 }
3944 spin_unlock(&all_mddevs_lock);
3945 }
3946
3947 error = -ENOMEM;
3948 mddev->queue = blk_alloc_queue(GFP_KERNEL);
3949 if (!mddev->queue)
3950 goto abort;
3951 mddev->queue->queuedata = mddev;
3952
3953 /* Can be unlocked because the queue is new: no concurrency */
3954 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
3955
3956 blk_queue_make_request(mddev->queue, md_make_request);
3957
3958 disk = alloc_disk(1 << shift);
3959 if (!disk) {
3960 blk_cleanup_queue(mddev->queue);
3961 mddev->queue = NULL;
3962 goto abort;
3963 }
3964 disk->major = MAJOR(mddev->unit);
3965 disk->first_minor = unit << shift;
3966 if (name)
3967 strcpy(disk->disk_name, name);
3968 else if (partitioned)
3969 sprintf(disk->disk_name, "md_d%d", unit);
3970 else
3971 sprintf(disk->disk_name, "md%d", unit);
3972 disk->fops = &md_fops;
3973 disk->private_data = mddev;
3974 disk->queue = mddev->queue;
3975 /* Allow extended partitions. This makes the
3976 * 'mdp' device redundant, but we can't really
3977 * remove it now.
3978 */
3979 disk->flags |= GENHD_FL_EXT_DEVT;
3980 add_disk(disk);
3981 mddev->gendisk = disk;
3982 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
3983 &disk_to_dev(disk)->kobj, "%s", "md");
3984 if (error) {
3985 /* This isn't possible, but as kobject_init_and_add is marked
3986 * __must_check, we must do something with the result
3987 */
3988 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3989 disk->disk_name);
3990 error = 0;
3991 }
3992 abort:
3993 mutex_unlock(&disks_mutex);
3994 if (!error) {
3995 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3996 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state");
3997 }
3998 mddev_put(mddev);
3999 return error;
4000 }
4001
4002 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4003 {
4004 md_alloc(dev, NULL);
4005 return NULL;
4006 }
4007
4008 static int add_named_array(const char *val, struct kernel_param *kp)
4009 {
4010 /* val must be "md_*" where * is not all digits.
4011 * We allocate an array with a large free minor number, and
4012 * set the name to val. val must not already be an active name.
4013 */
4014 int len = strlen(val);
4015 char buf[DISK_NAME_LEN];
4016
4017 while (len && val[len-1] == '\n')
4018 len--;
4019 if (len >= DISK_NAME_LEN)
4020 return -E2BIG;
4021 strlcpy(buf, val, len+1);
4022 if (strncmp(buf, "md_", 3) != 0)
4023 return -EINVAL;
4024 return md_alloc(0, buf);
4025 }
4026
4027 static void md_safemode_timeout(unsigned long data)
4028 {
4029 mddev_t *mddev = (mddev_t *) data;
4030
4031 if (!atomic_read(&mddev->writes_pending)) {
4032 mddev->safemode = 1;
4033 if (mddev->external)
4034 sysfs_notify_dirent(mddev->sysfs_state);
4035 }
4036 md_wakeup_thread(mddev->thread);
4037 }
4038
4039 static int start_dirty_degraded;
4040
4041 static int do_md_run(mddev_t * mddev)
4042 {
4043 int err;
4044 mdk_rdev_t *rdev;
4045 struct gendisk *disk;
4046 struct mdk_personality *pers;
4047
4048 if (list_empty(&mddev->disks))
4049 /* cannot run an array with no devices.. */
4050 return -EINVAL;
4051
4052 if (mddev->pers)
4053 return -EBUSY;
4054
4055 /*
4056 * Analyze all RAID superblock(s)
4057 */
4058 if (!mddev->raid_disks) {
4059 if (!mddev->persistent)
4060 return -EINVAL;
4061 analyze_sbs(mddev);
4062 }
4063
4064 if (mddev->level != LEVEL_NONE)
4065 request_module("md-level-%d", mddev->level);
4066 else if (mddev->clevel[0])
4067 request_module("md-%s", mddev->clevel);
4068
4069 /*
4070 * Drop all container device buffers, from now on
4071 * the only valid external interface is through the md
4072 * device.
4073 */
4074 list_for_each_entry(rdev, &mddev->disks, same_set) {
4075 if (test_bit(Faulty, &rdev->flags))
4076 continue;
4077 sync_blockdev(rdev->bdev);
4078 invalidate_bdev(rdev->bdev);
4079
4080 /* perform some consistency tests on the device.
4081 * We don't want the data to overlap the metadata,
4082 * Internal Bitmap issues have been handled elsewhere.
4083 */
4084 if (rdev->data_offset < rdev->sb_start) {
4085 if (mddev->dev_sectors &&
4086 rdev->data_offset + mddev->dev_sectors
4087 > rdev->sb_start) {
4088 printk("md: %s: data overlaps metadata\n",
4089 mdname(mddev));
4090 return -EINVAL;
4091 }
4092 } else {
4093 if (rdev->sb_start + rdev->sb_size/512
4094 > rdev->data_offset) {
4095 printk("md: %s: metadata overlaps data\n",
4096 mdname(mddev));
4097 return -EINVAL;
4098 }
4099 }
4100 sysfs_notify_dirent(rdev->sysfs_state);
4101 }
4102
4103 md_probe(mddev->unit, NULL, NULL);
4104 disk = mddev->gendisk;
4105 if (!disk)
4106 return -ENOMEM;
4107
4108 spin_lock(&pers_lock);
4109 pers = find_pers(mddev->level, mddev->clevel);
4110 if (!pers || !try_module_get(pers->owner)) {
4111 spin_unlock(&pers_lock);
4112 if (mddev->level != LEVEL_NONE)
4113 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4114 mddev->level);
4115 else
4116 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4117 mddev->clevel);
4118 return -EINVAL;
4119 }
4120 mddev->pers = pers;
4121 spin_unlock(&pers_lock);
4122 if (mddev->level != pers->level) {
4123 mddev->level = pers->level;
4124 mddev->new_level = pers->level;
4125 }
4126 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4127
4128 if (mddev->reshape_position != MaxSector &&
4129 pers->start_reshape == NULL) {
4130 /* This personality cannot handle reshaping... */
4131 mddev->pers = NULL;
4132 module_put(pers->owner);
4133 return -EINVAL;
4134 }
4135
4136 if (pers->sync_request) {
4137 /* Warn if this is a potentially silly
4138 * configuration.
4139 */
4140 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4141 mdk_rdev_t *rdev2;
4142 int warned = 0;
4143
4144 list_for_each_entry(rdev, &mddev->disks, same_set)
4145 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4146 if (rdev < rdev2 &&
4147 rdev->bdev->bd_contains ==
4148 rdev2->bdev->bd_contains) {
4149 printk(KERN_WARNING
4150 "%s: WARNING: %s appears to be"
4151 " on the same physical disk as"
4152 " %s.\n",
4153 mdname(mddev),
4154 bdevname(rdev->bdev,b),
4155 bdevname(rdev2->bdev,b2));
4156 warned = 1;
4157 }
4158 }
4159
4160 if (warned)
4161 printk(KERN_WARNING
4162 "True protection against single-disk"
4163 " failure might be compromised.\n");
4164 }
4165
4166 mddev->recovery = 0;
4167 /* may be over-ridden by personality */
4168 mddev->resync_max_sectors = mddev->dev_sectors;
4169
4170 mddev->barriers_work = 1;
4171 mddev->ok_start_degraded = start_dirty_degraded;
4172
4173 if (start_readonly)
4174 mddev->ro = 2; /* read-only, but switch on first write */
4175
4176 err = mddev->pers->run(mddev);
4177 if (err)
4178 printk(KERN_ERR "md: pers->run() failed ...\n");
4179 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4180 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4181 " but 'external_size' not in effect?\n", __func__);
4182 printk(KERN_ERR
4183 "md: invalid array_size %llu > default size %llu\n",
4184 (unsigned long long)mddev->array_sectors / 2,
4185 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4186 err = -EINVAL;
4187 mddev->pers->stop(mddev);
4188 }
4189 if (err == 0 && mddev->pers->sync_request) {
4190 err = bitmap_create(mddev);
4191 if (err) {
4192 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4193 mdname(mddev), err);
4194 mddev->pers->stop(mddev);
4195 }
4196 }
4197 if (err) {
4198 module_put(mddev->pers->owner);
4199 mddev->pers = NULL;
4200 bitmap_destroy(mddev);
4201 return err;
4202 }
4203 if (mddev->pers->sync_request) {
4204 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4205 printk(KERN_WARNING
4206 "md: cannot register extra attributes for %s\n",
4207 mdname(mddev));
4208 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4209 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4210 mddev->ro = 0;
4211
4212 atomic_set(&mddev->writes_pending,0);
4213 mddev->safemode = 0;
4214 mddev->safemode_timer.function = md_safemode_timeout;
4215 mddev->safemode_timer.data = (unsigned long) mddev;
4216 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4217 mddev->in_sync = 1;
4218
4219 list_for_each_entry(rdev, &mddev->disks, same_set)
4220 if (rdev->raid_disk >= 0) {
4221 char nm[20];
4222 sprintf(nm, "rd%d", rdev->raid_disk);
4223 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4224 printk("md: cannot register %s for %s\n",
4225 nm, mdname(mddev));
4226 }
4227
4228 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4229
4230 if (mddev->flags)
4231 md_update_sb(mddev, 0);
4232
4233 set_capacity(disk, mddev->array_sectors);
4234
4235 /* If there is a partially-recovered drive we need to
4236 * start recovery here. If we leave it to md_check_recovery,
4237 * it will remove the drives and not do the right thing
4238 */
4239 if (mddev->degraded && !mddev->sync_thread) {
4240 int spares = 0;
4241 list_for_each_entry(rdev, &mddev->disks, same_set)
4242 if (rdev->raid_disk >= 0 &&
4243 !test_bit(In_sync, &rdev->flags) &&
4244 !test_bit(Faulty, &rdev->flags))
4245 /* complete an interrupted recovery */
4246 spares++;
4247 if (spares && mddev->pers->sync_request) {
4248 mddev->recovery = 0;
4249 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4250 mddev->sync_thread = md_register_thread(md_do_sync,
4251 mddev,
4252 "resync");
4253 if (!mddev->sync_thread) {
4254 printk(KERN_ERR "%s: could not start resync"
4255 " thread...\n",
4256 mdname(mddev));
4257 /* leave the spares where they are, it shouldn't hurt */
