projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
md/raid5: avoid oops when number of devices is reduced then increased.
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / drivers / md / md.c
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/string.h>
43 #include <linux/hdreg.h>
44 #include <linux/proc_fs.h>
45 #include <linux/random.h>
46 #include <linux/reboot.h>
47 #include <linux/file.h>
48 #include <linux/compat.h>
49 #include <linux/delay.h>
50 #include <linux/raid/md_p.h>
51 #include <linux/raid/md_u.h>
52 #include <linux/slab.h>
53 #include "md.h"
54 #include "bitmap.h"
55
56 #define DEBUG 0
57 #define dprintk(x...) ((void)(DEBUG && printk(x)))
58
59
60 #ifndef MODULE
61 static void autostart_arrays(int part);
62 #endif
63
64 static LIST_HEAD(pers_list);
65 static DEFINE_SPINLOCK(pers_lock);
66
67 static void md_print_devices(void);
68
69 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
70
71 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
72
73 /*
74 * Default number of read corrections we'll attempt on an rdev
75 * before ejecting it from the array. We divide the read error
76 * count by 2 for every hour elapsed between read errors.
77 */
78 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
79 /*
80 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
81 * is 1000 KB/sec, so the extra system load does not show up that much.
82 * Increase it if you want to have more _guaranteed_ speed. Note that
83 * the RAID driver will use the maximum available bandwidth if the IO
84 * subsystem is idle. There is also an 'absolute maximum' reconstruction
85 * speed limit - in case reconstruction slows down your system despite
86 * idle IO detection.
87 *
88 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
89 * or /sys/block/mdX/md/sync_speed_{min,max}
90 */
91
92 static int sysctl_speed_limit_min = 1000;
93 static int sysctl_speed_limit_max = 200000;
94 static inline int speed_min(mddev_t *mddev)
95 {
96 return mddev->sync_speed_min ?
97 mddev->sync_speed_min : sysctl_speed_limit_min;
98 }
99
100 static inline int speed_max(mddev_t *mddev)
101 {
102 return mddev->sync_speed_max ?
103 mddev->sync_speed_max : sysctl_speed_limit_max;
104 }
105
106 static struct ctl_table_header *raid_table_header;
107
108 static ctl_table raid_table[] = {
109 {
110 .procname = "speed_limit_min",
111 .data = &sysctl_speed_limit_min,
112 .maxlen = sizeof(int),
113 .mode = S_IRUGO|S_IWUSR,
114 .proc_handler = proc_dointvec,
115 },
116 {
117 .procname = "speed_limit_max",
118 .data = &sysctl_speed_limit_max,
119 .maxlen = sizeof(int),
120 .mode = S_IRUGO|S_IWUSR,
121 .proc_handler = proc_dointvec,
122 },
123 { }
124 };
125
126 static ctl_table raid_dir_table[] = {
127 {
128 .procname = "raid",
129 .maxlen = 0,
130 .mode = S_IRUGO|S_IXUGO,
131 .child = raid_table,
132 },
133 { }
134 };
135
136 static ctl_table raid_root_table[] = {
137 {
138 .procname = "dev",
139 .maxlen = 0,
140 .mode = 0555,
141 .child = raid_dir_table,
142 },
143 { }
144 };
145
146 static const struct block_device_operations md_fops;
147
148 static int start_readonly;
149
150 /*
151 * We have a system wide 'event count' that is incremented
152 * on any 'interesting' event, and readers of /proc/mdstat
153 * can use 'poll' or 'select' to find out when the event
154 * count increases.
155 *
156 * Events are:
157 * start array, stop array, error, add device, remove device,
158 * start build, activate spare
159 */
160 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
161 static atomic_t md_event_count;
162 void md_new_event(mddev_t *mddev)
163 {
164 atomic_inc(&md_event_count);
165 wake_up(&md_event_waiters);
166 }
167 EXPORT_SYMBOL_GPL(md_new_event);
168
169 /* Alternate version that can be called from interrupts
170 * when calling sysfs_notify isn't needed.
171 */
172 static void md_new_event_inintr(mddev_t *mddev)
173 {
174 atomic_inc(&md_event_count);
175 wake_up(&md_event_waiters);
176 }
177
178 /*
179 * Enables to iterate over all existing md arrays
180 * all_mddevs_lock protects this list.
181 */
182 static LIST_HEAD(all_mddevs);
183 static DEFINE_SPINLOCK(all_mddevs_lock);
184
185
186 /*
187 * iterates through all used mddevs in the system.
188 * We take care to grab the all_mddevs_lock whenever navigating
189 * the list, and to always hold a refcount when unlocked.
190 * Any code which breaks out of this loop while own
191 * a reference to the current mddev and must mddev_put it.
192 */
193 #define for_each_mddev(mddev,tmp) \
194 \
195 for (({ spin_lock(&all_mddevs_lock); \
196 tmp = all_mddevs.next; \
197 mddev = NULL;}); \
198 ({ if (tmp != &all_mddevs) \
199 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
200 spin_unlock(&all_mddevs_lock); \
201 if (mddev) mddev_put(mddev); \
202 mddev = list_entry(tmp, mddev_t, all_mddevs); \
203 tmp != &all_mddevs;}); \
204 ({ spin_lock(&all_mddevs_lock); \
205 tmp = tmp->next;}) \
206 )
207
208
209 /* Rather than calling directly into the personality make_request function,
210 * IO requests come here first so that we can check if the device is
211 * being suspended pending a reconfiguration.
212 * We hold a refcount over the call to ->make_request. By the time that
213 * call has finished, the bio has been linked into some internal structure
214 * and so is visible to ->quiesce(), so we don't need the refcount any more.
215 */
216 static int md_make_request(struct request_queue *q, struct bio *bio)
217 {
218 const int rw = bio_data_dir(bio);
219 mddev_t *mddev = q->queuedata;
220 int rv;
221 int cpu;
222
223 if (mddev == NULL || mddev->pers == NULL) {
224 bio_io_error(bio);
225 return 0;
226 }
227 rcu_read_lock();
228 if (mddev->suspended || mddev->barrier) {
229 DEFINE_WAIT(__wait);
230 for (;;) {
231 prepare_to_wait(&mddev->sb_wait, &__wait,
232 TASK_UNINTERRUPTIBLE);
233 if (!mddev->suspended && !mddev->barrier)
234 break;
235 rcu_read_unlock();
236 schedule();
237 rcu_read_lock();
238 }
239 finish_wait(&mddev->sb_wait, &__wait);
240 }
241 atomic_inc(&mddev->active_io);
242 rcu_read_unlock();
243
244 rv = mddev->pers->make_request(mddev, bio);
245
246 cpu = part_stat_lock();
247 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
248 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
249 bio_sectors(bio));
250 part_stat_unlock();
251
252 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
253 wake_up(&mddev->sb_wait);
254
255 return rv;
256 }
257
258 /* mddev_suspend makes sure no new requests are submitted
259 * to the device, and that any requests that have been submitted
260 * are completely handled.
261 * Once ->stop is called and completes, the module will be completely
262 * unused.
263 */
264 static void mddev_suspend(mddev_t *mddev)
265 {
266 BUG_ON(mddev->suspended);
267 mddev->suspended = 1;
268 synchronize_rcu();
269 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
270 mddev->pers->quiesce(mddev, 1);
271 }
272
273 static void mddev_resume(mddev_t *mddev)
274 {
275 mddev->suspended = 0;
276 wake_up(&mddev->sb_wait);
277 mddev->pers->quiesce(mddev, 0);
278 }
279
280 int mddev_congested(mddev_t *mddev, int bits)
281 {
282 if (mddev->barrier)
283 return 1;
284 return mddev->suspended;
285 }
286 EXPORT_SYMBOL(mddev_congested);
287
288 /*
289 * Generic barrier handling for md
290 */
291
292 #define POST_REQUEST_BARRIER ((void*)1)
293
294 static void md_end_barrier(struct bio *bio, int err)
295 {
296 mdk_rdev_t *rdev = bio->bi_private;
297 mddev_t *mddev = rdev->mddev;
298 if (err == -EOPNOTSUPP && mddev->barrier != POST_REQUEST_BARRIER)
299 set_bit(BIO_EOPNOTSUPP, &mddev->barrier->bi_flags);
300
301 rdev_dec_pending(rdev, mddev);
302
303 if (atomic_dec_and_test(&mddev->flush_pending)) {
304 if (mddev->barrier == POST_REQUEST_BARRIER) {
305 /* This was a post-request barrier */
306 mddev->barrier = NULL;
307 wake_up(&mddev->sb_wait);
308 } else
309 /* The pre-request barrier has finished */
310 schedule_work(&mddev->barrier_work);
311 }
312 bio_put(bio);
313 }
314
315 static void submit_barriers(mddev_t *mddev)
316 {
317 mdk_rdev_t *rdev;
318
319 rcu_read_lock();
320 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
321 if (rdev->raid_disk >= 0 &&
322 !test_bit(Faulty, &rdev->flags)) {
323 /* Take two references, one is dropped
324 * when request finishes, one after
325 * we reclaim rcu_read_lock
326 */
327 struct bio *bi;
328 atomic_inc(&rdev->nr_pending);
329 atomic_inc(&rdev->nr_pending);
330 rcu_read_unlock();
331 bi = bio_alloc(GFP_KERNEL, 0);
332 bi->bi_end_io = md_end_barrier;
333 bi->bi_private = rdev;
334 bi->bi_bdev = rdev->bdev;
335 atomic_inc(&mddev->flush_pending);
336 submit_bio(WRITE_BARRIER, bi);
337 rcu_read_lock();
338 rdev_dec_pending(rdev, mddev);
339 }
340 rcu_read_unlock();
341 }
342
343 static void md_submit_barrier(struct work_struct *ws)
344 {
345 mddev_t *mddev = container_of(ws, mddev_t, barrier_work);
346 struct bio *bio = mddev->barrier;
347
348 atomic_set(&mddev->flush_pending, 1);
349
350 if (test_bit(BIO_EOPNOTSUPP, &bio->bi_flags))
351 bio_endio(bio, -EOPNOTSUPP);
352 else if (bio->bi_size == 0)
353 /* an empty barrier - all done */
354 bio_endio(bio, 0);
355 else {
356 bio->bi_rw &= ~(1<<BIO_RW_BARRIER);
357 if (mddev->pers->make_request(mddev, bio))
358 generic_make_request(bio);
359 mddev->barrier = POST_REQUEST_BARRIER;
360 submit_barriers(mddev);
361 }
362 if (atomic_dec_and_test(&mddev->flush_pending)) {
363 mddev->barrier = NULL;
364 wake_up(&mddev->sb_wait);
365 }
366 }
367
368 void md_barrier_request(mddev_t *mddev, struct bio *bio)
369 {
370 spin_lock_irq(&mddev->write_lock);
371 wait_event_lock_irq(mddev->sb_wait,
372 !mddev->barrier,
373 mddev->write_lock, /*nothing*/);
374 mddev->barrier = bio;
375 spin_unlock_irq(&mddev->write_lock);
376
377 atomic_set(&mddev->flush_pending, 1);
378 INIT_WORK(&mddev->barrier_work, md_submit_barrier);
379
380 submit_barriers(mddev);
381
382 if (atomic_dec_and_test(&mddev->flush_pending))
383 schedule_work(&mddev->barrier_work);
384 }
385 EXPORT_SYMBOL(md_barrier_request);
386
387 static inline mddev_t *mddev_get(mddev_t *mddev)
388 {
389 atomic_inc(&mddev->active);
390 return mddev;
391 }
392
393 static void mddev_delayed_delete(struct work_struct *ws);
394
395 static void mddev_put(mddev_t *mddev)
396 {
397 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
398 return;
399 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
400 mddev->ctime == 0 && !mddev->hold_active) {
401 /* Array is not configured at all, and not held active,
402 * so destroy it */
403 list_del(&mddev->all_mddevs);
404 if (mddev->gendisk) {
405 /* we did a probe so need to clean up.
406 * Call schedule_work inside the spinlock
407 * so that flush_scheduled_work() after
408 * mddev_find will succeed in waiting for the
409 * work to be done.
410 */
411 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
412 schedule_work(&mddev->del_work);
413 } else
414 kfree(mddev);
415 }
416 spin_unlock(&all_mddevs_lock);
417 }
418
419 static void mddev_init(mddev_t *mddev)
420 {
421 mutex_init(&mddev->open_mutex);
422 mutex_init(&mddev->reconfig_mutex);
423 mutex_init(&mddev->bitmap_info.mutex);
424 INIT_LIST_HEAD(&mddev->disks);
425 INIT_LIST_HEAD(&mddev->all_mddevs);
426 init_timer(&mddev->safemode_timer);
427 atomic_set(&mddev->active, 1);
428 atomic_set(&mddev->openers, 0);
429 atomic_set(&mddev->active_io, 0);
430 spin_lock_init(&mddev->write_lock);
431 atomic_set(&mddev->flush_pending, 0);
432 init_waitqueue_head(&mddev->sb_wait);
433 init_waitqueue_head(&mddev->recovery_wait);
434 mddev->reshape_position = MaxSector;
435 mddev->resync_min = 0;
436 mddev->resync_max = MaxSector;
437 mddev->level = LEVEL_NONE;
438 }
439
440 static mddev_t * mddev_find(dev_t unit)
441 {
442 mddev_t *mddev, *new = NULL;
443
444 retry:
445 spin_lock(&all_mddevs_lock);
446
447 if (unit) {
448 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
449 if (mddev->unit == unit) {
450 mddev_get(mddev);
451 spin_unlock(&all_mddevs_lock);
452 kfree(new);
453 return mddev;
454 }
455
456 if (new) {
457 list_add(&new->all_mddevs, &all_mddevs);
458 spin_unlock(&all_mddevs_lock);
459 new->hold_active = UNTIL_IOCTL;
460 return new;
461 }
462 } else if (new) {
463 /* find an unused unit number */
464 static int next_minor = 512;
465 int start = next_minor;
466 int is_free = 0;
467 int dev = 0;
468 while (!is_free) {
469 dev = MKDEV(MD_MAJOR, next_minor);
470 next_minor++;
471 if (next_minor > MINORMASK)
472 next_minor = 0;
473 if (next_minor == start) {
474 /* Oh dear, all in use. */
475 spin_unlock(&all_mddevs_lock);
476 kfree(new);
477 return NULL;
478 }
479
480 is_free = 1;
481 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
482 if (mddev->unit == dev) {
483 is_free = 0;
484 break;
485 }
486 }
487 new->unit = dev;
488 new->md_minor = MINOR(dev);
489 new->hold_active = UNTIL_STOP;
490 list_add(&new->all_mddevs, &all_mddevs);
491 spin_unlock(&all_mddevs_lock);
492 return new;
493 }
494 spin_unlock(&all_mddevs_lock);
495
496 new = kzalloc(sizeof(*new), GFP_KERNEL);
497 if (!new)
498 return NULL;
499
500 new->unit = unit;
501 if (MAJOR(unit) == MD_MAJOR)
502 new->md_minor = MINOR(unit);
503 else
504 new->md_minor = MINOR(unit) >> MdpMinorShift;
505
506 mddev_init(new);
507
508 goto retry;
509 }
510
511 static inline int mddev_lock(mddev_t * mddev)
512 {
513 return mutex_lock_interruptible(&mddev->reconfig_mutex);
514 }
515
516 static inline int mddev_is_locked(mddev_t *mddev)
517 {
518 return mutex_is_locked(&mddev->reconfig_mutex);
519 }
520
521 static inline int mddev_trylock(mddev_t * mddev)
522 {
523 return mutex_trylock(&mddev->reconfig_mutex);
524 }
525
526 static struct attribute_group md_redundancy_group;
527
528 static void mddev_unlock(mddev_t * mddev)
529 {
530 if (mddev->to_remove) {
531 /* These cannot be removed under reconfig_mutex as
532 * an access to the files will try to take reconfig_mutex
533 * while holding the file unremovable, which leads to
534 * a deadlock.
535 * So hold open_mutex instead - we are allowed to take
536 * it while holding reconfig_mutex, and md_run can
537 * use it to wait for the remove to complete.
538 */
539 struct attribute_group *to_remove = mddev->to_remove;
540 mddev->to_remove = NULL;
541 mutex_lock(&mddev->open_mutex);
542 mutex_unlock(&mddev->reconfig_mutex);
543
544 if (to_remove != &md_redundancy_group)
545 sysfs_remove_group(&mddev->kobj, to_remove);
546 if (mddev->pers == NULL ||
547 mddev->pers->sync_request == NULL) {
548 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
549 if (mddev->sysfs_action)
550 sysfs_put(mddev->sysfs_action);
551 mddev->sysfs_action = NULL;
552 }
553 mutex_unlock(&mddev->open_mutex);
554 } else
555 mutex_unlock(&mddev->reconfig_mutex);
556
557 md_wakeup_thread(mddev->thread);
558 }
559
560 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
561 {
562 mdk_rdev_t *rdev;
563
564 list_for_each_entry(rdev, &mddev->disks, same_set)
565 if (rdev->desc_nr == nr)
566 return rdev;
567
568 return NULL;
569 }
570
571 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
572 {
573 mdk_rdev_t *rdev;
574
575 list_for_each_entry(rdev, &mddev->disks, same_set)
576 if (rdev->bdev->bd_dev == dev)
577 return rdev;
578
579 return NULL;
580 }
581
582 static struct mdk_personality *find_pers(int level, char *clevel)
583 {
584 struct mdk_personality *pers;
585 list_for_each_entry(pers, &pers_list, list) {
586 if (level != LEVEL_NONE && pers->level == level)
587 return pers;
588 if (strcmp(pers->name, clevel)==0)
589 return pers;
590 }
591 return NULL;
592 }
593
594 /* return the offset of the super block in 512byte sectors */
595 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
596 {
597 sector_t num_sectors = bdev->bd_inode->i_size / 512;
598 return MD_NEW_SIZE_SECTORS(num_sectors);
599 }
600
601 static int alloc_disk_sb(mdk_rdev_t * rdev)
602 {
603 if (rdev->sb_page)
604 MD_BUG();
605
606 rdev->sb_page = alloc_page(GFP_KERNEL);
607 if (!rdev->sb_page) {
608 printk(KERN_ALERT "md: out of memory.\n");
609 return -ENOMEM;
610 }
611
612 return 0;
613 }
614
615 static void free_disk_sb(mdk_rdev_t * rdev)
616 {
617 if (rdev->sb_page) {
618 put_page(rdev->sb_page);
619 rdev->sb_loaded = 0;
620 rdev->sb_page = NULL;
621 rdev->sb_start = 0;
622 rdev->sectors = 0;
623 }
624 }
625
626
627 static void super_written(struct bio *bio, int error)
628 {
629 mdk_rdev_t *rdev = bio->bi_private;
630 mddev_t *mddev = rdev->mddev;
631
632 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
633 printk("md: super_written gets error=%d, uptodate=%d\n",
634 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
635 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
636 md_error(mddev, rdev);
637 }
638
639 if (atomic_dec_and_test(&mddev->pending_writes))
640 wake_up(&mddev->sb_wait);
641 bio_put(bio);
642 }
643
644 static void super_written_barrier(struct bio *bio, int error)
645 {
646 struct bio *bio2 = bio->bi_private;
647 mdk_rdev_t *rdev = bio2->bi_private;
648 mddev_t *mddev = rdev->mddev;
649
650 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
651 error == -EOPNOTSUPP) {
652 unsigned long flags;
653 /* barriers don't appear to be supported :-( */
654 set_bit(BarriersNotsupp, &rdev->flags);
655 mddev->barriers_work = 0;
656 spin_lock_irqsave(&mddev->write_lock, flags);
657 bio2->bi_next = mddev->biolist;
658 mddev->biolist = bio2;
659 spin_unlock_irqrestore(&mddev->write_lock, flags);
660 wake_up(&mddev->sb_wait);
661 bio_put(bio);
662 } else {
663 bio_put(bio2);
664 bio->bi_private = rdev;
665 super_written(bio, error);
666 }
667 }
668
669 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
670 sector_t sector, int size, struct page *page)
671 {
672 /* write first size bytes of page to sector of rdev
673 * Increment mddev->pending_writes before returning
674 * and decrement it on completion, waking up sb_wait
675 * if zero is reached.
676 * If an error occurred, call md_error
677 *
678 * As we might need to resubmit the request if BIO_RW_BARRIER
679 * causes ENOTSUPP, we allocate a spare bio...
680 */
681 struct bio *bio = bio_alloc(GFP_NOIO, 1);
682 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG);
683
684 bio->bi_bdev = rdev->bdev;
685 bio->bi_sector = sector;
686 bio_add_page(bio, page, size, 0);
687 bio->bi_private = rdev;
688 bio->bi_end_io = super_written;
689 bio->bi_rw = rw;
690
691 atomic_inc(&mddev->pending_writes);
692 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
693 struct bio *rbio;
694 rw |= (1<<BIO_RW_BARRIER);
695 rbio = bio_clone(bio, GFP_NOIO);
696 rbio->bi_private = bio;
697 rbio->bi_end_io = super_written_barrier;
698 submit_bio(rw, rbio);
699 } else
700 submit_bio(rw, bio);
701 }
702
703 void md_super_wait(mddev_t *mddev)
704 {
705 /* wait for all superblock writes that were scheduled to complete.
706 * if any had to be retried (due to BARRIER problems), retry them
707 */
708 DEFINE_WAIT(wq);
709 for(;;) {
710 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
711 if (atomic_read(&mddev->pending_writes)==0)
712 break;
713 while (mddev->biolist) {
714 struct bio *bio;
715 spin_lock_irq(&mddev->write_lock);
716 bio = mddev->biolist;
717 mddev->biolist = bio->bi_next ;
718 bio->bi_next = NULL;
719 spin_unlock_irq(&mddev->write_lock);
720 submit_bio(bio->bi_rw, bio);
721 }
722 schedule();
723 }
724 finish_wait(&mddev->sb_wait, &wq);
725 }
726
727 static void bi_complete(struct bio *bio, int error)
728 {
729 complete((struct completion*)bio->bi_private);
730 }
731
732 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
733 struct page *page, int rw)
734 {
735 struct bio *bio = bio_alloc(GFP_NOIO, 1);
736 struct completion event;
737 int ret;
738
739 rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
740
741 bio->bi_bdev = bdev;
742 bio->bi_sector = sector;
743 bio_add_page(bio, page, size, 0);
744 init_completion(&event);
745 bio->bi_private = &event;
746 bio->bi_end_io = bi_complete;
747 submit_bio(rw, bio);
748 wait_for_completion(&event);
749
750 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
751 bio_put(bio);
752 return ret;
753 }
754 EXPORT_SYMBOL_GPL(sync_page_io);
755
756 static int read_disk_sb(mdk_rdev_t * rdev, int size)
757 {
758 char b[BDEVNAME_SIZE];
759 if (!rdev->sb_page) {
760 MD_BUG();
761 return -EINVAL;
762 }
763 if (rdev->sb_loaded)
764 return 0;
765
766
767 if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
768 goto fail;
769 rdev->sb_loaded = 1;
770 return 0;
771
772 fail:
773 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
774 bdevname(rdev->bdev,b));
775 return -EINVAL;
776 }
777
778 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
779 {
780 return sb1->set_uuid0 == sb2->set_uuid0 &&
781 sb1->set_uuid1 == sb2->set_uuid1 &&
782 sb1->set_uuid2 == sb2->set_uuid2 &&
783 sb1->set_uuid3 == sb2->set_uuid3;
784 }
785
786 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
787 {
788 int ret;
789 mdp_super_t *tmp1, *tmp2;
790
791 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
792 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
793
794 if (!tmp1 || !tmp2) {
795 ret = 0;
796 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
797 goto abort;
798 }
799
800 *tmp1 = *sb1;
801 *tmp2 = *sb2;
802
803 /*
804 * nr_disks is not constant
805 */
806 tmp1->nr_disks = 0;
807 tmp2->nr_disks = 0;
808
809 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
810 abort:
811 kfree(tmp1);
812 kfree(tmp2);
813 return ret;
814 }
815
816
817 static u32 md_csum_fold(u32 csum)
818 {
819 csum = (csum & 0xffff) + (csum >> 16);
820 return (csum & 0xffff) + (csum >> 16);
821 }
822
823 static unsigned int calc_sb_csum(mdp_super_t * sb)
824 {
825 u64 newcsum = 0;
826 u32 *sb32 = (u32*)sb;
827 int i;
828 unsigned int disk_csum, csum;
829
830 disk_csum = sb->sb_csum;
831 sb->sb_csum = 0;
832
833 for (i = 0; i < MD_SB_BYTES/4 ; i++)
834 newcsum += sb32[i];
835 csum = (newcsum & 0xffffffff) + (newcsum>>32);
836
837
838 #ifdef CONFIG_ALPHA
839 /* This used to use csum_partial, which was wrong for several
840 * reasons including that different results are returned on
841 * different architectures. It isn't critical that we get exactly
842 * the same return value as before (we always csum_fold before
843 * testing, and that removes any differences). However as we
844 * know that csum_partial always returned a 16bit value on
845 * alphas, do a fold to maximise conformity to previous behaviour.
846 */
847 sb->sb_csum = md_csum_fold(disk_csum);
848 #else
849 sb->sb_csum = disk_csum;
850 #endif
851 return csum;
852 }
853
854
855 /*
856 * Handle superblock details.
857 * We want to be able to handle multiple superblock formats
858 * so we have a common interface to them all, and an array of
859 * different handlers.
860 * We rely on user-space to write the initial superblock, and support
861 * reading and updating of superblocks.
862 * Interface methods are:
863 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
864 * loads and validates a superblock on dev.
865 * if refdev != NULL, compare superblocks on both devices
866 * Return:
867 * 0 - dev has a superblock that is compatible with refdev
868 * 1 - dev has a superblock that is compatible and newer than refdev
869 * so dev should be used as the refdev in future
870 * -EINVAL superblock incompatible or invalid
871 * -othererror e.g. -EIO
872 *
873 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
874 * Verify that dev is acceptable into mddev.
875 * The first time, mddev->raid_disks will be 0, and data from
876 * dev should be merged in. Subsequent calls check that dev
877 * is new enough. Return 0 or -EINVAL
878 *
879 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
880 * Update the superblock for rdev with data in mddev
881 * This does not write to disc.
882 *
883 */
884
885 struct super_type {
886 char *name;
887 struct module *owner;
888 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
889 int minor_version);
890 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
891 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
892 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
893 sector_t num_sectors);
894 };
895
896 /*
897 * Check that the given mddev has no bitmap.
898 *
899 * This function is called from the run method of all personalities that do not
900 * support bitmaps. It prints an error message and returns non-zero if mddev
901 * has a bitmap. Otherwise, it returns 0.
902 *
903 */
904 int md_check_no_bitmap(mddev_t *mddev)
905 {
906 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
907 return 0;
908 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
909 mdname(mddev), mddev->pers->name);
910 return 1;
911 }
912 EXPORT_SYMBOL(md_check_no_bitmap);
913
914 /*
915 * load_super for 0.90.0
916 */
917 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
918 {
919 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
920 mdp_super_t *sb;
921 int ret;
922
923 /*
924 * Calculate the position of the superblock (512byte sectors),
925 * it's at the end of the disk.
926 *
927 * It also happens to be a multiple of 4Kb.