4258 mddev->recovery = 0;
4259 }
4260 }
4261 }
4262 md_wakeup_thread(mddev->thread);
4263 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4264
4265 revalidate_disk(mddev->gendisk);
4266 mddev->changed = 1;
4267 md_new_event(mddev);
4268 sysfs_notify_dirent(mddev->sysfs_state);
4269 if (mddev->sysfs_action)
4270 sysfs_notify_dirent(mddev->sysfs_action);
4271 sysfs_notify(&mddev->kobj, NULL, "degraded");
4272 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4273 return 0;
4274 }
4275
4276 static int restart_array(mddev_t *mddev)
4277 {
4278 struct gendisk *disk = mddev->gendisk;
4279
4280 /* Complain if it has no devices */
4281 if (list_empty(&mddev->disks))
4282 return -ENXIO;
4283 if (!mddev->pers)
4284 return -EINVAL;
4285 if (!mddev->ro)
4286 return -EBUSY;
4287 mddev->safemode = 0;
4288 mddev->ro = 0;
4289 set_disk_ro(disk, 0);
4290 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4291 mdname(mddev));
4292 /* Kick recovery or resync if necessary */
4293 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4294 md_wakeup_thread(mddev->thread);
4295 md_wakeup_thread(mddev->sync_thread);
4296 sysfs_notify_dirent(mddev->sysfs_state);
4297 return 0;
4298 }
4299
4300 /* similar to deny_write_access, but accounts for our holding a reference
4301 * to the file ourselves */
4302 static int deny_bitmap_write_access(struct file * file)
4303 {
4304 struct inode *inode = file->f_mapping->host;
4305
4306 spin_lock(&inode->i_lock);
4307 if (atomic_read(&inode->i_writecount) > 1) {
4308 spin_unlock(&inode->i_lock);
4309 return -ETXTBSY;
4310 }
4311 atomic_set(&inode->i_writecount, -1);
4312 spin_unlock(&inode->i_lock);
4313
4314 return 0;
4315 }
4316
4317 static void restore_bitmap_write_access(struct file *file)
4318 {
4319 struct inode *inode = file->f_mapping->host;
4320
4321 spin_lock(&inode->i_lock);
4322 atomic_set(&inode->i_writecount, 1);
4323 spin_unlock(&inode->i_lock);
4324 }
4325
4326 /* mode:
4327 * 0 - completely stop and dis-assemble array
4328 * 1 - switch to readonly
4329 * 2 - stop but do not disassemble array
4330 */
4331 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4332 {
4333 int err = 0;
4334 struct gendisk *disk = mddev->gendisk;
4335 mdk_rdev_t *rdev;
4336
4337 mutex_lock(&mddev->open_mutex);
4338 if (atomic_read(&mddev->openers) > is_open) {
4339 printk("md: %s still in use.\n",mdname(mddev));
4340 err = -EBUSY;
4341 } else if (mddev->pers) {
4342
4343 if (mddev->sync_thread) {
4344 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4345 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4346 md_unregister_thread(mddev->sync_thread);
4347 mddev->sync_thread = NULL;
4348 }
4349
4350 del_timer_sync(&mddev->safemode_timer);
4351
4352 switch(mode) {
4353 case 1: /* readonly */
4354 err = -ENXIO;
4355 if (mddev->ro==1)
4356 goto out;
4357 mddev->ro = 1;
4358 break;
4359 case 0: /* disassemble */
4360 case 2: /* stop */
4361 bitmap_flush(mddev);
4362 md_super_wait(mddev);
4363 if (mddev->ro)
4364 set_disk_ro(disk, 0);
4365
4366 mddev->pers->stop(mddev);
4367 mddev->queue->merge_bvec_fn = NULL;
4368 mddev->queue->unplug_fn = NULL;
4369 mddev->queue->backing_dev_info.congested_fn = NULL;
4370 module_put(mddev->pers->owner);
4371 if (mddev->pers->sync_request)
4372 mddev->private = &md_redundancy_group;
4373 mddev->pers = NULL;
4374 /* tell userspace to handle 'inactive' */
4375 sysfs_notify_dirent(mddev->sysfs_state);
4376
4377 list_for_each_entry(rdev, &mddev->disks, same_set)
4378 if (rdev->raid_disk >= 0) {
4379 char nm[20];
4380 sprintf(nm, "rd%d", rdev->raid_disk);
4381 sysfs_remove_link(&mddev->kobj, nm);
4382 }
4383
4384 set_capacity(disk, 0);
4385 mddev->changed = 1;
4386
4387 if (mddev->ro)
4388 mddev->ro = 0;
4389 }
4390 if (!mddev->in_sync || mddev->flags) {
4391 /* mark array as shutdown cleanly */
4392 mddev->in_sync = 1;
4393 md_update_sb(mddev, 1);
4394 }
4395 if (mode == 1)
4396 set_disk_ro(disk, 1);
4397 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4398 err = 0;
4399 }
4400 out:
4401 mutex_unlock(&mddev->open_mutex);
4402 if (err)
4403 return err;
4404 /*
4405 * Free resources if final stop
4406 */
4407 if (mode == 0) {
4408
4409 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4410
4411 bitmap_destroy(mddev);
4412 if (mddev->bitmap_file) {
4413 restore_bitmap_write_access(mddev->bitmap_file);
4414 fput(mddev->bitmap_file);
4415 mddev->bitmap_file = NULL;
4416 }
4417 mddev->bitmap_offset = 0;
4418
4419 /* make sure all md_delayed_delete calls have finished */
4420 flush_scheduled_work();
4421
4422 export_array(mddev);
4423
4424 mddev->array_sectors = 0;
4425 mddev->external_size = 0;
4426 mddev->dev_sectors = 0;
4427 mddev->raid_disks = 0;
4428 mddev->recovery_cp = 0;
4429 mddev->resync_min = 0;
4430 mddev->resync_max = MaxSector;
4431 mddev->reshape_position = MaxSector;
4432 mddev->external = 0;
4433 mddev->persistent = 0;
4434 mddev->level = LEVEL_NONE;
4435 mddev->clevel[0] = 0;
4436 mddev->flags = 0;
4437 mddev->ro = 0;
4438 mddev->metadata_type[0] = 0;
4439 mddev->chunk_sectors = 0;
4440 mddev->ctime = mddev->utime = 0;
4441 mddev->layout = 0;
4442 mddev->max_disks = 0;
4443 mddev->events = 0;
4444 mddev->delta_disks = 0;
4445 mddev->new_level = LEVEL_NONE;
4446 mddev->new_layout = 0;
4447 mddev->new_chunk_sectors = 0;
4448 mddev->curr_resync = 0;
4449 mddev->resync_mismatches = 0;
4450 mddev->suspend_lo = mddev->suspend_hi = 0;
4451 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4452 mddev->recovery = 0;
4453 mddev->in_sync = 0;
4454 mddev->changed = 0;
4455 mddev->degraded = 0;
4456 mddev->barriers_work = 0;
4457 mddev->safemode = 0;
4458 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4459 if (mddev->hold_active == UNTIL_STOP)
4460 mddev->hold_active = 0;
4461
4462 } else if (mddev->pers)
4463 printk(KERN_INFO "md: %s switched to read-only mode.\n",
4464 mdname(mddev));
4465 err = 0;
4466 blk_integrity_unregister(disk);
4467 md_new_event(mddev);
4468 sysfs_notify_dirent(mddev->sysfs_state);
4469 return err;
4470 }
4471
4472 #ifndef MODULE
4473 static void autorun_array(mddev_t *mddev)
4474 {
4475 mdk_rdev_t *rdev;
4476 int err;
4477
4478 if (list_empty(&mddev->disks))
4479 return;
4480
4481 printk(KERN_INFO "md: running: ");
4482
4483 list_for_each_entry(rdev, &mddev->disks, same_set) {
4484 char b[BDEVNAME_SIZE];
4485 printk("<%s>", bdevname(rdev->bdev,b));
4486 }
4487 printk("\n");
4488
4489 err = do_md_run(mddev);
4490 if (err) {
4491 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4492 do_md_stop(mddev, 0, 0);
4493 }
4494 }
4495
4496 /*
4497 * lets try to run arrays based on all disks that have arrived
4498 * until now. (those are in pending_raid_disks)
4499 *
4500 * the method: pick the first pending disk, collect all disks with
4501 * the same UUID, remove all from the pending list and put them into
4502 * the 'same_array' list. Then order this list based on superblock
4503 * update time (freshest comes first), kick out 'old' disks and
4504 * compare superblocks. If everything's fine then run it.
4505 *
4506 * If "unit" is allocated, then bump its reference count
4507 */
4508 static void autorun_devices(int part)
4509 {
4510 mdk_rdev_t *rdev0, *rdev, *tmp;
4511 mddev_t *mddev;
4512 char b[BDEVNAME_SIZE];
4513
4514 printk(KERN_INFO "md: autorun ...\n");
4515 while (!list_empty(&pending_raid_disks)) {
4516 int unit;
4517 dev_t dev;
4518 LIST_HEAD(candidates);
4519 rdev0 = list_entry(pending_raid_disks.next,
4520 mdk_rdev_t, same_set);
4521
4522 printk(KERN_INFO "md: considering %s ...\n",
4523 bdevname(rdev0->bdev,b));
4524 INIT_LIST_HEAD(&candidates);
4525 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4526 if (super_90_load(rdev, rdev0, 0) >= 0) {
4527 printk(KERN_INFO "md: adding %s ...\n",
4528 bdevname(rdev->bdev,b));
4529 list_move(&rdev->same_set, &candidates);
4530 }
4531 /*
4532 * now we have a set of devices, with all of them having
4533 * mostly sane superblocks. It's time to allocate the
4534 * mddev.
4535 */
4536 if (part) {
4537 dev = MKDEV(mdp_major,
4538 rdev0->preferred_minor << MdpMinorShift);
4539 unit = MINOR(dev) >> MdpMinorShift;
4540 } else {
4541 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4542 unit = MINOR(dev);
4543 }
4544 if (rdev0->preferred_minor != unit) {
4545 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4546 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4547 break;
4548 }
4549
4550 md_probe(dev, NULL, NULL);
4551 mddev = mddev_find(dev);
4552 if (!mddev || !mddev->gendisk) {
4553 if (mddev)
4554 mddev_put(mddev);
4555 printk(KERN_ERR
4556 "md: cannot allocate memory for md drive.\n");
4557 break;
4558 }
4559 if (mddev_lock(mddev))
4560 printk(KERN_WARNING "md: %s locked, cannot run\n",
4561 mdname(mddev));
4562 else if (mddev->raid_disks || mddev->major_version
4563 || !list_empty(&mddev->disks)) {
4564 printk(KERN_WARNING
4565 "md: %s already running, cannot run %s\n",
4566 mdname(mddev), bdevname(rdev0->bdev,b));
4567 mddev_unlock(mddev);
4568 } else {
4569 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4570 mddev->persistent = 1;
4571 rdev_for_each_list(rdev, tmp, &candidates) {
4572 list_del_init(&rdev->same_set);
4573 if (bind_rdev_to_array(rdev, mddev))
4574 export_rdev(rdev);
4575 }
4576 autorun_array(mddev);
4577 mddev_unlock(mddev);
4578 }
4579 /* on success, candidates will be empty, on error
4580 * it won't...
4581 */
4582 rdev_for_each_list(rdev, tmp, &candidates) {
4583 list_del_init(&rdev->same_set);
4584 export_rdev(rdev);
4585 }
4586 mddev_put(mddev);
4587 }
4588 printk(KERN_INFO "md: ... autorun DONE.\n");
4589 }
4590 #endif /* !MODULE */
4591
4592 static int get_version(void __user * arg)
4593 {
4594 mdu_version_t ver;
4595
4596 ver.major = MD_MAJOR_VERSION;
4597 ver.minor = MD_MINOR_VERSION;
4598 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4599
4600 if (copy_to_user(arg, &ver, sizeof(ver)))
4601 return -EFAULT;
4602
4603 return 0;
4604 }
4605
4606 static int get_array_info(mddev_t * mddev, void __user * arg)
4607 {
4608 mdu_array_info_t info;
4609 int nr,working,insync,failed,spare;
4610 mdk_rdev_t *rdev;
4611
4612 nr=working=insync=failed=spare=0;
4613 list_for_each_entry(rdev, &mddev->disks, same_set) {
4614 nr++;
4615 if (test_bit(Faulty, &rdev->flags))
4616 failed++;
4617 else {
4618 working++;
4619 if (test_bit(In_sync, &rdev->flags))
4620 insync++;
4621 else
4622 spare++;
4623 }
4624 }
4625
4626 info.major_version = mddev->major_version;
4627 info.minor_version = mddev->minor_version;
4628 info.patch_version = MD_PATCHLEVEL_VERSION;
4629 info.ctime = mddev->ctime;
4630 info.level = mddev->level;
4631 info.size = mddev->dev_sectors / 2;
4632 if (info.size != mddev->dev_sectors / 2) /* overflow */
4633 info.size = -1;
4634 info.nr_disks = nr;
4635 info.raid_disks = mddev->raid_disks;
4636 info.md_minor = mddev->md_minor;
4637 info.not_persistent= !mddev->persistent;
4638
4639 info.utime = mddev->utime;
4640 info.state = 0;
4641 if (mddev->in_sync)
4642 info.state = (1<<MD_SB_CLEAN);
4643 if (mddev->bitmap && mddev->bitmap_offset)
4644 info.state = (1<<MD_SB_BITMAP_PRESENT);
4645 info.active_disks = insync;
4646 info.working_disks = working;
4647 info.failed_disks = failed;
4648 info.spare_disks = spare;
4649
4650 info.layout = mddev->layout;
4651 info.chunk_size = mddev->chunk_sectors << 9;
4652
4653 if (copy_to_user(arg, &info, sizeof(info)))
4654 return -EFAULT;
4655
4656 return 0;
4657 }
4658
4659 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4660 {
4661 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4662 char *ptr, *buf = NULL;
4663 int err = -ENOMEM;
4664
4665 if (md_allow_write(mddev))
4666 file = kmalloc(sizeof(*file), GFP_NOIO);
4667 else
4668 file = kmalloc(sizeof(*file), GFP_KERNEL);
4669
4670 if (!file)
4671 goto out;
4672
4673 /* bitmap disabled, zero the first byte and copy out */
4674 if (!mddev->bitmap || !mddev->bitmap->file) {
4675 file->pathname[0] = '\0';
4676 goto copy_out;
4677 }
4678
4679 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4680 if (!buf)
4681 goto out;
4682
4683 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4684 if (IS_ERR(ptr))
4685 goto out;
4686
4687 strcpy(file->pathname, ptr);
4688
4689 copy_out:
4690 err = 0;
4691 if (copy_to_user(arg, file, sizeof(*file)))
4692 err = -EFAULT;
4693 out:
4694 kfree(buf);
4695 kfree(file);
4696 return err;
4697 }
4698
4699 static int get_disk_info(mddev_t * mddev, void __user * arg)
4700 {
4701 mdu_disk_info_t info;
4702 mdk_rdev_t *rdev;
4703
4704 if (copy_from_user(&info, arg, sizeof(info)))
4705 return -EFAULT;
4706
4707 rdev = find_rdev_nr(mddev, info.number);
4708 if (rdev) {
4709 info.major = MAJOR(rdev->bdev->bd_dev);
4710 info.minor = MINOR(rdev->bdev->bd_dev);
4711 info.raid_disk = rdev->raid_disk;
4712 info.state = 0;
4713 if (test_bit(Faulty, &rdev->flags))
4714 info.state |= (1<<MD_DISK_FAULTY);
4715 else if (test_bit(In_sync, &rdev->flags)) {
4716 info.state |= (1<<MD_DISK_ACTIVE);
4717 info.state |= (1<<MD_DISK_SYNC);
4718 }
4719 if (test_bit(WriteMostly, &rdev->flags))
4720 info.state |= (1<<MD_DISK_WRITEMOSTLY);
4721 } else {
4722 info.major = info.minor = 0;
4723 info.raid_disk = -1;
4724 info.state = (1<<MD_DISK_REMOVED);
4725 }
4726
4727 if (copy_to_user(arg, &info, sizeof(info)))
4728 return -EFAULT;
4729
4730 return 0;
4731 }
4732
4733 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4734 {
4735 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4736 mdk_rdev_t *rdev;
4737 dev_t dev = MKDEV(info->major,info->minor);
4738
4739 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4740 return -EOVERFLOW;
4741
4742 if (!mddev->raid_disks) {
4743 int err;
4744 /* expecting a device which has a superblock */
4745 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4746 if (IS_ERR(rdev)) {
4747 printk(KERN_WARNING
4748 "md: md_import_device returned %ld\n",
4749 PTR_ERR(rdev));
4750 return PTR_ERR(rdev);
4751 }
4752 if (!list_empty(&mddev->disks)) {
4753 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4754 mdk_rdev_t, same_set);
4755 err = super_types[mddev->major_version]
4756 .load_super(rdev, rdev0, mddev->minor_version);
4757 if (err < 0) {
4758 printk(KERN_WARNING
4759 "md: %s has different UUID to %s\n",
4760 bdevname(rdev->bdev,b),
4761 bdevname(rdev0->bdev,b2));
4762 export_rdev(rdev);
4763 return -EINVAL;
4764 }
4765 }
4766 err = bind_rdev_to_array(rdev, mddev);
4767 if (err)
4768 export_rdev(rdev);
4769 return err;
4770 }
4771
4772 /*
4773 * add_new_disk can be used once the array is assembled
4774 * to add "hot spares". They must already have a superblock
4775 * written
4776 */
4777 if (mddev->pers) {
4778 int err;
4779 if (!mddev->pers->hot_add_disk) {
4780 printk(KERN_WARNING
4781 "%s: personality does not support diskops!\n",
4782 mdname(mddev));
4783 return -EINVAL;
4784 }
4785 if (mddev->persistent)
4786 rdev = md_import_device(dev, mddev->major_version,
4787 mddev->minor_version);
4788 else
4789 rdev = md_import_device(dev, -1, -1);
4790 if (IS_ERR(rdev)) {
4791 printk(KERN_WARNING
4792 "md: md_import_device returned %ld\n",
4793 PTR_ERR(rdev));
4794 return PTR_ERR(rdev);
4795 }
4796 /* set save_raid_disk if appropriate */
4797 if (!mddev->persistent) {
4798 if (info->state & (1<<MD_DISK_SYNC) &&
4799 info->raid_disk < mddev->raid_disks)
4800 rdev->raid_disk = info->raid_disk;
4801 else
4802 rdev->raid_disk = -1;
4803 } else
4804 super_types[mddev->major_version].
4805 validate_super(mddev, rdev);
4806 rdev->saved_raid_disk = rdev->raid_disk;
4807
4808 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4809 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4810 set_bit(WriteMostly, &rdev->flags);
4811 else
4812 clear_bit(WriteMostly, &rdev->flags);
4813
4814 rdev->raid_disk = -1;
4815 err = bind_rdev_to_array(rdev, mddev);
4816 if (!err && !mddev->pers->hot_remove_disk) {
4817 /* If there is hot_add_disk but no hot_remove_disk
4818 * then added disks for geometry changes,
4819 * and should be added immediately.
4820 */
4821 super_types[mddev->major_version].
4822 validate_super(mddev, rdev);
4823 err = mddev->pers->hot_add_disk(mddev, rdev);
4824 if (err)
4825 unbind_rdev_from_array(rdev);
4826 }
4827 if (err)
4828 export_rdev(rdev);
4829 else
4830 sysfs_notify_dirent(rdev->sysfs_state);
4831
4832 md_update_sb(mddev, 1);
4833 if (mddev->degraded)
4834 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4835 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4836 md_wakeup_thread(mddev->thread);
4837 return err;
4838 }
4839
4840 /* otherwise, add_new_disk is only allowed
4841 * for major_version==0 superblocks
4842 */
4843 if (mddev->major_version != 0) {
4844 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4845 mdname(mddev));
4846 return -EINVAL;
4847 }
4848
4849 if (!(info->state & (1<<MD_DISK_FAULTY))) {
4850 int err;
4851 rdev = md_import_device(dev, -1, 0);
4852 if (IS_ERR(rdev)) {
4853 printk(KERN_WARNING
4854 "md: error, md_import_device() returned %ld\n",
4855 PTR_ERR(rdev));
4856 return PTR_ERR(rdev);
4857 }
4858 rdev->desc_nr = info->number;
4859 if (info->raid_disk < mddev->raid_disks)
4860 rdev->raid_disk = info->raid_disk;
4861 else
4862 rdev->raid_disk = -1;
4863
4864 if (rdev->raid_disk < mddev->raid_disks)
4865 if (info->state & (1<<MD_DISK_SYNC))
4866 set_bit(In_sync, &rdev->flags);
4867
4868 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4869 set_bit(WriteMostly, &rdev->flags);
4870
4871 if (!mddev->persistent) {
4872 printk(KERN_INFO "md: nonpersistent superblock ...\n");
4873 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4874 } else
4875 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4876 rdev->sectors = rdev->sb_start;
4877
4878 err = bind_rdev_to_array(rdev, mddev);
4879 if (err) {
4880 export_rdev(rdev);
4881 return err;
4882 }
4883 }
4884
4885 return 0;
4886 }
4887
4888 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4889 {
4890 char b[BDEVNAME_SIZE];
4891 mdk_rdev_t *rdev;
4892
4893 rdev = find_rdev(mddev, dev);
4894 if (!rdev)
4895 return -ENXIO;
4896
4897 if (rdev->raid_disk >= 0)
4898 goto busy;
4899
4900 kick_rdev_from_array(rdev);
4901 md_update_sb(mddev, 1);
4902 md_new_event(mddev);
4903
4904 return 0;
4905 busy:
4906 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4907 bdevname(rdev->bdev,b), mdname(mddev));
4908 return -EBUSY;
4909 }
4910
4911 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4912 {
4913 char b[BDEVNAME_SIZE];
4914 int err;
4915 mdk_rdev_t *rdev;
4916
4917 if (!mddev->pers)
4918 return -ENODEV;
4919
4920 if (mddev->major_version != 0) {
4921 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4922 " version-0 superblocks.\n",
4923 mdname(mddev));
4924 return -EINVAL;
4925 }
4926 if (!mddev->pers->hot_add_disk) {
4927 printk(KERN_WARNING
4928 "%s: personality does not support diskops!\n",
4929 mdname(mddev));
4930 return -EINVAL;
4931 }
4932
4933 rdev = md_import_device(dev, -1, 0);
4934 if (IS_ERR(rdev)) {
4935 printk(KERN_WARNING
4936 "md: error, md_import_device() returned %ld\n",
4937 PTR_ERR(rdev));
4938 return -EINVAL;
4939 }
4940
4941 if (mddev->persistent)
4942 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4943 else
4944 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4945
4946 rdev->sectors = rdev->sb_start;
4947
4948 if (test_bit(Faulty, &rdev->flags)) {
4949 printk(KERN_WARNING
4950 "md: can not hot-add faulty %s disk to %s!\n",
4951 bdevname(rdev->bdev,b), mdname(mddev));
4952 err = -EINVAL;
4953 goto abort_export;
4954 }
4955 clear_bit(In_sync, &rdev->flags);
4956 rdev->desc_nr = -1;
4957 rdev->saved_raid_disk = -1;
4958 err = bind_rdev_to_array(rdev, mddev);
4959 if (err)
4960 goto abort_export;
4961
4962 /*
4963 * The rest should better be atomic, we can have disk failures
4964 * noticed in interrupt contexts ...
4965 */
4966
4967 rdev->raid_disk = -1;
4968
4969 md_update_sb(mddev, 1);
4970
4971 /*
4972 * Kick recovery, maybe this spare has to be added to the
4973 * array immediately.
4974 */
4975 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4976 md_wakeup_thread(mddev->thread);
4977 md_new_event(mddev);
4978 return 0;
4979
4980 abort_export:
4981 export_rdev(rdev);
4982 return err;
4983 }
4984
4985 static int set_bitmap_file(mddev_t *mddev, int fd)
4986 {
4987 int err;
4988
4989 if (mddev->pers) {
4990 if (!mddev->pers->quiesce)
4991 return -EBUSY;
4992 if (mddev->recovery || mddev->sync_thread)
4993 return -EBUSY;
4994 /* we should be able to change the bitmap.. */
4995 }
4996
4997
4998 if (fd >= 0) {
4999 if (mddev->bitmap)
5000 return -EEXIST; /* cannot add when bitmap is present */
5001 mddev->bitmap_file = fget(fd);
5002
5003 if (mddev->bitmap_file == NULL) {
5004 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5005 mdname(mddev));
5006 return -EBADF;
5007 }
5008
5009 err = deny_bitmap_write_access(mddev->bitmap_file);
5010 if (err) {
5011 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5012 mdname(mddev));
5013 fput(mddev->bitmap_file);
5014 mddev->bitmap_file = NULL;
5015 return err;
5016 }
5017 mddev->bitmap_offset = 0; /* file overrides offset */
5018 } else if (mddev->bitmap == NULL)
5019 return -ENOENT; /* cannot remove what isn't there */
5020 err = 0;
5021 if (mddev->pers) {
5022 mddev->pers->quiesce(mddev, 1);
5023 if (fd >= 0)
5024 err = bitmap_create(mddev);
5025 if (fd < 0 || err) {
5026 bitmap_destroy(mddev);
5027 fd = -1; /* make sure to put the file */
5028 }
5029 mddev->pers->quiesce(mddev, 0);
5030 }
5031 if (fd < 0) {
5032 if (mddev->bitmap_file) {
5033 restore_bitmap_write_access(mddev->bitmap_file);
5034 fput(mddev->bitmap_file);
5035 }
5036 mddev->bitmap_file = NULL;
5037 }
5038
5039 return err;
5040 }
5041
5042 /*
5043 * set_array_info is used two different ways
5044 * The original usage is when creating a new array.
5045 * In this usage, raid_disks is > 0 and it together with
5046 * level, size, not_persistent,layout,chunksize determine the
5047 * shape of the array.
5048 * This will always create an array with a type-0.90.0 superblock.
5049 * The newer usage is when assembling an array.