928 */
929 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
930
931 ret = read_disk_sb(rdev, MD_SB_BYTES);
932 if (ret) return ret;
933
934 ret = -EINVAL;
935
936 bdevname(rdev->bdev, b);
937 sb = (mdp_super_t*)page_address(rdev->sb_page);
938
939 if (sb->md_magic != MD_SB_MAGIC) {
940 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
941 b);
942 goto abort;
943 }
944
945 if (sb->major_version != 0 ||
946 sb->minor_version < 90 ||
947 sb->minor_version > 91) {
948 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
949 sb->major_version, sb->minor_version,
950 b);
951 goto abort;
952 }
953
954 if (sb->raid_disks <= 0)
955 goto abort;
956
957 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
958 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
959 b);
960 goto abort;
961 }
962
963 rdev->preferred_minor = sb->md_minor;
964 rdev->data_offset = 0;
965 rdev->sb_size = MD_SB_BYTES;
966
967 if (sb->level == LEVEL_MULTIPATH)
968 rdev->desc_nr = -1;
969 else
970 rdev->desc_nr = sb->this_disk.number;
971
972 if (!refdev) {
973 ret = 1;
974 } else {
975 __u64 ev1, ev2;
976 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
977 if (!uuid_equal(refsb, sb)) {
978 printk(KERN_WARNING "md: %s has different UUID to %s\n",
979 b, bdevname(refdev->bdev,b2));
980 goto abort;
981 }
982 if (!sb_equal(refsb, sb)) {
983 printk(KERN_WARNING "md: %s has same UUID"
984 " but different superblock to %s\n",
985 b, bdevname(refdev->bdev, b2));
986 goto abort;
987 }
988 ev1 = md_event(sb);
989 ev2 = md_event(refsb);
990 if (ev1 > ev2)
991 ret = 1;
992 else
993 ret = 0;
994 }
995 rdev->sectors = rdev->sb_start;
996
997 if (rdev->sectors < sb->size * 2 && sb->level > 1)
998 /* "this cannot possibly happen" ... */
999 ret = -EINVAL;
1000
1001 abort:
1002 return ret;
1003 }
1004
1005 /*
1006 * validate_super for 0.90.0
1007 */
1008 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1009 {
1010 mdp_disk_t *desc;
1011 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
1012 __u64 ev1 = md_event(sb);
1013
1014 rdev->raid_disk = -1;
1015 clear_bit(Faulty, &rdev->flags);
1016 clear_bit(In_sync, &rdev->flags);
1017 clear_bit(WriteMostly, &rdev->flags);
1018 clear_bit(BarriersNotsupp, &rdev->flags);
1019
1020 if (mddev->raid_disks == 0) {
1021 mddev->major_version = 0;
1022 mddev->minor_version = sb->minor_version;
1023 mddev->patch_version = sb->patch_version;
1024 mddev->external = 0;
1025 mddev->chunk_sectors = sb->chunk_size >> 9;
1026 mddev->ctime = sb->ctime;
1027 mddev->utime = sb->utime;
1028 mddev->level = sb->level;
1029 mddev->clevel[0] = 0;
1030 mddev->layout = sb->layout;
1031 mddev->raid_disks = sb->raid_disks;
1032 mddev->dev_sectors = sb->size * 2;
1033 mddev->events = ev1;
1034 mddev->bitmap_info.offset = 0;
1035 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1036
1037 if (mddev->minor_version >= 91) {
1038 mddev->reshape_position = sb->reshape_position;
1039 mddev->delta_disks = sb->delta_disks;
1040 mddev->new_level = sb->new_level;
1041 mddev->new_layout = sb->new_layout;
1042 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1043 } else {
1044 mddev->reshape_position = MaxSector;
1045 mddev->delta_disks = 0;
1046 mddev->new_level = mddev->level;
1047 mddev->new_layout = mddev->layout;
1048 mddev->new_chunk_sectors = mddev->chunk_sectors;
1049 }
1050
1051 if (sb->state & (1<<MD_SB_CLEAN))
1052 mddev->recovery_cp = MaxSector;
1053 else {
1054 if (sb->events_hi == sb->cp_events_hi &&
1055 sb->events_lo == sb->cp_events_lo) {
1056 mddev->recovery_cp = sb->recovery_cp;
1057 } else
1058 mddev->recovery_cp = 0;
1059 }
1060
1061 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1062 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1063 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1064 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1065
1066 mddev->max_disks = MD_SB_DISKS;
1067
1068 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1069 mddev->bitmap_info.file == NULL)
1070 mddev->bitmap_info.offset =
1071 mddev->bitmap_info.default_offset;
1072
1073 } else if (mddev->pers == NULL) {
1074 /* Insist on good event counter while assembling, except
1075 * for spares (which don't need an event count) */
1076 ++ev1;
1077 if (sb->disks[rdev->desc_nr].state & (
1078 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1079 if (ev1 < mddev->events)
1080 return -EINVAL;
1081 } else if (mddev->bitmap) {
1082 /* if adding to array with a bitmap, then we can accept an
1083 * older device ... but not too old.
1084 */
1085 if (ev1 < mddev->bitmap->events_cleared)
1086 return 0;
1087 } else {
1088 if (ev1 < mddev->events)
1089 /* just a hot-add of a new device, leave raid_disk at -1 */
1090 return 0;
1091 }
1092
1093 if (mddev->level != LEVEL_MULTIPATH) {
1094 desc = sb->disks + rdev->desc_nr;
1095
1096 if (desc->state & (1<<MD_DISK_FAULTY))
1097 set_bit(Faulty, &rdev->flags);
1098 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1099 desc->raid_disk < mddev->raid_disks */) {
1100 set_bit(In_sync, &rdev->flags);
1101 rdev->raid_disk = desc->raid_disk;
1102 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1103 /* active but not in sync implies recovery up to
1104 * reshape position. We don't know exactly where
1105 * that is, so set to zero for now */
1106 if (mddev->minor_version >= 91) {
1107 rdev->recovery_offset = 0;
1108 rdev->raid_disk = desc->raid_disk;
1109 }
1110 }
1111 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1112 set_bit(WriteMostly, &rdev->flags);
1113 } else /* MULTIPATH are always insync */
1114 set_bit(In_sync, &rdev->flags);
1115 return 0;
1116 }
1117
1118 /*
1119 * sync_super for 0.90.0
1120 */
1121 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1122 {
1123 mdp_super_t *sb;
1124 mdk_rdev_t *rdev2;
1125 int next_spare = mddev->raid_disks;
1126
1127
1128 /* make rdev->sb match mddev data..
1129 *
1130 * 1/ zero out disks
1131 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1132 * 3/ any empty disks < next_spare become removed
1133 *
1134 * disks[0] gets initialised to REMOVED because
1135 * we cannot be sure from other fields if it has
1136 * been initialised or not.
1137 */
1138 int i;
1139 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1140
1141 rdev->sb_size = MD_SB_BYTES;
1142
1143 sb = (mdp_super_t*)page_address(rdev->sb_page);
1144
1145 memset(sb, 0, sizeof(*sb));
1146
1147 sb->md_magic = MD_SB_MAGIC;
1148 sb->major_version = mddev->major_version;
1149 sb->patch_version = mddev->patch_version;
1150 sb->gvalid_words = 0; /* ignored */
1151 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1152 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1153 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1154 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1155
1156 sb->ctime = mddev->ctime;
1157 sb->level = mddev->level;
1158 sb->size = mddev->dev_sectors / 2;
1159 sb->raid_disks = mddev->raid_disks;
1160 sb->md_minor = mddev->md_minor;
1161 sb->not_persistent = 0;
1162 sb->utime = mddev->utime;
1163 sb->state = 0;
1164 sb->events_hi = (mddev->events>>32);
1165 sb->events_lo = (u32)mddev->events;
1166
1167 if (mddev->reshape_position == MaxSector)
1168 sb->minor_version = 90;
1169 else {
1170 sb->minor_version = 91;
1171 sb->reshape_position = mddev->reshape_position;
1172 sb->new_level = mddev->new_level;
1173 sb->delta_disks = mddev->delta_disks;
1174 sb->new_layout = mddev->new_layout;
1175 sb->new_chunk = mddev->new_chunk_sectors << 9;
1176 }
1177 mddev->minor_version = sb->minor_version;
1178 if (mddev->in_sync)
1179 {
1180 sb->recovery_cp = mddev->recovery_cp;
1181 sb->cp_events_hi = (mddev->events>>32);
1182 sb->cp_events_lo = (u32)mddev->events;
1183 if (mddev->recovery_cp == MaxSector)
1184 sb->state = (1<< MD_SB_CLEAN);
1185 } else
1186 sb->recovery_cp = 0;
1187
1188 sb->layout = mddev->layout;
1189 sb->chunk_size = mddev->chunk_sectors << 9;
1190
1191 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1192 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1193
1194 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1195 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1196 mdp_disk_t *d;
1197 int desc_nr;
1198 int is_active = test_bit(In_sync, &rdev2->flags);
1199
1200 if (rdev2->raid_disk >= 0 &&
1201 sb->minor_version >= 91)
1202 /* we have nowhere to store the recovery_offset,
1203 * but if it is not below the reshape_position,
1204 * we can piggy-back on that.
1205 */
1206 is_active = 1;
1207 if (rdev2->raid_disk < 0 ||
1208 test_bit(Faulty, &rdev2->flags))
1209 is_active = 0;
1210 if (is_active)
1211 desc_nr = rdev2->raid_disk;
1212 else
1213 desc_nr = next_spare++;
1214 rdev2->desc_nr = desc_nr;
1215 d = &sb->disks[rdev2->desc_nr];
1216 nr_disks++;
1217 d->number = rdev2->desc_nr;
1218 d->major = MAJOR(rdev2->bdev->bd_dev);
1219 d->minor = MINOR(rdev2->bdev->bd_dev);
1220 if (is_active)
1221 d->raid_disk = rdev2->raid_disk;
1222 else
1223 d->raid_disk = rdev2->desc_nr; /* compatibility */
1224 if (test_bit(Faulty, &rdev2->flags))
1225 d->state = (1<<MD_DISK_FAULTY);
1226 else if (is_active) {
1227 d->state = (1<<MD_DISK_ACTIVE);
1228 if (test_bit(In_sync, &rdev2->flags))
1229 d->state |= (1<<MD_DISK_SYNC);
1230 active++;
1231 working++;
1232 } else {
1233 d->state = 0;
1234 spare++;
1235 working++;
1236 }
1237 if (test_bit(WriteMostly, &rdev2->flags))
1238 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1239 }
1240 /* now set the "removed" and "faulty" bits on any missing devices */
1241 for (i=0 ; i < mddev->raid_disks ; i++) {
1242 mdp_disk_t *d = &sb->disks[i];
1243 if (d->state == 0 && d->number == 0) {
1244 d->number = i;
1245 d->raid_disk = i;
1246 d->state = (1<<MD_DISK_REMOVED);
1247 d->state |= (1<<MD_DISK_FAULTY);
1248 failed++;
1249 }
1250 }
1251 sb->nr_disks = nr_disks;
1252 sb->active_disks = active;
1253 sb->working_disks = working;
1254 sb->failed_disks = failed;
1255 sb->spare_disks = spare;
1256
1257 sb->this_disk = sb->disks[rdev->desc_nr];
1258 sb->sb_csum = calc_sb_csum(sb);
1259 }
1260
1261 /*
1262 * rdev_size_change for 0.90.0
1263 */
1264 static unsigned long long
1265 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1266 {
1267 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1268 return 0; /* component must fit device */
1269 if (rdev->mddev->bitmap_info.offset)
1270 return 0; /* can't move bitmap */
1271 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1272 if (!num_sectors || num_sectors > rdev->sb_start)
1273 num_sectors = rdev->sb_start;
1274 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1275 rdev->sb_page);
1276 md_super_wait(rdev->mddev);
1277 return num_sectors / 2; /* kB for sysfs */
1278 }
1279
1280
1281 /*
1282 * version 1 superblock
1283 */
1284
1285 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1286 {
1287 __le32 disk_csum;
1288 u32 csum;
1289 unsigned long long newcsum;
1290 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1291 __le32 *isuper = (__le32*)sb;
1292 int i;
1293
1294 disk_csum = sb->sb_csum;
1295 sb->sb_csum = 0;
1296 newcsum = 0;
1297 for (i=0; size>=4; size -= 4 )
1298 newcsum += le32_to_cpu(*isuper++);
1299
1300 if (size == 2)
1301 newcsum += le16_to_cpu(*(__le16*) isuper);
1302
1303 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1304 sb->sb_csum = disk_csum;
1305 return cpu_to_le32(csum);
1306 }
1307
1308 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1309 {
1310 struct mdp_superblock_1 *sb;
1311 int ret;
1312 sector_t sb_start;
1313 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1314 int bmask;
1315
1316 /*
1317 * Calculate the position of the superblock in 512byte sectors.
1318 * It is always aligned to a 4K boundary and
1319 * depeding on minor_version, it can be:
1320 * 0: At least 8K, but less than 12K, from end of device
1321 * 1: At start of device
1322 * 2: 4K from start of device.
1323 */
1324 switch(minor_version) {
1325 case 0:
1326 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1327 sb_start -= 8*2;
1328 sb_start &= ~(sector_t)(4*2-1);
1329 break;
1330 case 1:
1331 sb_start = 0;
1332 break;
1333 case 2:
1334 sb_start = 8;
1335 break;
1336 default:
1337 return -EINVAL;
1338 }
1339 rdev->sb_start = sb_start;
1340
1341 /* superblock is rarely larger than 1K, but it can be larger,
1342 * and it is safe to read 4k, so we do that
1343 */
1344 ret = read_disk_sb(rdev, 4096);
1345 if (ret) return ret;
1346
1347
1348 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1349
1350 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1351 sb->major_version != cpu_to_le32(1) ||
1352 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1353 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1354 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1355 return -EINVAL;
1356
1357 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1358 printk("md: invalid superblock checksum on %s\n",
1359 bdevname(rdev->bdev,b));
1360 return -EINVAL;
1361 }
1362 if (le64_to_cpu(sb->data_size) < 10) {
1363 printk("md: data_size too small on %s\n",
1364 bdevname(rdev->bdev,b));
1365 return -EINVAL;
1366 }
1367
1368 rdev->preferred_minor = 0xffff;
1369 rdev->data_offset = le64_to_cpu(sb->data_offset);
1370 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1371
1372 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1373 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1374 if (rdev->sb_size & bmask)
1375 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1376
1377 if (minor_version
1378 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1379 return -EINVAL;
1380
1381 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1382 rdev->desc_nr = -1;
1383 else
1384 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1385
1386 if (!refdev) {
1387 ret = 1;
1388 } else {
1389 __u64 ev1, ev2;
1390 struct mdp_superblock_1 *refsb =
1391 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1392
1393 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1394 sb->level != refsb->level ||
1395 sb->layout != refsb->layout ||
1396 sb->chunksize != refsb->chunksize) {
1397 printk(KERN_WARNING "md: %s has strangely different"
1398 " superblock to %s\n",
1399 bdevname(rdev->bdev,b),
1400 bdevname(refdev->bdev,b2));
1401 return -EINVAL;
1402 }
1403 ev1 = le64_to_cpu(sb->events);
1404 ev2 = le64_to_cpu(refsb->events);
1405
1406 if (ev1 > ev2)
1407 ret = 1;
1408 else
1409 ret = 0;
1410 }
1411 if (minor_version)
1412 rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
1413 le64_to_cpu(sb->data_offset);
1414 else
1415 rdev->sectors = rdev->sb_start;
1416 if (rdev->sectors < le64_to_cpu(sb->data_size))
1417 return -EINVAL;
1418 rdev->sectors = le64_to_cpu(sb->data_size);
1419 if (le64_to_cpu(sb->size) > rdev->sectors)
1420 return -EINVAL;
1421 return ret;
1422 }
1423
1424 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1425 {
1426 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1427 __u64 ev1 = le64_to_cpu(sb->events);
1428
1429 rdev->raid_disk = -1;
1430 clear_bit(Faulty, &rdev->flags);
1431 clear_bit(In_sync, &rdev->flags);
1432 clear_bit(WriteMostly, &rdev->flags);
1433 clear_bit(BarriersNotsupp, &rdev->flags);
1434
1435 if (mddev->raid_disks == 0) {
1436 mddev->major_version = 1;
1437 mddev->patch_version = 0;
1438 mddev->external = 0;
1439 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1440 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1441 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1442 mddev->level = le32_to_cpu(sb->level);
1443 mddev->clevel[0] = 0;
1444 mddev->layout = le32_to_cpu(sb->layout);
1445 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1446 mddev->dev_sectors = le64_to_cpu(sb->size);
1447 mddev->events = ev1;
1448 mddev->bitmap_info.offset = 0;
1449 mddev->bitmap_info.default_offset = 1024 >> 9;
1450
1451 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1452 memcpy(mddev->uuid, sb->set_uuid, 16);
1453
1454 mddev->max_disks = (4096-256)/2;
1455
1456 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1457 mddev->bitmap_info.file == NULL )
1458 mddev->bitmap_info.offset =
1459 (__s32)le32_to_cpu(sb->bitmap_offset);
1460
1461 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1462 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1463 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1464 mddev->new_level = le32_to_cpu(sb->new_level);
1465 mddev->new_layout = le32_to_cpu(sb->new_layout);
1466 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1467 } else {
1468 mddev->reshape_position = MaxSector;
1469 mddev->delta_disks = 0;
1470 mddev->new_level = mddev->level;
1471 mddev->new_layout = mddev->layout;
1472 mddev->new_chunk_sectors = mddev->chunk_sectors;
1473 }
1474
1475 } else if (mddev->pers == NULL) {
1476 /* Insist of good event counter while assembling, except for
1477 * spares (which don't need an event count) */
1478 ++ev1;
1479 if (rdev->desc_nr >= 0 &&
1480 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1481 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1482 if (ev1 < mddev->events)
1483 return -EINVAL;
1484 } else if (mddev->bitmap) {
1485 /* If adding to array with a bitmap, then we can accept an
1486 * older device, but not too old.
1487 */
1488 if (ev1 < mddev->bitmap->events_cleared)
1489 return 0;
1490 } else {
1491 if (ev1 < mddev->events)
1492 /* just a hot-add of a new device, leave raid_disk at -1 */
1493 return 0;
1494 }
1495 if (mddev->level != LEVEL_MULTIPATH) {
1496 int role;
1497 if (rdev->desc_nr < 0 ||
1498 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1499 role = 0xffff;
1500 rdev->desc_nr = -1;
1501 } else
1502 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1503 switch(role) {
1504 case 0xffff: /* spare */
1505 break;
1506 case 0xfffe: /* faulty */
1507 set_bit(Faulty, &rdev->flags);
1508 break;
1509 default:
1510 if ((le32_to_cpu(sb->feature_map) &
1511 MD_FEATURE_RECOVERY_OFFSET))
1512 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1513 else
1514 set_bit(In_sync, &rdev->flags);
1515 rdev->raid_disk = role;
1516 break;
1517 }
1518 if (sb->devflags & WriteMostly1)
1519 set_bit(WriteMostly, &rdev->flags);
1520 } else /* MULTIPATH are always insync */
1521 set_bit(In_sync, &rdev->flags);
1522
1523 return 0;
1524 }
1525
1526 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1527 {
1528 struct mdp_superblock_1 *sb;
1529 mdk_rdev_t *rdev2;
1530 int max_dev, i;
1531 /* make rdev->sb match mddev and rdev data. */
1532
1533 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1534
1535 sb->feature_map = 0;
1536 sb->pad0 = 0;
1537 sb->recovery_offset = cpu_to_le64(0);
1538 memset(sb->pad1, 0, sizeof(sb->pad1));
1539 memset(sb->pad2, 0, sizeof(sb->pad2));
1540 memset(sb->pad3, 0, sizeof(sb->pad3));
1541
1542 sb->utime = cpu_to_le64((__u64)mddev->utime);
1543 sb->events = cpu_to_le64(mddev->events);
1544 if (mddev->in_sync)
1545 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1546 else
1547 sb->resync_offset = cpu_to_le64(0);
1548
1549 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1550
1551 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1552 sb->size = cpu_to_le64(mddev->dev_sectors);
1553 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1554 sb->level = cpu_to_le32(mddev->level);
1555 sb->layout = cpu_to_le32(mddev->layout);
1556
1557 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1558 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1559 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1560 }
1561
1562 if (rdev->raid_disk >= 0 &&
1563 !test_bit(In_sync, &rdev->flags)) {
1564 sb->feature_map |=
1565 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1566 sb->recovery_offset =
1567 cpu_to_le64(rdev->recovery_offset);
1568 }
1569
1570 if (mddev->reshape_position != MaxSector) {
1571 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1572 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1573 sb->new_layout = cpu_to_le32(mddev->new_layout);
1574 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1575 sb->new_level = cpu_to_le32(mddev->new_level);
1576 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1577 }
1578
1579 max_dev = 0;
1580 list_for_each_entry(rdev2, &mddev->disks, same_set)
1581 if (rdev2->desc_nr+1 > max_dev)
1582 max_dev = rdev2->desc_nr+1;
1583
1584 if (max_dev > le32_to_cpu(sb->max_dev)) {
1585 int bmask;
1586 sb->max_dev = cpu_to_le32(max_dev);
1587 rdev->sb_size = max_dev * 2 + 256;
1588 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1589 if (rdev->sb_size & bmask)
1590 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1591 }
1592 for (i=0; i<max_dev;i++)
1593 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1594
1595 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1596 i = rdev2->desc_nr;
1597 if (test_bit(Faulty, &rdev2->flags))
1598 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1599 else if (test_bit(In_sync, &rdev2->flags))
1600 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1601 else if (rdev2->raid_disk >= 0)
1602 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1603 else
1604 sb->dev_roles[i] = cpu_to_le16(0xffff);
1605 }
1606
1607 sb->sb_csum = calc_sb_1_csum(sb);
1608 }
1609
1610 static unsigned long long
1611 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1612 {
1613 struct mdp_superblock_1 *sb;
1614 sector_t max_sectors;
1615 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1616 return 0; /* component must fit device */
1617 if (rdev->sb_start < rdev->data_offset) {
1618 /* minor versions 1 and 2; superblock before data */
1619 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1620 max_sectors -= rdev->data_offset;
1621 if (!num_sectors || num_sectors > max_sectors)
1622 num_sectors = max_sectors;
1623 } else if (rdev->mddev->bitmap_info.offset) {
1624 /* minor version 0 with bitmap we can't move */
1625 return 0;
1626 } else {
1627 /* minor version 0; superblock after data */
1628 sector_t sb_start;
1629 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1630 sb_start &= ~(sector_t)(4*2 - 1);
1631 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1632 if (!num_sectors || num_sectors > max_sectors)
1633 num_sectors = max_sectors;
1634 rdev->sb_start = sb_start;
1635 }
1636 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1637 sb->data_size = cpu_to_le64(num_sectors);
1638 sb->super_offset = rdev->sb_start;
1639 sb->sb_csum = calc_sb_1_csum(sb);
1640 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1641 rdev->sb_page);
1642 md_super_wait(rdev->mddev);
1643 return num_sectors / 2; /* kB for sysfs */
1644 }
1645
1646 static struct super_type super_types[] = {
1647 [0] = {
1648 .name = "0.90.0",
1649 .owner = THIS_MODULE,
1650 .load_super = super_90_load,
1651 .validate_super = super_90_validate,
1652 .sync_super = super_90_sync,
1653 .rdev_size_change = super_90_rdev_size_change,
1654 },
1655 [1] = {
1656 .name = "md-1",
1657 .owner = THIS_MODULE,
1658 .load_super = super_1_load,
1659 .validate_super = super_1_validate,
1660 .sync_super = super_1_sync,
1661 .rdev_size_change = super_1_rdev_size_change,
1662 },
1663 };
1664
1665 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1666 {
1667 mdk_rdev_t *rdev, *rdev2;
1668
1669 rcu_read_lock();
1670 rdev_for_each_rcu(rdev, mddev1)
1671 rdev_for_each_rcu(rdev2, mddev2)
1672 if (rdev->bdev->bd_contains ==
1673 rdev2->bdev->bd_contains) {
1674 rcu_read_unlock();
1675 return 1;
1676 }
1677 rcu_read_unlock();
1678 return 0;
1679 }
1680
1681 static LIST_HEAD(pending_raid_disks);
1682
1683 /*
1684 * Try to register data integrity profile for an mddev
1685 *
1686 * This is called when an array is started and after a disk has been kicked
1687 * from the array. It only succeeds if all working and active component devices
1688 * are integrity capable with matching profiles.
1689 */
1690 int md_integrity_register(mddev_t *mddev)
1691 {
1692 mdk_rdev_t *rdev, *reference = NULL;
1693
1694 if (list_empty(&mddev->disks))
1695 return 0; /* nothing to do */
1696 if (blk_get_integrity(mddev->gendisk))
1697 return 0; /* already registered */
1698 list_for_each_entry(rdev, &mddev->disks, same_set) {
1699 /* skip spares and non-functional disks */
1700 if (test_bit(Faulty, &rdev->flags))
1701 continue;
1702 if (rdev->raid_disk < 0)
1703 continue;
1704 /*
1705 * If at least one rdev is not integrity capable, we can not
1706 * enable data integrity for the md device.
1707 */
1708 if (!bdev_get_integrity(rdev->bdev))
1709 return -EINVAL;
1710 if (!reference) {
1711 /* Use the first rdev as the reference */
1712 reference = rdev;
1713 continue;
1714 }
1715 /* does this rdev's profile match the reference profile? */
1716 if (blk_integrity_compare(reference->bdev->bd_disk,
1717 rdev->bdev->bd_disk) < 0)
1718 return -EINVAL;
1719 }
1720 /*
1721 * All component devices are integrity capable and have matching
1722 * profiles, register the common profile for the md device.
1723 */
1724 if (blk_integrity_register(mddev->gendisk,
1725 bdev_get_integrity(reference->bdev)) != 0) {
1726 printk(KERN_ERR "md: failed to register integrity for %s\n",
1727 mdname(mddev));
1728 return -EINVAL;
1729 }
1730 printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1731 mdname(mddev));
1732 return 0;
1733 }
1734 EXPORT_SYMBOL(md_integrity_register);
1735
1736 /* Disable data integrity if non-capable/non-matching disk is being added */
1737 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
1738 {
1739 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1740 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1741
1742 if (!bi_mddev) /* nothing to do */
1743 return;
1744 if (rdev->raid_disk < 0) /* skip spares */
1745 return;
1746 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1747 rdev->bdev->bd_disk) >= 0)
1748 return;
1749 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1750 blk_integrity_unregister(mddev->gendisk);
1751 }
1752 EXPORT_SYMBOL(md_integrity_add_rdev);
1753
1754 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1755 {
1756 char b[BDEVNAME_SIZE];
1757 struct kobject *ko;
1758 char *s;
1759 int err;
1760
1761 if (rdev->mddev) {
1762 MD_BUG();
1763 return -EINVAL;
1764 }
1765
1766 /* prevent duplicates */
1767 if (find_rdev(mddev, rdev->bdev->bd_dev))
1768 return -EEXIST;
1769
1770 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1771 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1772 rdev->sectors < mddev->dev_sectors)) {
1773 if (mddev->pers) {
1774 /* Cannot change size, so fail
1775 * If mddev->level <= 0, then we don't care
1776 * about aligning sizes (e.g. linear)
1777 */
1778 if (mddev->level > 0)
1779 return -ENOSPC;
1780 } else
1781 mddev->dev_sectors = rdev->sectors;
1782 }
1783
1784 /* Verify rdev->desc_nr is unique.
1785 * If it is -1, assign a free number, else
1786 * check number is not in use
1787 */
1788 if (rdev->desc_nr < 0) {
1789 int choice = 0;
1790 if (mddev->pers) choice = mddev->raid_disks;
1791 while (find_rdev_nr(mddev, choice))
1792 choice++;
1793 rdev->desc_nr = choice;
1794 } else {
1795 if (find_rdev_nr(mddev, rdev->desc_nr))
1796 return -EBUSY;
1797 }
1798 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1799 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1800 mdname(mddev), mddev->max_disks);
1801 return -EBUSY;
1802 }
1803 bdevname(rdev->bdev,b);
1804 while ( (s=strchr(b, '/')) != NULL)
1805 *s = '!';
1806
1807 rdev->mddev = mddev;
1808 printk(KERN_INFO "md: bind<%s>\n", b);
1809
1810 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1811 goto fail;
1812
1813 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1814 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1815 kobject_del(&rdev->kobj);
1816 goto fail;
1817 }
1818 rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, NULL, "state");
1819
1820 list_add_rcu(&rdev->same_set, &mddev->disks);
1821 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1822
1823 /* May as well allow recovery to be retried once */
1824 mddev->recovery_disabled = 0;
1825
1826 return 0;
1827
1828 fail:
1829 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1830 b, mdname(mddev));
1831 return err;
1832 }
1833
1834 static void md_delayed_delete(struct work_struct *ws)
1835 {
1836 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1837 kobject_del(&rdev->kobj);
1838 kobject_put(&rdev->kobj);
1839 }
1840
1841 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1842 {
1843 char b[BDEVNAME_SIZE];
1844 if (!rdev->mddev) {
1845 MD_BUG();
1846 return;
1847 }
1848 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1849 list_del_rcu(&rdev->same_set);
1850 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1851 rdev->mddev = NULL;
1852 sysfs_remove_link(&rdev->kobj, "block");
1853 sysfs_put(rdev->sysfs_state);
1854 rdev->sysfs_state = NULL;
1855 /* We need to delay this, otherwise we can deadlock when
1856 * writing to 'remove' to "dev/state". We also need
1857 * to delay it due to rcu usage.
1858 */
1859 synchronize_rcu();
1860 INIT_WORK(&rdev->del_work, md_delayed_delete);
1861 kobject_get(&rdev->kobj);
1862 schedule_work(&rdev->del_work);
1863 }
1864
1865 /*
1866 * prevent the device from being mounted, repartitioned or
1867 * otherwise reused by a RAID array (or any other kernel
1868 * subsystem), by bd_claiming the device.
1869 */
1870 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1871 {
1872 int err = 0;
1873 struct block_device *bdev;
1874 char b[BDEVNAME_SIZE];
1875
1876 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1877 if (IS_ERR(bdev)) {
1878 printk(KERN_ERR "md: could not open %s.\n",
1879 __bdevname(dev, b));
1880 return PTR_ERR(bdev);
1881 }
1882 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1883 if (err) {
1884 printk(KERN_ERR "md: could not bd_claim %s.\n",
1885 bdevname(bdev, b));
1886 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1887 return err;
1888 }
1889 if (!shared)
1890 set_bit(AllReserved, &rdev->flags);
1891 rdev->bdev = bdev;
1892 return err;
1893 }
1894
1895 static void unlock_rdev(mdk_rdev_t *rdev)
1896 {
1897 struct block_device *bdev = rdev->bdev;
1898 rdev->bdev = NULL;
1899 if (!bdev)
1900 MD_BUG();
1901 bd_release(bdev);
1902 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1903 }
1904
1905 void md_autodetect_dev(dev_t dev);
1906
1907 static void export_rdev(mdk_rdev_t * rdev)
1908 {
1909 char b[BDEVNAME_SIZE];
1910 printk(KERN_INFO "md: export_rdev(%s)\n",
1911 bdevname(rdev->bdev,b));
1912 if (rdev->mddev)
1913 MD_BUG();
1914 free_disk_sb(rdev);
1915 #ifndef MODULE
1916 if (test_bit(AutoDetected, &rdev->flags))
1917 md_autodetect_dev(rdev->bdev->bd_dev);
1918 #endif
1919 unlock_rdev(rdev);
1920 kobject_put(&rdev->kobj);
1921 }
1922
1923 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1924 {
1925 unbind_rdev_from_array(rdev);
1926 export_rdev(rdev);
1927 }
1928
1929 static void export_array(mddev_t *mddev)
1930 {
1931 mdk_rdev_t *rdev, *tmp;
1932
1933 rdev_for_each(rdev, tmp, mddev) {
1934 if (!rdev->mddev) {
1935 MD_BUG();
1936 continue;
1937 }
1938 kick_rdev_from_array(rdev);
1939 }
1940 if (!list_empty(&mddev->disks))
1941 MD_BUG();
1942 mddev->raid_disks = 0;
1943 mddev->major_version = 0;
1944 }
1945
1946 static void print_desc(mdp_disk_t *desc)
1947 {
1948 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1949 desc->major,desc->minor,desc->raid_disk,desc->state);
1950 }
1951
1952 static void print_sb_90(mdp_super_t *sb)
1953 {
1954 int i;
1955
1956 printk(KERN_INFO
1957 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1958 sb->major_version, sb->minor_version, sb->patch_version,
1959 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1960 sb->ctime);
1961 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1962 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1963 sb->md_minor, sb->layout, sb->chunk_size);
1964 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1965 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1966 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1967 sb->failed_disks, sb->spare_disks,
1968 sb->sb_csum, (unsigned long)sb->events_lo);
1969
1970 printk(KERN_INFO);
1971 for (i = 0; i < MD_SB_DISKS; i++) {
1972 mdp_disk_t *desc;
1973
1974 desc = sb->disks + i;
1975 if (desc->number || desc->major || desc->minor ||
1976 desc->raid_disk || (desc->state && (desc->state != 4))) {
1977 printk(" D %2d: ", i);
1978 print_desc(desc);
1979 }
1980 }
1981 printk(KERN_INFO "md: THIS: ");
1982 print_desc(&sb->this_disk);
1983 }
1984
1985 static void print_sb_1(struct mdp_superblock_1 *sb)
1986 {
1987 __u8 *uuid;
1988
1989 uuid = sb->set_uuid;
1990 printk(KERN_INFO
1991 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
1992 "md: Name: \"%s\" CT:%llu\n",
1993 le32_to_cpu(sb->major_version),
1994 le32_to_cpu(sb->feature_map),
1995 uuid,
1996 sb->set_name,
1997 (unsigned long long)le64_to_cpu(sb->ctime)
1998 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
1999
2000 uuid = sb->device_uuid;
2001 printk(KERN_INFO
2002 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2003 " RO:%llu\n"
2004 "md: Dev:%08x UUID: %pU\n"
2005 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2006 "md: (MaxDev:%u) \n",
2007 le32_to_cpu(sb->level),
2008 (unsigned long long)le64_to_cpu(sb->size),
2009 le32_to_cpu(sb->raid_disks),
2010 le32_to_cpu(sb->layout),
2011 le32_to_cpu(sb->chunksize),
2012 (unsigned long long)le64_to_cpu(sb->data_offset),
2013 (unsigned long long)le64_to_cpu(sb->data_size),
2014 (unsigned long long)le64_to_cpu(sb->super_offset),
2015 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2016 le32_to_cpu(sb->dev_number),
2017 uuid,
2018 sb->devflags,
2019 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2020 (unsigned long long)le64_to_cpu(sb->events),
2021 (unsigned long long)le64_to_cpu(sb->resync_offset),
2022 le32_to_cpu(sb->sb_csum),
2023 le32_to_cpu(sb->max_dev)
2024 );
2025 }
2026
2027 static void print_rdev(mdk_rdev_t *rdev, int major_version)
2028 {
2029 char b[BDEVNAME_SIZE];
2030 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2031 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2032 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2033 rdev->desc_nr);
2034 if (rdev->sb_loaded) {
2035 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2036 switch (major_version) {
2037 case 0:
2038 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
2039 break;
2040 case 1:
2041 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
2042 break;
2043 }
2044 } else
2045 printk(KERN_INFO "md: no rdev superblock!\n");
2046 }
2047
2048 static void md_print_devices(void)
2049 {
2050 struct list_head *tmp;
2051 mdk_rdev_t *rdev;
2052 mddev_t *mddev;
2053 char b[BDEVNAME_SIZE];
2054
2055 printk("\n");
2056 printk("md: **********************************\n");
2057 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2058 printk("md: **********************************\n");
2059 for_each_mddev(mddev, tmp) {
2060
2061 if (mddev->bitmap)
2062 bitmap_print_sb(mddev->bitmap);
2063 else
2064 printk("%s: ", mdname(mddev));
2065 list_for_each_entry(rdev, &mddev->disks, same_set)
2066 printk("<%s>", bdevname(rdev->bdev,b));
2067 printk("\n");
2068
2069 list_for_each_entry(rdev, &mddev->disks, same_set)
2070 print_rdev(rdev, mddev->major_version);
2071 }
2072 printk("md: **********************************\n");
2073 printk("\n");
2074 }
2075
2076
2077 static void sync_sbs(mddev_t * mddev, int nospares)
2078 {
2079 /* Update each superblock (in-memory image), but
2080 * if we are allowed to, skip spares which already
2081 * have the right event counter, or have one earlier
2082 * (which would mean they aren't being marked as dirty
2083 * with the rest of the array)
2084 */
2085 mdk_rdev_t *rdev;
2086
2087 /* First make sure individual recovery_offsets are correct */
2088 list_for_each_entry(rdev, &mddev->disks, same_set) {
2089 if (rdev->raid_disk >= 0 &&
2090 !test_bit(In_sync, &rdev->flags) &&
2091 mddev->curr_resync_completed > rdev->recovery_offset)
2092 rdev->recovery_offset = mddev->curr_resync_completed;
2093
2094 }
2095 list_for_each_entry(rdev, &mddev->disks, same_set) {
2096 if (rdev->sb_events == mddev->events ||
2097 (nospares &&
2098 rdev->raid_disk < 0 &&
2099 rdev->sb_events+1 == mddev->events)) {
2100 /* Don't update this superblock */
2101 rdev->sb_loaded = 2;
2102 } else {
2103 super_types[mddev->major_version].
2104 sync_super(mddev, rdev);
2105 rdev->sb_loaded = 1;
2106 }
2107 }
2108 }
2109
2110 static void md_update_sb(mddev_t * mddev, int force_change)
2111 {
2112 mdk_rdev_t *rdev;
2113 int sync_req;
2114 int nospares = 0;
2115
2116 mddev->utime = get_seconds();
2117 if (mddev->external)
2118 return;
2119 repeat:
2120 spin_lock_irq(&mddev->write_lock);
2121
2122 set_bit(MD_CHANGE_PENDING, &mddev->flags);
2123 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2124 force_change = 1;
2125 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2126 /* just a clean<-> dirty transition, possibly leave spares alone,
2127 * though if events isn't the right even/odd, we will have to do
2128 * spares after all
2129 */
2130 nospares = 1;
2131 if (force_change)
2132 nospares = 0;
2133 if (mddev->degraded)
2134 /* If the array is degraded, then skipping spares is both
2135 * dangerous and fairly pointless.
2136 * Dangerous because a device that was removed from the array
2137 * might have a event_count that still looks up-to-date,
2138 * so it can be re-added without a resync.
2139 * Pointless because if there are any spares to skip,
2140 * then a recovery will happen and soon that array won't
2141 * be degraded any more and the spare can go back to sleep then.
2142 */
2143 nospares = 0;
2144
2145 sync_req = mddev->in_sync;
2146
2147 /* If this is just a dirty<->clean transition, and the array is clean
2148 * and 'events' is odd, we can roll back to the previous clean state */
2149 if (nospares
2150 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2151 && mddev->can_decrease_events
2152 && mddev->events != 1) {
2153 mddev->events--;
2154 mddev->can_decrease_events = 0;
2155 } else {
2156 /* otherwise we have to go forward and ... */
2157 mddev->events ++;
2158 mddev->can_decrease_events = nospares;
2159 }
2160
2161 if (!mddev->events) {
2162 /*
2163 * oops, this 64-bit counter should never wrap.
2164 * Either we are in around ~1 trillion A.C., assuming
2165 * 1 reboot per second, or we have a bug:
2166 */
2167 MD_BUG();
2168 mddev->events --;
2169 }
2170
2171 /*
2172 * do not write anything to disk if using
2173 * nonpersistent superblocks
2174 */
2175 if (!mddev->persistent) {
2176 if (!mddev->external)
2177 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2178
2179 spin_unlock_irq(&mddev->write_lock);
2180 wake_up(&mddev->sb_wait);
2181 return;
2182 }
2183 sync_sbs(mddev, nospares);
2184 spin_unlock_irq(&mddev->write_lock);
2185
2186 dprintk(KERN_INFO
2187 "md: updating %s RAID superblock on device (in sync %d)\n",
2188 mdname(mddev),mddev->in_sync);
2189
2190 bitmap_update_sb(mddev->bitmap);
2191 list_for_each_entry(rdev, &mddev->disks, same_set) {
2192 char b[BDEVNAME_SIZE];
2193 dprintk(KERN_INFO "md: ");
2194 if (rdev->sb_loaded != 1)
2195 continue; /* no noise on spare devices */
2196 if (test_bit(Faulty, &rdev->flags))
2197 dprintk("(skipping faulty ");
2198
2199 dprintk("%s ", bdevname(rdev->bdev,b));
2200 if (!test_bit(Faulty, &rdev->flags)) {
2201 md_super_write(mddev,rdev,
2202 rdev->sb_start, rdev->sb_size,
2203 rdev->sb_page);
2204 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2205 bdevname(rdev->bdev,b),
2206 (unsigned long long)rdev->sb_start);
2207 rdev->sb_events = mddev->events;
2208
2209 } else
2210 dprintk(")\n");
2211 if (mddev->level == LEVEL_MULTIPATH)
2212 /* only need to write one superblock... */
2213 break;
2214 }
2215 md_super_wait(mddev);
2216 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2217
2218 spin_lock_irq(&mddev->write_lock);
2219 if (mddev->in_sync != sync_req ||
2220 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2221 /* have to write it out again */
2222 spin_unlock_irq(&mddev->write_lock);
2223 goto repeat;
2224 }
2225 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2226 spin_unlock_irq(&mddev->write_lock);
2227 wake_up(&mddev->sb_wait);
2228 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2229 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2230
2231 }
2232
2233 /* words written to sysfs files may, or may not, be \n terminated.
2234 * We want to accept with case. For this we use cmd_match.
2235 */
2236 static int cmd_match(const char *cmd, const char *str)
2237 {
2238 /* See if cmd, written into a sysfs file, matches
2239 * str. They must either be the same, or cmd can
2240 * have a trailing newline
2241 */
2242 while (*cmd && *str && *cmd == *str) {
2243 cmd++;
2244 str++;
2245 }
2246 if (*cmd == '\n')
2247 cmd++;
2248 if (*str || *cmd)
2249 return 0;
2250 return 1;
2251 }
2252
2253 struct rdev_sysfs_entry {
2254 struct attribute attr;
2255 ssize_t (*show)(mdk_rdev_t *, char *);
2256 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2257 };
2258
2259 static ssize_t
2260 state_show(mdk_rdev_t *rdev, char *page)
2261 {
2262 char *sep = "";
2263 size_t len = 0;
2264
2265 if (test_bit(Faulty, &rdev->flags)) {
2266 len+= sprintf(page+len, "%sfaulty",sep);
2267 sep = ",";
2268 }
2269 if (test_bit(In_sync, &rdev->flags)) {
2270 len += sprintf(page+len, "%sin_sync",sep);
2271 sep = ",";
2272 }
2273 if (test_bit(WriteMostly, &rdev->flags)) {
2274 len += sprintf(page+len, "%swrite_mostly",sep);
2275 sep = ",";
2276 }
2277 if (test_bit(Blocked, &rdev->flags)) {
2278 len += sprintf(page+len, "%sblocked", sep);
2279 sep = ",";
2280 }
2281 if (!test_bit(Faulty, &rdev->flags) &&
2282 !test_bit(In_sync, &rdev->flags)) {
2283 len += sprintf(page+len, "%sspare", sep);
2284 sep = ",";
2285 }
2286 return len+sprintf(page+len, "\n");
2287 }
2288
2289 static ssize_t
2290 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2291 {
2292 /* can write
2293 * faulty - simulates and error
2294 * remove - disconnects the device
2295 * writemostly - sets write_mostly
2296 * -writemostly - clears write_mostly
2297 * blocked - sets the Blocked flag
2298 * -blocked - clears the Blocked flag
2299 * insync - sets Insync providing device isn't active
2300 */
2301 int err = -EINVAL;
2302 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2303 md_error(rdev->mddev, rdev);
2304 err = 0;
2305 } else if (cmd_match(buf, "remove")) {
2306 if (rdev->raid_disk >= 0)
2307 err = -EBUSY;
2308 else {
2309 mddev_t *mddev = rdev->mddev;
2310 kick_rdev_from_array(rdev);
2311 if (mddev->pers)
2312 md_update_sb(mddev, 1);
2313 md_new_event(mddev);
2314 err = 0;
2315 }
2316 } else if (cmd_match(buf, "writemostly")) {
2317 set_bit(WriteMostly, &rdev->flags);
2318 err = 0;
2319 } else if (cmd_match(buf, "-writemostly")) {
2320 clear_bit(WriteMostly, &rdev->flags);
2321 err = 0;
2322 } else if (cmd_match(buf, "blocked")) {
2323 set_bit(Blocked, &rdev->flags);
2324 err = 0;
2325 } else if (cmd_match(buf, "-blocked")) {
2326 clear_bit(Blocked, &rdev->flags);
2327 wake_up(&rdev->blocked_wait);
2328 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2329 md_wakeup_thread(rdev->mddev->thread);
2330
2331 err = 0;
2332 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2333 set_bit(In_sync, &rdev->flags);
2334 err = 0;
2335 }
2336 if (!err && rdev->sysfs_state)
2337 sysfs_notify_dirent(rdev->sysfs_state);
2338 return err ? err : len;
2339 }
2340 static struct rdev_sysfs_entry rdev_state =
2341 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2342
2343 static ssize_t
2344 errors_show(mdk_rdev_t *rdev, char *page)
2345 {
2346 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2347 }
2348
2349 static ssize_t
2350 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2351 {
2352 char *e;
2353 unsigned long n = simple_strtoul(buf, &e, 10);
2354 if (*buf && (*e == 0 || *e == '\n')) {
2355 atomic_set(&rdev->corrected_errors, n);
2356 return len;
2357 }
2358 return -EINVAL;
2359 }
2360 static struct rdev_sysfs_entry rdev_errors =
2361 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2362
2363 static ssize_t
2364 slot_show(mdk_rdev_t *rdev, char *page)
2365 {
2366 if (rdev->raid_disk < 0)
2367 return sprintf(page, "none\n");
2368 else
2369 return sprintf(page, "%d\n", rdev->raid_disk);
2370 }
2371
2372 static ssize_t
2373 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2374 {
2375 char *e;
2376 int err;
2377 char nm[20];
2378 int slot = simple_strtoul(buf, &e, 10);
2379 if (strncmp(buf, "none", 4)==0)
2380 slot = -1;
2381 else if (e==buf || (*e && *e!= '\n'))
2382 return -EINVAL;
2383 if (rdev->mddev->pers && slot == -1) {
2384 /* Setting 'slot' on an active array requires also
2385 * updating the 'rd%d' link, and communicating
2386 * with the personality with ->hot_*_disk.
2387 * For now we only support removing
2388 * failed/spare devices. This normally happens automatically,
2389 * but not when the metadata is externally managed.
2390 */
2391 if (rdev->raid_disk == -1)
2392 return -EEXIST;
2393 /* personality does all needed checks */
2394 if (rdev->mddev->pers->hot_add_disk == NULL)
2395 return -EINVAL;
2396 err = rdev->mddev->pers->
2397 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2398 if (err)
2399 return err;
2400 sprintf(nm, "rd%d", rdev->raid_disk);
2401 sysfs_remove_link(&rdev->mddev->kobj, nm);
2402 rdev->raid_disk = -1;
2403 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2404 md_wakeup_thread(rdev->mddev->thread);
2405 } else if (rdev->mddev->pers) {
2406 mdk_rdev_t *rdev2;
2407 /* Activating a spare .. or possibly reactivating
2408 * if we ever get bitmaps working here.
2409 */
2410
2411 if (rdev->raid_disk != -1)
2412 return -EBUSY;
2413
2414 if (rdev->mddev->pers->hot_add_disk == NULL)
2415 return -EINVAL;
2416
2417 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2418 if (rdev2->raid_disk == slot)
2419 return -EEXIST;
2420
2421 rdev->raid_disk = slot;
2422 if (test_bit(In_sync, &rdev->flags))
2423 rdev->saved_raid_disk = slot;
2424 else
2425 rdev->saved_raid_disk = -1;
2426 err = rdev->mddev->pers->
2427 hot_add_disk(rdev->mddev, rdev);
2428 if (err) {
2429 rdev->raid_disk = -1;
2430 return err;
2431 } else
2432 sysfs_notify_dirent(rdev->sysfs_state);
2433 sprintf(nm, "rd%d", rdev->raid_disk);
2434 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2435 printk(KERN_WARNING
2436 "md: cannot register "
2437 "%s for %s\n",
2438 nm, mdname(rdev->mddev));
2439
2440 /* don't wakeup anyone, leave that to userspace. */
2441 } else {
2442 if (slot >= rdev->mddev->raid_disks)
2443 return -ENOSPC;
2444 rdev->raid_disk = slot;
2445 /* assume it is working */
2446 clear_bit(Faulty, &rdev->flags);
2447 clear_bit(WriteMostly, &rdev->flags);
2448 set_bit(In_sync, &rdev->flags);
2449 sysfs_notify_dirent(rdev->sysfs_state);
2450 }
2451 return len;
2452 }
2453
2454
2455 static struct rdev_sysfs_entry rdev_slot =
2456 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2457
2458 static ssize_t
2459 offset_show(mdk_rdev_t *rdev, char *page)
2460 {
2461 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2462 }
2463
2464 static ssize_t
2465 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2466 {
2467 char *e;
2468 unsigned long long offset = simple_strtoull(buf, &e, 10);
2469 if (e==buf || (*e && *e != '\n'))
2470 return -EINVAL;
2471 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2472 return -EBUSY;
2473 if (rdev->sectors && rdev->mddev->external)
2474 /* Must set offset before size, so overlap checks
2475 * can be sane */
2476 return -EBUSY;
2477 rdev->data_offset = offset;
2478 return len;
2479 }
2480
2481 static struct rdev_sysfs_entry rdev_offset =
2482 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2483
2484 static ssize_t
2485 rdev_size_show(mdk_rdev_t *rdev, char *page)
2486 {
2487 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2488 }
2489
2490 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2491 {
2492 /* check if two start/length pairs overlap */
2493 if (s1+l1 <= s2)
2494 return 0;
2495 if (s2+l2 <= s1)
2496 return 0;
2497 return 1;
2498 }
2499
2500 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2501 {
2502 unsigned long long blocks;
2503 sector_t new;
2504
2505 if (strict_strtoull(buf, 10, &blocks) < 0)
2506 return -EINVAL;
2507
2508 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2509 return -EINVAL; /* sector conversion overflow */
2510
2511 new = blocks * 2;
2512 if (new != blocks * 2)
2513 return -EINVAL; /* unsigned long long to sector_t overflow */
2514
2515 *sectors = new;
2516 return 0;
2517 }
2518
2519 static ssize_t
2520 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2521 {
2522 mddev_t *my_mddev = rdev->mddev;
2523 sector_t oldsectors = rdev->sectors;
2524 sector_t sectors;
2525
2526 if (strict_blocks_to_sectors(buf, &sectors) < 0)
2527 return -EINVAL;
2528 if (my_mddev->pers && rdev->raid_disk >= 0) {
2529 if (my_mddev->persistent) {
2530 sectors = super_types[my_mddev->major_version].
2531 rdev_size_change(rdev, sectors);
2532 if (!sectors)
2533 return -EBUSY;
2534 } else if (!sectors)
2535 sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2536 rdev->data_offset;
2537 }
2538 if (sectors < my_mddev->dev_sectors)
2539 return -EINVAL; /* component must fit device */
2540
2541 rdev->sectors = sectors;
2542 if (sectors > oldsectors && my_mddev->external) {
2543 /* need to check that all other rdevs with the same ->bdev
2544 * do not overlap. We need to unlock the mddev to avoid
2545 * a deadlock. We have already changed rdev->sectors, and if
2546 * we have to change it back, we will have the lock again.
2547 */
2548 mddev_t *mddev;
2549 int overlap = 0;
2550 struct list_head *tmp;
2551
2552 mddev_unlock(my_mddev);
2553 for_each_mddev(mddev, tmp) {
2554 mdk_rdev_t *rdev2;
2555
2556 mddev_lock(mddev);
2557 list_for_each_entry(rdev2, &mddev->disks, same_set)
2558 if (test_bit(AllReserved, &rdev2->flags) ||
2559 (rdev->bdev == rdev2->bdev &&
2560 rdev != rdev2 &&
2561 overlaps(rdev->data_offset, rdev->sectors,
2562 rdev2->data_offset,
2563 rdev2->sectors))) {
2564 overlap = 1;
2565 break;
2566 }
2567 mddev_unlock(mddev);
2568 if (overlap) {
2569 mddev_put(mddev);
2570 break;
2571 }
2572 }
2573 mddev_lock(my_mddev);
2574 if (overlap) {
2575 /* Someone else could have slipped in a size
2576 * change here, but doing so is just silly.
2577 * We put oldsectors back because we *know* it is
2578 * safe, and trust userspace not to race with
2579 * itself
2580 */
2581 rdev->sectors = oldsectors;
2582 return -EBUSY;
2583 }
2584 }
2585 return len;
2586 }
2587
2588 static struct rdev_sysfs_entry rdev_size =
2589 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2590
2591
2592 static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page)
2593 {
2594 unsigned long long recovery_start = rdev->recovery_offset;
2595
2596 if (test_bit(In_sync, &rdev->flags) ||
2597 recovery_start == MaxSector)
2598 return sprintf(page, "none\n");
2599
2600 return sprintf(page, "%llu\n", recovery_start);
2601 }
2602
2603 static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2604 {
2605 unsigned long long recovery_start;
2606
2607 if (cmd_match(buf, "none"))
2608 recovery_start = MaxSector;
2609 else if (strict_strtoull(buf, 10, &recovery_start))
2610 return -EINVAL;
2611
2612 if (rdev->mddev->pers &&
2613 rdev->raid_disk >= 0)
2614 return -EBUSY;
2615
2616 rdev->recovery_offset = recovery_start;
2617 if (recovery_start == MaxSector)
2618 set_bit(In_sync, &rdev->flags);
2619 else
2620 clear_bit(In_sync, &rdev->flags);
2621 return len;
2622 }
2623
2624 static struct rdev_sysfs_entry rdev_recovery_start =
2625 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2626
2627 static struct attribute *rdev_default_attrs[] = {
2628 &rdev_state.attr,
2629 &rdev_errors.attr,
2630 &rdev_slot.attr,
2631 &rdev_offset.attr,
2632 &rdev_size.attr,
2633 &rdev_recovery_start.attr,
2634 NULL,
2635 };
2636 static ssize_t
2637 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2638 {
2639 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2640 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2641 mddev_t *mddev = rdev->mddev;
2642 ssize_t rv;
2643
2644 if (!entry->show)
2645 return -EIO;
2646
2647 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2648 if (!rv) {
2649 if (rdev->mddev == NULL)
2650 rv = -EBUSY;
2651 else
2652 rv = entry->show(rdev, page);
2653 mddev_unlock(mddev);
2654 }
2655 return rv;
2656 }
2657
2658 static ssize_t
2659 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2660 const char *page, size_t length)
2661 {
2662 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2663 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2664 ssize_t rv;
2665 mddev_t *mddev = rdev->mddev;
2666
2667 if (!entry->store)
2668 return -EIO;
2669 if (!capable(CAP_SYS_ADMIN))
2670 return -EACCES;
2671 rv = mddev ? mddev_lock(mddev): -EBUSY;
2672 if (!rv) {
2673 if (rdev->mddev == NULL)
2674 rv = -EBUSY;
2675 else
2676 rv = entry->store(rdev, page, length);
2677 mddev_unlock(mddev);
2678 }
2679 return rv;
2680 }
2681
2682 static void rdev_free(struct kobject *ko)
2683 {
2684 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2685 kfree(rdev);
2686 }
2687 static const struct sysfs_ops rdev_sysfs_ops = {
2688 .show = rdev_attr_show,
2689 .store = rdev_attr_store,
2690 };
2691 static struct kobj_type rdev_ktype = {
2692 .release = rdev_free,
2693 .sysfs_ops = &rdev_sysfs_ops,
2694 .default_attrs = rdev_default_attrs,
2695 };
2696
2697 /*
2698 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2699 *
2700 * mark the device faulty if:
2701 *
2702 * - the device is nonexistent (zero size)
2703 * - the device has no valid superblock
2704 *
2705 * a faulty rdev _never_ has rdev->sb set.
2706 */
2707 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2708 {
2709 char b[BDEVNAME_SIZE];
2710 int err;
2711 mdk_rdev_t *rdev;
2712 sector_t size;
2713
2714 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2715 if (!rdev) {
2716 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2717 return ERR_PTR(-ENOMEM);
2718 }
2719
2720 if ((err = alloc_disk_sb(rdev)))
2721 goto abort_free;
2722
2723 err = lock_rdev(rdev, newdev, super_format == -2);
2724 if (err)
2725 goto abort_free;
2726
2727 kobject_init(&rdev->kobj, &rdev_ktype);
2728
2729 rdev->desc_nr = -1;
2730 rdev->saved_raid_disk = -1;
2731 rdev->raid_disk = -1;
2732 rdev->flags = 0;
2733 rdev->data_offset = 0;
2734 rdev->sb_events = 0;
2735 rdev->last_read_error.tv_sec = 0;
2736 rdev->last_read_error.tv_nsec = 0;
2737 atomic_set(&rdev->nr_pending, 0);
2738 atomic_set(&rdev->read_errors, 0);
2739 atomic_set(&rdev->corrected_errors, 0);
2740
2741 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2742 if (!size) {
2743 printk(KERN_WARNING
2744 "md: %s has zero or unknown size, marking faulty!\n",
2745 bdevname(rdev->bdev,b));
2746 err = -EINVAL;
2747 goto abort_free;
2748 }
2749
2750 if (super_format >= 0) {
2751 err = super_types[super_format].