5050 * In this case raid_disks will be 0, and the major_version field is
5051 * use to determine which style super-blocks are to be found on the devices.
5052 * The minor and patch _version numbers are also kept incase the
5053 * super_block handler wishes to interpret them.
5054 */
5055 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5056 {
5057
5058 if (info->raid_disks == 0) {
5059 /* just setting version number for superblock loading */
5060 if (info->major_version < 0 ||
5061 info->major_version >= ARRAY_SIZE(super_types) ||
5062 super_types[info->major_version].name == NULL) {
5063 /* maybe try to auto-load a module? */
5064 printk(KERN_INFO
5065 "md: superblock version %d not known\n",
5066 info->major_version);
5067 return -EINVAL;
5068 }
5069 mddev->major_version = info->major_version;
5070 mddev->minor_version = info->minor_version;
5071 mddev->patch_version = info->patch_version;
5072 mddev->persistent = !info->not_persistent;
5073 return 0;
5074 }
5075 mddev->major_version = MD_MAJOR_VERSION;
5076 mddev->minor_version = MD_MINOR_VERSION;
5077 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5078 mddev->ctime = get_seconds();
5079
5080 mddev->level = info->level;
5081 mddev->clevel[0] = 0;
5082 mddev->dev_sectors = 2 * (sector_t)info->size;
5083 mddev->raid_disks = info->raid_disks;
5084 /* don't set md_minor, it is determined by which /dev/md* was
5085 * openned
5086 */
5087 if (info->state & (1<<MD_SB_CLEAN))
5088 mddev->recovery_cp = MaxSector;
5089 else
5090 mddev->recovery_cp = 0;
5091 mddev->persistent = ! info->not_persistent;
5092 mddev->external = 0;
5093
5094 mddev->layout = info->layout;
5095 mddev->chunk_sectors = info->chunk_size >> 9;
5096
5097 mddev->max_disks = MD_SB_DISKS;
5098
5099 if (mddev->persistent)
5100 mddev->flags = 0;
5101 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5102
5103 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
5104 mddev->bitmap_offset = 0;
5105
5106 mddev->reshape_position = MaxSector;
5107
5108 /*
5109 * Generate a 128 bit UUID
5110 */
5111 get_random_bytes(mddev->uuid, 16);
5112
5113 mddev->new_level = mddev->level;
5114 mddev->new_chunk_sectors = mddev->chunk_sectors;
5115 mddev->new_layout = mddev->layout;
5116 mddev->delta_disks = 0;
5117
5118 return 0;
5119 }
5120
5121 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5122 {
5123 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5124
5125 if (mddev->external_size)
5126 return;
5127
5128 mddev->array_sectors = array_sectors;
5129 }
5130 EXPORT_SYMBOL(md_set_array_sectors);
5131
5132 static int update_size(mddev_t *mddev, sector_t num_sectors)
5133 {
5134 mdk_rdev_t *rdev;
5135 int rv;
5136 int fit = (num_sectors == 0);
5137
5138 if (mddev->pers->resize == NULL)
5139 return -EINVAL;
5140 /* The "num_sectors" is the number of sectors of each device that
5141 * is used. This can only make sense for arrays with redundancy.
5142 * linear and raid0 always use whatever space is available. We can only
5143 * consider changing this number if no resync or reconstruction is
5144 * happening, and if the new size is acceptable. It must fit before the
5145 * sb_start or, if that is <data_offset, it must fit before the size
5146 * of each device. If num_sectors is zero, we find the largest size
5147 * that fits.
5148
5149 */
5150 if (mddev->sync_thread)
5151 return -EBUSY;
5152 if (mddev->bitmap)
5153 /* Sorry, cannot grow a bitmap yet, just remove it,
5154 * grow, and re-add.
5155 */
5156 return -EBUSY;
5157 list_for_each_entry(rdev, &mddev->disks, same_set) {
5158 sector_t avail = rdev->sectors;
5159
5160 if (fit && (num_sectors == 0 || num_sectors > avail))
5161 num_sectors = avail;
5162 if (avail < num_sectors)
5163 return -ENOSPC;
5164 }
5165 rv = mddev->pers->resize(mddev, num_sectors);
5166 if (!rv)
5167 revalidate_disk(mddev->gendisk);
5168 return rv;
5169 }
5170
5171 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5172 {
5173 int rv;
5174 /* change the number of raid disks */
5175 if (mddev->pers->check_reshape == NULL)
5176 return -EINVAL;
5177 if (raid_disks <= 0 ||
5178 raid_disks >= mddev->max_disks)
5179 return -EINVAL;
5180 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5181 return -EBUSY;
5182 mddev->delta_disks = raid_disks - mddev->raid_disks;
5183
5184 rv = mddev->pers->check_reshape(mddev);
5185 return rv;
5186 }
5187
5188
5189 /*
5190 * update_array_info is used to change the configuration of an
5191 * on-line array.
5192 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5193 * fields in the info are checked against the array.
5194 * Any differences that cannot be handled will cause an error.
5195 * Normally, only one change can be managed at a time.
5196 */
5197 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5198 {
5199 int rv = 0;
5200 int cnt = 0;
5201 int state = 0;
5202
5203 /* calculate expected state,ignoring low bits */
5204 if (mddev->bitmap && mddev->bitmap_offset)
5205 state |= (1 << MD_SB_BITMAP_PRESENT);
5206
5207 if (mddev->major_version != info->major_version ||
5208 mddev->minor_version != info->minor_version ||
5209 /* mddev->patch_version != info->patch_version || */
5210 mddev->ctime != info->ctime ||
5211 mddev->level != info->level ||
5212 /* mddev->layout != info->layout || */
5213 !mddev->persistent != info->not_persistent||
5214 mddev->chunk_sectors != info->chunk_size >> 9 ||
5215 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5216 ((state^info->state) & 0xfffffe00)
5217 )
5218 return -EINVAL;
5219 /* Check there is only one change */
5220 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5221 cnt++;
5222 if (mddev->raid_disks != info->raid_disks)
5223 cnt++;
5224 if (mddev->layout != info->layout)
5225 cnt++;
5226 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5227 cnt++;
5228 if (cnt == 0)
5229 return 0;
5230 if (cnt > 1)
5231 return -EINVAL;
5232
5233 if (mddev->layout != info->layout) {
5234 /* Change layout
5235 * we don't need to do anything at the md level, the
5236 * personality will take care of it all.
5237 */
5238 if (mddev->pers->check_reshape == NULL)
5239 return -EINVAL;
5240 else {
5241 mddev->new_layout = info->layout;
5242 rv = mddev->pers->check_reshape(mddev);
5243 if (rv)
5244 mddev->new_layout = mddev->layout;
5245 return rv;
5246 }
5247 }
5248 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5249 rv = update_size(mddev, (sector_t)info->size * 2);
5250
5251 if (mddev->raid_disks != info->raid_disks)
5252 rv = update_raid_disks(mddev, info->raid_disks);
5253
5254 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5255 if (mddev->pers->quiesce == NULL)
5256 return -EINVAL;
5257 if (mddev->recovery || mddev->sync_thread)
5258 return -EBUSY;
5259 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5260 /* add the bitmap */
5261 if (mddev->bitmap)
5262 return -EEXIST;
5263 if (mddev->default_bitmap_offset == 0)
5264 return -EINVAL;
5265 mddev->bitmap_offset = mddev->default_bitmap_offset;
5266 mddev->pers->quiesce(mddev, 1);
5267 rv = bitmap_create(mddev);
5268 if (rv)
5269 bitmap_destroy(mddev);
5270 mddev->pers->quiesce(mddev, 0);
5271 } else {
5272 /* remove the bitmap */
5273 if (!mddev->bitmap)
5274 return -ENOENT;
5275 if (mddev->bitmap->file)
5276 return -EINVAL;
5277 mddev->pers->quiesce(mddev, 1);
5278 bitmap_destroy(mddev);
5279 mddev->pers->quiesce(mddev, 0);
5280 mddev->bitmap_offset = 0;
5281 }
5282 }
5283 md_update_sb(mddev, 1);
5284 return rv;
5285 }
5286
5287 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5288 {
5289 mdk_rdev_t *rdev;
5290
5291 if (mddev->pers == NULL)
5292 return -ENODEV;
5293
5294 rdev = find_rdev(mddev, dev);
5295 if (!rdev)
5296 return -ENODEV;
5297
5298 md_error(mddev, rdev);
5299 return 0;
5300 }
5301
5302 /*
5303 * We have a problem here : there is no easy way to give a CHS
5304 * virtual geometry. We currently pretend that we have a 2 heads
5305 * 4 sectors (with a BIG number of cylinders...). This drives
5306 * dosfs just mad... ;-)
5307 */
5308 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5309 {
5310 mddev_t *mddev = bdev->bd_disk->private_data;
5311
5312 geo->heads = 2;
5313 geo->sectors = 4;
5314 geo->cylinders = get_capacity(mddev->gendisk) / 8;
5315 return 0;
5316 }
5317
5318 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5319 unsigned int cmd, unsigned long arg)
5320 {
5321 int err = 0;
5322 void __user *argp = (void __user *)arg;
5323 mddev_t *mddev = NULL;
5324
5325 if (!capable(CAP_SYS_ADMIN))
5326 return -EACCES;
5327
5328 /*
5329 * Commands dealing with the RAID driver but not any
5330 * particular array:
5331 */
5332 switch (cmd)
5333 {
5334 case RAID_VERSION:
5335 err = get_version(argp);
5336 goto done;
5337
5338 case PRINT_RAID_DEBUG:
5339 err = 0;
5340 md_print_devices();
5341 goto done;
5342
5343 #ifndef MODULE
5344 case RAID_AUTORUN:
5345 err = 0;
5346 autostart_arrays(arg);
5347 goto done;
5348 #endif
5349 default:;
5350 }
5351
5352 /*
5353 * Commands creating/starting a new array:
5354 */
5355
5356 mddev = bdev->bd_disk->private_data;
5357
5358 if (!mddev) {
5359 BUG();
5360 goto abort;
5361 }
5362
5363 err = mddev_lock(mddev);
5364 if (err) {
5365 printk(KERN_INFO
5366 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5367 err, cmd);
5368 goto abort;
5369 }
5370
5371 switch (cmd)
5372 {
5373 case SET_ARRAY_INFO:
5374 {
5375 mdu_array_info_t info;
5376 if (!arg)
5377 memset(&info, 0, sizeof(info));
5378 else if (copy_from_user(&info, argp, sizeof(info))) {
5379 err = -EFAULT;
5380 goto abort_unlock;
5381 }
5382 if (mddev->pers) {
5383 err = update_array_info(mddev, &info);
5384 if (err) {
5385 printk(KERN_WARNING "md: couldn't update"
5386 " array info. %d\n", err);
5387 goto abort_unlock;
5388 }
5389 goto done_unlock;
5390 }
5391 if (!list_empty(&mddev->disks)) {
5392 printk(KERN_WARNING
5393 "md: array %s already has disks!\n",
5394 mdname(mddev));
5395 err = -EBUSY;
5396 goto abort_unlock;
5397 }
5398 if (mddev->raid_disks) {
5399 printk(KERN_WARNING
5400 "md: array %s already initialised!\n",
5401 mdname(mddev));
5402 err = -EBUSY;
5403 goto abort_unlock;
5404 }
5405 err = set_array_info(mddev, &info);
5406 if (err) {
5407 printk(KERN_WARNING "md: couldn't set"
5408 " array info. %d\n", err);
5409 goto abort_unlock;
5410 }
5411 }
5412 goto done_unlock;
5413
5414 default:;
5415 }
5416
5417 /*
5418 * Commands querying/configuring an existing array:
5419 */
5420 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5421 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5422 if ((!mddev->raid_disks && !mddev->external)
5423 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5424 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5425 && cmd != GET_BITMAP_FILE) {
5426 err = -ENODEV;
5427 goto abort_unlock;
5428 }
5429
5430 /*
5431 * Commands even a read-only array can execute:
5432 */
5433 switch (cmd)
5434 {
5435 case GET_ARRAY_INFO:
5436 err = get_array_info(mddev, argp);
5437 goto done_unlock;
5438
5439 case GET_BITMAP_FILE:
5440 err = get_bitmap_file(mddev, argp);
5441 goto done_unlock;
5442
5443 case GET_DISK_INFO:
5444 err = get_disk_info(mddev, argp);
5445 goto done_unlock;
5446
5447 case RESTART_ARRAY_RW:
5448 err = restart_array(mddev);
5449 goto done_unlock;
5450
5451 case STOP_ARRAY:
5452 err = do_md_stop(mddev, 0, 1);
5453 goto done_unlock;
5454
5455 case STOP_ARRAY_RO:
5456 err = do_md_stop(mddev, 1, 1);
5457 goto done_unlock;
5458
5459 }
5460
5461 /*
5462 * The remaining ioctls are changing the state of the
5463 * superblock, so we do not allow them on read-only arrays.