2752 load_super(rdev, NULL, super_minor);
2753 if (err == -EINVAL) {
2754 printk(KERN_WARNING
2755 "md: %s does not have a valid v%d.%d "
2756 "superblock, not importing!\n",
2757 bdevname(rdev->bdev,b),
2758 super_format, super_minor);
2759 goto abort_free;
2760 }
2761 if (err < 0) {
2762 printk(KERN_WARNING
2763 "md: could not read %s's sb, not importing!\n",
2764 bdevname(rdev->bdev,b));
2765 goto abort_free;
2766 }
2767 }
2768
2769 INIT_LIST_HEAD(&rdev->same_set);
2770 init_waitqueue_head(&rdev->blocked_wait);
2771
2772 return rdev;
2773
2774 abort_free:
2775 if (rdev->sb_page) {
2776 if (rdev->bdev)
2777 unlock_rdev(rdev);
2778 free_disk_sb(rdev);
2779 }
2780 kfree(rdev);
2781 return ERR_PTR(err);
2782 }
2783
2784 /*
2785 * Check a full RAID array for plausibility
2786 */
2787
2788
2789 static void analyze_sbs(mddev_t * mddev)
2790 {
2791 int i;
2792 mdk_rdev_t *rdev, *freshest, *tmp;
2793 char b[BDEVNAME_SIZE];
2794
2795 freshest = NULL;
2796 rdev_for_each(rdev, tmp, mddev)
2797 switch (super_types[mddev->major_version].
2798 load_super(rdev, freshest, mddev->minor_version)) {
2799 case 1:
2800 freshest = rdev;
2801 break;
2802 case 0:
2803 break;
2804 default:
2805 printk( KERN_ERR \
2806 "md: fatal superblock inconsistency in %s"
2807 " -- removing from array\n",
2808 bdevname(rdev->bdev,b));
2809 kick_rdev_from_array(rdev);
2810 }
2811
2812
2813 super_types[mddev->major_version].
2814 validate_super(mddev, freshest);
2815
2816 i = 0;
2817 rdev_for_each(rdev, tmp, mddev) {
2818 if (mddev->max_disks &&
2819 (rdev->desc_nr >= mddev->max_disks ||
2820 i > mddev->max_disks)) {
2821 printk(KERN_WARNING
2822 "md: %s: %s: only %d devices permitted\n",
2823 mdname(mddev), bdevname(rdev->bdev, b),
2824 mddev->max_disks);
2825 kick_rdev_from_array(rdev);
2826 continue;
2827 }
2828 if (rdev != freshest)
2829 if (super_types[mddev->major_version].
2830 validate_super(mddev, rdev)) {
2831 printk(KERN_WARNING "md: kicking non-fresh %s"
2832 " from array!\n",
2833 bdevname(rdev->bdev,b));
2834 kick_rdev_from_array(rdev);
2835 continue;
2836 }
2837 if (mddev->level == LEVEL_MULTIPATH) {
2838 rdev->desc_nr = i++;
2839 rdev->raid_disk = rdev->desc_nr;
2840 set_bit(In_sync, &rdev->flags);
2841 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
2842 rdev->raid_disk = -1;
2843 clear_bit(In_sync, &rdev->flags);
2844 }
2845 }
2846 }
2847
2848 /* Read a fixed-point number.
2849 * Numbers in sysfs attributes should be in "standard" units where
2850 * possible, so time should be in seconds.
2851 * However we internally use a a much smaller unit such as
2852 * milliseconds or jiffies.
2853 * This function takes a decimal number with a possible fractional
2854 * component, and produces an integer which is the result of
2855 * multiplying that number by 10^'scale'.
2856 * all without any floating-point arithmetic.
2857 */
2858 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
2859 {
2860 unsigned long result = 0;
2861 long decimals = -1;
2862 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
2863 if (*cp == '.')
2864 decimals = 0;
2865 else if (decimals < scale) {
2866 unsigned int value;
2867 value = *cp - '0';
2868 result = result * 10 + value;
2869 if (decimals >= 0)
2870 decimals++;
2871 }
2872 cp++;
2873 }
2874 if (*cp == '\n')
2875 cp++;
2876 if (*cp)
2877 return -EINVAL;
2878 if (decimals < 0)
2879 decimals = 0;
2880 while (decimals < scale) {
2881 result *= 10;
2882 decimals ++;
2883 }
2884 *res = result;
2885 return 0;
2886 }
2887
2888
2889 static void md_safemode_timeout(unsigned long data);
2890
2891 static ssize_t
2892 safe_delay_show(mddev_t *mddev, char *page)
2893 {
2894 int msec = (mddev->safemode_delay*1000)/HZ;
2895 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2896 }
2897 static ssize_t
2898 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2899 {
2900 unsigned long msec;
2901
2902 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
2903 return -EINVAL;
2904 if (msec == 0)
2905 mddev->safemode_delay = 0;
2906 else {
2907 unsigned long old_delay = mddev->safemode_delay;
2908 mddev->safemode_delay = (msec*HZ)/1000;
2909 if (mddev->safemode_delay == 0)
2910 mddev->safemode_delay = 1;
2911 if (mddev->safemode_delay < old_delay)
2912 md_safemode_timeout((unsigned long)mddev);
2913 }
2914 return len;
2915 }
2916 static struct md_sysfs_entry md_safe_delay =
2917 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2918
2919 static ssize_t
2920 level_show(mddev_t *mddev, char *page)
2921 {
2922 struct mdk_personality *p = mddev->pers;
2923 if (p)
2924 return sprintf(page, "%s\n", p->name);
2925 else if (mddev->clevel[0])
2926 return sprintf(page, "%s\n", mddev->clevel);
2927 else if (mddev->level != LEVEL_NONE)
2928 return sprintf(page, "%d\n", mddev->level);
2929 else
2930 return 0;
2931 }
2932
2933 static ssize_t
2934 level_store(mddev_t *mddev, const char *buf, size_t len)
2935 {
2936 char clevel[16];
2937 ssize_t rv = len;
2938 struct mdk_personality *pers;
2939 long level;
2940 void *priv;
2941 mdk_rdev_t *rdev;
2942
2943 if (mddev->pers == NULL) {
2944 if (len == 0)
2945 return 0;
2946 if (len >= sizeof(mddev->clevel))
2947 return -ENOSPC;
2948 strncpy(mddev->clevel, buf, len);
2949 if (mddev->clevel[len-1] == '\n')
2950 len--;
2951 mddev->clevel[len] = 0;
2952 mddev->level = LEVEL_NONE;
2953 return rv;
2954 }
2955
2956 /* request to change the personality. Need to ensure:
2957 * - array is not engaged in resync/recovery/reshape
2958 * - old personality can be suspended
2959 * - new personality will access other array.
2960 */
2961
2962 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
2963 return -EBUSY;
2964
2965 if (!mddev->pers->quiesce) {
2966 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
2967 mdname(mddev), mddev->pers->name);
2968 return -EINVAL;
2969 }
2970
2971 /* Now find the new personality */
2972 if (len == 0 || len >= sizeof(clevel))
2973 return -EINVAL;
2974 strncpy(clevel, buf, len);
2975 if (clevel[len-1] == '\n')
2976 len--;
2977 clevel[len] = 0;
2978 if (strict_strtol(clevel, 10, &level))
2979 level = LEVEL_NONE;
2980
2981 if (request_module("md-%s", clevel) != 0)
2982 request_module("md-level-%s", clevel);
2983 spin_lock(&pers_lock);
2984 pers = find_pers(level, clevel);
2985 if (!pers || !try_module_get(pers->owner)) {
2986 spin_unlock(&pers_lock);
2987 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
2988 return -EINVAL;
2989 }
2990 spin_unlock(&pers_lock);
2991
2992 if (pers == mddev->pers) {
2993 /* Nothing to do! */
2994 module_put(pers->owner);
2995 return rv;
2996 }
2997 if (!pers->takeover) {
2998 module_put(pers->owner);
2999 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3000 mdname(mddev), clevel);
3001 return -EINVAL;
3002 }
3003
3004 list_for_each_entry(rdev, &mddev->disks, same_set)
3005 rdev->new_raid_disk = rdev->raid_disk;
3006
3007 /* ->takeover must set new_* and/or delta_disks
3008 * if it succeeds, and may set them when it fails.
3009 */
3010 priv = pers->takeover(mddev);
3011 if (IS_ERR(priv)) {
3012 mddev->new_level = mddev->level;
3013 mddev->new_layout = mddev->layout;
3014 mddev->new_chunk_sectors = mddev->chunk_sectors;
3015 mddev->raid_disks -= mddev->delta_disks;
3016 mddev->delta_disks = 0;
3017 module_put(pers->owner);
3018 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3019 mdname(mddev), clevel);
3020 return PTR_ERR(priv);
3021 }
3022
3023 /* Looks like we have a winner */
3024 mddev_suspend(mddev);
3025 mddev->pers->stop(mddev);
3026
3027 if (mddev->pers->sync_request == NULL &&
3028 pers->sync_request != NULL) {
3029 /* need to add the md_redundancy_group */
3030 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3031 printk(KERN_WARNING
3032 "md: cannot register extra attributes for %s\n",
3033 mdname(mddev));
3034 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
3035 }
3036 if (mddev->pers->sync_request != NULL &&
3037 pers->sync_request == NULL) {
3038 /* need to remove the md_redundancy_group */
3039 if (mddev->to_remove == NULL)
3040 mddev->to_remove = &md_redundancy_group;
3041 }
3042
3043 if (mddev->pers->sync_request == NULL &&
3044 mddev->external) {
3045 /* We are converting from a no-redundancy array
3046 * to a redundancy array and metadata is managed
3047 * externally so we need to be sure that writes
3048 * won't block due to a need to transition
3049 * clean->dirty
3050 * until external management is started.
3051 */
3052 mddev->in_sync = 0;
3053 mddev->safemode_delay = 0;
3054 mddev->safemode = 0;
3055 }
3056
3057 list_for_each_entry(rdev, &mddev->disks, same_set) {
3058 char nm[20];
3059 if (rdev->raid_disk < 0)
3060 continue;
3061 if (rdev->new_raid_disk > mddev->raid_disks)
3062 rdev->new_raid_disk = -1;
3063 if (rdev->new_raid_disk == rdev->raid_disk)
3064 continue;
3065 sprintf(nm, "rd%d", rdev->raid_disk);
3066 sysfs_remove_link(&mddev->kobj, nm);
3067 }
3068 list_for_each_entry(rdev, &mddev->disks, same_set) {
3069 if (rdev->raid_disk < 0)
3070 continue;
3071 if (rdev->new_raid_disk == rdev->raid_disk)
3072 continue;
3073 rdev->raid_disk = rdev->new_raid_disk;
3074 if (rdev->raid_disk < 0)
3075 clear_bit(In_sync, &rdev->flags);
3076 else {
3077 char nm[20];
3078 sprintf(nm, "rd%d", rdev->raid_disk);
3079 if(sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3080 printk("md: cannot register %s for %s after level change\n",
3081 nm, mdname(mddev));
3082 }
3083 }
3084
3085 module_put(mddev->pers->owner);
3086 mddev->pers = pers;
3087 mddev->private = priv;
3088 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3089 mddev->level = mddev->new_level;
3090 mddev->layout = mddev->new_layout;
3091 mddev->chunk_sectors = mddev->new_chunk_sectors;
3092 mddev->delta_disks = 0;
3093 if (mddev->pers->sync_request == NULL) {
3094 /* this is now an array without redundancy, so
3095 * it must always be in_sync
3096 */
3097 mddev->in_sync = 1;
3098 del_timer_sync(&mddev->safemode_timer);
3099 }
3100 pers->run(mddev);
3101 mddev_resume(mddev);
3102 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3103 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3104 md_wakeup_thread(mddev->thread);
3105 sysfs_notify(&mddev->kobj, NULL, "level");
3106 md_new_event(mddev);
3107 return rv;
3108 }
3109
3110 static struct md_sysfs_entry md_level =
3111 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3112
3113
3114 static ssize_t
3115 layout_show(mddev_t *mddev, char *page)
3116 {
3117 /* just a number, not meaningful for all levels */
3118 if (mddev->reshape_position != MaxSector &&
3119 mddev->layout != mddev->new_layout)
3120 return sprintf(page, "%d (%d)\n",
3121 mddev->new_layout, mddev->layout);
3122 return sprintf(page, "%d\n", mddev->layout);
3123 }
3124
3125 static ssize_t
3126 layout_store(mddev_t *mddev, const char *buf, size_t len)
3127 {
3128 char *e;
3129 unsigned long n = simple_strtoul(buf, &e, 10);
3130
3131 if (!*buf || (*e && *e != '\n'))
3132 return -EINVAL;
3133
3134 if (mddev->pers) {
3135 int err;
3136 if (mddev->pers->check_reshape == NULL)
3137 return -EBUSY;
3138 mddev->new_layout = n;
3139 err = mddev->pers->check_reshape(mddev);
3140 if (err) {
3141 mddev->new_layout = mddev->layout;
3142 return err;
3143 }
3144 } else {
3145 mddev->new_layout = n;
3146 if (mddev->reshape_position == MaxSector)
3147 mddev->layout = n;
3148 }
3149 return len;
3150 }
3151 static struct md_sysfs_entry md_layout =
3152 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3153
3154
3155 static ssize_t
3156 raid_disks_show(mddev_t *mddev, char *page)
3157 {
3158 if (mddev->raid_disks == 0)
3159 return 0;
3160 if (mddev->reshape_position != MaxSector &&
3161 mddev->delta_disks != 0)
3162 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3163 mddev->raid_disks - mddev->delta_disks);
3164 return sprintf(page, "%d\n", mddev->raid_disks);
3165 }
3166
3167 static int update_raid_disks(mddev_t *mddev, int raid_disks);
3168
3169 static ssize_t
3170 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
3171 {
3172 char *e;
3173 int rv = 0;
3174 unsigned long n = simple_strtoul(buf, &e, 10);
3175
3176 if (!*buf || (*e && *e != '\n'))
3177 return -EINVAL;
3178
3179 if (mddev->pers)
3180 rv = update_raid_disks(mddev, n);
3181 else if (mddev->reshape_position != MaxSector) {
3182 int olddisks = mddev->raid_disks - mddev->delta_disks;
3183 mddev->delta_disks = n - olddisks;
3184 mddev->raid_disks = n;
3185 } else
3186 mddev->raid_disks = n;
3187 return rv ? rv : len;
3188 }
3189 static struct md_sysfs_entry md_raid_disks =
3190 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3191
3192 static ssize_t
3193 chunk_size_show(mddev_t *mddev, char *page)
3194 {
3195 if (mddev->reshape_position != MaxSector &&
3196 mddev->chunk_sectors != mddev->new_chunk_sectors)
3197 return sprintf(page, "%d (%d)\n",
3198 mddev->new_chunk_sectors << 9,
3199 mddev->chunk_sectors << 9);
3200 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3201 }
3202
3203 static ssize_t
3204 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
3205 {
3206 char *e;
3207 unsigned long n = simple_strtoul(buf, &e, 10);
3208
3209 if (!*buf || (*e && *e != '\n'))
3210 return -EINVAL;
3211
3212 if (mddev->pers) {
3213 int err;
3214 if (mddev->pers->check_reshape == NULL)
3215 return -EBUSY;
3216 mddev->new_chunk_sectors = n >> 9;
3217 err = mddev->pers->check_reshape(mddev);
3218 if (err) {
3219 mddev->new_chunk_sectors = mddev->chunk_sectors;
3220 return err;
3221 }
3222 } else {
3223 mddev->new_chunk_sectors = n >> 9;
3224 if (mddev->reshape_position == MaxSector)
3225 mddev->chunk_sectors = n >> 9;
3226 }
3227 return len;
3228 }
3229 static struct md_sysfs_entry md_chunk_size =
3230 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3231
3232 static ssize_t
3233 resync_start_show(mddev_t *mddev, char *page)
3234 {
3235 if (mddev->recovery_cp == MaxSector)
3236 return sprintf(page, "none\n");
3237 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3238 }
3239
3240 static ssize_t
3241 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
3242 {
3243 char *e;
3244 unsigned long long n = simple_strtoull(buf, &e, 10);
3245
3246 if (mddev->pers)
3247 return -EBUSY;
3248 if (cmd_match(buf, "none"))
3249 n = MaxSector;
3250 else if (!*buf || (*e && *e != '\n'))
3251 return -EINVAL;
3252
3253 mddev->recovery_cp = n;
3254 return len;
3255 }
3256 static struct md_sysfs_entry md_resync_start =
3257 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3258
3259 /*
3260 * The array state can be:
3261 *
3262 * clear
3263 * No devices, no size, no level
3264 * Equivalent to STOP_ARRAY ioctl
3265 * inactive
3266 * May have some settings, but array is not active
3267 * all IO results in error
3268 * When written, doesn't tear down array, but just stops it
3269 * suspended (not supported yet)
3270 * All IO requests will block. The array can be reconfigured.
3271 * Writing this, if accepted, will block until array is quiescent
3272 * readonly
3273 * no resync can happen. no superblocks get written.
3274 * write requests fail
3275 * read-auto
3276 * like readonly, but behaves like 'clean' on a write request.
3277 *
3278 * clean - no pending writes, but otherwise active.
3279 * When written to inactive array, starts without resync
3280 * If a write request arrives then
3281 * if metadata is known, mark 'dirty' and switch to 'active'.
3282 * if not known, block and switch to write-pending
3283 * If written to an active array that has pending writes, then fails.
3284 * active
3285 * fully active: IO and resync can be happening.
3286 * When written to inactive array, starts with resync
3287 *
3288 * write-pending
3289 * clean, but writes are blocked waiting for 'active' to be written.
3290 *
3291 * active-idle
3292 * like active, but no writes have been seen for a while (100msec).
3293 *
3294 */
3295 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3296 write_pending, active_idle, bad_word};
3297 static char *array_states[] = {
3298 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3299 "write-pending", "active-idle", NULL };
3300
3301 static int match_word(const char *word, char **list)
3302 {
3303 int n;
3304 for (n=0; list[n]; n++)
3305 if (cmd_match(word, list[n]))
3306 break;
3307 return n;
3308 }
3309
3310 static ssize_t
3311 array_state_show(mddev_t *mddev, char *page)
3312 {
3313 enum array_state st = inactive;
3314
3315 if (mddev->pers)
3316 switch(mddev->ro) {
3317 case 1:
3318 st = readonly;
3319 break;
3320 case 2:
3321 st = read_auto;
3322 break;
3323 case 0:
3324 if (mddev->in_sync)
3325 st = clean;
3326 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
3327 st = write_pending;
3328 else if (mddev->safemode)
3329 st = active_idle;
3330 else
3331 st = active;
3332 }
3333 else {
3334 if (list_empty(&mddev->disks) &&
3335 mddev->raid_disks == 0 &&
3336 mddev->dev_sectors == 0)
3337 st = clear;
3338 else
3339 st = inactive;
3340 }
3341 return sprintf(page, "%s\n", array_states[st]);
3342 }
3343
3344 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3345 static int md_set_readonly(mddev_t * mddev, int is_open);
3346 static int do_md_run(mddev_t * mddev);
3347 static int restart_array(mddev_t *mddev);
3348
3349 static ssize_t
3350 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3351 {
3352 int err = -EINVAL;
3353 enum array_state st = match_word(buf, array_states);
3354 switch(st) {
3355 case bad_word:
3356 break;
3357 case clear:
3358 /* stopping an active array */
3359 if (atomic_read(&mddev->openers) > 0)
3360 return -EBUSY;
3361 err = do_md_stop(mddev, 0, 0);
3362 break;
3363 case inactive:
3364 /* stopping an active array */
3365 if (mddev->pers) {
3366 if (atomic_read(&mddev->openers) > 0)
3367 return -EBUSY;
3368 err = do_md_stop(mddev, 2, 0);
3369 } else
3370 err = 0; /* already inactive */
3371 break;
3372 case suspended:
3373 break; /* not supported yet */
3374 case readonly:
3375 if (mddev->pers)
3376 err = md_set_readonly(mddev, 0);
3377 else {
3378 mddev->ro = 1;
3379 set_disk_ro(mddev->gendisk, 1);
3380 err = do_md_run(mddev);
3381 }
3382 break;
3383 case read_auto:
3384 if (mddev->pers) {
3385 if (mddev->ro == 0)
3386 err = md_set_readonly(mddev, 0);
3387 else if (mddev->ro == 1)
3388 err = restart_array(mddev);
3389 if (err == 0) {
3390 mddev->ro = 2;
3391 set_disk_ro(mddev->gendisk, 0);
3392 }
3393 } else {
3394 mddev->ro = 2;
3395 err = do_md_run(mddev);
3396 }
3397 break;
3398 case clean:
3399 if (mddev->pers) {
3400 restart_array(mddev);
3401 spin_lock_irq(&mddev->write_lock);
3402 if (atomic_read(&mddev->writes_pending) == 0) {
3403 if (mddev->in_sync == 0) {
3404 mddev->in_sync = 1;
3405 if (mddev->safemode == 1)
3406 mddev->safemode = 0;
3407 if (mddev->persistent)
3408 set_bit(MD_CHANGE_CLEAN,
3409 &mddev->flags);
3410 }
3411 err = 0;
3412 } else
3413 err = -EBUSY;
3414 spin_unlock_irq(&mddev->write_lock);
3415 } else
3416 err = -EINVAL;
3417 break;
3418 case active:
3419 if (mddev->pers) {
3420 restart_array(mddev);
3421 if (mddev->external)
3422 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3423 wake_up(&mddev->sb_wait);
3424 err = 0;
3425 } else {
3426 mddev->ro = 0;
3427 set_disk_ro(mddev->gendisk, 0);
3428 err = do_md_run(mddev);
3429 }
3430 break;
3431 case write_pending:
3432 case active_idle:
3433 /* these cannot be set */
3434 break;
3435 }
3436 if (err)
3437 return err;
3438 else {
3439 sysfs_notify_dirent(mddev->sysfs_state);
3440 return len;
3441 }
3442 }
3443 static struct md_sysfs_entry md_array_state =
3444 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3445
3446 static ssize_t
3447 max_corrected_read_errors_show(mddev_t *mddev, char *page) {
3448 return sprintf(page, "%d\n",
3449 atomic_read(&mddev->max_corr_read_errors));
3450 }
3451
3452 static ssize_t
3453 max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len)
3454 {
3455 char *e;
3456 unsigned long n = simple_strtoul(buf, &e, 10);
3457
3458 if (*buf && (*e == 0 || *e == '\n')) {
3459 atomic_set(&mddev->max_corr_read_errors, n);
3460 return len;
3461 }
3462 return -EINVAL;
3463 }
3464
3465 static struct md_sysfs_entry max_corr_read_errors =
3466 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3467 max_corrected_read_errors_store);
3468
3469 static ssize_t
3470 null_show(mddev_t *mddev, char *page)
3471 {
3472 return -EINVAL;
3473 }
3474
3475 static ssize_t
3476 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3477 {
3478 /* buf must be %d:%d\n? giving major and minor numbers */
3479 /* The new device is added to the array.
3480 * If the array has a persistent superblock, we read the
3481 * superblock to initialise info and check validity.
3482 * Otherwise, only checking done is that in bind_rdev_to_array,
3483 * which mainly checks size.
3484 */
3485 char *e;
3486 int major = simple_strtoul(buf, &e, 10);
3487 int minor;
3488 dev_t dev;
3489 mdk_rdev_t *rdev;
3490 int err;
3491
3492 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3493 return -EINVAL;
3494 minor = simple_strtoul(e+1, &e, 10);
3495 if (*e && *e != '\n')
3496 return -EINVAL;
3497 dev = MKDEV(major, minor);
3498 if (major != MAJOR(dev) ||
3499 minor != MINOR(dev))
3500 return -EOVERFLOW;
3501
3502
3503 if (mddev->persistent) {
3504 rdev = md_import_device(dev, mddev->major_version,
3505 mddev->minor_version);
3506 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3507 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3508 mdk_rdev_t, same_set);
3509 err = super_types[mddev->major_version]
3510 .load_super(rdev, rdev0, mddev->minor_version);
3511 if (err < 0)
3512 goto out;
3513 }
3514 } else if (mddev->external)
3515 rdev = md_import_device(dev, -2, -1);
3516 else
3517 rdev = md_import_device(dev, -1, -1);
3518
3519 if (IS_ERR(rdev))
3520 return PTR_ERR(rdev);
3521 err = bind_rdev_to_array(rdev, mddev);
3522 out:
3523 if (err)
3524 export_rdev(rdev);
3525 return err ? err : len;
3526 }
3527
3528 static struct md_sysfs_entry md_new_device =
3529 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3530
3531 static ssize_t
3532 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3533 {
3534 char *end;
3535 unsigned long chunk, end_chunk;
3536
3537 if (!mddev->bitmap)
3538 goto out;
3539 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3540 while (*buf) {
3541 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3542 if (buf == end) break;
3543 if (*end == '-') { /* range */
3544 buf = end + 1;
3545 end_chunk = simple_strtoul(buf, &end, 0);
3546 if (buf == end) break;
3547 }
3548 if (*end && !isspace(*end)) break;
3549 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3550 buf = skip_spaces(end);
3551 }
3552 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3553 out:
3554 return len;
3555 }
3556
3557 static struct md_sysfs_entry md_bitmap =
3558 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3559
3560 static ssize_t
3561 size_show(mddev_t *mddev, char *page)
3562 {
3563 return sprintf(page, "%llu\n",
3564 (unsigned long long)mddev->dev_sectors / 2);
3565 }
3566
3567 static int update_size(mddev_t *mddev, sector_t num_sectors);
3568
3569 static ssize_t
3570 size_store(mddev_t *mddev, const char *buf, size_t len)
3571 {
3572 /* If array is inactive, we can reduce the component size, but
3573 * not increase it (except from 0).
3574 * If array is active, we can try an on-line resize
3575 */
3576 sector_t sectors;
3577 int err = strict_blocks_to_sectors(buf, &sectors);
3578
3579 if (err < 0)
3580 return err;
3581 if (mddev->pers) {
3582 err = update_size(mddev, sectors);
3583 md_update_sb(mddev, 1);
3584 } else {
3585 if (mddev->dev_sectors == 0 ||
3586 mddev->dev_sectors > sectors)
3587 mddev->dev_sectors = sectors;
3588 else
3589 err = -ENOSPC;
3590 }
3591 return err ? err : len;
3592 }
3593
3594 static struct md_sysfs_entry md_size =
3595 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3596
3597
3598 /* Metdata version.
3599 * This is one of
3600 * 'none' for arrays with no metadata (good luck...)
3601 * 'external' for arrays with externally managed metadata,
3602 * or N.M for internally known formats
3603 */
3604 static ssize_t
3605 metadata_show(mddev_t *mddev, char *page)
3606 {
3607 if (mddev->persistent)
3608 return sprintf(page, "%d.%d\n",
3609 mddev->major_version, mddev->minor_version);
3610 else if (mddev->external)
3611 return sprintf(page, "external:%s\n", mddev->metadata_type);
3612 else
3613 return sprintf(page, "none\n");
3614 }
3615
3616 static ssize_t
3617 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3618 {
3619 int major, minor;
3620 char *e;
3621 /* Changing the details of 'external' metadata is
3622 * always permitted. Otherwise there must be
3623 * no devices attached to the array.
3624 */
3625 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3626 ;
3627 else if (!list_empty(&mddev->disks))
3628 return -EBUSY;
3629
3630 if (cmd_match(buf, "none")) {
3631 mddev->persistent = 0;
3632 mddev->external = 0;
3633 mddev->major_version = 0;
3634 mddev->minor_version = 90;
3635 return len;
3636 }
3637 if (strncmp(buf, "external:", 9) == 0) {
3638 size_t namelen = len-9;
3639 if (namelen >= sizeof(mddev->metadata_type))
3640 namelen = sizeof(mddev->metadata_type)-1;
3641 strncpy(mddev->metadata_type, buf+9, namelen);
3642 mddev->metadata_type[namelen] = 0;
3643 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3644 mddev->metadata_type[--namelen] = 0;
3645 mddev->persistent = 0;
3646 mddev->external = 1;
3647 mddev->major_version = 0;
3648 mddev->minor_version = 90;
3649 return len;
3650 }
3651 major = simple_strtoul(buf, &e, 10);
3652 if (e==buf || *e != '.')
3653 return -EINVAL;
3654 buf = e+1;
3655 minor = simple_strtoul(buf, &e, 10);
3656 if (e==buf || (*e && *e != '\n') )
3657 return -EINVAL;
3658 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3659 return -ENOENT;
3660 mddev->major_version = major;
3661 mddev->minor_version = minor;
3662 mddev->persistent = 1;
3663 mddev->external = 0;
3664 return len;
3665 }
3666
3667 static struct md_sysfs_entry md_metadata =
3668 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3669
3670 static ssize_t
3671 action_show(mddev_t *mddev, char *page)
3672 {
3673 char *type = "idle";
3674 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3675 type = "frozen";
3676 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3677 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3678 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3679 type = "reshape";
3680 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3681 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3682 type = "resync";
3683 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3684 type = "check";
3685 else
3686 type = "repair";
3687 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3688 type = "recover";
3689 }
3690 return sprintf(page, "%s\n", type);
3691 }
3692
3693 static ssize_t
3694 action_store(mddev_t *mddev, const char *page, size_t len)
3695 {
3696 if (!mddev->pers || !mddev->pers->sync_request)
3697 return -EINVAL;
3698
3699 if (cmd_match(page, "frozen"))
3700 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3701 else
3702 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3703
3704 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3705 if (mddev->sync_thread) {
3706 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3707 md_unregister_thread(mddev->sync_thread);
3708 mddev->sync_thread = NULL;
3709 mddev->recovery = 0;
3710 }
3711 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3712 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3713 return -EBUSY;
3714 else if (cmd_match(page, "resync"))
3715 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3716 else if (cmd_match(page, "recover")) {
3717 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3718 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3719 } else if (cmd_match(page, "reshape")) {
3720 int err;
3721 if (mddev->pers->start_reshape == NULL)
3722 return -EINVAL;
3723 err = mddev->pers->start_reshape(mddev);
3724 if (err)
3725 return err;
3726 sysfs_notify(&mddev->kobj, NULL, "degraded");
3727 } else {
3728 if (cmd_match(page, "check"))
3729 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3730 else if (!cmd_match(page, "repair"))
3731 return -EINVAL;
3732 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3733 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3734 }
3735 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3736 md_wakeup_thread(mddev->thread);
3737 sysfs_notify_dirent(mddev->sysfs_action);
3738 return len;
3739 }
3740
3741 static ssize_t
3742 mismatch_cnt_show(mddev_t *mddev, char *page)
3743 {
3744 return sprintf(page, "%llu\n",
3745 (unsigned long long) mddev->resync_mismatches);
3746 }
3747
3748 static struct md_sysfs_entry md_scan_mode =
3749 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3750
3751
3752 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3753
3754 static ssize_t
3755 sync_min_show(mddev_t *mddev, char *page)
3756 {
3757 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3758 mddev->sync_speed_min ? "local": "system");
3759 }
3760
3761 static ssize_t
3762 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3763 {
3764 int min;
3765 char *e;
3766 if (strncmp(buf, "system", 6)==0) {
3767 mddev->sync_speed_min = 0;
3768 return len;
3769 }
3770 min = simple_strtoul(buf, &e, 10);
3771 if (buf == e || (*e && *e != '\n') || min <= 0)
3772 return -EINVAL;
3773 mddev->sync_speed_min = min;
3774 return len;
3775 }
3776
3777 static struct md_sysfs_entry md_sync_min =
3778 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3779
3780 static ssize_t
3781 sync_max_show(mddev_t *mddev, char *page)
3782 {
3783 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3784 mddev->sync_speed_max ? "local": "system");
3785 }
3786
3787 static ssize_t
3788 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3789 {
3790 int max;
3791 char *e;
3792 if (strncmp(buf, "system", 6)==0) {
3793 mddev->sync_speed_max = 0;
3794 return len;
3795 }
3796 max = simple_strtoul(buf, &e, 10);
3797 if (buf == e || (*e && *e != '\n') || max <= 0)
3798 return -EINVAL;
3799 mddev->sync_speed_max = max;
3800 return len;
3801 }
3802
3803 static struct md_sysfs_entry md_sync_max =
3804 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3805
3806 static ssize_t
3807 degraded_show(mddev_t *mddev, char *page)
3808 {
3809 return sprintf(page, "%d\n", mddev->degraded);
3810 }
3811 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3812
3813 static ssize_t
3814 sync_force_parallel_show(mddev_t *mddev, char *page)
3815 {
3816 return sprintf(page, "%d\n", mddev->parallel_resync);
3817 }
3818
3819 static ssize_t
3820 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3821 {
3822 long n;
3823
3824 if (strict_strtol(buf, 10, &n))
3825 return -EINVAL;
3826
3827 if (n != 0 && n != 1)
3828 return -EINVAL;
3829
3830 mddev->parallel_resync = n;
3831
3832 if (mddev->sync_thread)
3833 wake_up(&resync_wait);
3834
3835 return len;
3836 }
3837
3838 /* force parallel resync, even with shared block devices */
3839 static struct md_sysfs_entry md_sync_force_parallel =
3840 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3841 sync_force_parallel_show, sync_force_parallel_store);
3842
3843 static ssize_t
3844 sync_speed_show(mddev_t *mddev, char *page)
3845 {
3846 unsigned long resync, dt, db;
3847 if (mddev->curr_resync == 0)
3848 return sprintf(page, "none\n");
3849 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3850 dt = (jiffies - mddev->resync_mark) / HZ;
3851 if (!dt) dt++;
3852 db = resync - mddev->resync_mark_cnt;
3853 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3854 }
3855
3856 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3857
3858 static ssize_t
3859 sync_completed_show(mddev_t *mddev, char *page)
3860 {
3861 unsigned long max_sectors, resync;
3862
3863 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3864 return sprintf(page, "none\n");
3865
3866 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3867 max_sectors = mddev->resync_max_sectors;
3868 else
3869 max_sectors = mddev->dev_sectors;
3870
3871 resync = mddev->curr_resync_completed;
3872 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3873 }
3874
3875 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3876
3877 static ssize_t
3878 min_sync_show(mddev_t *mddev, char *page)
3879 {
3880 return sprintf(page, "%llu\n",
3881 (unsigned long long)mddev->resync_min);
3882 }
3883 static ssize_t
3884 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3885 {
3886 unsigned long long min;
3887 if (strict_strtoull(buf, 10, &min))
3888 return -EINVAL;
3889 if (min > mddev->resync_max)
3890 return -EINVAL;
3891 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3892 return -EBUSY;
3893
3894 /* Must be a multiple of chunk_size */
3895 if (mddev->chunk_sectors) {
3896 sector_t temp = min;
3897 if (sector_div(temp, mddev->chunk_sectors))
3898 return -EINVAL;
3899 }
3900 mddev->resync_min = min;
3901
3902 return len;
3903 }
3904
3905 static struct md_sysfs_entry md_min_sync =
3906 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3907
3908 static ssize_t
3909 max_sync_show(mddev_t *mddev, char *page)
3910 {
3911 if (mddev->resync_max == MaxSector)
3912 return sprintf(page, "max\n");
3913 else
3914 return sprintf(page, "%llu\n",
3915 (unsigned long long)mddev->resync_max);
3916 }
3917 static ssize_t
3918 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3919 {
3920 if (strncmp(buf, "max", 3) == 0)
3921 mddev->resync_max = MaxSector;
3922 else {
3923 unsigned long long max;
3924 if (strict_strtoull(buf, 10, &max))
3925 return -EINVAL;
3926 if (max < mddev->resync_min)
3927 return -EINVAL;
3928 if (max < mddev->resync_max &&
3929 mddev->ro == 0 &&
3930 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3931 return -EBUSY;
3932
3933 /* Must be a multiple of chunk_size */
3934 if (mddev->chunk_sectors) {
3935 sector_t temp = max;
3936 if (sector_div(temp, mddev->chunk_sectors))
3937 return -EINVAL;
3938 }
3939 mddev->resync_max = max;
3940 }
3941 wake_up(&mddev->recovery_wait);
3942 return len;
3943 }
3944
3945 static struct md_sysfs_entry md_max_sync =
3946 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3947
3948 static ssize_t
3949 suspend_lo_show(mddev_t *mddev, char *page)
3950 {
3951 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3952 }
3953
3954 static ssize_t
3955 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3956 {
3957 char *e;
3958 unsigned long long new = simple_strtoull(buf, &e, 10);
3959
3960 if (mddev->pers == NULL ||
3961 mddev->pers->quiesce == NULL)
3962 return -EINVAL;
3963 if (buf == e || (*e && *e != '\n'))
3964 return -EINVAL;
3965 if (new >= mddev->suspend_hi ||
3966 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3967 mddev->suspend_lo = new;
3968 mddev->pers->quiesce(mddev, 2);
3969 return len;
3970 } else
3971 return -EINVAL;
3972 }
3973 static struct md_sysfs_entry md_suspend_lo =
3974 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3975
3976
3977 static ssize_t
3978 suspend_hi_show(mddev_t *mddev, char *page)
3979 {
3980 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3981 }
3982
3983 static ssize_t
3984 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3985 {
3986 char *e;
3987 unsigned long long new = simple_strtoull(buf, &e, 10);
3988
3989 if (mddev->pers == NULL ||
3990 mddev->pers->quiesce == NULL)
3991 return -EINVAL;
3992 if (buf == e || (*e && *e != '\n'))
3993 return -EINVAL;
3994 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3995 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3996 mddev->suspend_hi = new;
3997 mddev->pers->quiesce(mddev, 1);
3998 mddev->pers->quiesce(mddev, 0);
3999 return len;
4000 } else
4001 return -EINVAL;
4002 }
4003 static struct md_sysfs_entry md_suspend_hi =
4004 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4005
4006 static ssize_t
4007 reshape_position_show(mddev_t *mddev, char *page)
4008 {
4009 if (mddev->reshape_position != MaxSector)
4010 return sprintf(page, "%llu\n",
4011 (unsigned long long)mddev->reshape_position);
4012 strcpy(page, "none\n");
4013 return 5;
4014 }
4015
4016 static ssize_t
4017 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
4018 {
4019 char *e;
4020 unsigned long long new = simple_strtoull(buf, &e, 10);
4021 if (mddev->pers)
4022 return -EBUSY;
4023 if (buf == e || (*e && *e != '\n'))
4024 return -EINVAL;
4025 mddev->reshape_position = new;
4026 mddev->delta_disks = 0;
4027 mddev->new_level = mddev->level;
4028 mddev->new_layout = mddev->layout;
4029 mddev->new_chunk_sectors = mddev->chunk_sectors;
4030 return len;
4031 }
4032
4033 static struct md_sysfs_entry md_reshape_position =
4034 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4035 reshape_position_store);
4036
4037 static ssize_t
4038 array_size_show(mddev_t *mddev, char *page)
4039 {
4040 if (mddev->external_size)
4041 return sprintf(page, "%llu\n",
4042 (unsigned long long)mddev->array_sectors/2);
4043 else
4044 return sprintf(page, "default\n");
4045 }
4046
4047 static ssize_t
4048 array_size_store(mddev_t *mddev, const char *buf, size_t len)
4049 {
4050 sector_t sectors;
4051
4052 if (strncmp(buf, "default", 7) == 0) {
4053 if (mddev->pers)
4054 sectors = mddev->pers->size(mddev, 0, 0);
4055 else
4056 sectors = mddev->array_sectors;
4057
4058 mddev->external_size = 0;
4059 } else {
4060 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4061 return -EINVAL;
4062 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4063 return -E2BIG;
4064
4065 mddev->external_size = 1;
4066 }
4067
4068 mddev->array_sectors = sectors;
4069 set_capacity(mddev->gendisk, mddev->array_sectors);
4070 if (mddev->pers)
4071 revalidate_disk(mddev->gendisk);
4072
4073 return len;
4074 }
4075
4076 static struct md_sysfs_entry md_array_size =
4077 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4078 array_size_store);
4079
4080 static struct attribute *md_default_attrs[] = {
4081 &md_level.attr,
4082 &md_layout.attr,
4083 &md_raid_disks.attr,
4084 &md_chunk_size.attr,
4085 &md_size.attr,
4086 &md_resync_start.attr,
4087 &md_metadata.attr,
4088 &md_new_device.attr,
4089 &md_safe_delay.attr,
4090 &md_array_state.attr,
4091 &md_reshape_position.attr,
4092 &md_array_size.attr,
4093 &max_corr_read_errors.attr,
4094 NULL,
4095 };
4096
4097 static struct attribute *md_redundancy_attrs[] = {
4098 &md_scan_mode.attr,
4099 &md_mismatches.attr,
4100 &md_sync_min.attr,
4101 &md_sync_max.attr,
4102 &md_sync_speed.attr,
4103 &md_sync_force_parallel.attr,
4104 &md_sync_completed.attr,
4105 &md_min_sync.attr,
4106 &md_max_sync.attr,
4107 &md_suspend_lo.attr,
4108 &md_suspend_hi.attr,
4109 &md_bitmap.attr,
4110 &md_degraded.attr,
4111 NULL,
4112 };
4113 static struct attribute_group md_redundancy_group = {
4114 .name = NULL,
4115 .attrs = md_redundancy_attrs,
4116 };
4117
4118
4119 static ssize_t
4120 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4121 {
4122 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4123 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4124 ssize_t rv;
4125
4126 if (!entry->show)
4127 return -EIO;
4128 rv = mddev_lock(mddev);
4129 if (!rv) {
4130 rv = entry->show(mddev, page);
4131 mddev_unlock(mddev);
4132 }
4133 return rv;
4134 }
4135
4136 static ssize_t
4137 md_attr_store(struct kobject *kobj, struct attribute *attr,
4138 const char *page, size_t length)
4139 {
4140 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4141 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4142 ssize_t rv;
4143
4144 if (!entry->store)
4145 return -EIO;
4146 if (!capable(CAP_SYS_ADMIN))
4147 return -EACCES;
4148 rv = mddev_lock(mddev);
4149 if (mddev->hold_active == UNTIL_IOCTL)
4150 mddev->hold_active = 0;
4151 if (!rv) {
4152 rv = entry->store(mddev, page, length);
4153 mddev_unlock(mddev);
4154 }
4155 return rv;
4156 }
4157
4158 static void md_free(struct kobject *ko)
4159 {
4160 mddev_t *mddev = container_of(ko, mddev_t, kobj);
4161
4162 if (mddev->sysfs_state)
4163 sysfs_put(mddev->sysfs_state);
4164
4165 if (mddev->gendisk) {
4166 del_gendisk(mddev->gendisk);
4167 put_disk(mddev->gendisk);
4168 }
4169 if (mddev->queue)
4170 blk_cleanup_queue(mddev->queue);
4171
4172 kfree(mddev);
4173 }
4174
4175 static const struct sysfs_ops md_sysfs_ops = {
4176 .show = md_attr_show,
4177 .store = md_attr_store,
4178 };
4179 static struct kobj_type md_ktype = {
4180 .release = md_free,
4181 .sysfs_ops = &md_sysfs_ops,
4182 .default_attrs = md_default_attrs,
4183 };
4184
4185 int mdp_major = 0;
4186
4187 static void mddev_delayed_delete(struct work_struct *ws)
4188 {
4189 mddev_t *mddev = container_of(ws, mddev_t, del_work);
4190
4191 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4192 kobject_del(&mddev->kobj);
4193 kobject_put(&mddev->kobj);
4194 }
4195
4196 static int md_alloc(dev_t dev, char *name)
4197 {
4198 static DEFINE_MUTEX(disks_mutex);
4199 mddev_t *mddev = mddev_find(dev);
4200 struct gendisk *disk;
4201 int partitioned;
4202 int shift;
4203 int unit;
4204 int error;
4205
4206 if (!mddev)
4207 return -ENODEV;
4208
4209 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4210 shift = partitioned ? MdpMinorShift : 0;
4211 unit = MINOR(mddev->unit) >> shift;
4212
4213 /* wait for any previous instance if this device
4214 * to be completed removed (mddev_delayed_delete).
4215 */
4216 flush_scheduled_work();
4217
4218 mutex_lock(&disks_mutex);
4219 error = -EEXIST;
4220 if (mddev->gendisk)
4221 goto abort;
4222
4223 if (name) {
4224 /* Need to ensure that 'name' is not a duplicate.
4225 */
4226 mddev_t *mddev2;
4227 spin_lock(&all_mddevs_lock);
4228
4229 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4230 if (mddev2->gendisk &&
4231 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4232 spin_unlock(&all_mddevs_lock);
4233 goto abort;
4234 }
4235 spin_unlock(&all_mddevs_lock);
4236 }
4237
4238 error = -ENOMEM;
4239 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4240 if (!mddev->queue)
4241 goto abort;
4242 mddev->queue->queuedata = mddev;
4243
4244 /* Can be unlocked because the queue is new: no concurrency */
4245 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
4246
4247 blk_queue_make_request(mddev->queue, md_make_request);
4248
4249 disk = alloc_disk(1 << shift);
4250 if (!disk) {
4251 blk_cleanup_queue(mddev->queue);
4252 mddev->queue = NULL;
4253 goto abort;
4254 }
4255 disk->major = MAJOR(mddev->unit);
4256 disk->first_minor = unit << shift;
4257 if (name)
4258 strcpy(disk->disk_name, name);
4259 else if (partitioned)
4260 sprintf(disk->disk_name, "md_d%d", unit);
4261 else
4262 sprintf(disk->disk_name, "md%d", unit);
4263 disk->fops = &md_fops;
4264 disk->private_data = mddev;
4265 disk->queue = mddev->queue;
4266 /* Allow extended partitions. This makes the
4267 * 'mdp' device redundant, but we can't really
4268 * remove it now.
4269 */
4270 disk->flags |= GENHD_FL_EXT_DEVT;
4271 add_disk(disk);
4272 mddev->gendisk = disk;
4273 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4274 &disk_to_dev(disk)->kobj, "%s", "md");
4275 if (error) {
4276 /* This isn't possible, but as kobject_init_and_add is marked
4277 * __must_check, we must do something with the result
4278 */
4279 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4280 disk->disk_name);
4281 error = 0;
4282 }
4283 if (sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4284 printk(KERN_DEBUG "pointless warning\n");
4285 abort:
4286 mutex_unlock(&disks_mutex);
4287 if (!error) {
4288 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4289 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, NULL, "array_state");
4290 }
4291 mddev_put(mddev);
4292 return error;
4293 }
4294
4295 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4296 {
4297 md_alloc(dev, NULL);
4298 return NULL;
4299 }
4300
4301 static int add_named_array(const char *val, struct kernel_param *kp)
4302 {
4303 /* val must be "md_*" where * is not all digits.
4304 * We allocate an array with a large free minor number, and
4305 * set the name to val. val must not already be an active name.
4306 */
4307 int len = strlen(val);
4308 char buf[DISK_NAME_LEN];
4309
4310 while (len && val[len-1] == '\n')
4311 len--;
4312 if (len >= DISK_NAME_LEN)
4313 return -E2BIG;
4314 strlcpy(buf, val, len+1);
4315 if (strncmp(buf, "md_", 3) != 0)
4316 return -EINVAL;
4317 return md_alloc(0, buf);
4318 }
4319
4320 static void md_safemode_timeout(unsigned long data)
4321 {
4322 mddev_t *mddev = (mddev_t *) data;
4323
4324 if (!atomic_read(&mddev->writes_pending)) {
4325 mddev->safemode = 1;
4326 if (mddev->external)
4327 sysfs_notify_dirent(mddev->sysfs_state);
4328 }
4329 md_wakeup_thread(mddev->thread);
4330 }
4331
4332 static int start_dirty_degraded;
4333
4334 static int md_run(mddev_t *mddev)
4335 {
4336 int err;
4337 mdk_rdev_t *rdev;
4338 struct mdk_personality *pers;
4339
4340 if (list_empty(&mddev->disks))
4341 /* cannot run an array with no devices.. */
4342 return -EINVAL;
4343
4344 if (mddev->pers)
4345 return -EBUSY;
4346
4347 /* These two calls synchronise us with the
4348 * sysfs_remove_group calls in mddev_unlock,
4349 * so they must have completed.
4350 */
4351 mutex_lock(&mddev->open_mutex);
4352 mutex_unlock(&mddev->open_mutex);
4353
4354 /*
4355 * Analyze all RAID superblock(s)
4356 */
4357 if (!mddev->raid_disks) {
4358 if (!mddev->persistent)
4359 return -EINVAL;
4360 analyze_sbs(mddev);
4361 }
4362
4363 if (mddev->level != LEVEL_NONE)
4364 request_module("md-level-%d", mddev->level);
4365 else if (mddev->clevel[0])
4366 request_module("md-%s", mddev->clevel);
4367
4368 /*
4369 * Drop all container device buffers, from now on
4370 * the only valid external interface is through the md
4371 * device.
4372 */
4373 list_for_each_entry(rdev, &mddev->disks, same_set) {
4374 if (test_bit(Faulty, &rdev->flags))
4375 continue;
4376 sync_blockdev(rdev->bdev);
4377 invalidate_bdev(rdev->bdev);
4378
4379 /* perform some consistency tests on the device.
4380 * We don't want the data to overlap the metadata,
4381 * Internal Bitmap issues have been handled elsewhere.
4382 */
4383 if (rdev->data_offset < rdev->sb_start) {
4384 if (mddev->dev_sectors &&
4385 rdev->data_offset + mddev->dev_sectors
4386 > rdev->sb_start) {
4387 printk("md: %s: data overlaps metadata\n",
4388 mdname(mddev));
4389 return -EINVAL;
4390 }
4391 } else {
4392 if (rdev->sb_start + rdev->sb_size/512
4393 > rdev->data_offset) {
4394 printk("md: %s: metadata overlaps data\n",
4395 mdname(mddev));
4396 return -EINVAL;
4397 }
4398 }
4399 sysfs_notify_dirent(rdev->sysfs_state);
4400 }
4401
4402 spin_lock(&pers_lock);
4403 pers = find_pers(mddev->level, mddev->clevel);
4404 if (!pers || !try_module_get(pers->owner)) {
4405 spin_unlock(&pers_lock);
4406 if (mddev->level != LEVEL_NONE)
4407 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4408 mddev->level);
4409 else
4410 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4411 mddev->clevel);
4412 return -EINVAL;
4413 }
4414 mddev->pers = pers;
4415 spin_unlock(&pers_lock);
4416 if (mddev->level != pers->level) {
4417 mddev->level = pers->level;
4418 mddev->new_level = pers->level;
4419 }
4420 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4421
4422 if (mddev->reshape_position != MaxSector &&
4423 pers->start_reshape == NULL) {
4424 /* This personality cannot handle reshaping... */
4425 mddev->pers = NULL;
4426 module_put(pers->owner);
4427 return -EINVAL;
4428 }
4429
4430 if (pers->sync_request) {
4431 /* Warn if this is a potentially silly
4432 * configuration.
4433 */
4434 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4435 mdk_rdev_t *rdev2;
4436 int warned = 0;
4437
4438 list_for_each_entry(rdev, &mddev->disks, same_set)
4439 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4440 if (rdev < rdev2 &&
4441 rdev->bdev->bd_contains ==
4442 rdev2->bdev->bd_contains) {
4443 printk(KERN_WARNING
4444 "%s: WARNING: %s appears to be"
4445 " on the same physical disk as"
4446 " %s.\n",
4447 mdname(mddev),
4448 bdevname(rdev->bdev,b),
4449 bdevname(rdev2->bdev,b2));
4450 warned = 1;
4451 }
4452 }
4453
4454 if (warned)
4455 printk(KERN_WARNING
4456 "True protection against single-disk"
4457 " failure might be compromised.\n");
4458 }
4459
4460 mddev->recovery = 0;
4461 /* may be over-ridden by personality */
4462 mddev->resync_max_sectors = mddev->dev_sectors;
4463
4464 mddev->barriers_work = 1;
4465 mddev->ok_start_degraded = start_dirty_degraded;
4466
4467 if (start_readonly && mddev->ro == 0)
4468 mddev->ro = 2; /* read-only, but switch on first write */
4469
4470 err = mddev->pers->run(mddev);
4471 if (err)
4472 printk(KERN_ERR "md: pers->run() failed ...\n");
4473 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4474 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4475 " but 'external_size' not in effect?\n", __func__);
4476 printk(KERN_ERR
4477 "md: invalid array_size %llu > default size %llu\n",
4478 (unsigned long long)mddev->array_sectors / 2,
4479 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4480 err = -EINVAL;
4481 mddev->pers->stop(mddev);
4482 }
4483 if (err == 0 && mddev->pers->sync_request) {
4484 err = bitmap_create(mddev);
4485 if (err) {
4486 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4487 mdname(mddev), err);
4488 mddev->pers->stop(mddev);
4489 }
4490 }
4491 if (err) {
4492 module_put(mddev->pers->owner);
4493 mddev->pers = NULL;
4494 bitmap_destroy(mddev);
4495 return err;
4496 }
4497 if (mddev->pers->sync_request) {
4498 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4499 printk(KERN_WARNING
4500 "md: cannot register extra attributes for %s\n",
4501 mdname(mddev));
4502 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
4503 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4504 mddev->ro = 0;
4505
4506 atomic_set(&mddev->writes_pending,0);
4507 atomic_set(&mddev->max_corr_read_errors,
4508 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
4509 mddev->safemode = 0;
4510 mddev->safemode_timer.function = md_safemode_timeout;
4511 mddev->safemode_timer.data = (unsigned long) mddev;
4512 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4513 mddev->in_sync = 1;
4514
4515 list_for_each_entry(rdev, &mddev->disks, same_set)
4516 if (rdev->raid_disk >= 0) {
4517 char nm[20];
4518 sprintf(nm, "rd%d", rdev->raid_disk);
4519 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4520 printk("md: cannot register %s for %s\n",
4521 nm, mdname(mddev));
4522 }
4523
4524 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4525
4526 if (mddev->flags)
4527 md_update_sb(mddev, 0);
4528
4529 md_wakeup_thread(mddev->thread);
4530 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4531
4532 md_new_event(mddev);
4533 sysfs_notify_dirent(mddev->sysfs_state);
4534 if (mddev->sysfs_action)
4535 sysfs_notify_dirent(mddev->sysfs_action);
4536 sysfs_notify(&mddev->kobj, NULL, "degraded");
4537 return 0;
4538 }
4539
4540 static int do_md_run(mddev_t *mddev)
4541 {
4542 int err;
4543
4544 err = md_run(mddev);
4545 if (err)
4546 goto out;
4547
4548 set_capacity(mddev->gendisk, mddev->array_sectors);
4549 revalidate_disk(mddev->gendisk);
4550 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4551 out:
4552 return err;
4553 }
4554
4555 static int restart_array(mddev_t *mddev)
4556 {
4557 struct gendisk *disk = mddev->gendisk;
4558
4559 /* Complain if it has no devices */
4560 if (list_empty(&mddev->disks))
4561 return -ENXIO;
4562 if (!mddev->pers)
4563 return -EINVAL;
4564 if (!mddev->ro)
4565 return -EBUSY;
4566 mddev->safemode = 0;
4567 mddev->ro = 0;
4568 set_disk_ro(disk, 0);
4569 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4570 mdname(mddev));
4571 /* Kick recovery or resync if necessary */
4572 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4573 md_wakeup_thread(mddev->thread);
4574 md_wakeup_thread(mddev->sync_thread);
4575 sysfs_notify_dirent(mddev->sysfs_state);
4576 return 0;
4577 }
4578
4579 /* similar to deny_write_access, but accounts for our holding a reference
4580 * to the file ourselves */
4581 static int deny_bitmap_write_access(struct file * file)
4582 {
4583 struct inode *inode = file->f_mapping->host;
4584
4585 spin_lock(&inode->i_lock);
4586 if (atomic_read(&inode->i_writecount) > 1) {
4587 spin_unlock(&inode->i_lock);
4588 return -ETXTBSY;
4589 }
4590 atomic_set(&inode->i_writecount, -1);
4591 spin_unlock(&inode->i_lock);
4592
4593 return 0;
4594 }
4595
4596 void restore_bitmap_write_access(struct file *file)
4597 {
4598 struct inode *inode = file->f_mapping->host;
4599
4600 spin_lock(&inode->i_lock);
4601 atomic_set(&inode->i_writecount, 1);
4602 spin_unlock(&inode->i_lock);
4603 }
4604
4605 static void md_clean(mddev_t *mddev)
4606 {
4607 mddev->array_sectors = 0;
4608 mddev->external_size = 0;
4609 mddev->dev_sectors = 0;
4610 mddev->raid_disks = 0;
4611 mddev->recovery_cp = 0;
4612 mddev->resync_min = 0;
4613 mddev->resync_max = MaxSector;
4614 mddev->reshape_position = MaxSector;
4615 mddev->external = 0;
4616 mddev->persistent = 0;
4617 mddev->level = LEVEL_NONE;
4618 mddev->clevel[0] = 0;
4619 mddev->flags = 0;
4620 mddev->ro = 0;
4621 mddev->metadata_type[0] = 0;
4622 mddev->chunk_sectors = 0;
4623 mddev->ctime = mddev->utime = 0;
4624 mddev->layout = 0;
4625 mddev->max_disks = 0;
4626 mddev->events = 0;
4627 mddev->can_decrease_events = 0;
4628 mddev->delta_disks = 0;
4629 mddev->new_level = LEVEL_NONE;
4630 mddev->new_layout = 0;
4631 mddev->new_chunk_sectors = 0;
4632 mddev->curr_resync = 0;
4633 mddev->resync_mismatches = 0;
4634 mddev->suspend_lo = mddev->suspend_hi = 0;
4635 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4636 mddev->recovery = 0;
4637 mddev->in_sync = 0;
4638 mddev->degraded = 0;
4639 mddev->barriers_work = 0;
4640 mddev->safemode = 0;
4641 mddev->bitmap_info.offset = 0;
4642 mddev->bitmap_info.default_offset = 0;
4643 mddev->bitmap_info.chunksize = 0;
4644 mddev->bitmap_info.daemon_sleep = 0;
4645 mddev->bitmap_info.max_write_behind = 0;
4646 }
4647
4648 static void md_stop_writes(mddev_t *mddev)
4649 {
4650 if (mddev->sync_thread) {
4651 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4652 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4653 md_unregister_thread(mddev->sync_thread);
4654 mddev->sync_thread = NULL;
4655 }
4656
4657 del_timer_sync(&mddev->safemode_timer);
4658
4659 bitmap_flush(mddev);
4660 md_super_wait(mddev);
4661
4662 if (!mddev->in_sync || mddev->flags) {
4663 /* mark array as shutdown cleanly */
4664 mddev->in_sync = 1;
4665 md_update_sb(mddev, 1);
4666 }
4667 }
4668
4669 static void md_stop(mddev_t *mddev)
4670 {
4671 md_stop_writes(mddev);
4672
4673 mddev->pers->stop(mddev);
4674 if (mddev->pers->sync_request && mddev->to_remove == NULL)
4675 mddev->to_remove = &md_redundancy_group;
4676 module_put(mddev->pers->owner);
4677 mddev->pers = NULL;
4678 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4679 }
4680
4681 static int md_set_readonly(mddev_t *mddev, int is_open)
4682 {
4683 int err = 0;
4684 mutex_lock(&mddev->open_mutex);
4685 if (atomic_read(&mddev->openers) > is_open) {
4686 printk("md: %s still in use.\n",mdname(mddev));
4687 err = -EBUSY;
4688 goto out;
4689 }
4690 if (mddev->pers) {
4691 md_stop_writes(mddev);
4692
4693 err = -ENXIO;
4694 if (mddev->ro==1)
4695 goto out;
4696 mddev->ro = 1;
4697 set_disk_ro(mddev->gendisk, 1);
4698 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4699 sysfs_notify_dirent(mddev->sysfs_state);
4700 err = 0;
4701 }
4702 out:
4703 mutex_unlock(&mddev->open_mutex);
4704 return err;
4705 }
4706
4707 /* mode:
4708 * 0 - completely stop and dis-assemble array
4709 * 2 - stop but do not disassemble array
4710 */
4711 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4712 {
4713 int err = 0;
4714 struct gendisk *disk = mddev->gendisk;
4715 mdk_rdev_t *rdev;
4716
4717 mutex_lock(&mddev->open_mutex);
4718 if (atomic_read(&mddev->openers) > is_open) {
4719 printk("md: %s still in use.\n",mdname(mddev));
4720 err = -EBUSY;
4721 } else if (mddev->pers) {
4722
4723 if (mddev->ro)
4724 set_disk_ro(disk, 0);
4725
4726 md_stop(mddev);
4727 mddev->queue->merge_bvec_fn = NULL;
4728 mddev->queue->unplug_fn = NULL;
4729 mddev->queue->backing_dev_info.congested_fn = NULL;
4730
4731 /* tell userspace to handle 'inactive' */
4732 sysfs_notify_dirent(mddev->sysfs_state);
4733
4734 list_for_each_entry(rdev, &mddev->disks, same_set)
4735 if (rdev->raid_disk >= 0) {
4736 char nm[20];
4737 sprintf(nm, "rd%d", rdev->raid_disk);
4738 sysfs_remove_link(&mddev->kobj, nm);
4739 }
4740
4741 set_capacity(disk, 0);
4742 revalidate_disk(disk);
4743
4744 if (mddev->ro)
4745 mddev->ro = 0;
4746
4747 err = 0;
4748 }
4749 mutex_unlock(&mddev->open_mutex);
4750 if (err)
4751 return err;
4752 /*
4753 * Free resources if final stop
4754 */
4755 if (mode == 0) {
4756
4757 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4758
4759 bitmap_destroy(mddev);
4760 if (mddev->bitmap_info.file) {
4761 restore_bitmap_write_access(mddev->bitmap_info.file);
4762 fput(mddev->bitmap_info.file);
4763 mddev->bitmap_info.file = NULL;
4764 }
4765 mddev->bitmap_info.offset = 0;
4766
4767 export_array(mddev);
4768
4769 md_clean(mddev);
4770 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4771 if (mddev->hold_active == UNTIL_STOP)
4772 mddev->hold_active = 0;
4773
4774 }
4775 err = 0;
4776 blk_integrity_unregister(disk);
4777 md_new_event(mddev);
4778 sysfs_notify_dirent(mddev->sysfs_state);
4779 return err;
4780 }
4781
4782 #ifndef MODULE
4783 static void autorun_array(mddev_t *mddev)
4784 {
4785 mdk_rdev_t *rdev;
4786 int err;
4787
4788 if (list_empty(&mddev->disks))
4789 return;
4790
4791 printk(KERN_INFO "md: running: ");
4792
4793 list_for_each_entry(rdev, &mddev->disks, same_set) {
4794 char b[BDEVNAME_SIZE];
4795 printk("<%s>", bdevname(rdev->bdev,b));
4796 }
4797 printk("\n");
4798
4799 err = do_md_run(mddev);
4800 if (err) {
4801 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4802 do_md_stop(mddev, 0, 0);
4803 }
4804 }
4805
4806 /*
4807 * lets try to run arrays based on all disks that have arrived
4808 * until now. (those are in pending_raid_disks)
4809 *
4810 * the method: pick the first pending disk, collect all disks with
4811 * the same UUID, remove all from the pending list and put them into
4812 * the 'same_array' list. Then order this list based on superblock
4813 * update time (freshest comes first), kick out 'old' disks and
4814 * compare superblocks. If everything's fine then run it.