5464 * However non-MD ioctls (e.g. get-size) will still come through
5465 * here and hit the 'default' below, so only disallow
5466 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5467 */
5468 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5469 if (mddev->ro == 2) {
5470 mddev->ro = 0;
5471 sysfs_notify_dirent(mddev->sysfs_state);
5472 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5473 md_wakeup_thread(mddev->thread);
5474 } else {
5475 err = -EROFS;
5476 goto abort_unlock;
5477 }
5478 }
5479
5480 switch (cmd)
5481 {
5482 case ADD_NEW_DISK:
5483 {
5484 mdu_disk_info_t info;
5485 if (copy_from_user(&info, argp, sizeof(info)))
5486 err = -EFAULT;
5487 else
5488 err = add_new_disk(mddev, &info);
5489 goto done_unlock;
5490 }
5491
5492 case HOT_REMOVE_DISK:
5493 err = hot_remove_disk(mddev, new_decode_dev(arg));
5494 goto done_unlock;
5495
5496 case HOT_ADD_DISK:
5497 err = hot_add_disk(mddev, new_decode_dev(arg));
5498 goto done_unlock;
5499
5500 case SET_DISK_FAULTY:
5501 err = set_disk_faulty(mddev, new_decode_dev(arg));
5502 goto done_unlock;
5503
5504 case RUN_ARRAY:
5505 err = do_md_run(mddev);
5506 goto done_unlock;
5507
5508 case SET_BITMAP_FILE:
5509 err = set_bitmap_file(mddev, (int)arg);
5510 goto done_unlock;
5511
5512 default:
5513 err = -EINVAL;
5514 goto abort_unlock;
5515 }
5516
5517 done_unlock:
5518 abort_unlock:
5519 if (mddev->hold_active == UNTIL_IOCTL &&
5520 err != -EINVAL)
5521 mddev->hold_active = 0;
5522 mddev_unlock(mddev);
5523
5524 return err;
5525 done:
5526 if (err)
5527 MD_BUG();
5528 abort:
5529 return err;
5530 }
5531
5532 static int md_open(struct block_device *bdev, fmode_t mode)
5533 {
5534 /*
5535 * Succeed if we can lock the mddev, which confirms that
5536 * it isn't being stopped right now.
5537 */
5538 mddev_t *mddev = mddev_find(bdev->bd_dev);
5539 int err;
5540
5541 if (mddev->gendisk != bdev->bd_disk) {
5542 /* we are racing with mddev_put which is discarding this
5543 * bd_disk.
5544 */
5545 mddev_put(mddev);
5546 /* Wait until bdev->bd_disk is definitely gone */
5547 flush_scheduled_work();
5548 /* Then retry the open from the top */
5549 return -ERESTARTSYS;
5550 }
5551 BUG_ON(mddev != bdev->bd_disk->private_data);
5552
5553 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
5554 goto out;
5555
5556 err = 0;
5557 atomic_inc(&mddev->openers);
5558 mutex_unlock(&mddev->open_mutex);
5559
5560 check_disk_change(bdev);
5561 out:
5562 return err;
5563 }
5564
5565 static int md_release(struct gendisk *disk, fmode_t mode)
5566 {
5567 mddev_t *mddev = disk->private_data;
5568
5569 BUG_ON(!mddev);
5570 atomic_dec(&mddev->openers);
5571 mddev_put(mddev);
5572
5573 return 0;
5574 }
5575
5576 static int md_media_changed(struct gendisk *disk)
5577 {
5578 mddev_t *mddev = disk->private_data;
5579
5580 return mddev->changed;
5581 }
5582
5583 static int md_revalidate(struct gendisk *disk)
5584 {
5585 mddev_t *mddev = disk->private_data;
5586
5587 mddev->changed = 0;
5588 return 0;
5589 }
5590 static const struct block_device_operations md_fops =
5591 {
5592 .owner = THIS_MODULE,
5593 .open = md_open,
5594 .release = md_release,
5595 .ioctl = md_ioctl,
5596 .getgeo = md_getgeo,
5597 .media_changed = md_media_changed,
5598 .revalidate_disk= md_revalidate,
5599 };
5600
5601 static int md_thread(void * arg)
5602 {
5603 mdk_thread_t *thread = arg;
5604
5605 /*
5606 * md_thread is a 'system-thread', it's priority should be very
5607 * high. We avoid resource deadlocks individually in each
5608 * raid personality. (RAID5 does preallocation) We also use RR and
5609 * the very same RT priority as kswapd, thus we will never get
5610 * into a priority inversion deadlock.
5611 *
5612 * we definitely have to have equal or higher priority than
5613 * bdflush, otherwise bdflush will deadlock if there are too
5614 * many dirty RAID5 blocks.
5615 */
5616
5617 allow_signal(SIGKILL);
5618 while (!kthread_should_stop()) {
5619
5620 /* We need to wait INTERRUPTIBLE so that
5621 * we don't add to the load-average.
5622 * That means we need to be sure no signals are
5623 * pending
5624 */
5625 if (signal_pending(current))
5626 flush_signals(current);
5627
5628 wait_event_interruptible_timeout
5629 (thread->wqueue,
5630 test_bit(THREAD_WAKEUP, &thread->flags)
5631 || kthread_should_stop(),
5632 thread->timeout);
5633
5634 clear_bit(THREAD_WAKEUP, &thread->flags);
5635
5636 thread->run(thread->mddev);
5637 }
5638
5639 return 0;
5640 }
5641
5642 void md_wakeup_thread(mdk_thread_t *thread)
5643 {
5644 if (thread) {
5645 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5646 set_bit(THREAD_WAKEUP, &thread->flags);
5647 wake_up(&thread->wqueue);
5648 }
5649 }
5650
5651 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5652 const char *name)
5653 {
5654 mdk_thread_t *thread;
5655
5656 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5657 if (!thread)
5658 return NULL;
5659
5660 init_waitqueue_head(&thread->wqueue);
5661
5662 thread->run = run;
5663 thread->mddev = mddev;
5664 thread->timeout = MAX_SCHEDULE_TIMEOUT;
5665 thread->tsk = kthread_run(md_thread, thread,
5666 "%s_%s",
5667 mdname(thread->mddev),
5668 name ?: mddev->pers->name);
5669 if (IS_ERR(thread->tsk)) {
5670 kfree(thread);
5671 return NULL;
5672 }
5673 return thread;
5674 }
5675
5676 void md_unregister_thread(mdk_thread_t *thread)
5677 {
5678 if (!thread)
5679 return;
5680 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5681
5682 kthread_stop(thread->tsk);
5683 kfree(thread);
5684 }
5685
5686 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5687 {
5688 if (!mddev) {
5689 MD_BUG();
5690 return;
5691 }
5692
5693 if (!rdev || test_bit(Faulty, &rdev->flags))
5694 return;
5695
5696 if (mddev->external)
5697 set_bit(Blocked, &rdev->flags);
5698 /*
5699 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5700 mdname(mddev),
5701 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5702 __builtin_return_address(0),__builtin_return_address(1),
5703 __builtin_return_address(2),__builtin_return_address(3));
5704 */
5705 if (!mddev->pers)
5706 return;
5707 if (!mddev->pers->error_handler)
5708 return;
5709 mddev->pers->error_handler(mddev,rdev);
5710 if (mddev->degraded)
5711 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5712 set_bit(StateChanged, &rdev->flags);
5713 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5714 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5715 md_wakeup_thread(mddev->thread);
5716 md_new_event_inintr(mddev);
5717 }
5718
5719 /* seq_file implementation /proc/mdstat */
5720
5721 static void status_unused(struct seq_file *seq)
5722 {
5723 int i = 0;
5724 mdk_rdev_t *rdev;
5725
5726 seq_printf(seq, "unused devices: ");
5727
5728 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
5729 char b[BDEVNAME_SIZE];
5730 i++;
5731 seq_printf(seq, "%s ",
5732 bdevname(rdev->bdev,b));
5733 }
5734 if (!i)
5735 seq_printf(seq, "<none>");
5736
5737 seq_printf(seq, "\n");
5738 }
5739
5740
5741 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5742 {
5743 sector_t max_sectors, resync, res;
5744 unsigned long dt, db;
5745 sector_t rt;
5746 int scale;
5747 unsigned int per_milli;
5748
5749 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
5750
5751 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5752 max_sectors = mddev->resync_max_sectors;
5753 else
5754 max_sectors = mddev->dev_sectors;
5755
5756 /*
5757 * Should not happen.
5758 */
5759 if (!max_sectors) {
5760 MD_BUG();
5761 return;
5762 }
5763 /* Pick 'scale' such that (resync>>scale)*1000 will fit
5764 * in a sector_t, and (max_sectors>>scale) will fit in a
5765 * u32, as those are the requirements for sector_div.
5766 * Thus 'scale' must be at least 10
5767 */
5768 scale = 10;
5769 if (sizeof(sector_t) > sizeof(unsigned long)) {
5770 while ( max_sectors/2 > (1ULL<<(scale+32)))
5771 scale++;
5772 }
5773 res = (resync>>scale)*1000;
5774 sector_div(res, (u32)((max_sectors>>scale)+1));
5775
5776 per_milli = res;
5777 {
5778 int i, x = per_milli/50, y = 20-x;
5779 seq_printf(seq, "[");
5780 for (i = 0; i < x; i++)
5781 seq_printf(seq, "=");
5782 seq_printf(seq, ">");
5783 for (i = 0; i < y; i++)
5784 seq_printf(seq, ".");
5785 seq_printf(seq, "] ");
5786 }
5787 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5788 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5789 "reshape" :
5790 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5791 "check" :
5792 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5793 "resync" : "recovery"))),
5794 per_milli/10, per_milli % 10,
5795 (unsigned long long) resync/2,
5796 (unsigned long long) max_sectors/2);
5797
5798 /*
5799 * dt: time from mark until now
5800 * db: blocks written from mark until now
5801 * rt: remaining time
5802 *
5803 * rt is a sector_t, so could be 32bit or 64bit.
5804 * So we divide before multiply in case it is 32bit and close
5805 * to the limit.
5806 * We scale the divisor (db) by 32 to avoid loosing precision
5807 * near the end of resync when the number of remaining sectors
5808 * is close to 'db'.
5809 * We then divide rt by 32 after multiplying by db to compensate.
5810 * The '+1' avoids division by zero if db is very small.