4815 *
4816 * If "unit" is allocated, then bump its reference count
4817 */
4818 static void autorun_devices(int part)
4819 {
4820 mdk_rdev_t *rdev0, *rdev, *tmp;
4821 mddev_t *mddev;
4822 char b[BDEVNAME_SIZE];
4823
4824 printk(KERN_INFO "md: autorun ...\n");
4825 while (!list_empty(&pending_raid_disks)) {
4826 int unit;
4827 dev_t dev;
4828 LIST_HEAD(candidates);
4829 rdev0 = list_entry(pending_raid_disks.next,
4830 mdk_rdev_t, same_set);
4831
4832 printk(KERN_INFO "md: considering %s ...\n",
4833 bdevname(rdev0->bdev,b));
4834 INIT_LIST_HEAD(&candidates);
4835 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4836 if (super_90_load(rdev, rdev0, 0) >= 0) {
4837 printk(KERN_INFO "md: adding %s ...\n",
4838 bdevname(rdev->bdev,b));
4839 list_move(&rdev->same_set, &candidates);
4840 }
4841 /*
4842 * now we have a set of devices, with all of them having
4843 * mostly sane superblocks. It's time to allocate the
4844 * mddev.
4845 */
4846 if (part) {
4847 dev = MKDEV(mdp_major,
4848 rdev0->preferred_minor << MdpMinorShift);
4849 unit = MINOR(dev) >> MdpMinorShift;
4850 } else {
4851 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4852 unit = MINOR(dev);
4853 }
4854 if (rdev0->preferred_minor != unit) {
4855 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4856 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4857 break;
4858 }
4859
4860 md_probe(dev, NULL, NULL);
4861 mddev = mddev_find(dev);
4862 if (!mddev || !mddev->gendisk) {
4863 if (mddev)
4864 mddev_put(mddev);
4865 printk(KERN_ERR
4866 "md: cannot allocate memory for md drive.\n");
4867 break;
4868 }
4869 if (mddev_lock(mddev))
4870 printk(KERN_WARNING "md: %s locked, cannot run\n",
4871 mdname(mddev));
4872 else if (mddev->raid_disks || mddev->major_version
4873 || !list_empty(&mddev->disks)) {
4874 printk(KERN_WARNING
4875 "md: %s already running, cannot run %s\n",
4876 mdname(mddev), bdevname(rdev0->bdev,b));
4877 mddev_unlock(mddev);
4878 } else {
4879 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4880 mddev->persistent = 1;
4881 rdev_for_each_list(rdev, tmp, &candidates) {
4882 list_del_init(&rdev->same_set);
4883 if (bind_rdev_to_array(rdev, mddev))
4884 export_rdev(rdev);
4885 }
4886 autorun_array(mddev);
4887 mddev_unlock(mddev);
4888 }
4889 /* on success, candidates will be empty, on error
4890 * it won't...
4891 */
4892 rdev_for_each_list(rdev, tmp, &candidates) {
4893 list_del_init(&rdev->same_set);
4894 export_rdev(rdev);
4895 }
4896 mddev_put(mddev);
4897 }
4898 printk(KERN_INFO "md: ... autorun DONE.\n");
4899 }
4900 #endif /* !MODULE */
4901
4902 static int get_version(void __user * arg)
4903 {
4904 mdu_version_t ver;
4905
4906 ver.major = MD_MAJOR_VERSION;
4907 ver.minor = MD_MINOR_VERSION;
4908 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4909
4910 if (copy_to_user(arg, &ver, sizeof(ver)))
4911 return -EFAULT;
4912
4913 return 0;
4914 }
4915
4916 static int get_array_info(mddev_t * mddev, void __user * arg)
4917 {
4918 mdu_array_info_t info;
4919 int nr,working,insync,failed,spare;
4920 mdk_rdev_t *rdev;
4921
4922 nr=working=insync=failed=spare=0;
4923 list_for_each_entry(rdev, &mddev->disks, same_set) {
4924 nr++;
4925 if (test_bit(Faulty, &rdev->flags))
4926 failed++;
4927 else {
4928 working++;
4929 if (test_bit(In_sync, &rdev->flags))
4930 insync++;
4931 else
4932 spare++;
4933 }
4934 }
4935
4936 info.major_version = mddev->major_version;
4937 info.minor_version = mddev->minor_version;
4938 info.patch_version = MD_PATCHLEVEL_VERSION;
4939 info.ctime = mddev->ctime;
4940 info.level = mddev->level;
4941 info.size = mddev->dev_sectors / 2;
4942 if (info.size != mddev->dev_sectors / 2) /* overflow */
4943 info.size = -1;
4944 info.nr_disks = nr;
4945 info.raid_disks = mddev->raid_disks;
4946 info.md_minor = mddev->md_minor;
4947 info.not_persistent= !mddev->persistent;
4948
4949 info.utime = mddev->utime;
4950 info.state = 0;
4951 if (mddev->in_sync)
4952 info.state = (1<<MD_SB_CLEAN);
4953 if (mddev->bitmap && mddev->bitmap_info.offset)
4954 info.state = (1<<MD_SB_BITMAP_PRESENT);
4955 info.active_disks = insync;
4956 info.working_disks = working;
4957 info.failed_disks = failed;
4958 info.spare_disks = spare;
4959
4960 info.layout = mddev->layout;
4961 info.chunk_size = mddev->chunk_sectors << 9;
4962
4963 if (copy_to_user(arg, &info, sizeof(info)))
4964 return -EFAULT;
4965
4966 return 0;
4967 }
4968
4969 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4970 {
4971 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4972 char *ptr, *buf = NULL;
4973 int err = -ENOMEM;
4974
4975 if (md_allow_write(mddev))
4976 file = kmalloc(sizeof(*file), GFP_NOIO);
4977 else
4978 file = kmalloc(sizeof(*file), GFP_KERNEL);
4979
4980 if (!file)
4981 goto out;
4982
4983 /* bitmap disabled, zero the first byte and copy out */
4984 if (!mddev->bitmap || !mddev->bitmap->file) {
4985 file->pathname[0] = '\0';
4986 goto copy_out;
4987 }
4988
4989 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4990 if (!buf)
4991 goto out;
4992
4993 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4994 if (IS_ERR(ptr))
4995 goto out;
4996
4997 strcpy(file->pathname, ptr);
4998
4999 copy_out:
5000 err = 0;
5001 if (copy_to_user(arg, file, sizeof(*file)))
5002 err = -EFAULT;
5003 out:
5004 kfree(buf);
5005 kfree(file);
5006 return err;
5007 }
5008
5009 static int get_disk_info(mddev_t * mddev, void __user * arg)
5010 {
5011 mdu_disk_info_t info;
5012 mdk_rdev_t *rdev;
5013
5014 if (copy_from_user(&info, arg, sizeof(info)))
5015 return -EFAULT;
5016
5017 rdev = find_rdev_nr(mddev, info.number);
5018 if (rdev) {
5019 info.major = MAJOR(rdev->bdev->bd_dev);
5020 info.minor = MINOR(rdev->bdev->bd_dev);
5021 info.raid_disk = rdev->raid_disk;
5022 info.state = 0;
5023 if (test_bit(Faulty, &rdev->flags))
5024 info.state |= (1<<MD_DISK_FAULTY);
5025 else if (test_bit(In_sync, &rdev->flags)) {
5026 info.state |= (1<<MD_DISK_ACTIVE);
5027 info.state |= (1<<MD_DISK_SYNC);
5028 }
5029 if (test_bit(WriteMostly, &rdev->flags))
5030 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5031 } else {
5032 info.major = info.minor = 0;
5033 info.raid_disk = -1;
5034 info.state = (1<<MD_DISK_REMOVED);
5035 }
5036
5037 if (copy_to_user(arg, &info, sizeof(info)))
5038 return -EFAULT;
5039
5040 return 0;
5041 }
5042
5043 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
5044 {
5045 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5046 mdk_rdev_t *rdev;
5047 dev_t dev = MKDEV(info->major,info->minor);
5048
5049 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5050 return -EOVERFLOW;
5051
5052 if (!mddev->raid_disks) {
5053 int err;
5054 /* expecting a device which has a superblock */
5055 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5056 if (IS_ERR(rdev)) {
5057 printk(KERN_WARNING
5058 "md: md_import_device returned %ld\n",
5059 PTR_ERR(rdev));
5060 return PTR_ERR(rdev);
5061 }
5062 if (!list_empty(&mddev->disks)) {
5063 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
5064 mdk_rdev_t, same_set);
5065 err = super_types[mddev->major_version]
5066 .load_super(rdev, rdev0, mddev->minor_version);
5067 if (err < 0) {
5068 printk(KERN_WARNING
5069 "md: %s has different UUID to %s\n",
5070 bdevname(rdev->bdev,b),
5071 bdevname(rdev0->bdev,b2));
5072 export_rdev(rdev);
5073 return -EINVAL;
5074 }
5075 }
5076 err = bind_rdev_to_array(rdev, mddev);
5077 if (err)
5078 export_rdev(rdev);
5079 return err;
5080 }
5081
5082 /*
5083 * add_new_disk can be used once the array is assembled
5084 * to add "hot spares". They must already have a superblock
5085 * written
5086 */
5087 if (mddev->pers) {
5088 int err;
5089 if (!mddev->pers->hot_add_disk) {
5090 printk(KERN_WARNING
5091 "%s: personality does not support diskops!\n",
5092 mdname(mddev));
5093 return -EINVAL;
5094 }
5095 if (mddev->persistent)
5096 rdev = md_import_device(dev, mddev->major_version,
5097 mddev->minor_version);
5098 else
5099 rdev = md_import_device(dev, -1, -1);
5100 if (IS_ERR(rdev)) {
5101 printk(KERN_WARNING
5102 "md: md_import_device returned %ld\n",
5103 PTR_ERR(rdev));
5104 return PTR_ERR(rdev);
5105 }
5106 /* set save_raid_disk if appropriate */
5107 if (!mddev->persistent) {
5108 if (info->state & (1<<MD_DISK_SYNC) &&
5109 info->raid_disk < mddev->raid_disks)
5110 rdev->raid_disk = info->raid_disk;
5111 else
5112 rdev->raid_disk = -1;
5113 } else
5114 super_types[mddev->major_version].
5115 validate_super(mddev, rdev);
5116 rdev->saved_raid_disk = rdev->raid_disk;
5117
5118 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5119 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5120 set_bit(WriteMostly, &rdev->flags);
5121 else
5122 clear_bit(WriteMostly, &rdev->flags);
5123
5124 rdev->raid_disk = -1;
5125 err = bind_rdev_to_array(rdev, mddev);
5126 if (!err && !mddev->pers->hot_remove_disk) {
5127 /* If there is hot_add_disk but no hot_remove_disk
5128 * then added disks for geometry changes,
5129 * and should be added immediately.
5130 */
5131 super_types[mddev->major_version].
5132 validate_super(mddev, rdev);
5133 err = mddev->pers->hot_add_disk(mddev, rdev);
5134 if (err)
5135 unbind_rdev_from_array(rdev);
5136 }
5137 if (err)
5138 export_rdev(rdev);
5139 else
5140 sysfs_notify_dirent(rdev->sysfs_state);
5141
5142 md_update_sb(mddev, 1);
5143 if (mddev->degraded)
5144 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5145 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5146 md_wakeup_thread(mddev->thread);
5147 return err;
5148 }
5149
5150 /* otherwise, add_new_disk is only allowed
5151 * for major_version==0 superblocks
5152 */
5153 if (mddev->major_version != 0) {
5154 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5155 mdname(mddev));
5156 return -EINVAL;
5157 }
5158
5159 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5160 int err;
5161 rdev = md_import_device(dev, -1, 0);
5162 if (IS_ERR(rdev)) {
5163 printk(KERN_WARNING
5164 "md: error, md_import_device() returned %ld\n",
5165 PTR_ERR(rdev));
5166 return PTR_ERR(rdev);
5167 }
5168 rdev->desc_nr = info->number;
5169 if (info->raid_disk < mddev->raid_disks)
5170 rdev->raid_disk = info->raid_disk;
5171 else
5172 rdev->raid_disk = -1;
5173
5174 if (rdev->raid_disk < mddev->raid_disks)
5175 if (info->state & (1<<MD_DISK_SYNC))
5176 set_bit(In_sync, &rdev->flags);
5177
5178 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5179 set_bit(WriteMostly, &rdev->flags);
5180
5181 if (!mddev->persistent) {
5182 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5183 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5184 } else
5185 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5186 rdev->sectors = rdev->sb_start;
5187
5188 err = bind_rdev_to_array(rdev, mddev);
5189 if (err) {
5190 export_rdev(rdev);
5191 return err;
5192 }
5193 }
5194
5195 return 0;
5196 }
5197
5198 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
5199 {
5200 char b[BDEVNAME_SIZE];
5201 mdk_rdev_t *rdev;
5202
5203 rdev = find_rdev(mddev, dev);
5204 if (!rdev)
5205 return -ENXIO;
5206
5207 if (rdev->raid_disk >= 0)
5208 goto busy;
5209
5210 kick_rdev_from_array(rdev);
5211 md_update_sb(mddev, 1);
5212 md_new_event(mddev);
5213
5214 return 0;
5215 busy:
5216 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5217 bdevname(rdev->bdev,b), mdname(mddev));
5218 return -EBUSY;
5219 }
5220
5221 static int hot_add_disk(mddev_t * mddev, dev_t dev)
5222 {
5223 char b[BDEVNAME_SIZE];
5224 int err;
5225 mdk_rdev_t *rdev;
5226
5227 if (!mddev->pers)
5228 return -ENODEV;
5229
5230 if (mddev->major_version != 0) {
5231 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5232 " version-0 superblocks.\n",
5233 mdname(mddev));
5234 return -EINVAL;
5235 }
5236 if (!mddev->pers->hot_add_disk) {
5237 printk(KERN_WARNING
5238 "%s: personality does not support diskops!\n",
5239 mdname(mddev));
5240 return -EINVAL;
5241 }
5242
5243 rdev = md_import_device(dev, -1, 0);
5244 if (IS_ERR(rdev)) {
5245 printk(KERN_WARNING
5246 "md: error, md_import_device() returned %ld\n",
5247 PTR_ERR(rdev));
5248 return -EINVAL;
5249 }
5250
5251 if (mddev->persistent)
5252 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5253 else
5254 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5255
5256 rdev->sectors = rdev->sb_start;
5257
5258 if (test_bit(Faulty, &rdev->flags)) {
5259 printk(KERN_WARNING
5260 "md: can not hot-add faulty %s disk to %s!\n",
5261 bdevname(rdev->bdev,b), mdname(mddev));
5262 err = -EINVAL;
5263 goto abort_export;
5264 }
5265 clear_bit(In_sync, &rdev->flags);
5266 rdev->desc_nr = -1;
5267 rdev->saved_raid_disk = -1;
5268 err = bind_rdev_to_array(rdev, mddev);
5269 if (err)
5270 goto abort_export;
5271
5272 /*
5273 * The rest should better be atomic, we can have disk failures
5274 * noticed in interrupt contexts ...
5275 */
5276
5277 rdev->raid_disk = -1;
5278
5279 md_update_sb(mddev, 1);
5280
5281 /*
5282 * Kick recovery, maybe this spare has to be added to the
5283 * array immediately.
5284 */
5285 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5286 md_wakeup_thread(mddev->thread);
5287 md_new_event(mddev);
5288 return 0;
5289
5290 abort_export:
5291 export_rdev(rdev);
5292 return err;
5293 }
5294
5295 static int set_bitmap_file(mddev_t *mddev, int fd)
5296 {
5297 int err;
5298
5299 if (mddev->pers) {
5300 if (!mddev->pers->quiesce)
5301 return -EBUSY;
5302 if (mddev->recovery || mddev->sync_thread)
5303 return -EBUSY;
5304 /* we should be able to change the bitmap.. */
5305 }
5306
5307
5308 if (fd >= 0) {
5309 if (mddev->bitmap)
5310 return -EEXIST; /* cannot add when bitmap is present */
5311 mddev->bitmap_info.file = fget(fd);
5312
5313 if (mddev->bitmap_info.file == NULL) {
5314 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5315 mdname(mddev));
5316 return -EBADF;
5317 }
5318
5319 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5320 if (err) {
5321 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5322 mdname(mddev));
5323 fput(mddev->bitmap_info.file);
5324 mddev->bitmap_info.file = NULL;
5325 return err;
5326 }
5327 mddev->bitmap_info.offset = 0; /* file overrides offset */
5328 } else if (mddev->bitmap == NULL)
5329 return -ENOENT; /* cannot remove what isn't there */
5330 err = 0;
5331 if (mddev->pers) {
5332 mddev->pers->quiesce(mddev, 1);
5333 if (fd >= 0)
5334 err = bitmap_create(mddev);
5335 if (fd < 0 || err) {
5336 bitmap_destroy(mddev);
5337 fd = -1; /* make sure to put the file */
5338 }
5339 mddev->pers->quiesce(mddev, 0);
5340 }
5341 if (fd < 0) {
5342 if (mddev->bitmap_info.file) {
5343 restore_bitmap_write_access(mddev->bitmap_info.file);
5344 fput(mddev->bitmap_info.file);
5345 }
5346 mddev->bitmap_info.file = NULL;
5347 }
5348
5349 return err;
5350 }
5351
5352 /*
5353 * set_array_info is used two different ways
5354 * The original usage is when creating a new array.
5355 * In this usage, raid_disks is > 0 and it together with
5356 * level, size, not_persistent,layout,chunksize determine the
5357 * shape of the array.
5358 * This will always create an array with a type-0.90.0 superblock.
5359 * The newer usage is when assembling an array.
5360 * In this case raid_disks will be 0, and the major_version field is
5361 * use to determine which style super-blocks are to be found on the devices.
5362 * The minor and patch _version numbers are also kept incase the
5363 * super_block handler wishes to interpret them.
5364 */
5365 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5366 {
5367
5368 if (info->raid_disks == 0) {
5369 /* just setting version number for superblock loading */
5370 if (info->major_version < 0 ||
5371 info->major_version >= ARRAY_SIZE(super_types) ||
5372 super_types[info->major_version].name == NULL) {
5373 /* maybe try to auto-load a module? */
5374 printk(KERN_INFO
5375 "md: superblock version %d not known\n",
5376 info->major_version);
5377 return -EINVAL;
5378 }
5379 mddev->major_version = info->major_version;
5380 mddev->minor_version = info->minor_version;
5381 mddev->patch_version = info->patch_version;
5382 mddev->persistent = !info->not_persistent;
5383 /* ensure mddev_put doesn't delete this now that there
5384 * is some minimal configuration.
5385 */
5386 mddev->ctime = get_seconds();
5387 return 0;
5388 }
5389 mddev->major_version = MD_MAJOR_VERSION;
5390 mddev->minor_version = MD_MINOR_VERSION;
5391 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5392 mddev->ctime = get_seconds();
5393
5394 mddev->level = info->level;
5395 mddev->clevel[0] = 0;
5396 mddev->dev_sectors = 2 * (sector_t)info->size;
5397 mddev->raid_disks = info->raid_disks;
5398 /* don't set md_minor, it is determined by which /dev/md* was
5399 * openned
5400 */
5401 if (info->state & (1<<MD_SB_CLEAN))
5402 mddev->recovery_cp = MaxSector;
5403 else
5404 mddev->recovery_cp = 0;
5405 mddev->persistent = ! info->not_persistent;
5406 mddev->external = 0;
5407
5408 mddev->layout = info->layout;
5409 mddev->chunk_sectors = info->chunk_size >> 9;
5410
5411 mddev->max_disks = MD_SB_DISKS;
5412
5413 if (mddev->persistent)
5414 mddev->flags = 0;
5415 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5416
5417 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5418 mddev->bitmap_info.offset = 0;
5419
5420 mddev->reshape_position = MaxSector;
5421
5422 /*
5423 * Generate a 128 bit UUID
5424 */
5425 get_random_bytes(mddev->uuid, 16);
5426
5427 mddev->new_level = mddev->level;
5428 mddev->new_chunk_sectors = mddev->chunk_sectors;
5429 mddev->new_layout = mddev->layout;
5430 mddev->delta_disks = 0;
5431
5432 return 0;
5433 }
5434
5435 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5436 {
5437 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5438
5439 if (mddev->external_size)
5440 return;
5441
5442 mddev->array_sectors = array_sectors;
5443 }
5444 EXPORT_SYMBOL(md_set_array_sectors);
5445
5446 static int update_size(mddev_t *mddev, sector_t num_sectors)
5447 {
5448 mdk_rdev_t *rdev;
5449 int rv;
5450 int fit = (num_sectors == 0);
5451
5452 if (mddev->pers->resize == NULL)
5453 return -EINVAL;
5454 /* The "num_sectors" is the number of sectors of each device that
5455 * is used. This can only make sense for arrays with redundancy.
5456 * linear and raid0 always use whatever space is available. We can only
5457 * consider changing this number if no resync or reconstruction is
5458 * happening, and if the new size is acceptable. It must fit before the
5459 * sb_start or, if that is <data_offset, it must fit before the size
5460 * of each device. If num_sectors is zero, we find the largest size
5461 * that fits.
5462
5463 */
5464 if (mddev->sync_thread)
5465 return -EBUSY;
5466 if (mddev->bitmap)
5467 /* Sorry, cannot grow a bitmap yet, just remove it,
5468 * grow, and re-add.
5469 */
5470 return -EBUSY;
5471 list_for_each_entry(rdev, &mddev->disks, same_set) {
5472 sector_t avail = rdev->sectors;
5473
5474 if (fit && (num_sectors == 0 || num_sectors > avail))
5475 num_sectors = avail;
5476 if (avail < num_sectors)
5477 return -ENOSPC;
5478 }
5479 rv = mddev->pers->resize(mddev, num_sectors);
5480 if (!rv)
5481 revalidate_disk(mddev->gendisk);
5482 return rv;
5483 }
5484
5485 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5486 {
5487 int rv;
5488 /* change the number of raid disks */
5489 if (mddev->pers->check_reshape == NULL)
5490 return -EINVAL;
5491 if (raid_disks <= 0 ||
5492 (mddev->max_disks && raid_disks >= mddev->max_disks))
5493 return -EINVAL;
5494 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5495 return -EBUSY;
5496 mddev->delta_disks = raid_disks - mddev->raid_disks;
5497
5498 rv = mddev->pers->check_reshape(mddev);
5499 return rv;
5500 }
5501
5502
5503 /*
5504 * update_array_info is used to change the configuration of an
5505 * on-line array.
5506 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5507 * fields in the info are checked against the array.
5508 * Any differences that cannot be handled will cause an error.
5509 * Normally, only one change can be managed at a time.
5510 */
5511 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5512 {
5513 int rv = 0;
5514 int cnt = 0;
5515 int state = 0;
5516
5517 /* calculate expected state,ignoring low bits */
5518 if (mddev->bitmap && mddev->bitmap_info.offset)
5519 state |= (1 << MD_SB_BITMAP_PRESENT);
5520
5521 if (mddev->major_version != info->major_version ||
5522 mddev->minor_version != info->minor_version ||
5523 /* mddev->patch_version != info->patch_version || */
5524 mddev->ctime != info->ctime ||
5525 mddev->level != info->level ||
5526 /* mddev->layout != info->layout || */
5527 !mddev->persistent != info->not_persistent||
5528 mddev->chunk_sectors != info->chunk_size >> 9 ||
5529 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5530 ((state^info->state) & 0xfffffe00)
5531 )
5532 return -EINVAL;
5533 /* Check there is only one change */
5534 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5535 cnt++;
5536 if (mddev->raid_disks != info->raid_disks)
5537 cnt++;
5538 if (mddev->layout != info->layout)
5539 cnt++;
5540 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5541 cnt++;
5542 if (cnt == 0)
5543 return 0;
5544 if (cnt > 1)
5545 return -EINVAL;
5546
5547 if (mddev->layout != info->layout) {
5548 /* Change layout
5549 * we don't need to do anything at the md level, the
5550 * personality will take care of it all.
5551 */
5552 if (mddev->pers->check_reshape == NULL)
5553 return -EINVAL;
5554 else {
5555 mddev->new_layout = info->layout;
5556 rv = mddev->pers->check_reshape(mddev);
5557 if (rv)
5558 mddev->new_layout = mddev->layout;
5559 return rv;
5560 }
5561 }
5562 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5563 rv = update_size(mddev, (sector_t)info->size * 2);
5564
5565 if (mddev->raid_disks != info->raid_disks)
5566 rv = update_raid_disks(mddev, info->raid_disks);
5567
5568 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5569 if (mddev->pers->quiesce == NULL)
5570 return -EINVAL;
5571 if (mddev->recovery || mddev->sync_thread)
5572 return -EBUSY;
5573 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5574 /* add the bitmap */
5575 if (mddev->bitmap)
5576 return -EEXIST;
5577 if (mddev->bitmap_info.default_offset == 0)
5578 return -EINVAL;
5579 mddev->bitmap_info.offset =
5580 mddev->bitmap_info.default_offset;
5581 mddev->pers->quiesce(mddev, 1);
5582 rv = bitmap_create(mddev);
5583 if (rv)
5584 bitmap_destroy(mddev);
5585 mddev->pers->quiesce(mddev, 0);
5586 } else {
5587 /* remove the bitmap */
5588 if (!mddev->bitmap)
5589 return -ENOENT;
5590 if (mddev->bitmap->file)
5591 return -EINVAL;
5592 mddev->pers->quiesce(mddev, 1);
5593 bitmap_destroy(mddev);
5594 mddev->pers->quiesce(mddev, 0);
5595 mddev->bitmap_info.offset = 0;
5596 }
5597 }
5598 md_update_sb(mddev, 1);
5599 return rv;
5600 }
5601
5602 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5603 {
5604 mdk_rdev_t *rdev;
5605
5606 if (mddev->pers == NULL)
5607 return -ENODEV;
5608
5609 rdev = find_rdev(mddev, dev);
5610 if (!rdev)
5611 return -ENODEV;
5612
5613 md_error(mddev, rdev);
5614 return 0;
5615 }
5616
5617 /*
5618 * We have a problem here : there is no easy way to give a CHS
5619 * virtual geometry. We currently pretend that we have a 2 heads
5620 * 4 sectors (with a BIG number of cylinders...). This drives
5621 * dosfs just mad... ;-)
5622 */
5623 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5624 {
5625 mddev_t *mddev = bdev->bd_disk->private_data;
5626
5627 geo->heads = 2;
5628 geo->sectors = 4;
5629 geo->cylinders = mddev->array_sectors / 8;
5630 return 0;
5631 }
5632
5633 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5634 unsigned int cmd, unsigned long arg)
5635 {
5636 int err = 0;
5637 void __user *argp = (void __user *)arg;
5638 mddev_t *mddev = NULL;
5639 int ro;
5640
5641 if (!capable(CAP_SYS_ADMIN))
5642 return -EACCES;
5643
5644 /*
5645 * Commands dealing with the RAID driver but not any
5646 * particular array:
5647 */
5648 switch (cmd)
5649 {
5650 case RAID_VERSION:
5651 err = get_version(argp);
5652 goto done;
5653
5654 case PRINT_RAID_DEBUG:
5655 err = 0;
5656 md_print_devices();
5657 goto done;
5658
5659 #ifndef MODULE
5660 case RAID_AUTORUN:
5661 err = 0;
5662 autostart_arrays(arg);
5663 goto done;
5664 #endif
5665 default:;
5666 }
5667
5668 /*
5669 * Commands creating/starting a new array:
5670 */
5671
5672 mddev = bdev->bd_disk->private_data;
5673
5674 if (!mddev) {
5675 BUG();
5676 goto abort;
5677 }
5678
5679 err = mddev_lock(mddev);
5680 if (err) {
5681 printk(KERN_INFO
5682 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5683 err, cmd);
5684 goto abort;
5685 }
5686
5687 switch (cmd)
5688 {
5689 case SET_ARRAY_INFO:
5690 {
5691 mdu_array_info_t info;
5692 if (!arg)
5693 memset(&info, 0, sizeof(info));
5694 else if (copy_from_user(&info, argp, sizeof(info))) {
5695 err = -EFAULT;
5696 goto abort_unlock;
5697 }
5698 if (mddev->pers) {
5699 err = update_array_info(mddev, &info);
5700 if (err) {
5701 printk(KERN_WARNING "md: couldn't update"
5702 " array info. %d\n", err);
5703 goto abort_unlock;
5704 }
5705 goto done_unlock;
5706 }
5707 if (!list_empty(&mddev->disks)) {
5708 printk(KERN_WARNING
5709 "md: array %s already has disks!\n",
5710 mdname(mddev));
5711 err = -EBUSY;
5712 goto abort_unlock;
5713 }
5714 if (mddev->raid_disks) {
5715 printk(KERN_WARNING
5716 "md: array %s already initialised!\n",
5717 mdname(mddev));
5718 err = -EBUSY;
5719 goto abort_unlock;
5720 }
5721 err = set_array_info(mddev, &info);
5722 if (err) {
5723 printk(KERN_WARNING "md: couldn't set"
5724 " array info. %d\n", err);
5725 goto abort_unlock;
5726 }
5727 }
5728 goto done_unlock;
5729
5730 default:;
5731 }
5732
5733 /*
5734 * Commands querying/configuring an existing array:
5735 */
5736 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5737 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5738 if ((!mddev->raid_disks && !mddev->external)
5739 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5740 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5741 && cmd != GET_BITMAP_FILE) {
5742 err = -ENODEV;
5743 goto abort_unlock;
5744 }
5745
5746 /*
5747 * Commands even a read-only array can execute:
5748 */
5749 switch (cmd)
5750 {
5751 case GET_ARRAY_INFO:
5752 err = get_array_info(mddev, argp);
5753 goto done_unlock;
5754
5755 case GET_BITMAP_FILE:
5756 err = get_bitmap_file(mddev, argp);
5757 goto done_unlock;
5758
5759 case GET_DISK_INFO:
5760 err = get_disk_info(mddev, argp);
5761 goto done_unlock;
5762
5763 case RESTART_ARRAY_RW:
5764 err = restart_array(mddev);
5765 goto done_unlock;
5766
5767 case STOP_ARRAY:
5768 err = do_md_stop(mddev, 0, 1);
5769 goto done_unlock;
5770
5771 case STOP_ARRAY_RO:
5772 err = md_set_readonly(mddev, 1);
5773 goto done_unlock;
5774
5775 case BLKROSET:
5776 if (get_user(ro, (int __user *)(arg))) {
5777 err = -EFAULT;
5778 goto done_unlock;
5779 }
5780 err = -EINVAL;
5781
5782 /* if the bdev is going readonly the value of mddev->ro
5783 * does not matter, no writes are coming
5784 */
5785 if (ro)
5786 goto done_unlock;
5787
5788 /* are we are already prepared for writes? */
5789 if (mddev->ro != 1)
5790 goto done_unlock;
5791
5792 /* transitioning to readauto need only happen for
5793 * arrays that call md_write_start
5794 */
5795 if (mddev->pers) {
5796 err = restart_array(mddev);
5797 if (err == 0) {
5798 mddev->ro = 2;
5799 set_disk_ro(mddev->gendisk, 0);
5800 }
5801 }
5802 goto done_unlock;
5803 }
5804
5805 /*
5806 * The remaining ioctls are changing the state of the
5807 * superblock, so we do not allow them on read-only arrays.
5808 * However non-MD ioctls (e.g. get-size) will still come through
5809 * here and hit the 'default' below, so only disallow
5810 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5811 */
5812 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5813 if (mddev->ro == 2) {
5814 mddev->ro = 0;
5815 sysfs_notify_dirent(mddev->sysfs_state);
5816 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5817 md_wakeup_thread(mddev->thread);
5818 } else {
5819 err = -EROFS;
5820 goto abort_unlock;
5821 }
5822 }
5823
5824 switch (cmd)
5825 {
5826 case ADD_NEW_DISK:
5827 {
5828 mdu_disk_info_t info;
5829 if (copy_from_user(&info, argp, sizeof(info)))
5830 err = -EFAULT;
5831 else
5832 err = add_new_disk(mddev, &info);
5833 goto done_unlock;
5834 }
5835
5836 case HOT_REMOVE_DISK:
5837 err = hot_remove_disk(mddev, new_decode_dev(arg));
5838 goto done_unlock;
5839
5840 case HOT_ADD_DISK:
5841 err = hot_add_disk(mddev, new_decode_dev(arg));
5842 goto done_unlock;
5843
5844 case SET_DISK_FAULTY:
5845 err = set_disk_faulty(mddev, new_decode_dev(arg));
5846 goto done_unlock;
5847
5848 case RUN_ARRAY:
5849 err = do_md_run(mddev);
5850 goto done_unlock;
5851
5852 case SET_BITMAP_FILE:
5853 err = set_bitmap_file(mddev, (int)arg);
5854 goto done_unlock;
5855
5856 default:
5857 err = -EINVAL;
5858 goto abort_unlock;
5859 }
5860
5861 done_unlock:
5862 abort_unlock:
5863 if (mddev->hold_active == UNTIL_IOCTL &&
5864 err != -EINVAL)
5865 mddev->hold_active = 0;
5866 mddev_unlock(mddev);
5867
5868 return err;
5869 done:
5870 if (err)
5871 MD_BUG();
5872 abort:
5873 return err;
5874 }
5875 #ifdef CONFIG_COMPAT
5876 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
5877 unsigned int cmd, unsigned long arg)
5878 {
5879 switch (cmd) {
5880 case HOT_REMOVE_DISK:
5881 case HOT_ADD_DISK:
5882 case SET_DISK_FAULTY:
5883 case SET_BITMAP_FILE:
5884 /* These take in integer arg, do not convert */
5885 break;
5886 default:
5887 arg = (unsigned long)compat_ptr(arg);
5888 break;
5889 }
5890
5891 return md_ioctl(bdev, mode, cmd, arg);
5892 }
5893 #endif /* CONFIG_COMPAT */
5894
5895 static int md_open(struct block_device *bdev, fmode_t mode)
5896 {
5897 /*
5898 * Succeed if we can lock the mddev, which confirms that
5899 * it isn't being stopped right now.
5900 */
5901 mddev_t *mddev = mddev_find(bdev->bd_dev);
5902 int err;
5903
5904 if (mddev->gendisk != bdev->bd_disk) {
5905 /* we are racing with mddev_put which is discarding this
5906 * bd_disk.
5907 */
5908 mddev_put(mddev);
5909 /* Wait until bdev->bd_disk is definitely gone */
5910 flush_scheduled_work();
5911 /* Then retry the open from the top */
5912 return -ERESTARTSYS;
5913 }
5914 BUG_ON(mddev != bdev->bd_disk->private_data);
5915
5916 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
5917 goto out;
5918
5919 err = 0;
5920 atomic_inc(&mddev->openers);
5921 mutex_unlock(&mddev->open_mutex);
5922
5923 check_disk_size_change(mddev->gendisk, bdev);
5924 out:
5925 return err;
5926 }
5927
5928 static int md_release(struct gendisk *disk, fmode_t mode)
5929 {
5930 mddev_t *mddev = disk->private_data;
5931
5932 BUG_ON(!mddev);
5933 atomic_dec(&mddev->openers);
5934 mddev_put(mddev);
5935
5936 return 0;
5937 }
5938 static const struct block_device_operations md_fops =
5939 {
5940 .owner = THIS_MODULE,
5941 .open = md_open,
5942 .release = md_release,
5943 .ioctl = md_ioctl,
5944 #ifdef CONFIG_COMPAT
5945 .compat_ioctl = md_compat_ioctl,
5946 #endif
5947 .getgeo = md_getgeo,
5948 };
5949
5950 static int md_thread(void * arg)
5951 {
5952 mdk_thread_t *thread = arg;
5953
5954 /*
5955 * md_thread is a 'system-thread', it's priority should be very
5956 * high. We avoid resource deadlocks individually in each
5957 * raid personality. (RAID5 does preallocation) We also use RR and
5958 * the very same RT priority as kswapd, thus we will never get
5959 * into a priority inversion deadlock.
5960 *
5961 * we definitely have to have equal or higher priority than
5962 * bdflush, otherwise bdflush will deadlock if there are too
5963 * many dirty RAID5 blocks.
5964 */
5965
5966 allow_signal(SIGKILL);
5967 while (!kthread_should_stop()) {
5968
5969 /* We need to wait INTERRUPTIBLE so that
5970 * we don't add to the load-average.
5971 * That means we need to be sure no signals are
5972 * pending
5973 */
5974 if (signal_pending(current))
5975 flush_signals(current);
5976
5977 wait_event_interruptible_timeout
5978 (thread->wqueue,
5979 test_bit(THREAD_WAKEUP, &thread->flags)
5980 || kthread_should_stop(),
5981 thread->timeout);
5982
5983 clear_bit(THREAD_WAKEUP, &thread->flags);
5984
5985 thread->run(thread->mddev);
5986 }
5987
5988 return 0;
5989 }
5990
5991 void md_wakeup_thread(mdk_thread_t *thread)
5992 {
5993 if (thread) {
5994 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5995 set_bit(THREAD_WAKEUP, &thread->flags);
5996 wake_up(&thread->wqueue);
5997 }
5998 }
5999
6000 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
6001 const char *name)
6002 {
6003 mdk_thread_t *thread;
6004
6005 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
6006 if (!thread)
6007 return NULL;
6008
6009 init_waitqueue_head(&thread->wqueue);
6010
6011 thread->run = run;
6012 thread->mddev = mddev;
6013 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6014 thread->tsk = kthread_run(md_thread, thread,
6015 "%s_%s",
6016 mdname(thread->mddev),
6017 name ?: mddev->pers->name);
6018 if (IS_ERR(thread->tsk)) {
6019 kfree(thread);
6020 return NULL;
6021 }
6022 return thread;
6023 }
6024
6025 void md_unregister_thread(mdk_thread_t *thread)
6026 {
6027 if (!thread)
6028 return;
6029 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6030
6031 kthread_stop(thread->tsk);
6032 kfree(thread);
6033 }
6034
6035 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
6036 {
6037 if (!mddev) {
6038 MD_BUG();
6039 return;
6040 }
6041
6042 if (!rdev || test_bit(Faulty, &rdev->flags))
6043 return;
6044
6045 if (mddev->external)
6046 set_bit(Blocked, &rdev->flags);
6047 /*
6048 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
6049 mdname(mddev),
6050 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
6051 __builtin_return_address(0),__builtin_return_address(1),
6052 __builtin_return_address(2),__builtin_return_address(3));
6053 */
6054 if (!mddev->pers)
6055 return;
6056 if (!mddev->pers->error_handler)
6057 return;
6058 mddev->pers->error_handler(mddev,rdev);
6059 if (mddev->degraded)
6060 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6061 sysfs_notify_dirent(rdev->sysfs_state);
6062 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6063 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6064 md_wakeup_thread(mddev->thread);
6065 md_new_event_inintr(mddev);
6066 }
6067
6068 /* seq_file implementation /proc/mdstat */
6069
6070 static void status_unused(struct seq_file *seq)
6071 {
6072 int i = 0;
6073 mdk_rdev_t *rdev;
6074
6075 seq_printf(seq, "unused devices: ");
6076
6077 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6078 char b[BDEVNAME_SIZE];
6079 i++;
6080 seq_printf(seq, "%s ",
6081 bdevname(rdev->bdev,b));
6082 }
6083 if (!i)
6084 seq_printf(seq, "<none>");
6085
6086 seq_printf(seq, "\n");
6087 }
6088
6089
6090 static void status_resync(struct seq_file *seq, mddev_t * mddev)
6091 {
6092 sector_t max_sectors, resync, res;
6093 unsigned long dt, db;
6094 sector_t rt;
6095 int scale;
6096 unsigned int per_milli;
6097
6098 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
6099
6100 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6101 max_sectors = mddev->resync_max_sectors;
6102 else
6103 max_sectors = mddev->dev_sectors;
6104
6105 /*
6106 * Should not happen.
6107 */
6108 if (!max_sectors) {
6109 MD_BUG();
6110 return;
6111 }
6112 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6113 * in a sector_t, and (max_sectors>>scale) will fit in a
6114 * u32, as those are the requirements for sector_div.
6115 * Thus 'scale' must be at least 10
6116 */
6117 scale = 10;
6118 if (sizeof(sector_t) > sizeof(unsigned long)) {
6119 while ( max_sectors/2 > (1ULL<<(scale+32)))
6120 scale++;
6121 }
6122 res = (resync>>scale)*1000;
6123 sector_div(res, (u32)((max_sectors>>scale)+1));
6124
6125 per_milli = res;
6126 {
6127 int i, x = per_milli/50, y = 20-x;
6128 seq_printf(seq, "[");
6129 for (i = 0; i < x; i++)
6130 seq_printf(seq, "=");
6131 seq_printf(seq, ">");
6132 for (i = 0; i < y; i++)
6133 seq_printf(seq, ".");
6134 seq_printf(seq, "] ");
6135 }
6136 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6137 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6138 "reshape" :
6139 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6140 "check" :
6141 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6142 "resync" : "recovery"))),
6143 per_milli/10, per_milli % 10,
6144 (unsigned long long) resync/2,
6145 (unsigned long long) max_sectors/2);
6146
6147 /*
6148 * dt: time from mark until now
6149 * db: blocks written from mark until now
6150 * rt: remaining time
6151 *
6152 * rt is a sector_t, so could be 32bit or 64bit.
6153 * So we divide before multiply in case it is 32bit and close
6154 * to the limit.
6155 * We scale the divisor (db) by 32 to avoid loosing precision
6156 * near the end of resync when the number of remaining sectors
6157 * is close to 'db'.
6158 * We then divide rt by 32 after multiplying by db to compensate.
6159 * The '+1' avoids division by zero if db is very small.
6160 */
6161 dt = ((jiffies - mddev->resync_mark) / HZ);
6162 if (!dt) dt++;
6163 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6164 - mddev->resync_mark_cnt;
6165
6166 rt = max_sectors - resync; /* number of remaining sectors */
6167 sector_div(rt, db/32+1);
6168 rt *= dt;
6169 rt >>= 5;
6170
6171 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6172 ((unsigned long)rt % 60)/6);
6173
6174 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6175 }
6176
6177 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6178 {
6179 struct list_head *tmp;
6180 loff_t l = *pos;
6181 mddev_t *mddev;
6182
6183 if (l >= 0x10000)
6184 return NULL;
6185 if (!l--)
6186 /* header */
6187 return (void*)1;
6188
6189 spin_lock(&all_mddevs_lock);
6190 list_for_each(tmp,&all_mddevs)
6191 if (!l--) {
6192 mddev = list_entry(tmp, mddev_t, all_mddevs);
6193 mddev_get(mddev);
6194 spin_unlock(&all_mddevs_lock);
6195 return mddev;
6196 }
6197 spin_unlock(&all_mddevs_lock);
6198 if (!l--)
6199 return (void*)2;/* tail */
6200 return NULL;
6201 }
6202
6203 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6204 {
6205 struct list_head *tmp;
6206 mddev_t *next_mddev, *mddev = v;
6207
6208 ++*pos;
6209 if (v == (void*)2)
6210 return NULL;
6211
6212 spin_lock(&all_mddevs_lock);
6213 if (v == (void*)1)
6214 tmp = all_mddevs.next;
6215 else
6216 tmp = mddev->all_mddevs.next;
6217 if (tmp != &all_mddevs)
6218 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
6219 else {
6220 next_mddev = (void*)2;
6221 *pos = 0x10000;
6222 }
6223 spin_unlock(&all_mddevs_lock);
6224
6225 if (v != (void*)1)
6226 mddev_put(mddev);
6227 return next_mddev;
6228
6229 }
6230
6231 static void md_seq_stop(struct seq_file *seq, void *v)
6232 {
6233 mddev_t *mddev = v;
6234
6235 if (mddev && v != (void*)1 && v != (void*)2)
6236 mddev_put(mddev);
6237 }
6238
6239 struct mdstat_info {
6240 int event;
6241 };
6242
6243 static int md_seq_show(struct seq_file *seq, void *v)
6244 {
6245 mddev_t *mddev = v;
6246 sector_t sectors;
6247 mdk_rdev_t *rdev;
6248 struct mdstat_info *mi = seq->private;
6249 struct bitmap *bitmap;
6250
6251 if (v == (void*)1) {
6252 struct mdk_personality *pers;
6253 seq_printf(seq, "Personalities : ");
6254 spin_lock(&pers_lock);
6255 list_for_each_entry(pers, &pers_list, list)
6256 seq_printf(seq, "[%s] ", pers->name);
6257
6258 spin_unlock(&pers_lock);
6259 seq_printf(seq, "\n");
6260 mi->event = atomic_read(&md_event_count);
6261 return 0;
6262 }
6263 if (v == (void*)2) {
6264 status_unused(seq);
6265 return 0;
6266 }
6267
6268 if (mddev_lock(mddev) < 0)
6269 return -EINTR;
6270
6271 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6272 seq_printf(seq, "%s : %sactive", mdname(mddev),
6273 mddev->pers ? "" : "in");
6274 if (mddev->pers) {
6275 if (mddev->ro==1)
6276 seq_printf(seq, " (read-only)");
6277 if (mddev->ro==2)
6278 seq_printf(seq, " (auto-read-only)");
6279 seq_printf(seq, " %s", mddev->pers->name);
6280 }
6281
6282 sectors = 0;
6283 list_for_each_entry(rdev, &mddev->disks, same_set) {
6284 char b[BDEVNAME_SIZE];
6285 seq_printf(seq, " %s[%d]",
6286 bdevname(rdev->bdev,b), rdev->desc_nr);
6287 if (test_bit(WriteMostly, &rdev->flags))
6288 seq_printf(seq, "(W)");
6289 if (test_bit(Faulty, &rdev->flags)) {
6290 seq_printf(seq, "(F)");
6291 continue;
6292 } else if (rdev->raid_disk < 0)
6293 seq_printf(seq, "(S)"); /* spare */
6294 sectors += rdev->sectors;
6295 }
6296
6297 if (!list_empty(&mddev->disks)) {
6298 if (mddev->pers)
6299 seq_printf(seq, "\n %llu blocks",
6300 (unsigned long long)
6301 mddev->array_sectors / 2);
6302 else
6303 seq_printf(seq, "\n %llu blocks",
6304 (unsigned long long)sectors / 2);
6305 }
6306 if (mddev->persistent) {
6307 if (mddev->major_version != 0 ||
6308 mddev->minor_version != 90) {
6309 seq_printf(seq," super %d.%d",
6310 mddev->major_version,
6311 mddev->minor_version);
6312 }
6313 } else if (mddev->external)
6314 seq_printf(seq, " super external:%s",
6315 mddev->metadata_type);
6316 else
6317 seq_printf(seq, " super non-persistent");
6318
6319 if (mddev->pers) {
6320 mddev->pers->status(seq, mddev);
6321 seq_printf(seq, "\n ");
6322 if (mddev->pers->sync_request) {
6323 if (mddev->curr_resync > 2) {
6324 status_resync(seq, mddev);
6325 seq_printf(seq, "\n ");
6326 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6327 seq_printf(seq, "\tresync=DELAYED\n ");
6328 else if (mddev->recovery_cp < MaxSector)
6329 seq_printf(seq, "\tresync=PENDING\n ");
6330 }
6331 } else
6332 seq_printf(seq, "\n ");
6333
6334 if ((bitmap = mddev->bitmap)) {
6335 unsigned long chunk_kb;
6336 unsigned long flags;
6337 spin_lock_irqsave(&bitmap->lock, flags);
6338 chunk_kb = mddev->bitmap_info.chunksize >> 10;
6339 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6340 "%lu%s chunk",
6341 bitmap->pages - bitmap->missing_pages,
6342 bitmap->pages,
6343 (bitmap->pages - bitmap->missing_pages)
6344 << (PAGE_SHIFT - 10),
6345 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
6346 chunk_kb ? "KB" : "B");
6347 if (bitmap->file) {
6348 seq_printf(seq, ", file: ");
6349 seq_path(seq, &bitmap->file->f_path, " \t\n");
6350 }
6351
6352 seq_printf(seq, "\n");
6353 spin_unlock_irqrestore(&bitmap->lock, flags);
6354 }
6355
6356 seq_printf(seq, "\n");
6357 }
6358 mddev_unlock(mddev);
6359
6360 return 0;
6361 }
6362
6363 static const struct seq_operations md_seq_ops = {
6364 .start = md_seq_start,
6365 .next = md_seq_next,
6366 .stop = md_seq_stop,
6367 .show = md_seq_show,
6368 };
6369
6370 static int md_seq_open(struct inode *inode, struct file *file)
6371 {
6372 int error;
6373 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6374 if (mi == NULL)
6375 return -ENOMEM;
6376
6377 error = seq_open(file, &md_seq_ops);
6378 if (error)
6379 kfree(mi);
6380 else {
6381 struct seq_file *p = file->private_data;
6382 p->private = mi;
6383 mi->event = atomic_read(&md_event_count);
6384 }
6385 return error;
6386 }
6387
6388 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6389 {
6390 struct seq_file *m = filp->private_data;
6391 struct mdstat_info *mi = m->private;
6392 int mask;
6393
6394 poll_wait(filp, &md_event_waiters, wait);
6395
6396 /* always allow read */
6397 mask = POLLIN | POLLRDNORM;
6398
6399 if (mi->event != atomic_read(&md_event_count))
6400 mask |= POLLERR | POLLPRI;
6401 return mask;
6402 }
6403
6404 static const struct file_operations md_seq_fops = {
6405 .owner = THIS_MODULE,
6406 .open = md_seq_open,
6407 .read = seq_read,
6408 .llseek = seq_lseek,
6409 .release = seq_release_private,
6410 .poll = mdstat_poll,
6411 };
6412
6413 int register_md_personality(struct mdk_personality *p)
6414 {
6415 spin_lock(&pers_lock);
6416 list_add_tail(&p->list, &pers_list);
6417 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6418 spin_unlock(&pers_lock);
6419 return 0;
6420 }
6421
6422 int unregister_md_personality(struct mdk_personality *p)
6423 {
6424 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6425 spin_lock(&pers_lock);
6426 list_del_init(&p->list);
6427 spin_unlock(&pers_lock);
6428 return 0;
6429 }
6430
6431 static int is_mddev_idle(mddev_t *mddev, int init)
6432 {
6433 mdk_rdev_t * rdev;
6434 int idle;
6435 int curr_events;
6436
6437 idle = 1;
6438 rcu_read_lock();
6439 rdev_for_each_rcu(rdev, mddev) {
6440 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6441 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6442 (int)part_stat_read(&disk->part0, sectors[1]) -
6443 atomic_read(&disk->sync_io);
6444 /* sync IO will cause sync_io to increase before the disk_stats
6445 * as sync_io is counted when a request starts, and
6446 * disk_stats is counted when it completes.