5811 */
5812 dt = ((jiffies - mddev->resync_mark) / HZ);
5813 if (!dt) dt++;
5814 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5815 - mddev->resync_mark_cnt;
5816
5817 rt = max_sectors - resync; /* number of remaining sectors */
5818 sector_div(rt, db/32+1);
5819 rt *= dt;
5820 rt >>= 5;
5821
5822 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
5823 ((unsigned long)rt % 60)/6);
5824
5825 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5826 }
5827
5828 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5829 {
5830 struct list_head *tmp;
5831 loff_t l = *pos;
5832 mddev_t *mddev;
5833
5834 if (l >= 0x10000)
5835 return NULL;
5836 if (!l--)
5837 /* header */
5838 return (void*)1;
5839
5840 spin_lock(&all_mddevs_lock);
5841 list_for_each(tmp,&all_mddevs)
5842 if (!l--) {
5843 mddev = list_entry(tmp, mddev_t, all_mddevs);
5844 mddev_get(mddev);
5845 spin_unlock(&all_mddevs_lock);
5846 return mddev;
5847 }
5848 spin_unlock(&all_mddevs_lock);
5849 if (!l--)
5850 return (void*)2;/* tail */
5851 return NULL;
5852 }
5853
5854 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5855 {
5856 struct list_head *tmp;
5857 mddev_t *next_mddev, *mddev = v;
5858
5859 ++*pos;
5860 if (v == (void*)2)
5861 return NULL;
5862
5863 spin_lock(&all_mddevs_lock);
5864 if (v == (void*)1)
5865 tmp = all_mddevs.next;
5866 else
5867 tmp = mddev->all_mddevs.next;
5868 if (tmp != &all_mddevs)
5869 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5870 else {
5871 next_mddev = (void*)2;
5872 *pos = 0x10000;
5873 }
5874 spin_unlock(&all_mddevs_lock);
5875
5876 if (v != (void*)1)
5877 mddev_put(mddev);
5878 return next_mddev;
5879
5880 }
5881
5882 static void md_seq_stop(struct seq_file *seq, void *v)
5883 {
5884 mddev_t *mddev = v;
5885
5886 if (mddev && v != (void*)1 && v != (void*)2)
5887 mddev_put(mddev);
5888 }
5889
5890 struct mdstat_info {
5891 int event;
5892 };
5893
5894 static int md_seq_show(struct seq_file *seq, void *v)
5895 {
5896 mddev_t *mddev = v;
5897 sector_t sectors;
5898 mdk_rdev_t *rdev;
5899 struct mdstat_info *mi = seq->private;
5900 struct bitmap *bitmap;
5901
5902 if (v == (void*)1) {
5903 struct mdk_personality *pers;
5904 seq_printf(seq, "Personalities : ");
5905 spin_lock(&pers_lock);
5906 list_for_each_entry(pers, &pers_list, list)
5907 seq_printf(seq, "[%s] ", pers->name);
5908
5909 spin_unlock(&pers_lock);
5910 seq_printf(seq, "\n");
5911 mi->event = atomic_read(&md_event_count);
5912 return 0;
5913 }
5914 if (v == (void*)2) {
5915 status_unused(seq);
5916 return 0;
5917 }
5918
5919 if (mddev_lock(mddev) < 0)
5920 return -EINTR;
5921
5922 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5923 seq_printf(seq, "%s : %sactive", mdname(mddev),
5924 mddev->pers ? "" : "in");
5925 if (mddev->pers) {
5926 if (mddev->ro==1)
5927 seq_printf(seq, " (read-only)");
5928 if (mddev->ro==2)
5929 seq_printf(seq, " (auto-read-only)");
5930 seq_printf(seq, " %s", mddev->pers->name);
5931 }
5932
5933 sectors = 0;
5934 list_for_each_entry(rdev, &mddev->disks, same_set) {
5935 char b[BDEVNAME_SIZE];
5936 seq_printf(seq, " %s[%d]",
5937 bdevname(rdev->bdev,b), rdev->desc_nr);
5938 if (test_bit(WriteMostly, &rdev->flags))
5939 seq_printf(seq, "(W)");
5940 if (test_bit(Faulty, &rdev->flags)) {
5941 seq_printf(seq, "(F)");
5942 continue;
5943 } else if (rdev->raid_disk < 0)
5944 seq_printf(seq, "(S)"); /* spare */
5945 sectors += rdev->sectors;
5946 }
5947
5948 if (!list_empty(&mddev->disks)) {
5949 if (mddev->pers)
5950 seq_printf(seq, "\n %llu blocks",
5951 (unsigned long long)
5952 mddev->array_sectors / 2);
5953 else
5954 seq_printf(seq, "\n %llu blocks",
5955 (unsigned long long)sectors / 2);
5956 }
5957 if (mddev->persistent) {
5958 if (mddev->major_version != 0 ||
5959 mddev->minor_version != 90) {
5960 seq_printf(seq," super %d.%d",
5961 mddev->major_version,
5962 mddev->minor_version);
5963 }
5964 } else if (mddev->external)
5965 seq_printf(seq, " super external:%s",
5966 mddev->metadata_type);
5967 else
5968 seq_printf(seq, " super non-persistent");
5969
5970 if (mddev->pers) {
5971 mddev->pers->status(seq, mddev);
5972 seq_printf(seq, "\n ");
5973 if (mddev->pers->sync_request) {
5974 if (mddev->curr_resync > 2) {
5975 status_resync(seq, mddev);
5976 seq_printf(seq, "\n ");
5977 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5978 seq_printf(seq, "\tresync=DELAYED\n ");
5979 else if (mddev->recovery_cp < MaxSector)
5980 seq_printf(seq, "\tresync=PENDING\n ");
5981 }
5982 } else
5983 seq_printf(seq, "\n ");
5984
5985 if ((bitmap = mddev->bitmap)) {
5986 unsigned long chunk_kb;
5987 unsigned long flags;
5988 spin_lock_irqsave(&bitmap->lock, flags);
5989 chunk_kb = bitmap->chunksize >> 10;
5990 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5991 "%lu%s chunk",
5992 bitmap->pages - bitmap->missing_pages,
5993 bitmap->pages,
5994 (bitmap->pages - bitmap->missing_pages)
5995 << (PAGE_SHIFT - 10),
5996 chunk_kb ? chunk_kb : bitmap->chunksize,
5997 chunk_kb ? "KB" : "B");
5998 if (bitmap->file) {
5999 seq_printf(seq, ", file: ");
6000 seq_path(seq, &bitmap->file->f_path, " \t\n");
6001 }
6002
6003 seq_printf(seq, "\n");
6004 spin_unlock_irqrestore(&bitmap->lock, flags);
6005 }
6006
6007 seq_printf(seq, "\n");
6008 }
6009 mddev_unlock(mddev);
6010
6011 return 0;
6012 }
6013
6014 static const struct seq_operations md_seq_ops = {
6015 .start = md_seq_start,
6016 .next = md_seq_next,
6017 .stop = md_seq_stop,
6018 .show = md_seq_show,
6019 };
6020
6021 static int md_seq_open(struct inode *inode, struct file *file)
6022 {
6023 int error;
6024 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6025 if (mi == NULL)
6026 return -ENOMEM;
6027
6028 error = seq_open(file, &md_seq_ops);
6029 if (error)
6030 kfree(mi);
6031 else {
6032 struct seq_file *p = file->private_data;
6033 p->private = mi;
6034 mi->event = atomic_read(&md_event_count);
6035 }
6036 return error;
6037 }
6038
6039 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6040 {
6041 struct seq_file *m = filp->private_data;
6042 struct mdstat_info *mi = m->private;
6043 int mask;
6044
6045 poll_wait(filp, &md_event_waiters, wait);
6046
6047 /* always allow read */
6048 mask = POLLIN | POLLRDNORM;
6049
6050 if (mi->event != atomic_read(&md_event_count))
6051 mask |= POLLERR | POLLPRI;
6052 return mask;
6053 }
6054
6055 static const struct file_operations md_seq_fops = {
6056 .owner = THIS_MODULE,
6057 .open = md_seq_open,
6058 .read = seq_read,
6059 .llseek = seq_lseek,
6060 .release = seq_release_private,
6061 .poll = mdstat_poll,
6062 };
6063
6064 int register_md_personality(struct mdk_personality *p)
6065 {
6066 spin_lock(&pers_lock);
6067 list_add_tail(&p->list, &pers_list);
6068 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6069 spin_unlock(&pers_lock);
6070 return 0;
6071 }
6072
6073 int unregister_md_personality(struct mdk_personality *p)
6074 {
6075 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6076 spin_lock(&pers_lock);
6077 list_del_init(&p->list);
6078 spin_unlock(&pers_lock);
6079 return 0;
6080 }
6081
6082 static int is_mddev_idle(mddev_t *mddev, int init)
6083 {
6084 mdk_rdev_t * rdev;
6085 int idle;
6086 int curr_events;
6087
6088 idle = 1;
6089 rcu_read_lock();
6090 rdev_for_each_rcu(rdev, mddev) {
6091 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6092 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6093 (int)part_stat_read(&disk->part0, sectors[1]) -
6094 atomic_read(&disk->sync_io);
6095 /* sync IO will cause sync_io to increase before the disk_stats
6096 * as sync_io is counted when a request starts, and
6097 * disk_stats is counted when it completes.
6098 * So resync activity will cause curr_events to be smaller than
6099 * when there was no such activity.
6100 * non-sync IO will cause disk_stat to increase without
6101 * increasing sync_io so curr_events will (eventually)
6102 * be larger than it was before. Once it becomes
6103 * substantially larger, the test below will cause
6104 * the array to appear non-idle, and resync will slow
6105 * down.
6106 * If there is a lot of outstanding resync activity when
6107 * we set last_event to curr_events, then all that activity
6108 * completing might cause the array to appear non-idle
6109 * and resync will be slowed down even though there might
6110 * not have been non-resync activity. This will only
6111 * happen once though. 'last_events' will soon reflect
6112 * the state where there is little or no outstanding
6113 * resync requests, and further resync activity will
6114 * always make curr_events less than last_events.
6115 *
6116 */
6117 if (init || curr_events - rdev->last_events > 64) {
6118 rdev->last_events = curr_events;
6119 idle = 0;
6120 }
6121 }
6122 rcu_read_unlock();
6123 return idle;
6124 }
6125
6126 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6127 {
6128 /* another "blocks" (512byte) blocks have been synced */
6129 atomic_sub(blocks, &mddev->recovery_active);
6130 wake_up(&mddev->recovery_wait);
6131 if (!ok) {
6132 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6133 md_wakeup_thread(mddev->thread);
6134 // stop recovery, signal do_sync ....
6135 }
6136 }
6137
6138
6139 /* md_write_start(mddev, bi)
6140 * If we need to update some array metadata (e.g. 'active' flag
6141 * in superblock) before writing, schedule a superblock update
6142 * and wait for it to complete.
6143 */
6144 void md_write_start(mddev_t *mddev, struct bio *bi)
6145 {
6146 int did_change = 0;
6147 if (bio_data_dir(bi) != WRITE)
6148 return;
6149
6150 BUG_ON(mddev->ro == 1);
6151 if (mddev->ro == 2) {
6152 /* need to switch to read/write */
6153 mddev->ro = 0;
6154 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6155 md_wakeup_thread(mddev->thread);
6156 md_wakeup_thread(mddev->sync_thread);
6157 did_change = 1;
6158 }
6159 atomic_inc(&mddev->writes_pending);
6160 if (mddev->safemode == 1)
6161 mddev->safemode = 0;
6162 if (mddev->in_sync) {
6163 spin_lock_irq(&mddev->write_lock);
6164 if (mddev->in_sync) {
6165 mddev->in_sync = 0;
6166 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6167 md_wakeup_thread(mddev->thread);
6168 did_change = 1;
6169 }
6170 spin_unlock_irq(&mddev->write_lock);
6171 }
6172 if (did_change)
6173 sysfs_notify_dirent(mddev->sysfs_state);
6174 wait_event(mddev->sb_wait,
6175 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
6176 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6177 }
6178
6179 void md_write_end(mddev_t *mddev)
6180 {
6181 if (atomic_dec_and_test(&mddev->writes_pending)) {
6182 if (mddev->safemode == 2)
6183 md_wakeup_thread(mddev->thread);
6184 else if (mddev->safemode_delay)
6185 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6186 }
6187 }
6188
6189 /* md_allow_write(mddev)
6190 * Calling this ensures that the array is marked 'active' so that writes
6191 * may proceed without blocking. It is important to call this before
6192 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6193 * Must be called with mddev_lock held.
6194 *
6195 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6196 * is dropped, so return -EAGAIN after notifying userspace.
6197 */
6198 int md_allow_write(mddev_t *mddev)
6199 {
6200 if (!mddev->pers)
6201 return 0;
6202 if (mddev->ro)
6203 return 0;
6204 if (!mddev->pers->sync_request)
6205 return 0;
6206
6207 spin_lock_irq(&mddev->write_lock);
6208 if (mddev->in_sync) {
6209 mddev->in_sync = 0;
6210 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6211 if (mddev->safemode_delay &&
6212 mddev->safemode == 0)
6213 mddev->safemode = 1;
6214 spin_unlock_irq(&mddev->write_lock);
6215 md_update_sb(mddev, 0);
6216 sysfs_notify_dirent(mddev->sysfs_state);
6217 } else
6218 spin_unlock_irq(&mddev->write_lock);
6219
6220 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
6221 return -EAGAIN;
6222 else
6223 return 0;
6224 }
6225 EXPORT_SYMBOL_GPL(md_allow_write);
6226
6227 #define SYNC_MARKS 10
6228 #define SYNC_MARK_STEP (3*HZ)
6229 void md_do_sync(mddev_t *mddev)
6230 {
6231 mddev_t *mddev2;
6232 unsigned int currspeed = 0,
6233 window;
6234 sector_t max_sectors,j, io_sectors;
6235 unsigned long mark[SYNC_MARKS];
6236 sector_t mark_cnt[SYNC_MARKS];
6237 int last_mark,m;
6238 struct list_head *tmp;
6239 sector_t last_check;
6240 int skipped = 0;
6241 mdk_rdev_t *rdev;
6242 char *desc;
6243
6244 /* just incase thread restarts... */
6245 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6246 return;
6247 if (mddev->ro) /* never try to sync a read-only array */
6248 return;
6249
6250 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6251 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6252 desc = "data-check";
6253 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6254 desc = "requested-resync";
6255 else
6256 desc = "resync";
6257 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6258 desc = "reshape";
6259 else
6260 desc = "recovery";
6261
6262 /* we overload curr_resync somewhat here.
6263 * 0 == not engaged in resync at all
6264 * 2 == checking that there is no conflict with another sync
6265 * 1 == like 2, but have yielded to allow conflicting resync to
6266 * commense
6267 * other == active in resync - this many blocks
6268 *
6269 * Before starting a resync we must have set curr_resync to
6270 * 2, and then checked that every "conflicting" array has curr_resync
6271 * less than ours. When we find one that is the same or higher
6272 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6273 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6274 * This will mean we have to start checking from the beginning again.
6275 *
6276 */
6277
6278 do {
6279 mddev->curr_resync = 2;
6280
6281 try_again:
6282 if (kthread_should_stop()) {
6283 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6284 goto skip;
6285 }
6286 for_each_mddev(mddev2, tmp) {
6287 if (mddev2 == mddev)
6288 continue;
6289 if (!mddev->parallel_resync
6290 && mddev2->curr_resync
6291 && match_mddev_units(mddev, mddev2)) {
6292 DEFINE_WAIT(wq);
6293 if (mddev < mddev2 && mddev->curr_resync == 2) {
6294 /* arbitrarily yield */
6295 mddev->curr_resync = 1;
6296 wake_up(&resync_wait);
6297 }
6298 if (mddev > mddev2 && mddev->curr_resync == 1)
6299 /* no need to wait here, we can wait the next
6300 * time 'round when curr_resync == 2
6301 */
6302 continue;
6303 /* We need to wait 'interruptible' so as not to
6304 * contribute to the load average, and not to
6305 * be caught by 'softlockup'
6306 */
6307 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6308 if (!kthread_should_stop() &&
6309 mddev2->curr_resync >= mddev->curr_resync) {
6310 printk(KERN_INFO "md: delaying %s of %s"
6311 " until %s has finished (they"
6312 " share one or more physical units)\n",
6313 desc, mdname(mddev), mdname(mddev2));
6314 mddev_put(mddev2);
6315 if (signal_pending(current))
6316 flush_signals(current);
6317 schedule();
6318 finish_wait(&resync_wait, &wq);
6319 goto try_again;
6320 }
6321 finish_wait(&resync_wait, &wq);
6322 }
6323 }
6324 } while (mddev->curr_resync < 2);
6325
6326 j = 0;
6327 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6328 /* resync follows the size requested by the personality,
6329 * which defaults to physical size, but can be virtual size
6330 */
6331 max_sectors = mddev->resync_max_sectors;
6332 mddev->resync_mismatches = 0;
6333 /* we don't use the checkpoint if there's a bitmap */
6334 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6335 j = mddev->resync_min;
6336 else if (!mddev->bitmap)
6337 j = mddev->recovery_cp;
6338
6339 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6340 max_sectors = mddev->dev_sectors;
6341 else {
6342 /* recovery follows the physical size of devices */
6343 max_sectors = mddev->dev_sectors;
6344 j = MaxSector;
6345 list_for_each_entry(rdev, &mddev->disks, same_set)
6346 if (rdev->raid_disk >= 0 &&
6347 !test_bit(Faulty, &rdev->flags) &&
6348 !test_bit(In_sync, &rdev->flags) &&
6349 rdev->recovery_offset < j)
6350 j = rdev->recovery_offset;
6351 }
6352
6353 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6354 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6355 " %d KB/sec/disk.\n", speed_min(mddev));
6356 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6357 "(but not more than %d KB/sec) for %s.\n",
6358 speed_max(mddev), desc);
6359
6360 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6361
6362 io_sectors = 0;
6363 for (m = 0; m < SYNC_MARKS; m++) {
6364 mark[m] = jiffies;
6365 mark_cnt[m] = io_sectors;
6366 }
6367 last_mark = 0;
6368 mddev->resync_mark = mark[last_mark];
6369 mddev->resync_mark_cnt = mark_cnt[last_mark];
6370
6371 /*
6372 * Tune reconstruction:
6373 */
6374 window = 32*(PAGE_SIZE/512);
6375 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6376 window/2,(unsigned long long) max_sectors/2);
6377
6378 atomic_set(&mddev->recovery_active, 0);
6379 last_check = 0;
6380
6381 if (j>2) {
6382 printk(KERN_INFO
6383 "md: resuming %s of %s from checkpoint.\n",
6384 desc, mdname(mddev));
6385 mddev->curr_resync = j;
6386 }
6387
6388 while (j < max_sectors) {
6389 sector_t sectors;
6390
6391 skipped = 0;
6392
6393 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6394 ((mddev->curr_resync > mddev->curr_resync_completed &&
6395 (mddev->curr_resync - mddev->curr_resync_completed)
6396 > (max_sectors >> 4)) ||
6397 (j - mddev->curr_resync_completed)*2
6398 >= mddev->resync_max - mddev->curr_resync_completed
6399 )) {
6400 /* time to update curr_resync_completed */
6401 blk_unplug(mddev->queue);
6402 wait_event(mddev->recovery_wait,
6403 atomic_read(&mddev->recovery_active) == 0);
6404 mddev->curr_resync_completed =
6405 mddev->curr_resync;
6406 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6407 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6408 }
6409
6410 while (j >= mddev->resync_max && !kthread_should_stop()) {
6411 /* As this condition is controlled by user-space,
6412 * we can block indefinitely, so use '_interruptible'
6413 * to avoid triggering warnings.
6414 */
6415 flush_signals(current); /* just in case */
6416 wait_event_interruptible(mddev->recovery_wait,
6417 mddev->resync_max > j
6418 || kthread_should_stop());
6419 }
6420
6421 if (kthread_should_stop())
6422 goto interrupted;
6423
6424 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6425 currspeed < speed_min(mddev));
6426 if (sectors == 0) {
6427 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6428 goto out;
6429 }
6430
6431 if (!skipped) { /* actual IO requested */
6432 io_sectors += sectors;
6433 atomic_add(sectors, &mddev->recovery_active);
6434 }
6435
6436 j += sectors;
6437 if (j>1) mddev->curr_resync = j;
6438 mddev->curr_mark_cnt = io_sectors;
6439 if (last_check == 0)
6440 /* this is the earliers that rebuilt will be
6441 * visible in /proc/mdstat
6442 */
6443 md_new_event(mddev);
6444
6445 if (last_check + window > io_sectors || j == max_sectors)
6446 continue;
6447
6448 last_check = io_sectors;
6449
6450 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6451 break;
6452
6453 repeat:
6454 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6455 /* step marks */
6456 int next = (last_mark+1) % SYNC_MARKS;
6457
6458 mddev->resync_mark = mark[next];
6459 mddev->resync_mark_cnt = mark_cnt[next];
6460 mark[next] = jiffies;
6461 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6462 last_mark = next;
6463 }
6464
6465
6466 if (kthread_should_stop())
6467 goto interrupted;
6468
6469
6470 /*
6471 * this loop exits only if either when we are slower than
6472 * the 'hard' speed limit, or the system was IO-idle for
6473 * a jiffy.
6474 * the system might be non-idle CPU-wise, but we only care
6475 * about not overloading the IO subsystem. (things like an
6476 * e2fsck being done on the RAID array should execute fast)
6477 */
6478 blk_unplug(mddev->queue);
6479 cond_resched();
6480
6481 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6482 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6483
6484 if (currspeed > speed_min(mddev)) {
6485 if ((currspeed > speed_max(mddev)) ||
6486 !is_mddev_idle(mddev, 0)) {
6487 msleep(500);
6488 goto repeat;
6489 }
6490 }
6491 }
6492 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6493 /*
6494 * this also signals 'finished resyncing' to md_stop
6495 */
6496 out:
6497 blk_unplug(mddev->queue);
6498
6499 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6500
6501 /* tell personality that we are finished */
6502 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6503
6504 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6505 mddev->curr_resync > 2) {
6506 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6507 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6508 if (mddev->curr_resync >= mddev->recovery_cp) {
6509 printk(KERN_INFO
6510 "md: checkpointing %s of %s.\n",
6511 desc, mdname(mddev));
6512 mddev->recovery_cp = mddev->curr_resync;
6513 }
6514 } else
6515 mddev->recovery_cp = MaxSector;
6516 } else {
6517 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6518 mddev->curr_resync = MaxSector;
6519 list_for_each_entry(rdev, &mddev->disks, same_set)
6520 if (rdev->raid_disk >= 0 &&
6521 !test_bit(Faulty, &rdev->flags) &&
6522 !test_bit(In_sync, &rdev->flags) &&
6523 rdev->recovery_offset < mddev->curr_resync)
6524 rdev->recovery_offset = mddev->curr_resync;
6525 }
6526 }
6527 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6528
6529 skip:
6530 mddev->curr_resync = 0;
6531 mddev->curr_resync_completed = 0;
6532 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6533 /* We completed so max setting can be forgotten. */
6534 mddev->resync_max = MaxSector;
6535 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6536 wake_up(&resync_wait);
6537 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6538 md_wakeup_thread(mddev->thread);
6539 return;
6540
6541 interrupted:
6542 /*
6543 * got a signal, exit.
6544 */
6545 printk(KERN_INFO
6546 "md: md_do_sync() got signal ... exiting\n");
6547 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6548 goto out;
6549
6550 }
6551 EXPORT_SYMBOL_GPL(md_do_sync);
6552
6553
6554 static int remove_and_add_spares(mddev_t *mddev)
6555 {
6556 mdk_rdev_t *rdev;
6557 int spares = 0;
6558
6559 mddev->curr_resync_completed = 0;
6560
6561 list_for_each_entry(rdev, &mddev->disks, same_set)
6562 if (rdev->raid_disk >= 0 &&
6563 !test_bit(Blocked, &rdev->flags) &&
6564 (test_bit(Faulty, &rdev->flags) ||
6565 ! test_bit(In_sync, &rdev->flags)) &&
6566 atomic_read(&rdev->nr_pending)==0) {
6567 if (mddev->pers->hot_remove_disk(
6568 mddev, rdev->raid_disk)==0) {
6569 char nm[20];
6570 sprintf(nm,"rd%d", rdev->raid_disk);
6571 sysfs_remove_link(&mddev->kobj, nm);
6572 rdev->raid_disk = -1;
6573 }
6574 }
6575
6576 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
6577 list_for_each_entry(rdev, &mddev->disks, same_set) {
6578 if (rdev->raid_disk >= 0 &&
6579 !test_bit(In_sync, &rdev->flags) &&
6580 !test_bit(Blocked, &rdev->flags))
6581 spares++;
6582 if (rdev->raid_disk < 0
6583 && !test_bit(Faulty, &rdev->flags)) {
6584 rdev->recovery_offset = 0;
6585 if (mddev->pers->
6586 hot_add_disk(mddev, rdev) == 0) {
6587 char nm[20];
6588 sprintf(nm, "rd%d", rdev->raid_disk);
6589 if (sysfs_create_link(&mddev->kobj,
6590 &rdev->kobj, nm))
6591 printk(KERN_WARNING
6592 "md: cannot register "
6593 "%s for %s\n",
6594 nm, mdname(mddev));
6595 spares++;
6596 md_new_event(mddev);
6597 } else
6598 break;
6599 }
6600 }
6601 }
6602 return spares;
6603 }
6604 /*
6605 * This routine is regularly called by all per-raid-array threads to
6606 * deal with generic issues like resync and super-block update.