6447 * So resync activity will cause curr_events to be smaller than
6448 * when there was no such activity.
6449 * non-sync IO will cause disk_stat to increase without
6450 * increasing sync_io so curr_events will (eventually)
6451 * be larger than it was before. Once it becomes
6452 * substantially larger, the test below will cause
6453 * the array to appear non-idle, and resync will slow
6454 * down.
6455 * If there is a lot of outstanding resync activity when
6456 * we set last_event to curr_events, then all that activity
6457 * completing might cause the array to appear non-idle
6458 * and resync will be slowed down even though there might
6459 * not have been non-resync activity. This will only
6460 * happen once though. 'last_events' will soon reflect
6461 * the state where there is little or no outstanding
6462 * resync requests, and further resync activity will
6463 * always make curr_events less than last_events.
6464 *
6465 */
6466 if (init || curr_events - rdev->last_events > 64) {
6467 rdev->last_events = curr_events;
6468 idle = 0;
6469 }
6470 }
6471 rcu_read_unlock();
6472 return idle;
6473 }
6474
6475 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6476 {
6477 /* another "blocks" (512byte) blocks have been synced */
6478 atomic_sub(blocks, &mddev->recovery_active);
6479 wake_up(&mddev->recovery_wait);
6480 if (!ok) {
6481 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6482 md_wakeup_thread(mddev->thread);
6483 // stop recovery, signal do_sync ....
6484 }
6485 }
6486
6487
6488 /* md_write_start(mddev, bi)
6489 * If we need to update some array metadata (e.g. 'active' flag
6490 * in superblock) before writing, schedule a superblock update
6491 * and wait for it to complete.
6492 */
6493 void md_write_start(mddev_t *mddev, struct bio *bi)
6494 {
6495 int did_change = 0;
6496 if (bio_data_dir(bi) != WRITE)
6497 return;
6498
6499 BUG_ON(mddev->ro == 1);
6500 if (mddev->ro == 2) {
6501 /* need to switch to read/write */
6502 mddev->ro = 0;
6503 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6504 md_wakeup_thread(mddev->thread);
6505 md_wakeup_thread(mddev->sync_thread);
6506 did_change = 1;
6507 }
6508 atomic_inc(&mddev->writes_pending);
6509 if (mddev->safemode == 1)
6510 mddev->safemode = 0;
6511 if (mddev->in_sync) {
6512 spin_lock_irq(&mddev->write_lock);
6513 if (mddev->in_sync) {
6514 mddev->in_sync = 0;
6515 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6516 md_wakeup_thread(mddev->thread);
6517 did_change = 1;
6518 }
6519 spin_unlock_irq(&mddev->write_lock);
6520 }
6521 if (did_change)
6522 sysfs_notify_dirent(mddev->sysfs_state);
6523 wait_event(mddev->sb_wait,
6524 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
6525 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6526 }
6527
6528 void md_write_end(mddev_t *mddev)
6529 {
6530 if (atomic_dec_and_test(&mddev->writes_pending)) {
6531 if (mddev->safemode == 2)
6532 md_wakeup_thread(mddev->thread);
6533 else if (mddev->safemode_delay)
6534 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6535 }
6536 }
6537
6538 /* md_allow_write(mddev)
6539 * Calling this ensures that the array is marked 'active' so that writes
6540 * may proceed without blocking. It is important to call this before
6541 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6542 * Must be called with mddev_lock held.
6543 *
6544 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6545 * is dropped, so return -EAGAIN after notifying userspace.
6546 */
6547 int md_allow_write(mddev_t *mddev)
6548 {
6549 if (!mddev->pers)
6550 return 0;
6551 if (mddev->ro)
6552 return 0;
6553 if (!mddev->pers->sync_request)
6554 return 0;
6555
6556 spin_lock_irq(&mddev->write_lock);
6557 if (mddev->in_sync) {
6558 mddev->in_sync = 0;
6559 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6560 if (mddev->safemode_delay &&
6561 mddev->safemode == 0)
6562 mddev->safemode = 1;
6563 spin_unlock_irq(&mddev->write_lock);
6564 md_update_sb(mddev, 0);
6565 sysfs_notify_dirent(mddev->sysfs_state);
6566 } else
6567 spin_unlock_irq(&mddev->write_lock);
6568
6569 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
6570 return -EAGAIN;
6571 else
6572 return 0;
6573 }
6574 EXPORT_SYMBOL_GPL(md_allow_write);
6575
6576 #define SYNC_MARKS 10
6577 #define SYNC_MARK_STEP (3*HZ)
6578 void md_do_sync(mddev_t *mddev)
6579 {
6580 mddev_t *mddev2;
6581 unsigned int currspeed = 0,
6582 window;
6583 sector_t max_sectors,j, io_sectors;
6584 unsigned long mark[SYNC_MARKS];
6585 sector_t mark_cnt[SYNC_MARKS];
6586 int last_mark,m;
6587 struct list_head *tmp;
6588 sector_t last_check;
6589 int skipped = 0;
6590 mdk_rdev_t *rdev;
6591 char *desc;
6592
6593 /* just incase thread restarts... */
6594 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6595 return;
6596 if (mddev->ro) /* never try to sync a read-only array */
6597 return;
6598
6599 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6600 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6601 desc = "data-check";
6602 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6603 desc = "requested-resync";
6604 else
6605 desc = "resync";
6606 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6607 desc = "reshape";
6608 else
6609 desc = "recovery";
6610
6611 /* we overload curr_resync somewhat here.
6612 * 0 == not engaged in resync at all
6613 * 2 == checking that there is no conflict with another sync
6614 * 1 == like 2, but have yielded to allow conflicting resync to
6615 * commense
6616 * other == active in resync - this many blocks
6617 *
6618 * Before starting a resync we must have set curr_resync to
6619 * 2, and then checked that every "conflicting" array has curr_resync
6620 * less than ours. When we find one that is the same or higher
6621 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6622 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6623 * This will mean we have to start checking from the beginning again.
6624 *
6625 */
6626
6627 do {
6628 mddev->curr_resync = 2;
6629
6630 try_again:
6631 if (kthread_should_stop())
6632 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6633
6634 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6635 goto skip;
6636 for_each_mddev(mddev2, tmp) {
6637 if (mddev2 == mddev)
6638 continue;
6639 if (!mddev->parallel_resync
6640 && mddev2->curr_resync
6641 && match_mddev_units(mddev, mddev2)) {
6642 DEFINE_WAIT(wq);
6643 if (mddev < mddev2 && mddev->curr_resync == 2) {
6644 /* arbitrarily yield */
6645 mddev->curr_resync = 1;
6646 wake_up(&resync_wait);
6647 }
6648 if (mddev > mddev2 && mddev->curr_resync == 1)
6649 /* no need to wait here, we can wait the next
6650 * time 'round when curr_resync == 2
6651 */
6652 continue;
6653 /* We need to wait 'interruptible' so as not to
6654 * contribute to the load average, and not to
6655 * be caught by 'softlockup'
6656 */
6657 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6658 if (!kthread_should_stop() &&
6659 mddev2->curr_resync >= mddev->curr_resync) {
6660 printk(KERN_INFO "md: delaying %s of %s"
6661 " until %s has finished (they"
6662 " share one or more physical units)\n",
6663 desc, mdname(mddev), mdname(mddev2));
6664 mddev_put(mddev2);
6665 if (signal_pending(current))
6666 flush_signals(current);
6667 schedule();
6668 finish_wait(&resync_wait, &wq);
6669 goto try_again;
6670 }
6671 finish_wait(&resync_wait, &wq);
6672 }
6673 }
6674 } while (mddev->curr_resync < 2);
6675
6676 j = 0;
6677 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6678 /* resync follows the size requested by the personality,
6679 * which defaults to physical size, but can be virtual size
6680 */
6681 max_sectors = mddev->resync_max_sectors;
6682 mddev->resync_mismatches = 0;
6683 /* we don't use the checkpoint if there's a bitmap */
6684 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6685 j = mddev->resync_min;
6686 else if (!mddev->bitmap)
6687 j = mddev->recovery_cp;
6688
6689 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6690 max_sectors = mddev->dev_sectors;
6691 else {
6692 /* recovery follows the physical size of devices */
6693 max_sectors = mddev->dev_sectors;
6694 j = MaxSector;
6695 rcu_read_lock();
6696 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6697 if (rdev->raid_disk >= 0 &&
6698 !test_bit(Faulty, &rdev->flags) &&
6699 !test_bit(In_sync, &rdev->flags) &&
6700 rdev->recovery_offset < j)
6701 j = rdev->recovery_offset;
6702 rcu_read_unlock();
6703 }
6704
6705 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6706 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6707 " %d KB/sec/disk.\n", speed_min(mddev));
6708 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6709 "(but not more than %d KB/sec) for %s.\n",
6710 speed_max(mddev), desc);
6711
6712 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6713
6714 io_sectors = 0;
6715 for (m = 0; m < SYNC_MARKS; m++) {
6716 mark[m] = jiffies;
6717 mark_cnt[m] = io_sectors;
6718 }
6719 last_mark = 0;
6720 mddev->resync_mark = mark[last_mark];
6721 mddev->resync_mark_cnt = mark_cnt[last_mark];
6722
6723 /*
6724 * Tune reconstruction:
6725 */
6726 window = 32*(PAGE_SIZE/512);
6727 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6728 window/2,(unsigned long long) max_sectors/2);
6729
6730 atomic_set(&mddev->recovery_active, 0);
6731 last_check = 0;
6732
6733 if (j>2) {
6734 printk(KERN_INFO
6735 "md: resuming %s of %s from checkpoint.\n",
6736 desc, mdname(mddev));
6737 mddev->curr_resync = j;
6738 }
6739 mddev->curr_resync_completed = mddev->curr_resync;
6740
6741 while (j < max_sectors) {
6742 sector_t sectors;
6743
6744 skipped = 0;
6745
6746 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6747 ((mddev->curr_resync > mddev->curr_resync_completed &&
6748 (mddev->curr_resync - mddev->curr_resync_completed)
6749 > (max_sectors >> 4)) ||
6750 (j - mddev->curr_resync_completed)*2
6751 >= mddev->resync_max - mddev->curr_resync_completed
6752 )) {
6753 /* time to update curr_resync_completed */
6754 blk_unplug(mddev->queue);
6755 wait_event(mddev->recovery_wait,
6756 atomic_read(&mddev->recovery_active) == 0);
6757 mddev->curr_resync_completed =
6758 mddev->curr_resync;
6759 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6760 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6761 }
6762
6763 while (j >= mddev->resync_max && !kthread_should_stop()) {
6764 /* As this condition is controlled by user-space,
6765 * we can block indefinitely, so use '_interruptible'
6766 * to avoid triggering warnings.
6767 */
6768 flush_signals(current); /* just in case */
6769 wait_event_interruptible(mddev->recovery_wait,
6770 mddev->resync_max > j
6771 || kthread_should_stop());
6772 }
6773
6774 if (kthread_should_stop())
6775 goto interrupted;
6776
6777 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6778 currspeed < speed_min(mddev));
6779 if (sectors == 0) {
6780 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6781 goto out;
6782 }
6783
6784 if (!skipped) { /* actual IO requested */
6785 io_sectors += sectors;
6786 atomic_add(sectors, &mddev->recovery_active);
6787 }
6788
6789 j += sectors;
6790 if (j>1) mddev->curr_resync = j;
6791 mddev->curr_mark_cnt = io_sectors;
6792 if (last_check == 0)
6793 /* this is the earliers that rebuilt will be
6794 * visible in /proc/mdstat
6795 */
6796 md_new_event(mddev);
6797
6798 if (last_check + window > io_sectors || j == max_sectors)
6799 continue;
6800
6801 last_check = io_sectors;
6802
6803 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6804 break;
6805
6806 repeat:
6807 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6808 /* step marks */
6809 int next = (last_mark+1) % SYNC_MARKS;
6810
6811 mddev->resync_mark = mark[next];
6812 mddev->resync_mark_cnt = mark_cnt[next];
6813 mark[next] = jiffies;
6814 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6815 last_mark = next;
6816 }
6817
6818
6819 if (kthread_should_stop())
6820 goto interrupted;
6821
6822
6823 /*
6824 * this loop exits only if either when we are slower than
6825 * the 'hard' speed limit, or the system was IO-idle for
6826 * a jiffy.
6827 * the system might be non-idle CPU-wise, but we only care
6828 * about not overloading the IO subsystem. (things like an
6829 * e2fsck being done on the RAID array should execute fast)
6830 */
6831 blk_unplug(mddev->queue);
6832 cond_resched();
6833
6834 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6835 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6836
6837 if (currspeed > speed_min(mddev)) {
6838 if ((currspeed > speed_max(mddev)) ||
6839 !is_mddev_idle(mddev, 0)) {
6840 msleep(500);
6841 goto repeat;
6842 }
6843 }
6844 }
6845 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6846 /*
6847 * this also signals 'finished resyncing' to md_stop
6848 */
6849 out:
6850 blk_unplug(mddev->queue);
6851
6852 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6853
6854 /* tell personality that we are finished */
6855 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6856
6857 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6858 mddev->curr_resync > 2) {
6859 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6860 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6861 if (mddev->curr_resync >= mddev->recovery_cp) {
6862 printk(KERN_INFO
6863 "md: checkpointing %s of %s.\n",
6864 desc, mdname(mddev));
6865 mddev->recovery_cp = mddev->curr_resync;
6866 }
6867 } else
6868 mddev->recovery_cp = MaxSector;
6869 } else {
6870 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6871 mddev->curr_resync = MaxSector;
6872 rcu_read_lock();
6873 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6874 if (rdev->raid_disk >= 0 &&
6875 !test_bit(Faulty, &rdev->flags) &&
6876 !test_bit(In_sync, &rdev->flags) &&
6877 rdev->recovery_offset < mddev->curr_resync)
6878 rdev->recovery_offset = mddev->curr_resync;
6879 rcu_read_unlock();
6880 }
6881 }
6882 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6883
6884 skip:
6885 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6886 /* We completed so min/max setting can be forgotten if used. */
6887 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6888 mddev->resync_min = 0;
6889 mddev->resync_max = MaxSector;
6890 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6891 mddev->resync_min = mddev->curr_resync_completed;
6892 mddev->curr_resync = 0;
6893 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6894 mddev->curr_resync_completed = 0;
6895 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6896 wake_up(&resync_wait);
6897 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6898 md_wakeup_thread(mddev->thread);
6899 return;
6900
6901 interrupted:
6902 /*
6903 * got a signal, exit.
6904 */
6905 printk(KERN_INFO
6906 "md: md_do_sync() got signal ... exiting\n");
6907 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6908 goto out;
6909
6910 }
6911 EXPORT_SYMBOL_GPL(md_do_sync);
6912
6913
6914 static int remove_and_add_spares(mddev_t *mddev)
6915 {
6916 mdk_rdev_t *rdev;
6917 int spares = 0;
6918
6919 mddev->curr_resync_completed = 0;
6920
6921 list_for_each_entry(rdev, &mddev->disks, same_set)
6922 if (rdev->raid_disk >= 0 &&
6923 !test_bit(Blocked, &rdev->flags) &&
6924 (test_bit(Faulty, &rdev->flags) ||
6925 ! test_bit(In_sync, &rdev->flags)) &&
6926 atomic_read(&rdev->nr_pending)==0) {
6927 if (mddev->pers->hot_remove_disk(
6928 mddev, rdev->raid_disk)==0) {
6929 char nm[20];
6930 sprintf(nm,"rd%d", rdev->raid_disk);
6931 sysfs_remove_link(&mddev->kobj, nm);
6932 rdev->raid_disk = -1;
6933 }
6934 }
6935
6936 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
6937 list_for_each_entry(rdev, &mddev->disks, same_set) {
6938 if (rdev->raid_disk >= 0 &&
6939 !test_bit(In_sync, &rdev->flags) &&
6940 !test_bit(Blocked, &rdev->flags))
6941 spares++;
6942 if (rdev->raid_disk < 0
6943 && !test_bit(Faulty, &rdev->flags)) {
6944 rdev->recovery_offset = 0;
6945 if (mddev->pers->
6946 hot_add_disk(mddev, rdev) == 0) {
6947 char nm[20];
6948 sprintf(nm, "rd%d", rdev->raid_disk);
6949 if (sysfs_create_link(&mddev->kobj,
6950 &rdev->kobj, nm))
6951 printk(KERN_WARNING
6952 "md: cannot register "
6953 "%s for %s\n",
6954 nm, mdname(mddev));
6955 spares++;
6956 md_new_event(mddev);
6957 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6958 } else
6959 break;
6960 }
6961 }
6962 }
6963 return spares;
6964 }
6965 /*
6966 * This routine is regularly called by all per-raid-array threads to
6967 * deal with generic issues like resync and super-block update.
6968 * Raid personalities that don't have a thread (linear/raid0) do not
6969 * need this as they never do any recovery or update the superblock.
6970 *
6971 * It does not do any resync itself, but rather "forks" off other threads
6972 * to do that as needed.
6973 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6974 * "->recovery" and create a thread at ->sync_thread.
6975 * When the thread finishes it sets MD_RECOVERY_DONE
6976 * and wakeups up this thread which will reap the thread and finish up.
6977 * This thread also removes any faulty devices (with nr_pending == 0).
6978 *
6979 * The overall approach is:
6980 * 1/ if the superblock needs updating, update it.
6981 * 2/ If a recovery thread is running, don't do anything else.
6982 * 3/ If recovery has finished, clean up, possibly marking spares active.
6983 * 4/ If there are any faulty devices, remove them.
6984 * 5/ If array is degraded, try to add spares devices
6985 * 6/ If array has spares or is not in-sync, start a resync thread.
6986 */
6987 void md_check_recovery(mddev_t *mddev)
6988 {
6989 mdk_rdev_t *rdev;
6990
6991
6992 if (mddev->bitmap)
6993 bitmap_daemon_work(mddev);
6994
6995 if (mddev->ro)
6996 return;
6997
6998 if (signal_pending(current)) {
6999 if (mddev->pers->sync_request && !mddev->external) {
7000 printk(KERN_INFO "md: %s in immediate safe mode\n",
7001 mdname(mddev));
7002 mddev->safemode = 2;
7003 }
7004 flush_signals(current);
7005 }
7006
7007 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7008 return;
7009 if ( ! (
7010 (mddev->flags && !mddev->external) ||
7011 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7012 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7013 (mddev->external == 0 && mddev->safemode == 1) ||
7014 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7015 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7016 ))
7017 return;
7018
7019 if (mddev_trylock(mddev)) {
7020 int spares = 0;
7021
7022 if (mddev->ro) {
7023 /* Only thing we do on a ro array is remove
7024 * failed devices.
7025 */
7026 remove_and_add_spares(mddev);
7027 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7028 goto unlock;
7029 }
7030
7031 if (!mddev->external) {
7032 int did_change = 0;
7033 spin_lock_irq(&mddev->write_lock);
7034 if (mddev->safemode &&
7035 !atomic_read(&mddev->writes_pending) &&
7036 !mddev->in_sync &&
7037 mddev->recovery_cp == MaxSector) {
7038 mddev->in_sync = 1;
7039 did_change = 1;
7040 if (mddev->persistent)
7041 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7042 }
7043 if (mddev->safemode == 1)
7044 mddev->safemode = 0;
7045 spin_unlock_irq(&mddev->write_lock);
7046 if (did_change)
7047 sysfs_notify_dirent(mddev->sysfs_state);
7048 }
7049
7050 if (mddev->flags)
7051 md_update_sb(mddev, 0);
7052
7053 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7054 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7055 /* resync/recovery still happening */
7056 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7057 goto unlock;
7058 }
7059 if (mddev->sync_thread) {
7060 /* resync has finished, collect result */
7061 md_unregister_thread(mddev->sync_thread);
7062 mddev->sync_thread = NULL;
7063 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7064 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7065 /* success...*/
7066 /* activate any spares */
7067 if (mddev->pers->spare_active(mddev))
7068 sysfs_notify(&mddev->kobj, NULL,
7069 "degraded");
7070 }
7071 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7072 mddev->pers->finish_reshape)
7073 mddev->pers->finish_reshape(mddev);
7074 md_update_sb(mddev, 1);
7075
7076 /* if array is no-longer degraded, then any saved_raid_disk
7077 * information must be scrapped
7078 */
7079 if (!mddev->degraded)
7080 list_for_each_entry(rdev, &mddev->disks, same_set)
7081 rdev->saved_raid_disk = -1;
7082
7083 mddev->recovery = 0;
7084 /* flag recovery needed just to double check */
7085 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7086 sysfs_notify_dirent(mddev->sysfs_action);
7087 md_new_event(mddev);
7088 goto unlock;
7089 }
7090 /* Set RUNNING before clearing NEEDED to avoid
7091 * any transients in the value of "sync_action".
7092 */
7093 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7094 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7095 /* Clear some bits that don't mean anything, but
7096 * might be left set
7097 */
7098 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7099 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7100
7101 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7102 goto unlock;
7103 /* no recovery is running.
7104 * remove any failed drives, then
7105 * add spares if possible.
7106 * Spare are also removed and re-added, to allow
7107 * the personality to fail the re-add.
7108 */
7109
7110 if (mddev->reshape_position != MaxSector) {
7111 if (mddev->pers->check_reshape == NULL ||
7112 mddev->pers->check_reshape(mddev) != 0)
7113 /* Cannot proceed */
7114 goto unlock;
7115 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7116 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7117 } else if ((spares = remove_and_add_spares(mddev))) {
7118 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7119 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7120 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7121 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7122 } else if (mddev->recovery_cp < MaxSector) {
7123 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7124 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7125 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7126 /* nothing to be done ... */
7127 goto unlock;
7128
7129 if (mddev->pers->sync_request) {
7130 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7131 /* We are adding a device or devices to an array
7132 * which has the bitmap stored on all devices.
7133 * So make sure all bitmap pages get written
7134 */
7135 bitmap_write_all(mddev->bitmap);
7136 }
7137 mddev->sync_thread = md_register_thread(md_do_sync,
7138 mddev,
7139 "resync");
7140 if (!mddev->sync_thread) {
7141 printk(KERN_ERR "%s: could not start resync"
7142 " thread...\n",
7143 mdname(mddev));
7144 /* leave the spares where they are, it shouldn't hurt */
7145 mddev->recovery = 0;
7146 } else
7147 md_wakeup_thread(mddev->sync_thread);
7148 sysfs_notify_dirent(mddev->sysfs_action);
7149 md_new_event(mddev);
7150 }
7151 unlock:
7152 if (!mddev->sync_thread) {
7153 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7154 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7155 &mddev->recovery))
7156 if (mddev->sysfs_action)
7157 sysfs_notify_dirent(mddev->sysfs_action);
7158 }
7159 mddev_unlock(mddev);
7160 }
7161 }
7162
7163 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
7164 {
7165 sysfs_notify_dirent(rdev->sysfs_state);
7166 wait_event_timeout(rdev->blocked_wait,
7167 !test_bit(Blocked, &rdev->flags),
7168 msecs_to_jiffies(5000));
7169 rdev_dec_pending(rdev, mddev);
7170 }
7171 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7172
7173 static int md_notify_reboot(struct notifier_block *this,
7174 unsigned long code, void *x)
7175 {
7176 struct list_head *tmp;
7177 mddev_t *mddev;
7178
7179 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
7180
7181 printk(KERN_INFO "md: stopping all md devices.\n");
7182
7183 for_each_mddev(mddev, tmp)
7184 if (mddev_trylock(mddev)) {
7185 /* Force a switch to readonly even array
7186 * appears to still be in use. Hence
7187 * the '100'.
7188 */
7189 md_set_readonly(mddev, 100);
7190 mddev_unlock(mddev);
7191 }
7192 /*
7193 * certain more exotic SCSI devices are known to be
7194 * volatile wrt too early system reboots. While the
7195 * right place to handle this issue is the given
7196 * driver, we do want to have a safe RAID driver ...
7197 */
7198 mdelay(1000*1);
7199 }
7200 return NOTIFY_DONE;
7201 }
7202
7203 static struct notifier_block md_notifier = {
7204 .notifier_call = md_notify_reboot,
7205 .next = NULL,
7206 .priority = INT_MAX, /* before any real devices */
7207 };
7208
7209 static void md_geninit(void)
7210 {
7211 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
7212
7213 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
7214 }
7215
7216 static int __init md_init(void)
7217 {
7218 if (register_blkdev(MD_MAJOR, "md"))
7219 return -1;
7220 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
7221 unregister_blkdev(MD_MAJOR, "md");
7222 return -1;
7223 }
7224 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
7225 md_probe, NULL, NULL);
7226 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
7227 md_probe, NULL, NULL);
7228
7229 register_reboot_notifier(&md_notifier);
7230 raid_table_header = register_sysctl_table(raid_root_table);
7231
7232 md_geninit();
7233 return 0;
7234 }
7235
7236
7237 #ifndef MODULE
7238
7239 /*
7240 * Searches all registered partitions for autorun RAID arrays
7241 * at boot time.
7242 */
7243
7244 static LIST_HEAD(all_detected_devices);
7245 struct detected_devices_node {
7246 struct list_head list;
7247 dev_t dev;
7248 };
7249
7250 void md_autodetect_dev(dev_t dev)
7251 {
7252 struct detected_devices_node *node_detected_dev;
7253
7254 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
7255 if (node_detected_dev) {
7256 node_detected_dev->dev = dev;
7257 list_add_tail(&node_detected_dev->list, &all_detected_devices);
7258 } else {
7259 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
7260 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
7261 }
7262 }
7263
7264
7265 static void autostart_arrays(int part)
7266 {
7267 mdk_rdev_t *rdev;
7268 struct detected_devices_node *node_detected_dev;
7269 dev_t dev;
7270 int i_scanned, i_passed;
7271
7272 i_scanned = 0;
7273 i_passed = 0;
7274
7275 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
7276
7277 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
7278 i_scanned++;
7279 node_detected_dev = list_entry(all_detected_devices.next,
7280 struct detected_devices_node, list);
7281 list_del(&node_detected_dev->list);
7282 dev = node_detected_dev->dev;
7283 kfree(node_detected_dev);
7284 rdev = md_import_device(dev,0, 90);
7285 if (IS_ERR(rdev))
7286 continue;
7287
7288 if (test_bit(Faulty, &rdev->flags)) {
7289 MD_BUG();
7290 continue;
7291 }
7292 set_bit(AutoDetected, &rdev->flags);
7293 list_add(&rdev->same_set, &pending_raid_disks);
7294 i_passed++;
7295 }
7296
7297 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
7298 i_scanned, i_passed);
7299
7300 autorun_devices(part);
7301 }
7302
7303 #endif /* !MODULE */
7304
7305 static __exit void md_exit(void)
7306 {
7307 mddev_t *mddev;
7308 struct list_head *tmp;
7309
7310 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
7311 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
7312
7313 unregister_blkdev(MD_MAJOR,"md");
7314 unregister_blkdev(mdp_major, "mdp");
7315 unregister_reboot_notifier(&md_notifier);
7316 unregister_sysctl_table(raid_table_header);
7317 remove_proc_entry("mdstat", NULL);
7318 for_each_mddev(mddev, tmp) {
7319 export_array(mddev);
7320 mddev->hold_active = 0;
7321 }
7322 }
7323
7324 subsys_initcall(md_init);
7325 module_exit(md_exit)
7326
7327 static int get_ro(char *buffer, struct kernel_param *kp)
7328 {
7329 return sprintf(buffer, "%d", start_readonly);
7330 }
7331 static int set_ro(const char *val, struct kernel_param *kp)
7332 {
7333 char *e;
7334 int num = simple_strtoul(val, &e, 10);
7335 if (*val && (*e == '\0' || *e == '\n')) {
7336 start_readonly = num;
7337 return 0;
7338 }
7339 return -EINVAL;
7340 }
7341
7342 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
7343 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
7344
7345 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
7346
7347 EXPORT_SYMBOL(register_md_personality);
7348 EXPORT_SYMBOL(unregister_md_personality);
7349 EXPORT_SYMBOL(md_error);
7350 EXPORT_SYMBOL(md_done_sync);
7351 EXPORT_SYMBOL(md_write_start);
7352 EXPORT_SYMBOL(md_write_end);
7353 EXPORT_SYMBOL(md_register_thread);
7354 EXPORT_SYMBOL(md_unregister_thread);
7355 EXPORT_SYMBOL(md_wakeup_thread);
7356 EXPORT_SYMBOL(md_check_recovery);
7357 MODULE_LICENSE("GPL");
7358 MODULE_DESCRIPTION("MD RAID framework");
7359 MODULE_ALIAS("md");
7360 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
kernel source for telekom puls (alcatel/mediatek)