6607 * Raid personalities that don't have a thread (linear/raid0) do not
6608 * need this as they never do any recovery or update the superblock.
6609 *
6610 * It does not do any resync itself, but rather "forks" off other threads
6611 * to do that as needed.
6612 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6613 * "->recovery" and create a thread at ->sync_thread.
6614 * When the thread finishes it sets MD_RECOVERY_DONE
6615 * and wakeups up this thread which will reap the thread and finish up.
6616 * This thread also removes any faulty devices (with nr_pending == 0).
6617 *
6618 * The overall approach is:
6619 * 1/ if the superblock needs updating, update it.
6620 * 2/ If a recovery thread is running, don't do anything else.
6621 * 3/ If recovery has finished, clean up, possibly marking spares active.
6622 * 4/ If there are any faulty devices, remove them.
6623 * 5/ If array is degraded, try to add spares devices
6624 * 6/ If array has spares or is not in-sync, start a resync thread.
6625 */
6626 void md_check_recovery(mddev_t *mddev)
6627 {
6628 mdk_rdev_t *rdev;
6629
6630
6631 if (mddev->bitmap)
6632 bitmap_daemon_work(mddev->bitmap);
6633
6634 if (mddev->ro)
6635 return;
6636
6637 if (signal_pending(current)) {
6638 if (mddev->pers->sync_request && !mddev->external) {
6639 printk(KERN_INFO "md: %s in immediate safe mode\n",
6640 mdname(mddev));
6641 mddev->safemode = 2;
6642 }
6643 flush_signals(current);
6644 }
6645
6646 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
6647 return;
6648 if ( ! (
6649 (mddev->flags && !mddev->external) ||
6650 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
6651 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
6652 (mddev->external == 0 && mddev->safemode == 1) ||
6653 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6654 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6655 ))
6656 return;
6657
6658 if (mddev_trylock(mddev)) {
6659 int spares = 0;
6660
6661 if (mddev->ro) {
6662 /* Only thing we do on a ro array is remove
6663 * failed devices.
6664 */
6665 remove_and_add_spares(mddev);
6666 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6667 goto unlock;
6668 }
6669
6670 if (!mddev->external) {
6671 int did_change = 0;
6672 spin_lock_irq(&mddev->write_lock);
6673 if (mddev->safemode &&
6674 !atomic_read(&mddev->writes_pending) &&
6675 !mddev->in_sync &&
6676 mddev->recovery_cp == MaxSector) {
6677 mddev->in_sync = 1;
6678 did_change = 1;
6679 if (mddev->persistent)
6680 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6681 }
6682 if (mddev->safemode == 1)
6683 mddev->safemode = 0;
6684 spin_unlock_irq(&mddev->write_lock);
6685 if (did_change)
6686 sysfs_notify_dirent(mddev->sysfs_state);
6687 }
6688
6689 if (mddev->flags)
6690 md_update_sb(mddev, 0);
6691
6692 list_for_each_entry(rdev, &mddev->disks, same_set)
6693 if (test_and_clear_bit(StateChanged, &rdev->flags))
6694 sysfs_notify_dirent(rdev->sysfs_state);
6695
6696
6697 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6698 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6699 /* resync/recovery still happening */
6700 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6701 goto unlock;
6702 }
6703 if (mddev->sync_thread) {
6704 /* resync has finished, collect result */
6705 md_unregister_thread(mddev->sync_thread);
6706 mddev->sync_thread = NULL;
6707 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
6708 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
6709 /* success...*/
6710 /* activate any spares */
6711 if (mddev->pers->spare_active(mddev))
6712 sysfs_notify(&mddev->kobj, NULL,
6713 "degraded");
6714 }
6715 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6716 mddev->pers->finish_reshape)
6717 mddev->pers->finish_reshape(mddev);
6718 md_update_sb(mddev, 1);
6719
6720 /* if array is no-longer degraded, then any saved_raid_disk
6721 * information must be scrapped
6722 */
6723 if (!mddev->degraded)
6724 list_for_each_entry(rdev, &mddev->disks, same_set)
6725 rdev->saved_raid_disk = -1;
6726
6727 mddev->recovery = 0;
6728 /* flag recovery needed just to double check */
6729 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6730 sysfs_notify_dirent(mddev->sysfs_action);
6731 md_new_event(mddev);
6732 goto unlock;
6733 }
6734 /* Set RUNNING before clearing NEEDED to avoid
6735 * any transients in the value of "sync_action".
6736 */
6737 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6738 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6739 /* Clear some bits that don't mean anything, but
6740 * might be left set
6741 */
6742 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
6743 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
6744
6745 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6746 goto unlock;
6747 /* no recovery is running.
6748 * remove any failed drives, then
6749 * add spares if possible.
6750 * Spare are also removed and re-added, to allow
6751 * the personality to fail the re-add.
6752 */
6753
6754 if (mddev->reshape_position != MaxSector) {
6755 if (mddev->pers->check_reshape == NULL ||
6756 mddev->pers->check_reshape(mddev) != 0)
6757 /* Cannot proceed */
6758 goto unlock;
6759 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6760 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6761 } else if ((spares = remove_and_add_spares(mddev))) {
6762 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6763 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6764 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
6765 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6766 } else if (mddev->recovery_cp < MaxSector) {
6767 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6768 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6769 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6770 /* nothing to be done ... */
6771 goto unlock;
6772
6773 if (mddev->pers->sync_request) {
6774 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6775 /* We are adding a device or devices to an array
6776 * which has the bitmap stored on all devices.
6777 * So make sure all bitmap pages get written
6778 */
6779 bitmap_write_all(mddev->bitmap);
6780 }
6781 mddev->sync_thread = md_register_thread(md_do_sync,
6782 mddev,
6783 "resync");
6784 if (!mddev->sync_thread) {
6785 printk(KERN_ERR "%s: could not start resync"
6786 " thread...\n",
6787 mdname(mddev));
6788 /* leave the spares where they are, it shouldn't hurt */
6789 mddev->recovery = 0;
6790 } else
6791 md_wakeup_thread(mddev->sync_thread);
6792 sysfs_notify_dirent(mddev->sysfs_action);
6793 md_new_event(mddev);
6794 }
6795 unlock:
6796 if (!mddev->sync_thread) {
6797 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6798 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
6799 &mddev->recovery))
6800 if (mddev->sysfs_action)
6801 sysfs_notify_dirent(mddev->sysfs_action);
6802 }
6803 mddev_unlock(mddev);
6804 }
6805 }
6806
6807 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
6808 {
6809 sysfs_notify_dirent(rdev->sysfs_state);
6810 wait_event_timeout(rdev->blocked_wait,
6811 !test_bit(Blocked, &rdev->flags),
6812 msecs_to_jiffies(5000));
6813 rdev_dec_pending(rdev, mddev);
6814 }
6815 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
6816
6817 static int md_notify_reboot(struct notifier_block *this,
6818 unsigned long code, void *x)
6819 {
6820 struct list_head *tmp;
6821 mddev_t *mddev;
6822
6823 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
6824
6825 printk(KERN_INFO "md: stopping all md devices.\n");
6826
6827 for_each_mddev(mddev, tmp)
6828 if (mddev_trylock(mddev)) {
6829 /* Force a switch to readonly even array
6830 * appears to still be in use. Hence
6831 * the '100'.
6832 */
6833 do_md_stop(mddev, 1, 100);
6834 mddev_unlock(mddev);
6835 }
6836 /*
6837 * certain more exotic SCSI devices are known to be
6838 * volatile wrt too early system reboots. While the
6839 * right place to handle this issue is the given
6840 * driver, we do want to have a safe RAID driver ...
6841 */
6842 mdelay(1000*1);
6843 }
6844 return NOTIFY_DONE;
6845 }
6846
6847 static struct notifier_block md_notifier = {
6848 .notifier_call = md_notify_reboot,
6849 .next = NULL,
6850 .priority = INT_MAX, /* before any real devices */
6851 };
6852
6853 static void md_geninit(void)
6854 {
6855 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
6856
6857 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
6858 }
6859
6860 static int __init md_init(void)
6861 {
6862 if (register_blkdev(MD_MAJOR, "md"))
6863 return -1;
6864 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
6865 unregister_blkdev(MD_MAJOR, "md");
6866 return -1;
6867 }
6868 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
6869 md_probe, NULL, NULL);
6870 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
6871 md_probe, NULL, NULL);
6872
6873 register_reboot_notifier(&md_notifier);
6874 raid_table_header = register_sysctl_table(raid_root_table);
6875
6876 md_geninit();
6877 return 0;
6878 }
6879
6880
6881 #ifndef MODULE
6882
6883 /*
6884 * Searches all registered partitions for autorun RAID arrays
6885 * at boot time.
6886 */
6887
6888 static LIST_HEAD(all_detected_devices);
6889 struct detected_devices_node {
6890 struct list_head list;
6891 dev_t dev;
6892 };
6893
6894 void md_autodetect_dev(dev_t dev)
6895 {
6896 struct detected_devices_node *node_detected_dev;
6897
6898 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6899 if (node_detected_dev) {
6900 node_detected_dev->dev = dev;
6901 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6902 } else {
6903 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6904 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6905 }
6906 }
6907
6908
6909 static void autostart_arrays(int part)
6910 {
6911 mdk_rdev_t *rdev;
6912 struct detected_devices_node *node_detected_dev;
6913 dev_t dev;
6914 int i_scanned, i_passed;
6915
6916 i_scanned = 0;
6917 i_passed = 0;
6918
6919 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6920
6921 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6922 i_scanned++;
6923 node_detected_dev = list_entry(all_detected_devices.next,
6924 struct detected_devices_node, list);
6925 list_del(&node_detected_dev->list);
6926 dev = node_detected_dev->dev;
6927 kfree(node_detected_dev);
6928 rdev = md_import_device(dev,0, 90);
6929 if (IS_ERR(rdev))
6930 continue;
6931
6932 if (test_bit(Faulty, &rdev->flags)) {
6933 MD_BUG();
6934 continue;
6935 }
6936 set_bit(AutoDetected, &rdev->flags);
6937 list_add(&rdev->same_set, &pending_raid_disks);
6938 i_passed++;
6939 }
6940
6941 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6942 i_scanned, i_passed);
6943
6944 autorun_devices(part);
6945 }
6946
6947 #endif /* !MODULE */
6948
6949 static __exit void md_exit(void)
6950 {
6951 mddev_t *mddev;
6952 struct list_head *tmp;
6953
6954 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
6955 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6956
6957 unregister_blkdev(MD_MAJOR,"md");
6958 unregister_blkdev(mdp_major, "mdp");
6959 unregister_reboot_notifier(&md_notifier);
6960 unregister_sysctl_table(raid_table_header);
6961 remove_proc_entry("mdstat", NULL);
6962 for_each_mddev(mddev, tmp) {
6963 export_array(mddev);
6964 mddev->hold_active = 0;
6965 }
6966 }
6967
6968 subsys_initcall(md_init);
6969 module_exit(md_exit)
6970
6971 static int get_ro(char *buffer, struct kernel_param *kp)
6972 {
6973 return sprintf(buffer, "%d", start_readonly);
6974 }
6975 static int set_ro(const char *val, struct kernel_param *kp)
6976 {
6977 char *e;
6978 int num = simple_strtoul(val, &e, 10);
6979 if (*val && (*e == '\0' || *e == '\n')) {
6980 start_readonly = num;
6981 return 0;
6982 }
6983 return -EINVAL;
6984 }
6985
6986 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6987 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6988
6989 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
6990
6991 EXPORT_SYMBOL(register_md_personality);
6992 EXPORT_SYMBOL(unregister_md_personality);
6993 EXPORT_SYMBOL(md_error);
6994 EXPORT_SYMBOL(md_done_sync);
6995 EXPORT_SYMBOL(md_write_start);
6996 EXPORT_SYMBOL(md_write_end);
6997 EXPORT_SYMBOL(md_register_thread);
6998 EXPORT_SYMBOL(md_unregister_thread);
6999 EXPORT_SYMBOL(md_wakeup_thread);
7000 EXPORT_SYMBOL(md_check_recovery);
7001 MODULE_LICENSE("GPL");
7002 MODULE_ALIAS("md");
7003 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree